Re: [Qemu-devel] [PATCH v4 2/3] kvm: Add support to KVM_GET_MSR_FEATURE_INDEX_LIST and KVM_GET_MSRS system ioctl

[Robert Hoo]

On Thu, 2018-09-20 at 00:07 -0300, Eduardo Habkost wrote:
> Hi,
> 
> Thanks for the patch and sorry for taking so long to review it.

Never mind. I understand you're really busy. :-)
> 
> On Sun, Sep 02, 2018 at 07:46:06PM +0800, Robert Hoo wrote:
> > Add kvm_get_supported_feature_msrs() to get supported MSR feature
> > index list.
> > Add kvm_arch_get_supported_msr_feature() to get each MSR features
> > value.
> > 
> > Signed-off-by: Robert Hoo <address@hidden>
> > ---
> >  include/sysemu/kvm.h |  2 ++
> >  target/i386/cpu.c    |  7 ++---
> >  target/i386/kvm.c    | 72
> > ++++++++++++++++++++++++++++++++++++++++++++++++++++
> >  3 files changed, 78 insertions(+), 3 deletions(-)
> > 
> > diff --git a/include/sysemu/kvm.h b/include/sysemu/kvm.h
> > index 0b64b8e..97d8d9d 100644
> > --- a/include/sysemu/kvm.h
> > +++ b/include/sysemu/kvm.h
> > @@ -463,6 +463,8 @@ int kvm_vm_check_extension(KVMState *s,
> > unsigned int extension);
> >  
> >  uint32_t kvm_arch_get_supported_cpuid(KVMState *env, uint32_t
> > function,
> >                                        uint32_t index, int reg);
> > +uint32_t kvm_arch_get_supported_msr_feature(KVMState *s, uint32_t
> > index);
> > +
> >  
> >  void kvm_set_sigmask_len(KVMState *s, unsigned int sigmask_len);
> >  
> > diff --git a/target/i386/cpu.c b/target/i386/cpu.c
> > index a252c26..0160e97 100644
> > --- a/target/i386/cpu.c
> > +++ b/target/i386/cpu.c
> > @@ -3670,7 +3670,7 @@ static uint32_t
> > x86_cpu_get_supported_feature_word(FeatureWord w,
> >                                                     bool
> > migratable_only)
> >  {
> >      FeatureWordInfo *wi = &feature_word_info[w];
> > -    uint32_t r;
> > +    uint32_t r = 0;
> >  
> >      if (kvm_enabled()) {
> >          switch (wi->type) {
> > @@ -3679,8 +3679,9 @@ static uint32_t
> > x86_cpu_get_supported_feature_word(FeatureWord w,
> >                                                  wi->cpuid.ecx,
> >                                                  wi->cpuid.reg);
> >              break;
> > -        default:
> > -            r = 0;
> > +        case MSR_FEATURE_WORD:
> > +            r = kvm_arch_get_supported_msr_feature(kvm_state,
> > +                        wi->msr.index);
> 
> If you move this patch before patch 1/3, this hunk could be part
> of patch 1/3.
> 
I'm afraid that if I moved this hunk, because of the dependency, I
would have to move the definition of
kvm_arch_get_supported_msr_feature() to patch 1/3, then in turn, it
used kvm_feature_msrs, I've to put its definition and initialization
function kvm_get_supported_feature_msrs() to patch 1/3 as well. Then
actually, this makes patch 1/3 and 2/3 merged into 1. Would you like me
to do so?

> >              break;
> >          }
> >      } else if (hvf_enabled()) {
> > diff --git a/target/i386/kvm.c b/target/i386/kvm.c
> > index 0b2a07d..bfd8088 100644
> > --- a/target/i386/kvm.c
> > +++ b/target/i386/kvm.c
> > @@ -107,6 +107,7 @@ static int has_pit_state2;
> >  static bool has_msr_mcg_ext_ctl;
> >  
> >  static struct kvm_cpuid2 *cpuid_cache;
> > +static struct kvm_msr_list *kvm_feature_msrs;
> >  
> >  int kvm_has_pit_state2(void)
> >  {
> > @@ -420,6 +421,33 @@ uint32_t kvm_arch_get_supported_cpuid(KVMState
> > *s, uint32_t function,
> >      return ret;
> >  }
> >  
> > +uint32_t kvm_arch_get_supported_msr_feature(KVMState *s, uint32_t
> > index)
> > +{
> > +    struct {
> > +        struct kvm_msrs info;
> > +        struct kvm_msr_entry entries[1];
> > +    } msr_data;
> > +    uint32_t ret;
> > +
> > +    if (kvm_feature_msrs == NULL) { /*ARCH doesn't support feature
> > MSRs*/
> 
> Nit: normally comments have spaces after "/*" and before "*/".
> 
> Also: what do you mean by "ARCH"?  Do you mean "host kernel"?
> 
Going to say "Host".

