On 06/02/2016 03:13 PM, Michal Hocko wrote:
On Thu 02-06-16 13:59:49, Michal Hocko wrote:
[CCing Ebru]
On Thu 02-06-16 13:50:46, Michal Hocko wrote:
[CCing Andrea and Kirill]
Hmm, thinking about it little bit more it might be related to "mm, thp:
make swapin readahead under down_read of mmap_sem". I didn't get to look
closer at the patch but maybe revalidate after mmap sem is dropped is
not sufficient.
so hugepage_vma_revalidate does this:
hstart = (vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK;
hend = vma->vm_end & HPAGE_PMD_MASK;
if (address < hstart || address + HPAGE_PMD_SIZE > hend)
return SCAN_ADDRESS_RANGE;
I really do not see why we have to play with hstart and hend.
AFAIU the point of these tests is to see whether the vma is large enough
to contain the given address within a huge page. And it kind of silently
assumes that address is already huge-page aligned.
But
address + HPAGE_PMD_SIZE > hend part looks suspicious. address
always have to vm_start <= address && address < vm_end AFAICS.
I think the check is fine after a deeper look. The code itself is also
quite old, the recent patches just abstracted it to the new function.
and the above allows address + HPAGE_PMD_MASK == end.
--
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>