Posting receive buffer operation can fail, thus we should make
sure to have an error flow during initialization phase. While
we're here, add a debug print in case of a failure.
Signed-off-by: Max Gurtovoy <maxg@mellanox.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
ib_post_send operation should succeed unless something unusual
happened to the ib device.
Signed-off-by: Max Gurtovoy <maxg@mellanox.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
The patch enables inline data sizes using up to 4 recv sges, and capping
the size at 16KB or at least 1 page size. So on a 4K page system, up to
16KB is supported, and for a 64K page system 1 page of 64KB is supported.
We avoid > 0 order page allocations for the inline buffers by using
multiple recv sges, one for each page. If the device cannot support
the configured inline data size due to lack of enough recv sges, then
log a warning and reduce the inline size.
Add a new configfs port attribute, called param_inline_data_size,
to allow configuring the size of inline data for a given nvmf port.
The maximum size allowed is still enforced by nvmet-rdma with
NVMET_RDMA_MAX_INLINE_DATA_SIZE, which is now max(16KB, PAGE_SIZE).
And the default size, if not specified via configfs, is still PAGE_SIZE.
This preserves the existing behavior, but allows larger inline sizes
for small page systems. If the configured inline data size exceeds
NVMET_RDMA_MAX_INLINE_DATA_SIZE, a warning is logged and the size is
reduced. If param_inline_data_size is set to 0, then inline data is
disabled for that nvmf port.
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Allow up to 4 segments of inline data for NVMF WRITE operations. This
reduces latency for small WRITEs by removing the need for the target to
issue a READ WR for IB, or a REG_MR + READ WR chain for iWarp.
Also cap the inline segments used based on the limitations of the
device.
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Add a new "buffered_io" attribute, which disabled direct I/O and thus
enables page cache based caching when enabled. The attribute can only
be changed when the namespace is disabled as the file has to be reopend
for the change to take effect.
The possibly blocking read/write are deferred to a newly introduced
global workqueue.
Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
This patch adds support for Commands Supported and Effects log page
(Log Identifier 05h) for NVMeOF. This also makes it easier to find
which commands are supported, e.g. :-
subnqn : testnqn1
Admin Command Set
ACS2 [Get Log Page ] 00000001
ACS6 [Identify ] 00000001
ACS8 [Abort ] 00000001
ACS9 [Set Features ] 00000001
ACS10 [Get Features ] 00000001
ACS12 [Asynchronous Event Request ] 00000001
ACS24 [Keep Alive ] 00000001
NVM Command Set
IOCS0 [Flush ] 00000001
IOCS1 [Write ] 00000001
IOCS2 [Read ] 00000001
IOCS8 [Write Zeroes ] 00000001
IOCS9 [Dataset Management ] 00000001
This partticular functionality can be used from the host side to examine
the NVMeOF ctrl commands supported.
Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Currently, the code initializes the keep alive work item whenever
nvme_start_keep_alive() is called. However, this routine is called
several times while reconnecting, etc. Although it's hoped that keep
alive is always disabled and not scheduled when start is called,
re-initing if it were scheduled or completing can have very bad
side effects. There's no need for re-initialization.
Move the keep_alive work item and cmd struct initialization to
controller init.
Signed-off-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Added some feature ids present in nvme-cli but not kernel.
Signed-off-by: Revanth Rajashekar <revanth.rajashekar@intel.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Inside blk_mq_try_issue_list_directly(), if the request is issued as
failed, we shouldn't try to do it again, otherwise the warning in
blk_mq_start_request() will be triggered. This change is aligned to
behaviour of other ways of request issue & dispatch.
Fixes: 6ce3dd6eec ("blk-mq: issue directly if hw queue isn't busy in case of 'none'")
Cc: Kashyap Desai <kashyap.desai@broadcom.com>
Cc: Laurence Oberman <loberman@redhat.com>
Cc: Omar Sandoval <osandov@fb.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Bart Van Assche <bart.vanassche@wdc.com>
Cc: Hannes Reinecke <hare@suse.de>
Cc: Kashyap Desai <kashyap.desai@broadcom.com>
Cc: kernel test robot <rong.a.chen@intel.com>
Cc: LKP <lkp@01.org>
Reported-by: kernel test robot <rong.a.chen@intel.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
With the change to use UINT_MAX I broke the depth check as any value of
inflight (ie 0) would be less than (int)UINT_MAX. Fix this by changing
everything to unsigned int to match the depth.
