Re: [PATCH] x86: limit irq affinity

[Dimitri Sivanich <sivanich@sgi.com>]

On Sun, Sep 27, 2009 at 06:19:09PM -0700, Yinghai Lu wrote:
> On Sun, Sep 27, 2009 at 6:02 PM, Dimitri Sivanich <sivanich@sgi.com> wrote:
> > This patch allows for restrictions to irq affinity via a new cpumask and
> > device node value in the irq_cfg structure.  The node value can then be
> > used by specific x86 architectures to determine the cpumask for the
> > desired cpu irq affinity domain.
> >
> > The mask forces IRQ affinity to remain within the specified cpu domain.
> > On some UV systems, this domain will be limited to the nodes accessible
> > to the given node.  Currently other X86 systems will have all bits in
> > the cpumask set, so non-UV systems will remain unaffected at this time.
> >
> > Signed-off-by: Dimitri Sivanich <sivanich@sgi.com>
> >
> > ---
> >
> >  arch/x86/Kconfig                   |    1
> >  arch/x86/include/asm/uv/uv_irq.h   |    2
> >  arch/x86/include/asm/uv/uv_mmrs.h  |   25 ++++
> >  arch/x86/kernel/apic/io_apic.c     |  166 +++++++++++++++++++++++++++------
> >  arch/x86/kernel/apic/x2apic_uv_x.c |    2
> >  arch/x86/kernel/uv_irq.c           |   68 +++++++++++++
> >  6 files changed, 235 insertions(+), 29 deletions(-)
> >
> > Index: linux/arch/x86/kernel/apic/io_apic.c
> > ===================================================================
> > --- linux.orig/arch/x86/kernel/apic/io_apic.c   2009-09-26 15:28:04.000000000 -0500
> > +++ linux/arch/x86/kernel/apic/io_apic.c        2009-09-26 16:20:04.000000000 -0500
> > @@ -62,6 +62,7 @@
> >  #include <asm/hw_irq.h>
> >  #include <asm/uv/uv_hub.h>
> >  #include <asm/uv/uv_irq.h>
> > +#include <asm/uv/uv.h>
> >
> >  #include <asm/apic.h>
> >
> > @@ -149,6 +150,8 @@ struct irq_cfg {
> >        struct irq_pin_list *irq_2_pin;
> >        cpumask_var_t domain;
> >        cpumask_var_t old_domain;
> > +       cpumask_var_t allowed;
> > +       int node;
> >        unsigned move_cleanup_count;
> >        u8 vector;
> >        u8 move_in_progress : 1;
> > @@ -184,6 +187,18 @@ void __init io_apic_disable_legacy(void)
> >        nr_irqs_gsi = 0;
> >  }
> >
> 
> original irq affinity in irq_desc does not work?
>

The allowed cpumask allows us to store the complete set of possible cpus allowed for the life of the irq, without having to recalculate that everytime IRQ affinity changes.   The irq_desc affinity is a dynamically changing value.