From patchwork Mon Mar 27 20:11:24 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Mike Snitzer X-Patchwork-Id: 13190046 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 vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id 4E26CC77B6C for ; Mon, 27 Mar 2023 20:14:01 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S231841AbjC0UOA (ORCPT ); Mon, 27 Mar 2023 16:14:00 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:38764 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S231892AbjC0UN7 (ORCPT ); Mon, 27 Mar 2023 16:13:59 -0400 Received: from mail-qt1-f171.google.com (mail-qt1-f171.google.com [209.85.160.171]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id 8ACC135A7 for ; Mon, 27 Mar 2023 13:13:09 -0700 (PDT) Received: by mail-qt1-f171.google.com with SMTP id hf2so9810792qtb.3 for ; Mon, 27 Mar 2023 13:13:09 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; t=1679947988; h=content-transfer-encoding:mime-version:references:in-reply-to :message-id:date:subject:cc:to:from:x-gm-message-state:from:to:cc :subject:date:message-id:reply-to; bh=pNKDXzqz05HkFrPteMNUMY0uT7UJYCgf272zyuZUZl4=; b=P2bP3zxprxxRbhvZPiIj0ofz6A9NEGq1/HdfjVyX+dWLO4+sjCFNLzksQ7KMgC1O7l /eoeV0UfoPtsZfjWgWzvLotchU43qk/sb94saZqwYteSqUYMF6qKsFlgKdcgOLJIdDJP awLrXsSC+O7Jr6JRI4r/p9HlnMEL1OPTCcl2Z4I92WhPUX6fqjbxGALCDIBhXa5yj8NA VYcW3KX1f0dMwxw97VeVM8tnVrArMi4pgu5v2CHBSK0u95Rr1iFwahmMpaBBA280j+vy /x7d98NoIWClOgU0+OkD7RyOUAJZe0hnwWImBHChF2CUzyQL5Gpphg8BW5HV6PoEqabz Czkg== X-Gm-Message-State: AO0yUKUV9SWGAtY4ClYksf/qKym+kTzBYhUkEmWHAUIak/CcV4TxRH8D gUdbQGSZgQnm25oo4Lo1v1+A X-Google-Smtp-Source: AK7set+7MMoq6Vd0Xdcf9qnBj1WT4JAtfpuUnRj7pcmV6oOeji14qNAy6w9DZt923oR8XoqLbUpKMQ== X-Received: by 2002:ac8:5f0b:0:b0:3e1:c341:f618 with SMTP id x11-20020ac85f0b000000b003e1c341f618mr20781332qta.65.1679947988341; Mon, 27 Mar 2023 13:13:08 -0700 (PDT) Received: from localhost (pool-68-160-166-30.bstnma.fios.verizon.net. [68.160.166.30]) by smtp.gmail.com with ESMTPSA id 128-20020a370486000000b00746772d78a6sm11397199qke.2.2023.03.27.13.13.07 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Mon, 27 Mar 2023 13:13:07 -0700 (PDT) From: Mike Snitzer To: dm-devel@redhat.com Cc: linux-block@vger.kernel.org, axboe@kernel.dk, ejt@redhat.com, mpatocka@redhat.com, heinzm@redhat.com, nhuck@google.com, ebiggers@kernel.org, keescook@chromium.org, luomeng12@huawei.com, Mike Snitzer Subject: [dm-6.4 PATCH v3 01/20] dm bufio: remove unused dm_bufio_release_move interface Date: Mon, 27 Mar 2023 16:11:24 -0400 Message-Id: <20230327201143.51026-2-snitzer@kernel.org> X-Mailer: git-send-email 2.40.0 In-Reply-To: <20230327201143.51026-1-snitzer@kernel.org> References: <20230327201143.51026-1-snitzer@kernel.org> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-block@vger.kernel.org From: Joe Thornber Was used by multi-snapshot DM target that never went upstream. Signed-off-by: Joe Thornber Acked-by: Mikulas Patocka Signed-off-by: Mike Snitzer --- drivers/md/dm-bufio.c | 77 ---------------------------------------- include/linux/dm-bufio.h | 6 ---- 2 files changed, 83 deletions(-) diff --git a/drivers/md/dm-bufio.c b/drivers/md/dm-bufio.c index cf077f9b30c3..79434b38f368 100644 --- a/drivers/md/dm-bufio.c +++ b/drivers/md/dm-bufio.c @@ -1415,83 +1415,6 @@ int dm_bufio_issue_discard(struct dm_bufio_client *c, sector_t block, sector_t c } EXPORT_SYMBOL_GPL(dm_bufio_issue_discard); -/* - * We first delete any other buffer that may be at that new location. - * - * Then, we write the buffer to the original location if it was dirty. - * - * Then, if we are the only one who is holding the buffer, relink the buffer - * in the buffer tree for the new location. - * - * If there was someone else holding the buffer, we write it to the new - * location but not relink it, because that other user needs to have the buffer - * at the same place. - */ -void dm_bufio_release_move(struct dm_buffer *b, sector_t new_block) -{ - struct dm_bufio_client *c = b->c; - struct dm_buffer *new; - - BUG_ON(dm_bufio_in_request()); - - dm_bufio_lock(c); - -retry: - new = __find(c, new_block); - if (new) { - if (new->hold_count) { - __wait_for_free_buffer(c); - goto retry; - } - - /* - * FIXME: Is there any point waiting for a write that's going - * to be overwritten in a bit? - */ - __make_buffer_clean(new); - __unlink_buffer(new); - __free_buffer_wake(new); - } - - BUG_ON(!b->hold_count); - BUG_ON(test_bit(B_READING, &b->state)); - - __write_dirty_buffer(b, NULL); - if (b->hold_count == 1) { - wait_on_bit_io(&b->state, B_WRITING, - TASK_UNINTERRUPTIBLE); - set_bit(B_DIRTY, &b->state); - b->dirty_start = 0; - b->dirty_end = c->block_size; - __unlink_buffer(b); - __link_buffer(b, new_block, LIST_DIRTY); - } else { - sector_t old_block; - - wait_on_bit_lock_io(&b->state, B_WRITING, - TASK_UNINTERRUPTIBLE); - /* - * Relink buffer to "new_block" so that write_callback - * sees "new_block" as a block number. - * After the write, link the buffer back to old_block. - * All this must be done in bufio lock, so that block number - * change isn't visible to other threads. - */ - old_block = b->block; - __unlink_buffer(b); - __link_buffer(b, new_block, b->list_mode); - submit_io(b, REQ_OP_WRITE, write_endio); - wait_on_bit_io(&b->state, B_WRITING, - TASK_UNINTERRUPTIBLE); - __unlink_buffer(b); - __link_buffer(b, old_block, b->list_mode); - } - - dm_bufio_unlock(c); - dm_bufio_release(b); -} -EXPORT_SYMBOL_GPL(dm_bufio_release_move); - static void forget_buffer_locked(struct dm_buffer *b) { if (likely(!b->hold_count) && likely(!smp_load_acquire(&b->state))) { diff --git a/include/linux/dm-bufio.h b/include/linux/dm-bufio.h index 2056743aaaaa..681656a1c03d 100644 --- a/include/linux/dm-bufio.h +++ b/include/linux/dm-bufio.h @@ -130,12 +130,6 @@ int dm_bufio_issue_flush(struct dm_bufio_client *c); */ int dm_bufio_issue_discard(struct dm_bufio_client *c, sector_t block, sector_t count); -/* - * Like dm_bufio_release but also move the buffer to the new - * block. dm_bufio_write_dirty_buffers is needed to commit the new block. - */ -void dm_bufio_release_move(struct dm_buffer *b, sector_t new_block); - /* * Free the given buffer. * This is just a hint, if the buffer is in use or dirty, this function From patchwork Mon Mar 27 20:11:25 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Mike Snitzer X-Patchwork-Id: 13190047 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 vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id CA7B1C77B60 for ; Mon, 27 Mar 2023 20:14:00 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S229940AbjC0UN7 (ORCPT ); Mon, 27 Mar 2023 16:13:59 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:38814 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S231841AbjC0UN7 (ORCPT ); Mon, 27 Mar 2023 16:13:59 -0400 Received: from mail-qv1-f51.google.com (mail-qv1-f51.google.com [209.85.219.51]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id A3E0B359C for ; Mon, 27 Mar 2023 13:13:10 -0700 (PDT) Received: by mail-qv1-f51.google.com with SMTP id cu4so7692645qvb.3 for ; Mon, 27 Mar 2023 13:13:10 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; t=1679947990; h=content-transfer-encoding:mime-version:references:in-reply-to :message-id:date:subject:cc:to:from:x-gm-message-state:from:to:cc :subject:date:message-id:reply-to; bh=K6rad/3rQ5uNwjgAbrRZn8viSEaiHhaJsk57wekX4Yc=; b=wVstvHlM3mXjpfnzeRKpiVOg4slUAX7toeFE8xNnbrThgO9z5p9BIfLNafqKCmv4fB B4mT3lm/qZU9ouwApFL7H5+VoNM3Ymh81+dis1/fG6bAOi6qm6nGH9izCn+OSzDtS/C9 gWe4wxvNQmkU54rGwkoljPll6hub/WQ8L4xwIoMYToTAJxBLRD+MaU8atsN0Cc5k13Q7 10gcdGgezTLFg8rrVLzxwFGH6buFOCdu/8OasZN/BrGPXg09OKXzkkjEZxomPSiiCAXj El4QXpaQ8mUCoa5P+ZBv/TNVowyT2BYpftvOlNx7A2xf3NNDH5y2Zla3qmRkTGdRrz0o QkLA== X-Gm-Message-State: AAQBX9c2UPS8f/8f8JxUnGb1DTBGOAyDXI5pdQBt6TosWe2i/E/XMtZI BmaxjZtkSfpJRLAHCQzKkFmQ X-Google-Smtp-Source: AKy350aGPv+m7+Nvtksg720HQ297PPqoxtEn/b3IGQEXHEExxXsRvNIRRAb6T0t3WIyj8iHk9azUFg== X-Received: by 2002:a05:6214:b64:b0:5b8:1f61:a20 with SMTP id ey4-20020a0562140b6400b005b81f610a20mr25404563qvb.35.1679947989792; Mon, 27 Mar 2023 13:13:09 -0700 (PDT) Received: from localhost (pool-68-160-166-30.bstnma.fios.verizon.net. [68.160.166.30]) by smtp.gmail.com with ESMTPSA id f16-20020ac86ed0000000b003e390b48958sm4723408qtv.55.2023.03.27.13.13.09 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Mon, 27 Mar 2023 13:13:09 -0700 (PDT) From: Mike Snitzer To: dm-devel@redhat.com Cc: linux-block@vger.kernel.org, axboe@kernel.dk, ejt@redhat.com, mpatocka@redhat.com, heinzm@redhat.com, nhuck@google.com, ebiggers@kernel.org, keescook@chromium.org, luomeng12@huawei.com, Mike Snitzer Subject: [dm-6.4 PATCH v3 02/20] dm bufio: use WARN_ON in dm_bufio_client_destroy and dm_bufio_exit Date: Mon, 27 Mar 2023 16:11:25 -0400 Message-Id: <20230327201143.51026-3-snitzer@kernel.org> X-Mailer: git-send-email 2.40.0 In-Reply-To: <20230327201143.51026-1-snitzer@kernel.org> References: <20230327201143.51026-1-snitzer@kernel.org> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-block@vger.kernel.org Using BUG_ON when tearing down is excessive. Relax these to WARN_ONs. Signed-off-by: Mike Snitzer --- drivers/md/dm-bufio.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/drivers/md/dm-bufio.c b/drivers/md/dm-bufio.c index 79434b38f368..dac9f1f84c34 100644 --- a/drivers/md/dm-bufio.c +++ b/drivers/md/dm-bufio.c @@ -1828,8 +1828,8 @@ void dm_bufio_client_destroy(struct dm_bufio_client *c) mutex_unlock(&dm_bufio_clients_lock); - BUG_ON(!RB_EMPTY_ROOT(&c->buffer_tree)); - BUG_ON(c->need_reserved_buffers); + WARN_ON(!RB_EMPTY_ROOT(&c->buffer_tree)); + WARN_ON(c->need_reserved_buffers); while (!list_empty(&c->reserved_buffers)) { struct dm_buffer *b = list_entry(c->reserved_buffers.next, @@ -1843,7 +1843,7 @@ void dm_bufio_client_destroy(struct dm_bufio_client *c) DMERR("leaked buffer count %d: %ld", i, c->n_buffers[i]); for (i = 0; i < LIST_SIZE; i++) - BUG_ON(c->n_buffers[i]); + WARN_ON(c->n_buffers[i]); kmem_cache_destroy(c->slab_cache); kmem_cache_destroy(c->slab_buffer); @@ -2082,7 +2082,7 @@ static void __exit dm_bufio_exit(void) bug = 1; } - BUG_ON(bug); + WARN_ON(bug); /* leaks are not worth crashing the system */ } module_init(dm_bufio_init) From patchwork Mon Mar 27 20:11:26 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Mike Snitzer X-Patchwork-Id: 13190050 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 vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id C4D3EC77B6C for ; Mon, 27 Mar 2023 20:14:05 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S231892AbjC0UOE (ORCPT ); Mon, 27 Mar 2023 16:14:04 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:38854 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S232071AbjC0UOD (ORCPT ); Mon, 27 Mar 2023 16:14:03 -0400 Received: from mail-qv1-f44.google.com (mail-qv1-f44.google.com [209.85.219.44]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id 37E9B1BF8 for ; Mon, 27 Mar 2023 13:13:12 -0700 (PDT) Received: by mail-qv1-f44.google.com with SMTP id jl13so7608649qvb.10 for ; Mon, 27 Mar 2023 13:13:12 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; t=1679947991; h=content-transfer-encoding:mime-version:references:in-reply-to :message-id:date:subject:cc:to:from:x-gm-message-state:from:to:cc :subject:date:message-id:reply-to; bh=38Sgxe5/q+hJ0JKVFdOX1S2/H0qUacEb3wrSiTwviHY=; b=U0wP3Nugs8NAQG9ZTr1OX2UVkJRtnEISnGUfIeLnB0VvBvrhdTSQdlLBzD1R378pm/ SdMOPoO7IR/EAfmtGrc7Ti8ytbwcSx0FAYE0AQLfJUSYKCWddK6ujdpUXJny9T5qIh8S KfVnSZEkRCQiS/Uu8dkLtslO2tWV4wOseHCjrLzgnthOPuZur1l7LuQrxjsub8oxu1Ye dpGjy9ddBuzKhNYNVLQy8Q40+z5WX+MJ8+CTTC+1LxsN21YOqfoTrnkY92GOkVvzSlBT bYA4t9diB7LBBrXPOClK9G9THkiilf+7/DI93sztTrMLTC801tl3J4Q+S0KP0W2qje8S Y5mw== X-Gm-Message-State: AAQBX9e7++MQm/p+WochG/OHNl3JIwaKh4Jys0TXyxCsepfXfu3Rfwa5 tfjdkJxI9iXEPVhXUdq82ErH X-Google-Smtp-Source: AKy350Z/SX3NEFg9tn3FY6AM6zZMl6uXUeBw6GhmbRlqi/g3goSm+tkfNNjvAC/CMiQFi0MxHMsbFw== X-Received: by 2002:ad4:5d69:0:b0:5db:e06f:e0f9 with SMTP id fn9-20020ad45d69000000b005dbe06fe0f9mr19745867qvb.13.1679947991288; Mon, 27 Mar 2023 13:13:11 -0700 (PDT) Received: from localhost (pool-68-160-166-30.bstnma.fios.verizon.net. [68.160.166.30]) by smtp.gmail.com with ESMTPSA id a200-20020ae9e8d1000000b00747d211536dsm3327118qkg.107.2023.03.27.13.13.10 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Mon, 27 Mar 2023 13:13:10 -0700 (PDT) From: Mike Snitzer To: dm-devel@redhat.com Cc: linux-block@vger.kernel.org, axboe@kernel.dk, ejt@redhat.com, mpatocka@redhat.com, heinzm@redhat.com, nhuck@google.com, ebiggers@kernel.org, keescook@chromium.org, luomeng12@huawei.com, Mike Snitzer Subject: [dm-6.4 PATCH v3 03/20] dm bufio: never crash if dm_bufio_in_request() Date: Mon, 27 Mar 2023 16:11:26 -0400 Message-Id: <20230327201143.51026-4-snitzer@kernel.org> X-Mailer: git-send-email 2.40.0 In-Reply-To: <20230327201143.51026-1-snitzer@kernel.org> References: <20230327201143.51026-1-snitzer@kernel.org> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-block@vger.kernel.org All these instances are entirely avoidable given that they speak to coding mistakes that result in inappropriate use. Proper testing during development will catch any such coding bug so its best to relax all of these from BUG_ON to WARN_ON_ONCE. Signed-off-by: Mike Snitzer --- drivers/md/dm-bufio.c | 18 ++++++++++++------ 1 file changed, 12 insertions(+), 6 deletions(-) diff --git a/drivers/md/dm-bufio.c b/drivers/md/dm-bufio.c index dac9f1f84c34..d63f94ab1d9f 100644 --- a/drivers/md/dm-bufio.c +++ b/drivers/md/dm-bufio.c @@ -1152,7 +1152,8 @@ EXPORT_SYMBOL_GPL(dm_bufio_get); void *dm_bufio_read(struct dm_bufio_client *c, sector_t block, struct dm_buffer **bp) { - BUG_ON(dm_bufio_in_request()); + if (WARN_ON_ONCE(dm_bufio_in_request())) + return ERR_PTR(-EINVAL); return new_read(c, block, NF_READ, bp); } @@ -1161,7 +1162,8 @@ EXPORT_SYMBOL_GPL(dm_bufio_read); void *dm_bufio_new(struct dm_bufio_client *c, sector_t block, struct dm_buffer **bp) { - BUG_ON(dm_bufio_in_request()); + if (WARN_ON_ONCE(dm_bufio_in_request())) + return ERR_PTR(-EINVAL); return new_read(c, block, NF_FRESH, bp); } @@ -1174,7 +1176,8 @@ void dm_bufio_prefetch(struct dm_bufio_client *c, LIST_HEAD(write_list); - BUG_ON(dm_bufio_in_request()); + if (WARN_ON_ONCE(dm_bufio_in_request())) + return; /* should never happen */ blk_start_plug(&plug); dm_bufio_lock(c); @@ -1281,7 +1284,8 @@ void dm_bufio_write_dirty_buffers_async(struct dm_bufio_client *c) { LIST_HEAD(write_list); - BUG_ON(dm_bufio_in_request()); + if (WARN_ON_ONCE(dm_bufio_in_request())) + return; /* should never happen */ dm_bufio_lock(c); __write_dirty_buffers_async(c, 0, &write_list); @@ -1386,7 +1390,8 @@ int dm_bufio_issue_flush(struct dm_bufio_client *c) .count = 0, }; - BUG_ON(dm_bufio_in_request()); + if (WARN_ON_ONCE(dm_bufio_in_request())) + return -EINVAL; return dm_io(&io_req, 1, &io_reg, NULL); } @@ -1409,7 +1414,8 @@ int dm_bufio_issue_discard(struct dm_bufio_client *c, sector_t block, sector_t c .count = block_to_sector(c, count), }; - BUG_ON(dm_bufio_in_request()); + if (WARN_ON_ONCE(dm_bufio_in_request())) + return -EINVAL; /* discards are optional */ return dm_io(&io_req, 1, &io_reg, NULL); } From patchwork Mon Mar 27 20:11:27 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Mike Snitzer X-Patchwork-Id: 13190048 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 vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id A29C3C76195 for ; Mon, 27 Mar 2023 20:14:04 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S232120AbjC0UOD (ORCPT ); Mon, 27 Mar 2023 16:14:03 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:38786 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S231892AbjC0UOC (ORCPT ); Mon, 27 Mar 2023 16:14:02 -0400 Received: from mail-qt1-f181.google.com (mail-qt1-f181.google.com [209.85.160.181]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id C62021FE9 for ; Mon, 27 Mar 2023 13:13:13 -0700 (PDT) Received: by mail-qt1-f181.google.com with SMTP id cr18so5981373qtb.0 for ; Mon, 27 Mar 2023 13:13:13 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; t=1679947993; h=content-transfer-encoding:mime-version:references:in-reply-to :message-id:date:subject:cc:to:from:x-gm-message-state:from:to:cc :subject:date:message-id:reply-to; bh=hmqlLAmAcoDYlM73GWg+waCa1PP8Bn6cws6QITx68xo=; b=ScCmgyHA2M/AS8xpVObhUJXpXwjXBbm4S5OJaXCOvvcRwiiKMFOZxhJ+mVvGQLIyMR vkM0RxyZRY4+1gVW4OzY6SysNkU0pRpcVj2tV6VSzb3Sx9mmPKq9XYu7UUoC+BgQVI36 pgZM67ei1Hr9OtFgtGOktYOIFNPx/uolhIs5LXzT2wAiyUAlxXe6JCJS79XWw8h+ZCCh TT7rtuGKf+vsc2J3dBKDnf/N/4MPfX2kBflakgThAb5kkxFjVro24rjD5nSse0mx2vl0 rsAspN8/CsJ/qxkI3lDjFk9k7/2Mr3VWXxv/wsQGQo/zq7AxOLBeeWCyG+2hNeQfFIvX vmJg== X-Gm-Message-State: AAQBX9dZUOlRgPDeWbB5dT5T93dBulf+BtcIqlAJkxGdKxOqoAGJCOlp bDCMHmCIDSvdsDXC2UiCeoZib5aLh5uYVJfv7RGT X-Google-Smtp-Source: AKy350b9yynzcGT7R923zG1Nqhk3Mn4VUNQG7FVDF/2AYEYBKvcrKmzGtyrZXsaF2wG0qO6dp7ngxQ== X-Received: by 2002:a05:622a:1002:b0:3e4:dbfa:ee2b with SMTP id d2-20020a05622a100200b003e4dbfaee2bmr15465044qte.17.1679947992941; Mon, 27 Mar 2023 13:13:12 -0700 (PDT) Received: from localhost (pool-68-160-166-30.bstnma.fios.verizon.net. [68.160.166.30]) by smtp.gmail.com with ESMTPSA id 141-20020a370793000000b0074672975d5csm12784571qkh.91.2023.03.27.13.13.12 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Mon, 27 Mar 2023 13:13:12 -0700 (PDT) From: Mike Snitzer To: dm-devel@redhat.com Cc: linux-block@vger.kernel.org, axboe@kernel.dk, ejt@redhat.com, mpatocka@redhat.com, heinzm@redhat.com, nhuck@google.com, ebiggers@kernel.org, keescook@chromium.org, luomeng12@huawei.com, Mike Snitzer Subject: [dm-6.4 PATCH v3 04/20] dm bufio: don't bug for clear developer oversight Date: Mon, 27 Mar 2023 16:11:27 -0400 Message-Id: <20230327201143.51026-5-snitzer@kernel.org> X-Mailer: git-send-email 2.40.0 In-Reply-To: <20230327201143.51026-1-snitzer@kernel.org> References: <20230327201143.51026-1-snitzer@kernel.org> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-block@vger.kernel.org Reasonable to relax to WARN_ON because these are easily avoided but do offer some assurance future coding mistakes won't occur (if changes tested properly). Signed-off-by: Mike Snitzer --- drivers/md/dm-bufio.c | 13 ++++++++----- 1 file changed, 8 insertions(+), 5 deletions(-) diff --git a/drivers/md/dm-bufio.c b/drivers/md/dm-bufio.c index d63f94ab1d9f..64fb5fd39bd9 100644 --- a/drivers/md/dm-bufio.c +++ b/drivers/md/dm-bufio.c @@ -378,8 +378,10 @@ static void adjust_total_allocated(struct dm_buffer *b, bool unlink) */ static void __cache_size_refresh(void) { - BUG_ON(!mutex_is_locked(&dm_bufio_clients_lock)); - BUG_ON(dm_bufio_client_count < 0); + if (WARN_ON(!mutex_is_locked(&dm_bufio_clients_lock))) + return; + if (WARN_ON(dm_bufio_client_count < 0)) + return; dm_bufio_cache_size_latch = READ_ONCE(dm_bufio_cache_size); @@ -1536,7 +1538,8 @@ static void drop_buffers(struct dm_bufio_client *c) int i; bool warned = false; - BUG_ON(dm_bufio_in_request()); + if (WARN_ON(dm_bufio_in_request())) + return; /* should never happen */ /* * An optimization so that the buffers are not written one-by-one. @@ -1556,7 +1559,7 @@ static void drop_buffers(struct dm_bufio_client *c) (unsigned long long)b->block, b->hold_count, i); #ifdef CONFIG_DM_DEBUG_BLOCK_STACK_TRACING stack_trace_print(b->stack_entries, b->stack_len, 1); - /* mark unclaimed to avoid BUG_ON below */ + /* mark unclaimed to avoid WARN_ON below */ b->hold_count = 0; #endif } @@ -1567,7 +1570,7 @@ static void drop_buffers(struct dm_bufio_client *c) #endif for (i = 0; i < LIST_SIZE; i++) - BUG_ON(!list_empty(&c->lru[i])); + WARN_ON(!list_empty(&c->lru[i])); dm_bufio_unlock(c); } From patchwork Mon Mar 27 20:11:28 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Mike Snitzer X-Patchwork-Id: 13190049 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 vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id 2D1DDC77B6E for ; Mon, 27 Mar 2023 20:14:05 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S232134AbjC0UOD (ORCPT ); Mon, 27 Mar 2023 16:14:03 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:38802 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S231929AbjC0UOC (ORCPT ); Mon, 27 Mar 2023 16:14:02 -0400 Received: from mail-qv1-f44.google.com (mail-qv1-f44.google.com [209.85.219.44]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id E17BA211E for ; Mon, 27 Mar 2023 13:13:14 -0700 (PDT) Received: by mail-qv1-f44.google.com with SMTP id jl13so7608756qvb.10 for ; Mon, 27 Mar 2023 13:13:14 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; t=1679947994; h=content-transfer-encoding:mime-version:references:in-reply-to :message-id:date:subject:cc:to:from:x-gm-message-state:from:to:cc :subject:date:message-id:reply-to; bh=HnRXjPFL8ryVwQIZKj/idU8AtLLYydcfQZ5/bZxBP7E=; b=fq4sTO2+62wAlRztKwokTFj/Lv4Myvm/HFc5PVqLDCwpPyxrah80GE9YfHiKLNEiPh lPocyoxnIud2ZMr3pYYS7pZaLtkwylenLtD5D2IX8nYVZfmyOgzUCXnr8p6pTzwKr67A /PgrBUAUm0KppihrH8rz72DrunV1fntxCqxZothI//aD6ovwIpgzAhCuyp/reTMzDiwD kSDr4ddGlEGeekRz1ZgtSExQgcEZvXmJzDmVyM/iCEvVx3FLNk7D0gvt1SRk/kjRbErm vVQ9Bh8xej1gO8vCBbjK2+ao6KQzT1JrjuINl7teYOOB8TZMB1rb6i7/EbUTT2E3+MLi k77A== X-Gm-Message-State: AAQBX9dDobLdYBdYJvEr571ddbPcAtWYGYMIAjq9VplMcNX7xu/iAksA Og8ScJgfnxaai8qe1m16q4f+ X-Google-Smtp-Source: AKy350aV2shDFEGq+y0+P6CuzF0QHTPYTOXNK5SCBfMQ/nybOe1LXkh957qYAcLDjaI6537aBHt8UA== X-Received: by 2002:ad4:5dec:0:b0:5ad:5698:848e with SMTP id jn12-20020ad45dec000000b005ad5698848emr19825525qvb.48.1679947994512; Mon, 27 Mar 2023 13:13:14 -0700 (PDT) Received: from localhost (pool-68-160-166-30.bstnma.fios.verizon.net. [68.160.166.30]) by smtp.gmail.com with ESMTPSA id d13-20020a0cf6cd000000b005dd8b9345d3sm3161240qvo.107.2023.03.27.13.13.13 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Mon, 27 Mar 2023 13:13:14 -0700 (PDT) From: Mike Snitzer