linux/drivers/scsi/scsi_lib.c

3225 lines
84 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0-only
/*
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
* Copyright (C) 1999 Eric Youngdale
* Copyright (C) 2014 Christoph Hellwig
*
* SCSI queueing library.
* Initial versions: Eric Youngdale (eric@andante.org).
* Based upon conversations with large numbers
* of people at Linux Expo.
*/
#include <linux/bio.h>
#include <linux/bitops.h>
#include <linux/blkdev.h>
#include <linux/completion.h>
#include <linux/kernel.h>
#include <linux/export.h>
#include <linux/init.h>
#include <linux/pci.h>
#include <linux/delay.h>
#include <linux/hardirq.h>
#include <linux/scatterlist.h>
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
#include <linux/blk-mq.h>
#include <linux/ratelimit.h>
#include <asm/unaligned.h>
#include <scsi/scsi.h>
#include <scsi/scsi_cmnd.h>
#include <scsi/scsi_dbg.h>
#include <scsi/scsi_device.h>
#include <scsi/scsi_driver.h>
#include <scsi/scsi_eh.h>
#include <scsi/scsi_host.h>
#include <scsi/scsi_transport.h> /* __scsi_init_queue() */
#include <scsi/scsi_dh.h>
#include <trace/events/scsi.h>
#include "scsi_debugfs.h"
#include "scsi_priv.h"
#include "scsi_logging.h"
/*
* Size of integrity metadata is usually small, 1 inline sg should
* cover normal cases.
*/
#ifdef CONFIG_ARCH_NO_SG_CHAIN
#define SCSI_INLINE_PROT_SG_CNT 0
#define SCSI_INLINE_SG_CNT 0
#else
#define SCSI_INLINE_PROT_SG_CNT 1
#define SCSI_INLINE_SG_CNT 2
#endif
static struct kmem_cache *scsi_sense_cache;
static DEFINE_MUTEX(scsi_sense_cache_mutex);
static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd);
int scsi_init_sense_cache(struct Scsi_Host *shost)
{
int ret = 0;
mutex_lock(&scsi_sense_cache_mutex);
if (!scsi_sense_cache) {
scsi_sense_cache =
kmem_cache_create_usercopy("scsi_sense_cache",
SCSI_SENSE_BUFFERSIZE, 0, SLAB_HWCACHE_ALIGN,
0, SCSI_SENSE_BUFFERSIZE, NULL);
if (!scsi_sense_cache)
ret = -ENOMEM;
}
mutex_unlock(&scsi_sense_cache_mutex);
return ret;
}
[SCSI] bidirectional command support At the block level bidi request uses req->next_rq pointer for a second bidi_read request. At Scsi-midlayer a second scsi_data_buffer structure is used for the bidi_read part. This bidi scsi_data_buffer is put on request->next_rq->special. Struct scsi_cmnd is not changed. - Define scsi_bidi_cmnd() to return true if it is a bidi request and a second sgtable was allocated. - Define scsi_in()/scsi_out() to return the in or out scsi_data_buffer from this command This API is to isolate users from the mechanics of bidi. - Define scsi_end_bidi_request() to do what scsi_end_request() does but for a bidi request. This is necessary because bidi commands are a bit tricky here. (See comments in body) - scsi_release_buffers() will also release the bidi_read scsi_data_buffer - scsi_io_completion() on bidi commands will now call scsi_end_bidi_request() and return. - The previous work done in scsi_init_io() is now done in a new scsi_init_sgtable() (which is 99% identical to old scsi_init_io()) The new scsi_init_io() will call the above twice if needed also for the bidi_read command. Only at this point is a command bidi. - In scsi_error.c at scsi_eh_prep/restore_cmnd() make sure bidi-lld is not confused by a get-sense command that looks like bidi. This is done by puting NULL at request->next_rq, and restoring. [jejb: update to sg_table and resolve conflicts also update to blk-end-request and resolve conflicts] Signed-off-by: Boaz Harrosh <bharrosh@panasas.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2007-12-13 11:50:53 +00:00
/*
* When to reinvoke queueing after a resource shortage. It's 3 msecs to
* not change behaviour from the previous unplug mechanism, experimentation
* may prove this needs changing.
*/
#define SCSI_QUEUE_DELAY 3
static void
scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
{
struct Scsi_Host *host = cmd->device->host;
struct scsi_device *device = cmd->device;
[SCSI] Add helper code so transport classes/driver can control queueing (v3) SCSI-ml manages the queueing limits for the device and host, but does not do so at the target level. However something something similar can come in userful when a driver is transitioning a transport object to the the blocked state, becuase at that time we do not want to queue io and we do not want the queuecommand to be called again. The patch adds code similar to the exisiting SCSI_ML_*BUSY handlers. You can now return SCSI_MLQUEUE_TARGET_BUSY when we hit a transport level queueing issue like the hw cannot allocate some resource at the iscsi session/connection level, or the target has temporarily closed or shrunk the queueing window, or if we are transitioning to the blocked state. bnx2i, when they rework their firmware according to netdev developers requests, will also need to be able to limit queueing at this level. bnx2i will hook into libiscsi, but will allocate a scsi host per netdevice/hba, so unlike pure software iscsi/iser which is allocating a host per session, it cannot set the scsi_host->can_queue and return SCSI_MLQUEUE_HOST_BUSY to reflect queueing limits on the transport. The iscsi class/driver can also set a scsi_target->can_queue value which reflects the max commands the driver/class can support. For iscsi this reflects the number of commands we can support for each session due to session/connection hw limits, driver limits, and to also reflect the session/targets's queueing window. Changes: v1 - initial patch. v2 - Fix scsi_run_queue handling of multiple blocked targets. Previously we would break from the main loop if a device was added back on the starved list. We now run over the list and check if any target is blocked. v3 - Rediff for scsi-misc. Signed-off-by: Mike Christie <michaelc@cs.wisc.edu> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-08-17 20:24:38 +00:00
struct scsi_target *starget = scsi_target(device);
/*
* Set the appropriate busy bit for the device/host.
*
* If the host/device isn't busy, assume that something actually
* completed, and that we should be able to queue a command now.
*
* Note that the prior mid-layer assumption that any host could
* always queue at least one command is now broken. The mid-layer
* will implement a user specifiable stall (see
* scsi_host.max_host_blocked and scsi_device.max_device_blocked)
* if a command is requeued with no other commands outstanding
* either for the device or for the host.
*/
[SCSI] Add helper code so transport classes/driver can control queueing (v3) SCSI-ml manages the queueing limits for the device and host, but does not do so at the target level. However something something similar can come in userful when a driver is transitioning a transport object to the the blocked state, becuase at that time we do not want to queue io and we do not want the queuecommand to be called again. The patch adds code similar to the exisiting SCSI_ML_*BUSY handlers. You can now return SCSI_MLQUEUE_TARGET_BUSY when we hit a transport level queueing issue like the hw cannot allocate some resource at the iscsi session/connection level, or the target has temporarily closed or shrunk the queueing window, or if we are transitioning to the blocked state. bnx2i, when they rework their firmware according to netdev developers requests, will also need to be able to limit queueing at this level. bnx2i will hook into libiscsi, but will allocate a scsi host per netdevice/hba, so unlike pure software iscsi/iser which is allocating a host per session, it cannot set the scsi_host->can_queue and return SCSI_MLQUEUE_HOST_BUSY to reflect queueing limits on the transport. The iscsi class/driver can also set a scsi_target->can_queue value which reflects the max commands the driver/class can support. For iscsi this reflects the number of commands we can support for each session due to session/connection hw limits, driver limits, and to also reflect the session/targets's queueing window. Changes: v1 - initial patch. v2 - Fix scsi_run_queue handling of multiple blocked targets. Previously we would break from the main loop if a device was added back on the starved list. We now run over the list and check if any target is blocked. v3 - Rediff for scsi-misc. Signed-off-by: Mike Christie <michaelc@cs.wisc.edu> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-08-17 20:24:38 +00:00
switch (reason) {
case SCSI_MLQUEUE_HOST_BUSY:
atomic_set(&host->host_blocked, host->max_host_blocked);
[SCSI] Add helper code so transport classes/driver can control queueing (v3) SCSI-ml manages the queueing limits for the device and host, but does not do so at the target level. However something something similar can come in userful when a driver is transitioning a transport object to the the blocked state, becuase at that time we do not want to queue io and we do not want the queuecommand to be called again. The patch adds code similar to the exisiting SCSI_ML_*BUSY handlers. You can now return SCSI_MLQUEUE_TARGET_BUSY when we hit a transport level queueing issue like the hw cannot allocate some resource at the iscsi session/connection level, or the target has temporarily closed or shrunk the queueing window, or if we are transitioning to the blocked state. bnx2i, when they rework their firmware according to netdev developers requests, will also need to be able to limit queueing at this level. bnx2i will hook into libiscsi, but will allocate a scsi host per netdevice/hba, so unlike pure software iscsi/iser which is allocating a host per session, it cannot set the scsi_host->can_queue and return SCSI_MLQUEUE_HOST_BUSY to reflect queueing limits on the transport. The iscsi class/driver can also set a scsi_target->can_queue value which reflects the max commands the driver/class can support. For iscsi this reflects the number of commands we can support for each session due to session/connection hw limits, driver limits, and to also reflect the session/targets's queueing window. Changes: v1 - initial patch. v2 - Fix scsi_run_queue handling of multiple blocked targets. Previously we would break from the main loop if a device was added back on the starved list. We now run over the list and check if any target is blocked. v3 - Rediff for scsi-misc. Signed-off-by: Mike Christie <michaelc@cs.wisc.edu> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-08-17 20:24:38 +00:00
break;
case SCSI_MLQUEUE_DEVICE_BUSY:
[SCSI] scsi_lib: pause between error retries During cable pull tests on our 16G FC adapter, we are seeing errors, typically reads to close targets, which fail due to CRC or framing errors caused by the cable being pull (return status DID_ERROR). The adapter detects the error on one of the first frames received, marks the FC exchange as dead (further frames go to bit bucket) and signals the host of the error. This action is so quick, and coupled with fast host CPUs, creates a scenario in which the midlayer sees the failure and retries the io almost immediately. We've seen link traces with the retry on the link while the original i/o is still being processed by the target. We're also seeing the time window for the "link to pull-apart" and the physical interface to report disconnected to be in the few millisecond range. Which means, we're encountering scenarios where the full retry count is exhausted (all with error) by the midlayer before the link disconnect state is detected. We looked at 8G FC behavior and occasionally see the same behavior, but as the link was slower, it rarely could exhaust all retries before the link reported disconnect. What is needed is a slight delay between io retries due to DID_ERROR to cover this error. It is inappropriate to put this delay in the driver, as the error is indistinguishable from other link-related errors, nor does the driver track whether the io is a retry or not. This is also easier than tracking between-io-error bursts that are seen in this scenario. The patch below updates the retry path so that it inserts a delay as if the target was busy. The busy delay is on the order of 6ms. This delay is sufficient to ensure the link down condition is reported before the retry count is exhausted (at most 1 retry is seen). Signed-off-by: Alex Iannicelli <alex.iannicelli@emulex.com> Signed-off-by: James Smart <james.smart@emulex.com> Signed-off-by: James Bottomley <JBottomley@Parallels.com>
2011-07-06 16:45:17 +00:00
case SCSI_MLQUEUE_EH_RETRY:
atomic_set(&device->device_blocked,
device->max_device_blocked);
[SCSI] Add helper code so transport classes/driver can control queueing (v3) SCSI-ml manages the queueing limits for the device and host, but does not do so at the target level. However something something similar can come in userful when a driver is transitioning a transport object to the the blocked state, becuase at that time we do not want to queue io and we do not want the queuecommand to be called again. The patch adds code similar to the exisiting SCSI_ML_*BUSY handlers. You can now return SCSI_MLQUEUE_TARGET_BUSY when we hit a transport level queueing issue like the hw cannot allocate some resource at the iscsi session/connection level, or the target has temporarily closed or shrunk the queueing window, or if we are transitioning to the blocked state. bnx2i, when they rework their firmware according to netdev developers requests, will also need to be able to limit queueing at this level. bnx2i will hook into libiscsi, but will allocate a scsi host per netdevice/hba, so unlike pure software iscsi/iser which is allocating a host per session, it cannot set the scsi_host->can_queue and return SCSI_MLQUEUE_HOST_BUSY to reflect queueing limits on the transport. The iscsi class/driver can also set a scsi_target->can_queue value which reflects the max commands the driver/class can support. For iscsi this reflects the number of commands we can support for each session due to session/connection hw limits, driver limits, and to also reflect the session/targets's queueing window. Changes: v1 - initial patch. v2 - Fix scsi_run_queue handling of multiple blocked targets. Previously we would break from the main loop if a device was added back on the starved list. We now run over the list and check if any target is blocked. v3 - Rediff for scsi-misc. Signed-off-by: Mike Christie <michaelc@cs.wisc.edu> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-08-17 20:24:38 +00:00
break;
case SCSI_MLQUEUE_TARGET_BUSY:
atomic_set(&starget->target_blocked,
starget->max_target_blocked);
[SCSI] Add helper code so transport classes/driver can control queueing (v3) SCSI-ml manages the queueing limits for the device and host, but does not do so at the target level. However something something similar can come in userful when a driver is transitioning a transport object to the the blocked state, becuase at that time we do not want to queue io and we do not want the queuecommand to be called again. The patch adds code similar to the exisiting SCSI_ML_*BUSY handlers. You can now return SCSI_MLQUEUE_TARGET_BUSY when we hit a transport level queueing issue like the hw cannot allocate some resource at the iscsi session/connection level, or the target has temporarily closed or shrunk the queueing window, or if we are transitioning to the blocked state. bnx2i, when they rework their firmware according to netdev developers requests, will also need to be able to limit queueing at this level. bnx2i will hook into libiscsi, but will allocate a scsi host per netdevice/hba, so unlike pure software iscsi/iser which is allocating a host per session, it cannot set the scsi_host->can_queue and return SCSI_MLQUEUE_HOST_BUSY to reflect queueing limits on the transport. The iscsi class/driver can also set a scsi_target->can_queue value which reflects the max commands the driver/class can support. For iscsi this reflects the number of commands we can support for each session due to session/connection hw limits, driver limits, and to also reflect the session/targets's queueing window. Changes: v1 - initial patch. v2 - Fix scsi_run_queue handling of multiple blocked targets. Previously we would break from the main loop if a device was added back on the starved list. We now run over the list and check if any target is blocked. v3 - Rediff for scsi-misc. Signed-off-by: Mike Christie <michaelc@cs.wisc.edu> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-08-17 20:24:38 +00:00
break;
}
}
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
{
if (cmd->request->rq_flags & RQF_DONTPREP) {
cmd->request->rq_flags &= ~RQF_DONTPREP;
scsi_mq_uninit_cmd(cmd);
} else {
WARN_ON_ONCE(true);
}
blk_mq_requeue_request(cmd->request, true);
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
}
/**
* __scsi_queue_insert - private queue insertion
* @cmd: The SCSI command being requeued
* @reason: The reason for the requeue
* @unbusy: Whether the queue should be unbusied
*
* This is a private queue insertion. The public interface
* scsi_queue_insert() always assumes the queue should be unbusied
* because it's always called before the completion. This function is
* for a requeue after completion, which should only occur in this
* file.
*/
static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, bool unbusy)
{
struct scsi_device *device = cmd->device;
SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
"Inserting command %p into mlqueue\n", cmd));
scsi_set_blocked(cmd, reason);
/*
* Decrement the counters, since these commands are no longer
* active on the host/device.
*/
if (unbusy)
scsi_device_unbusy(device, cmd);
/*
* Requeue this command. It will go before all other commands
* that are already in the queue. Schedule requeue work under
* lock such that the kblockd_schedule_work() call happens
* before blk_cleanup_queue() finishes.
*/
cmd->result = 0;
blk_mq_requeue_request(cmd->request, true);
}
/**
* scsi_queue_insert - Reinsert a command in the queue.
* @cmd: command that we are adding to queue.
* @reason: why we are inserting command to queue.
*
* We do this for one of two cases. Either the host is busy and it cannot accept
* any more commands for the time being, or the device returned QUEUE_FULL and
* can accept no more commands.
*
* Context: This could be called either from an interrupt context or a normal
* process context.
*/
void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
{
__scsi_queue_insert(cmd, reason, true);
}
/**
* __scsi_execute - insert request and wait for the result
* @sdev: scsi device
* @cmd: scsi command
* @data_direction: data direction
* @buffer: data buffer
* @bufflen: len of buffer
* @sense: optional sense buffer
* @sshdr: optional decoded sense header
* @timeout: request timeout in HZ
* @retries: number of times to retry request
* @flags: flags for ->cmd_flags
* @rq_flags: flags for ->rq_flags
* @resid: optional residual length
*
* Returns the scsi_cmnd result field if a command was executed, or a negative
* Linux error code if we didn't get that far.
*/
int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
int data_direction, void *buffer, unsigned bufflen,
unsigned char *sense, struct scsi_sense_hdr *sshdr,
int timeout, int retries, u64 flags, req_flags_t rq_flags,
int *resid)
{
struct request *req;
struct scsi_request *rq;
int ret;
req = blk_get_request(sdev->request_queue,
data_direction == DMA_TO_DEVICE ?
REQ_OP_DRV_OUT : REQ_OP_DRV_IN,
rq_flags & RQF_PM ? BLK_MQ_REQ_PM : 0);
if (IS_ERR(req))
return PTR_ERR(req);
rq = scsi_req(req);
if (bufflen) {
ret = blk_rq_map_kern(sdev->request_queue, req,
buffer, bufflen, GFP_NOIO);
if (ret)
goto out;
}
rq->cmd_len = COMMAND_SIZE(cmd[0]);
memcpy(rq->cmd, cmd, rq->cmd_len);
rq->retries = retries;
req->timeout = timeout;
req->cmd_flags |= flags;
req->rq_flags |= rq_flags | RQF_QUIET;
/*
* head injection *required* here otherwise quiesce won't work
*/
blk_execute_rq(NULL, req, 1);
/*
* Some devices (USB mass-storage in particular) may transfer
* garbage data together with a residue indicating that the data
* is invalid. Prevent the garbage from being misinterpreted
* and prevent security leaks by zeroing out the excess data.
*/
if (unlikely(rq->resid_len > 0 && rq->resid_len <= bufflen))
memset(buffer + (bufflen - rq->resid_len), 0, rq->resid_len);
if (resid)
*resid = rq->resid_len;
if (sense && rq->sense_len)
memcpy(sense, rq->sense, SCSI_SENSE_BUFFERSIZE);
if (sshdr)
scsi_normalize_sense(rq->sense, rq->sense_len, sshdr);
ret = rq->result;
out:
blk_put_request(req);
return ret;
}
EXPORT_SYMBOL(__scsi_execute);
/*
* Wake up the error handler if necessary. Avoid as follows that the error
* handler is not woken up if host in-flight requests number ==
* shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
* with an RCU read lock in this function to ensure that this function in
* its entirety either finishes before scsi_eh_scmd_add() increases the
* host_failed counter or that it notices the shost state change made by
* scsi_eh_scmd_add().
*/
static void scsi_dec_host_busy(struct Scsi_Host *shost, struct scsi_cmnd *cmd)
{
unsigned long flags;
rcu_read_lock();
__clear_bit(SCMD_STATE_INFLIGHT, &cmd->state);
if (unlikely(scsi_host_in_recovery(shost))) {
spin_lock_irqsave(shost->host_lock, flags);
if (shost->host_failed || shost->host_eh_scheduled)
scsi_eh_wakeup(shost);
spin_unlock_irqrestore(shost->host_lock, flags);
}
rcu_read_unlock();
}
void scsi_device_unbusy(struct scsi_device *sdev, struct scsi_cmnd *cmd)
{
struct Scsi_Host *shost = sdev->host;
struct scsi_target *starget = scsi_target(sdev);
scsi_dec_host_busy(shost, cmd);
if (starget->can_queue > 0)
atomic_dec(&starget->target_busy);
scsi: core: Replace sdev->device_busy with sbitmap SCSI currently uses an atomic variable to track queue depth for each attached device. The queue depth depends on many factors such as transport type and device implementation. In addition, the SCSI device queue depth is not a static entity but changes over time as a result of congestion management. While blk-mq currently tracks queue depth for each hctx, it can't easily be changed to accommodate the SCSI per-device requirement. The current approach of using an atomic variable doesn't scale well when there are lots of CPU cores and the disk is very fast. IOPS can be substantially impacted by the atomic in the hot path. Replace the atomic variable sdev->device_busy with an sbitmap for tracking the SCSI device queue depth. It has been observed that IOPS is improved ~30% by this patchset in the following test: 1) test machine(32 logical CPU cores) Thread(s) per core: 2 Core(s) per socket: 8 Socket(s): 2 NUMA node(s): 2 Model name: Intel(R) Xeon(R) Silver 4110 CPU @ 2.10GHz 2) setup scsi_debug: modprobe scsi_debug virtual_gb=128 max_luns=1 submit_queues=32 delay=0 max_queue=256 3) fio script: fio --rw=randread --size=128G --direct=1 --ioengine=libaio --iodepth=2048 \ --numjobs=32 --bs=4k --group_reporting=1 --group_reporting=1 --runtime=60 \ --loops=10000 --name=job1 --filename=/dev/sdN [mkp: fix device_busy reference in mpt3sas] Link: https://lore.kernel.org/r/20210122023317.687987-14-ming.lei@redhat.com Link: https://lore.kernel.org/linux-block/20200119071432.18558-6-ming.lei@redhat.com/ Cc: Omar Sandoval <osandov@fb.com> Cc: Kashyap Desai <kashyap.desai@broadcom.com> Cc: Sumanesh Samanta <sumanesh.samanta@broadcom.com> Cc: Ewan D. Milne <emilne@redhat.com> Cc: Hannes Reinecke <hare@suse.de> Tested-by: Sumanesh Samanta <sumanesh.samanta@broadcom.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2021-01-22 02:33:17 +00:00
sbitmap_put(&sdev->budget_map, cmd->budget_token);
cmd->budget_token = -1;
}
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
static void scsi_kick_queue(struct request_queue *q)
{
blk_mq_run_hw_queues(q, false);
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
}
/*
* Called for single_lun devices on IO completion. Clear starget_sdev_user,
* and call blk_run_queue for all the scsi_devices on the target -
* including current_sdev first.
*
* Called with *no* scsi locks held.
*/
static void scsi_single_lun_run(struct scsi_device *current_sdev)
{
struct Scsi_Host *shost = current_sdev->host;
struct scsi_device *sdev, *tmp;
struct scsi_target *starget = scsi_target(current_sdev);
unsigned long flags;
spin_lock_irqsave(shost->host_lock, flags);
starget->starget_sdev_user = NULL;
spin_unlock_irqrestore(shost->host_lock, flags);
/*
* Call blk_run_queue for all LUNs on the target, starting with
* current_sdev. We race with others (to set starget_sdev_user),
* but in most cases, we will be first. Ideally, each LU on the
* target would get some limited time or requests on the target.
*/
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
scsi_kick_queue(current_sdev->request_queue);
spin_lock_irqsave(shost->host_lock, flags);
if (starget->starget_sdev_user)
goto out;
list_for_each_entry_safe(sdev, tmp, &starget->devices,
same_target_siblings) {
if (sdev == current_sdev)
continue;
if (scsi_device_get(sdev))
continue;
spin_unlock_irqrestore(shost->host_lock, flags);
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
scsi_kick_queue(sdev->request_queue);
spin_lock_irqsave(shost->host_lock, flags);
scsi_device_put(sdev);
}
out:
spin_unlock_irqrestore(shost->host_lock, flags);
}
static inline bool scsi_device_is_busy(struct scsi_device *sdev)
{
if (scsi_device_busy(sdev) >= sdev->queue_depth)
return true;
if (atomic_read(&sdev->device_blocked) > 0)
return true;
return false;
}
static inline bool scsi_target_is_busy(struct scsi_target *starget)
[SCSI] Add helper code so transport classes/driver can control queueing (v3) SCSI-ml manages the queueing limits for the device and host, but does not do so at the target level. However something something similar can come in userful when a driver is transitioning a transport object to the the blocked state, becuase at that time we do not want to queue io and we do not want the queuecommand to be called again. The patch adds code similar to the exisiting SCSI_ML_*BUSY handlers. You can now return SCSI_MLQUEUE_TARGET_BUSY when we hit a transport level queueing issue like the hw cannot allocate some resource at the iscsi session/connection level, or the target has temporarily closed or shrunk the queueing window, or if we are transitioning to the blocked state. bnx2i, when they rework their firmware according to netdev developers requests, will also need to be able to limit queueing at this level. bnx2i will hook into libiscsi, but will allocate a scsi host per netdevice/hba, so unlike pure software iscsi/iser which is allocating a host per session, it cannot set the scsi_host->can_queue and return SCSI_MLQUEUE_HOST_BUSY to reflect queueing limits on the transport. The iscsi class/driver can also set a scsi_target->can_queue value which reflects the max commands the driver/class can support. For iscsi this reflects the number of commands we can support for each session due to session/connection hw limits, driver limits, and to also reflect the session/targets's queueing window. Changes: v1 - initial patch. v2 - Fix scsi_run_queue handling of multiple blocked targets. Previously we would break from the main loop if a device was added back on the starved list. We now run over the list and check if any target is blocked. v3 - Rediff for scsi-misc. Signed-off-by: Mike Christie <michaelc@cs.wisc.edu> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-08-17 20:24:38 +00:00
{
if (starget->can_queue > 0) {
if (atomic_read(&starget->target_busy) >= starget->can_queue)
return true;
if (atomic_read(&starget->target_blocked) > 0)
return true;
}
return false;
[SCSI] Add helper code so transport classes/driver can control queueing (v3) SCSI-ml manages the queueing limits for the device and host, but does not do so at the target level. However something something similar can come in userful when a driver is transitioning a transport object to the the blocked state, becuase at that time we do not want to queue io and we do not want the queuecommand to be called again. The patch adds code similar to the exisiting SCSI_ML_*BUSY handlers. You can now return SCSI_MLQUEUE_TARGET_BUSY when we hit a transport level queueing issue like the hw cannot allocate some resource at the iscsi session/connection level, or the target has temporarily closed or shrunk the queueing window, or if we are transitioning to the blocked state. bnx2i, when they rework their firmware according to netdev developers requests, will also need to be able to limit queueing at this level. bnx2i will hook into libiscsi, but will allocate a scsi host per netdevice/hba, so unlike pure software iscsi/iser which is allocating a host per session, it cannot set the scsi_host->can_queue and return SCSI_MLQUEUE_HOST_BUSY to reflect queueing limits on the transport. The iscsi class/driver can also set a scsi_target->can_queue value which reflects the max commands the driver/class can support. For iscsi this reflects the number of commands we can support for each session due to session/connection hw limits, driver limits, and to also reflect the session/targets's queueing window. Changes: v1 - initial patch. v2 - Fix scsi_run_queue handling of multiple blocked targets. Previously we would break from the main loop if a device was added back on the starved list. We now run over the list and check if any target is blocked. v3 - Rediff for scsi-misc. Signed-off-by: Mike Christie <michaelc@cs.wisc.edu> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-08-17 20:24:38 +00:00
}
static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
{
if (atomic_read(&shost->host_blocked) > 0)
return true;
if (shost->host_self_blocked)
return true;
return false;
}
static void scsi_starved_list_run(struct Scsi_Host *shost)
{
LIST_HEAD(starved_list);
struct scsi_device *sdev;
unsigned long flags;
spin_lock_irqsave(shost->host_lock, flags);
list_splice_init(&shost->starved_list, &starved_list);
while (!list_empty(&starved_list)) {
struct request_queue *slq;
/*
* As long as shost is accepting commands and we have
* starved queues, call blk_run_queue. scsi_request_fn
* drops the queue_lock and can add us back to the
* starved_list.
*
* host_lock protects the starved_list and starved_entry.
* scsi_request_fn must get the host_lock before checking
* or modifying starved_list or starved_entry.
