[LTP] [PATCH] futex: skip futex_wake04.c if system not support huge pages

[Wanlong Gao]

On Tue, Nov 17, 2015 at 03:52:29PM +0800, Li Wang wrote:
> Hi,
> 
> On Mon, Nov 16, 2015 at 5:01 PM, Li Wang <liwang@redhat.com> wrote:
> 
>     Hi,
> 
>     On Mon, Nov 16, 2015 at 3:37 PM, Wanlong Gao <wanlong.gao@gmail.com> wrote:
>    
>         On Mon, Nov 16, 2015 at 02:00:52PM +0800, Li Wang wrote:
>         >
>         > In my test system, there exist file "/proc/sys/vm/nr_hugepages" with
>         normal
>         > permission, but reject any R/W operation.
>         >
>         > So, if I use   if (access(PATH_NR_HUGEPAGES, F_OK) == -1),  It
>         doesn't work.
>         >
>         > # ll /proc/sys/vm/nr_hugepages
>         > -rw-r--r--. 1 root root 0 Nov 15 09:03 /proc/sys/vm/nr_hugepages
>         >
>         > # cat /proc/sys/vm/nr_hugepages
>         > cat: write error: Bad address
> 
>         Would you show your kernel version or investigate why this happens?
>         This should not exist if we disable the hugetlbfs in kernel config.
>    
> 
>     I looked close to my test bed, it is a KVM Guest which enabled hugetlbfs on
>     IBM POWER8E,
> 
>     the config is:
>     CONFIG_CGROUP_HUGETLB=y
>     CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE=y
>     CONFIG_HUGETLB_PAGE_SIZE_VARIABLE=y
>     CONFIG_SYS_SUPPORTS_HUGETLBFS=y
>     CONFIG_TRANSPARENT_HUGEPAGE=y
>     # CONFIG_TRANSPARENT_HUGEPAGE_ALWAYS is not set
>     CONFIG_TRANSPARENT_HUGEPAGE_MADVISE=y
>     CONFIG_HUGETLBFS=y
>     CONFIG_HUGETLB_PAGE=y
> 
>     # uname -r
>     3.10.0-*.el7.ppc64le
>      
> 
> 
>         > VmallocChunk:   8589850624 kB
>         > HardwareCorrupted:     0 kB
> 
>         > AnonHugePages:         0 kB
> 
>         This tells that your kernel support THP.
> 
>      
>     yes, the reason is that my kvm guest XML was not added these below line:
> 
>     guest XML:
>      <memoryBacking>
>           <hugepages/>
>      </memoryBacking>
> 
>     after adding them, the kvm guest can works well with hugepages.
> 
>     it is a configuration issue of guest, not a kernel bug.  so, I propose my
>     patch as previously.
> 
> 
> IMO, if we use the original "/proc/sys/vm/nr_hugepages" to check the hugetlb
> support, it does not work on the situation which I mentioned above.
> 
> So, I suggest to add new Marco 'PATH_SYS_HUGEPAGES'  to skip the testcase.
> 
> what do you think?

It's just a workaround, anyway, I have no objection.

Thanks,
Wanlong Gao

> 
> --
> Regards,
> Li Wang
> Email: liwang@redhat.com