To: dm-devel@redhat.com Cc: linux-block@vger.kernel.org, axboe@kernel.dk, ejt@redhat.com, mpatocka@redhat.com, heinzm@redhat.com, nhuck@google.com, ebiggers@kernel.org, keescook@chromium.org, luomeng12@huawei.com, Mike Snitzer Subject: [dm-6.4 PATCH v3 05/20] dm bufio: add LRU abstraction Date: Mon, 27 Mar 2023 16:11:28 -0400 Message-Id: <20230327201143.51026-6-snitzer@kernel.org> X-Mailer: git-send-email 2.40.0 In-Reply-To: <20230327201143.51026-1-snitzer@kernel.org> References: <20230327201143.51026-1-snitzer@kernel.org> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-block@vger.kernel.org From: Joe Thornber A CLOCK algorithm is used in this LRU abstraction. This avoids relinking list nodes, which would require a write lock protecting it. None of the LRU methods are threadsafe; locking must be done at a higher level. Code that uses this new LRU will be introduced in the next 2 commits. As such, this commit will cause "defined but not used" compiler warnings that will be resolved by the next 2 commits. Signed-off-by: Joe Thornber Signed-off-by: Mike Snitzer --- drivers/md/dm-bufio.c | 235 ++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 235 insertions(+) diff --git a/drivers/md/dm-bufio.c b/drivers/md/dm-bufio.c index 64fb5fd39bd9..a0bb0de0d4e7 100644 --- a/drivers/md/dm-bufio.c +++ b/drivers/md/dm-bufio.c @@ -66,6 +66,241 @@ #define LIST_DIRTY 1 #define LIST_SIZE 2 +/*--------------------------------------------------------------*/ + +/* + * Rather than use an LRU list, we use a clock algorithm where entries + * are held in a circular list. When an entry is 'hit' a reference bit + * is set. The least recently used entry is approximated by running a + * cursor around the list selecting unreferenced entries. Referenced + * entries have their reference bit cleared as the cursor passes them. + */ +struct lru_entry { + struct list_head list; + atomic_t referenced; +}; + +struct lru_iter { + struct lru *lru; + struct list_head list; + struct lru_entry *stop; + struct lru_entry *e; +}; + +struct lru { + struct list_head *cursor; + unsigned long count; + + struct list_head iterators; +}; + +/*--------------*/ + +static void lru_init(struct lru *lru) +{ + lru->cursor = NULL; + lru->count = 0; + INIT_LIST_HEAD(&lru->iterators); +} + +static void lru_destroy(struct lru *lru) +{ + WARN_ON_ONCE(lru->cursor); + WARN_ON_ONCE(!list_empty(&lru->iterators)); +} + +/* + * Insert a new entry into the lru. + */ +static void lru_insert(struct lru *lru, struct lru_entry *le) +{ + /* + * Don't be tempted to set to 1, makes the lru aspect + * perform poorly. + */ + atomic_set(&le->referenced, 0); + + if (lru->cursor) { + list_add_tail(&le->list, lru->cursor); + } else { + INIT_LIST_HEAD(&le->list); + lru->cursor = &le->list; + } + lru->count++; +} + +/*--------------*/ + +/* + * Convert a list_head pointer to an lru_entry pointer. + */ +static inline struct lru_entry *to_le(struct list_head *l) +{ + return container_of(l, struct lru_entry, list); +} + +/* + * Initialize an lru_iter and add it to the list of cursors in the lru. + */ +static void lru_iter_begin(struct lru *lru, struct lru_iter *it) +{ + it->lru = lru; + it->stop = lru->cursor ? to_le(lru->cursor->prev) : NULL; + it->e = lru->cursor ? to_le(lru->cursor) : NULL; + list_add(&it->list, &lru->iterators); +} + +/* + * Remove an lru_iter from the list of cursors in the lru. + */ +static inline void lru_iter_end(struct lru_iter *it) +{ + list_del(&it->list); +} + +/* Predicate function type to be used with lru_iter_next */ +typedef bool (*iter_predicate)(struct lru_entry *le, void *context); + +/* + * Advance the cursor to the next entry that passes the + * predicate, and return that entry. Returns NULL if the + * iteration is complete. + */ +static struct lru_entry *lru_iter_next(struct lru_iter *it, + iter_predicate pred, void *context) +{ + struct lru_entry *e; + + while (it->e) { + e = it->e; + + /* advance the cursor */ + if (it->e == it->stop) + it->e = NULL; + else + it->e = to_le(it->e->list.next); + + if (pred(e, context)) + return e; + } + + return NULL; +} + +/* + * Invalidate a specific lru_entry and update all cursors in + * the lru accordingly. + */ +static void lru_iter_invalidate(struct lru *lru, struct lru_entry *e) +{ + struct lru_iter *it; + + list_for_each_entry(it, &lru->iterators, list) { + /* Move c->e forwards if necc. */ + if (it->e == e) { + it->e = to_le(it->e->list.next); + if (it->e == e) + it->e = NULL; + } + + /* Move it->stop backwards if necc. */ + if (it->stop == e) { + it->stop = to_le(it->stop->list.prev); + if (it->stop == e) + it->stop = NULL; + } + } +} + +/*--------------*/ + +/* + * Remove a specific entry from the lru. + */ +static void lru_remove(struct lru *lru, struct lru_entry *le) +{ + lru_iter_invalidate(lru, le); + if (lru->count == 1) { + lru->cursor = NULL; + } else { + if (lru->cursor == &le->list) + lru->cursor = lru->cursor->next; + list_del(&le->list); + } + lru->count--; +} + +/* + * Mark as referenced. + */ +static inline void lru_reference(struct lru_entry *le) +{ + atomic_set(&le->referenced, 1); +} + +/*--------------*/ + +/* + * Remove the least recently used entry (approx), that passes the predicate. + * Returns NULL on failure. + */ +enum evict_result { + ER_EVICT, + ER_DONT_EVICT, + ER_STOP, /* stop looking for something to evict */ +}; + +typedef enum evict_result (*le_predicate)(struct lru_entry *le, void *context); + +static struct lru_entry *lru_evict(struct lru *lru, le_predicate pred, void *context) +{ + unsigned long tested = 0; + struct list_head *h = lru->cursor; + struct lru_entry *le; + + if (!h) + return NULL; + /* + * In the worst case we have to loop around twice. Once to clear + * the reference flags, and then again to discover the predicate + * fails for all entries. + */ + while (tested < lru->count) { + le = container_of(h, struct lru_entry, list); + + if (atomic_read(&le->referenced)) { + atomic_set(&le->referenced, 0); + } else { + tested++; + switch (pred(le, context)) { + case ER_EVICT: + /* + * Adjust the cursor, so we start the next + * search from here. + */ + lru->cursor = le->list.next; + lru_remove(lru, le); + return le; + + case ER_DONT_EVICT: + break; + + case ER_STOP: + lru->cursor = le->list.next; + return NULL; + } + } + + h = h->next; + + cond_resched(); + } + + return NULL; +} + +/*--------------------------------------------------------------*/ + /* * Linking of buffers: * All buffers are linked to buffer_tree with their node field. From patchwork Mon Mar 27 20:11:29 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Mike Snitzer X-Patchwork-Id: 13190052 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 vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id 129E3C761A6 for ; Mon, 27 Mar 2023 20:14:10 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S232144AbjC0UOI (ORCPT ); Mon, 27 Mar 2023 16:14:08 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:38766 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S232071AbjC0UOH (ORCPT ); Mon, 27 Mar 2023 16:14:07 -0400 Received: from mail-qt1-f181.google.com (mail-qt1-f181.google.com [209.85.160.181]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id 6D0851FCB for ; Mon, 27 Mar 2023 13:13:17 -0700 (PDT) Received: by mail-qt1-f181.google.com with SMTP id cn12so6456808qtb.8 for ; Mon, 27 Mar 2023 13:13:17 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; t=1679947996; h=content-transfer-encoding:mime-version:references:in-reply-to :message-id:date:subject:cc:to:from:x-gm-message-state:from:to:cc :subject:date:message-id:reply-to; bh=uAVSvRsxMnPc21jXZ4LrLYAk6KvY+7YTUeeMl0wWC/w=; b=CLUP7oeFAYfHJQqtvwc+bP57rWZpbMpzxklW95ppp5HYKRcK//rH6CJ34E5k5J0N/0 HWqK2gkoWwcc34e7nveHlcMCo2EEbvTrFVq1MMrMmzevsqSOs0QKor7cyT26Iih6FxrC VU9BeNGrSbLQZrycX6uQPEtaECiqLdSf/Dxi4aCcWJbrpoButwapMGJawyEeN33Q9TCF 0k6qxmMsUje0Xp5zusxfEGlENP3iGqdHNjfsGDMFjH4YqodAg5ZdGUSDDR5rioMPkEYg 8n5RfjgvtI5fEPBOn/wl+NcYCk1sjJvCMIMQXoGF/FzYBvOoJmTTljn71fy/GJbzOZji 4kvQ== X-Gm-Message-State: AO0yUKWj3TQynyBrpW53uoIWzoowp82cujrmYHLoJcfcMdCQXxP+J3zW sNsriHpozjKU6cTfguNPXB1E X-Google-Smtp-Source: AK7set8E1sQHoMgS/ejqC0UID98TZdqr1b8EC2TrHyC7jnO0V77TRODkBdML8gGustfuewA+h/qb3Q== X-Received: by 2002:ac8:5dd0:0:b0:3e1:65f5:4a85 with SMTP id e16-20020ac85dd0000000b003e165f54a85mr17645462qtx.58.1679947996400; Mon, 27 Mar 2023 13:13:16 -0700 (PDT) Received: from localhost (pool-68-160-166-30.bstnma.fios.verizon.net. [68.160.166.30]) by smtp.gmail.com with ESMTPSA id he8-20020a05622a600800b003b835e7e283sm380964qtb.44.2023.03.27.13.13.15 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Mon, 27 Mar 2023 13:13:16 -0700 (PDT) From: Mike Snitzer To: dm-devel@redhat.com Cc: linux-block@vger.kernel.org, axboe@kernel.dk, ejt@redhat.com, mpatocka@redhat.com, heinzm@redhat.com, nhuck@google.com, ebiggers@kernel.org, keescook@chromium.org, luomeng12@huawei.com, Mike Snitzer Subject: [dm-6.4 PATCH v3 06/20] dm bufio: add dm_buffer_cache abstraction Date: Mon, 27 Mar 2023 16:11:29 -0400 Message-Id: <20230327201143.51026-7-snitzer@kernel.org> X-Mailer: git-send-email 2.40.0 In-Reply-To: <20230327201143.51026-1-snitzer@kernel.org> References: <20230327201143.51026-1-snitzer@kernel.org> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-block@vger.kernel.org From: Joe Thornber The buffer cache is responsible for managing the holder count, tracking clean/dirty state, and choosing buffers via predicates. Higher level code is responsible for allocation of buffers, IO and eviction/cache sizing. The buffer cache has thread safe methods for acquiring a reference to an existing buffer. All other methods in buffer cache are _not_ threadsafe, and only contain enough locking to guarantee the safe methods. Rather than a single mutex, sharded rw_semaphores are used to allow concurrent threads to 'get' buffers. Each rw_semaphore protects its own rbtree of buffer entries. Code that uses this new dm_buffer_cache abstraction will be introduced in a following commit. This commit moves the dm_buffer struct in preparation for finer grained dm_buffer changes, in the next commit, to be more easily seen. It also introduces temporary dm_buffer struct members to allow compilation of this intermediate commit (they will be elided in the next commit). This commit will cause "defined but not used" compiler warnings that will be resolved by the next commit. Signed-off-by: Joe Thornber Signed-off-by: Mike Snitzer --- drivers/md/dm-bufio.c | 588 +++++++++++++++++++++++++++++++++++++----- 1 file changed, 526 insertions(+), 62 deletions(-) diff --git a/drivers/md/dm-bufio.c b/drivers/md/dm-bufio.c index a0bb0de0d4e7..9daff9b77cee 100644 --- a/drivers/md/dm-bufio.c +++ b/drivers/md/dm-bufio.c @@ -301,6 +301,517 @@ static struct lru_entry *lru_evict(struct lru *lru, le_predicate pred, void *con /*--------------------------------------------------------------*/ +/* + * Buffer state bits. + */ +#define B_READING 0 +#define B_WRITING 1 +#define B_DIRTY 2 + +/* + * Describes how the block was allocated: + * kmem_cache_alloc(), __get_free_pages() or vmalloc(). + * See the comment at alloc_buffer_data. + */ +enum data_mode { + DATA_MODE_SLAB = 0, + DATA_MODE_GET_FREE_PAGES = 1, + DATA_MODE_VMALLOC = 2, + DATA_MODE_LIMIT = 3 +}; + +struct dm_buffer { + struct rb_node node; + struct list_head lru_list; + struct list_head global_list; + + sector_t block; + void *data; + unsigned char data_mode; /* DATA_MODE_* */ + + unsigned int accessed; + unsigned int __hold_count; + unsigned long last_accessed; + + unsigned char list_mode; /* LIST_* */ + blk_status_t read_error; + blk_status_t write_error; + unsigned long state; + unsigned int dirty_start; + unsigned int dirty_end; + unsigned int write_start; + unsigned int write_end; + struct dm_bufio_client *c; + struct list_head write_list; + void (*end_io)(struct dm_buffer *b, blk_status_t bs); +#ifdef CONFIG_DM_DEBUG_BLOCK_STACK_TRACING +#define MAX_STACK 10 + unsigned int stack_len; + unsigned long stack_entries[MAX_STACK]; +#endif + /* Temp members to allow dm_buffer_cache code to compile */ + atomic_t hold_count; + struct lru_entry lru; +}; + +/*--------------------------------------------------------------*/ + +/* + * The buffer cache manages buffers, particularly: + * - inc/dec of holder count + * - setting the last_accessed field + * - maintains clean/dirty state along with lru + * - selecting buffers that match predicates + * + * It does *not* handle: + * - allocation/freeing of buffers. + * - IO + * - Eviction or cache sizing. + * + * cache_get() and cache_put() are threadsafe, you do not need to + * protect these calls with a surrounding mutex. All the other + * methods are not threadsafe; they do use locking primitives, but + * only enough to ensure get/put are threadsafe. + */ + +#define NR_LOCKS 64 +#define LOCKS_MASK (NR_LOCKS - 1) + +struct buffer_tree { + struct rw_semaphore lock; + struct rb_root root; +} ____cacheline_aligned_in_smp; + +struct dm_buffer_cache { + /* + * We spread entries across multiple trees to reduce contention + * on the locks. + */ + struct buffer_tree trees[NR_LOCKS]; + struct lru lru[LIST_SIZE]; +}; + +static inline unsigned int cache_index(sector_t block) +{ + return block & LOCKS_MASK; +} + +static inline void cache_read_lock(struct dm_buffer_cache *bc, sector_t block) +{ + down_read(&bc->trees[cache_index(block)].lock); +} + +static inline void cache_read_unlock(struct dm_buffer_cache *bc, sector_t block) +{ + up_read(&bc->trees[cache_index(block)].lock); +} + +static inline void cache_write_lock(struct dm_buffer_cache *bc, sector_t block) +{ + down_write(&bc->trees[cache_index(block)].lock); +} + +static inline void cache_write_unlock(struct dm_buffer_cache *bc, sector_t block) +{ + up_write(&bc->trees[cache_index(block)].lock); +} + +static inline struct dm_buffer *le_to_buffer(struct lru_entry *le) +{ + return container_of(le, struct dm_buffer, lru); +} + +static struct dm_buffer *list_to_buffer(struct list_head *l) +{ + struct lru_entry *le = list_entry(l, struct lru_entry, list); + + if (!le) + return NULL; + + return le_to_buffer(le); +} + +static void cache_init(struct dm_buffer_cache *bc) +{ + unsigned int i; + + for (i = 0; i < NR_LOCKS; i++) { + init_rwsem(&bc->trees[i].lock); + bc->trees[i].root = RB_ROOT; + } + + lru_init(&bc->lru[LIST_CLEAN]); + lru_init(&bc->lru[LIST_DIRTY]); +} + +static void cache_destroy(struct dm_buffer_cache *bc) +{ + unsigned int i; + + for (i = 0; i < NR_LOCKS; i++) + WARN_ON_ONCE(!RB_EMPTY_ROOT(&bc->trees[i].root)); + + lru_destroy(&bc->lru[LIST_CLEAN]); + lru_destroy(&bc->lru[LIST_DIRTY]); +} + +/*--------------*/ + +/* + * not threadsafe, or racey depending how you look at it + */ +static inline unsigned long cache_count(struct dm_buffer_cache *bc, int list_mode) +{ + return bc->lru[list_mode].count; +} + +static inline unsigned long cache_total(struct dm_buffer_cache *bc) +{ + return cache_count(bc, LIST_CLEAN) + cache_count(bc, LIST_DIRTY); +} + +/*--------------*/ + +/* + * Gets a specific buffer, indexed by block. + * If the buffer is found then its holder count will be incremented and + * lru_reference will be called. + * + * threadsafe + */ +static struct dm_buffer *__cache_get(const struct rb_root *root, sector_t block) +{ + struct rb_node *n = root->rb_node; + struct dm_buffer *b; + + while (n) { + b = container_of(n, struct dm_buffer, node); + + if (b->block == block) + return b; + + n = block < b->block ? n->rb_left : n->rb_right; + } + + return NULL; +} + +static void __cache_inc_buffer(struct dm_buffer *b) +{ + atomic_inc(&b->hold_count); + WRITE_ONCE(b->last_accessed, jiffies); +} + +static struct dm_buffer *cache_get(struct dm_buffer_cache *bc, sector_t block) +{ + struct dm_buffer *b; + + cache_read_lock(bc, block); + b = __cache_get(&bc->trees[cache_index(block)].root, block); + if (b) { + lru_reference(&b->lru); + __cache_inc_buffer(b); + } + cache_read_unlock(bc, block); + + return b; +} + +/*--------------*/ + +/* + * Returns true if the hold count hits zero. + * threadsafe + */ +static bool cache_put(struct dm_buffer_cache *bc, struct dm_buffer *b) +{ + bool r; + + cache_read_lock(bc, b->block); + BUG_ON(!atomic_read(&b->hold_count)); + r = atomic_dec_and_test(&b->hold_count); + cache_read_unlock(bc, b->block); + + return r; +} + +/*--------------*/ + +typedef enum evict_result (*b_predicate)(struct dm_buffer *, void *); + +/* + * Evicts a buffer based on a predicate. The oldest buffer that + * matches the predicate will be selected. In addition to the + * predicate the hold_count of the selected buffer will be zero. + */ +struct evict_wrapper { + b_predicate pred; + void *context; +}; + +/* + * Wraps the buffer predicate turning it into an lru predicate. Adds + * extra test for hold_count. + */ +static enum evict_result __evict_pred(struct lru_entry *le, void *context) +{ + struct evict_wrapper *w = context; + struct dm_buffer *b = le_to_buffer(le); + + if (atomic_read(&b->hold_count)) + return ER_DONT_EVICT; + + return w->pred(b, w->context); +} + +static struct dm_buffer *cache_evict(struct dm_buffer_cache *bc, int list_mode, + b_predicate pred, void *context) +{ + struct evict_wrapper w = {.pred = pred, .context = context}; + struct lru_entry *le; + struct dm_buffer *b; + + le = lru_evict(&bc->lru[list_mode], __evict_pred, &w); + if (!le) + return NULL; + + b = le_to_buffer(le); + /* __evict_pred will have locked the appropriate tree. */ + rb_erase(&b->node, &bc->trees[cache_index(b->block)].root); + + return b; +} + +/*--------------*/ + +/* + * Mark a buffer as clean or dirty. Not threadsafe. + */ +static void cache_mark(struct dm_buffer_cache *bc, struct dm_buffer *b, int list_mode) +{ + cache_write_lock(bc, b->block); + if (list_mode != b->list_mode) { + lru_remove(&bc->lru[b->list_mode], &b->lru); + b->list_mode = list_mode; + lru_insert(&bc->lru[b->list_mode], &b->lru); + } + cache_write_unlock(bc, b->block); +} + +/*--------------*/ + +/* + * Runs through the lru associated with 'old_mode', if the predicate matches then + * it moves them to 'new_mode'. Not threadsafe. + */ +static void cache_mark_many(struct dm_buffer_cache *bc, int old_mode, int new_mode, + b_predicate pred, void *context) +{ + struct lru_entry *le; + struct dm_buffer *b; + struct evict_wrapper w = {.pred = pred, .context = context}; + + while (true) { + le = lru_evict(&bc->lru[old_mode], __evict_pred, &w); + if (!le) + break; + + b = le_to_buffer(le); + b->list_mode = new_mode; + lru_insert(&bc->lru[b->list_mode], &b->lru); + } +} + +/*--------------*/ + +/* + * Iterates through all clean or dirty entries calling a function for each + * entry. The callback may terminate the iteration early. Not threadsafe. + */ + +/* + * Iterator functions should return one of these actions to indicate + * how the iteration should proceed. + */ +enum it_action { + IT_NEXT, + IT_COMPLETE, +}; + +typedef enum it_action (*iter_fn)(struct dm_buffer *b, void *context); + +static void cache_iterate(struct dm_buffer_cache *bc, int list_mode, + iter_fn fn, void *context) +{ + struct lru *lru = &bc->lru[list_mode]; + struct lru_entry *le, *first; + + if (!lru->cursor) + return; + + first = le = to_le(lru->cursor); + do { + struct dm_buffer *b = le_to_buffer(le); + + switch (fn(b, context)) { + case IT_NEXT: + break; + + case IT_COMPLETE: + return; + } + cond_resched(); + + le = to_le(le->list.next); + } while (le != first); +} + +/*--------------*/ + +/* + * Passes ownership of the buffer to the cache. Returns false if the + * buffer was already present (in which case ownership does not pass). + * eg, a race with another thread. + * + * Holder count should be 1 on insertion. + * + * Not threadsafe. + */ +static bool __cache_insert(struct rb_root *root, struct dm_buffer *b) +{ + struct rb_node **new = &root->rb_node, *parent = NULL; + struct dm_buffer *found; + + while (*new) { + found = container_of(*new, struct dm_buffer, node); + + if (found->block == b->block) + return false; + + parent = *new; + new = b->block < found->block ? + &found->node.rb_left : &found->node.rb_right; + } + + rb_link_node(&b->node, parent, new); + rb_insert_color(&b->node, root); + + return true; +} + +static bool cache_insert(struct dm_buffer_cache *bc, struct dm_buffer *b) +{ + bool r; + + if (WARN_ON_ONCE(b->list_mode >= LIST_SIZE)) + return false; + + cache_write_lock(bc, b->block); + BUG_ON(atomic_read(&b->hold_count) != 1); + r = __cache_insert(&bc->trees[cache_index(b->block)].root, b); + if (r) + lru_insert(&bc->lru[b->list_mode], &b->lru); + cache_write_unlock(bc, b->block); + + return r; +} + +/*--------------*/ + +/* + * Removes buffer from cache, ownership of the buffer passes back to the caller. + * Fails if the hold_count is not one (ie. the caller holds the only reference). + * + * Not threadsafe. + */ +static bool cache_remove(struct dm_buffer_cache *bc, struct dm_buffer *b) +{ + bool r; + + cache_write_lock(bc, b->block); + + if (atomic_read(&b->hold_count) != 1) { + r = false; + } else { + r = true; + rb_erase(&b->node, &bc->trees[cache_index(b->block)].root); + lru_remove(&bc->lru[b->list_mode], &b->lru); + } + + cache_write_unlock(bc, b->block); + + return r; +} + +/*--------------*/ + +typedef void (*b_release)(struct dm_buffer *); + +static struct dm_buffer *__find_next(struct rb_root *root, sector_t block) +{ + struct rb_node *n = root->rb_node; + struct dm_buffer *b; + struct dm_buffer *best = NULL; + + while (n) { + b = container_of(n, struct dm_buffer, node); + + if (b->block == block) + return b; + + if (block <= b->block) { + n = n->rb_left; + best = b; + } else { + n = n->rb_right; + } + } + + return best; +} + +static void __remove_range(struct dm_buffer_cache *bc, + struct rb_root *root, + sector_t begin, sector_t end, + b_predicate