When we requeue a request, we can always insert the request
back to the scheduler instead of doing it when restarting
the queues and kicking the requeue work, so get rid of
the requeue kick in nvme (core and drivers).
Also, now there is no need start hw queues in nvme_kill_queues
We don't stop the hw queues anymore, so no need to
start them.
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
unlike blk_mq_stop_hw_queues and blk_mq_start_stopped_hw_queues
quiescing/unquiescing respects the submission path rcu grace.
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
unlike blk_mq_stop_hw_queues and blk_mq_start_stopped_hw_queues
quiescing/unquiescing respects the submission path rcu grace.
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
unlike blk_mq_stop_hw_queues and blk_mq_start_stopped_hw_queues
quiescing/unquiescing respects the submission path rcu grace.
Also, make sure to unquiesce before cleanup the admin queue.
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-By: James Smart <james.smart@broadcom.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
unlike blk_mq_stop_hw_queues and blk_mq_start_stopped_hw_queues
quiescing/unquiescing respects the submission path rcu grace.
Also make sure to kick the requeue list when appropriate.
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
This patch improves the way the RDMA IB signalling is done by using atomic
operations for the signalling variable. This avoids race conditions on
sig_count.
The signalling interval changes slightly and is now the largest power of
two not larger than queue depth / 2.
ilog() usage idea by Bart Van Assche.
Signed-off-by: Marta Rybczynska <marta.rybczynska@kalray.eu>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Cc: stable@vger.kernel.org
We might have more/less queues once we reconnect/reset. For
example due to cpu going online/offline or controller constraints.
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
We might have more/less queues once we reconnect/reset. For
example due to cpu going online/offline
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
We might have more/less queues once we reconnect/reset. For
example due to cpu going online/offline or controller constraints.
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Its what the user passed, so its probably a better
idea to keep it intact. Also, limit the number of
I/O queues to max online cpus and the lport maximum
hw queues.
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
we are going to need the name for the core routine...
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
All transports use either a private cache of controller cap or an on-stack
copy, move it to the generic struct nvme_ctrl. In the future it will also
be maintained by the core.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
All all transports use the queue_count in exactly the same, so move it to
the generic struct nvme_ctrl. In the future it will also be maintained by
the core.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-By: James Smart <james.smart@broadcom.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
PM1725 controllers have a couple of quirks that need to be handled in
the driver:
- I/O queue depth must be limited to 64 entries on controllers that do
not report MQES.
- The host interface registers go offline briefly while resetting the
chip. Thus a delay is needed before checking whether the controller
is ready.
Note that the admin queue depth is also limited to 64 on older versions
of this board. Since our NVME_AQ_DEPTH is now 32 that is no longer an
issue.
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Do bitmap checks only when debug mode is enable. The line bitmap used
for mapping to physical addresses is fairly large (~512KB) and it is
expensive to do this checks on the fast path.
Signed-off-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Matias Bjørling <matias@cnexlabs.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
When a read is directed to the cache, we risk that the lba has been
updated during the time we made the L2P table lookup and the time we are
actually reading form the cache. We intentionally not hold the L2P lock
not to block other threads.
While strict ordering is not a guarantee at this level (unless REQ_FLUSH
has been previously issued), we have experience that some databases that
have recently implemented direct I/O support, issue metadata reads very
close to the writes, without issuing a fsync in the middle. An easy way
to support them while they is to make an extra effort and check the L2P
map right before reading the cache.
Signed-off-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Matias Bjørling <matias@cnexlabs.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Add a sanity check to the pblk initialization sequence in order to
ensure that enough LUNs have been allocated to store the line metadata.
Signed-off-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Matias Bjørling <matias@cnexlabs.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
When removing a pblk instance, pad the current line using asynchronous
I/O. This reduces the removal time from ~1 minute in the worst case to a
couple of seconds.
Signed-off-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Matias Bjørling <matias@cnexlabs.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
For now, we allocate a per I/O buffer for GC data. Since the potential
size of the buffer is 256KB and GC is not in the fast path, do this
allocation with vmalloc. This puts lets pressure on the memory
allocator at no performance cost.
