From 325f85f37e5b35807d86185bdf2c64d2980c44ba Mon Sep 17 00:00:00 2001 From: Michael Chan Date: Tue, 10 Dec 2019 02:49:08 -0500 Subject: bnxt_en: Free context memory in the open path if firmware has been reset. This will trigger new context memory to be rediscovered and allocated during the re-probe process after a firmware reset. Without this, the newly reset firmware does not have valid context memory and the driver will eventually fail to allocate some resources. Fixes: ec5d31e3c15d ("bnxt_en: Handle firmware reset status during IF_UP.") Signed-off-by: Michael Chan Signed-off-by: David S. Miller --- drivers/net/ethernet/broadcom/bnxt/bnxt.c | 3 +++ 1 file changed, 3 insertions(+) (limited to 'drivers/net/ethernet/broadcom/bnxt/bnxt.c') diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c index 85983f0e3134..65c1c4e26f4f 100644 --- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c +++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c @@ -8796,6 +8796,9 @@ static int bnxt_hwrm_if_change(struct bnxt *bp, bool up) if (fw_reset) { if (!test_bit(BNXT_STATE_IN_FW_RESET, &bp->state)) bnxt_ulp_stop(bp); + bnxt_free_ctx_mem(bp); + kfree(bp->ctx); + bp->ctx = NULL; rc = bnxt_fw_init_one(bp); if (rc) { set_bit(BNXT_STATE_ABORT_ERR, &bp->state); -- cgit From 8280b38e01f71e0f89389ccad3fa43b79e57c604 Mon Sep 17 00:00:00 2001 From: Vasundhara Volam Date: Tue, 10 Dec 2019 02:49:10 -0500 Subject: bnxt_en: Fix bp->fw_health allocation and free logic. bp->fw_health needs to be allocated for either the firmware initiated reset feature or the driver initiated error recovery feature. The current code is not allocating bp->fw_health for all the necessary cases. This patch corrects the logic to allocate bp->fw_health correctly when needed. If allocation fails, we clear the feature flags. We also add the the missing kfree(bp->fw_health) when the driver is unloaded. If we get an async reset message from the firmware, we also need to make sure that we have a valid bp->fw_health before proceeding. Fixes: 07f83d72d238 ("bnxt_en: Discover firmware error recovery capabilities.") Signed-off-by: Vasundhara Volam Signed-off-by: Michael Chan Signed-off-by: David S. Miller --- drivers/net/ethernet/broadcom/bnxt/bnxt.c | 36 ++++++++++++++++++++++--------- 1 file changed, 26 insertions(+), 10 deletions(-) (limited to 'drivers/net/ethernet/broadcom/bnxt/bnxt.c') diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c index 65c1c4e26f4f..d6a5fce1b06e 100644 --- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c +++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c @@ -2001,6 +2001,9 @@ static int bnxt_async_event_process(struct bnxt *bp, case ASYNC_EVENT_CMPL_EVENT_ID_RESET_NOTIFY: { u32 data1 = le32_to_cpu(cmpl->event_data1); + if (!bp->fw_health) + goto async_event_process_exit; + bp->fw_reset_timestamp = jiffies; bp->fw_reset_min_dsecs = cmpl->timestamp_lo; if (!bp->fw_reset_min_dsecs) @@ -4421,8 +4424,9 @@ int bnxt_hwrm_func_drv_rgtr(struct bnxt *bp, unsigned long *bmap, int bmap_size, FUNC_DRV_RGTR_REQ_ENABLES_ASYNC_EVENT_FWD); req.os_type = cpu_to_le16(FUNC_DRV_RGTR_REQ_OS_TYPE_LINUX); - flags = FUNC_DRV_RGTR_REQ_FLAGS_16BIT_VER_MODE | - FUNC_DRV_RGTR_REQ_FLAGS_HOT_RESET_SUPPORT; + flags = FUNC_DRV_RGTR_REQ_FLAGS_16BIT_VER_MODE; + if (bp->fw_cap & BNXT_FW_CAP_HOT_RESET) + flags |= FUNC_DRV_RGTR_REQ_FLAGS_HOT_RESET_SUPPORT; if (bp->fw_cap & BNXT_FW_CAP_ERROR_RECOVERY) flags |= FUNC_DRV_RGTR_REQ_FLAGS_ERROR_RECOVERY_SUPPORT | FUNC_DRV_RGTR_REQ_FLAGS_MASTER_SUPPORT; @@ -7115,14 +7119,6 @@ static int bnxt_hwrm_error_recovery_qcfg(struct bnxt *bp) rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT); if (rc) goto err_recovery_out; - if (!fw_health) { - fw_health = kzalloc(sizeof(*fw_health), GFP_KERNEL); - bp->fw_health = fw_health; - if (!fw_health) { - rc = -ENOMEM; - goto err_recovery_out; - } - } fw_health->flags = le32_to_cpu(resp->flags); if ((fw_health->flags & ERROR_RECOVERY_QCFG_RESP_FLAGS_CO_CPU) && !