From patchwork Mon Jun 27 13:35:31 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Hao Xu X-Patchwork-Id: 12896647 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 09BBFC43334 for ; Mon, 27 Jun 2022 13:35:57 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S234561AbiF0Nf4 (ORCPT ); Mon, 27 Jun 2022 09:35:56 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:44090 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S235951AbiF0Nfz (ORCPT ); Mon, 27 Jun 2022 09:35:55 -0400 Received: from out0.migadu.com (out0.migadu.com [94.23.1.103]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id 3DFEA63A5 for ; Mon, 27 Jun 2022 06:35:55 -0700 (PDT) X-Report-Abuse: Please report any abuse attempt to abuse@migadu.com and include these headers. DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=linux.dev; s=key1; t=1656336953; h=from:from:reply-to:subject:subject:date:date:message-id:message-id: to:to:cc:cc:mime-version:mime-version: content-transfer-encoding:content-transfer-encoding: in-reply-to:in-reply-to:references:references; bh=weyq1RViEPpLZyQ6CiIHnd3SKxHsPNn7JBltw99c5cY=; b=bWSWdTFeskdmjOWI4QzkCEYKnOdDk/C/50LdmsSMAvKVTKtrGS3m3Bz2FfACNjceLd4Qnc UnVR2+rIuxXidAqCw+tSDkvfr+dJykyqDq5o6Vjsh+RbBwAvN8Hcv3wAB/V/ODGowKUvUv MTJlCD1opGJUmiWFfaN9iV6+1Y7n3/M= From: Hao Xu To: io-uring@vger.kernel.org Cc: Jens Axboe , Pavel Begunkov Subject: [PATCH 01/11] io-wq: add a worker flag for individual exit Date: Mon, 27 Jun 2022 21:35:31 +0800 Message-Id: <20220627133541.15223-2-hao.xu@linux.dev> In-Reply-To: <20220627133541.15223-1-hao.xu@linux.dev> References: <20220627133541.15223-1-hao.xu@linux.dev> MIME-Version: 1.0 X-Migadu-Flow: FLOW_OUT X-Migadu-Auth-User: linux.dev Precedence: bulk List-ID: X-Mailing-List: io-uring@vger.kernel.org From: Hao Xu Add a worker flag to control exit of an individual worker, this is needed for fixed worker in the next patches but also as a generic functionality. Signed-off-by: Hao Xu --- io_uring/io-wq.c | 11 ++++++++++- 1 file changed, 10 insertions(+), 1 deletion(-) diff --git a/io_uring/io-wq.c b/io_uring/io-wq.c index 824623bcf1a5..0c26805ca6de 100644 --- a/io_uring/io-wq.c +++ b/io_uring/io-wq.c @@ -26,6 +26,7 @@ enum { IO_WORKER_F_RUNNING = 2, /* account as running */ IO_WORKER_F_FREE = 4, /* worker on free list */ IO_WORKER_F_BOUND = 8, /* is doing bounded work */ + IO_WORKER_F_EXIT = 32, /* worker is exiting */ }; enum { @@ -639,8 +640,12 @@ static int io_wqe_worker(void *data) while (!test_bit(IO_WQ_BIT_EXIT, &wq->state)) { long ret; + if (worker->flags & IO_WORKER_F_EXIT) + break; + set_current_state(TASK_INTERRUPTIBLE); - while (io_acct_run_queue(acct)) + while (!(worker->flags & IO_WORKER_F_EXIT) && + io_acct_run_queue(acct)) io_worker_handle_work(worker); raw_spin_lock(&wqe->lock); @@ -656,6 +661,10 @@ static int io_wqe_worker(void *data) raw_spin_unlock(&wqe->lock); if (io_flush_signals()) continue; + if (worker->flags & IO_WORKER_F_EXIT) { + __set_current_state(TASK_RUNNING); + break; + } ret = schedule_timeout(WORKER_IDLE_TIMEOUT); if (signal_pending(current)) { struct ksignal ksig; From patchwork Mon Jun 27 13:35:32 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Hao Xu X-Patchwork-Id: 12896648 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 5CDE0C43334 for ; Mon, 27 Jun 2022 13:36:00 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S235951AbiF0Nf7 (ORCPT ); Mon, 27 Jun 2022 09:35:59 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:44114 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S235629AbiF0Nf7 (ORCPT ); Mon, 27 Jun 2022 09:35:59 -0400 Received: from out0.migadu.com (out0.migadu.com [IPv6:2001:41d0:2:267::]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id A73006362 for ; Mon, 27 Jun 2022 06:35:58 -0700 (PDT) X-Report-Abuse: Please report any abuse attempt to abuse@migadu.com and include these headers. DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=linux.dev; s=key1; t=1656336956; h=from:from:reply-to:subject:subject:date:date:message-id:message-id: to:to:cc:cc:mime-version:mime-version: content-transfer-encoding:content-transfer-encoding: in-reply-to:in-reply-to:references:references; bh=eFAqcjlFMxWhM2a8jmHB1NuKlvvBP5MivvkGdb6YA/0=; b=lIwHRVRc+Ey3e/ExskNuvvSrisVKSq+4aLWfDch6gc7HzTx7h9rvspcocLWXR6CGodctVl 7bwCPMPVZXIwRWy8hoGoLj9oT8O/NZY5omfwcJf9JwZueGs2xZ5p9rKiU1XP8VLRG0aTrA nlphyrTfgxw4bceLEvKWfEnNnkzrrNY= From: Hao Xu To: io-uring@vger.kernel.org Cc: Jens Axboe , Pavel Begunkov Subject: [PATCH 02/11] io-wq: change argument of create_io_worker() for convienence Date: Mon, 27 Jun 2022 21:35:32 +0800 Message-Id: <20220627133541.15223-3-hao.xu@linux.dev> In-Reply-To: <20220627133541.15223-1-hao.xu@linux.dev> References: <20220627133541.15223-1-hao.xu@linux.dev> MIME-Version: 1.0 X-Migadu-Flow: FLOW_OUT X-Migadu-Auth-User: linux.dev Precedence: bulk List-ID: X-Mailing-List: io-uring@vger.kernel.org From: Hao Xu Change index to acct itself for create_io_worker() for convienence in the next patches. Signed-off-by: Hao Xu --- io_uring/io-wq.c | 12 +++++++----- 1 file changed, 7 insertions(+), 5 deletions(-) diff --git a/io_uring/io-wq.c b/io_uring/io-wq.c index 0c26805ca6de..35ce622f77ba 100644 --- a/io_uring/io-wq.c +++ b/io_uring/io-wq.c @@ -139,7 +139,8 @@ struct io_cb_cancel_data { bool cancel_all; }; -static bool create_io_worker(struct io_wq *wq, struct io_wqe *wqe, int index); +static bool create_io_worker(struct io_wq *wq, struct io_wqe *wqe, + struct io_wqe_acct *acct); static void io_wqe_dec_running(struct io_worker *worker); static bool io_acct_cancel_pending_work(struct io_wqe *wqe, struct io_wqe_acct *acct, @@ -306,7 +307,7 @@ static bool io_wqe_create_worker(struct io_wqe *wqe, struct io_wqe_acct *acct) raw_spin_unlock(&wqe->lock); atomic_inc(&acct->nr_running); atomic_inc(&wqe->wq->worker_refs); - return create_io_worker(wqe->wq, wqe, acct->index); + return create_io_worker(wqe->wq, wqe, acct); } static void io_wqe_inc_running(struct io_worker *worker) @@ -335,7 +336,7 @@ static void create_worker_cb(struct callback_head *cb) } raw_spin_unlock(&wqe->lock); if (do_create) { - create_io_worker(wq, wqe, worker->create_index); + create_io_worker(wq, wqe, acct); } else { atomic_dec(&acct->nr_running); io_worker_ref_put(wq); @@ -812,9 +813,10 @@ static void io_workqueue_create(struct work_struct *work) kfree(worker); } -static bool create_io_worker(struct io_wq *wq, struct io_wqe *wqe, int index) +static bool create_io_worker(struct io_wq *wq, struct io_wqe *wqe, + struct io_wqe_acct *acct) { - struct io_wqe_acct *acct = &wqe->acct[index]; + int index = acct->index; struct io_worker *worker; struct task_struct *tsk; From patchwork Mon Jun 27 13:35:33 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Hao Xu X-Patchwork-Id: 12896649 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 11B02C433EF for ; Mon, 27 Jun 2022 13:36:03 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S235629AbiF0NgC (ORCPT ); Mon, 27 Jun 2022 09:36:02 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:44130 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S234557AbiF0NgC (ORCPT ); Mon, 27 Jun 2022 09:36:02 -0400 Received: from out0.migadu.com (out0.migadu.com [IPv6:2001:41d0:2:267::]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id 1C76F6362 for ; Mon, 27 Jun 2022 06:36:01 -0700 (PDT) X-Report-Abuse: Please report any abuse attempt to abuse@migadu.com and include these headers. DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=linux.dev; s=key1; t=1656336959; h=from:from:reply-to:subject:subject:date:date:message-id:message-id: to:to:cc:cc:mime-version:mime-version: content-transfer-encoding:content-transfer-encoding: in-reply-to:in-reply-to:references:references; bh=H0JYJ8IO2S6E5ZdsH649EidNbXdzvlMspRSyF8OEuE4=; b=h1eUMLt0ZkF8JsyVRkQuD2NIp0yZ/OfZrYt8oR4/lZ8Z4fgfy3wf79999BTdoPdYJXqw9u 1IOaPIdn3+bL41DVIlaZZx0dJorjTTjH0u7t9rb2+UefuDemhS7plT98YJWDApB2nzdPS7 ra0flMxH6Wh1pCWhRcXoPen9gvoxgto= From: Hao Xu To: io-uring@vger.kernel.org Cc: Jens Axboe , Pavel Begunkov Subject: [PATCH 03/11] io-wq: add infra data structure for fixed workers Date: Mon, 27 Jun 2022 21:35:33 +0800 Message-Id: <20220627133541.15223-4-hao.xu@linux.dev> In-Reply-To: <20220627133541.15223-1-hao.xu@linux.dev> References: <20220627133541.15223-1-hao.xu@linux.dev> MIME-Version: 1.0 X-Migadu-Flow: FLOW_OUT X-Migadu-Auth-User: linux.dev Precedence: bulk List-ID: X-Mailing-List: io-uring@vger.kernel.org From: Hao Xu Add data sttructure and basic initialization for fixed worker. Signed-off-by: Hao Xu --- io_uring/io-wq.c | 103 ++++++++++++++++++++++++++++++++++++++++++----- 1 file changed, 92 insertions(+), 11 deletions(-) diff --git a/io_uring/io-wq.c b/io_uring/io-wq.c index 35ce622f77ba..d9b3aeea2c6a 100644 --- a/io_uring/io-wq.c +++ b/io_uring/io-wq.c @@ -26,6 +26,7 @@ enum { IO_WORKER_F_RUNNING = 2, /* account as running */ IO_WORKER_F_FREE = 4, /* worker on free list */ IO_WORKER_F_BOUND = 8, /* is doing bounded work */ + IO_WORKER_F_FIXED = 16, /* is a fixed worker */ IO_WORKER_F_EXIT = 32, /* worker is exiting */ }; @@ -37,6 +38,66 @@ enum { IO_ACCT_STALLED_BIT = 0, /* stalled on hash */ }; +struct io_wqe_acct { +/* + * union { + * // (1) for wqe->acct (normal worker) + * struct { + * unsigned nr_workers; + * unsigned max_workers; + * struct io_wq_work_list work_list; + * }; + * // (2) for wqe->fixed_acct (fixed worker) + * struct { + * unsigned nr_workers; + * unsigned max_workers; + * unsigned nr_fixed; + * unsigned max_works; + * struct io_worker **fixed_workers; + * }; + * // (3) for fixed worker's private acct + * struct { + * unsigned nr_works; + * unsigned max_works; + * struct io_wq_work_list work_list; + * }; + * }; + */ + union { + struct { + unsigned nr_workers; + unsigned max_workers; + }; + unsigned nr_works; + }; + unsigned max_works; + union { + struct io_wq_work_list work_list; + struct { + unsigned nr_fixed; + struct io_worker **fixed_workers; + }; + }; + + /* + * nr_running is not meaningful for fixed worker + * but still keep the same logic for it for the + * convinence for now. So do nr_workers and + * max_workers. + */ + atomic_t nr_running; + /* + * For 1), it protects the work_list, the other two member nr_workers + * and max_workers are protected by wqe->lock. + * For 2), it protects nr_fixed, max_works, fixed_workers + * For 3), it protects nr_works, max_works and work_list. + */ + raw_spinlock_t lock; + int index; + unsigned long flags; + bool fixed_worker_registered; +}; + /* * One for each thread in a wqe pool */ @@ -62,6 +123,8 @@ struct io_worker { struct rcu_head rcu; struct work_struct work; }; + int index; + struct io_wqe_acct acct; }; #if BITS_PER_LONG == 64 @@ -72,16 +135,6 @@ struct io_worker { #define IO_WQ_NR_HASH_BUCKETS (1u << IO_WQ_HASH_ORDER) -struct io_wqe_acct { - unsigned nr_workers; - unsigned max_workers; - int index; - atomic_t nr_running; - raw_spinlock_t lock; - struct io_wq_work_list work_list; - unsigned long flags; -}; - enum { IO_WQ_ACCT_BOUND, IO_WQ_ACCT_UNBOUND, @@ -94,6 +147,7 @@ enum { struct io_wqe { raw_spinlock_t lock; struct io_wqe_acct acct[IO_WQ_ACCT_NR]; + struct io_wqe_acct fixed_acct[IO_WQ_ACCT_NR]; int node; @@ -1205,6 +1259,31 @@ struct io_wq *io_wq_create(unsigned bounded, struct io_wq_data *data) atomic_set(&acct->nr_running, 0); INIT_WQ_LIST(&acct->work_list); raw_spin_lock_init(&acct->lock); + + acct = &wqe->fixed_acct[i]; + acct->index = i; + INIT_WQ_LIST(&acct->work_list); + raw_spin_lock_init(&acct->lock); + /* + * nr_running for a fixed worker is meaningless + * for now, init it to 1 to wround around the + * io_wqe_dec_running logic + */ + atomic_set(&acct->nr_running, 1); + /* + * max_workers for a fixed worker is meaningless + * for now, init it so since number of fixed workers + * should be controlled by users. + */ + acct->max_workers = task_rlimit(current, RLIMIT_NPROC); + raw_spin_lock_init(&acct->lock); + /* + * For fixed worker, not necessary + * but do it explicitly for clearity + */ + acct->nr_fixed = 0; + acct->max_works = 0; + acct->fixed_workers = NULL; } wqe->wq = wq; raw_spin_lock_init(&wqe->lock); @@ -1287,7 +1366,7 @@ static void io_wq_exit_workers(struct io_wq *wq) static void io_wq_destroy(struct io_wq *wq) { - int node; + int i, node; cpuhp_state_remove_instance_nocalls(io_wq_online, &wq->cpuhp_node); @@ -1299,6 +1378,8 @@ static void io_wq_destroy(struct io_wq *wq) }; io_wqe_cancel_pending_work(wqe, &match); free_cpumask_var(wqe->cpu_mask); + for (i = 0; i < IO_WQ_ACCT_NR; i++) + kfree(wqe->fixed_acct[i].fixed_workers); kfree(wqe); } io_wq_put_hash(wq->hash); From patchwork Mon Jun 27 13:35:34 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Hao Xu X-Patchwork-Id: 12896650 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 E005FC43334 for ; Mon, 27 Jun 2022 13:36:05 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S234154AbiF0NgF (ORCPT ); Mon, 27 Jun 2022 09:36:05 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:44158 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S234557AbiF0NgE (ORCPT ); Mon, 27 Jun 2022 09:36:04 -0400 Received: from out0.migadu.com (out0.migadu.com [IPv6:2001:41d0:2:267::]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id E06CB6362 for ; Mon, 27 Jun 2022 06:36:03 -0700 (PDT) X-Report-Abuse: Please report any abuse attempt to abuse@migadu.com and include these headers. DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=linux.dev; s=key1; t=1656336962; h=from:from:reply-to:subject:subject:date:date:message-id:message-id: to:to:cc:cc:mime-version:mime-version: content-transfer-encoding:content-transfer-encoding: in-reply-to:in-reply-to:references:references; bh=vMfnbP3LS47Qq96xStumF96UHoQqwFDSioFSA6kvU68=; b=jTGWWy2Qy7MUPzEOnvVZN3nrOIMDFxObuoLeHSh6mSsvqe9OzEAsT9Fsd4F6fJGAjLc2ZK lB0SLlEhK65uKFsSYW69b3HBfm2mBWr/K02RAWsNhw80MvYpmgKMWmVFLhwMzsWa2Wl2pF jXQslnxrDV4GxYnQmrvoRyd8+4ktHH8= From: Hao Xu To: io-uring@vger.kernel.org Cc: Jens Axboe , Pavel Begunkov Subject: [PATCH 04/11] io-wq: tweak io_get_acct() Date: Mon, 27 Jun 2022 21:35:34 +0800 Message-Id: <20220627133541.15223-5-hao.xu@linux.dev> In-Reply-To: <20220627133541.15223-1-hao.xu@linux.dev> References: <20220627133541.15223-1-hao.xu@linux.dev> MIME-Version: 1.0 X-Migadu-Flow: FLOW_OUT X-Migadu-Auth-User: linux.dev Precedence: bulk List-ID: X-Mailing-List: io-uring@vger.kernel.org From: Hao Xu Add an argument for io_get_acct() to indicate fixed or normal worker Signed-off-by: Hao Xu --- io_uring/io-wq.c | 14 +++++++++----- 1 file changed, 9 insertions(+), 5 deletions(-) diff --git a/io_uring/io-wq.c b/io_uring/io-wq.c index d9b3aeea2c6a..7775ba5fddba 100644 --- a/io_uring/io-wq.c +++ b/io_uring/io-wq.c @@ -213,20 +213,24 @@ static void io_worker_release(struct io_worker *worker) complete(&worker->ref_done); } -static inline struct io_wqe_acct *io_get_acct(struct io_wqe *wqe, bool bound) +static inline struct io_wqe_acct *io_get_acct(struct io_wqe *wqe, bool bound, + bool fixed) { - return &wqe->acct[bound ? IO_WQ_ACCT_BOUND : IO_WQ_ACCT_UNBOUND]; + unsigned index = bound ? IO_WQ_ACCT_BOUND : IO_WQ_ACCT_UNBOUND; + + return fixed ? &wqe->fixed_acct[index] : &wqe->acct[index]; } static inline struct io_wqe_acct *io_work_get_acct(struct io_wqe *wqe, struct io_wq_work *work) { - return io_get_acct(wqe, !