diff options
author | Nishanth Aravamudan <nacc@us.ibm.com> | 2007-12-17 19:20:12 -0500 |
---|---|---|
committer | Linus Torvalds <torvalds@woody.linux-foundation.org> | 2007-12-17 22:28:17 -0500 |
commit | d1c3fb1f8f29c41b0d098d7cfb3c32939043631f (patch) | |
tree | b91983662da7ec4c28ac0788e835c2d51eea20e1 /mm/hugetlb.c | |
parent | 7a3f595cc8298df14a7c71b0d876bafd8e9e1cbf (diff) |
hugetlb: introduce nr_overcommit_hugepages sysctl
hugetlb: introduce nr_overcommit_hugepages sysctl
While examining the code to support /proc/sys/vm/hugetlb_dynamic_pool, I
became convinced that having a boolean sysctl was insufficient:
1) To support per-node control of hugepages, I have previously submitted
patches to add a sysfs attribute related to nr_hugepages. However, with
a boolean global value and per-mount quota enforcement constraining the
dynamic pool, adding corresponding control of the dynamic pool on a
per-node basis seems inconsistent to me.
2) Administration of the hugetlb dynamic pool with multiple hugetlbfs
mount points is, arguably, more arduous than it needs to be. Each quota
would need to be set separately, and the sum would need to be monitored.
To ease the administration, and to help make the way for per-node
control of the static & dynamic hugepage pool, I added a separate
sysctl, nr_overcommit_hugepages. This value serves as a high watermark
for the overall hugepage pool, while nr_hugepages serves as a low
watermark. The boolean sysctl can then be removed, as the condition
nr_overcommit_hugepages > 0
indicates the same administrative setting as
hugetlb_dynamic_pool == 1
Quotas still serve as local enforcement of the size of the pool on a
per-mount basis.
A few caveats:
1) There is a race whereby the global surplus huge page counter is
incremented before a hugepage has allocated. Another process could then
try grow the pool, and fail to convert a surplus huge page to a normal
huge page and instead allocate a fresh huge page. I believe this is
benign, as no memory is leaked (the actual pages are still tracked
correctly) and the counters won't go out of sync.
2) Shrinking the static pool while a surplus is in effect will allow the
number of surplus huge pages to exceed the overcommit value. As long as
this condition holds, however, no more surplus huge pages will be
allowed on the system until one of the two sysctls are increased
sufficiently, or the surplus huge pages go out of use and are freed.
Successfully tested on x86_64 with the current libhugetlbfs snapshot,
modified to use the new sysctl.
Signed-off-by: Nishanth Aravamudan <nacc@us.ibm.com>
Acked-by: Adam Litke <agl@us.ibm.com>
Cc: William Lee Irwin III <wli@holomorphy.com>
Cc: Dave Hansen <haveblue@us.ibm.com>
Cc: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Diffstat (limited to 'mm/hugetlb.c')
-rw-r--r-- | mm/hugetlb.c | 67 |
1 files changed, 61 insertions, 6 deletions
diff --git a/mm/hugetlb.c b/mm/hugetlb.c index 6f978218c2c..3a790651475 100644 --- a/mm/hugetlb.c +++ b/mm/hugetlb.c | |||
@@ -32,6 +32,7 @@ static unsigned int surplus_huge_pages_node[MAX_NUMNODES]; | |||
32 | static gfp_t htlb_alloc_mask = GFP_HIGHUSER; | 32 | static gfp_t htlb_alloc_mask = GFP_HIGHUSER; |
33 | unsigned long hugepages_treat_as_movable; | 33 | unsigned long hugepages_treat_as_movable; |
34 | int hugetlb_dynamic_pool; | 34 | int hugetlb_dynamic_pool; |
35 | unsigned long nr_overcommit_huge_pages; | ||
35 | static int hugetlb_next_nid; | 36 | static int hugetlb_next_nid; |
36 | 37 | ||
37 | /* | 38 | /* |
@@ -227,22 +228,62 @@ static struct page *alloc_buddy_huge_page(struct vm_area_struct *vma, | |||
227 | unsigned long address) | 228 | unsigned long address) |
228 | { | 229 | { |
229 | struct page *page; | 230 | struct page *page; |
231 | unsigned int nid; | ||
230 | 232 | ||
231 | /* Check if the dynamic pool is enabled */ | 233 | /* Check if the dynamic pool is enabled */ |
232 | if (!hugetlb_dynamic_pool) | 234 | if (!hugetlb_dynamic_pool) |
233 | return NULL; | 235 | return NULL; |
234 | 236 | ||
237 | /* | ||
238 | * Assume we will successfully allocate the surplus page to | ||
239 | * prevent racing processes from causing the surplus to exceed | ||
240 | * overcommit | ||
241 | * | ||
242 | * This however introduces a different race, where a process B | ||
243 | * tries to grow the static hugepage pool while alloc_pages() is | ||
244 | * called by process A. B will only examine the per-node | ||
245 | * counters in determining if surplus huge pages can be | ||
