From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mga03.intel.com (mga03.intel.com [134.134.136.65]) by dpdk.org (Postfix) with ESMTP id 69629959 for ; Tue, 16 Dec 2014 09:23:29 +0100 (CET) Received: from orsmga002.jf.intel.com ([10.7.209.21]) by orsmga103.jf.intel.com with ESMTP; 16 Dec 2014 00:21:13 -0800 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.07,585,1413270000"; d="scan'208";a="654707912" Received: from shvmail01.sh.intel.com ([10.239.29.42]) by orsmga002.jf.intel.com with ESMTP; 16 Dec 2014 00:23:27 -0800 Received: from shecgisg004.sh.intel.com (shecgisg004.sh.intel.com [10.239.29.89]) by shvmail01.sh.intel.com with ESMTP id sBG8NONu018346; Tue, 16 Dec 2014 16:23:24 +0800 Received: from shecgisg004.sh.intel.com (localhost [127.0.0.1]) by shecgisg004.sh.intel.com (8.13.6/8.13.6/SuSE Linux 0.8) with ESMTP id sBG8NMSC031958; Tue, 16 Dec 2014 16:23:24 +0800 Received: (from hzhan75@localhost) by shecgisg004.sh.intel.com (8.13.6/8.13.6/Submit) id sBG8NMP4031954; Tue, 16 Dec 2014 16:23:22 +0800 From: Helin Zhang To: dev@dpdk.org Date: Tue, 16 Dec 2014 16:23:20 +0800 Message-Id: <1418718200-31924-1-git-send-email-helin.zhang@intel.com> X-Mailer: git-send-email 1.7.4.1 In-Reply-To: <1418630187-28917-1-git-send-email-helin.zhang@intel.com> References: <1418630187-28917-1-git-send-email-helin.zhang@intel.com> Cc: aaron.f.rowden@intel.com Subject: [dpdk-dev] [PATCH v3] i40e: workaround for X710 performance issues X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: patches and discussions about DPDK List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Tue, 16 Dec 2014 08:23:30 -0000 On X710, performance number is far from the expectation on recent firmware versions. The fix for this issue may not be integrated in the following firmware version. So the workaround in software driver is needed. It needs to modify the initial values of 3 internal only registers. Note that the workaround can be removed when it is fixed in firmware in the future. Signed-off-by: Helin Zhang --- lib/librte_pmd_i40e/i40e_ethdev.c | 89 +++++++++++++++++++++++++++++++++++++++ 1 file changed, 89 insertions(+) v2 changes: * Added a compile error fix. v3 changes: * Used PRIx32 and PRIx64 instead for printing uint32_t and uint64_t variables. * Re-worded annotations, and commit logs. diff --git a/lib/librte_pmd_i40e/i40e_ethdev.c b/lib/librte_pmd_i40e/i40e_ethdev.c index 008d62c..624f0ce 100644 --- a/lib/librte_pmd_i40e/i40e_ethdev.c +++ b/lib/librte_pmd_i40e/i40e_ethdev.c @@ -198,6 +198,7 @@ static int i40e_dev_filter_ctrl(struct rte_eth_dev *dev, enum rte_filter_type filter_type, enum rte_filter_op filter_op, void *arg); +static void i40e_configure_registers(struct i40e_hw *hw); /* Default hash key buffer for RSS */ static uint32_t rss_key_default[I40E_PFQF_HKEY_MAX_INDEX + 1]; @@ -443,6 +444,16 @@ eth_i40e_dev_init(__rte_unused struct eth_driver *eth_drv, /* Clear PXE mode */ i40e_clear_pxe_mode(hw); + /* + * On X710, performance number is far from the expectation on recent + * firmware versions. The fix for this issue may not be integrated in + * the following firmware version. So the workaround in software driver + * is needed. It needs to modify the initial values of 3 internal only + * registers. Note that the workaround can be removed when it is fixed + * in firmware in the future. + */ + i40e_configure_registers(hw); + /* Get hw capabilities */ ret = i40e_get_cap(hw); if (ret != I40E_SUCCESS) { @@ -5294,3 +5305,81 @@ i40e_pctype_to_flowtype(enum i40e_filter_pctype pctype) return flowtype_table[pctype]; } + +static int +i40e_debug_read_register(struct i40e_hw *hw, uint32_t addr, uint64_t *val) +{ + struct i40e_aq_desc desc; + struct i40e_aqc_debug_reg_read_write *cmd = + (struct i40e_aqc_debug_reg_read_write *)&desc.params.raw; + enum i40e_status_code status; + + i40e_fill_default_direct_cmd_desc(&desc, i40e_aqc_opc_debug_read_reg); + cmd->address = rte_cpu_to_le_32(addr); + status = i40e_asq_send_command(hw, &desc, NULL, 0, NULL); + if (status < 0) + return status; + + *val = ((uint64_t)(rte_le_to_cpu_32(cmd->value_high)) << (CHAR_BIT * + sizeof(uint32_t))) + rte_le_to_cpu_32(cmd->value_low); + + return status; +} + +/* + * On X710, performance number is far from the expectation on recent firmware + * versions. The fix for this issue may not be integrated in the following + * firmware version. So the workaround in software driver is needed. It needs + * to modify the initial values of 3 internal only registers. Note that the + * workaround can be removed when it is fixed in firmware in the future. + */ +static void +i40e_configure_registers(struct i40e_hw *hw) +{ +#define I40E_GL_SWR_PRI_JOIN_MAP_0 0x26CE00 +#define I40E_GL_SWR_PRI_JOIN_MAP_2 0x26CE08 +#define I40E_GL_SWR_PM_UP_THR 0x269FBC +#define I40E_GL_SWR_PRI_JOIN_MAP_0_VALUE 0x10000200 +#define I40E_GL_SWR_PRI_JOIN_MAP_2_VALUE 0x011f0200 +#define I40E_GL_SWR_PM_UP_THR_VALUE 0x03030303 + + static const struct { + uint32_t addr; + uint64_t val; + } reg_table[] = { + {I40E_GL_SWR_PRI_JOIN_MAP_0, I40E_GL_SWR_PRI_JOIN_MAP_0_VALUE}, + {I40E_GL_SWR_PRI_JOIN_MAP_2, I40E_GL_SWR_PRI_JOIN_MAP_2_VALUE}, + {I40E_GL_SWR_PM_UP_THR, I40E_GL_SWR_PM_UP_THR_VALUE}, + }; + uint64_t reg; + uint32_t i; + int ret; + + /* Below fix is for X710 only */ + if (i40e_is_40G_device(hw->device_id)) + return; + + for (i = 0; i < RTE_DIM(reg_table); i++) { + ret = i40e_debug_read_register(hw, reg_table[i].addr, ®); + if (ret < 0) { + PMD_DRV_LOG(ERR, "Failed to read from 0x%"PRIx32, + reg_table[i].addr); + break; + } + PMD_DRV_LOG(DEBUG, "Read from 0x%"PRIx32": 0x%"PRIx64, + reg_table[i].addr, reg); + if (reg == reg_table[i].val) + continue; + + ret = i40e_aq_debug_write_register(hw, reg_table[i].addr, + reg_table[i].val, NULL); + if (ret < 0) { + PMD_DRV_LOG(ERR, "Failed to write 0x%"PRIx64" to the " + "address of 0x%"PRIx32, reg_table[i].val, + reg_table[i].addr); + break; + } + PMD_DRV_LOG(DEBUG, "Write 0x%"PRIx64" to the address of " + "0x%"PRIx32, reg_table[i].val, reg_table[i].addr); + } +} -- 1.8.1.4