From patchwork Thu May 19 03:03:16 2016 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Chen Feng X-Patchwork-Id: 68064 Delivered-To: patch@linaro.org Received: by 10.140.92.199 with SMTP id b65csp2961761qge; Wed, 18 May 2016 20:04:07 -0700 (PDT) X-Received: by 10.66.221.167 with SMTP id qf7mr15924912pac.94.1463627047294; Wed, 18 May 2016 20:04:07 -0700 (PDT) Return-Path: Received: from vger.kernel.org (vger.kernel.org. [209.132.180.67]) by mx.google.com with ESMTP id u2si16457559pac.207.2016.05.18.20.04.06; Wed, 18 May 2016 20:04:07 -0700 (PDT) Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67; Authentication-Results: mx.google.com; spf=pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) smtp.mailfrom=linux-kernel-owner@vger.kernel.org Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1752153AbcESDD5 (ORCPT + 29 others); Wed, 18 May 2016 23:03:57 -0400 Received: from szxga01-in.huawei.com ([58.251.152.64]:63708 "EHLO szxga01-in.huawei.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751761AbcESDDz (ORCPT ); Wed, 18 May 2016 23:03:55 -0400 Received: from 172.24.1.136 (EHLO szxeml428-hub.china.huawei.com) ([172.24.1.136]) by szxrg01-dlp.huawei.com (MOS 4.3.7-GA FastPath queued) with ESMTP id DKS07169; Thu, 19 May 2016 11:03:27 +0800 (CST) Received: from vm163-62.huawei.com (10.184.163.62) by szxeml428-hub.china.huawei.com (10.82.67.183) with Microsoft SMTP Server id 14.3.235.1; Thu, 19 May 2016 11:03:17 +0800 From: Chen Feng To: , , , , , , CC: , , , , , , , , Subject: [PATCH v2] ION: Sys_heap: Add cached pool to spead up cached buffer alloc Date: Thu, 19 May 2016 11:03:16 +0800 Message-ID: <1463626996-51176-1-git-send-email-puck.chen@hisilicon.com> X-Mailer: git-send-email 1.9.1 MIME-Version: 1.0 X-Originating-IP: [10.184.163.62] X-CFilter-Loop: Reflected X-Mirapoint-Virus-RAPID-Raw: score=unknown(0), refid=str=0001.0A090203.573D2D03.004C, ss=1, re=0.000, recu=0.000, reip=0.000, cl=1, cld=1, fgs=0, ip=0.0.0.0, so=2013-06-18 04:22:30, dmn=2013-03-21 17:37:32 X-Mirapoint-Loop-Id: 0f130cf25d64da7cfa8cb04bd21a48f7 Sender: linux-kernel-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Add ion cached pool in system heap. This patch add a cached pool in system heap. It has a great improvement of alloc for cached buffer. With memory pressue alloc test 800MB in userspace used iontest. The result avg is 577ms. Without patch it's avg is about 883ms. v1: Makes the cached buffer zeroed before going to pool v2: Add cached param in pool to distinguish wheather need to flush cache at a fresh alloc. Rework the shrink function. Signed-off-by: Chen Feng Signed-off-by: Xia Qing Reviewed-by: Fu Jun --- drivers/staging/android/ion/ion_page_pool.c | 10 +- drivers/staging/android/ion/ion_priv.h | 5 +- drivers/staging/android/ion/ion_system_heap.c | 183 +++++++++++++++++--------- 3 files changed, 133 insertions(+), 65 deletions(-) -- 1.9.1 diff --git a/drivers/staging/android/ion/ion_page_pool.c b/drivers/staging/android/ion/ion_page_pool.c index 1fe8016..2c5e5c5 100644 --- a/drivers/staging/android/ion/ion_page_pool.c +++ b/drivers/staging/android/ion/ion_page_pool.c @@ -30,8 +30,9 @@ static void *ion_page_pool_alloc_pages(struct ion_page_pool *pool) if (!page) return NULL; - ion_pages_sync_for_device(NULL, page, PAGE_SIZE << pool->order, - DMA_BIDIRECTIONAL); + if (!pool->cached) + ion_pages_sync_for_device(NULL, page, PAGE_SIZE << pool->order, + DMA_BIDIRECTIONAL); return