From ccc67ef50b9085b895738d7720840eb6fe98745e Mon Sep 17 00:00:00 2001 From: Tomer Tayar Date: Mon, 28 Jan 2019 19:27:56 +0200 Subject: qede: Error recovery process This patch adds the error recovery process in the qede driver. The process includes a partial/customized driver unload and load, which allows it to look like a short suspend period to the kernel while preserving the net devices' state. Signed-off-by: Tomer Tayar Signed-off-by: Ariel Elior Signed-off-by: Michal Kalderon Signed-off-by: David S. Miller --- drivers/net/ethernet/qlogic/qede/qede.h | 3 + drivers/net/ethernet/qlogic/qede/qede_main.c | 292 ++++++++++++++++++++++----- drivers/net/ethernet/qlogic/qede/qede_rdma.c | 63 ++++-- 3 files changed, 288 insertions(+), 70 deletions(-) (limited to 'drivers/net/ethernet/qlogic') diff --git a/drivers/net/ethernet/qlogic/qede/qede.h b/drivers/net/ethernet/qlogic/qede/qede.h index 613249d1e967..843416404aeb 100644 --- a/drivers/net/ethernet/qlogic/qede/qede.h +++ b/drivers/net/ethernet/qlogic/qede/qede.h @@ -162,6 +162,7 @@ struct qede_rdma_dev { struct list_head entry; struct list_head rdma_event_list; struct workqueue_struct *rdma_wq; + bool exp_recovery; }; struct qede_ptp; @@ -264,6 +265,7 @@ struct qede_dev { enum QEDE_STATE { QEDE_STATE_CLOSED, QEDE_STATE_OPEN, + QEDE_STATE_RECOVERY, }; #define HILO_U64(hi, lo) ((((u64)(hi)) << 32) + (lo)) @@ -462,6 +464,7 @@ struct qede_fastpath { #define QEDE_CSUM_UNNECESSARY BIT(1) #define QEDE_TUNN_CSUM_UNNECESSARY BIT(2) +#define QEDE_SP_RECOVERY 0 #define QEDE_SP_RX_MODE 1 #ifdef CONFIG_RFS_ACCEL diff --git a/drivers/net/ethernet/qlogic/qede/qede_main.c b/drivers/net/ethernet/qlogic/qede/qede_main.c index 5a74fcbdbc2b..6b4d96635238 100644 --- a/drivers/net/ethernet/qlogic/qede/qede_main.c +++ b/drivers/net/ethernet/qlogic/qede/qede_main.c @@ -133,23 +133,12 @@ static int qede_probe(struct pci_dev *pdev, const struct pci_device_id *id); static void qede_remove(struct pci_dev *pdev); static void qede_shutdown(struct pci_dev *pdev); static void qede_link_update(void *dev, struct qed_link_output *link); +static void qede_schedule_recovery_handler(void *dev); +static void qede_recovery_handler(struct qede_dev *edev); static void qede_get_eth_tlv_data(void *edev, void *data); static void qede_get_generic_tlv_data(void *edev, struct qed_generic_tlvs *data); -/* The qede lock is used to protect driver state change and driver flows that - * are not reentrant. - */ -void __qede_lock(struct qede_dev *edev) -{ - mutex_lock(&edev->qede_lock); -} - -void __qede_unlock(struct qede_dev *edev) -{ - mutex_unlock(&edev->qede_lock); -} - #ifdef CONFIG_QED_SRIOV static int qede_set_vf_vlan(struct net_device *ndev, int vf, u16 vlan, u8 qos, __be16 vlan_proto) @@ -231,6 +220,7 @@ static struct qed_eth_cb_ops qede_ll_ops = { .arfs_filter_op = qede_arfs_filter_op, #endif .link_update = qede_link_update, + .schedule_recovery_handler = qede_schedule_recovery_handler, .get_generic_tlv_data = qede_get_generic_tlv_data, .get_protocol_tlv_data = qede_get_eth_tlv_data, }, @@ -950,11 +940,57 @@ err: return -ENOMEM; } +/* The qede lock is used to protect driver state change and driver flows that + * are not reentrant. + */ +void __qede_lock(struct qede_dev *edev) +{ + mutex_lock(&edev->qede_lock); +} + +void __qede_unlock(struct qede_dev *edev) +{ + mutex_unlock(&edev->qede_lock); +} + +/* This version of the lock should be used when