fio.git
2 months agoMerge branch 'master' of https://github.com/bvanassche/fio
Jens Axboe [Fri, 22 Oct 2021 16:19:04 +0000 (10:19 -0600)]
Merge branch 'master' of https://github.com/bvanassche/fio

* 'master' of https://github.com/bvanassche/fio:
  Android: Add io_uring support

3 months agoAndroid: Add io_uring support
Bart Van Assche [Thu, 21 Oct 2021 21:41:40 +0000 (14:41 -0700)]
Android: Add io_uring support

This patch has been tested on a recent Android phone. Compilation of this
patch has been verified as follows:

    NDK=/usr/lib/android-ndk
    export LIBS="-landroid"
    export UNAME=Android
    for ((i=23;i<=30;i++)); do
        echo "==== i = $i ===="
        export CC=$NDK/toolchains/llvm/prebuilt/linux-x86_64/bin/aarch64-linux-android${i}-clang
        [ -e "$CC" ] || continue
        ./configure && make -j$(nproc) fio || break
    done

Signed-off-by: Bart Van Assche <bvanassche@acm.org>
3 months agoMerge branch 'patch-1' of https://github.com/sweettea/fio
Jens Axboe [Tue, 19 Oct 2021 22:09:21 +0000 (16:09 -0600)]
Merge branch 'patch-1' of https://github.com/sweettea/fio

* 'patch-1' of https://github.com/sweettea/fio:
  t/fuzz: Clean up generated dependency makefiles

3 months agot/fuzz: Clean up generated dependency makefiles
Sweet Tea Dorminy [Tue, 19 Oct 2021 20:31:27 +0000 (16:31 -0400)]
t/fuzz: Clean up generated dependency makefiles

Currently, the 'clean' target cleans up the t/ directory, but not its
subdirectories. As t/fuzz contains c files, though, dependency makefiles
are created there and should be cleaned up.

Signed-off-by: Sweet Tea Dorminy <sweettea@dorminy.me>
3 months agoMerge branch 'fixes_1290' of https://github.com/rthardin/fio
Jens Axboe [Tue, 19 Oct 2021 01:29:46 +0000 (19:29 -0600)]
Merge branch 'fixes_1290' of https://github.com/rthardin/fio

* 'fixes_1290' of https://github.com/rthardin/fio:
  Use min_bs in rate_process=poisson

3 months agoUse min_bs in rate_process=poisson
Ryan Hardin [Mon, 18 Oct 2021 20:43:22 +0000 (16:43 -0400)]
Use min_bs in rate_process=poisson

This fixes an issue where IOPS targets were not met
when the `bs` parameter was not given explicitly, such
as when using `bssplit`.

Fixes #1290

Signed-off-by: Ryan Hardin <ryan.hardin@nutanix.com>
3 months agorun-fio-tests: make test runs more resilient
Vincent Fu [Tue, 21 Sep 2021 21:27:11 +0000 (21:27 +0000)]
run-fio-tests: make test runs more resilient

Catch exceptions that occur during test setup/running/evaluation. This
makes it more likely that the entire test suite can run to completion
even if some tests fail in an unexpected fashion.

In particular I have seen failures in FioJobTest_t0014() when the test
is run on a bare metal machine. Without this patch these failures make
the entire script grind to a halt.

Signed-off-by: Vincent Fu <vincent.fu@samsung.com>
Link: https://lore.kernel.org/r/20210921212639.61319-1-vincent.fu@samsung.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 months agot/zbd: Add -w option to ensure no open zone before write tests
Shin'ichiro Kawasaki [Wed, 13 Oct 2021 06:09:03 +0000 (15:09 +0900)]
t/zbd: Add -w option to ensure no open zone before write tests

The commit b34eb155e4a6 ("t/zbd: Reset all zones before test when max
open zones is specified") introduced -o max_open_zones option to the
script t/zbd/test-zbd-support. It passes max_open_zones value to fio and
resets all zones of the test target device before each test case run
with write operation. This zone reset by the script ensures that no zone
out of the IO range is in open status and the write operation do not
exceed the max_open_zones limit.

On the other hand, since commit d2f442bc0bd5 ("ioengines: add
get_max_open_zones zoned block device operation"), fio automatically
fetches the max_open_zones value. So it is no longer required to pass
the max_open_zones value from the script to fio. To simplify the script
usage, introduce -w option which does not require max_open_zones value.
This option just resets zones before test cases with write operation.

Of note is that fio itself resets the zones exceeding max_open_zones
limit since the commit 954217b90191 ("zbd: Initialize open zones list
referring zone status at fio start"), but it just resets zones within
the fio IO range. Still zone reset by the test script is required for
zones out of IO range. Zone reset out of IO range by fio is not
implemented since it may cause unexpected data erasure.

Suggested-by: Niklas Cassel <niklas.cassel@wdc.com>
Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Reviewed-by: Niklas Cassel <niklas.cassel@wdc.com>
Link: https://lore.kernel.org/r/20211013060903.166543-6-shinichiro.kawasaki@wdc.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 months agot/zbd: Align block size to zone capacity
Shin'ichiro Kawasaki [Wed, 13 Oct 2021 06:09:02 +0000 (15:09 +0900)]
t/zbd: Align block size to zone capacity

The test cases #5, #6, #15 and #37 writes data and read it back (or
write with verify option for read back). When test target zones have
zone capacity unaligned to the block size, read request can not be made
to all of the written data, and the test cases fail.

To avoid the failures, check zone capacity of zones and get block size
which can align to the zone capacities. Then use the block size for the
test cases.

Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Reviewed-by: Niklas Cassel <niklas.cassel@wdc.com>
Link: https://lore.kernel.org/r/20211013060903.166543-5-shinichiro.kawasaki@wdc.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 months agot/zbd: Do not use too large block size in test case #4
Shin'ichiro Kawasaki [Wed, 13 Oct 2021 06:09:01 +0000 (15:09 +0900)]
t/zbd: Do not use too large block size in test case #4

The test case #4 specifies zone size as block size to read a zone. For
some devices, zone size is very large in GB order, then single pread64
system call can not complete the request. This makes the test case fail.

To avoid the failure, keep the block size adequate. If zone size is too
large, use logical_block_size * 256 as the block size.

Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Reviewed-by: Niklas Cassel <niklas.cassel@wdc.com>
Link: https://lore.kernel.org/r/20211013060903.166543-4-shinichiro.kawasaki@wdc.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 months agozbd: Fix type of local variable min_bs
Shin'ichiro Kawasaki [Wed, 13 Oct 2021 06:09:00 +0000 (15:09 +0900)]
zbd: Fix type of local variable min_bs

In zbd.c, thread option min_bs[] is referred and stored in the local
variable min_bs. Elements of min_bs[] have type unsigned long long, but
the local variable min_bs has type uint32_t. When an element of min_bs[]
has value larger than UINT32_MAX, it overflows on assignment to min_bs.

To avoid the overflow, fix type of the local variable min_bs from
uint32_t to uint64_t. Use uint64_t rather than unsigned long long to be
more specific about data size and consistency in zbd.c. The variable is
passed to the helper function zbd_find_zone(), then fix the type of the
argument of the function also.

Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Reviewed-by: Niklas Cassel <niklas.cassel@wdc.com>
Link: https://lore.kernel.org/r/20211013060903.166543-3-shinichiro.kawasaki@wdc.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 months agozbd: Remove cast to unsigned long long for printf
Shin'ichiro Kawasaki [Wed, 13 Oct 2021 06:08:59 +0000 (15:08 +0900)]
zbd: Remove cast to unsigned long long for printf

Many of the variables in zbd.c have type uint64_t. They are casted to
unsigned long long and printed with printf %llu format to handle
uint64_t types difference among architectures. This requires many
lengthy casts to unsigned long long.

To simplify the code, remove the casts to unsigned long long. Some of
the casts are simply unnecessary. To remove other casts, replace the
printf format %llu with PRIu64 so that uint64_t type difference among
architectures is handled accordingly.

Fio build pass of this change was confirmed with 32bit ARM cross
compiler and 64bit x86 compiler.

Suggested-by: Niklas Cassel <niklas.cassel@wdc.com>
Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Reviewed-by: Niklas Cassel <niklas.cassel@wdc.com>
Link: https://lore.kernel.org/r/20211013060903.166543-2-shinichiro.kawasaki@wdc.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 months agoengines/http.c: add fallthrough annotation to _curl_trace
Rebecca Cran [Sat, 16 Oct 2021 06:17:38 +0000 (00:17 -0600)]
engines/http.c: add fallthrough annotation to _curl_trace

To avoid the warning from clang "warning: unannotated fall-through
between switch labels [-Wimplicit-fallthrough]" swap the "fall through"
comment with the "fallthrough;" annotation from compiler.h.

Since the second "fall through" comment isn't really a new fall-through,
remove it.

Signed-off-by: Rebecca Cran <rebecca@bsdio.com>
Link: https://lore.kernel.org/r/20211016061738.76654-1-rebecca@bsdio.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 months agot/io_uring: Fix the parameters calculation for multiple threads scenario
Pankaj Raghav [Fri, 15 Oct 2021 12:09:56 +0000 (14:09 +0200)]
t/io_uring: Fix the parameters calculation for multiple threads scenario

The this_done, this_call and this_reap parameter should be a summation of
the corresponding field from all the submitters.

Currently, we are adding the done, calls and reaps param of the last used
submitter nthread times.

Signed-off-by: Pankaj Raghav <p.raghav@samsung.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 months agoMerge branch 'evelu-typo' of https://github.com/ErwanAliasr1/fio
Jens Axboe [Thu, 14 Oct 2021 21:01:30 +0000 (15:01 -0600)]
Merge branch 'evelu-typo' of https://github.com/ErwanAliasr1/fio

* 'evelu-typo' of https://github.com/ErwanAliasr1/fio:
  t/io_uring: Fixing typo

3 months agot/io_uring: Fixing typo
Erwan Velu [Thu, 14 Oct 2021 20:38:36 +0000 (22:38 +0200)]
t/io_uring: Fixing typo

s/Maxiumum/Maximum/g

Signed-off-by: Erwan Velu <erwanaliasr1@gmail.com>
3 months agot/io_uring: include a maximum IOPS seen when exiting
Jens Axboe [Thu, 14 Oct 2021 13:56:56 +0000 (07:56 -0600)]
t/io_uring: include a maximum IOPS seen when exiting

Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 months agot/io_uring: don't append 'K' to IOPS if we don't divide by 1000
Jens Axboe [Wed, 13 Oct 2021 12:17:44 +0000 (06:17 -0600)]
t/io_uring: don't append 'K' to IOPS if we don't divide by 1000

Impressive two errors in that silly change.

Reported-by: Erwan Velu <erwanaliasr1@gmail.com>
Fixes: dc10c23ab9a7 ("t/io_uring: show IOPS in increments of 1000 IOPS if necessary")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 months agot/io_uring: update for new DMA map buffers API
Jens Axboe [Wed, 13 Oct 2021 00:41:14 +0000 (18:41 -0600)]
t/io_uring: update for new DMA map buffers API

Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 months agot/io_uring: add test support for pre mapping DMA buffers
Jens Axboe [Tue, 12 Oct 2021 20:25:23 +0000 (14:25 -0600)]
t/io_uring: add test support for pre mapping DMA buffers

This is in no shape or form the final evolution or API of this, but
easier to stuff it in here for testing.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 months agot/io_uring: fix silly identical branch error
Jens Axboe [Tue, 12 Oct 2021 20:09:33 +0000 (14:09 -0600)]
t/io_uring: fix silly identical branch error

The previous change inadvertently added the / 1000 to both branches, it
should of course only be done on the first one.

Fixes: dc10c23ab9a7 ("t/io_uring: show IOPS in increments of 1000 IOPS if necessary")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 months agoMerge branch 'evelu-onecore' of https://github.com/ErwanAliasr1/fio
Jens Axboe [Tue, 12 Oct 2021 19:50:54 +0000 (13:50 -0600)]
Merge branch 'evelu-onecore' of https://github.com/ErwanAliasr1/fio

* 'evelu-onecore' of https://github.com/ErwanAliasr1/fio:
  t/one-core-peak: Improving check_sysblock_value error handling

3 months agot/io_uring: show IOPS in increments of 1000 IOPS if necessary
Jens Axboe [Tue, 12 Oct 2021 19:48:45 +0000 (13:48 -0600)]
t/io_uring: show IOPS in increments of 1000 IOPS if necessary

It's a bit hard to read the millions of IOPS, so if we're above 100K
IOPS, scale by 1000 and add a K instead. This is easier to read:

IOPS=7235K, BW=3532MiB/s, IOS/call=31/31, inflight=(78 114)
IOPS=7218K, BW=3524MiB/s, IOS/call=32/32, inflight=(79 105)

Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 months agot/one-core-peak: Improving check_sysblock_value error handling
Erwan Velu [Tue, 12 Oct 2021 19:39:58 +0000 (21:39 +0200)]
t/one-core-peak: Improving check_sysblock_value error handling

The current code was reporting the following output:
cat: /sys/block/nvme0n1/queue/wbt_lat_usec: Argument invalide
nvme0n1: /sys/block/nvme0n1/queue/wbt_lat_usec set to 0.
Warning: nvme0n1: Cannot set 0 on /sys/block/nvme0n1/queue/wbt_lat_usec

This is problematic for several reasons:
- cat reports an error at reading wbt_lat_usec
- a message says it set wbt_lat_usec to 0
- a warning reports it cannot set wbt_lat_usec to 0

This commit:
- prevents the first error to be printed
- only report wbt_lat_usec is set to 0 if succeed unles it print the Warning message.

