Skip to contentSkip to conversation

ufs: core: mcq: Add ufshcd_abort() and error handler support in MCQ mode

PATCH Superseded Bao D. Nguyen 4 participants 14 messages Mar 29, 2023 – May 4, 2023

Subsystems (4)

v1 of 7 Superseded Latest is v7 →

lore
Share & tools
lore
#0 lore raw

cover ufs: core: mcq: Add ufshcd_abort() and error handler support in MCQ mode

This patch series enable support for ufshcd_abort() and error handler in MCQ mode. The first 3 patches are for supporting ufshcd_abort().
The 4th and 5th patches are for supporting error handler.

Bao D. Nguyen (5):
ufs: mcq: Add supporting functions for mcq abort
ufs: mcq: Add support for clean up mcq resources
ufs: mcq: Added ufshcd_mcq_abort()
ufs: mcq: Use ufshcd_mcq_poll_cqe_lock() in mcq mode
ufs: core: Add error handling for MCQ mode

 drivers/ufs/core/ufs-mcq.c     | 242 ++++++++++++++++++++++++++++++++++++++++-
 drivers/ufs/core/ufshcd-priv.h |  15 ++-
 drivers/ufs/core/ufshcd.c      | 140 ++++++++++++++++++++----
 drivers/ufs/host/ufs-qcom.c    |   2 +-
 include/ufs/ufshcd.h           |   2 +-
 include/ufs/ufshci.h           |  17 +++
 6 files changed, 388 insertions(+), 30 deletions(-)
Signature · 1 line
2.7.4
#1 lore raw

PATCH 1/5 ufs: mcq: Add supporting functions for mcq abort

Add supporting functions to handle ufs abort in mcq mode.

---
 drivers/ufs/core/ufs-mcq.c     | 174 +++++++++++++++++++++++++++++++++++++++++
 drivers/ufs/core/ufshcd-priv.h |  10 +++
 drivers/ufs/core/ufshcd.c      |   2 -
 include/ufs/ufshci.h           |  17 ++++
 4 files changed, 201 insertions(+), 2 deletions(-)
drivers/ufs/core/ufs-mcq.c +174 -0
index 31df052..6f6b22c 100644
@@ -12,6 +12,10 @@
 #include <linux/module.h>
 #include <linux/platform_device.h>
 #include "ufshcd-priv.h"
+#include <linux/delay.h>
+#include <scsi/scsi_cmnd.h>
+#include <linux/bitfield.h>
+#include <linux/iopoll.h>
 
 #define MAX_QUEUE_SUP GENMASK(7, 0)
 #define UFS_MCQ_MIN_RW_QUEUES 2
@@ -27,6 +31,9 @@
 #define MCQ_ENTRY_SIZE_IN_DWORD	8
 #define CQE_UCD_BA GENMASK_ULL(63, 7)
 
+/* Max mcq register polling time in micro second unit */
+#define MCQ_POLL_US 500000
+
 static int rw_queue_count_set(const char *val, const struct kernel_param *kp)
 {
 	return param_set_uint_minmax(val, kp, UFS_MCQ_MIN_RW_QUEUES,
@@ -429,3 +436,170 @@ int ufshcd_mcq_init(struct ufs_hba *hba)
 	host->host_tagset = 1;
 	return 0;
 }
+
+static int ufshcd_mcq_sq_stop(struct ufs_hba *hba, struct ufs_hw_queue *hwq)
+{
+	void __iomem *reg;
+	u32 i = hwq->id, val;
+	int err;
+
+	writel(SQ_STOP, mcq_opr_base(hba, OPR_SQD, i) + REG_SQRTC);
+	reg = mcq_opr_base(hba, OPR_SQD, i) + REG_SQRTS;
+	err = read_poll_timeout(readl, val, val & SQ_STS, 20,
+				MCQ_POLL_US, false, reg);
+	if (err)
+		dev_err(hba->dev, "%s: failed. hwq-id=%d, err=%d\n",
+			__func__, i, err);
+	return err;
+}
+
+static int ufshcd_mcq_sq_start(struct ufs_hba *hba, struct ufs_hw_queue *hwq)
+{
+	void __iomem *reg;
+	u32 i = hwq->id, val;
+	int err;
+
+	writel(SQ_START, mcq_opr_base(hba, OPR_SQD, i) + REG_SQRTC);
+	reg = mcq_opr_base(hba, OPR_SQD, i) + REG_SQRTS;
+	err = read_poll_timeout(readl, val, !(val & SQ_STS), 20,
+				MCQ_POLL_US, false, reg);
+	if (err)
+		dev_err(hba->dev, "%s: failed. hwq-id=%d, err=%d\n",
+			__func__, i, err);
+	return err;
+}
+
+/**
+ * ufshcd_mcq_sq_cleanup - Clean up Submission Queue resources
+ * associated with the pending command.
+ * @hba - per adapter instance.
+ * @task_tag - The command's task tag.
+ * @result - Result of the Clean up operation.
+ *
+ * Returns 0 and result on completion. Returns error code if
+ * the operation fails.
+ */
+int ufshcd_mcq_sq_cleanup(struct ufs_hba *hba, int task_tag, int *result)
+{
+	struct ufshcd_lrb *lrbp = &hba->lrb[task_tag];
+	struct scsi_cmnd *cmd = lrbp->cmd;
+	struct ufs_hw_queue *hwq;
+	void __iomem *reg, *opr_sqd_base;
+	u32 nexus, i, val;
+	int err;
+
+	if (task_tag != hba->nutrs - UFSHCD_NUM_RESERVED) {
+		if (!cmd)
+			return FAILED;
+		hwq = ufshcd_mcq_req_to_hwq(hba, scsi_cmd_to_rq(cmd));
+	} else {
+		hwq = hba->dev_cmd_queue;
+	}
+
+	i = hwq->id;
+
+	spin_lock(&hwq->sq_lock);
+
+	/* stop the SQ fetching before working on it */
+	err = ufshcd_mcq_sq_stop(hba, hwq);
+	if (err)
+		goto unlock;
+
+	/* SQCTI = EXT_IID, IID, LUN, Task Tag */
+	nexus = lrbp->lun << 8 | task_tag;
+	opr_sqd_base = mcq_opr_base(hba, OPR_SQD, i);
+	writel(nexus, opr_sqd_base + REG_SQCTI);
+
+	/* SQRTCy.ICU = 1 */
+	writel(SQ_ICU, opr_sqd_base + REG_SQRTC);
+
+	/* Poll SQRTSy.CUS = 1. Return result from SQRTSy.RTC */
+	reg = opr_sqd_base + REG_SQRTS;
+	err = read_poll_timeout(readl, val, val & SQ_CUS, 20,
+				MCQ_POLL_US, false, reg);
+	if (err)
+		dev_err(hba->dev, "%s: failed. hwq=%d, lun=0x%x, tag=%d\n",
+			__func__, i, lrbp->lun, task_tag);
+
+	*result = FIELD_GET(SQ_ICU_ERR_CODE_MASK, readl(reg));
+
+	if (ufshcd_mcq_sq_start(hba, hwq))
+		err = FAILED;
+
+unlock:
+	spin_unlock(&hwq->sq_lock);
+	return err;
+}
+
+static u64 ufshcd_mcq_get_cmd_desc_addr(struct ufs_hba *hba,
+					int task_tag)
+{
+	struct ufshcd_lrb *lrbp = &hba->lrb[task_tag];
+	__le32 hi = lrbp->utr_descriptor_ptr->command_desc_base_addr_hi;
+	__le32 lo = lrbp->utr_descriptor_ptr->command_desc_base_addr_lo;
+
+	return le64_to_cpu((__le64)hi << 32 | lo);
+}
+
+/**
+ * ufshcd_mcq_nullify_cmd - Nullify utrd. Host controller does not fetch
+ * transfer with Command Type = 0xF. post the Completion Queue with OCS=ABORTED.
+ * @hba - per adapter instance.
+ * @hwq - Hardware Queue of the nullified utrd.
+ */
+static void ufshcd_mcq_nullify_cmd(struct ufs_hba *hba, struct ufs_hw_queue *hwq)
+{
+	struct utp_transfer_req_desc *utrd;
+	u32 dword_0;
+
+	utrd = (struct utp_transfer_req_desc *)(hwq->sqe_base_addr +
+			hwq->id * sizeof(struct utp_transfer_req_desc));
+	dword_0 = le32_to_cpu(utrd->header.dword_0);
+	dword_0 &= ~UPIU_COMMAND_TYPE_MASK;
+	dword_0 |= FIELD_PREP(UPIU_COMMAND_TYPE_MASK, 0xF);
+	utrd->header.dword_0 = cpu_to_le32(dword_0);
+}
+
+/**
+ * ufshcd_mcq_sqe_search - Search for the cmd in the Submission Queue (SQ)
+ * @hba - per adapter instance.
+ * @hwq - Hardware Queue to be searched.
+ * @task_tag - The command's task tag.
+ *
+ * Returns true if the SQE containing the command is present in the SQ
+ * (not fetched by the controller); returns false if the SQE is not in the SQ.
+ */
+static bool ufshcd_mcq_sqe_search(struct ufs_hba *hba,
+		struct ufs_hw_queue *hwq, int task_tag)
+{
+	struct utp_transfer_req_desc *utrd;
+	u32 mask = hwq->max_entries - 1;
+	bool ret = false;
+	u64 addr, match;
+	u32 sq_head_slot;
+
+	spin_lock(&hwq->sq_lock);
+
+	ufshcd_mcq_sq_stop(hba, hwq);
+	sq_head_slot = ufshcd_mcq_get_sq_head_slot(hwq);
+	if (sq_head_slot == hwq->sq_tail_slot)
+		goto out;
+
+	addr = ufshcd_mcq_get_cmd_desc_addr(hba, task_tag);
+	while (sq_head_slot != hwq->sq_tail_slot) {
+		utrd = (struct utp_transfer_req_desc *)(hwq->sqe_base_addr +
+				sq_head_slot * sizeof(struct utp_transfer_req_desc));
+		match = le64_to_cpu((__le64)utrd->command_desc_base_addr_hi << 32 |
+				utrd->command_desc_base_addr_lo) & CQE_UCD_BA;
+		if (addr == match) {
+			ret = true;
+			goto out;
+		}
+		sq_head_slot = (sq_head_slot + 1) & mask;
+	}
+
+out:
+	ufshcd_mcq_sq_start(hba, hwq);
+	spin_unlock(&hwq->sq_lock);
+	return ret;
+}
drivers/ufs/core/ufshcd-priv.h +10 -0
index 529f850..1a40cf7 100644
@@ -78,6 +78,8 @@ struct ufs_hw_queue *ufshcd_mcq_req_to_hwq(struct ufs_hba *hba,
 unsigned long ufshcd_mcq_poll_cqe_lock(struct ufs_hba *hba,
 				       struct ufs_hw_queue *hwq);
 