acquiring the RTNL lock is also + * needed in addition to the internal qede lock. + */ +void qede_lock(struct qede_dev *edev) +{ + rtnl_lock(); + __qede_lock(edev); +} + +void qede_unlock(struct qede_dev *edev) +{ + __qede_unlock(edev); + rtnl_unlock(); +} + static void qede_sp_task(struct work_struct *work) { struct qede_dev *edev = container_of(work, struct qede_dev, sp_task.work); + /* The locking scheme depends on the specific flag: + * In case of QEDE_SP_RECOVERY, acquiring the RTNL lock is required to + * ensure that ongoing flows are ended and new ones are not started. + * In other cases - only the internal qede lock should be acquired. + */ + + if (test_and_clear_bit(QEDE_SP_RECOVERY, &edev->sp_flags)) { +#ifdef CONFIG_QED_SRIOV + /* SRIOV must be disabled outside the lock to avoid a deadlock. + * The recovery of the active VFs is currently not supported. + */ + qede_sriov_configure(edev->pdev, 0); +#endif + qede_lock(edev); + qede_recovery_handler(edev); + qede_unlock(edev); + } + __qede_lock(edev); if (test_and_clear_bit(QEDE_SP_RX_MODE, &edev->sp_flags)) @@ -1031,6 +1067,7 @@ static void qede_log_probe(struct qede_dev *edev) enum qede_probe_mode { QEDE_PROBE_NORMAL, + QEDE_PROBE_RECOVERY, }; static int __qede_probe(struct pci_dev *pdev, u32 dp_module, u8 dp_level, @@ -1051,6 +1088,7 @@ static int __qede_probe(struct pci_dev *pdev, u32 dp_module, u8 dp_level, probe_params.dp_module = dp_module; probe_params.dp_level = dp_level; probe_params.is_vf = is_vf; + probe_params.recov_in_prog = (mode == QEDE_PROBE_RECOVERY); cdev = qed_ops->common->probe(pdev, &probe_params); if (!cdev) { rc = -ENODEV; @@ -1078,11 +1116,20 @@ static int __qede_probe(struct pci_dev *pdev, u32 dp_module, u8 dp_level, if (rc) goto err2; - edev = qede_alloc_etherdev(cdev, pdev, &dev_info, dp_module, - dp_level); - if (!edev) { - rc = -ENOMEM; - goto err2; + if (mode != QEDE_PROBE_RECOVERY) { + edev = qede_alloc_etherdev(cdev, pdev, &dev_info, dp_module, + dp_level); + if (!edev) { + rc = -ENOMEM; + goto err2; + } + } else { + struct net_device *ndev = pci_get_drvdata(pdev); + + edev = netdev_priv(ndev); + edev->cdev = cdev; + memset(&edev->stats, 0, sizeof(edev->stats)); + memcpy(&edev->dev_info, &dev_info, sizeof(dev_info)); } if (is_vf) @@ -1090,28 +1137,31 @@ static int __qede_probe(struct pci_dev *pdev, u32 dp_module, u8 dp_level, qede_init_ndev(edev); - rc = qede_rdma_dev_add(edev); + rc = qede_rdma_dev_add(edev, (mode == QEDE_PROBE_RECOVERY)); if (rc) goto err3; - /* Prepare the lock prior to the registration of the netdev, - * as once it's registered we might reach flows requiring it - * [it's even possible to reach a flow needing it directly - * from there, although it's unlikely]. - */ - INIT_DELAYED_WORK(&edev->sp_task, qede_sp_task); - mutex_init(&edev->qede_lock); - rc = register_netdev(edev->ndev); - if (rc) { - DP_NOTICE(edev, "Cannot register net-device\n"); - goto err4; + if (mode != QEDE_PROBE_RECOVERY) { + /* Prepare the lock prior to the registration of the netdev, + * as once it's registered we might reach flows requiring it + * [it's even possible to reach a flow needing it directly + * from there, although it's unlikely]. + */ + INIT_DELAYED_WORK(&edev->sp_task, qede_sp_task); + mutex_init(&edev->qede_lock); + + rc = register_netdev(edev->ndev); + if (rc) { + DP_NOTICE(edev, "Cannot register net-device\n"); + goto