Signed-off-by: Erwan Velu <erwanaliasr1@gmail.com>
3 months agoMerge branch 'windows-res' of https://github.com/bjpaupor/fio
Jens Axboe [Tue, 12 Oct 2021 19:20:56 +0000 (13:20 -0600)]
Merge branch 'windows-res' of https://github.com/bjpaupor/fio

* 'windows-res' of https://github.com/bjpaupor/fio:
  Query Windows clock frequency and use reported max

3 months agoQuery Windows clock frequency and use reported max
Brandon Paupore [Tue, 12 Oct 2021 19:00:41 +0000 (14:00 -0500)]
Query Windows clock frequency and use reported max

Previously FIO used the Windows lower-bound clock frequency of 64 Hz for
its helper-thread. This caused IOPS/BW logs to have large drift between
timestamps when not using per-unit logging for those measurements.

Now query the current resolution and set to use the maximum for more
accurate timestamps. Note that the resolution is automatically restored
after FIO terminates.

Signed-off-by: Brandon Paupore <brandon.paupore@wdc.com>
3 months agoio_u: don't attempt to requeue for full residual
Jens Axboe [Mon, 11 Oct 2021 15:49:21 +0000 (09:49 -0600)]
io_u: don't attempt to requeue for full residual

If we get zero bytes transferred, then don't attempt to re-set the
io_u and requeue the IO. That's a fatal condition for this IO.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 months agot/io_uring: fix latency stats for depth == 1
Jens Axboe [Sat, 9 Oct 2021 18:56:11 +0000 (12:56 -0600)]
t/io_uring: fix latency stats for depth == 1

Two issues here:

- Stat increment accounting was off-by-one, causing no stats added
  for depth == 1
- The stat batch count should be a minimum of 2, since it's really
  a mask.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 months agoMerge branch 'evelu-ocp' of https://github.com/ErwanAliasr1/fio
Jens Axboe [Thu, 7 Oct 2021 12:18:21 +0000 (06:18 -0600)]
Merge branch 'evelu-ocp' of https://github.com/ErwanAliasr1/fio

* 'evelu-ocp' of https://github.com/ErwanAliasr1/fio:
  t/io_uring: Add -r option to control the runtime
  t/one-core-peak: Reporting RETPOLINE & PAGE_TABLE_ISOLATION
  t/one-core-peak: Reporting kernel cmdline
  t/one-core-peak: Reporting BLK_WBT_MQ
  t/one-core-peak: Reporting BLK_CGROUP

3 months agot/io_uring: Add -r option to control the runtime
Erwan Velu [Wed, 6 Oct 2021 21:40:27 +0000 (23:40 +0200)]
t/io_uring: Add -r option to control the runtime

By default the test is running until someone press Ctrl-C.
This commit add an option to define the expected runtime.

Signed-off-by: Erwan Velu <erwanaliasr1@gmail.com>
3 months agot/one-core-peak: Reporting RETPOLINE & PAGE_TABLE_ISOLATION
Erwan Velu [Wed, 6 Oct 2021 21:42:29 +0000 (23:42 +0200)]
t/one-core-peak: Reporting RETPOLINE & PAGE_TABLE_ISOLATION

These settings can influence the max perf if enabled.
Let's report them.

Signed-off-by: Erwan Velu <erwanaliasr1@gmail.com>
3 months agot/one-core-peak: Reporting kernel cmdline
Erwan Velu [Wed, 6 Oct 2021 21:25:49 +0000 (23:25 +0200)]
t/one-core-peak: Reporting kernel cmdline

The cmdline can contain many interesting options that were set and could
influence the final result/one-core-peak: Reporting kernel cmdline

The cmdline can contain many interesting options that were set and could
influence the final result

Signed-off-by: Erwan Velu <erwanaliasr1@gmail.com>
3 months agot/one-core-peak: Reporting BLK_WBT_MQ
Erwan Velu [Wed, 6 Oct 2021 21:02:15 +0000 (23:02 +0200)]
t/one-core-peak: Reporting BLK_WBT_MQ

If BLK_WBT_MQ is set, some ktime_get() call can be seen in the io path.
Let's report the value of this setting and disable it if present.

Signed-off-by: Erwan Velu <erwanaliasr1@gmail.com>
3 months agot/one-core-peak: Reporting BLK_CGROUP
Erwan Velu [Wed, 6 Oct 2021 20:19:30 +0000 (22:19 +0200)]
t/one-core-peak: Reporting BLK_CGROUP

When BLK_CGROUP is enabled, it induces some rdtsc calls which reduce the
overall performance.

Let's report if this option is enabled.

The tool was reporting BLK_CGROUP_IOCOST which wasn't the right one.

Signed-off-by: Erwan Velu <erwanaliasr1@gmail.com>
3 months agot/io_uring: get rid of old debug printfs
Jens Axboe [Tue, 5 Oct 2021 12:58:07 +0000 (06:58 -0600)]
t/io_uring: get rid of old debug printfs

We don't really care about the sq/cq ring pointers, that was something
I originally added as this test tool was the first one that I wrote to
bring up io_uring and help debug ring issues.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 months agot/io_uring: print submitter id with tid on startup
Jens Axboe [Tue, 5 Oct 2021 12:38:41 +0000 (06:38 -0600)]
t/io_uring: print submitter id with tid on startup

Makes it easier to match up multiple threads with the stats.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 months agot/io_uring: clean up aio wait loop
Jens Axboe [Mon, 4 Oct 2021 23:04:04 +0000 (17:04 -0600)]
t/io_uring: clean up aio wait loop

No functional changes, just makes it easier to read and gets rid of
an indentation.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 months agot/io_uring: check for valid clock_index and finish state for stats
Jens Axboe [Mon, 4 Oct 2021 22:35:15 +0000 (16:35 -0600)]
t/io_uring: check for valid clock_index and finish state for stats

If the clock_index isn't non-zero, it's not valid and we should disregard
the sample. Ditto if an exit signal has been sent, we're done at that
point and aren't interested in the last samples.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 months agot/io_uring: don't track IO latencies the first second of runtime
Jens Axboe [Mon, 4 Oct 2021 22:18:39 +0000 (16:18 -0600)]
t/io_uring: don't track IO latencies the first second of runtime

The most variation is usually seen at startup, so don't start tracking
latencies until we've done the first reporting run. Things should be
nice and stable at that point.

To make this cheaper on the fast path, clock_index is only valid if
it's non-zero. This makes checking for stats cheap in the reap path.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 months agot/io_uring: don't print partial IOPS etc output if exit signal was received
Jens Axboe [Mon, 4 Oct 2021 22:16:01 +0000 (16:16 -0600)]
t/io_uring: don't print partial IOPS etc output if exit signal was received

The run always terminates with what looks like a much slower cycle than
the previous seconds. That's not really the case, it's just that the
sleep() got interrupted by the signal and we slept less than we thought
we did, yet we still account it as a full second.

Just make it cleaner and break if finish is set.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 months agot/io_uring: add support for legacy AIO
Jens Axboe [Mon, 4 Oct 2021 18:42:01 +0000 (12:42 -0600)]
t/io_uring: add support for legacy AIO

Just as a comparison point, not really interesting otherwise. It doesn't
support any of the advanced features, just basic IO.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 months agot/io_uring: remove extra add_stat() call
Jens Axboe [Mon, 4 Oct 2021 18:33:40 +0000 (12:33 -0600)]
t/io_uring: remove extra add_stat() call

If we're batching the stat updates, it's incorrect to add the individual
stat. Would have skewed the percentiles, and make -t1 run slower than it
otherwise would have.