pred, b_release release) +{ + struct dm_buffer *b; + + while (true) { + cond_resched(); + + b = __find_next(root, begin); + if (!b || (b->block >= end)) + break; + + begin = b->block + 1; + + if (atomic_read(&b->hold_count)) + continue; + + if (pred(b, NULL) == ER_EVICT) { + rb_erase(&b->node, root); + lru_remove(&bc->lru[b->list_mode], &b->lru); + release(b); + } + } +} + +static void cache_remove_range(struct dm_buffer_cache *bc, + sector_t begin, sector_t end, + b_predicate pred, b_release release) +{ + unsigned int i; + + for (i = 0; i < NR_LOCKS; i++) { + down_write(&bc->trees[i].lock); + __remove_range(bc, &bc->trees[i].root, begin, end, pred, release); + up_write(&bc->trees[i].lock); + } +} + +/*----------------------------------------------------------------*/ + /* * Linking of buffers: * All buffers are linked to buffer_tree with their node field. @@ -352,53 +863,6 @@ struct dm_bufio_client { atomic_long_t need_shrink; }; -/* - * Buffer state bits. - */ -#define B_READING 0 -#define B_WRITING 1 -#define B_DIRTY 2 - -/* - * Describes how the block was allocated: - * kmem_cache_alloc(), __get_free_pages() or vmalloc(). - * See the comment at alloc_buffer_data. - */ -enum data_mode { - DATA_MODE_SLAB = 0, - DATA_MODE_GET_FREE_PAGES = 1, - DATA_MODE_VMALLOC = 2, - DATA_MODE_LIMIT = 3 -}; - -struct dm_buffer { - struct rb_node node; - struct list_head lru_list; - struct list_head global_list; - sector_t block; - void *data; - unsigned char data_mode; /* DATA_MODE_* */ - unsigned char list_mode; /* LIST_* */ - blk_status_t read_error; - blk_status_t write_error; - unsigned int accessed; - unsigned int hold_count; - unsigned long state; - unsigned long last_accessed; - unsigned int dirty_start; - unsigned int dirty_end; - unsigned int write_start; - unsigned int write_end; - struct dm_bufio_client *c; - struct list_head write_list; - void (*end_io)(struct dm_buffer *buf, blk_status_t stat); -#ifdef CONFIG_DM_DEBUG_BLOCK_STACK_TRACING -#define MAX_STACK 10 - unsigned int stack_len; - unsigned long stack_entries[MAX_STACK]; -#endif -}; - static DEFINE_STATIC_KEY_FALSE(no_sleep_enabled); /*----------------------------------------------------------------*/ @@ -516,7 +980,7 @@ static struct dm_buffer *__find(struct dm_bufio_client *c, sector_t block) return NULL; } -static struct dm_buffer *__find_next(struct dm_bufio_client *c, sector_t block) +static struct dm_buffer *__find_next_old(struct dm_bufio_client *c, sector_t block) { struct rb_node *n = c->buffer_tree.rb_node; struct dm_buffer *b; @@ -1040,7 +1504,7 @@ static void __flush_write_list(struct list_head *write_list) */ static void __make_buffer_clean(struct dm_buffer *b) { - BUG_ON(b->hold_count); + BUG_ON(b->__hold_count); /* smp_load_acquire() pairs with read_endio()'s smp_mb__before_atomic() */ if (!smp_load_acquire(&b->state)) /* fast case */ @@ -1067,7 +1531,7 @@ static struct dm_buffer *__get_unclaimed_buffer(struct dm_bufio_client *c) unlikely(test_bit_acquire(B_READING, &b->state))) continue; - if (!b->hold_count) { + if (!b->__hold_count) { __make_buffer_clean(b); __unlink_buffer(b); return b; @@ -1081,7 +1545,7 @@ static struct dm_buffer *__get_unclaimed_buffer(struct dm_bufio_client *c) list_for_each_entry_reverse(b, &c->lru[LIST_DIRTY], lru_list) { BUG_ON(test_bit(B_READING, &b->state)); - if (!b->hold_count) { + if (!b->__hold_count) { __make_buffer_clean(b); __unlink_buffer(b); return b; @@ -1283,7 +1747,7 @@ static struct dm_buffer *__bufio_new(struct dm_bufio_client *c, sector_t block, __check_watermark(c, write_list); b = new_b; - b->hold_count = 1; + b->__hold_count = 1; b->read_error = 0; b->write_error = 0; __link_buffer(b, block, LIST_CLEAN); @@ -1311,7 +1775,7 @@ static struct dm_buffer *__bufio_new(struct dm_bufio_client *c, sector_t block, if (nf == NF_GET && unlikely(test_bit_acquire(B_READING, &b->state))) return NULL; - b->hold_count++; + b->__hold_count++; __relink_lru(b, test_bit(B_DIRTY, &b->state) || test_bit(B_WRITING, &b->state)); return b; @@ -1460,10 +1924,10 @@ void dm_bufio_release(struct dm_buffer *b) dm_bufio_lock(c); - BUG_ON(!b->hold_count); + BUG_ON(!b->__hold_count); - b->hold_count--; - if (!b->hold_count) { + b->__hold_count--; + if (!b->__hold_count) { wake_up(&c->free_buffer_wait); /* @@ -1564,12 +2028,12 @@ int dm_bufio_write_dirty_buffers(struct dm_bufio_client *c) if (test_bit(B_WRITING, &b->state)) { if (buffers_processed < c->n_buffers[LIST_DIRTY]) { dropped_lock = 1; - b->hold_count++; + b->__hold_count++; dm_bufio_unlock(c); wait_on_bit_io(&b->state, B_WRITING, TASK_UNINTERRUPTIBLE); dm_bufio_lock(c); - b->hold_count--; + b->__hold_count--; } else wait_on_bit_io(&b->state, B_WRITING, TASK_UNINTERRUPTIBLE); @@ -1660,7 +2124,7 @@ EXPORT_SYMBOL_GPL(dm_bufio_issue_discard); static void forget_buffer_locked(struct dm_buffer *b) { - if (likely(!b->hold_count) && likely(!smp_load_acquire(&b->state))) { + if (likely(!b->__hold_count) && likely(!smp_load_acquire(&b->state))) { __unlink_buffer(b); __free_buffer_wake(b); } @@ -1694,7 +2158,7 @@ void dm_bufio_forget_buffers(struct dm_bufio_client *c, sector_t block, sector_t while (block < end_block) { dm_bufio_lock(c); - b = __find_next(c, block); + b = __find_next_old(c, block); if (b) { block = b->block + 1; forget_buffer_locked(b); @@ -1791,7 +2255,7 @@ static void drop_buffers(struct dm_bufio_client *c) WARN_ON(!warned); warned = true; DMERR("leaked buffer %llx, hold count %u, list %d", - (unsigned long long)b->block, b->hold_count, i); + (unsigned long long)b->block, b->__hold_count, i); #ifdef CONFIG_DM_DEBUG_BLOCK_STACK_TRACING stack_trace_print(b->stack_entries, b->stack_len, 1); /* mark unclaimed to avoid WARN_ON below */ @@ -1828,7 +2292,7 @@ static bool __try_evict_buffer(struct dm_buffer *b, gfp_t gfp) return false; } - if (b->hold_count) + if (b->__hold_count) return false; __make_buffer_clean(b); From patchwork Mon Mar 27 20:11:30 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 8bit X-Patchwork-Submitter: Mike Snitzer X-Patchwork-Id: 13190056 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 vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id CE085C76195 for ; Mon, 27 Mar 2023 20:14:15 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S232354AbjC0UOO (ORCPT ); Mon, 27 Mar 2023 16:14:14 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:39000 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S231929AbjC0UOL (ORCPT ); Mon, 27 Mar 2023 16:14:11 -0400 Received: from mail-qt1-f169.google.com (mail-qt1-f169.google.com [209.85.160.169]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id 77C791BF6 for ; Mon, 27 Mar 2023 13:13:19 -0700 (PDT) Received: by mail-qt1-f169.google.com with SMTP id t19so9790941qta.12 for ; Mon, 27 Mar 2023 13:13:19 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; t=1679947998; h=content-transfer-encoding:mime-version:references:in-reply-to :message-id:date:subject:cc:to:from:x-gm-message-state:from:to:cc :subject:date:message-id:reply-to; bh=KZEmP2Lma0r81Bos8w4U89OEq/V+OweIm/gue2zysiY=; b=zatRm5vcsBOJvKzHkYbhsp4LXTmtC97qp1dB94djR7hu2Zyq8pvmP8wWZT5f//RZhr hRXoPY80mMFPz8DZjlu8QhPok+2Gd9eUkw3WxSfJWgEi3AuKe9nbzLnTLFuQn01Tjxug jUWS1CoHapGBWEDmfJPJqVIiQg47mCg5/ANjIqLBsnPESr5KuhRTE9ivq6NFa6MPgN4P 20WnCSN1fhTSMwvsrCJDL9gvRa6G6N6RkQ78KKduIip6+eVomTNRI+H8kB/jy9Fjuwkt 5McdXjAo7Fi5dGHXwctBIL9LIQokfZF3mzlyBPFgxUjpHz1srjr2Q+psm+XkFRI7XN7A au+w== X-Gm-Message-State: AO0yUKUBJWgwAtAkMsa6p4tDFNEsHv7tZMHr7ty1na2UsScN1CyoxdIH 5Bu/DtRQy+TCNH+KOtDoXjBV X-Google-Smtp-Source: AK7set/4u5wtNp1+7i337s6/269TkCLiJgXwRPcmr5VJh9wfIB7rZepSb8dvmBloMyznyDlN/1yP2Q== X-Received: by 2002:a05:622a:1109:b0:3e3:82c4:db44 with SMTP id e9-20020a05622a110900b003e382c4db44mr23802990qty.52.1679947998246; Mon, 27 Mar 2023 13:13:18 -0700 (PDT) Received: from localhost (pool-68-160-166-30.bstnma.fios.verizon.net. [68.160.166.30]) by smtp.gmail.com with ESMTPSA id u23-20020a37ab17000000b0071f0d0aaef7sm13735581qke.80.2023.03.27.13.13.17 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Mon, 27 Mar 2023 13:13:17 -0700 (PDT) From: Mike Snitzer To: dm-devel@redhat.com Cc: linux-block@vger.kernel.org, axboe@kernel.dk, ejt@redhat.com, mpatocka@redhat.com, heinzm@redhat.com, nhuck@google.com, ebiggers@kernel.org, keescook@chromium.org, luomeng12@huawei.com, Mike Snitzer Subject: [dm-6.4 PATCH v3 07/20] dm bufio: improve concurrent IO performance Date: Mon, 27 Mar 2023 16:11:30 -0400 Message-Id: <20230327201143.51026-8-snitzer@kernel.org> X-Mailer: git-send-email 2.40.0 In-Reply-To: <20230327201143.51026-1-snitzer@kernel.org> References: <20230327201143.51026-1-snitzer@kernel.org> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-block@vger.kernel.org From: Joe Thornber When multiple threads perform IO to a thin device, the underlying dm_bufio object can become a bottleneck; slowing down access to btree nodes that store the thin metadata. Prior to this commit, each bufio instance had a single mutex that was taken for every bufio operation. This commit concentrates on improving the common case where: a user of dm_bufio wishes to access, but not modify, a buffer which is already within the dm_bufio cache. Implementation:: The code has been refactored; pulling out an 'lru' abstraction and a 'buffer cache' abstraction (see 2 previous commits). This commit updates higher level bufio code (that performs allocation of buffers, IO and eviction/cache sizing) to leverage both abstractions. It also deals with the delicate locking requirements of both abstractions to provide finer grained locking. The result is significantly better concurrent IO performance. Before this commit, bufio has a global lru list it used to evict the oldest, clean buffers from _all_ clients. With the new locking we don’t want different ways to access the same buffer, so instead do_global_cleanup() loops around the clients asking them to free buffers older than a certain time. This commit also converts many old BUG_ONs to WARN_ON_ONCE, see the lru_evict and cache_evict code in particular. They will return ER_DONT_EVICT if a given buffer somehow meets the invariants that should _never_ happen. [Aside from revising this commit's header and fixing coding style and whitespace nits: this switching to WARN_ON_ONCE is Mike Snitzer's lone contribution to this commit] Testing:: Some of the low level functions have been unit tested using dm-unit: https://github.com/jthornber/dm-unit/blob/main/src/tests/bufio.rs Higher level concurrency and IO is tested via a test only target found here: https://github.com/jthornber/linux/blob/2023-03-24-thin-concurrency-9/drivers/md/dm-bufio-test.c The associated userland side of these tests is here: https://github.com/jthornber/dmtest-python/blob/main/src/dmtest/bufio/bufio_tests.py In addition the full dmtest suite of tests (dm-thin, dm-cache, etc) has been run (~450 tests). Performance:: Most bufio operations have unchanged performance. But if multiple threads are attempting to get buffers concurrently, and these buffers are already in the cache then there's a big speed up. Eg, one test has 16 'hotspot' threads simulating btree lookups while another thread dirties the whole device. In this case the hotspot threads acquire the buffers about 25 times faster. Signed-off-by: Joe Thornber Signed-off-by: Mike Snitzer --- drivers/md/dm-bufio.c | 959 +++++++++++++++++++++--------------------- 1 file changed, 487 insertions(+), 472 deletions(-) diff --git a/drivers/md/dm-bufio.c b/drivers/md/dm-bufio.c index 9daff9b77cee..1e000ec73bd6 100644 --- a/drivers/md/dm-bufio.c +++ b/drivers/md/dm-bufio.c @@ -321,37 +321,42 @@ enum data_mode { }; struct dm_buffer { + /* protected by the locks in dm_buffer_cache */ struct rb_node node; - struct list_head lru_list; - struct list_head global_list; + /* immutable, so don't need protecting */ sector_t block; void *data; unsigned char data_mode; /* DATA_MODE_* */ - unsigned int accessed; - unsigned int __hold_count; + /* + * These two fields are used in isolation, so do not need + * a surrounding lock. + */ + atomic_t hold_count; unsigned long last_accessed; + /* + * Everything else is protected by the mutex in + * dm_bufio_client + */ + unsigned long state; + struct lru_entry lru; unsigned char list_mode; /* LIST_* */ blk_status_t read_error; blk_status_t write_error; - unsigned long state; unsigned int dirty_start; unsigned int dirty_end; unsigned int write_start; unsigned int write_end; - struct dm_bufio_client *c; struct list_head write_list; + struct dm_bufio_client *c; void (*end_io)(struct dm_buffer *b, blk_status_t bs); #ifdef CONFIG_DM_DEBUG_BLOCK_STACK_TRACING #define MAX_STACK 10 unsigned int stack_len; unsigned long stack_entries[MAX_STACK]; #endif - /* Temp members to allow dm_buffer_cache code to compile */ - atomic_t hold_count; - struct lru_entry lru; }; /*--------------------------------------------------------------*/ @@ -814,7 +819,7 @@ static void cache_remove_range(struct dm_buffer_cache *bc, /* * Linking of buffers: - * All buffers are linked to buffer_tree with their node field. + * All buffers are linked to buffer_cache with their node field. * * Clean buffers that are not being written (B_WRITING not set) * are linked to lru[LIST_CLEAN] with their lru_list field. @@ -832,9 +837,6 @@ struct dm_bufio_client { spinlock_t spinlock; bool no_sleep; - struct list_head lru[LIST_SIZE]; - unsigned long n_buffers[LIST_SIZE]; - struct block_device *bdev; unsigned int block_size; s8 sectors_per_block_bits; @@ -849,7 +851,7 @@ struct dm_bufio_client { unsigned int minimum_buffers; - struct rb_root buffer_tree; + struct dm_buffer_cache cache; wait_queue_head_t free_buffer_wait; sector_t start; @@ -861,6 +863,11 @@ struct dm_bufio_client { struct shrinker shrinker; struct work_struct shrink_work; atomic_long_t need_shrink; + + /* + * Used by global_cleanup to sort the clients list. + */ + unsigned long oldest_buffer; }; static DEFINE_STATIC_KEY_FALSE(no_sleep_enabled); @@ -877,14 +884,6 @@ static void dm_bufio_lock(struct dm_bufio_client *c) mutex_lock_nested(&c->lock, dm_bufio_in_request()); } -static int dm_bufio_trylock(struct dm_bufio_client *c) -{ - if (static_branch_unlikely(&no_sleep_enabled) && c->no_sleep) - return spin_trylock_bh(&c->spinlock); - else - return mutex_trylock(&c->lock); -} - static void dm_bufio_unlock(struct dm_bufio_client *c) { if (static_branch_unlikely(&no_sleep_enabled) && c->no_sleep) @@ -913,10 +912,6 @@ static unsigned long dm_bufio_cache_size_latch; static DEFINE_SPINLOCK(global_spinlock); -static LIST_HEAD(global_queue); - -static unsigned long global_num; - /* * Buffers are freed after this timeout */ @@ -958,78 +953,6 @@ static void buffer_record_stack(struct dm_buffer *b) } #endif -/* - *---------------------------------------------------------------- - * A red/black tree acts as an index for all the buffers. - *---------------------------------------------------------------- - */ -static struct dm_buffer *__find(struct dm_bufio_client *c, sector_t block) -{ - struct rb_node *n = c->buffer_tree.rb_node; - struct dm_buffer *b; - - while (n) { - b = container_of(n, struct dm_buffer, node); - - if (b->block == block) - return b; - - n = block < b->block ? n->rb_left : n->rb_right; - } - - return NULL; -} - -static struct dm_buffer *__find_next_old(struct dm_bufio_client *c, sector_t block) -{ - struct rb_node *n = c->buffer_tree.rb_node; - struct dm_buffer *b; - struct dm_buffer *best = NULL; - - while (n) { - b = container_of(n, struct dm_buffer, node); - - if (b->block == block) - return b; - - if (block <= b->block) { - n = n->rb_left; - best = b; - } else { - n = n->rb_right; - } - } - - return best; -} - -static void __insert(struct dm_bufio_client *c, struct dm_buffer *b) -{ - struct rb_node **new = &c->buffer_tree.rb_node, *parent = NULL; - struct dm_buffer *found; - - while (*new) { - found = container_of(*new, struct dm_buffer, node); - - if (found->block == b->block) { - BUG_ON(found != b); - return; - } - - parent = *new; - new = b->block < found->block ? - &found->node.rb_left : &found->node.rb_right; - } - - rb_link_node(&b->node, parent, new); - rb_insert_color(&b->node, &c->buffer_tree); -} - -static void __remove(struct dm_bufio_client *c, struct dm_buffer *b) -{ - rb_erase(&b->node, &c->buffer_tree); -} - /*----------------------------------------------------------------*/ static void adjust_total_allocated(struct dm_buffer *b, bool unlink) @@ -1057,16 +980,9 @@ static void adjust_total_allocated(struct dm_buffer *b, bool unlink) if (dm_bufio_current_allocated > dm_bufio_peak_allocated) dm_bufio_peak_allocated = dm_bufio_current_allocated; - b->accessed = 1; - if (!unlink) { - list_add(&b->global_list, &global_queue); - global_num++; if (dm_bufio_current_allocated > dm_bufio_cache_size) queue_work(dm_bufio_wq, &dm_bufio_replacement_work); - } else { - list_del(&b->global_list); - global_num--; } spin_unlock(&global_spinlock); @@ -1196,6 +1112,7 @@ static struct dm_buffer *alloc_buffer(struct dm_bufio_client *c, gfp_t gfp_mask) kmem_cache_free(c->slab_buffer, b); return NULL; } + adjust_total_allocated(b, false); #ifdef CONFIG_DM_DEBUG_BLOCK_STACK_TRACING b->stack_len = 0; @@ -1210,61 +1127,11 @@ static void free_buffer(struct dm_buffer *b) { struct dm_bufio_client *c = b->c; + adjust_total_allocated(b, true); free_buffer_data(c, b->data, b->data_mode); kmem_cache_free(c->slab_buffer, b); } -/* - * Link buffer to the buffer tree and clean or dirty queue. - */ -static void __link_buffer(struct dm_buffer *b, sector_t block, int dirty) -{ - struct dm_bufio_client *c = b->c; - - c->n_buffers[dirty]++; - b->block = block; - b->list_mode = dirty; - list_add(&b->lru_list, &c->lru[dirty]); - __insert(b->c, b); - b->last_accessed = jiffies; - - adjust_total_allocated(b, false); -} - -/* - * Unlink buffer from the buffer tree and dirty or clean queue. - */ -static void __unlink_buffer(struct dm_buffer *b) -{ - struct dm_bufio_client *c = b->c; - - BUG_ON(!c->n_buffers[b->list_mode]); - - c->n_buffers[b->list_mode]--; - __remove(b->c, b); - list_del(&b->lru_list); - - adjust_total_allocated(b, true); -} - -/* - * Place the buffer to the head of dirty or clean LRU queue. - */ -static void __relink_lru(struct dm_buffer *b, int dirty) -{ - struct dm_bufio_client *c = b->c; - - b->accessed = 1; - - BUG_ON(!c->n_buffers[b->list_mode]); - - c->n_buffers[b->list_mode]--; - c->n_buffers[dirty]++; - b->list_mode = dirty; - list_move(&b->lru_list, &c->lru[dirty]); - b->last_accessed = jiffies; -} - /* *-------------------------------------------------------------------------- * Submit I/O on the buffer. @@ -1504,7 +1371,7 @@ static void __flush_write_list(struct list_head *write_list) */ static void __make_buffer_clean(struct dm_buffer *b) { - BUG_ON(b->__hold_count); + BUG_ON(atomic_read(&b->hold_count)); /* smp_load_acquire() pairs with read_endio()'s smp_mb__before_atomic() */ if (!smp_load_acquire(&b->state)) /* fast case */ @@ -1515,6 +1382,36 @@ static void __make_buffer_clean(struct dm_buffer *b) wait_on_bit_io(&b->state, B_WRITING, TASK_UNINTERRUPTIBLE); } +static enum evict_result is_clean(struct dm_buffer *b, void *context) +{ + struct dm_bufio_client *c = context; + + /* These should never happen */ + if (WARN_ON_ONCE(test_bit(B_WRITING, &b->state))) + return ER_DONT_EVICT; + if (WARN_ON_ONCE(test_bit(B_DIRTY, &b->state))) + return ER_DONT_EVICT; + if (WARN_ON_ONCE(b->list_mode != LIST_CLEAN)) + return ER_DONT_EVICT; + + if (static_branch_unlikely(&no_sleep_enabled) && c->no_sleep && + unlikely(test_bit(B_READING, &b->state))) + return ER_DONT_EVICT; + + return ER_EVICT; +} + +static enum evict_result is_dirty(struct dm_buffer *b, void *context) +{ + /* These should never happen */ + if (WARN_ON_ONCE(test_bit(B_READING, &b->state))) + return ER_DONT_EVICT; + if (WARN_ON_ONCE(b->list_mode != LIST_DIRTY)) + return ER_DONT_EVICT; + + return ER_EVICT; +} + /* * Find some buffer that is not held by anybody, clean it, unlink it and * return it. @@ -1523,34 +1420,20 @@ static struct dm_buffer *__get_unclaimed_buffer(struct dm_bufio_client *c) { struct dm_buffer *b; - list_for_each_entry_reverse(b, &c->lru[LIST_CLEAN], lru_list) { - BUG_ON(test_bit(B_WRITING, &b->state)); - BUG_ON(test_bit(B_DIRTY, &b->state)); - - if (static_branch_unlikely(&no_sleep_enabled) && c->no_sleep && - unlikely(test_bit_acquire(B_READING, &b->state))) - continue; - - if (!b->__hold_count) { - __make_buffer_clean(b); - __unlink_buffer(b); - return b; - } - cond_resched(); + b = cache_evict(&c->cache, LIST_CLEAN, is_clean, c); + if (b) { + /* this also waits for pending reads */ + __make_buffer_clean(b); + return b; } if (static_branch_unlikely(&no_sleep_enabled) && c->no_sleep) return NULL; - list_for_each_entry_reverse(b, &c->lru[LIST_DIRTY], lru_list) { - BUG_ON(test_bit(B_READING, &b->state)); - - if (!b->__hold_count) { - __make_buffer_clean(b); - __unlink_buffer(b); - return b; - } - cond_resched(); + b = cache_evict(&c->cache, LIST_DIRTY, is_dirty, NULL); + if (b) { + __make_buffer_clean(b); + return b; } return NULL; @@ -1571,7 +1454,12 @@ static void __wait_for_free_buffer(struct dm_bufio_client *c) set_current_state(TASK_UNINTERRUPTIBLE); dm_bufio_unlock(c); - io_schedule(); + /* + * It's possible to miss a wake up event since we don't always + * hold c->lock when wake_up is called. So we have a timeout here, + * just in case. + */ + io_schedule_timeout(5 * HZ); remove_wait_queue(&c->free_buffer_wait, &wait); @@ -1629,9 +1517,8 @@ static struct dm_buffer *__alloc_buffer_wait_no_callback(struct dm_bufio_client } if (!list_empty(&c->reserved_buffers)) { - b = list_entry(c->reserved_buffers.next, - struct dm_buffer, lru_list); - list_del(&b->lru_list); + b = list_to_buffer(c->reserved_buffers.next); + list_del(&b->lru.list); c->need_reserved_buffers++; return