+int ufshcd_mcq_sq_cleanup(struct ufs_hba *hba, int task_tag, int *result);
+
 #define UFSHCD_MCQ_IO_QUEUE_OFFSET	1
 #define SD_ASCII_STD true
 #define SD_RAW false
@@ -403,4 +405,12 @@ static inline struct cq_entry *ufshcd_mcq_cur_cqe(struct ufs_hw_queue *q)
 
 	return cqe + q->cq_head_slot;
 }
+
+static inline u32 ufshcd_mcq_get_sq_head_slot(struct ufs_hw_queue *q)
+{
+	u32 val = readl(q->mcq_sq_head);
+
+	return val / sizeof(struct utp_transfer_req_desc);
+}
+
 #endif /* _UFSHCD_PRIV_H_ */
drivers/ufs/core/ufshcd.c +0 -2
index 35a3bd9..808387c 100644
@@ -56,7 +56,6 @@
 #define NOP_OUT_RETRIES    10
 /* Timeout after 50 msecs if NOP OUT hangs without response */
 #define NOP_OUT_TIMEOUT    50 /* msecs */
-
 /* Query request retries */
 #define QUERY_REQ_RETRIES 3
 /* Query request timeout */
@@ -173,7 +172,6 @@ EXPORT_SYMBOL_GPL(ufshcd_dump_regs);
 enum {
 	UFSHCD_MAX_CHANNEL	= 0,
 	UFSHCD_MAX_ID		= 1,
-	UFSHCD_NUM_RESERVED	= 1,
 	UFSHCD_CMD_PER_LUN	= 32 - UFSHCD_NUM_RESERVED,
 	UFSHCD_CAN_QUEUE	= 32 - UFSHCD_NUM_RESERVED,
 };
include/ufs/ufshci.h +17 -0
index 11424bb..ceb8664 100644
@@ -99,6 +99,9 @@ enum {
 enum {
 	REG_SQHP		= 0x0,
 	REG_SQTP		= 0x4,
+	REG_SQRTC		= 0x8,
+	REG_SQCTI		= 0xC,
+	REG_SQRTS		= 0x10,
 };
 
 enum {
@@ -111,12 +114,26 @@ enum {
 	REG_CQIE		= 0x4,
 };
 
+enum {
+	SQ_START		= 0x0,
+	SQ_STOP			= 0x1,
+	SQ_ICU			= 0x2,
+};
+
+enum {
+	SQ_STS			= 0x1,
+	SQ_CUS			= 0x2,
+};
+
+#define SQ_ICU_ERR_CODE_MASK		GENMASK(7, 4)
+#define UPIU_COMMAND_TYPE_MASK		GENMASK(31, 28)
 #define UFS_MASK(mask, offset)		((mask) << (offset))
 
 /* UFS Version 08h */
 #define MINOR_VERSION_NUM_MASK		UFS_MASK(0xFFFF, 0)
 #define MAJOR_VERSION_NUM_MASK		UFS_MASK(0xFFFF, 16)
 
+#define UFSHCD_NUM_RESERVED	1
 /*
  * Controller UFSHCI version
  * - 2.x and newer use the following scheme:
Signature · 1 line
2.7.4
#2 lore raw

Re: [PATCH v1 1/5] ufs: mcq: Add supporting functions for mcq abort

Hi Bao D.,

We have done some test based on your RFC v3 patches and an issue is reported.

kworker/u16:4: BUG: scheduling while atomic:
kworker/u16:4/5736/0x00000002
kworker/u16:4: [name:core&]Preemption disabled at:
kworker/u16:4: [<ffffffef97e33024>] ufshcd_mcq_sq_cleanup+0x9c/0x27c kworker/u16:4: CPU: 2 PID: 5736 Comm: kworker/u16:4 Tainted: G
S W OE
kworker/u16:4: Workqueue: ufs_eh_wq_0 ufshcd_err_handler kworker/u16:4: Call trace:

   kworker/u16:4:  dump_backtrace+0x108/0x15c
   kworker/u16:4:  show_stack+0x20/0x30
   kworker/u16:4:  dump_stack_lvl+0x6c/0x8c
   kworker/u16:4:  dump_stack+0x20/0x44
   kworker/u16:4:  __schedule_bug+0xd4/0x100
   kworker/u16:4:  __schedule+0x660/0xa5c
   kworker/u16:4:  schedule+0x80/0xec
   kworker/u16:4:  schedule_hrtimeout_range_clock+0xa0/0x140
   kworker/u16:4:  schedule_hrtimeout_range+0x1c/0x30
   kworker/u16:4:  usleep_range_state+0x88/0xd8
   kworker/u16:4:  ufshcd_mcq_sq_cleanup+0x170/0x27c
   kworker/u16:4:  ufshcd_clear_cmds+0x78/0x184
   kworker/u16:4:  ufshcd_wait_for_dev_cmd+0x234/0x348
   kworker/u16:4:  ufshcd_exec_dev_cmd+0x220/0x298
   kworker/u16:4:  ufshcd_verify_dev_init+0x68/0x124
   kworker/u16:4:  ufshcd_probe_hba+0x390/0x9bc
   kworker/u16:4:  ufshcd_host_reset_and_restore+0x74/0x158
   kworker/u16:4:  ufshcd_reset_and_restore+0x70/0x31c
   kworker/u16:4:  ufshcd_err_handler+0xad4/0xe58
   kworker/u16:4:  process_one_work+0x214/0x5b8
   kworker/u16:4:  worker_thread+0x2d4/0x448
   kworker/u16:4:  kthread+0x110/0x1e0
   kworker/u16:4:  ret_from_fork+0x10/0x20

kworker/u16:4: ------------[ cut here ]------------

Quoted 25 lines · Bao D. Nguyen
drivers/ufs/core/ufs-mcq.c +11 -0
+/**
+ * ufshcd_mcq_sq_cleanup - Clean up Submission Queue resources
+ * associated with the pending command.
+ * @hba - per adapter instance.
+ * @task_tag - The command's task tag.
+ * @result - Result of the Clean up operation.
+ *
+ * Returns 0 and result on completion. Returns error code if
+ * the operation fails.
+ */
+int ufshcd_mcq_sq_cleanup(struct ufs_hba *hba, int task_tag, int