Signed-off-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Matias Bjørling <matias@cnexlabs.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
When user threads place data into the write buffer, they reserve space
and do the memory copy out of the lock. As a consequence, when the write
thread starts persisting data, there is a chance that it is not copied
yet. In this case, avoid polling, and schedule before retrying.
Signed-off-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Matias Bjørling <matias@cnexlabs.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Prevent pblk->lines being double freed in case of an error during pblk
initialization.
Fixes: dd2a434373: "lightnvm: pblk: sched. metadata on write thread"
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Matias Bjørling <matias@cnexlabs.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Use the right types and conversions on le64 variables. Reported by
sparse.
Signed-off-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Matias Bjørling <matias@cnexlabs.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Remove dead build rule for drivers/nvme/host/scsi.c which has been
removed by commit ("nvme: Remove SCSI translations").
Signed-off-by: Valentin Rothberg <vrothberg@suse.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This patch performs sequential mapping between CPUs and queues.
In case the system has more CPUs than HWQs then there are still
CPUs to map to HWQs. In hyperthreaded system, map the unmapped CPUs
and their siblings to the same HWQ.
This actually fixes a bug that found unmapped HWQs in a system with
2 sockets, 18 cores per socket, 2 threads per core (total 72 CPUs)
running NVMEoF (opens upto maximum of 64 HWQs).
Performance results running fio (72 jobs, 128 iodepth)
using null_blk (w/w.o patch):
bs IOPS(read submit_queues=72) IOPS(write submit_queues=72) IOPS(read submit_queues=24) IOPS(write submit_queues=24)
----- ---------------------------- ------------------------------ ---------------------------- -----------------------------
512 4890.4K/4723.5K 4524.7K/4324.2K 4280.2K/4264.3K 3902.4K/3909.5K
1k 4910.1K/4715.2K 4535.8K/4309.6K 4296.7K/4269.1K 3906.8K/3914.9K
2k 4906.3K/4739.7K 4526.7K/4330.6K 4301.1K/4262.4K 3890.8K/3900.1K
4k 4918.6K/4730.7K 4556.1K/4343.6K 4297.6K/4264.5K 3886.9K/3893.9K
8k 4906.4K/4748.9K 4550.9K/4346.7K 4283.2K/4268.8K 3863.4K/3858.2K
16k 4903.8K/4782.6K 4501.5K/4233.9K 4292.3K/4282.3K 3773.1K/3773.5K
32k 4885.8K/4782.4K 4365.9K/4184.2K 4307.5K/4289.4K 3780.3K/3687.3K
64k 4822.5K/4762.7K 2752.8K/2675.1K 4308.8K/4312.3K 2651.5K/2655.7K
128k 2388.5K/2313.8K 1391.9K/1375.7K 2142.8K/2152.2K 1395.5K/1374.2K
Signed-off-by: Max Gurtovoy <maxg@mellanox.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We can deadlock in case we got to a device removal
event on a queue which is already in the process of
destroying the cm_id is this is blocking until all
events on this cm_id will drain. On the other hand
we cannot guarantee that rdma_destroy_id was invoked
as we only have indication that the queue disconnect
flow has been queued (the queue state is updated before
the realease work has been queued).
So, we leave all the queue removal to a separate ib_client
to avoid this deadlock as ib_client device removal is in
a different context than the cm_id itself.
Reported-by: Shiraz Saleem <shiraz.saleem@intel.com>
Tested-by: Shiraz Saleem <shiraz.saleem@intel.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Currently, the fc transport invokes nvme_fc_error_recovery() on every
io in which the transport detects an error. Which means:
a) it's really noisy on large io loads that all get hit by a link down.
b) we repeatively call nvme_stop_queues() even though queues are
stopped upon the first error or as first steps of reset_work.
Correct by:
Errors are only meaningful if the controller is in the LIVE state.
Thus, enact the reset_work only if LIVE. If called repeatively, state
will have already transitioned.