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Add tracking of REQ_OP_DISCARD ios to the per-cgroup io.stat. Two
fields, dbytes and dios, to respectively count the total bytes and
number of discards are added.
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Andy Newell <newella@fb.com>
Cc: Michael Callahan <michaelcallahan@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Add tracking of REQ_OP_DISCARD ios to the partition statistics and
append them to the various stat files in /sys as well as
/proc/diskstats. These are tracked with the same four stats as reads
and writes:
Number of discard ios completed.
Number of discard ios merged
Number of discard sectors completed
Milliseconds spent on discard requests
This is done via adding a new STAT_DISCARD define to genhd.h and then
using it to index that stat field for discard requests.
tj: Refreshed on top of v4.17 and other previous updates.
Signed-off-by: Michael Callahan <michaelcallahan@fb.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Andy Newell <newella@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Add and use a new op_stat_group() function for indexing partition stat
fields rather than indexing them by rq_data_dir() or bio_data_dir().
This function works similarly to op_is_sync() in that it takes the
request::cmd_flags or bio::bi_opf flags and determines which stats
should et updated.
In addition, the second parameter to generic_start_io_acct() and
generic_end_io_acct() is now a REQ_OP rather than simply a read or
write bit and it uses op_stat_group() on the parameter to determine
the stat group.
Note that the partition in_flight counts are not part of the per-cpu
statistics and as such are not indexed via this function. It's now
indexed by op_is_write().
tj: Refreshed on top of v4.17. Updated to pass around REQ_OP.
Signed-off-by: Michael Callahan <michaelcallahan@fb.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Joshua Morris <josh.h.morris@us.ibm.com>
Cc: Philipp Reisner <philipp.reisner@linbit.com>
Cc: Matias Bjorling <mb@lightnvm.io>
Cc: Kent Overstreet <kent.overstreet@gmail.com>
Cc: Alasdair Kergon <agk@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Add defines for STAT_READ and STAT_WRITE for indexing the partition
stat entries. This clarifies some fs/ code which has hardcoded 1 for
STAT_WRITE and will make it easier to extend the stats with additional
fields.
tj: Refreshed on top of v4.17.
Signed-off-by: Michael Callahan <michaelcallahan@fb.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Cc: Jaegeuk Kim <jaegeuk@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Add a part_stat_read_accum macro to genhd.h to read and sum across
field entries. For example to sum up the number read and write
sectors completed. In addition to being ar reasonable cleanup by
itself this will make it easier to add new stat fields in the future.
tj: Refreshed on top of v4.17.
Signed-off-by: Michael Callahan <michaelcallahan@fb.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
c11f0c0b5b ("block/mm: make bdev_ops->rw_page() take a bool for
read/write") replaced @op with boolean @is_write, which limited the
amount of information going into ->rw_page() and more importantly
page_endio(), which removed the need to expose block internals to mm.
Unfortunately, we want to track discards separately and @is_write
isn't enough information. This patch updates bdev_ops->rw_page() to
take REQ_OP instead but leaves page_endio() to take bool @is_write.
This allows the block part of operations to have enough information
while not leaking it to mm.
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Mike Christie <mchristi@redhat.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
* Remove checkpatch errors caused due to assignment operation in if
condition
Signed-off-by: RAGHU Halharvi <raghuhack78@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
In case of 'none' io scheduler, when hw queue isn't busy, it isn't
necessary to enqueue request to sw queue and dequeue it from
sw queue because request may be submitted to hw queue asap without
extra cost, meantime there shouldn't be much request in sw queue,
and we don't need to worry about effect on IO merge.
There are still some single hw queue SCSI HBAs(HPSA, megaraid_sas, ...)
which may connect high performance devices, so 'none' is often required
for obtaining good performance.
This patch improves IOPS and decreases CPU unilization on megaraid_sas,
per Kashyap's test.
