Skip to contentSkip to conversation

pm80xx updates

PATCH Applied Viswas G 2 participants 11 messages Apr 15, 2021 – Apr 20, 2021

Subsystems (2)
lore
Share & tools
lore

Viswas G

author
#0 lore raw

v4 cover pm80xx updates

This patch set include some bug fixes and enhancements for pm80xx driver.

Changes from v3 to v4:
- For sysfs attribute patches
- Removed the unwanted variable 'c'
- For "Add sysfs attribute to check mpi state" patch
- split the ctl_mpi_state to two (ctl_mpi_state, ctl_hmi_error) as suggested.
- Changed static char mpiStateText[] to static const char *const mpiStateText[]

Changes from v2 to v3:
- For "Add sysfs attribute to check mpi state" patch
- Fixed "warn: mask and shift to zero"
- Made mpiStateText static, as suggested by buildbot

Changes from v1 to v2:
- For sysfs attribute patches
- Used sysfs_emit instead of sprintf for sysfs attribute patches.
- Removed debug message for sysfs attribute patches.
- For "Reset PI and CI memory during re-initialize" patch
- Improved commit message.

Ruksar Devadi (1):
pm80xx: Completing pending IO after fatal error

Vishakha Channapattan (5):
pm80xx: Add sysfs attribute to check mpi state
pm80xx: Add sysfs attribute to check controller hmi error
pm80xx: Add sysfs attribute to track RAAE count
pm80xx: Add sysfs attribute to track iop0 count
pm80xx: Add sysfs attribute to track iop1 count

Viswas G (2):
pm80xx: Reset PI and CI memory during re-initialize
pm80xx: remove global lock from outbound queue processing

 drivers/scsi/pm8001/pm8001_ctl.c  | 121 +++++++++++++++++++++++++++++++++++++-
 drivers/scsi/pm8001/pm8001_hwi.c  |  68 +++++++++++++++++++--
 drivers/scsi/pm8001/pm8001_hwi.h  |   1 +
 drivers/scsi/pm8001/pm8001_init.c |   9 ++-
 drivers/scsi/pm8001/pm8001_sas.c  |   2 +-
 drivers/scsi/pm8001/pm8001_sas.h  |   2 +
 drivers/scsi/pm8001/pm80xx_hwi.c  |   7 ++-
 drivers/scsi/pm8001/pm80xx_hwi.h  |   1 +
 8 files changed, 198 insertions(+), 13 deletions(-)
Signature · 1 line
2.16.3
author
#1 lore raw

PATCH v4 1/8 pm80xx: Add sysfs attribute to check mpi state

From: Vishakha Channapattan <[email protected]>

A new sysfs variable 'ctl_mpi_state' is being introduced to check the state of mpi.

Tested: Using 'ctl_mpi_state' sysfs variable we check the mpi state

linux-2dq0:~# cat /sys/class/scsi_host/host*/ctl_mpi_state
MPI is successfully initialized
---
 drivers/scsi/pm8001/pm8001_ctl.c | 32 +++++++++++++++++++++++++++++++-
 1 file changed, 31 insertions(+), 1 deletion(-)
drivers/scsi/pm8001/pm8001_ctl.c +31 -1
index 12035baf0997..a93799bd3a32 100644
@@ -41,6 +41,7 @@
 #include <linux/slab.h>
 #include "pm8001_sas.h"
 #include "pm8001_ctl.h"
+#include "pm8001_chips.h"
 
