Hard coded "mm01 mmap001 -m 10000" in runtest/mm causes test failure

[ywangwrs]

In runtest/mm, mm01 is defined as "mmap001 -m 10000", here the pages is set to 10000.
Normally, PAGE_SIZE = 4KB, so 40MB of mmap will be tested, but if a kernel has set PAGE_SIZE to 64KB, 10000 x 64KB = 640MB, this may exceed available memory in the system and cause failures.
For example: in a aarch64 system:
<<<test_start>>>
tag=mm01 stime=1565786038
cmdline="mmap001 -m 10000"
contacts=""
analysis=exit
<<<test_output>>>
mmap001 0 TINFO : mmap()ing file of 10000 pages or 655360000 bytes
mmap001 1 TPASS : mmap() completed successfully.
mmap001 0 TINFO : touching mmaped memory
mmap001 2 TBROK : tst_sig.c:233: unexpected signal SIGBUS(7) received (pid = 27075).
mmap001 3 TBROK : tst_sig.c:233: Remaining cases broken
<<<execution_status>>>
initiation_status="ok"
duration=36 termination_type=exited termination_id=2 corefile=no
cutime=48 cstime=53
<<<test_end>>>

[xuyang0410]

Maybe we can lower this value to 5000， I guess most systems have so much available memory.

[ywangwrs]

Hmmm... That could be too large for some aarch64 board as well.
Is that possible to use variable and test available size of memory?

[xuyang0410]

Maybe we can add a check in mmap01.c, ie hugetlb: limit hugepages according to MemAvailable.

[pevik]

There could be even some API function to use % size of the memory.

[metan-ucw]

Indeed there should be an API for figuring out how much memory a test can use, then we could use this function and divide it by a page size in this case.

[xuyang0410]

I will cleanup this case and add api to figure out the available memory .

[ywangwrs]

Any update on this issue?

[metan-ucw]

@xuyang0410 have you had time to look into this?

[xuyang0410]

I plan to pick up msgstress case again and then look into this case in November if nobody does it.