From patchwork Tue Feb 27 03:02:54 2024 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Chengming Zhou X-Patchwork-Id: 13573219 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from kanga.kvack.org (kanga.kvack.org [205.233.56.17]) by smtp.lore.kernel.org (Postfix) with ESMTP id AF1ABC5478C for ; Tue, 27 Feb 2024 03:03:17 +0000 (UTC) Received: by kanga.kvack.org (Postfix) id 33CB04401E3; Mon, 26 Feb 2024 22:03:17 -0500 (EST) Received: by kanga.kvack.org (Postfix, from userid 40) id 2EBF244017F; Mon, 26 Feb 2024 22:03:17 -0500 (EST) X-Delivered-To: int-list-linux-mm@kvack.org Received: by kanga.kvack.org (Postfix, from userid 63042) id 13EF84401E3; Mon, 26 Feb 2024 22:03:17 -0500 (EST) X-Delivered-To: linux-mm@kvack.org Received: from relay.hostedemail.com (smtprelay0011.hostedemail.com [216.40.44.11]) by kanga.kvack.org (Postfix) with ESMTP id 0472744017F for ; Mon, 26 Feb 2024 22:03:17 -0500 (EST) Received: from smtpin30.hostedemail.com (a10.router.float.18 [10.200.18.1]) by unirelay08.hostedemail.com (Postfix) with ESMTP id D4F4B140A72 for ; Tue, 27 Feb 2024 03:03:16 +0000 (UTC) X-FDA: 81836087592.30.69AC8C2 Received: from out-177.mta1.migadu.com (out-177.mta1.migadu.com [95.215.58.177]) by imf21.hostedemail.com (Postfix) with ESMTP id 1501F1C0012 for ; Tue, 27 Feb 2024 03:03:14 +0000 (UTC) Authentication-Results: imf21.hostedemail.com; dkim=none; dmarc=fail reason="SPF not aligned (relaxed), No valid DKIM" header.from=bytedance.com (policy=quarantine); spf=pass (imf21.hostedemail.com: domain of chengming.zhou@linux.dev designates 95.215.58.177 as permitted sender) smtp.mailfrom=chengming.zhou@linux.dev ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=hostedemail.com; s=arc-20220608; t=1709002995; h=from:from:sender:reply-to:subject:subject:date:date: message-id:message-id:to:to:cc:cc:mime-version:mime-version: content-type:content-type: content-transfer-encoding:content-transfer-encoding: in-reply-to:in-reply-to:references:references; bh=+AeoM9bXyNXnW2+Ff6qKZg8BHvCGSUPxBlTPsvhNgxg=; b=XPio75yyvxYwhJFaWRNIlWkVFF/1NS4XuByjsnt/sg86/0wlVj4dVCaowR477PDcpFEwCU dqI+XIvLEL8Hy55VzG9HZZ4qTrRkDLYAT21s0mxrq0b2/TNKDi6LE+RdjEsLedXInSSI3c BkQu/ZqStLWMBN4gaSu/UNRloTHuzD8= ARC-Authentication-Results: i=1; imf21.hostedemail.com; dkim=none; dmarc=fail reason="SPF not aligned (relaxed), No valid DKIM" header.from=bytedance.com (policy=quarantine); spf=pass (imf21.hostedemail.com: domain of chengming.zhou@linux.dev designates 95.215.58.177 as permitted sender) smtp.mailfrom=chengming.zhou@linux.dev ARC-Seal: i=1; s=arc-20220608; d=hostedemail.com; t=1709002995; a=rsa-sha256; cv=none; b=v67jnbxEecLQ9a+xGHcy4OeEqAu3MXUZ0AiQTvHI7K04IPdUuixte/HIKN/XujQxaepR1i zKuANtU9HTxrDWu9i3tWXOxPwZEM5s0OHyOCJ6DasfOjbqu0fwqbcePLVcupsbyOGmQotH 0n+3845Pqf7D1RKR/kSBlU/WYvsPDJ0= X-Report-Abuse: Please report any abuse attempt to abuse@migadu.com and include these headers. From: Chengming Zhou Date: Tue, 27 Feb 2024 03:02:54 +0000 Subject: [PATCH 1/2] mm/zsmalloc: don't hold locks of all pages when free_zspage() MIME-Version: 1.0 Message-Id: <20240226-zsmalloc-zspage-rcu-v1-1-456b0ef1a89d@bytedance.com> References: <20240226-zsmalloc-zspage-rcu-v1-0-456b0ef1a89d@bytedance.com> In-Reply-To: <20240226-zsmalloc-zspage-rcu-v1-0-456b0ef1a89d@bytedance.com> To: yosryahmed@google.com, Sergey Senozhatsky , hannes@cmpxchg.org, nphamcs@gmail.com, Andrew Morton , Minchan Kim Cc: linux-kernel@vger.kernel.org, linux-mm@kvack.org, Chengming Zhou X-Developer-Signature: v=1; a=ed25519-sha256; t=1709002985; l=7649; i=zhouchengming@bytedance.com; s=20240220; h=from:subject:message-id; bh=/Y+8Rr5r7R4qG0VPWtmHKI3CmlgVA1JjzF5IldWYPOo=; b=Ipe8SF91JeqdJP/MV2fbyh7lbm0zV5Funv1K3Qy6bYmAr/Mo669Z/C6TNDwxz1AL+WdI8ItTm pWaD8WRhpRYDB285bxUKML+G8G1G1EV72v/91+bR9W3Kf9H97jTCK7k X-Developer-Key: i=zhouchengming@bytedance.com; a=ed25519; pk=5+68Wfci+T30FoQos5RH+hfToF6SlC+S9LMPSPBFWuw= X-Migadu-Flow: FLOW_OUT X-Rspam-User: X-Rspamd-Queue-Id: 1501F1C0012 X-Rspamd-Server: rspam12 X-Stat-Signature: 4aksuxe936mn9pixqq36zdqx16io3ecz X-Rspamd-Pre-Result: action=add header; module=dmarc; Action set by DMARC X-Rspam: Yes X-HE-Tag: 1709002994-447531 X-HE-Meta: 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 LY4AqYzE xQYNJEpz9FHOuX99vMdWjRZEpFBZsIcXZFaBOipAzWxTHpqkyCPnBlTwHuQ== X-Bogosity: Ham, tests=bogofilter, spamicity=0.000000, version=1.2.4 Sender: owner-linux-mm@kvack.org Precedence: bulk X-Loop: owner-majordomo@kvack.org List-ID: List-Subscribe: List-Unsubscribe: free_zspage() has to hold locks of all pages, since zs_page_migrate() path rely on this page lock to protect the race between zs_free() and it, so it can safely get zspage from page->private. But this way is not good and simple enough: 1. Since zs_free() couldn't be sleepable, it can only trylock pages, or has to kick_deferred_free() to defer that to a work. 2. Even in the worker context, async_free_zspage() can't simply lock all pages in lock_zspage(), it's still trylock because of the race between zs_free() and zs_page_migrate(). Please see the commit 2505a981114d ("zsmalloc: fix races between asynchronous zspage free and page migration") for details. Actually, all free_zspage() needs is to get zspage from page safely, we can use RCU to achieve it easily. Then free_zspage() don't need to hold locks of all pages, so don't need the deferred free mechanism at all. The updated zs_page_migrate() now has two more cases to consider: 1. get_zspage_lockless() return NULL: it means free_zspage() has used reset_page() on this page and its reference of page. 2. get_zspage_lockless() return zspage but it's not on pool list: it means zspage has been removed from list and in process of free. I'm not sure what value should be returned in these cases? -EINVAL or -EAGAIN or other value? If the migration caller can find that page has no extra referenced and can just free it, I think we should return -EAGAIN to let the migration caller retry this page later to free it. Now I choose to use -EINVAL to skip migration of this page, it seems not a big deal to fail migration of some pages? Signed-off-by: Chengming Zhou --- mm/zsmalloc.c | 97 ++++++++++++++++++++++++++++++++++------------------------- 1 file changed, 56 insertions(+), 41 deletions(-) diff --git a/mm/zsmalloc.c b/mm/zsmalloc.c index 63ec385cd670..b153f2e5fc0f 100644 --- a/mm/zsmalloc.c +++ b/mm/zsmalloc.c @@ -253,6 +253,7 @@ struct zspage { struct list_head list; /* fullness list */ struct zs_pool *pool; rwlock_t lock; + struct rcu_head rcu_head; }; struct mapping_area { @@ -310,6 +311,8 @@ static int create_cache(struct zs_pool *pool) static void destroy_cache(struct zs_pool *pool) { kmem_cache_destroy(pool->handle_cachep); + /* Synchronize RCU to free zspage. */ + synchronize_rcu(); kmem_cache_destroy(pool->zspage_cachep); } @@ -335,6 +338,14 @@ static void cache_free_zspage(struct zs_pool *pool, struct zspage *zspage) kmem_cache_free(pool->zspage_cachep, zspage); } +static void rcu_free_zspage(struct rcu_head *h) +{ + struct zspage *zspage = container_of(h, struct zspage, rcu_head); + struct zs_pool *pool = zspage->pool; + + kmem_cache_free(pool->zspage_cachep, zspage); +} + /* pool->lock(which owns the handle) synchronizes races */ static void record_obj(unsigned long handle, unsigned long obj) { @@ -710,14 +721,31 @@ static int fix_fullness_group(struct size_class *class, struct zspage *zspage) return newfg; } +static void set_zspage(struct page *page, struct zspage *zspage) +{ + struct zspage __rcu **private = (struct zspage __rcu **)&page->private; + + rcu_assign_pointer(*private, zspage); +} + static struct zspage *get_zspage(struct page *page) { - struct zspage *zspage = (struct zspage *)page_private(page); + struct zspage __rcu **private = (struct