Re: [PATCH v6 2/4] KVM: X86: Add Paravirt TLB Shootdown

[Radim KrÄmÃÅ]

2017-11-30 14:24+0800, Wanpeng Li:
> 2017-11-30 0:21 GMT+08:00 Radim KrÄmÃÅ <rkrcmar@xxxxxxxxxx>:
> > 2017-11-27 20:05-0800, Wanpeng Li:
> >> From: Wanpeng Li <wanpeng.li@xxxxxxxxxxx>
> >> diff --git a/arch/x86/kernel/kvm.c b/arch/x86/kernel/kvm.c
> >> @@ -498,6 +498,37 @@ static void __init kvm_apf_trap_init(void)
> >>       update_intr_gate(X86_TRAP_PF, async_page_fault);
> >>  }
> >>
> >> +static DEFINE_PER_CPU(cpumask_t, __pv_tlb_mask);
> >> +
> >> +static void kvm_flush_tlb_others(const struct cpumask *cpumask,
> >> +                     const struct flush_tlb_info *info)
> >> +{
> >> +     u8 state;
> >> +     int cpu;
> >> +     struct kvm_steal_time *src;
> >> +     cpumask_t *flushmask = &per_cpu(__pv_tlb_mask, smp_processor_id());
> >> +
> >> +     if (unlikely(!flushmask))
> >> +             return;
> >
> > I don't see how this can be NULL and if it could, we'd have to call
> > native_flush_tlb_others() instead of returning anyway.
> >
> > Also, Peter mentioned that we're wasting memory (default is 1k per CPU)
> > when not running on KVM.  Hyper-V hijacks x86_platform.apic_post_init()
> > to achieve late allocation.  smp_ops.smp_prepare_cpus seems slightly
> > better for our purposes, but I don't really like either.
> >
> > Couldn't we use use arch_initcall(), or early_initcall() if there are
> > complications with allocating after smp_init()?
> 
> Do it in v7. In addition, move pv_mmu_ops.flush_tlb_others =
> kvm_flush_tlb_others to the arch_initcall() fails to work even if I
> disable rodata through grub. So I continue to keep the callback
> replacement in kvm_guest_init() and late allocation in
> arch_initcall().

I think it has to do with the patching -- you'd need to re-patch
flush_tlb_others callsites for the change to take effect or add a
hypervisor late init just before check_bugs(), where the patching is
currently done.

Not sure how either of those is acceptable, though.