ATA and SATA drives have had built-in retries for media errors
for as long as they've been commonplace in computers (early 1990s).
When libata stumbles across a bad sector, it can waste minutes
sitting there doing retry after retry before finally giving up
and letting the higher layers deal with it.
This patch removes retries for media errors only.
Signed-off-by: Mark Lord <mlord@pobox.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
Commit d902747("[libata] Add ATA transport class") introduced
ATA_EFLAG_OLD_ER to mark entries in the error ring as cleared.
But ata_count_probe_trials_cb() didn't check this flag and it still
counts the old error history. So wrong probe trials count is returned
and it causes problem, for example, SATA link speed is slowed down from
3.0Gbps to 1.5Gbps.
Fix it by checking ATA_EFLAG_OLD_ER in ata_count_probe_trials_cb().
Cc: stable <stable@vger.kernel.org> # 2.6.37+
Signed-off-by: Lin Ming <ming.m.lin@intel.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
Link resets leave ata affiliations intact, so arrange for libsas to make
an effort to avoid dropping the device due to a slow-to-recover link.
Towards this end carry out reset in the host workqueue so that it can
check for ata devices and kick the reset request to libata. Hard
resets, in contrast, bypass libata since they are meant for associating
an ata device with another initiator in the domain (tears down
affiliations).
Need to add a new transport_sas_phy_reset() since the current
sas_phy_reset() is a utility function to libsas lldds. They are not
prepared for it to loop back into eh.
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: James Bottomley <JBottomley@Parallels.com>
Reenable sending SRST to devices connected behind a Sil3726 PMP.
This allow staggered spinups and handles drives that spins up slowly.
While the drives spin up, the PMP will not accept SRST.
Most controller reissues the reset until the drive is ready, while
some [Sil3124] returns an error.
In ata_eh_error, wait 10s before reset the ATA port and try again.
Signed-off-by: Gwendal Grignou <gwendal@google.com>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
They were getting this implicitly by an include of module.h
from device.h -- but we are going to clean that up and break
that include chain, so include export.h explicitly now.
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
libata EH intentionally left a port frozen if it failed
ata_eh_reset(). The intention was avoiding continuous loop of resets
when the controller or attached device is flaky and reporting spurious
hotplug events. Once port enters this state, it can be recovered with
manual rescan, which seemed reasonable.
However, outside of my convoluted test setup, there have been very few
reports justifying this choice while there have been more cases where
the automatic freezing of the port after hotplug attempt of a faulty
device caused confusion and led to unnecessary resets.
This patch changes the behavior so that the port is thawed after reset
failure. This change doesn't necessarily solve but makes it easier
and more intuitive to work around hotplug related problems
(ie. re-pluggin or power cycling the device) as reported in the
followings.
https://bugzilla.kernel.org/show_bug.cgi?id=34712http://thread.gmane.org/gmane.linux.kernel/1123265/focus=49548
Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Reartes Guillermo <rtguille@gmail.com>
Reported-by: Bruce Stenning <b.stenning@indigovision.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
To work around controllers which can't properly plug events while
reset, ata_eh_reset() clears error states and ATA_PFLAG_EH_PENDING
after reset but before RESET is marked done. As reset is the final
recovery action and full verification of devices including onlineness
and classfication match is done afterwards, this shouldn't lead to
lost devices or missed hotplug events.
Unfortunately, it forgot to thaw the port when clearing EH_PENDING, so
if the condition happens after resetting an empty port, the port could
be left frozen and EH will end without thawing it, making the port
unresponsive to further hotplug events.
Thaw if the port is frozen after clearing EH_PENDING. This problem is
reported by Bruce Stenning in the following thread.
http://thread.gmane.org/gmane.linux.kernel/1123265
stable: I think we should weather this patch a bit longer in -rcX
before sending it to -stable. Please wait at least a month
after this patch makes upstream. Thanks.
-v2: Fixed spelling in the comment per Dave Howorth.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Bruce Stenning <b.stenning@indigovision.com>
Cc: stable@kernel.org
Cc: Dave Howorth <dhoworth@mrc-lmb.cam.ac.uk>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
Give users the option of completely powering off unoccupied
SATA ports using the existing min_power link_power_management_policy
option. When the use selects this option on an empty port, we
will power the port off by setting DET to off. For occupied ports,
behavior is unchanged.
