From patchwork Mon Nov 8 11:49:49 2021 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Xuan Zhuo X-Patchwork-Id: 12608315 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 mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id 342ECC433FE for ; Mon, 8 Nov 2021 11:49:58 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by mail.kernel.org (Postfix) with ESMTP id 1558361355 for ; Mon, 8 Nov 2021 11:49:58 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S239333AbhKHLwl (ORCPT ); Mon, 8 Nov 2021 06:52:41 -0500 Received: from out30-45.freemail.mail.aliyun.com ([115.124.30.45]:50327 "EHLO out30-45.freemail.mail.aliyun.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S239317AbhKHLwi (ORCPT ); Mon, 8 Nov 2021 06:52:38 -0500 X-Alimail-AntiSpam: AC=PASS;BC=-1|-1;BR=01201311R471e4;CH=green;DM=||false|;DS=||;FP=0|-1|-1|-1|0|-1|-1|-1;HT=e01e04407;MF=xuanzhuo@linux.alibaba.com;NM=1;PH=DS;RN=6;SR=0;TI=SMTPD_---0Uvb9Wct_1636372191; Received: from localhost(mailfrom:xuanzhuo@linux.alibaba.com fp:SMTPD_---0Uvb9Wct_1636372191) by smtp.aliyun-inc.com(127.0.0.1); Mon, 08 Nov 2021 19:49:51 +0800 From: Xuan Zhuo To: virtualization@lists.linux-foundation.org, netdev@vger.kernel.org Cc: "Michael S. Tsirkin" , Jason Wang , "David S. Miller" , Jakub Kicinski Subject: [PATCH v4 1/3] virtio: cache indirect desc for split Date: Mon, 8 Nov 2021 19:49:49 +0800 Message-Id: <20211108114951.92862-2-xuanzhuo@linux.alibaba.com> X-Mailer: git-send-email 2.31.0 In-Reply-To: <20211108114951.92862-1-xuanzhuo@linux.alibaba.com> References: <20211108114951.92862-1-xuanzhuo@linux.alibaba.com> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: netdev@vger.kernel.org In the case of using indirect, indirect desc must be allocated and released each time, which increases a lot of cpu overhead. Here, a cache is added for indirect. If the number of indirect desc to be applied for is less than desc_cache_thr, the desc array with the size of desc_cache_thr is fixed and cached for reuse. Signed-off-by: Xuan Zhuo --- drivers/virtio/virtio_ring.c | 126 ++++++++++++++++++++++++++++++++--- include/linux/virtio.h | 17 +++++ 2 files changed, 135 insertions(+), 8 deletions(-) diff --git a/drivers/virtio/virtio_ring.c b/drivers/virtio/virtio_ring.c index dd95dfd85e98..a4a91c497a83 100644 --- a/drivers/virtio/virtio_ring.c +++ b/drivers/virtio/virtio_ring.c @@ -85,6 +85,19 @@ struct vring_desc_extra { u16 next; /* The next desc state in a list. */ }; +struct vring_desc_cache { + /* desc cache chain */ + struct list_head list; + + void *array; + + /* desc cache threshold + * 0 - disable desc cache + * > 0 - enable desc cache. As the threshold of the desc cache. + */ + u32 threshold; +}; + struct vring_virtqueue { struct virtqueue vq; @@ -117,6 +130,8 @@ struct vring_virtqueue { /* Hint for event idx: already triggered no need to disable. */ bool event_triggered; + struct vring_desc_cache desc_cache; + union { /* Available for split ring */ struct { @@ -423,7 +438,50 @@ static unsigned int vring_unmap_one_split(const struct vring_virtqueue *vq, return extra[i].next; } -static struct vring_desc *alloc_indirect_split(struct virtqueue *_vq, +static void desc_cache_init(struct vring_virtqueue *vq) +{ + vq->desc_cache.array = NULL; + vq->desc_cache.threshold = 0; + INIT_LIST_HEAD(&vq->desc_cache.list); +} + +static void desc_cache_free(struct vring_virtqueue *vq) +{ + kfree(vq->desc_cache.array); +} + +static void __desc_cache_put(struct vring_virtqueue *vq, + struct list_head *node, int n) +{ + if (n <= vq->desc_cache.threshold) + list_add(node, &vq->desc_cache.list); + else + kfree(node); +} + +#define desc_cache_put(vq, desc, n) \ + __desc_cache_put(vq, (struct list_head *)desc, n) + +static void *desc_cache_get(struct vring_virtqueue *vq, + int size, int n, gfp_t gfp) +{ + struct list_head *node; + + if (n > vq->desc_cache.threshold) + return kmalloc_array(n, size, gfp); + + node = vq->desc_cache.list.next; + list_del(node); + return node; +} + +#define _desc_cache_get(vq, n, gfp, tp) \ + ((tp *)desc_cache_get(vq, (sizeof(tp)), n, gfp)) + +#define desc_cache_get_split(vq, n, gfp) \ + _desc_cache_get(vq, n, gfp, struct vring_desc) + +static struct vring_desc *alloc_indirect_split(struct vring_virtqueue *vq, unsigned int total_sg, gfp_t gfp) { @@ -437,12 +495,12 @@ static struct vring_desc *alloc_indirect_split(struct virtqueue *_vq, */ gfp &= ~__GFP_HIGHMEM; - desc = kmalloc_array(total_sg, sizeof(struct vring_desc), gfp); + desc = desc_cache_get_split(vq, total_sg, gfp); if (!desc) return NULL; for (i = 0; i < total_sg; i++) - desc[i].next = cpu_to_virtio16(_vq->vdev, i + 1); + desc[i].next = cpu_to_virtio16(vq->vq.vdev, i + 1); return desc; } @@ -508,7 +566,7 @@ static inline int virtqueue_add_split(struct virtqueue *_vq, head = vq->free_head; if (virtqueue_use_indirect(_vq, total_sg)) - desc = alloc_indirect_split(_vq, total_sg, gfp); + desc = alloc_indirect_split(vq, total_sg, gfp); else { desc = NULL; WARN_ON_ONCE(total_sg > vq->split.vring.num && !vq->indirect); @@ -652,7 +710,7 @@ static inline int virtqueue_add_split(struct virtqueue *_vq, } if (indirect) - kfree(desc); + desc_cache_put(vq, desc, total_sg); END_USE(vq); return -ENOMEM; @@ -717,7 +775,7 @@ static void detach_buf_split(struct vring_virtqueue *vq, unsigned int head, if (vq->indirect) { struct vring_desc *indir_desc = vq->split.desc_state[head].indir_desc; - u32 len; + u32 len, n; /* Free the indirect table, if any, now that it's unmapped. */ if (!indir_desc) @@ -729,10 +787,12 @@ static void detach_buf_split(struct vring_virtqueue *vq, unsigned int head, VRING_DESC_F_INDIRECT)); BUG_ON(len == 0 || len % sizeof(struct vring_desc)); - for (j = 0; j < len / sizeof(struct vring_desc); j++) + n = len / sizeof(struct vring_desc); + + for (j = 0; j < n; j++) vring_unmap_one_split_indirect(vq, &indir_desc[j]); - kfree(indir_desc); + desc_cache_put(vq, indir_desc, n); vq->split.desc_state[head].indir_desc = NULL; } else if (ctx) { *ctx = vq->split.desc_state[head].indir_desc; @@ -2200,6 +2260,8 @@ struct virtqueue *__vring_new_virtqueue(unsigned int index, !context; vq->event = virtio_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX); + desc_cache_init(vq); + if (virtio_has_feature(vdev, VIRTIO_F_ORDER_PLATFORM)) vq->weak_barriers = false; @@ -2329,6 +2391,7 @@ void vring_del_virtqueue(struct virtqueue *_vq) if (!vq->packed_ring) { kfree(vq->split.desc_state); kfree(vq->split.desc_extra); + desc_cache_free(vq); } kfree(vq); } @@ -2445,6 +2508,53 @@ dma_addr_t virtqueue_get_used_addr(struct virtqueue *_vq) } EXPORT_SYMBOL_GPL(virtqueue_get_used_addr); +int virtqueue_set_desc_cache(struct virtqueue *_vq, u32 threshold) +{ + struct vring_virtqueue *vq = to_vvq(_vq); + struct list_head *node; + int size, num, i; + + BUILD_BUG_ON(sizeof(struct list_head) > sizeof(struct vring_desc)); + BUILD_BUG_ON(sizeof(struct list_head) > sizeof(struct vring_packed_desc)); + + BUG_ON(!vq->we_own_ring); + + if (!vq->indirect) + return 0; + + vq->desc_cache.threshold = threshold; + + if (!threshold) + return 0; + + if (vq->packed_ring) { + size = sizeof(struct vring_packed_desc); + num = vq->packed.vring.num; + } else { + size = sizeof(struct vring_desc); + num = vq->split.vring.num; + } + + size = size * vq->desc_cache.threshold; + + vq->desc_cache.array = kmalloc_array(num, size, GFP_KERNEL); + if (!vq->desc_cache.array) { + vq->desc_cache.threshold = 0; + dev_warn(&vq->vq.vdev->dev, + "queue[%d] alloc desc cache fail. turn off it.