b; @@ -1665,36 +1552,56 @@ static void __free_buffer_wake(struct dm_buffer *b) { struct dm_bufio_client *c = b->c; + b->block = -1; if (!c->need_reserved_buffers) free_buffer(b); else { - list_add(&b->lru_list, &c->reserved_buffers); + list_add(&b->lru.list, &c->reserved_buffers); c->need_reserved_buffers--; } wake_up(&c->free_buffer_wait); } +static enum evict_result cleaned(struct dm_buffer *b, void *context) +{ + if (WARN_ON_ONCE(test_bit(B_READING, &b->state))) + return ER_DONT_EVICT; /* should never happen */ + + if (test_bit(B_DIRTY, &b->state) || test_bit(B_WRITING, &b->state)) + return ER_DONT_EVICT; + else + return ER_EVICT; +} + +static void __move_clean_buffers(struct dm_bufio_client *c) +{ + cache_mark_many(&c->cache, LIST_DIRTY, LIST_CLEAN, cleaned, NULL); +} + +struct write_context { + int no_wait; + struct list_head *write_list; +}; + +static enum it_action write_one(struct dm_buffer *b, void *context) +{ + struct write_context *wc = context; + + if (wc->no_wait && test_bit(B_WRITING, &b->state)) + return IT_COMPLETE; + + __write_dirty_buffer(b, wc->write_list); + return IT_NEXT; +} + static void __write_dirty_buffers_async(struct dm_bufio_client *c, int no_wait, struct list_head *write_list) { - struct dm_buffer *b, *tmp; + struct write_context wc = {.no_wait = no_wait, .write_list = write_list}; - list_for_each_entry_safe_reverse(b, tmp, &c->lru[LIST_DIRTY], lru_list) { - BUG_ON(test_bit(B_READING, &b->state)); - - if (!test_bit(B_DIRTY, &b->state) && - !test_bit(B_WRITING, &b->state)) { - __relink_lru(b, LIST_CLEAN); - continue; - } - - if (no_wait && test_bit(B_WRITING, &b->state)) - return; - - __write_dirty_buffer(b, write_list); - cond_resched(); - } + __move_clean_buffers(c); + cache_iterate(&c->cache, LIST_DIRTY, write_one, &wc); } /* @@ -1705,7 +1612,8 @@ static void __write_dirty_buffers_async(struct dm_bufio_client *c, int no_wait, static void __check_watermark(struct dm_bufio_client *c, struct list_head *write_list) { - if (c->n_buffers[LIST_DIRTY] > c->n_buffers[LIST_CLEAN] * DM_BUFIO_WRITEBACK_RATIO) + if (cache_count(&c->cache, LIST_DIRTY) > + cache_count(&c->cache, LIST_CLEAN) * DM_BUFIO_WRITEBACK_RATIO) __write_dirty_buffers_async(c, 1, write_list); } @@ -1715,6 +1623,21 @@ static void __check_watermark(struct dm_bufio_client *c, *-------------------------------------------------------------- */ +static void cache_put_and_wake(struct dm_bufio_client *c, struct dm_buffer *b) +{ + /* + * Relying on waitqueue_active() is racey, but we sleep + * with schedule_timeout anyway. + */ + if (cache_put(&c->cache, b) && + unlikely(waitqueue_active(&c->free_buffer_wait))) + wake_up(&c->free_buffer_wait); +} + +/* + * This assumes you have already checked the cache to see if the buffer + * is already present (it will recheck after dropping the lock for allocation). + */ static struct dm_buffer *__bufio_new(struct dm_bufio_client *c, sector_t block, enum new_flag nf, int *need_submit, struct list_head *write_list) @@ -1723,11 +1646,8 @@ static struct dm_buffer *__bufio_new(struct dm_bufio_client *c, sector_t block, *need_submit = 0; - b = __find(c, block); - if (b) - goto found_buffer; - - if (nf == NF_GET) + /* This can't be called with NF_GET */ + if (WARN_ON_ONCE(nf == NF_GET)) return NULL; new_b = __alloc_buffer_wait(c, nf); @@ -1738,7 +1658,7 @@ static struct dm_buffer *__bufio_new(struct dm_bufio_client *c, sector_t block, * We've had a period where the mutex was unlocked, so need to * recheck the buffer tree. */ - b = __find(c, block); + b = cache_get(&c->cache, block); if (b) { __free_buffer_wake(new_b); goto found_buffer; @@ -1747,24 +1667,35 @@ static struct dm_buffer *__bufio_new(struct dm_bufio_client *c, sector_t block, __check_watermark(c, write_list); b = new_b; - b->__hold_count = 1; + atomic_set(&b->hold_count, 1); + WRITE_ONCE(b->last_accessed, jiffies); + b->block = block; b->read_error = 0; b->write_error = 0; - __link_buffer(b, block, LIST_CLEAN); + b->list_mode = LIST_CLEAN; - if (nf == NF_FRESH) { + if (nf == NF_FRESH) b->state = 0; - return b; + else { + b->state = 1 << B_READING; + *need_submit = 1; } - b->state = 1 << B_READING; - *need_submit = 1; + /* + * We mustn't insert into the cache until the B_READING state + * is set. Otherwise another thread could get it and use + * it before it had been read. + */ + cache_insert(&c->cache, b); return b; found_buffer: - if (nf == NF_PREFETCH) + if (nf == NF_PREFETCH) { + cache_put_and_wake(c, b); return NULL; + } + /* * Note: it is essential that we don't wait for the buffer to be * read if dm_bufio_get function is used. Both dm_bufio_get and @@ -1772,12 +1703,11 @@ static struct dm_buffer *__bufio_new(struct dm_bufio_client *c, sector_t block, * If the user called both dm_bufio_prefetch and dm_bufio_get on * the same buffer, it would deadlock if we waited. */ - if (nf == NF_GET && unlikely(test_bit_acquire(B_READING, &b->state))) + if (nf == NF_GET && unlikely(test_bit_acquire(B_READING, &b->state))) { + cache_put_and_wake(c, b); return NULL; + } - b->__hold_count++; - __relink_lru(b, test_bit(B_DIRTY, &b->state) || - test_bit(B_WRITING, &b->state)); return b; } @@ -1807,18 +1737,50 @@ static void read_endio(struct dm_buffer *b, blk_status_t status) static void *new_read(struct dm_bufio_client *c, sector_t block, enum new_flag nf, struct dm_buffer **bp) { - int need_submit; + int need_submit = 0; struct dm_buffer *b; LIST_HEAD(write_list); - dm_bufio_lock(c); - b = __bufio_new(c, block, nf, &need_submit, &write_list); + *bp = NULL; + + /* + * Fast path, hopefully the block is already in the cache. No need + * to get the client lock for this. + */ + b = cache_get(&c->cache, block); + if (b) { + if (nf == NF_PREFETCH) { + cache_put_and_wake(c, b); + return NULL; + } + + /* + * Note: it is essential that we don't wait for the buffer to be + * read if dm_bufio_get function is used. Both dm_bufio_get and + * dm_bufio_prefetch can be used in the driver request routine. + * If the user called both dm_bufio_prefetch and dm_bufio_get on + * the same buffer, it would deadlock if we waited. + */ + if (nf == NF_GET && unlikely(test_bit_acquire(B_READING, &b->state))) { + cache_put_and_wake(c, b); + return NULL; + } + } + + if (!b) { + if (nf == NF_GET) + return NULL; + + dm_bufio_lock(c); + b = __bufio_new(c, block, nf, &need_submit, &write_list); + dm_bufio_unlock(c); + } + #ifdef CONFIG_DM_DEBUG_BLOCK_STACK_TRACING - if (b && b->hold_count == 1) + if (b && (atomic_read(&b->hold_count) == 1)) buffer_record_stack(b); #endif - dm_bufio_unlock(c); __flush_write_list(&write_list); @@ -1881,12 +1843,19 @@ void dm_bufio_prefetch(struct dm_bufio_client *c, return; /* should never happen */ blk_start_plug(&plug); - dm_bufio_lock(c); for (; n_blocks--; block++) { int need_submit; struct dm_buffer *b; + b = cache_get(&c->cache, block); + if (b) { + /* already in cache */ + cache_put_and_wake(c, b); + continue; + } + + dm_bufio_lock(c); b = __bufio_new(c, block, NF_PREFETCH, &need_submit, &write_list); if (unlikely(!list_empty(&write_list))) { @@ -1909,10 +1878,9 @@ void dm_bufio_prefetch(struct dm_bufio_client *c, goto flush_plug; dm_bufio_lock(c); } + dm_bufio_unlock(c); } - dm_bufio_unlock(c); - flush_plug: blk_finish_plug(&plug); } @@ -1922,29 +1890,28 @@ void dm_bufio_release(struct dm_buffer *b) { struct dm_bufio_client *c = b->c; - dm_bufio_lock(c); + /* + * If there were errors on the buffer, and the buffer is not + * to be written, free the buffer. There is no point in caching + * invalid buffer. + */ + if ((b->read_error || b->write_error) && + !test_bit_acquire(B_READING, &b->state) && + !test_bit(B_WRITING, &b->state) && + !test_bit(B_DIRTY, &b->state)) { + dm_bufio_lock(c); - BUG_ON(!b->__hold_count); - - b->__hold_count--; - if (!b->__hold_count) { - wake_up(&c->free_buffer_wait); - - /* - * If there were errors on the buffer, and the buffer is not - * to be written, free the buffer. There is no point in caching - * invalid buffer. - */ - if ((b->read_error || b->write_error) && - !test_bit_acquire(B_READING, &b->state) && - !test_bit(B_WRITING, &b->state) && - !test_bit(B_DIRTY, &b->state)) { - __unlink_buffer(b); + /* cache remove can fail if there are other holders */ + if (cache_remove(&c->cache, b)) { __free_buffer_wake(b); + dm_bufio_unlock(c); + return; } + + dm_bufio_unlock(c); } - dm_bufio_unlock(c); + cache_put_and_wake(c, b); } EXPORT_SYMBOL_GPL(dm_bufio_release); @@ -1963,7 +1930,7 @@ void dm_bufio_mark_partial_buffer_dirty(struct dm_buffer *b, if (!test_and_set_bit(B_DIRTY, &b->state)) { b->dirty_start = start; b->dirty_end = end; - __relink_lru(b, LIST_DIRTY); + cache_mark(&c->cache, b, LIST_DIRTY); } else { if (start < b->dirty_start) b->dirty_start = start; @@ -2002,11 +1969,19 @@ EXPORT_SYMBOL_GPL(dm_bufio_write_dirty_buffers_async); * * Finally, we flush hardware disk cache. */ +static bool is_writing(struct lru_entry *e, void *context) +{ + struct dm_buffer *b = le_to_buffer(e); + + return test_bit(B_WRITING, &b->state); +} + int dm_bufio_write_dirty_buffers(struct dm_bufio_client *c) { int a, f; - unsigned long buffers_processed = 0; - struct dm_buffer *b, *tmp; + unsigned long nr_buffers; + struct lru_entry *e; + struct lru_iter it; LIST_HEAD(write_list); @@ -2016,52 +1991,32 @@ int dm_bufio_write_dirty_buffers(struct dm_bufio_client *c) __flush_write_list(&write_list); dm_bufio_lock(c); -again: - list_for_each_entry_safe_reverse(b, tmp, &c->lru[LIST_DIRTY], lru_list) { - int dropped_lock = 0; - - if (buffers_processed < c->n_buffers[LIST_DIRTY]) - buffers_processed++; + nr_buffers = cache_count(&c->cache, LIST_DIRTY); + lru_iter_begin(&c->cache.lru[LIST_DIRTY], &it); + while ((e = lru_iter_next(&it, is_writing, c))) { + struct dm_buffer *b = le_to_buffer(e); + __cache_inc_buffer(b); BUG_ON(test_bit(B_READING, &b->state)); - if (test_bit(B_WRITING, &b->state)) { - if (buffers_processed < c->n_buffers[LIST_DIRTY]) { - dropped_lock = 1; - b->__hold_count++; - dm_bufio_unlock(c); - wait_on_bit_io(&b->state, B_WRITING, - TASK_UNINTERRUPTIBLE); - dm_bufio_lock(c); - b->__hold_count--; - } else - wait_on_bit_io(&b->state, B_WRITING, - TASK_UNINTERRUPTIBLE); + if (nr_buffers) { + nr_buffers--; + dm_bufio_unlock(c); + wait_on_bit_io(&b->state, B_WRITING, TASK_UNINTERRUPTIBLE); + dm_bufio_lock(c); + } else { + wait_on_bit_io(&b->state, B_WRITING, TASK_UNINTERRUPTIBLE); } - if (!test_bit(B_DIRTY, &b->state) && - !test_bit(B_WRITING, &b->state)) - __relink_lru(b, LIST_CLEAN); + if (!test_bit(B_DIRTY, &b->state) && !test_bit(B_WRITING, &b->state)) + cache_mark(&c->cache, b, LIST_CLEAN); + + cache_put_and_wake(c, b); cond_resched(); - - /* - * If we dropped the lock, the list is no longer consistent, - * so we must restart the search. - * - * In the most common case, the buffer just processed is - * relinked to the clean list, so we won't loop scanning the - * same buffer again and again. - * - * This may livelock if there is another thread simultaneously - * dirtying buffers, so we count the number of buffers walked - * and if it exceeds the total number of buffers, it means that - * someone is doing some writes simultaneously with us. In - * this case, stop, dropping the lock. - */ - if (dropped_lock) - goto again; } + lru_iter_end(&it); + wake_up(&c->free_buffer_wait); dm_bufio_unlock(c); @@ -2122,12 +2077,23 @@ int dm_bufio_issue_discard(struct dm_bufio_client *c, sector_t block, sector_t c } EXPORT_SYMBOL_GPL(dm_bufio_issue_discard); -static void forget_buffer_locked(struct dm_buffer *b) +static bool forget_buffer(struct dm_bufio_client *c, sector_t block) { - if (likely(!b->__hold_count) && likely(!smp_load_acquire(&b->state))) { - __unlink_buffer(b); - __free_buffer_wake(b); + struct dm_buffer *b; + + b = cache_get(&c->cache, block); + if (b) { + if (likely(!smp_load_acquire(&b->state))) { + if (cache_remove(&c->cache, b)) + __free_buffer_wake(b); + else + cache_put_and_wake(c, b); + } else { + cache_put_and_wake(c, b); + } } + + return b ? true : false; } /* @@ -2138,38 +2104,22 @@ static void forget_buffer_locked(struct dm_buffer *b) */ void dm_bufio_forget(struct dm_bufio_client *c, sector_t block) { - struct dm_buffer *b; - dm_bufio_lock(c); - - b = __find(c, block); - if (b) - forget_buffer_locked(b); - + forget_buffer(c, block); dm_bufio_unlock(c); } EXPORT_SYMBOL_GPL(dm_bufio_forget); +static enum evict_result idle(struct dm_buffer *b, void *context) +{ + return b->state ? ER_DONT_EVICT : ER_EVICT; +} + void dm_bufio_forget_buffers(struct dm_bufio_client *c, sector_t block, sector_t n_blocks) { - struct dm_buffer *b; - sector_t end_block = block + n_blocks; - - while (block < end_block) { - dm_bufio_lock(c); - - b = __find_next_old(c, block); - if (b) { - block = b->block + 1; - forget_buffer_locked(b); - } - - dm_bufio_unlock(c); - - if (!b) - break; - } - + dm_bufio_lock(c); + cache_remove_range(&c->cache, block, block + n_blocks, idle, __free_buffer_wake); + dm_bufio_unlock(c); } EXPORT_SYMBOL_GPL(dm_bufio_forget_buffers); @@ -2231,11 +2181,26 @@ struct dm_bufio_client *dm_bufio_get_client(struct dm_buffer *b) } EXPORT_SYMBOL_GPL(dm_bufio_get_client); +static enum it_action warn_leak(struct dm_buffer *b, void *context) +{ + bool *warned = context; + + WARN_ON(!(*warned)); + *warned = true; + DMERR("leaked buffer %llx, hold count %u, list %d", + (unsigned long long)b->block, atomic_read(&b->hold_count), b->list_mode); +#ifdef CONFIG_DM_DEBUG_BLOCK_STACK_TRACING + stack_trace_print(b->stack_entries, b->stack_len, 1); + /* mark unclaimed to avoid WARN_ON at end of drop_buffers() */ + atomic_set(&b->hold_count, 0); +#endif + return IT_NEXT; +} + static void drop_buffers(struct dm_bufio_client *c) { - struct dm_buffer *b; int i; - bool warned = false; + struct dm_buffer *b; if (WARN_ON(dm_bufio_in_request())) return; /* should never happen */ @@ -2250,18 +2215,11 @@ static void drop_buffers(struct dm_bufio_client *c) while ((b = __get_unclaimed_buffer(c))) __free_buffer_wake(b); - for (i = 0; i < LIST_SIZE; i++) - list_for_each_entry(b, &c->lru[i], lru_list) { - WARN_ON(!warned); - warned = true; - DMERR("leaked buffer %llx, hold count %u, list %d", - (unsigned long long)b->block, b->__hold_count, i); -#ifdef CONFIG_DM_DEBUG_BLOCK_STACK_TRACING - stack_trace_print(b->stack_entries, b->stack_len, 1); - /* mark unclaimed to avoid WARN_ON below */ - b->hold_count = 0; -#endif - } + for (i = 0; i < LIST_SIZE; i++) { + bool warned = false; + + cache_iterate(&c->cache, i, warn_leak, &warned); + } #ifdef CONFIG_DM_DEBUG_BLOCK_STACK_TRACING while ((b = __get_unclaimed_buffer(c))) @@ -2269,39 +2227,11 @@ static void drop_buffers(struct dm_bufio_client *c) #endif for (i = 0; i < LIST_SIZE; i++) - WARN_ON(!list_empty(&c->lru[i])); + WARN_ON(cache_count(&c->cache, i)); dm_bufio_unlock(c); } -/* - * We may not be able to evict this buffer if IO pending or the client - * is still using it. Caller is expected to know buffer is too old. - * - * And if GFP_NOFS is used, we must not do any I/O because we hold - * dm_bufio_clients_lock and we would risk deadlock if the I/O gets - * rerouted to different bufio client. - */ -static bool __try_evict_buffer(struct dm_buffer *b, gfp_t gfp) -{ - if (!(gfp & __GFP_FS) || - (static_branch_unlikely(&no_sleep_enabled) && b->c->no_sleep)) { - if (test_bit_acquire(B_READING, &b->state) || - test_bit(B_WRITING, &b->state) || - test_bit(B_DIRTY, &b->state)) - return false; - } - - if (b->__hold_count) - return false; - - __make_buffer_clean(b); - __unlink_buffer(b); - __free_buffer_wake(b); - - return true; -} - static unsigned long get_retain_buffers(struct dm_bufio_client *c) { unsigned long retain_bytes = READ_ONCE(dm_bufio_retain_bytes); @@ -2317,22 +2247,28 @@ static unsigned long get_retain_buffers(struct dm_bufio_client *c) static void __scan(struct dm_bufio_client *c) { int l; - struct dm_buffer *b, *tmp; + struct dm_buffer *b; unsigned long freed = 0; - unsigned long count = c->n_buffers[LIST_CLEAN] + - c->n_buffers[LIST_DIRTY]; unsigned long retain_target = get_retain_buffers(c); + unsigned long count = cache_total(&c->cache); for (l = 0; l < LIST_SIZE; l++) { - list_for_each_entry_safe_reverse(b, tmp, &c->lru[l], lru_list) { + while (true) { if (count - freed <= retain_target) atomic_long_set(&c->need_shrink, 0); if (!atomic_long_read(&c->need_shrink)) - return; - if (__try_evict_buffer(b, GFP_KERNEL)) { - atomic_long_dec(&c->need_shrink); - freed++; - } + break; + + b = cache_evict(&c->cache, l, + l == LIST_CLEAN ? is_clean : is_dirty, c); + if (!b) + break; + + __make_buffer_clean(b); + __free_buffer_wake(b); + + atomic_long_dec(&c->need_shrink); + freed++; cond_resched(); } } @@ -2361,8 +2297,7 @@ static unsigned long dm_bufio_shrink_scan(struct shrinker *shrink, struct shrink static unsigned long dm_bufio_shrink_count(struct shrinker *shrink, struct shrink_control *sc) { struct dm_bufio_client *c = container_of(shrink, struct dm_bufio_client, shrinker); - unsigned long count = READ_ONCE(c->n_buffers[LIST_CLEAN]) + - READ_ONCE(c->n_buffers[LIST_DIRTY]); + unsigned long count = cache_total(&c->cache); unsigned long retain_target = get_retain_buffers(c); unsigned long queued_for_cleanup = atomic_long_read(&c->need_shrink); @@ -2390,7 +2325,6 @@ struct dm_bufio_client *dm_bufio_client_create(struct block_device *bdev, unsign { int r; struct dm_bufio_client *c; - unsigned int i; char slab_name[27]; if (!block_size || block_size & ((1 << SECTOR_SHIFT) - 1)) { @@ -2404,7 +2338,7 @@ struct dm_bufio_client *dm_bufio_client_create(struct block_device *bdev, unsign r = -ENOMEM; goto bad_client; } - c->buffer_tree = RB_ROOT; + cache_init(&c->cache); c->bdev = bdev; c->block_size = block_size; @@ -2421,11 +2355,6 @@ struct dm_bufio_client *dm_bufio_client_create(struct block_device *bdev, unsign static_branch_inc(&no_sleep_enabled); } - for (i = 0; i < LIST_SIZE; i++) { - INIT_LIST_HEAD(&c->lru[i]); - c->n_buffers[i] = 0; - } - mutex_init(&c->lock); spin_lock_init(&c->spinlock); INIT_LIST_HEAD(&c->reserved_buffers); @@ -2497,9 +2426,9 @@ struct dm_bufio_client *dm_bufio_client_create(struct block_device *bdev, unsign bad: while (!list_empty(&c->reserved_buffers)) { - struct dm_buffer *b = list_entry(c->reserved_buffers.next, - struct dm_buffer, lru_list); - list_del(&b->lru_list); + struct dm_buffer *b = list_to_buffer(c->reserved_buffers.next); + + list_del(&b->lru.list); free_buffer(b); } kmem_cache_destroy(c->slab_cache); @@ -2536,23 +2465,23 @@ void dm_bufio_client_destroy(struct dm_bufio_client *c) mutex_unlock(&dm_bufio_clients_lock); - WARN_ON(!RB_EMPTY_ROOT(&c->buffer_tree)); WARN_ON(c->need_reserved_buffers); while (!list_empty(&c->reserved_buffers)) { - struct dm_buffer *b = list_entry(c->reserved_buffers.next, - struct dm_buffer, lru_list); - list_del(&b->lru_list); + struct dm_buffer *b = list_to_buffer(c->reserved_buffers.next); + + list_del(&b->lru.list); free_buffer(b); } for (i = 0; i < LIST_SIZE; i++) - if (c->n_buffers[i]) - DMERR("leaked buffer count %d: %ld", i, c->n_buffers[i]); + if (cache_count(&c->cache, i)) + DMERR("leaked buffer count %d: %lu", i, cache_count(&c->cache, i)); for (i = 0; i < LIST_SIZE; i++) - WARN_ON(c->n_buffers[i]); + WARN_ON(cache_count(&c->cache, i)); + cache_destroy(&c->cache); kmem_cache_destroy(c->slab_cache); kmem_cache_destroy(c->slab_buffer); dm_io_client_destroy(c->dm_io); @@ -2569,6 +2498,8 @@ void dm_bufio_set_sector_offset(struct dm_bufio_client *c, sector_t start) } EXPORT_SYMBOL_GPL(dm_bufio_set_sector_offset); +/*--------------------------------------------------------------*/ + static unsigned int get_max_age_hz(void) { unsigned int max_age = READ_ONCE(dm_bufio_max_age); @@ -2581,13 +2512,74 @@ static unsigned int get_max_age_hz(void) static bool older_than(struct dm_buffer *b, unsigned long age_hz) { - return time_after_eq(jiffies, b->last_accessed + age_hz); + return time_after_eq(jiffies, READ_ONCE(b->last_accessed) + age_hz); } -static void __evict_old_buffers(struct dm_bufio_client *c, unsigned long age_hz) +struct evict_params { + gfp_t gfp; + unsigned long age_hz; + + /* + * This gets updated with the largest last_accessed (ie. most + * recently used) of the evicted buffers. It will not be reinitialised + * by __evict_many(), so you can use it across multiple invocations. + */ + unsigned long last_accessed; +}; + +/* + * We may not be able to evict this buffer if IO pending or the client + * is still using it. + * + * And if GFP_NOFS is used, we must not do any I/O because we hold + * dm_bufio_clients_lock and we would risk deadlock if the I/O gets + * rerouted to different bufio client. + */ +static enum evict_result select_for_evict(struct dm_buffer *b, void *context) +{ + struct evict_params *params = context; + + if (!