Signed-off-by: Kristen Carlson Accardi <kristen@linux.intel.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
NVIDIA mcp65 familiy of controllers cause command timeouts when DIPM
is used. Implement ATA_FLAG_NO_DIPM and apply it.
This problem was reported by Stefan Bader in the following thread.
http://thread.gmane.org/gmane.linux.ide/48841
stable: applicable to 2.6.37 and 38.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Stefan Bader <stefan.bader@canonical.com>
Cc: stable@kernel.org
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>
Right at the moment, the libata error handler is incredibly
monolithic. This makes it impossible to use from composite drivers
like libsas and ipr which have to handle error themselves in the first
instance.
The essence of the change is to split the monolithic error handler
into two components: one which handles a queue of ata commands for
processing and the other which handles the back end of readying a
port. This allows the upper error handler fine grained control in
calling libsas functions (and making sure they only get called for ATA
commands whose lower errors have been fixed up).
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
The SCSI host eh_cmd_q should be protected by the host lock (not the
port lock). This probably doesn't matter that much at the moment,
since we try to serialise the add and eh pieces, but it might matter
in future for more convenient error handling. Plus this switches
libata to the standard eh pattern where you lock, remove from the cmd
queue to a local list and unlock and then operate on the local list.
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
ata_eh_analyze_serror() suppresses hotplug notifications if LPM is
being used because LPM generates spurious hotplug events. It compared
whether link->lpm_policy was different from ATA_LPM_MAX_POWER to
determine whether LPM is enabled; however, this is incorrect as for
drivers which don't implement LPM, lpm_policy is always
ATA_LPM_UNKNOWN. This disabled hotplug detection for all drivers
which don't implement LPM.
Fix it by comparing whether lpm_policy is greater than
ATA_LPM_MAX_POWER.
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: stable@kernel.org
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
Low level drivers may behave differently depending on the current
link->lpm_policy. During ata_eh_set_lpm(), DIPM enable commands are
issued after the successful completion of ap->ops->set_lpm(), which
means that the controller is already in the target state. This causes
DIPM enable commands to be processed with mismatching controller power
state and link->lpm_policy value.
In ahci, link->lpm_policy is used to ignore certain PHY events if LPM
is enabled; however, as DIPM commands are issued with stale
link->lpm_policy, they sometimes end up triggering these conditions
and get aborted leading to LPM configuration failure.
Fix it by updating link->lpm_policy before issuing DIPM enable
commands.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Kyle McMartin <kyle@mcmartin.ca>
Cc: stable@kernel.org
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
In libata, the non-EH code paths should always take and release
ap->lock explicitly when accessing hardware or shared data structures.
However, once EH is active, it's assumed that the port is owned by EH
and EH methods don't explicitly take ap->lock unless race from irq
handler or other code paths are expected. However, libata EH didn't
guarantee exclusion among EHs for ports of the same host. IOW,
multiple EHs may execute in parallel on multiple ports of the same
controller.
In many cases, especially in SATA, the ports are completely
independent of each other and this doesn't cause problems; however,
there are cases where different ports share the same resource, which
lead to obscure timing related bugs such as the one fixed by commit
213373cf (ata_piix: fix locking around SIDPR access).
This patch implements exclusion among EHs of the same host. When EH
begins, it acquires per-host EH ownership by calling ata_eh_acquire().
When EH finishes, the ownership is released by calling
ata_eh_release(). EH ownership is also released whenever the EH
thread goes to sleep from ata_msleep() or explicitly and reacquired
after waking up.
This ensures that while EH is actively accessing the hardware, it has
exclusive access to it while allowing EHs to interleave and progress
in parallel as they hit waiting stages, which dominate the time spent
in EH. This achieves cross-port EH exclusion without pervasive and
fragile changes while still allowing parallel EH for the most part.
This was first reported by yuanding02@gmail.com more than three years
ago in the following bugzilla. :-)
https://bugzilla.kernel.org/show_bug.cgi?id=8223
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Reported-by: yuanding02@gmail.com
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
Add optional @ap argument to ata_wait_register() and replace msleep()
calls with ata_msleep() which take optional @ap in addition to the
duration. These will be used to implement EH exclusion.
This patch doesn't cause any behavior difference.