(work->flags & IO_WQ_WORK_UNBOUND)); + return io_get_acct(wqe, !(work->flags & IO_WQ_WORK_UNBOUND), false); } static inline struct io_wqe_acct *io_wqe_get_acct(struct io_worker *worker) { - return io_get_acct(worker->wqe, worker->flags & IO_WORKER_F_BOUND); + return io_get_acct(worker->wqe, worker->flags & IO_WORKER_F_BOUND, + worker->flags & IO_WORKER_F_FIXED); } static void io_worker_ref_put(struct io_wq *wq) @@ -1129,7 +1133,7 @@ static void io_wqe_cancel_pending_work(struct io_wqe *wqe, int i; retry: for (i = 0; i < IO_WQ_ACCT_NR; i++) { - struct io_wqe_acct *acct = io_get_acct(wqe, i == 0); + struct io_wqe_acct *acct = io_get_acct(wqe, i == 0, false); if (io_acct_cancel_pending_work(wqe, acct, match)) { if (match->cancel_all) From patchwork Mon Jun 27 13:35:35 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Hao Xu X-Patchwork-Id: 12896651 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 95349C433EF for ; Mon, 27 Jun 2022 13:36:08 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S235823AbiF0NgI (ORCPT ); Mon, 27 Jun 2022 09:36:08 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:44176 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S234557AbiF0NgH (ORCPT ); Mon, 27 Jun 2022 09:36:07 -0400 Received: from out0.migadu.com (out0.migadu.com [94.23.1.103]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id 820EE6362 for ; Mon, 27 Jun 2022 06:36:06 -0700 (PDT) X-Report-Abuse: Please report any abuse attempt to abuse@migadu.com and include these headers. DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=linux.dev; s=key1; t=1656336965; h=from:from:reply-to:subject:subject:date:date:message-id:message-id: to:to:cc:cc:mime-version:mime-version: content-transfer-encoding:content-transfer-encoding: in-reply-to:in-reply-to:references:references; bh=lq9xSQoJ1dWu1UNDvb4YbQASOJjgErq67w2TQXDxjEU=; b=n9ZdrxS+YNHp6JNV0IfqLJYpzZaE8j/EFfmFnubR9dRee1g5GND8kkeaz8SNCXAes2CBiz bGj/HNpGPMUsAQs5AzMteJ4Y2pjSfFrNGJDszvZoGRSLs6ua4IlJksKFlwNYCmj2gqF5dA rWVi+WU7dUPyN/oHnxHa931073g+OzA= From: Hao Xu To: io-uring@vger.kernel.org Cc: Jens Axboe , Pavel Begunkov Subject: [PATCH 05/11] io-wq: fixed worker initialization Date: Mon, 27 Jun 2022 21:35:35 +0800 Message-Id: <20220627133541.15223-6-hao.xu@linux.dev> In-Reply-To: <20220627133541.15223-1-hao.xu@linux.dev> References: <20220627133541.15223-1-hao.xu@linux.dev> MIME-Version: 1.0 X-Migadu-Flow: FLOW_OUT X-Migadu-Auth-User: linux.dev Precedence: bulk List-ID: X-Mailing-List: io-uring@vger.kernel.org From: Hao Xu Implementation of the fixed worker initialization. Signed-off-by: Hao Xu --- io_uring/io-wq.c | 24 ++++++++++++++++++++++++ 1 file changed, 24 insertions(+) diff --git a/io_uring/io-wq.c b/io_uring/io-wq.c index 7775ba5fddba..70240b847112 100644 --- a/io_uring/io-wq.c +++ b/io_uring/io-wq.c @@ -779,6 +779,26 @@ void io_wq_worker_sleeping(struct task_struct *tsk) io_wqe_dec_running(worker); } +static void io_init_new_fixed_worker(struct io_wqe *wqe, + struct io_worker *worker) +{ + struct io_wqe_acct *acct = io_wqe_get_acct(worker); + struct io_wqe_acct *iw_acct = &worker->acct; + unsigned index = acct->index; + unsigned *nr_fixed; + + raw_spin_lock(&acct->lock); + nr_fixed = &acct->nr_fixed; + acct->fixed_workers[*nr_fixed] = worker; + worker->index = (*nr_fixed)++; + iw_acct->nr_works = 0; + iw_acct->max_works = acct->max_works; + iw_acct->index = index; + INIT_WQ_LIST(&iw_acct->work_list); + raw_spin_lock_init(&iw_acct->lock); + raw_spin_unlock(&acct->lock); +} + static void io_init_new_worker(struct io_wqe *wqe, struct io_worker *worker, struct task_struct *tsk) { @@ -792,6 +812,8 @@ static void io_init_new_worker(struct io_wqe *wqe, struct io_worker *worker, list_add_tail_rcu(&worker->all_list, &wqe->all_list); worker->flags |= IO_WORKER_F_FREE; raw_spin_unlock(&wqe->lock); + if (worker->flags & IO_WORKER_F_FIXED) + io_init_new_fixed_worker(wqe, worker); wake_up_new_task(tsk); } @@ -898,6 +920,8 @@ static bool create_io_worker(struct io_wq *wq, struct io_wqe *wqe, if (index == IO_WQ_ACCT_BOUND) worker->flags |= IO_WORKER_F_BOUND; + if (&wqe->fixed_acct[index] == acct) + worker->flags |= IO_WORKER_F_FIXED; tsk = create_io_thread(io_wqe_worker, worker, wqe->node); if (!IS_ERR(tsk)) { From patchwork Mon Jun 27 13:35:36 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Hao Xu X-Patchwork-Id: 12896652 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 5BB61C43334 for ; Mon, 27 Jun 2022 13:36:10 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S235950AbiF0NgJ (ORCPT ); Mon, 27 Jun 2022 09:36:09 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:44184 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S234557AbiF0NgJ (ORCPT ); Mon, 27 Jun 2022 09:36:09 -0400 Received: from out0.migadu.com (out0.migadu.com [IPv6:2001:41d0:2:267::]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id BBCF56362 for ; Mon, 27 Jun 2022 06:36:08 -0700 (PDT) X-Report-Abuse: Please report any abuse attempt to abuse@migadu.com and include these headers. DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=linux.dev; s=key1; t=1656336967; h=from:from:reply-to:subject:subject:date:date:message-id:message-id: to:to:cc:cc:mime-version:mime-version: content-transfer-encoding:content-transfer-encoding: in-reply-to:in-reply-to:references:references; bh=BX61HmQlVCCZmq3PJ/mL1vyIpiqCl6HtuPRTd4Rejt4=; b=gWj/IMcf4ROC5UwAFuMrE+lgGOPW0eaQRxChHddtXm7VJLP8IfBQ/677VckKQswlFG9srC PrD3wxlF3PnI7og8UxH77joIkjC0PLmNObM+CfGXE/9CgxGJvaVqYEddqSd7g6vKyGQ7EF FHId7i+tE+jVam8AaKBU3Zc7+uQ1IWE= From: Hao Xu To: io-uring@vger.kernel.org Cc: Jens Axboe , Pavel Begunkov Subject: [PATCH 06/11] io-wq: fixed worker exit Date: Mon, 27 Jun 2022 21:35:36 +0800 Message-Id: <20220627133541.15223-7-hao.xu@linux.dev> In-Reply-To: <20220627133541.15223-1-hao.xu@linux.dev> References: <20220627133541.15223-1-hao.xu@linux.dev> MIME-Version: 1.0 X-Migadu-Flow: FLOW_OUT X-Migadu-Auth-User: linux.dev Precedence: bulk List-ID: X-Mailing-List: io-uring@vger.kernel.org From: Hao Xu Implement the fixed worker exit Signed-off-by: Hao Xu --- io_uring/io-wq.c | 26 ++++++++++++++++++++++++++ 1 file changed, 26 insertions(+) diff --git a/io_uring/io-wq.c b/io_uring/io-wq.c index 70240b847112..11fe5787c409 100644 --- a/io_uring/io-wq.c +++ b/io_uring/io-wq.c @@ -264,6 +264,29 @@ static