From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from dpdk.org (dpdk.org [92.243.14.124]) by inbox.dpdk.org (Postfix) with ESMTP id 65314A0613 for ; Tue, 24 Sep 2019 10:45:04 +0200 (CEST) Received: from [92.243.14.124] (localhost [127.0.0.1]) by dpdk.org (Postfix) with ESMTP id 304701BE90; Tue, 24 Sep 2019 10:44:49 +0200 (CEST) Received: from mga05.intel.com (mga05.intel.com [192.55.52.43]) by dpdk.org (Postfix) with ESMTP id 3C1952BF2 for ; Tue, 24 Sep 2019 10:44:43 +0200 (CEST) X-Amp-Result: SKIPPED(no attachment in message) X-Amp-File-Uploaded: False Received: from fmsmga003.fm.intel.com ([10.253.24.29]) by fmsmga105.fm.intel.com with ESMTP/TLS/DHE-RSA-AES256-GCM-SHA384; 24 Sep 2019 01:44:42 -0700 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.64,543,1559545200"; d="scan'208";a="195631465" Received: from dpdk-xiaoyun3.sh.intel.com ([10.67.118.162]) by FMSMGA003.fm.intel.com with ESMTP; 24 Sep 2019 01:44:41 -0700 From: Xiaoyun Li To: jingjing.wu@intel.com, keith.wiles@intel.com, omkar.maslekar@intel.com, cunming.liang@intel.com Cc: dev@dpdk.org, Xiaoyun Li Date: Tue, 24 Sep 2019 16:43:44 +0800 Message-Id: <20190924084345.93255-4-xiaoyun.li@intel.com> X-Mailer: git-send-email 2.17.1 In-Reply-To: <20190924084345.93255-1-xiaoyun.li@intel.com> References: <20190909032730.29718-1-xiaoyun.li@intel.com> <20190924084345.93255-1-xiaoyun.li@intel.com> Subject: [dpdk-dev] [PATCH v5 3/4] raw/ntb: add enqueue and dequeue functions X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org Sender: "dev" Introduce enqueue and dequeue functions to support packet based processing. And enable write-combining for ntb driver since it can improve the performance a lot. Signed-off-by: Xiaoyun Li --- doc/guides/rawdevs/ntb.rst | 28 ++++ drivers/raw/ntb/ntb.c | 242 ++++++++++++++++++++++++++++++--- drivers/raw/ntb/ntb.h | 2 + drivers/raw/ntb/ntb_hw_intel.c | 22 +++ 4 files changed, 275 insertions(+), 19 deletions(-) diff --git a/doc/guides/rawdevs/ntb.rst b/doc/guides/rawdevs/ntb.rst index 99e7db441..afd5769fc 100644 --- a/doc/guides/rawdevs/ntb.rst +++ b/doc/guides/rawdevs/ntb.rst @@ -45,6 +45,24 @@ to use, i.e. igb_uio, vfio. The ``dpdk-devbind.py`` script can be used to show devices status and to bind them to a suitable kernel driver. They will appear under the category of "Misc (rawdev) devices". +Prerequisites +------------- +NTB PMD needs kernel PCI driver to support write combining (WC) to get +better performance. The difference will be more than 10 times. +To enable WC, there are 2 ways. +- Insert igb_uio with ``wc_active=1`` flag if use igb_uio driver. + insmod igb_uio.ko wc_active=1 +- Enable WC for NTB device's Bar 2 and Bar 4 (Mapped memory) manually. + Get bar base address using ``lspci -vvv -s ae:00.0 | grep Region``. + Region 0: Memory at 39bfe0000000 (64-bit, prefetchable) [size=64K] + Region 2: Memory at 39bfa0000000 (64-bit, prefetchable) [size=512M] + Region 4: Memory at 39bfc0000000 (64-bit, prefetchable) [size=512M] + Using the following command to enable WC. + echo "base=0x39bfa0000000 size=0x400000 type=write-combining" >> /proc/mtrr + echo "base=0x39bfc0000000 size=0x400000 type=write-combining" >> /proc/mtrr + To disable WC for these regions, using the following. + echo "disable=1" >> /proc/mtrr + Ring Layout ----------- @@ -83,6 +101,16 @@ like the following: +------------------------+ +------------------------+ <---------traffic--------- +- Enqueue and Dequeue + Based on this ring layout, enqueue reads rx_tail to get how many free + buffers and writes used_ring and tx_tail to tell the peer which buffers + are filled with data. + And dequeue reads tx_tail to get how many packets are arrived, and + writes desc_ring and rx_tail to tell the peer about the new allocated + buffers. + So in this way, only remote write happens and remote read can be avoid + to get better performance. + Limitation ---------- diff --git a/drivers/raw/ntb/ntb.c b/drivers/raw/ntb/ntb.c index 1367b2f59..d0f4f1af0 100644 --- a/drivers/raw/ntb/ntb.c +++ b/drivers/raw/ntb/ntb.c @@ -562,26 +562,140 @@ ntb_queue_init(struct rte_rawdev *dev, uint16_t qp_id) return 0; } +static inline void +ntb_enqueue_cleanup(struct ntb_tx_queue *txq) +{ + struct ntb_tx_entry *sw_ring = txq->sw_ring; + uint16_t tx_free = txq->last_avail; + uint16_t nb_to_clean, i; + + /* avail_cnt + 1 represents where to rx next in the peer. */ + nb_to_clean = (*txq->avail_cnt - txq->last_avail + 1 + + txq->nb_tx_desc) & (txq->nb_tx_desc - 1); + nb_to_clean = RTE_MIN(nb_to_clean, txq->tx_free_thresh); + for (i = 0; i < nb_to_clean; i++) { + if (sw_ring[tx_free].mbuf) + rte_pktmbuf_free_seg(sw_ring[tx_free].mbuf); + tx_free = (tx_free + 1) & (txq->nb_tx_desc - 1); + } + + txq->nb_tx_free += nb_to_clean; + txq->last_avail = tx_free; +} + static int ntb_enqueue_bufs(struct rte_rawdev *dev, struct rte_rawdev_buf **buffers, unsigned int count, rte_rawdev_obj_t context) { - /* Not FIFO right now. Just for testing memory write. */ struct ntb_hw *hw = dev->dev_private; - unsigned int i; - void *bar_addr; - size_t size; + struct ntb_tx_queue *txq = hw->tx_queues[(size_t)context]; + struct ntb_tx_entry *sw_ring = txq->sw_ring; + struct rte_mbuf *txm; + struct ntb_used tx_used[NTB_MAX_DESC_SIZE]; + volatile struct ntb_desc *tx_item; + uint16_t tx_last, nb_segs, off, last_used, avail_cnt; + uint16_t nb_mbufs = 0; + uint16_t nb_tx = 0; + uint64_t bytes = 0; + void *buf_addr; + int i; - if (hw->ntb_ops->get_peer_mw_addr == NULL) - return -ENOTSUP; - bar_addr = (*hw->ntb_ops->get_peer_mw_addr)(dev, 0); - size = (size_t)context; + if (unlikely(hw->ntb_ops->ioremap == NULL)) { + NTB_LOG(ERR, "Ioremap not supported."); + return nb_tx; + } - for (i = 0; i < count; i++) - rte_memcpy(bar_addr, buffers[i]->buf_addr, size); - return 0; + if (unlikely(dev->started == 0 || hw->peer_dev_up == 0)) { + NTB_LOG(DEBUG, "Link is not up."); + return nb_tx; + } + + if (txq->nb_tx_free < txq->tx_free_thresh) + ntb_enqueue_cleanup(txq); + + off = NTB_XSTATS_NUM * ((size_t)context + 1); + last_used = txq->last_used; + avail_cnt = *txq->avail_cnt;/* Where to alloc next. */ + for (nb_tx = 0; nb_tx < count; nb_tx++) { + txm = (struct rte_mbuf *)(buffers[nb_tx]->buf_addr); + if (txm == NULL || txq->nb_tx_free < txm->nb_segs) + break; + + tx_last = (txq->last_used + txm->nb_segs - 1) & + (txq->nb_tx_desc - 1); + nb_segs = txm->nb_segs; + for (i = 0; i < nb_segs; i++) { + /* Not enough ring space for tx. */ + if (txq->last_used == avail_cnt) + goto end_of_tx; + sw_ring[txq->last_used].mbuf = txm; + tx_item = txq->tx_desc_ring + txq->last_used; + + if (!tx_item->len) { + (hw->ntb_xstats[NTB_TX_ERRS_ID + off])++; + goto end_of_tx; + } + if (txm->data_len > tx_item->len) { + NTB_LOG(ERR, "Data length exceeds buf length." + " Only %u data would be