Skip to contentSkip to conversation

qedi: Misc bug fixes and enhancements

PATCH Applied Manish Rangankar 2 participants 11 messages Sep 8, 2020 – Sep 16, 2020

Subsystems (2)
lore
Share & tools
lore

Manish Rangankar

author maintainer
#0 lore raw

v2 cover qedi: Misc bug fixes and enhancements

Hi Martin,

Please apply the qedi miscellaneous bug fixes and enhancement patches to the scsi tree at your convenience.

v1->v2:
Fix warning reported by kernel test robot

Thanks,
Manish

Manish Rangankar (5):
qedi: Use qed count from set_fp_int in msix allocation.
qedi: Skip f/w connection termination for pci shutdown handler.
qedi: Use snprintf instead of sprintf
qedi: Add firmware error recovery invocation support.
qedi: Add support for handling the pcie errors.

Nilesh Javali (3):
qedi: Fix list_del corruption while removing active IO
qedi: Protect active command list to avoid list corruption
qedi: Mark all connections for recovery on link down event

 drivers/scsi/qedi/qedi.h       |   5 ++
 drivers/scsi/qedi/qedi_fw.c    |  30 +++++++--
 drivers/scsi/qedi/qedi_iscsi.c |   7 +++
 drivers/scsi/qedi/qedi_main.c  | 108 +++++++++++++++++++++++++++++++--
 4 files changed, 140 insertions(+), 10 deletions(-)
Signature · 1 line
2.25.0
author maintainer
#1 lore raw

PATCH v2 1/8 qedi: Use qed count from set_fp_int in msix allocation.

To avoid unnecessary vector allocation when number of fast-path queues are less then available msix vectors, use return count from module qed->set_fp_int.

---
 drivers/scsi/qedi/qedi.h      | 1 +
 drivers/scsi/qedi/qedi_main.c | 9 +++++++--
 2 files changed, 8 insertions(+), 2 deletions(-)
drivers/scsi/qedi/qedi.h +1 -0
index 9498279ae80d..9c19ec9dc682 100644
@@ -305,6 +305,7 @@ struct qedi_ctx {
 	u32 max_sqes;
 	u8 num_queues;
 	u32 max_active_conns;
+	s32 msix_count;
 
 	struct iscsi_cid_queue cid_que;
 	struct qedi_endpoint **ep_tbl;
drivers/scsi/qedi/qedi_main.c +7 -2
index 6f038ae5efca..e1ec22d7d699 100644
@@ -1357,7 +1357,7 @@ static int qedi_request_msix_irq(struct qedi_ctx *qedi)
 	u16 idx;
 
 	cpu = cpumask_first(cpu_online_mask);
-	for (i = 0; i < qedi->int_info.msix_cnt; i++) {
+	for (i = 0; i < qedi->msix_count; i++) {
 		idx = i * qedi->dev_info.common.num_hwfns +
 			  qedi_ops->common->get_affin_hwfn_idx(qedi->cdev);
 
@@ -1387,7 +1387,12 @@ static int qedi_setup_int(struct qedi_ctx *qedi)
 {
 	int rc = 0;
 
-	rc = qedi_ops->common->set_fp_int(qedi->cdev, num_online_cpus());
+	rc = qedi_ops->common->set_fp_int(qedi->cdev, qedi->num_queues);
+	if (rc < 0)
+		goto exit_setup_int;
+
+	qedi->msix_count = rc;
+
 	rc = qedi_ops->common->get_fp_int(qedi->cdev, &qedi->int_info);
 	if (rc)
 		goto exit_setup_int;
Signature · 1 line
2.25.0
author maintainer
#2 lore raw

PATCH v2 2/8 qedi: Skip f/w connection termination for pci shutdown handler.

In reply to Manish Rangankar

In boot from san scenario when qedi pci shutdown handler is called with active iSCSI sessions, sometimes target takes too long time to respond to f/w connection termination request. Instead skip sending termination ramrod and progress with unload path.

