[PATCH v2] scsi: ufs: Fix the polling implementation

[PATCH v2] scsi: ufs: Fix the polling implementation

[Bart Van Assche]

Fix the following issues in ufshcd_poll():
- If polling succeeds, return a positive value.
- Do not complete polling requests from interrupt context because the
  block layer expects these requests to be completed from thread
  context. From block/bio.c:

    If REQ_ALLOC_CACHE is set, the final put of the bio MUST be done
    from process context, not hard/soft IRQ.

Fixes: eaab9b573054 ("scsi: ufs: Implement polling support")
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
---

Changes compared to v1:
- Made sure that polled requests are not completed from interrupt context.

 drivers/ufs/core/ufshcd.c | 28 ++++++++++++++++++++++++++--
 1 file changed, 26 insertions(+), 2 deletions(-)

diff --git a/drivers/ufs/core/ufshcd.c b/drivers/ufs/core/ufshcd.c
index 768cb49d269c..b4bf3c3bef0c 100644
--- a/drivers/ufs/core/ufshcd.c
+++ b/drivers/ufs/core/ufshcd.c
@@ -5344,6 +5344,26 @@ static void __ufshcd_transfer_req_compl(struct ufs_hba *hba,
 	}
 }
 
+/* Any value that is not an existing queue number is fine for this constant. */
+enum {
+	UFSHCD_POLL_FROM_INTERRUPT_CONTEXT = -1
+};
+
+static void ufshcd_clear_polled(struct ufs_hba *hba,
+				unsigned long *completed_reqs)
+{
+	int tag;
+
+	for_each_set_bit(tag, completed_reqs, hba->nutrs) {
+		struct scsi_cmnd *cmd = hba->lrb[tag].cmd;
+
+		if (!cmd)
+			continue;
+		if (scsi_cmd_to_rq(cmd)->cmd_flags & REQ_POLLED)
+			__clear_bit(tag, completed_reqs);
+	}
+}
+
 /*
  * Returns > 0 if one or more commands have been completed or 0 if no
  * requests have been completed.
@@ -5360,13 +5380,17 @@ static int ufshcd_poll(struct Scsi_Host *shost, unsigned int queue_num)
 	WARN_ONCE(completed_reqs & ~hba->outstanding_reqs,
 		  "completed: %#lx; outstanding: %#lx\n", completed_reqs,
 		  hba->outstanding_reqs);
+	if (queue_num == UFSHCD_POLL_FROM_INTERRUPT_CONTEXT) {
+		/* Do not complete polled requests from interrupt context. */
+		ufshcd_clear_polled(hba, &completed_reqs);
+	}
 	hba->outstanding_reqs &= ~completed_reqs;
 	spin_unlock_irqrestore(&hba->outstanding_lock, flags);
 
 	if (completed_reqs)
 		__ufshcd_transfer_req_compl(hba, completed_reqs);
 
-	return completed_reqs;
+	return completed_reqs != 0;
 }
 
 /**
@@ -5397,7 +5421,7 @@ 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, 0);
+	ufshcd_poll(hba->host, UFSHCD_POLL_FROM_INTERRUPT_CONTEXT);
 
 	return IRQ_HANDLED;
 }

Re: [PATCH v2] scsi: ufs: Fix the polling implementation

[Adrian Hunter]

On 19/11/22 01:37, Bart Van Assche wrote:
> Fix the following issues in ufshcd_poll():
> - If polling succeeds, return a positive value.
> - Do not complete polling requests from interrupt context because the
>   block layer expects these requests to be completed from thread
>   context. From block/bio.c:
> 
>     If REQ_ALLOC_CACHE is set, the final put of the bio MUST be done
>     from process context, not hard/soft IRQ.
> 
> Fixes: eaab9b573054 ("scsi: ufs: Implement polling support")
> Signed-off-by: Bart Van Assche <bvanassche@acm.org>