*result)

drivers/ufs/core/ufs-mcq.c +12 -0
+{
+       struct ufshcd_lrb *lrbp = &hba->lrb[task_tag];
+       struct scsi_cmnd *cmd = lrbp->cmd;
+       struct ufs_hw_queue *hwq;
+       void __iomem *reg, *opr_sqd_base;
+       u32 nexus, i, val;
+       int err;
+
+       if (task_tag != hba->nutrs - UFSHCD_NUM_RESERVED) {
+               if (!cmd)
+                       return FAILED;
+               hwq = ufshcd_mcq_req_to_hwq(hba,

scsi_cmd_to_rq(cmd));

drivers/ufs/core/ufs-mcq.c int ufshcd_mcq_init(struct ufs_hba *hba) · line 501

Quoted 7 lines
drivers/ufs/core/ufs-mcq.c +7 -0
+       } else {
+               hwq = hba->dev_cmd_queue;
+       }
+
+       i = hwq->id;
+
+       spin_lock(&hwq->sq_lock);

As spin_lock() disable preemption

Quoted 8 lines
drivers/ufs/core/ufs-mcq.c +8 -0
+
+       /* stop the SQ fetching before working on it */
+       err = ufshcd_mcq_sq_stop(hba, hwq);
+       if (err)
+               goto unlock;
+
+       /* SQCTI = EXT_IID, IID, LUN, Task Tag */
+       nexus = lrbp->lun << 8 | task_tag;

drivers/ufs/core/ufs-mcq.c int ufshcd_mcq_init(struct ufs_hba *hba) · line 519

Quoted 10 lines
drivers/ufs/core/ufs-mcq.c +10 -0
+       opr_sqd_base = mcq_opr_base(hba, OPR_SQD, i);
+       writel(nexus, opr_sqd_base + REG_SQCTI);
+
+       /* SQRTCy.ICU = 1 */
+       writel(SQ_ICU, opr_sqd_base + REG_SQRTC);
+
+       /* Poll SQRTSy.CUS = 1. Return result from SQRTSy.RTC */
+       reg = opr_sqd_base + REG_SQRTS;
+       err = read_poll_timeout(readl, val, val & SQ_CUS, 20,
+                               MCQ_POLL_US, false, reg);

read_poll_timeout() was ufshcd_mcq_poll_register() in last patch, right? ufshcd_mcq_poll_register() calls usleep_range() causing KE as reported above. Same issue seems to still exist as read_poll_timeout() sleeps.

Skipping ufshcd_mcq_sq_cleanup() by returning FAILED directly to trigger reset in ufshcd error handler successfully recover host.

BTW, is there maybe a change list between RFC v3 and this v1 patch? :) Thanks

Po-Wen

#3 lore raw

Re: [PATCH v1 1/5] ufs: mcq: Add supporting functions for mcq abort

Quoted 4 lines · Powen Kao (高伯文)

Hi Bao D.,

We have done some test based on your RFC v3 patches and an issue is reported.

Hi Powen,

Thank you very much for catching this issue. It seems that I cannot use read_poll_timeout() because it sleeps while holding the spin_lock().

I will make the change to poll the registers in a tight loop with udelay(20) polling interval in the next revision.

Thanks,
Bao

Quoted 103 lines
    kworker/u16:4: BUG: scheduling while atomic:
    kworker/u16:4/5736/0x00000002
    kworker/u16:4: [name:core&]Preemption disabled at:
    kworker/u16:4: [<ffffffef97e33024>] ufshcd_mcq_sq_cleanup+0x9c/0x27c
    kworker/u16:4: CPU: 2 PID: 5736 Comm: kworker/u16:4 Tainted: G
    S      W  OE
    kworker/u16:4: Workqueue: ufs_eh_wq_0 ufshcd_err_handler
    kworker/u16:4: Call trace:
    kworker/u16:4:  dump_backtrace+0x108/0x15c
    kworker/u16:4:  show_stack+0x20/0x30
    kworker/u16:4:  dump_stack_lvl+0x6c/0x8c
    kworker/u16:4:  dump_stack+0x20/0x44
    kworker/u16:4:  __schedule_bug+0xd4/0x100
    kworker/u16:4:  __schedule+0x660/0xa5c
    kworker/u16:4:  schedule+0x80/0xec
    kworker/u16:4:  schedule_hrtimeout_range_clock+0xa0/0x140
    kworker/u16:4:  schedule_hrtimeout_range+0x1c/0x30
    kworker/u16:4:  usleep_range_state+0x88/0xd8
    kworker/u16:4:  ufshcd_mcq_sq_cleanup+0x170/0x27c
    kworker/u16:4:  ufshcd_clear_cmds+0x78/0x184
    kworker/u16:4:  ufshcd_wait_for_dev_cmd+0x234/0x348
    kworker/u16:4:  ufshcd_exec_dev_cmd+0x220/0x298
    kworker/u16:4:  ufshcd_verify_dev_init+0x68/0x124
    kworker/u16:4:  ufshcd_probe_hba+0x390/0x9bc
    kworker/u16:4:  ufshcd_host_reset_and_restore+0x74/0x158
    kworker/u16:4:  ufshcd_reset_and_restore+0x70/0x31c
    kworker/u16:4:  ufshcd_err_handler+0xad4/0xe58
    kworker/u16:4:  process_one_work+0x214/0x5b8
    kworker/u16:4:  worker_thread+0x2d4/0x448
    kworker/u16:4:  kthread+0x110/0x1e0
    kworker/u16:4:  ret_from_fork+0x10/0x20
    kworker/u16:4: ------------[ cut here ]------------
Quoted 25 lines · Bao D. Nguyen
drivers/ufs/core/ufs-mcq.c +11 -0
+/**
+ * ufshcd_mcq_sq_cleanup - Clean up Submission Queue resources
+ * associated with the pending command.
+ * @hba - per adapter instance.
+ * @task_tag - The command's task tag.
+ * @result - Result of the Clean up operation.
+ *
+ * Returns 0 and result on completion. Returns error code if
+ * the operation fails.
+ */
+int ufshcd_mcq_sq_cleanup(struct ufs_hba *hba, int task_tag, int

*result)

drivers/ufs/core/ufs-mcq.c +12 -0
+{
+       struct ufshcd_lrb *lrbp = &hba->lrb[task_tag];
+       struct scsi_cmnd *cmd = lrbp->cmd;
+       struct ufs_hw_queue *hwq;
+       void __iomem *reg, *opr_sqd_base;
+       u32 nexus, i, val;
+       int err;
+
+       if (task_tag != hba->nutrs - UFSHCD_NUM_RESERVED) {
+               if (!cmd)
+                       return FAILED;
+               hwq = ufshcd_mcq_req_to_hwq(hba,

scsi_cmd_to_rq(cmd));

drivers/ufs/core/ufs-mcq.c int ufshcd_mcq_init(struct ufs_hba *hba) · line 501

Quoted 7 lines
drivers/ufs/core/ufs-mcq.c +7 -0
+       } else {
+               hwq = hba->dev_cmd_queue;
+       }
+
+       i = hwq->id;
+
+       spin_lock(&hwq->sq_lock);

