[PATCH v2 0/3] ufs: Fix a race between the interrupt handler and the reset handler

[PATCH v2 0/3] ufs: Fix a race between the interrupt handler and the reset handler

[Bart Van Assche]

Hi Martin,

This patch series is version two of a fix between the UFS interrupt handler and
reset handlers. Please consider this patch series for kernel v5.20.

Changes compared to v1:
- Converted a single patch into three patches.
- Modified patch 3/3 such that only cleared requests are completed.

Bart Van Assche (3):
  scsi: ufs: Simplify ufshcd_clear_cmd()
  scsi: ufs: Support clearing multiple commands at once
  scsi: ufs: Fix a race between the interrupt handler and the reset
    handler

 drivers/ufs/core/ufshcd.c | 76 ++++++++++++++++++++++++---------------
 1 file changed, 48 insertions(+), 28 deletions(-)

[PATCH v2 1/3] scsi: ufs: Simplify ufshcd_clear_cmd()

[Bart Van Assche]

Remove the local variable 'err'. This patch does not change any
functionality.

Signed-off-by: Bart Van Assche <bvanassche@acm.org>
---
 drivers/ufs/core/ufshcd.c | 8 ++------
 1 file changed, 2 insertions(+), 6 deletions(-)

diff --git a/drivers/ufs/core/ufshcd.c b/drivers/ufs/core/ufshcd.c
index bb6cbd514a69..4ed2f6c29e80 100644
--- a/drivers/ufs/core/ufshcd.c
+++ b/drivers/ufs/core/ufshcd.c
@@ -2868,7 +2868,6 @@ static int ufshcd_compose_dev_cmd(struct ufs_hba *hba,
 static int
 ufshcd_clear_cmd(struct ufs_hba *hba, int tag)
 {
-	int err = 0;
 	unsigned long flags;
 	u32 mask = 1 << tag;
 
@@ -2881,11 +2880,8 @@ ufshcd_clear_cmd(struct ufs_hba *hba, int tag)
 	 * wait for h/w to clear corresponding bit in door-bell.
 	 * max. wait is 1 sec.
 	 */
-	err = ufshcd_wait_for_register(hba,
-			REG_UTP_TRANSFER_REQ_DOOR_BELL,
-			mask, ~mask, 1000, 1000);
-
-	return err;
+	return ufshcd_wait_for_register(hba, REG_UTP_TRANSFER_REQ_DOOR_BELL,
+					mask, ~mask, 1000, 1000);
 }
 
 static int

[PATCH v2 2/3] scsi: ufs: Support clearing multiple commands at once

[Bart Van Assche]

Modify ufshcd_clear_cmd() such that it supports clearing multiple
commands at once instead of one command at a time. This change will be
used in a later patch to reduce the time spent in the reset handler.

Signed-off-by: Bart Van Assche <bvanassche@acm.org>
---
 drivers/ufs/core/ufshcd.c | 42 ++++++++++++++++++++++++++-------------
 1 file changed, 28 insertions(+), 14 deletions(-)

diff --git a/drivers/ufs/core/ufshcd.c b/drivers/ufs/core/ufshcd.c
index 4ed2f6c29e80..b7120c5f0920 100644
--- a/drivers/ufs/core/ufshcd.c
+++ b/drivers/ufs/core/ufshcd.c
@@ -748,17 +748,28 @@ static enum utp_ocs ufshcd_get_tr_ocs(struct ufshcd_lrb *lrbp)
 }
 
 /**
- * ufshcd_utrl_clear - Clear a bit in UTRLCLR register
+ * ufshcd_utrl_clear() - Clear requests from the controller request list.
  * @hba: per adapter instance
- * @pos: position of the bit to be cleared
+ * @mask: mask with one bit set for each request to be cleared
  */
-static inline void ufshcd_utrl_clear(struct ufs_hba *hba, u32 pos)
+static inline void ufshcd_utrl_clear(struct ufs_hba *hba, u32 mask)
 {
 	if (hba->quirks & UFSHCI_QUIRK_BROKEN_REQ_LIST_CLR)
-		ufshcd_writel(hba, (1 << pos), REG_UTP_TRANSFER_REQ_LIST_CLEAR);
-	else
-		ufshcd_writel(hba, ~(1 << pos),
-				REG_UTP_TRANSFER_REQ_LIST_CLEAR);
+		mask = ~mask;
+	/*
+	 * From the UFSHCI specification: "UTP Transfer Request List CLear
+	 * Register (UTRLCLR): This field is bit significant. Each bit
+	 * corresponds to a slot in the UTP Transfer Request List, where bit 0
+	 * corresponds to request slot 0. A bit in this field is set to ‘0’
+	 * by host software to indicate to the host controller that a transfer
+	 * request slot is cleared. The host controller
+	 * shall free up any resources associated to the request slot
+	 * immediately, and shall set the associated bit in UTRLDBR to ‘0’. The
+	 * host software indicates no change to request slots by setting the
+	 * associated bits in this field to ‘1’. Bits in this field shall only
+	 * be set ‘1’ or ‘0’ by host software when UTRLRSR is set to ‘1’."
+	 */
+	ufshcd_writel(hba, ~mask, REG_UTP_TRANSFER_REQ_LIST_CLEAR);
 }
 