---
 drivers/scsi/qedi/qedi_iscsi.c | 4 ++++
 1 file changed, 4 insertions(+)
drivers/scsi/qedi/qedi_iscsi.c +4 -0
index c14ac7882afa..f815845fc568 100644
@@ -1069,6 +1069,10 @@ static void qedi_ep_disconnect(struct iscsi_endpoint *ep)
 		wait_delay += qedi->pf_params.iscsi_pf_params.two_msl_timer;
 
 	qedi_ep->state = EP_STATE_DISCONN_START;
+
+	if (test_bit(QEDI_IN_SHUTDOWN, &qedi->flags))
+		goto ep_release_conn;
+
 	ret = qedi_ops->destroy_conn(qedi->cdev, qedi_ep->handle, abrt_conn);
 	if (ret) {
 		QEDI_WARN(&qedi->dbg_ctx,
Signature · 1 line
2.25.0
author maintainer
#3 lore raw

PATCH v2 3/8 qedi: Fix list_del corruption while removing active IO

In reply to Manish Rangankar

From: Nilesh Javali <[email protected]>

While aborting the IO, the FW cleanup task timed out and driver deleted the IO from active command list. Some time later, the FW sends the cleanup task response and driver again deletes the IO from active command list causing FW to send IO completion for non-existent IO and list_del corruption of active command list.

Add fix to check if IO is present before deleting the i/o from active command list ensuring FW sends valid i/o completion and protect the list_del corruption.

---
 drivers/scsi/qedi/qedi_fw.c | 15 +++++++++++----
 1 file changed, 11 insertions(+), 4 deletions(-)
drivers/scsi/qedi/qedi_fw.c +11 -4
index 6ed74583b1b9..34c477bda8a4 100644
@@ -816,8 +816,11 @@ static void qedi_process_cmd_cleanup_resp(struct qedi_ctx *qedi,
 			qedi_clear_task_idx(qedi_conn->qedi, rtid);
 
 			spin_lock(&qedi_conn->list_lock);
-			list_del_init(&dbg_cmd->io_cmd);
-			qedi_conn->active_cmd_count--;
+			if (likely(dbg_cmd->io_cmd_in_list)) {
+				dbg_cmd->io_cmd_in_list = false;
+				list_del_init(&dbg_cmd->io_cmd);
+				qedi_conn->active_cmd_count--;
+			}
 			spin_unlock(&qedi_conn->list_lock);
 			qedi_cmd->state = CLEANUP_RECV;
 			wake_up_interruptible(&qedi_conn->wait_queue);
@@ -1235,6 +1238,7 @@ int qedi_cleanup_all_io(struct qedi_ctx *qedi, struct qedi_conn *qedi_conn,
 		qedi_conn->cmd_cleanup_req++;
 		qedi_iscsi_cleanup_task(ctask, true);
 
+		cmd->io_cmd_in_list = false;
 		list_del_init(&cmd->io_cmd);
 		qedi_conn->active_cmd_count--;
 		QEDI_WARN(&qedi->dbg_ctx,
@@ -1446,8 +1450,11 @@ static void qedi_tmf_work(struct work_struct *work)
 	spin_unlock_bh(&qedi_conn->tmf_work_lock);
 
 	spin_lock(&qedi_conn->list_lock);
-	list_del_init(&cmd->io_cmd);
-	qedi_conn->active_cmd_count--;
+	if (likely(cmd->io_cmd_in_list)) {
+		cmd->io_cmd_in_list = false;
+		list_del_init(&cmd->io_cmd);
+		qedi_conn->active_cmd_count--;
+	}
 	spin_unlock(&qedi_conn->list_lock);
 
 	clear_bit(QEDI_CONN_FW_CLEANUP, &qedi_conn->flags);
Signature · 1 line
2.25.0
author maintainer
#4 lore raw

PATCH v2 4/8 qedi: Protect active command list to avoid list corruption

In reply to Manish Rangankar

From: Nilesh Javali <[email protected]>

Protect active command list for non i/o commands like, login response, logout response, text response and recovery cleanup of active list to avoid list corruption.