bool io_task_worker_match(struct callback_head *cb, void *data) return worker == data; } +static void io_fixed_worker_exit(struct io_worker *worker) +{ + int *nr_fixed; + int index = worker->acct.index; + struct io_wqe *wqe = worker->wqe; + struct io_wqe_acct *acct = io_get_acct(wqe, index == 0, true); + struct io_worker **fixed_workers; + + raw_spin_lock(&acct->lock); + fixed_workers = acct->fixed_workers; + if (!fixed_workers || worker->index == -1) { + raw_spin_unlock(&acct->lock); + return; + } + nr_fixed = &acct->nr_fixed; + /* reuse variable index to represent fixed worker index in its array */ + index = worker->index; + fixed_workers[index] = fixed_workers[*nr_fixed - 1]; + (*nr_fixed)--; + fixed_workers[index]->index = index; + raw_spin_unlock(&acct->lock); +} + static void io_worker_exit(struct io_worker *worker) { struct io_wqe *wqe = worker->wqe; @@ -687,6 +710,7 @@ static int io_wqe_worker(void *data) struct io_wqe *wqe = worker->wqe; struct io_wq *wq = wqe->wq; bool last_timeout = false; + bool fixed = worker->flags & IO_WORKER_F_FIXED; char buf[TASK_COMM_LEN]; worker->flags |= (IO_WORKER_F_UP | IO_WORKER_F_RUNNING); @@ -737,6 +761,8 @@ static int io_wqe_worker(void *data) if (test_bit(IO_WQ_BIT_EXIT, &wq->state)) io_worker_handle_work(worker); + if (fixed) + io_fixed_worker_exit(worker); audit_free(current); io_worker_exit(worker); From patchwork Mon Jun 27 13:35:37 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Hao Xu X-Patchwork-Id: 12896653 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 17DC0C433EF for ; Mon, 27 Jun 2022 13:36:15 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S235927AbiF0NgO (ORCPT ); Mon, 27 Jun 2022 09:36:14 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:44226 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S234557AbiF0NgN (ORCPT ); Mon, 27 Jun 2022 09:36:13 -0400 Received: from out0.migadu.com (out0.migadu.com [IPv6:2001:41d0:2:267::]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id 578A06362 for ; Mon, 27 Jun 2022 06:36:12 -0700 (PDT) X-Report-Abuse: Please report any abuse attempt to abuse@migadu.com and include these headers. DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=linux.dev; s=key1; t=1656336971; h=from:from:reply-to:subject:subject:date:date:message-id:message-id: to:to:cc:cc:mime-version:mime-version: content-transfer-encoding:content-transfer-encoding: in-reply-to:in-reply-to:references:references; bh=RniOcn9yPuvipwzwt28WCa9KHB7H+Sp1eVey/+p4yNE=; b=eRtI2NderyvAJesaCu69Ue46FrYS91bh7OZep2B5GUS7Sa3LN0gkMC7Frt992WxedDionK QpKjh4NK60esrY750ooCRCtG1U6M3Y+CkErRU3lSNdcPNf6lHu5y1HgXRpvqj3jumbvfiE gbaQ0G9qmB7Ae1RTDJbKYpT5AqRRxGY= From: Hao Xu To: io-uring@vger.kernel.org Cc: Jens Axboe , Pavel Begunkov Subject: [PATCH 07/11] io-wq: implement fixed worker logic Date: Mon, 27 Jun 2022 21:35:37 +0800 Message-Id: <20220627133541.15223-8-hao.xu@linux.dev> In-Reply-To: <20220627133541.15223-1-hao.xu@linux.dev> References: <20220627133541.15223-1-hao.xu@linux.dev> MIME-Version: 1.0 X-Migadu-Flow: FLOW_OUT X-Migadu-Auth-User: linux.dev Precedence: bulk List-ID: X-Mailing-List: io-uring@vger.kernel.org From: Hao Xu The current implementation of io-wq has big spinlock contension. The main reason is the single work list model. All producers(who insert works) and consumers(io-workers) have to grap wqe->lock to move ahead. Set max_worker to 3 or 4, do a fio read test, we can see 40%~50% lock contension. Introduce fixed io-workers which sticks there to handle works and have their own work list. previous: producer0 ---insert---> work_list ---get---> io-worker0,1,2 now: ---> private work_list0 --get--> fixed-worker0 / producer0 --insert----> private work_list1 --get--> fixed-worker1 | \ | ---> private work_list2 --get--> fixed-worker2 | |---insert---> public work_list --get--> (normal)io-worker Since each fixed-worker has a private work list, the contension will be limited to a smaller range(the private work list). Logic of fixed-worker: first handle private works then public ones. Logic of normal io-worker: only handle public works. Logic of producer: 1) randomly pick a private work list and check if it is full, insert the work if it's not 2) insert the work to the public work list if 1) fails. The get logic of a private list: fixed-worker grab all the works in its private work list(like what tctx_task_work() does) rather than one by one.(this code is in the next patches as a optimization) To achieve this, we need to add an io_wqe_acct for each fixed-worker struct, and though this we can leverage the old code as much as possible, which makes the new design clean and compatible. Good things of this feature: 1) bound and unbound work lists now have different spinlocks. 2) much smaller contension between work producers and consumers. 3) fixed workers are friendly for users to control: binding cpus, reset priority etc. Wrote a nop test program to test it, 3 fixed-workers VS 3 normal workers. normal workers: ./run_nop_wqe.sh nop_wqe_normal 200000 100 3 1-3 time spent: 17314274 usecs IOPS: 1155116 time spent: 17016942 usecs IOPS: 1175299 time spent: 17908684 usecs IOPS: 1116776 fixed workers: ./run_nop_wqe.sh nop_wqe_fixed 200000 100 3 1-3 time spent: 10464397 usecs IOPS: 1911242 time spent: 9610976 usecs IOPS: 2080954 time spent: 9807361 usecs IOPS: 2039284 About 2x improvement. From perf result, almost no acct->lock contension. Test program: https://github.com/HowHsu/liburing/tree/fixed_worker Signed-off-by: Hao Xu --- io_uring/io-wq.c | 148 ++++++++++++++++++++++++++++++++++++++--------- 1 file changed, 122 insertions(+), 26 deletions(-) diff --git a/io_uring/io-wq.c b/io_uring/io-wq.c index 11fe5787c409..c87ba38f27b1 100644 --- a/io_uring/io-wq.c +++ b/io_uring/io-wq.c @@ -71,6 +71,7 @@ struct io_wqe_acct { unsigned nr_works; }; unsigned max_works; + unsigned work_seq; union { struct io_wq_work_list work_list; struct { @@ -629,9 +630,9 @@ static void io_assign_current_work(struct io_worker *worker, static void io_wqe_enqueue(struct io_wqe *wqe, struct io_wq_work *work); -static void io_worker_handle_work(struct io_worker *worker) +static void io_worker_handle_work(struct io_worker *worker, + struct io_wqe_acct *acct) { - struct io_wqe_acct *acct = io_wqe_get_acct(worker); struct io_wqe *wqe = worker->wqe; struct io_wq *wq = wqe->wq; bool do_kill = test_bit(IO_WQ_BIT_EXIT, &wq->state); @@ -703,19 +704,31 @@ static void io_worker_handle_work(struct io_worker *worker) } while (1); } +static inline void io_worker_handle_private_work(struct io_worker *worker) +{ + io_worker_handle_work(worker, &worker->acct); +} + +static inline void io_worker_handle_public_work(struct io_worker *worker) +{ + io_worker_handle_work(worker, io_wqe_get_acct(worker)); +} + static int io_wqe_worker(void *data) { struct io_worker *worker = data; - struct io_wqe_acct *acct = io_wqe_get_acct(worker); struct io_wqe *wqe = worker->wqe; struct io_wq *wq = wqe->wq; - bool last_timeout = false; + struct io_wqe_acct *acct = + io_get_acct(wqe, worker->flags & IO_WORKER_F_BOUND, false); bool fixed = worker->flags & IO_WORKER_F_FIXED; + bool last_timeout = false; char buf[TASK_COMM_LEN]; worker->flags |= (IO_WORKER_F_UP | IO_WORKER_F_RUNNING); - snprintf(buf, sizeof(buf), "iou-wrk-%d", wq->task->pid); + snprintf(buf, sizeof(buf), fixed ? "iou-fix-%d" : "iou-wrk-%d", + wq->task->pid); set_task_comm(current, buf); audit_alloc_kernel(current); @@ -727,13 +740,24 @@ static int io_wqe_worker(void *data) break; set_current_state(TASK_INTERRUPTIBLE); - while (!