As spin_lock() disable preemption

Quoted 8 lines
drivers/ufs/core/ufs-mcq.c +8 -0
+
+       /* stop the SQ fetching before working on it */
+       err = ufshcd_mcq_sq_stop(hba, hwq);
+       if (err)
+               goto unlock;
+
+       /* SQCTI = EXT_IID, IID, LUN, Task Tag */
+       nexus = lrbp->lun << 8 | task_tag;

drivers/ufs/core/ufs-mcq.c int ufshcd_mcq_init(struct ufs_hba *hba) · line 519

Quoted 10 lines
drivers/ufs/core/ufs-mcq.c +10 -0
+       opr_sqd_base = mcq_opr_base(hba, OPR_SQD, i);
+       writel(nexus, opr_sqd_base + REG_SQCTI);
+
+       /* SQRTCy.ICU = 1 */
+       writel(SQ_ICU, opr_sqd_base + REG_SQRTC);
+
+       /* Poll SQRTSy.CUS = 1. Return result from SQRTSy.RTC */
+       reg = opr_sqd_base + REG_SQRTS;
+       err = read_poll_timeout(readl, val, val & SQ_CUS, 20,
+                               MCQ_POLL_US, false, reg);

read_poll_timeout() was ufshcd_mcq_poll_register() in last patch, right? ufshcd_mcq_poll_register() calls usleep_range() causing KE as reported above. Same issue seems to still exist as read_poll_timeout() sleeps.

Skipping ufshcd_mcq_sq_cleanup() by returning FAILED directly to trigger reset in ufshcd error handler successfully recover host.

BTW, is there maybe a change list between RFC v3 and this v1 patch? :) Thanks

Po-Wen

#4 lore raw

PATCH 2/5 ufs: mcq: Add support for clean up mcq resources

In reply to Bao D. Nguyen

Update ufshcd_clear_cmds() to clean up the mcq resources similar to the function ufshcd_utrl_clear() does for sdb mode.

Update ufshcd_try_to_abort_task() to support mcq mode so that this function can be invoked in either mcq or sdb mode.

---
 drivers/ufs/core/ufshcd.c | 45 +++++++++++++++++++++++++++++++++++++++------
 1 file changed, 39 insertions(+), 6 deletions(-)
drivers/ufs/core/ufshcd.c +39 -6
index 808387c..ffccb91 100644
@@ -3007,10 +3007,28 @@ static int ufshcd_compose_dev_cmd(struct ufs_hba *hba,
  * @mask and wait until the controller confirms that these requests have been
  * cleared.
  */
-static int ufshcd_clear_cmds(struct ufs_hba *hba, u32 mask)
+static int ufshcd_clear_cmds(struct ufs_hba *hba, unsigned long mask)
 {
 	unsigned long flags;
+	int err, tag, result = FAILED;
 
+	if (is_mcq_enabled(hba)) {
+		/*
+		 * MCQ mode. Clean up the MCQ resources similar to
+		 * what the ufshcd_utrl_clear() does for SDB mode.
+		 */
+		for_each_set_bit(tag, &mask, hba->nutrs) {
+			err = ufshcd_mcq_sq_cleanup(hba, tag, &result);
+			if (err || result) {
+				dev_err(hba->dev, "%s: failed tag=%d. err=%d, result=%d\n",
+					__func__, tag, err, result);
+				return FAILED;
+			}
+		}
+		return 0;
+	}
+
+	/* Single Doorbell Mode */
 	/* clear outstanding transaction before retry */
 	spin_lock_irqsave(hba->host->host_lock, flags);
 	ufshcd_utrl_clear(hba, mask);
@@ -3110,7 +3128,7 @@ static int ufshcd_wait_for_dev_cmd(struct ufs_hba *hba,
 		err = -ETIMEDOUT;
 		dev_dbg(hba->dev, "%s: dev_cmd request timedout, tag %d\n",
 			__func__, lrbp->task_tag);
-		if (ufshcd_clear_cmds(hba, 1U << lrbp->task_tag) == 0) {
+		if (ufshcd_clear_cmds(hba, 1UL << lrbp->task_tag) == 0) {
 			/* successfully cleared the command, retry if needed */
 			err = -EAGAIN;
 			/*
@@ -7379,6 +7397,20 @@ static int ufshcd_try_to_abort_task(struct ufs_hba *hba, int tag)
 			 */
 			dev_err(hba->dev, "%s: cmd at tag %d not pending in the device.\n",
 				__func__, tag);
+			if (is_mcq_enabled(hba)) {
+				/* MCQ mode */
+				if (lrbp->cmd) {
+					/* sleep for max. 200us to stabilize */
+					usleep_range(100, 200);
+					continue;
+				}
+				/* command completed already */
+				dev_err(hba->dev, "%s: cmd at tag=%d is cleared.\n",
+					__func__, tag);
+				goto out;
+			}
+
+			/* Single Doorbell Mode */
 			reg = ufshcd_readl(hba, REG_UTP_TRANSFER_REQ_DOOR_BELL);
 			if (reg & (1 << tag)) {
 				/* sleep for max. 200us to stabilize */
@@ -7415,7 +7447,7 @@ static int ufshcd_try_to_abort_task(struct ufs_hba *hba, int tag)
 		goto out;
 	}
 
-	err = ufshcd_clear_cmds(hba, 1U << tag);
+	err = ufshcd_clear_cmds(hba, 1UL << tag);
 	if (err)
 		dev_err(hba->dev, "%s: Failed clearing cmd at tag %d, err %d\n",
 			__func__, tag, err);
@@ -7445,8 +7477,8 @@ static int ufshcd_abort(struct scsi_cmnd *cmd)
 
 	ufshcd_hold(hba, false);
 	reg = ufshcd_readl(hba, REG_UTP_TRANSFER_REQ_DOOR_BELL);
-	/* If command is already aborted/completed, return FAILED. */
-	if (!(test_bit(tag, &hba->outstanding_reqs))) {
+	if (!is_mcq_enabled(hba) && !(test_bit(tag, &hba->outstanding_reqs))) {
+		/* If command is already aborted/completed, return FAILED. */
 		dev_err(hba->dev,
 			"%s: cmd at tag %d already completed, outstanding=0x%lx, doorbell=0x%x\n",
 			__func__, tag, hba->outstanding_reqs, reg);
@@ -7475,7 +7507,8 @@ static int ufshcd_abort(struct scsi_cmnd *cmd)
 	}
 	hba->req_abort_count++;
 
-	if (!(reg & (1 << tag))) {
+	if (!is_mcq_enabled(hba) && !(reg & (1 << tag))) {
+		/* only execute this code in single doorbell mode */
 		dev_err(hba->dev,
 		"%s: cmd was completed, but without a notifying intr, tag = %d",
 		__func__, tag);
Signature · 1 line
2.7.4
#5 lore raw

PATCH 3/5 ufs: mcq: Added ufshcd_mcq_abort()

In reply to Bao D. Nguyen

Add ufshcd_mcq_abort() to support ufs abort in mcq mode.

---
 drivers/ufs/core/ufs-mcq.c     | 62 ++++++++++++++++++++++++++++++++++++++++++
 drivers/ufs/core/ufshcd-priv.h |  5 +++-
 drivers/ufs/core/ufshcd.c      | 11 ++++++--
 3 files changed, 74 insertions(+), 4 deletions(-)
drivers/ufs/core/ufs-mcq.c +62 -0
index 6f6b22c..78bd5d6 100644
@@ -603,3 +603,65 @@ static bool ufshcd_mcq_sqe_search(struct ufs_hba *hba,
 	spin_unlock(&hwq->sq_lock);
 	return ret;
 }
+
+/**
+ * ufshcd_mcq_abort - Abort the command in MCQ.
+ * @cmd - The command to be aborted.
+ *
+ * Returns SUCCESS or FAILED error codes
+ */
+int ufshcd_mcq_abort(struct scsi_cmnd *cmd)
+{
+	struct Scsi_Host *host = cmd->device->host;
+	struct ufs_hba *hba = shost_priv(host);
+	int tag = scsi_cmd_to_rq(cmd)->tag;
+	struct ufshcd_lrb *lrbp = &hba->lrb[tag];
+	struct ufs_hw_queue *hwq;
+	int err = FAILED;
+
+	if (!lrbp->cmd) {
+		dev_err(hba->dev,
+			"%s: skip abort. cmd at tag %d already completed.\n",
+			__func__, tag);
+		goto out;
+	}
+
+	/* Skip task abort in case previous aborts failed and report failure */
+	if (lrbp->req_abort_skip) {
+		dev_err(hba->dev, "%s: skip abort. tag %d failed earlier\n",
+			__func__, tag);
+		goto out;
+	}
+
+	hwq = ufshcd_mcq_req_to_hwq(hba, scsi_cmd_to_rq(cmd));
+
+	if (ufshcd_mcq_sqe_search(hba, hwq, tag)) {
+		/*
+		 * Failure. The command should not be "stuck" in SQ for
+		 * a long time which resulted in command being aborted.
+		 */
+		dev_err(hba->dev, "%s: cmd found in sq. hwq=%d, tag=%d\n",
+				__func__, hwq->id, tag);
+		/* Set the Command Type to 0xF per the spec */
+		ufshcd_mcq_nullify_cmd(hba, hwq);
+		goto out;
+	}
+
+	/*
+	 * The command is not in the Submission Queue, and it is not
+	 * in the Completion Queue either. Query the device to see if
+	 * the command is being processed in the device.
+	 */
+	if (ufshcd_try_to_abort_task(hba, tag)) {
+		dev_err(hba->dev, "%s: device abort failed %d\n", __func__, err);
+		lrbp->req_abort_skip = true;
+		goto out;
+	}
+
+	err = SUCCESS;
+	if (lrbp->cmd)
+		ufshcd_release_scsi_cmd(hba, lrbp);
+
+out:
+	return err;
+}
drivers/ufs/core/ufshcd-priv.h +4 -1
index 1a40cf7..ef66151 100644
@@ -79,7 +79,10 @@ unsigned long ufshcd_mcq_poll_cqe_lock(struct ufs_hba *hba,
 				       struct ufs_hw_queue *hwq);
 
 int ufshcd_mcq_sq_cleanup(struct ufs_hba *hba, int task_tag, int *result);
-
+int ufshcd_mcq_abort(struct scsi_cmnd *cmd);
+int ufshcd_try_to_abort_task(struct ufs_hba *hba, int tag);
+void ufshcd_release_scsi_cmd(struct ufs_hba *hba,
+				struct ufshcd_lrb *lrbp);
 #define UFSHCD_MCQ_IO_QUEUE_OFFSET	1
 #define SD_ASCII_STD true
 #define SD_RAW false
drivers/ufs/core/ufshcd.c +8 -3
index ffccb91..84efb52 100644
@@ -299,7 +299,6 @@ static int ufshcd_setup_hba_vreg(struct ufs_hba *hba, bool on);
 static int ufshcd_setup_vreg(struct ufs_hba *hba, bool on);
 static inline int ufshcd_config_vreg_hpm(struct ufs_hba *hba,
 					 struct ufs_vreg *vreg);
-static int ufshcd_try_to_abort_task(struct ufs_hba *hba, int tag);
 static void ufshcd_wb_toggle_buf_flush_during_h8(struct ufs_hba *hba,
 						 bool enable);
 static void ufshcd_hba_vreg_set_lpm(struct ufs_hba *hba);
@@ -5438,7 +5437,7 @@ static irqreturn_t ufshcd_uic_cmd_compl(struct ufs_hba *hba, u32 intr_status)
 }
 
 /* Release the resources allocated for processing a SCSI command. */
-static void ufshcd_release_scsi_cmd(struct ufs_hba *hba,
+void ufshcd_release_scsi_cmd(struct ufs_hba *hba,
 				    struct ufshcd_lrb *lrbp)
 {
 	struct scsi_cmnd *cmd = lrbp->cmd;
@@ -7374,7 +7373,7 @@ static void ufshcd_set_req_abort_skip(struct ufs_hba *hba, unsigned long bitmap)
  *
  * Returns zero on success, non-zero on failure
  */
-static int ufshcd_try_to_abort_task(struct ufs_hba *hba, int tag)
+int ufshcd_try_to_abort_task(struct ufs_hba *hba, int tag)
 {
 	struct ufshcd_lrb *lrbp = &hba->lrb[tag];
 	int err = 0;
@@ -7534,6 +7533,12 @@ static int ufshcd_abort(struct scsi_cmnd *cmd)
 		goto release;
 	}
 
+	if (is_mcq_enabled(hba)) {
+		/* MCQ mode. Branch off to handle abort for mcq mode */
+		err = ufshcd_mcq_abort(cmd);
+		goto release;
+	}
+
 	/* Skip task abort in case previous aborts failed and report failure */
 	if (lrbp->req_abort_skip) {
 		dev_err(hba->dev, "%s: skipping abort\n", __func__);
Signature · 1 line
2.7.4
#6 lore raw

PATCH 4/5 ufs: mcq: Use ufshcd_mcq_poll_cqe_lock() in mcq mode

In reply to Bao D. Nguyen

In preparation for adding mcq error handler support, update the mcq code to use the ufshcd_mcq_poll_cqe_lock() in interrupt context instead of using ufshcd_mcq_poll_cqe_nolock(). This is to keep synchronization between mcq interrupt and error handler contexts because both need to access the mcq hardware in separate contexts.

---
 drivers/ufs/core/ufs-mcq.c     | 6 +++---
 drivers/ufs/core/ufshcd-priv.h | 2 --
 drivers/ufs/core/ufshcd.c      | 2 +-
 drivers/ufs/host/ufs-qcom.c    | 2 +-
 include/ufs/ufshcd.h           | 2 +-
 5 files changed, 6 insertions(+), 8 deletions(-)
drivers/ufs/core/ufs-mcq.c +3 -3
index 78bd5d6..1ae2ec4 100644
@@ -284,8 +284,8 @@ static void ufshcd_mcq_process_cqe(struct ufs_hba *hba,
 	ufshcd_compl_one_cqe(hba, tag, cqe);
 }
 
-unsigned long ufshcd_mcq_poll_cqe_nolock(struct ufs_hba *hba,
-					 struct ufs_hw_queue *hwq)
+static unsigned long ufshcd_mcq_poll_cqe_nolock(struct ufs_hba *hba,
+						struct ufs_hw_queue *hwq)
 {
 	unsigned long completed_reqs = 0;
 
@@ -301,7 +301,6 @@ unsigned long ufshcd_mcq_poll_cqe_nolock(struct ufs_hba *hba,
 
 	return completed_reqs;
 }
-EXPORT_SYMBOL_GPL(ufshcd_mcq_poll_cqe_nolock);
 
 unsigned long ufshcd_mcq_poll_cqe_lock(struct ufs_hba *hba,
 				       struct ufs_hw_queue *hwq)
@@ -314,6 +313,7 @@ unsigned long ufshcd_mcq_poll_cqe_lock(struct ufs_hba *hba,
 
 	return completed_reqs;
 }
+EXPORT_SYMBOL_GPL(ufshcd_mcq_poll_cqe_lock);
 
 void ufshcd_mcq_make_queues_operational(struct ufs_hba *hba)
 {
drivers/ufs/core/ufshcd-priv.h +0 -2
index ef66151..70b7ffc 100644
@@ -71,8 +71,6 @@ void ufshcd_mcq_config_mac(struct ufs_hba *hba, u32 max_active_cmds);
 void ufshcd_mcq_select_mcq_mode(struct ufs_hba *hba);
 u32 ufshcd_mcq_read_cqis(struct ufs_hba *hba, int i);
 void ufshcd_mcq_write_cqis(struct ufs_hba *hba, u32 val, int i);
-unsigned long ufshcd_mcq_poll_cqe_nolock(struct ufs_hba *hba,
-					 struct ufs_hw_queue *hwq);
 struct ufs_hw_queue *ufshcd_mcq_req_to_hwq(struct ufs_hba *hba,
 					   struct request *req);
 unsigned long ufshcd_mcq_poll_cqe_lock(struct ufs_hba *hba,
drivers/ufs/core/ufshcd.c +1 -1
index 84efb52..fef1907 100644
@@ -6792,7 +6792,7 @@ static irqreturn_t ufshcd_handle_mcq_cq_events(struct ufs_hba *hba)
 			ufshcd_mcq_write_cqis(hba, events, i);
 
 		if (events & UFSHCD_MCQ_CQIS_TAIL_ENT_PUSH_STS)
-			ufshcd_mcq_poll_cqe_nolock(hba, hwq);
+			ufshcd_mcq_poll_cqe_lock(hba, hwq);
 	}
 
 	return IRQ_HANDLED;
drivers/ufs/host/ufs-qcom.c +1 -1
index 6568fdf..34d3814 100644
@@ -1558,7 +1558,7 @@ static irqreturn_t ufs_qcom_mcq_esi_handler(int irq, void *__hba)
 	struct ufs_hw_queue *hwq = &hba->uhq[id];
 
 	ufshcd_mcq_write_cqis(hba, 0x1, id);
-	ufshcd_mcq_poll_cqe_nolock(hba, hwq);
+	ufshcd_mcq_poll_cqe_lock(hba, hwq);
 
 	return IRQ_HANDLED;
 }
include/ufs/ufshcd.h +1 -1
index 25aab8e..3fca558 100644
@@ -1246,7 +1246,7 @@ void ufshcd_update_evt_hist(struct ufs_hba *hba, u32 id, u32 val);
 void ufshcd_hba_stop(struct ufs_hba *hba);
 void ufshcd_schedule_eh_work(struct ufs_hba *hba);
 void ufshcd_mcq_write_cqis(struct ufs_hba *hba, u32 val, int i);
-unsigned long ufshcd_mcq_poll_cqe_nolock(struct ufs_hba *hba,
+unsigned long ufshcd_mcq_poll_cqe_lock(struct ufs_hba *hba,
 					 struct ufs_hw_queue *hwq);
 void ufshcd_mcq_enable_esi(struct ufs_hba *hba);
 void ufshcd_mcq_config_esi(struct ufs_hba *hba, struct msi_msg *msg);
Signature · 1 line
2.7.4
#7 lore raw

PATCH 5/5 ufs: core: Add error handling for MCQ mode

In reply to Bao D. Nguyen

Add support for error handling for MCQ mode.

---
 drivers/ufs/core/ufshcd.c | 80 ++++++++++++++++++++++++++++++++++++++++-------
 1 file changed, 69 insertions(+), 11 deletions(-)
drivers/ufs/core/ufshcd.c +69 -11
index fef1907..e947f7f 100644
@@ -3127,6 +3127,12 @@ static int ufshcd_wait_for_dev_cmd(struct ufs_hba *hba,
 		err = -ETIMEDOUT;
 		dev_dbg(hba->dev, "%s: dev_cmd request timedout, tag %d\n",
 			__func__, lrbp->task_tag);
+
+		/* MCQ mode */
+		if (is_mcq_enabled(hba))
+			return ufshcd_clear_cmds(hba, 1UL << lrbp->task_tag);
+
+		/* SDB mode */
 		if (ufshcd_clear_cmds(hba, 1UL << lrbp->task_tag) == 0) {
 			/* successfully cleared the command, retry if needed */
 			err = -EAGAIN;
@@ -5562,6 +5568,10 @@ static int ufshcd_poll(struct Scsi_Host *shost, unsigned int queue_num)
  */
 static irqreturn_t ufshcd_transfer_req_compl(struct ufs_hba *hba)
 {
+	struct ufshcd_lrb *lrbp;
+	u32 hwq_num, utag;
+	int tag;
+
 	/* Resetting interrupt aggregation counters first and reading the
 	 * DOOR_BELL afterward allows us to handle all the completed requests.
 	 * In order to prevent other interrupts starvation the DB is read once
@@ -5580,7 +5590,22 @@ static irqreturn_t ufshcd_transfer_req_compl(struct ufs_hba *hba)
 	 * Ignore the ufshcd_poll() return value and return IRQ_HANDLED since we
 	 * do not want polling to trigger spurious interrupt complaints.
 	 */
-	ufshcd_poll(hba->host, UFSHCD_POLL_FROM_INTERRUPT_CONTEXT);
+	if (!is_mcq_enabled(hba)) {
+		ufshcd_poll(hba->host, UFSHCD_POLL_FROM_INTERRUPT_CONTEXT);
+		goto out;
+	}
+
+	/* MCQ mode */
+	for (tag = 0; tag < hba->nutrs; tag++) {
+		lrbp = &hba->lrb[tag];
+		if (lrbp->cmd) {
+			utag = blk_mq_unique_tag(scsi_cmd_to_rq(lrbp->cmd));
+			hwq_num = blk_mq_unique_tag_to_hwq(utag);
+			ufshcd_poll(hba->host, hwq_num);
+		}
+	}
+
+out:
 
 	return IRQ_HANDLED;
 }
@@ -6359,18 +6384,36 @@ static bool ufshcd_abort_all(struct ufs_hba *hba)
 	bool needs_reset = false;
 	int tag, ret;
 
-	/* Clear pending transfer requests */
-	for_each_set_bit(tag, &hba->outstanding_reqs, hba->nutrs) {
-		ret = ufshcd_try_to_abort_task(hba, tag);
-		dev_err(hba->dev, "Aborting tag %d / CDB %#02x %s\n", tag,
-			hba->lrb[tag].cmd ? hba->lrb[tag].cmd->cmnd[0] : -1,
-			ret ? "failed" : "succeeded");
-		if (ret) {
-			needs_reset = true;
-			goto out;
+	if (is_mcq_enabled(hba)) {
+		struct ufshcd_lrb *lrbp;
+		int tag;
+
+		for (tag = 0; tag < hba->nutrs; tag++) {
+			lrbp = &hba->lrb[tag];
+			if (lrbp->cmd) {
+				ret = ufshcd_try_to_abort_task(hba, tag);
+				dev_err(hba->dev, "Aborting tag %d / CDB %#02x %s\n", tag,
+					hba->lrb[tag].cmd ? hba->lrb[tag].cmd->cmnd[0] : -1,
+					ret ? "failed" : "succeeded");
+			}
+			if (ret) {
+				needs_reset = true;
+				goto out;
+			}
+		}
+	} else {
+		/* Clear pending transfer requests */
+		for_each_set_bit(tag, &hba->outstanding_reqs, hba->nutrs) {
+			ret = ufshcd_try_to_abort_task(hba, tag);
+			dev_err(hba->dev, "Aborting tag %d / CDB %#02x %s\n", tag,
+				hba->lrb[tag].cmd ? hba->lrb[tag].cmd->cmnd[0] : -1,
+				ret ? "failed" : "succeeded");
+			if (ret) {
+				needs_reset = true;
+				goto out;
+			}
 		}
 	}
-
 	/* Clear pending task management requests */
 	for_each_set_bit(tag, &hba->outstanding_tasks, hba->nutmrs) {
 		if (ufshcd_clear_tm_cmd(hba, tag)) {
@@ -7302,6 +7345,8 @@ static int ufshcd_eh_device_reset_handler(struct scsi_cmnd *cmd)
 	unsigned long flags, pending_reqs = 0, not_cleared = 0;
 	struct Scsi_Host *host;
 	struct ufs_hba *hba;
+	struct ufs_hw_queue *hwq;
+	struct ufshcd_lrb *lrbp;
 	u32 pos;
 	int err;
 	u8 resp = 0xF, lun;
@@ -7317,6 +7362,19 @@ static int ufshcd_eh_device_reset_handler(struct scsi_cmnd *cmd)
 		goto out;
 	}
 
+	if (is_mcq_enabled(hba)) {
+		for (pos = 0; pos < hba->nutrs; pos++) {
+			lrbp = &hba->lrb[pos];
+			if (lrbp->cmd && lrbp->lun == lun) {
+				ufshcd_clear_cmds(hba, 1UL << pos);
+				hwq = ufshcd_mcq_req_to_hwq(hba, scsi_cmd_to_rq(lrbp->cmd));
+				ufshcd_mcq_poll_cqe_lock(hba, hwq);
+			}
+		}
+		err = 0;
+		goto out;
+	}
+
 	/* clear the commands that were pending for corresponding LUN */
 	spin_lock_irqsave(&hba->outstanding_lock, flags);
 	for_each_set_bit(pos, &hba->outstanding_reqs, hba->nutrs)
Signature · 1 line
2.7.4
#8 lore raw

Re: [PATCH v1 5/5] ufs: core: Add error handling for MCQ mode

Hi Bao,

Bao D. Nguyen <[email protected]> 於 2023年3月29日 週三 下午6:14寫道:

Quoted 13 lines

Add support for error handling for MCQ mode.

---
 drivers/ufs/core/ufshcd.c | 80 ++++++++++++++++++++++++++++++++++++++++-------
 1 file changed, 69 insertions(+), 11 deletions(-)
drivers/ufs/core/ufshcd.c +0 -0
index fef1907..e947f7f 100644

drivers/ufs/core/ufshcd.c static int ufshcd_wait_for_dev_cmd(struct ufs_hba *hba, · line 3133

Quoted 7 lines
drivers/ufs/core/ufshcd.c +4 -0
@@ -3127,6 +3127,12 @@ static int ufshcd_wait_for_dev_cmd(struct ufs_hba *hba,
                err = -ETIMEDOUT;
                dev_dbg(hba->dev, "%s: dev_cmd request timedout, tag %d\n",
                        __func__, lrbp->task_tag);
+
+               /* MCQ mode */
+               if (is_mcq_enabled(hba))
+                       return ufshcd_clear_cmds(hba, 1UL << lrbp->task_tag);

When a time-out occurs during the command-clearing process, it appears that the MCQ path does not properly clear 'hba->dev_cmd.complete'. This could result in a null pointer reference if the device command interrupt arrives at a later time.

Could you please help check this?

Thanks,
Stanley Chu

Quoted 119 lines
drivers/ufs/core/ufshcd.c +65 -12
+
+               /* SDB mode */
                if (ufshcd_clear_cmds(hba, 1UL << lrbp->task_tag) == 0) {
                        /* successfully cleared the command, retry if needed */
                        err = -EAGAIN;
@@ -5562,6 +5568,10 @@ static int ufshcd_poll(struct Scsi_Host *shost, unsigned int queue_num)
  */
 static irqreturn_t ufshcd_transfer_req_compl(struct ufs_hba *hba)
 {
+       struct ufshcd_lrb *lrbp;
+       u32 hwq_num, utag;
+       int tag;
+
        /* Resetting interrupt aggregation counters first and reading the
         * DOOR_BELL afterward allows us to handle all the completed requests.
         * In order to prevent other interrupts starvation the DB is read once
@@ -5580,7 +5590,22 @@ static irqreturn_t ufshcd_transfer_req_compl(struct ufs_hba *hba)
         * Ignore the ufshcd_poll() return value and return IRQ_HANDLED since we
         * do not want polling to trigger spurious interrupt complaints.
         */
-       ufshcd_poll(hba->host, UFSHCD_POLL_FROM_INTERRUPT_CONTEXT);
+       if (!is_mcq_enabled(hba)) {
+               ufshcd_poll(hba->host, UFSHCD_POLL_FROM_INTERRUPT_CONTEXT);
+               goto out;
+       }
+
+       /* MCQ mode */
+       for (tag = 0; tag < hba->nutrs; tag++) {
+               lrbp = &hba->lrb[tag];
+               if (lrbp->cmd) {
+                       utag = blk_mq_unique_tag(scsi_cmd_to_rq(lrbp->cmd));
+                       hwq_num = blk_mq_unique_tag_to_hwq(utag);
+                       ufshcd_poll(hba->host, hwq_num);
+               }
+       }
+
+out:
 
        return IRQ_HANDLED;
 }
@@ -6359,18 +6384,36 @@ static bool ufshcd_abort_all(struct ufs_hba *hba)
        bool needs_reset = false;
        int tag, ret;
 
-       /* Clear pending transfer requests */
-       for_each_set_bit(tag, &hba->outstanding_reqs, hba->nutrs) {
-               ret = ufshcd_try_to_abort_task(hba, tag);
-               dev_err(hba->dev, "Aborting tag %d / CDB %#02x %s\n", tag,
-                       hba->lrb[tag].cmd ? hba->lrb[tag].cmd->cmnd[0] : -1,
-                       ret ? "failed" : "succeeded");
-               if (ret) {
-                       needs_reset = true;
-                       goto out;
+       if (is_mcq_enabled(hba)) {
+               struct ufshcd_lrb *lrbp;
+               int tag;
+
+               for (tag = 0; tag < hba->nutrs; tag++) {
+                       lrbp = &hba->lrb[tag];
+                       if (lrbp->cmd) {
+                               ret = ufshcd_try_to_abort_task(hba, tag);
+                               dev_err(hba->dev, "Aborting tag %d / CDB %#02x %s\n", tag,
+                                       hba->lrb[tag].cmd ? hba->lrb[tag].cmd->cmnd[0] : -1,
+                                       ret ? "failed" : "succeeded");
+                       }
+                       if (ret) {
+                               needs_reset = true;
+                               goto out;
+                       }
+               }
+       } else {
+               /* Clear pending transfer requests */
+               for_each_set_bit(tag, &hba->outstanding_reqs, hba->nutrs) {
+                       ret = ufshcd_try_to_abort_task(hba, tag);
+                       dev_err(hba->dev, "Aborting tag %d / CDB %#02x %s\n", tag,
+                               hba->lrb[tag].cmd ? hba->lrb[tag].cmd->cmnd[0] : -1,
+                               ret ? "failed" : "succeeded");
+                       if (ret) {
+                               needs_reset = true;
+                               goto out;
+                       }
                }
        }
-
        /* Clear pending task management requests */
        for_each_set_bit(tag, &hba->outstanding_tasks, hba->nutmrs) {
                if (ufshcd_clear_tm_cmd(hba, tag)) {
@@ -7302,6 +7345,8 @@ static int ufshcd_eh_device_reset_handler(struct scsi_cmnd *cmd)
        unsigned long flags, pending_reqs = 0, not_cleared = 0;
        struct Scsi_Host *host;
        struct ufs_hba *hba;
+       struct ufs_hw_queue *hwq;
+       struct ufshcd_lrb *lrbp;
        u32 pos;
        int err;
        u8 resp = 0xF, lun;
@@ -7317,6 +7362,19 @@ static int ufshcd_eh_device_reset_handler(struct scsi_cmnd *cmd)
                goto out;
        }
 
+       if (is_mcq_enabled(hba)) {
+               for (pos = 0; pos < hba->nutrs; pos++) {
+                       lrbp = &hba->lrb[pos];
+                       if (lrbp->cmd && lrbp->lun == lun) {
+                               ufshcd_clear_cmds(hba, 1UL << pos);
+                               hwq = ufshcd_mcq_req_to_hwq(hba, scsi_cmd_to_rq(lrbp->cmd));
+                               ufshcd_mcq_poll_cqe_lock(hba, hwq);
+                       }
+               }
+               err = 0;
+               goto out;
+       }
+
        /* clear the commands that were pending for corresponding LUN */
        spin_lock_irqsave(&hba->outstanding_lock, flags);
        for_each_set_bit(pos, &hba->outstanding_reqs, hba->nutrs)
--

2.7.4

#9 lore raw

Re: [PATCH v1 5/5] ufs: core: Add error handling for MCQ mode

Quoted 29 lines · Stanley Chu

Hi Bao,

Bao D. Nguyen <[email protected]> 於 2023年3月29日 週三 下午6:14寫道:

Quoted 13 lines

Add support for error handling for MCQ mode.