> 
> > +        return 0;
> > +    }
> > +
> > +    msr_data.info.nmsrs = 1;
> > +    msr_data.entries[0].index = index;
> > +
> > +    ret = kvm_ioctl(s, KVM_GET_MSRS, &msr_data);
> > +
> > +    if (ret != 1) {
> 
> If the MSR is not supported by the host kernel, it must not be a
> fatal error.  We should just return 0 on that case.
> 
> Probably the best way to ensure that is to check if the MSR is
> listed on kvm_feature_msrs before calling KVM_GET_MSRS (and
> return 0 if the MSR is not on the list).

Yes. Will do in this way in v5. 
> 
> 
> > +        fprintf(stderr, "KVM get MSR (index=0x%x) feature failed,
> > %s\n",
> > +            index, strerror(-ret));
> 
> Please use error_report() instead of fprintf(stderr).
> 
> > +        exit(1);
> 
> I'm unsure if exit(1) is the best option here, but at least this
> is consistent with error handling kvm_arch_get_supported_cpuid().
> 
> 
> > +    }
> > +
> > +    return msr_data.entries[0].data;
> > +}
> > +
> > +
> >  typedef struct HWPoisonPage {
> >      ram_addr_t ram_addr;
> >      QLIST_ENTRY(HWPoisonPage) list;
> > @@ -1239,6 +1267,45 @@ void kvm_arch_do_init_vcpu(X86CPU *cpu)
> >      }
> >  }
> >  
> > +static int kvm_get_supported_feature_msrs(KVMState *s)
> > +{
> > +    int ret = 0;
> > +
> > +    if (kvm_feature_msrs != NULL) {
> > +        return 0;
> > +    }
> > +
> > +    if (!kvm_check_extension(s, KVM_CAP_GET_MSR_FEATURES)) {
> > +        return -1;
> 
> There's nothing wrong with not supporting
> KVM_CAP_GET_MSR_FEATURES.  Why not return 0?
> 
OK
> 
> > +    }
> > +
> > +    struct kvm_msr_list msr_list;
> > +
> > +    msr_list.nmsrs = 0;
> > +    ret = kvm_ioctl(s, KVM_GET_MSR_FEATURE_INDEX_LIST, &msr_list);
> > +    if (ret < 0 && ret != -E2BIG) {
> 
> You print an error to stderr if (ret < 0) below, but don't print
> anything here.  Seems inconsistent.

Going to be consistent.
> 
> > +        return ret;
> > +    }
> > +
> > +    assert(msr_list.nmsrs > 0);
> > +    kvm_feature_msrs = (struct kvm_msr_list *) \
> > +        g_malloc0(sizeof(msr_list) +
> > +                 msr_list.nmsrs * sizeof(msr_list.indices[0]));
> > +
> > +    kvm_feature_msrs->nmsrs = msr_list.nmsrs;
> > +    ret = kvm_ioctl(s, KVM_GET_MSR_FEATURE_INDEX_LIST,
> > kvm_feature_msrs);
> 
> kvm_arch_get_supported_msr_feature() is only checking if
> kvm_feature_msrs is NULL, and nothing else.
> 
> What exactly is the point of calling
> KVM_GET_MSR_FEATURE_INDEX_LIST and saving data at
> kvm_feature_msrs, if no other code is ever looking at the
> returned data?
> 
Now above change make it useful for checking requested feature MSR's
validity.
> 
> > +
> > +    if (ret < 0) {
> > +        fprintf(stderr, "Fetch KVM feature MSRs failed: %s\n",
> > +            strerror(-ret));
> 
> Please use error_report() instead of fprintf(stderr).

OK
> 
> > +        g_free(kvm_feature_msrs);
> > +        kvm_feature_msrs = NULL;
> > +        return ret;
> > +    }
> > +
> > +    return 0;
> > +}
> > +
> >  static int kvm_get_supported_msrs(KVMState *s)
> >  {
> >      static int kvm_supported_msrs;
> > @@ -1392,6 +1459,11 @@ int kvm_arch_init(MachineState *ms, KVMState
> > *s)
> >          return ret;
> >      }
> >  
> > +    ret = kvm_get_supported_feature_msrs(s);
> > +    if (ret < 0) { /*if MSR based features aren't supported,
> > ignore it.*/
> > +        warn_report("Get supported feature MSRs failed.");
> 
> We must not print a warning only because KVM_CAP_GET_MSR_FEATURES
> isn't supported by the host kernel.
> 
> If KVM_GET_MSR_FEATURE_INDEX_LIST fails, on the other hand, we
> probably should make it a fatal error and not a warning.
> 
OK, remove the check.

> > +    }
> > +
> >      uname(&utsname);
> >      lm_capable_kernel = strcmp(utsname.machine, "x86_64") == 0;
> >  
> > -- 
> > 1.8.3.1
> > 
> > 
> 
>