Cc: Kashyap Desai <kashyap.desai@broadcom.com>
Cc: Laurence Oberman <loberman@redhat.com>
Cc: Omar Sandoval <osandov@fb.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Bart Van Assche <bart.vanassche@wdc.com>
Cc: Hannes Reinecke <hare@suse.de>
Reported-by: Kashyap Desai <kashyap.desai@broadcom.com>
Tested-by: Kashyap Desai <kashyap.desai@broadcom.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
In our longer tests we noticed that some boxes would degrade to the
point of uselessness. This is because we truncate the current time when
saving it in our bio, but I was using the raw current time to subtract
from. So once the box had been up a certain amount of time it would
appear as if our IO's were taking several years to complete. Fix this
by truncating the current time so it matches the issue time. Verified
this worked by running with this patch for a week on our test tier.
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Early versions of these patches had us waiting for seconds at a time
during submission, so we had to adjust the timing window we monitored
for latency. Now we don't do things like that so this is unnecessary
code.
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We can't know if a block is closed or not on 1.2 devices, so assume
closed state to make sure that blocks are erased before writing.
Fixes: 32ef9412c1 ("lightnvm: pblk: implement get log report chunk")
Signed-off-by: Hans Holmberg <hans.holmberg@cnexlabs.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
In the read path, partial reads are currently performed synchronously
which affects performance for workloads that generate many partial
reads. This patch adds an asynchronous partial read path as well as
the required partial read ctx.
Signed-off-by: Heiner Litz <hlitz@ucsc.edu>
Reviewed-by: Igor Konopko <igor.j.konopko@intel.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
In preparation to enabling -Wimplicit-fallthrough, mark switch cases
where we are expecting to fall through.
Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The error messages in pblk does not say which pblk instance that
a message occurred from. Update each error message to reflect the
instance it belongs to, and also prefix it with pblk, so we know
the message comes from the pblk module.
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Reviewed-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
For devices that does not specify a limit on its transfer size, the
get_chk_meta command may send down a single I/O retrieving the full
chunk metadata table. Resulting in large 2-4MB I/O requests. Instead,
split up the I/Os to a maximum of 256KB and issue them separately to
reduce memory requirements.
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Reviewed-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
If using pblk on a 32bit architecture, and there is a need to
perform a partial read, the partial read bitmap will only have
allocated 32 entries, where as 64 are needed.
Make sure that the read_bitmap is initialized to 64bits on 32bit
architectures as well.
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Reviewed-by: Igor Konopko <igor.j.konopko@intel.com>
Reviewed-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Since both blk_old_get_request() and blk_mq_alloc_request() initialize
rq->__data_len to zero, it is not necessary to initialize that member
in nvme_nvm_alloc_request(). Hence remove the rq->__data_len
initialization from nvme_nvm_alloc_request().
Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
When recovering a line, an extra check was added when debugging was
active, such that minor version where also checked. Unfortunately,
this used the ifdef NVM_DEBUG, which is not correct.
Instead use the proper DEBUG def, and now that it compiles, also fix
the variable.
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Fixes: d0ab0b1ab9 ("lightnvm: pblk: check data lines version on recovery")
Reviewed-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
There is no users of CONFIG_NVM_DEBUG in the LightNVM subsystem. All
users are in pblk. Rename NVM_DEBUG to NVM_PBLK_DEBUG and enable
only for pblk.
Also fix up the CONFIG_NVM_PBLK entry to follow the code style for
Kconfig files.
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Reviewed-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Some devices can expose mw_cunits equal to 0, it can cause the
creation of too small write buffer and cause performance to drop
on write workloads.
Additionally, write buffer size must cover write data requirements,
such as WS_MIN and MW_CUNITS - it must be greater than or equal to
the larger one multiplied by the number of PUs. However, for
performance reasons, use the WS_OPT value to calculation instead of
WS_MIN.
Because the place where buffer size is calculated was changed, this
patch also removes pgs_in_buffer filed in pblk structure.
Signed-off-by: Marcin Dziegielewski <marcin.dziegielewski@intel.com>
Signed-off-by: Igor Konopko <igor.j.konopko@intel.com>
Reviewed-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Remove blkdev_entry_to_request() macro, which remained unused through
the observable history, also note that it repeats list_entry_rq() macro
verbatim.
