From: Long Li <[email protected]>
storvsc doesn't use a dedicated hardware queue for a given CPU queue. When
issuing I/O, it selects returning CPU (hardware queue) dynamically based on
vmbus channel usage across all channels.
This patch sets up a 1:1 mapping between hardware queue and CPU queue, thus
avoiding unnecessary locking at upper layer when issuing I/O.
Signed-off-by: Long Li <[email protected]>
---
drivers/scsi/storvsc_drv.c | 16 ++++++++++++++--
1 file changed, 14 insertions(+), 2 deletions(-)
diff --git a/drivers/scsi/storvsc_drv.c b/drivers/scsi/storvsc_drv.c
index b89269120a2d..26c16d40ec46 100644
--- a/drivers/scsi/storvsc_drv.c
+++ b/drivers/scsi/storvsc_drv.c
@@ -1682,6 +1682,18 @@ static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd)
return 0;
}
+static int storvsc_map_queues(struct Scsi_Host *shost)
+{
+ unsigned int cpu;
+ struct blk_mq_queue_map *qmap = &shost->tag_set.map[HCTX_TYPE_DEFAULT];
+
+ for_each_possible_cpu(cpu) {
+ qmap->mq_map[cpu] = cpu;
+ }
+
+ return 0;
+}
+
static struct scsi_host_template scsi_driver = {
.module = THIS_MODULE,
.name = "storvsc_host_t",
@@ -1697,6 +1709,7 @@ static struct scsi_host_template scsi_driver = {
.this_id = -1,
/* Make sure we dont get a sg segment crosses a page boundary */
.dma_boundary = PAGE_SIZE-1,
+ .map_queues = storvsc_map_queues,
.no_write_same = 1,
.track_queue_depth = 1,
};
@@ -1836,8 +1849,7 @@ static int storvsc_probe(struct hv_device *device,
/*
* Set the number of HW queues we are supporting.
*/
- if (stor_device->num_sc != 0)
- host->nr_hw_queues = stor_device->num_sc + 1;
+ host->nr_hw_queues = num_possible_cpus();
/*
* Set the error handler work queue.
--
2.17.1
On Tue, Aug 20, 2019 at 3:36 AM <[email protected]> wrote:
>
> From: Long Li <[email protected]>
>
> storvsc doesn't use a dedicated hardware queue for a given CPU queue. When
> issuing I/O, it selects returning CPU (hardware queue) dynamically based on
> vmbus channel usage across all channels.
>
> This patch sets up a 1:1 mapping between hardware queue and CPU queue, thus
> avoiding unnecessary locking at upper layer when issuing I/O.
>
> Signed-off-by: Long Li <[email protected]>
> ---
> drivers/scsi/storvsc_drv.c | 16 ++++++++++++++--
> 1 file changed, 14 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/scsi/storvsc_drv.c b/drivers/scsi/storvsc_drv.c
> index b89269120a2d..26c16d40ec46 100644
> --- a/drivers/scsi/storvsc_drv.c
> +++ b/drivers/scsi/storvsc_drv.c
> @@ -1682,6 +1682,18 @@ static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd)
> return 0;
> }
>
> +static int storvsc_map_queues(struct Scsi_Host *shost)
> +{
> + unsigned int cpu;
> + struct blk_mq_queue_map *qmap = &shost->tag_set.map[HCTX_TYPE_DEFAULT];
> +
> + for_each_possible_cpu(cpu) {
> + qmap->mq_map[cpu] = cpu;
> + }
Block layer provides the helper of blk_mq_map_queues(), so suggest you to use
the default cpu mapping, instead of inventing a new one.
thanks,
Ming Lei
>>>Subject: Re: [PATCH] storvsc: setup 1:1 mapping between hardware queue
>>>and CPU queue
>>>
>>>On Tue, Aug 20, 2019 at 3:36 AM <[email protected]> wrote:
>>>>
>>>> From: Long Li <[email protected]>
>>>>
>>>> storvsc doesn't use a dedicated hardware queue for a given CPU queue.
>>>> When issuing I/O, it selects returning CPU (hardware queue)
>>>> dynamically based on vmbus channel usage across all channels.
>>>>
>>>> This patch sets up a 1:1 mapping between hardware queue and CPU
>>>queue,
>>>> thus avoiding unnecessary locking at upper layer when issuing I/O.
>>>>
>>>> Signed-off-by: Long Li <[email protected]>
>>>> ---
>>>> drivers/scsi/storvsc_drv.c | 16 ++++++++++++++--
>>>> 1 file changed, 14 insertions(+), 2 deletions(-)
>>>>
>>>> diff --git a/drivers/scsi/storvsc_drv.c b/drivers/scsi/storvsc_drv.c
>>>> index b89269120a2d..26c16d40ec46 100644
>>>> --- a/drivers/scsi/storvsc_drv.c
>>>> +++ b/drivers/scsi/storvsc_drv.c
>>>> @@ -1682,6 +1682,18 @@ static int storvsc_queuecommand(struct
>>>Scsi_Host *host, struct scsi_cmnd *scmnd)
>>>> return 0;
>>>> }
>>>>
>>>> +static int storvsc_map_queues(struct Scsi_Host *shost) {
>>>> + unsigned int cpu;
>>>> + struct blk_mq_queue_map *qmap =
>>>> +&shost->tag_set.map[HCTX_TYPE_DEFAULT];
>>>> +
>>>> + for_each_possible_cpu(cpu) {
>>>> + qmap->mq_map[cpu] = cpu;
>>>> + }
>>>
>>>Block layer provides the helper of blk_mq_map_queues(), so suggest you
>>>to use the default cpu mapping, instead of inventing a new one.
Thanks for the pointer. I'm sending a v2.
Long
>>>
>>>thanks,
>>>Ming Lei