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

> ---
> 
> Changes compared to v1:
> - Made sure that polled requests are not completed from interrupt context.
> 
>  drivers/ufs/core/ufshcd.c | 28 ++++++++++++++++++++++++++--
>  1 file changed, 26 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/ufs/core/ufshcd.c b/drivers/ufs/core/ufshcd.c
> index 768cb49d269c..b4bf3c3bef0c 100644
> --- a/drivers/ufs/core/ufshcd.c
> +++ b/drivers/ufs/core/ufshcd.c
> @@ -5344,6 +5344,26 @@ static void __ufshcd_transfer_req_compl(struct ufs_hba *hba,
>  	}
>  }
>  
> +/* Any value that is not an existing queue number is fine for this constant. */
> +enum {
> +	UFSHCD_POLL_FROM_INTERRUPT_CONTEXT = -1
> +};
> +
> +static void ufshcd_clear_polled(struct ufs_hba *hba,
> +				unsigned long *completed_reqs)
> +{
> +	int tag;
> +
> +	for_each_set_bit(tag, completed_reqs, hba->nutrs) {
> +		struct scsi_cmnd *cmd = hba->lrb[tag].cmd;
> +
> +		if (!cmd)
> +			continue;
> +		if (scsi_cmd_to_rq(cmd)->cmd_flags & REQ_POLLED)
> +			__clear_bit(tag, completed_reqs);
> +	}
> +}
> +
>  /*
>   * Returns > 0 if one or more commands have been completed or 0 if no
>   * requests have been completed.
> @@ -5360,13 +5380,17 @@ static int ufshcd_poll(struct Scsi_Host *shost, unsigned int queue_num)
>  	WARN_ONCE(completed_reqs & ~hba->outstanding_reqs,
>  		  "completed: %#lx; outstanding: %#lx\n", completed_reqs,
>  		  hba->outstanding_reqs);
> +	if (queue_num == UFSHCD_POLL_FROM_INTERRUPT_CONTEXT) {
> +		/* Do not complete polled requests from interrupt context. */
> +		ufshcd_clear_polled(hba, &completed_reqs);
> +	}
>  	hba->outstanding_reqs &= ~completed_reqs;
>  	spin_unlock_irqrestore(&hba->outstanding_lock, flags);
>  
>  	if (completed_reqs)
>  		__ufshcd_transfer_req_compl(hba, completed_reqs);
>  
> -	return completed_reqs;
> +	return completed_reqs != 0;
>  }
>  
>  /**
> @@ -5397,7 +5421,7 @@ 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, 0);
> +	ufshcd_poll(hba->host, UFSHCD_POLL_FROM_INTERRUPT_CONTEXT);
>  
>  	return IRQ_HANDLED;
>  }

Re: [PATCH v2] scsi: ufs: Fix the polling implementation

[Martin K. Petersen]

Bart,

> Fix the following issues in ufshcd_poll():
> - If polling succeeds, return a positive value.
> - Do not complete polling requests from interrupt context because the
>   block layer expects these requests to be completed from thread
>   context. From block/bio.c:

Applied to 6.2/scsi-staging, thanks!

-- 
Martin K. Petersen	Oracle Linux Engineering

Re: [PATCH v2] scsi: ufs: Fix the polling implementation

[Martin K. Petersen]

On Fri, 18 Nov 2022 15:37:03 -0800, Bart Van Assche wrote:

> Fix the following issues in ufshcd_poll():
> - If polling succeeds, return a positive value.
> - Do not complete polling requests from interrupt context because the
>   block layer expects these requests to be completed from thread
>   context. From block/bio.c:
> 
>     If REQ_ALLOC_CACHE is set, the final put of the bio MUST be done
>     from process context, not hard/soft IRQ.
> 
> [...]

Applied to 6.2/scsi-queue, thanks!

[1/1] scsi: ufs: Fix the polling implementation
      https://git.kernel.org/mkp/scsi/c/ee8c88cab4af

-- 
Martin K. Petersen	Oracle Linux Engineering