(params->gfp & __GFP_FS) || + (static_branch_unlikely(&no_sleep_enabled) && b->c->no_sleep)) { + if (test_bit_acquire(B_READING, &b->state) || + test_bit(B_WRITING, &b->state) || + test_bit(B_DIRTY, &b->state)) + return ER_DONT_EVICT; + } + + return older_than(b, params->age_hz) ? ER_EVICT : ER_STOP; +} + +static unsigned long __evict_many(struct dm_bufio_client *c, + struct evict_params *params, + int list_mode, unsigned long max_count) +{ + unsigned long count; + unsigned long last_accessed; + struct dm_buffer *b; + + for (count = 0; count < max_count; count++) { + b = cache_evict(&c->cache, list_mode, select_for_evict, params); + if (!b) + break; + + last_accessed = READ_ONCE(b->last_accessed); + if (time_after_eq(params->last_accessed, last_accessed)) + params->last_accessed = last_accessed; + + __make_buffer_clean(b); + __free_buffer_wake(b); + + cond_resched(); + } + + return count; +} + +static void evict_old_buffers(struct dm_bufio_client *c, unsigned long age_hz) { - struct dm_buffer *b, *tmp; - unsigned long retain_target = get_retain_buffers(c); + struct evict_params params = {.gfp = 0, .age_hz = age_hz, .last_accessed = 0}; + unsigned long retain = get_retain_buffers(c); unsigned long count; LIST_HEAD(write_list); @@ -2600,91 +2592,13 @@ static void __evict_old_buffers(struct dm_bufio_client *c, unsigned long age_hz) dm_bufio_lock(c); } - count = c->n_buffers[LIST_CLEAN] + c->n_buffers[LIST_DIRTY]; - list_for_each_entry_safe_reverse(b, tmp, &c->lru[LIST_CLEAN], lru_list) { - if (count <= retain_target) - break; - - if (!older_than(b, age_hz)) - break; - - if (__try_evict_buffer(b, 0)) - count--; - - cond_resched(); - } + count = cache_total(&c->cache); + if (count > retain) + __evict_many(c, ¶ms, LIST_CLEAN, count - retain); dm_bufio_unlock(c); } -static void do_global_cleanup(struct work_struct *w) -{ - struct dm_bufio_client *locked_client = NULL; - struct dm_bufio_client *current_client; - struct dm_buffer *b; - unsigned int spinlock_hold_count; - unsigned long threshold = dm_bufio_cache_size - - dm_bufio_cache_size / DM_BUFIO_LOW_WATERMARK_RATIO; - unsigned long loops = global_num * 2; - - mutex_lock(&dm_bufio_clients_lock); - - while (1) { - cond_resched(); - - spin_lock(&global_spinlock); - if (unlikely(dm_bufio_current_allocated <= threshold)) - break; - - spinlock_hold_count = 0; -get_next: - if (!loops--) - break; - if (unlikely(list_empty(&global_queue))) - break; - b = list_entry(global_queue.prev, struct dm_buffer, global_list); - - if (b->accessed) { - b->accessed = 0; - list_move(&b->global_list, &global_queue); - if (likely(++spinlock_hold_count < 16)) - goto get_next; - spin_unlock(&global_spinlock); - continue; - } - - current_client = b->c; - if (unlikely(current_client != locked_client)) { - if (locked_client) - dm_bufio_unlock(locked_client); - - if (!dm_bufio_trylock(current_client)) { - spin_unlock(&global_spinlock); - dm_bufio_lock(current_client); - locked_client = current_client; - continue; - } - - locked_client = current_client; - } - - spin_unlock(&global_spinlock); - - if (unlikely(!__try_evict_buffer(b, GFP_KERNEL))) { - spin_lock(&global_spinlock); - list_move(&b->global_list, &global_queue); - spin_unlock(&global_spinlock); - } - } - - spin_unlock(&global_spinlock); - - if (locked_client) - dm_bufio_unlock(locked_client); - - mutex_unlock(&dm_bufio_clients_lock); -} - static void cleanup_old_buffers(void) { unsigned long max_age_hz = get_max_age_hz(); @@ -2695,7 +2609,7 @@ static void cleanup_old_buffers(void) __cache_size_refresh(); list_for_each_entry(c, &dm_bufio_all_clients, client_list) - __evict_old_buffers(c, max_age_hz); + evict_old_buffers(c, max_age_hz); mutex_unlock(&dm_bufio_clients_lock); } @@ -2708,6 +2622,107 @@ static void work_fn(struct work_struct *w) DM_BUFIO_WORK_TIMER_SECS * HZ); } +/*--------------------------------------------------------------*/ + +/* + * Global cleanup tries to evict the oldest buffers from across _all_ + * the clients. It does this by repeatedly evicting a few buffers from + * the client that holds the oldest buffer. It's approximate, but hopefully + * good enough. + */ +static struct dm_bufio_client *__pop_client(void) +{ + struct list_head *h; + + if (list_empty(&dm_bufio_all_clients)) + return NULL; + + h = dm_bufio_all_clients.next; + list_del(h); + return container_of(h, struct dm_bufio_client, client_list); +} + +/* + * Inserts the client in the global client list based on its + * 'oldest_buffer' field. + */ +static void __insert_client(struct dm_bufio_client *new_client) +{ + struct dm_bufio_client *c; + struct list_head *h = dm_bufio_all_clients.next; + + while (h != &dm_bufio_all_clients) { + c = container_of(h, struct dm_bufio_client, client_list); + if (time_after_eq(c->oldest_buffer, new_client->oldest_buffer)) + break; + h = h->next; + } + + list_add_tail(&new_client->client_list, h); +} + +static unsigned long __evict_a_few(unsigned long nr_buffers) +{ + unsigned long count; + struct dm_bufio_client *c; + struct evict_params params = { + .gfp = GFP_KERNEL, + .age_hz = 0, + /* set to jiffies in case there are no buffers in this client */ + .last_accessed = jiffies + }; + + c = __pop_client(); + if (!c) + return 0; + + dm_bufio_lock(c); + count = __evict_many(c, ¶ms, LIST_CLEAN, nr_buffers); + dm_bufio_unlock(c); + + if (count) + c->oldest_buffer = params.last_accessed; + __insert_client(c); + + return count; +} + +static void check_watermarks(void) +{ + LIST_HEAD(write_list); + struct dm_bufio_client *c; + + mutex_lock(&dm_bufio_clients_lock); + list_for_each_entry(c, &dm_bufio_all_clients, client_list) { + dm_bufio_lock(c); + __check_watermark(c, &write_list); + dm_bufio_unlock(c); + } + mutex_unlock(&dm_bufio_clients_lock); + + __flush_write_list(&write_list); +} + +static void evict_old(void) +{ + unsigned long threshold = dm_bufio_cache_size - + dm_bufio_cache_size / DM_BUFIO_LOW_WATERMARK_RATIO; + + mutex_lock(&dm_bufio_clients_lock); + while (dm_bufio_current_allocated > threshold) { + if (!__evict_a_few(64)) + break; + cond_resched(); + } + mutex_unlock(&dm_bufio_clients_lock); +} + +static void do_global_cleanup(struct work_struct *w) +{ + check_watermarks(); + evict_old(); +} + /* *-------------------------------------------------------------- * Module setup From patchwork Mon Mar 27 20:11:31 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Mike Snitzer X-Patchwork-Id: 13190053 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 vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id D5C90C77B60 for ; Mon, 27 Mar 2023 20:14:12 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S232071AbjC0UOK (ORCPT ); Mon, 27 Mar 2023 16:14:10 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:38740 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S231929AbjC0UOJ (ORCPT ); Mon, 27 Mar 2023 16:14:09 -0400 Received: from mail-qt1-f175.google.com (mail-qt1-f175.google.com [209.85.160.175]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id 1213BA6 for ; Mon, 27 Mar 2023 13:13:21 -0700 (PDT) Received: by mail-qt1-f175.google.com with SMTP id s12so6188811qtx.11 for ; Mon, 27 Mar 2023 13:13:21 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; t=1679948000; h=content-transfer-encoding:mime-version:references:in-reply-to :message-id:date:subject:cc:to:from:x-gm-message-state:from:to:cc :subject:date:message-id:reply-to; bh=dgaYGvMwSGrKRZ2WwE4xFce2cAfLE729BA8da+BNlMU=; b=hiwtCD/N/Ui4N4yjROOUzgjGal9lmJHqdhlVfdAdLWepF4ivwcXPLoJOGNc5doC+W4 yg/xgvLLJ+c06mPfxM4C55qFyJG0ii4E3x8g0OGc1oImSCHkqqnsdIn+ICAkvbtAQmQv +xFD1TnZ+4iMNUm9fBodYF2XpA0DO7TqNgAuK6U/P/G74UfojFWbsc6OT3qrAxaBHvUB BeRmZ+BS/zrPO9fsR5K4F+iJ51JMimA2oWdtUmsLClU+NbwYOYEW0PnaFvqy+9y7hkYA rluLUuO8yn1+E8hcr2MewP+mwEQgavdDa8U4cJPxqF3CKJL+jcGS687bR0KQNiZ45jQi lWdA== X-Gm-Message-State: AO0yUKU92yK2SGQoMHakM9DCEUPIOwHXnc+XSlrNL29LnymNQ2L4Efwf Q/BquYjAvWWMUZuUWroxvbok X-Google-Smtp-Source: AK7set9AVHWFXkb3npjJh0xtSby8NWtVc9+BAEPPLkZMxF17dzC/dc7TdH75nxfrpPvYHfLhi0sMtQ== X-Received: by 2002:a05:622a:198b:b0:3e2:6a40:5631 with SMTP id u11-20020a05622a198b00b003e26a405631mr22093845qtc.10.1679948000120; Mon, 27 Mar 2023 13:13:20 -0700 (PDT) Received: from localhost (pool-68-160-166-30.bstnma.fios.verizon.net. [68.160.166.30]) by smtp.gmail.com with ESMTPSA id t23-20020ac865d7000000b003b635a5d56csm15119212qto.30.2023.03.27.13.13.19 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Mon, 27 Mar 2023 13:13:19 -0700 (PDT) From: Mike Snitzer To: dm-devel@redhat.com Cc: linux-block@vger.kernel.org, axboe@kernel.dk, ejt@redhat.com, mpatocka@redhat.com, heinzm@redhat.com, nhuck@google.com, ebiggers@kernel.org, keescook@chromium.org, luomeng12@huawei.com, Mike Snitzer Subject: [dm-6.4 PATCH v3 08/20] dm bufio: add lock_history optimization for cache iterators Date: Mon, 27 Mar 2023 16:11:31 -0400 Message-Id: <20230327201143.51026-9-snitzer@kernel.org> X-Mailer: git-send-email 2.40.0 In-Reply-To: <20230327201143.51026-1-snitzer@kernel.org> References: <20230327201143.51026-1-snitzer@kernel.org> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-block@vger.kernel.org From: Joe Thornber Sometimes it is beneficial to repeatedly get and drop locks as part of an iteration. Introduce lock_history struct to help avoid redundant drop and gets of the same lock. Optimizes cache_iterate, cache_mark_many and cache_evict. Signed-off-by: Joe Thornber Signed-off-by: Mike Snitzer --- drivers/md/dm-bufio.c | 119 +++++++++++++++++++++++++++++++++++++++--- 1 file changed, 111 insertions(+), 8 deletions(-) diff --git a/drivers/md/dm-bufio.c b/drivers/md/dm-bufio.c index 1e000ec73bd6..9ac50024006d 100644 --- a/drivers/md/dm-bufio.c +++ b/drivers/md/dm-bufio.c @@ -421,6 +421,70 @@ static inline void cache_write_unlock(struct dm_buffer_cache *bc, sector_t block up_write(&bc->trees[cache_index(block)].lock); } +/* + * Sometimes we want to repeatedly get and drop locks as part of an iteration. + * This struct helps avoid redundant drop and gets of the same lock. + */ +struct lock_history { + struct dm_buffer_cache *cache; + bool write; + unsigned int previous; +}; + +static void lh_init(struct lock_history *lh, struct dm_buffer_cache *cache, bool write) +{ + lh->cache = cache; + lh->write = write; + lh->previous = NR_LOCKS; /* indicates no previous */ +} + +static void __lh_lock(struct lock_history *lh, unsigned int index) +{ + if (lh->write) + down_write(&lh->cache->trees[index].lock); + else + down_read(&lh->cache->trees[index].lock); +} + +static void __lh_unlock(struct lock_history *lh, unsigned int index) +{ + if (lh->write) + up_write(&lh->cache->trees[index].lock); + else + up_read(&lh->cache->trees[index].lock); +} + +/* + * Make sure you call this since it will unlock the final lock. + */ +static void lh_exit(struct lock_history *lh) +{ + if (lh->previous != NR_LOCKS) { + __lh_unlock(lh, lh->previous); + lh->previous = NR_LOCKS; + } +} + +/* + * Named 'next' because there is no corresponding + * 'up/unlock' call since it's done automatically. + */ +static void lh_next(struct lock_history *lh, sector_t b) +{ + unsigned int index = cache_index(b); + + if (lh->previous != NR_LOCKS) { + if (lh->previous != index) { + __lh_unlock(lh, lh->previous); + __lh_lock(lh, index); + lh->previous = index; + } + } else { + __lh_lock(lh, index); + lh->previous = index; + } +} + static inline struct dm_buffer *le_to_buffer(struct lru_entry *le) { return container_of(le, struct dm_buffer, lru); @@ -550,6 +614,7 @@ typedef enum evict_result (*b_predicate)(struct dm_buffer *, void *); * predicate the hold_count of the selected buffer will be zero. */ struct evict_wrapper { + struct lock_history *lh; b_predicate pred; void *context; }; @@ -563,16 +628,19 @@ static enum evict_result __evict_pred(struct lru_entry *le, void *context) struct evict_wrapper *w = context; struct dm_buffer *b = le_to_buffer(le); + lh_next(w->lh, b->block); + if (atomic_read(&b->hold_count)) return ER_DONT_EVICT; return w->pred(b, w->context); } -static struct dm_buffer *cache_evict(struct dm_buffer_cache *bc, int list_mode, - b_predicate pred, void *context) +static struct dm_buffer *__cache_evict(struct dm_buffer_cache *bc, int list_mode, + b_predicate pred, void *context, + struct lock_history *lh) { - struct evict_wrapper w = {.pred = pred, .context = context}; + struct evict_wrapper w = {.lh = lh, .pred = pred, .context = context}; struct lru_entry *le; struct dm_buffer *b; @@ -587,6 +655,19 @@ static struct dm_buffer *cache_evict(struct dm_buffer_cache *bc, int list_mode, return b; } +static struct dm_buffer *cache_evict(struct dm_buffer_cache *bc, int list_mode, + b_predicate pred, void *context) +{ + struct dm_buffer *b; + struct lock_history lh; + + lh_init(&lh, bc, true); + b = __cache_evict(bc, list_mode, pred, context, &lh); + lh_exit(&lh); + + return b; +} + /*--------------*/ /* @@ -609,12 +690,12 @@ static void cache_mark(struct dm_buffer_cache *bc, struct dm_buffer *b, int list * Runs through the lru associated with 'old_mode', if the predicate matches then * it moves them to 'new_mode'. Not threadsafe. */ -static void cache_mark_many(struct dm_buffer_cache *bc, int old_mode, int new_mode, - b_predicate pred, void *context) +static void __cache_mark_many(struct dm_buffer_cache *bc, int old_mode, int new_mode, + b_predicate pred, void *context, struct lock_history *lh) { struct lru_entry *le; struct dm_buffer *b; - struct evict_wrapper w = {.pred = pred, .context = context}; + struct evict_wrapper w = {.lh = lh, .pred = pred, .context = context}; while (true) { le = lru_evict(&bc->lru[old_mode], __evict_pred, &w); @@ -627,6 +708,16 @@ static void cache_mark_many(struct dm_buffer_cache *bc, int old_mode, int new_mo } } +static void cache_mark_many(struct dm_buffer_cache *bc, int old_mode, int new_mode, + b_predicate pred, void *context) +{ + struct lock_history lh; + + lh_init(&lh, bc, true); + __cache_mark_many(bc, old_mode, new_mode, pred, context, &lh); + lh_exit(&lh); +} + /*--------------*/ /* @@ -645,8 +736,8 @@ enum it_action { typedef enum it_action (*iter_fn)(struct dm_buffer *b, void *context); -static void cache_iterate(struct dm_buffer_cache *bc, int list_mode, - iter_fn fn, void *context) +static void __cache_iterate(struct dm_buffer_cache *bc, int list_mode, + iter_fn fn, void *context, struct lock_history *lh) { struct lru *lru = &bc->lru[list_mode]; struct lru_entry *le, *first; @@ -658,6 +749,8 @@ static void cache_iterate(struct dm_buffer_cache *bc, int list_mode, do { struct dm_buffer *b = le_to_buffer(le); + lh_next(lh, b->block); + switch (fn(b, context)) { case IT_NEXT: break; @@ -671,6 +764,16 @@ static void cache_iterate(struct dm_buffer_cache *bc, int list_mode, } while (le != first); } +static void cache_iterate(struct dm_buffer_cache *bc, int list_mode, + iter_fn fn, void *context) +{ + struct lock_history lh; + + lh_init(&lh, bc, false); + __cache_iterate(bc, list_mode, fn, context, &lh); + lh_exit(&lh); +} + /*--------------*/ /* From patchwork Mon Mar 27 20:11:32 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Mike Snitzer X-Patchwork-Id: 13190051 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 vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id 1BBB9C6FD1D for ; Mon, 27 Mar 2023 20:14:09 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S232140AbjC0UOH (ORCPT ); Mon, 27 Mar 2023 16:14:07 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:39004 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S231929AbjC0UOG (ORCPT ); Mon, 27 Mar 2023 16:14:06 -0400 Received: from mail-qt1-f181.google.com (mail-qt1-f181.google.com [209.85.160.181]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id 060C81BFD for ; Mon, 27 Mar 2023 13:13:22 -0700 (PDT) Received: by mail-qt1-f181.google.com with SMTP id cr18so5981723qtb.0 for ; Mon, 27 Mar 2023 13:13:21 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; t=1679948001; h=content-transfer-encoding:mime-version:references:in-reply-to :message-id:date:subject:cc:to:from:x-gm-message-state:from:to:cc :subject:date:message-id:reply-to; bh=H9SedxOrdczzTjA4k7blz5Fc7tZ0dfZriRfxA3Ti63I=; b=o6M0OlzKJOuribECt8iVPIfFtmtO4Tc7V0v/FXjM/gzCbtTc/LYxdQSZLh+2mcvHfU 90Do/GtYfocPQ6UvY7JFWRn1/OD9IxDDw6wyMfx1YCE+ncufDUlLVPZAsNoyTXUrk0qc md/gsBr0KCTi3CLTk4UYWpi4VkTJ+MGvJ+uuFtnQfkym/PtGGF7Cx94HFfYhmhtW43wS 1qeyNYyxuhFt0vMEXGlk+qRaaoc1qcDWlcYcA02BgdAGiE3+vhM5qTODgqZ1P8fDmarU SgAgHW+nNMM2Vekeie55QxHhc8sczgjCCQzySJlSV+Wrjs0mHwqisARRqteLnoEoCl// 8b+w== X-Gm-Message-State: AO0yUKV8IcwQovcs9Xn4J8Y3KbxK0Hi5rdx/kcUTUoaYskH9MQAUf/ye 3SXl9B34EQCqX/h6fwozA0Cr X-Google-Smtp-Source: AK7set98Y7dLUBXYMxUAgPYJRwSwe5G/4pJt7wTHoABVMGSpSmarcdzS7b1UEdwaYPt5vQLyrZsh/w== X-Received: by 2002:a05:622a:170b:b0:3e3:7e24:a35a with SMTP id h11-20020a05622a170b00b003e37e24a35amr21832421qtk.9.1679948001588; Mon, 27 Mar 2023 13:13:21 -0700 (PDT) Received: from localhost (pool-68-160-166-30.bstnma.fios.verizon.net. [68.160.166.30]) by smtp.gmail.com with ESMTPSA id 201-20020a3708d2000000b007456efa7f73sm12278213qki.85.2023.03.27.13.13.20 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Mon, 27 Mar 2023 13:13:21 -0700 (PDT) From: Mike Snitzer To: dm-devel@redhat.com Cc: linux-block@vger.kernel.org, axboe@kernel.dk, ejt@redhat.com, mpatocka@redhat.com, heinzm@redhat.com, nhuck@google.com, ebiggers@kernel.org, keescook@chromium.org, luomeng12@huawei.com, Mike Snitzer Subject: [dm-6.4 PATCH v3 09/20] dm bufio: move dm_bufio_client members to avoid spanning cachelines Date: Mon, 27 Mar 2023 16:11:32 -0400 Message-Id: <20230327201143.51026-10-snitzer@kernel.org> X-Mailer: git-send-email 2.40.0 In-Reply-To: <20230327201143.51026-1-snitzer@kernel.org> References: <20230327201143.51026-1-snitzer@kernel.org> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-block@vger.kernel.org Movement also consolidates holes in dm_bufio_client struct. But the overall size of the struct isn't changed. Signed-off-by: Mike Snitzer --- drivers/md/dm-bufio.c | 24 +++++++++++++----------- 1 file changed, 13 insertions(+), 11 deletions(-) diff --git a/drivers/md/dm-bufio.c b/drivers/md/dm-bufio.c index 9ac50024006d..e5459741335d 100644 --- a/drivers/md/dm-bufio.c +++ b/drivers/md/dm-bufio.c @@ -936,13 +936,16 @@ static void cache_remove_range(struct dm_buffer_cache *bc, * context. */ struct dm_bufio_client { - struct mutex lock; - spinlock_t spinlock; - bool no_sleep; - struct block_device *bdev; unsigned int block_size; s8 sectors_per_block_bits; + + bool no_sleep; + struct mutex lock; + spinlock_t spinlock; + + int async_write_error; + void (*alloc_callback)(struct dm_buffer *buf); void (*write_callback)(struct dm_buffer *buf); struct kmem_cache *slab_buffer; @@ -954,23 +957,22 @@ struct dm_bufio_client { unsigned int minimum_buffers; - struct dm_buffer_cache cache; - wait_queue_head_t free_buffer_wait; - sector_t start; - int async_write_error; - - struct list_head client_list; - struct shrinker shrinker; struct work_struct shrink_work; atomic_long_t need_shrink; + wait_queue_head_t free_buffer_wait; + + struct list_head client_list; + /* * Used by global_cleanup to sort the clients list. */ unsigned long oldest_buffer; + + struct dm_buffer_cache cache; }; static DEFINE_STATIC_KEY_FALSE(no_sleep_enabled); From patchwork Mon Mar 27 20:11:33 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Mike Snitzer X-Patchwork-Id: 13190054 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 vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id 42D7CC761A6 for ; Mon, 27 Mar 2023 20:14:15 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S232225AbjC0UOO (ORCPT ); Mon, 27 Mar 2023 16:14:14 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:39002 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S232183AbjC0UOK (ORCPT ); Mon, 27 Mar 2023 16:14:10 -0400 Received: from mail-qt1-f175.google.com (mail-qt1-f175.google.com [209.85.160.175]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id 8A4F61BFB for ; Mon, 27 Mar 2023 13:13:23 -0700 (PDT) Received: by mail-qt1-f175.google.com with SMTP id s12so6188935qtx.11 for ; Mon, 27 Mar 2023 13:13:23 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; t=1679948003; h=content-transfer-encoding:mime-version:references:in-reply-to :message-id:date:subject:cc:to:from:x-gm-message-state:from:to:cc :subject:date:message-id:reply-to; bh=W8N6ipnCV5hoHhg/CORvER6nf9MSwcb0kouKMhOxJx4=; b=nAU6YSBQ7rhroO43s5qQXgewW+numV/72F5Clec5PdwRG5bMYR1nnyCvKKsaa/NQQb zMdVv55mwsm8cGV+s0PNRMjq05tB8oi9oIdIKg08t7G03h/SRgH1pbH4nlzpKdTGIVUE g7oBCI/f+z4xFPkgUyxOV430XpfoMFnLNfn6DaLcArIuN47NTjxdPdIBtCwaHrCU8eHn 2v7PUKjYWWlw3x8EoSZ+tAyz8qMvPv1dsVVvGDNH4n05S0STYYKPueuZgQb2hykbXCNq 2A6t8SjD1TP0ZhbwJNknCp3IFJAxJwv9z9Hbt0ci9/ffpzzIQLIoQwKcUWOEWlTG14s5 nAUQ== X-Gm-Message-State: AO0yUKV1VzXuzfOcgxZ/pQFHVJ88URXQEgZtSBmr18n+4UcrMusTucks vTiGkx2uWgSGUex1pEcENo6P X-Google-Smtp-Source: AK7set9+sfwAJb0hRAS53hihYhDVSWnGpR49MCgCFLmIkZazfZgDsImOqvtZXurStI3emQaDQ0Dvkw== X-Received: by 2002:a05:622a:1315:b0:3d5:500a:4819 with SMTP id v21-20020a05622a131500b003d5500a4819mr22550964qtk.23.1679948003245; Mon, 27 Mar 2023 13:13:23 -0700 (PDT) Received: from localhost (pool-68-160-166-30.bstnma.fios.verizon.net. [68.160.166.30]) by smtp.gmail.com with ESMTPSA id y126-20020a376484000000b007468765b411sm11876942qkb.45.2023.03.27.13.13.22 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Mon, 27 Mar 2023 13:13:22 -0700 (PDT) From: Mike Snitzer To: dm-devel@redhat.com Cc: linux-block@vger.kernel.org, axboe@kernel.dk, ejt@redhat.com, mpatocka@redhat.com, heinzm@redhat.com, nhuck@google.com, ebiggers@kernel.org, keescook@chromium.org, luomeng12@huawei.com, Mike Snitzer Subject: [dm-6.4 PATCH v3 10/20] dm bufio: use waitqueue_active in __free_buffer_wake Date: Mon, 27 Mar 2023 16:11:33 -0400 Message-Id: <20230327201143.51026-11-snitzer@kernel.org> X-Mailer: git-send-email 2.40.0 In-Reply-To: <20230327201143.51026-1-snitzer@kernel.org> References: <20230327201143.51026-1-snitzer@kernel.org> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-block@vger.kernel.org From: Mikulas Patocka Save one spinlock by using waitqueue_active. We hold the bufio lock at this place, so no one can add entries to the waitqueue at this point. Signed-off-by: Mikulas Patocka Signed-off-by: Mike Snitzer --- drivers/md/dm-bufio.