*/
if (scsi_host_is_busy(shost))
[SCSI] Add helper code so transport classes/driver can control queueing (v3) SCSI-ml manages the queueing limits for the device and host, but does not do so at the target level. However something something similar can come in userful when a driver is transitioning a transport object to the the blocked state, becuase at that time we do not want to queue io and we do not want the queuecommand to be called again. The patch adds code similar to the exisiting SCSI_ML_*BUSY handlers. You can now return SCSI_MLQUEUE_TARGET_BUSY when we hit a transport level queueing issue like the hw cannot allocate some resource at the iscsi session/connection level, or the target has temporarily closed or shrunk the queueing window, or if we are transitioning to the blocked state. bnx2i, when they rework their firmware according to netdev developers requests, will also need to be able to limit queueing at this level. bnx2i will hook into libiscsi, but will allocate a scsi host per netdevice/hba, so unlike pure software iscsi/iser which is allocating a host per session, it cannot set the scsi_host->can_queue and return SCSI_MLQUEUE_HOST_BUSY to reflect queueing limits on the transport. The iscsi class/driver can also set a scsi_target->can_queue value which reflects the max commands the driver/class can support. For iscsi this reflects the number of commands we can support for each session due to session/connection hw limits, driver limits, and to also reflect the session/targets's queueing window. Changes: v1 - initial patch. v2 - Fix scsi_run_queue handling of multiple blocked targets. Previously we would break from the main loop if a device was added back on the starved list. We now run over the list and check if any target is blocked. v3 - Rediff for scsi-misc. Signed-off-by: Mike Christie <michaelc@cs.wisc.edu> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-08-17 20:24:38 +00:00
break;
sdev = list_entry(starved_list.next,
struct scsi_device, starved_entry);
list_del_init(&sdev->starved_entry);
[SCSI] Add helper code so transport classes/driver can control queueing (v3) SCSI-ml manages the queueing limits for the device and host, but does not do so at the target level. However something something similar can come in userful when a driver is transitioning a transport object to the the blocked state, becuase at that time we do not want to queue io and we do not want the queuecommand to be called again. The patch adds code similar to the exisiting SCSI_ML_*BUSY handlers. You can now return SCSI_MLQUEUE_TARGET_BUSY when we hit a transport level queueing issue like the hw cannot allocate some resource at the iscsi session/connection level, or the target has temporarily closed or shrunk the queueing window, or if we are transitioning to the blocked state. bnx2i, when they rework their firmware according to netdev developers requests, will also need to be able to limit queueing at this level. bnx2i will hook into libiscsi, but will allocate a scsi host per netdevice/hba, so unlike pure software iscsi/iser which is allocating a host per session, it cannot set the scsi_host->can_queue and return SCSI_MLQUEUE_HOST_BUSY to reflect queueing limits on the transport. The iscsi class/driver can also set a scsi_target->can_queue value which reflects the max commands the driver/class can support. For iscsi this reflects the number of commands we can support for each session due to session/connection hw limits, driver limits, and to also reflect the session/targets's queueing window. Changes: v1 - initial patch. v2 - Fix scsi_run_queue handling of multiple blocked targets. Previously we would break from the main loop if a device was added back on the starved list. We now run over the list and check if any target is blocked. v3 - Rediff for scsi-misc. Signed-off-by: Mike Christie <michaelc@cs.wisc.edu> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-08-17 20:24:38 +00:00
if (scsi_target_is_busy(scsi_target(sdev))) {
list_move_tail(&sdev->starved_entry,
&shost->starved_list);
continue;
}
/*
* Once we drop the host lock, a racing scsi_remove_device()
* call may remove the sdev from the starved list and destroy
* it and the queue. Mitigate by taking a reference to the
* queue and never touching the sdev again after we drop the
* host lock. Note: if __scsi_remove_device() invokes
* blk_cleanup_queue() before the queue is run from this
* function then blk_run_queue() will return immediately since
* blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
*/
slq = sdev->request_queue;
if (!blk_get_queue(slq))
continue;
spin_unlock_irqrestore(shost->host_lock, flags);
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
scsi_kick_queue(slq);
blk_put_queue(slq);
spin_lock_irqsave(shost->host_lock, flags);
}
/* put any unprocessed entries back */
list_splice(&starved_list, &shost->starved_list);
spin_unlock_irqrestore(shost->host_lock, flags);
}
/**
* scsi_run_queue - Select a proper request queue to serve next.
* @q: last request's queue
*
* The previous command was completely finished, start a new one if possible.
*/
static void scsi_run_queue(struct request_queue *q)
{
struct scsi_device *sdev = q->queuedata;
if (scsi_target(sdev)->single_lun)
scsi_single_lun_run(sdev);
if (!list_empty(&sdev->host->starved_list))
scsi_starved_list_run(sdev->host);
blk_mq_run_hw_queues(q, false);
}
void scsi_requeue_run_queue(struct work_struct *work)
{
struct scsi_device *sdev;
struct request_queue *q;
sdev = container_of(work, struct scsi_device, requeue_work);
q = sdev->request_queue;
scsi_run_queue(q);
}
void scsi_run_host_queues(struct Scsi_Host *shost)
{
struct scsi_device *sdev;
shost_for_each_device(sdev, shost)
scsi_run_queue(sdev->request_queue);
}
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
{
if (!blk_rq_is_passthrough(cmd->request)) {
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
if (drv->uninit_command)
drv->uninit_command(cmd);
}
}
void scsi_free_sgtables(struct scsi_cmnd *cmd)
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
{
if (cmd->sdb.table.nents)
sg_free_table_chained(&cmd->sdb.table,
SCSI_INLINE_SG_CNT);
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
if (scsi_prot_sg_count(cmd))
sg_free_table_chained(&cmd->prot_sdb->table,
SCSI_INLINE_PROT_SG_CNT);
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
}
EXPORT_SYMBOL_GPL(scsi_free_sgtables);
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
{
scsi_free_sgtables(cmd);
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
scsi_uninit_cmd(cmd);
}
scsi: core: Run queue in case of I/O resource contention failure I/O requests may be held in scheduler queue because of resource contention. The starvation scenario was handled properly in the regular completion path but we failed to account for it during I/O submission. This lead to the hang captured below. Make sure we run the queue when resource contention is encountered in the submission path. [ 39.054963] scsi 13:0:0:0: rejecting I/O to dead device [ 39.058700] scsi 13:0:0:0: rejecting I/O to dead device [ 39.087855] sd 13:0:0:1: [sdd] Synchronizing SCSI cache [ 39.088909] scsi 13:0:0:1: rejecting I/O to dead device [ 39.095351] scsi 13:0:0:1: rejecting I/O to dead device [ 39.096962] scsi 13:0:0:1: rejecting I/O to dead device [ 247.021859] INFO: task scsi-stress-rem:813 blocked for more than 122 seconds. [ 247.023258] Not tainted 5.8.0-rc2 #8 [ 247.024069] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 247.025331] scsi-stress-rem D 0 813 802 0x00004000 [ 247.025334] Call Trace: [ 247.025354] __schedule+0x504/0x55f [ 247.027987] schedule+0x72/0xa8 [ 247.027991] blk_mq_freeze_queue_wait+0x63/0x8c [ 247.027994] ? do_wait_intr_irq+0x7a/0x7a [ 247.027996] blk_cleanup_queue+0x4b/0xc9 [ 247.028000] __scsi_remove_device+0xf6/0x14e [ 247.028002] scsi_remove_device+0x21/0x2b [ 247.029037] sdev_store_delete+0x58/0x7c [ 247.029041] kernfs_fop_write+0x10d/0x14f [ 247.031281] vfs_write+0xa2/0xdf [ 247.032670] ksys_write+0x6b/0xb3 [ 247.032673] do_syscall_64+0x56/0x82 [ 247.034053] entry_SYSCALL_64_after_hwframe+0x44/0xa9 [ 247.034059] RIP: 0033:0x7f69f39e9008 [ 247.036330] Code: Bad RIP value. [ 247.036331] RSP: 002b:00007ffdd8116498 EFLAGS: 00000246 ORIG_RAX: 0000000000000001 [ 247.037613] RAX: ffffffffffffffda RBX: 0000000000000002 RCX: 00007f69f39e9008 [ 247.039714] RDX: 0000000000000002 RSI: 000055cde92a0ab0 RDI: 0000000000000001 [ 247.039715] RBP: 000055cde92a0ab0 R08: 000000000000000a R09: 00007f69f3a79e80 [ 247.039716] R10: 000000000000000a R11: 0000000000000246 R12: 00007f69f3abb780 [ 247.039717] R13: 0000000000000002 R14: 00007f69f3ab6740 R15: 0000000000000002 Link: https://lore.kernel.org/r/20200720025435.812030-1-ming.lei@redhat.com Cc: linux-block@vger.kernel.org Cc: Christoph Hellwig <hch@lst.de> Reviewed-by: Bart Van Assche <bvanassche@acm.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2020-07-20 02:54:35 +00:00
static void scsi_run_queue_async(struct scsi_device *sdev)
{
if (scsi_target(sdev)->single_lun ||
!list_empty(&sdev->host->starved_list)) {
scsi: core: Run queue in case of I/O resource contention failure I/O requests may be held in scheduler queue because of resource contention. The starvation scenario was handled properly in the regular completion path but we failed to account for it during I/O submission. This lead to the hang captured below. Make sure we run the queue when resource contention is encountered in the submission path. [ 39.054963] scsi 13:0:0:0: rejecting I/O to dead device [ 39.058700] scsi 13:0:0:0: rejecting I/O to dead device [ 39.087855] sd 13:0:0:1: [sdd] Synchronizing SCSI cache [ 39.088909] scsi 13:0:0:1: rejecting I/O to dead device [ 39.095351] scsi 13:0:0:1: rejecting I/O to dead device [ 39.096962] scsi 13:0:0:1: rejecting I/O to dead device [ 247.021859] INFO: task scsi-stress-rem:813 blocked for more than 122 seconds. [ 247.023258] Not tainted 5.8.0-rc2 #8 [ 247.024069] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 247.025331] scsi-stress-rem D 0 813 802 0x00004000 [ 247.025334] Call Trace: [ 247.025354] __schedule+0x504/0x55f [ 247.027987] schedule+0x72/0xa8 [ 247.027991] blk_mq_freeze_queue_wait+0x63/0x8c [ 247.027994] ? do_wait_intr_irq+0x7a/0x7a [ 247.027996] blk_cleanup_queue+0x4b/0xc9 [ 247.028000] __scsi_remove_device+0xf6/0x14e [ 247.028002] scsi_remove_device+0x21/0x2b [ 247.029037] sdev_store_delete+0x58/0x7c [ 247.029041] kernfs_fop_write+0x10d/0x14f [ 247.031281] vfs_write+0xa2/0xdf [ 247.032670] ksys_write+0x6b/0xb3 [ 247.032673] do_syscall_64+0x56/0x82 [ 247.034053] entry_SYSCALL_64_after_hwframe+0x44/0xa9 [ 247.034059] RIP: 0033:0x7f69f39e9008 [ 247.036330] Code: Bad RIP value. [ 247.036331] RSP: 002b:00007ffdd8116498 EFLAGS: 00000246 ORIG_RAX: 0000000000000001 [ 247.037613] RAX: ffffffffffffffda RBX: 0000000000000002 RCX: 00007f69f39e9008 [ 247.039714] RDX: 0000000000000002 RSI: 000055cde92a0ab0 RDI: 0000000000000001 [ 247.039715] RBP: 000055cde92a0ab0 R08: 000000000000000a R09: 00007f69f3a79e80 [ 247.039716] R10: 000000000000000a R11: 0000000000000246 R12: 00007f69f3abb780 [ 247.039717] R13: 0000000000000002 R14: 00007f69f3ab6740 R15: 0000000000000002 Link: https://lore.kernel.org/r/20200720025435.812030-1-ming.lei@redhat.com Cc: linux-block@vger.kernel.org Cc: Christoph Hellwig <hch@lst.de> Reviewed-by: Bart Van Assche <bvanassche@acm.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2020-07-20 02:54:35 +00:00
kblockd_schedule_work(&sdev->requeue_work);
} else {
/*
* smp_mb() present in sbitmap_queue_clear() or implied in
* .end_io is for ordering writing .device_busy in
* scsi_device_unbusy() and reading sdev->restarts.
*/
int old = atomic_read(&sdev->restarts);
/*
* ->restarts has to be kept as non-zero if new budget
* contention occurs.
*
* No need to run queue when either another re-run
* queue wins in updating ->restarts or a new budget
* contention occurs.
*/
if (old && atomic_cmpxchg(&sdev->restarts, old, 0) == old)
blk_mq_run_hw_queues(sdev->request_queue, true);
}
scsi: core: Run queue in case of I/O resource contention failure I/O requests may be held in scheduler queue because of resource contention. The starvation scenario was handled properly in the regular completion path but we failed to account for it during I/O submission. This lead to the hang captured below. Make sure we run the queue when resource contention is encountered in the submission path. [ 39.054963] scsi 13:0:0:0: rejecting I/O to dead device [ 39.058700] scsi 13:0:0:0: rejecting I/O to dead device [ 39.087855] sd 13:0:0:1: [sdd] Synchronizing SCSI cache [ 39.088909] scsi 13:0:0:1: rejecting I/O to dead device [ 39.095351] scsi 13:0:0:1: rejecting I/O to dead device [ 39.096962] scsi 13:0:0:1: rejecting I/O to dead device [ 247.021859] INFO: task scsi-stress-rem:813 blocked for more than 122 seconds. [ 247.023258] Not tainted 5.8.0-rc2 #8 [ 247.024069] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 247.025331] scsi-stress-rem D 0 813 802 0x00004000 [ 247.025334] Call Trace: [ 247.025354] __schedule+0x504/0x55f [ 247.027987] schedule+0x72/0xa8 [ 247.027991] blk_mq_freeze_queue_wait+0x63/0x8c [ 247.027994] ? do_wait_intr_irq+0x7a/0x7a [ 247.027996] blk_cleanup_queue+0x4b/0xc9 [ 247.028000] __scsi_remove_device+0xf6/0x14e [ 247.028002] scsi_remove_device+0x21/0x2b [ 247.029037] sdev_store_delete+0x58/0x7c [ 247.029041] kernfs_fop_write+0x10d/0x14f [ 247.031281] vfs_write+0xa2/0xdf [ 247.032670] ksys_write+0x6b/0xb3 [ 247.032673] do_syscall_64+0x56/0x82 [ 247.034053] entry_SYSCALL_64_after_hwframe+0x44/0xa9 [ 247.034059] RIP: 0033:0x7f69f39e9008 [ 247.036330] Code: Bad RIP value. [ 247.036331] RSP: 002b:00007ffdd8116498 EFLAGS: 00000246 ORIG_RAX: 0000000000000001 [ 247.037613] RAX: ffffffffffffffda RBX: 0000000000000002 RCX: 00007f69f39e9008 [ 247.039714] RDX: 0000000000000002 RSI: 000055cde92a0ab0 RDI: 0000000000000001 [ 247.039715] RBP: 000055cde92a0ab0 R08: 000000000000000a R09: 00007f69f3a79e80 [ 247.039716] R10: 000000000000000a R11: 0000000000000246 R12: 00007f69f3abb780 [ 247.039717] R13: 0000000000000002 R14: 00007f69f3ab6740 R15: 0000000000000002 Link: https://lore.kernel.org/r/20200720025435.812030-1-ming.lei@redhat.com Cc: linux-block@vger.kernel.org Cc: Christoph Hellwig <hch@lst.de> Reviewed-by: Bart Van Assche <bvanassche@acm.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2020-07-20 02:54:35 +00:00
}
/* Returns false when no more bytes to process, true if there are more */
static bool scsi_end_request(struct request *req, blk_status_t error,
unsigned int bytes)
{
struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
struct scsi_device *sdev = cmd->device;
struct request_queue *q = sdev->request_queue;
if (blk_update_request(req, error, bytes))
return true;
if (blk_queue_add_random(q))
add_disk_randomness(req->rq_disk);
if (!blk_rq_is_passthrough(req)) {
WARN_ON_ONCE(!(cmd->flags & SCMD_INITIALIZED));
cmd->flags &= ~SCMD_INITIALIZED;
}
/*
* Calling rcu_barrier() is not necessary here because the
* SCSI error handler guarantees that the function called by
* call_rcu() has been called before scsi_end_request() is
* called.
*/
destroy_rcu_head(&cmd->rcu);
/*
* In the MQ case the command gets freed by __blk_mq_end_request,
* so we have to do all cleanup that depends on it earlier.
*
* We also can't kick the queues from irq context, so we
* will have to defer it to a workqueue.
*/
scsi_mq_uninit_cmd(cmd);
/*
* queue is still alive, so grab the ref for preventing it
* from being cleaned up during running queue.
*/
percpu_ref_get(&q->q_usage_counter);
__blk_mq_end_request(req, error);
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
scsi: core: Run queue in case of I/O resource contention failure I/O requests may be held in scheduler queue because of resource contention. The starvation scenario was handled properly in the regular completion path but we failed to account for it during I/O submission. This lead to the hang captured below. Make sure we run the queue when resource contention is encountered in the submission path. [ 39.054963] scsi 13:0:0:0: rejecting I/O to dead device [ 39.058700] scsi 13:0:0:0: rejecting I/O to dead device [ 39.087855] sd 13:0:0:1: [sdd] Synchronizing SCSI cache [ 39.088909] scsi 13:0:0:1: rejecting I/O to dead device [ 39.095351] scsi 13:0:0:1: rejecting I/O to dead device [ 39.096962] scsi 13:0:0:1: rejecting I/O to dead device [ 247.021859] INFO: task scsi-stress-rem:813 blocked for more than 122 seconds. [ 247.023258] Not tainted 5.8.0-rc2 #8 [ 247.024069] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 247.025331] scsi-stress-rem D 0 813 802 0x00004000 [ 247.025334] Call Trace: [ 247.025354] __schedule+0x504/0x55f [ 247.027987] schedule+0x72/0xa8 [ 247.027991] blk_mq_freeze_queue_wait+0x63/0x8c [ 247.027994] ? do_wait_intr_irq+0x7a/0x7a [ 247.027996] blk_cleanup_queue+0x4b/0xc9 [ 247.028000] __scsi_remove_device+0xf6/0x14e [ 247.028002] scsi_remove_device+0x21/0x2b [ 247.029037] sdev_store_delete+0x58/0x7c [ 247.029041] kernfs_fop_write+0x10d/0x14f [ 247.031281] vfs_write+0xa2/0xdf [ 247.032670] ksys_write+0x6b/0xb3 [ 247.032673] do_syscall_64+0x56/0x82 [ 247.034053] entry_SYSCALL_64_after_hwframe+0x44/0xa9 [ 247.034059] RIP: 0033:0x7f69f39e9008 [ 247.036330] Code: Bad RIP value. [ 247.036331] RSP: 002b:00007ffdd8116498 EFLAGS: 00000246 ORIG_RAX: 0000000000000001 [ 247.037613] RAX: ffffffffffffffda RBX: 0000000000000002 RCX: 00007f69f39e9008 [ 247.039714] RDX: 0000000000000002 RSI: 000055cde92a0ab0 RDI: 0000000000000001 [ 247.039715] RBP: 000055cde92a0ab0 R08: 000000000000000a R09: 00007f69f3a79e80 [ 247.039716] R10: 000000000000000a R11: 0000000000000246 R12: 00007f69f3abb780 [ 247.039717] R13: 0000000000000002 R14: 00007f69f3ab6740 R15: 0000000000000002 Link: https://lore.kernel.org/r/20200720025435.812030-1-ming.lei@redhat.com Cc: linux-block@vger.kernel.org Cc: Christoph Hellwig <hch@lst.de> Reviewed-by: Bart Van Assche <bvanassche@acm.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2020-07-20 02:54:35 +00:00
scsi_run_queue_async(sdev);
percpu_ref_put(&q->q_usage_counter);
return false;
}
/**
* scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
* @cmd: SCSI command
* @result: scsi error code
*
* Translate a SCSI result code into a blk_status_t value. May reset the host
* byte of @cmd->result.
*/
static blk_status_t scsi_result_to_blk_status(struct scsi_cmnd *cmd, int result)
{
switch (host_byte(result)) {
case DID_OK:
if (scsi_status_is_good(result))
return BLK_STS_OK;
return BLK_STS_IOERR;
case DID_TRANSPORT_FAILFAST:
case DID_TRANSPORT_MARGINAL:
return BLK_STS_TRANSPORT;
case DID_TARGET_FAILURE:
set_host_byte(cmd, DID_OK);
return BLK_STS_TARGET;
case DID_NEXUS_FAILURE:
set_host_byte(cmd, DID_OK);
return BLK_STS_NEXUS;
case DID_ALLOC_FAILURE:
set_host_byte(cmd, DID_OK);
return BLK_STS_NOSPC;
case DID_MEDIUM_ERROR:
set_host_byte(cmd, DID_OK);
return BLK_STS_MEDIUM;
default:
return BLK_STS_IOERR;
}
}
/* Helper for scsi_io_completion() when "reprep" action required. */
static void scsi_io_completion_reprep(struct scsi_cmnd *cmd,
struct request_queue *q)
{
/* A new command will be prepared and issued. */
scsi_mq_requeue_cmd(cmd);
}
static bool scsi_cmd_runtime_exceeced(struct scsi_cmnd *cmd)
{
struct request *req = cmd->request;
unsigned long wait_for;
if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
return false;
wait_for = (cmd->allowed + 1) * req->timeout;
if (time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
scmd_printk(KERN_ERR, cmd, "timing out command, waited %lus\n",
wait_for/HZ);
return true;
}
return false;
}
/* Helper for scsi_io_completion() when special action required. */
static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
{
struct request_queue *q = cmd->device->request_queue;
struct request *req = cmd->request;
int level = 0;
enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
ACTION_DELAYED_RETRY} action;
struct scsi_sense_hdr sshdr;
bool sense_valid;
bool sense_current = true; /* false implies "deferred sense" */
blk_status_t blk_stat;
sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
if (sense_valid)
sense_current = !scsi_sense_is_deferred(&sshdr);
blk_stat = scsi_result_to_blk_status(cmd, result);
if (host_byte(result) == DID_RESET) {
/* Third party bus reset or reset for error recovery
* reasons. Just retry the command and see what
* happens.
*/
action = ACTION_RETRY;
} else if (sense_valid && sense_current) {
switch (sshdr.sense_key) {
case UNIT_ATTENTION:
if (cmd->device->removable) {
/* Detected disc change. Set a bit
* and quietly refuse further access.
*/
cmd->device->changed = 1;
action = ACTION_FAIL;
} else {
/* Must have been a power glitch, or a
* bus reset. Could not have been a
* media change, so we just retry the
* command and see what happens.
*/
action = ACTION_RETRY;
}
break;
case ILLEGAL_REQUEST:
/* If we had an ILLEGAL REQUEST returned, then
* we may have performed an unsupported
* command. The only thing this should be
* would be a ten byte read where only a six
* byte read was supported. Also, on a system
* where READ CAPACITY failed, we may have
* read past the end of the disk.
*/
if ((cmd->device->use_10_for_rw &&
sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
(cmd->cmnd[0] == READ_10 ||
cmd->cmnd[0] == WRITE_10)) {
/* This will issue a new 6-byte command. */
cmd->device->use_10_for_rw = 0;
action = ACTION_REPREP;
} else if (sshdr.asc == 0x10) /* DIX */ {
action = ACTION_FAIL;
blk_stat = BLK_STS_PROTECTION;
/* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
} else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
action = ACTION_FAIL;
blk_stat = BLK_STS_TARGET;
} else
action = ACTION_FAIL;
break;
case ABORTED_COMMAND:
action = ACTION_FAIL;
if (sshdr.asc == 0x10) /* DIF */
blk_stat = BLK_STS_PROTECTION;
break;
case NOT_READY:
/* If the device is in the process of becoming
* ready, or has a temporary blockage, retry.
*/
if (sshdr.asc == 0x04) {
switch (sshdr.ascq) {
case 0x01: /* becoming ready */
case 0x04: /* format in progress */
case 0x05: /* rebuild in progress */
case 0x06: /* recalculation in progress */
case 0x07: /* operation in progress */
case 0x08: /* Long write in progress */
case 0x09: /* self test in progress */
case 0x11: /* notify (enable spinup) required */
case 0x14: /* space allocation in progress */
case 0x1a: /* start stop unit in progress */
case 0x1b: /* sanitize in progress */
case 0x1d: /* configuration in progress */
case 0x24: /* depopulation in progress */
action = ACTION_DELAYED_RETRY;
break;
case 0x0a: /* ALUA state transition */
blk_stat = BLK_STS_AGAIN;
fallthrough;
default:
action = ACTION_FAIL;
break;
}
} else
action = ACTION_FAIL;
break;
case VOLUME_OVERFLOW:
/* See SSC3rXX or current. */
action = ACTION_FAIL;
break;
case DATA_PROTECT:
action = ACTION_FAIL;
if ((sshdr.asc == 0x0C && sshdr.ascq == 0x12) ||
(sshdr.asc == 0x55 &&
(sshdr.ascq == 0x0E || sshdr.ascq == 0x0F))) {
/* Insufficient zone resources */
blk_stat = BLK_STS_ZONE_OPEN_RESOURCE;
}
break;
default:
action = ACTION_FAIL;
break;
}
} else
action = ACTION_FAIL;
if (action != ACTION_FAIL && scsi_cmd_runtime_exceeced(cmd))
action = ACTION_FAIL;
switch (action) {
case ACTION_FAIL:
/* Give up and fail the remainder of the request */
if (!(req->rq_flags & RQF_QUIET)) {
static DEFINE_RATELIMIT_STATE(_rs,
DEFAULT_RATELIMIT_INTERVAL,
DEFAULT_RATELIMIT_BURST);
if (unlikely(scsi_logging_level))
level =
SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
SCSI_LOG_MLCOMPLETE_BITS);
/*
* if logging is enabled the failure will be printed
* in scsi_log_completion(), so avoid duplicate messages
*/
if (!level && __ratelimit(&_rs)) {
scsi_print_result(cmd, NULL, FAILED);
if (sense_valid)
scsi_print_sense(cmd);
scsi_print_command(cmd);
}
}
if (!scsi_end_request(req, blk_stat, blk_rq_err_bytes(req)))
return;
fallthrough;
case ACTION_REPREP:
scsi_io_completion_reprep(cmd, q);
break;
case ACTION_RETRY:
/* Retry the same command immediately */
__scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
break;
case ACTION_DELAYED_RETRY:
/* Retry the same command after a delay */
__scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
break;
}
}
/*
* Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
* new result that may suppress further error checking. Also modifies
* *blk_statp in some cases.
*/
static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
blk_status_t *blk_statp)
{
bool sense_valid;
bool sense_current = true; /* false implies "deferred sense" */
struct request *req = cmd->request;
struct scsi_sense_hdr sshdr;
sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
if (sense_valid)
sense_current = !scsi_sense_is_deferred(&sshdr);
if (blk_rq_is_passthrough(req)) {
if (sense_valid) {
/*
* SG_IO wants current and deferred errors
*/
scsi_req(req)->sense_len =
min(8 + cmd->sense_buffer[7],
SCSI_SENSE_BUFFERSIZE);
}
if (sense_current)
*blk_statp = scsi_result_to_blk_status(cmd, result);
} else if (blk_rq_bytes(req) == 0 && sense_current) {
/*
* Flush commands do not transfers any data, and thus cannot use
* good_bytes != blk_rq_bytes(req) as the signal for an error.
* This sets *blk_statp explicitly for the problem case.
*/
*blk_statp = scsi_result_to_blk_status(cmd, result);
}
/*
* Recovered errors need reporting, but they're always treated as
* success, so fiddle the result code here. For passthrough requests
* we already took a copy of the original into sreq->result which
* is what gets returned to the user
*/
if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
bool do_print = true;
/*
* if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
* skip print since caller wants ATA registers. Only occurs
* on SCSI ATA PASS_THROUGH commands when CK_COND=1
*/
if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
do_print = false;
else if (req->rq_flags & RQF_QUIET)
do_print = false;
if (do_print)
scsi_print_sense(cmd);
result = 0;
/* for passthrough, *blk_statp may be set */
*blk_statp = BLK_STS_OK;
}
/*
* Another corner case: the SCSI status byte is non-zero but 'good'.
* Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
* it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
* if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
* intermediate statuses (both obsolete in SAM-4) as good.
*/
if ((result & 0xff) && scsi_status_is_good(result)) {
result = 0;
*blk_statp = BLK_STS_OK;
}
return result;
}
/**
* scsi_io_completion - Completion processing for SCSI commands.
* @cmd: command that is finished.
* @good_bytes: number of processed bytes.
*
* We will finish off the specified number of sectors. If we are done, the
* command block will be released and the queue function will be goosed. If we
* are not done then we have to figure out what to do next:
*
* a) We can call scsi_io_completion_reprep(). The request will be
* unprepared and put back on the queue. Then a new command will
* be created for it. This should be used if we made forward
* progress, or if we want to switch from READ(10) to READ(6) for
* example.
*
* b) We can call scsi_io_completion_action(). The request will be
* put back on the queue and retried using the same command as
* before, possibly after a delay.
*
* c) We can call scsi_end_request() with blk_stat other than
* BLK_STS_OK, to fail the remainder of the request.