page; } @@ -147,7 +148,8 @@ int ion_page_pool_shrink(struct ion_page_pool *pool, gfp_t gfp_mask, return freed; } -struct ion_page_pool *ion_page_pool_create(gfp_t gfp_mask, unsigned int order) +struct ion_page_pool *ion_page_pool_create(gfp_t gfp_mask, unsigned int order, + bool cached) { struct ion_page_pool *pool = kmalloc(sizeof(*pool), GFP_KERNEL); @@ -161,6 +163,8 @@ struct ion_page_pool *ion_page_pool_create(gfp_t gfp_mask, unsigned int order) pool->order = order; mutex_init(&pool->mutex); plist_node_init(&pool->list, order); + if (cached) + pool->cached = true; return pool; } diff --git a/drivers/staging/android/ion/ion_priv.h b/drivers/staging/android/ion/ion_priv.h index 0239883..f3cb8b4 100644 --- a/drivers/staging/android/ion/ion_priv.h +++ b/drivers/staging/android/ion/ion_priv.h @@ -360,6 +360,7 @@ void ion_carveout_free(struct ion_heap *heap, ion_phys_addr_t addr, * @gfp_mask: gfp_mask to use from alloc * @order: order of pages in the pool * @list: plist node for list of pools + * @cached: it's cached pool or not * * Allows you to keep a pool of pre allocated pages to use from your heap. * Keeping a pool of pages that is ready for dma, ie any cached mapping have @@ -369,6 +370,7 @@ void ion_carveout_free(struct ion_heap *heap, ion_phys_addr_t addr, struct ion_page_pool { int high_count; int low_count; + bool cached; struct list_head high_items; struct list_head low_items; struct mutex mutex; @@ -377,7 +379,8 @@ struct ion_page_pool { struct plist_node list; }; -struct ion_page_pool *ion_page_pool_create(gfp_t gfp_mask, unsigned int order); +struct ion_page_pool *ion_page_pool_create(gfp_t gfp_mask, unsigned int order, + bool cached); void ion_page_pool_destroy(struct ion_page_pool *); struct page *ion_page_pool_alloc(struct ion_page_pool *); void ion_page_pool_free(struct ion_page_pool *, struct page *); diff --git a/drivers/staging/android/ion/ion_system_heap.c b/drivers/staging/android/ion/ion_system_heap.c index b69dfc7..32a0ebf 100644 --- a/drivers/staging/android/ion/ion_system_heap.c +++ b/drivers/staging/android/ion/ion_system_heap.c @@ -26,16 +26,18 @@ #include "ion.h" #include "ion_priv.h" +#define NUM_ORDERS ARRAY_SIZE(orders) + static gfp_t high_order_gfp_flags = (GFP_HIGHUSER | __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY) & ~__GFP_RECLAIM; static gfp_t low_order_gfp_flags = (GFP_HIGHUSER | __GFP_ZERO | __GFP_NOWARN); static const unsigned int orders[] = {8, 4, 0}; -static const int num_orders = ARRAY_SIZE(orders); + static int order_to_index(unsigned int order) { int i; - for (i = 0; i < num_orders; i++) + for (i = 0; i < NUM_ORDERS; i++) if (order == orders[i]) return i; BUG(); @@ -49,47 +51,55 @@ static inline unsigned int order_to_size(int order) struct ion_system_heap { struct ion_heap heap; - struct ion_page_pool *pools[0]; + struct ion_page_pool *uncached_pools[NUM_ORDERS]; + struct ion_page_pool *cached_pools[NUM_ORDERS]; }; +/** + * The page from page-pool are all zeroed before. We need do cache + * clean for cached buffer. The uncached buffer are always non-cached + * since it's allocated. So no need for non-cached pages. + */ static struct page *alloc_buffer_page(struct ion_system_heap *heap, struct ion_buffer *buffer, unsigned long order) { bool cached = ion_buffer_cached(buffer); - struct ion_page_pool *pool = heap->pools[order_to_index(order)]; + struct ion_page_pool *pool; struct