Re: [tip:x86/apic] x86/ioapic.c: unify ioapic_retrigger_irq()

Re: [tip:x86/apic] x86/ioapic.c: unify ioapic_retrigger_irq()

[Jeremy Fitzhardinge]

On 07/18/09 05:05, tip-bot for Jeremy Fitzhardinge wrote:
> Commit-ID:  e25371d60cb06a44d7a32d7966ab9bfbeacb9390
> Gitweb:     http://git.kernel.org/tip/e25371d60cb06a44d7a32d7966ab9bfbeacb9390
> Author:     Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
> AuthorDate: Mon, 8 Jun 2009 03:49:01 -0700
> Committer:  Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
> CommitDate: Tue, 14 Jul 2009 13:32:51 -0700
>
> x86/ioapic.c: unify ioapic_retrigger_irq()
>
> The 32 and 64-bit versions of ioapic_retrigger_irq() are identical
> except the 64-bit one takes vector_lock.  vector_lock is defined and
> used on 32-bit too, so just use a common ioapic_retrigger_irq().
>   

Having another look at this patch, the other difference is that the
32-bit version just does:

    apic->send_IPI_self(irq_cfg(irq)->vector);

whereas the 64-bit does:

    apic->send_IPI_mask(cpumask_of(cpumask_first(cfg->domain)),
    cfg->vector);


Does 32-bit do the whole vector domain thing now?  Are these actually
equivalent?  Sending to self seems like it should be more efficient.

    J

Re: [tip:x86/apic] x86/ioapic.c: unify ioapic_retrigger_irq()

[Eric W. Biederman]

Jeremy Fitzhardinge <jeremy@goop.org> writes:

> On 07/18/09 05:05, tip-bot for Jeremy Fitzhardinge wrote:
>> Commit-ID:  e25371d60cb06a44d7a32d7966ab9bfbeacb9390
>> Gitweb:     http://git.kernel.org/tip/e25371d60cb06a44d7a32d7966ab9bfbeacb9390
>> Author:     Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
>> AuthorDate: Mon, 8 Jun 2009 03:49:01 -0700
>> Committer:  Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
>> CommitDate: Tue, 14 Jul 2009 13:32:51 -0700
>>
>> x86/ioapic.c: unify ioapic_retrigger_irq()
>>
>> The 32 and 64-bit versions of ioapic_retrigger_irq() are identical
>> except the 64-bit one takes vector_lock.  vector_lock is defined and
>> used on 32-bit too, so just use a common ioapic_retrigger_irq().
>>   
>
> Having another look at this patch, the other difference is that the
> 32-bit version just does:
>
>     apic->send_IPI_self(irq_cfg(irq)->vector);
>
> whereas the 64-bit does:
>
>     apic->send_IPI_mask(cpumask_of(cpumask_first(cfg->domain)),
>     cfg->vector);
>
>
> Does 32-bit do the whole vector domain thing now?  Are these actually
> equivalent?  Sending to self seems like it should be more efficient.

It should.  I believe YH did that work when he merged the x86_64 and
i386 versions io_apic.c.  Sending to self is a problem if we retrigger
this from the wrong irq.

Eric