 /* scsi host attributes */
 
@@ -883,9 +884,37 @@ static ssize_t pm8001_show_update_fw(struct device *cdev,
 			flash_error_table[i].err_code,
 			flash_error_table[i].reason);
 }
-
 static DEVICE_ATTR(update_fw, S_IRUGO|S_IWUSR|S_IWGRP,
 	pm8001_show_update_fw, pm8001_store_update_fw);
+
+/**
+ * ctl_mpi_state_show - controller MPI state check
+ * @cdev: pointer to embedded class device
+ * @buf: the buffer returned
+ *
+ * A sysfs 'read-only' shost attribute.
+ */
+
+static const char *const mpiStateText[] = {
+	"MPI is not initialized",
+	"MPI is successfully initialized",
+	"MPI termination is in progress",
+	"MPI initialization failed with error in [31:16]"
+};
+
+static ssize_t ctl_mpi_state_show(struct device *cdev,
+		struct device_attribute *attr, char *buf)
+{
+	struct Scsi_Host *shost = class_to_shost(cdev);
+	struct sas_ha_struct *sha = SHOST_TO_SAS_HA(shost);
+	struct pm8001_hba_info *pm8001_ha = sha->lldd_ha;
+	unsigned int mpidw0;
+
+	mpidw0 = pm8001_mr32(pm8001_ha->general_stat_tbl_addr, 0);
+	return sysfs_emit(buf, "%s\n", mpiStateText[mpidw0 & 0x0003]);
+}
+static DEVICE_ATTR_RO(ctl_mpi_state);
+
 struct device_attribute *pm8001_host_attrs[] = {
 	&dev_attr_interface_rev,
 	&dev_attr_controller_fatal_error,
@@ -909,6 +938,7 @@ struct device_attribute *pm8001_host_attrs[] = {
 	&dev_attr_ob_log,
 	&dev_attr_ila_version,
 	&dev_attr_inc_fw_ver,
+	&dev_attr_ctl_mpi_state,
 	NULL,
 };
 
Signature · 1 line
2.16.3
author
#2 lore raw

PATCH v4 2/8 pm80xx: Add sysfs attribute to check controller hmi error

In reply to Viswas G

From: Vishakha Channapattan <[email protected]>

A new sysfs variable 'ctl_hmi_error' is being introduced to give the error details if the MPI initialization fails

Tested: Using 'ctl_hmi_error' sysfs variable we check the error details

linux-2dq0:~# cat /sys/class/scsi_host/host*/ctl_hmi_error
0x00000000
0x00000000
0x00000000
---
 drivers/scsi/pm8001/pm8001_ctl.c | 22 ++++++++++++++++++++++
 1 file changed, 22 insertions(+)
drivers/scsi/pm8001/pm8001_ctl.c +22 -0
index a93799bd3a32..b1f105253a12 100644
@@ -915,6 +915,27 @@ static ssize_t ctl_mpi_state_show(struct device *cdev,
 }
 static DEVICE_ATTR_RO(ctl_mpi_state);
 
+/**
+ * ctl_hmi_error_show - controller MPI initialization fails
+ * @cdev: pointer to embedded class device
+ * @buf: the buffer returned
+ *
+ * A sysfs 'read-only' shost attribute.
+ */
+
+static ssize_t ctl_hmi_error_show(struct device *cdev,
+		struct device_attribute *attr, char *buf)
+{
+	struct Scsi_Host *shost = class_to_shost(cdev);
+	struct sas_ha_struct *sha = SHOST_TO_SAS_HA(shost);
+	struct pm8001_hba_info *pm8001_ha = sha->lldd_ha;
+	unsigned int mpidw0;
+
+	mpidw0 = pm8001_mr32(pm8001_ha->general_stat_tbl_addr, 0);
+	return sysfs_emit(buf, "0x%08x\n", (mpidw0 >> 16));
+}
+static DEVICE_ATTR_RO(ctl_hmi_error);
+
 struct device_attribute *pm8001_host_attrs[] = {
 	&dev_attr_interface_rev,
 	&dev_attr_controller_fatal_error,
@@ -939,6 +960,7 @@ struct device_attribute *pm8001_host_attrs[] = {
 	&dev_attr_ila_version,
 	&dev_attr_inc_fw_ver,
 	&dev_attr_ctl_mpi_state,
+	&dev_attr_ctl_hmi_error,
 	NULL,
 };
 
Signature · 1 line
2.16.3
author
#3 lore raw

PATCH v4 3/8 pm80xx: Add sysfs attribute to track RAAE count

In reply to Viswas G

From: Vishakha Channapattan <[email protected]>

A new sysfs variable 'ctl_raae_count' is being introduced that tells if the controller is alive by indicating controller ticks. If on subsequent run we see the ticks changing in RAAE count that indicates that controller is not dead.

