From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-5.2 required=3.0 tests=HEADER_FROM_DIFFERENT_DOMAINS, INCLUDES_PATCH,MAILING_LIST_MULTI,SPF_HELO_NONE,SPF_PASS,UNPARSEABLE_RELAY, USER_AGENT_SANE_1 autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id 8179BC433E0 for ; Wed, 8 Jul 2020 16:51:49 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by mail.kernel.org (Postfix) with ESMTP id 5F6CB2063A for ; Wed, 8 Jul 2020 16:51:49 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1730452AbgGHQvt (ORCPT ); Wed, 8 Jul 2020 12:51:49 -0400 Received: from out30-44.freemail.mail.aliyun.com ([115.124.30.44]:56392 "EHLO out30-44.freemail.mail.aliyun.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1730349AbgGHQvs (ORCPT ); Wed, 8 Jul 2020 12:51:48 -0400 X-Alimail-AntiSpam: AC=PASS;BC=-1|-1;BR=01201311R121e4;CH=green;DM=||false|;DS=||;FP=0|-1|-1|-1|0|-1|-1|-1;HT=e01e07425;MF=xiaoguang.wang@linux.alibaba.com;NM=1;PH=DS;RN=3;SR=0;TI=SMTPD_---0U27slkj_1594227104; Received: from 30.39.158.199(mailfrom:xiaoguang.wang@linux.alibaba.com fp:SMTPD_---0U27slkj_1594227104) by smtp.aliyun-inc.com(127.0.0.1); Thu, 09 Jul 2020 00:51:44 +0800 Subject: Re: [PATCH] io_uring: export cq overflow status to userspace To: Jens Axboe Cc: io-uring@vger.kernel.org, joseph.qi@linux.alibaba.com References: <83a3a0eb-8dea-20ad-ffc5-619226b47998@linux.alibaba.com> <54ce9903-4016-5b30-2fe9-397da9161bfe@linux.alibaba.com> <6c770628-34bd-f75a-3d4a-c1810f652054@kernel.dk> From: Xiaoguang Wang Message-ID: Date: Thu, 9 Jul 2020 00:51:44 +0800 User-Agent: Mozilla/5.0 (Windows NT 10.0; WOW64; rv:68.0) Gecko/20100101 Thunderbird/68.10.0 MIME-Version: 1.0 In-Reply-To: <6c770628-34bd-f75a-3d4a-c1810f652054@kernel.dk> Content-Type: text/plain; charset=utf-8; format=flowed Content-Transfer-Encoding: 7bit Sender: io-uring-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: io-uring@vger.kernel.org hi, > On 7/8/20 9:39 AM, Xiaoguang Wang wrote: >> hi, >> >>> On 7/7/20 11:29 PM, Xiaoguang Wang wrote: >>>> I modify above test program a bit: >>>> #include >>>> #include >>>> #include >>>> #include >>>> #include >>>> #include >>>> #include >>>> >>>> #include "liburing.h" >>>> >>>> static void test_cq_overflow(struct io_uring *ring) >>>> { >>>> struct io_uring_cqe *cqe; >>>> struct io_uring_sqe *sqe; >>>> int issued = 0; >>>> int ret = 0; >>>> int i; >>>> >>>> for (i = 0; i < 33; i++) { >>>> sqe = io_uring_get_sqe(ring); >>>> if (!sqe) { >>>> fprintf(stderr, "get sqe failed\n"); >>>> break;; >>>> } >>>> ret = io_uring_submit(ring); >>>> if (ret <= 0) { >>>> if (ret != -EBUSY) >>>> fprintf(stderr, "sqe submit failed: %d\n", ret); >>>> break; >>>> } >>>> issued++; >>>> } >>>> >>>> printf("issued requests: %d\n", issued); >>>> >>>> while (issued) { >>>> ret = io_uring_peek_cqe(ring, &cqe); >>>> if (ret) { >>>> if (ret != -EAGAIN) { >>>> fprintf(stderr, "peek completion failed: %s\n", >>>> strerror(ret)); >>>> break; >>>> } >>>> printf("left requets: %d %d\n", issued, IO_URING_READ_ONCE(*ring->sq.kflags)); >>>> continue; >>>> } >>>> io_uring_cqe_seen(ring, cqe); >>>> issued--; >>>> printf("left requets: %d\n", issued); >>>> } >>>> } >>>> >>>> int main(int argc, char *argv[]) >>>> { >>>> int ret; >>>> struct io_uring ring; >>>> >>>> ret = io_uring_queue_init(16, &ring, 0); >>>> if (ret) { >>>> fprintf(stderr, "ring setup failed: %d\n", ret); >>>> return 1; >>>> } >>>> >>>> test_cq_overflow(&ring); >>>> return 0; >>>> } >>>> >>>> Though with your patches applied, we still can not peek the last cqe. >>>> This test program will only issue 33 sqes, so it won't get EBUSY error. >>> >>> How about we make this even simpler, then - make the >>> IORING_SQ_CQ_OVERFLOW actually track the state, rather than when we fail >>> on submission. The liburing change would be the same, the kernel side >>> would then look like the below. >>> >>> >>> diff --git a/fs/io_uring.c b/fs/io_uring.c >>> index 4c9a494c9f9f..01981926cdf4 100644 >>> --- a/fs/io_uring.c >>> +++ b/fs/io_uring.c >>> @@ -1342,6 +1342,7 @@ static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) >>> if (cqe) { >>> clear_bit(0, &ctx->sq_check_overflow); >>> clear_bit(0, &ctx->cq_check_overflow); >>> + ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW; >>> } >>> spin_unlock_irqrestore(&ctx->completion_lock, flags); >>> io_cqring_ev_posted(ctx); >>> @@ -1379,6 +1380,7 @@ static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags) >>> if (list_empty(&ctx->cq_overflow_list)) { >>> set_bit(0, &ctx->sq_check_overflow); >>> set_bit(0, &ctx->cq_check_overflow); >>> + ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW; Some callers to __io_cqring_fill_event() don't hold completion_lock, for example: ==> io_iopoll_complete ====> __io_cqring_fill_event() So this patch maybe still not safe when SQPOLL is enabled. Do you perfer adding a new lock or just do completion_lock here only when cq ring is overflowed? Regards, Xiaoguang Wang >>> } >>> req->flags |= REQ_F_OVERFLOW; >>> refcount_inc(&req->refs); >>> @@ -6375,9 +6377,9 @@ static int io_sq_thread(void *data) >>> } >>> >>> /* Tell userspace we may need a wakeup call */ >>> + spin_lock_irq(&ctx->completion_lock); >>> ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; >>> - /* make sure to read SQ tail after writing flags */ >>> - smp_mb(); >>> + spin_unlock_irq(&ctx->completion_lock); >>> >>> to_submit = io_sqring_entries(ctx); >>> if (!to_submit || ret == -EBUSY) { >>> @@ -6400,7 +6402,9 @@ static int io_sq_thread(void *data) >>> } >>> finish_wait(&ctx->sqo_wait, &wait); >>> >>> + spin_lock_irq(&ctx->completion_lock); >>> ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; >>> + spin_unlock_irq(&ctx->completion_lock); >>> } >>> >>> mutex_lock(&ctx->uring_lock); >>> diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h >>> index 8d033961cb78..91953b543125 100644 >>> --- a/include/uapi/linux/io_uring.h >>> +++ b/include/uapi/linux/io_uring.h >>> @@ -198,6 +198,7 @@ struct io_sqring_offsets { >>> * sq_ring->flags >>> */ >>> #define IORING_SQ_NEED_WAKEUP (1U << 0) /* needs io_uring_enter wakeup */ >>> +#define IORING_SQ_CQ_OVERFLOW (1U << 1) /* app needs to enter kernel */ >>> >>> struct io_cqring_offsets { >>> __u32 head; >>> >> Looks good, I think it'll work now. >> I'll test and send patches soon, thanks. > > One missing bit clear, and I corrected that last comment. Just base on > this one, thanks! > > > diff --git a/fs/io_uring.c b/fs/io_uring.c > index 4c9a494c9f9f..0b6a4b2d7e76 100644 > --- a/fs/io_uring.c > +++ b/fs/io_uring.c > @@ -1342,6 +1342,7 @@ static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force) > if (cqe) { > clear_bit(0, &ctx->sq_check_overflow); > clear_bit(0, &ctx->cq_check_overflow); > + ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW; > } > spin_unlock_irqrestore(&ctx->completion_lock, flags); > io_cqring_ev_posted(ctx); > @@ -1379,6 +1380,7 @@ static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags) > if (list_empty(&ctx->cq_overflow_list)) { > set_bit(0, &ctx->sq_check_overflow); > set_bit(0, &ctx->cq_check_overflow); > + ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW; > } > req->flags |= REQ_F_OVERFLOW; > refcount_inc(&req->refs); > @@ -6375,9 +6377,9 @@ static int io_sq_thread(void *data) > } > > /* Tell userspace we may need a wakeup call */ > + spin_lock_irq(&ctx->completion_lock); > ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP; > - /* make sure to read SQ tail after writing flags */ > - smp_mb(); > + spin_unlock_irq(&ctx->completion_lock); > > to_submit = io_sqring_entries(ctx); > if (!to_submit || ret == -EBUSY) { > @@ -6400,7 +6402,9 @@ static int io_sq_thread(void *data) > } > finish_wait(&ctx->sqo_wait, &wait); > > + spin_lock_irq(&ctx->completion_lock); > ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP; > + spin_unlock_irq(&ctx->completion_lock); > } > > mutex_lock(&ctx->uring_lock); > @@ -7810,6 +7814,7 @@ static void io_uring_cancel_files(struct io_ring_ctx *ctx, > if (list_empty(&ctx->cq_overflow_list)) { > clear_bit(0, &ctx->sq_check_overflow); > clear_bit(0, &ctx->cq_check_overflow); > + ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW; > } > spin_unlock_irq(&ctx->completion_lock); > > diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h > index 8d033961cb78..d65fde732518 100644 > --- a/include/uapi/linux/io_uring.h > +++ b/include/uapi/linux/io_uring.h > @@ -198,6 +198,7 @@ struct io_sqring_offsets { > * sq_ring->flags > */ > #define IORING_SQ_NEED_WAKEUP (1U << 0) /* needs io_uring_enter wakeup */ > +#define IORING_SQ_CQ_OVERFLOW (1U << 1) /* CQ ring is overflown */ > > struct io_cqring_offsets { > __u32 head; >