246 | * converted to normal huge pages in adjust_pool_surplus(). A | ||
247 | * won't be able to increment the per-node counter, until the | ||
248 | * lock is dropped by B, but B doesn't drop hugetlb_lock until | ||
249 | * no more huge pages can be converted from surplus to normal | ||
250 | * state (and doesn't try to convert again). Thus, we have a | ||
251 | * case where a surplus huge page exists, the pool is grown, and | ||
252 | * the surplus huge page still exists after, even though it | ||
253 | * should just have been converted to a normal huge page. This | ||
254 | * does not leak memory, though, as the hugepage will be freed | ||
255 | * once it is out of use. It also does not allow the counters to | ||
256 | * go out of whack in adjust_pool_surplus() as we don't modify | ||
257 | * the node values until we've gotten the hugepage and only the | ||
258 | * per-node value is checked there. | ||
259 | */ | ||
260 | spin_lock(&hugetlb_lock); | ||
261 | if (surplus_huge_pages >= nr_overcommit_huge_pages) { | ||
262 | spin_unlock(&hugetlb_lock); | ||
263 | return NULL; | ||
264 | } else { | ||
265 | nr_huge_pages++; | ||
266 | surplus_huge_pages++; | ||
267 | } | ||
268 | spin_unlock(&hugetlb_lock); | ||
269 | |||
235 | page = alloc_pages(htlb_alloc_mask|__GFP_COMP|__GFP_NOWARN, | 270 | page = alloc_pages(htlb_alloc_mask|__GFP_COMP|__GFP_NOWARN, |
236 | HUGETLB_PAGE_ORDER); | 271 | HUGETLB_PAGE_ORDER); |
272 | |||
273 | spin_lock(&hugetlb_lock); | ||
237 | if (page) { | 274 | if (page) { |
275 | nid = page_to_nid(page); | ||
238 | set_compound_page_dtor(page, free_huge_page); | 276 | set_compound_page_dtor(page, free_huge_page); |
239 | spin_lock(&hugetlb_lock); | 277 | /* |
240 | nr_huge_pages++; | 278 | * We incremented the global counters already |
241 | nr_huge_pages_node[page_to_nid(page)]++; | 279 | */ |
242 | surplus_huge_pages++; | 280 | nr_huge_pages_node[nid]++; |
243 | surplus_huge_pages_node[page_to_nid(page)]++; | 281 | surplus_huge_pages_node[nid]++; |
244 | spin_unlock(&hugetlb_lock); | 282 | } else { |
283 | nr_huge_pages--; | ||
284 | surplus_huge_pages--; | ||
245 | } | 285 | } |
286 | spin_unlock(&hugetlb_lock); | ||
246 | 287 | ||
247 | return page; | 288 | return page; |
248 | } | 289 | } |
@@ -481,6 +522,12 @@ static unsigned long set_max_huge_pages(unsigned long count) | |||
481 | * Increase the pool size | 522 | * Increase the pool size |
482 | * First take pages out of surplus state. Then make up the | 523 | * First take pages out of surplus state. Then make up the |
483 | * remaining difference by allocating fresh huge pages. | 524 | * remaining difference by allocating fresh huge pages. |
525 | * | ||
526 | * We might race with alloc_buddy_huge_page() here and be unable | ||
527 | * to convert a surplus huge page to a normal huge page. That is | ||
528 | * not critical, though, it just means the overall size of the | ||
529 | * pool might be one hugepage larger than it needs to be, but | ||
530 | * within all the constraints specified by the sysctls. | ||
484 | */ | 531 | */ |
485 | spin_lock(&hugetlb_lock); | 532 | spin_lock(&hugetlb_lock); |
486 | while (surplus_huge_pages && count > persistent_huge_pages) { | 533 | while (surplus_huge_pages && count > persistent_huge_pages) { |
@@ -509,6 +556,14 @@ static unsigned long set_max_huge_pages(unsigned long count) | |||
509 | * to keep enough around to satisfy reservations). Then place | 556 | * to keep enough around to satisfy reservations). Then place |
510 | * pages into surplus state as needed so the pool will shrink | 557 | * pages into surplus state as needed so the pool will shrink |
511 | * to the desired size as pages become free. | 558 | * to the desired size as pages become free. |
559 | * | ||
560 | * By placing pages into the surplus state independent of the | ||
561 | * overcommit value, we are allowing the surplus pool size to | ||
562 | * exceed overcommit. There are few sane options here. Since | ||
563 | * alloc_buddy_huge_page() is checking the global counter, | ||
564 | * though, we'll note that we're not allowed to exceed surplus | ||
565 | * and won't grow the pool anywhere else. Not until one of the | ||
566 | * sysctls are changed, or the surplus pages go out of use. | ||
512 | */ | 567 | */ |
513 | min_count = resv_huge_pages + nr_huge_pages - free_huge_pages; | 568 | min_count = resv_huge_pages + nr_huge_pages - free_huge_pages; |
514 | min_count = max(count, min_count); | 569 | min_count = max(count, min_count); |