There's no need to stop the queues here. Let the first steps of
reset_work do the queue stopping.
Signed-off-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
if a nvme command is issued with an opcode that is not supported by
the target (example: opcode 21 - detach namespace), the target
crashes due to a null pointer.
nvmet_req_init() detects the bad opcode and immediately calls the nvme
command done routine with an error status, allowing the transport to
send the response. However, the FC transport was aborting the command
on error, so the abort freed the lldd point, but the rsp transmit path
referenced it psot the free.
Fix by removing the abort call on nvmet_req_init() failure.
The completion response will be sent with an error status code.
As the completion path will terminate the io, ensure the data_sg
lists show an unused state so that teardown paths are successful.
Signed-off-by: Paul Ely <Paul.Ely@broadcom.com>
Signed-off-by: James Smart <james.smart@broadcom.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
If a controller connection is attempted (say to a subsystem that
does not exist), the first attempt errors out. If another connect
is attempted, it crashes.
Issue is the prior controller has yet execute it's final put, thus
its still on lists. However, opts points on it have been cleared, thus
causing the crash if they are referenced.
Fix is to add the missing put after the nvme_uninit_ctrl() call on
the attachment failure.
Signed-off-by: Paul Ely <Paul.Ely@broadcom.com>
Signed-off-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Per the recommendation by Sagi on:
http://lists.infradead.org/pipermail/linux-nvme/2017-April/009261.html
Wait for io aborts to complete wait converted from msleep look to
using a struct completion.
Signed-off-by: James Smart <james.smart@broadcom.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Current fc transport code, on io termination, is calling
nvme_cleanup_cmd() followed by the transport dma unmap routine
which also calls nvme_cleanup_cmd(). Which means two kfrees occur
on the same address, raising havoc. This resulted in odd data errors,
effectively corruption..
Fix by removing the extraneous double calls. Call now occurs only in
teardown paths and as part of dma unmap routine.
Signed-off-by: James Smart <james.smart@broadcom.com>
Reviewed-by: Ewan D. Milne <emilne@redhat.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
NVMe 1.2.1 or later requires controllers to provide a subsystem NQN in the
Identify controller data structures. Use this NQN for the subsysnqn
sysfs attribute by storing it in the nvme_ctrl structure after verifying
it. For older controllers we generate a "fake" NQN per non-normative
text in the NVMe 1.3 spec.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <keith.busch@intel.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
While a NVMe Namespace is somewhat similar to a SCSI Logical Unit (and not
a Logical Unit Number anyway) there are subtile differences. Remove the
misleading comment.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Sagi Grimberg <sagi@grmberg.me>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
A user reports APST is enabled, even when the NVMe is quirked or with
option "default_ps_max_latency_us=0".
The current logic will not set APST if the device is quirked. But the
NVMe in question will enable APST automatically.
Separate the logic "apst is supported" and "to enable apst", so we can
use the latter one to explicitly disable APST at initialiaztion.
BugLink: https://bugs.launchpad.net/bugs/1699004
Signed-off-by: Kai-Heng Feng <kai.heng.feng@canonical.com>
Reviewed-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
No need to differentiate fabrics from pci/loop, also lower
it to 32 as we don't really need 256 inflight admin commands.
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Currently we have no way to define a stable host-id but always use the one
which is randomly generated when we add the host or use the default host.
Provide a "hostid=%s" for user-space to pass in a persistent host-id which
overrides the randomly generated one.
Signed-off-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The SCSI-to-NVMe translations were added to assist storage applications
utilizing SG_IO transitioning to NVMe. It was always recommended,
however, to use native NVMe for device management as too much is lost
in translation and the maintenance burden in keeping this kludgey
layer around has been neglected such that much of the translations are
completely broken.
This patch removes SG_IO handling from NVMe to avoid any confusion
regarding maintenance support for this interface. The config option for
NVMe SCSI emulation has been disabled by default since 4.5. The driver
has supported native nvme user commands since the beginning, and native
tooling is publicly available for use or as reference for anyone writing
their own tools, so there's no excuse for hanging onto a broken crutch.
