[PATCH 00/18] mm, hugetlb: remove a hugetlb_instantiation_mutex

From: Joonsoo Kim
Date: Mon Jul 29 2013 - 01:32:57 EST

Without a hugetlb_instantiation_mutex, if parallel fault occur, we can
fail to allocate a hugepage, because many threads dequeue a hugepage
to handle a fault of same address. This makes reserved pool shortage
just for a little while and this cause faulting thread who is ensured
to have enough reserved hugepages to get a SIGBUS signal.

To solve this problem, we already have a nice solution, that is,
a hugetlb_instantiation_mutex. This blocks other threads to dive into
a fault handler. This solve the problem clearly, but it introduce
performance degradation, because it serialize all fault handling.

Now, I try to remove a hugetlb_instantiation_mutex to get rid of
performance problem reported by Davidlohr Bueso [1].

It is implemented by following 3-steps.

Step 1. Protect region tracking via per region spin_lock.

Currently, region tracking is protected by a
hugetlb_instantiation_mutex, so before removing it, we should
replace it with another solution.

Step 2. Decide whether we use reserved page pool or not by an uniform way.

We need a graceful failure handling if there is no lock like as
hugetlb_instantiation_mutex. To decide whether we need to handle
a failure or not, we need to know current status properly.

Step 3. Graceful failure handling if we failed with reserved page or
failed to allocate with use_reserve.

Failure handling consist of two cases. One is if we failed with
having reserved page, we return back to reserved pool properly.
Current code doesn't recover a reserve count properly, so we need
to fix it. The other is if we failed to allocate a new huge page
with use_reserve indicator, we return 0 to fault handler,
instead of SIGBUS. This makes this thread retrying fault handling.
With above handlings, we can succeed to handle a fault
on any situation without a hugetlb_instantiation_mutex.

Patch 1: Fix a minor problem
Patch 2-5: Implement Step 1.
Patch 6-11: Implement Step 2.
Patch 12-18: Implement Step 3.

These patches are based on my previous patchset [2].
[2] is based on v3.10.

With applying these, I passed a libhugetlbfs test suite clearly which
have allocation-instantiation race test cases.

If there is a something I should consider, please let me know!

[1] http://lwn.net/Articles/558863/
"[PATCH] mm/hugetlb: per-vma instantiation mutexes"
[2] https://lkml.org/lkml/2013/7/22/96
"[PATCH v2 00/10] mm, hugetlb: clean-up and possible bug fix"

Joonsoo Kim (18):
mm, hugetlb: protect reserved pages when softofflining requests the
mm, hugetlb: change variable name reservations to resv
mm, hugetlb: unify region structure handling
mm, hugetlb: region manipulation functions take resv_map rather
mm, hugetlb: protect region tracking via newly introduced resv_map
mm, hugetlb: remove vma_need_reservation()
mm, hugetlb: pass has_reserve to dequeue_huge_page_vma()
mm, hugetlb: do hugepage_subpool_get_pages() when avoid_reserve
mm, hugetlb: unify has_reserve and avoid_reserve to use_reserve
mm, hugetlb: call vma_has_reserve() before entering alloc_huge_page()
mm, hugetlb: move down outside_reserve check
mm, hugetlb: remove a check for return value of alloc_huge_page()
mm, hugetlb: grab a page_table_lock after page_cache_release
mm, hugetlb: clean-up error handling in hugetlb_cow()
mm, hugetlb: move up anon_vma_prepare()
mm, hugetlb: return a reserved page to a reserved pool if failed
mm, hugetlb: retry if we fail to allocate a hugepage with use_reserve
mm, hugetlb: remove a hugetlb_instantiation_mutex

fs/hugetlbfs/inode.c | 12 +-
include/linux/hugetlb.h | 10 ++
mm/hugetlb.c | 361 +++++++++++++++++++++++++----------------------
3 files changed, 217 insertions(+), 166 deletions(-)


To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/