Tested: Using 'ctl_raae_count' sysfs variable we can see ticks incrementing

linux-9saw:~# cat /sys/class/scsi_host/host*/ctl_raae_count
0x00002245
0x00002253
0x0000225e
---
 drivers/scsi/pm8001/pm8001_ctl.c | 22 ++++++++++++++++++++++
 1 file changed, 22 insertions(+)
drivers/scsi/pm8001/pm8001_ctl.c +22 -0
index b1f105253a12..a39cfd9d2a6b 100644
@@ -936,6 +936,27 @@ static ssize_t ctl_hmi_error_show(struct device *cdev,
 }
 static DEVICE_ATTR_RO(ctl_hmi_error);
 
+/**
+ * ctl_raae_count_show - controller raae count check
+ * @cdev: pointer to embedded class device
+ * @buf: the buffer returned
+ *
+ * A sysfs 'read-only' shost attribute.
+ */
+
+static ssize_t ctl_raae_count_show(struct device *cdev,
+		struct device_attribute *attr, char *buf)
+{
+	struct Scsi_Host *shost = class_to_shost(cdev);
+	struct sas_ha_struct *sha = SHOST_TO_SAS_HA(shost);
+	struct pm8001_hba_info *pm8001_ha = sha->lldd_ha;
+	unsigned int raaecnt;
+
+	raaecnt = pm8001_mr32(pm8001_ha->general_stat_tbl_addr, 12);
+	return sysfs_emit(buf, "0x%08x\n", raaecnt);
+}
+static DEVICE_ATTR_RO(ctl_raae_count);
+
 struct device_attribute *pm8001_host_attrs[] = {
 	&dev_attr_interface_rev,
 	&dev_attr_controller_fatal_error,
@@ -961,6 +982,7 @@ struct device_attribute *pm8001_host_attrs[] = {
 	&dev_attr_inc_fw_ver,
 	&dev_attr_ctl_mpi_state,
 	&dev_attr_ctl_hmi_error,
+	&dev_attr_ctl_raae_count,
 	NULL,
 };
 
Signature · 1 line
2.16.3
author
#4 lore raw

PATCH v4 4/8 pm80xx: Add sysfs attribute to track iop0 count

In reply to Viswas G

From: Vishakha Channapattan <[email protected]>

A new sysfs variable 'ctl_iop0_count' is being introduced that tells if the controller is alive by indicating controller ticks. If on subsequent run we see the ticks changing that indicates that controller is not dead.

Tested: Using 'ctl_iop0_count' sysfs variable we can see ticks incrementing

linux-9saw:~# cat  /sys/class/scsi_host/host*/ctl_iop0_count
0x000000a3
0x000001db
0x000001e4
0x000001e7
---
 drivers/scsi/pm8001/pm8001_ctl.c | 22 ++++++++++++++++++++++
 1 file changed, 22 insertions(+)
drivers/scsi/pm8001/pm8001_ctl.c +22 -0
index a39cfd9d2a6b..b1c351dd2704 100644
@@ -957,6 +957,27 @@ static ssize_t ctl_raae_count_show(struct device *cdev,
 }
 static DEVICE_ATTR_RO(ctl_raae_count);
 
+/**
+ * ctl_iop0_count_show - controller iop0 count check
+ * @cdev: pointer to embedded class device
+ * @buf: the buffer returned
+ *
+ * A sysfs 'read-only' shost attribute.
+ */
+
+static ssize_t ctl_iop0_count_show(struct device *cdev,
+		struct device_attribute *attr, char *buf)
+{
+	struct Scsi_Host *shost = class_to_shost(cdev);
+	struct sas_ha_struct *sha = SHOST_TO_SAS_HA(shost);
+	struct pm8001_hba_info *pm8001_ha = sha->lldd_ha;
+	unsigned int iop0cnt;
+
+	iop0cnt = pm8001_mr32(pm8001_ha->general_stat_tbl_addr, 16);
+	return sysfs_emit(buf, "0x%08x\n", iop0cnt);
+}
+static DEVICE_ATTR_RO(ctl_iop0_count);
+
 struct device_attribute *pm8001_host_attrs[] = {
 	&dev_attr_interface_rev,
 	&dev_attr_controller_fatal_error,
@@ -983,6 +1004,7 @@ struct device_attribute *pm8001_host_attrs[] = {
 	&dev_attr_ctl_mpi_state,
 	&dev_attr_ctl_hmi_error,
 	&dev_attr_ctl_raae_count,
+	&dev_attr_ctl_iop0_count,
 	NULL,
 };
 