transmitted.", + tx_item->len); + txm->data_len = tx_item->len; + } + + /* translate remote virtual addr to bar virtual addr */ + buf_addr = (*hw->ntb_ops->ioremap)(dev, tx_item->addr); + if (buf_addr == NULL) { + (hw->ntb_xstats[NTB_TX_ERRS_ID + off])++; + NTB_LOG(ERR, "Null remap addr."); + goto end_of_tx; + } + rte_memcpy(buf_addr, rte_pktmbuf_mtod(txm, void *), + txm->data_len); + + tx_used[nb_mbufs].len = txm->data_len; + tx_used[nb_mbufs++].flags = (txq->last_used == + tx_last) ? + NTB_FLAG_EOP : 0; + + /* update stats */ + bytes += txm->data_len; + + txm = txm->next; + + sw_ring[txq->last_used].next_id = (txq->last_used + 1) & + (txq->nb_tx_desc - 1); + sw_ring[txq->last_used].last_id = tx_last; + txq->last_used = (txq->last_used + 1) & + (txq->nb_tx_desc - 1); + } + txq->nb_tx_free -= nb_segs; + } + +end_of_tx: + if (nb_tx) { + uint16_t nb1, nb2; + if (nb_mbufs > txq->nb_tx_desc - last_used) { + nb1 = txq->nb_tx_desc - last_used; + nb2 = nb_mbufs - txq->nb_tx_desc + last_used; + } else { + nb1 = nb_mbufs; + nb2 = 0; + } + rte_memcpy(txq->tx_used_ring + last_used, tx_used, + sizeof(struct ntb_used) * nb1); + rte_memcpy(txq->tx_used_ring, tx_used + nb1, + sizeof(struct ntb_used) * nb2); + *txq->used_cnt = txq->last_used; + rte_wmb(); + + /* update queue stats */ + hw->ntb_xstats[NTB_TX_BYTES_ID + off] += bytes; + hw->ntb_xstats[NTB_TX_PKTS_ID + off] += nb_tx; + } + + return nb_tx; } static int @@ -590,16 +704,106 @@ ntb_dequeue_bufs(struct rte_rawdev *dev, unsigned int count, rte_rawdev_obj_t context) { - /* Not FIFO. Just for testing memory read. */ struct ntb_hw *hw = dev->dev_private; - unsigned int i; - size_t size; + struct ntb_rx_queue *rxq = hw->rx_queues[(size_t)context]; + struct ntb_rx_entry *sw_ring = rxq->sw_ring; + struct ntb_desc rx_desc[NTB_MAX_DESC_SIZE]; + struct rte_mbuf *first, *rxm_t; + struct rte_mbuf *prev = NULL; + volatile struct ntb_used *rx_item; + uint16_t nb_mbufs = 0; + uint16_t nb_rx = 0; + uint64_t bytes = 0; + uint16_t off, last_avail, used_cnt, used_nb; + int i; + + if (unlikely(dev->started == 0 || hw->peer_dev_up == 0)) { + NTB_LOG(DEBUG, "Link is not up"); + return nb_rx; + } + + used_cnt = *rxq->used_cnt; + + if (rxq->last_used == used_cnt) + return nb_rx; + + last_avail = rxq->last_avail; + used_nb = (used_cnt - rxq->last_used) & (rxq->nb_rx_desc - 1); + count = RTE_MIN(count, used_nb); + for (nb_rx = 0; nb_rx < count; nb_rx++) { + i = 0; + while (true) { + rx_item = rxq->rx_used_ring + rxq->last_used; + rxm_t = sw_ring[rxq->last_used].mbuf; + rxm_t->data_len = rx_item->len; + rxm_t->data_off = RTE_PKTMBUF_HEADROOM; + rxm_t->port = rxq->port_id; + + if (!i) { + rxm_t->nb_segs = 1; + first = rxm_t; + first->pkt_len = 0; + buffers[nb_rx]->buf_addr = rxm_t; + } else { + prev->next = rxm_t; + first->nb_segs++; + } - size = (size_t)context; + prev = rxm_t; + first->pkt_len += prev->data_len; + rxq->last_used = (rxq->last_used + 1) & + (rxq->nb_rx_desc - 1); - for (i = 0; i < count; i++) - rte_memcpy(buffers[i]->buf_addr, hw->mz[i]->addr, size); - return 0; + /* alloc new mbuf */ + rxm_t = rte_mbuf_raw_alloc(rxq->mpool); + if (unlikely(rxm_t == NULL)) { + NTB_LOG(ERR, "recv alloc mbuf failed."); + goto end_of_rx; + } + rxm_t->port = rxq->port_id; + sw_ring[rxq->last_avail].mbuf = rxm_t; + i++; + + /* fill new desc */ + rx_desc[nb_mbufs].addr = + rte_pktmbuf_mtod(rxm_t, size_t); + rx_desc[nb_mbufs++].len = rxm_t->buf_len - + RTE_PKTMBUF_HEADROOM; + rxq->last_avail = (rxq->last_avail + 1) & + (rxq->nb_rx_desc - 1); + + if (rx_item->flags & NTB_FLAG_EOP) + break; + } + /* update stats */ + bytes += first->pkt_len; + } + +end_of_rx: + if (nb_rx) { + uint16_t nb1, nb2; + if (nb_mbufs > rxq->nb_rx_desc - last_avail) { + nb1 = rxq->nb_rx_desc - last_avail; + nb2 = nb_mbufs - rxq->nb_rx_desc + last_avail; + } else { + nb1 = nb_mbufs; + nb2 = 0; + } + rte_memcpy(rxq->rx_desc_ring + last_avail, rx_desc, + sizeof(struct ntb_desc) * nb1); + rte_memcpy(rxq->rx_desc_ring, rx_desc + nb1, + sizeof(struct ntb_desc) * nb2); + *rxq->avail_cnt = rxq->last_avail; + rte_wmb(); + + /* update queue stats */ + off = NTB_XSTATS_NUM * ((size_t)context + 1); + hw->ntb_xstats[NTB_RX_BYTES_ID + off] += bytes; + hw->ntb_xstats[NTB_RX_PKTS_ID + off] += nb_rx; + hw->ntb_xstats[NTB_RX_MISS_ID + off] += (count - nb_rx); + } + + return nb_rx; } static void @@ -1296,7 +1500,7 @@ ntb_remove(struct rte_pci_device *pci_dev) static struct rte_pci_driver rte_ntb_pmd = { .id_table = pci_id_ntb_map, - .drv_flags = RTE_PCI_DRV_NEED_MAPPING, + .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_WC_ACTIVATE, .probe = ntb_probe, .remove = ntb_remove, }; diff --git a/drivers/raw/ntb/ntb.h b/drivers/raw/ntb/ntb.h index 3cc160680..a561c42d1 100644 --- a/drivers/raw/ntb/ntb.h +++ b/drivers/raw/ntb/ntb.h @@ -87,6 +87,7 @@ enum ntb_spad_idx { * @ntb_dev_init: Init ntb dev. * @get_peer_mw_addr: To get the addr of peer mw[mw_idx]. * @mw_set_trans: Set translation of internal memory that remote can access. + * @ioremap: Translate the remote host address to bar address. * @get_link_status: get link status, link speed and link width. * @set_link: Set local side up/down. * @spad_read: Read local/peer spad register val. @@ -103,6 +104,7 @@ struct ntb_dev_ops { void *(*get_peer_mw_addr)(const struct rte_rawdev *dev, int mw_idx); int (*mw_set_trans)(const struct rte_rawdev *dev, int mw_idx, uint64_t addr, uint64_t size); + void *(*ioremap)(const struct rte_rawdev *dev, uint64_t addr); int (*get_link_status)(const struct rte_rawdev *dev); int (*set_link)(const struct rte_rawdev *dev, bool up); uint32_t (*spad_read)(const struct rte_rawdev *dev, int spad, diff --git a/drivers/raw/ntb/ntb_hw_intel.c b/drivers/raw/ntb/ntb_hw_intel.c index 0e73f1609..e7f8667cd 100644 --- a/drivers/raw/ntb/ntb_hw_intel.c +++ b/drivers/raw/ntb/ntb_hw_intel.c @@ -162,6 +162,27 @@ intel_ntb_mw_set_trans(const struct rte_rawdev *dev, int mw_idx, return 0; } +static void * +intel_ntb_ioremap(const struct rte_rawdev *dev, uint64_t addr) +{ + struct ntb_hw *hw = dev->dev_private; + void *mapped = NULL; + void *base; + int i; + + for (i = 0; i < hw->peer_used_mws; i++) { + if (addr >= hw->peer_mw_base[i] && + addr <= hw->peer_mw_base[i] + hw->mw_size[i]) { + base = intel_ntb_get_peer_mw_addr(dev, i); + mapped = (void *)(size_t)(addr - hw->peer_mw_base[i] + + (size_t)base); + break; + } + } + + return mapped; +} + static int intel_ntb_get_link_status(const struct rte_rawdev *dev) { @@ -357,6 +378,7 @@ const struct ntb_dev_ops intel_ntb_ops = { .ntb_dev_init = intel_ntb_dev_init, .get_peer_mw_addr = intel_ntb_get_peer_mw_addr, .mw_set_trans = intel_ntb_mw_set_trans, + .ioremap = intel_ntb_ioremap, .get_link_status = intel_ntb_get_link_status, .set_link = intel_ntb_set_link, .spad_read = intel_ntb_spad_read, -- 2.17.1