[v3,0/7] 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.

Bao D. Nguyen (7):
  ufs: core: Combine 32-bit command_desc_base_addr_lo/hi
  ufs: core: Update the ufshcd_clear_cmds() functionality
  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     | 236 ++++++++++++++++++++++++++++++++++++++++-
 drivers/ufs/core/ufshcd-priv.h |  17 ++-
 drivers/ufs/core/ufshcd.c      | 200 +++++++++++++++++++++++++++-------
 drivers/ufs/host/ufs-qcom.c    |   2 +-
 include/ufs/ufshcd.h           |   5 +-
 include/ufs/ufshci.h           |  23 +++-
 6 files changed, 431 insertions(+), 52 deletions(-)

---
v2->v3:
patch #1:
  New patch per Bart's comment. Helps process utp cmd
  descriptor addr easier.
patch #2:
  New patch to address Bart's comment regarding potentialoverflow 
  when mcq queue depth becomes greater than 64.
patch #3:
  Address Bart's comments
  . Replaced ufshcd_mcq_poll_register() with read_poll_timeout()
  . Replace spin_lock(sq_lock) with mutex(sq_mutex)
  . Updated ufshcd_mcq_nullify_cmd() and renamed to ufshcd_mcq_nullify_sqe()
  . Minor cosmetic changes
patch #4:
  Adress Bart's comments. Added new function ufshcd_cmd_inflight()
  to replace the usage of lrbp->cmd
patch #5:
  Continue replacing lrbp->cmd with ufshcd_cmd_inflight()
patch #6:
  No change
patch #7:
  Address Stanley Chu's comment about clearing hba->dev_cmd.complete
  in clear ufshcd_wait_for_dev_cmd()
  Address Bart's comment.
---
v1->v2:
patch #1: Addressed Powen's comment. Replaced read_poll_timeout()
with ufshcd_mcq_poll_register(). The function read_poll_timeout()
may sleep while the caller is holding a spin_lock(). Poll the registers
in a tight loop instead. 
---
2.7.4

On 5/9/23 22:24, Bao D. Nguyen wrote:
> The utp command descriptor base address is a 57-bit field in the utp
> utp transfer request descriptor. Combine the two 32-bit
> command_desc_base_addr_lo/hi fields into a 64-bit for better handling
> of this field.

The "utp utp" in the patch description probably should be changed into 
"UTP"? Otherwise this patch looks fine to me. Hence:

Reviewed-by: Bart Van Assche <bvanassche@acm.org>

On 5/9/23 22:24, Bao D. Nguyen wrote:
> +/**
> + * 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, id, 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;
> +	}
> +
> +	id = hwq->id;
> +
> +	mutex_lock(&hwq->sq_mutex);
> +
> +	/* 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, id);
> +	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__, id, lrbp->lun, task_tag);
> +
> +	*result = FIELD_GET(SQ_ICU_ERR_CODE_MASK, readl(reg));
> +
> +	if (ufshcd_mcq_sq_start(hba, hwq))
> +		err = FAILED;
> +
> +unlock:
> +	mutex_unlock(&hwq->sq_mutex);
> +	return err;
> +}

Please do not use the FAILED / SUCCESS return values in this function. 
These values should only be returned by functions related to SCSI error 
handling. Please do the following:
* Return a negative Unix error code in case of failure.
* Return zero upon success.
* Return the FIELD_GET() result as a positive value.
* Remove the 'int *result' argument.

> +	/* prevent concurrent access to sq hw */
> +	struct mutex sq_mutex;

Hmm ... a submission queue (SQ) exists in host memory so I'm not sure 
the "hw" part of the above comment is correct.

Thanks,

Bart.

On 5/9/23 22:24, Bao D. Nguyen wrote:
> +bool ufshcd_cmd_inflight(struct scsi_cmnd *cmd)
> +{
> +	struct request *rq;
> +
> +	if (!cmd)
> +		return false;
> +
> +	rq = scsi_cmd_to_rq(cmd);
> +	if (!rq || !blk_mq_request_started(rq))
> +		return false;
> +
> +	return true;

The return value of scsi_cmd_to_rq() is never NULL so please remove the 
!rq test.

> @@ -7450,8 +7499,8 @@ 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);

With the above change, the doorbell register is read even in MCQ mode. 
Shouldn't reading the doorbell register be skipped in MCQ mode?

Thanks,

Bart.