scsi: mpi3mr: Record and report controller firmware faults

Capture and retain firmware fault codes and extended fault information
whenever the controller enters a fault state.

Maintain a persistent firmware fault counter, expose it via sysfs, and
generate uevents to aid userspace diagnostics and failure analysis.

Co-developed-by: Salomon Dushimirimana <salomondush@google.com>
Signed-off-by: Salomon Dushimirimana <salomondush@google.com>
Signed-off-by: Ranjan Kumar <ranjan.kumar@broadcom.com>
Link: https://patch.msgid.link/20260116060719.32937-7-ranjan.kumar@broadcom.com
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

Author: Ranjan Kumar

drivers/scsi/mpi3mr/mpi3mr.h

@@ -1137,6 +1137,10 @@ struct scmd_priv {
  * @default_qcount: Total Default queues
  * @active_poll_qcount: Currently active poll queue count
  * @requested_poll_qcount: User requested poll queue count
+ * @fault_during_init: Indicates a firmware fault occurred during initialization
+ * @saved_fault_code: Firmware fault code captured at the time of failure
+ * @saved_fault_info: Additional firmware-provided fault information
+ * @fwfault_counter: Count of firmware faults detected by the driver
  * @bsg_dev: BSG device structure
  * @bsg_queue: Request queue for BSG device
  * @stop_bsgs: Stop BSG request flag
@@ -1340,6 +1344,10 @@ struct mpi3mr_ioc {
 	u16 default_qcount;
 	u16 active_poll_qcount;
 	u16 requested_poll_qcount;
+	u8 fault_during_init;
+	u32 saved_fault_code;
+	u32 saved_fault_info[3];
+	u64 fwfault_counter;
 
 	struct device bsg_dev;
 	struct request_queue *bsg_queue;

drivers/scsi/mpi3mr/mpi3mr_app.c

@@ -3255,6 +3255,29 @@ adp_state_show(struct device *dev, struct device_attribute *attr,
 
 static DEVICE_ATTR_RO(adp_state);
 
+/**
+ * fwfault_count_show() - SysFS callback to show firmware fault count
+ * @dev: class device
+ * @attr: Device attribute
+ * @buf: Buffer to copy data into
+ *
+ * Displays the total number of firmware faults detected by the driver
+ * since the controller was initialized.
+ *
+ * Return: Number of bytes written to @buf
+ */
+
+static ssize_t
+fwfault_count_show(struct device *dev, struct device_attribute *attr,
+	char *buf)
+{
+	struct Scsi_Host *shost = class_to_shost(dev);
+	struct mpi3mr_ioc *mrioc = shost_priv(shost);
+
+	return snprintf(buf, PAGE_SIZE, "%llu\n", mrioc->fwfault_counter);
+}
+static DEVICE_ATTR_RO(fwfault_count);
+
 static struct attribute *mpi3mr_host_attrs[] = {
 	&dev_attr_version_fw.attr,
 	&dev_attr_fw_queue_depth.attr,
@@ -3263,6 +3286,7 @@ static struct attribute *mpi3mr_host_attrs[] = {
 	&dev_attr_reply_qfull_count.attr,
 	&dev_attr_logging_level.attr,
 	&dev_attr_adp_state.attr,
+	&dev_attr_fwfault_count.attr,
 	NULL,
 };
 

drivers/scsi/mpi3mr/mpi3mr_fw.c

@@ -1108,6 +1108,31 @@ void mpi3mr_print_fault_info(struct mpi3mr_ioc *mrioc)
 	}
 }
 
+/**
+ * mpi3mr_save_fault_info - Save fault information
+ * @mrioc: Adapter instance reference
+ *
+ * Save the controller fault information if there is a
+ * controller fault.
+ *
+ * Return: Nothing.
+ */
+static void mpi3mr_save_fault_info(struct mpi3mr_ioc *mrioc)
+{
+	u32 ioc_status, i;
+
+	ioc_status = readl(&mrioc->sysif_regs->ioc_status);
+
+	if (ioc_status & MPI3_SYSIF_IOC_STATUS_FAULT) {
+		mrioc->saved_fault_code = readl(&mrioc->sysif_regs->fault) &
+		    MPI3_SYSIF_FAULT_CODE_MASK;
+		for (i = 0; i < 3; i++) {
+			mrioc->saved_fault_info[i] =
+			readl(&mrioc->sysif_regs->fault_info[i]);
+		}
+	}
+}
+
 /**
  * mpi3mr_get_iocstate - Get IOC State
  * @mrioc: Adapter instance reference
@@ -1249,6 +1274,60 @@ static void mpi3mr_alloc_ioctl_dma_memory(struct mpi3mr_ioc *mrioc)
 	mpi3mr_free_ioctl_dma_memory(mrioc);
 }
 
