[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [PATCH] x86/vLAPIC: only accept interrupts on enabled LAPICs



On Thu, 19 Nov 2026, Jan Beulich wrote:
> On 19.11.2026 13:49, Andrew Cooper wrote:
> > On 19/11/2026 12:18 pm, Jan Beulich wrote:
> >> --- a/xen/arch/x86/include/asm/hvm/vlapic.h
> >> +++ b/xen/arch/x86/include/asm/hvm/vlapic.h
> >> @@ -136,6 +136,6 @@ struct vlapic *vlapic_lowest_prio(
> >>  
> >>  bool vlapic_match_dest(
> >>      const struct vlapic *target, const struct vlapic *source,
> >> -    int short_hand, uint32_t dest, bool dest_mode);
> >> +    unsigned int short_hand, uint32_t dest, bool dest_mode, unsigned int 
> >> dm);
> >>  
> > 
> > Even before this change, I was thinking that 3 fragments of ICR were too
> > many parameters to be passing, especially as some are bits as per their
> > ICR position, but dest_mode is shifted down to bit 0.
> > 
> > I think passing a real ICR pair will be an improvement both for
> > legibility and for code generation.
> 
> Are you saying this with just two of the three uses in vlapic.c in mind,
> or did you mean to suggest to have all the other call sites end up more
> complicated to achieve this goal?

Only the IPI paths have an ICR, and even there the destination has to be
decoded in the sender's mode first. The vIO-APIC, MSI and passthrough
callers would have to build a fake ICR.

On the patch itself:


default: needs fallthrough;

In _hvm_dpci_msi_eoi() if the guest EOIs after disabling its LAPIC in
software, the match now fails and the physical EOI is left to eoi_timer.
This is a change compared to existing behavior, was this intended?


This patch is better than my "[PATCH 2/2] x86/passthrough: skip disabled
vLAPICs in vector_hashing_dest()". I'll drop it.



 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.