The default nvme_tcp_wq will use all CPUs to process tasks. Sometimes it is
necessary to set CPU affinity to improve performance.
A new module parameter wq_unbound is added here. If set to true, users can
configure cpu affinity through
/sys/devices/virtual/workqueue/nvme_tcp_wq/cpumask.
Signed-off-by: Li Feng <fengli@smartx.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Make the workqueue userspace visible for easy viewing and configuration.
Signed-off-by: Li Feng <fengli@smartx.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Keith Busch <kbusch@kernel.org>
This commit adds NVME_QUIRK_NO_DEEPEST_PS and NVME_QUIRK_BOGUS_NID for
device [126f:2262], which appears to be a generic VID:PID pair used for
many SSDs based on the Silicon Motion SM2262/SM2262EN controller.
Two of my SSDs with this VID:PID pair exhibit the same behavior:
* They frequently have trouble exiting the deepest power state (5),
resulting in the entire disk unresponsive.
Verified by setting nvme_core.default_ps_max_latency_us=10000 and
observing them behaving normally.
* They produce all-zero nguid and eui64 with `nvme id-ns` command.
The offending products are:
* HP SSD EX950 1TB
* HIKVISION C2000Pro 2TB
Signed-off-by: Jiawei Fu <i@ibugone.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Add the parse of format command's lbafu to calculate lbaf.
Signed-off-by: Guixin Liu <kanie@linux.alibaba.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Add detailed parsing of reservation commands to make the trace log
more consistent and human-readable.
Signed-off-by: Guixin Liu <kanie@linux.alibaba.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Parse zone mgmt send commands's zsa and receive command's
zrasf to string to make the trace log more human-readable.
Signed-off-by: Guixin Liu <kanie@linux.alibaba.com>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Use nvme_disk_is_ns_head helper instead of check fops directly,
and also drop CONFIG_NVME_MULTIPATH check.
Signed-off-by: Guixin Liu <kanie@linux.alibaba.com>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
We found a issue on production environment while using NVMe over RDMA,
admin_q reconnect failed forever while remote target and network is ok.
After dig into it, we found it may caused by a ABBA deadlock due to tag
allocation. In my case, the tag was hold by a keep alive request
waiting inside admin_q, as we quiesced admin_q while reset ctrl, so the
request maked as idle and will not process before reset success. As
fabric_q shares tagset with admin_q, while reconnect remote target, we
need a tag for connect command, but the only one reserved tag was held
by keep alive command which waiting inside admin_q. As a result, we
failed to reconnect admin_q forever. In order to fix this issue, I
think we should keep two reserved tags for admin queue.
Fixes: ed01fee283 ("nvme-fabrics: only reserve a single tag")
Signed-off-by: Chunguang Xu <chunguang.xu@shopee.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
The .remove() callback for a platform driver returns an int which makes
many driver authors wrongly assume it's possible to do error handling by
returning an error code. However the value returned is ignored (apart
from emitting a warning) and this typically results in resource leaks.
To improve here there is a quest to make the remove callback return
void. In the first step of this quest all drivers are converted to
.remove_new(), which already returns void. Eventually after all drivers
are converted, .remove_new() will be renamed to .remove().
Trivially convert this driver from always returning zero in the remove
callback to the void returning variant.
Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
When the length check for an icreq sqe fails we should not
continue processing but rather return immediately as all
other contents of that sqe cannot be relied on.
Signed-off-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Keith Busch <kbusch@kernel.org>
User visible messages containing the word "timeout" can be alarming.
This one from nvme is just reporting a potentially informative device
configuration, and everything is working as designed. Change the text to
report the less concerning "D3 entry latency", which is where this value
comes from anyway.
Reported-by: Len Brown <lenb@kernel.org>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
The .remove() callback for a platform driver returns an int which makes
many driver authors wrongly assume it's possible to do error handling by
returning an error code. However the value returned is ignored (apart
from emitting a warning) and this typically results in resource leaks.
To improve here there is a quest to make the remove callback return
void. In the first step of this quest all drivers are converted to
.remove_new(), which already returns void. Eventually after all drivers
are converted, .remove_new() will be renamed to .remove().
Trivially convert this driver from always returning zero in the remove
callback to the void returning variant.
Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Link: https://lore.kernel.org/r/20240307180837.190626-2-u.kleine-koenig@pengutronix.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Pull NVMe updates from Keith:
"nvme updates for Linux 6.9
- RDMA target enhancements (Max)
- Fabrics fixes (Max, Guixin, Hannes)
- Atomic queue_limits usage (Christoph)
- Const use for class_register (Ricardo)
- Identification error handling fixes (Shin'ichiro, Keith)"
* tag 'nvme-6.9-2024-03-07' of git://git.infradead.org/nvme: (31 commits)
nvme: clear caller pointer on identify failure
nvme: host: fix double-free of struct nvme_id_ns in ns_update_nuse()
nvme: fcloop: make fcloop_class constant
nvme: fabrics: make nvmf_class constant
nvme: core: constify struct class usage
nvme-fabrics: typo in nvmf_parse_key()
nvme-multipath: use atomic queue limits API for stacking limits
nvme-multipath: pass queue_limits to blk_alloc_disk
nvme: use the atomic queue limits update API
nvme: cleanup nvme_configure_metadata
nvme: don't query identify data in configure_metadata
nvme: split out a nvme_identify_ns_nvm helper
nvme: move common logic into nvme_update_ns_info
nvme: move setting the write cache flags out of nvme_set_queue_limits
nvme: move a few things out of nvme_update_disk_info
nvme: don't use nvme_update_disk_info for the multipath disk
nvme: move blk_integrity_unregister into nvme_init_integrity
nvme: cleanup the nvme_init_integrity calling conventions
nvme: move max_integrity_segments handling out of nvme_init_integrity
nvme: remove nvme_revalidate_zones
...