+/**
+ * mpi3mr_fault_uevent_emit - Emit uevent for any controller
+ * fault
+ * @mrioc: Pointer to the mpi3mr_ioc structure for the controller instance
+ *
+ * This function is invoked when the controller undergoes any
+ * type of fault.
+ */
+
+static void mpi3mr_fault_uevent_emit(struct mpi3mr_ioc *mrioc)
+{
+	struct kobj_uevent_env *env;
+	int ret;
+
+	env = kzalloc(sizeof(*env), GFP_KERNEL);
+	if (!env)
+		return;
+
+	ret = add_uevent_var(env, "DRIVER=%s", mrioc->driver_name);
+	if (ret)
+		goto out_free;
+
+	ret = add_uevent_var(env, "IOC_ID=%u", mrioc->id);
+	if (ret)
+		goto out_free;
+
+	ret = add_uevent_var(env, "FAULT_CODE=0x%08x",
+			    mrioc->saved_fault_code);
+	if (ret)
+		goto out_free;
+
+	ret = add_uevent_var(env, "FAULT_INFO0=0x%08x",
+			     mrioc->saved_fault_info[0]);
+	if (ret)
+		goto out_free;
+
+	ret = add_uevent_var(env, "FAULT_INFO1=0x%08x",
+			    mrioc->saved_fault_info[1]);
+	if (ret)
+		goto out_free;
+
+	ret = add_uevent_var(env, "FAULT_INFO2=0x%08x",
+			    mrioc->saved_fault_info[2]);
+	if (ret)
+		goto out_free;
+
+	kobject_uevent_env(&mrioc->shost->shost_gendev.kobj,
+			KOBJ_CHANGE, env->envp);
+
+out_free:
+	kfree(env);
+
+}
+
 /**
  * mpi3mr_clear_reset_history - clear reset history
  * @mrioc: Adapter instance reference
@@ -1480,6 +1559,10 @@ static int mpi3mr_bring_ioc_ready(struct mpi3mr_ioc *mrioc)
 		if (ioc_state == MRIOC_STATE_FAULT) {
 			timeout = MPI3_SYSIF_DIAG_SAVE_TIMEOUT * 10;
 			mpi3mr_print_fault_info(mrioc);
+			mpi3mr_save_fault_info(mrioc);
+			mrioc->fault_during_init = 1;
+			mrioc->fwfault_counter++;
+
 			do {
 				host_diagnostic =
 					readl(&mrioc->sysif_regs->host_diagnostic);
@@ -2577,6 +2660,9 @@ void mpi3mr_check_rh_fault_ioc(struct mpi3mr_ioc *mrioc, u32 reason_code)
 		mpi3mr_set_trigger_data_in_all_hdb(mrioc,
 		    MPI3MR_HDB_TRIGGER_TYPE_FAULT, &trigger_data, 0);
 		mpi3mr_print_fault_info(mrioc);
+		mpi3mr_save_fault_info(mrioc);
+		mrioc->fault_during_init = 1;
+		mrioc->fwfault_counter++;
 		return;
 	}
 
@@ -2594,6 +2680,10 @@ void mpi3mr_check_rh_fault_ioc(struct mpi3mr_ioc *mrioc, u32 reason_code)
 			break;
 		msleep(100);
 	} while (--timeout);
+
+	mpi3mr_save_fault_info(mrioc);
+	mrioc->fault_during_init = 1;
+	mrioc->fwfault_counter++;
 }
 
 /**
@@ -2770,6 +2860,11 @@ static void mpi3mr_watchdog_work(struct work_struct *work)
 	union mpi3mr_trigger_data trigger_data;
 	u16 reset_reason = MPI3MR_RESET_FROM_FAULT_WATCH;
 
+	if (mrioc->fault_during_init) {
+		mpi3mr_fault_uevent_emit(mrioc);
+		mrioc->fault_during_init = 0;
+	}
+
 	if (mrioc->reset_in_progress || mrioc->pci_err_recovery)
 		return;
 
@@ -2842,6 +2937,10 @@ static void mpi3mr_watchdog_work(struct work_struct *work)
 		goto schedule_work;
 	}
 
+	mpi3mr_save_fault_info(mrioc);
+	mpi3mr_fault_uevent_emit(mrioc);
+	mrioc->fwfault_counter++;
+
 	switch (trigger_data.fault) {
 	case MPI3_SYSIF_FAULT_CODE_COMPLETE_RESET_NEEDED:
 	case MPI3_SYSIF_FAULT_CODE_POWER_CYCLE_REQUIRED:
@@ -5478,6 +5577,10 @@ int mpi3mr_soft_reset_handler(struct mpi3mr_ioc *mrioc,
 					break;
 				msleep(100);
 			} while (--timeout);
+
+			mpi3mr_save_fault_info(mrioc);
+			mpi3mr_fault_uevent_emit(mrioc);
+			mrioc->fwfault_counter++;
 			mpi3mr_set_trigger_data_in_all_hdb(mrioc,
 			    MPI3MR_HDB_TRIGGER_TYPE_FAULT, &trigger_data, 0);
 		}