Currently vm_area_alloc_pages() contains two cond_resched() points. However, the page allocator already has its own in slow path so an extra resched is not optimal because it delays the loops. The place where a CPU time can be stolen is a VA-space search in the alloc_vmap_area(), especially if the space is really fragmented using synthetic stress tests, after a fast path falls back to a slow one. Move a single cond_resched() there, after dropping free_vmap_area_lock in a slow path. This keeps fairness where it matters while removing redundant yields from the page-allocation path. Signed-off-by: Uladzislau Rezki (Sony) --- mm/vmalloc.c | 9 +++++++-- 1 file changed, 7 insertions(+), 2 deletions(-) diff --git a/mm/vmalloc.c b/mm/vmalloc.c index 5edd536ba9d2..58204728123d 100644 --- a/mm/vmalloc.c +++ b/mm/vmalloc.c @@ -2057,6 +2057,13 @@ static struct vmap_area *alloc_vmap_area(unsigned long size, addr = __alloc_vmap_area(&free_vmap_area_root, &free_vmap_area_list, size, align, vstart, vend); spin_unlock(&free_vmap_area_lock); + + /* + * This is not a fast path. Check if yielding is + * needed. This is the only one reschedule point + * in vmalloc() path. + */ + cond_resched(); } trace_alloc_vmap_area(addr, size, align, vstart, vend, IS_ERR_VALUE(addr)); @@ -3622,7 +3629,6 @@ vm_area_alloc_pages(gfp_t gfp, int nid, pages + nr_allocated); nr_allocated += nr; - cond_resched(); /* * If zero or pages were obtained partly, @@ -3664,7 +3670,6 @@ vm_area_alloc_pages(gfp_t gfp, int nid, for (i = 0; i < (1U << order); i++) pages[nr_allocated + i] = page + i; - cond_resched(); nr_allocated += 1U << order; } -- 2.47.3