*/
void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
{
int result = cmd->result;
struct request_queue *q = cmd->device->request_queue;
struct request *req = cmd->request;
blk_status_t blk_stat = BLK_STS_OK;
if (unlikely(result)) /* a nz result may or may not be an error */
result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
if (unlikely(blk_rq_is_passthrough(req))) {
[SCSI] scsi_lib: fix scsi_io_completion's SG_IO error propagation The following v3.4-rc1 commit unmasked an existing bug in scsi_io_completion's SG_IO error handling: 47ac56d [SCSI] scsi_error: classify some ILLEGAL_REQUEST sense as a permanent TARGET_ERROR Given that certain ILLEGAL_REQUEST are now properly categorized as TARGET_ERROR the host_byte is being set (before host_byte wasn't ever set for these ILLEGAL_REQUEST). In scsi_io_completion, initialize req->errors with cmd->result _after_ the SG_IO block that calls __scsi_error_from_host_byte (which may modify the host_byte). Before this fix: cdb to send: 12 01 01 00 00 00 ioctl(3, SG_IO, {'S', SG_DXFER_NONE, cmd[6]=[12, 01, 01, 00, 00, 00], mx_sb_len=32, iovec_count=0, dxfer_len=0, timeout=20000, flags=0, status=02, masked_status=01, sb[19]=[70, 00, 05, 00, 00, 00, 00, 0b, 00, 00, 00, 00, 24, 00, 00, 00, 00, 00, 00], host_status=0x10, driver_status=0x8, resid=0, duration=0, info=0x1}) = 0 SCSI Status: Check Condition Sense Information: sense buffer empty After: cdb to send: 12 01 01 00 00 00 ioctl(3, SG_IO, {'S', SG_DXFER_NONE, cmd[6]=[12, 01, 01, 00, 00, 00], mx_sb_len=32, iovec_count=0, dxfer_len=0, timeout=20000, flags=0, status=02, masked_status=01, sb[19]=[70, 00, 05, 00, 00, 00, 00, 0b, 00, 00, 00, 00, 24, 00, 00, 00, 00, 00, 00], host_status=0, driver_status=0x8, resid=0, duration=0, info=0x1}) = 0 SCSI Status: Check Condition Sense Information: Fixed format, current; Sense key: Illegal Request Additional sense: Invalid field in cdb Raw sense data (in hex): 70 00 05 00 00 00 00 0b 00 00 00 00 24 00 00 00 00 00 00 Reported-by: Paolo Bonzini <pbonzini@redhat.com> Tested-by: Paolo Bonzini <pbonzini@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Reviewed-by: Babu Moger <babu.moger@netapp.com> Cc: stable@vger.kernel.org # 3.4 Signed-off-by: James Bottomley <JBottomley@Parallels.com>
2012-05-31 19:05:33 +00:00
/*
* scsi_result_to_blk_status may have reset the host_byte
[SCSI] scsi_lib: fix scsi_io_completion's SG_IO error propagation The following v3.4-rc1 commit unmasked an existing bug in scsi_io_completion's SG_IO error handling: 47ac56d [SCSI] scsi_error: classify some ILLEGAL_REQUEST sense as a permanent TARGET_ERROR Given that certain ILLEGAL_REQUEST are now properly categorized as TARGET_ERROR the host_byte is being set (before host_byte wasn't ever set for these ILLEGAL_REQUEST). In scsi_io_completion, initialize req->errors with cmd->result _after_ the SG_IO block that calls __scsi_error_from_host_byte (which may modify the host_byte). Before this fix: cdb to send: 12 01 01 00 00 00 ioctl(3, SG_IO, {'S', SG_DXFER_NONE, cmd[6]=[12, 01, 01, 00, 00, 00], mx_sb_len=32, iovec_count=0, dxfer_len=0, timeout=20000, flags=0, status=02, masked_status=01, sb[19]=[70, 00, 05, 00, 00, 00, 00, 0b, 00, 00, 00, 00, 24, 00, 00, 00, 00, 00, 00], host_status=0x10, driver_status=0x8, resid=0, duration=0, info=0x1}) = 0 SCSI Status: Check Condition Sense Information: sense buffer empty After: cdb to send: 12 01 01 00 00 00 ioctl(3, SG_IO, {'S', SG_DXFER_NONE, cmd[6]=[12, 01, 01, 00, 00, 00], mx_sb_len=32, iovec_count=0, dxfer_len=0, timeout=20000, flags=0, status=02, masked_status=01, sb[19]=[70, 00, 05, 00, 00, 00, 00, 0b, 00, 00, 00, 00, 24, 00, 00, 00, 00, 00, 00], host_status=0, driver_status=0x8, resid=0, duration=0, info=0x1}) = 0 SCSI Status: Check Condition Sense Information: Fixed format, current; Sense key: Illegal Request Additional sense: Invalid field in cdb Raw sense data (in hex): 70 00 05 00 00 00 00 0b 00 00 00 00 24 00 00 00 00 00 00 Reported-by: Paolo Bonzini <pbonzini@redhat.com> Tested-by: Paolo Bonzini <pbonzini@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Reviewed-by: Babu Moger <babu.moger@netapp.com> Cc: stable@vger.kernel.org # 3.4 Signed-off-by: James Bottomley <JBottomley@Parallels.com>
2012-05-31 19:05:33 +00:00
*/
scsi_req(req)->result = cmd->result;
}
/*
* Next deal with any sectors which we were able to correctly
* handle.
*/
SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
"%u sectors total, %d bytes done.\n",
blk_rq_sectors(req), good_bytes));
/*
* Failed, zero length commands always need to drop down
* to retry code. Fast path should return in this block.
*/
if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
return; /* no bytes remaining */
}
/* Kill remainder if no retries. */
if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
WARN_ONCE(true,
"Bytes remaining after failed, no-retry command");
return;
}
/*
* If there had been no error, but we have leftover bytes in the
* requeues just queue the command up again.
*/
if (likely(result == 0))
scsi_io_completion_reprep(cmd, q);
else
scsi_io_completion_action(cmd, result);
}
static inline bool scsi_cmd_needs_dma_drain(struct scsi_device *sdev,
struct request *rq)
{
return sdev->dma_drain_len && blk_rq_is_passthrough(rq) &&
!op_is_write(req_op(rq)) &&
sdev->host->hostt->dma_need_drain(rq);
}
[SCSI] bidirectional command support At the block level bidi request uses req->next_rq pointer for a second bidi_read request. At Scsi-midlayer a second scsi_data_buffer structure is used for the bidi_read part. This bidi scsi_data_buffer is put on request->next_rq->special. Struct scsi_cmnd is not changed. - Define scsi_bidi_cmnd() to return true if it is a bidi request and a second sgtable was allocated. - Define scsi_in()/scsi_out() to return the in or out scsi_data_buffer from this command This API is to isolate users from the mechanics of bidi. - Define scsi_end_bidi_request() to do what scsi_end_request() does but for a bidi request. This is necessary because bidi commands are a bit tricky here. (See comments in body) - scsi_release_buffers() will also release the bidi_read scsi_data_buffer - scsi_io_completion() on bidi commands will now call scsi_end_bidi_request() and return. - The previous work done in scsi_init_io() is now done in a new scsi_init_sgtable() (which is 99% identical to old scsi_init_io()) The new scsi_init_io() will call the above twice if needed also for the bidi_read command. Only at this point is a command bidi. - In scsi_error.c at scsi_eh_prep/restore_cmnd() make sure bidi-lld is not confused by a get-sense command that looks like bidi. This is done by puting NULL at request->next_rq, and restoring. [jejb: update to sg_table and resolve conflicts also update to blk-end-request and resolve conflicts] Signed-off-by: Boaz Harrosh <bharrosh@panasas.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2007-12-13 11:50:53 +00:00
/**
* scsi_alloc_sgtables - Allocate and initialize data and integrity scatterlists
* @cmd: SCSI command data structure to initialize.
*
* Initializes @cmd->sdb and also @cmd->prot_sdb if data integrity is enabled
* for @cmd.
[SCSI] bidirectional command support At the block level bidi request uses req->next_rq pointer for a second bidi_read request. At Scsi-midlayer a second scsi_data_buffer structure is used for the bidi_read part. This bidi scsi_data_buffer is put on request->next_rq->special. Struct scsi_cmnd is not changed. - Define scsi_bidi_cmnd() to return true if it is a bidi request and a second sgtable was allocated. - Define scsi_in()/scsi_out() to return the in or out scsi_data_buffer from this command This API is to isolate users from the mechanics of bidi. - Define scsi_end_bidi_request() to do what scsi_end_request() does but for a bidi request. This is necessary because bidi commands are a bit tricky here. (See comments in body) - scsi_release_buffers() will also release the bidi_read scsi_data_buffer - scsi_io_completion() on bidi commands will now call scsi_end_bidi_request() and return. - The previous work done in scsi_init_io() is now done in a new scsi_init_sgtable() (which is 99% identical to old scsi_init_io()) The new scsi_init_io() will call the above twice if needed also for the bidi_read command. Only at this point is a command bidi. - In scsi_error.c at scsi_eh_prep/restore_cmnd() make sure bidi-lld is not confused by a get-sense command that looks like bidi. This is done by puting NULL at request->next_rq, and restoring. [jejb: update to sg_table and resolve conflicts also update to blk-end-request and resolve conflicts] Signed-off-by: Boaz Harrosh <bharrosh@panasas.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2007-12-13 11:50:53 +00:00
*
* Returns:
* * BLK_STS_OK - on success
* * BLK_STS_RESOURCE - if the failure is retryable
* * BLK_STS_IOERR - if the failure is fatal
[SCSI] bidirectional command support At the block level bidi request uses req->next_rq pointer for a second bidi_read request. At Scsi-midlayer a second scsi_data_buffer structure is used for the bidi_read part. This bidi scsi_data_buffer is put on request->next_rq->special. Struct scsi_cmnd is not changed. - Define scsi_bidi_cmnd() to return true if it is a bidi request and a second sgtable was allocated. - Define scsi_in()/scsi_out() to return the in or out scsi_data_buffer from this command This API is to isolate users from the mechanics of bidi. - Define scsi_end_bidi_request() to do what scsi_end_request() does but for a bidi request. This is necessary because bidi commands are a bit tricky here. (See comments in body) - scsi_release_buffers() will also release the bidi_read scsi_data_buffer - scsi_io_completion() on bidi commands will now call scsi_end_bidi_request() and return. - The previous work done in scsi_init_io() is now done in a new scsi_init_sgtable() (which is 99% identical to old scsi_init_io()) The new scsi_init_io() will call the above twice if needed also for the bidi_read command. Only at this point is a command bidi. - In scsi_error.c at scsi_eh_prep/restore_cmnd() make sure bidi-lld is not confused by a get-sense command that looks like bidi. This is done by puting NULL at request->next_rq, and restoring. [jejb: update to sg_table and resolve conflicts also update to blk-end-request and resolve conflicts] Signed-off-by: Boaz Harrosh <bharrosh@panasas.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2007-12-13 11:50:53 +00:00
*/
blk_status_t scsi_alloc_sgtables(struct scsi_cmnd *cmd)
[SCSI] bidirectional command support At the block level bidi request uses req->next_rq pointer for a second bidi_read request. At Scsi-midlayer a second scsi_data_buffer structure is used for the bidi_read part. This bidi scsi_data_buffer is put on request->next_rq->special. Struct scsi_cmnd is not changed. - Define scsi_bidi_cmnd() to return true if it is a bidi request and a second sgtable was allocated. - Define scsi_in()/scsi_out() to return the in or out scsi_data_buffer from this command This API is to isolate users from the mechanics of bidi. - Define scsi_end_bidi_request() to do what scsi_end_request() does but for a bidi request. This is necessary because bidi commands are a bit tricky here. (See comments in body) - scsi_release_buffers() will also release the bidi_read scsi_data_buffer - scsi_io_completion() on bidi commands will now call scsi_end_bidi_request() and return. - The previous work done in scsi_init_io() is now done in a new scsi_init_sgtable() (which is 99% identical to old scsi_init_io()) The new scsi_init_io() will call the above twice if needed also for the bidi_read command. Only at this point is a command bidi. - In scsi_error.c at scsi_eh_prep/restore_cmnd() make sure bidi-lld is not confused by a get-sense command that looks like bidi. This is done by puting NULL at request->next_rq, and restoring. [jejb: update to sg_table and resolve conflicts also update to blk-end-request and resolve conflicts] Signed-off-by: Boaz Harrosh <bharrosh@panasas.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2007-12-13 11:50:53 +00:00
{
struct scsi_device *sdev = cmd->device;
struct request *rq = cmd->request;
unsigned short nr_segs = blk_rq_nr_phys_segments(rq);
struct scatterlist *last_sg = NULL;
blk_status_t ret;
bool need_drain = scsi_cmd_needs_dma_drain(sdev, rq);
int count;
if (WARN_ON_ONCE(!nr_segs))
return BLK_STS_IOERR;
/*
* Make sure there is space for the drain. The driver must adjust
* max_hw_segments to be prepared for this.
*/
if (need_drain)
nr_segs++;
/*
* If sg table allocation fails, requeue request later.
*/
if (unlikely(sg_alloc_table_chained(&cmd->sdb.table, nr_segs,
cmd->sdb.table.sgl, SCSI_INLINE_SG_CNT)))
return BLK_STS_RESOURCE;
/*
* Next, walk the list, and fill in the addresses and sizes of
* each segment.
*/
count = __blk_rq_map_sg(rq->q, rq, cmd->sdb.table.sgl, &last_sg);
if (blk_rq_bytes(rq) & rq->q->dma_pad_mask) {
unsigned int pad_len =
(rq->q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
last_sg->length += pad_len;
cmd->extra_len += pad_len;
}
if (need_drain) {
sg_unmark_end(last_sg);
last_sg = sg_next(last_sg);
sg_set_buf(last_sg, sdev->dma_drain_buf, sdev->dma_drain_len);
sg_mark_end(last_sg);
cmd->extra_len += sdev->dma_drain_len;
count++;
}
BUG_ON(count > cmd->sdb.table.nents);
cmd->sdb.table.nents = count;
cmd->sdb.length = blk_rq_payload_bytes(rq);
[SCSI] bidirectional command support At the block level bidi request uses req->next_rq pointer for a second bidi_read request. At Scsi-midlayer a second scsi_data_buffer structure is used for the bidi_read part. This bidi scsi_data_buffer is put on request->next_rq->special. Struct scsi_cmnd is not changed. - Define scsi_bidi_cmnd() to return true if it is a bidi request and a second sgtable was allocated. - Define scsi_in()/scsi_out() to return the in or out scsi_data_buffer from this command This API is to isolate users from the mechanics of bidi. - Define scsi_end_bidi_request() to do what scsi_end_request() does but for a bidi request. This is necessary because bidi commands are a bit tricky here. (See comments in body) - scsi_release_buffers() will also release the bidi_read scsi_data_buffer - scsi_io_completion() on bidi commands will now call scsi_end_bidi_request() and return. - The previous work done in scsi_init_io() is now done in a new scsi_init_sgtable() (which is 99% identical to old scsi_init_io()) The new scsi_init_io() will call the above twice if needed also for the bidi_read command. Only at this point is a command bidi. - In scsi_error.c at scsi_eh_prep/restore_cmnd() make sure bidi-lld is not confused by a get-sense command that looks like bidi. This is done by puting NULL at request->next_rq, and restoring. [jejb: update to sg_table and resolve conflicts also update to blk-end-request and resolve conflicts] Signed-off-by: Boaz Harrosh <bharrosh@panasas.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2007-12-13 11:50:53 +00:00
if (blk_integrity_rq(rq)) {
struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
int ivecs;
if (WARN_ON_ONCE(!prot_sdb)) {
/*
* This can happen if someone (e.g. multipath)
* queues a command to a device on an adapter
* that does not support DIX.
*/
ret = BLK_STS_IOERR;
goto out_free_sgtables;
}
ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
prot_sdb->table.sgl,
SCSI_INLINE_PROT_SG_CNT)) {
ret = BLK_STS_RESOURCE;
goto out_free_sgtables;
}
count = blk_rq_map_integrity_sg(rq->q, rq->bio,
prot_sdb->table.sgl);
BUG_ON(count > ivecs);
BUG_ON(count > queue_max_integrity_segments(rq->q));
cmd->prot_sdb = prot_sdb;
cmd->prot_sdb->table.nents = count;
}
return BLK_STS_OK;
out_free_sgtables:
scsi_free_sgtables(cmd);
return ret;
[SCSI] bidirectional command support At the block level bidi request uses req->next_rq pointer for a second bidi_read request. At Scsi-midlayer a second scsi_data_buffer structure is used for the bidi_read part. This bidi scsi_data_buffer is put on request->next_rq->special. Struct scsi_cmnd is not changed. - Define scsi_bidi_cmnd() to return true if it is a bidi request and a second sgtable was allocated. - Define scsi_in()/scsi_out() to return the in or out scsi_data_buffer from this command This API is to isolate users from the mechanics of bidi. - Define scsi_end_bidi_request() to do what scsi_end_request() does but for a bidi request. This is necessary because bidi commands are a bit tricky here. (See comments in body) - scsi_release_buffers() will also release the bidi_read scsi_data_buffer - scsi_io_completion() on bidi commands will now call scsi_end_bidi_request() and return. - The previous work done in scsi_init_io() is now done in a new scsi_init_sgtable() (which is 99% identical to old scsi_init_io()) The new scsi_init_io() will call the above twice if needed also for the bidi_read command. Only at this point is a command bidi. - In scsi_error.c at scsi_eh_prep/restore_cmnd() make sure bidi-lld is not confused by a get-sense command that looks like bidi. This is done by puting NULL at request->next_rq, and restoring. [jejb: update to sg_table and resolve conflicts also update to blk-end-request and resolve conflicts] Signed-off-by: Boaz Harrosh <bharrosh@panasas.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2007-12-13 11:50:53 +00:00
}
EXPORT_SYMBOL(scsi_alloc_sgtables);
/**
* scsi_initialize_rq - initialize struct scsi_cmnd partially
* @rq: Request associated with the SCSI command to be initialized.
*
* This function initializes the members of struct scsi_cmnd that must be
* initialized before request processing starts and that won't be
* reinitialized if a SCSI command is requeued.
*
* Called from inside blk_get_request() for pass-through requests and from
* inside scsi_init_command() for filesystem requests.
*/
static void scsi_initialize_rq(struct request *rq)
{
struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
scsi_req_init(&cmd->req);
init_rcu_head(&cmd->rcu);
cmd->jiffies_at_alloc = jiffies;
cmd->retries = 0;
}
/*
* Only called when the request isn't completed by SCSI, and not freed by
* SCSI
*/
static void scsi_cleanup_rq(struct request *rq)
{
if (rq->rq_flags & RQF_DONTPREP) {
scsi_mq_uninit_cmd(blk_mq_rq_to_pdu(rq));
rq->rq_flags &= ~RQF_DONTPREP;
}
}
/* Called before a request is prepared. See also scsi_mq_prep_fn(). */
void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
{
void *buf = cmd->sense_buffer;
void *prot = cmd->prot_sdb;
struct request *rq = blk_mq_rq_from_pdu(cmd);
unsigned int flags = cmd->flags & SCMD_PRESERVED_FLAGS;
unsigned long jiffies_at_alloc;
int retries, to_clear;
bool in_flight;
int budget_token = cmd->budget_token;
if (!blk_rq_is_passthrough(rq) && !(flags & SCMD_INITIALIZED)) {
flags |= SCMD_INITIALIZED;
scsi_initialize_rq(rq);
}
jiffies_at_alloc = cmd->jiffies_at_alloc;
retries = cmd->retries;
in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state);
/*
* Zero out the cmd, except for the embedded scsi_request. Only clear
* the driver-private command data if the LLD does not supply a
* function to initialize that data.
*/
to_clear = sizeof(*cmd) - sizeof(cmd->req);
if (!dev->host->hostt->init_cmd_priv)
to_clear += dev->host->hostt->cmd_size;
memset((char *)cmd + sizeof(cmd->req), 0, to_clear);
cmd->device = dev;
cmd->sense_buffer = buf;
cmd->prot_sdb = prot;
cmd->flags = flags;
INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
cmd->jiffies_at_alloc = jiffies_at_alloc;
cmd->retries = retries;
if (in_flight)
__set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
cmd->budget_token = budget_token;
[SCSI] Let scsi_cmnd->cmnd use request->cmd buffer - struct scsi_cmnd had a 16 bytes command buffer of its own. This is an unnecessary duplication and copy of request's cmd. It is probably left overs from the time that scsi_cmnd could function without a request attached. So clean that up. - Once above is done, few places, apart from scsi-ml, needed adjustments due to changing the data type of scsi_cmnd->cmnd. - Lots of drivers still use MAX_COMMAND_SIZE. So I have left that #define but equate it to BLK_MAX_CDB. The way I see it and is reflected in the patch below is. MAX_COMMAND_SIZE - means: The longest fixed-length (*) SCSI CDB as per the SCSI standard and is not related to the implementation. BLK_MAX_CDB. - The allocated space at the request level - I have audit all ISA drivers and made sure none use ->cmnd in a DMA Operation. Same audit was done by Andi Kleen. (*)fixed-length here means commands that their size can be determined by their opcode and the CDB does not carry a length specifier, (unlike the VARIABLE_LENGTH_CMD(0x7f) command). This is actually not exactly true and the SCSI standard also defines extended commands and vendor specific commands that can be bigger than 16 bytes. The kernel will support these using the same infrastructure used for VARLEN CDB's. So in effect MAX_COMMAND_SIZE means the maximum size command scsi-ml supports without specifying a cmd_len by ULD's Signed-off-by: Boaz Harrosh <bharrosh@panasas.com> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-04-30 08:19:47 +00:00
}
static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
struct request *req)
{
struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
/*
* Passthrough requests may transfer data, in which case they must
* a bio attached to them. Or they might contain a SCSI command
* that does not transfer data, in which case they may optionally
* submit a request without an attached bio.
*/
if (req->bio) {
blk_status_t ret = scsi_alloc_sgtables(cmd);
if (unlikely(ret != BLK_STS_OK))
return ret;
} else {
BUG_ON(blk_rq_bytes(req));
memset(&cmd->sdb, 0, sizeof(cmd->sdb));
}
cmd->cmd_len = scsi_req(req)->cmd_len;
if (cmd->cmd_len == 0)
cmd->cmd_len = scsi_command_size(cmd->cmnd);
cmd->cmnd = scsi_req(req)->cmd;
cmd->transfersize = blk_rq_bytes(req);
cmd->allowed = scsi_req(req)->retries;
return BLK_STS_OK;
}
static blk_status_t
scsi_device_state_check(struct scsi_device *sdev, struct request *req)
{
switch (sdev->sdev_state) {
case SDEV_CREATED:
return BLK_STS_OK;
case SDEV_OFFLINE:
case SDEV_TRANSPORT_OFFLINE:
/*
* If the device is offline we refuse to process any
* commands. The device must be brought online
* before trying any recovery commands.
*/
if (!sdev->offline_already) {
sdev->offline_already = true;
sdev_printk(KERN_ERR, sdev,
"rejecting I/O to offline device\n");
}
return BLK_STS_IOERR;
case SDEV_DEL:
/*
* If the device is fully deleted, we refuse to
* process any commands as well.
*/
sdev_printk(KERN_ERR, sdev,
"rejecting I/O to dead device\n");
return BLK_STS_IOERR;
case SDEV_BLOCK:
case SDEV_CREATED_BLOCK:
return BLK_STS_RESOURCE;
case SDEV_QUIESCE:
/*
* If the device is blocked we only accept power management
* commands.
*/
if (req && WARN_ON_ONCE(!(req->rq_flags & RQF_PM)))
return BLK_STS_RESOURCE;
return BLK_STS_OK;
default:
/*
* For any other not fully online state we only allow
* power management commands.
*/
if (req && !(req->rq_flags & RQF_PM))
return BLK_STS_IOERR;
return BLK_STS_OK;
}
}
/*
scsi: core: Replace sdev->device_busy with sbitmap SCSI currently uses an atomic variable to track queue depth for each attached device. The queue depth depends on many factors such as transport type and device implementation. In addition, the SCSI device queue depth is not a static entity but changes over time as a result of congestion management. While blk-mq currently tracks queue depth for each hctx, it can't easily be changed to accommodate the SCSI per-device requirement. The current approach of using an atomic variable doesn't scale well when there are lots of CPU cores and the disk is very fast. IOPS can be substantially impacted by the atomic in the hot path. Replace the atomic variable sdev->device_busy with an sbitmap for tracking the SCSI device queue depth. It has been observed that IOPS is improved ~30% by this patchset in the following test: 1) test machine(32 logical CPU cores) Thread(s) per core: 2 Core(s) per socket: 8 Socket(s): 2 NUMA node(s): 2 Model name: Intel(R) Xeon(R) Silver 4110 CPU @ 2.10GHz 2) setup scsi_debug: modprobe scsi_debug virtual_gb=128 max_luns=1 submit_queues=32 delay=0 max_queue=256 3) fio script: fio --rw=randread --size=128G --direct=1 --ioengine=libaio --iodepth=2048 \ --numjobs=32 --bs=4k --group_reporting=1 --group_reporting=1 --runtime=60 \ --loops=10000 --name=job1 --filename=/dev/sdN [mkp: fix device_busy reference in mpt3sas] Link: https://lore.kernel.org/r/20210122023317.687987-14-ming.lei@redhat.com Link: https://lore.kernel.org/linux-block/20200119071432.18558-6-ming.lei@redhat.com/ Cc: Omar Sandoval <osandov@fb.com> Cc: Kashyap Desai <kashyap.desai@broadcom.com> Cc: Sumanesh Samanta <sumanesh.samanta@broadcom.com> Cc: Ewan D. Milne <emilne@redhat.com> Cc: Hannes Reinecke <hare@suse.de> Tested-by: Sumanesh Samanta <sumanesh.samanta@broadcom.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2021-01-22 02:33:17 +00:00
* scsi_dev_queue_ready: if we can send requests to sdev, assign one token
* and return the token else return -1.