(worker->flags & IO_WORKER_F_EXIT) && - io_acct_run_queue(acct)) - io_worker_handle_work(worker); - + if (fixed) { + while (io_acct_run_queue(&worker->acct)) + io_worker_handle_private_work(worker); + if (io_acct_run_queue(acct)) + io_worker_handle_public_work(worker); + } else { + while (io_acct_run_queue(acct)) + io_worker_handle_public_work(worker); + } raw_spin_lock(&wqe->lock); - /* timed out, exit unless we're the last worker */ - if (last_timeout && acct->nr_workers > 1) { + /* timed out, a worker will exit only if: + * - not a fixed worker + * - not the last non-fixed worker + * + * the second condition is due to we need at least one worker to + * handle the public work list. + */ + if (last_timeout && !fixed && acct->nr_workers > 1) { acct->nr_workers--; raw_spin_unlock(&wqe->lock); __set_current_state(TASK_RUNNING); @@ -759,10 +783,18 @@ static int io_wqe_worker(void *data) last_timeout = !ret; } - if (test_bit(IO_WQ_BIT_EXIT, &wq->state)) - io_worker_handle_work(worker); - if (fixed) + if (test_bit(IO_WQ_BIT_EXIT, &wq->state) && !fixed) + io_worker_handle_public_work(worker); + if (fixed) { io_fixed_worker_exit(worker); + /* + * Check and handle private work list again + * to avoid race with private work insertion + * TODO: an alternative way is to deliver + * works to the public work list + */ + io_worker_handle_private_work(worker); + } audit_free(current); io_worker_exit(worker); @@ -1006,9 +1038,9 @@ static void io_run_cancel(struct io_wq_work *work, struct io_wqe *wqe) } while (work); } -static void io_wqe_insert_work(struct io_wqe *wqe, struct io_wq_work *work) +static void io_wqe_insert_work(struct io_wqe *wqe, struct io_wq_work *work, + struct io_wqe_acct *acct) { - struct io_wqe_acct *acct = io_work_get_acct(wqe, work); unsigned int hash; struct io_wq_work *tail; @@ -1027,6 +1059,45 @@ static void io_wqe_insert_work(struct io_wqe *wqe, struct io_wq_work *work) wq_list_add_after(&work->list, &tail->list, &acct->work_list); } +static bool io_wqe_insert_private_work(struct io_wqe *wqe, + struct io_wq_work *work, + struct io_wqe_acct *acct) +{ + unsigned int nr_fixed; + struct io_worker *fixed_worker; + struct io_wqe_acct *iw_acct; + unsigned int fixed_worker_index; + + raw_spin_lock(&acct->lock); + nr_fixed = acct->nr_fixed; + if (!nr_fixed) { + raw_spin_unlock(&acct->lock); + return false; + } + + fixed_worker_index = (acct->work_seq++) % nr_fixed; + fixed_worker = acct->fixed_workers[fixed_worker_index]; + if (!fixed_worker || fixed_worker->flags & IO_WORKER_F_EXIT) { + raw_spin_unlock(&acct->lock); + return false; + } + iw_acct = &fixed_worker->acct; + + raw_spin_lock(&iw_acct->lock); + if (iw_acct->nr_works < iw_acct->max_works) { + io_wqe_insert_work(wqe, work, iw_acct); + iw_acct->nr_works++; + raw_spin_unlock(&iw_acct->lock); + wake_up_process(fixed_worker->task); + raw_spin_unlock(&acct->lock); + return true; + } + raw_spin_unlock(&iw_acct->lock); + raw_spin_unlock(&acct->lock); + + return false; +} + static bool io_wq_work_match_item(struct io_wq_work *work, void *data) { return work == data; @@ -1035,6 +1106,7 @@ static bool io_wq_work_match_item(struct io_wq_work *work, void *data) static void io_wqe_enqueue(struct io_wqe *wqe, struct io_wq_work *work) { struct io_wqe_acct *acct = io_work_get_acct(wqe, work); + struct io_wqe_acct *fixed_acct; struct io_cb_cancel_data match; unsigned work_flags = work->flags; bool do_create; @@ -1049,8 +1121,14 @@ static void io_wqe_enqueue(struct io_wqe *wqe, struct io_wq_work *work) return; } + fixed_acct = io_get_acct(wqe, !acct->index, true); + if (fixed_acct->fixed_worker_registered && !io_wq_is_hashed(work)) { + if (io_wqe_insert_private_work(wqe, work, fixed_acct)) + return; + } + raw_spin_lock(&acct->lock); - io_wqe_insert_work(wqe, work); + io_wqe_insert_work(wqe, work, acct); clear_bit(IO_ACCT_STALLED_BIT, &acct->flags); raw_spin_unlock(&acct->lock); @@ -1136,9 +1214,9 @@ static bool io_wq_worker_cancel(struct io_worker *worker, void *data) static inline void io_wqe_remove_pending(struct io_wqe *wqe, struct io_wq_work *work, - struct io_wq_work_node *prev) + struct io_wq_work_node *prev, + struct io_wqe_acct *acct) { - struct io_wqe_acct *acct = io_work_get_acct(wqe, work); unsigned int hash = io_get_work_hash(work); struct io_wq_work *prev_work = NULL; @@ -1165,7 +1243,7 @@ static bool io_acct_cancel_pending_work(struct io_wqe *wqe, work = container_of(node, struct io_wq_work, list); if (!match->fn(work, match->data)) continue; - io_wqe_remove_pending(wqe, work, prev); + io_wqe_remove_pending(wqe, work, prev, acct); raw_spin_unlock(&acct->lock); io_run_cancel(work, wqe); match->nr_pending++; @@ -1180,17 +1258,35 @@ static bool io_acct_cancel_pending_work(struct io_wqe *wqe, static void io_wqe_cancel_pending_work(struct io_wqe *wqe, struct io_cb_cancel_data *match) { - int i; -retry: - for (i = 0; i < IO_WQ_ACCT_NR; i++) { - struct io_wqe_acct *acct = io_get_acct(wqe, i == 0, false); + int i, j; + struct io_wqe_acct *acct, *iw_acct; +retry_public: + for (i = 0; i < IO_WQ_ACCT_NR; i++) { + acct = io_get_acct(wqe, i == 0, false); if (io_acct_cancel_pending_work(wqe, acct, match)) { if (match->cancel_all) - goto retry; - break; + goto retry_public; + return; } } + +retry_private: + for (i = 0; i < IO_WQ_ACCT_NR; i++) { + acct = io_get_acct(wqe, i == 0, true); + raw_spin_lock(&acct->lock); + for (j = 0; j < acct->nr_fixed; j++) { + iw_acct = &acct->fixed_workers[j]->acct; + if (io_acct_cancel_pending_work(wqe, iw_acct, match)) { + if (match->cancel_all) { + raw_spin_unlock(&acct->lock); + goto retry_private; + } + break; + } + } + raw_spin_unlock(&acct->lock); + } } static void io_wqe_cancel_running_work(struct io_wqe *wqe, From patchwork Mon Jun 27 13:35:38 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Hao Xu X-Patchwork-Id: 12896654 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 BDD71C43334 for ; Mon, 27 Jun 2022 13:36:16 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S235949AbiF0NgP (ORCPT ); Mon, 27 Jun 2022 09:36:15 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:44240 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S234557AbiF0NgP (ORCPT ); Mon, 27 Jun 2022 09:36:15 -0400 Received: from out0.migadu.com (out0.migadu.com [IPv6:2001:41d0:2:267::]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id 8FF246362 for ; Mon, 27 Jun 2022 06:36:14 -0700 (PDT) X-Report-Abuse: Please report any abuse attempt to abuse@migadu.com and include these headers. DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=linux.dev; s=key1; t=1656336973; h=from:from:reply-to:subject:subject:date:date:message-id:message-id: to:to:cc:cc:mime-version:mime-version: content-transfer-encoding:content-transfer-encoding: in-reply-to:in-reply-to:references:references; bh=21GvqUh+eSxTwP7XE55aCv2mhTFoxlA4FnB4sAjQh+s=; b=BgPsTRcNrFcEgbyuW29P9KFbgM+a9001rA2WKXtw3rQP8PYr2qWvV5+XBO/VV2pMm9RgKe