On 04/06/2016 10:26 AM, Nikolay Borisov wrote:
>
>
> On 04/06/2016 04:25 AM, David Rientjes wrote:
>> The page_counter rounds limits down to page size values. This makes
>> sense, except in the case of hugetlb_cgroup where it's not possible to
>> charge partial hugepages.
>>
>> Round the hugetlb_cgroup limit down to hugepage size.
>>
>> Signed-off-by: David Rientjes <rient...@google.com>
>> ---
>> mm/hugetlb_cgroup.c | 1 +
>> 1 file changed, 1 insertion(+)
>>
>> diff --git a/mm/hugetlb_cgroup.c b/mm/hugetlb_cgroup.c
>> --- a/mm/hugetlb_cgroup.c
>> +++ b/mm/hugetlb_cgroup.c
>> @@ -288,6 +288,7 @@ static ssize_t hugetlb_cgroup_write(struct
>> kernfs_open_file *of,
>>
>> switch (MEMFILE_ATTR(of_cft(of)->private)) {
>> case RES_LIMIT:
>> + nr_pages &= ~((1 << huge_page_order(&hstates[idx])) - 1);
>
> Why not:
>
> nr_pages = round_down(nr_pages, huge_page_order(&hstates[idx]));
Oops, that should be:
round_down(nr_pages, 1 << huge_page_order(&hstates[idx]));
>
>
>> mutex_lock(&hugetlb_limit_mutex);
>> ret = page_counter_limit(&h_cg->hugepage[idx], nr_pages);
>> mutex_unlock(&hugetlb_limit_mutex);
>>