---
  drivers/ufs/core/ufshcd.c | 80 ++++++++++++++++++++++++++++++++++++++++-------
  1 file changed, 69 insertions(+), 11 deletions(-)
drivers/ufs/core/ufshcd.c +0 -0
index fef1907..e947f7f 100644

drivers/ufs/core/ufshcd.c static int ufshcd_wait_for_dev_cmd(struct ufs_hba *hba, · line 3133

Quoted 7 lines
drivers/ufs/core/ufshcd.c +4 -0
@@ -3127,6 +3127,12 @@ static int ufshcd_wait_for_dev_cmd(struct ufs_hba *hba,
                 err = -ETIMEDOUT;
                 dev_dbg(hba->dev, "%s: dev_cmd request timedout, tag %d\n",
                         __func__, lrbp->task_tag);
+
+               /* MCQ mode */
+               if (is_mcq_enabled(hba))
+                       return ufshcd_clear_cmds(hba, 1UL << lrbp->task_tag);

When a time-out occurs during the command-clearing process, it appears that the MCQ path does not properly clear 'hba->dev_cmd.complete'. This could result in a null pointer reference if the device command interrupt arrives at a later time.

Quoted 1 line

Could you please help check this?

Thanks Stanley. I will take a look.

Quoted 123 lines

Thanks,
Stanley Chu

Quoted 119 lines
drivers/ufs/core/ufshcd.c +65 -12
+
+               /* SDB mode */
                 if (ufshcd_clear_cmds(hba, 1UL << lrbp->task_tag) == 0) {
                         /* successfully cleared the command, retry if needed */
                         err = -EAGAIN;
@@ -5562,6 +5568,10 @@ static int ufshcd_poll(struct Scsi_Host *shost, unsigned int queue_num)
   */
  static irqreturn_t ufshcd_transfer_req_compl(struct ufs_hba *hba)
  {
+       struct ufshcd_lrb *lrbp;
+       u32 hwq_num, utag;
+       int tag;
+
         /* Resetting interrupt aggregation counters first and reading the
          * DOOR_BELL afterward allows us to handle all the completed requests.
          * In order to prevent other interrupts starvation the DB is read once
@@ -5580,7 +5590,22 @@ static irqreturn_t ufshcd_transfer_req_compl(struct ufs_hba *hba)
          * Ignore the ufshcd_poll() return value and return IRQ_HANDLED since we
          * do not want polling to trigger spurious interrupt complaints.
          */
-       ufshcd_poll(hba->host, UFSHCD_POLL_FROM_INTERRUPT_CONTEXT);
+       if (!is_mcq_enabled(hba)) {
+               ufshcd_poll(hba->host, UFSHCD_POLL_FROM_INTERRUPT_CONTEXT);
+               goto out;
+       }
+
+       /* MCQ mode */
+       for (tag = 0; tag < hba->nutrs; tag++) {
+               lrbp = &hba->lrb[tag];
+               if (lrbp->cmd) {
+                       utag = blk_mq_unique_tag(scsi_cmd_to_rq(lrbp->cmd));
+                       hwq_num = blk_mq_unique_tag_to_hwq(utag);
+                       ufshcd_poll(hba->host, hwq_num);
+               }
+       }
+
+out:
 
         return IRQ_HANDLED;
  }
@@ -6359,18 +6384,36 @@ static bool ufshcd_abort_all(struct ufs_hba *hba)
         bool needs_reset = false;
         int tag, ret;
 
-       /* Clear pending transfer requests */
-       for_each_set_bit(tag, &hba->outstanding_reqs, hba->nutrs) {
-               ret = ufshcd_try_to_abort_task(hba, tag);
-               dev_err(hba->dev, "Aborting tag %d / CDB %#02x %s\n", tag,
-                       hba->lrb[tag].cmd ? hba->lrb[tag].cmd->cmnd[0] : -1,
-                       ret ? "failed" : "succeeded");
-               if (ret) {
-                       needs_reset = true;
-                       goto out;
+       if (is_mcq_enabled(hba)) {
+               struct ufshcd_lrb *lrbp;
+               int tag;
+
+               for (tag = 0; tag < hba->nutrs; tag++) {
+                       lrbp = &hba->lrb[tag];
+                       if (lrbp->cmd) {
+                               ret = ufshcd_try_to_abort_task(hba, tag);
+                               dev_err(hba->dev, "Aborting tag %d / CDB %#02x %s\n", tag,
+                                       hba->lrb[tag].cmd ? hba->lrb[tag].cmd->cmnd[0] : -1,
+                                       ret ? "failed" : "succeeded");
+                       }
+                       if (ret) {
+                               needs_reset = true;
+                               goto out;
+                       }
+               }
+       } else {
+               /* Clear pending transfer requests */
+               for_each_set_bit(tag, &hba->outstanding_reqs, hba->nutrs) {
+                       ret = ufshcd_try_to_abort_task(hba, tag);
+                       dev_err(hba->dev, "Aborting tag %d / CDB %#02x %s\n", tag,
+                               hba->lrb[tag].cmd ? hba->lrb[tag].cmd->cmnd[0] : -1,
+                               ret ? "failed" : "succeeded");
+                       if (ret) {
+                               needs_reset = true;
+                               goto out;
+                       }
                 }
         }
-
         /* Clear pending task management requests */
         for_each_set_bit(tag, &hba->outstanding_tasks, hba->nutmrs) {
                 if (ufshcd_clear_tm_cmd(hba, tag)) {
@@ -7302,6 +7345,8 @@ static int ufshcd_eh_device_reset_handler(struct scsi_cmnd *cmd)
         unsigned long flags, pending_reqs = 0, not_cleared = 0;
         struct Scsi_Host *host;
         struct ufs_hba *hba;
+       struct ufs_hw_queue *hwq;
+       struct ufshcd_lrb *lrbp;
         u32 pos;
         int err;
         u8 resp = 0xF, lun;
@@ -7317,6 +7362,19 @@ static int ufshcd_eh_device_reset_handler(struct scsi_cmnd *cmd)
                 goto out;
         }
 
+       if (is_mcq_enabled(hba)) {
+               for (pos = 0; pos < hba->nutrs; pos++) {
+                       lrbp = &hba->lrb[pos];
+                       if (lrbp->cmd && lrbp->lun == lun) {
+                               ufshcd_clear_cmds(hba, 1UL << pos);
+                               hwq = ufshcd_mcq_req_to_hwq(hba, scsi_cmd_to_rq(lrbp->cmd));
+                               ufshcd_mcq_poll_cqe_lock(hba, hwq);
+                       }
+               }
+               err = 0;
+               goto out;
+       }
+
         /* clear the commands that were pending for corresponding LUN */
         spin_lock_irqsave(&hba->outstanding_lock, flags);
         for_each_set_bit(pos, &hba->outstanding_reqs, hba->nutrs)
--

2.7.4

#10 lore raw
In reply to Bao D. Nguyen
Quoted 3 lines · Bao D. Nguyen

This patch series enable support for ufshcd_abort() and error handler in MCQ mode. The first 3 patches are for supporting ufshcd_abort().
The 4th and 5th patches are for supporting error handler.

Is this perhaps a resend of v1? Last time this series was posted it had version number 3.

Thanks,

Bart.

#11 lore raw
Quoted 7 lines · Bart Van Assche
Quoted 3 lines · Bao D. Nguyen

This patch series enable support for ufshcd_abort() and error handler in MCQ mode. The first 3 patches are for supporting ufshcd_abort().
The 4th and 5th patches are for supporting error handler.

Is this perhaps a resend of v1? Last time this series was posted it had version number 3.

Hi Bart, I am not sure if I did it correctly or not. The last time I posted, it was with RFC tag in the subject. I have made some minor changes to address the comments and some bug fixes. I am posting the patches again without "RFC" tag now.

Thanks,

Bao

Quoted 3 lines

Thanks,

Bart.

#12 lore raw
Quoted 1 line · Bao D. Nguyen

Hi Bart, I am not sure if I did it correctly or not. The last time I posted, it was with RFC tag in the subject. I have made some minor changes to address the comments and some bug fixes. I am posting the patches again without "RFC" tag now.

The change history is missing. This could have been made clear in the change history. Anyway, I will take a closer look at these patches.

Bart.

#13 lore raw
Quoted 3 lines · Bart Van Assche

The change history is missing. This could have been made clear in the change history. Anyway, I will take a closer look at these patches.

Bart.

Got it. I will add the change history back in the next revision.

Thanks,
Bao