\n", + vq->vq.index); + return -1; + } + + for (i = 0; i < num; ++i) { + node = vq->desc_cache.array + (i * size); + list_add(node, &vq->desc_cache.list); + } + + return 0; +} +EXPORT_SYMBOL_GPL(virtqueue_set_desc_cache); + /* Only available for split ring */ const struct vring *virtqueue_get_vring(struct virtqueue *vq) { diff --git a/include/linux/virtio.h b/include/linux/virtio.h index 41edbc01ffa4..e24b2e90dd42 100644 --- a/include/linux/virtio.h +++ b/include/linux/virtio.h @@ -89,6 +89,23 @@ dma_addr_t virtqueue_get_desc_addr(struct virtqueue *vq); dma_addr_t virtqueue_get_avail_addr(struct virtqueue *vq); dma_addr_t virtqueue_get_used_addr(struct virtqueue *vq); +/** + * virtqueue_set_desc_cache - set virtio ring desc cache threshold + * + * virtio will allocate ring.num desc arrays of size threshold in advance. If + * total_sg exceeds the threshold, use kmalloc/kfree allocation indirect desc, + * if total_sg is less than or equal to the threshold, use these pre-allocated + * desc arrays. + * + * This function must be called immediately after find_vqs and before device + * ready. + * + * @threshold: + * 0 - disable desc cache + * > 0 - enable desc cache. As the threshold of the desc cache. + */ +int virtqueue_set_desc_cache(struct virtqueue *_vq, u32 threshold); + /** * virtio_device - representation of a device using virtio * @index: unique position on the virtio bus From patchwork Mon Nov 8 11:49:50 2021 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Xuan Zhuo X-Patchwork-Id: 12608313 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 mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id AEB28C433EF for ; Mon, 8 Nov 2021 11:49:57 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by mail.kernel.org (Postfix) with ESMTP id 958D061288 for ; Mon, 8 Nov 2021 11:49:57 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S239335AbhKHLwk (ORCPT ); Mon, 8 Nov 2021 06:52:40 -0500 Received: from out30-42.freemail.mail.aliyun.com ([115.124.30.42]:37818 "EHLO out30-42.freemail.mail.aliyun.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S239330AbhKHLwi (ORCPT ); Mon, 8 Nov 2021 06:52:38 -0500 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=e01e04407;MF=xuanzhuo@linux.alibaba.com;NM=1;PH=DS;RN=6;SR=0;TI=SMTPD_---0Uvb-rcW_1636372192; Received: from localhost(mailfrom:xuanzhuo@linux.alibaba.com fp:SMTPD_---0Uvb-rcW_1636372192) by smtp.aliyun-inc.com(127.0.0.1); Mon, 08 Nov 2021 19:49:52 +0800 From: Xuan Zhuo To: virtualization@lists.linux-foundation.org, netdev@vger.kernel.org Cc: "Michael S. Tsirkin" , Jason Wang , "David S. Miller" , Jakub Kicinski Subject: [PATCH v4 2/3] virtio: cache indirect desc for packed Date: Mon, 8 Nov 2021 19:49:50 +0800 Message-Id: <20211108114951.92862-3-xuanzhuo@linux.alibaba.com> X-Mailer: git-send-email 2.31.0 In-Reply-To: <20211108114951.92862-1-xuanzhuo@linux.alibaba.com> References: <20211108114951.92862-1-xuanzhuo@linux.alibaba.com> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: netdev@vger.kernel.org In the case of using indirect, indirect desc must be allocated and released each time, which increases a lot of cpu overhead. Here, a cache is added for indirect. If the number of indirect desc to be applied for is less than desc_cache_thr, the desc array with the size of desc_cache_thr is fixed and cached for reuse. Signed-off-by: Xuan Zhuo --- drivers/virtio/virtio_ring.c | 28 ++++++++++++++++++---------- 1 file changed, 18 insertions(+), 10 deletions(-) diff --git a/drivers/virtio/virtio_ring.c