(bp->fw_cap & BNXT_FW_CAP_KONG_MB_CHNL)) { @@ -10485,6 +10481,23 @@ static void bnxt_init_dflt_coal(struct bnxt *bp) bp->stats_coal_ticks = BNXT_DEF_STATS_COAL_TICKS; } +static void bnxt_alloc_fw_health(struct bnxt *bp) +{ + if (bp->fw_health) + return; + + if (!(bp->fw_cap & BNXT_FW_CAP_HOT_RESET) && + !(bp->fw_cap & BNXT_FW_CAP_ERROR_RECOVERY)) + return; + + bp->fw_health = kzalloc(sizeof(*bp->fw_health), GFP_KERNEL); + if (!bp->fw_health) { + netdev_warn(bp->dev, "Failed to allocate fw_health\n"); + bp->fw_cap &= ~BNXT_FW_CAP_HOT_RESET; + bp->fw_cap &= ~BNXT_FW_CAP_ERROR_RECOVERY; + } +} + static int bnxt_fw_init_one_p1(struct bnxt *bp) { int rc; @@ -10531,6 +10544,7 @@ static int bnxt_fw_init_one_p2(struct bnxt *bp) netdev_warn(bp->dev, "hwrm query adv flow mgnt failure rc: %d\n", rc); + bnxt_alloc_fw_health(bp); rc = bnxt_hwrm_error_recovery_qcfg(bp); if (rc) netdev_warn(bp->dev, "hwrm query error recovery failure rc: %d\n", @@ -11418,6 +11432,8 @@ static void bnxt_remove_one(struct pci_dev *pdev) bnxt_dcb_free(bp); kfree(bp->edev); bp->edev = NULL; + kfree(bp->fw_health); + bp->fw_health = NULL; bnxt_cleanup_pci(bp); bnxt_free_ctx_mem(bp); kfree(bp->ctx); -- cgit From 0797c10d2d1fa0d6f14612404781b348fc757c3e Mon Sep 17 00:00:00 2001 From: Vasundhara Volam Date: Tue, 10 Dec 2019 02:49:11 -0500 Subject: bnxt_en: Remove unnecessary NULL checks for fw_health After fixing the allocation of bp->fw_health in the previous patch, the driver will not go through the fw reset and recovery code paths if bp->fw_health allocation fails. So we can now remove the unnecessary NULL checks. Signed-off-by: Vasundhara Volam Signed-off-by: Michael Chan Signed-off-by: David S. Miller --- drivers/net/ethernet/broadcom/bnxt/bnxt.c | 6 ++---- 1 file changed, 2 insertions(+), 4 deletions(-) (limited to 'drivers/net/ethernet/broadcom/bnxt/bnxt.c') diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c index d6a5fce1b06e..2a100ff3da8b 100644 --- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c +++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c @@ -9989,8 +9989,7 @@ static void bnxt_fw_health_check(struct bnxt *bp) struct bnxt_fw_health *fw_health = bp->fw_health; u32 val; - if (!fw_health || !fw_health->enabled || - test_bit(BNXT_STATE_IN_FW_RESET, &bp->state)) + if (!fw_health->enabled || test_bit(BNXT_STATE_IN_FW_RESET, &bp->state)) return; if (fw_health->tmr_counter) { @@ -10768,8 +10767,7 @@ static void bnxt_fw_reset_task(struct work_struct *work) bnxt_queue_fw_reset_work(bp, bp->fw_reset_min_dsecs * HZ / 10); return; case BNXT_FW_RESET_STATE_ENABLE_DEV: - if (test_bit(BNXT_STATE_FW_FATAL_COND, &bp->state) && - bp->fw_health) { + if (test_bit(BNXT_STATE_FW_FATAL_COND, &bp->state)) { u32 val; val = bnxt_fw_health_readl(bp, -- cgit From 937f188c1f4f89b3fa93ba31fc8587dc1fb14a22 Mon Sep 17 00:00:00 2001 From: Vasundhara Volam Date: Tue, 10 Dec 2019 02:49:12 -0500 Subject: bnxt_en: Fix the logic that creates the health reporters. Fix the logic to properly check the fw capabilities and create the devlink health reporters only when needed. The current code creates the reporters unconditionally as long as bp->fw_health is valid, and that's not correct. Call bnxt_dl_fw_reporters_create() directly from the init and reset code path instead of from bnxt_dl_register(). This allows the reporters to be adjusted when capabilities change. The same applies to bnxt_dl_fw_reporters_destroy(). Fixes: 6763c779c2d8 ("bnxt_en: Add new FW devlink_health_reporter") Signed-off-by: Vasundhara Volam Signed-off-by: Michael Chan Signed-off-by: David S. Miller --- drivers/net/ethernet/broadcom/bnxt/bnxt.c | 11 ++++++++++- 1 file changed, 10 insertions(+), 1 deletion(-) (limited to 'drivers/net/ethernet/broadcom/bnxt/bnxt.c') diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c index 2a100ff3da8b..819b7d72f9de 100644 --- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c +++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c @@ -10625,6 +10625,12 @@ static int bnxt_fw_init_one(struct bnxt *bp) rc = bnxt_approve_mac(bp, bp->dev->dev_addr, false); if (rc) return