err4; + } } edev->ops->common->set_name(cdev, edev->ndev->name); /* PTP not supported on VFs */ if (!is_vf) - qede_ptp_enable(edev, true); + qede_ptp_enable(edev, (mode == QEDE_PROBE_NORMAL)); edev->ops->register_ops(cdev, &qede_ll_ops, edev); @@ -1126,7 +1176,7 @@ static int __qede_probe(struct pci_dev *pdev, u32 dp_module, u8 dp_level, return 0; err4: - qede_rdma_dev_remove(edev); + qede_rdma_dev_remove(edev, (mode == QEDE_PROBE_RECOVERY)); err3: free_netdev(edev->ndev); err2: @@ -1162,6 +1212,7 @@ static int qede_probe(struct pci_dev *pdev, const struct pci_device_id *id) enum qede_remove_mode { QEDE_REMOVE_NORMAL, + QEDE_REMOVE_RECOVERY, }; static void __qede_remove(struct pci_dev *pdev, enum qede_remove_mode mode) @@ -1172,15 +1223,19 @@ static void __qede_remove(struct pci_dev *pdev, enum qede_remove_mode mode) DP_INFO(edev, "Starting qede_remove\n"); - qede_rdma_dev_remove(edev); - unregister_netdev(ndev); - cancel_delayed_work_sync(&edev->sp_task); + qede_rdma_dev_remove(edev, (mode == QEDE_REMOVE_RECOVERY)); - qede_ptp_disable(edev); + if (mode != QEDE_REMOVE_RECOVERY) { + unregister_netdev(ndev); - edev->ops->common->set_power_state(cdev, PCI_D0); + cancel_delayed_work_sync(&edev->sp_task); - pci_set_drvdata(pdev, NULL); + edev->ops->common->set_power_state(cdev, PCI_D0); + + pci_set_drvdata(pdev, NULL); + } + + qede_ptp_disable(edev); /* Use global ops since we've freed edev */ qed_ops->common->slowpath_stop(cdev); @@ -1194,7 +1249,8 @@ static void __qede_remove(struct pci_dev *pdev, enum qede_remove_mode mode) * [e.g., QED register callbacks] won't break anything when * accessing the netdevice. */ - free_netdev(ndev); + if (mode != QEDE_REMOVE_RECOVERY) + free_netdev(ndev); dev_info(&pdev->dev, "Ending qede_remove successfully\n"); } @@ -1539,6 +1595,58 @@ static int qede_alloc_mem_load(struct qede_dev *edev) return 0; } +static void qede_empty_tx_queue(struct qede_dev *edev, + struct qede_tx_queue *txq) +{ + unsigned int pkts_compl = 0, bytes_compl = 0; + struct netdev_queue *netdev_txq; + int rc, len = 0; + + netdev_txq = netdev_get_tx_queue(edev->ndev, txq->ndev_txq_id); + + while (qed_chain_get_cons_idx(&txq->tx_pbl) != + qed_chain_get_prod_idx(&txq->tx_pbl)) { + DP_VERBOSE(edev, NETIF_MSG_IFDOWN, + "Freeing a packet on tx queue[%d]: chain_cons 0x%x, chain_prod 0x%x\n", + txq->index, qed_chain_get_cons_idx(&txq->tx_pbl), + qed_chain_get_prod_idx(&txq->tx_pbl)); + + rc = qede_free_tx_pkt(edev, txq, &len); + if (rc) { + DP_NOTICE(edev, + "Failed to free a packet on tx queue[%d]: chain_cons 0x%x, chain_prod 0x%x\n", + txq->index, + qed_chain_get_cons_idx(&txq->tx_pbl), + qed_chain_get_prod_idx(&txq->tx_pbl)); + break; + } + + bytes_compl += len; + pkts_compl++; + txq->sw_tx_cons++; + } + + netdev_tx_completed_queue(netdev_txq, pkts_compl, bytes_compl); +} + +static void qede_empty_tx_queues(struct qede_dev *edev) +{ + int i; + + for_each_queue(i) + if (edev->fp_array[i].type & QEDE_FASTPATH_TX) { + int cos; + + for_each_cos_in_txq(edev, cos) { + struct qede_fastpath *fp; + + fp = &edev->fp_array[i]; + qede_empty_tx_queue(edev, + &fp->txq[cos]); + } + } +} + /* This function inits fp content and resets the SB, RXQ and TXQ structures */ static void qede_init_fp(struct qede_dev *edev) { @@ -2053,6 +2161,7 @@ out: enum qede_unload_mode { QEDE_UNLOAD_NORMAL, + QEDE_UNLOAD_RECOVERY, }; static void qede_unload(struct qede_dev *edev, enum qede_unload_mode mode, @@ -2068,7 +2177,8 @@ static void qede_unload(struct qede_dev *edev, enum qede_unload_mode mode, clear_bit(QEDE_FLAGS_LINK_REQUESTED, &edev->flags); - edev->state = QEDE_STATE_CLOSED; + if (mode != QEDE_UNLOAD_RECOVERY) + edev->state = QEDE_STATE_CLOSED; qede_rdma_dev_event_close(edev); @@ -2076,17 +2186,20 @@ static void qede_unload(struct qede_dev *edev, enum qede_unload_mode mode, netif_tx_disable(edev->ndev); netif_carrier_off(edev->ndev); - /* Reset the link */ - memset(&link_params, 0, sizeof(link_params)); - link_params.link_up = false; - edev->ops->common->set_link(edev->cdev, &link_params); - rc = qede_stop_queues(edev); - if (rc) { - qede_sync_free_irqs(edev); - goto out; - } + if (mode != QEDE_UNLOAD_RECOVERY) { + /* Reset the link */ + memset(&link_params, 0, sizeof(link_params)); + link_params.link_up = false; + edev->ops->common->set_link(edev->cdev, &link_params); - DP_INFO(edev, "Stopped Queues\n"); + rc = qede_stop_queues(edev); + if (rc) { + qede_sync_free_irqs(edev); + goto out; + } + + DP_INFO(edev, "Stopped Queues\n"); + } qede_vlan_mark_nonconfigured(edev); edev->ops->fastpath_stop(edev->cdev); @@ -2102,18 +2215,26 @@ static void qede_unload(struct qede_dev *edev, enum qede_unload_mode mode, qede_napi_disable_remove(edev); + if (mode == QEDE_UNLOAD_RECOVERY) + qede_empty_tx_queues(edev); + qede_free_mem_load(edev); qede_free_fp_array(edev); out: if (!is_locked) __qede_unlock(edev); + + if (mode != QEDE_UNLOAD_RECOVERY) + DP_NOTICE(edev, "Link is down\n"); + DP_INFO(edev, "Ending qede unload\n"); } enum qede_load_mode { QEDE_LOAD_NORMAL, QEDE_LOAD_RELOAD, + QEDE_LOAD_RECOVERY, }; static int qede_load(struct qede_dev *edev, enum qede_load_mode mode, @@ -2293,6 +2414,77 @@ static void qede_link_update(void *dev, struct qed_link_output *link) } } +static void qede_schedule_recovery_handler(void *dev) +{ + struct qede_dev *edev = dev; + + if (edev->state == QEDE_STATE_RECOVERY) { + DP_NOTICE(edev, + "Avoid scheduling a recovery handling since already in recovery state\n"); + return; + } + + set_bit(QEDE_SP_RECOVERY, &edev->sp_flags); + schedule_delayed_work(&edev->sp_task, 0); + + DP_INFO(edev, "Scheduled a recovery handler\n"); +} + +static void qede_recovery_failed(struct qede_dev *edev) +{ + netdev_err(edev->ndev, "Recovery handling has failed. Power cycle is needed.\n"); + + netif_device_detach(edev->ndev); + + if (edev->cdev) + edev->ops->common->set_power_state(edev->cdev, PCI_D3hot); +} + +static void qede_recovery_handler(struct qede_dev *edev) +{ + u32 curr_state = edev->state; + int rc; + + DP_NOTICE(edev, "Starting a recovery process\n"); + + /* No need to acquire first the qede_lock since is done by qede_sp_task + * before calling this function. + */ + edev->state = QEDE_STATE_RECOVERY; + + edev->ops->common->recovery_prolog(edev->cdev); + + if (curr_state == QEDE_STATE_OPEN) + qede_unload(edev, QEDE_UNLOAD_RECOVERY, true); + + __qede_remove(edev->pdev, QEDE_REMOVE_RECOVERY); + + rc = __qede_probe(edev->pdev, edev->dp_module, edev->dp_level, + IS_VF(edev), QEDE_PROBE_RECOVERY); + if (rc) { + edev->cdev = NULL; + goto err; + } + + if (curr_state == QEDE_STATE_OPEN) { + rc = qede_load(edev, QEDE_LOAD_RECOVERY, true); + if (rc) + goto err; + + qede_config_rx_mode(edev->ndev); + udp_tunnel_get_rx_info(edev->ndev); + } + + edev->state = curr_state; + + DP_NOTICE(edev, "Recovery handling is done\n"); + + return; + +err: + qede_recovery_failed(edev); +} + static