Signature · 1 line
2.16.3
author
#5 lore raw

PATCH v4 5/8 pm80xx: Add sysfs attribute to track iop1 count

In reply to Viswas G

From: Vishakha Channapattan <[email protected]>

A new sysfs variable 'ctl_iop1_count' is being introduced that tells if the controller is alive by indicating controller ticks. If on subsequent run we see the ticks changing that indicates that controller is not dead.

Tested: Using 'ctl_iop1_count' sysfs variable we can see ticks incrementing

linux-9saw:~# cat  /sys/class/scsi_host/host*/ctl_iop1_count
0x00000069
0x0000006b
0x0000006d
0x00000072
---
 drivers/scsi/pm8001/pm8001_ctl.c | 23 +++++++++++++++++++++++
 1 file changed, 23 insertions(+)
drivers/scsi/pm8001/pm8001_ctl.c +23 -0
index b1c351dd2704..f3c9fd067272 100644
@@ -978,6 +978,28 @@ static ssize_t ctl_iop0_count_show(struct device *cdev,
 }
 static DEVICE_ATTR_RO(ctl_iop0_count);
 
+/**
+ * ctl_iop1_count_show - controller iop1 count check
+ * @cdev: pointer to embedded class device
+ * @buf: the buffer returned
+ *
+ * A sysfs 'read-only' shost attribute.
+ */
+
+static ssize_t ctl_iop1_count_show(struct device *cdev,
+		struct device_attribute *attr, char *buf)
+{
+	struct Scsi_Host *shost = class_to_shost(cdev);
+	struct sas_ha_struct *sha = SHOST_TO_SAS_HA(shost);
+	struct pm8001_hba_info *pm8001_ha = sha->lldd_ha;
+	unsigned int iop1cnt;
+
+	iop1cnt = pm8001_mr32(pm8001_ha->general_stat_tbl_addr, 20);
+	return sysfs_emit(buf, "0x%08x\n", iop1cnt);
+
+}
+static DEVICE_ATTR_RO(ctl_iop1_count);
+
 struct device_attribute *pm8001_host_attrs[] = {
 	&dev_attr_interface_rev,
 	&dev_attr_controller_fatal_error,
@@ -1005,6 +1027,7 @@ struct device_attribute *pm8001_host_attrs[] = {
 	&dev_attr_ctl_hmi_error,
 	&dev_attr_ctl_raae_count,
 	&dev_attr_ctl_iop0_count,
+	&dev_attr_ctl_iop1_count,
 	NULL,
 };
 
Signature · 1 line
2.16.3
author
#6 lore raw

PATCH v4 6/8 pm80xx: Completing pending IO after fatal error

In reply to Viswas G

From: Ruksar Devadi <[email protected]>

When controller runs into fatal error, IOs get stuck with no response, handler event is defined to complete the pending IOs
(SAS task and internal task) and also perform the cleanup for the drives.