rc; + + /* In case fw capabilities have changed, destroy the unneeded + * reporters and create newly capable ones. + */ + bnxt_dl_fw_reporters_destroy(bp, false); + bnxt_dl_fw_reporters_create(bp); bnxt_fw_init_one_p3(bp); return 0; } @@ -11413,6 +11419,7 @@ static void bnxt_remove_one(struct pci_dev *pdev) if (BNXT_PF(bp)) { bnxt_sriov_disable(bp); + bnxt_dl_fw_reporters_destroy(bp, true); bnxt_dl_unregister(bp); } @@ -11892,8 +11899,10 @@ static int bnxt_init_one(struct pci_dev *pdev, const struct pci_device_id *ent) if (rc) goto init_err_cleanup_tc; - if (BNXT_PF(bp)) + if (BNXT_PF(bp)) { bnxt_dl_register(bp); + bnxt_dl_fw_reporters_create(bp); + } netdev_info(dev, "%s found at mem %lx, node addr %pM\n", board_info[ent->driver_data].name, -- cgit From 7e334fc8003c7a38372cc98e7be6082670a47d29 Mon Sep 17 00:00:00 2001 From: Vasundhara Volam Date: Tue, 10 Dec 2019 02:49:13 -0500 Subject: bnxt_en: Add missing devlink health reporters for VFs. The VF driver also needs to create the health reporters since VFs are also involved in firmware reset and recovery. Modify bnxt_dl_register() and bnxt_dl_unregister() so that they can be called by the VFs to register/unregister devlink. Only the PF will register the devlink parameters. With devlink registered, we can now create the health reporters on the VFs. Fixes: 6763c779c2d8 ("bnxt_en: Add new FW devlink_health_reporter") Signed-off-by: Vasundhara Volam Signed-off-by: Michael Chan Signed-off-by: David S. Miller --- drivers/net/ethernet/broadcom/bnxt/bnxt.c | 13 +++++-------- 1 file changed, 5 insertions(+), 8 deletions(-) (limited to 'drivers/net/ethernet/broadcom/bnxt/bnxt.c') diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c index 819b7d72f9de..a754903b27ee 100644 --- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c +++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c @@ -11417,12 +11417,11 @@ static void bnxt_remove_one(struct pci_dev *pdev) struct net_device *dev = pci_get_drvdata(pdev); struct bnxt *bp = netdev_priv(dev); - if (BNXT_PF(bp)) { + if (BNXT_PF(bp)) bnxt_sriov_disable(bp); - bnxt_dl_fw_reporters_destroy(bp, true); - bnxt_dl_unregister(bp); - } + bnxt_dl_fw_reporters_destroy(bp, true); + bnxt_dl_unregister(bp); pci_disable_pcie_error_reporting(pdev); unregister_netdev(dev); bnxt_shutdown_tc(bp); @@ -11899,10 +11898,8 @@ static int bnxt_init_one(struct pci_dev *pdev, const struct pci_device_id *ent) if (rc) goto init_err_cleanup_tc; - if (BNXT_PF(bp)) { - bnxt_dl_register(bp); - bnxt_dl_fw_reporters_create(bp); - } + bnxt_dl_register(bp); + bnxt_dl_fw_reporters_create(bp); netdev_info(dev, "%s found at mem %lx, node addr %pM\n", board_info[ent->driver_data].name, -- cgit From 6adc4601c2a1ac87b4ab8ed0cb55db6efd0264e8 Mon Sep 17 00:00:00 2001 From: Jonathan Lemon Date: Tue, 10 Dec 2019 08:39:46 -0800 Subject: bnxt: apply computed clamp value for coalece parameter After executing "ethtool -C eth0 rx-usecs-irq 0", the box becomes unresponsive, likely due to interrupt livelock. It appears that a minimum clamp value for the irq timer is computed, but is never applied. Fix by applying the corrected clamp value. Fixes: 74706afa712d ("bnxt_en: Update interrupt coalescing logic.") Signed-off-by: Jonathan Lemon Signed-off-by: Michael Chan Signed-off-by: Jakub Kicinski --- drivers/net/ethernet/broadcom/bnxt/bnxt.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers/net/ethernet/broadcom/bnxt/bnxt.c') diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c index a754903b27ee..c779f9cf8822 100644 --- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c +++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c @@ -6190,7 +6190,7 @@ static void bnxt_hwrm_set_coal_params(struct bnxt *bp, tmr = bnxt_usec_to_coal_tmr(bp, hw_coal->coal_ticks_irq); val = clamp_t(u16, tmr, 1, coal_cap->cmpl_aggr_dma_tmr_during_int_max); - req->cmpl_aggr_dma_tmr_during_int = cpu_to_le16(tmr); + req->cmpl_aggr_dma_tmr_during_int = cpu_to_le16(val); req->enables |= cpu_to_le16(BNXT_COAL_CMPL_AGGR_TMR_DURING_INT_ENABLE); } -- cgit