---
 drivers/scsi/qedi/qedi_fw.c    | 8 ++++++++
 drivers/scsi/qedi/qedi_iscsi.c | 2 ++
 2 files changed, 10 insertions(+)
drivers/scsi/qedi/qedi_fw.c +8 -0
index 34c477bda8a4..f158fde0a43c 100644
@@ -59,6 +59,7 @@ static void qedi_process_logout_resp(struct qedi_ctx *qedi,
 		  "Freeing tid=0x%x for cid=0x%x\n",
 		  cmd->task_id, qedi_conn->iscsi_conn_id);
 
+	spin_lock(&qedi_conn->list_lock);
 	if (likely(cmd->io_cmd_in_list)) {
 		cmd->io_cmd_in_list = false;
 		list_del_init(&cmd->io_cmd);
@@ -69,6 +70,7 @@ static void qedi_process_logout_resp(struct qedi_ctx *qedi,
 			  cmd->task_id, qedi_conn->iscsi_conn_id,
 			  &cmd->io_cmd);
 	}
+	spin_unlock(&qedi_conn->list_lock);
 
 	cmd->state = RESPONSE_RECEIVED;
 	qedi_clear_task_idx(qedi, cmd->task_id);
@@ -122,6 +124,7 @@ static void qedi_process_text_resp(struct qedi_ctx *qedi,
 		  "Freeing tid=0x%x for cid=0x%x\n",
 		  cmd->task_id, qedi_conn->iscsi_conn_id);
 
+	spin_lock(&qedi_conn->list_lock);
 	if (likely(cmd->io_cmd_in_list)) {
 		cmd->io_cmd_in_list = false;
 		list_del_init(&cmd->io_cmd);
@@ -132,6 +135,7 @@ static void qedi_process_text_resp(struct qedi_ctx *qedi,
 			  cmd->task_id, qedi_conn->iscsi_conn_id,
 			  &cmd->io_cmd);
 	}
+	spin_unlock(&qedi_conn->list_lock);
 
 	cmd->state = RESPONSE_RECEIVED;
 	qedi_clear_task_idx(qedi, cmd->task_id);
@@ -222,11 +226,13 @@ static void qedi_process_tmf_resp(struct qedi_ctx *qedi,
 
 	tmf_hdr = (struct iscsi_tm *)qedi_cmd->task->hdr;
 
+	spin_lock(&qedi_conn->list_lock);
 	if (likely(qedi_cmd->io_cmd_in_list)) {
 		qedi_cmd->io_cmd_in_list = false;
 		list_del_init(&qedi_cmd->io_cmd);
 		qedi_conn->active_cmd_count--;
 	}
+	spin_unlock(&qedi_conn->list_lock);
 
 	if (((tmf_hdr->flags & ISCSI_FLAG_TM_FUNC_MASK) ==
 	      ISCSI_TM_FUNC_LOGICAL_UNIT_RESET) ||
@@ -288,11 +294,13 @@ static void qedi_process_login_resp(struct qedi_ctx *qedi,
 		  ISCSI_LOGIN_RESPONSE_HDR_DATA_SEG_LEN_MASK;
 	qedi_conn->gen_pdu.resp_wr_ptr = qedi_conn->gen_pdu.resp_buf + pld_len;
 
+	spin_lock(&qedi_conn->list_lock);
 	if (likely(cmd->io_cmd_in_list)) {
 		cmd->io_cmd_in_list = false;
 		list_del_init(&cmd->io_cmd);
 		qedi_conn->active_cmd_count--;
 	}
+	spin_unlock(&qedi_conn->list_lock);
 
 	memset(task_ctx, '\0', sizeof(*task_ctx));
 
drivers/scsi/qedi/qedi_iscsi.c +2 -0
index f815845fc568..ae86a40ca040 100644
@@ -975,11 +975,13 @@ static void qedi_cleanup_active_cmd_list(struct qedi_conn *qedi_conn)
 {
 	struct qedi_cmd *cmd, *cmd_tmp;
 
+	spin_lock(&qedi_conn->list_lock);
 	list_for_each_entry_safe(cmd, cmd_tmp, &qedi_conn->active_cmd_list,
 				 io_cmd) {
 		list_del_init(&cmd->io_cmd);
 		qedi_conn->active_cmd_count--;
 	}
+	spin_unlock(&qedi_conn->list_lock);
 }
 
 static void qedi_ep_disconnect(struct iscsi_endpoint *ep)
Signature · 1 line
2.25.0
author maintainer
#5 lore raw

PATCH v2 5/8 qedi: Use snprintf instead of sprintf

In reply to Manish Rangankar

Use snprintf to limit max number of bytes to the buffer.

---
 drivers/scsi/qedi/qedi_main.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/scsi/qedi/qedi_main.c +1 -1
index e1ec22d7d699..2db99613b8a9 100644
@@ -2538,7 +2538,7 @@ static int __qedi_probe(struct pci_dev *pdev, int mode)
 	QEDI_INFO(&qedi->dbg_ctx, QEDI_LOG_DISC, "MAC address is %pM.\n",
 		  qedi->mac);
 
-	sprintf(host_buf, "host_%d", qedi->shost->host_no);
+	snprintf(host_buf, sizeof(host_buf), "host_%d", qedi->shost->host_no);
 	qedi_ops->common->set_name(qedi->cdev, host_buf);
 
 	qedi_ops->register_ops(qedi->cdev, &qedi_cb_ops, qedi);
Signature · 1 line
2.25.0
author maintainer
#6 lore raw

PATCH v2 6/8 qedi: Mark all connections for recovery on link down event

In reply to Manish Rangankar

From: Nilesh Javali <[email protected]>

For short time cable pulls, the in-flight i/o to the FW,
is never cleaned up, resulting in the behaviour of stale i/o completion causing list_del corruption and soft lockup of the system. So, on link down event, mark all the connections for recovery, causing cleanup of all the in-flight i/o immediately.

---
 drivers/scsi/qedi/qedi_main.c | 10 ++++++++++
 1 file changed, 10 insertions(+)
drivers/scsi/qedi/qedi_main.c +10 -0
index 2db99613b8a9..4f43e2a24b50 100644
@@ -1127,6 +1127,15 @@ static void qedi_schedule_recovery_handler(void *dev)
 	schedule_delayed_work(&qedi->recovery_work, 0);
 }
 
+static void qedi_set_conn_recovery(struct iscsi_cls_session *cls_session)
+{
+	struct iscsi_session *session = cls_session->dd_data;
+	struct iscsi_conn *conn = session->leadconn;
+	struct qedi_conn *qedi_conn = conn->dd_data;
+
+	qedi_start_conn_recovery(qedi_conn->qedi, qedi_conn);
+}
+
 static void qedi_link_update(void *dev, struct qed_link_output *link)
 {
 	struct qedi_ctx *qedi = (struct qedi_ctx *)dev;
@@ -1138,6 +1147,7 @@ static void qedi_link_update(void *dev, struct qed_link_output *link)
 		QEDI_INFO(&qedi->dbg_ctx, QEDI_LOG_INFO,
 			  "Link Down event.\n");
 		atomic_set(&qedi->link_state, QEDI_LINK_DOWN);
+		iscsi_host_for_each_session(qedi->shost, qedi_set_conn_recovery);
 	}
 }
 
Signature · 1 line
2.25.0
author maintainer
#7 lore raw

PATCH v2 7/8 qedi: Add firmware error recovery invocation support.

In reply to Manish Rangankar

Add support to initiate MFW process recovery for all the devices if storage function receive the event first.

