Re: [Patch 0/2] Disable SW irqbalance/irqaffinity for E7520/E7320/E7525

[Suresh Siddha <suresh.b.siddha@intel.com>]

Zwane,

As far as irq_affinity is concerned, workaround fits nicely in quirks.c
infrastructure.

x86 irqbalance is the one which is causing the pain. Workaround will be more
cleaner if we can move balanced_irq_init() to late_initcall.

If there is no objection, I can post a new patch with that change.

thanks,
suresh

On Sat, Sep 25, 2004 at 12:15:27PM +0300, Zwane Mwaikambo wrote:
> On Fri, 24 Sep 2004, Suresh Siddha wrote:
> > +#ifdef CONFIG_X86_IO_APIC
> > +#include <asm/hw_irq.h>
> > +#ifdef CONFIG_IRQBALANCE
> > +extern int irqbalance_disable(char *str);
> > +#endif
> > +extern int no_irq_affinity;
> > +extern int noirqdebug_setup(char *str);
> 
> Ok this is sort of ugly, but it's not your fault, i understand that the
> PCI quirks code is too late after IOAPIC setup, x86_64 has some early PCI
> bridge detection code which helps in doing IOAPIC quirks.
> 
> > +void __devinit quirk_intel_irqbalance(struct pci_dev *dev)
> 
> This may as well be moved elsewhere since it's not actually going to be 
> used in PCI quirks. I think you should just do the chipset detection in 
> io_apic.c and then do the disable from there, it's racy and strange 
> (although it may seem natural) to do it in quirks.c