From c596d9f320aaf30d28c1d793ff3a976dee1db8f5 Mon Sep 17 00:00:00 2001 From: David Rientjes Date: Sun, 6 May 2007 14:49:32 -0700 Subject: [PATCH] cpusets: allow TIF_MEMDIE threads to allocate anywhere OOM killed tasks have access to memory reserves as specified by the TIF_MEMDIE flag in the hopes that it will quickly exit. If such a task has memory allocations constrained by cpusets, we may encounter a deadlock if a blocking task cannot exit because it cannot allocate the necessary memory. We allow tasks that have the TIF_MEMDIE flag to allocate memory anywhere, including outside its cpuset restriction, so that it can quickly die regardless of whether it is __GFP_HARDWALL. Cc: Andi Kleen Cc: Paul Jackson Cc: Christoph Lameter Signed-off-by: David Rientjes Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- kernel/cpuset.c | 22 ++++++++++++++++++++-- 1 file changed, 20 insertions(+), 2 deletions(-) diff --git a/kernel/cpuset.c b/kernel/cpuset.c index f382b0f775e..d240349cbf0 100644 --- a/kernel/cpuset.c +++ b/kernel/cpuset.c @@ -2351,6 +2351,8 @@ static const struct cpuset *nearest_exclusive_ancestor(const struct cpuset *cs) * z's node is in our tasks mems_allowed, yes. If it's not a * __GFP_HARDWALL request and this zone's nodes is in the nearest * mem_exclusive cpuset ancestor to this tasks cpuset, yes. + * If the task has been OOM killed and has access to memory reserves + * as specified by the TIF_MEMDIE flag, yes. * Otherwise, no. * * If __GFP_HARDWALL is set, cpuset_zone_allowed_softwall() @@ -2368,7 +2370,8 @@ static const struct cpuset *nearest_exclusive_ancestor(const struct cpuset *cs) * calls get to this routine, we should just shut up and say 'yes'. * * GFP_USER allocations are marked with the __GFP_HARDWALL bit, - * and do not allow allocations outside the current tasks cpuset. + * and do not allow allocations outside the current tasks cpuset + * unless the task has been OOM killed as is marked TIF_MEMDIE. * GFP_KERNEL allocations are not so marked, so can escape to the * nearest enclosing mem_exclusive ancestor cpuset. * @@ -2392,6 +2395,7 @@ static const struct cpuset *nearest_exclusive_ancestor(const struct cpuset *cs) * affect that: * in_interrupt - any node ok (current task context irrelevant) * GFP_ATOMIC - any node ok + * TIF_MEMDIE - any node ok * GFP_KERNEL - any node in enclosing mem_exclusive cpuset ok * GFP_USER - only nodes in current tasks mems allowed ok. * @@ -2413,6 +2417,12 @@ int __cpuset_zone_allowed_softwall(struct zone *z, gfp_t gfp_mask) might_sleep_if(!(gfp_mask & __GFP_HARDWALL)); if (node_isset(node, current->mems_allowed)) return 1; + /* + * Allow tasks that have access to memory reserves because they have + * been OOM killed to get memory anywhere. + */ + if (unlikely(test_thread_flag(TIF_MEMDIE))) + return 1; if (gfp_mask & __GFP_HARDWALL) /* If hardwall request, stop here */ return 0; @@ -2438,7 +2448,9 @@ int __cpuset_zone_allowed_softwall(struct zone *z, gfp_t gfp_mask) * * If we're in interrupt, yes, we can always allocate. * If __GFP_THISNODE is set, yes, we can always allocate. If zone - * z's node is in our tasks mems_allowed, yes. Otherwise, no. + * z's node is in our tasks mems_allowed, yes. If the task has been + * OOM killed and has access to memory reserves as specified by the + * TIF_MEMDIE flag, yes. Otherwise, no. * * The __GFP_THISNODE placement logic is really handled elsewhere, * by forcibly using a zonelist starting at a specified node, and by @@ -2462,6 +2474,12 @@ int __cpuset_zone_allowed_hardwall(struct zone *z, gfp_t gfp_mask) node = zone_to_nid(z); if (node_isset(node, current->mems_allowed)) return 1; + /* + * Allow tasks that have access to memory reserves because they have + * been OOM killed to get memory anywhere. + */ + if (unlikely(test_thread_flag(TIF_MEMDIE))) + return 1; return 0; } -- 2.20.1