Also added fix for kernel test robot warning,

Quoted 2 lines
Quoted 2 lines

drivers/scsi/qedi/qedi_main.c:1119:6: warning: no previous prototype for 'qedi_schedule_hw_err_handler' [-Wmissing-prototypes]

---
 drivers/scsi/qedi/qedi.h       |  4 +++
 drivers/scsi/qedi/qedi_fw.c    |  7 ++--
 drivers/scsi/qedi/qedi_iscsi.c |  3 +-
 drivers/scsi/qedi/qedi_main.c  | 63 +++++++++++++++++++++++++++++++++-
 4 files changed, 73 insertions(+), 4 deletions(-)
drivers/scsi/qedi/qedi.h +4 -0
index 9c19ec9dc682..7e59d50f2fab 100644
@@ -274,6 +274,10 @@ struct qedi_ctx {
 	spinlock_t ll2_lock;	/* Light L2 lock */
 	spinlock_t hba_lock;	/* per port lock */
 	struct task_struct *ll2_recv_thread;
+	unsigned long qedi_err_flags;
+#define QEDI_ERR_ATTN_CLR_EN	0
+#define QEDI_ERR_IS_RECOVERABLE	2
+#define QEDI_ERR_OVERRIDE_EN	31
 	unsigned long flags;
 #define UIO_DEV_OPENED		1
 #define QEDI_IOTHREAD_WAKE	2
drivers/scsi/qedi/qedi_fw.c +5 -2
index f158fde0a43c..440ddd2309f1 100644
@@ -1267,7 +1267,8 @@ int qedi_cleanup_all_io(struct qedi_ctx *qedi, struct qedi_conn *qedi_conn,
 	rval  = wait_event_interruptible_timeout(qedi_conn->wait_queue,
 						 ((qedi_conn->cmd_cleanup_req ==
 						 qedi_conn->cmd_cleanup_cmpl) ||
-						 qedi_conn->ep),
+						 test_bit(QEDI_IN_RECOVERY,
+							  &qedi->flags)),
 						 5 * HZ);
 	if (rval) {
 		QEDI_INFO(&qedi->dbg_ctx, QEDI_LOG_SCSI_TM,
@@ -1292,7 +1293,9 @@ int qedi_cleanup_all_io(struct qedi_ctx *qedi, struct qedi_conn *qedi_conn,
 	/* Enable IOs for all other sessions except current.*/
 	if (!wait_event_interruptible_timeout(qedi_conn->wait_queue,
 					      (qedi_conn->cmd_cleanup_req ==
-					       qedi_conn->cmd_cleanup_cmpl),
+					       qedi_conn->cmd_cleanup_cmpl) ||
+					       test_bit(QEDI_IN_RECOVERY,
+							&qedi->flags),
 					      5 * HZ)) {
 		iscsi_host_for_each_session(qedi->shost,
 					    qedi_mark_device_available);
drivers/scsi/qedi/qedi_iscsi.c +2 -1
index ae86a40ca040..08c05403cd72 100644
@@ -1072,7 +1072,8 @@ static void qedi_ep_disconnect(struct iscsi_endpoint *ep)
 
 	qedi_ep->state = EP_STATE_DISCONN_START;
 
-	if (test_bit(QEDI_IN_SHUTDOWN, &qedi->flags))
+	if (test_bit(QEDI_IN_SHUTDOWN, &qedi->flags) ||
+	    test_bit(QEDI_IN_RECOVERY, &qedi->flags))
 		goto ep_release_conn;
 
 	ret = qedi_ops->destroy_conn(qedi->cdev, qedi_ep->handle, abrt_conn);
drivers/scsi/qedi/qedi_main.c +62 -1

122 lines not on this page · raw message

Signature · 1 line
2.25.0
author maintainer
#8 lore raw

PATCH v2 8/8 qedi: Add support for handling the pcie errors.

In reply to Manish Rangankar

The error recovery is handled by management firmware (MFW) with the help of qed/qedi drivers. Upon detecting the errors, driver informs MFW about this event which in turn starts a recovery process. MFW sends ERROR_RECOVERY notification to the driver which performs the required cleanup/recovery from the driver side.

