ION system heap keeps pages in its pool for better performance. When the system is under memory pressure, slab shrinker calls the callback registered and then the pages pooled get freed. When the shrinker is called, it checks gfp_mask and determines whether the pages from highmem need to be freed or the pages from lowmem. Usually, slab shrinker is invoked on kswapd context which gfp_mask is always GFP_KERNEL, so only lowmem pages are released on kswapd context. This means that highmem pages in the pool are never reclaimed until direct reclaim occurs. This can be problematic when the page pool holds excessive amounts of highmem. For now, the shrinker callback cannot know exactly which zone should be targeted for reclamation, as enough information are not passed to. Thus, it makes sense to shrink both lowmem and highmem zone on kswapd context. Reported-by: Wonseo Choi <wonseo.choi@xxxxxxxxxxx> Signed-off-by: Heesub Shin <heesub.shin@xxxxxxxxxxx> --- drivers/staging/android/ion/ion_page_pool.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) diff --git a/drivers/staging/android/ion/ion_page_pool.c b/drivers/staging/android/ion/ion_page_pool.c index c1cea42b..5864f3d 100644 --- a/drivers/staging/android/ion/ion_page_pool.c +++ b/drivers/staging/android/ion/ion_page_pool.c @@ -21,6 +21,7 @@ #include <linux/list.h> #include <linux/module.h> #include <linux/slab.h> +#include <linux/swap.h> #include "ion_priv.h" static void *ion_page_pool_alloc_pages(struct ion_page_pool *pool) @@ -118,7 +119,10 @@ int ion_page_pool_shrink(struct ion_page_pool *pool, gfp_t gfp_mask, int freed; bool high; - high = !!(gfp_mask & __GFP_HIGHMEM); + if (current_is_kswapd()) + high = 1; + else + high = !!(gfp_mask & __GFP_HIGHMEM); if (nr_to_scan == 0) return ion_page_pool_total(pool, high); -- 1.9.1 _______________________________________________ devel mailing list devel@xxxxxxxxxxxxxxxxxxxxxx http://driverdev.linuxdriverproject.org/mailman/listinfo/driverdev-devel
