On Wed, Jan 23, 2013 at 11:56:04AM -0600, Scott Wood wrote:
> On 01/22/2013 11:05:00 PM, David Gibson wrote:
> >Currently qemu does not get and put the state of the floating
> >point and
> >vector registers to KVM. This is obviously a problem for savevm,
> >as well
> >as possibly being problematic for debugging of FP-using guests.
> >
> >This patch fixes this by using new extensions to the ONE_REG
> >interface to
> >synchronize the qemu floating point state with KVM.
> >
> >Signed-off-by: David Gibson <da...@gibson.dropbear.id.au>
> >---
> > target-ppc/kvm.c | 123
> >++++++++++++++++++++++++++++++++++++++++++++++++++++++
> > 1 file changed, 123 insertions(+)
> >
> >diff --git a/target-ppc/kvm.c b/target-ppc/kvm.c
> >index e84b993..7ed76be 100644
> >--- a/target-ppc/kvm.c
> >+++ b/target-ppc/kvm.c
> >@@ -518,6 +518,125 @@ static void kvm_put_one_spr(CPUState *cs,
> >uint64_t id, int spr)
> > }
> > }
> >
> >+static void kvm_put_fp(CPUState *cs)
> >+{
> >+ PowerPCCPU *cpu = POWERPC_CPU(cs);
> >+ CPUPPCState *env = &cpu->env;
> >+ struct kvm_one_reg reg;
> >+ int i;
> >+ int ret;
> >+
> >+ if (env->insns_flags & PPC_FLOAT) {
> >+ uint64_t fpscr = env->fpscr;
> >+ bool vsx = !!(env->insns_flags2 & PPC2_VSX);
> >+
> >+ reg.id = KVM_REG_PPC_FPSCR;
> >+ reg.addr = (uintptr_t)&fpscr;
> >+ ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®);
> >+ if (ret < 0) {
> >+ fprintf(stderr, "Warning: Unable to set FPSCR to KVM:
> >%s\n",
> >+ strerror(errno));
> >+ }
> >+ for (i = 0; i < 32; i++) {
> >+ uint64_t vsr[2];
> >+
> >+ vsr[0] = float64_val(env->fpr[i]);
> >+ vsr[1] = env->vsr[i];
> >+ reg.addr = (uintptr_t) &vsr;
> >+ reg.id = vsx ? KVM_REG_PPC_VSR(i) : KVM_REG_PPC_FPR(i);
> >+
> >+ ret = kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, ®);
> >+ if (ret < 0) {
> >+ fprintf(stderr, "Warning Unable to set %s%d to
> >KVM: %s\n",
> >+ vsx ? "VSR" : "FPR", i, strerror(errno));
> >+ }
> >+ }
> >+ }
>
> I don't see where you check a capability, so users of pre-one-reg
> KVM now get (at least) 33 warning messages every time context gets
> synchronized?
Given that a bunch of important state is missing from that
synchronization, I think a slew of warnings is entirely appropriate.
There are no per-register capabilities so there's nothing we can check
here.
--
David Gibson | I'll have my music baroque, and my code
david AT gibson.dropbear.id.au | minimalist, thank you. NOT _the_ _other_
| _way_ _around_!
http://www.ozlabs.org/~dgibson
signature.asc
Description: Digital signature
