RE: [v3 23/26] KVM: Update Posted-Interrupts Descriptor when vCPU is preempted

From: Wu, Feng
Date: Wed Dec 17 2014 - 22:16:57 EST




> -----Original Message-----
> From: Paolo Bonzini [mailto:paolo.bonzini@xxxxxxxxx] On Behalf Of Paolo
> Bonzini
> Sent: Thursday, December 18, 2014 1:11 AM
> To: Wu, Feng; Thomas Gleixner; Ingo Molnar; H. Peter Anvin; x86@xxxxxxxxxx;
> Gleb Natapov; Paolo Bonzini; dwmw2@xxxxxxxxxxxxx; joro@xxxxxxxxxx; Alex
> Williamson; jiang.liu-VuQAYsv1563Yd54FQh9/CA@xxxxxxxxxxxxxxxx
> Cc: iommu@xxxxxxxxxxxxxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx; KVM list;
> Eric Auger
> Subject: Re: [v3 23/26] KVM: Update Posted-Interrupts Descriptor when vCPU
> is preempted
>
>
>
> On 12/12/2014 16:14, Feng Wu wrote:
> > + if (irq_remapping_cap(IRQ_POSTING_CAP)) {
> > + struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
> > + struct pi_desc old, new;
> > + unsigned int dest;
> > +
> > + memset(&old, 0, sizeof(old));
> > + memset(&new, 0, sizeof(new));
>
> This is quite expensive. Just use an u64 for old_control and
> new_control, instead of a full struct.
>
> >
> > + pi_clear_sn(&new);
>
> This can be simply new.sn = 0. It does not need atomic operations.

Thanks for your comments, Paolo!

If we use u64 new_control, we cannot use new.sn any more.
Maybe we can change the struct pi_desc {} like this:

typedef struct pid_control{
u64 on : 1,
sn : 1,
rsvd_1 : 13,
ndm : 1,
nv : 8,
rsvd_2 : 8,
ndst : 32;
}pid_control_t;

struct pi_desc {
u32 pir[8]; /* Posted interrupt requested */
pid_control_t control;
u32 rsvd[6];
} __aligned(64);


Then we can define pid_control_t new_control, old_control. And use new_control.sn = 0.

What is your opinon?

Thanks,
Feng

>
> Same in patch 24 (if needed at all there---see the reply there).
>
> >
> > + if (irq_remapping_cap(IRQ_POSTING_CAP)) {
> > + struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
> > + struct pi_desc old, new;
> > +
> > + memset(&old, 0, sizeof(old));
> > + memset(&new, 0, sizeof(new));
> > +
>
> Here you do not need old/new at all because...
>
> > + if (vcpu->preempted) {
> > + do {
> > + old.control = new.control = pi_desc->control;
> > + pi_set_sn(&new);
> > + } while (cmpxchg(&pi_desc->control, old.control,
> > + new.control) != old.control);
>
> this can do pi_set_sn directly on pi_desc, without the cmpxchg.
>
> Paolo
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/