*/
static inline int scsi_dev_queue_ready(struct request_queue *q,
struct scsi_device *sdev)
{
scsi: core: Replace sdev->device_busy with sbitmap SCSI currently uses an atomic variable to track queue depth for each attached device. The queue depth depends on many factors such as transport type and device implementation. In addition, the SCSI device queue depth is not a static entity but changes over time as a result of congestion management. While blk-mq currently tracks queue depth for each hctx, it can't easily be changed to accommodate the SCSI per-device requirement. The current approach of using an atomic variable doesn't scale well when there are lots of CPU cores and the disk is very fast. IOPS can be substantially impacted by the atomic in the hot path. Replace the atomic variable sdev->device_busy with an sbitmap for tracking the SCSI device queue depth. It has been observed that IOPS is improved ~30% by this patchset in the following test: 1) test machine(32 logical CPU cores) Thread(s) per core: 2 Core(s) per socket: 8 Socket(s): 2 NUMA node(s): 2 Model name: Intel(R) Xeon(R) Silver 4110 CPU @ 2.10GHz 2) setup scsi_debug: modprobe scsi_debug virtual_gb=128 max_luns=1 submit_queues=32 delay=0 max_queue=256 3) fio script: fio --rw=randread --size=128G --direct=1 --ioengine=libaio --iodepth=2048 \ --numjobs=32 --bs=4k --group_reporting=1 --group_reporting=1 --runtime=60 \ --loops=10000 --name=job1 --filename=/dev/sdN [mkp: fix device_busy reference in mpt3sas] Link: https://lore.kernel.org/r/20210122023317.687987-14-ming.lei@redhat.com Link: https://lore.kernel.org/linux-block/20200119071432.18558-6-ming.lei@redhat.com/ Cc: Omar Sandoval <osandov@fb.com> Cc: Kashyap Desai <kashyap.desai@broadcom.com> Cc: Sumanesh Samanta <sumanesh.samanta@broadcom.com> Cc: Ewan D. Milne <emilne@redhat.com> Cc: Hannes Reinecke <hare@suse.de> Tested-by: Sumanesh Samanta <sumanesh.samanta@broadcom.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2021-01-22 02:33:17 +00:00
int token;
scsi: core: Replace sdev->device_busy with sbitmap SCSI currently uses an atomic variable to track queue depth for each attached device. The queue depth depends on many factors such as transport type and device implementation. In addition, the SCSI device queue depth is not a static entity but changes over time as a result of congestion management. While blk-mq currently tracks queue depth for each hctx, it can't easily be changed to accommodate the SCSI per-device requirement. The current approach of using an atomic variable doesn't scale well when there are lots of CPU cores and the disk is very fast. IOPS can be substantially impacted by the atomic in the hot path. Replace the atomic variable sdev->device_busy with an sbitmap for tracking the SCSI device queue depth. It has been observed that IOPS is improved ~30% by this patchset in the following test: 1) test machine(32 logical CPU cores) Thread(s) per core: 2 Core(s) per socket: 8 Socket(s): 2 NUMA node(s): 2 Model name: Intel(R) Xeon(R) Silver 4110 CPU @ 2.10GHz 2) setup scsi_debug: modprobe scsi_debug virtual_gb=128 max_luns=1 submit_queues=32 delay=0 max_queue=256 3) fio script: fio --rw=randread --size=128G --direct=1 --ioengine=libaio --iodepth=2048 \ --numjobs=32 --bs=4k --group_reporting=1 --group_reporting=1 --runtime=60 \ --loops=10000 --name=job1 --filename=/dev/sdN [mkp: fix device_busy reference in mpt3sas] Link: https://lore.kernel.org/r/20210122023317.687987-14-ming.lei@redhat.com Link: https://lore.kernel.org/linux-block/20200119071432.18558-6-ming.lei@redhat.com/ Cc: Omar Sandoval <osandov@fb.com> Cc: Kashyap Desai <kashyap.desai@broadcom.com> Cc: Sumanesh Samanta <sumanesh.samanta@broadcom.com> Cc: Ewan D. Milne <emilne@redhat.com> Cc: Hannes Reinecke <hare@suse.de> Tested-by: Sumanesh Samanta <sumanesh.samanta@broadcom.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2021-01-22 02:33:17 +00:00
token = sbitmap_get(&sdev->budget_map);
if (atomic_read(&sdev->device_blocked)) {
scsi: core: Replace sdev->device_busy with sbitmap SCSI currently uses an atomic variable to track queue depth for each attached device. The queue depth depends on many factors such as transport type and device implementation. In addition, the SCSI device queue depth is not a static entity but changes over time as a result of congestion management. While blk-mq currently tracks queue depth for each hctx, it can't easily be changed to accommodate the SCSI per-device requirement. The current approach of using an atomic variable doesn't scale well when there are lots of CPU cores and the disk is very fast. IOPS can be substantially impacted by the atomic in the hot path. Replace the atomic variable sdev->device_busy with an sbitmap for tracking the SCSI device queue depth. It has been observed that IOPS is improved ~30% by this patchset in the following test: 1) test machine(32 logical CPU cores) Thread(s) per core: 2 Core(s) per socket: 8 Socket(s): 2 NUMA node(s): 2 Model name: Intel(R) Xeon(R) Silver 4110 CPU @ 2.10GHz 2) setup scsi_debug: modprobe scsi_debug virtual_gb=128 max_luns=1 submit_queues=32 delay=0 max_queue=256 3) fio script: fio --rw=randread --size=128G --direct=1 --ioengine=libaio --iodepth=2048 \ --numjobs=32 --bs=4k --group_reporting=1 --group_reporting=1 --runtime=60 \ --loops=10000 --name=job1 --filename=/dev/sdN [mkp: fix device_busy reference in mpt3sas] Link: https://lore.kernel.org/r/20210122023317.687987-14-ming.lei@redhat.com Link: https://lore.kernel.org/linux-block/20200119071432.18558-6-ming.lei@redhat.com/ Cc: Omar Sandoval <osandov@fb.com> Cc: Kashyap Desai <kashyap.desai@broadcom.com> Cc: Sumanesh Samanta <sumanesh.samanta@broadcom.com> Cc: Ewan D. Milne <emilne@redhat.com> Cc: Hannes Reinecke <hare@suse.de> Tested-by: Sumanesh Samanta <sumanesh.samanta@broadcom.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2021-01-22 02:33:17 +00:00
if (token < 0)
goto out;
if (scsi_device_busy(sdev) > 1)
goto out_dec;
/*
* unblock after device_blocked iterates to zero
*/
if (atomic_dec_return(&sdev->device_blocked) > 0)
goto out_dec;
SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
"unblocking device at zero depth\n"));
}
scsi: core: Replace sdev->device_busy with sbitmap SCSI currently uses an atomic variable to track queue depth for each attached device. The queue depth depends on many factors such as transport type and device implementation. In addition, the SCSI device queue depth is not a static entity but changes over time as a result of congestion management. While blk-mq currently tracks queue depth for each hctx, it can't easily be changed to accommodate the SCSI per-device requirement. The current approach of using an atomic variable doesn't scale well when there are lots of CPU cores and the disk is very fast. IOPS can be substantially impacted by the atomic in the hot path. Replace the atomic variable sdev->device_busy with an sbitmap for tracking the SCSI device queue depth. It has been observed that IOPS is improved ~30% by this patchset in the following test: 1) test machine(32 logical CPU cores) Thread(s) per core: 2 Core(s) per socket: 8 Socket(s): 2 NUMA node(s): 2 Model name: Intel(R) Xeon(R) Silver 4110 CPU @ 2.10GHz 2) setup scsi_debug: modprobe scsi_debug virtual_gb=128 max_luns=1 submit_queues=32 delay=0 max_queue=256 3) fio script: fio --rw=randread --size=128G --direct=1 --ioengine=libaio --iodepth=2048 \ --numjobs=32 --bs=4k --group_reporting=1 --group_reporting=1 --runtime=60 \ --loops=10000 --name=job1 --filename=/dev/sdN [mkp: fix device_busy reference in mpt3sas] Link: https://lore.kernel.org/r/20210122023317.687987-14-ming.lei@redhat.com Link: https://lore.kernel.org/linux-block/20200119071432.18558-6-ming.lei@redhat.com/ Cc: Omar Sandoval <osandov@fb.com> Cc: Kashyap Desai <kashyap.desai@broadcom.com> Cc: Sumanesh Samanta <sumanesh.samanta@broadcom.com> Cc: Ewan D. Milne <emilne@redhat.com> Cc: Hannes Reinecke <hare@suse.de> Tested-by: Sumanesh Samanta <sumanesh.samanta@broadcom.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2021-01-22 02:33:17 +00:00
return token;
out_dec:
scsi: core: Replace sdev->device_busy with sbitmap SCSI currently uses an atomic variable to track queue depth for each attached device. The queue depth depends on many factors such as transport type and device implementation. In addition, the SCSI device queue depth is not a static entity but changes over time as a result of congestion management. While blk-mq currently tracks queue depth for each hctx, it can't easily be changed to accommodate the SCSI per-device requirement. The current approach of using an atomic variable doesn't scale well when there are lots of CPU cores and the disk is very fast. IOPS can be substantially impacted by the atomic in the hot path. Replace the atomic variable sdev->device_busy with an sbitmap for tracking the SCSI device queue depth. It has been observed that IOPS is improved ~30% by this patchset in the following test: 1) test machine(32 logical CPU cores) Thread(s) per core: 2 Core(s) per socket: 8 Socket(s): 2 NUMA node(s): 2 Model name: Intel(R) Xeon(R) Silver 4110 CPU @ 2.10GHz 2) setup scsi_debug: modprobe scsi_debug virtual_gb=128 max_luns=1 submit_queues=32 delay=0 max_queue=256 3) fio script: fio --rw=randread --size=128G --direct=1 --ioengine=libaio --iodepth=2048 \ --numjobs=32 --bs=4k --group_reporting=1 --group_reporting=1 --runtime=60 \ --loops=10000 --name=job1 --filename=/dev/sdN [mkp: fix device_busy reference in mpt3sas] Link: https://lore.kernel.org/r/20210122023317.687987-14-ming.lei@redhat.com Link: https://lore.kernel.org/linux-block/20200119071432.18558-6-ming.lei@redhat.com/ Cc: Omar Sandoval <osandov@fb.com> Cc: Kashyap Desai <kashyap.desai@broadcom.com> Cc: Sumanesh Samanta <sumanesh.samanta@broadcom.com> Cc: Ewan D. Milne <emilne@redhat.com> Cc: Hannes Reinecke <hare@suse.de> Tested-by: Sumanesh Samanta <sumanesh.samanta@broadcom.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2021-01-22 02:33:17 +00:00
if (token >= 0)
sbitmap_put(&sdev->budget_map, token);
out:
return -1;
}
[SCSI] Add helper code so transport classes/driver can control queueing (v3) SCSI-ml manages the queueing limits for the device and host, but does not do so at the target level. However something something similar can come in userful when a driver is transitioning a transport object to the the blocked state, becuase at that time we do not want to queue io and we do not want the queuecommand to be called again. The patch adds code similar to the exisiting SCSI_ML_*BUSY handlers. You can now return SCSI_MLQUEUE_TARGET_BUSY when we hit a transport level queueing issue like the hw cannot allocate some resource at the iscsi session/connection level, or the target has temporarily closed or shrunk the queueing window, or if we are transitioning to the blocked state. bnx2i, when they rework their firmware according to netdev developers requests, will also need to be able to limit queueing at this level. bnx2i will hook into libiscsi, but will allocate a scsi host per netdevice/hba, so unlike pure software iscsi/iser which is allocating a host per session, it cannot set the scsi_host->can_queue and return SCSI_MLQUEUE_HOST_BUSY to reflect queueing limits on the transport. The iscsi class/driver can also set a scsi_target->can_queue value which reflects the max commands the driver/class can support. For iscsi this reflects the number of commands we can support for each session due to session/connection hw limits, driver limits, and to also reflect the session/targets's queueing window. Changes: v1 - initial patch. v2 - Fix scsi_run_queue handling of multiple blocked targets. Previously we would break from the main loop if a device was added back on the starved list. We now run over the list and check if any target is blocked. v3 - Rediff for scsi-misc. Signed-off-by: Mike Christie <michaelc@cs.wisc.edu> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-08-17 20:24:38 +00:00
/*
* scsi_target_queue_ready: checks if there we can send commands to target
* @sdev: scsi device on starget to check.
*/
static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
struct scsi_device *sdev)
{
struct scsi_target *starget = scsi_target(sdev);
unsigned int busy;
[SCSI] Add helper code so transport classes/driver can control queueing (v3) SCSI-ml manages the queueing limits for the device and host, but does not do so at the target level. However something something similar can come in userful when a driver is transitioning a transport object to the the blocked state, becuase at that time we do not want to queue io and we do not want the queuecommand to be called again. The patch adds code similar to the exisiting SCSI_ML_*BUSY handlers. You can now return SCSI_MLQUEUE_TARGET_BUSY when we hit a transport level queueing issue like the hw cannot allocate some resource at the iscsi session/connection level, or the target has temporarily closed or shrunk the queueing window, or if we are transitioning to the blocked state. bnx2i, when they rework their firmware according to netdev developers requests, will also need to be able to limit queueing at this level. bnx2i will hook into libiscsi, but will allocate a scsi host per netdevice/hba, so unlike pure software iscsi/iser which is allocating a host per session, it cannot set the scsi_host->can_queue and return SCSI_MLQUEUE_HOST_BUSY to reflect queueing limits on the transport. The iscsi class/driver can also set a scsi_target->can_queue value which reflects the max commands the driver/class can support. For iscsi this reflects the number of commands we can support for each session due to session/connection hw limits, driver limits, and to also reflect the session/targets's queueing window. Changes: v1 - initial patch. v2 - Fix scsi_run_queue handling of multiple blocked targets. Previously we would break from the main loop if a device was added back on the starved list. We now run over the list and check if any target is blocked. v3 - Rediff for scsi-misc. Signed-off-by: Mike Christie <michaelc@cs.wisc.edu> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-08-17 20:24:38 +00:00
if (starget->single_lun) {
spin_lock_irq(shost->host_lock);
[SCSI] Add helper code so transport classes/driver can control queueing (v3) SCSI-ml manages the queueing limits for the device and host, but does not do so at the target level. However something something similar can come in userful when a driver is transitioning a transport object to the the blocked state, becuase at that time we do not want to queue io and we do not want the queuecommand to be called again. The patch adds code similar to the exisiting SCSI_ML_*BUSY handlers. You can now return SCSI_MLQUEUE_TARGET_BUSY when we hit a transport level queueing issue like the hw cannot allocate some resource at the iscsi session/connection level, or the target has temporarily closed or shrunk the queueing window, or if we are transitioning to the blocked state. bnx2i, when they rework their firmware according to netdev developers requests, will also need to be able to limit queueing at this level. bnx2i will hook into libiscsi, but will allocate a scsi host per netdevice/hba, so unlike pure software iscsi/iser which is allocating a host per session, it cannot set the scsi_host->can_queue and return SCSI_MLQUEUE_HOST_BUSY to reflect queueing limits on the transport. The iscsi class/driver can also set a scsi_target->can_queue value which reflects the max commands the driver/class can support. For iscsi this reflects the number of commands we can support for each session due to session/connection hw limits, driver limits, and to also reflect the session/targets's queueing window. Changes: v1 - initial patch. v2 - Fix scsi_run_queue handling of multiple blocked targets. Previously we would break from the main loop if a device was added back on the starved list. We now run over the list and check if any target is blocked. v3 - Rediff for scsi-misc. Signed-off-by: Mike Christie <michaelc@cs.wisc.edu> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-08-17 20:24:38 +00:00
if (starget->starget_sdev_user &&
starget->starget_sdev_user != sdev) {
spin_unlock_irq(shost->host_lock);
return 0;
}
[SCSI] Add helper code so transport classes/driver can control queueing (v3) SCSI-ml manages the queueing limits for the device and host, but does not do so at the target level. However something something similar can come in userful when a driver is transitioning a transport object to the the blocked state, becuase at that time we do not want to queue io and we do not want the queuecommand to be called again. The patch adds code similar to the exisiting SCSI_ML_*BUSY handlers. You can now return SCSI_MLQUEUE_TARGET_BUSY when we hit a transport level queueing issue like the hw cannot allocate some resource at the iscsi session/connection level, or the target has temporarily closed or shrunk the queueing window, or if we are transitioning to the blocked state. bnx2i, when they rework their firmware according to netdev developers requests, will also need to be able to limit queueing at this level. bnx2i will hook into libiscsi, but will allocate a scsi host per netdevice/hba, so unlike pure software iscsi/iser which is allocating a host per session, it cannot set the scsi_host->can_queue and return SCSI_MLQUEUE_HOST_BUSY to reflect queueing limits on the transport. The iscsi class/driver can also set a scsi_target->can_queue value which reflects the max commands the driver/class can support. For iscsi this reflects the number of commands we can support for each session due to session/connection hw limits, driver limits, and to also reflect the session/targets's queueing window. Changes: v1 - initial patch. v2 - Fix scsi_run_queue handling of multiple blocked targets. Previously we would break from the main loop if a device was added back on the starved list. We now run over the list and check if any target is blocked. v3 - Rediff for scsi-misc. Signed-off-by: Mike Christie <michaelc@cs.wisc.edu> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-08-17 20:24:38 +00:00
starget->starget_sdev_user = sdev;
spin_unlock_irq(shost->host_lock);
[SCSI] Add helper code so transport classes/driver can control queueing (v3) SCSI-ml manages the queueing limits for the device and host, but does not do so at the target level. However something something similar can come in userful when a driver is transitioning a transport object to the the blocked state, becuase at that time we do not want to queue io and we do not want the queuecommand to be called again. The patch adds code similar to the exisiting SCSI_ML_*BUSY handlers. You can now return SCSI_MLQUEUE_TARGET_BUSY when we hit a transport level queueing issue like the hw cannot allocate some resource at the iscsi session/connection level, or the target has temporarily closed or shrunk the queueing window, or if we are transitioning to the blocked state. bnx2i, when they rework their firmware according to netdev developers requests, will also need to be able to limit queueing at this level. bnx2i will hook into libiscsi, but will allocate a scsi host per netdevice/hba, so unlike pure software iscsi/iser which is allocating a host per session, it cannot set the scsi_host->can_queue and return SCSI_MLQUEUE_HOST_BUSY to reflect queueing limits on the transport. The iscsi class/driver can also set a scsi_target->can_queue value which reflects the max commands the driver/class can support. For iscsi this reflects the number of commands we can support for each session due to session/connection hw limits, driver limits, and to also reflect the session/targets's queueing window. Changes: v1 - initial patch. v2 - Fix scsi_run_queue handling of multiple blocked targets. Previously we would break from the main loop if a device was added back on the starved list. We now run over the list and check if any target is blocked. v3 - Rediff for scsi-misc. Signed-off-by: Mike Christie <michaelc@cs.wisc.edu> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-08-17 20:24:38 +00:00
}
if (starget->can_queue <= 0)
return 1;
busy = atomic_inc_return(&starget->target_busy) - 1;
if (atomic_read(&starget->target_blocked) > 0) {
if (busy)
goto starved;
[SCSI] Add helper code so transport classes/driver can control queueing (v3) SCSI-ml manages the queueing limits for the device and host, but does not do so at the target level. However something something similar can come in userful when a driver is transitioning a transport object to the the blocked state, becuase at that time we do not want to queue io and we do not want the queuecommand to be called again. The patch adds code similar to the exisiting SCSI_ML_*BUSY handlers. You can now return SCSI_MLQUEUE_TARGET_BUSY when we hit a transport level queueing issue like the hw cannot allocate some resource at the iscsi session/connection level, or the target has temporarily closed or shrunk the queueing window, or if we are transitioning to the blocked state. bnx2i, when they rework their firmware according to netdev developers requests, will also need to be able to limit queueing at this level. bnx2i will hook into libiscsi, but will allocate a scsi host per netdevice/hba, so unlike pure software iscsi/iser which is allocating a host per session, it cannot set the scsi_host->can_queue and return SCSI_MLQUEUE_HOST_BUSY to reflect queueing limits on the transport. The iscsi class/driver can also set a scsi_target->can_queue value which reflects the max commands the driver/class can support. For iscsi this reflects the number of commands we can support for each session due to session/connection hw limits, driver limits, and to also reflect the session/targets's queueing window. Changes: v1 - initial patch. v2 - Fix scsi_run_queue handling of multiple blocked targets. Previously we would break from the main loop if a device was added back on the starved list. We now run over the list and check if any target is blocked. v3 - Rediff for scsi-misc. Signed-off-by: Mike Christie <michaelc@cs.wisc.edu> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-08-17 20:24:38 +00:00
/*
* unblock after target_blocked iterates to zero
*/
if (atomic_dec_return(&starget->target_blocked) > 0)
goto out_dec;
SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
"unblocking target at zero depth\n"));
[SCSI] Add helper code so transport classes/driver can control queueing (v3) SCSI-ml manages the queueing limits for the device and host, but does not do so at the target level. However something something similar can come in userful when a driver is transitioning a transport object to the the blocked state, becuase at that time we do not want to queue io and we do not want the queuecommand to be called again. The patch adds code similar to the exisiting SCSI_ML_*BUSY handlers. You can now return SCSI_MLQUEUE_TARGET_BUSY when we hit a transport level queueing issue like the hw cannot allocate some resource at the iscsi session/connection level, or the target has temporarily closed or shrunk the queueing window, or if we are transitioning to the blocked state. bnx2i, when they rework their firmware according to netdev developers requests, will also need to be able to limit queueing at this level. bnx2i will hook into libiscsi, but will allocate a scsi host per netdevice/hba, so unlike pure software iscsi/iser which is allocating a host per session, it cannot set the scsi_host->can_queue and return SCSI_MLQUEUE_HOST_BUSY to reflect queueing limits on the transport. The iscsi class/driver can also set a scsi_target->can_queue value which reflects the max commands the driver/class can support. For iscsi this reflects the number of commands we can support for each session due to session/connection hw limits, driver limits, and to also reflect the session/targets's queueing window. Changes: v1 - initial patch. v2 - Fix scsi_run_queue handling of multiple blocked targets. Previously we would break from the main loop if a device was added back on the starved list. We now run over the list and check if any target is blocked. v3 - Rediff for scsi-misc. Signed-off-by: Mike Christie <michaelc@cs.wisc.edu> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-08-17 20:24:38 +00:00
}
if (busy >= starget->can_queue)
goto starved;
[SCSI] Add helper code so transport classes/driver can control queueing (v3) SCSI-ml manages the queueing limits for the device and host, but does not do so at the target level. However something something similar can come in userful when a driver is transitioning a transport object to the the blocked state, becuase at that time we do not want to queue io and we do not want the queuecommand to be called again. The patch adds code similar to the exisiting SCSI_ML_*BUSY handlers. You can now return SCSI_MLQUEUE_TARGET_BUSY when we hit a transport level queueing issue like the hw cannot allocate some resource at the iscsi session/connection level, or the target has temporarily closed or shrunk the queueing window, or if we are transitioning to the blocked state. bnx2i, when they rework their firmware according to netdev developers requests, will also need to be able to limit queueing at this level. bnx2i will hook into libiscsi, but will allocate a scsi host per netdevice/hba, so unlike pure software iscsi/iser which is allocating a host per session, it cannot set the scsi_host->can_queue and return SCSI_MLQUEUE_HOST_BUSY to reflect queueing limits on the transport. The iscsi class/driver can also set a scsi_target->can_queue value which reflects the max commands the driver/class can support. For iscsi this reflects the number of commands we can support for each session due to session/connection hw limits, driver limits, and to also reflect the session/targets's queueing window. Changes: v1 - initial patch. v2 - Fix scsi_run_queue handling of multiple blocked targets. Previously we would break from the main loop if a device was added back on the starved list. We now run over the list and check if any target is blocked. v3 - Rediff for scsi-misc. Signed-off-by: Mike Christie <michaelc@cs.wisc.edu> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-08-17 20:24:38 +00:00
return 1;
starved:
spin_lock_irq(shost->host_lock);
list_move_tail(&sdev->starved_entry, &shost->starved_list);
spin_unlock_irq(shost->host_lock);
out_dec:
if (starget->can_queue > 0)
atomic_dec(&starget->target_busy);
return 0;
[SCSI] Add helper code so transport classes/driver can control queueing (v3) SCSI-ml manages the queueing limits for the device and host, but does not do so at the target level. However something something similar can come in userful when a driver is transitioning a transport object to the the blocked state, becuase at that time we do not want to queue io and we do not want the queuecommand to be called again. The patch adds code similar to the exisiting SCSI_ML_*BUSY handlers. You can now return SCSI_MLQUEUE_TARGET_BUSY when we hit a transport level queueing issue like the hw cannot allocate some resource at the iscsi session/connection level, or the target has temporarily closed or shrunk the queueing window, or if we are transitioning to the blocked state. bnx2i, when they rework their firmware according to netdev developers requests, will also need to be able to limit queueing at this level. bnx2i will hook into libiscsi, but will allocate a scsi host per netdevice/hba, so unlike pure software iscsi/iser which is allocating a host per session, it cannot set the scsi_host->can_queue and return SCSI_MLQUEUE_HOST_BUSY to reflect queueing limits on the transport. The iscsi class/driver can also set a scsi_target->can_queue value which reflects the max commands the driver/class can support. For iscsi this reflects the number of commands we can support for each session due to session/connection hw limits, driver limits, and to also reflect the session/targets's queueing window. Changes: v1 - initial patch. v2 - Fix scsi_run_queue handling of multiple blocked targets. Previously we would break from the main loop if a device was added back on the starved list. We now run over the list and check if any target is blocked. v3 - Rediff for scsi-misc. Signed-off-by: Mike Christie <michaelc@cs.wisc.edu> Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
2008-08-17 20:24:38 +00:00
}
/*
* scsi_host_queue_ready: if we can send requests to shost, return 1 else
* return 0. We must end up running the queue again whenever 0 is
* returned, else IO can hang.
*/
static inline int scsi_host_queue_ready(struct request_queue *q,
struct Scsi_Host *shost,
struct scsi_device *sdev,
struct scsi_cmnd *cmd)
{
if (scsi_host_in_recovery(shost))
return 0;
if (atomic_read(&shost->host_blocked) > 0) {
if (scsi_host_busy(shost) > 0)
goto starved;
/*
* unblock after host_blocked iterates to zero
*/
if (atomic_dec_return(&shost->host_blocked) > 0)
goto out_dec;
SCSI_LOG_MLQUEUE(3,
shost_printk(KERN_INFO, shost,
"unblocking host at zero depth\n"));
}
if (shost->host_self_blocked)
goto starved;
/* We're OK to process the command, so we can't be starved */
if (!list_empty(&sdev->starved_entry)) {
spin_lock_irq(shost->host_lock);
if (!list_empty(&sdev->starved_entry))
list_del_init(&sdev->starved_entry);
spin_unlock_irq(shost->host_lock);
}
__set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
return 1;
starved:
spin_lock_irq(shost->host_lock);
if (list_empty(&sdev->starved_entry))
list_add_tail(&sdev->starved_entry, &shost->starved_list);
spin_unlock_irq(shost->host_lock);
out_dec:
scsi_dec_host_busy(shost, cmd);
return 0;
}
/*
* Busy state exporting function for request stacking drivers.
*
* For efficiency, no lock is taken to check the busy state of
* shost/starget/sdev, since the returned value is not guaranteed and
* may be changed after request stacking drivers call the function,
* regardless of taking lock or not.
*
* When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
* needs to return 'not busy'. Otherwise, request stacking drivers
* may hold requests forever.
*/
static bool scsi_mq_lld_busy(struct request_queue *q)
{
struct scsi_device *sdev = q->queuedata;
struct Scsi_Host *shost;
if (blk_queue_dying(q))
return false;
shost = sdev->host;
/*
* Ignore host/starget busy state.
* Since block layer does not have a concept of fairness across
* multiple queues, congestion of host/starget needs to be handled
* in SCSI layer.
*/
if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
return true;
return false;
}
/*
* Block layer request completion callback. May be called from interrupt
* context.
*/
static void scsi_complete(struct request *rq)
{
struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
enum scsi_disposition disposition;
INIT_LIST_HEAD(&cmd->eh_entry);
atomic_inc(&cmd->device->iodone_cnt);
if (cmd->result)
atomic_inc(&cmd->device->ioerr_cnt);
disposition = scsi_decide_disposition(cmd);
if (disposition != SUCCESS && scsi_cmd_runtime_exceeced(cmd))
disposition = SUCCESS;
scsi_log_completion(cmd, disposition);
switch (disposition) {
case SUCCESS:
scsi_finish_command(cmd);
break;
case NEEDS_RETRY:
scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
break;
case ADD_TO_MLQUEUE:
scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
break;
default:
scsi_eh_scmd_add(cmd);
break;
}
}
/**
* scsi_dispatch_cmd - Dispatch a command to the low-level driver.
* @cmd: command block we are dispatching.
*
* Return: nonzero return request was rejected and device's queue needs to be
* plugged.
*/
static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
{
struct Scsi_Host *host = cmd->device->host;
int rtn = 0;
atomic_inc(&cmd->device->iorequest_cnt);
/* check if the device is still usable */
if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
/* in SDEV_DEL we error all commands. DID_NO_CONNECT
* returns an immediate error upwards, and signals
* that the device is no longer present */
cmd->result = DID_NO_CONNECT << 16;
goto done;
}
/* Check to see if the scsi lld made this device blocked. */
if (unlikely(scsi_device_blocked(cmd->device))) {
/*
* in blocked state, the command is just put back on
* the device queue. The suspend state has already
* blocked the queue so future requests should not
* occur until the device transitions out of the
* suspend state.
*/
SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
"queuecommand : device blocked\n"));
return SCSI_MLQUEUE_DEVICE_BUSY;
}
/* Store the LUN value in cmnd, if needed. */
if (cmd->device->lun_in_cdb)
cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
(cmd->device->lun << 5 & 0xe0);
scsi_log_send(cmd);
/*
* Before we queue this command, check if the command
* length exceeds what the host adapter can handle.