 /**
@@ -2865,15 +2876,18 @@ static int ufshcd_compose_dev_cmd(struct ufs_hba *hba,
 	return ufshcd_compose_devman_upiu(hba, lrbp);
 }
 
-static int
-ufshcd_clear_cmd(struct ufs_hba *hba, int tag)
+/*
+ * Clear all the requests from the controller for which a bit has been set in
+ * @mask and wait until the controller confirms that these requests have been
+ * cleared.
+ */
+static int ufshcd_clear_cmds(struct ufs_hba *hba, u32 mask)
 {
 	unsigned long flags;
-	u32 mask = 1 << tag;
 
 	/* clear outstanding transaction before retry */
 	spin_lock_irqsave(hba->host->host_lock, flags);
-	ufshcd_utrl_clear(hba, tag);
+	ufshcd_utrl_clear(hba, mask);
 	spin_unlock_irqrestore(hba->host->host_lock, flags);
 
 	/*
@@ -2961,7 +2975,7 @@ 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_cmd(hba, lrbp->task_tag))
+		if (!ufshcd_clear_cmds(hba, 1U << lrbp->task_tag))
 			/* successfully cleared the command, retry if needed */
 			err = -EAGAIN;
 		/*
@@ -6989,7 +7003,7 @@ static int ufshcd_eh_device_reset_handler(struct scsi_cmnd *cmd)
 	/* clear the commands that were pending for corresponding LUN */
 	for_each_set_bit(pos, &hba->outstanding_reqs, hba->nutrs) {
 		if (hba->lrb[pos].lun == lun) {
-			err = ufshcd_clear_cmd(hba, pos);
+			err = ufshcd_clear_cmds(hba, 1U << pos);
 			if (err)
 				break;
 			__ufshcd_transfer_req_compl(hba, 1U << pos);
@@ -7091,7 +7105,7 @@ static int ufshcd_try_to_abort_task(struct ufs_hba *hba, int tag)
 		goto out;
 	}
 
-	err = ufshcd_clear_cmd(hba, tag);
+	err = ufshcd_clear_cmds(hba, 1U << tag);
 	if (err)
 		dev_err(hba->dev, "%s: Failed clearing cmd at tag %d, err %d\n",
 			__func__, tag, err);

[PATCH v2 3/3] scsi: ufs: Fix a race between the interrupt handler and the reset handler

[Bart Van Assche]

Prevent that both the interrupt handler and the reset handler try to
complete a request at the same time. This patch is the result of an
analysis of the following crash:

Unable to handle kernel NULL pointer dereference at virtual address 0000000000000120
CPU: 0 PID: 0 Comm: swapper/0 Tainted: G           OE     5.10.107-android13-4-00051-g1e48e8970cca-ab8664745 #1
pc : ufshcd_release_scsi_cmd+0x30/0x46c
lr : __ufshcd_transfer_req_compl+0x4fc/0x9c0
Call trace:
 ufshcd_release_scsi_cmd+0x30/0x46c
 __ufshcd_transfer_req_compl+0x4fc/0x9c0
 ufshcd_poll+0xf0/0x208
 ufshcd_sl_intr+0xb8/0xf0
 ufshcd_intr+0x168/0x2f4
 __handle_irq_event_percpu+0xa0/0x30c
 handle_irq_event+0x84/0x178
 handle_fasteoi_irq+0x150/0x2e8
 __handle_domain_irq+0x114/0x1e4
 gic_handle_irq.31846+0x58/0x300
 el1_irq+0xe4/0x1c0
 cpuidle_enter_state+0x3ac/0x8c4
 do_idle+0x2fc/0x55c
 cpu_startup_entry+0x84/0x90
 kernel_init+0x0/0x310
 start_kernel+0x0/0x608
 start_kernel+0x4ec/0x608

Signed-off-by: Bart Van Assche <bvanassche@acm.org>
---
 drivers/ufs/core/ufshcd.c | 28 +++++++++++++++++++---------
 1 file changed, 19 insertions(+), 9 deletions(-)

diff --git a/drivers/ufs/core/ufshcd.c b/drivers/ufs/core/ufshcd.c
index b7120c5f0920..ff3abda91bb6 100644
--- a/drivers/ufs/core/ufshcd.c
+++ b/drivers/ufs/core/ufshcd.c
@@ -6975,14 +6975,14 @@ int ufshcd_exec_raw_upiu_cmd(struct ufs_hba *hba,
 }
 