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) diff --git a/drivers/md/dm-bufio.c b/drivers/md/dm-bufio.c index e5459741335d..cca43ed13fd1 100644 --- a/drivers/md/dm-bufio.c +++ b/drivers/md/dm-bufio.c @@ -1665,7 +1665,12 @@ static void __free_buffer_wake(struct dm_buffer *b) c->need_reserved_buffers--; } - wake_up(&c->free_buffer_wait); + /* + * We hold the bufio lock here, so no one can add entries to the + * wait queue anyway. + */ + if (unlikely(waitqueue_active(&c->free_buffer_wait))) + wake_up(&c->free_buffer_wait); } static enum evict_result cleaned(struct dm_buffer *b, void *context) From patchwork Mon Mar 27 20:11:34 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Mike Snitzer X-Patchwork-Id: 13190055 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 vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id 60209C761A6 for ; Mon, 27 Mar 2023 20:14:17 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S232245AbjC0UOQ (ORCPT ); Mon, 27 Mar 2023 16:14:16 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:38834 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S232329AbjC0UON (ORCPT ); Mon, 27 Mar 2023 16:14:13 -0400 Received: from mail-qt1-f176.google.com (mail-qt1-f176.google.com [209.85.160.176]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id 230011BF9 for ; Mon, 27 Mar 2023 13:13:26 -0700 (PDT) Received: by mail-qt1-f176.google.com with SMTP id cj15so2316440qtb.5 for ; Mon, 27 Mar 2023 13:13:26 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; t=1679948005; h=content-transfer-encoding:mime-version:references:in-reply-to :message-id:date:subject:cc:to:from:x-gm-message-state:from:to:cc :subject:date:message-id:reply-to; bh=pVnf9GI1VNVce+UTfVz3CQAGt+fJmIpbfH8UFMjTV9Y=; b=X0OPFAQC5zI90eoYmVVxYxCLkuR5Cr7MbqXGe/AE0wFlY8tPPyvJFj3YH25OLCu2gn maDSvZ6JXRnjwx2Ls3ucBct8SShgilGcBc42A68GDQXhzYQxMnNLvGpcl1Hu4y6CUnk3 N+Lov+zHPgkmM1orjUWi0t8zAbZUahJdsu2u/UzIT94Qmf+tAu/2OyjHrxz4A6PSmGTi cEtKK+1g96dZ4UxWhDTRP9/tNJO/CNhjkxz1wzjQklC4rY3uMW6zNXfxqJZqAeszJRQx ky9FByi4O5tMKR/P/laYQ21wwSYjlD7bnk7Ik4Ju8QQHwrEViKhj42EEpTxwQvLCgeuT yq4A== X-Gm-Message-State: AO0yUKUU64HeJ1cCXxCdEIWdGF7eadSbeV8Yq+ZDU0p1BiaCHo13yY5R 8CYUhTpZE+/z+7vRECCpIbTL X-Google-Smtp-Source: AK7set/QhxRjmAYqq9vwOTVV6ZCx9K0pKe5GG2wkdvifb91iLlhC2h6StgNwdnfJ1O/ta0hPff4PHw== X-Received: by 2002:a05:622a:1a03:b0:3e3:8427:fb56 with SMTP id f3-20020a05622a1a0300b003e38427fb56mr22558795qtb.63.1679948004961; Mon, 27 Mar 2023 13:13:24 -0700 (PDT) Received: from localhost (pool-68-160-166-30.bstnma.fios.verizon.net. [68.160.166.30]) by smtp.gmail.com with ESMTPSA id 75-20020a370a4e000000b006ff8a122a1asm11508110qkk.78.2023.03.27.13.13.24 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Mon, 27 Mar 2023 13:13:24 -0700 (PDT) From: Mike Snitzer To: dm-devel@redhat.com Cc: linux-block@vger.kernel.org, axboe@kernel.dk, ejt@redhat.com, mpatocka@redhat.com, heinzm@redhat.com, nhuck@google.com, ebiggers@kernel.org, keescook@chromium.org, luomeng12@huawei.com, Mike Snitzer Subject: [dm-6.4 PATCH v3 11/20] dm bufio: use multi-page bio vector Date: Mon, 27 Mar 2023 16:11:34 -0400 Message-Id: <20230327201143.51026-12-snitzer@kernel.org> X-Mailer: git-send-email 2.40.0 In-Reply-To: <20230327201143.51026-1-snitzer@kernel.org> References: <20230327201143.51026-1-snitzer@kernel.org> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-block@vger.kernel.org From: Mikulas Patocka The kernel supports multi page bio vector entries, so we can use them in dm-bufio as an optimization. Signed-off-by: Mikulas Patocka Signed-off-by: Mike Snitzer --- drivers/md/dm-bufio.c | 24 ++++-------------------- 1 file changed, 4 insertions(+), 20 deletions(-) diff --git a/drivers/md/dm-bufio.c b/drivers/md/dm-bufio.c index cca43ed13fd1..ae552644a0b4 100644 --- a/drivers/md/dm-bufio.c +++ b/drivers/md/dm-bufio.c @@ -1312,19 +1312,14 @@ static void use_bio(struct dm_buffer *b, enum req_op op, sector_t sector, { struct bio *bio; char *ptr; - unsigned int vec_size, len; + unsigned int len; - vec_size = b->c->block_size >> PAGE_SHIFT; - if (unlikely(b->c->sectors_per_block_bits < PAGE_SHIFT - SECTOR_SHIFT)) - vec_size += 2; - - bio = bio_kmalloc(vec_size, GFP_NOWAIT | __GFP_NORETRY | __GFP_NOWARN); + bio = bio_kmalloc(1, GFP_NOWAIT | __GFP_NORETRY | __GFP_NOWARN); if (!bio) { -dmio: use_dmio(b, op, sector, n_sectors, offset); return; } - bio_init(bio, b->c->bdev, bio->bi_inline_vecs, vec_size, op); + bio_init(bio, b->c->bdev, bio->bi_inline_vecs, 1, op); bio->bi_iter.bi_sector = sector; bio->bi_end_io = bio_complete; bio->bi_private = b; @@ -1332,18 +1327,7 @@ static void use_bio(struct dm_buffer *b, enum req_op op, sector_t sector, ptr = (char *)b->data + offset; len = n_sectors << SECTOR_SHIFT; - do { - unsigned int this_step = min((unsigned int)(PAGE_SIZE - offset_in_page(ptr)), len); - - if (!bio_add_page(bio, virt_to_page(ptr), this_step, - offset_in_page(ptr))) { - bio_put(bio); - goto dmio; - } - - len -= this_step; - ptr += this_step; - } while (len > 0); + __bio_add_page(bio, virt_to_page(ptr), len, offset_in_page(ptr)); submit_bio(bio); } From patchwork Mon Mar 27 20:11:35 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Mike Snitzer X-Patchwork-Id: 13190057 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 vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id B5061C77B6C for ; Mon, 27 Mar 2023 20:14:18 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S232183AbjC0UOR (ORCPT ); Mon, 27 Mar 2023 16:14:17 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:39032 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S232244AbjC0UOP (ORCPT ); Mon, 27 Mar 2023 16:14:15 -0400 Received: from mail-qv1-f52.google.com (mail-qv1-f52.google.com [209.85.219.52]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id A926F1FCF for ; Mon, 27 Mar 2023 13:13:27 -0700 (PDT) Received: by mail-qv1-f52.google.com with SMTP id t13so7704629qvn.2 for ; Mon, 27 Mar 2023 13:13:27 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; t=1679948007; h=content-transfer-encoding:mime-version:references:in-reply-to :message-id:date:subject:cc:to:from:x-gm-message-state:from:to:cc :subject:date:message-id:reply-to; bh=7GATmDpp2LqxaO3YBsTr83VB1ZcPRtnkKD4xtYp0nm8=; b=PJ5YUgxpkooPAStG43yFlylYAELAzvmxYMxeHotDyrq6ubmyuIHXTsbE97Tbp0Hh1b mkZeEV08/40qdLpKRAxgSmp10/+tkwen98gm787U3qF5OQVHlMMVS423/t2YQBhLo45r x7TqYe09xtjc3P4+xM2gP9W6okGgwwWePqVwegKjzIgvrPcF0SOlATexLcvu8q9ssqx9 OFt9s6kK2TP7oErTfkOE1zSOsjtkMYK8YBh42+02oJSzy0pM+9W0qgbeuToMNFdsE20w 0yMJL8SJmhI0PMNMWpLRq5DCbVHGPO/SfXg6KmNRyO85RoAm299nWMlA7ji04U/H3RJx i0Tg== X-Gm-Message-State: AAQBX9dtTKLgLefd3SluPNUaqXpaNgG1q4G+0w3fbCSu9UOVGixDRXic gDA4RmoW2pxf3B6AaVDRZV4M X-Google-Smtp-Source: AKy350ZI9rUbEEtVYKNa1rqSBi0yYegoZt54hv+kbNekJScyZbdA+0A6vj49+yGatxgvJEDQcYubbw== X-Received: by 2002:ad4:5bc7:0:b0:5b8:d0b5:9a46 with SMTP id t7-20020ad45bc7000000b005b8d0b59a46mr20840019qvt.37.1679948006778; Mon, 27 Mar 2023 13:13:26 -0700 (PDT) Received: from localhost (pool-68-160-166-30.bstnma.fios.verizon.net. [68.160.166.30]) by smtp.gmail.com with ESMTPSA id jy3-20020a0562142b4300b005dd8b934595sm3228566qvb.45.2023.03.27.13.13.26 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Mon, 27 Mar 2023 13:13:26 -0700 (PDT) From: Mike Snitzer To: dm-devel@redhat.com Cc: linux-block@vger.kernel.org, axboe@kernel.dk, ejt@redhat.com, mpatocka@redhat.com, heinzm@redhat.com, nhuck@google.com, ebiggers@kernel.org, keescook@chromium.org, luomeng12@huawei.com, Mike Snitzer Subject: [dm-6.4 PATCH v3 12/20] dm thin: speed up cell_defer_no_holder() Date: Mon, 27 Mar 2023 16:11:35 -0400 Message-Id: <20230327201143.51026-13-snitzer@kernel.org> X-Mailer: git-send-email 2.40.0 In-Reply-To: <20230327201143.51026-1-snitzer@kernel.org> References: <20230327201143.51026-1-snitzer@kernel.org> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-block@vger.kernel.org From: Joe Thornber Reduce the time that a spinlock is held in cell_defer_no_holder(). Signed-off-by: Joe Thornber Signed-off-by: Mike Snitzer --- drivers/md/dm-thin.c | 14 ++++++++------ 1 file changed, 8 insertions(+), 6 deletions(-) diff --git a/drivers/md/dm-thin.c b/drivers/md/dm-thin.c index 13d4677baafd..00323428919e 100644 --- a/drivers/md/dm-thin.c +++ b/drivers/md/dm-thin.c @@ -883,15 +883,17 @@ static void cell_defer_no_holder(struct thin_c *tc, struct dm_bio_prison_cell *c { struct pool *pool = tc->pool; unsigned long flags; - int has_work; + struct bio_list bios; - spin_lock_irqsave(&tc->lock, flags); - cell_release_no_holder(pool, cell, &tc->deferred_bio_list); - has_work = !bio_list_empty(&tc->deferred_bio_list); - spin_unlock_irqrestore(&tc->lock, flags); + bio_list_init(&bios); + cell_release_no_holder(pool, cell, &bios); - if (has_work) + if (!bio_list_empty(&bios)) { + spin_lock_irqsave(&tc->lock, flags); + bio_list_merge(&tc->deferred_bio_list, &bios); + spin_unlock_irqrestore(&tc->lock, flags); wake_worker(pool); + } } static void thin_defer_bio(struct thin_c *tc, struct bio *bio); From patchwork Mon Mar 27 20:11:36 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Mike Snitzer X-Patchwork-Id: 13190058 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 vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id 21980C76195 for ; Mon, 27 Mar 2023 20:14:21 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S231929AbjC0UOT (ORCPT ); Mon, 27 Mar 2023 16:14:19 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:39046 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S232310AbjC0UOT (ORCPT ); Mon, 27 Mar 2023 16:14:19 -0400 Received: from mail-qt1-f172.google.com (mail-qt1-f172.google.com [209.85.160.172]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id C3FAE1FD0 for ; Mon, 27 Mar 2023 13:13:29 -0700 (PDT) Received: by mail-qt1-f172.google.com with SMTP id bz27so9824350qtb.1 for ; Mon, 27 Mar 2023 13:13:29 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; t=1679948009; h=content-transfer-encoding:mime-version:references:in-reply-to :message-id:date:subject:cc:to:from:x-gm-message-state:from:to:cc :subject:date:message-id:reply-to; bh=6PKqdJFHlMUFZNUEvpL9gBzLYs3G0zKQPB+NnAYo5GY=; b=LNhGbnQ872T6kfvd8On5BE3swsvg6RPnCAsE2Cxsk20+d8Fv/S8cKRHP0n7319K436 CDBGd6AmWvjb4cbXM1f6HtnTTG2Ize3oXs5AdZCeS2NnqbIKFhnhgu2ZoRqUjdHyaSVh SPXMPtI7lJtn32PFtyJuA84LXoAQ81C9+VtD5rnI0HotZPbm/vL90EYitFPNMdjUHhaX HesIw/sGsVrdWTG5vlI1vF+CHCEb1q2R++tccDUGKHBrYo26lOWVT3pF2SAI6C38RQHr RNndI93i4b+5IO643GTlgP9GmbZQtkcf4gfLjiwUCUgMwQvTIzR7knwKRHv4tSRSOkjU VwPA== X-Gm-Message-State: AO0yUKX63OkLZkX1EWy96u6LmkUvGB3k9IyhQ1Sk/CsMZUnvzlJQRIZO VU3ouB7wAAfQpRYpepRvAY9Y X-Google-Smtp-Source: AK7set8CW+cpPq4Pp9FJfsBjh5ZA7tWZLyqOfAoMlX+sA6HGXom7lYbcNpi3u7woOx27vLDKG9NLYA== X-Received: by 2002:a05:622a:1a98:b0:3d8:9b45:d362 with SMTP id s24-20020a05622a1a9800b003d89b45d362mr21296148qtc.28.1679948008860; Mon, 27 Mar 2023 13:13:28 -0700 (PDT) Received: from localhost (pool-68-160-166-30.bstnma.fios.verizon.net. [68.160.166.30]) by smtp.gmail.com with ESMTPSA id m64-20020a375843000000b0073b8512d2dbsm16899758qkb.72.2023.03.27.13.13.28 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Mon, 27 Mar 2023 13:13:28 -0700 (PDT) From: Mike Snitzer To: dm-devel@redhat.com Cc: linux-block@vger.kernel.org, axboe@kernel.dk, ejt@redhat.com, mpatocka@redhat.com, heinzm@redhat.com, nhuck@google.com, ebiggers@kernel.org, keescook@chromium.org, luomeng12@huawei.com, Mike Snitzer Subject: [dm-6.4 PATCH v3 13/20] dm: split discards further if target sets max_discard_granularity Date: Mon, 27 Mar 2023 16:11:36 -0400 Message-Id: <20230327201143.51026-14-snitzer@kernel.org> X-Mailer: git-send-email 2.40.0 In-Reply-To: <20230327201143.51026-1-snitzer@kernel.org> References: <20230327201143.51026-1-snitzer@kernel.org> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-block@vger.kernel.org The block core (bio_split_discard) will already split discards based on the 'discard_granularity' and 'max_discard_sectors' queue_limits. But the DM thin target also needs to ensure that it doesn't receive a discard that spans a 'max_discard_sectors' boundary. Introduce a dm_target 'max_discard_granularity' flag that if set will cause DM core to split discard bios relative to 'max_discard_sectors'. This treats 'discard_granularity' as a "min_discard_granularity" and 'max_discard_sectors' as a "max_discard_granularity". Requested-by: Joe Thornber Signed-off-by: Mike Snitzer --- drivers/md/dm.c | 25 +++++++++++++++++++------ include/linux/device-mapper.h | 6 ++++++ include/uapi/linux/dm-ioctl.h | 4 ++-- 3 files changed, 27 insertions(+), 8 deletions(-) diff --git a/drivers/md/dm.c b/drivers/md/dm.c index b6ace995b9ca..6eb0748a3bb2 100644 --- a/drivers/md/dm.c +++ b/drivers/md/dm.c @@ -1162,7 +1162,8 @@ static inline sector_t max_io_len_target_boundary(struct dm_target *ti, return ti->len - target_offset; } -static sector_t max_io_len(struct dm_target *ti, sector_t sector) +static sector_t __max_io_len(struct dm_target *ti, sector_t sector, + unsigned int max_granularity) { sector_t target_offset = dm_target_offset(ti, sector); sector_t len = max_io_len_target_boundary(ti, target_offset); @@ -1173,11 +1174,16 @@ static sector_t max_io_len(struct dm_target *ti, sector_t sector) * explains why stacked chunk_sectors based splitting via * bio_split_to_limits() isn't possible here. */ - if (!ti->max_io_len) + if (!max_granularity) return len; return min_t(sector_t, len, min(queue_max_sectors(ti->table->md->queue), - blk_chunk_sectors_left(target_offset, ti->max_io_len))); + blk_chunk_sectors_left(target_offset, max_granularity))); +} + +static inline sector_t max_io_len(struct dm_target *ti, sector_t sector) +{ + return __max_io_len(ti, sector, ti->max_io_len); } int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) @@ -1562,12 +1568,13 @@ static void __send_empty_flush(struct clone_info *ci) } static void __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti, - unsigned int num_bios) + unsigned int num_bios, + unsigned int max_granularity) { unsigned int len, bios; len = min_t(sector_t, ci->sector_count, - max_io_len_target_boundary(ti, dm_target_offset(ti, ci->sector))); + __max_io_len(ti, ci->sector, max_granularity)); atomic_add(num_bios, &ci->io->io_count); bios = __send_duplicate_bios(ci, ti, num_bios, &len); @@ -1603,10 +1610,16 @@ static blk_status_t __process_abnormal_io(struct clone_info *ci, struct dm_target *ti) { unsigned int num_bios = 0; + unsigned int max_granularity = 0; switch (bio_op(ci->bio)) { case REQ_OP_DISCARD: num_bios = ti->num_discard_bios; + if (ti->max_discard_granularity) { + struct queue_limits *limits = + dm_get_queue_limits(ti->table->md); + max_granularity = limits->max_discard_sectors; + } break; case REQ_OP_SECURE_ERASE: num_bios = ti->num_secure_erase_bios; @@ -1627,7 +1640,7 @@ static blk_status_t __process_abnormal_io(struct clone_info *ci, if (unlikely(!num_bios)) return BLK_STS_NOTSUPP; - __send_changing_extent_only(ci, ti, num_bios); + __send_changing_extent_only(ci, ti, num_bios, max_granularity); return BLK_STS_OK; } diff --git a/include/linux/device-mapper.h b/include/linux/device-mapper.h index 7975483816e4..8aa6b3ea91fa 100644 --- a/include/linux/device-mapper.h +++ b/include/linux/device-mapper.h @@ -358,6 +358,12 @@ struct dm_target { */ bool discards_supported:1; + /* + * Set if this target requires that discards be split on both + * 'discard_granularity' and 'max_discard_sectors' boundaries. + */ + bool max_discard_granularity:1; + /* * Set if we need to limit the number of in-flight bios when swapping. */ diff --git a/include/uapi/linux/dm-ioctl.h b/include/uapi/linux/dm-ioctl.h index 7edf335778ba..1990b5700f69 100644 --- a/include/uapi/linux/dm-ioctl.h +++ b/include/uapi/linux/dm-ioctl.h @@ -286,9 +286,9 @@ enum { #define DM_DEV_SET_GEOMETRY _IOWR(DM_IOCTL, DM_DEV_SET_GEOMETRY_CMD, struct dm_ioctl) #define DM_VERSION_MAJOR 4 -#define DM_VERSION_MINOR 47 +#define DM_VERSION_MINOR 48 #define DM_VERSION_PATCHLEVEL 0 -#define DM_VERSION_EXTRA "-ioctl (2022-07-28)" +#define DM_VERSION_EXTRA "-ioctl (2023-03-01)" /* Status bits */ #define DM_READONLY_FLAG (1 << 0) /* In/Out */ From patchwork Mon Mar 27 20:11:37 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Mike Snitzer X-Patchwork-Id: 13190061 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 vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id D28EAC77B61 for ; Mon, 27 Mar 2023 20:14:24 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S232349AbjC0UOX (ORCPT ); Mon, 27 Mar 2023 16:14:23 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:39058 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S232300AbjC0UOW (ORCPT ); Mon, 27 Mar 2023 16:14:22 -0400 Received: from mail-qt1-f172.google.com (mail-qt1-f172.google.com [209.85.160.172]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id 91D7F1FF7 for ; Mon, 27 Mar 2023 13:13:31 -0700 (PDT) Received: by mail-qt1-f172.google.com with SMTP id hf2so9811807qtb.3 for ; Mon, 27 Mar 2023 13:13:31 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; t=1679948010; h=content-transfer-encoding:mime-version:references:in-reply-to :message-id:date:subject:cc:to:from:x-gm-message-state:from:to:cc :subject:date:message-id:reply-to; bh=VKZdoNXFigBuaARZaAptCA6/0fq1wlNkMtf1bLoETbM=; b=p/QNZslkrX4o/4mjthzpEq1SoUKkuVf8stN+Zn7w2BU0DsFkfyR8Hv9B48KDbQaqRg Ip3Mxw6Xv/XTsAjl3ieQVTgPMHSMv+aKEkUFARaphBeQsCCFwasLSpDZOMWWiZZoVqIk ihImU6qGUa8FVLfNQ3AosfTKqgVwc1VctkAnrnqt5XRji/g7n9QEGVQeaoNRK3qqJWzW B9CPlslUsbK/duUqq/kZRJPbOQDg3xd/jYjl9eZKFsbyRFOiPlRJB+YS2LpzKjmCExMA XF9ufVSNZ2eKdx0jLVD2JLnbawwzKq7b2WI3jLlpFQGhjwfODmVUaXiyl38tF/zeEpjh taEw== X-Gm-Message-State: AO0yUKWEk405oKLfDVmEZ5eon6wMyGD4psMBtm2XDuGlJFaqv09QkNBc 4E6NcssaoRYjxyykGh6wPTa9 X-Google-Smtp-Source: AK7set9KXUMs1rqs7kAqxN9Yk6OnkJzXX9Ry0bTCMOPwZUEbPgMi3QtSYmMLZ60y5R3fllURrlRHWA== X-Received: by 2002:a05:622a:647:b0:3bf:d9ee:882d with SMTP id a7-20020a05622a064700b003bfd9ee882dmr22445529qtb.40.1679948010612; Mon, 27 Mar 2023 13:13:30 -0700 (PDT) Received: from localhost (pool-68-160-166-30.bstnma.fios.verizon.net. [68.160.166.30]) by smtp.gmail.com with ESMTPSA id j9-20020a05620a288900b0070648cf78bdsm16879656qkp.54.2023.03.27.13.13.29 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Mon, 27 Mar 2023 13:13:30 -0700 (PDT) From: Mike Snitzer To: dm-devel@redhat.com Cc: linux-block@vger.kernel.org, axboe@kernel.dk, ejt@redhat.com, mpatocka@redhat.com, heinzm@redhat.com, nhuck@google.com, ebiggers@kernel.org, keescook@chromium.org, luomeng12@huawei.com, Mike Snitzer Subject: [dm-6.4 PATCH v3 14/20] dm bio prison v1: improve concurrent IO performance Date: Mon, 27 Mar 2023 16:11:37 -0400 Message-Id: <20230327201143.51026-15-snitzer@kernel.org> X-Mailer: git-send-email 2.40.0 In-Reply-To: <20230327201143.51026-1-snitzer@kernel.org> References: <20230327201143.51026-1-snitzer@kernel.org> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-block@vger.kernel.org From: Joe Thornber Split the bio prison into multiple regions, with a separate rbtree and associated lock for each region. To get fast bio prison locking and not damage the performance of discards too much the bio-prison now stipulates that discards should not cross a BIO_PRISON_MAX_RANGE boundary. Because the range of a key (block_end - block_begin) must not exceed BIO_PRISON_MAX_RANGE: break_up_discard_bio() now ensures the data range reflected in PHYSICAL key doesn't exceed BIO_PRISON_MAX_RANGE. And splitting the thin target's discards (handled with VIRTUAL key) is achieved by updating dm-thin.c to set limits->max_discard_sectors in terms of BIO_PRISON_MAX_RANGE _and_ setting the thin and thin-pool targets' max_discard_granularity to true. Signed-off-by: Joe Thornber Signed-off-by: Mike Snitzer --- drivers/md/dm-bio-prison-v1.c | 87 +++++++++++++++++++++------------ drivers/md/dm-bio-prison-v1.h | 10 ++++ drivers/md/dm-thin.c | 92 ++++++++++++++++++++--------------- 3 files changed, 121 insertions(+), 68 deletions(-) diff --git a/drivers/md/dm-bio-prison-v1.c b/drivers/md/dm-bio-prison-v1.c index c4c05d5d8909..2b8af861e5f6 100644 --- a/drivers/md/dm-bio-prison-v1.c +++ b/drivers/md/dm-bio-prison-v1.c @@ -16,11 +16,17 @@ /*----------------------------------------------------------------*/ +#define NR_LOCKS 64 +#define LOCK_MASK (NR_LOCKS - 1) #define MIN_CELLS 1024 +struct prison_region { + spinlock_t lock; + struct rb_root cell; +} ____cacheline_aligned_in_smp; + struct dm_bio_prison { - spinlock_t lock; - struct rb_root cells; + struct prison_region regions[NR_LOCKS]; mempool_t cell_pool; }; @@ -34,13 +40,17 @@ static struct kmem_cache *_cell_cache; */ struct dm_bio_prison *dm_bio_prison_create(void) { - struct dm_bio_prison *prison = kzalloc(sizeof(*prison), GFP_KERNEL); int ret; + unsigned i; + struct dm_bio_prison *prison = kzalloc(sizeof(*prison), GFP_KERNEL); if (!prison) return NULL; - spin_lock_init(&prison->lock); + for (i = 0; i < NR_LOCKS; i++) { + spin_lock_init(&prison->regions[i].lock); + prison->regions[i].cell = RB_ROOT; + } ret = mempool_init_slab_pool(&prison->cell_pool, MIN_CELLS, _cell_cache); if (ret) { @@ -48,8 +58,6 @@ struct dm_bio_prison *dm_bio_prison_create(void) return NULL; } - prison->cells = RB_ROOT; - return prison; } EXPORT_SYMBOL_GPL(dm_bio_prison_create); @@ -107,14 +115,26 @@ static int cmp_keys(struct dm_cell_key *lhs, return 0; } -static int __bio_detain(struct dm_bio_prison *prison, +static unsigned lock_nr(struct dm_cell_key *key) +{ + return (key->block_begin >> BIO_PRISON_MAX_RANGE_SHIFT) & LOCK_MASK; +} + +static void check_range(struct dm_cell_key *key) +{ + BUG_ON(key->block_end - key->block_begin > BIO_PRISON_MAX_RANGE); + BUG_ON((key->block_begin >> BIO_PRISON_MAX_RANGE_SHIFT) != + ((key->block_end - 1) >> BIO_PRISON_MAX_RANGE_SHIFT)); +} + +static int __bio_detain(struct rb_root *root, struct dm_cell_key *key, struct bio *inmate, struct dm_bio_prison_cell *cell_prealloc, struct dm_bio_prison_cell **cell_result) { int r; - struct rb_node **new = &prison->cells.rb_node, *parent = NULL; + struct rb_node **new = &root->rb_node, *parent = NULL; while (*new) { struct dm_bio_prison_cell *cell = @@ -139,7 +159,7 @@ static int __bio_detain(struct dm_bio_prison *prison, *cell_result = cell_prealloc; rb_link_node(&cell_prealloc->node, parent, new); - rb_insert_color(&cell_prealloc->node, &prison->cells); + rb_insert_color(&cell_prealloc->node, root); return 0; } @@ -151,10 +171,12 @@ static int bio_detain(struct dm_bio_prison *prison, struct dm_bio_prison_cell **cell_result) { int r; + unsigned l = lock_nr(key); + check_range(key); - spin_lock_irq(&prison->lock); - r = __bio_detain(prison, key, inmate, cell_prealloc, cell_result); - spin_unlock_irq(&prison->lock); + spin_lock_irq(&prison->regions[l].lock); + r = __bio_detain(&prison->regions[l].cell, key, inmate, cell_prealloc, cell_result); + spin_unlock_irq(&prison->regions[l].lock); return r; } @@ -181,11 +203,11 @@ EXPORT_SYMBOL_GPL(dm_get_cell); /* * @inmates must have been initialised prior to this call */ -static void __cell_release(struct dm_bio_prison *prison, +static void __cell_release(struct rb_root *root, struct dm_bio_prison_cell *cell, struct bio_list *inmates) { - rb_erase(&cell->node, &prison->cells); + rb_erase(&cell->node, root); if (inmates) { if (cell->holder) @@ -198,20 +220,22 @@ void dm_cell_release(struct dm_bio_prison *prison, struct dm_bio_prison_cell *cell, struct bio_list *bios) { - spin_lock_irq(&prison->lock); - __cell_release(prison, cell, bios); - spin_unlock_irq(&prison->lock); + unsigned l = lock_nr(&cell->key); + + spin_lock_irq(&prison->regions[l].lock); + __cell_release(&prison->regions[l].cell, cell, bios); + spin_unlock_irq(&prison->regions[l].lock); } EXPORT_SYMBOL_GPL(dm_cell_release); /* * Sometimes we don't want the holder, just the additional bios. */ -static void __cell_release_no_holder(struct dm_bio_prison *prison, +static void __cell_release_no_holder(struct rb_root *root, struct dm_bio_prison_cell *cell, struct bio_list *inmates) { - rb_erase(&cell->node, &prison->cells); + rb_erase(&cell->node, root); bio_list_merge(inmates, &cell->bios); } @@ -219,11 +243,12 @@ void dm_cell_release_no_holder(struct dm_bio_prison *prison, struct dm_bio_prison_cell *cell, struct bio_list *inmates) { + unsigned l = lock_nr(&cell->key); unsigned long flags; - spin_lock_irqsave(&prison->lock, flags); - __cell_release_no_holder(prison, cell, inmates); - spin_unlock_irqrestore(&prison->lock, flags); + spin_lock_irqsave(&prison->regions[l].lock, flags); + __cell_release_no_holder(&prison->regions[l].cell, cell, inmates); + spin_unlock_irqrestore(&prison->regions[l].lock, flags); } EXPORT_SYMBOL_GPL(dm_cell_release_no_holder); @@ -248,18 +273,19 @@ void dm_cell_visit_release(struct dm_bio_prison *prison, void *context, struct dm_bio_prison_cell *cell) { - spin_lock_irq(&prison->lock); + unsigned l = lock_nr(&cell->key); + spin_lock_irq(&prison->regions[l].lock); visit_fn(context, cell); - rb_erase(&cell->node, &prison->cells); - spin_unlock_irq(&prison->lock); + rb_erase(&cell->node, &prison->regions[l].cell); + spin_unlock_irq(&prison->regions[l].lock); } EXPORT_SYMBOL_GPL(dm_cell_visit_release); -static int __promote_or_release(struct dm_bio_prison *prison, +static int __promote_or_release(struct rb_root *root, struct dm_bio_prison_cell *cell) { if (bio_list_empty(&cell->bios)) { - rb_erase(&cell->node, &prison->cells); + rb_erase(&cell->node, root); return 1; } @@ -271,10 +297,11 @@ int dm_cell_promote_or_release(struct dm_bio_prison *prison, struct dm_bio_prison_cell *cell) { int r; + unsigned l = lock_nr(&cell->key); - spin_lock_irq(&prison->lock); - r = __promote_or_release(prison, cell); - spin_unlock_irq(&prison->lock); + spin_lock_irq(&prison->regions[l].lock); + r = __promote_or_release(&prison->regions[l].cell, cell); + spin_unlock_irq(&prison->regions[l].lock); return r; } diff --git a/drivers/md/dm-bio-prison-v1.h b/drivers/md/dm-bio-prison-v1.h index dfbf1e94cb75..0b8acd6708fb 100644 --- a/drivers/md/dm-bio-prison-v1.h +++ b/drivers/md/dm-bio-prison-v1.h @@ -34,6 +34,16 @@ struct dm_cell_key { dm_block_t block_begin, block_end; }; +/* + * The range of a key (block_end - block_begin) must not + * exceed BIO_PRISON_MAX_RANGE. Also the range must not + * cross a similarly sized boundary. + * + * Must be a power of 2. + */ +#define BIO_PRISON_MAX_RANGE 1024 +#define BIO_PRISON_MAX_RANGE_SHIFT 10 + /* * Treat this as opaque, only in header so callers can manage allocation * themselves. diff --git a/drivers/md/dm-thin.c b/drivers/md/dm-thin.c index 00323428919e..33ad5695f959 100644 --- a/drivers/md/dm-thin.c +++ b/drivers/md/dm-thin.c @@ -1674,54 +1674,69 @@ static void break_up_discard_bio(struct thin_c *tc, dm_block_t begin, dm_block_t struct dm_cell_key data_key; struct dm_bio_prison_cell *data_cell; struct dm_thin_new_mapping *m; - dm_block_t virt_begin, virt_end, data_begin; + dm_block_t virt_begin, virt_end, data_begin, data_end; + dm_block_t len, next_boundary; while (begin != end) { - r = ensure_next_mapping(pool); - if (r) - /* we did our best */ - return; - r = dm_thin_find_mapped_range(tc->td, begin, end, &virt_begin, &virt_end, &data_begin, &maybe_shared); - if (r) + if (r) { /* * Silently fail, letting any mappings we've * created complete. */ break; - - build_key(tc->td, PHYSICAL, data_begin, data_begin + (virt_end - virt_begin), &data_key); - if (bio_detain(tc->pool, &data_key, NULL, &data_cell)) { - /* contention, we'll give up with this range */ - begin = virt_end; - continue; } - /* - * IO may still be going to the destination block. We must - * quiesce before we can do the removal. - */ - m = get_next_mapping(pool); - m->tc = tc; - m->maybe_shared = maybe_shared; - m->virt_begin = virt_begin; - m->virt_end = virt_end; - m->data_block = data_begin; - m->cell = data_cell; - m->bio = bio; + data_end = data_begin + (virt_end - virt_begin); /* - * The parent bio must not complete before sub discard bios are - * chained to it (see end_discard's bio_chain)! - * - * This per-mapping bi_remaining increment is paired with - * the implicit decrement that occurs via bio_endio() in - * end_discard(). + * Make sure the data region obeys the bio prison restrictions. */ - bio_inc_remaining(bio); - if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) - pool->process_prepared_discard(m); + while (data_begin < data_end) { + r = ensure_next_mapping(pool); + if (r) + return; /* we did our best */ + + next_boundary = ((data_begin >> BIO_PRISON_MAX_RANGE_SHIFT) + 1) + << BIO_PRISON_MAX_RANGE_SHIFT; + len = min_t(sector_t, data_end - data_begin, next_boundary - data_begin); + + build_key(tc->td, PHYSICAL, data_begin, data_begin + len, &data_key); + if (bio_detain(tc->pool, &data_key, NULL, &data_cell)) { + /* contention, we'll give up with this range */ + data_begin += len; + continue; + } + + /* + * IO may still be going to the destination block. We must + * quiesce before we can do the removal. + */ + m = get_next_mapping(pool); + m->tc = tc; + m->maybe_shared = maybe_shared; + m->virt_begin = virt_begin; + m->virt_end = virt_begin + len; + m->data_block = data_begin; + m->cell = data_cell; + m->bio = bio; + + /* + * The parent bio must not complete before sub discard bios are + * chained to it (see end_discard's bio_chain)! + * + * This per-mapping bi_remaining increment is paired with + * the implicit decrement that occurs via bio_endio() in + * end_discard(). + */ + bio_inc_remaining(bio); + if (!dm_deferred_set_add_work(pool->all_io_ds, &m->list)) + pool->process_prepared_discard(m); + + virt_begin += len; + data_begin += len; + } begin = virt_end; } @@ -3380,13 +3395,13 @@ static int pool_ctr(struct dm_target *ti, unsigned int argc, char **argv) */ if (pf.discard_enabled && pf.discard_passdown) { ti->num_discard_bios = 1; - /* * Setting 'discards_supported' circumvents the normal * stacking of discard limits (this keeps the pool and * thin devices' discard limits consistent). */ ti->discards_supported = true; + ti->max_discard_granularity = true; } ti->private = pt; @@ -4096,7 +4111,7 @@ static struct target_type pool_target = { .name = "thin-pool", .features = DM_TARGET_SINGLETON | DM_TARGET_ALWAYS_WRITEABLE | DM_TARGET_IMMUTABLE, - .version = {1, 22, 0}, + .version = {1, 23, 0}, .module = THIS_MODULE, .ctr = pool_ctr, .dtr = pool_dtr, @@ -4261,6 +4276,7 @@ static int thin_ctr(struct dm_target *ti, unsigned int argc, char **argv) if (tc->pool->pf.discard_enabled) { ti->discards_supported = true; ti->num_discard_bios = 1; + ti->max_discard_granularity = true; } mutex_unlock(&dm_thin_pool_table.mutex); @@ -4476,12 +4492,12 @@ static void thin_io_hints(struct dm_target *ti, struct queue_limits *limits) return; limits->discard_granularity = pool->sectors_per_block << SECTOR_SHIFT; - limits->max_discard_sectors = 2048 * 1024 * 16; /* 16G */ + limits->max_discard_sectors = pool->sectors_per_block * BIO_PRISON_MAX_RANGE; } static struct target_type thin_target = { .name = "thin", - .version = {1, 22, 0}, + .version = {1, 23, 0}, .module = THIS_MODULE, .ctr = thin_ctr, .dtr = thin_dtr, From patchwork Mon Mar 27 20:11:38 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Mike Snitzer X-Patchwork-Id: 13190059 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 vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id E8842C761A6 for ; Mon, 27 Mar 2023 20:14:22 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S232443AbjC0UOV (ORCPT ); Mon, 27 Mar 2023 16:14:21 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:39082 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S232200AbjC0UOU (ORCPT ); Mon, 27 Mar 2023 16:14:20 -0400 Received: from mail-qt1-f175.google.com (mail-qt1-f175.google.com [209.85.160.175]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id E1B021FDC for ; Mon, 27 Mar 2023 13:13:32 -0700 (PDT) Received: by mail-qt1-f175.google.com with SMTP id h16so3007841qtn.7 for ; Mon, 27 Mar 2023 13:13:32 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; t=1679948012; h=content-transfer-encoding:mime-version:references:in-reply-to :message-id:date:subject:cc:to:from:x-gm-message-state:from:to:cc :subject:date:message-id:reply-to; bh=SBN3e8d5Qq0hFXuPFkD52ybusL049Itzx+LBtkZ2VOM=; b=V+2bhA9WBWWqxWYYjWNHW0UGNRybuMKby2/pkKPBfjyOm8e9DpFdYbkYGpwGU3747j z8Nv9ZQlHq6pPjq3TvO8lzN6tUmQaYlMC47KTYmYAf1hRkwkfBTtRRWG+hxySN333NAA 4c2mHq6ptS88rxXxzCO912c5gDan573ELCOK3HK3qhCOo2ZFSMn0/4onqdXlDg78lwd0 8UV1f1x9BEw4AwM+G4/dDo5gS3mC+CBuJTmTuKn2wD82YvBNs233MRKPt+Dheh7+8ezP 6wHYgMewTvQeo8PH5HXtvXzi25LIEwnBxplEMGbLRKTlrEikbYQvSgtLA7BzpV7m0Qld pflA== X-Gm-Message-State: AO0yUKUUFbkG6Yu4TlFWJwu9XW7ZjO/jaWWrNIEKyuBuPmrv5Ny0cZuh m5ZT09w0hh2Hs76u1gY+Klzn X-Google-Smtp-Source: AK7set/KcR8zPzvyZmCoD66iZdu4uOgZL06cjihbV5XKhpEFExLT5A/6x5Rr5XaBiwEgxAgiHfTc4g== X-Received: by 2002:ac8:58c6:0:b0:3e3:9199:d27 with SMTP id u6-20020ac858c6000000b003e391990d27mr22314214qta.53.1679948012077; Mon, 27 Mar 2023 13:13:32 -0700 (PDT) Received: from localhost (pool-68-160-166-30.bstnma.fios.verizon.net. [68.160.166.30]) by smtp.gmail.com with ESMTPSA id b126-20020ae9eb84000000b007468b183a65sm11345416qkg.30.2023.03.27.13.13.31 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Mon, 27 Mar 2023 13:13:31 -0700 (PDT) From: Mike Snitzer To: dm-devel@redhat.com Cc: linux-block@vger.kernel.org, axboe@kernel.dk, ejt@redhat.com, mpatocka@redhat.com, heinzm@redhat.com, nhuck@google.com, ebiggers@kernel.org, keescook@chromium.org, luomeng12@huawei.com, Mike Snitzer Subject: [dm-6.4 PATCH v3 15/20] dm bio prison v1: add dm_cell_key_has_valid_range Date: Mon, 27 Mar 2023 16:11:38 -0400 Message-Id: <20230327201143.51026-16-snitzer@kernel.org> X-Mailer: git-send-email 2.40.0 In-Reply-To: <20230327201143.51026-1-snitzer@kernel.org> References: <20230327201143.51026-1-snitzer@kernel.org> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-block@vger.kernel.org Don't have bio_detain() BUG_ON if a dm_cell_key is beyond BIO_PRISON_MAX_RANGE or spans a boundary. Update dm-thin.c:build_key() to use dm_cell_key_has_valid_range() which will do this checking without using BUG_ON. Also update process_discard_bio() to check the discard bio that DM core passes in (having first imposed max_discard_granularity based splitting). dm_cell_key_has_valid_range() will merely WARN_ON_ONCE if it returns false because if it does: it is programmer error that should be caught with proper testing. So relax the BUG_ONs to be WARN_ON_ONCE. Signed-off-by: Mike Snitzer --- drivers/md/dm-bio-prison-v1.c | 14 +++++++++----- drivers/md/dm-bio-prison-v1.h | 5 +++++ drivers/md/dm-thin.c | 21 +++++++++++++++------ 3 files changed, 29 insertions(+), 11 deletions(-) diff --git a/drivers/md/dm-bio-prison-v1.c b/drivers/md/dm-bio-prison-v1.c index 2b8af861e5f6..78bb559b521c 100644 --- a/drivers/md/dm-bio-prison-v1.c +++ b/drivers/md/dm-bio-prison-v1.c @@ -120,12 +120,17 @@ static unsigned lock_nr(struct dm_cell_key *key) return (key->block_begin >> BIO_PRISON_MAX_RANGE_SHIFT) & LOCK_MASK; } -static void check_range(struct dm_cell_key *key) +bool dm_cell_key_has_valid_range(struct dm_cell_key *key) { - BUG_ON(key->block_end - key->block_begin > BIO_PRISON_MAX_RANGE); - BUG_ON((key->block_begin >> BIO_PRISON_MAX_RANGE_SHIFT) != - ((key->block_end - 1) >> BIO_PRISON_MAX_RANGE_SHIFT)); + if (WARN_ON_ONCE(key->block_end - key->block_begin > BIO_PRISON_MAX_RANGE)) + return false; + if (WARN_ON_ONCE((key->block_begin >> BIO_PRISON_MAX_RANGE_SHIFT) != + (key->block_end - 1) >> BIO_PRISON_MAX_RANGE_SHIFT)) + return false; + + return true; } +EXPORT_SYMBOL(dm_cell_key_has_valid_range); static int __bio_detain(struct rb_root *root, struct dm_cell_key *key, @@ -172,7 +177,6 @@ static int bio_detain(struct dm_bio_prison *prison, { int r; unsigned l = lock_nr(key); - check_range(key); spin_lock_irq(&prison->regions[l].lock); r = __bio_detain(&prison->regions[l].cell, key, inmate, cell_prealloc, cell_result); diff --git a/drivers/md/dm-bio-prison-v1.h b/drivers/md/dm-bio-prison-v1.h index 0b8acd6708fb..2a097ed0d85e 100644 --- a/drivers/md/dm-bio-prison-v1.h +++ b/drivers/md/dm-bio-prison-v1.h @@ -83,6 +83,11 @@ int dm_get_cell(struct dm_bio_prison *prison, struct dm_bio_prison_cell *cell_prealloc, struct dm_bio_prison_cell **cell_result); +/* + * Returns false if key is beyond BIO_PRISON_MAX_RANGE or spans a boundary. + */ +bool dm_cell_key_has_valid_range(struct dm_cell_key *key); + /* * An atomic op that combines retrieving or creating a cell, and adding a * bio to it. diff --git a/drivers/md/dm-thin.c b/drivers/md/dm-thin.c index 33ad5695f959..2b13c949bd72 100644 --- a/drivers/md/dm-thin.c +++ b/drivers/md/dm-thin.c @@ -118,25 +118,27 @@ enum lock_space { PHYSICAL }; -static void build_key(struct dm_thin_device *td, enum lock_space ls, +static bool build_key(struct dm_thin_device *td, enum lock_space ls, dm_block_t b, dm_block_t e, struct dm_cell_key *key) { key->virtual = (ls == VIRTUAL); key->dev = dm_thin_dev_id(td); key->block_begin = b; key->block_end = e; + + return dm_cell_key_has_valid_range(key); } static void build_data_key(struct dm_thin_device *td, dm_block_t b, struct dm_cell_key *key) { - build_key(td, PHYSICAL, b, b + 1llu, key); + (void) build_key(td, PHYSICAL, b, b + 1llu, key); } static void build_virtual_key(struct dm_thin_device *td, dm_block_t b, struct dm_cell_key *key) { - build_key(td, VIRTUAL, b, b + 1llu, key); + (void) build_key(td, VIRTUAL, b, b + 1llu, key); } /*----------------------------------------------------------------*/ @@ -1702,7 +1704,8 @@ static void break_up_discard_bio(struct thin_c *tc, dm_block_t begin, dm_block_t << BIO_PRISON_MAX_RANGE_SHIFT; len = min_t(sector_t, data_end - data_begin, next_boundary - data_begin); - build_key(tc->td, PHYSICAL, data_begin, data_begin + len, &data_key); + /* This key is certainly within range given the above splitting */ + (void) build_key(tc->td, PHYSICAL, data_begin, data_begin + len, &data_key); if (bio_detain(tc->pool, &data_key, NULL, &data_cell)) { /* contention, we'll give up with this range */ data_begin += len; @@ -1778,8 +1781,13 @@ static void process_discard_bio(struct thin_c *tc, struct bio *bio) return; } - build_key(tc->td, VIRTUAL, begin, end, &virt_key); - if (bio_detain(tc->pool, &virt_key, bio, &virt_cell)) + if (unlikely(!build_key(tc->td, VIRTUAL, begin, end, &virt_key))) { + DMERR_LIMIT("Discard doesn't respect bio prison limits"); + bio_endio(bio); + return; + } + + if (bio_detain(tc->pool, &virt_key, bio, &virt_cell)) { /* * Potential starvation issue: We're relying on the * fs/application being well behaved, and not trying to @@ -1788,6 +1796,7 @@ static void process_discard_bio(struct thin_c *tc, struct bio *bio) * cell will never be granted. */ return; + } tc->pool->process_discard_cell(tc, virt_cell); } From patchwork Mon Mar 27 20:11:39 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Mike Snitzer X-Patchwork-Id: 13190060 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 vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id E723AC6FD1D for ; Mon, 27 Mar 2023 20:14:23 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S232200AbjC0UOW (ORCPT ); Mon, 27 Mar 2023 16:14:22 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:39116 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S232349AbjC0UOV (ORCPT ); Mon, 27 Mar 2023 16:14:21 -0400 Received: from mail-qt1-f178.google.com (mail-qt1-f178.google.com [209.85.160.178]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id 857213583 for ; Mon, 27 Mar 2023 13:13:34 -0700 (PDT) Received: by mail-qt1-f178.google.com with SMTP id cr18so5982226qtb.0 for ; Mon, 27 Mar 2023 13:13:34 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; t=1679948013; h=content-transfer-encoding:mime-version:references:in-reply-to :message-id:date:subject:cc:to:from:x-gm-message-state:from:to:cc :subject:date:message-id:reply-to; bh=P0nH8y5nrVQVdjQ+7MWEvssVfY/cEbcGbEWp6ylrqyQ=; b=ScdDx1DqN86Dz3YI124dbxilOBRwykvRhF62zEdEnm92B2AHG4TIeAV5UXI8ggN8jQ Y7+XDVq3idZljunFoh7WLJks+8dx6jy1QvNs+Evu06aik05yGPHc2ixN/5A9Mf1Jqdpp HBYYpswubxNAcCTFckCokxa+n2YvxdH1gmiV5ruP8/8blMEzJrQ70pJpZ6/QSDU8fFng qli1+oOOGrPTfB0BZsmrJXZWy0+REFvk3trDAmXmZr7sHbFi2/Ij80+ZoXIBJIGTXcc+ aoxHdyYtxo9sI3lN26R0n+1cGiByd4GN8kc36U6g2JcXUsOhfwkrEzlrnDjvmlEVfNOy 8Z5w== X-Gm-Message-State: AAQBX9eypFW929qszY+uxo/7XIj/tkYv43w/a/dxjRxTzdHOhVTzXG6l BHNbjjObJyg6VJhiF7wMTljp X-Google-Smtp-Source: AKy350Y96gqJ8gm1eUDkWBdhHVPSIF25C9Z/7pTPicBHvic+yBssV8gx7JTYQ5NQ4P/SYw+5vk3tRA== X-Received: by 2002:ac8:5c46:0:b0:3e4:e8a8:f235 with SMTP id j6-20020ac85c46000000b003e4e8a8f235mr8388241qtj.36.1679948013682; Mon, 27 Mar 2023 13:13:33 -0700 (PDT) Received: from localhost (pool-68-160-166-30.bstnma.fios.verizon.net. [68.160.166.30]) by smtp.gmail.com with ESMTPSA id 6-20020ac85646000000b003e3897d8505sm6400099qtt.54.2023.03.27.13.13.33 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Mon, 27 Mar 2023 13:13:33 -0700 (PDT) From: Mike Snitzer To: dm-devel@redhat.com Cc: linux-block@vger.kernel.org, axboe@kernel.dk, ejt@redhat.com, mpatocka@redhat.com, heinzm@redhat.com, nhuck@google.com, ebiggers@kernel.org, keescook@chromium.org, luomeng12@huawei.com, Mike Snitzer Subject: [dm-6.4 PATCH v3 16/20] dm: add dm_num_sharded_locks() Date: Mon, 27 Mar 2023 16:11:39 -0400 Message-Id: <20230327201143.51026-17-snitzer@kernel.org> X-Mailer: git-send-email 2.40.0 In-Reply-To: <20230327201143.51026-1-snitzer@kernel.org> References: <20230327201143.51026-1-snitzer@kernel.org> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-block@vger.kernel.org Simple helper to use when DM core code needs to appropriately size, based on num_online_cpus(), its data structures that split locks. dm_num_sharded_locks() rounds up num_online_cpus() to next power of 2 but caps return at DM_MAX_SHARDED_LOCKS (64). This heuristic may evolve as warranted, but as-is it will serve as a more informed basis for sizing the sharded lock structs in dm-bufio's dm_buffer_cache (buffer_trees) and dm-bio-prison-v1's dm_bio_prison (prison_regions). Signed-off-by: Mike Snitzer --- drivers/md/dm.h | 10 ++++++++++ 1 file changed, 10 insertions(+) diff --git a/drivers/md/dm.h b/drivers/md/dm.h index 22eaed188907..18450282d0d9 100644 --- a/drivers/md/dm.h +++ b/drivers/md/dm.h @@ -20,6 +20,7 @@ #include #include #include +#include #include "dm-stats.h" @@ -228,4 +229,13 @@ void dm_free_md_mempools(struct dm_md_mempools *pools); */ unsigned int dm_get_reserved_bio_based_ios(void); +#define DM_MAX_SHARDED_LOCKS 64 + +static inline unsigned int dm_num_sharded_locks(void) +{ + unsigned int num_locks = roundup_pow_of_two(num_online_cpus()); + + return min_t(unsigned int, num_locks, DM_MAX_SHARDED_LOCKS); +} + #endif From patchwork Mon Mar 27 20:11:40 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Mike Snitzer X-Patchwork-Id: 13190064 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 vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id E79F9C6FD1D for ; Mon, 27 Mar 2023 20:14:28 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S232394AbjC0UO1 (ORCPT ); Mon, 27 Mar 2023 16:14:27 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:39196 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S232417AbjC0UO0 (ORCPT ); Mon, 27 Mar 2023 16:14:26 -0400 Received: from mail-qv1-f49.google.com (mail-qv1-f49.google.com [209.85.219.49]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id 31BEA3C00 for ; Mon, 27 Mar 2023 13:13:36 -0700 (PDT) Received: by mail-qv1-f49.google.com with SMTP id 59so7589111qva.11 for ; Mon, 27 Mar 2023 13:13:36 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; t=1679948015; h=content-transfer-encoding:mime-version:references:in-reply-to :message-id:date:subject:cc:to:from:x-gm-message-state:from:to:cc :subject:date:message-id:reply-to; bh=Xis1PZIHxgZc33JUTcnyVvy12Vhl77i8OslXpZXLdgA=; b=ieP7hbFbub5hiLTsWAfr55rcpqyCGnVQTT9Ok0Q2V/CUTuj+7sJYZyJi6NP/bJuDEy ZaEREWOFMyUBIjOAtxV+cdEwVXpLDbrj2I6l5HRvCNwZ7wSIDDmyFNObsS2Qi1CPIXCi z9Ka9vwpSpw6wPVfBHUDfdT3seybodNWqd8Gh2MECg0pruTMQj9NZN3/BU9BFum/cYs+ SbvHRbMN5u8rIZMyOEdl3GorqwRKq7vUclcfpL+Kcm2ZyxCdhxwjzsEMBWXsHRl3bbT0 ThE28A83DF6BJNRZ0kh3KwKWyBuMSsepsUt/05NwK6QruFl7aHhf3Rbbw/3naSABep5N z0QQ== X-Gm-Message-State: AAQBX9d1XWYo34TqPv4vMWWQOrcCxl9mcfrofTu94/rgouNMs6yWqwq3 rMTRSB/nQITaoQsfuKf0UyVT X-Google-Smtp-Source: AKy350bdI7fPLiZRLW4GPQlt6CDNO3b2uDbMBtu6JVcWz3roMzyEeTD8mnUG2okibZRJCTBwmlMNmw== X-Received: by 2002:a05:6214:27c6:b0:56b:fb18:adcd with SMTP id ge6-20020a05621427c600b0056bfb18adcdmr19738100qvb.8.1679948015146; Mon, 27 Mar 2023 13:13:35 -0700 (PDT) Received: from localhost (pool-68-160-166-30.bstnma.fios.verizon.net. [68.160.166.30]) by smtp.gmail.com with ESMTPSA id cw2-20020ad44dc2000000b005dd8b9345aesm3214903qvb.70.2023.03.27.13.13.34 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Mon, 27 Mar 2023 13:13:34 -0700 (PDT) From: Mike Snitzer To: dm-devel@redhat.com Cc: linux-block@vger.kernel.org, axboe@kernel.dk, ejt@redhat.com, mpatocka@redhat.com, heinzm@redhat.com, nhuck@google.com, ebiggers@kernel.org, keescook@chromium.org, luomeng12@huawei.com, Mike Snitzer Subject: [dm-6.4 PATCH v3 17/20] dm bufio: prepare to intelligently size dm_buffer_cache's buffer_trees Date: Mon, 27 Mar 2023 16:11:40 -0400 Message-Id: <20230327201143.51026-18-snitzer@kernel.org> X-Mailer: git-send-email 2.40.0 In-Reply-To: <20230327201143.51026-1-snitzer@kernel.org> References: <20230327201143.51026-1-snitzer@kernel.org> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-block@vger.kernel.org Add num_locks member to dm_buffer_cache struct and use it rather than the NR_LOCKS magic value (64). Next commit will size the dm_buffer_cache's buffer_trees according to dm_num_sharded_locks(). Signed-off-by: Mike Snitzer --- drivers/md/dm-bufio.c | 48 +++++++++++++++++++++++-------------------- 1 file changed, 26 insertions(+), 22 deletions(-) diff --git a/drivers/md/dm-bufio.c b/drivers/md/dm-bufio.c index ae552644a0b4..2250799a70e4 100644 --- a/drivers/md/dm-bufio.c +++ b/drivers/md/dm-bufio.c @@ -380,7 +380,6 @@ struct dm_buffer { */ #define NR_LOCKS 64 -#define LOCKS_MASK (NR_LOCKS - 1) struct buffer_tree { struct rw_semaphore lock; @@ -388,37 +387,38 @@ struct buffer_tree { } ____cacheline_aligned_in_smp; struct dm_buffer_cache { + struct lru lru[LIST_SIZE]; /* * We spread entries across multiple trees to reduce contention * on the locks. */ + unsigned int num_locks; struct buffer_tree trees[NR_LOCKS]; - struct lru lru[LIST_SIZE]; }; -static inline unsigned int cache_index(sector_t block) +static inline unsigned int cache_index(sector_t block, unsigned int num_locks) { - return block & LOCKS_MASK; + return block & (num_locks - 1); } static inline void cache_read_lock(struct dm_buffer_cache *bc, sector_t block) { - down_read(&bc->trees[cache_index(block)].lock); + down_read(&bc->trees[cache_index(block, bc->num_locks)].lock); } static inline void cache_read_unlock(struct dm_buffer_cache *bc, sector_t block) { - up_read(&bc->trees[cache_index(block)].lock); + up_read(&bc->trees[cache_index(block, bc->num_locks)].lock); } static inline void cache_write_lock(struct dm_buffer_cache *bc, sector_t block) { - down_write(&bc->trees[cache_index(block)].lock); + down_write(&bc->trees[cache_index(block, bc->num_locks)].lock); } static inline void cache_write_unlock(struct dm_buffer_cache *bc, sector_t block) { - up_write(&bc->trees[cache_index(block)].lock); + up_write(&bc->trees[cache_index(block, bc->num_locks)].lock); } /* @@ -429,13 +429,15 @@ struct lock_history { struct dm_buffer_cache *cache; bool write; unsigned int previous; + unsigned int no_previous; }; static void lh_init(struct lock_history *lh, struct dm_buffer_cache *cache, bool write) { lh->cache = cache; lh->write = write; - lh->previous = NR_LOCKS; /* indicates no previous */ + lh->no_previous = cache->num_locks; + lh->previous = lh->no_previous; } static void __lh_lock(struct lock_history *lh, unsigned int index) @@ -459,9 +461,9 @@ static void __lh_unlock(struct lock_history *lh, unsigned int index) */ static void lh_exit(struct lock_history *lh) { - if (lh->previous != NR_LOCKS) { + if (lh->previous != lh->no_previous) { __lh_unlock(lh, lh->previous); - lh->previous = NR_LOCKS; + lh->previous = lh->no_previous; } } @@ -471,9 +473,9 @@ static void lh_exit(struct lock_history *lh) */ static void lh_next(struct lock_history *lh, sector_t b) { - unsigned int index = cache_index(b); + unsigned int index = cache_index(b, lh->no_previous); /* no_previous is num_locks */ - if (lh->previous != NR_LOCKS) { + if (lh->previous != lh->no_previous) { if (lh->previous != index) { __lh_unlock(lh, lh->previous); __lh_lock(lh, index); @@ -500,11 +502,13 @@ static struct dm_buffer *list_to_buffer(struct list_head *l) return le_to_buffer(le); } -static void cache_init(struct dm_buffer_cache *bc) +static void cache_init(struct dm_buffer_cache *bc, unsigned int num_locks) { unsigned int i; - for (i = 0; i < NR_LOCKS; i++) { + bc->num_locks = num_locks; + + for (i = 0; i < bc->num_locks; i++) { init_rwsem(&bc->trees[i].lock); bc->trees[i].root = RB_ROOT; } @@ -517,7 +521,7 @@ static void cache_destroy(struct dm_buffer_cache *bc) { unsigned int i; - for (i = 0; i < NR_LOCKS; i++) + for (i = 0; i < bc->num_locks; i++) WARN_ON_ONCE(!RB_EMPTY_ROOT(&bc->trees[i].root)); lru_destroy(&bc->lru[LIST_CLEAN]); @@ -576,7 +580,7 @@ static struct dm_buffer *cache_get(struct dm_buffer_cache *bc, sector_t block) struct dm_buffer *b; cache_read_lock(bc, block); - b = __cache_get(&bc->trees[cache_index(block)].root, block); + b = __cache_get(&bc->trees[cache_index(block, bc->num_locks)].root, block); if (b) { lru_reference(&b->lru); __cache_inc_buffer(b); @@ -650,7 +654,7 @@ static struct dm_buffer *__cache_evict(struct dm_buffer_cache *bc, int list_mode b = le_to_buffer(le); /* __evict_pred will have locked the appropriate tree. */ - rb_erase(&b->node, &bc->trees[cache_index(b->block)].root); + rb_erase(&b->node, &bc->trees[cache_index(b->block, bc->num_locks)].root); return b; } @@ -816,7 +820,7 @@ static bool cache_insert(struct dm_buffer_cache *bc, struct dm_buffer *b) cache_write_lock(bc, b->block); BUG_ON(atomic_read(&b->hold_count) != 1); - r = __cache_insert(&bc->trees[cache_index(b->block)].root, b); + r = __cache_insert(&bc->trees[cache_index(b->block, bc->num_locks)].root, b); if (r) lru_insert(&bc->lru[b->list_mode], &b->lru); cache_write_unlock(bc, b->block); @@ -842,7 +846,7 @@ static bool cache_remove(struct dm_buffer_cache *bc, struct dm_buffer *b) r = false; } else { r = true; - rb_erase(&b->node, &bc->trees[cache_index(b->block)].root); + rb_erase(&b->node, &bc->trees[cache_index(b->block, bc->num_locks)].root); lru_remove(&bc->lru[b->list_mode], &b->lru); } @@ -911,7 +915,7 @@ static void cache_remove_range(struct dm_buffer_cache *bc, { unsigned int i; - for (i = 0; i < NR_LOCKS; i++) { + for (i = 0; i < bc->num_locks; i++) { down_write(&bc->trees[i].lock); __remove_range(bc, &bc->trees[i].root, begin, end, pred, release); up_write(&bc->trees[i].lock); @@ -2432,7 +2436,7 @@ struct dm_bufio_client *dm_bufio_client_create(struct block_device *bdev, unsign r = -ENOMEM; goto bad_client; } - cache_init(&c->cache); + cache_init(&c->cache, NR_LOCKS); c->bdev = bdev; c->block_size = block_size; From patchwork Mon Mar 27 20:11:41 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Mike Snitzer X-Patchwork-Id: 13190062 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 vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id 0E11DC76195 for ; Mon, 27 Mar 2023 20:14:27 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S232329AbjC0UOZ (ORCPT ); Mon, 27 Mar 2023 16:14:25 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:39210 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S232300AbjC0UOZ (ORCPT ); Mon, 27 Mar 2023 16:14:25 -0400 Received: from mail-qt1-f179.google.com (mail-qt1-f179.google.com [209.85.160.179]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id B77562100 for ; Mon, 27 Mar 2023 13:13:37 -0700 (PDT) Received: by mail-qt1-f179.google.com with SMTP id cn12so6457696qtb.8 for ; Mon, 27 Mar 2023 13:13:37 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; t=1679948017; h=content-transfer-encoding:mime-version:references:in-reply-to :message-id:date:subject:cc:to:from:x-gm-message-state:from:to:cc :subject:date:message-id:reply-to; bh=PwFipSc82OIEFDU8Id4luDtLttcFOJWrHTnvxSKayKA=; b=mLYedgBLA5EDUqW7ZZnumufDDDQ2VubdDrhvPw5q6Bn2tQIbOgdzq3TkAaWUT6Fl3o ykM7PgetnBwq69g9AeWkSgX/OTJhk402C7d/nUYOL0q3CFjIFYwELk9nYALHbNkXJwU0 dLSjrBpzn7lUaonMHTUwLlYCD8I+cwmf14URDVJsHkYGYgPG0khEHZyy1daGlgoLAbYU eLqrwiZ7Dzp5Z5Y33AbT9A4tt5z+jhuTvgx9Sa/9oaeYr2EsWrtXnH6LsSiGbJ3E4PyJ 4dq5gn54iAzYpSfPOcv0L3iwRHLjq1w0dQ63tgxABttY858AeE/f9fbXSaj7LVuBJI9p lmtg== X-Gm-Message-State: AO0yUKUjB17E4jmHYZe5ZdMg4ajbpWgtaLaN4ZTQCF0tf/w3WS/0AIer SWO/MHDfxMkaDFRMseXqvjYw X-Google-Smtp-Source: AK7set/FmDxDs4w1j0e5PA+jX3MwJRj6+AOGAADIBIksQBRZljk9BW4Q1CQMIOVrgNxvcG4up41zZQ== X-Received: by 2002:a05:622a:c6:b0:3b6:2c3b:8c00 with SMTP id p6-20020a05622a00c600b003b62c3b8c00mr23270455qtw.66.1679948016921; Mon, 27 Mar 2023 13:13:36 -0700 (PDT) Received: from localhost (pool-68-160-166-30.bstnma.fios.verizon.net. [68.160.166.30]) by smtp.gmail.com with ESMTPSA id f8-20020a05620a280800b0074269db4699sm10226828qkp.46.2023.03.27.13.13.36 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Mon, 27 Mar 2023 13:13:36 -0700 (PDT) From: Mike Snitzer To: dm-devel@redhat.com Cc: linux-block@vger.kernel.org, axboe@kernel.dk, ejt@redhat.com, mpatocka@redhat.com, heinzm@redhat.com, nhuck@google.com, ebiggers@kernel.org, keescook@chromium.org, luomeng12@huawei.com, Mike Snitzer Subject: [dm-6.4 PATCH v3 18/20] dm bufio: intelligently size dm_buffer_cache's buffer_trees Date: Mon, 27 Mar 2023 16:11:41 -0400 Message-Id: <20230327201143.51026-19-snitzer@kernel.org> X-Mailer: git-send-email 2.40.0 In-Reply-To: <20230327201143.51026-1-snitzer@kernel.org> References: <20230327201143.51026-1-snitzer@kernel.org> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-block@vger.kernel.org Size the dm_buffer_cache's number of buffer_tree structs using dm_num_sharded_locks(). Signed-off-by: Mike Snitzer --- drivers/md/dm-bufio.c | 14 ++++++++------ 1 file changed, 8 insertions(+), 6 deletions(-) diff --git a/drivers/md/dm-bufio.c b/drivers/md/dm-bufio.c index 2250799a70e4..7dc53f3d0739 100644 --- a/drivers/md/dm-bufio.c +++ b/drivers/md/dm-bufio.c @@ -21,6 +21,8 @@ #include #include +#include "dm.h" + #define DM_MSG_PREFIX "bufio" /* @@ -379,8 +381,6 @@ struct dm_buffer { * only enough to ensure get/put are threadsafe. */ -#define NR_LOCKS 64 - struct buffer_tree { struct rw_semaphore lock; struct rb_root root; @@ -393,7 +393,7 @@ struct dm_buffer_cache { * on the locks. */ unsigned int num_locks; - struct buffer_tree trees[NR_LOCKS]; + struct buffer_tree trees[]; }; static inline unsigned int cache_index(sector_t block, unsigned int num_locks) @@ -976,7 +976,7 @@ struct dm_bufio_client { */ unsigned long oldest_buffer; - struct dm_buffer_cache cache; + struct dm_buffer_cache cache; /* must be last member */ }; static DEFINE_STATIC_KEY_FALSE(no_sleep_enabled); @@ -2422,6 +2422,7 @@ struct dm_bufio_client *dm_bufio_client_create(struct block_device *bdev, unsign unsigned int flags) { int r; + unsigned int num_locks; struct dm_bufio_client *c; char slab_name[27]; @@ -2431,12 +2432,13 @@ struct dm_bufio_client *dm_bufio_client_create(struct block_device *bdev, unsign goto bad_client; } - c = kzalloc(sizeof(*c), GFP_KERNEL); + num_locks = dm_num_sharded_locks(); + c = kzalloc(sizeof(*c) + (num_locks * sizeof(struct buffer_tree)), GFP_KERNEL); if (!c) { r = -ENOMEM; goto bad_client; } - cache_init(&c->cache, NR_LOCKS); + cache_init(&c->cache, num_locks); c->bdev = bdev; c->block_size = block_size; From patchwork Mon Mar 27 20:11:42 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Mike Snitzer X-Patchwork-Id: 13190063 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 vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id 3D7FEC76195 for ; Mon, 27 Mar 2023 20:14:29 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S232479AbjC0UO2 (ORCPT ); Mon, 27 Mar 2023 16:14:28 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:39200 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S232244AbjC0UO0 (ORCPT ); Mon, 27 Mar 2023 16:14:26 -0400 Received: from mail-qt1-f175.google.com (mail-qt1-f175.google.com [209.85.160.175]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id B47E32D44 for ; Mon, 27 Mar 2023 13:13:38 -0700 (PDT) Received: by mail-qt1-f175.google.com with SMTP id h16so3008089qtn.7 for ; Mon, 27 Mar 2023 13:13:38 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; t=1679948018; h=content-transfer-encoding:mime-version:references:in-reply-to :message-id:date:subject:cc:to:from:x-gm-message-state:from:to:cc :subject:date:message-id:reply-to; bh=RS/kygwvcLIIpB3MPZf9iFXCT5/ZQtXgTxSpDCvveGI=; b=Eby2HfXlEZwdm1i4OAzfhOSjeJt7Y0jx0gYckqKKZmG5/B3xVphRqxPGzhlUmrj4oG gZZLEm6gsK17W2sVUavdzXvebu6J2kauNB6x122vg+PwCljKNnxVayEiDQAL5AFF5HOn 7sFDm8Rs8gYdchOuxTLTQELIJIODN6BNqYLVm9fzxWxaHxKjqwM06/BCdGXCFXe4rlyp Ax+c21clg9rSrxK543Cypf5t2+BFF/601AgiT42NjQXAf21ixlaLX2ftb+ytefTLhMUz KWxi6/2eT/OFo299gj7VrKhe0SVjJnD53Fo4qhf/eEmmAlLUsqFd6UwuO6acE2VX9oHa tBAg== X-Gm-Message-State: AO0yUKUYpIrrHu+1Mwv/WlSQVdwZQmFv6tgCuH4VF89UY7BhIIm7oAfa 88VjrsZkh5uZpk5tRon3GuWT X-Google-Smtp-Source: AK7set9gY+NOfaP2EPxXeQTEmFaMokIIbBlTZWwiEUt/LmRMwCYwgGOWnXBndmFgQ57KlO5UkZnONg== X-Received: by 2002:a05:622a:1d3:b0:3e3:87a2:e7f5 with SMTP id t19-20020a05622a01d300b003e387a2e7f5mr20519135qtw.11.1679948018387; Mon, 27 Mar 2023 13:13:38 -0700 (PDT) Received: from localhost (pool-68-160-166-30.bstnma.fios.verizon.net. [68.160.166.30]) by smtp.gmail.com with ESMTPSA id 207-20020a3703d8000000b007468ec2e5dcsm10816449qkd.87.2023.03.27.13.13.37 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Mon, 27 Mar 2023 13:13:38 -0700 (PDT) From: Mike Snitzer To: dm-devel@redhat.com Cc: linux-block@vger.kernel.org, axboe@kernel.dk, ejt@redhat.com, mpatocka@redhat.com, heinzm@redhat.com, nhuck@google.com, ebiggers@kernel.org, keescook@chromium.org, luomeng12@huawei.com, Mike Snitzer Subject: [dm-6.4 PATCH v3 19/20] dm bio prison v1: prepare to intelligently size dm_bio_prison's prison_regions Date: Mon, 27 Mar 2023 16:11:42 -0400 Message-Id: <20230327201143.51026-20-snitzer@kernel.org> X-Mailer: git-send-email 2.40.0 In-Reply-To: <20230327201143.51026-1-snitzer@kernel.org> References: <20230327201143.51026-1-snitzer@kernel.org> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-block@vger.kernel.org Add num_locks member to dm_bio_prison struct and use it rather than the NR_LOCKS magic value (64). Next commit will size the dm_bio_prison's prison_regions according to dm_num_sharded_locks(). Signed-off-by: Mike Snitzer --- drivers/md/dm-bio-prison-v1.c | 21 +++++++++++---------- 1 file changed, 11 insertions(+), 10 deletions(-) diff --git a/drivers/md/dm-bio-prison-v1.c b/drivers/md/dm-bio-prison-v1.c index 78bb559b521c..a7930ad1878b 100644 --- a/drivers/md/dm-bio-prison-v1.c +++ b/drivers/md/dm-bio-prison-v1.c @@ -17,7 +17,6 @@ /*----------------------------------------------------------------*/ #define NR_LOCKS 64 -#define LOCK_MASK (NR_LOCKS - 1) #define MIN_CELLS 1024 struct prison_region { @@ -26,8 +25,9 @@ struct prison_region { } ____cacheline_aligned_in_smp; struct dm_bio_prison { - struct prison_region regions[NR_LOCKS]; mempool_t cell_pool; + unsigned int num_locks; + struct prison_region regions[NR_LOCKS]; }; static struct kmem_cache *_cell_cache; @@ -46,8 +46,9 @@ struct dm_bio_prison *dm_bio_prison_create(void) if (!prison) return NULL; + prison->num_locks = NR_LOCKS; - for (i = 0; i < NR_LOCKS; i++) { + for (i = 0; i < prison->num_locks; i++) { spin_lock_init(&prison->regions[i].lock); prison->regions[i].cell = RB_ROOT; } @@ -115,9 +116,9 @@ static int cmp_keys(struct dm_cell_key *lhs, return 0; } -static unsigned lock_nr(struct dm_cell_key *key) +static unsigned lock_nr(struct dm_cell_key *key, unsigned int num_locks) { - return (key->block_begin >> BIO_PRISON_MAX_RANGE_SHIFT) & LOCK_MASK; + return (key->block_begin >> BIO_PRISON_MAX_RANGE_SHIFT) & (num_locks - 1); } bool dm_cell_key_has_valid_range(struct dm_cell_key *key) @@ -176,7 +177,7 @@ static int bio_detain(struct dm_bio_prison *prison, struct dm_bio_prison_cell **cell_result) { int r; - unsigned l = lock_nr(key); + unsigned l = lock_nr(key, prison->num_locks); spin_lock_irq(&prison->regions[l].lock); r = __bio_detain(&prison->regions[l].cell, key, inmate, cell_prealloc, cell_result); @@ -224,7 +225,7 @@ void dm_cell_release(struct dm_bio_prison *prison, struct dm_bio_prison_cell *cell, struct bio_list *bios) { - unsigned l = lock_nr(&cell->key); + unsigned l = lock_nr(&cell->key, prison->num_locks); spin_lock_irq(&prison->regions[l].lock); __cell_release(&prison->regions[l].cell, cell, bios); @@ -247,7 +248,7 @@ void dm_cell_release_no_holder(struct dm_bio_prison *prison, struct dm_bio_prison_cell *cell, struct bio_list *inmates) { - unsigned l = lock_nr(&cell->key); + unsigned l = lock_nr(&cell->key, prison->num_locks); unsigned long flags; spin_lock_irqsave(&prison->regions[l].lock, flags); @@ -277,7 +278,7 @@ void dm_cell_visit_release(struct dm_bio_prison *prison, void *context, struct dm_bio_prison_cell *cell) { - unsigned l = lock_nr(&cell->key); + unsigned l = lock_nr(&cell->key, prison->num_locks); spin_lock_irq(&prison->regions[l].lock); visit_fn(context, cell); rb_erase(&cell->node, &prison->regions[l].cell); @@ -301,7 +302,7 @@ int dm_cell_promote_or_release(struct dm_bio_prison *prison, struct dm_bio_prison_cell *cell) { int r; - unsigned l = lock_nr(&cell->key); + unsigned l = lock_nr(&cell->key, prison->num_locks); spin_lock_irq(&prison->regions[l].lock); r = __promote_or_release(&prison->regions[l].cell, cell); From patchwork Mon Mar 27 20:11:43 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Mike Snitzer X-Patchwork-Id: 13190065 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 vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id BBF7DC6FD1D for ; Mon, 27 Mar 2023 20:14:33 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S232509AbjC0UOc (ORCPT ); Mon, 27 Mar 2023 16:14:32 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:39276 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S232573AbjC0UOb (ORCPT ); Mon, 27 Mar 2023 16:14:31 -0400 Received: from mail-qv1-f43.google.com (mail-qv1-f43.google.com [209.85.219.43]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id B02892129 for ; Mon, 27 Mar 2023 13:13:40 -0700 (PDT) Received: by mail-qv1-f43.google.com with SMTP id l7so7670042qvh.5 for ; Mon, 27 Mar 2023 13:13:40 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; t=1679948020; h=content-transfer-encoding:mime-version:references:in-reply-to :message-id:date:subject:cc:to:from:x-gm-message-state:from:to:cc :subject:date:message-id:reply-to; bh=mQyYjWMmcas91qVtcY8jvp2wBSEYoGBjkcnNGzLUcAg=; b=ag7EmRvTXsRG5wNkya8nPzToIkGtjT92iacdQeW6vDZt/5arl450KClfiHZv/wJC82 5oUYfn4ZZKZpD40nJA3wpK+FPEa83pMctFJvdi88QQa/LmdNqdgdEMOnDq1ZkrdW+Gyz BBpkIBvePeqbEDgVP+GrYeeORgVciegdfPvrmVpFI36WhW2SEDengnrcH8lW2VfPhxTX owip9xM8e/9+TRlMXz+Us8jYZvIhdhMs//7zP7oEKsyiBpi1OVg+peLgorTwebNUV1JG 6Z5ne/N/YZJMd313UnBauqFyyoxylFFKLn1YRmMTLbaEOY3Kc4xbmn+mqCSJrPYx5osK 9OOQ== X-Gm-Message-State: AAQBX9f3gZcJR/m5skDXe1vGqwRkxyr9CkDV0WEv6FhxbsdPz1K9KjY8 DKXt7RGPC67Ue95HiEoqcsb1 X-Google-Smtp-Source: AKy350aUOhS5a7pfiFoOC/vCDrOmXamVaEKNC4KXqyNQGje3o8Ci+KX+zebywjT0o5kNzZBKf3poOw== X-Received: by 2002:a05:6214:2aa6:b0:56c:376:3191 with SMTP id js6-20020a0562142aa600b0056c03763191mr21114410qvb.44.1679948019857; Mon, 27 Mar 2023 13:13:39 -0700 (PDT) Received: from localhost (pool-68-160-166-30.bstnma.fios.verizon.net. [68.160.166.30]) by smtp.gmail.com with ESMTPSA id a8-20020a0ce908000000b005dd8b9345a3sm3224293qvo.59.2023.03.27.13.13.39 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Mon, 27 Mar 2023 13:13:39 -0700 (PDT) From: Mike Snitzer To: dm-devel@redhat.com Cc: linux-block@vger.kernel.org, axboe@kernel.dk, ejt@redhat.com, mpatocka@redhat.com, heinzm@redhat.com, nhuck@google.com, ebiggers@kernel.org, keescook@chromium.org, luomeng12@huawei.com, Mike Snitzer Subject: [dm-6.4 PATCH v3 20/20] dm bio prison v1: intelligently size dm_bio_prison's prison_regions Date: Mon, 27 Mar 2023 16:11:43 -0400 Message-Id: <20230327201143.51026-21-snitzer@kernel.org> X-Mailer: git-send-email 2.40.0 In-Reply-To: <20230327201143.51026-1-snitzer@kernel.org> References: <20230327201143.51026-1-snitzer@kernel.org> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: linux-block@vger.kernel.org Size the dm_bio_prison's number of prison_region structs using dm_num_sharded_locks(). Signed-off-by: Mike Snitzer --- drivers/md/dm-bio-prison-v1.c | 11 ++++++----- 1 file changed, 6 insertions(+), 5 deletions(-) diff --git a/drivers/md/dm-bio-prison-v1.c b/drivers/md/dm-bio-prison-v1.c index a7930ad1878b..1d560a83b144 100644 --- a/drivers/md/dm-bio-prison-v1.c +++ b/drivers/md/dm-bio-prison-v1.c @@ -16,7 +16,6 @@ /*----------------------------------------------------------------*/ -#define NR_LOCKS 64 #define MIN_CELLS 1024 struct prison_region { @@ -27,7 +26,7 @@ struct prison_region { struct dm_bio_prison { mempool_t cell_pool; unsigned int num_locks; - struct prison_region regions[NR_LOCKS]; + struct prison_region regions[]; }; static struct kmem_cache *_cell_cache; @@ -41,12 +40,14 @@ static struct kmem_cache *_cell_cache; struct dm_bio_prison *dm_bio_prison_create(void) { int ret; - unsigned i; - struct dm_bio_prison *prison = kzalloc(sizeof(*prison), GFP_KERNEL); + unsigned int i, num_locks; + struct dm_bio_prison *prison; + num_locks = dm_num_sharded_locks(); + prison = kzalloc(struct_size(prison, regions, num_locks), GFP_KERNEL); if (!prison) return NULL; - prison->num_locks = NR_LOCKS; + prison->num_locks = num_locks; for (i = 0; i < prison->num_locks; i++) { spin_lock_init(&prison->regions[i].lock);