Signed-off-by: Keith Busch <keith.busch@intel.com>
Acked-by: Jens Axboe <axboe@kernel.dk>
Reviewed-by: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Guan Junxiong <guanjunxiong@huawei.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Given that the code is simple enough it seems better
then passing a tag by reference for each call site, also
we can now get rid of __nvme_process_cq.
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Also, maintain a consumed counter to rely on for doorbell and
cqe_seen update instead of directly relying on the cq head and phase.
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Makes the code slightly more readable.
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Nice abstraction of the actual mechanics of how to do it.
Note the change that we call it after we assign nvmeq->cq_head
to avoid passing it.
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Some architectures (at least PPC) doesn't like get/put_user with
64-bit types on a 32-bit system. Use the variably sized copy
to/from user variants instead.
Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Fixes: c75b1d9421 ("fs: add fcntl() interface for setting/getting write life time hints")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Drop static on a local variable, when the variable is initialized before
any use, on every possible execution path through the function. The
static has no benefit, and dropping it reduces the code size.
The semantic patch that fixes this problem is as follows:
(http://coccinelle.lip6.fr/)
// <smpl>
@bad exists@
position p;
identifier x;
type T;
@@
static T x@p;
...
x = <+...x...+>
@@
identifier x;
expression e;
type T;
position p != bad.p;
@@
-static
T x@p;
... when != x
when strict
?x = e;
// </smpl>
The change in code size is indicates by the following output from the size
command.
before:
text data bss dec hex filename
67299 2291 1056 70646 113f6 drivers/block/drbd/drbd_nl.o
after:
text data bss dec hex filename
67283 2291 1056 70630 113e6 drivers/block/drbd/drbd_nl.o
Signed-off-by: Julia Lawall <Julia.Lawall@lip6.fr>
Signed-off-by: Roland Kammerer <roland.kammerer@linbit.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This commit fixes a bug triggered by a non-trivial sequence of
events. These events are briefly described in the next two
paragraphs. The impatiens, or those who are familiar with queue
merging and splitting, can jump directly to the last paragraph.
On each I/O-request arrival for a shared bfq_queue, i.e., for a
bfq_queue that is the result of the merge of two or more bfq_queues,
BFQ checks whether the shared bfq_queue has become seeky (i.e., if too
many random I/O requests have arrived for the bfq_queue; if the device
is non rotational, then random requests must be also small for the
bfq_queue to be tagged as seeky). If the shared bfq_queue is actually
detected as seeky, then a split occurs: the bfq I/O context of the
process that has issued the request is redirected from the shared
bfq_queue to a new non-shared bfq_queue. As a degenerate case, if the
shared bfq_queue actually happens to be shared only by one process
(because of previous splits), then no new bfq_queue is created: the
state of the shared bfq_queue is just changed from shared to non
shared.
Regardless of whether a brand new non-shared bfq_queue is created, or
the pre-existing shared bfq_queue is just turned into a non-shared
bfq_queue, several parameters of the non-shared bfq_queue are set
(restored) to the original values they had when the bfq_queue
associated with the bfq I/O context of the process (that has just
issued an I/O request) was merged with the shared bfq_queue. One of
these parameters is the weight-raising state.
If, on the split of a shared bfq_queue,
1) a pre-existing shared bfq_queue is turned into a non-shared
bfq_queue;
2) the previously shared bfq_queue happens to be busy;
3) the weight-raising state of the previously shared bfq_queue happens
to change;
the number of weight-raised busy queues changes. The field
wr_busy_queues must then be updated accordingly, but such an update
was missing. This commit adds the missing update.
Reported-by: Luca Miccio <lucmiccio@gmail.com>
Signed-off-by: Paolo Valente <paolo.valente@linaro.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
BLK_BOUNCE_ANY is the defauly now, so the call is superflous.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Now all queues allocators come without abounce limit by default,
dm doesn't have to override this anymore.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