 /**
- * ufshcd_eh_device_reset_handler - device reset handler registered to
- *                                    scsi layer.
+ * ufshcd_eh_device_reset_handler() - Reset a single logical unit.
  * @cmd: SCSI command pointer
  *
  * Returns SUCCESS/FAILED
  */
 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;
 	u32 pos;
@@ -7001,14 +7001,24 @@ static int ufshcd_eh_device_reset_handler(struct scsi_cmnd *cmd)
 	}
 
 	/* clear the commands that were pending for corresponding LUN */
-	for_each_set_bit(pos, &hba->outstanding_reqs, hba->nutrs) {
-		if (hba->lrb[pos].lun == lun) {
-			err = ufshcd_clear_cmds(hba, 1U << pos);
-			if (err)
-				break;
-			__ufshcd_transfer_req_compl(hba, 1U << pos);
-		}
+	spin_lock_irqsave(&hba->outstanding_lock, flags);
+	for_each_set_bit(pos, &hba->outstanding_reqs, hba->nutrs)
+		if (hba->lrb[pos].lun == lun)
+			__set_bit(pos, &pending_reqs);
+	hba->outstanding_reqs &= ~pending_reqs;
+	spin_unlock_irqrestore(&hba->outstanding_lock, flags);
+
+	if (ufshcd_clear_cmds(hba, pending_reqs) < 0) {
+		spin_lock_irqsave(&hba->outstanding_lock, flags);
+		not_cleared = pending_reqs &
+			ufshcd_readl(hba, REG_UTP_TRANSFER_REQ_DOOR_BELL);
+		hba->outstanding_reqs |= not_cleared;
+		spin_unlock_irqrestore(&hba->outstanding_lock, flags);
+
+		dev_err(hba->dev, "%s: failed to clear requests %#lx\n",
+			__func__, not_cleared);
 	}
+	__ufshcd_transfer_req_compl(hba, pending_reqs & ~not_cleared);
 
 out:
 	hba->req_abort_count = 0;

Re: [PATCH v2 0/3] ufs: Fix a race between the interrupt handler and the reset handler

[Stanley Chu]

Hi Bart,

I am not sure if this series could solve the racing you met.
However this series looks good to me.

Reviewed-by: Stanley Chu <stanley.chu@mediatek.com>

On Mon, 2022-06-13 at 14:44 -0700, Bart Van Assche wrote:
> Hi Martin,
> 
> This patch series is version two of a fix between the UFS interrupt
> handler and
> reset handlers. Please consider this patch series for kernel v5.20.
> 
> Changes compared to v1:
> - Converted a single patch into three patches.
> - Modified patch 3/3 such that only cleared requests are completed.
> 
> Bart Van Assche (3):
>   scsi: ufs: Simplify ufshcd_clear_cmd()
>   scsi: ufs: Support clearing multiple commands at once
>   scsi: ufs: Fix a race between the interrupt handler and the reset
>     handler
> 
>  drivers/ufs/core/ufshcd.c | 76 ++++++++++++++++++++++++-------------
> --
>  1 file changed, 48 insertions(+), 28 deletions(-)
> 

Re: [PATCH v2 0/3] ufs: Fix a race between the interrupt handler and the reset handler

[Adrian Hunter]

On 14/06/22 00:44, Bart Van Assche wrote:
> Hi Martin,
> 
> This patch series is version two of a fix between the UFS interrupt handler and
> reset handlers. Please consider this patch series for kernel v5.20.
> 
> Changes compared to v1:
> - Converted a single patch into three patches.
> - Modified patch 3/3 such that only cleared requests are completed.
> 
> Bart Van Assche (3):
>   scsi: ufs: Simplify ufshcd_clear_cmd()
>   scsi: ufs: Support clearing multiple commands at once
>   scsi: ufs: Fix a race between the interrupt handler and the reset
>     handler
> 
>  drivers/ufs/core/ufshcd.c | 76 ++++++++++++++++++++++++---------------
>  1 file changed, 48 insertions(+), 28 deletions(-)
> 

Reviewed-by: Adrian Hunter <adrian.hunter@intel.com>

Re: [PATCH v2 0/3] ufs: Fix a race between the interrupt handler and the reset handler

[Martin K. Petersen]

On Mon, 13 Jun 2022 14:44:39 -0700, Bart Van Assche wrote:

> This patch series is version two of a fix between the UFS interrupt handler and
> reset handlers. Please consider this patch series for kernel v5.20.
> 
> Changes compared to v1:
> - Converted a single patch into three patches.
> - Modified patch 3/3 such that only cleared requests are completed.
> 
> [...]

Applied to 5.19/scsi-fixes, thanks!

[1/3] scsi: ufs: Simplify ufshcd_clear_cmd()
      https://git.kernel.org/mkp/scsi/c/da8badd7d358
[2/3] scsi: ufs: Support clearing multiple commands at once
      https://git.kernel.org/mkp/scsi/c/d1a7644648b7
[3/3] scsi: ufs: Fix a race between the interrupt handler and the reset handler
      https://git.kernel.org/mkp/scsi/c/2acd76e7b859

-- 
Martin K. Petersen	Oracle Linux Engineering