Fixes: ab85494f8bf0 ("t/io_uring: batch stat updates")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 months agoMerge branch 'evelu-fixes2' of https://github.com/ErwanAliasr1/fio
Jens Axboe [Fri, 1 Oct 2021 19:55:52 +0000 (13:55 -0600)]
Merge branch 'evelu-fixes2' of https://github.com/ErwanAliasr1/fio

* 'evelu-fixes2' of https://github.com/ErwanAliasr1/fio:
  t/one-core-peak: nvme-cli as optional tooling
  t/one-core-peak: Report numa as off if missing

3 months agot/one-core-peak: nvme-cli as optional tooling
Erwan Velu [Fri, 1 Oct 2021 19:43:07 +0000 (21:43 +0200)]
t/one-core-peak: nvme-cli as optional tooling

Not all systems has nvme-cli installed.

If present then let's print additional low-level info,
If not, let's ignore and continue.

Signed-off-by: Erwan Velu <erwanaliasr1@gmail.com>
3 months agot/one-core-peak: Report numa as off if missing
Erwan Velu [Fri, 1 Oct 2021 19:37:29 +0000 (21:37 +0200)]
t/one-core-peak: Report numa as off if missing

Some systems doesn't have numa enabled,
if so don't report an error but report numa as off.

Signed-off-by: Erwan Velu <erwanaliasr1@gmail.com>
3 months agoRefer td->loops instead of td->o.loops to fix loop count issue
Shin'ichiro Kawasaki [Fri, 1 Oct 2021 10:32:57 +0000 (19:32 +0900)]
Refer td->loops instead of td->o.loops to fix loop count issue

In the github issues #1093 and #1278, it was reported that the loops
option does not work as expected when do_verify=0 option is specified.
Per analysis by Sowmya Ravi, the cause was as follows:

1) keep_running() decrements td->o.loops at job repetition, then
   td->o.loops has zero value when the last loop is executed.
2) clear_io_state() is called at the beginning of the thread_main loop
   for each repetition for loops option.
3) clear_io_state() calls reset_io_counters() which resets
   td->nr_done_files to zero when td->o.loops is non-zero.
4) For the last loop of loops option, clear_io_state() call does not
   clear td->nr_done_files since td->l.loops is zero. This results in a
   setup error in do_io().

To fix the issue, modify reset_io_counters() to refer td->loops instead
of td->o.loops. td->o.loops is not a good reference since it is updated
in keep_running(). td->loops is not updated during fio run, and safe to
refer.

Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Link: https://lore.kernel.org/r/20211001103257.4130231-3-shinichiro.kawasaki@wdc.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 months agoRevert "Fix for loop count issue when do_verify=0 (#1093)"
Shin'ichiro Kawasaki [Fri, 1 Oct 2021 10:32:56 +0000 (19:32 +0900)]
Revert "Fix for loop count issue when do_verify=0 (#1093)"

This reverts commit 499cded5f435a0a7c379b606eb3e903d7f43c360.

The commit enabled clear_io_state() call in the loop of thread_main()
after completion of IOs, regardless of verify option. This sets zero to
td->nr_done_files even when the IOs are sequential workload with holes.
Such IOs depend on td->nr_done_files to judge job completion in
__get_next_file(). With zero value in td->nr_done_files, the sequential
IOs do not complete as expected, and results in failure of a test case

Revert the commit to avoid the failure. Regarding the loop count issue
with do_verify=0 option, another fix patch follows.

Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Link: https://lore.kernel.org/r/20211001103257.4130231-2-shinichiro.kawasaki@wdc.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 months agot/io_uring: correct percentile ranking
Jens Axboe [Fri, 1 Oct 2021 17:11:53 +0000 (11:11 -0600)]
t/io_uring: correct percentile ranking

Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 months agozbd: Fix unexpected job termination by open zone search failure
Shin'ichiro Kawasaki [Thu, 30 Sep 2021 00:02:36 +0000 (09:02 +0900)]
zbd: Fix unexpected job termination by open zone search failure

Test case #46 in t/zbd/test-zbd-support fails when it is repeated
hundreds of times on null_blk zoned devices. The test case uses libaio
IO engine to run 8 random write jobs on 4 sequential write required
zones. When all of the 4 zones get almost full but still open for
in-flight writes, the helper function zbd_convert_to_open_zone() fails
to get an opened zone for next write. This results in unexpected job
termination.

To avoid the unexpected job termination, retry the steps in
zbd_convert_to_open_zone(). Before retry, call io_u_quiesce() to ensure
that the in-flight writes get completed.

To prevent infinite loop by the retry, retry only when any IOs are
in-flight or in-flight IOs get completed. To check in-flight IO count of
all jobs, add a new helper function any_io_in_flight().

Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Reviewed-by: Niklas Cassel <niklas.cassel@wdc.com>
Reviewed-by: Dmitry Fomichev <dmitry.fomichev@wdc.com>
Link: https://lore.kernel.org/r/20210930000236.4116945-1-shinichiro.kawasaki@wdc.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 months agot/io_uring: store TSC rate in local file
Jens Axboe [Thu, 30 Sep 2021 02:15:45 +0000 (20:15 -0600)]
t/io_uring: store TSC rate in local file

Doesn't change on a single machine, so let's just cache the value instead
of requiring it to be specified every time. If we specify the rate, the
local data is updated. If we don't specify it, we check the file, and use
the rate in there if it exists.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 months agoMerge branch 'patch-1' of https://github.com/ravisowmya/fio
Jens Axboe [Wed, 29 Sep 2021 17:38:58 +0000 (11:38 -0600)]
Merge branch 'patch-1' of https://github.com/ravisowmya/fio

* 'patch-1' of https://github.com/ravisowmya/fio:
  Fix for loop count issue when do_verify=0 (#1093)

3 months agoFix for loop count issue when do_verify=0 (#1093)
ravisowmya [Tue, 28 Sep 2021 19:09:38 +0000 (12:09 -0700)]
Fix for loop count issue when do_verify=0 (#1093)

'clear_io_state' is called twice and resets the nr_done_files.
'clear_io_state' resets the nr_done_files if loop>=1.
This API is called twice with in thread_main and the second call is
skipped if do_verify=0. We rely on the first call for setup management.
So, for the very last loop, we would have skipped reseting
'nr_done_files' because loops=0 resulting in an IO error
in do_io and we exit without performing any IOs. Fix will invoke
the second call to clear_io_state

Signed-off-by: Sowmya Ravi sowmyaravi.92@gmail.com
3 months agoMerge branch 'sigbreak' of https://github.com/bjpaupor/fio
Jens Axboe [Tue, 28 Sep 2021 19:28:18 +0000 (13:28 -0600)]
Merge branch 'sigbreak' of https://github.com/bjpaupor/fio

* 'sigbreak' of https://github.com/bjpaupor/fio:
  add signal handlers for Windows SIGBREAK