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
Port multipliers can do DIPM on fan-out links fine. Implement support
for it. Tested w/ SIMG 57xx and marvell PMPs. Both the host and
fan-out links enter power save modes nicely.
SIMG 37xx and 47xx report link offline on SStatus causing EH to detach
the devices. Blacklisted.
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
The current LPM implementation has the following issues.
* Operation order isn't well thought-out. e.g. HIPM should be
configured after IPM in SControl is properly configured. Not the
other way around.
* Suspend/resume paths call ata_lpm_enable/disable() which must only
be called from EH context directly. Also, ata_lpm_enable/disable()
were called whether LPM was in use or not.
* Implementation is per-port when it should be per-link. As a result,
it can't be used for controllers with slave links or PMP.
* LPM state isn't managed consistently. After a link reset for
whatever reason including suspend/resume the actual LPM state would
be reset leaving ap->lpm_policy inconsistent.
* Generic/driver-specific logic boundary isn't clear. Currently,
libahci has to mangle stuff which libata EH proper should be
handling. This makes the implementation unnecessarily complex and
fragile.
* Tied to ALPM. Doesn't consider DIPM only cases and doesn't check
whether the device allows HIPM.
* Error handling isn't implemented.
Given the extent of mismatch with the rest of libata, I don't think
trying to fix it piecewise makes much sense. This patch reimplements
LPM support.
* The new implementation is per-link. The target policy is still
port-wide (ap->target_lpm_policy) but all the mechanisms and states
are per-link and integrate well with the rest of link abstraction
and can work with slave and PMP links.
* Core EH has proper control of LPM state. LPM state is reconfigured
when and only when reconfiguration is necessary. It makes sure that
LPM state is reset when probing for new device on the link.
Controller agnostic logic is now implemented in libata EH proper and
driver implementation only has to deal with controller specifics.
* Proper error handling. LPM config failure is attributed to the
device on the link and LPM is disabled for the link if it fails
repeatedly.
* ops->enable/disable_pm() are replaced with single ops->set_lpm()
which takes @policy and @hints. This simplifies driver specific
implementation.
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
Link power management related symbols are in confusing state w/ mixed
usages of lpm, ipm and pm. This patch cleans up lpm related symbols
and sysfs show/store functions as follows.
* lpm states - NOT_AVAILABLE, MIN_POWER, MAX_PERFORMANCE and
MEDIUM_POWER are renamed to ATA_LPM_UNKNOWN and
ATA_LPM_{MIN|MAX|MED}_POWER.
* Pre/postfixes are unified to lpm.
* sysfs show/store functions for link_power_management_policy were
curiously named get/put and unnecessarily complex. Renamed to
show/store and simplified.
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
This is a scheleton for libata transport class.
All information is read only, exporting information from libata:
- ata_port class: one per ATA port
- ata_link class: one per ATA port or 15 for SATA Port Multiplier
- ata_device class: up to 2 for PATA link, usually one for SATA.
Signed-off-by: Gwendal Grignou <gwendal@google.com>
Reviewed-by: Grant Grundler <grundler@google.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
For some mysterious reason, certain hardware reacts badly to usual EH
actions while the system is going for suspend. As the devices won't
be needed until the system is resumed, ask EH to skip usual autopsy
and recovery and proceed directly to suspend.
Signed-off-by: Tejun Heo <tj@kernel.org>
Tested-by: Stephan Diestelhorst <stephan.diestelhorst@amd.com>
Cc: stable@kernel.org
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/wq: (55 commits)
workqueue: mark init_workqueues() as early_initcall()
workqueue: explain for_each_*cwq_cpu() iterators
fscache: fix build on !CONFIG_SYSCTL
slow-work: kill it
gfs2: use workqueue instead of slow-work
drm: use workqueue instead of slow-work
cifs: use workqueue instead of slow-work
fscache: drop references to slow-work
fscache: convert operation to use workqueue instead of slow-work
fscache: convert object to use workqueue instead of slow-work
workqueue: fix how cpu number is stored in work->data
workqueue: fix mayday_mask handling on UP
workqueue: fix build problem on !CONFIG_SMP
workqueue: fix locking in retry path of maybe_create_worker()
async: use workqueue for worker pool
workqueue: remove WQ_SINGLE_CPU and use WQ_UNBOUND instead
workqueue: implement unbound workqueue
workqueue: prepare for WQ_UNBOUND implementation
libata: take advantage of cmwq and remove concurrency limitations
workqueue: fix worker management invocation without pending works
...