---
 drivers/scsi/qedi/qedi_main.c | 24 ++++++++++++++++++++++++
 1 file changed, 24 insertions(+)
drivers/scsi/qedi/qedi_main.c +24 -0
index 1d8c78c41405..10a7ee055552 100644
@@ -2391,6 +2391,25 @@ static int qedi_setup_boot_info(struct qedi_ctx *qedi)
 	return -ENOMEM;
 }
 
+static pci_ers_result_t qedi_io_error_detected(struct pci_dev *pdev,
+					       pci_channel_state_t state)
+{
+	struct qedi_ctx *qedi = pci_get_drvdata(pdev);
+
+	QEDI_ERR(&qedi->dbg_ctx, "%s: PCI error detected [%d]\n",
+		 __func__, state);
+
+	if (test_and_set_bit(QEDI_IN_RECOVERY, &qedi->flags)) {
+		QEDI_INFO(&qedi->dbg_ctx, QEDI_LOG_INFO,
+			  "Recovery already in progress.\n");
+		return PCI_ERS_RESULT_NONE;
+	}
+
+	qedi_ops->common->recovery_process(qedi->cdev);
+
+	return PCI_ERS_RESULT_CAN_RECOVER;
+}
+
 static void __qedi_remove(struct pci_dev *pdev, int mode)
 {
 	struct qedi_ctx *qedi = pci_get_drvdata(pdev);
@@ -2820,12 +2839,17 @@ MODULE_DEVICE_TABLE(pci, qedi_pci_tbl);
 
 static enum cpuhp_state qedi_cpuhp_state;
 
+static struct pci_error_handlers qedi_err_handler = {
+	.error_detected = qedi_io_error_detected,
+};
+
 static struct pci_driver qedi_pci_driver = {
 	.name = QEDI_MODULE_NAME,
 	.id_table = qedi_pci_tbl,
 	.probe = qedi_probe,
 	.remove = qedi_remove,
 	.shutdown = qedi_shutdown,
+	.err_handler = &qedi_err_handler,
 };
 
 static int __init qedi_init(void)
Signature · 1 line
2.25.0
#9 lore raw
In reply to Manish Rangankar

Manish,

Quoted 2 lines

Please apply the qedi miscellaneous bug fixes and enhancement patches to the scsi tree at your convenience.

Applied to the 5.10 SCSI staging tree. Thanks!

Signature · 1 line
Martin K. Petersen	Oracle Linux Engineering
#10 lore raw
In reply to Manish Rangankar
Quoted 10 lines · Manish Rangankar

Please apply the qedi miscellaneous bug fixes and enhancement patches to the scsi tree at your convenience.

v1->v2:
Fix warning reported by kernel test robot

Thanks,
Manish

[...]

Applied to 5.10/scsi-queue, thanks!

[1/8] scsi: qedi: Use qed count from set_fp_int in msix allocation

      https://git.kernel.org/mkp/scsi/c/3f8ad0072bf7
[2/8] scsi: qedi: Skip firmware connection termination for PCI shutdown handler
      https://git.kernel.org/mkp/scsi/c/5c35e4646566
[3/8] scsi: qedi: Fix list_del corruption while removing active I/O
      https://git.kernel.org/mkp/scsi/c/28b35d17f9f8
[4/8] scsi: qedi: Protect active command list to avoid list corruption
      https://git.kernel.org/mkp/scsi/c/c0650e28448d
[5/8] scsi: qedi: Use snprintf instead of sprintf
      https://git.kernel.org/mkp/scsi/c/5a2e69af16ce
[6/8] scsi: qedi: Mark all connections for recovery on link down event
      https://git.kernel.org/mkp/scsi/c/4118879be375
[7/8] scsi: qedi: Add firmware error recovery invocation support
      https://git.kernel.org/mkp/scsi/c/f4ba4e55db6d
[8/8] scsi: qedi: Add support for handling PCIe errors
      https://git.kernel.org/mkp/scsi/c/96a766a789eb
Signature · 1 line
Martin K. Petersen	Oracle Linux Engineering