P6U7LMlWSlhnldOMGKnnKtldRinnnWJL6e09PZBqIkl+9Z8A7HPZj1HqzQQATV54ACHNFk Rz6+dsjpZhEQw7Wct/pUeQioR09+lyU= From: Hao Xu To: io-uring@vger.kernel.org Cc: Jens Axboe , Pavel Begunkov Subject: [PATCH 08/11] io-wq: batch the handling of fixed worker private works Date: Mon, 27 Jun 2022 21:35:38 +0800 Message-Id: <20220627133541.15223-9-hao.xu@linux.dev> In-Reply-To: <20220627133541.15223-1-hao.xu@linux.dev> References: <20220627133541.15223-1-hao.xu@linux.dev> MIME-Version: 1.0 X-Migadu-Flow: FLOW_OUT X-Migadu-Auth-User: linux.dev Precedence: bulk List-ID: X-Mailing-List: io-uring@vger.kernel.org From: Hao Xu Reduce acct->lock contension by batching the handling of private work list for fixed_workers. Signed-off-by: Hao Xu --- io_uring/io-wq.c | 42 +++++++++++++++++++++++++++++++++--------- io_uring/io-wq.h | 5 +++++ 2 files changed, 38 insertions(+), 9 deletions(-) diff --git a/io_uring/io-wq.c b/io_uring/io-wq.c index c87ba38f27b1..ce754c78ecac 100644 --- a/io_uring/io-wq.c +++ b/io_uring/io-wq.c @@ -545,8 +545,23 @@ static bool io_wait_on_hash(struct io_wqe *wqe, unsigned int hash) return ret; } +static inline void conditional_acct_lock(struct io_wqe_acct *acct, + bool needs_lock) +{ + if (needs_lock) + raw_spin_lock(&acct->lock); +} + +static inline void conditional_acct_unlock(struct io_wqe_acct *acct, + bool needs_lock) +{ + if (needs_lock) + raw_spin_unlock(&acct->lock); +} + static struct io_wq_work *io_get_next_work(struct io_wqe_acct *acct, - struct io_worker *worker) + struct io_worker *worker, + bool needs_lock) __must_hold(acct->lock) { struct io_wq_work_node *node, *prev; @@ -554,6 +569,7 @@ static struct io_wq_work *io_get_next_work(struct io_wqe_acct *acct, unsigned int stall_hash = -1U; struct io_wqe *wqe = worker->wqe; + conditional_acct_lock(acct, needs_lock); wq_list_for_each(node, prev, &acct->work_list) { unsigned int hash; @@ -562,6 +578,7 @@ static struct io_wq_work *io_get_next_work(struct io_wqe_acct *acct, /* not hashed, can run anytime */ if (!io_wq_is_hashed(work)) { wq_list_del(&acct->work_list, node, prev); + conditional_acct_unlock(acct, needs_lock); return work; } @@ -573,6 +590,7 @@ static struct io_wq_work *io_get_next_work(struct io_wqe_acct *acct, if (!test_and_set_bit(hash, &wqe->wq->hash->map)) { wqe->hash_tail[hash] = NULL; wq_list_cut(&acct->work_list, &tail->list, prev); + conditional_acct_unlock(acct, needs_lock); return work; } if (stall_hash == -1U) @@ -589,15 +607,16 @@ static struct io_wq_work *io_get_next_work(struct io_wqe_acct *acct, * work being added and clearing the stalled bit. */ set_bit(IO_ACCT_STALLED_BIT, &acct->flags); - raw_spin_unlock(&acct->lock); + conditional_acct_unlock(acct, needs_lock); unstalled = io_wait_on_hash(wqe, stall_hash); - raw_spin_lock(&acct->lock); + conditional_acct_lock(acct, needs_lock); if (unstalled) { clear_bit(IO_ACCT_STALLED_BIT, &acct->flags); if (wq_has_sleeper(&wqe->wq->hash->wait)) wake_up(&wqe->wq->hash->wait); } } + conditional_acct_unlock(acct, needs_lock); return NULL; } @@ -631,7 +650,7 @@ static void io_assign_current_work(struct io_worker *worker, static void io_wqe_enqueue(struct io_wqe *wqe, struct io_wq_work *work); static void io_worker_handle_work(struct io_worker *worker, - struct io_wqe_acct *acct) + struct io_wqe_acct *acct, bool needs_lock) { struct io_wqe *wqe = worker->wqe; struct io_wq *wq = wqe->wq; @@ -647,9 +666,7 @@ static void io_worker_handle_work(struct io_worker *worker, * can't make progress, any work completion or insertion will * clear the stalled flag. */ - raw_spin_lock(&acct->lock); - work = io_get_next_work(acct, worker); - raw_spin_unlock(&acct->lock); + work = io_get_next_work(acct, worker, needs_lock); if (work) { __io_worker_busy(wqe, worker); @@ -706,12 +723,19 @@ static void io_worker_handle_work(struct io_worker *worker, static inline void io_worker_handle_private_work(struct io_worker *worker) { - io_worker_handle_work(worker, &worker->acct); + struct io_wqe_acct acct; + + raw_spin_lock(&worker->acct.lock); + acct = worker->acct; + wq_list_clean(&worker->acct.work_list); + worker->acct.nr_works = 0; + raw_spin_unlock(&worker->acct.lock); + io_worker_handle_work(worker, &acct, false); } static inline void io_worker_handle_public_work(struct io_worker *worker) { - io_worker_handle_work(worker, io_wqe_get_acct(worker)); + io_worker_handle_work(worker, io_wqe_get_acct(worker), true); } static int io_wqe_worker(void *data) diff --git a/io_uring/io-wq.h b/io_uring/io-wq.h index 10b80ef78bb8..78efbb8c53f0 100644 --- a/io_uring/io-wq.h +++ b/io_uring/io-wq.h @@ -32,6 +32,11 @@ enum io_wq_cancel { (list)->first = NULL; \ } while (0) +static inline void wq_list_clean(struct io_wq_work_list *list) +{ + list->first = list->last = NULL; +} + static inline void wq_list_add_after(struct io_wq_work_node *node, struct io_wq_work_node *pos, struct io_wq_work_list *list) From patchwork Mon Jun 27 13:35:39 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Hao Xu X-Patchwork-Id: 12896655 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 6C049C43334 for ; Mon, 27 Jun 2022 13:36:24 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S235933AbiF0NgW (ORCPT ); Mon, 27 Jun 2022 09:36:22 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:44260 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S233017AbiF0NgT (ORCPT ); Mon, 27 Jun 2022 09:36:19 -0400 Received: from out0.migadu.com (out0.migadu.com [94.23.1.103]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id 3F0416399 for ; Mon, 27 Jun 2022 06:36:18 -0700 (PDT) X-Report-Abuse: Please report any abuse attempt to abuse@migadu.com and include these headers. DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=linux.dev; s=key1; t=1656336976; h=from:from:reply-to:subject:subject:date:date:message-id:message-id: to:to:cc:cc:mime-version:mime-version: content-transfer-encoding:content-transfer-encoding: in-reply-to:in-reply-to:references:references; bh=AMJzuZSGA5zOVeL0+SHdCg5RQuGZG9/gdjYNzyrAZ6g=; b=bQ30DJtbnKJL2gUsytXfTjebVs7rrhPWXZ02aOsuh74uQJE8BkSKZuBBcsoovZMMAMk6pk n50JNAjmUfEQBGvPA+C+0g6QWpleKrjVqB3NdrZ9VO1A4AXKJarC2e1MQ628GYd8iA+Fd1 6dkU+ZSr/er3g3lke4Po83fiMrXLomk= From: Hao Xu To: io-uring@vger.kernel.org Cc: Jens Axboe , Pavel Begunkov Subject: [PATCH 09/11] io_uring: add register fixed worker interface Date: Mon, 27 Jun 2022 21:35:39 +0800 Message-Id: <20220627133541.15223-10-hao.xu@linux.dev> In-Reply-To: <20220627133541.15223-1-hao.xu@linux.dev> References: <20220627133541.15223-1-hao.xu@linux.dev> MIME-Version: 1.0 X-Migadu-Flow: FLOW_OUT X-Migadu-Auth-User: linux.dev Precedence: bulk List-ID: X-Mailing-List: io-uring@vger.kernel.org From: Hao Xu Add an io_uring_register() interface to register fixed workers and indicate its work capacity. The argument is an array of two elements each is struct { __s32 nr_workers; __s32 max_works; } (nr_workers, max_works) meaning nr_workers or max_works < -1 invalid nr_workers or max_works == -1 get the old value back nr_workers or max_works >= 0 get the old value and set to the value Signed-off-by: Hao Xu --- include/uapi/linux/io_uring.h | 11 ++++ io_uring/io-wq.c | 101 ++++++++++++++++++++++++++++++++++ io_uring/io-wq.h | 3 + io_uring/io_uring.c | 71 ++++++++++++++++++++++++ 4 files changed, 186 insertions(+) diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h index 8715f0942ec2..5480829d07c0 100644 --- a/include/uapi/linux/io_uring.h +++ b/include/uapi/linux/io_uring.h @@ -423,6 +423,12 @@ enum { IORING_REGISTER_PBUF_RING = 22, IORING_UNREGISTER_PBUF_RING = 23, + /* set number of fixed workers and number + * of works in a private work list which + * belongs to a fixed worker + */ + IORING_REGISTER_IOWQ_FIXED_WORKERS = 24, + /* this goes last */ IORING_REGISTER_LAST }; @@ -558,4 +564,9 @@ struct io_uring_getevents_arg { __u64 ts; }; +struct io_uring_fixed_worker_arg { + __s32 nr_workers; + __s32 max_works; +}; + #endif diff --git a/io_uring/io-wq.c b/io_uring/io-wq.c index ce754c78ecac..d54056b98e2b 100644 --- a/io_uring/io-wq.c +++ b/io_uring/io-wq.c @@ -1676,6 +1676,107 @@ int io_wq_max_workers(struct io_wq *wq, int *new_count) return 0; } +/* + * Set max number of fixed workers and the capacity of private work list, + * returns old value. If new_count is -1, then just return the old value. + */ +int io_wq_fixed_workers(struct io_wq *wq, + struct io_uring_fixed_worker_arg *new_count) +{ + struct io_uring_fixed_worker_arg prev[IO_WQ_ACCT_NR]; + bool first_node = true; + int i, node; + bool readonly[2] = { + (new_count[0].nr_workers == -1 && new_count[0].max_works == -1), + (new_count[1].nr_workers == -1 && new_count[1].max_works == -1), + }; + + BUILD_BUG_ON((int) IO_WQ_ACCT_BOUND != (int) IO_WQ_BOUND); + BUILD_BUG_ON((int) IO_WQ_ACCT_UNBOUND != (int) IO_WQ_UNBOUND); + BUILD_BUG_ON((int) IO_WQ_ACCT_NR != 2); + + for (i = 0; i < IO_WQ_ACCT_NR; i++) { + if (new_count[i].nr_workers > task_rlimit(current, RLIMIT_NPROC)) + new_count[i].nr_workers = + task_rlimit(current, RLIMIT_NPROC); + } + + rcu_read_lock(); + for_each_node(node) { + int j; + struct io_wqe *wqe = wq->wqes[node]; + + for (i = 0; i < IO_WQ_ACCT_NR; i++) { + struct io_wqe_acct *acct = &wqe->fixed_acct[i]; + int *nr_fixed, *max_works; + struct io_worker **fixed_workers; + int nr = new_count[i].nr_workers; + + raw_spin_lock(&acct->lock); + nr_fixed = &acct->nr_fixed; + max_works = &acct->max_works; + fixed_workers = acct->fixed_workers; + if (first_node) { + prev[i].nr_workers = *nr_fixed; + prev[i].max_works = *max_works; + } + if (readonly[i]) { + raw_spin_unlock(&acct->lock); + continue; + } + if (*nr_fixed == nr || nr == -1) { + *max_works = new_count[i].max_works; + raw_spin_unlock(&acct->lock); + continue; + } + for (j = 0; j < *nr_fixed; j++) { + struct io_worker *worker = fixed_workers[j]; + + if (!worker) + continue; + worker->flags |= IO_WORKER_F_EXIT; + /* + * Mark index to -1 to avoid false deletion + * in io_fixed_worker_exit() + */ + worker->index = -1; + /* + * Once a worker is in fixed_workers array + * it is definitely there before we release + * the acct->lock below. That's why we don't + * need to increment the worker->ref here. + */ + wake_up_process(worker->task); + } + kfree(fixed_workers); + acct->fixed_workers = NULL; + *nr_fixed = 0; + *max_works = new_count[i].max_works; + acct->fixed_workers = kzalloc_node( + sizeof(*fixed_workers) * nr, + GFP_KERNEL, wqe->node); + if (!acct->fixed_workers) { + raw_spin_unlock(&acct->lock); + return -ENOMEM; + } + raw_spin_unlock(&acct->lock); + for (j = 0; j < nr; j++) + io_wqe_create_worker(wqe, acct); + + acct->fixed_worker_registered = !!nr; + } + first_node = false; + } + rcu_read_unlock(); + + for (i = 0; i < IO_WQ_ACCT_NR; i++) { + new_count[i].nr_workers = prev[i].nr_workers; + new_count[i].max_works = prev[i].max_works; + } + + return 0; +} + static __init int io_wq_init(void) { int ret; diff --git a/io_uring/io-wq.h b/io_uring/io-wq.h index 78efbb8c53f0..fbbe13d75595 100644 --- a/io_uring/io-wq.h +++ b/io_uring/io-wq.h @@ -3,6 +3,7 @@ #include #include +#include struct io_wq; @@ -188,6 +189,8 @@ void io_wq_hash_work(struct io_wq_work *work, void *val); int io_wq_cpu_affinity(struct io_wq *wq, cpumask_var_t mask); int io_wq_max_workers(struct io_wq *wq, int *new_count); +int io_wq_fixed_workers(struct io_wq *wq, + struct io_uring_fixed_worker_arg *new_count); static inline bool io_wq_is_hashed(struct io_wq_work *work) { diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c index afda42246d12..637c5a50c97f 100644 --- a/io_uring/io_uring.c +++ b/io_uring/io_uring.c @@ -3764,6 +3764,71 @@ static __cold int io_register_iowq_max_workers(struct io_ring_ctx *ctx, return ret; } +static __cold int io_register_iowq_fixed_workers(struct io_ring_ctx *ctx, + void __user *arg) + __must_hold(&ctx->uring_lock) +{ + struct io_uring_task *tctx = NULL; + struct io_sq_data *sqd = NULL; + struct io_uring_fixed_worker_arg new_count[2]; + int i, ret; + + if (copy_from_user(new_count, arg, sizeof(new_count))) + return -EFAULT; + for (i = 0; i < ARRAY_SIZE(new_count); i++) { + int nr_workers = new_count[i].nr_workers; + int max_works = new_count[i].max_works; + + if (nr_workers < -1 || max_works < -1) + return -EINVAL; + } + + if (ctx->flags & IORING_SETUP_SQPOLL) { + sqd = ctx->sq_data; + if (sqd) { + /* + * Observe the correct sqd->lock -> ctx->uring_lock + * ordering. Fine to drop uring_lock here, we hold + * a ref to the ctx. + */ + refcount_inc(&sqd->refs); + mutex_unlock(&ctx->uring_lock); + mutex_lock(&sqd->lock); + mutex_lock(&ctx->uring_lock); + if (sqd->thread) + tctx = sqd->thread->io_uring; + } + } else { + tctx = current->io_uring; + } + + if (tctx && tctx->io_wq) { + ret = io_wq_fixed_workers(tctx->io_wq, new_count); + if (ret) + goto err; + } else { + memset(new_count, -1, sizeof(new_count)); + } + + if (sqd) { + mutex_unlock(&sqd->lock); + io_put_sq_data(sqd); + } + + if (copy_to_user(arg, new_count, sizeof(new_count))) + return -EFAULT; + + /* that's it for SQPOLL, only the SQPOLL task creates requests */ + if (sqd) + return 0; + +err: + if (sqd) { + mutex_unlock(&sqd->lock); + io_put_sq_data(sqd); + } + return ret; +} static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, void __user *arg, unsigned nr_args) __releases(ctx->uring_lock) @@ -3910,6 +3975,12 @@ static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode, break; ret = io_unregister_pbuf_ring(ctx, arg); break; + case IORING_REGISTER_IOWQ_FIXED_WORKERS: + ret = -EINVAL; + if (!arg || nr_args != 2) + break; + ret = io_register_iowq_fixed_workers(ctx, arg); + break; default: ret = -EINVAL; break; From patchwork Mon Jun 27 13:35:40 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Hao Xu X-Patchwork-Id: 12896656 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 C9DBFC433EF for ; Mon, 27 Jun 2022 13:36:24 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S233017AbiF0NgX (ORCPT ); Mon, 27 Jun 2022 09:36:23 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:44294 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S235941AbiF0NgX (ORCPT ); Mon, 27 Jun 2022 09:36:23 -0400 Received: from out0.migadu.com (out0.migadu.com [IPv6:2001:41d0:2:267::]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id 3910863DD for ; Mon, 27 Jun 2022 06:36:22 -0700 (PDT) X-Report-Abuse: Please report any abuse attempt to abuse@migadu.com