public inbox for linux-scsi@vger.kernel.org
 help / color / mirror / Atom feed
From: Douglas Gilbert <dgilbert@interlog.com>
To: John Garry <john.garry@huawei.com>,
	jejb@linux.vnet.ibm.com, martin.petersen@oracle.com
Cc: linux-scsi@vger.kernel.org, hare@suse.com, ming.lei@redhat.com,
	kashyap.desai@broadcom.com
Subject: Re: [PATCH v2 2/2] scsi: scsi_debug: Support hostwide tags
Date: Thu, 9 Jul 2020 11:25:43 -0400	[thread overview]
Message-ID: <0eebd28b-ca5e-6ddc-1a16-6097cb1b5a08@interlog.com> (raw)
In-Reply-To: <1594297400-24756-3-git-send-email-john.garry@huawei.com>

On 2020-07-09 8:23 a.m., John Garry wrote:
> Many SCSI HBAs support a hostwide tagset, whereby each command submitted
> to the HW from all submission queues must have a unique tag identifier.
> 
> Normally this unique tag will be in the range [0, max queue), where "max
> queue" is the depth of each of the submission queues.
> 
> Add support for this hostwide tag feature, via module parameter
> "host_max_queue". A non-zero value means that the feature is enabled. In
> this case, the submission queues are not exposed to upper layer, i.e. from
> blk-mq prespective, the device has a single hw queue. There are 2 reasons
> for this:
> a. it is assumed that the host can support nr_hw_queues * can_queue
>     commands, but this is not true for hostwide tags
> b. for nr_hw_queues != 0, the request tag is not unique over all HW queues,
>     and some HBA drivers want to use this tag for the hostwide tag
> 
> However, like many SCSI HBA drivers today - megaraid sas being an example -
> the full set of HW submission queues are still used in the LLDD driver. So
> instead of using a complicated "reply_map" to create a per-CPU submission
> queue mapping like megaraid sas (as it depends on a PCI device + MSIs) -
> use a simple algorithm:
> 
> hwq = cpu % queue count
> 
> If the host max queue param is set non-zero, then the max queue depth is
> fixed at this value also.
> 
> If and when hostwide shared tags are supported in blk-mq/scsi mid-layer,
> then the policy to set nr_hw_queues = 0 for hostwide tags can be revised.
> 
> Signed-off-by: John Garry <john.garry@huawei.com>

Acked-by: Douglas Gilbert <dgilbert@interlog.com>

I have updated this page to reflect this addition:
      http://sg.danny.cz/sg/scsi_debug.html