---
 drivers/scsi/pm8001/pm8001_hwi.c | 66 ++++++++++++++++++++++++++++++++++++----
 drivers/scsi/pm8001/pm8001_hwi.h |  1 +
 drivers/scsi/pm8001/pm8001_sas.c |  2 +-
 drivers/scsi/pm8001/pm8001_sas.h |  1 +
 drivers/scsi/pm8001/pm80xx_hwi.c |  1 +
 drivers/scsi/pm8001/pm80xx_hwi.h |  1 +
 6 files changed, 65 insertions(+), 7 deletions(-)
drivers/scsi/pm8001/pm8001_hwi.c +60 -6
index 31e5455d280c..4e97075d2ad8 100644
@@ -1499,12 +1499,14 @@ void pm8001_work_fn(struct work_struct *work)
 	 * was cancelled. This nullification happens when the device
 	 * goes away.
 	 */
-	pm8001_dev = pw->data; /* Most stash device structure */
-	if ((pm8001_dev == NULL)
-	 || ((pw->handler != IO_XFER_ERROR_BREAK)
-	  && (pm8001_dev->dev_type == SAS_PHY_UNUSED))) {
-		kfree(pw);
-		return;
+	if (pw->handler != IO_FATAL_ERROR) {
+		pm8001_dev = pw->data; /* Most stash device structure */
+		if ((pm8001_dev == NULL)
+		 || ((pw->handler != IO_XFER_ERROR_BREAK)
+			 && (pm8001_dev->dev_type == SAS_PHY_UNUSED))) {
+			kfree(pw);
+			return;
+		}
 	}
 
 	switch (pw->handler) {
@@ -1668,6 +1670,58 @@ void pm8001_work_fn(struct work_struct *work)
 		dev = pm8001_dev->sas_device;
 		pm8001_I_T_nexus_reset(dev);
 		break;
+	case IO_FATAL_ERROR:
+	{
+		struct pm8001_hba_info *pm8001_ha = pw->pm8001_ha;
+		struct pm8001_ccb_info *ccb;
+		struct task_status_struct *ts;
+		struct sas_task *task;
+		int i;
+		u32 tag, device_id;
+
+		for (i = 0; ccb = NULL, i < PM8001_MAX_CCB; i++) {
+			ccb = &pm8001_ha->ccb_info[i];
+			task = ccb->task;
+			ts = &task->task_status;
+			tag = ccb->ccb_tag;
+			/* check if tag is NULL */
+			if (!tag) {
+				pm8001_dbg(pm8001_ha, FAIL,
+					"tag Null\n");
+				continue;
+			}
+			if (task != NULL) {
+				dev = task->dev;
+				if (!dev) {
+					pm8001_dbg(pm8001_ha, FAIL,
+						"dev is NULL\n");
+					continue;
+				}
+				/*complete sas task and update to top layer */
+				pm8001_ccb_task_free(pm8001_ha, task, ccb, tag);
+				ts->resp = SAS_TASK_COMPLETE;
+				task->task_done(task);
+			} else if (tag != 0xFFFFFFFF) {
+				/* complete the internal commands/non-sas task */
+				pm8001_dev = ccb->device;
+				if (pm8001_dev->dcompletion) {
+					complete(pm8001_dev->dcompletion);
+					pm8001_dev->dcompletion = NULL;
+				}
+				complete(pm8001_ha->nvmd_completion);
+				pm8001_tag_free(pm8001_ha, tag);
+			}
+		}
+		/* Deregsiter all the device ids  */
+		for (i = 0; i < PM8001_MAX_DEVICES; i++) {
+			pm8001_dev = &pm8001_ha->devices[i];
+			device_id = pm8001_dev->device_id;
+			if (device_id) {
+				PM8001_CHIP_DISP->dereg_dev_req(pm8001_ha, device_id);
+				pm8001_free_dev(pm8001_dev);
+			}
+		}
+	}	break;
 	}
 	kfree(pw);
 }
drivers/scsi/pm8001/pm8001_hwi.h +1 -0
index 6d91e2446542..d1f3aa93325b 100644
@@ -805,6 +805,7 @@ struct set_dev_state_resp {
 #define IO_ABORT_IN_PROGRESS				0x40
 #define IO_ABORT_DELAYED				0x41
 #define IO_INVALID_LENGTH				0x42
+#define IO_FATAL_ERROR					0x51
 
 /* WARNING: This error code must always be the last number.
  * If you add error code, modify this code also
drivers/scsi/pm8001/pm8001_sas.c +1 -1
index a98d4496ff8b..edec599ac641 100644
@@ -590,7 +590,7 @@ struct pm8001_device *pm8001_find_dev(struct pm8001_hba_info *pm8001_ha,
 	return NULL;
 }
 
-static void pm8001_free_dev(struct pm8001_device *pm8001_dev)
+void pm8001_free_dev(struct pm8001_device *pm8001_dev)
 {
 	u32 id = pm8001_dev->id;
 	memset(pm8001_dev, 0, sizeof(*pm8001_dev));
drivers/scsi/pm8001/pm8001_sas.h +1 -0
index 039ed91e9841..36cd37c8c29a 100644
@@ -727,6 +727,7 @@ ssize_t pm80xx_get_non_fatal_dump(struct device *cdev,
 		struct device_attribute *attr, char *buf);
 ssize_t pm8001_get_gsm_dump(struct device *cdev, u32, char *buf);
 int pm80xx_fatal_errors(struct pm8001_hba_info *pm8001_ha);
+void pm8001_free_dev(struct pm8001_device *pm8001_dev);
 /* ctl shared API */
 extern struct device_attribute *pm8001_host_attrs[];
 
drivers/scsi/pm8001/pm80xx_hwi.c +1 -0
index 84315560e8e1..1aa3a499c85a 100644
@@ -4126,6 +4126,7 @@ static int process_oq(struct pm8001_hba_info *pm8001_ha, u8 vec)
 			pm8001_dbg(pm8001_ha, FAIL,
 				   "Firmware Fatal error! Regval:0x%x\n",
 				   regval);
+			pm8001_handle_event(pm8001_ha, NULL, IO_FATAL_ERROR);
 			print_scratchpad_registers(pm8001_ha);
 			return ret;
 		}
drivers/scsi/pm8001/pm80xx_hwi.h +1 -0
index 2c8e85cfdbc4..c7e5d93bea92 100644
@@ -1272,6 +1272,7 @@ typedef struct SASProtocolTimerConfig SASProtocolTimerConfig_t;
 #define IO_OPEN_CNX_ERROR_IT_NEXUS_LOSS_OPEN_COLLIDE	0x47
 #define IO_OPEN_CNX_ERROR_IT_NEXUS_LOSS_PATHWAY_BLOCKED	0x48
 #define IO_DS_INVALID					0x49
+#define IO_FATAL_ERROR					0x51
 /* WARNING: the value is not contiguous from here */
 #define IO_XFER_ERR_LAST_PIO_DATAIN_CRC_ERR	0x52
 #define IO_XFER_DMA_ACTIVATE_TIMEOUT		0x53
Signature · 1 line
2.16.3
author
#7 lore raw

PATCH v4 7/8 pm80xx: Reset PI and CI memory during re-initialize

In reply to Viswas G

Producer index(PI) outbound queue and consumer index(CI) for Outbound queue are in DMA memory. During resume(), the stale PI and CI Values will leads to unexpected behavior. These values should be reset to 0 during driver reinitialization.

---
 drivers/scsi/pm8001/pm8001_hwi.c | 2 ++
 drivers/scsi/pm8001/pm80xx_hwi.c | 2 ++
 2 files changed, 4 insertions(+)
drivers/scsi/pm8001/pm8001_hwi.c +2 -0
index 4e97075d2ad8..42e8a2e1e284 100644
@@ -240,6 +240,7 @@ static void init_default_table_values(struct pm8001_hba_info *pm8001_ha)
 			pm8001_ha->memoryMap.region[ci_offset + i].phys_addr_lo;
 		pm8001_ha->inbnd_q_tbl[i].ci_virt		=
 			pm8001_ha->memoryMap.region[ci_offset + i].virt_ptr;
+		pm8001_write_32(pm8001_ha->inbnd_q_tbl[i].ci_virt, 0, 0);
 		offsetib = i * 0x20;
 		pm8001_ha->inbnd_q_tbl[i].pi_pci_bar		=
 			get_pci_bar_index(pm8001_mr32(addressib,
@@ -268,6 +269,7 @@ static void init_default_table_values(struct pm8001_hba_info *pm8001_ha)
 			0 | (10 << 16) | (i << 24);
 		pm8001_ha->outbnd_q_tbl[i].pi_virt		=
 			pm8001_ha->memoryMap.region[pi_offset + i].virt_ptr;
+		pm8001_write_32(pm8001_ha->outbnd_q_tbl[i].pi_virt, 0, 0);
 		offsetob = i * 0x24;
 		pm8001_ha->outbnd_q_tbl[i].ci_pci_bar		=
 			get_pci_bar_index(pm8001_mr32(addressob,
drivers/scsi/pm8001/pm80xx_hwi.c +2 -0
index 1aa3a499c85a..0f2c57e054ac 100644
@@ -787,6 +787,7 @@ static void init_default_table_values(struct pm8001_hba_info *pm8001_ha)
 			pm8001_ha->memoryMap.region[ci_offset + i].phys_addr_lo;
 		pm8001_ha->inbnd_q_tbl[i].ci_virt		=
 			pm8001_ha->memoryMap.region[ci_offset + i].virt_ptr;
+		pm8001_write_32(pm8001_ha->inbnd_q_tbl[i].ci_virt, 0, 0);
 		offsetib = i * 0x20;
 		pm8001_ha->inbnd_q_tbl[i].pi_pci_bar		=
 			get_pci_bar_index(pm8001_mr32(addressib,
@@ -820,6 +821,7 @@ static void init_default_table_values(struct pm8001_hba_info *pm8001_ha)
 		pm8001_ha->outbnd_q_tbl[i].interrup_vec_cnt_delay = (i << 24);
 		pm8001_ha->outbnd_q_tbl[i].pi_virt		=
 			pm8001_ha->memoryMap.region[pi_offset + i].virt_ptr;
+		pm8001_write_32(pm8001_ha->outbnd_q_tbl[i].pi_virt, 0, 0);
 		offsetob = i * 0x24;
 		pm8001_ha->outbnd_q_tbl[i].ci_pci_bar		=
 			get_pci_bar_index(pm8001_mr32(addressob,
Signature · 1 line
2.16.3
author
#8 lore raw

PATCH v4 8/8 pm80xx: remove global lock from outbound queue processing

In reply to Viswas G

Introduced spin lock for outbound queue. With this, driver need not acquire hba global lock for outbound queue processing.

---
 drivers/scsi/pm8001/pm8001_init.c | 9 ++++++---
 drivers/scsi/pm8001/pm8001_sas.h  | 1 +
 drivers/scsi/pm8001/pm80xx_hwi.c  | 4 ++--
 3 files changed, 9 insertions(+), 5 deletions(-)
drivers/scsi/pm8001/pm8001_init.c +6 -3
index bd626ef876da..a3c8fb9a885f 100644
@@ -267,7 +267,8 @@ static int pm8001_alloc(struct pm8001_hba_info *pm8001_ha,
 {
 	int i, count = 0, rc = 0;
 	u32 ci_offset, ib_offset, ob_offset, pi_offset;
-	struct inbound_queue_table *circularQ;
+	struct inbound_queue_table *ibq;
+	struct outbound_queue_table *obq;
 
 	spin_lock_init(&pm8001_ha->lock);
 	spin_lock_init(&pm8001_ha->bitmap_lock);
@@ -315,8 +316,8 @@ static int pm8001_alloc(struct pm8001_hba_info *pm8001_ha,
 	pm8001_ha->memoryMap.region[IOP].alignment = 32;
 
 	for (i = 0; i < count; i++) {
-		circularQ = &pm8001_ha->inbnd_q_tbl[i];
-		spin_lock_init(&circularQ->iq_lock);
+		ibq = &pm8001_ha->inbnd_q_tbl[i];
+		spin_lock_init(&ibq->iq_lock);
 		/* MPI Memory region 3 for consumer Index of inbound queues */
 		pm8001_ha->memoryMap.region[ci_offset+i].num_elements = 1;
 		pm8001_ha->memoryMap.region[ci_offset+i].element_size = 4;
@@ -345,6 +346,8 @@ static int pm8001_alloc(struct pm8001_hba_info *pm8001_ha,
 	}
 
 	for (i = 0; i < count; i++) {
+		obq = &pm8001_ha->outbnd_q_tbl[i];
+		spin_lock_init(&obq->oq_lock);
 		/* MPI Memory region 4 for producer Index of outbound queues */
 		pm8001_ha->memoryMap.region[pi_offset+i].num_elements = 1;
 		pm8001_ha->memoryMap.region[pi_offset+i].element_size = 4;
drivers/scsi/pm8001/pm8001_sas.h +1 -0
index 36cd37c8c29a..f835557ee354 100644
@@ -457,6 +457,7 @@ struct outbound_queue_table {
 	u32			dinterrup_to_pci_offset;
 	__le32			producer_index;
 	u32			consumer_idx;
+	spinlock_t		oq_lock;
 };
 struct pm8001_hba_memspace {
 	void __iomem  		*memvirtaddr;
drivers/scsi/pm8001/pm80xx_hwi.c +2 -2
index 0f2c57e054ac..f1276baebe1d 100644
@@ -4133,8 +4133,8 @@ static int process_oq(struct pm8001_hba_info *pm8001_ha, u8 vec)
 			return ret;
 		}
 	}
-	spin_lock_irqsave(&pm8001_ha->lock, flags);
 	circularQ = &pm8001_ha->outbnd_q_tbl[vec];
+	spin_lock_irqsave(&circularQ->oq_lock, flags);
 	do {
 		/* spurious interrupt during setup if kexec-ing and
 		 * driver doing a doorbell access w/ the pre-kexec oq
@@ -4160,7 +4160,7 @@ static int process_oq(struct pm8001_hba_info *pm8001_ha, u8 vec)
 				break;
 		}
 	} while (1);
-	spin_unlock_irqrestore(&pm8001_ha->lock, flags);
+	spin_unlock_irqrestore(&circularQ->oq_lock, flags);
 	return ret;
 }
 
Signature · 1 line
2.16.3
#9 lore raw
In reply to Viswas G

Viswas,

Quoted 2 lines

This patch set include some bug fixes and enhancements for pm80xx driver.

Applied to 5.13/scsi-staging, thanks!

Signature · 1 line
Martin K. Petersen	Oracle Linux Engineering
#10 lore raw
In reply to Viswas G
Quoted 11 lines · Viswas G

This patch set include some bug fixes and enhancements for pm80xx driver.

Changes from v3 to v4:
- For sysfs attribute patches
- Removed the unwanted variable 'c'
- For "Add sysfs attribute to check mpi state" patch
- split the ctl_mpi_state to two (ctl_mpi_state, ctl_hmi_error) as suggested.
- Changed static char mpiStateText[] to static const char *const mpiStateText[]

Quoted 1 line

[...]

Applied to 5.13/scsi-queue, thanks!

[1/8] pm80xx: Add sysfs attribute to check mpi state

      https://git.kernel.org/mkp/scsi/c/4ddbea1b6f51
[2/8] pm80xx: Add sysfs attribute to check controller hmi error
      https://git.kernel.org/mkp/scsi/c/a4c55e16c500
[3/8] pm80xx: Add sysfs attribute to track RAAE count
      https://git.kernel.org/mkp/scsi/c/dd49ded8aa43
[4/8] pm80xx: Add sysfs attribute to track iop0 count
      https://git.kernel.org/mkp/scsi/c/0602624ace23
[5/8] pm80xx: Add sysfs attribute to track iop1 count
      https://git.kernel.org/mkp/scsi/c/b0c306e62167
[6/8] pm80xx: Completing pending IO after fatal error
      https://git.kernel.org/mkp/scsi/c/4f5deeb40f9c
[7/8] pm80xx: Reset PI and CI memory during re-initialize
      https://git.kernel.org/mkp/scsi/c/b431472bc88b
[8/8] pm80xx: remove global lock from outbound queue processing
      https://git.kernel.org/mkp/scsi/c/1f02beff224e
Signature · 1 line
Martin K. Petersen	Oracle Linux Engineering