Fixed up conflicts in fs/cifs/* as per Tejun. Other trivial conflicts in
include/linux/workqueue.h, kernel/trace/Kconfig and kernel/workqueue.c
libata has two concurrency related limitations.
a. ata_wq which is used for polling PIO has single thread per CPU. If
there are multiple devices doing polling PIO on the same CPU, they
can't be executed simultaneously.
b. ata_aux_wq which is used for SCSI probing has single thread. In
cases where SCSI probing is stalled for extended period of time
which is possible for ATAPI devices, this will stall all probing.
#a is solved by increasing maximum concurrency of ata_wq. Please note
that polling PIO might be used under allocation path and thus needs to
be served by a separate wq with a rescuer.
#b is solved by using the default wq instead and achieving exclusion
via per-port mutex.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Jeff Garzik <jgarzik@pobox.com>
Some of error handling logic in ata_sff_error_handler() and all of
ata_sff_post_internal_cmd() are for BMDMA. Create
ata_bmdma_error_handler() and ata_bmdma_post_internal_cmd() and move
BMDMA part into those.
While at it, change DMA protocol check to ata_is_dma(), fix
post_internal_cmd to call ap->ops->bmdma_stop instead of directly
calling ata_bmdma_stop() and open code hardreset selection so that
ata_std_error_handler() doesn't have to know about sff hardreset.
As these two functions are BMDMA specific, there's no reason to check
for bmdma_addr before calling bmdma methods if the protocol of the
failed command is DMA. sata_mv and pata_mpc52xx now don't need to set
.post_internal_cmd to ATA_OP_NULL and pata_icside and sata_qstor don't
need to set it to their bmdma_stop routines.
ata_sff_post_internal_cmd() becomes noop and is removed.
This fixes p3 described in clean-up-BMDMA-initialization patch.
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
port_task is tightly bound to the standard SFF PIO HSM implementation.
Using it for any other purpose would be error-prone and there's no
such user and if some drivers need such feature, it would be much
better off using its own. Move it inside CONFIG_ATA_SFF and rename it
to sff_pio_task.
The only function which is exposed to the core layer is
ata_sff_flush_pio_task() which is renamed from ata_port_flush_task()
and now also takes care of resetting hsm_task_state to HSM_ST_IDLE,
which is possible as it's now specific to PIO HSM.
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
blk_abort_request() expectes queue lock to be held by the caller.
Grab it before calling the function.
Lack of this synchronization led to infinite loop on corrupt
q->timeout_list.
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: stable@kernel.org
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
libata currently doesn't retry if a command fails with AC_ERR_INVALID
assuming that retrying won't get it any further even if retried.
However, a failure may be classified as invalid through hardware
glitch (incorrect reading of the error register or firmware bug) and
there isn't whole lot to gain by not retrying as actually invalid
commands will be failed immediately. Also, commands serving FS IOs
are extremely unlikely to be invalid. Retry FS IOs even if it's
marked invalid.
Transient and incorrect invalid failure was seen while debugging
firmware related issue on Samsung n130 on bko#14314.
http://bugzilla.kernel.org/show_bug.cgi?id=14314
Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Johannes Stezenbach <js@sig21.net>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
If ATA device failed FLUSH, it means that the device failed to write
out some amount of data and the error needs to be reported to upper
layers. As retries can't recover the lost data, FLUSH failures need to
be reported immediately in general.
However, if FLUSH fails due to transmission errors, the FLUSH needs to
be retried; otherwise, filesystems may switch to RO mode and/or raid
array may drop a drive for a random transmission glitch.
This condition can be rather easily reproduced on certain ahci
controllers which go through a PHY event after powersave mode switch +
ext4 combination. Powersave mode switch is often closely followed by
flush from the filesystem failing the FLUSH with ATA bus error which
makes the filesystem code believe that data is lost and drop to RO
mode. This was reported in the following bugzilla bug.
http://bugzilla.kernel.org/show_bug.cgi?id=14543
This patch makes libata EH retry FLUSH if it wasn't failed by the
device.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Andrey Vihrov <andrey.vihrov@gmail.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
Commit 842faa6c1a fixed error handling
during attach by not committing detected device class to dev->class
while attaching a new device. However, this change missed the PMP
class check in the configuration loop causing a new PMP device to go
through ata_dev_configure() as if it were an ATA or ATAPI device.
