[LTP] 答复: [PATCH] Reduce loop count to meet need of low performance terminals

[Cyril Hrubis]

Hi!
> Are you sure that your kernel isn't affected by the bug the test checks for? The test timeouts on a buggy kernel.
> -- Is there any place where I can see more relevant information? which kind of bug you have found?

It's right in the top level comment in the testcase source where the
test description is.

> the test timeout on myside where there still 640000 loop,
> ZZL pid = 29466 this is while 646533 times loop = 646533 -------
> ZZL pid = 29466 this is while 646532 times loop = 646532 -------
> ZZL pid = 29466 this is while 646531 times loop = 646531 -------
> ZZL pid = 29466 this is while 646530 times loop = 646530 -------
> ZZL pid = 29466 this is while 646529 times loop = 646529 -------
> ZZL pid = 29466 this is while 646528 times loop = 646528 -------
> ZZL pid = 29466 this is while 646527 times loop = 646527 -------
> ZZL pid = 29466 this is while 646526 times loop = 646526 -------
> ZZL pid = 29466 this is while 646525 times loop = 646525 -------
> ZZL pid = 29466 this is while 646524 times loop = 646524 -------
> ZZL pid = 29466 this is while 646523 times loop = 646523 -------
> ZZL pid = 29466 this is while 646522 times loop = 646522 -------
> ZZL pid = 29466 this is while 646521 times loop = 646521 -------
> Test timeouted, sending SIGKILL!
> external/ltp/lib/tst_test.c:1641: TINFO: Killed the leftover descendant processes
> external/ltp/lib/tst_test.c:1648: TINFO: If you are running on slow machine, try exporting LTP_TIMEOUT_MUL > 1
> external/ltp/lib/tst_test.c:1649: TBROK: Test killed! (timeout?)
> 
> Summary:
> passed   1
> failed   0
> broken   1
> skipped  0
> warnings 0
> 
> 
> How long does the test run if you disable timeouts (by setting static int timeout = -1 in the source)?
> -- about 370 seconds

Looking at this it seems that your CPU is just a bit slower, if the test
fails it's supposed to be at least one order of magnitude slower.

> Also what kind of system it is? How fast is the CPU?
> --android 14 ,  kenrel 6.1.43-android14-11-o-g9bbfe9009b86 ,
> 
> cat /proc/cpuinfo
> processor       : 0
> BogoMIPS        : 38.40
> Features        : fp asimd evtstrm aes pmull sha1 sha2 crc32 atomics fphp asimdhp cpuid asimdrdm lrcpc dcpop asimddp
> CPU implementer : 0x41
> CPU architecture: 8
> CPU variant     : 0x2
> CPU part        : 0xd05  （ARM_CPU_PART_CORTEX_A55)
> CPU revision    : 0

So this is single core ARM cortex A55, it really seems to be just a case
of slower procesor, so it does make sense to lower the number of
iterations or increase the test timeout.

However the problem is that if we divide the default number of
iterations by two and keep the timeout the test would stil pass even on
broken kernel on an x86_64 desktop. So we probably need a callibration
loop that would be able to estimate the CPU speed to set the
expectations right.

-- 
Cyril Hrubis
chrubis@suse.cz