Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1758866AbaKUQTB (ORCPT ); Fri, 21 Nov 2014 11:19:01 -0500 Received: from mail-pa0-f53.google.com ([209.85.220.53]:61348 "EHLO mail-pa0-f53.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1758177AbaKUQTA (ORCPT ); Fri, 21 Nov 2014 11:19:00 -0500 From: Mahendran Ganesh To: minchan@kernel.org, ngupta@vflare.org, ddstreet@ieee.org, sergey.senozhatsky@gmail.com Cc: linux-mm@kvack.org, linux-kernel@vger.kernel.org, Mahendran Ganesh Subject: [PATCH] mm/zsmalloc: support max obj with length of PAGE_SIZE Date: Sat, 22 Nov 2014 00:18:39 +0800 Message-Id: <1416586719-10125-1-git-send-email-opensource.ganesh@gmail.com> X-Mailer: git-send-email 1.7.9.5 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org 2014-11-21 18:32 GMT+08:00 Minchan Kim : > On Fri, Nov 21, 2014 at 06:48:49AM +0000, Minchan Kim wrote: >> On Fri, Nov 21, 2014 at 01:33:26PM +0800, Ganesh Mahendran wrote: >> > Hello >> > >> > 2014-11-21 11:54 GMT+08:00 Minchan Kim : >> > > On Thu, Nov 20, 2014 at 09:21:56PM +0800, Mahendran Ganesh wrote: >> > >> ZS_SIZE_CLASSES is calc by: >> > >> ((ZS_MAX_ALLOC_SIZE - ZS_MIN_ALLOC_SIZE) / ZS_SIZE_CLASS_DELTA + 1) >> > >> >> > >> So when i is in [0, ZS_SIZE_CLASSES - 1), the size: >> > >> size = ZS_MIN_ALLOC_SIZE + i * ZS_SIZE_CLASS_DELTA >> > >> will not be greater than ZS_MAX_ALLOC_SIZE >> > >> >> > >> This patch removes the unnecessary check. >> > > >> > > It depends on ZS_MIN_ALLOC_SIZE. >> > > For example, we would change min to 8 but MAX is still 4096. >> > > ZS_SIZE_CLASSES is (4096 - 8) / 16 + 1 = 256 so 8 + 255 * 16 = 4088, >> > > which exceeds the max. >> > Here, 4088 is less than MAX(4096). >> > >> > ZS_SIZE_CLASSES = (MAX - MIN) / Delta + 1 >> > So, I think the value of >> > MIN + (ZS_SIZE_CLASSES - 1) * Delta = >> > MIN + ((MAX - MIN) / Delta) * Delta = >> > MAX >> > will not exceed the MAX >> >> You're right. It was complext math for me. >> I should go back to elementary school. >> >> Thanks! >> >> Acked-by: Minchan Kim > > I catch a nasty cold but above my poor math makes me think more. > ZS_SIZE_CLASSES is broken. In above my example, current code cannot > allocate 4096 size class so we should correct ZS_SIZE_CLASSES > at first. > > zs_size_classes = zs_max - zs_min / delta + 1; > if ((zs_max - zs_min) % delta) > zs_size_classes += 1; > > Then, we need to code piece you removed. > As well, we need to fix below. > > - area->vm_buf = (char *)__get_free_page(GFP_KERNEL); > + area->vm_buf = kmalloc(ZS_MAX_ALLOC_SIZE); > > Hope I am sane in this time :( how about something like this? In zsmalloc, if ZS_MIN_ALLOC_SIZE is less than ZS_SIZE_CLASS_DELTA. Max obj size (ZS_MIN_ALLOC_SIZE + ZS_SIZE_CLASSES * ZS_SIZE_CLASS_DELTA) will be less than ZS_MAX_ALLOC_SIZE(page size). And in zs_malloc(), we thought we can put an obj(len == PAGE_SIZE) into zsmalloc. But actually we can not. This will make user confused. This patch takes Minchan Kim's suggestion. https://lkml.org/lkml/2014/11/21/172 When (ZS_MAX_ALLOC_SIZE - ZS_MIN_ALLOC_SIZE) % ZS_SIZE_CLASS_DELTA) != 0, we add increase sz_size_class by 1 to make the size of max obj equal to ZS_MAX_ALLOC_SIZE(PAGE_SIZE). Signed-off-by: Mahendran Ganesh --- mm/zsmalloc.c | 22 +++++++++++++++++----- 1 file changed, 17 insertions(+), 5 deletions(-) diff --git a/mm/zsmalloc.c b/mm/zsmalloc.c index 810eda1..bcea72d 100644 --- a/mm/zsmalloc.c +++ b/mm/zsmalloc.c @@ -155,8 +155,6 @@ * (reason above) */ #define ZS_SIZE_CLASS_DELTA (PAGE_SIZE >> 8) -#define ZS_SIZE_CLASSES ((ZS_MAX_ALLOC_SIZE - ZS_MIN_ALLOC_SIZE) / \ - ZS_SIZE_CLASS_DELTA + 1) /* * We do not maintain any list for completely empty or full pages @@ -214,7 +212,8 @@ struct link_free { }; struct zs_pool { - struct size_class *size_class[ZS_SIZE_CLASSES]; + struct size_class **size_class; + int nr_size_classes; gfp_t flags; /* allocation flags used when growing pool */ atomic_long_t pages_allocated; @@ -956,17 +955,28 @@ struct zs_pool *zs_create_pool(gfp_t flags) { int i, ovhd_size; struct zs_pool *pool; + int nr; ovhd_size = roundup(sizeof(*pool), PAGE_SIZE); pool = kzalloc(ovhd_size, GFP_KERNEL); if (!pool) return NULL; + nr = (ZS_MAX_ALLOC_SIZE - ZS_MIN_ALLOC_SIZE) / ZS_SIZE_CLASS_DELTA + 1; + if ((ZS_MAX_ALLOC_SIZE - ZS_MIN_ALLOC_SIZE) % ZS_SIZE_CLASS_DELTA) + nr += 1; + + pool->size_class = kcalloc(nr, sizeof(struct size_class *), + GFP_KERNEL); + if (!pool->size_class) + goto err; + pool->nr_size_classes = nr; + /* * Iterate reversly, because, size of size_class that we want to use * for merging should be larger or equal to current size. */ - for (i = ZS_SIZE_CLASSES - 1; i >= 0; i--) { + for (i = pool->nr_size_classes - 1; i >= 0; i--) { int size; int pages_per_zspage; struct size_class *class; @@ -1020,7 +1030,7 @@ void zs_destroy_pool(struct zs_pool *pool) { int i; - for (i = 0; i < ZS_SIZE_CLASSES; i++) { + for (i = 0; i < pool->nr_size_classes; i++) { int fg; struct size_class *class = pool->size_class[i]; @@ -1038,6 +1048,8 @@ void zs_destroy_pool(struct zs_pool *pool) } kfree(class); } + + kfree(pool->size_class); kfree(pool); } EXPORT_SYMBOL_GPL(zs_destroy_pool); -- 1.7.9.5 -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/