As PMP device doesn't have a regular IDENTIFY data, this makes
ata_dev_configure() tries to configure a PMP device using an invalid
data. For the most part, it wasn't too harmful and went unnoticed but
this ends up clearing dev->flags which may have ATA_DFLAG_AN set by
sata_pmp_attach(). This means that SATA_PMP_FEAT_NOTIFY ends up being
disabled on PMPs and on PMPs which honor the flag breaks hotplug
support.
This problem was discovered and reported by Ethan Hsiao.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Ethan Hsiao <ethanhsiao@jmicron.com>
Cc: stable@kernel.org
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
While trying to work around spurious detection retries for
non-existent devices on slave links, commit
816ab89782 incorrectly added link
offline check logic before ata_eh_thaw() was called. This means that
if an occupied link goes down briefly at the time that offline check
was performed, device class will be cleared to ATA_DEV_NONE and libata
wouldn't retry thus failing detection of the device.
The offline check should be done after the port is thawed together
with online check so that such link glitches can be detected by the
interrupt handler and handled properly.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Tim Blechmann <tim@klingt.org>
Cc: stable@kernel.org
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
This patch improve libata's output for error/notification messages
to allow easier comprehension and debugging:
When ATAPI commands issued through the SCSI layer fail, use SCSI
functions to print the CDB in human-readable form instead of just
dumping out the CDB in hex.
Print out the name of the failed command (as defined by the ATA
specification) in error handling output along with the raw register
contents.
When reporting status of ACPI taskfile commands executed on resume,
also output the names of the commands being executed (or not) in
readable form.
Since the extra data for printing command names increases kernel
size slightly, a config option has been added to allow disabling
command name output (as well as some of the error register parsing)
for those highly sensitive to kernel text size.
Signed-off-by: Robert Hancock <hancockrwd@gmail.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
Resets are done with port frozen but some controllers still issue
interrupts during reset and they may end up recording error conditions
in ehi leading to unnecessary EH retrials.
This patch makes ata_eh_reset() clear ehi on reset completion. As
reset is the most severe recovery action, there's nothing to lose by
clearing ehi on its completion.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Zdenek Kaspar <zkaspar82@gmail.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
Call the ->freeze() hook before aborting qc's, because some hardware
requires special handling prior to accessing the taskfile registers
(for diagnosis/analysis/reset). Most notably, hardware may wish to
disable the DMA engine or interrupts in the ->freeze() hook.
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
drivers/ata/libata-eh.c +2403 ata_eh_reset(80) warning: variable derefenced before check 'slave'
Please note that this is _not_ a real bug at the moment since ata_eh_context
structure is embedded into ata_list structure and the code alwas checks for
'slave' before accessing 'sehc'.
Anyway lets add missing check and always have a valid 'sehc' pointer (which
makes code easier to understand and prevents introducing some possible bugs
in the future).
Reported-by: Dan Carpenter <error27@gmail.com>
Cc: corbet@lwn.net
Cc: eteo@redhat.com
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
ata_eh_reset() was missing error return handling after follow-up SRST
allowing EH to continue the normal probing path after reset failure.
This was discovered while testing new WD 2TB drives which take longer
than 10 secs to spin up and cause the first follow-up SRST to time
out.
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
Error timestamps are in jiffies which doesn't run while suspended and
PHY events during resume isn't too uncommon. When the two are
combined, it can lead to unnecessary speed downs if the machine is
suspended and resumed repeatedly. Clear error history on resume.
This was reported and verified in bnc#486803 by Vladimir Botka.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Vladimir Botka <vbotka@novell.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
New device attach path in ata_eh_revalidate_and_attach() is divided
into two separate loops because ATA requires IDENTIFY to be issued to
slave first while the user expects to see device probe messages from
the master device. new_mask is used to track which devices are the
new ones between the first loop and the second.
