[LTP] [PATCH v2 02/13] thp: ensure THP/hugetlbfs is available

[Punit Agrawal]

Hi Cyril,

Thanks for taking a look.

Cyril Hrubis <chrubis@suse.cz> writes:

> Hi!
>> diff --git a/testcases/kernel/mem/thp/thp01.c b/testcases/kernel/mem/thp/thp01.c
>> index 2f2bb2bdb..939c01b98 100644
>> --- a/testcases/kernel/mem/thp/thp01.c
>> +++ b/testcases/kernel/mem/thp/thp01.c
>> @@ -116,6 +116,9 @@ static void setup(void)
>>  	int i;
>>  	long arg_len, arg_count;
>>  
>> +	if (access(PATH_THP, F_OK) == -1)
>> +		tst_brk(TCONF, "THP not enabled in kernel?");
>> +
>>  	bst = SAFE_MMAP(NULL, sizeof(*bst),
>>  			   PROT_READ | PROT_WRITE,
>>  			   MAP_SHARED | MAP_ANONYMOUS, -1, 0);
>
> I do not get what is the problem with this test on a machine without
> hugepages or transparent hugepages. The test just tries to execute true
> with excessively large array of command line arguments, if that fails
> without hugepages it's a kernel bug.

The test passes even if THP is not enabled on the test kernel. This
gives the false impression that THP is working as expected.

The explicit check is making sure the functionality under test is is
enabled in the test kernel.

>
>> diff --git a/testcases/kernel/mem/thp/thp02.c b/testcases/kernel/mem/thp/thp02.c
>> index acc70e215..6502210e5 100644
>> --- a/testcases/kernel/mem/thp/thp02.c
>> +++ b/testcases/kernel/mem/thp/thp02.c
>> @@ -90,6 +90,8 @@ static void setup(void)
>>  	if (access(PATH_THP, F_OK) == -1)
>>  		tst_brk(TCONF, "THP not enabled in kernel?");
>>  
>> +	check_hugepage();
>> +
>>  	ps = sysconf(_SC_PAGESIZE);
>>  	hps = SAFE_READ_MEMINFO("Hugepagesize:") * 1024;
>>  	size = hps * 4;
>> diff --git a/testcases/kernel/mem/thp/thp03.c b/testcases/kernel/mem/thp/thp03.c
>> index 19db44944..c6062505f 100644
>> --- a/testcases/kernel/mem/thp/thp03.c
>> +++ b/testcases/kernel/mem/thp/thp03.c
>> @@ -82,6 +82,8 @@ static void setup(void)
>>  	if (access(PATH_THP, F_OK) == -1)
>>  		tst_brk(TCONF, "THP not enabled in kernel?");
>>  
>> +	check_hugepage();
>> +
>>  	hugepage_size = SAFE_READ_MEMINFO("Hugepagesize:") * KB;
>>  	unaligned_size = hugepage_size * 4 - 1;
>>  	page_size = SAFE_SYSCONF(_SC_PAGESIZE);
>
> Does this two fail on an attempt to find "Hugepagesize:" in meminfo
> file? Otherwise I do not see how they can fail.

Yes, "Hugepagesize:" goes missing in /proc/meminfo. When the kernel
doesn't have hugepage support enabled, the tests fail with message like
-

safe_file_ops.c:220: BROK: Expected 1 conversions got 0 at thp03.c:85

With the patch, the message helps identify the cause of the failure.

Thanks,
Punit