bcache currently calculates the stripe size for the non-cached_dev
case directly in bcache_device_init, but for the cached_dev case it does
it in the caller. Consolidate it in one places, which also enables
setting the io_opt queue_limit before allocating the gendisk so that it
can be passed in instead of changing the limit just after the allocation.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Coly Li <colyli@suse.de>
Link: https://lore.kernel.org/r/20240226104826.283067-2-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
In virtblk_read_zoned_limits(), setting a zoned block device maximum
number of open and active zones using the functions
disk_set_max_open_zones() and disk_set_max_active_zones() is incorrect
as setting the limits for the request queue is now done atomically when
the gendisk is created (with blk_mq_alloc_disk()). The value set by the
disk_set_max_open/active_zones() functions will be overwritten.
Fix this by setting the maximum number of open and active zones directly
in the queue_limits structure passed to virtblk_read_zoned_limits().
Fixes: 8b83725656 ("virtio_blk: pass queue_limits to blk_mq_alloc_disk")
Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Link: https://lore.kernel.org/r/20240301192639.410183-2-dlemoal@kernel.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This patch is against CVE-2023-6270. The description of cve is:
A flaw was found in the ATA over Ethernet (AoE) driver in the Linux
kernel. The aoecmd_cfg_pkts() function improperly updates the refcnt on
`struct net_device`, and a use-after-free can be triggered by racing
between the free on the struct and the access through the `skbtxq`
global queue. This could lead to a denial of service condition or
potential code execution.
In aoecmd_cfg_pkts(), it always calls dev_put(ifp) when skb initial
code is finished. But the net_device ifp will still be used in
later tx()->dev_queue_xmit() in kthread. Which means that the
dev_put(ifp) should NOT be called in the success path of skb
initial code in aoecmd_cfg_pkts(). Otherwise tx() may run into
use-after-free because the net_device is freed.
This patch removed the dev_put(ifp) in the success path in
aoecmd_cfg_pkts(), and added dev_put() after skb xmit in tx().
Link: https://nvd.nist.gov/vuln/detail/CVE-2023-6270
Fixes: 7562f876cd ("[NET]: Rework dev_base via list_head (v3)")
Signed-off-by: Chun-Yi Lee <jlee@suse.com>
Link: https://lore.kernel.org/r/20240305082048.25526-1-jlee@suse.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Switch drbd_reconsider_queue_parameters to set up the queue parameters
in an on-stack queue_limits structure and apply the atomically. Remove
various helpers that have become so trivial that they can be folded into
drbd_reconsider_queue_parameters.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20240305134041.137006-8-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Pull MD fixes from Song:
"This set fixes two issues:
1. dmraid regression since 6.7 kernels. This issue was initially
reported in [1]. This set of fix has been reviewed and tested by
md and dm folks.
2. raid5 hang since 6.7 kernel, reported in [2]. We haven't got a
better fix for this issue yet. This revert is a workaround. It has
been applied to 6.7 stable kernels [3], and proved to be affective.
We will look more into this issue for a better fix.
[1] https://lore.kernel.org/linux-raid/e5e8afe2-e9a8-49a2-5ab0-958d4065c55e@redhat.com/
[2] https://lore.kernel.org/linux-raid/20240123005700.9302-1-dan@danm.net/
[3] 87165c64fe1a in linux-6.7.y branch."
* tag 'md-6.9-20240305' of https://git.kernel.org/pub/scm/linux/kernel/git/song/md:
dm-raid: fix lockdep waring in "pers->hot_add_disk"
dm-raid456, md/raid456: fix a deadlock for dm-raid456 while io concurrent with reshape
dm-raid: add a new helper prepare_suspend() in md_personality
md/dm-raid: don't call md_reap_sync_thread() directly
dm-raid: really frozen sync_thread during suspend
md: add a new helper reshape_interrupted()
md: export helper md_is_rdwr()
md: export helpers to stop sync_thread
md: don't clear MD_RECOVERY_FROZEN for new dm-raid until resume
Revert "Revert "md/raid5: Wait for MD_SB_CHANGE_PENDING in raid5d""
Pass the constant limits directly to blk_mq_alloc_disk, set the nonrot
flag there as well, and then use the commit API to change the transfer
size and logical block size dependent values.
This relies on the assumption that no I/O can be pending before the
devices moves into the ready state and doesn't need extra freezing
for changes to the queue limits.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Stefan Haberland <sth@linux.ibm.com>
Link: https://lore.kernel.org/r/20240228133742.806274-4-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Most of the code in setup_blk_queue is shared between all disciplines.
Move it to common code and leave a method to query the maximum number
of transferable blocks, and a flag to indicate discard support.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Stefan Haberland <sth@linux.ibm.com>
Link: https://lore.kernel.org/r/20240228133742.806274-3-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The memory allocated for the identification is freed on failure. Set
it to NULL so the caller doesn't have a pointer to that freed address.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