This usually works well but if an error occurs during configuration
stage, ata_dev_revalidate_and_attach() returns with error code and
forgets new_mask. On the retry run, dev->class is set and new_mask
for the device is clear, so the device just gets revalidated and thus
ends up skipping post-configuration procedure including scheduling of
SCSI_HOTPLUG for the device. When this occurs, ATA part of probing
works fine but SCSI probing usually doesn't happen and makes the
device unreachable.
The behavior has been around for a very long time but it has been
uncovered with the recent addition of 1_5_GBPS horkage which uses
-EAGAIN return value from ata_dev_configure() to restart the probing
sequence after forcing cable speed.
This can be fixed by making sure dev->class is permanently set only
after all configurations are successfully complete. Fix it.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Tim Connors <tconnors+linuxkml@astro.swin.edu.au>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
On a timeout call a device specific handler early in the recovery so that
we can complete and process successful commands which timed out due to IRQ
loss or the like rather more elegantly.
[Revised to exclude the timeout handling on a few devices that inherit from
SFF but are not SFF enough to use the default timeout handler]
Signed-off-by: Alan Cox <alan@redhat.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
When SCR access is available and the link is offline, softreset is
skipped as it only wastes time and some controllers don't respond very
well. However, the skip path forgot to thaw the port, which not only
blocks further event notification from the port but also causes
repeated EH invocations on the same event on drivers which rely on
->thaw() to clear events if the IRQ is shared with another device or
port.
This problem has always been there but is uncovered by recent sata_nv
nf2/3 change which dropped hardreset support while maintaining SCR
access. nf2/3 doesn't clear hotplug event mask from the interrupt
handler but relies on ->thaw() to clear them. When the hardreset was
there, the reset action was never skipped and the port was always
thawed but, with the hardreset gone, ->prereset() determines that
there's no need for softreset and both ->softreset() and ->thaw() are
skipped. This leads to stuck hotplug event in the IRQ status register
triggering hotplug event whenever IRQ is delieverd on the same IRQ.
As the controller shares the same IRQ for both ports, this happens on
every IO if one port is occpupied and the other isn't.
This patch fixes the problem by making sure that the port is thawed on
reset-skip path.
bko#11615 reports this problem.
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Robert Hancock <hancockrwd@gmail.com>
Reported-by: Dan Andresan <danyer@gmail.com>
Reported-by: Arne Woerner <arne_woerner@yahoo.com>
Reported-by: Stefan Lippers-Hollmann <s.L-H@gmx.de>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
sense_buffer is used as DMA target and shouldn't be allocated on
stack. Use ap->sector_buf instead. This problem is spotted by Chuck
Ebbert.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Chuck Ebbert <cebbert@redhat.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
Let -EAGAIN from EH device handling routines trigger EH retry without
consuming its tries count. This will be used to implement link SPD
horkage which requires hardreset to adjust SPD without affecting other
EH decisions. As it bypasses the forward progress guarantee provided
by the tries count, the requester is responsible for ensuring forward
progress.
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
When link is flaky at high speed, it isn't uncommon for a device to
repeatedly fail probing sequence early after successfully negotiating
high link speed. This often leads to consecutive hotplug events
without successful probing.
This patch improves libata EH such that it remembers probing trials
and if there have been more than two unsuccessful trials in the past
60 seconds, slows down link speed to 1.5Gbps.
As link speed negotiation is the duty of the PHY layer proper, the
goal of this fallback mechanism is to provide the last resort when
everything else fails, which unfortunately happens not too
infrequently, so no fancy 6->3->1.5 speeding down or highest
successful transmission speed seen kind of logics (yet).
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
Add @spd_limit to sata_down_spd_limit() so that the caller can specify
the SPD limit it wants. This parameter doesn't get in the way even
when it's too low. The closest possible limit is applied.
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
dev->ering used to be cleared together with the rest of ata_device in
ata_dev_init() which is called whenever a probing event occurs.
dev->ering is about to be used to track probing failures so it needs
to remain persistent over multiple porbing events. This patch
achieves this by doing the following.
* Instead of CLEAR_OFFSET, define CLEAR_BEGIN and CLEAR_END and only
clear between BEGIN and END. ering is moved after END. The split
of persistent area is to allow hotter items remain at the head.
* ering is explicitly cleared on ata_dev_disable() and when device
attach succeeds. So, ering is persistent throug a device's life
time (unless explicitly cleared of course) and also through periods
inbetween disablement of an attached device and successful detection
of the next one.
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
