Re: [PATCH] KVM: PPC: Book3S HV: Fix r3 corruption in h_set_dabr()

[Michael Neuling <mikey@neuling.org>]

> > > 3:
> > >         /* Emulate H_SET_DABR/X on P8 for the sake of compat mode
> > > guests */
> > >         rlwimi  r5, r4, 5, DAWRX_DR | DAWRX_DW
> > > c00000000010b03c:       74 2e 85 50     rlwimi  r5,r4,5,25,26
> > >         rlwimi  r5, r4, 2, DAWRX_WT
> > > c00000000010b040:       f6 16 85 50     rlwimi  r5,r4,2,27,27
> > >         clrrdi  r4, r4, 3
> > > c00000000010b044:       24 07 84 78     rldicr  r4,r4,0,60
> > >         std     r4, VCPU_DAWR(r3)
> > > c00000000010b048:       c0 13 83 f8     std     r4,5056(r3)
> > >         std     r5, VCPU_DAWRX(r3)
> > > c00000000010b04c:       c8 13 a3 f8     std     r5,5064(r3)
> > >         mtspr   SPRN_DAWR, r4
> > > c00000000010b050:       a6 2b 94 7c     mtspr   180,r4
> > >         mtspr   SPRN_DAWRX, r5
> > > c00000000010b054:       a6 2b bc 7c     mtspr   188,r5
> > >         li      r3, 0
> > > c00000000010b058:       00 00 60 38     li      r3,0
> > >         blr
> > > c00000000010b05c:       20 00 80 4e     blr
> > 
> > It's the `mtspr   SPRN_DAWR, r4` as you're HV=0.  I'm not sure how
> > nested works
> > in that regard. Is the level above suppose to trap and emulate
> > that?  
> > 
> 
> Yeah so as a nested hypervisor we need to avoid that call to mtspr
> SPRN_DAWR since it's HV privileged and we run with HV = 0.
> 
> The fix will be to check kvmhv_on_pseries() before doing the write. In
> fact we should avoid the write any time we call the function from _not_
> real mode.
> 
> I'll submit a fix for the KVM side. Doesn't look like this is anything
> to do with Mikey's patch, was always broken as far as I can tell.

Thanks Suraj.

Mikey