b/drivers/virtio/virtio_ring.c index a4a91c497a83..76a974219ffd 100644 --- a/drivers/virtio/virtio_ring.c +++ b/drivers/virtio/virtio_ring.c @@ -1092,7 +1092,11 @@ static void vring_unmap_desc_packed(const struct vring_virtqueue *vq, } } -static struct vring_packed_desc *alloc_indirect_packed(unsigned int total_sg, +#define desc_cache_get_packed(vq, n, gfp) \ + _desc_cache_get(vq, n, gfp, struct vring_packed_desc) + +static struct vring_packed_desc *alloc_indirect_packed(struct vring_virtqueue *vq, + unsigned int total_sg, gfp_t gfp) { struct vring_packed_desc *desc; @@ -1104,7 +1108,7 @@ static struct vring_packed_desc *alloc_indirect_packed(unsigned int total_sg, */ gfp &= ~__GFP_HIGHMEM; - desc = kmalloc_array(total_sg, sizeof(struct vring_packed_desc), gfp); + desc = desc_cache_get_packed(vq, total_sg, gfp); return desc; } @@ -1124,7 +1128,7 @@ static int virtqueue_add_indirect_packed(struct vring_virtqueue *vq, dma_addr_t addr; head = vq->packed.next_avail_idx; - desc = alloc_indirect_packed(total_sg, gfp); + desc = alloc_indirect_packed(vq, total_sg, gfp); if (unlikely(vq->vq.num_free < 1)) { pr_debug("Can't add buf len 1 - avail = 0\n"); @@ -1215,7 +1219,7 @@ static int virtqueue_add_indirect_packed(struct vring_virtqueue *vq, for (i = 0; i < err_idx; i++) vring_unmap_desc_packed(vq, &desc[i]); - kfree(desc); + desc_cache_put(vq, desc, total_sg); END_USE(vq); return -ENOMEM; @@ -1440,20 +1444,22 @@ static void detach_buf_packed(struct vring_virtqueue *vq, } if (vq->indirect) { - u32 len; + u32 len, n; /* Free the indirect table, if any, now that it's unmapped. */ desc = state->indir_desc; if (!desc) return; + len = vq->packed.desc_extra[id].len; + n = len / sizeof(struct vring_packed_desc); + if (vq->use_dma_api) { - len = vq->packed.desc_extra[id].len; - for (i = 0; i < len / sizeof(struct vring_packed_desc); - i++) + for (i = 0; i < n; i++) vring_unmap_desc_packed(vq, &desc[i]); } - kfree(desc); + + desc_cache_put(vq, desc, n); state->indir_desc = NULL; } else if (ctx) { *ctx = state->indir_desc; @@ -1772,6 +1778,8 @@ static struct virtqueue *vring_create_virtqueue_packed( !context; vq->event = virtio_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX); + desc_cache_init(vq); + if (virtio_has_feature(vdev, VIRTIO_F_ORDER_PLATFORM)) vq->weak_barriers = false; @@ -2391,8 +2399,8 @@ void vring_del_virtqueue(struct virtqueue *_vq) if (!vq->packed_ring) { kfree(vq->split.desc_state); kfree(vq->split.desc_extra); - desc_cache_free(vq); } + desc_cache_free(vq); kfree(vq); } EXPORT_SYMBOL_GPL(vring_del_virtqueue); From patchwork Mon Nov 8 11:49:51 2021 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Xuan Zhuo X-Patchwork-Id: 12608317 X-Patchwork-Delegate: kuba@kernel.org 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 mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id 7DF67C433EF for ; Mon, 8 Nov 2021 11:49:59 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by mail.kernel.org (Postfix) with ESMTP id 61D6361181 for ; Mon, 8 Nov 2021 11:49:59 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S239336AbhKHLwl (ORCPT ); Mon, 8 Nov 2021 06:52:41 -0500 Received: from out30-43.freemail.mail.aliyun.com ([115.124.30.43]:34867 "EHLO out30-43.freemail.mail.aliyun.