3 months agoadd signal handlers for Windows SIGBREAK
Brandon Paupore [Tue, 28 Sep 2021 17:12:15 +0000 (12:12 -0500)]
add signal handlers for Windows SIGBREAK

Signed-off-by: Brandon Paupore <brandon.paupore@wdc.com>
3 months agoMerge branch 'onecore' of https://github.com/ByteHamster/fio
Jens Axboe [Sun, 26 Sep 2021 22:32:32 +0000 (16:32 -0600)]
Merge branch 'onecore' of https://github.com/ByteHamster/fio

* 'onecore' of https://github.com/ByteHamster/fio:
  Pick core for running t/one-core-peak.sh

3 months agoMerge branch 'evelu-fio' of https://github.com/ErwanAliasr1/fio
Jens Axboe [Sun, 26 Sep 2021 22:32:05 +0000 (16:32 -0600)]
Merge branch 'evelu-fio' of https://github.com/ErwanAliasr1/fio

* 'evelu-fio' of https://github.com/ErwanAliasr1/fio:
  one-core-peak: Reporting NVME features
  t/one-core-peak: Reporting kernel config
  one-core-peak.sh: Fixing bash

3 months agoone-core-peak: Reporting NVME features
Erwan Velu [Sun, 26 Sep 2021 20:26:27 +0000 (22:26 +0200)]
one-core-peak: Reporting NVME features

This commit get some low-level features of NVME drives and report them.
It includes, temperature, apste, power state and submission & completion queues

A typical output looks like :
  nvme0n1: MODEL=Samsung SSD 970 EVO Plus 2TB FW=2B2QEXM7 serial=S59CNM0R417706B PCI=0000:01:00.0@8.0 GT/s PCIe IRQ=62 NUMA=0 CPUS=0-31
  nvme0n1: Temp:34 C, Autonomous Power State Transition: Enabled, PowerState:4, Completion Queues:32, Submission Queues:32

Signed-off-by: Erwan Velu <erwanaliasr1@gmail.com>
3 months agot/one-core-peak: Reporting kernel config
Erwan Velu [Sun, 26 Sep 2021 19:43:39 +0000 (21:43 +0200)]
t/one-core-peak: Reporting kernel config

This patch add a reporting of some items of the kernel config.

A typical output looks like :
system: KERNEL: 5.15.0-rc2+
system: KERNEL: CONFIG_BLK_CGROUP_IOCOST=y
system: KERNEL: CONFIG_HZ=1000

Signed-off-by: Erwan Velu <erwanaliasr1@gmail.com>
3 months agoPick core for running t/one-core-peak.sh
ByteHamster [Wed, 22 Sep 2021 14:30:35 +0000 (16:30 +0200)]
Pick core for running t/one-core-peak.sh

3 months agoone-core-peak.sh: Fixing bash
Erwan Velu [Sun, 26 Sep 2021 19:03:36 +0000 (21:03 +0200)]
one-core-peak.sh: Fixing bash

This commit fixes some warning around the bash syntax

Signed-off-by: Erwan Velu <erwanaliasr1@gmail.com>
3 months agoMerge branch 'tsc' of https://github.com/ErwanAliasr1/fio
Jens Axboe [Sun, 26 Sep 2021 15:58:05 +0000 (09:58 -0600)]
Merge branch 'tsc' of https://github.com/ErwanAliasr1/fio

* 'tsc' of https://github.com/ErwanAliasr1/fio:
  one-core-peak: Adding option to reporting latencies
  one-core-peak: Avoid reporting Unknown memory speed

3 months agoone-core-peak: Adding option to reporting latencies
Erwan Velu [Sat, 25 Sep 2021 21:51:24 +0000 (23:51 +0200)]
one-core-peak: Adding option to reporting latencies

Since commit 932131c944b10f2a03f4028318c454c98eca489f,
it is now possible to report the io_uring benchmark latencies.

This patch detects the current TSC value and enable the latency feature if requested.

Signed-off-by: Erwan Velu <e.velu@criteo.com>
3 months agoone-core-peak: Avoid reporting Unknown memory speed
Erwan Velu [Sat, 25 Sep 2021 21:49:12 +0000 (23:49 +0200)]
one-core-peak: Avoid reporting Unknown memory speed

Some BIOSes, reports the configured mem speed to unknown making the report useless.
Adding a match on a real speed to avoid this.

Before: system: MEMORY: Unknown
After:  system: MEMORY: 3466 MT/s

Signed-off-by: Erwan Velu <e.velu@criteo.com>
3 months agoMerge branch 'evelu-uring' of https://github.com/ErwanAliasr1/fio
Jens Axboe [Sat, 25 Sep 2021 20:56:14 +0000 (14:56 -0600)]
Merge branch 'evelu-uring' of https://github.com/ErwanAliasr1/fio

* 'evelu-uring' of https://github.com/ErwanAliasr1/fio:
  t/io_uring.c: Adding \n on help

3 months agot/io_uring.c: Adding \n on help
Erwan Velu [Sat, 25 Sep 2021 20:45:51 +0000 (22:45 +0200)]
t/io_uring.c: Adding \n on help

Without these \n, the new options were baddly printed

Signed-off-by: Erwan Velu <e.velu@criteo.com>
3 months agot/io_uring: batch stat updates
Jens Axboe [Sat, 25 Sep 2021 20:38:10 +0000 (14:38 -0600)]
t/io_uring: batch stat updates

Track the last clock_index, and batch increments if at all possible.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 months agot/io_uring: add support for latency tracking
Jens Axboe [Sat, 25 Sep 2021 20:25:05 +0000 (14:25 -0600)]
t/io_uring: add support for latency tracking

This will display the latency percentiles for the run when done, per
submitter thread. It takes two arguments:

-t<x> Enable latency tracking if x is non-zero
-T<Y> Set TSC clock rate to Y Hz

The tsc rate can be programatically deduced (fio does this), for now
pass it in. dmesg will generally tell you:

tsc: Refined TSC clocksource calibration: 3699.889 MHz

and you'd then do:

-t1 -T3699889000

for that. Here's an example, synchronous optane gen2 read:

[...]
IOPS=254118, BW=124MiB/s, IOS/call=0/0, inflight=(1)
IOPS=255024, BW=124MiB/s, IOS/call=0/0, inflight=(1)
IOPS=255100, BW=124MiB/s, IOS/call=0/0, inflight=(1)
IOPS=254791, BW=124MiB/s, IOS/call=0/0, inflight=(1)
^CExiting on signal 2
IOPS=100086, BW=48MiB/s, IOS/call=1/1, inflight=(1)
515102: Latency percentiles:
    percentiles (nsec):
     |  1.0000th=[ 3857],  5.0000th=[ 3857], 10.0000th=[ 3857],
     | 20.0000th=[ 3857], 30.0000th=[ 3857], 40.0000th=[ 3892],
     | 50.0000th=[ 3892], 60.0000th=[ 3892], 70.0000th=[ 3892],
     | 80.0000th=[ 3892], 90.0000th=[ 3961], 95.0000th=[ 3961],
     | 99.9000th=[ 8752], 99.5000th=[ 8752], 99.9000th=[ 8752],
     | 99.9500th=[ 9064], 99.9900th=[ 9755]

Or a higher depth run:

IOPS=3549568, BW=1733MiB/s, IOS/call=32/32, inflight=(64)
IOPS=3547712, BW=1732MiB/s, IOS/call=32/31, inflight=(111)
IOPS=3549504, BW=1733MiB/s, IOS/call=32/31, inflight=(128)
^CExiting on signal 2
IOPS=1413600, BW=690MiB/s, IOS/call=32/32, inflight=(35)
515078: Latency percentiles:
    percentiles (nsec):
     |  1.0000th=[13630],  5.0000th=[14322], 10.0000th=[15291],
     | 20.0000th=[16121], 30.0000th=[20065], 40.0000th=[21726],
     | 50.0000th=[22279], 60.0000th=[26154], 70.0000th=[27814],
     | 80.0000th=[28368], 90.0000th=[33903], 95.0000th=[34180],
     | 99.9000th=[52862], 99.5000th=[52862], 99.9000th=[52862],
     | 99.9500th=[56183], 99.9900th=[67807]

Note that latency tracking isn't cheap, even if we tried to do it in the
cheapest way possible. The peak workload shown here will run at ~3.7M
IOPS without tracking, and as shown about 3.55M with tracking enabled.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 months agot/io_uring: don't print BW numbers for do_nop
Jens Axboe [Fri, 24 Sep 2021 21:17:44 +0000 (15:17 -0600)]
t/io_uring: don't print BW numbers for do_nop

They don't mean anything for nops, we're just interested in IOPS here.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
3 months agot/io_uring: ensure batch counts are smaller or equal to depth
Jens Axboe [Thu, 23 Sep 2021 15:15:16 +0000 (09:15 -0600)]
t/io_uring: ensure batch counts are smaller or equal to depth

If you use a batch submit or complete count that's larger than the
depth, then t/io_uring will stall. Make sure to sanitize the counts
so that any batch values is always <= total depth.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 months agoMerge branch 'one-core' of https://github.com/ErwanAliasr1/fio
Jens Axboe [Tue, 21 Sep 2021 00:29:40 +0000 (18:29 -0600)]
Merge branch 'one-core' of https://github.com/ErwanAliasr1/fio

* 'one-core' of https://github.com/ErwanAliasr1/fio:
  t/one-core.sh: Adding script to run the one-core io benchmark

4 months agot/one-core.sh: Adding script to run the one-core io benchmark
Erwan Velu [Thu, 16 Sep 2021 20:52:22 +0000 (22:52 +0200)]
t/one-core.sh: Adding script to run the one-core io benchmark

Associated to fio, the t/io_uring test is used to compute the max IOPS a
single core can get.

Jens published several times the procedure he uses, but trying to
reproduce this setup is error-prone. It's easy to miss a configuration
and get a different result.

This script is about setting up a common setup to reproduce these runs.

From the fio directory, execute like the folliowing :

[user@fio] t/one-core.sh /dev/nvme0n1 [other drives]
        ##################################################:
system: CPU: AMD EPYC 7502P 32-Core Processor
system: MEMORY: 2933 MT/s
system: KERNEL: 5.10.35-1.el8.x86_64
nvme0n1: MODEL=Samsung SSD 970 EVO Plus 2TB FW=2B2QEXM7 serial=S59CNM0R417706B PCI=0000:01:00.0@8.0 GT/s PCIe IRQ=64 NUMA=0 CPUS=0-23
nvme0n1: set none as io scheduler
nvme0n1: iostats set to 1.
nvme0n1: nomerge set to 0.
Warning: For better performance, you should enable nvme poll queues by setting nvme.poll_queues=32 on the kernel commande line
        ##################################################:

io_uring: Running taskset -c 0,12 t/io_uring -b512 -d128 -c32 -s32 -p1 -F1 -B1 -n4 /dev/nvme0n1
        [...]
IOPS=731008, BWPS=356 MB IOS/call=32/31, inflight=(108 127 126 106)

This script will take care of the following items:
- nvme poll queues
- io scheduler
- iostats
- io_poll
- nomerge
- finding the logical cores running on the first physical core
- cpu frequency governor on performance
- cpu idle governor on menu
- calling t/io_uring with the proper parameters in 512 bytes fashion
- reporting the nvme & pci configuration

Signed-off-by: Erwan Velu <erwanaliasr1@gmail.com>
4 months agot/io_uring: fix bandwidth calculation
Jens Axboe [Thu, 16 Sep 2021 17:41:06 +0000 (11:41 -0600)]
t/io_uring: fix bandwidth calculation

Fixes: 22fd35012cea ("t/io_uring: Reporting bandwidth")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 months agoMerge branch 'bwps' of https://github.com/ErwanAliasr1/fio
Jens Axboe [Thu, 16 Sep 2021 17:29:57 +0000 (11:29 -0600)]
Merge branch 'bwps' of https://github.com/ErwanAliasr1/fio

* 'bwps' of https://github.com/ErwanAliasr1/fio:
  t/io_uring: Reporting bandwidth

4 months agot/io_uring: Reporting bandwidth
Erwan Velu [Thu, 16 Sep 2021 16:46:30 +0000 (18:46 +0200)]
t/io_uring: Reporting bandwidth

When performing tests at various block size, it's sometimes a bit
difficult to estimate if we reach the limit of the datapath.

This commit offer to simply prints the resulting bandwitdh of the IOPS
multiplied by the block size.

A typical output looks like :

  [user@hosŧ] t/io_uring -b512 -d128 -c32 -s32 -p1 -F1 -B1 -n4 /dev/nvme0n1
   ...
   IOPS=729856, BW=356 MiB/s, IOS/call=32/32, inflight=(105 119 108 109)

  [user@host] t/io_uring -b4096 -d128 -c32 -s32 -p1 -F1 -B1 -n4 /dev/nvme0n1
   ...
   IOPS=746368, BW=2915 MiB/s, IOS/call=32/31, inflight=(121 115 122 122)

In the 4K case, as for a PCI Gen3 product, we are clearly limited by the
bandwidth while in the 512 case we hit latency issues.

BW is expressed in MiB/sec.

Signed-off-by: Erwan Velu <e.velu@criteo.com>
4 months agot/io_uring: add switch -O for O_DIRECT vs buffered
Jens Axboe [Wed, 15 Sep 2021 12:51:01 +0000 (06:51 -0600)]
t/io_uring: add switch -O for O_DIRECT vs buffered

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 months agozbd: remove dead zone retrieval call
Jens Axboe [Mon, 13 Sep 2021 20:09:01 +0000 (14:09 -0600)]
zbd: remove dead zone retrieval call

A previous commit missed to realize that not only was the assignment
useless, that also made the very call to zbd_zone_nr() useless as
well. Remove it.