Signed-off-by: Vladimir Zapolskiy <vz@mleia.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Use the existing %pad printk format to print dma_addr_t values.
This avoids the following warnings when compiling on the parisc64 platform:
drivers/block/skd_main.c: In function 'skd_preop_sg_list':
drivers/block/skd_main.c:660:4: warning: format '%llx' expects argument of type 'long long unsigned int', but argument 6 has type 'dma_addr_t {aka unsigned int}' [-Wformat=]
Reviewed-by: Bart Van Assche <bart.vanassche@wdc.com>
Signed-off-by: Helge Deller <deller@gmx.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The code poses a security risk due to user memory access in ->release
and had an API that can't be used reliably. As far as we know it was
never used for real, but if that turns out wrong we'll have to revert
this commit and come up with a band aid.
Jann Horn did look software archives for users of this interface,
and the only users found were example code in sg3_utils, and optional
support in an optional module of the tgt user space iscsi target,
which looks like a proof of concept extension of the /dev/sg
read/write support.
Tony Battersby chimes in that the code is basically unsafe to use in
general:
The read/write interface on /dev/bsg is impossible to use safely
because the list of completed commands is per-device (bd->done_list)
rather than per-fd like it is with /dev/sg. So if program A and
program B are both using the write/read interface on the same bsg
device, then their command responses will get mixed up, and program
A will read() some command results from program B and vice versa.
So no, I don't use read/write on /dev/bsg. From a security standpoint,
it should definitely be fixed or removed.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
max_depth used to be a u64, but I changed it to a unsigned int but
didn't convert my comparisons over everywhere. Fix by using UINT_MAX
everywhere instead of (u64)-1.
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
On 32-bit architectures, dividing a 64-bit number needs to use the
do_div() function or something like it to avoid a link failure:
block/blk-iolatency.o: In function `iolatency_prfill_limit':
blk-iolatency.c:(.text+0x8cc): undefined reference to `__aeabi_uldivmod'
Using div_u64() gives us the best output and avoids the need for an
explicit cast.
Fixes: d706751215 ("block: introduce blk-iolatency io controller")
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Fix build warnings in DAC960.c when CONFIG_PROC_FS is not enabled
by marking the unused functions as __maybe_unused.
../drivers/block/DAC960.c:6429:12: warning: 'dac960_proc_show' defined but not used [-Wunused-function]
../drivers/block/DAC960.c:6449:12: warning: 'dac960_initial_status_proc_show' defined but not used [-Wunused-function]
../drivers/block/DAC960.c:6456:12: warning: 'dac960_current_status_proc_show' defined but not used [-Wunused-function]
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: linux-block@vger.kernel.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Adds support for exposing a null_blk device through the zone device
interface.
The interface is managed with the parameters zoned and zone_size.
If zoned is set, the null_blk instance registers as a zoned block
device. The zone_size parameter defines how big each zone will be.
Signed-off-by: Matias Bjørling <matias.bjorling@wdc.com>
Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Split the null_blk device driver, such that it can prepare for
zoned block interface support.
Signed-off-by: Matias Bjørling <matias.bjorling@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
With gcc 4.9.0 and 7.3.0:
block/blk-core.c: In function 'blk_pm_allow_request':
block/blk-core.c:2747:2: warning: enumeration value 'RPM_ACTIVE' not handled in switch [-Wswitch]
switch (rq->q->rpm_status) {
^
Convert the return statement below the switch() block into a default
case to fix this.
Fixes: e4f36b249b ("block: fix peeking requests during PM")
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
If __blkdev_issue_discard is in progress and a device mapper device is
reloaded with a table that doesn't support discard,
q->limits.max_discard_sectors is set to zero. This results in infinite
loop in __blkdev_issue_discard.
This patch checks if max_discard_sectors is zero and aborts with
-EOPNOTSUPP.
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Tested-by: Zdenek Kabelac <mpatocka@redhat.com>
Cc: stable@vger.kernel.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The flag GFP_ATOMIC already contains __GFP_HIGH. There is no need to
explicitly or __GFP_HIGH again. So, just remove unnecessary __GFP_HIGH.