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S231401AbhKHLwj (ORCPT ); Mon, 8 Nov 2021 06:52:39 -0500 X-Alimail-AntiSpam: AC=PASS;BC=-1|-1;BR=01201311R551e4;CH=green;DM=||false|;DS=||;FP=0|-1|-1|-1|0|-1|-1|-1;HT=e01e01424;MF=xuanzhuo@linux.alibaba.com;NM=1;PH=DS;RN=6;SR=0;TI=SMTPD_---0Uvb9Wd2_1636372192; Received: from localhost(mailfrom:xuanzhuo@linux.alibaba.com fp:SMTPD_---0Uvb9Wd2_1636372192) by smtp.aliyun-inc.com(127.0.0.1); Mon, 08 Nov 2021 19:49:53 +0800 From: Xuan Zhuo To: virtualization@lists.linux-foundation.org, netdev@vger.kernel.org Cc: "Michael S. Tsirkin" , Jason Wang , "David S. Miller" , Jakub Kicinski Subject: [PATCH v4 3/3] virtio-net: enable virtio desc cache Date: Mon, 8 Nov 2021 19:49:51 +0800 Message-Id: <20211108114951.92862-4-xuanzhuo@linux.alibaba.com> X-Mailer: git-send-email 2.31.0 In-Reply-To: <20211108114951.92862-1-xuanzhuo@linux.alibaba.com> References: <20211108114951.92862-1-xuanzhuo@linux.alibaba.com> MIME-Version: 1.0 Precedence: bulk List-ID: X-Mailing-List: netdev@vger.kernel.org X-Patchwork-Delegate: kuba@kernel.org If the VIRTIO_RING_F_INDIRECT_DESC negotiation succeeds, and the number of sgs used for sending packets is greater than 1. We must constantly call __kmalloc/kfree to allocate/release desc. In the case of extremely fast package delivery, the overhead cannot be ignored: 27.46% [kernel] [k] virtqueue_add 16.66% [kernel] [k] detach_buf_split 16.51% [kernel] [k] virtnet_xsk_xmit 14.04% [kernel] [k] virtqueue_add_outbuf > 5.18% [kernel] [k] __kmalloc > 4.08% [kernel] [k] kfree 2.80% [kernel] [k] virtqueue_get_buf_ctx 2.22% [kernel] [k] xsk_tx_peek_desc 2.08% [kernel] [k] memset_erms 0.83% [kernel] [k] virtqueue_kick_prepare 0.76% [kernel] [k] virtnet_xsk_run 0.62% [kernel] [k] __free_old_xmit_ptr 0.60% [kernel] [k] vring_map_one_sg 0.53% [kernel] [k] native_apic_mem_write 0.46% [kernel] [k] sg_next 0.43% [kernel] [k] sg_init_table > 0.41% [kernel] [k] kmalloc_slab Compared to not using virtio indirect cache, virtio-net can get a 16% performance improvement when using virtio desc cache. In the test case, the CPU where the package is sent has reached 100%. The following are the PPS in two cases: indirect desc cache | no cache 3074658 | 2685132 3111866 | 2666118 3152527 | 2653632 3125867 | 2669820 3027147 | 2644464 3069211 | 2669777 3038522 | 2675645 3034507 | 2671302 3102257 | 2685504 3083712 | 2692800 3051771 | 2676928 3080684 | 2695040 3147816 | 2720876 3123887 | 2705492 3180963 | 2699520 3191579 | 2676480 3161670 | 2686272 3189768 | 2692588 3174272 | 2686692 3143434 | 2682416 Signed-off-by: Xuan Zhuo --- drivers/net/virtio_net.c | 12 +++++++++++- 1 file changed, 11 insertions(+), 1 deletion(-) diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c index 9ff2ef9dceca..193c8b38433e 100644 --- a/drivers/net/virtio_net.c +++ b/drivers/net/virtio_net.c @@ -42,6 +42,9 @@ module_param(csum, bool, 0444); module_param(gso, bool, 0444); module_param(napi_tx, bool, 0644); +static u32 virtio_desc_cache_threshold = MAX_SKB_FRAGS + 2; +module_param(virtio_desc_cache_threshold, uint, 0644); + /* FIXME: MTU in config. */ #define GOOD_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN) #define GOOD_COPY_LEN 128 @@ -3350,10 +3353,10 @@ static unsigned int mergeable_min_buf_len(struct virtnet_info *vi, struct virtqu static int virtnet_find_vqs(struct virtnet_info *vi) { + int i, total_vqs, threshold; vq_callback_t **callbacks; struct virtqueue **vqs; int ret = -ENOMEM; - int i, total_vqs; const char **names; bool *ctx; @@ -3411,10 +3414,17 @@ static int virtnet_find_vqs(struct virtnet_info *vi) vi->dev->features |= NETIF_F_HW_VLAN_CTAG_FILTER; } + threshold = min_t(u32, virtio_desc_cache_threshold, 2 + MAX_SKB_FRAGS); + for (i = 0; i < vi->max_queue_pairs; i++) { vi->rq[i].vq = vqs[rxq2vq(i)]; vi->rq[i].min_buf_len = mergeable_min_buf_len(vi, vi->rq[i].vq); vi->sq[i].vq = vqs[txq2vq(i)]; + + if (!vi->mergeable_rx_bufs && vi->big_packets) + virtqueue_set_desc_cache(vi->rq[i].vq, MAX_SKB_FRAGS + 2); + + virtqueue_set_desc_cache(vi->sq[i].vq, threshold); } /* run here: ret == 0. */