bool qede_is_txq_full(struct qede_dev *edev, struct qede_tx_queue *txq) { struct netdev_queue *netdev_txq; diff --git a/drivers/net/ethernet/qlogic/qede/qede_rdma.c b/drivers/net/ethernet/qlogic/qede/qede_rdma.c index 1900bf7e67d1..ffabc2d2f082 100644 --- a/drivers/net/ethernet/qlogic/qede/qede_rdma.c +++ b/drivers/net/ethernet/qlogic/qede/qede_rdma.c @@ -50,6 +50,8 @@ static void _qede_rdma_dev_add(struct qede_dev *edev) if (!qedr_drv) return; + /* Leftovers from previous error recovery */ + edev->rdma_info.exp_recovery = false; edev->rdma_info.qedr_dev = qedr_drv->add(edev->cdev, edev->pdev, edev->ndev); } @@ -87,21 +89,26 @@ static void qede_rdma_destroy_wq(struct qede_dev *edev) destroy_workqueue(edev->rdma_info.rdma_wq); } -int qede_rdma_dev_add(struct qede_dev *edev) +int qede_rdma_dev_add(struct qede_dev *edev, bool recovery) { - int rc = 0; + int rc; - if (qede_rdma_supported(edev)) { - rc = qede_rdma_create_wq(edev); - if (rc) - return rc; + if (!qede_rdma_supported(edev)) + return 0; - INIT_LIST_HEAD(&edev->rdma_info.entry); - mutex_lock(&qedr_dev_list_lock); - list_add_tail(&edev->rdma_info.entry, &qedr_dev_list); - _qede_rdma_dev_add(edev); - mutex_unlock(&qedr_dev_list_lock); - } + /* Cannot start qedr while recovering since it wasn't fully stopped */ + if (recovery) + return 0; + + rc = qede_rdma_create_wq(edev); + if (rc) + return rc; + + INIT_LIST_HEAD(&edev->rdma_info.entry); + mutex_lock(&qedr_dev_list_lock); + list_add_tail(&edev->rdma_info.entry, &qedr_dev_list); + _qede_rdma_dev_add(edev); + mutex_unlock(&qedr_dev_list_lock); return rc; } @@ -110,19 +117,30 @@ static void _qede_rdma_dev_remove(struct qede_dev *edev) { if (qedr_drv && qedr_drv->remove && edev->rdma_info.qedr_dev) qedr_drv->remove(edev->rdma_info.qedr_dev); - edev->rdma_info.qedr_dev = NULL; } -void qede_rdma_dev_remove(struct qede_dev *edev) +void qede_rdma_dev_remove(struct qede_dev *edev, bool recovery) { if (!qede_rdma_supported(edev)) return; - qede_rdma_destroy_wq(edev); - mutex_lock(&qedr_dev_list_lock); - _qede_rdma_dev_remove(edev); - list_del(&edev->rdma_info.entry); - mutex_unlock(&qedr_dev_list_lock); + /* Cannot remove qedr while recovering since it wasn't fully stopped */ + if (!recovery) { + qede_rdma_destroy_wq(edev); + mutex_lock(&qedr_dev_list_lock); + if (!edev->rdma_info.exp_recovery) + _qede_rdma_dev_remove(edev); + edev->rdma_info.qedr_dev = NULL; + list_del(&edev->rdma_info.entry); + mutex_unlock(&qedr_dev_list_lock); + } else { + if (!edev->rdma_info.exp_recovery) { + mutex_lock(&qedr_dev_list_lock); + _qede_rdma_dev_remove(edev); + mutex_unlock(&qedr_dev_list_lock); + } + edev->rdma_info.exp_recovery = true; + } } static void _qede_rdma_dev_open(struct qede_dev *edev) @@ -204,7 +222,8 @@ void qede_rdma_unregister_driver(struct qedr_driver *drv) mutex_lock(&qedr_dev_list_lock); list_for_each_entry(edev, &qedr_dev_list, rdma_info.entry) { - if (edev->rdma_info.qedr_dev) + /* If device has experienced recovery it was already removed */ + if (edev->rdma_info.qedr_dev && !edev->rdma_info.exp_recovery) _qede_rdma_dev_remove(edev); } qedr_drv = NULL; @@ -284,6 +303,10 @@ static void qede_rdma_add_event(struct qede_dev *edev, { struct qede_rdma_event_work *event_node; + /* If a recovery was experienced avoid adding the event */ + if (edev->rdma_info.exp_recovery) + return; + if (!edev->rdma_info.qedr_dev) return; -- cgit v1.2.3