Re: [RFC PATCH] tests/tcg: drop -cpu max from s390x sha512-mvx invocation

[Thomas Huth <thuth@redhat.com>]

On 10/03/2022 09.36, Thomas Huth wrote:
> On 09/03/2022 12.22, Alex Bennée wrote:
>> With -cpu max we get a warning:
>>
>>    qemu-s390x: warning: 'msa5-base' requires 'kimd-sha-512'.
>>
>> But dropping the -cpu max and it still runs fine.
>>
>> Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
>> Cc: Thomas Huth <thuth@redhat.com>
>> ---
>>   tests/tcg/s390x/Makefile.target | 2 --
>>   1 file changed, 2 deletions(-)
>>
>> diff --git a/tests/tcg/s390x/Makefile.target 
>> b/tests/tcg/s390x/Makefile.target
>> index 257c568c58..7aa502a557 100644
>> --- a/tests/tcg/s390x/Makefile.target
>> +++ b/tests/tcg/s390x/Makefile.target
>> @@ -34,6 +34,4 @@ sha512-mvx: CFLAGS=-march=z13 -mvx -O3
>>   sha512-mvx: sha512.c
>>       $(CC) $(CFLAGS) $(EXTRA_CFLAGS) $< -o $@ $(LDFLAGS)
>> -run-sha512-mvx: QEMU_OPTS+=-cpu max
>> -
>>   TESTS+=sha512-mvx
> 
> Ok, this helps to get rid of the warnings, thus feel free to add:
> 
> Tested-by: Thomas Huth <thuth@redhat.com>
> 
> However, the error still persists. I now had a closer look by running the 
> test with "qemu-s390x" on my z15 machine directly, and all tests are 
> failing! The problem happens with both, gcc v11.2.1 and clang v13.0.1. The 
> problem goes away (i.e. test works fine) if I compile the code with -O1 
> instead of -O3, or if I use GCC v8.5 instead. I'll try to find out more, but 
> as a temporary workaround, we could also switch to -O1 instead of -O3.

I noticed that the problem does not occur when running the test natively
on my s390x box, or via qemu-s390x on my x86 laptop, or when running it
via qemu-s390x v6.1.0 on my s390x box. So it's something wrong in the TCG
backend, I think, and I was able to bisect the issue down to this commit
here:

$ git bisect bad
9bca986df88b8ea46b100e3d21cc9e653c83e0b3 is the first bad commit
commit 9bca986df88b8ea46b100e3d21cc9e653c83e0b3
Author: Richard Henderson <richard.henderson@linaro.org>
Date:   Mon Sep 14 20:36:36 2020 -0700

     tcg/s390x: Implement TCG_TARGET_HAS_bitsel_vec

Richard, could you please have a look? I've uploaded my statically linked
binary that shows the problem here:

  http://people.redhat.com/~thuth/data/sha512

When running on a z15 s390x host with current QEMU master, the test fails:

$ ./qemu-s390x -cpu max ~/sha512
qemu-s390x: warning: 'msa5-base' requires 'kimd-sha-512'.
qemu-s390x: warning: 'msa5-base' requires 'klmd-sha-512'.
1..1
h    : 0000:  38 31 a6 a6  15 5e 50 9d  ee 59 a7 f4  51 eb 35 32 81...^P..Y..Q.52
h    : 0010:  4d 8f 8f 2d  f6 e3 70 88  94 74 0f 98  fd ee 23 88 M..-..p..t....#.
h    : 0020:  9f 4d e5 ad  b0 c5 01 0d  fb 55 5c da  77 c8 ab 5d .M.......U\.w..]
h    : 0030:  c9 02 09 4c  52 de 32 78  f3 5a 75 eb  c2 5f 09 3a ...LR.2x.Zu.._.:
got  : 0000:  7b 47 7b 48  1a 49 1a 49  79 48 7d 7f  7d 47 79 7d {G{H.I.IyH}.}Gy}
got  : 0010:  1a 7f 7d 40  1a 48 19 4b  7d 48 1a 7b  7b 48 7b 49 ..}@.H.K}H.{{H{I
got  : 0020:  7f 7d 7f 7f  7f 7f 7b 7d  19 49 1a 48  7f 47 7f 47 .}....{}.I.H.G.G
got  : 0030:  79 4b 7f 4b  77 7f 79 47  19 7d 1a 7f  7b 48 7f 47 yK.Kw.yG.}..{H.G
got  : 0040:  79 7f 7f 7d  1a 48 1a 7d  19 47 1c 48  77 48 77 7d y..}.H.}.G.HwHw}
got  : 0050:  19 7b 7d 48  7d 7f 19 7b  7f 48 1c 47  1a 7b 7d 7d .{}H}..{.H.G.{}}
got  : 0060:  1c 40 77 49  77 40 7f 7f  7d 49 19 7f  7b 49 7f 47 .@wIw@..}I..{I.G
got  : 0070:  19 48 7d 7b  7f 48 1a 7b  1c 49 7d 7f  77 40 7b 7b .H}{.H.{.I}.w@{{
got  : 0080:  00                                                 .
not ok 1 - do_test(&tests[0])
#     Failed test (sha512.c:main() at line 1046)
# Looks like you failed 1 tests of 1.

When running on my x86 laptop, the test succeeds:

$ ./qemu-s390x -cpu max ~/sha512
qemu-s390x: warning: 'msa5-base' requires 'kimd-sha-512'.
qemu-s390x: warning: 'msa5-base' requires 'klmd-sha-512'.
1..1
h    : 0000:  38 31 a6 a6  15 5e 50 9d  ee 59 a7 f4  51 eb 35 32 81...^P..Y..Q.52
h    : 0010:  4d 8f 8f 2d  f6 e3 70 88  94 74 0f 98  fd ee 23 88 M..-..p..t....#.
h    : 0020:  9f 4d e5 ad  b0 c5 01 0d  fb 55 5c da  77 c8 ab 5d .M.......U\.w..]
h    : 0030:  c9 02 09 4c  52 de 32 78  f3 5a 75 eb  c2 5f 09 3a ...LR.2x.Zu.._.:
got  : 0000:  33 38 33 31  61 36 61 36  31 35 35 65  35 30 39 64 3831a6a6155e509d
got  : 0010:  65 65 35 39  61 37 66 34  35 31 65 62  33 35 33 32 ee59a7f451eb3532
got  : 0020:  34 64 38 66  38 66 32 64  66 36 65 33  37 30 38 38 4d8f8f2df6e37088
got  : 0030:  39 34 37 34  30 66 39 38  66 64 65 65  32 33 38 38 94740f98fdee2388
got  : 0040:  39 66 34 64  65 35 61 64  62 30 63 35  30 31 30 64 9f4de5adb0c5010d
got  : 0050:  66 62 35 35  35 63 64 61  37 37 63 38  61 62 35 64 fb555cda77c8ab5d
got  : 0060:  63 39 30 32  30 39 34 63  35 32 64 65  33 32 37 38 c902094c52de3278
got  : 0070:  66 33 35 61  37 35 65 62  63 32 35 66  30 39 33 61 f35a75ebc25f093a
got  : 0080:  00                                                 .
ok 1 - do_test(&tests[0])

BTW, we recently also saw some issues with the migration qtest on a s390x
host, maybe that's related:

  https://app.travis-ci.com/gitlab/qemu-project/qemu/jobs/562674331#L7834
  https://app.travis-ci.com/github/huth/qemu/jobs/562290832#L7783

?

  Thomas