On Thu, 23 Feb 2012 14:56:36 -0500
Rik van Riel [off-list ref] wrote:

When we look for a VMA smaller than the cached_hole_size, we set the
starting search address to mm->mmap_base, to try and find our hole.

However, even in the case where we fall through and found nothing at
the mm->free_area_cache, we still reset the search address to mm->mmap_base.
This bug results in quadratic behaviour, with observed mmap times of 0.4
seconds for processes that have very fragmented memory.

If there is no hole small enough for us to fit the VMA, and we have
no good spot for us right at mm->free_area_cache, we are much better
off continuing the search down from mm->free_area_cache, instead of
all the way from the top.

This has been at least partially addressed in recent patches from Xiao
Guangrong.  Please review his five-patch series starting with "[PATCH
1/5] hugetlbfs: fix hugetlb_get_unmapped_area".

I've already merged those patches and we need to work out what way to
go.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>