On Thu, Jan 10, 2019 at 5:05 AM <[email protected]> wrote: > > -----Original Message----- > Subject: Re: [PATCH] hpsa: add module parameter to disable irq affinity > > On 1/8/19 10:16 PM, [email protected] wrote: > > On 12/3/18 11:35 PM, Don Brace wrote: > >> The PCI_IRQ_AFFINITY flag prevents customers from changing the > >> smp_affinity and smp_affinity_list entries. > >> > >> - add a module parameter to allow this flag to be turned > >> off. > >> > >> - to turn off PCI_IRQ_AFFINITY: > >> flag hpsa_disable_irq_affinity=1 > >> > >> Reviewed-by: David Carroll <[email protected]> > >> Reviewed-by: Scott Teel <[email protected]> > >> Signed-off-by: Don Brace <[email protected]> > >> --- > >> drivers/scsi/hpsa.c | 13 ++++++++++--- > >> 1 file changed, 10 insertions(+), 3 deletions(-) > >> > >> diff --git a/drivers/scsi/hpsa.c b/drivers/scsi/hpsa.c index > >> c9cccf35e9d7..0aa5aa66151f 100644 > >> --- a/drivers/scsi/hpsa.c > >> +++ b/drivers/scsi/hpsa.c > >> @@ -87,6 +87,10 @@ static int hpsa_simple_mode; > >> module_param(hpsa_simple_mode, int, S_IRUGO|S_IWUSR); > >> MODULE_PARM_DESC(hpsa_simple_mode, > >> "Use 'simple mode' rather than 'performant mode'"); > >> +static bool hpsa_disable_irq_affinity; > >> +module_param(hpsa_disable_irq_affinity, bool, S_IRUGO|S_IWUSR); > >> +MODULE_PARM_DESC(hpsa_disable_irq_affinity, > >> + "Turn off managed irq affinity. Allows smp_affinity to be > >> +changed."); > >> > > > > Is there a better method? > > > Shouldn't we rather figure out what is triggering the I/O error? > Have you checked that we're not running into the issue discussed in the > thread 'scsi: hpsa: fix selection of reply queue' ? > > Cheers, > > Hannes > -- > Dr. Hannes Reinecke Teamlead Storage & Networking > [email protected] +49 911 74053 688 > SUSE LINUX GmbH, Maxfeldstr. 5, 90409 Nürnberg > GF: F. Imendörffer, J. Smithard, J. Guild, D. Upmanyu, G. Norton HRB 21284 > (AG Nürnberg) > --- > The issue is that in kernel/irq/proc.c function write_irq_affinity there is a > check to see if the > affinity can be changed. This fails with -EIO because interrupts are managed.
That is expected behavior for managed IRQ. > There are customers who want to be able to change affinity for application > performance reasons. Could you give examples in which customized affinity may improve application performance. > Note that there are other drivers which also unset the PCI_IRQ_AFFINITY flag > for similar reasons. Any SCSI drivers which unset PCI_IRQ_AFFINITY may break blk-mq. Thanks, Ming Lei
