[LTP] [PATCH v1 3/3] Add process_madvise03 test

[Richard Palethorpe]

Richard Palethorpe <rpalethorpe@suse.de> writes:

> Hello,
>
> Andrea Cervesato via ltp <ltp@lists.linux.it> writes:
>
>> Test for checking MADV_PAGEOUT functionality over memory-mapped file
>> in process_madvise syscall.
>
> So this one doesn't need swap, but it has some other issues.
>
>>
>> Signed-off-by: Andrea Cervesato <andrea.cervesato@suse.com>
>> ---
>>  testcases/kernel/syscalls/cma/.gitignore      |   1 +
>>  .../kernel/syscalls/cma/process_madvise03.c   | 139 ++++++++++++++++++
>>  2 files changed, 140 insertions(+)
>>  create mode 100644 testcases/kernel/syscalls/cma/process_madvise03.c
>>
>> diff --git a/testcases/kernel/syscalls/cma/.gitignore b/testcases/kernel/syscalls/cma/.gitignore
>> index 47ae3e445..147b03c48 100644
>> --- a/testcases/kernel/syscalls/cma/.gitignore
>> +++ b/testcases/kernel/syscalls/cma/.gitignore
>> @@ -4,3 +4,4 @@
>>  /process_vm_writev02
>>  /process_madvise01
>>  /process_madvise02
>> +/process_madvise03
>> diff --git a/testcases/kernel/syscalls/cma/process_madvise03.c b/testcases/kernel/syscalls/cma/process_madvise03.c
>> new file mode 100644
>> index 000000000..3f12ef530
>> --- /dev/null
>> +++ b/testcases/kernel/syscalls/cma/process_madvise03.c
>> @@ -0,0 +1,139 @@
>> +// SPDX-License-Identifier: GPL-2.0-or-later
>> +/*
>> + * Copyright (C) 2022 SUSE LLC Andrea Cervesato <andrea.cervesato@suse.com>
>> + */
>> +
>> +/*\
>> + * [Description]
>> + *
>> + * Spawn child inside cgroup and set max memory. Allocate file-backed memory
>> + * pages inside child and reclaim it with MADV_PAGEOUT. Then check if memory
>> + * pages have been written back to the backing storage.

Actually, one more thing. You don't check if it has been written to the
backing store and it's quite hard to check for this.

At best you could reopen the file in the parent and check the contents
are correct. Otherwise it requires checking the page cache has been
discarded using a side channel (e.g. timing loads).

So I would just not bother for this test.

Also process_madvise returns the number of bytes *advised* not what was
actually reclaimed. Even that is not guaranteed to be the same as the
amount requested.

-- 
Thank you,
Richard.