> ---
>   drivers/scsi/scsi_debug.c | 80 +++++++++++++++++++++++++++++++++------
>   1 file changed, 68 insertions(+), 12 deletions(-)
> 
> diff --git a/drivers/scsi/scsi_debug.c b/drivers/scsi/scsi_debug.c
> index 68534a23866e..2246d721ec88 100644
> --- a/drivers/scsi/scsi_debug.c
> +++ b/drivers/scsi/scsi_debug.c
> @@ -344,6 +344,7 @@ struct sdebug_defer {
>   	struct execute_work ew;
>   	int sqa_idx;	/* index of sdebug_queue array */
>   	int qc_idx;	/* index of sdebug_queued_cmd array within sqa_idx */
> +	int hc_idx;	/* hostwide tag index */
>   	int issuing_cpu;
>   	bool init_hrt;
>   	bool init_wq;
> @@ -759,6 +760,7 @@ static int sdebug_dsense = DEF_D_SENSE;
>   static int sdebug_every_nth = DEF_EVERY_NTH;
>   static int sdebug_fake_rw = DEF_FAKE_RW;
>   static unsigned int sdebug_guard = DEF_GUARD;
> +static int sdebug_host_max_queue;	/* per host */
>   static int sdebug_lowest_aligned = DEF_LOWEST_ALIGNED;
>   static int sdebug_max_luns = DEF_MAX_LUNS;
>   static int sdebug_max_queue = SDEBUG_CANQUEUE;	/* per submit queue */
> @@ -4707,15 +4709,28 @@ static int resp_rwp_zone(struct scsi_cmnd *scp, struct sdebug_dev_info *devip)
>   
>   static struct sdebug_queue *get_queue(struct scsi_cmnd *cmnd)
>   {
> -	u32 tag = blk_mq_unique_tag(cmnd->request);
> -	u16 hwq = blk_mq_unique_tag_to_hwq(tag);
> +	u16 hwq;
>   
> -	pr_debug("tag=%#x, hwq=%d\n", tag, hwq);
> -	if (WARN_ON_ONCE(hwq >= submit_queues))
> -		hwq = 0;
> +	if (sdebug_host_max_queue) {
> +		/* Provide a simple method to choose the hwq */
> +		hwq = smp_processor_id() % submit_queues;
> +	} else {
> +		u32 tag = blk_mq_unique_tag(cmnd->request);
> +
> +		hwq = blk_mq_unique_tag_to_hwq(tag);
> +
> +		pr_debug("tag=%#x, hwq=%d\n", tag, hwq);
> +		if (WARN_ON_ONCE(hwq >= submit_queues))
> +			hwq = 0;
> +	}
>   	return sdebug_q_arr + hwq;
>   }
>   
> +static u32 get_tag(struct scsi_cmnd *cmnd)
> +{
> +	return blk_mq_unique_tag(cmnd->request);
> +}
> +
>   /* Queued (deferred) command completions converge here. */
>   static void sdebug_q_cmd_complete(struct sdebug_defer *sd_dp)
>   {
> @@ -4747,8 +4762,8 @@ static void sdebug_q_cmd_complete(struct sdebug_defer *sd_dp)
>   	scp = sqcp->a_cmnd;
>   	if (unlikely(scp == NULL)) {
>   		spin_unlock_irqrestore(&sqp->qc_lock, iflags);
> -		pr_err("scp is NULL, sqa_idx=%d, qc_idx=%d\n",
> -		       sd_dp->sqa_idx, qc_idx);
> +		pr_err("scp is NULL, sqa_idx=%d, qc_idx=%d, hc_idx=%d\n",
> +		       sd_dp->sqa_idx, qc_idx, sd_dp->hc_idx);
>   		return;
>   	}
>   	devip = (struct sdebug_dev_info *)scp->device->hostdata;
> @@ -5451,6 +5466,10 @@ static int schedule_resp(struct scsi_cmnd *cmnd, struct sdebug_dev_info *devip,
>   		new_sd_dp = false;
>   	}
>   
> +	/* Set the hostwide tag */
> +	if (sdebug_host_max_queue)
> +		sd_dp->hc_idx = get_tag(cmnd);
> +
>   	if (ndelay > 0 && ndelay < INCLUSIVE_TIMING_MAX_NS)
>   		ns_from_boot = ktime_get_boottime_ns();
>   
> @@ -5572,6 +5591,7 @@ module_param_named(every_nth, sdebug_every_nth, int, S_IRUGO | S_IWUSR);
>   module_param_named(fake_rw, sdebug_fake_rw, int, S_IRUGO | S_IWUSR);
>   module_param_named(guard, sdebug_guard, uint, S_IRUGO);
>   module_param_named(host_lock, sdebug_host_lock, bool, S_IRUGO | S_IWUSR);
> +module_param_named(host_max_queue, sdebug_host_max_queue, int, S_IRUGO);
>   module_param_string(inq_product, sdebug_inq_product_id,
>   		    sizeof(sdebug_inq_product_id), S_IRUGO | S_IWUSR);
>   module_param_string(inq_rev, sdebug_inq_product_rev,
> @@ -5642,6 +5662,8 @@ MODULE_PARM_DESC(every_nth, "timeout every nth command(def=0)");
>   MODULE_PARM_DESC(fake_rw, "fake reads/writes instead of copying (def=0)");
>   MODULE_PARM_DESC(guard, "protection checksum: 0=crc, 1=ip (def=0)");
>   MODULE_PARM_DESC(host_lock, "host_lock is ignored (def=0)");
> +MODULE_PARM_DESC(host_max_queue,
> +		 "host max # of queued cmds (0 to max(def) [max_queue fixed equal for !0])");
>   MODULE_PARM_DESC(inq_product, "SCSI INQUIRY product string (def=\"scsi_debug\")");
>   MODULE_PARM_DESC(inq_rev, "SCSI INQUIRY revision string (def=\""
>   		 SDEBUG_VERSION "\")");
> @@ -6141,7 +6163,8 @@ static ssize_t max_queue_store(struct device_driver *ddp, const char *buf,