Fixes: 000ecb5fe36d ("zbd: Removing useless variable assignment")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 months agot/io_uring: add -N option for do_nop
Jens Axboe [Mon, 13 Sep 2021 19:18:26 +0000 (13:18 -0600)]
t/io_uring: add -N option for do_nop

Makes it easier than asking people to edit and compile.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 months agot/io_uring: don't require a file for do_nop runs
Jens Axboe [Mon, 13 Sep 2021 19:14:29 +0000 (13:14 -0600)]
t/io_uring: don't require a file for do_nop runs

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 months agoMerge branch 'ft' of https://github.com/ErwanAliasr1/fio
Jens Axboe [Wed, 8 Sep 2021 21:40:47 +0000 (15:40 -0600)]
Merge branch 'ft' of https://github.com/ErwanAliasr1/fio

* 'ft' of https://github.com/ErwanAliasr1/fio:
  log: Removing useless assignment
  zbd: Removing useless variable assignment
  lib/fls.h: Remove unused variable assignment
  engines/sg: Removing useless variable assignment
  stat: Avoid freeing null pointer
  filesetup: Removing unused variable usage
  engines/sg: Return error if generic_close_file fails

4 months agolog: Removing useless assignment
Erwan Velu [Wed, 8 Sep 2021 21:10:50 +0000 (23:10 +0200)]
log: Removing useless assignment

The last len assigment is never read which makes it useless.

Signed-off-by: Erwan Velu <erwanaliasr1@gmail.com>
4 months agozbd: Removing useless variable assignment
Erwan Velu [Wed, 8 Sep 2021 21:00:45 +0000 (23:00 +0200)]
zbd: Removing useless variable assignment

zone_idx_b is set but never read again.

Signed-off-by: Erwan Velu <erwanaliasr1@gmail.com>
4 months agolib/fls.h: Remove unused variable assignment
Erwan Velu [Wed, 8 Sep 2021 20:52:10 +0000 (22:52 +0200)]
lib/fls.h: Remove unused variable assignment

x is modified just before the last set of r but x is never used again.
Let's remove this useless assignment.

Signed-off-by: Erwan Velu <erwanaliasr1@gmail.com>
4 months agoengines/sg: Removing useless variable assignment
Erwan Velu [Wed, 8 Sep 2021 20:43:39 +0000 (22:43 +0200)]
engines/sg: Removing useless variable assignment

ret is set to -1 but the break statement will not use this value.

So let's remove this useless assignment which could be confusing.

Signed-off-by: Erwan Velu <erwanaliasr1@gmail.com>
4 months agostat: Avoid freeing null pointer
Erwan Velu [Wed, 8 Sep 2021 20:35:59 +0000 (22:35 +0200)]
stat: Avoid freeing null pointer

If ovals is NULL, the jump to out will free(ovals) and will trigger an error.
As the out label was only used for this condition, let's remove it and return immediately.
As out was also used as a variable name, this makes the function easier
to read and more robust.

Signed-off-by: Erwan Velu <erwanaliasr1@gmail.com>
4 months agoREADME: add link to new lore archive
Jens Axboe [Wed, 8 Sep 2021 20:31:04 +0000 (14:31 -0600)]
README: add link to new lore archive

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 months agofilesetup: Removing unused variable usage
Erwan Velu [Wed, 8 Sep 2021 20:22:56 +0000 (22:22 +0200)]
filesetup: Removing unused variable usage

done is set to true but this is useless as break will
stop the while loop.

So let's remove this useless assignment.

Signed-off-by: Erwan Velu <erwanaliasr1@gmail.com>
4 months agoengines/sg: Return error if generic_close_file fails
Erwan Velu [Wed, 8 Sep 2021 20:18:53 +0000 (22:18 +0200)]
engines/sg: Return error if generic_close_file fails

The current code was returning 1 if generic_close_file() fails.
The ret value was prepared with the real error, let's return this one as
the per generic_open_file() error handling.

Signed-off-by: Erwan Velu <erwanaliasr1@gmail.com>
4 months agot/io_uring: ensure that nthreads is > 0
Jens Axboe [Wed, 8 Sep 2021 20:12:16 +0000 (14:12 -0600)]
t/io_uring: ensure that nthreads is > 0

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 months agot/io_uring: allow flexible IO threads assignment
Andrzej Jakowski [Wed, 8 Sep 2021 18:35:00 +0000 (11:35 -0700)]
t/io_uring: allow flexible IO threads assignment

This patch allows to flexibly assign IO threads to fileset. When
you specify:

t/io_uring -n 5 /dev/dev1 dev/dev2

First file/device will get 3 IO threads and second file/device
remaining 2 IO threads. When there is more files then IO threads,
IO thread may get assigned multiple files/devices.

Signed-off-by: Andrzej Jakowski <andrzej.jakowski@intel.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 months agoFio 3.28 fio-3.28
Jens Axboe [Wed, 8 Sep 2021 14:59:48 +0000 (08:59 -0600)]
Fio 3.28

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 months agot/io_uring: don't make setrlimit() failing fatal
Jens Axboe [Wed, 8 Sep 2021 14:07:57 +0000 (08:07 -0600)]
t/io_uring: don't make setrlimit() failing fatal

We don't even need this on newer kernels, so just ignore it if it
fails. The worst that can happen is that buffer registration will
fail.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 months agot/io_uring: fixes in output
Andrzej Jakowski [Wed, 8 Sep 2021 04:10:44 +0000 (21:10 -0700)]
t/io_uring: fixes in output

Provide description of available options in usage command
and fix alignment so they look pretty.

Also remove debug output.

Signed-off-by: Andrzej Jakowski <andrzej.jakowski@intel.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 months agooptions: Add thinktime_iotime option
Shin'ichiro Kawasaki [Mon, 6 Sep 2021 01:50:00 +0000 (10:50 +0900)]
options: Add thinktime_iotime option

The thinktime option allows stalling a job for a specified amount of
time. Using the thinktime_blocks option, periodic stalls can be added
every thinktime_blocks IOs. However, with this option, the periodic
stall may not be repeated at equal time intervals as the time to execute
thinktime_blocks IOs may vary.

To control the thinktime interval by time, introduce the option
thinktime_iotime. With this new option, the thinktime stall is repeated
after IOs are executed for thinktime_iotime. If this option is used
together with the thinktime_blocks option, the thinktime pause is
repeated after thinktime_iotime or after thinktime_blocks IOs, whichever
happens first.

To support the new option, add a new member thinktime_iotime in the
struct thread_options and the struct thread_options_pack. Avoid size
increase of the struct thread_options_pack by replacing a padding 'pad5'
with the new member. To keep thinktime related members close, move the
members near the position where the padding was placed. Make same
changes to the struct thread_option also for consistency.

To track the time and IO block count at the last stall, add
last_thinktime variable and last_thinktime_blocks variable to struct
thread_data. Also, introduce the helper function init_thinktime()
to group thinktime related preparations.

Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 months agoexamples: add examples for cmdprio_* IO priority options
Damien Le Moal [Fri, 3 Sep 2021 15:20:27 +0000 (15:20 +0000)]
examples: add examples for cmdprio_* IO priority options

Add the example scripts cmdprio-percentage.fio and cmdprio-bssplit.fio
to illustrate the use of the cmdprio_percentage, cmdprio_class,
cmdprio and cmdprio_bssplit options. Also add the fiograph output
images for these example scripts.

Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 months agofio: Introduce the log_prio option
Damien Le Moal [Fri, 3 Sep 2021 15:20:26 +0000 (15:20 +0000)]
fio: Introduce the log_prio option

Introduce the log_prio option to expand priority logging from just a
single bit information (priority high vs low) to the full value of the
priority value used to execute IOs. When this option is set, the
priority value is printed as a 16-bits hexadecimal value combining
the I/O priority class and priority level as defined by the
ioprio_value() helper.

Similarly to the log_offset option, this option does not result in
actual I/O priority logging when log_avg_msec is set.

This patch also fixes a problem with the IO_U_F_PRIORITY flag, namely
that this flag is used to indicate that the IO is being executed with a
high priority on the device while at the same time indicating how to
account for the IO completion latency (high_prio clat vs low_prio clat).
With the introduction of the cmdprio_class and cmdprio options, these
assumptions are not necesarilly compatible anymore.

These problems are addressed as follows:
* The priority_bit field of struct iosample is replaced with the
  16-bits priority field representing the full io_u->ioprio value. When
  log_prio is set, the priority field value is logged as is. When
  log_prio is not set, 1 is logged as the entry's priority field if the
  sample priority class is IOPRIO_CLASS_RT, and 0 otherwise.
* IO_U_F_PRIORITY is renamed to IO_U_F_HIGH_PRIO to indicate that a job
  IO has the highest priority within the job context and so must be
  accounted as such using high_prio clat.

While fio final statistics only show accounting of high vs low IO
completion latency statistics, the log_prio option allows a user to
perform more detailed statistical analysis of a workload using
multiple different IO priorities.

Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 months agolibaio,io_uring: relax cmdprio_percentage constraints
Damien Le Moal [Fri, 3 Sep 2021 15:20:26 +0000 (15:20 +0000)]
libaio,io_uring: relax cmdprio_percentage constraints

In fio, a job IO priority is controlled with the prioclass and prio
options and these options cannot be used together with the
cmdprio_percentage option.

Allow a user to have async IO priorities default to the job defined
IO priority by removing the mutual exclusion between the options
cmdprio_percentage and prioclass/prio.

With the introduction of the cmdprio_class option, an async IO priority
may be lower than the job default priority, resulting in reversed clat
statistics showed for high and low priority IOs when fio completes.
Solve this by setting an io_u IO_U_F_PRIORITY flag depending on a
comparison between the async IO priority and job default IO priority.

When an async IO is issued without a priority set, Linux kernel will
execute it using the IO priority of the issuing context set with
ioprio_set(). This works fine for libaio, where the context will be
the same as the context that submitted the IO.

However, io_uring can be used with a kernel thread that performs
block device IO submissions (sqthread_poll). Therefore, for io_uring,
an IO sqe ioprio field must be set to the job default priority unless
the IO priority is set according to the job cmdprio_percentage value.

Because of this, IO uring already did set sqe->ioprio even when only
prio/prioclass was used. See commit b7ed2a862dda ("io_uring: set sqe
iopriority, if prio/prioclass is set"). In order to make the code easier
to maintain, handle all I/O priority preparations in the same function.

Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 months agolibaio,io_uring: introduce cmdprio_bssplit
Damien Le Moal [Fri, 3 Sep 2021 15:20:25 +0000 (15:20 +0000)]
libaio,io_uring: introduce cmdprio_bssplit

The cmdprio_percentage, cmdprio_class and cmdprio options allow
specifying different values for read and write operations. This enables
various IO priority issuing patterns even uner a mixed read-write
workload but does not allow differentiation within read and write
I/O operation types with different sizes when the bssplit option is
used.

Introduce the cmdprio_bssplit option to complement the use of the
bssplit option.  This new option has the same format as the bssplit
option, but the percentage values indicate the percentage of I/O
operations with a particular block size that must be issued with the
priority class and value specified by cmdprio_class and cmdprio.

Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 months agolibaio,io_uring: introduce cmdprio_class and cmdprio options
Damien Le Moal [Fri, 3 Sep 2021 15:20:24 +0000 (15:20 +0000)]
libaio,io_uring: introduce cmdprio_class and cmdprio options

When the cmdprio_percentage option is used, the specified percentage of
IO will be issued with the highest priority class IOPRIO_CLASS_RT. This
priority class maps to the ATA NCQ "high" priority level and allows
exercising a SATA device to measure its command latency characteristics
in the presence of low and high priority commands.

Beside ATA NCQ commands, Linux block IO schedulers also support IO
priorities and will behave differently in the presence of IOs with
different IO priority classes and values. However, cmdprio_percentage
does not allow specifying all possible priority classes and values.

To solve this, introduce libaio and io_uring engine specific options
cmdprio_class and cmdprio. These new options are the equivalent
of the prioclass and prio options and allow specifying the priority
class and priority value to use for asynchronous I/Os when the
cmdprio_percentage option is used. If not specified, the I/O priority
class defaults to IOPRIO_CLASS_RT and the I/O priority value to 0,
as before. Similarly to the cmdprio_percentage option, these options
can specify different values for read and write I/Os using a comma
separated list.

The manpage, HOWTO and fiograph configuration file are updated to
document these new options.

Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 months agolibaio,io_uring: improve cmdprio_percentage option
Damien Le Moal [Fri, 3 Sep 2021 15:20:24 +0000 (15:20 +0000)]
libaio,io_uring: improve cmdprio_percentage option

The cmdprio_percentage option of the libaio and io_uring engines defines
a single percentage that applies to all IO operations, regardless of
their direction. This prevents defining different high priority IO
percentages for reads and writes operations. This differentiation can
however be useful in the case of a mixed read-write workload (rwmixread
and rwmixwrite options).

Change the option definition to allow specifying a comma separated list
of percentages, 2 at most, one for reads and one for writes. If only a
single percentage is defined, it applies to both reads and writes as
before. The cmdprio_percentage option becomes an array of DDIR_RWDIR_CNT
elements indexed with enum fio_ddir values. The last entry of the array
(for DDIR_TRIM) is always 0.

Also create a new cmdprio helper file, engines/cmdprio.h,
such that we can avoid code duplication between io_uring and libaio
io engines. This helper file will be extended in subsequent patches.

Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
4 months agooptions: make parsing functions available to ioengines
Damien Le Moal [Fri, 3 Sep 2021 15:20:23 +0000 (15:20 +0000)]
options: make parsing functions available to ioengines

Move the declaration of split_parse_ddir(), str_split_parse() and
the split_parse_fn typedef to thread_options.h so that IO engines
can use these functions to parse options. The definition of struct
split is also moved to thread_options.h from options.c.

The type of the split_parse_fn callback function is changed to add a
void * argument that can be used for an option parsing callback to pass
a private data pointer to the split_parse_fn function. This can be used
by an IO engine to pass a pointer to its engine specific option
structure as td->eo is not yet set when options are being parsed.

Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>