From patchwork Thu Mar 14 15:08:51 2013 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Vipul Pandya X-Patchwork-Id: 2271971 Return-Path: X-Original-To: patchwork-linux-rdma@patchwork.kernel.org Delivered-To: patchwork-process-083081@patchwork1.kernel.org Received: from vger.kernel.org (vger.kernel.org [209.132.180.67]) by patchwork1.kernel.org (Postfix) with ESMTP id A78E93FCF6 for ; Thu, 14 Mar 2013 15:17:53 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1758062Ab3CNPRs (ORCPT ); Thu, 14 Mar 2013 11:17:48 -0400 Received: from stargate.chelsio.com ([67.207.112.58]:6815 "EHLO stargate.chelsio.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1757582Ab3CNPPS (ORCPT ); Thu, 14 Mar 2013 11:15:18 -0400 Received: from maui.asicdesigners.com (maui.asicdesigners.com [10.192.180.15]) by stargate.chelsio.com (8.13.1/8.13.1) with SMTP id r2EFFEwD007532; Thu, 14 Mar 2013 08:15:14 -0700 Received: from strawberry.asicdesigners.com ([10.193.185.96]) by maui.asicdesigners.com with Microsoft SMTPSVC(6.0.3790.4675); Thu, 14 Mar 2013 08:15:14 -0700 From: Vipul Pandya To: netdev@vger.kernel.org, linux-rdma@vger.kernel.org, linux-scsi@vger.kernel.org Cc: davem@davemloft.net, roland@purestorage.com, JBottomley@parallels.com, dm@chelsio.com, swise@opengridcomputing.com, leedom@chelsio.com, naresh@chelsio.com, divy@chelsio.com, santosh@chelsio.com, arvindb@chelsio.com, abhishek@chelsio.com, vipul@chelsio.com Subject: [PATCH v2 net-next 05/22] cxgb4: Add T5 write combining support Date: Thu, 14 Mar 2013 20:38:51 +0530 Message-Id: <1363273748-25330-6-git-send-email-vipul@chelsio.com> X-Mailer: git-send-email 1.7.1 In-Reply-To: <1363273748-25330-1-git-send-email-vipul@chelsio.com> References: <1363273748-25330-1-git-send-email-vipul@chelsio.com> X-OriginalArrivalTime: 14 Mar 2013 15:15:14.0457 (UTC) FILETIME=[BA7B2890:01CE20C6] Sender: linux-rdma-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: linux-rdma@vger.kernel.org From: Santosh Rastapur This patch implements a low latency Write Combining (aka Write Coalescing) work request path. PCIE maps User Space Doorbell BAR2 region writes to the new interface to SGE. SGE pulls a new message from PCIE new interface and if its a coalesced write work request then pushes it for processing. This patch copies coalesced work request to memory mapped BAR2 space. Signed-off-by: Santosh Rastapur Signed-off-by: Vipul Pandya --- v2: Replaced #ifdef with portable interface wmb in ring_tx_db drivers/net/ethernet/chelsio/cxgb4/cxgb4.h | 2 + drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c | 53 ++++++++++++++++++++++- drivers/net/ethernet/chelsio/cxgb4/sge.c | 52 +++++++++++++++++++++- 3 files changed, 102 insertions(+), 5 deletions(-) diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h b/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h index a91dea6..f8ff30e 100644 --- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h +++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h @@ -439,6 +439,7 @@ struct sge_txq { spinlock_t db_lock; int db_disabled; unsigned short db_pidx; + u64 udb; }; struct sge_eth_txq { /* state for an SGE Ethernet Tx queue */ @@ -543,6 +544,7 @@ enum chip_type { struct adapter { void __iomem *regs; + void __iomem *bar2; struct pci_dev *pdev; struct device *pdev_dev; unsigned int mbox; diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c index 3d6d23a..ce1451c 100644 --- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c +++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c @@ -1327,6 +1327,8 @@ static char stats_strings[][ETH_GSTRING_LEN] = { "VLANinsertions ", "GROpackets ", "GROmerged ", + "WriteCoalSuccess ", + "WriteCoalFail ", }; static int get_sset_count(struct net_device *dev, int sset) @@ -1422,11 +1424,25 @@ static void get_stats(struct net_device *dev, struct ethtool_stats *stats, { struct port_info *pi = netdev_priv(dev); struct adapter *adapter = pi->adapter; + u32 val1, val2; t4_get_port_stats(adapter, pi->tx_chan, (struct port_stats *)data); data += sizeof(struct port_stats) / sizeof(u64); collect_sge_port_stats(adapter, pi, (struct queue_port_stats *)data); + data += sizeof(struct queue_port_stats) / sizeof(u64); + if (!is_t4(adapter->chip)) { + t4_write_reg(adapter, SGE_STAT_CFG, STATSOURCE_T5(7)); + val1 = t4_read_reg(adapter, SGE_STAT_TOTAL); + val2 = t4_read_reg(adapter, SGE_STAT_MATCH); + *data = val1 - val2; + data++; + *data = val2; + data++; + } else { + memset(data, 0, 2 * sizeof(u64)); + *data += 2; + } } /* @@ -5337,10 +5353,11 @@ static void free_some_resources(struct adapter *adapter) #define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN) #define VLAN_FEAT (NETIF_F_SG | NETIF_F_IP_CSUM | TSO_FLAGS | \ NETIF_F_IPV6_CSUM | NETIF_F_HIGHDMA) +#define SEGMENT_SIZE 128 static int init_one(struct pci_dev *pdev, const struct pci_device_id *ent) { - int func, i, err; + int func, i, err, s_qpp, qpp, num_seg; struct port_info *pi; bool highdma = false; struct adapter *adapter = NULL; @@ -5420,7 +5437,34 @@ static int init_one(struct pci_dev *pdev, const struct pci_device_id *ent) err = t4_prep_adapter(adapter); if (err) - goto out_unmap_bar; + goto out_unmap_bar0; + + if (!is_t4(adapter->chip)) { + s_qpp = QUEUESPERPAGEPF1 * adapter->fn; + qpp = 1 << QUEUESPERPAGEPF0_GET(t4_read_reg(adapter, + SGE_EGRESS_QUEUES_PER_PAGE_PF) >> s_qpp); + num_seg = PAGE_SIZE / SEGMENT_SIZE; + + /* Each segment size is 128B. Write coalescing is enabled only + * when SGE_EGRESS_QUEUES_PER_PAGE_PF reg value for the + * queue is less no of segments that can be accommodated in + * a page size. + */ + if (qpp > num_seg) { + dev_err(&pdev->dev, + "Incorrect number of egress queues per page\n"); + err = -EINVAL; + goto out_unmap_bar0; + } + adapter->bar2 = ioremap_wc(pci_resource_start(pdev, 2), + pci_resource_len(pdev, 2)); + if (!adapter->bar2) { + dev_err(&pdev->dev, "cannot map device bar2 region\n"); + err = -ENOMEM; + goto out_unmap_bar0; + } + } + setup_memwin(adapter); err = adap_init0(adapter); setup_memwin_rdma(adapter); @@ -5552,6 +5596,9 @@ sriov: out_free_dev: free_some_resources(adapter); out_unmap_bar: + if (!is_t4(adapter->chip)) + iounmap(adapter->bar2); + out_unmap_bar0: iounmap(adapter->regs); out_free_adapter: kfree(adapter); @@ -5602,6 +5649,8 @@ static void remove_one(struct pci_dev *pdev) free_some_resources(adapter); iounmap(adapter->regs); + if (!is_t4(adapter->chip)) + iounmap(adapter->bar2); kfree(adapter); pci_disable_pcie_error_reporting(pdev); pci_disable_device(pdev); diff --git a/drivers/net/ethernet/chelsio/cxgb4/sge.c b/drivers/net/ethernet/chelsio/cxgb4/sge.c index 7b17623..8b47b25 100644 --- a/drivers/net/ethernet/chelsio/cxgb4/sge.c +++ b/drivers/net/ethernet/chelsio/cxgb4/sge.c @@ -816,6 +816,22 @@ static void write_sgl(const struct sk_buff *skb, struct sge_txq *q, *end = 0; } +/* This function copies 64 byte coalesced work request to + * memory mapped BAR2 space(user space writes). + * For coalesced WR SGE, fetches data from the FIFO instead of from Host. + */ +static void cxgb_pio_copy(u64 __iomem *dst, u64 *src) +{ + int count = 8; + + while (count) { + writeq(*src, dst); + src++; + dst++; + count--; + } +} + /** * ring_tx_db - check and potentially ring a Tx queue's doorbell * @adap: the adapter @@ -826,11 +842,25 @@ static void write_sgl(const struct sk_buff *skb, struct sge_txq *q, */ static inline void ring_tx_db(struct adapter *adap, struct sge_txq *q, int n) { + unsigned int *wr, index; + wmb(); /* write descriptors before telling HW */ spin_lock(&q->db_lock); if (!q->db_disabled) { - t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL), - QID(q->cntxt_id) | PIDX(n)); + if (is_t4(adap->chip)) { + t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL), + QID(q->cntxt_id) | PIDX(n)); + } else { + if (n == 1) { + index = q->pidx ? (q->pidx - 1) : (q->size - 1); + wr = (unsigned int *)&q->desc[index]; + cxgb_pio_copy((u64 __iomem *) + (adap->bar2 + q->udb + 64), + (u64 *)wr); + } else + writel(n, adap->bar2 + q->udb + 8); + wmb(); + } } q->db_pidx = q->pidx; spin_unlock(&q->db_lock); @@ -2151,11 +2181,27 @@ err: static void init_txq(struct adapter *adap, struct sge_txq *q, unsigned int id) { + q->cntxt_id = id; + if (!is_t4(adap->chip)) { + unsigned int s_qpp; + unsigned short udb_density; + unsigned long qpshift; + int page; + + s_qpp = QUEUESPERPAGEPF1 * adap->fn; + udb_density = 1 << QUEUESPERPAGEPF0_GET((t4_read_reg(adap, + SGE_EGRESS_QUEUES_PER_PAGE_PF) >> s_qpp)); + qpshift = PAGE_SHIFT - ilog2(udb_density); + q->udb = q->cntxt_id << qpshift; + q->udb &= PAGE_MASK; + page = q->udb / PAGE_SIZE; + q->udb += (q->cntxt_id - (page * udb_density)) * 128; + } + q->in_use = 0; q->cidx = q->pidx = 0; q->stops = q->restarts = 0; q->stat = (void *)&q->desc[q->size]; - q->cntxt_id = id; spin_lock_init(&q->db_lock); adap->sge.egr_map[id - adap->sge.egr_start] = q; }