[PATCH 2/6] scsi-mq: add operations about device_busy/target_busy/host_busy in scsi_mq_queue_rq().

[PATCH 2/6] scsi-mq: add operations about device_busy/target_busy/host_busy in scsi_mq_queue_rq().

[majianpeng]

To make the scsi-mq work, add those operations in scsi_mq_queue_rq().

Signed-off-by: Jianpeng Ma <majianpeng@gmail.com>
---
 drivers/scsi/scsi-mq.c |    8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/drivers/scsi/scsi-mq.c b/drivers/scsi/scsi-mq.c
index 918b443..eccdbd5 100644
--- a/drivers/scsi/scsi-mq.c
+++ b/drivers/scsi/scsi-mq.c
@@ -20,8 +20,10 @@ static int scsi_mq_queue_rq(struct blk_mq_hw_ctx *hctx, struct request *rq)
 	struct scsi_device *sdev = q->queuedata;
 	struct scsi_cmnd *sc = rq->special;
 	struct scatterlist *sg = sc->mq_sgl;
+	struct Scsi_Host *shost = sdev->host;
 	unsigned char *sense_buf = sc->sense_buffer;
 	int rc;
+
 	/*
 	 * With blk-mq rq->special pre-allocation, scsi_get_cmd_from_req()
 	 * is using the hardware context *sc descriptor
@@ -66,6 +68,12 @@ static int scsi_mq_queue_rq(struct blk_mq_hw_ctx *hctx, struct request *rq)
 	if (sdev->host->hostt->cmd_size)
 		sc->SCp.ptr = blk_mq_rq_to_pdu(rq) + sizeof(struct scsi_cmnd);
 
+	spin_lock_irq(shost->host_lock);
+	sdev->device_busy++;
+	scsi_target(sdev)->target_busy++;
+	shost->host_busy++;
+	spin_unlock_irq(shost->host_lock);
+
 	rc = scsi_dispatch_cmd(sc);
 	switch (rc) {
 	case 0:
-- 
1.7.10.4

Re: [PATCH 2/6] scsi-mq: add operations about device_busy/target_busy/host_busy in scsi_mq_queue_rq().

[Nicholas A. Bellinger]

On Fri, 2013-11-29 at 16:41 +0800, majianpeng wrote:
> To make the scsi-mq work, add those operations in scsi_mq_queue_rq().
> 
> Signed-off-by: Jianpeng Ma <majianpeng@gmail.com>
> ---
>  drivers/scsi/scsi-mq.c |    8 ++++++++
>  1 file changed, 8 insertions(+)
> 
> diff --git a/drivers/scsi/scsi-mq.c b/drivers/scsi/scsi-mq.c
> index 918b443..eccdbd5 100644
> --- a/drivers/scsi/scsi-mq.c
> +++ b/drivers/scsi/scsi-mq.c
> @@ -20,8 +20,10 @@ static int scsi_mq_queue_rq(struct blk_mq_hw_ctx *hctx, struct request *rq)
>  	struct scsi_device *sdev = q->queuedata;
>  	struct scsi_cmnd *sc = rq->special;
>  	struct scatterlist *sg = sc->mq_sgl;
> +	struct Scsi_Host *shost = sdev->host;
>  	unsigned char *sense_buf = sc->sense_buffer;
>  	int rc;
> +
>  	/*
>  	 * With blk-mq rq->special pre-allocation, scsi_get_cmd_from_req()
>  	 * is using the hardware context *sc descriptor
> @@ -66,6 +68,12 @@ static int scsi_mq_queue_rq(struct blk_mq_hw_ctx *hctx, struct request *rq)
>  	if (sdev->host->hostt->cmd_size)
>  		sc->SCp.ptr = blk_mq_rq_to_pdu(rq) + sizeof(struct scsi_cmnd);
>  
> +	spin_lock_irq(shost->host_lock);
> +	sdev->device_busy++;
> +	scsi_target(sdev)->target_busy++;
> +	shost->host_busy++;
> +	spin_unlock_irq(shost->host_lock);
> +
>  	rc = scsi_dispatch_cmd(sc);
>  	switch (rc) {
>  	case 0:

Same with this patch, it doesn't seem to be strictly required for the
QEMU AHCI + TYPE_ROM test case to function with v3.13-rc3 code.

I've been avoiding legacy host_lock access in the fast-path thus far,
and would not like to give that away just yet unless it's absolutely
required for prototype scsi-mq code to function.

--nab