On Wednesday 17 February 2016 03:29 PM, Kirill A. Shutemov wrote:
> On Wed, Feb 17, 2016 at 01:13:26AM +0530, Vaishali Thakkar wrote:
>> Currently incorrect default hugepage pool size is reported by proc
>> nr_hugepages when number of pages for the default huge page size is
>> specified twice.
>>
>> When multiple huge page sizes are supported, /proc/sys/vm/nr_hugepages
>> indicates the current number of pre-allocated huge pages of the default
>> size. Basically /proc/sys/vm/nr_hugepages displays default_hstate->
>> max_huge_pages and after boot time pre-allocation, max_huge_pages should
>> equal the number of pre-allocated pages (nr_hugepages).
>>
>> Test case:
>>
>> Note that this is specific to x86 architecture.
>>
>> Boot the kernel with command line option 'default_hugepagesz=1G
>> hugepages=X hugepagesz=2M hugepages=Y hugepagesz=1G hugepages=Z'. After
>> boot, 'cat /proc/sys/vm/nr_hugepages' and 'sysctl -a | grep hugepages'
>> returns the value X. However, dmesg output shows that Z huge pages were
>> pre-allocated.
>>
>> So, the root cause of the problem here is that the global variable
>> default_hstate_max_huge_pages is set if a default huge page size is
>> specified (directly or indirectly) on the command line. After the
>> command line processing in hugetlb_init, if default_hstate_max_huge_pages
>> is set, the value is assigned to default_hstae.max_huge_pages. However,
>> default_hstate.max_huge_pages may have already been set based on the
>> number of pre-allocated huge pages of default_hstate size.
>>
>> The solution to this problem is if hstate->max_huge_pages is already set
>> then it should not set as a result of global max_huge_pages value.
>> Basically if the value of the variable hugepages is set multiple times
>> on a command line for a specific supported hugepagesize then proc layer
>> should consider the last specified value.
>>
>> Signed-off-by: Vaishali Thakkar <vaishali.thakkar@xxxxxxxxxx>
>> ---
>> The patch contains one line over 80 characters as I think limiting that
>> line to 80 characters makes code look bit ugly. But if anyone is having
>> issue with that then I am fine with limiting it to 80 chracters.
> What about this?
>
> if (default_hstate_max_huge_pages && !default_hstate.max_huge_pages)
> default_hstate.max_huge_pages = default_hstate_max_huge_pages;
Yes, it does the same thing. I thought about it. But to me somehow it looks
hard to read. So, personally I don't prefer it.
Do you want me to change this?
>> ---
>> mm/hugetlb.c | 6 ++++--
>> 1 file changed, 4 insertions(+), 2 deletions(-)
>>
>> diff --git a/mm/hugetlb.c b/mm/hugetlb.c
>> index 06ae13e..01f2b48 100644
>> --- a/mm/hugetlb.c
>> +++ b/mm/hugetlb.c
>> @@ -2630,8 +2630,10 @@ static int __init hugetlb_init(void)
>> hugetlb_add_hstate(HUGETLB_PAGE_ORDER);
>> }
>> default_hstate_idx = hstate_index(size_to_hstate(default_hstate_size));
>> - if (default_hstate_max_huge_pages)
>> - default_hstate.max_huge_pages = default_hstate_max_huge_pages;
>> + if (default_hstate_max_huge_pages) {
>> + if (!default_hstate.max_huge_pages)
>> + default_hstate.max_huge_pages = default_hstate_max_huge_pages;
>> + }
>>
>> hugetlb_init_hstates();
>> gather_bootmem_prealloc();
>> --
>> 2.1.4
>>
>> --
>> To unsubscribe, send a message with 'unsubscribe linux-mm' in
>> the body to majordomo@xxxxxxxxx. For more info on Linux MM,
>> see: http://www.linux-mm.org/ .
>> Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>
--
Vaishali
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>