page *page; - if (!cached) { - page = ion_page_pool_alloc(pool); - } else { - gfp_t gfp_flags = low_order_gfp_flags; + if (!cached) + pool = heap->uncached_pools[order_to_index(order)]; + else + pool = heap->cached_pools[order_to_index(order)]; - if (order > 4) - gfp_flags = high_order_gfp_flags; - page = alloc_pages(gfp_flags | __GFP_COMP, order); - if (!page) - return NULL; - ion_pages_sync_for_device(NULL, page, PAGE_SIZE << order, - DMA_BIDIRECTIONAL); - } + page = ion_page_pool_alloc(pool); + if (cached) + ion_pages_sync_for_device(NULL, page, PAGE_SIZE << order, + DMA_BIDIRECTIONAL); return page; } static void free_buffer_page(struct ion_system_heap *heap, struct ion_buffer *buffer, struct page *page) { + struct ion_page_pool *pool; unsigned int order = compound_order(page); bool cached = ion_buffer_cached(buffer); - if (!cached && !(buffer->private_flags & ION_PRIV_FLAG_SHRINKER_FREE)) { - struct ion_page_pool *pool = heap->pools[order_to_index(order)]; - - ion_page_pool_free(pool, page); - } else { + /* go to system */ + if (buffer->private_flags & ION_PRIV_FLAG_SHRINKER_FREE) { __free_pages(page, order); + return; } + + if (!cached) + pool = heap->uncached_pools[order_to_index(order)]; + else + pool = heap->cached_pools[order_to_index(order)]; + + ion_page_pool_free(pool, page); } @@ -101,7 +111,7 @@ static struct page *alloc_largest_available(struct ion_system_heap *heap, struct page *page; int i; - for (i = 0; i < num_orders; i++) { + for (i = 0; i < NUM_ORDERS; i++) { if (size < order_to_size(orders[i])) continue; if (max_order < orders[i]) @@ -181,16 +191,11 @@ static void ion_system_heap_free(struct ion_buffer *buffer) struct ion_system_heap, heap); struct sg_table *table = buffer->sg_table; - bool cached = ion_buffer_cached(buffer); struct scatterlist *sg; int i; - /* - * uncached pages come from the page pools, zero them before returning - * for security purposes (other allocations are zerod at - * alloc time - */ - if (!cached && !(buffer->private_flags & ION_PRIV_FLAG_SHRINKER_FREE)) + /* zero the buffer before goto page pool */ + if (!(buffer->private_flags & ION_PRIV_FLAG_SHRINKER_FREE)) ion_heap_buffer_zero(buffer); for_each_sg(table->sgl, sg, table->nents, i) @@ -213,6 +218,8 @@ static void ion_system_heap_unmap_dma(struct ion_heap *heap, static int ion_system_heap_shrink(struct ion_heap *heap, gfp_t gfp_mask, int nr_to_scan) { + struct ion_page_pool *uncached_pool; + struct ion_page_pool *cached_pool; struct ion_system_heap *sys_heap; int nr_total = 0; int i, nr_freed; @@ -223,20 +230,35 @@ static int ion_system_heap_shrink(struct ion_heap *heap, gfp_t gfp_mask, if (!nr_to_scan) only_scan = 1; - for (i = 0; i < num_orders; i++) { - struct ion_page_pool *pool = sys_heap->pools[i]; - - nr_freed = ion_page_pool_shrink(pool, gfp_mask, nr_to_scan); - nr_total += nr_freed; - - if (!only_scan) { + for (i = 0; i < NUM_ORDERS; i++) { + uncached_pool = sys_heap->uncached_pools[i]; + cached_pool = sys_heap->cached_pools[i]; + + if (only_scan) { + nr_total += ion_page_pool_shrink(uncached_pool, + gfp_mask, + nr_to_scan); + + nr_total += ion_page_pool_shrink(cached_pool, + gfp_mask, + nr_to_scan); + } else { + nr_freed = ion_page_pool_shrink(uncached_pool, + gfp_mask, + nr_to_scan); nr_to_scan -= nr_freed; - /* shrink completed */ + nr_total += nr_freed; + if (nr_to_scan <= 0) + break; + nr_freed = ion_page_pool_shrink(cached_pool, + gfp_mask, + nr_to_scan); + nr_to_scan -= nr_freed; + nr_total += nr_freed; if (nr_to_scan <= 0) break; } } - return nr_total; } @@ -259,52 +281,89 @@ static int ion_system_heap_debug_show(struct ion_heap *heap, struct seq_file *s, struct ion_system_heap, heap); int i; + struct ion_page_pool *pool; - for (i = 0; i < num_orders; i++) { - struct ion_page_pool *pool = sys_heap->pools[i]; + for (i = 0; i < NUM_ORDERS; i++) { + pool = sys_heap->uncached_pools[i]; - seq_printf(s, "%d order %u highmem pages in pool = %lu total\n", + seq_printf(s, "%d order %u highmem pages uncached %lu total\n", pool->high_count, pool->order, (PAGE_SIZE << pool->order) * pool->high_count); - seq_printf(s, "%d order %u lowmem pages in pool = %lu total\n", + seq_printf(s, "%d order %u lowmem pages uncached %lu total\n", + pool->low_count, pool->order, + (PAGE_SIZE << pool->order) * pool->low_count); + } + + for (i = 0; i < NUM_ORDERS; i++) { + pool = sys_heap->cached_pools[i]; + + seq_printf(s, "%d order %u highmem pages cached %lu total\n", + pool->high_count, pool->order, + (PAGE_SIZE << pool->order) * pool->high_count); + seq_printf(s, "%d order %u lowmem pages cached %lu total\n", pool->low_count, pool->order, (PAGE_SIZE << pool->order) * pool->low_count); } return 0; } +static void ion_system_heap_destroy_pools(struct ion_page_pool **pools) +{ + int i; + + for (i = 0; i < NUM_ORDERS; i++) + if (pools[i]) + ion_page_pool_destroy(pools[i]); +} + +static int ion_system_heap_create_pools(struct ion_page_pool **pools, + bool cached) +{ + int i; + gfp_t gfp_flags = low_order_gfp_flags; + + for (i = 0; i < NUM_ORDERS; i++) { + struct ion_page_pool *pool; + + if (orders[i] > 4) + gfp_flags = high_order_gfp_flags; + + pool = ion_page_pool_create(gfp_flags, orders[i], cached); + if (!pool) + goto err_create_pool; + pools[i] = pool; + } + return 0; + +err_create_pool: + ion_system_heap_destroy_pools(pools); + return -ENOMEM; +} + struct ion_heap *ion_system_heap_create(struct ion_platform_heap *unused) { struct ion_system_heap *heap; - int i; - heap = kzalloc(sizeof(struct ion_system_heap) + - sizeof(struct ion_page_pool *) * num_orders, - GFP_KERNEL); + heap = kzalloc(sizeof(*heap), GFP_KERNEL); if (!heap) return ERR_PTR(-ENOMEM); heap->heap.ops = &system_heap_ops; heap->heap.type = ION_HEAP_TYPE_SYSTEM; heap->heap.flags = ION_HEAP_FLAG_DEFER_FREE; - for (i = 0; i < num_orders; i++) { - struct ion_page_pool *pool; - gfp_t gfp_flags = low_order_gfp_flags; + if (ion_system_heap_create_pools(heap->uncached_pools, false)) + goto free_heap; - if (orders[i] > 4) - gfp_flags = high_order_gfp_flags; - pool = ion_page_pool_create(gfp_flags, orders[i]); - if (!pool) - goto destroy_pools; - heap->pools[i] = pool; - } + if (ion_system_heap_create_pools(heap->cached_pools, true)) + goto destroy_uncached_pools; heap->heap.debug_show = ion_system_heap_debug_show; return &heap->heap; -destroy_pools: - while (i--) - ion_page_pool_destroy(heap->pools[i]); +destroy_uncached_pools: + ion_system_heap_destroy_pools(heap->uncached_pools); + +free_heap: kfree(heap); return ERR_PTR(-ENOMEM); } @@ -316,8 +375,10 @@ void ion_system_heap_destroy(struct ion_heap *heap) heap); int i; - for (i = 0; i < num_orders; i++) - ion_page_pool_destroy(sys_heap->pools[i]); + for (i = 0; i < NUM_ORDERS; i++) { + ion_page_pool_destroy(sys_heap->uncached_pools[i]); + ion_page_pool_destroy(sys_heap->cached_pools[i]); + } kfree(sys_heap); }