and include these headers. DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=linux.dev; s=key1; t=1656336980; h=from:from:reply-to:subject:subject:date:date:message-id:message-id: to:to:cc:cc:mime-version:mime-version: content-transfer-encoding:content-transfer-encoding: in-reply-to:in-reply-to:references:references; bh=AzQ2x8u9+8abs2ZXi1QKt0WNLkPxccTGlwQkpOrihIw=; b=ftTeICYQuJpQf+sYCtrw1qwRHw2SzRyPqxgrjCr8+1/LAvhJWJzkU40sNZ05ja6N4aEK9p CP2vArJzRqkl1c+lo7qfma+Zp70LpaUZmd5P0OEDi7u7HtAtaqMmhuzMdZe65pquB7BePx EcFlwnZ0tmbNBbXC3j3DpJ7ra0Cl0LQ= From: Hao Xu To: io-uring@vger.kernel.org Cc: Jens Axboe , Pavel Begunkov Subject: [PATCH 10/11] io-wq: add an work list for fixed worker Date: Mon, 27 Jun 2022 21:35:40 +0800 Message-Id: <20220627133541.15223-11-hao.xu@linux.dev> In-Reply-To: <20220627133541.15223-1-hao.xu@linux.dev> References: <20220627133541.15223-1-hao.xu@linux.dev> MIME-Version: 1.0 X-Migadu-Flow: FLOW_OUT X-Migadu-Auth-User: linux.dev Precedence: bulk List-ID: X-Mailing-List: io-uring@vger.kernel.org From: Hao Xu Previously when a fixed worker handles its private works, it get all of them from worker->acct.work_list to a temporary acct->work_list. This prevents work cancellation since the cancellation process cannot find works from this temporary work list. Thus add a new acct so to address it. Signed-off-by: Hao Xu --- io_uring/io-wq.c | 20 +++++++++++++------- 1 file changed, 13 insertions(+), 7 deletions(-) diff --git a/io_uring/io-wq.c b/io_uring/io-wq.c index d54056b98e2b..38d88da70a40 100644 --- a/io_uring/io-wq.c +++ b/io_uring/io-wq.c @@ -126,6 +126,7 @@ struct io_worker { }; int index; struct io_wqe_acct acct; + struct io_wqe_acct exec_acct; }; #if BITS_PER_LONG == 64 @@ -723,14 +724,17 @@ static void io_worker_handle_work(struct io_worker *worker, static inline void io_worker_handle_private_work(struct io_worker *worker) { - struct io_wqe_acct acct; + struct io_wqe_acct *acct = &worker->acct; + struct io_wqe_acct *exec_acct = &worker->exec_acct; - raw_spin_lock(&worker->acct.lock); - acct = worker->acct; - wq_list_clean(&worker->acct.work_list); - worker->acct.nr_works = 0; - raw_spin_unlock(&worker->acct.lock); - io_worker_handle_work(worker, &acct, false); + raw_spin_lock(&acct->lock); + exec_acct->nr_works = acct->nr_works; + exec_acct->max_works = acct->max_works; + exec_acct->work_list = acct->work_list; + wq_list_clean(&acct->work_list); + acct->nr_works = 0; + raw_spin_unlock(&acct->lock); + io_worker_handle_work(worker, exec_acct, false); } static inline void io_worker_handle_public_work(struct io_worker *worker) @@ -866,6 +870,7 @@ static void io_init_new_fixed_worker(struct io_wqe *wqe, { struct io_wqe_acct *acct = io_wqe_get_acct(worker); struct io_wqe_acct *iw_acct = &worker->acct; + struct io_wqe_acct *exec_acct = &worker->exec_acct; unsigned index = acct->index; unsigned *nr_fixed; @@ -878,6 +883,7 @@ static void io_init_new_fixed_worker(struct io_wqe *wqe, iw_acct->index = index; INIT_WQ_LIST(&iw_acct->work_list); raw_spin_lock_init(&iw_acct->lock); + raw_spin_lock_init(&exec_acct->lock); raw_spin_unlock(&acct->lock); } From patchwork Mon Jun 27 13:35:41 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Hao Xu X-Patchwork-Id: 12896657 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 BE232C433EF for ; Mon, 27 Jun 2022 13:36:27 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S234557AbiF0Ng1 (ORCPT ); Mon, 27 Jun 2022 09:36:27 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:44330 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S235941AbiF0NgZ (ORCPT ); Mon, 27 Jun 2022 09:36:25 -0400 Received: from out0.migadu.com (out0.migadu.com [IPv6:2001:41d0:2:267::]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id 3374A643A for ; Mon, 27 Jun 2022 06:36:25 -0700 (PDT) X-Report-Abuse: Please report any abuse attempt to abuse@migadu.com and include these headers. DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=linux.dev; s=key1; t=1656336983; h=from:from:reply-to:subject:subject:date:date:message-id:message-id: to:to:cc:cc:mime-version:mime-version: content-transfer-encoding:content-transfer-encoding: in-reply-to:in-reply-to:references:references; bh=4rzKJ453Ym13+cBpJt6sKAiNar/paaU20aGoTw9WDQs=; b=lv4E6YksKcduIemt1A1711YcQ/6VbuKZpMJbSt2raP32qHU8CIcih0b745eXTmMSLH2GjQ 6AKqC3bI9oAhVzhZcq16ZtPNEjDCRFOh7HoMr8K6foXHIVkqvEi+xpUBn+ja2KqovChInq hfBJ5PFaSr3dNyko8EcEf6Vf0kD6KRM= From: Hao Xu To: io-uring@vger.kernel.org Cc: Jens Axboe , Pavel Begunkov Subject: [PATCH 11/11] io_uring: cancel works in exec work list for fixed worker Date: Mon, 27 Jun 2022 21:35:41 +0800 Message-Id: <20220627133541.15223-12-hao.xu@linux.dev> In-Reply-To: <20220627133541.15223-1-hao.xu@linux.dev> References: <20220627133541.15223-1-hao.xu@linux.dev> MIME-Version: 1.0 X-Migadu-Flow: FLOW_OUT X-Migadu-Auth-User: linux.dev Precedence: bulk List-ID: X-Mailing-List: io-uring@vger.kernel.org From: Hao Xu When users want to cancel a request, look into the exec work list of fixed worker as well. It's not sane to ignore it. Signed-off-by: Hao Xu --- io_uring/io-wq.c | 44 +++++++++++++++++++++++++++++++++++--------- 1 file changed, 35 insertions(+), 9 deletions(-) diff --git a/io_uring/io-wq.c b/io_uring/io-wq.c index 38d88da70a40..9e8ad7455373 100644 --- a/io_uring/io-wq.c +++ b/io_uring/io-wq.c @@ -1285,32 +1285,44 @@ static bool io_acct_cancel_pending_work(struct io_wqe *wqe, return false; } -static void io_wqe_cancel_pending_work(struct io_wqe *wqe, - struct io_cb_cancel_data *match) +static void io_wqe_cancel_pending_work_normal(struct io_wqe *wqe, + struct io_cb_cancel_data *match) { - int i, j; - struct io_wqe_acct *acct, *iw_acct; + int i; + struct io_wqe_acct *acct; -retry_public: +retry: for (i = 0; i < IO_WQ_ACCT_NR; i++) { acct = io_get_acct(wqe, i == 0, false); if (io_acct_cancel_pending_work(wqe, acct, match)) { if (match->cancel_all) - goto retry_public; + goto retry; return; } } +} + +static void __io_wqe_cancel_pending_work_fixed(struct io_wqe *wqe, + struct io_cb_cancel_data *match, + bool exec) +{ + int i, j; + struct io_wqe_acct *acct, *iw_acct; -retry_private: +retry: for (i = 0; i < IO_WQ_ACCT_NR; i++) { acct = io_get_acct(wqe, i == 0, true); raw_spin_lock(&acct->lock); for (j = 0; j < acct->nr_fixed; j++) { - iw_acct = &acct->fixed_workers[j]->acct; + if (exec) + iw_acct = &acct->fixed_workers[j]->acct; + else + iw_acct = &acct->fixed_workers[j]->exec_acct; + if (io_acct_cancel_pending_work(wqe, iw_acct, match)) { if (match->cancel_all) { raw_spin_unlock(&acct->lock); - goto retry_private; + goto retry; } break; } @@ -1319,6 +1331,20 @@ static void io_wqe_cancel_pending_work(struct io_wqe *wqe, } } +static void io_wqe_cancel_pending_work_fixed(struct io_wqe *wqe, + struct io_cb_cancel_data *match) +{ + __io_wqe_cancel_pending_work_fixed(wqe, match, false); + __io_wqe_cancel_pending_work_fixed(wqe, match, true); +} + +static void io_wqe_cancel_pending_work(struct io_wqe *wqe, + struct io_cb_cancel_data *match) +{ + io_wqe_cancel_pending_work_normal(wqe, match); + io_wqe_cancel_pending_work_fixed(wqe, match); +} + static void io_wqe_cancel_running_work(struct io_wqe *wqe, struct io_cb_cancel_data *match) {