| Message ID | 1578730634-109961-1-git-send-email-longli@linuxonhyperv.com (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
| Series | scsi: storvsc: Correctly set number of hardware queues for IDE disk | expand |
On Sat, Jan 11, 2020 at 4:17 PM <longli@linuxonhyperv.com> wrote: > > From: Long Li <longli@microsoft.com> > > Commit 0ed881027690 ("scsi: storvsc: setup 1:1 mapping between hardware queue and CPU queue") > introduced a regression for disks attached to IDE. For these disks the host VSP only offers > one VMBUS channel. Setting multiple queues can overload the VMBUS channel and result in > performance drop for high queue depth workload on system with large number of CPUs. > > Fix it by leaving the number of hardware queues to 1 (default value) for IDE > disks. > > Fixes: 0ed881027690 ("scsi: storvsc: setup 1:1 mapping between hardware queue and CPU queue") > Signed-off-by: Long Li <longli@microsoft.com> > --- > drivers/scsi/storvsc_drv.c | 4 +++- > 1 file changed, 3 insertions(+), 1 deletion(-) > > diff --git a/drivers/scsi/storvsc_drv.c b/drivers/scsi/storvsc_drv.c > index f8faf8b3d965..992b28e40374 100644 > --- a/drivers/scsi/storvsc_drv.c > +++ b/drivers/scsi/storvsc_drv.c > @@ -1842,9 +1842,11 @@ static int storvsc_probe(struct hv_device *device, > */ > host->sg_tablesize = (stor_device->max_transfer_bytes >> PAGE_SHIFT); > /* > + * For non-IDE disks, the host supports multiple channels. > * Set the number of HW queues we are supporting. > */ > - host->nr_hw_queues = num_present_cpus(); > + if (dev_id->driver_data != IDE_GUID) > + host->nr_hw_queues = num_present_cpus(); > > /* > * Set the error handler work queue. > -- > 2.20.1 > Reviewed-by: Ming Lei <ming.lei@redhat.com>
From: Long Li <longli@microsoft.com> Sent: Saturday, January 11, 2020 12:17 AM > > Commit 0ed881027690 ("scsi: storvsc: setup 1:1 mapping between hardware queue and > CPU queue") > introduced a regression for disks attached to IDE. For these disks the host VSP only offers > one VMBUS channel. Setting multiple queues can overload the VMBUS channel and result > in > performance drop for high queue depth workload on system with large number of CPUs. > > Fix it by leaving the number of hardware queues to 1 (default value) for IDE > disks. > > Fixes: 0ed881027690 ("scsi: storvsc: setup 1:1 mapping between hardware queue and CPU > queue") > Signed-off-by: Long Li <longli@microsoft.com> > --- > drivers/scsi/storvsc_drv.c | 4 +++- > 1 file changed, 3 insertions(+), 1 deletion(-) > > diff --git a/drivers/scsi/storvsc_drv.c b/drivers/scsi/storvsc_drv.c > index f8faf8b3d965..992b28e40374 100644 > --- a/drivers/scsi/storvsc_drv.c > +++ b/drivers/scsi/storvsc_drv.c > @@ -1842,9 +1842,11 @@ static int storvsc_probe(struct hv_device *device, > */ > host->sg_tablesize = (stor_device->max_transfer_bytes >> PAGE_SHIFT); > /* > + * For non-IDE disks, the host supports multiple channels. > * Set the number of HW queues we are supporting. > */ > - host->nr_hw_queues = num_present_cpus(); > + if (dev_id->driver_data != IDE_GUID) This function already has a pre-computed value of this test in the local variable "dev_is_ide". It would be more consistent to just use it. Michael > + host->nr_hw_queues = num_present_cpus(); > > /* > * Set the error handler work queue. > -- > 2.20.1
>Subject: RE: [PATCH] scsi: storvsc: Correctly set number of hardware queues for >IDE disk > >From: Long Li <longli@microsoft.com> Sent: Saturday, January 11, 2020 12:17 >AM >> >> Commit 0ed881027690 ("scsi: storvsc: setup 1:1 mapping between >> hardware queue and CPU queue") introduced a regression for disks >> attached to IDE. For these disks the host VSP only offers one VMBUS >> channel. Setting multiple queues can overload the VMBUS channel and >> result in performance drop for high queue depth workload on system >> with large number of CPUs. >> >> Fix it by leaving the number of hardware queues to 1 (default value) >> for IDE disks. >> >> Fixes: 0ed881027690 ("scsi: storvsc: setup 1:1 mapping between >> hardware queue and CPU >> queue") >> Signed-off-by: Long Li <longli@microsoft.com> >> --- >> drivers/scsi/storvsc_drv.c | 4 +++- >> 1 file changed, 3 insertions(+), 1 deletion(-) >> >> diff --git a/drivers/scsi/storvsc_drv.c b/drivers/scsi/storvsc_drv.c >> index f8faf8b3d965..992b28e40374 100644 >> --- a/drivers/scsi/storvsc_drv.c >> +++ b/drivers/scsi/storvsc_drv.c >> @@ -1842,9 +1842,11 @@ static int storvsc_probe(struct hv_device *device, >> */ >> host->sg_tablesize = (stor_device->max_transfer_bytes >> PAGE_SHIFT); >> /* >> + * For non-IDE disks, the host supports multiple channels. >> * Set the number of HW queues we are supporting. >> */ >> - host->nr_hw_queues = num_present_cpus(); >> + if (dev_id->driver_data != IDE_GUID) > >This function already has a pre-computed value of this test in >the local variable "dev_is_ide". It would be more consistent >to just use it. > >Michael I will send v2 to address this. Long > >> + host->nr_hw_queues = num_present_cpus(); >> >> /* >> * Set the error handler work queue. >> -- >> 2.20.1
diff --git a/drivers/scsi/storvsc_drv.c b/drivers/scsi/storvsc_drv.c index f8faf8b3d965..992b28e40374 100644 --- a/drivers/scsi/storvsc_drv.c +++ b/drivers/scsi/storvsc_drv.c @@ -1842,9 +1842,11 @@ static int storvsc_probe(struct hv_device *device, */ host->sg_tablesize = (stor_device->max_transfer_bytes >> PAGE_SHIFT); /* + * For non-IDE disks, the host supports multiple channels. * Set the number of HW queues we are supporting. */ - host->nr_hw_queues = num_present_cpus(); + if (dev_id->driver_data != IDE_GUID) + host->nr_hw_queues = num_present_cpus(); /* * Set the error handler work queue.