*/
if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
"queuecommand : command too long. "
"cdb_size=%d host->max_cmd_len=%d\n",
cmd->cmd_len, cmd->device->host->max_cmd_len));
cmd->result = (DID_ABORT << 16);
goto done;
}
if (unlikely(host->shost_state == SHOST_DEL)) {
cmd->result = (DID_NO_CONNECT << 16);
goto done;
}
trace_scsi_dispatch_cmd_start(cmd);
rtn = host->hostt->queuecommand(host, cmd);
if (rtn) {
trace_scsi_dispatch_cmd_error(cmd, rtn);
if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
rtn != SCSI_MLQUEUE_TARGET_BUSY)
rtn = SCSI_MLQUEUE_HOST_BUSY;
SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
"queuecommand : request rejected\n"));
}
return rtn;
done:
cmd->scsi_done(cmd);
return 0;
}
/* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)
{
return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *
sizeof(struct scatterlist);
}
static blk_status_t scsi_prepare_cmd(struct request *req)
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
{
struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
struct scsi_device *sdev = req->q->queuedata;
struct Scsi_Host *shost = sdev->host;
struct scatterlist *sg;
scsi_init_command(sdev, cmd);
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
cmd->request = req;
cmd->tag = req->tag;
cmd->prot_op = SCSI_PROT_NORMAL;
if (blk_rq_bytes(req))
cmd->sc_data_direction = rq_dma_dir(req);
else
cmd->sc_data_direction = DMA_NONE;
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
cmd->sdb.table.sgl = sg;
if (scsi_host_get_prot(shost)) {
memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
cmd->prot_sdb->table.sgl =
(struct scatterlist *)(cmd->prot_sdb + 1);
}
/*
* Special handling for passthrough commands, which don't go to the ULP
* at all:
*/
if (blk_rq_is_passthrough(req))
return scsi_setup_scsi_cmnd(sdev, req);
if (sdev->handler && sdev->handler->prep_fn) {
blk_status_t ret = sdev->handler->prep_fn(sdev, req);
if (ret != BLK_STS_OK)
return ret;
}
cmd->cmnd = scsi_req(req)->cmd = scsi_req(req)->__cmd;
memset(cmd->cmnd, 0, BLK_MAX_CDB);
return scsi_cmd_to_driver(cmd)->init_command(cmd);
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
}
static void scsi_mq_done(struct scsi_cmnd *cmd)
{
if (unlikely(blk_should_fake_timeout(cmd->request->q)))
return;
if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
return;
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
trace_scsi_dispatch_cmd_done(cmd);
blk_mq_complete_request(cmd->request);
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
}
static void scsi_mq_put_budget(struct request_queue *q, int budget_token)
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
{
struct scsi_device *sdev = q->queuedata;
scsi: core: Replace sdev->device_busy with sbitmap SCSI currently uses an atomic variable to track queue depth for each attached device. The queue depth depends on many factors such as transport type and device implementation. In addition, the SCSI device queue depth is not a static entity but changes over time as a result of congestion management. While blk-mq currently tracks queue depth for each hctx, it can't easily be changed to accommodate the SCSI per-device requirement. The current approach of using an atomic variable doesn't scale well when there are lots of CPU cores and the disk is very fast. IOPS can be substantially impacted by the atomic in the hot path. Replace the atomic variable sdev->device_busy with an sbitmap for tracking the SCSI device queue depth. It has been observed that IOPS is improved ~30% by this patchset in the following test: 1) test machine(32 logical CPU cores) Thread(s) per core: 2 Core(s) per socket: 8 Socket(s): 2 NUMA node(s): 2 Model name: Intel(R) Xeon(R) Silver 4110 CPU @ 2.10GHz 2) setup scsi_debug: modprobe scsi_debug virtual_gb=128 max_luns=1 submit_queues=32 delay=0 max_queue=256 3) fio script: fio --rw=randread --size=128G --direct=1 --ioengine=libaio --iodepth=2048 \ --numjobs=32 --bs=4k --group_reporting=1 --group_reporting=1 --runtime=60 \ --loops=10000 --name=job1 --filename=/dev/sdN [mkp: fix device_busy reference in mpt3sas] Link: https://lore.kernel.org/r/20210122023317.687987-14-ming.lei@redhat.com Link: https://lore.kernel.org/linux-block/20200119071432.18558-6-ming.lei@redhat.com/ Cc: Omar Sandoval <osandov@fb.com> Cc: Kashyap Desai <kashyap.desai@broadcom.com> Cc: Sumanesh Samanta <sumanesh.samanta@broadcom.com> Cc: Ewan D. Milne <emilne@redhat.com> Cc: Hannes Reinecke <hare@suse.de> Tested-by: Sumanesh Samanta <sumanesh.samanta@broadcom.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2021-01-22 02:33:17 +00:00
sbitmap_put(&sdev->budget_map, budget_token);
}
static int scsi_mq_get_budget(struct request_queue *q)
{
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
struct scsi_device *sdev = q->queuedata;
scsi: core: Replace sdev->device_busy with sbitmap SCSI currently uses an atomic variable to track queue depth for each attached device. The queue depth depends on many factors such as transport type and device implementation. In addition, the SCSI device queue depth is not a static entity but changes over time as a result of congestion management. While blk-mq currently tracks queue depth for each hctx, it can't easily be changed to accommodate the SCSI per-device requirement. The current approach of using an atomic variable doesn't scale well when there are lots of CPU cores and the disk is very fast. IOPS can be substantially impacted by the atomic in the hot path. Replace the atomic variable sdev->device_busy with an sbitmap for tracking the SCSI device queue depth. It has been observed that IOPS is improved ~30% by this patchset in the following test: 1) test machine(32 logical CPU cores) Thread(s) per core: 2 Core(s) per socket: 8 Socket(s): 2 NUMA node(s): 2 Model name: Intel(R) Xeon(R) Silver 4110 CPU @ 2.10GHz 2) setup scsi_debug: modprobe scsi_debug virtual_gb=128 max_luns=1 submit_queues=32 delay=0 max_queue=256 3) fio script: fio --rw=randread --size=128G --direct=1 --ioengine=libaio --iodepth=2048 \ --numjobs=32 --bs=4k --group_reporting=1 --group_reporting=1 --runtime=60 \ --loops=10000 --name=job1 --filename=/dev/sdN [mkp: fix device_busy reference in mpt3sas] Link: https://lore.kernel.org/r/20210122023317.687987-14-ming.lei@redhat.com Link: https://lore.kernel.org/linux-block/20200119071432.18558-6-ming.lei@redhat.com/ Cc: Omar Sandoval <osandov@fb.com> Cc: Kashyap Desai <kashyap.desai@broadcom.com> Cc: Sumanesh Samanta <sumanesh.samanta@broadcom.com> Cc: Ewan D. Milne <emilne@redhat.com> Cc: Hannes Reinecke <hare@suse.de> Tested-by: Sumanesh Samanta <sumanesh.samanta@broadcom.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2021-01-22 02:33:17 +00:00
int token = scsi_dev_queue_ready(q, sdev);
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
scsi: core: Replace sdev->device_busy with sbitmap SCSI currently uses an atomic variable to track queue depth for each attached device. The queue depth depends on many factors such as transport type and device implementation. In addition, the SCSI device queue depth is not a static entity but changes over time as a result of congestion management. While blk-mq currently tracks queue depth for each hctx, it can't easily be changed to accommodate the SCSI per-device requirement. The current approach of using an atomic variable doesn't scale well when there are lots of CPU cores and the disk is very fast. IOPS can be substantially impacted by the atomic in the hot path. Replace the atomic variable sdev->device_busy with an sbitmap for tracking the SCSI device queue depth. It has been observed that IOPS is improved ~30% by this patchset in the following test: 1) test machine(32 logical CPU cores) Thread(s) per core: 2 Core(s) per socket: 8 Socket(s): 2 NUMA node(s): 2 Model name: Intel(R) Xeon(R) Silver 4110 CPU @ 2.10GHz 2) setup scsi_debug: modprobe scsi_debug virtual_gb=128 max_luns=1 submit_queues=32 delay=0 max_queue=256 3) fio script: fio --rw=randread --size=128G --direct=1 --ioengine=libaio --iodepth=2048 \ --numjobs=32 --bs=4k --group_reporting=1 --group_reporting=1 --runtime=60 \ --loops=10000 --name=job1 --filename=/dev/sdN [mkp: fix device_busy reference in mpt3sas] Link: https://lore.kernel.org/r/20210122023317.687987-14-ming.lei@redhat.com Link: https://lore.kernel.org/linux-block/20200119071432.18558-6-ming.lei@redhat.com/ Cc: Omar Sandoval <osandov@fb.com> Cc: Kashyap Desai <kashyap.desai@broadcom.com> Cc: Sumanesh Samanta <sumanesh.samanta@broadcom.com> Cc: Ewan D. Milne <emilne@redhat.com> Cc: Hannes Reinecke <hare@suse.de> Tested-by: Sumanesh Samanta <sumanesh.samanta@broadcom.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2021-01-22 02:33:17 +00:00
if (token >= 0)
return token;
atomic_inc(&sdev->restarts);
/*
* Orders atomic_inc(&sdev->restarts) and atomic_read(&sdev->device_busy).
* .restarts must be incremented before .device_busy is read because the
* code in scsi_run_queue_async() depends on the order of these operations.
*/
smp_mb__after_atomic();
/*
* If all in-flight requests originated from this LUN are completed
* before reading .device_busy, sdev->device_busy will be observed as
* zero, then blk_mq_delay_run_hw_queues() will dispatch this request
* soon. Otherwise, completion of one of these requests will observe
* the .restarts flag, and the request queue will be run for handling
* this request, see scsi_end_request().
*/
if (unlikely(scsi_device_busy(sdev) == 0 &&
!scsi_device_blocked(sdev)))
blk_mq_delay_run_hw_queues(sdev->request_queue, SCSI_QUEUE_DELAY);
return -1;
}
static void scsi_mq_set_rq_budget_token(struct request *req, int token)
{
struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
cmd->budget_token = token;
}
static int scsi_mq_get_rq_budget_token(struct request *req)
{
struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
return cmd->budget_token;
}
static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
const struct blk_mq_queue_data *bd)
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
{
struct request *req = bd->rq;
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
struct request_queue *q = req->q;
struct scsi_device *sdev = q->queuedata;
struct Scsi_Host *shost = sdev->host;
struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
blk_status_t ret;
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
int reason;
WARN_ON_ONCE(cmd->budget_token < 0);
/*
* If the device is not in running state we will reject some or all
* commands.
*/
if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
ret = scsi_device_state_check(sdev, req);
if (ret != BLK_STS_OK)
goto out_put_budget;
}
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
ret = BLK_STS_RESOURCE;
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
if (!scsi_target_queue_ready(shost, sdev))
goto out_put_budget;
if (!scsi_host_queue_ready(q, shost, sdev, cmd))
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
goto out_dec_target_busy;
if (!(req->rq_flags & RQF_DONTPREP)) {
ret = scsi_prepare_cmd(req);
if (ret != BLK_STS_OK)
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
goto out_dec_host_busy;
req->rq_flags |= RQF_DONTPREP;
} else {
clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
}
cmd->flags &= SCMD_PRESERVED_FLAGS;
if (sdev->simple_tags)
cmd->flags |= SCMD_TAGGED;
if (bd->last)
cmd->flags |= SCMD_LAST;
scsi_set_resid(cmd, 0);
memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
cmd->scsi_done = scsi_mq_done;
blk_mq_start_request(req);
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
reason = scsi_dispatch_cmd(cmd);
if (reason) {
scsi_set_blocked(cmd, reason);
ret = BLK_STS_RESOURCE;
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
goto out_dec_host_busy;
}
return BLK_STS_OK;
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
out_dec_host_busy:
scsi_dec_host_busy(shost, cmd);
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
out_dec_target_busy:
if (scsi_target(sdev)->can_queue > 0)
atomic_dec(&scsi_target(sdev)->target_busy);
out_put_budget:
scsi_mq_put_budget(q, cmd->budget_token);
cmd->budget_token = -1;
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
switch (ret) {
case BLK_STS_OK:
break;
case BLK_STS_RESOURCE:
block: Introduce REQ_OP_ZONE_APPEND Define REQ_OP_ZONE_APPEND to append-write sectors to a zone of a zoned block device. This is a no-merge write operation. A zone append write BIO must: * Target a zoned block device * Have a sector position indicating the start sector of the target zone * The target zone must be a sequential write zone * The BIO must not cross a zone boundary * The BIO size must not be split to ensure that a single range of LBAs is written with a single command. Implement these checks in generic_make_request_checks() using the helper function blk_check_zone_append(). To avoid write append BIO splitting, introduce the new max_zone_append_sectors queue limit attribute and ensure that a BIO size is always lower than this limit. Export this new limit through sysfs and check these limits in bio_full(). Also when a LLDD can't dispatch a request to a specific zone, it will return BLK_STS_ZONE_RESOURCE indicating this request needs to be delayed, e.g. because the zone it will be dispatched to is still write-locked. If this happens set the request aside in a local list to continue trying dispatching requests such as READ requests or a WRITE/ZONE_APPEND requests targetting other zones. This way we can still keep a high queue depth without starving other requests even if one request can't be served due to zone write-locking. Finally, make sure that the bio sector position indicates the actual write position as indicated by the device on completion. Signed-off-by: Keith Busch <kbusch@kernel.org> [ jth: added zone-append specific add_page and merge_page helpers ] Signed-off-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2020-05-12 08:55:47 +00:00
case BLK_STS_ZONE_RESOURCE:
if (scsi_device_blocked(sdev))
blk-mq: introduce BLK_STS_DEV_RESOURCE This status is returned from driver to block layer if device related resource is unavailable, but driver can guarantee that IO dispatch will be triggered in future when the resource is available. Convert some drivers to return BLK_STS_DEV_RESOURCE. Also, if driver returns BLK_STS_RESOURCE and SCHED_RESTART is set, rerun queue after a delay (BLK_MQ_DELAY_QUEUE) to avoid IO stalls. BLK_MQ_DELAY_QUEUE is 3 ms because both scsi-mq and nvmefc are using that magic value. If a driver can make sure there is in-flight IO, it is safe to return BLK_STS_DEV_RESOURCE because: 1) If all in-flight IOs complete before examining SCHED_RESTART in blk_mq_dispatch_rq_list(), SCHED_RESTART must be cleared, so queue is run immediately in this case by blk_mq_dispatch_rq_list(); 2) if there is any in-flight IO after/when examining SCHED_RESTART in blk_mq_dispatch_rq_list(): - if SCHED_RESTART isn't set, queue is run immediately as handled in 1) - otherwise, this request will be dispatched after any in-flight IO is completed via blk_mq_sched_restart() 3) if SCHED_RESTART is set concurently in context because of BLK_STS_RESOURCE, blk_mq_delay_run_hw_queue() will cover the above two cases and make sure IO hang can be avoided. One invariant is that queue will be rerun if SCHED_RESTART is set. Suggested-by: Jens Axboe <axboe@kernel.dk> Tested-by: Laurence Oberman <loberman@redhat.com> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Mike Snitzer <snitzer@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2018-01-31 03:04:57 +00:00
ret = BLK_STS_DEV_RESOURCE;
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
break;
case BLK_STS_AGAIN:
scsi_req(req)->result = DID_BUS_BUSY << 16;
if (req->rq_flags & RQF_DONTPREP)
scsi_mq_uninit_cmd(cmd);
break;
default:
if (unlikely(!scsi_device_online(sdev)))
scsi_req(req)->result = DID_NO_CONNECT << 16;
else
scsi_req(req)->result = DID_ERROR << 16;
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
/*
* Make sure to release all allocated resources when
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
* we hit an error, as we will never see this command
* again.
*/
if (req->rq_flags & RQF_DONTPREP)
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
scsi_mq_uninit_cmd(cmd);
scsi: core: Run queue in case of I/O resource contention failure I/O requests may be held in scheduler queue because of resource contention. The starvation scenario was handled properly in the regular completion path but we failed to account for it during I/O submission. This lead to the hang captured below. Make sure we run the queue when resource contention is encountered in the submission path. [ 39.054963] scsi 13:0:0:0: rejecting I/O to dead device [ 39.058700] scsi 13:0:0:0: rejecting I/O to dead device [ 39.087855] sd 13:0:0:1: [sdd] Synchronizing SCSI cache [ 39.088909] scsi 13:0:0:1: rejecting I/O to dead device [ 39.095351] scsi 13:0:0:1: rejecting I/O to dead device [ 39.096962] scsi 13:0:0:1: rejecting I/O to dead device [ 247.021859] INFO: task scsi-stress-rem:813 blocked for more than 122 seconds. [ 247.023258] Not tainted 5.8.0-rc2 #8 [ 247.024069] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 247.025331] scsi-stress-rem D 0 813 802 0x00004000 [ 247.025334] Call Trace: [ 247.025354] __schedule+0x504/0x55f [ 247.027987] schedule+0x72/0xa8 [ 247.027991] blk_mq_freeze_queue_wait+0x63/0x8c [ 247.027994] ? do_wait_intr_irq+0x7a/0x7a [ 247.027996] blk_cleanup_queue+0x4b/0xc9 [ 247.028000] __scsi_remove_device+0xf6/0x14e [ 247.028002] scsi_remove_device+0x21/0x2b [ 247.029037] sdev_store_delete+0x58/0x7c [ 247.029041] kernfs_fop_write+0x10d/0x14f [ 247.031281] vfs_write+0xa2/0xdf [ 247.032670] ksys_write+0x6b/0xb3 [ 247.032673] do_syscall_64+0x56/0x82 [ 247.034053] entry_SYSCALL_64_after_hwframe+0x44/0xa9 [ 247.034059] RIP: 0033:0x7f69f39e9008 [ 247.036330] Code: Bad RIP value. [ 247.036331] RSP: 002b:00007ffdd8116498 EFLAGS: 00000246 ORIG_RAX: 0000000000000001 [ 247.037613] RAX: ffffffffffffffda RBX: 0000000000000002 RCX: 00007f69f39e9008 [ 247.039714] RDX: 0000000000000002 RSI: 000055cde92a0ab0 RDI: 0000000000000001 [ 247.039715] RBP: 000055cde92a0ab0 R08: 000000000000000a R09: 00007f69f3a79e80 [ 247.039716] R10: 000000000000000a R11: 0000000000000246 R12: 00007f69f3abb780 [ 247.039717] R13: 0000000000000002 R14: 00007f69f3ab6740 R15: 0000000000000002 Link: https://lore.kernel.org/r/20200720025435.812030-1-ming.lei@redhat.com Cc: linux-block@vger.kernel.org Cc: Christoph Hellwig <hch@lst.de> Reviewed-by: Bart Van Assche <bvanassche@acm.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2020-07-20 02:54:35 +00:00
scsi_run_queue_async(sdev);
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
break;
}
return ret;
}
static enum blk_eh_timer_return scsi_timeout(struct request *req,
bool reserved)
{
if (reserved)
return BLK_EH_RESET_TIMER;
return scsi_times_out(req);
}
static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
unsigned int hctx_idx, unsigned int numa_node)
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
{
struct Scsi_Host *shost = set->driver_data;
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
struct scatterlist *sg;
int ret = 0;
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
cmd->sense_buffer =
kmem_cache_alloc_node(scsi_sense_cache, GFP_KERNEL, numa_node);
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
if (!cmd->sense_buffer)
return -ENOMEM;
cmd->req.sense = cmd->sense_buffer;
if (scsi_host_get_prot(shost)) {
sg = (void *)cmd + sizeof(struct scsi_cmnd) +
shost->hostt->cmd_size;
cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
}
if (shost->hostt->init_cmd_priv) {
ret = shost->hostt->init_cmd_priv(shost, cmd);
if (ret < 0)
kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
}
return ret;
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
}
static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
unsigned int hctx_idx)
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
{
struct Scsi_Host *shost = set->driver_data;
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
if (shost->hostt->exit_cmd_priv)
shost->hostt->exit_cmd_priv(shost, cmd);
kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
}
static int scsi_mq_poll(struct blk_mq_hw_ctx *hctx)
{
struct Scsi_Host *shost = hctx->driver_data;
if (shost->hostt->mq_poll)
return shost->hostt->mq_poll(shost, hctx->queue_num);
return 0;
}
static int scsi_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
unsigned int hctx_idx)
{
struct Scsi_Host *shost = data;
hctx->driver_data = shost;
return 0;
}
static int scsi_map_queues(struct blk_mq_tag_set *set)
{
struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
if (shost->hostt->map_queues)
return shost->hostt->map_queues(shost);
return blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
}
void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
{
struct device *dev = shost->dma_dev;
/*
* this limit is imposed by hardware restrictions
*/
blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
SG_MAX_SEGMENTS));
if (scsi_host_prot_dma(shost)) {
shost->sg_prot_tablesize =
min_not_zero(shost->sg_prot_tablesize,
(unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
}
if (dev->dma_mask) {
shost->max_sectors = min_t(unsigned int, shost->max_sectors,
dma_max_mapping_size(dev) >> SECTOR_SHIFT);
}
blk_queue_max_hw_sectors(q, shost->max_sectors);
blk_queue_segment_boundary(q, shost->dma_boundary);
dma_set_seg_boundary(dev, shost->dma_boundary);
blk_queue_max_segment_size(q, shost->max_segment_size);
blk_queue_virt_boundary(q, shost->virt_boundary_mask);
dma_set_max_seg_size(dev, queue_max_segment_size(q));
/*
* Set a reasonable default alignment: The larger of 32-byte (dword),
* which is a common minimum for HBAs, and the minimum DMA alignment,
* which is set by the platform.
*
* Devices that require a bigger alignment can increase it later.
*/
blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
}
EXPORT_SYMBOL_GPL(__scsi_init_queue);
static const struct blk_mq_ops scsi_mq_ops_no_commit = {
.get_budget = scsi_mq_get_budget,
.put_budget = scsi_mq_put_budget,
.queue_rq = scsi_queue_rq,
.complete = scsi_complete,
.timeout = scsi_timeout,
#ifdef CONFIG_BLK_DEBUG_FS
.show_rq = scsi_show_rq,
#endif
.init_request = scsi_mq_init_request,
.exit_request = scsi_mq_exit_request,
.initialize_rq_fn = scsi_initialize_rq,
.cleanup_rq = scsi_cleanup_rq,
.busy = scsi_mq_lld_busy,
.map_queues = scsi_map_queues,
.init_hctx = scsi_init_hctx,
.poll = scsi_mq_poll,
.set_rq_budget_token = scsi_mq_set_rq_budget_token,
.get_rq_budget_token = scsi_mq_get_rq_budget_token,
};
static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx)
{
struct Scsi_Host *shost = hctx->driver_data;
shost->hostt->commit_rqs(shost, hctx->queue_num);
}
static const struct blk_mq_ops scsi_mq_ops = {
.get_budget = scsi_mq_get_budget,
.put_budget = scsi_mq_put_budget,
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
.queue_rq = scsi_queue_rq,
.commit_rqs = scsi_commit_rqs,
.complete = scsi_complete,
.timeout = scsi_timeout,
#ifdef CONFIG_BLK_DEBUG_FS
.show_rq = scsi_show_rq,
#endif
.init_request = scsi_mq_init_request,
.exit_request = scsi_mq_exit_request,
.initialize_rq_fn = scsi_initialize_rq,
.cleanup_rq = scsi_cleanup_rq,
.busy = scsi_mq_lld_busy,
.map_queues = scsi_map_queues,
.init_hctx = scsi_init_hctx,
.poll = scsi_mq_poll,
.set_rq_budget_token = scsi_mq_set_rq_budget_token,
.get_rq_budget_token = scsi_mq_get_rq_budget_token,
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
};
int scsi_mq_setup_tags(struct Scsi_Host *shost)
{
unsigned int cmd_size, sgl_size;
struct blk_mq_tag_set *tag_set = &shost->tag_set;
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
sgl_size = max_t(unsigned int, sizeof(struct scatterlist),
scsi_mq_inline_sgl_size(shost));
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
if (scsi_host_get_prot(shost))
cmd_size += sizeof(struct scsi_data_buffer) +
sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
memset(tag_set, 0, sizeof(*tag_set));
if (shost->hostt->commit_rqs)
tag_set->ops = &scsi_mq_ops;
else
tag_set->ops = &scsi_mq_ops_no_commit;
tag_set->nr_hw_queues = shost->nr_hw_queues ? : 1;
tag_set->nr_maps = shost->nr_maps ? : 1;
tag_set->queue_depth = shost->can_queue;
tag_set->cmd_size = cmd_size;
tag_set->numa_node = NUMA_NO_NODE;
tag_set->flags = BLK_MQ_F_SHOULD_MERGE;
tag_set->flags |=
BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
tag_set->driver_data = shost;
if (shost->host_tagset)
tag_set->flags |= BLK_MQ_F_TAG_HCTX_SHARED;
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
return blk_mq_alloc_tag_set(tag_set);
scsi: add support for a blk-mq based I/O path. This patch adds support for an alternate I/O path in the scsi midlayer which uses the blk-mq infrastructure instead of the legacy request code. Use of blk-mq is fully transparent to drivers, although for now a host template field is provided to opt out of blk-mq usage in case any unforseen incompatibilities arise. In general replacing the legacy request code with blk-mq is a simple and mostly mechanical transformation. The biggest exception is the new code that deals with the fact the I/O submissions in blk-mq must happen from process context, which slightly complicates the I/O completion handler. The second biggest differences is that blk-mq is build around the concept of preallocated requests that also include driver specific data, which in SCSI context means the scsi_cmnd structure. This completely avoids dynamic memory allocations for the fast path through I/O submission. Due the preallocated requests the MQ code path exclusively uses the host-wide shared tag allocator instead of a per-LUN one. This only affects drivers actually using the block layer provided tag allocator instead of their own. Unlike the old path blk-mq always provides a tag, although drivers don't have to use it. For now the blk-mq path is disable by defauly and must be enabled using the "use_blk_mq" module parameter. Once the remaining work in the block layer to make blk-mq more suitable for slow devices is complete I hope to make it the default and eventually even remove the old code path. Based on the earlier scsi-mq prototype by Nicholas Bellinger. Thanks to Bart Van Assche and Robert Elliot for testing, benchmarking and various sugestions and code contributions. Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com> Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Webb Scales <webbnh@hp.com> Acked-by: Jens Axboe <axboe@kernel.dk> Tested-by: Bart Van Assche <bvanassche@acm.org> Tested-by: Robert Elliott <elliott@hp.com>
2014-01-17 11:06:53 +00:00
}
void scsi_mq_destroy_tags(struct Scsi_Host *shost)
{
blk_mq_free_tag_set(&shost->tag_set);
}
/**
* scsi_device_from_queue - return sdev associated with a request_queue
* @q: The request queue to return the sdev from
*
* Return the sdev associated with a request queue or NULL if the
* request_queue does not reference a SCSI device.
*/
struct scsi_device *scsi_device_from_queue(struct request_queue *q)
{
struct scsi_device *sdev = NULL;
if (q->mq_ops == &scsi_mq_ops_no_commit ||
q->mq_ops == &scsi_mq_ops)
sdev = q->queuedata;
if (!sdev || !get_device(&sdev->sdev_gendev))
sdev = NULL;
return sdev;
}
/**
* scsi_block_requests - Utility function used by low-level drivers to prevent
* further commands from being queued to the device.
* @shost: host in question
*
* There is no timer nor any other means by which the requests get unblocked
* other than the low-level driver calling scsi_unblock_requests().
*/
void scsi_block_requests(struct Scsi_Host *shost)
{
shost->host_self_blocked = 1;
}
EXPORT_SYMBOL(scsi_block_requests);
/**
* scsi_unblock_requests - Utility function used by low-level drivers to allow
* further commands to be queued to the device.
* @shost: host in question
*
* There is no timer nor any other means by which the requests get unblocked
* other than the low-level driver calling scsi_unblock_requests(). This is done
* as an API function so that changes to the internals of the scsi mid-layer
* won't require wholesale changes to drivers that use this feature.
*/
void scsi_unblock_requests(struct Scsi_Host *shost)
{
shost->host_self_blocked = 0;
scsi_run_host_queues(shost);
}
EXPORT_SYMBOL(scsi_unblock_requests);
void scsi_exit_queue(void)
{
kmem_cache_destroy(scsi_sense_cache);
}
/**
* scsi_mode_select - issue a mode select
* @sdev: SCSI device to be queried
* @pf: Page format bit (1 == standard, 0 == vendor specific)
* @sp: Save page bit (0 == don't save, 1 == save)
* @modepage: mode page being requested
* @buffer: request buffer (may not be smaller than eight bytes)
* @len: length of request buffer.
* @timeout: command timeout
* @retries: number of retries before failing
* @data: returns a structure abstracting the mode header data
* @sshdr: place to put sense data (or NULL if no sense to be collected).
* must be SCSI_SENSE_BUFFERSIZE big.
*
* Returns zero if successful; negative error number or scsi
* status on error
*
*/
int
scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
unsigned char *buffer, int len, int timeout, int retries,
struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
{
unsigned char cmd[10];
unsigned char *real_buffer;
int ret;
memset(cmd, 0, sizeof(cmd));
cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
if (sdev->use_10_for_ms) {
if (len > 65535)
return -EINVAL;
real_buffer = kmalloc(8 + len, GFP_KERNEL);
if (!real_buffer)
return -ENOMEM;
memcpy(real_buffer + 8, buffer, len);
len += 8;
real_buffer[0] = 0;
real_buffer[1] = 0;
real_buffer[2] = data->medium_type;
real_buffer[3] = data->device_specific;
real_buffer[4] = data->longlba ? 0x01 : 0;
real_buffer[5] = 0;
real_buffer[6] = data->block_descriptor_length >> 8;
real_buffer[7] = data->block_descriptor_length;
cmd[0] = MODE_SELECT_10;
cmd[7] = len >> 8;
cmd[8] = len;
} else {
if (len > 255 || data->block_descriptor_length > 255 ||
data->longlba)
return -EINVAL;
real_buffer = kmalloc(4 + len, GFP_KERNEL);
if (!real_buffer)
return -ENOMEM;
memcpy(real_buffer + 4, buffer, len);
len += 4;
real_buffer[0] = 0;
real_buffer[1] = data->medium_type;
real_buffer[2] = data->device_specific;
real_buffer[3] = data->block_descriptor_length;
cmd[0] = MODE_SELECT;
cmd[4] = len;
}
ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
sshdr, timeout, retries, NULL);
kfree(real_buffer);
return ret;
}
EXPORT_SYMBOL_GPL(scsi_mode_select);
/**
* scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
* @sdev: SCSI device to be queried
* @dbd: set if mode sense will allow block descriptors to be returned
* @modepage: mode page being requested
* @buffer: request buffer (may not be smaller than eight bytes)
* @len: length of request buffer.
* @timeout: command timeout
* @retries: number of retries before failing
* @data: returns a structure abstracting the mode header data
* @sshdr: place to put sense data (or NULL if no sense to be collected).
* must be SCSI_SENSE_BUFFERSIZE big.
*
* Returns zero if successful, or a negative error number on failure
*/
int
scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
unsigned char *buffer, int len, int timeout, int retries,
struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
{
unsigned char cmd[12];
int use_10_for_ms;
int header_length;
int result, retry_count = retries;
struct scsi_sense_hdr my_sshdr;
memset(data, 0, sizeof(*data));
memset(&cmd[0], 0, 12);
dbd = sdev->set_dbd_for_ms ? 8 : dbd;
cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
cmd[2] = modepage;
/* caller might not be interested in sense, but we need it */
if (!sshdr)
sshdr = &my_sshdr;
retry:
use_10_for_ms = sdev->use_10_for_ms;
if (use_10_for_ms) {
if (len < 8)
len = 8;
cmd[0] = MODE_SENSE_10;
cmd[8] = len;
header_length = 8;
} else {
if (len < 4)
len = 4;
cmd[0] = MODE_SENSE;
cmd[4] = len;
header_length = 4;
}
memset(buffer, 0, len);
result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
sshdr, timeout, retries, NULL);
if (result < 0)
return result;
/* This code looks awful: what it's doing is making sure an
* ILLEGAL REQUEST sense return identifies the actual command
* byte as the problem. MODE_SENSE commands can return
* ILLEGAL REQUEST if the code page isn't supported */
if (!scsi_status_is_good(result)) {
if (scsi_sense_valid(sshdr)) {
if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
(sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
/*
* Invalid command operation code
*/
if (use_10_for_ms) {
sdev->use_10_for_ms = 0;
goto retry;
}
}
if (scsi_status_is_check_condition(result) &&
sshdr->sense_key == UNIT_ATTENTION &&
retry_count) {
retry_count--;
goto retry;
}
}
return -EIO;
}
if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
(modepage == 6 || modepage == 8))) {
/* Initio breakage? */
header_length = 0;
data->length = 13;
data->medium_type = 0;
data->device_specific = 0;
data->longlba = 0;
data->block_descriptor_length = 0;
} else if (use_10_for_ms) {
data->length = buffer[0]*256 + buffer[1] + 2;
data->medium_type = buffer[2];
data->device_specific = buffer[3];
data->longlba = buffer[4] & 0x01;
data->block_descriptor_length = buffer[6]*256
+ buffer[7];
} else {
data->length = buffer[0] + 1;
data->medium_type = buffer[1];
data->device_specific = buffer[2];
data->block_descriptor_length = buffer[3];
}
data->header_length = header_length;
return 0;
}
EXPORT_SYMBOL(scsi_mode_sense);
/**
* scsi_test_unit_ready - test if unit is ready
* @sdev: scsi device to change the state of.
* @timeout: command timeout
* @retries: number of retries before failing
* @sshdr: outpout pointer for decoded sense information.
*
* Returns zero if unsuccessful or an error if TUR failed. For
* removable media, UNIT_ATTENTION sets ->changed flag.
**/
int
scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
struct scsi_sense_hdr *sshdr)
{
char cmd[] = {
TEST_UNIT_READY, 0, 0, 0, 0, 0,
};
int result;
/* try to eat the UNIT_ATTENTION if there are enough retries */
do {
result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
timeout, 1, NULL);
if (sdev->removable && scsi_sense_valid(sshdr) &&
sshdr->sense_key == UNIT_ATTENTION)
sdev->changed = 1;
} while (scsi_sense_valid(sshdr) &&
sshdr->sense_key == UNIT_ATTENTION && --retries);
return result;
}
EXPORT_SYMBOL(scsi_test_unit_ready);
/**
* scsi_device_set_state - Take the given device through the device state model.
* @sdev: scsi device to change the state of.
* @state: state to change to.
*
* Returns zero if successful or an error if the requested
* transition is illegal.
*/
int
scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
{
enum scsi_device_state oldstate = sdev->sdev_state;
if (state == oldstate)
return 0;
switch (state) {
case SDEV_CREATED:
switch (oldstate) {
case SDEV_CREATED_BLOCK:
break;
default:
goto illegal;
}
break;
case SDEV_RUNNING:
switch (oldstate) {
case SDEV_CREATED:
case SDEV_OFFLINE:
case SDEV_TRANSPORT_OFFLINE:
case SDEV_QUIESCE:
case SDEV_BLOCK:
break;
default:
goto illegal;
}
break;
case SDEV_QUIESCE:
switch (oldstate) {
case SDEV_RUNNING:
case SDEV_OFFLINE:
case SDEV_TRANSPORT_OFFLINE:
break;
default:
goto illegal;
}
break;
case SDEV_OFFLINE:
case SDEV_TRANSPORT_OFFLINE:
switch (oldstate) {
case SDEV_CREATED:
case SDEV_RUNNING:
case SDEV_QUIESCE:
case SDEV_BLOCK:
break;
default:
goto illegal;
}
break;
case SDEV_BLOCK:
switch (oldstate) {
case SDEV_RUNNING:
case SDEV_CREATED_BLOCK:
case SDEV_QUIESCE:
case SDEV_OFFLINE:
break;
default:
goto illegal;
}
break;
case SDEV_CREATED_BLOCK:
switch (oldstate) {
case SDEV_CREATED:
break;
default:
goto illegal;
}
break;
case SDEV_CANCEL:
switch (oldstate) {
case SDEV_CREATED:
case SDEV_RUNNING:
case SDEV_QUIESCE:
case SDEV_OFFLINE:
case SDEV_TRANSPORT_OFFLINE:
break;
default:
goto illegal;
}
break;
case SDEV_DEL:
switch (oldstate) {
[SCSI] scsi: Device scanning oops for offlined devices (resend) If a device gets offlined as a result of the Inquiry sent during scanning, the following oops can occur. After the disk gets put into the SDEV_OFFLINE state, the error handler sends back the failed inquiry, which wakes the thread doing the scan. This starts a race between the scanning thread freeing the scsi device and the error handler calling scsi_run_host_queues to restart the host. Since the disk is in the SDEV_OFFLINE state, scsi_device_get will still work, which results in __scsi_iterate_devices getting a reference to the scsi disk when it shouldn't. The following execution thread causes the oops: CPU 0 (scan) CPU 1 (eh) --------------------------------------------------------- scsi_probe_and_add_lun .... scsi_eh_offline_sdevs scsi_eh_flush_done_q scsi_destroy_sdev scsi_device_dev_release scsi_restart_operations scsi_run_host_queues __scsi_iterate_devices get_device scsi_device_dev_release_usercontext scsi_run_queue <---OOPS---> The patch fixes this by changing the state of the sdev to SDEV_DEL before doing the final put_device, which should prevent the race from occurring. Original oops follows: Badness in kref_get at lib/kref.c:32 Call Trace: [C00000002F4476D0] [C00000000000EE20] .show_stack+0x68/0x1b0 (unreliable) [C00000002F447770] [C00000000037515C] .program_check_exception+0x1cc/0x5a8 [C00000002F447840] [C00000000000446C] program_check_common+0xec/0x100 Exception: 700 at .kref_get+0x10/0x28 LR = .kobject_get+0x20/0x3c [C00000002F447B30] [C00000002F447BC0] 0xc00000002f447bc0 (unreliable) [C00000002F447BB0] [C000000000254BDC] .get_device+0x20/0x3c [C00000002F447C30] [D000000000063188] .scsi_device_get+0x34/0xdc [scsi_mod] [C00000002F447CC0] [D0000000000633EC] .__scsi_iterate_devices+0x50/0xbc [scsi_mod] [C00000002F447D60] [D00000000006A910] .scsi_run_host_queues+0x34/0x5c [scsi_mod] [C00000002F447DF0] [D000000000069054] .scsi_error_handler+0xdb4/0xe44 [scsi_mod] [C00000002F447EE0] [C00000000007B4E0] .kthread+0x128/0x178 [C00000002F447F90] [C000000000025E84] .kernel_thread+0x4c/0x68 Unable to handle kernel paging request for <7>PCI: Enabling device: (0002:41:01.1), cmd 143 data at address 0x000001b8 Faulting instruction address: 0xd0000000000698e4 sym1: <1010-66> rev 0x1 at pci 0002:41:01.1 irq 216 sym1: No NVRAM, ID 7, Fast-80, LVD, parity checking sym1: SCSI BUS has been reset. scsi2 : sym-2.2.2 cpu 0x0: Vector: 300 (Data Access) at [c00000002f447a30] pc: d0000000000698e4: .scsi_run_queue+0x2c/0x218 [scsi_mod] lr: d00000000006a904: .scsi_run_host_queues+0x28/0x5c [scsi_mod] sp: c00000002f447cb0 msr: 9000000000009032 dar: 1b8 dsisr: 40000000 current = 0xc0000000045fecd0 paca = 0xc00000000048ee80 pid = 1123, comm = scsi_eh_1 enter ? for help [c00000002f447d60] d00000000006a904 .scsi_run_host_queues+0x28/0x5c [scsi_mod] [c00000002f447df0] d000000000069054 .scsi_error_handler+0xdb4/0xe44 [scsi_mod] [c00000002f447ee0] c00000000007b4e0 .kthread+0x128/0x178 [c00000002f447f90] c000000000025e84 .kernel_thread+0x4c/0x68 Signed-off-by: Brian King <brking@us.ibm.com> Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
2006-06-27 16:10:43 +00:00
case SDEV_CREATED:
case SDEV_RUNNING:
case SDEV_OFFLINE:
case SDEV_TRANSPORT_OFFLINE:
case SDEV_CANCEL:
case SDEV_BLOCK:
case SDEV_CREATED_BLOCK:
break;
default:
goto illegal;
}
break;
}
sdev->offline_already = false;
sdev->sdev_state = state;
return 0;
illegal:
SCSI_LOG_ERROR_RECOVERY(1,
sdev_printk(KERN_ERR, sdev,
"Illegal state transition %s->%s",
scsi_device_state_name(oldstate),
scsi_device_state_name(state))
);
return -EINVAL;
}
EXPORT_SYMBOL(scsi_device_set_state);
/**
* scsi_evt_emit - emit a single SCSI device uevent
* @sdev: associated SCSI device
* @evt: event to emit
*
* Send a single uevent (scsi_event) to the associated scsi_device.
*/
static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
{
int idx = 0;
char *envp[3];
switch (evt->evt_type) {
case SDEV_EVT_MEDIA_CHANGE:
envp[idx++] = "SDEV_MEDIA_CHANGE=1";
break;
case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
scsi_rescan_device(&sdev->sdev_gendev);
envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
break;
case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
break;
case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
break;
case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
break;
case SDEV_EVT_LUN_CHANGE_REPORTED:
envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
break;
case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
break;
case SDEV_EVT_POWER_ON_RESET_OCCURRED:
envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
break;
default:
/* do nothing */
break;
}
envp[idx++] = NULL;
kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
}
/**
* scsi_evt_thread - send a uevent for each scsi event
* @work: work struct for scsi_device
*
* Dispatch queued events to their associated scsi_device kobjects
* as uevents.
*/
void scsi_evt_thread(struct work_struct *work)
{
struct scsi_device *sdev;
enum scsi_device_event evt_type;
LIST_HEAD(event_list);
sdev = container_of(work, struct scsi_device, event_work);
for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
if (test_and_clear_bit(evt_type, sdev->pending_events))
sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
while (1) {
struct scsi_event *evt;
struct list_head *this, *tmp;
unsigned long flags;
spin_lock_irqsave(&sdev->list_lock, flags);
list_splice_init(&sdev->event_list, &event_list);
spin_unlock_irqrestore(&sdev->list_lock, flags);
if (list_empty(&event_list))
break;
list_for_each_safe(this, tmp, &event_list) {
evt = list_entry(this, struct scsi_event, node);
list_del(&evt->node);
scsi_evt_emit(sdev, evt);
kfree(evt);
}
}
}
/**
* sdev_evt_send - send asserted event to uevent thread
* @sdev: scsi_device event occurred on
* @evt: event to send
*
* Assert scsi device event asynchronously.
*/
void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
{
unsigned long flags;
#if 0
/* FIXME: currently this check eliminates all media change events
* for polled devices. Need to update to discriminate between AN
* and polled events */
if (!test_bit(evt->evt_type, sdev->supported_events)) {
kfree(evt);
return;
}
#endif
spin_lock_irqsave(&sdev->list_lock, flags);
list_add_tail(&evt->node, &sdev->event_list);
schedule_work(&sdev->event_work);
spin_unlock_irqrestore(&sdev->list_lock, flags);
}
EXPORT_SYMBOL_GPL(sdev_evt_send);
/**
* sdev_evt_alloc - allocate a new scsi event
* @evt_type: type of event to allocate
* @gfpflags: GFP flags for allocation
*
* Allocates and returns a new scsi_event.
*/
struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
gfp_t gfpflags)
{
struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
if (!evt)
return NULL;
evt->evt_type = evt_type;
INIT_LIST_HEAD(&evt->node);
/* evt_type-specific initialization, if any */
switch (evt_type) {
case SDEV_EVT_MEDIA_CHANGE:
case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
case SDEV_EVT_LUN_CHANGE_REPORTED:
case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
case SDEV_EVT_POWER_ON_RESET_OCCURRED:
default:
/* do nothing */
break;
}
return evt;
}
EXPORT_SYMBOL_GPL(sdev_evt_alloc);
/**
* sdev_evt_send_simple - send asserted event to uevent thread
* @sdev: scsi_device event occurred on
* @evt_type: type of event to send
* @gfpflags: GFP flags for allocation
*
* Assert scsi device event asynchronously, given an event type.
*/
void sdev_evt_send_simple(struct scsi_device *sdev,
enum scsi_device_event evt_type, gfp_t gfpflags)
{
struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
if (!evt) {
sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
evt_type);
return;
}
sdev_evt_send(sdev, evt);
}
EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
/**
* scsi_device_quiesce - Block all commands except power management.
* @sdev: scsi device to quiesce.
*
* This works by trying to transition to the SDEV_QUIESCE state
* (which must be a legal transition). When the device is in this
* state, only power management requests will be accepted, all others will
* be deferred.
*
* Must be called with user context, may sleep.
*
* Returns zero if unsuccessful or an error if not.
*/
int
scsi_device_quiesce(struct scsi_device *sdev)
{
block, scsi: Make SCSI quiesce and resume work reliably The contexts from which a SCSI device can be quiesced or resumed are: * Writing into /sys/class/scsi_device/*/device/state. * SCSI parallel (SPI) domain validation. * The SCSI device power management methods. See also scsi_bus_pm_ops. It is essential during suspend and resume that neither the filesystem state nor the filesystem metadata in RAM changes. This is why while the hibernation image is being written or restored that SCSI devices are quiesced. The SCSI core quiesces devices through scsi_device_quiesce() and scsi_device_resume(). In the SDEV_QUIESCE state execution of non-preempt requests is deferred. This is realized by returning BLKPREP_DEFER from inside scsi_prep_state_check() for quiesced SCSI devices. Avoid that a full queue prevents power management requests to be submitted by deferring allocation of non-preempt requests for devices in the quiesced state. This patch has been tested by running the following commands and by verifying that after each resume the fio job was still running: for ((i=0; i<10; i++)); do ( cd /sys/block/md0/md && while true; do [ "$(<sync_action)" = "idle" ] && echo check > sync_action sleep 1 done ) & pids=($!) for d in /sys/class/block/sd*[a-z]; do bdev=${d#/sys/class/block/} hcil=$(readlink "$d/device") hcil=${hcil#../../../} echo 4 > "$d/queue/nr_requests" echo 1 > "/sys/class/scsi_device/$hcil/device/queue_depth" fio --name="$bdev" --filename="/dev/$bdev" --buffered=0 --bs=512 \ --rw=randread --ioengine=libaio --numjobs=4 --iodepth=16 \ --iodepth_batch=1 --thread --loops=$((2**31)) & pids+=($!) done sleep 1 echo "$(date) Hibernating ..." >>hibernate-test-log.txt systemctl hibernate sleep 10 kill "${pids[@]}" echo idle > /sys/block/md0/md/sync_action wait echo "$(date) Done." >>hibernate-test-log.txt done Reported-by: Oleksandr Natalenko <oleksandr@natalenko.name> References: "I/O hangs after resuming from suspend-to-ram" (https://marc.info/?l=linux-block&m=150340235201348). Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Martin Steigerwald <martin@lichtvoll.de> Tested-by: Oleksandr Natalenko <oleksandr@natalenko.name> Cc: Martin K. Petersen <martin.petersen@oracle.com> Cc: Ming Lei <ming.lei@redhat.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Johannes Thumshirn <jthumshirn@suse.de> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2017-11-09 18:49:58 +00:00
struct request_queue *q = sdev->request_queue;
int err;
block, scsi: Make SCSI quiesce and resume work reliably The contexts from which a SCSI device can be quiesced or resumed are: * Writing into /sys/class/scsi_device/*/device/state. * SCSI parallel (SPI) domain validation. * The SCSI device power management methods. See also scsi_bus_pm_ops. It is essential during suspend and resume that neither the filesystem state nor the filesystem metadata in RAM changes. This is why while the hibernation image is being written or restored that SCSI devices are quiesced. The SCSI core quiesces devices through scsi_device_quiesce() and scsi_device_resume(). In the SDEV_QUIESCE state execution of non-preempt requests is deferred. This is realized by returning BLKPREP_DEFER from inside scsi_prep_state_check() for quiesced SCSI devices. Avoid that a full queue prevents power management requests to be submitted by deferring allocation of non-preempt requests for devices in the quiesced state. This patch has been tested by running the following commands and by verifying that after each resume the fio job was still running: for ((i=0; i<10; i++)); do ( cd /sys/block/md0/md && while true; do [ "$(<sync_action)" = "idle" ] && echo check > sync_action sleep 1 done ) & pids=($!) for d in /sys/class/block/sd*[a-z]; do bdev=${d#/sys/class/block/} hcil=$(readlink "$d/device") hcil=${hcil#../../../} echo 4 > "$d/queue/nr_requests" echo 1 > "/sys/class/scsi_device/$hcil/device/queue_depth" fio --name="$bdev" --filename="/dev/$bdev" --buffered=0 --bs=512 \ --rw=randread --ioengine=libaio --numjobs=4 --iodepth=16 \ --iodepth_batch=1 --thread --loops=$((2**31)) & pids+=($!) done sleep 1 echo "$(date) Hibernating ..." >>hibernate-test-log.txt systemctl hibernate sleep 10 kill "${pids[@]}" echo idle > /sys/block/md0/md/sync_action wait echo "$(date) Done." >>hibernate-test-log.txt done Reported-by: Oleksandr Natalenko <oleksandr@natalenko.name> References: "I/O hangs after resuming from suspend-to-ram" (https://marc.info/?l=linux-block&m=150340235201348). Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Martin Steigerwald <martin@lichtvoll.de> Tested-by: Oleksandr Natalenko <oleksandr@natalenko.name> Cc: Martin K. Petersen <martin.petersen@oracle.com> Cc: Ming Lei <ming.lei@redhat.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Johannes Thumshirn <jthumshirn@suse.de> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2017-11-09 18:49:58 +00:00
/*
* It is allowed to call scsi_device_quiesce() multiple times from
* the same context but concurrent scsi_device_quiesce() calls are
* not allowed.
*/
WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
if (sdev->quiesced_by == current)
return 0;
blk_set_pm_only(q);
block, scsi: Make SCSI quiesce and resume work reliably The contexts from which a SCSI device can be quiesced or resumed are: * Writing into /sys/class/scsi_device/*/device/state. * SCSI parallel (SPI) domain validation. * The SCSI device power management methods. See also scsi_bus_pm_ops. It is essential during suspend and resume that neither the filesystem state nor the filesystem metadata in RAM changes. This is why while the hibernation image is being written or restored that SCSI devices are quiesced. The SCSI core quiesces devices through scsi_device_quiesce() and scsi_device_resume(). In the SDEV_QUIESCE state execution of non-preempt requests is deferred. This is realized by returning BLKPREP_DEFER from inside scsi_prep_state_check() for quiesced SCSI devices. Avoid that a full queue prevents power management requests to be submitted by deferring allocation of non-preempt requests for devices in the quiesced state. This patch has been tested by running the following commands and by verifying that after each resume the fio job was still running: for ((i=0; i<10; i++)); do ( cd /sys/block/md0/md && while true; do [ "$(<sync_action)" = "idle" ] && echo check > sync_action sleep 1 done ) & pids=($!) for d in /sys/class/block/sd*[a-z]; do bdev=${d#/sys/class/block/} hcil=$(readlink "$d/device") hcil=${hcil#../../../} echo 4 > "$d/queue/nr_requests" echo 1 > "/sys/class/scsi_device/$hcil/device/queue_depth" fio --name="$bdev" --filename="/dev/$bdev" --buffered=0 --bs=512 \ --rw=randread --ioengine=libaio --numjobs=4 --iodepth=16 \ --iodepth_batch=1 --thread --loops=$((2**31)) & pids+=($!) done sleep 1 echo "$(date) Hibernating ..." >>hibernate-test-log.txt systemctl hibernate sleep 10 kill "${pids[@]}" echo idle > /sys/block/md0/md/sync_action wait echo "$(date) Done." >>hibernate-test-log.txt done Reported-by: Oleksandr Natalenko <oleksandr@natalenko.name> References: "I/O hangs after resuming from suspend-to-ram" (https://marc.info/?l=linux-block&m=150340235201348). Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Martin Steigerwald <martin@lichtvoll.de> Tested-by: Oleksandr Natalenko <oleksandr@natalenko.name> Cc: Martin K. Petersen <martin.petersen@oracle.com> Cc: Ming Lei <ming.lei@redhat.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Johannes Thumshirn <jthumshirn@suse.de> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2017-11-09 18:49:58 +00:00
blk_mq_freeze_queue(q);
/*
* Ensure that the effect of blk_set_pm_only() will be visible
block, scsi: Make SCSI quiesce and resume work reliably The contexts from which a SCSI device can be quiesced or resumed are: * Writing into /sys/class/scsi_device/*/device/state. * SCSI parallel (SPI) domain validation. * The SCSI device power management methods. See also scsi_bus_pm_ops. It is essential during suspend and resume that neither the filesystem state nor the filesystem metadata in RAM changes. This is why while the hibernation image is being written or restored that SCSI devices are quiesced. The SCSI core quiesces devices through scsi_device_quiesce() and scsi_device_resume(). In the SDEV_QUIESCE state execution of non-preempt requests is deferred. This is realized by returning BLKPREP_DEFER from inside scsi_prep_state_check() for quiesced SCSI devices. Avoid that a full queue prevents power management requests to be submitted by deferring allocation of non-preempt requests for devices in the quiesced state. This patch has been tested by running the following commands and by verifying that after each resume the fio job was still running: for ((i=0; i<10; i++)); do ( cd /sys/block/md0/md && while true; do [ "$(<sync_action)" = "idle" ] && echo check > sync_action sleep 1 done ) & pids=($!) for d in /sys/class/block/sd*[a-z]; do bdev=${d#/sys/class/block/} hcil=$(readlink "$d/device") hcil=${hcil#../../../} echo 4 > "$d/queue/nr_requests" echo 1 > "/sys/class/scsi_device/$hcil/device/queue_depth" fio --name="$bdev" --filename="/dev/$bdev" --buffered=0 --bs=512 \ --rw=randread --ioengine=libaio --numjobs=4 --iodepth=16 \ --iodepth_batch=1 --thread --loops=$((2**31)) & pids+=($!) done sleep 1 echo "$(date) Hibernating ..." >>hibernate-test-log.txt systemctl hibernate sleep 10 kill "${pids[@]}" echo idle > /sys/block/md0/md/sync_action wait echo "$(date) Done." >>hibernate-test-log.txt done Reported-by: Oleksandr Natalenko <oleksandr@natalenko.name> References: "I/O hangs after resuming from suspend-to-ram" (https://marc.info/?l=linux-block&m=150340235201348). Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Martin Steigerwald <martin@lichtvoll.de> Tested-by: Oleksandr Natalenko <oleksandr@natalenko.name> Cc: Martin K. Petersen <martin.petersen@oracle.com> Cc: Ming Lei <ming.lei@redhat.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Johannes Thumshirn <jthumshirn@suse.de> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2017-11-09 18:49:58 +00:00
* for percpu_ref_tryget() callers that occur after the queue
* unfreeze even if the queue was already frozen before this function
* was called. See also https://lwn.net/Articles/573497/.
*/
synchronize_rcu();
blk_mq_unfreeze_queue(q);
mutex_lock(&sdev->state_mutex);
err = scsi_device_set_state(sdev, SDEV_QUIESCE);
block, scsi: Make SCSI quiesce and resume work reliably The contexts from which a SCSI device can be quiesced or resumed are: * Writing into /sys/class/scsi_device/*/device/state. * SCSI parallel (SPI) domain validation. * The SCSI device power management methods. See also scsi_bus_pm_ops. It is essential during suspend and resume that neither the filesystem state nor the filesystem metadata in RAM changes. This is why while the hibernation image is being written or restored that SCSI devices are quiesced. The SCSI core quiesces devices through scsi_device_quiesce() and scsi_device_resume(). In the SDEV_QUIESCE state execution of non-preempt requests is deferred. This is realized by returning BLKPREP_DEFER from inside scsi_prep_state_check() for quiesced SCSI devices. Avoid that a full queue prevents power management requests to be submitted by deferring allocation of non-preempt requests for devices in the quiesced state. This patch has been tested by running the following commands and by verifying that after each resume the fio job was still running: for ((i=0; i<10; i++)); do ( cd /sys/block/md0/md && while true; do [ "$(<sync_action)" = "idle" ] && echo check > sync_action sleep 1 done ) & pids=($!) for d in /sys/class/block/sd*[a-z]; do bdev=${d#/sys/class/block/} hcil=$(readlink "$d/device") hcil=${hcil#../../../} echo 4 > "$d/queue/nr_requests" echo 1 > "/sys/class/scsi_device/$hcil/device/queue_depth" fio --name="$bdev" --filename="/dev/$bdev" --buffered=0 --bs=512 \ --rw=randread --ioengine=libaio --numjobs=4 --iodepth=16 \ --iodepth_batch=1 --thread --loops=$((2**31)) & pids+=($!) done sleep 1 echo "$(date) Hibernating ..." >>hibernate-test-log.txt systemctl hibernate sleep 10 kill "${pids[@]}" echo idle > /sys/block/md0/md/sync_action wait echo "$(date) Done." >>hibernate-test-log.txt done Reported-by: Oleksandr Natalenko <oleksandr@natalenko.name> References: "I/O hangs after resuming from suspend-to-ram" (https://marc.info/?l=linux-block&m=150340235201348). Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Martin Steigerwald <martin@lichtvoll.de> Tested-by: Oleksandr Natalenko <oleksandr@natalenko.name> Cc: Martin K. Petersen <martin.petersen@oracle.com> Cc: Ming Lei <ming.lei@redhat.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Johannes Thumshirn <jthumshirn@suse.de> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2017-11-09 18:49:58 +00:00
if (err == 0)
sdev->quiesced_by = current;
else
blk_clear_pm_only(q);
mutex_unlock(&sdev->state_mutex);
block, scsi: Make SCSI quiesce and resume work reliably The contexts from which a SCSI device can be quiesced or resumed are: * Writing into /sys/class/scsi_device/*/device/state. * SCSI parallel (SPI) domain validation. * The SCSI device power management methods. See also scsi_bus_pm_ops. It is essential during suspend and resume that neither the filesystem state nor the filesystem metadata in RAM changes. This is why while the hibernation image is being written or restored that SCSI devices are quiesced. The SCSI core quiesces devices through scsi_device_quiesce() and scsi_device_resume(). In the SDEV_QUIESCE state execution of non-preempt requests is deferred. This is realized by returning BLKPREP_DEFER from inside scsi_prep_state_check() for quiesced SCSI devices. Avoid that a full queue prevents power management requests to be submitted by deferring allocation of non-preempt requests for devices in the quiesced state. This patch has been tested by running the following commands and by verifying that after each resume the fio job was still running: for ((i=0; i<10; i++)); do ( cd /sys/block/md0/md && while true; do [ "$(<sync_action)" = "idle" ] && echo check > sync_action sleep 1 done ) & pids=($!) for d in /sys/class/block/sd*[a-z]; do bdev=${d#/sys/class/block/} hcil=$(readlink "$d/device") hcil=${hcil#../../../} echo 4 > "$d/queue/nr_requests" echo 1 > "/sys/class/scsi_device/$hcil/device/queue_depth" fio --name="$bdev" --filename="/dev/$bdev" --buffered=0 --bs=512 \ --rw=randread --ioengine=libaio --numjobs=4 --iodepth=16 \ --iodepth_batch=1 --thread --loops=$((2**31)) & pids+=($!) done sleep 1 echo "$(date) Hibernating ..." >>hibernate-test-log.txt systemctl hibernate sleep 10 kill "${pids[@]}" echo idle > /sys/block/md0/md/sync_action wait echo "$(date) Done." >>hibernate-test-log.txt done Reported-by: Oleksandr Natalenko <oleksandr@natalenko.name> References: "I/O hangs after resuming from suspend-to-ram" (https://marc.info/?l=linux-block&m=150340235201348). Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com> Reviewed-by: Hannes Reinecke <hare@suse.com> Tested-by: Martin Steigerwald <martin@lichtvoll.de> Tested-by: Oleksandr Natalenko <oleksandr@natalenko.name> Cc: Martin K. Petersen <martin.petersen@oracle.com> Cc: Ming Lei <ming.lei@redhat.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Johannes Thumshirn <jthumshirn@suse.de> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2017-11-09 18:49:58 +00:00
return err;
}
EXPORT_SYMBOL(scsi_device_quiesce);
/**
* scsi_device_resume - Restart user issued commands to a quiesced device.
* @sdev: scsi device to resume.
*
* Moves the device from quiesced back to running and restarts the
* queues.
*
* Must be called with user context, may sleep.
*/
[SCSI] sd: limit the scope of the async probe domain sd injects and synchronizes probe work on the global kernel-wide domain. This runs into conflict with PM that wants to perform resume actions in async context: [ 494.237079] INFO: task kworker/u:3:554 blocked for more than 120 seconds. [ 494.294396] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 494.360809] kworker/u:3 D 0000000000000000 0 554 2 0x00000000 [ 494.420739] ffff88012e4d3af0 0000000000000046 ffff88013200c160 ffff88012e4d3fd8 [ 494.484392] ffff88012e4d3fd8 0000000000012500 ffff8801394ea0b0 ffff88013200c160 [ 494.548038] ffff88012e4d3ae0 00000000000001e3 ffffffff81a249e0 ffff8801321c5398 [ 494.611685] Call Trace: [ 494.632649] [<ffffffff8149dd25>] schedule+0x5a/0x5c [ 494.674687] [<ffffffff8104b968>] async_synchronize_cookie_domain+0xb6/0x112 [ 494.734177] [<ffffffff810461ff>] ? __init_waitqueue_head+0x50/0x50 [ 494.787134] [<ffffffff8131a224>] ? scsi_remove_target+0x48/0x48 [ 494.837900] [<ffffffff8104b9d9>] async_synchronize_cookie+0x15/0x17 [ 494.891567] [<ffffffff8104ba49>] async_synchronize_full+0x54/0x70 <-- here we wait for async contexts to complete [ 494.943783] [<ffffffff8104b9f5>] ? async_synchronize_full_domain+0x1a/0x1a [ 495.002547] [<ffffffffa00114b1>] sd_remove+0x2c/0xa2 [sd_mod] [ 495.051861] [<ffffffff812fe94f>] __device_release_driver+0x86/0xcf [ 495.104807] [<ffffffff812fe9bd>] device_release_driver+0x25/0x32 <-- here we take device_lock() [ 853.511341] INFO: task kworker/u:4:549 blocked for more than 120 seconds. [ 853.568693] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 853.635119] kworker/u:4 D ffff88013097b5d0 0 549 2 0x00000000 [ 853.695129] ffff880132773c40 0000000000000046 ffff880130790000 ffff880132773fd8 [ 853.758990] ffff880132773fd8 0000000000012500 ffff88013288a0b0 ffff880130790000 [ 853.822796] 0000000000000246 0000000000000040 ffff88013097b5c8 ffff880130790000 [ 853.886633] Call Trace: [ 853.907631] [<ffffffff8149dd25>] schedule+0x5a/0x5c [ 853.949670] [<ffffffff8149cc44>] __mutex_lock_common+0x220/0x351 [ 854.001225] [<ffffffff81304bd7>] ? device_resume+0x58/0x1c4 [ 854.049082] [<ffffffff81304bd7>] ? device_resume+0x58/0x1c4 [ 854.097011] [<ffffffff8149ce48>] mutex_lock_nested+0x2f/0x36 <-- here we wait for device_lock() [ 854.145591] [<ffffffff81304bd7>] device_resume+0x58/0x1c4 [ 854.192066] [<ffffffff81304d61>] async_resume+0x1e/0x45 [ 854.237019] [<ffffffff8104bc93>] async_run_entry_fn+0xc6/0x173 <-- ...while running in async context Provide a 'scsi_sd_probe_domain' so that async probe actions actions can be flushed without regard for the state of PM, and allow for the resume path to handle devices that have transitioned from SDEV_QUIESCE to SDEV_DEL prior to resume. Acked-by: Alan Stern <stern@rowland.harvard.edu> [alan: uplevel scsi_sd_probe_domain, clarify scsi_device_resume] Signed-off-by: Dan Williams <dan.j.williams@intel.com> [jejb: remove unneeded config guards in include file] Signed-off-by: James Bottomley <JBottomley@Parallels.com>
2012-03-23 00:05:11 +00:00
void scsi_device_resume(struct scsi_device *sdev)
{
[SCSI] sd: limit the scope of the async probe domain sd injects and synchronizes probe work on the global kernel-wide domain. This runs into conflict with PM that wants to perform resume actions in async context: [ 494.237079] INFO: task kworker/u:3:554 blocked for more than 120 seconds. [ 494.294396] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 494.360809] kworker/u:3 D 0000000000000000 0 554 2 0x00000000 [ 494.420739] ffff88012e4d3af0 0000000000000046 ffff88013200c160 ffff88012e4d3fd8 [ 494.484392] ffff88012e4d3fd8 0000000000012500 ffff8801394ea0b0 ffff88013200c160 [ 494.548038] ffff88012e4d3ae0 00000000000001e3 ffffffff81a249e0 ffff8801321c5398 [ 494.611685] Call Trace: [ 494.632649] [<ffffffff8149dd25>] schedule+0x5a/0x5c [ 494.674687] [<ffffffff8104b968>] async_synchronize_cookie_domain+0xb6/0x112 [ 494.734177] [<ffffffff810461ff>] ? __init_waitqueue_head+0x50/0x50 [ 494.787134] [<ffffffff8131a224>] ? scsi_remove_target+0x48/0x48 [ 494.837900] [<ffffffff8104b9d9>] async_synchronize_cookie+0x15/0x17 [ 494.891567] [<ffffffff8104ba49>] async_synchronize_full+0x54/0x70 <-- here we wait for async contexts to complete [ 494.943783] [<ffffffff8104b9f5>] ? async_synchronize_full_domain+0x1a/0x1a [ 495.002547] [<ffffffffa00114b1>] sd_remove+0x2c/0xa2 [sd_mod] [ 495.051861] [<ffffffff812fe94f>] __device_release_driver+0x86/0xcf [ 495.104807] [<ffffffff812fe9bd>] device_release_driver+0x25/0x32 <-- here we take device_lock() [ 853.511341] INFO: task kworker/u:4:549 blocked for more than 120 seconds. [ 853.568693] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 853.635119] kworker/u:4 D ffff88013097b5d0 0 549 2 0x00000000 [ 853.695129] ffff880132773c40 0000000000000046 ffff880130790000 ffff880132773fd8 [ 853.758990] ffff880132773fd8 0000000000012500 ffff88013288a0b0 ffff880130790000 [ 853.822796] 0000000000000246 0000000000000040 ffff88013097b5c8 ffff880130790000 [ 853.886633] Call Trace: [ 853.907631] [<ffffffff8149dd25>] schedule+0x5a/0x5c [ 853.949670] [<ffffffff8149cc44>] __mutex_lock_common+0x220/0x351 [ 854.001225] [<ffffffff81304bd7>] ? device_resume+0x58/0x1c4 [ 854.049082] [<ffffffff81304bd7>] ? device_resume+0x58/0x1c4 [ 854.097011] [<ffffffff8149ce48>] mutex_lock_nested+0x2f/0x36 <-- here we wait for device_lock() [ 854.145591] [<ffffffff81304bd7>] device_resume+0x58/0x1c4 [ 854.192066] [<ffffffff81304d61>] async_resume+0x1e/0x45 [ 854.237019] [<ffffffff8104bc93>] async_run_entry_fn+0xc6/0x173 <-- ...while running in async context Provide a 'scsi_sd_probe_domain' so that async probe actions actions can be flushed without regard for the state of PM, and allow for the resume path to handle devices that have transitioned from SDEV_QUIESCE to SDEV_DEL prior to resume. Acked-by: Alan Stern <stern@rowland.harvard.edu> [alan: uplevel scsi_sd_probe_domain, clarify scsi_device_resume] Signed-off-by: Dan Williams <dan.j.williams@intel.com> [jejb: remove unneeded config guards in include file] Signed-off-by: James Bottomley <JBottomley@Parallels.com>
2012-03-23 00:05:11 +00:00
/* check if the device state was mutated prior to resume, and if
* so assume the state is being managed elsewhere (for example
* device deleted during suspend)
*/
mutex_lock(&sdev->state_mutex);
if (sdev->sdev_state == SDEV_QUIESCE)
scsi_device_set_state(sdev, SDEV_RUNNING);
2019-03-15 23:27:58 +00:00
if (sdev->quiesced_by) {
sdev->quiesced_by = NULL;
blk_clear_pm_only(sdev->request_queue);
}
mutex_unlock(&sdev->state_mutex);
}
EXPORT_SYMBOL(scsi_device_resume);
static void
device_quiesce_fn(struct scsi_device *sdev, void *data)
{
scsi_device_quiesce(sdev);
}
void
scsi_target_quiesce(struct scsi_target *starget)
{
starget_for_each_device(starget, NULL, device_quiesce_fn);
}
EXPORT_SYMBOL(scsi_target_quiesce);
static void
device_resume_fn(struct scsi_device *sdev, void *data)
{
scsi_device_resume(sdev);
}
void
scsi_target_resume(struct scsi_target *starget)
{
starget_for_each_device(starget, NULL, device_resume_fn);
}
EXPORT_SYMBOL(scsi_target_resume);
/**
* scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
* @sdev: device to block
*
* Pause SCSI command processing on the specified device. Does not sleep.
*
* Returns zero if successful or a negative error code upon failure.
*
* Notes:
* This routine transitions the device to the SDEV_BLOCK state (which must be
* a legal transition). When the device is in this state, command processing
* is paused until the device leaves the SDEV_BLOCK state. See also
* scsi_internal_device_unblock_nowait().
*/
int scsi_internal_device_block_nowait(struct scsi_device *sdev)
{
struct request_queue *q = sdev->request_queue;
int err = 0;
err = scsi_device_set_state(sdev, SDEV_BLOCK);
if (err) {
err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
if (err)
return err;
}
/*
* The device has transitioned to SDEV_BLOCK. Stop the
* block layer from calling the midlayer with this device's
* request queue.
*/
blk_mq_quiesce_queue_nowait(q);
return 0;
}
EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
/**
* scsi_internal_device_block - try to transition to the SDEV_BLOCK state
* @sdev: device to block
*
* Pause SCSI command processing on the specified device and wait until all
* ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep.
*
* Returns zero if successful or a negative error code upon failure.
*
* Note:
* This routine transitions the device to the SDEV_BLOCK state (which must be
* a legal transition). When the device is in this state, command processing
* is paused until the device leaves the SDEV_BLOCK state. See also
* scsi_internal_device_unblock().
*/
static int scsi_internal_device_block(struct scsi_device *sdev)
{
struct request_queue *q = sdev->request_queue;
int err;
mutex_lock(&sdev->state_mutex);
err = scsi_internal_device_block_nowait(sdev);
if (err == 0)
blk_mq_quiesce_queue(q);
mutex_unlock(&sdev->state_mutex);
return err;
}
void scsi_start_queue(struct scsi_device *sdev)
{
struct request_queue *q = sdev->request_queue;
blk_mq_unquiesce_queue(q);
}
/**
* scsi_internal_device_unblock_nowait - resume a device after a block request
* @sdev: device to resume
* @new_state: state to set the device to after unblocking
*
* Restart the device queue for a previously suspended SCSI device. Does not
* sleep.
*
* Returns zero if successful or a negative error code upon failure.
*
* Notes:
* This routine transitions the device to the SDEV_RUNNING state or to one of
* the offline states (which must be a legal transition) allowing the midlayer
* to goose the queue for this device.
*/
int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
enum scsi_device_state new_state)
{
switch (new_state) {
case SDEV_RUNNING:
case SDEV_TRANSPORT_OFFLINE:
break;
default:
return -EINVAL;
}
/*
* Try to transition the scsi device to SDEV_RUNNING or one of the
* offlined states and goose the device queue if successful.
*/
switch (sdev->sdev_state) {
case SDEV_BLOCK:
case SDEV_TRANSPORT_OFFLINE:
sdev->sdev_state = new_state;
break;
case SDEV_CREATED_BLOCK:
if (new_state == SDEV_TRANSPORT_OFFLINE ||
new_state == SDEV_OFFLINE)
sdev->sdev_state = new_state;
else
sdev->sdev_state = SDEV_CREATED;
break;
case SDEV_CANCEL:
case SDEV_OFFLINE:
break;
default:
return -EINVAL;
}
scsi_start_queue(sdev);
return 0;
}
EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
/**
* scsi_internal_device_unblock - resume a device after a block request
* @sdev: device to resume
* @new_state: state to set the device to after unblocking
*
* Restart the device queue for a previously suspended SCSI device. May sleep.
*
* Returns zero if successful or a negative error code upon failure.
*
* Notes:
* This routine transitions the device to the SDEV_RUNNING state or to one of
* the offline states (which must be a legal transition) allowing the midlayer
* to goose the queue for this device.
*/
static int scsi_internal_device_unblock(struct scsi_device *sdev,
enum scsi_device_state new_state)
{
int ret;
mutex_lock(&sdev->state_mutex);
ret = scsi_internal_device_unblock_nowait(sdev, new_state);
mutex_unlock(&sdev->state_mutex);
return ret;
}
static void
device_block(struct scsi_device *sdev, void *data)
{
int ret;
ret = scsi_internal_device_block(sdev);
WARN_ONCE(ret, "scsi_internal_device_block(%s) failed: ret = %d\n",
dev_name(&sdev->sdev_gendev), ret);
}
static int
target_block(struct device *dev, void *data)
{
if (scsi_is_target_device(dev))
starget_for_each_device(to_scsi_target(dev), NULL,
device_block);
return 0;
}
void
scsi_target_block(struct device *dev)
{
if (scsi_is_target_device(dev))
starget_for_each_device(to_scsi_target(dev), NULL,
device_block);
else
device_for_each_child(dev, NULL, target_block);
}
EXPORT_SYMBOL_GPL(scsi_target_block);
static void
device_unblock(struct scsi_device *sdev, void *data)
{
scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
}
static int
target_unblock(struct device *dev, void *data)
{
if (scsi_is_target_device(dev))
starget_for_each_device(to_scsi_target(dev), data,
device_unblock);
return 0;
}
void
scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
{
if (scsi_is_target_device(dev))
starget_for_each_device(to_scsi_target(dev), &new_state,
device_unblock);
else
device_for_each_child(dev, &new_state, target_unblock);
}
EXPORT_SYMBOL_GPL(scsi_target_unblock);
[SCSI] dc395x: dynamically map scatter-gather for PIO The current dc395x driver uses PIO to transfer up to 4 bytes which do not get transferred by DMA (under unclear circumstances). For this the driver uses page_address() which is broken on highmem. Apart from this the actual calculation of the virtual address is wrong (even without highmem). So, e.g., for reading it reads bytes from the driver to a wrong address and returns wrong data, I guess, for writing it would just output random data to the device. The proper fix, as suggested by many, is to dynamically map data using kmap_atomic(page, KM_BIO_SRC_IRQ) / kunmap_atomic(virt). The reason why it has not been done until now, although I've done some preliminary patches more than a year ago was that nobody interested in fixing this problem was able to reliably reproduce it. Now it changed - with the help from Sebastian Frei (CC'ed) I was able to trigger the PIO path. Thus, I was also able to test and debug it. There are 4 cases when PIO is used in dc395x - data-in / -out with and without scatter-gather. I was able to reproduce and test only data-in with and without SG. So, the data-out path is still untested, but it is also somewhat simpler than the data-in. Fredrik Roubert (also CC'ed) also had PIO triggering on his system, and in his case it was data-out without SG. It would be great if he could test the attached patch on his system, but even if he cannot, I would still request to apply the patch and just wait if anybody cries... Implementation: I put 2 new functions in scsi_lib.c and their declarations in scsi_cmnd.h. I exported them without _GPL, although, I don't feel strongly about that - not many drivers are likely to use them. But there is at least one more - I want to use them in tmscsim.c. Whether these are the right files for the functions and their declarations - not sure either. Actually, they are not scsi-specific, so, might go somewhere around other scattergather magic? They are not platform specific either, and most SG functions are defined under arch/*/... As these issues were discussed previously there were some more routines suggested to manipulate scattergather buffers, I think, some of them were needed around crypto code... So, might be a common place reasonable, like lib/scattergather.c? I am open here. Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
2006-04-02 19:57:43 +00:00
int
scsi_host_block(struct Scsi_Host *shost)
{
struct scsi_device *sdev;
int ret = 0;
/*
* Call scsi_internal_device_block_nowait so we can avoid
* calling synchronize_rcu() for each LUN.
*/
shost_for_each_device(sdev, shost) {
mutex_lock(&sdev->state_mutex);
ret = scsi_internal_device_block_nowait(sdev);
mutex_unlock(&sdev->state_mutex);
if (ret) {
scsi_device_put(sdev);
break;
}
}
/*
* SCSI never enables blk-mq's BLK_MQ_F_BLOCKING flag so
* calling synchronize_rcu() once is enough.
*/
WARN_ON_ONCE(shost->tag_set.flags & BLK_MQ_F_BLOCKING);
if (!ret)
synchronize_rcu();
return ret;
}
EXPORT_SYMBOL_GPL(scsi_host_block);
int
scsi_host_unblock(struct Scsi_Host *shost, int new_state)
{
struct scsi_device *sdev;
int ret = 0;
shost_for_each_device(sdev, shost) {
ret = scsi_internal_device_unblock(sdev, new_state);
if (ret) {
scsi_device_put(sdev);
break;
}
}
return ret;
}
EXPORT_SYMBOL_GPL(scsi_host_unblock);
[SCSI] dc395x: dynamically map scatter-gather for PIO The current dc395x driver uses PIO to transfer up to 4 bytes which do not get transferred by DMA (under unclear circumstances). For this the driver uses page_address() which is broken on highmem. Apart from this the actual calculation of the virtual address is wrong (even without highmem). So, e.g., for reading it reads bytes from the driver to a wrong address and returns wrong data, I guess, for writing it would just output random data to the device. The proper fix, as suggested by many, is to dynamically map data using kmap_atomic(page, KM_BIO_SRC_IRQ) / kunmap_atomic(virt). The reason why it has not been done until now, although I've done some preliminary patches more than a year ago was that nobody interested in fixing this problem was able to reliably reproduce it. Now it changed - with the help from Sebastian Frei (CC'ed) I was able to trigger the PIO path. Thus, I was also able to test and debug it. There are 4 cases when PIO is used in dc395x - data-in / -out with and without scatter-gather. I was able to reproduce and test only data-in with and without SG. So, the data-out path is still untested, but it is also somewhat simpler than the data-in. Fredrik Roubert (also CC'ed) also had PIO triggering on his system, and in his case it was data-out without SG. It would be great if he could test the attached patch on his system, but even if he cannot, I would still request to apply the patch and just wait if anybody cries... Implementation: I put 2 new functions in scsi_lib.c and their declarations in scsi_cmnd.h. I exported them without _GPL, although, I don't feel strongly about that - not many drivers are likely to use them. But there is at least one more - I want to use them in tmscsim.c. Whether these are the right files for the functions and their declarations - not sure either. Actually, they are not scsi-specific, so, might go somewhere around other scattergather magic? They are not platform specific either, and most SG functions are defined under arch/*/... As these issues were discussed previously there were some more routines suggested to manipulate scattergather buffers, I think, some of them were needed around crypto code... So, might be a common place reasonable, like lib/scattergather.c? I am open here. Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
2006-04-02 19:57:43 +00:00
/**
* scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
* @sgl: scatter-gather list
[SCSI] dc395x: dynamically map scatter-gather for PIO The current dc395x driver uses PIO to transfer up to 4 bytes which do not get transferred by DMA (under unclear circumstances). For this the driver uses page_address() which is broken on highmem. Apart from this the actual calculation of the virtual address is wrong (even without highmem). So, e.g., for reading it reads bytes from the driver to a wrong address and returns wrong data, I guess, for writing it would just output random data to the device. The proper fix, as suggested by many, is to dynamically map data using kmap_atomic(page, KM_BIO_SRC_IRQ) / kunmap_atomic(virt). The reason why it has not been done until now, although I've done some preliminary patches more than a year ago was that nobody interested in fixing this problem was able to reliably reproduce it. Now it changed - with the help from Sebastian Frei (CC'ed) I was able to trigger the PIO path. Thus, I was also able to test and debug it. There are 4 cases when PIO is used in dc395x - data-in / -out with and without scatter-gather. I was able to reproduce and test only data-in with and without SG. So, the data-out path is still untested, but it is also somewhat simpler than the data-in. Fredrik Roubert (also CC'ed) also had PIO triggering on his system, and in his case it was data-out without SG. It would be great if he could test the attached patch on his system, but even if he cannot, I would still request to apply the patch and just wait if anybody cries... Implementation: I put 2 new functions in scsi_lib.c and their declarations in scsi_cmnd.h. I exported them without _GPL, although, I don't feel strongly about that - not many drivers are likely to use them. But there is at least one more - I want to use them in tmscsim.c. Whether these are the right files for the functions and their declarations - not sure either. Actually, they are not scsi-specific, so, might go somewhere around other scattergather magic? They are not platform specific either, and most SG functions are defined under arch/*/... As these issues were discussed previously there were some more routines suggested to manipulate scattergather buffers, I think, some of them were needed around crypto code... So, might be a common place reasonable, like lib/scattergather.c? I am open here. Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
2006-04-02 19:57:43 +00:00
* @sg_count: number of segments in sg
* @offset: offset in bytes into sg, on return offset into the mapped area
* @len: bytes to map, on return number of bytes mapped
*
* Returns virtual address of the start of the mapped page
*/
void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
[SCSI] dc395x: dynamically map scatter-gather for PIO The current dc395x driver uses PIO to transfer up to 4 bytes which do not get transferred by DMA (under unclear circumstances). For this the driver uses page_address() which is broken on highmem. Apart from this the actual calculation of the virtual address is wrong (even without highmem). So, e.g., for reading it reads bytes from the driver to a wrong address and returns wrong data, I guess, for writing it would just output random data to the device. The proper fix, as suggested by many, is to dynamically map data using kmap_atomic(page, KM_BIO_SRC_IRQ) / kunmap_atomic(virt). The reason why it has not been done until now, although I've done some preliminary patches more than a year ago was that nobody interested in fixing this problem was able to reliably reproduce it. Now it changed - with the help from Sebastian Frei (CC'ed) I was able to trigger the PIO path. Thus, I was also able to test and debug it. There are 4 cases when PIO is used in dc395x - data-in / -out with and without scatter-gather. I was able to reproduce and test only data-in with and without SG. So, the data-out path is still untested, but it is also somewhat simpler than the data-in. Fredrik Roubert (also CC'ed) also had PIO triggering on his system, and in his case it was data-out without SG. It would be great if he could test the attached patch on his system, but even if he cannot, I would still request to apply the patch and just wait if anybody cries... Implementation: I put 2 new functions in scsi_lib.c and their declarations in scsi_cmnd.h. I exported them without _GPL, although, I don't feel strongly about that - not many drivers are likely to use them. But there is at least one more - I want to use them in tmscsim.c. Whether these are the right files for the functions and their declarations - not sure either. Actually, they are not scsi-specific, so, might go somewhere around other scattergather magic? They are not platform specific either, and most SG functions are defined under arch/*/... As these issues were discussed previously there were some more routines suggested to manipulate scattergather buffers, I think, some of them were needed around crypto code... So, might be a common place reasonable, like lib/scattergather.c? I am open here. Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
2006-04-02 19:57:43 +00:00
size_t *offset, size_t *len)
{
int i;
size_t sg_len = 0, len_complete = 0;
struct scatterlist *sg;
[SCSI] dc395x: dynamically map scatter-gather for PIO The current dc395x driver uses PIO to transfer up to 4 bytes which do not get transferred by DMA (under unclear circumstances). For this the driver uses page_address() which is broken on highmem. Apart from this the actual calculation of the virtual address is wrong (even without highmem). So, e.g., for reading it reads bytes from the driver to a wrong address and returns wrong data, I guess, for writing it would just output random data to the device. The proper fix, as suggested by many, is to dynamically map data using kmap_atomic(page, KM_BIO_SRC_IRQ) / kunmap_atomic(virt). The reason why it has not been done until now, although I've done some preliminary patches more than a year ago was that nobody interested in fixing this problem was able to reliably reproduce it. Now it changed - with the help from Sebastian Frei (CC'ed) I was able to trigger the PIO path. Thus, I was also able to test and debug it. There are 4 cases when PIO is used in dc395x - data-in / -out with and without scatter-gather. I was able to reproduce and test only data-in with and without SG. So, the data-out path is still untested, but it is also somewhat simpler than the data-in. Fredrik Roubert (also CC'ed) also had PIO triggering on his system, and in his case it was data-out without SG. It would be great if he could test the attached patch on his system, but even if he cannot, I would still request to apply the patch and just wait if anybody cries... Implementation: I put 2 new functions in scsi_lib.c and their declarations in scsi_cmnd.h. I exported them without _GPL, although, I don't feel strongly about that - not many drivers are likely to use them. But there is at least one more - I want to use them in tmscsim.c. Whether these are the right files for the functions and their declarations - not sure either. Actually, they are not scsi-specific, so, might go somewhere around other scattergather magic? They are not platform specific either, and most SG functions are defined under arch/*/... As these issues were discussed previously there were some more routines suggested to manipulate scattergather buffers, I think, some of them were needed around crypto code... So, might be a common place reasonable, like lib/scattergather.c? I am open here. Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
2006-04-02 19:57:43 +00:00
struct page *page;
WARN_ON(!irqs_disabled());
for_each_sg(sgl, sg, sg_count, i) {
[SCSI] dc395x: dynamically map scatter-gather for PIO The current dc395x driver uses PIO to transfer up to 4 bytes which do not get transferred by DMA (under unclear circumstances). For this the driver uses page_address() which is broken on highmem. Apart from this the actual calculation of the virtual address is wrong (even without highmem). So, e.g., for reading it reads bytes from the driver to a wrong address and returns wrong data, I guess, for writing it would just output random data to the device. The proper fix, as suggested by many, is to dynamically map data using kmap_atomic(page, KM_BIO_SRC_IRQ) / kunmap_atomic(virt). The reason why it has not been done until now, although I've done some preliminary patches more than a year ago was that nobody interested in fixing this problem was able to reliably reproduce it. Now it changed - with the help from Sebastian Frei (CC'ed) I was able to trigger the PIO path. Thus, I was also able to test and debug it. There are 4 cases when PIO is used in dc395x - data-in / -out with and without scatter-gather. I was able to reproduce and test only data-in with and without SG. So, the data-out path is still untested, but it is also somewhat simpler than the data-in. Fredrik Roubert (also CC'ed) also had PIO triggering on his system, and in his case it was data-out without SG. It would be great if he could test the attached patch on his system, but even if he cannot, I would still request to apply the patch and just wait if anybody cries... Implementation: I put 2 new functions in scsi_lib.c and their declarations in scsi_cmnd.h. I exported them without _GPL, although, I don't feel strongly about that - not many drivers are likely to use them. But there is at least one more - I want to use them in tmscsim.c. Whether these are the right files for the functions and their declarations - not sure either. Actually, they are not scsi-specific, so, might go somewhere around other scattergather magic? They are not platform specific either, and most SG functions are defined under arch/*/... As these issues were discussed previously there were some more routines suggested to manipulate scattergather buffers, I think, some of them were needed around crypto code... So, might be a common place reasonable, like lib/scattergather.c? I am open here. Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
2006-04-02 19:57:43 +00:00
len_complete = sg_len; /* Complete sg-entries */
sg_len += sg->length;
[SCSI] dc395x: dynamically map scatter-gather for PIO The current dc395x driver uses PIO to transfer up to 4 bytes which do not get transferred by DMA (under unclear circumstances). For this the driver uses page_address() which is broken on highmem. Apart from this the actual calculation of the virtual address is wrong (even without highmem). So, e.g., for reading it reads bytes from the driver to a wrong address and returns wrong data, I guess, for writing it would just output random data to the device. The proper fix, as suggested by many, is to dynamically map data using kmap_atomic(page, KM_BIO_SRC_IRQ) / kunmap_atomic(virt). The reason why it has not been done until now, although I've done some preliminary patches more than a year ago was that nobody interested in fixing this problem was able to reliably reproduce it. Now it changed - with the help from Sebastian Frei (CC'ed) I was able to trigger the PIO path. Thus, I was also able to test and debug it. There are 4 cases when PIO is used in dc395x - data-in / -out with and without scatter-gather. I was able to reproduce and test only data-in with and without SG. So, the data-out path is still untested, but it is also somewhat simpler than the data-in. Fredrik Roubert (also CC'ed) also had PIO triggering on his system, and in his case it was data-out without SG. It would be great if he could test the attached patch on his system, but even if he cannot, I would still request to apply the patch and just wait if anybody cries... Implementation: I put 2 new functions in scsi_lib.c and their declarations in scsi_cmnd.h. I exported them without _GPL, although, I don't feel strongly about that - not many drivers are likely to use them. But there is at least one more - I want to use them in tmscsim.c. Whether these are the right files for the functions and their declarations - not sure either. Actually, they are not scsi-specific, so, might go somewhere around other scattergather magic? They are not platform specific either, and most SG functions are defined under arch/*/... As these issues were discussed previously there were some more routines suggested to manipulate scattergather buffers, I think, some of them were needed around crypto code... So, might be a common place reasonable, like lib/scattergather.c? I am open here. Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
2006-04-02 19:57:43 +00:00
if (sg_len > *offset)
break;
}
if (unlikely(i == sg_count)) {
printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
"elements %d\n",
__func__, sg_len, *offset, sg_count);
[SCSI] dc395x: dynamically map scatter-gather for PIO The current dc395x driver uses PIO to transfer up to 4 bytes which do not get transferred by DMA (under unclear circumstances). For this the driver uses page_address() which is broken on highmem. Apart from this the actual calculation of the virtual address is wrong (even without highmem). So, e.g., for reading it reads bytes from the driver to a wrong address and returns wrong data, I guess, for writing it would just output random data to the device. The proper fix, as suggested by many, is to dynamically map data using kmap_atomic(page, KM_BIO_SRC_IRQ) / kunmap_atomic(virt). The reason why it has not been done until now, although I've done some preliminary patches more than a year ago was that nobody interested in fixing this problem was able to reliably reproduce it. Now it changed - with the help from Sebastian Frei (CC'ed) I was able to trigger the PIO path. Thus, I was also able to test and debug it. There are 4 cases when PIO is used in dc395x - data-in / -out with and without scatter-gather. I was able to reproduce and test only data-in with and without SG. So, the data-out path is still untested, but it is also somewhat simpler than the data-in. Fredrik Roubert (also CC'ed) also had PIO triggering on his system, and in his case it was data-out without SG. It would be great if he could test the attached patch on his system, but even if he cannot, I would still request to apply the patch and just wait if anybody cries... Implementation: I put 2 new functions in scsi_lib.c and their declarations in scsi_cmnd.h. I exported them without _GPL, although, I don't feel strongly about that - not many drivers are likely to use them. But there is at least one more - I want to use them in tmscsim.c. Whether these are the right files for the functions and their declarations - not sure either. Actually, they are not scsi-specific, so, might go somewhere around other scattergather magic? They are not platform specific either, and most SG functions are defined under arch/*/... As these issues were discussed previously there were some more routines suggested to manipulate scattergather buffers, I think, some of them were needed around crypto code... So, might be a common place reasonable, like lib/scattergather.c? I am open here. Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
2006-04-02 19:57:43 +00:00
WARN_ON(1);
return NULL;
}
/* Offset starting from the beginning of first page in this sg-entry */
*offset = *offset - len_complete + sg->offset;
[SCSI] dc395x: dynamically map scatter-gather for PIO The current dc395x driver uses PIO to transfer up to 4 bytes which do not get transferred by DMA (under unclear circumstances). For this the driver uses page_address() which is broken on highmem. Apart from this the actual calculation of the virtual address is wrong (even without highmem). So, e.g., for reading it reads bytes from the driver to a wrong address and returns wrong data, I guess, for writing it would just output random data to the device. The proper fix, as suggested by many, is to dynamically map data using kmap_atomic(page, KM_BIO_SRC_IRQ) / kunmap_atomic(virt). The reason why it has not been done until now, although I've done some preliminary patches more than a year ago was that nobody interested in fixing this problem was able to reliably reproduce it. Now it changed - with the help from Sebastian Frei (CC'ed) I was able to trigger the PIO path. Thus, I was also able to test and debug it. There are 4 cases when PIO is used in dc395x - data-in / -out with and without scatter-gather. I was able to reproduce and test only data-in with and without SG. So, the data-out path is still untested, but it is also somewhat simpler than the data-in. Fredrik Roubert (also CC'ed) also had PIO triggering on his system, and in his case it was data-out without SG. It would be great if he could test the attached patch on his system, but even if he cannot, I would still request to apply the patch and just wait if anybody cries... Implementation: I put 2 new functions in scsi_lib.c and their declarations in scsi_cmnd.h. I exported them without _GPL, although, I don't feel strongly about that - not many drivers are likely to use them. But there is at least one more - I want to use them in tmscsim.c. Whether these are the right files for the functions and their declarations - not sure either. Actually, they are not scsi-specific, so, might go somewhere around other scattergather magic? They are not platform specific either, and most SG functions are defined under arch/*/... As these issues were discussed previously there were some more routines suggested to manipulate scattergather buffers, I think, some of them were needed around crypto code... So, might be a common place reasonable, like lib/scattergather.c? I am open here. Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
2006-04-02 19:57:43 +00:00
/* Assumption: contiguous pages can be accessed as "page + i" */
page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
[SCSI] dc395x: dynamically map scatter-gather for PIO The current dc395x driver uses PIO to transfer up to 4 bytes which do not get transferred by DMA (under unclear circumstances). For this the driver uses page_address() which is broken on highmem. Apart from this the actual calculation of the virtual address is wrong (even without highmem). So, e.g., for reading it reads bytes from the driver to a wrong address and returns wrong data, I guess, for writing it would just output random data to the device. The proper fix, as suggested by many, is to dynamically map data using kmap_atomic(page, KM_BIO_SRC_IRQ) / kunmap_atomic(virt). The reason why it has not been done until now, although I've done some preliminary patches more than a year ago was that nobody interested in fixing this problem was able to reliably reproduce it. Now it changed - with the help from Sebastian Frei (CC'ed) I was able to trigger the PIO path. Thus, I was also able to test and debug it. There are 4 cases when PIO is used in dc395x - data-in / -out with and without scatter-gather. I was able to reproduce and test only data-in with and without SG. So, the data-out path is still untested, but it is also somewhat simpler than the data-in. Fredrik Roubert (also CC'ed) also had PIO triggering on his system, and in his case it was data-out without SG. It would be great if he could test the attached patch on his system, but even if he cannot, I would still request to apply the patch and just wait if anybody cries... Implementation: I put 2 new functions in scsi_lib.c and their declarations in scsi_cmnd.h. I exported them without _GPL, although, I don't feel strongly about that - not many drivers are likely to use them. But there is at least one more - I want to use them in tmscsim.c. Whether these are the right files for the functions and their declarations - not sure either. Actually, they are not scsi-specific, so, might go somewhere around other scattergather magic? They are not platform specific either, and most SG functions are defined under arch/*/... As these issues were discussed previously there were some more routines suggested to manipulate scattergather buffers, I think, some of them were needed around crypto code... So, might be a common place reasonable, like lib/scattergather.c? I am open here. Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
2006-04-02 19:57:43 +00:00
*offset &= ~PAGE_MASK;
/* Bytes in this sg-entry from *offset to the end of the page */
sg_len = PAGE_SIZE - *offset;
if (*len > sg_len)
*len = sg_len;
return kmap_atomic(page);
[SCSI] dc395x: dynamically map scatter-gather for PIO The current dc395x driver uses PIO to transfer up to 4 bytes which do not get transferred by DMA (under unclear circumstances). For this the driver uses page_address() which is broken on highmem. Apart from this the actual calculation of the virtual address is wrong (even without highmem). So, e.g., for reading it reads bytes from the driver to a wrong address and returns wrong data, I guess, for writing it would just output random data to the device. The proper fix, as suggested by many, is to dynamically map data using kmap_atomic(page, KM_BIO_SRC_IRQ) / kunmap_atomic(virt). The reason why it has not been done until now, although I've done some preliminary patches more than a year ago was that nobody interested in fixing this problem was able to reliably reproduce it. Now it changed - with the help from Sebastian Frei (CC'ed) I was able to trigger the PIO path. Thus, I was also able to test and debug it. There are 4 cases when PIO is used in dc395x - data-in / -out with and without scatter-gather. I was able to reproduce and test only data-in with and without SG. So, the data-out path is still untested, but it is also somewhat simpler than the data-in. Fredrik Roubert (also CC'ed) also had PIO triggering on his system, and in his case it was data-out without SG. It would be great if he could test the attached patch on his system, but even if he cannot, I would still request to apply the patch and just wait if anybody cries... Implementation: I put 2 new functions in scsi_lib.c and their declarations in scsi_cmnd.h. I exported them without _GPL, although, I don't feel strongly about that - not many drivers are likely to use them. But there is at least one more - I want to use them in tmscsim.c. Whether these are the right files for the functions and their declarations - not sure either. Actually, they are not scsi-specific, so, might go somewhere around other scattergather magic? They are not platform specific either, and most SG functions are defined under arch/*/... As these issues were discussed previously there were some more routines suggested to manipulate scattergather buffers, I think, some of them were needed around crypto code... So, might be a common place reasonable, like lib/scattergather.c? I am open here. Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
2006-04-02 19:57:43 +00:00
}
EXPORT_SYMBOL(scsi_kmap_atomic_sg);
/**
* scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
[SCSI] dc395x: dynamically map scatter-gather for PIO The current dc395x driver uses PIO to transfer up to 4 bytes which do not get transferred by DMA (under unclear circumstances). For this the driver uses page_address() which is broken on highmem. Apart from this the actual calculation of the virtual address is wrong (even without highmem). So, e.g., for reading it reads bytes from the driver to a wrong address and returns wrong data, I guess, for writing it would just output random data to the device. The proper fix, as suggested by many, is to dynamically map data using kmap_atomic(page, KM_BIO_SRC_IRQ) / kunmap_atomic(virt). The reason why it has not been done until now, although I've done some preliminary patches more than a year ago was that nobody interested in fixing this problem was able to reliably reproduce it. Now it changed - with the help from Sebastian Frei (CC'ed) I was able to trigger the PIO path. Thus, I was also able to test and debug it. There are 4 cases when PIO is used in dc395x - data-in / -out with and without scatter-gather. I was able to reproduce and test only data-in with and without SG. So, the data-out path is still untested, but it is also somewhat simpler than the data-in. Fredrik Roubert (also CC'ed) also had PIO triggering on his system, and in his case it was data-out without SG. It would be great if he could test the attached patch on his system, but even if he cannot, I would still request to apply the patch and just wait if anybody cries... Implementation: I put 2 new functions in scsi_lib.c and their declarations in scsi_cmnd.h. I exported them without _GPL, although, I don't feel strongly about that - not many drivers are likely to use them. But there is at least one more - I want to use them in tmscsim.c. Whether these are the right files for the functions and their declarations - not sure either. Actually, they are not scsi-specific, so, might go somewhere around other scattergather magic? They are not platform specific either, and most SG functions are defined under arch/*/... As these issues were discussed previously there were some more routines suggested to manipulate scattergather buffers, I think, some of them were needed around crypto code... So, might be a common place reasonable, like lib/scattergather.c? I am open here. Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
2006-04-02 19:57:43 +00:00
* @virt: virtual address to be unmapped
*/
void scsi_kunmap_atomic_sg(void *virt)
{
kunmap_atomic(virt);
[SCSI] dc395x: dynamically map scatter-gather for PIO The current dc395x driver uses PIO to transfer up to 4 bytes which do not get transferred by DMA (under unclear circumstances). For this the driver uses page_address() which is broken on highmem. Apart from this the actual calculation of the virtual address is wrong (even without highmem). So, e.g., for reading it reads bytes from the driver to a wrong address and returns wrong data, I guess, for writing it would just output random data to the device. The proper fix, as suggested by many, is to dynamically map data using kmap_atomic(page, KM_BIO_SRC_IRQ) / kunmap_atomic(virt). The reason why it has not been done until now, although I've done some preliminary patches more than a year ago was that nobody interested in fixing this problem was able to reliably reproduce it. Now it changed - with the help from Sebastian Frei (CC'ed) I was able to trigger the PIO path. Thus, I was also able to test and debug it. There are 4 cases when PIO is used in dc395x - data-in / -out with and without scatter-gather. I was able to reproduce and test only data-in with and without SG. So, the data-out path is still untested, but it is also somewhat simpler than the data-in. Fredrik Roubert (also CC'ed) also had PIO triggering on his system, and in his case it was data-out without SG. It would be great if he could test the attached patch on his system, but even if he cannot, I would still request to apply the patch and just wait if anybody cries... Implementation: I put 2 new functions in scsi_lib.c and their declarations in scsi_cmnd.h. I exported them without _GPL, although, I don't feel strongly about that - not many drivers are likely to use them. But there is at least one more - I want to use them in tmscsim.c. Whether these are the right files for the functions and their declarations - not sure either. Actually, they are not scsi-specific, so, might go somewhere around other scattergather magic? They are not platform specific either, and most SG functions are defined under arch/*/... As these issues were discussed previously there were some more routines suggested to manipulate scattergather buffers, I think, some of them were needed around crypto code... So, might be a common place reasonable, like lib/scattergather.c? I am open here. Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
2006-04-02 19:57:43 +00:00
}
EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
void sdev_disable_disk_events(struct scsi_device *sdev)
{
atomic_inc(&sdev->disk_events_disable_depth);
}
EXPORT_SYMBOL(sdev_disable_disk_events);
void sdev_enable_disk_events(struct scsi_device *sdev)
{
if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
return;
atomic_dec(&sdev->disk_events_disable_depth);
}
EXPORT_SYMBOL(sdev_enable_disk_events);
scsi: core: Fix VPD LUN ID designator priorities The current implementation of scsi_vpd_lun_id() uses the designator length as an implicit measure of priority. This works most of the time, but not always. For example, some Hitachi storage arrays return this in VPD 0x83: VPD INQUIRY: Device Identification page Designation descriptor number 1, descriptor length: 24 designator_type: T10 vendor identification, code_set: ASCII associated with the Addressed logical unit vendor id: HITACHI vendor specific: 5030C3502025 Designation descriptor number 2, descriptor length: 6 designator_type: vendor specific [0x0], code_set: Binary associated with the Target port vendor specific: 08 03 Designation descriptor number 3, descriptor length: 20 designator_type: NAA, code_set: Binary associated with the Addressed logical unit NAA 6, IEEE Company_id: 0x60e8 Vendor Specific Identifier: 0x7c35000 Vendor Specific Identifier Extension: 0x30c35000002025 [0x60060e8007c350000030c35000002025] The current code would use the first descriptor because it's longer than the NAA descriptor. But this is wrong, the kernel is supposed to prefer NAA descriptors over T10 vendor ID. Designator length should only be used to compare designators of the same type. This patch addresses the issue by separating designator priority and length. Link: https://lore.kernel.org/r/20201029170846.14786-1-mwilck@suse.com Fixes: 9983bed3907c ("scsi: Add scsi_vpd_lun_id()") Reviewed-by: Hannes Reinecke <hare@suse.de> Signed-off-by: Martin Wilck <mwilck@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2020-10-29 17:08:45 +00:00
static unsigned char designator_prio(const unsigned char *d)
{
if (d[1] & 0x30)
/* not associated with LUN */
return 0;
if (d[3] == 0)
/* invalid length */
return 0;
/*
* Order of preference for lun descriptor:
* - SCSI name string
* - NAA IEEE Registered Extended
* - EUI-64 based 16-byte
* - EUI-64 based 12-byte
* - NAA IEEE Registered
* - NAA IEEE Extended
* - EUI-64 based 8-byte
* - SCSI name string (truncated)
* - T10 Vendor ID
* as longer descriptors reduce the likelyhood
* of identification clashes.
*/
switch (d[1] & 0xf) {
case 8:
/* SCSI name string, variable-length UTF-8 */
return 9;
case 3:
switch (d[4] >> 4) {
case 6:
/* NAA registered extended */
return 8;
case 5:
/* NAA registered */
return 5;
case 4:
/* NAA extended */
return 4;
case 3:
/* NAA locally assigned */
return 1;
default:
break;
}
break;
case 2:
switch (d[3]) {
case 16:
/* EUI64-based, 16 byte */
return 7;
case 12:
/* EUI64-based, 12 byte */
return 6;
case 8:
/* EUI64-based, 8 byte */
return 3;
default:
break;
}
break;
case 1:
/* T10 vendor ID */
return 1;
default:
break;
}
return 0;
}
/**
* scsi_vpd_lun_id - return a unique device identification
* @sdev: SCSI device
* @id: buffer for the identification
* @id_len: length of the buffer
*
* Copies a unique device identification into @id based
* on the information in the VPD page 0x83 of the device.
* The string will be formatted as a SCSI name string.
*
* Returns the length of the identification or error on failure.
* If the identifier is longer than the supplied buffer the actual
* identifier length is returned and the buffer is not zero-padded.
*/
int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
{
scsi: core: Fix VPD LUN ID designator priorities The current implementation of scsi_vpd_lun_id() uses the designator length as an implicit measure of priority. This works most of the time, but not always. For example, some Hitachi storage arrays return this in VPD 0x83: VPD INQUIRY: Device Identification page Designation descriptor number 1, descriptor length: 24 designator_type: T10 vendor identification, code_set: ASCII associated with the Addressed logical unit vendor id: HITACHI vendor specific: 5030C3502025 Designation descriptor number 2, descriptor length: 6 designator_type: vendor specific [0x0], code_set: Binary associated with the Target port vendor specific: 08 03 Designation descriptor number 3, descriptor length: 20 designator_type: NAA, code_set: Binary associated with the Addressed logical unit NAA 6, IEEE Company_id: 0x60e8 Vendor Specific Identifier: 0x7c35000 Vendor Specific Identifier Extension: 0x30c35000002025 [0x60060e8007c350000030c35000002025] The current code would use the first descriptor because it's longer than the NAA descriptor. But this is wrong, the kernel is supposed to prefer NAA descriptors over T10 vendor ID. Designator length should only be used to compare designators of the same type. This patch addresses the issue by separating designator priority and length. Link: https://lore.kernel.org/r/20201029170846.14786-1-mwilck@suse.com Fixes: 9983bed3907c ("scsi: Add scsi_vpd_lun_id()") Reviewed-by: Hannes Reinecke <hare@suse.de> Signed-off-by: Martin Wilck <mwilck@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2020-10-29 17:08:45 +00:00
u8 cur_id_prio = 0;
u8 cur_id_size = 0;
const unsigned char *d, *cur_id_str;
const struct scsi_vpd *vpd_pg83;
int id_size = -EINVAL;
rcu_read_lock();
vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
if (!vpd_pg83) {
rcu_read_unlock();
return -ENXIO;
}
/* The id string must be at least 20 bytes + terminating NULL byte */
if (id_len < 21) {
rcu_read_unlock();
return -EINVAL;
}
memset(id, 0, id_len);
for (d = vpd_pg83->data + 4;
d < vpd_pg83->data + vpd_pg83->len;
d += d[3] + 4) {
scsi: core: Fix VPD LUN ID designator priorities The current implementation of scsi_vpd_lun_id() uses the designator length as an implicit measure of priority. This works most of the time, but not always. For example, some Hitachi storage arrays return this in VPD 0x83: VPD INQUIRY: Device Identification page Designation descriptor number 1, descriptor length: 24 designator_type: T10 vendor identification, code_set: ASCII associated with the Addressed logical unit vendor id: HITACHI vendor specific: 5030C3502025 Designation descriptor number 2, descriptor length: 6 designator_type: vendor specific [0x0], code_set: Binary associated with the Target port vendor specific: 08 03 Designation descriptor number 3, descriptor length: 20 designator_type: NAA, code_set: Binary associated with the Addressed logical unit NAA 6, IEEE Company_id: 0x60e8 Vendor Specific Identifier: 0x7c35000 Vendor Specific Identifier Extension: 0x30c35000002025 [0x60060e8007c350000030c35000002025] The current code would use the first descriptor because it's longer than the NAA descriptor. But this is wrong, the kernel is supposed to prefer NAA descriptors over T10 vendor ID. Designator length should only be used to compare designators of the same type. This patch addresses the issue by separating designator priority and length. Link: https://lore.kernel.org/r/20201029170846.14786-1-mwilck@suse.com Fixes: 9983bed3907c ("scsi: Add scsi_vpd_lun_id()") Reviewed-by: Hannes Reinecke <hare@suse.de> Signed-off-by: Martin Wilck <mwilck@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2020-10-29 17:08:45 +00:00
u8 prio = designator_prio(d);
if (prio == 0 || cur_id_prio > prio)
continue;
switch (d[1] & 0xf) {
case 0x1:
/* T10 Vendor ID */
if (cur_id_size > d[3])
break;
scsi: core: Fix VPD LUN ID designator priorities The current implementation of scsi_vpd_lun_id() uses the designator length as an implicit measure of priority. This works most of the time, but not always. For example, some Hitachi storage arrays return this in VPD 0x83: VPD INQUIRY: Device Identification page Designation descriptor number 1, descriptor length: 24 designator_type: T10 vendor identification, code_set: ASCII associated with the Addressed logical unit vendor id: HITACHI vendor specific: 5030C3502025 Designation descriptor number 2, descriptor length: 6 designator_type: vendor specific [0x0], code_set: Binary associated with the Target port vendor specific: 08 03 Designation descriptor number 3, descriptor length: 20 designator_type: NAA, code_set: Binary associated with the Addressed logical unit NAA 6, IEEE Company_id: 0x60e8 Vendor Specific Identifier: 0x7c35000 Vendor Specific Identifier Extension: 0x30c35000002025 [0x60060e8007c350000030c35000002025] The current code would use the first descriptor because it's longer than the NAA descriptor. But this is wrong, the kernel is supposed to prefer NAA descriptors over T10 vendor ID. Designator length should only be used to compare designators of the same type. This patch addresses the issue by separating designator priority and length. Link: https://lore.kernel.org/r/20201029170846.14786-1-mwilck@suse.com Fixes: 9983bed3907c ("scsi: Add scsi_vpd_lun_id()") Reviewed-by: Hannes Reinecke <hare@suse.de> Signed-off-by: Martin Wilck <mwilck@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2020-10-29 17:08:45 +00:00
cur_id_prio = prio;
cur_id_size = d[3];
if (cur_id_size + 4 > id_len)
cur_id_size = id_len - 4;
cur_id_str = d + 4;
id_size = snprintf(id, id_len, "t10.%*pE",
cur_id_size, cur_id_str);
break;
case 0x2:
/* EUI-64 */
scsi: core: Fix VPD LUN ID designator priorities The current implementation of scsi_vpd_lun_id() uses the designator length as an implicit measure of priority. This works most of the time, but not always. For example, some Hitachi storage arrays return this in VPD 0x83: VPD INQUIRY: Device Identification page Designation descriptor number 1, descriptor length: 24 designator_type: T10 vendor identification, code_set: ASCII associated with the Addressed logical unit vendor id: HITACHI vendor specific: 5030C3502025 Designation descriptor number 2, descriptor length: 6 designator_type: vendor specific [0x0], code_set: Binary associated with the Target port vendor specific: 08 03 Designation descriptor number 3, descriptor length: 20 designator_type: NAA, code_set: Binary associated with the Addressed logical unit NAA 6, IEEE Company_id: 0x60e8 Vendor Specific Identifier: 0x7c35000 Vendor Specific Identifier Extension: 0x30c35000002025 [0x60060e8007c350000030c35000002025] The current code would use the first descriptor because it's longer than the NAA descriptor. But this is wrong, the kernel is supposed to prefer NAA descriptors over T10 vendor ID. Designator length should only be used to compare designators of the same type. This patch addresses the issue by separating designator priority and length. Link: https://lore.kernel.org/r/20201029170846.14786-1-mwilck@suse.com Fixes: 9983bed3907c ("scsi: Add scsi_vpd_lun_id()") Reviewed-by: Hannes Reinecke <hare@suse.de> Signed-off-by: Martin Wilck <mwilck@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2020-10-29 17:08:45 +00:00
cur_id_prio = prio;
cur_id_size = d[3];
cur_id_str = d + 4;
switch (cur_id_size) {
case 8:
id_size = snprintf(id, id_len,
"eui.%8phN",
cur_id_str);
break;
case 12:
id_size = snprintf(id, id_len,
"eui.%12phN",
cur_id_str);
break;
case 16:
id_size = snprintf(id, id_len,
"eui.%16phN",
cur_id_str);
break;
default:
break;
}
break;
case 0x3:
/* NAA */
scsi: core: Fix VPD LUN ID designator priorities The current implementation of scsi_vpd_lun_id() uses the designator length as an implicit measure of priority. This works most of the time, but not always. For example, some Hitachi storage arrays return this in VPD 0x83: VPD INQUIRY: Device Identification page Designation descriptor number 1, descriptor length: 24 designator_type: T10 vendor identification, code_set: ASCII associated with the Addressed logical unit vendor id: HITACHI vendor specific: 5030C3502025 Designation descriptor number 2, descriptor length: 6 designator_type: vendor specific [0x0], code_set: Binary associated with the Target port vendor specific: 08 03 Designation descriptor number 3, descriptor length: 20 designator_type: NAA, code_set: Binary associated with the Addressed logical unit NAA 6, IEEE Company_id: 0x60e8 Vendor Specific Identifier: 0x7c35000 Vendor Specific Identifier Extension: 0x30c35000002025 [0x60060e8007c350000030c35000002025] The current code would use the first descriptor because it's longer than the NAA descriptor. But this is wrong, the kernel is supposed to prefer NAA descriptors over T10 vendor ID. Designator length should only be used to compare designators of the same type. This patch addresses the issue by separating designator priority and length. Link: https://lore.kernel.org/r/20201029170846.14786-1-mwilck@suse.com Fixes: 9983bed3907c ("scsi: Add scsi_vpd_lun_id()") Reviewed-by: Hannes Reinecke <hare@suse.de> Signed-off-by: Martin Wilck <mwilck@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2020-10-29 17:08:45 +00:00
cur_id_prio = prio;
cur_id_size = d[3];
cur_id_str = d + 4;
switch (cur_id_size) {
case 8:
id_size = snprintf(id, id_len,
"naa.%8phN",
cur_id_str);
break;
case 16:
id_size = snprintf(id, id_len,
"naa.%16phN",
cur_id_str);
break;
default:
break;
}
break;
case 0x8:
/* SCSI name string */
scsi: core: Fix VPD LUN ID designator priorities The current implementation of scsi_vpd_lun_id() uses the designator length as an implicit measure of priority. This works most of the time, but not always. For example, some Hitachi storage arrays return this in VPD 0x83: VPD INQUIRY: Device Identification page Designation descriptor number 1, descriptor length: 24 designator_type: T10 vendor identification, code_set: ASCII associated with the Addressed logical unit vendor id: HITACHI vendor specific: 5030C3502025 Designation descriptor number 2, descriptor length: 6 designator_type: vendor specific [0x0], code_set: Binary associated with the Target port vendor specific: 08 03 Designation descriptor number 3, descriptor length: 20 designator_type: NAA, code_set: Binary associated with the Addressed logical unit NAA 6, IEEE Company_id: 0x60e8 Vendor Specific Identifier: 0x7c35000 Vendor Specific Identifier Extension: 0x30c35000002025 [0x60060e8007c350000030c35000002025] The current code would use the first descriptor because it's longer than the NAA descriptor. But this is wrong, the kernel is supposed to prefer NAA descriptors over T10 vendor ID. Designator length should only be used to compare designators of the same type. This patch addresses the issue by separating designator priority and length. Link: https://lore.kernel.org/r/20201029170846.14786-1-mwilck@suse.com Fixes: 9983bed3907c ("scsi: Add scsi_vpd_lun_id()") Reviewed-by: Hannes Reinecke <hare@suse.de> Signed-off-by: Martin Wilck <mwilck@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2020-10-29 17:08:45 +00:00
if (cur_id_size > d[3])
break;
/* Prefer others for truncated descriptor */
scsi: core: Fix VPD LUN ID designator priorities The current implementation of scsi_vpd_lun_id() uses the designator length as an implicit measure of priority. This works most of the time, but not always. For example, some Hitachi storage arrays return this in VPD 0x83: VPD INQUIRY: Device Identification page Designation descriptor number 1, descriptor length: 24 designator_type: T10 vendor identification, code_set: ASCII associated with the Addressed logical unit vendor id: HITACHI vendor specific: 5030C3502025 Designation descriptor number 2, descriptor length: 6 designator_type: vendor specific [0x0], code_set: Binary associated with the Target port vendor specific: 08 03 Designation descriptor number 3, descriptor length: 20 designator_type: NAA, code_set: Binary associated with the Addressed logical unit NAA 6, IEEE Company_id: 0x60e8 Vendor Specific Identifier: 0x7c35000 Vendor Specific Identifier Extension: 0x30c35000002025 [0x60060e8007c350000030c35000002025] The current code would use the first descriptor because it's longer than the NAA descriptor. But this is wrong, the kernel is supposed to prefer NAA descriptors over T10 vendor ID. Designator length should only be used to compare designators of the same type. This patch addresses the issue by separating designator priority and length. Link: https://lore.kernel.org/r/20201029170846.14786-1-mwilck@suse.com Fixes: 9983bed3907c ("scsi: Add scsi_vpd_lun_id()") Reviewed-by: Hannes Reinecke <hare@suse.de> Signed-off-by: Martin Wilck <mwilck@suse.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
2020-10-29 17:08:45 +00:00
if (d[3] > id_len) {
prio = 2;
if (cur_id_prio > prio)
break;
}
cur_id_prio = prio;
cur_id_size = id_size = d[3];
cur_id_str = d + 4;
if (cur_id_size >= id_len)
cur_id_size = id_len - 1;
memcpy(id, cur_id_str, cur_id_size);
break;
default:
break;
}
}
rcu_read_unlock();
return id_size;
}
EXPORT_SYMBOL(scsi_vpd_lun_id);
/*
* scsi_vpd_tpg_id - return a target port group identifier
* @sdev: SCSI device
*
* Returns the Target Port Group identifier from the information
* froom VPD page 0x83 of the device.
*
* Returns the identifier or error on failure.
*/
int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
{
const unsigned char *d;
const struct scsi_vpd *vpd_pg83;
int group_id = -EAGAIN, rel_port = -1;
rcu_read_lock();
vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
if (!vpd_pg83) {
rcu_read_unlock();
return -ENXIO;
}
d = vpd_pg83->data + 4;
while (d < vpd_pg83->data + vpd_pg83->len) {
switch (d[1] & 0xf) {
case 0x4:
/* Relative target port */
rel_port = get_unaligned_be16(&d[6]);
break;
case 0x5:
/* Target port group */
group_id = get_unaligned_be16(&d[6]);
break;
default:
break;
}
d += d[3] + 4;
}
rcu_read_unlock();
if (group_id >= 0 && rel_id && rel_port != -1)
*rel_id = rel_port;
return group_id;
}
EXPORT_SYMBOL(scsi_vpd_tpg_id);
/**
* scsi_build_sense - build sense data for a command
* @scmd: scsi command for which the sense should be formatted
* @desc: Sense format (non-zero == descriptor format,
* 0 == fixed format)
* @key: Sense key
* @asc: Additional sense code
* @ascq: Additional sense code qualifier
*
**/
void scsi_build_sense(struct scsi_cmnd *scmd, int desc, u8 key, u8 asc, u8 ascq)
{
scsi_build_sense_buffer(desc, scmd->sense_buffer, key, asc, ascq);
scmd->result = SAM_STAT_CHECK_CONDITION;
}
EXPORT_SYMBOL_GPL(scsi_build_sense);