[PATCH] generic/095: don't use 1k blocksize for DIO operations

[PATCH] generic/095: don't use 1k blocksize for DIO operations

[Luís Henriques]

This test fails when using zram because it's using bs=1k for DIO in fio
jobs.  Fix the test by using _min_dio_alignment instead to get the correct
blocksize.

Signed-off-by: Luís Henriques <lhenriques@suse.de>
---
 tests/generic/095 | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/tests/generic/095 b/tests/generic/095
index c4693917b3e1..80448200bb86 100755
--- a/tests/generic/095
+++ b/tests/generic/095
@@ -23,6 +23,7 @@ iodepth_batch=$((8 * LOAD_FACTOR))
 numjobs=$((5 * LOAD_FACTOR))
 fio_config=$tmp.fio
 fio_out=$tmp.fio.out
+blksz=$(_min_dio_alignment $SCRATCH_DEV)
 cat >$fio_config <<EOF
 [global]
 bs=8k
@@ -34,7 +35,7 @@ directory=$SCRATCH_MNT
 numjobs=$numjobs
 [job1]
 ioengine=sync
-bs=1k
+bs=$blksz
 direct=1
 rw=randread
 filename=file1:file2
@@ -44,7 +45,7 @@ rw=randwrite
 direct=1
 filename=file1:file2
 [job3]
-bs=1k
+bs=$blksz
 ioengine=posixaio
 rw=randwrite
 direct=1

Re: [PATCH] generic/095: don't use 1k blocksize for DIO operations

[Theodore Ts'o]

On Mon, Oct 11, 2021 at 02:55:04PM +0100, Luís Henriques wrote:
> This test fails when using zram because it's using bs=1k for DIO in fio
> jobs.  Fix the test by using _min_dio_alignment instead to get the correct
> blocksize.
> 
> Signed-off-by: Luís Henriques <lhenriques@suse.de>

I'd suggest using the commit summary:

generic/095: align DIO read/writes with the supported sector size for the device

It might also be worth noting that generic/095 would have been failing
not just for zram, but if run on other hardware devices that might
have a logical sector size > 1k --- e.g., advanced format HDD's with
4k sectors, DASD's on mainframes with 2k sectors, etc.

Note: this change will cause the fio stress workload to do its DIO
operations using a block size of 512, which I think is fine for this
test.  (There might be other tests which are making assumptions about
the supported DIO alignment of the scratch device where adjusting the
test might be more complicated; fortunately, this doesn't appear to be
one of them.)

Other than changes to the commit description, feel free to add:

Reviewed-by: Theodore Ts'o <tytso@mit.edu>

Thanks,

					- Ted