zspage __rcu **)&page->private; + struct zspage *zspage; + zspage = rcu_dereference_protected(*private, true); BUG_ON(zspage->magic != ZSPAGE_MAGIC); return zspage; } +/* Only used in zs_page_migrate() to get zspage locklessly. */ +static struct zspage *get_zspage_lockless(struct page *page) +{ + struct zspage __rcu **private = (struct zspage __rcu **)&page->private; + + return rcu_dereference(*private); +} + static struct page *get_next_page(struct page *page) { struct zspage *zspage = get_zspage(page); @@ -793,32 +821,11 @@ static void reset_page(struct page *page) { __ClearPageMovable(page); ClearPagePrivate(page); - set_page_private(page, 0); + set_zspage(page, NULL); page_mapcount_reset(page); page->index = 0; } -static int trylock_zspage(struct zspage *zspage) -{ - struct page *cursor, *fail; - - for (cursor = get_first_page(zspage); cursor != NULL; cursor = - get_next_page(cursor)) { - if (!trylock_page(cursor)) { - fail = cursor; - goto unlock; - } - } - - return 1; -unlock: - for (cursor = get_first_page(zspage); cursor != fail; cursor = - get_next_page(cursor)) - unlock_page(cursor); - - return 0; -} - static void __free_zspage(struct zs_pool *pool, struct size_class *class, struct zspage *zspage) { @@ -834,13 +841,12 @@ static void __free_zspage(struct zs_pool *pool, struct size_class *class, VM_BUG_ON_PAGE(!PageLocked(page), page); next = get_next_page(page); reset_page(page); - unlock_page(page); dec_zone_page_state(page, NR_ZSPAGES); put_page(page); page = next; } while (page != NULL); - cache_free_zspage(pool, zspage); + call_rcu(&zspage->rcu_head, rcu_free_zspage); class_stat_dec(class, ZS_OBJS_ALLOCATED, class->objs_per_zspage); atomic_long_sub(class->pages_per_zspage, &pool->pages_allocated); @@ -852,16 +858,6 @@ static void free_zspage(struct zs_pool *pool, struct size_class *class, VM_BUG_ON(get_zspage_inuse(zspage)); VM_BUG_ON(list_empty(&zspage->list)); - /* - * Since zs_free couldn't be sleepable, this function cannot call - * lock_page. The page locks trylock_zspage got will be released - * by __free_zspage. - */ - if (!trylock_zspage(zspage)) { - kick_deferred_free(pool); - return; - } - remove_zspage(class, zspage); __free_zspage(pool, class, zspage); } @@ -929,7 +925,7 @@ static void create_page_chain(struct size_class *class, struct zspage *zspage, */ for (i = 0; i < nr_pages; i++) { page = pages[i]; - set_page_private(page, (unsigned long)zspage); + set_zspage(page, zspage); page->index = 0; if (i == 0) { zspage->first_page = page; @@ -978,10 +974,11 @@ static struct zspage *alloc_zspage(struct zs_pool *pool, pages[i] = page; } - create_page_chain(class, zspage, pages); init_zspage(class, zspage); zspage->pool = pool; zspage->class = class->index; + /* RCU set_zspage() after zspage initialized. */ + create_page_chain(class, zspage, pages); return zspage; } @@ -1765,17 +1762,35 @@ static int zs_page_migrate(struct page *newpage, struct page *page, VM_BUG_ON_PAGE(!PageIsolated(page), page); - /* The page is locked, so this pointer must remain valid */ - zspage = get_zspage(page); - pool = zspage->pool; + rcu_read_lock(); + zspage = get_zspage_lockless(page); + if (!zspage) { + rcu_read_unlock(); + return -EINVAL; + } /* * The pool's lock protects the race between zpage migration - * and zs_free. + * and zs_free. We check if the zspage is still in pool with + * pool->lock protection. If the zspage isn't in pool anymore, + * it should be freed by RCU soon. */ + pool = zspage->pool; spin_lock(&pool->lock); class = zspage_class(pool, zspage); + if (list_empty(&zspage->list)) { + spin_unlock(&pool->lock); + rcu_read_unlock(); + return -EINVAL; + } + + /* + * Now the zspage is still on pool, and we held pool->lock, + * it can't be freed in the meantime. + */ + rcu_read_unlock(); + /* the migrate_write_lock protects zpage access via zs_map_object */ migrate_write_lock(zspage);