Signed-off-by: Shakeel Butt <shakeelb@google.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Currently mbps knob could only be set once before switching power knob to
on, after power knob has been set at least once, there is no way to set
mbps knob again due to -EBUSY.
As nullb is mainly used for testing, in order to make it flexible, this
removes the flag NULLB_DEV_FL_CONFIGURED so that mbps knob can be reset
when power knob is off, e.g.
echo 0 > /config/nullb/a/power
echo 40 > /config/nullb/a/mbps
echo 1 > /config/nullb/a/power
So does other knobs under /config/nullb/a.
Signed-off-by: Liu Bo <bo.liu@linux.alibaba.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We noticed in testing we'd get pretty bad latency stalls under heavy
pressure because read ahead would try to do its thing while the cgroup
was under severe pressure. If we're under this much pressure we want to
do as little IO as possible so we can still make progress on real work
if we're a throttled cgroup, so just skip readahead if our group is
under pressure.
Signed-off-by: Josef Bacik <jbacik@fb.com>
Acked-by: Tejun Heo <tj@kernel.org>
Acked-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
A basic documentation to describe the interface, statistics, and
behavior of io.latency.
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Current IO controllers for the block layer are less than ideal for our
use case. The io.max controller is great at hard limiting, but it is
not work conserving. This patch introduces io.latency. You provide a
latency target for your group and we monitor the io in short windows to
make sure we are not exceeding those latency targets. This makes use of
the rq-qos infrastructure and works much like the wbt stuff. There are
a few differences from wbt
- It's bio based, so the latency covers the whole block layer in addition to
the actual io.
- We will throttle all IO types that comes in here if we need to.
- We use the mean latency over the 100ms window. This is because writes can
be particularly fast, which could give us a false sense of the impact of
other workloads on our protected workload.
- By default there's no throttling, we set the queue_depth to INT_MAX so that
we can have as many outstanding bio's as we're allowed to. Only at
throttle time do we pay attention to the actual queue depth.
- We backcharge cgroups for root cg issued IO and induce artificial
delays in order to deal with cases like metadata only or swap heavy
workloads.
In testing this has worked out relatively well. Protected workloads
will throttle noisy workloads down to 1 io at time if they are doing
normal IO on their own, or induce up to a 1 second delay per syscall if
they are doing a lot of root issued IO (metadata/swap IO).
Our testing has revolved mostly around our production web servers where
we have hhvm (the web server application) in a protected group and
everything else in another group. We see slightly higher requests per
second (RPS) on the test tier vs the control tier, and much more stable
RPS across all machines in the test tier vs the control tier.
Another test we run is a slow memory allocator in the unprotected group.
Before this would eventually push us into swap and cause the whole box
to die and not recover at all. With these patches we see slight RPS
drops (usually 10-15%) before the memory consumer is properly killed and
things recover within seconds.
Signed-off-by: Josef Bacik <jbacik@fb.com>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
wbt cares only about request completion time, but controllers may need
information that is on the bio itself, so add a done_bio callback for
rq-qos so things like blk-iolatency can use it to have the bio when it
completes.
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We don't really need to save this stuff in the core block code, we can
just pass the bio back into the helpers later on to derive the same
flags and update the rq->wbt_flags appropriately.
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
blkcg-qos is going to do essentially what wbt does, only on a cgroup
basis. Break out the common code that will be shared between blkcg-qos
and wbt into blk-rq-qos.* so they can both utilize the same
infrastructure.
Signed-off-by: Josef Bacik <jbacik@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We need to use blk_rq_stat in the blkcg qos stuff, so export some of
these helpers so they can be used by other things.
Signed-off-by: Josef Bacik <jbacik@fb.com>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Memory allocations can induce swapping via kswapd or direct reclaim. If
we are having IO done for us by kswapd and don't actually go into direct
reclaim we may never get scheduled for throttling. So instead check to
see if our cgroup is congested, and if so schedule the throttling.
Before we return to user space the throttling stuff will only throttle
if we actually required it.
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Josef Bacik <jbacik@fb.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
