From: Paul Jackson The cpuset mems_allowed update code in alloc_pages_current could (in theory) put a task to sleep that didn't allow sleeping (did not have __GFP_WAIT flag set). In the rare circumstance that the current tasks mems_generation is outofdate compared to the tasks cpuset mems_generation, this mems_allowed update code needs to grap cpuset_sem, which can sleep. We avoid this by not trying to update mems_allowed here if we can't sleep (__GFP_WAIT not set). Thanks to Ray Bryant for noticing this. Signed-off-by: Paul Jackson Signed-off-by: Andrew Morton --- 25-akpm/mm/mempolicy.c | 6 +++++- 1 files changed, 5 insertions(+), 1 deletion(-) diff -puN mm/mempolicy.c~cpusets-alloc-gfp_wait-sleep-fix mm/mempolicy.c --- 25/mm/mempolicy.c~cpusets-alloc-gfp_wait-sleep-fix 2005-03-16 17:12:39.000000000 -0800 +++ 25-akpm/mm/mempolicy.c 2005-03-16 17:12:39.000000000 -0800 @@ -788,12 +788,16 @@ alloc_page_vma(unsigned gfp, struct vm_a * Allocate a page from the kernel page pool. When not in * interrupt context and apply the current process NUMA policy. * Returns NULL when no page can be allocated. + * + * Don't call cpuset_update_current_mems_allowed() unless + * 1) it's ok to take cpuset_sem (can WAIT), and + * 2) allocating for current task (not interrupt). */ struct page *alloc_pages_current(unsigned gfp, unsigned order) { struct mempolicy *pol = current->mempolicy; - if (!in_interrupt()) + if ((gfp & __GFP_WAIT) && !in_interrupt()) cpuset_update_current_mems_allowed(); if (!pol || in_interrupt()) pol = &default_policy; _