>   	struct sdebug_queue *sqp;
>   
>   	if ((count > 0) && (1 == sscanf(buf, "%d", &n)) && (n > 0) &&
> -	    (n <= SDEBUG_CANQUEUE)) {
> +	    (n <= SDEBUG_CANQUEUE) &&
> +	    (sdebug_host_max_queue == 0)) {
>   		block_unblock_all_queues(true);
>   		k = 0;
>   		for (j = 0, sqp = sdebug_q_arr; j < submit_queues;
> @@ -6164,6 +6187,17 @@ static ssize_t max_queue_store(struct device_driver *ddp, const char *buf,
>   }
>   static DRIVER_ATTR_RW(max_queue);
>   
> +static ssize_t host_max_queue_show(struct device_driver *ddp, char *buf)
> +{
> +	return scnprintf(buf, PAGE_SIZE, "%d\n", sdebug_host_max_queue);
> +}
> +
> +/*
> + * Since this is used for .can_queue, and we get the hc_idx tag from the bitmap
> + * in range [0, sdebug_host_max_queue), we can't change it.
> + */
> +static DRIVER_ATTR_RO(host_max_queue);
> +
>   static ssize_t no_uld_show(struct device_driver *ddp, char *buf)
>   {
>   	return scnprintf(buf, PAGE_SIZE, "%d\n", sdebug_no_uld);
> @@ -6503,6 +6537,7 @@ static struct attribute *sdebug_drv_attrs[] = {
>   	&driver_attr_ptype.attr,
>   	&driver_attr_dsense.attr,
>   	&driver_attr_fake_rw.attr,
> +	&driver_attr_host_max_queue.attr,
>   	&driver_attr_no_lun_0.attr,
>   	&driver_attr_num_tgts.attr,
>   	&driver_attr_dev_size_mb.attr,
> @@ -6619,6 +6654,20 @@ static int __init scsi_debug_init(void)
>   		return -EINVAL;
>   	}
>   
> +	if ((sdebug_host_max_queue > SDEBUG_CANQUEUE) ||
> +	    (sdebug_host_max_queue < 0)) {
> +		pr_err("host_max_queue must be in range [0 %d]\n",
> +		       SDEBUG_CANQUEUE);
> +		return -EINVAL;
> +	}
> +
> +	if (sdebug_host_max_queue &&
> +	    (sdebug_max_queue != sdebug_host_max_queue)) {
> +		sdebug_max_queue = sdebug_host_max_queue;
> +		pr_warn("fixing max submit queue depth to host max queue depth, %d\n",
> +			sdebug_max_queue);
> +	}
> +
>   	sdebug_q_arr = kcalloc(submit_queues, sizeof(struct sdebug_queue),
>   			       GFP_KERNEL);
>   	if (sdebug_q_arr == NULL)
> @@ -7257,7 +7306,10 @@ static int sdebug_driver_probe(struct device *dev)
>   
>   	sdbg_host = to_sdebug_host(dev);
>   
> -	sdebug_driver_template.can_queue = sdebug_max_queue;
> +	if (sdebug_host_max_queue)
> +		sdebug_driver_template.can_queue = sdebug_host_max_queue;
> +	else
> +		sdebug_driver_template.can_queue = sdebug_max_queue;
>   	if (!sdebug_clustering)
>   		sdebug_driver_template.dma_boundary = PAGE_SIZE - 1;
>   
> @@ -7272,9 +7324,13 @@ static int sdebug_driver_probe(struct device *dev)
>   			my_name, submit_queues, nr_cpu_ids);
>   		submit_queues = nr_cpu_ids;
>   	}
> -	/* Decide whether to tell scsi subsystem that we want mq */
> -	/* Following should give the same answer for each host */
> -	hpnt->nr_hw_queues = submit_queues;
> +	/*
> +	 * Decide whether to tell scsi subsystem that we want mq. The
> +	 * following should give the same answer for each host. If the host
> +	 * has a limit of hostwide max commands, then do not set.
> +	 */
> +	if (!sdebug_host_max_queue)
> +		hpnt->nr_hw_queues = submit_queues;
>   
>   	sdbg_host->shost = hpnt;
>   	*((struct sdebug_host_info **)hpnt->hostdata) = sdbg_host;
> 


  reply	other threads:[~2020-07-09 15:25 UTC|newest]

Thread overview: 7+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-07-09 12:23 [PATCH v2 0/2] scsi: scsi_debug: Support hostwide tags and a fix John Garry
2020-07-09 12:23 ` [PATCH v2 1/2] scsi: scsi_debug: Add check for sdebug_max_queue during module init John Garry
2020-07-09 13:18   ` Ming Lei
2020-07-09 12:23 ` [PATCH v2 2/2] scsi: scsi_debug: Support hostwide tags John Garry
2020-07-09 15:25   ` Douglas Gilbert [this message]
2020-07-09 15:42     ` John Garry
2020-07-14  4:58 ` [PATCH v2 0/2] scsi: scsi_debug: Support hostwide tags and a fix Martin K. Petersen

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=0eebd28b-ca5e-6ddc-1a16-6097cb1b5a08@interlog.com \
    --to=dgilbert@interlog.com \
    --cc=hare@suse.com \
    --cc=jejb@linux.vnet.ibm.com \
    --cc=john.garry@huawei.com \
    --cc=kashyap.desai@broadcom.com \
    --cc=linux-scsi@vger.kernel.org \
    --cc=martin.petersen@oracle.com \
    --cc=ming.lei@redhat.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox