Re: [PATCH v3 09/10] KVM: riscv: selftests: Skip some registers set operation

[Andrew Jones <ajones@xxxxxxxxxxxxxxxx>]

On Sat, Jun 10, 2023 at 10:35:24AM +0800, Haibo Xu wrote:
> On Fri, Jun 9, 2023 at 5:24 PM Andrew Jones <ajones@xxxxxxxxxxxxxxxx> wrote:
> >
> > On Fri, Jun 09, 2023 at 10:12:17AM +0800, Haibo Xu wrote:
> > > Set operation on some riscv registers(mostly pesudo ones) was not
> > > supported and should be skipped in the get-reg-list test. Just
> > > reuse the rejects_set utilities to handle it in riscv.
> > >
> > > Signed-off-by: Haibo Xu <haibo1.xu@xxxxxxxxx>
> > > Reviewed-by: Andrew Jones <ajones@xxxxxxxxxxxxxxxx>
> > > ---
> > >  tools/testing/selftests/kvm/get-reg-list.c | 20 +++++++++++++-------
> > >  1 file changed, 13 insertions(+), 7 deletions(-)
> > >
> > > diff --git a/tools/testing/selftests/kvm/get-reg-list.c b/tools/testing/selftests/kvm/get-reg-list.c
> > > index c4bd5a5259da..abacb95c21c6 100644
> > > --- a/tools/testing/selftests/kvm/get-reg-list.c
> > > +++ b/tools/testing/selftests/kvm/get-reg-list.c
> > > @@ -211,16 +211,22 @@ static void run_test(struct vcpu_reg_list *c)
> > >                       ++failed_get;
> > >               }
> > >
> > > -             /* rejects_set registers are rejected after KVM_ARM_VCPU_FINALIZE */
> > > +             /*
> > > +              * rejects_set registers are rejected after KVM_ARM_VCPU_FINALIZE on aarch64,
> > > +              * or registers that should skip set operation on riscv.
> > > +              */
> > >               for_each_sublist(c, s) {
> > >                       if (s->rejects_set && find_reg(s->rejects_set, s->rejects_set_n, reg.id)) {
> > >                               reject_reg = true;
> > > -                             ret = __vcpu_ioctl(vcpu, KVM_SET_ONE_REG, &reg);
> > > -                             if (ret != -1 || errno != EPERM) {
> > > -                                     printf("%s: Failed to reject (ret=%d, errno=%d) ", config_name(c), ret, errno);
> > > -                                     print_reg(config_name(c), reg.id);
> > > -                                     putchar('\n');
> > > -                                     ++failed_reject;
> > > +                             if ((reg.id & KVM_REG_ARCH_MASK) == KVM_REG_ARM64) {
> > > +                                     ret = __vcpu_ioctl(vcpu, KVM_SET_ONE_REG, &reg);
> > > +                                     if (ret != -1 || errno != EPERM) {
> > > +                                             printf("%s: Failed to reject (ret=%d, errno=%d) ",
> > > +                                                             config_name(c), ret, errno);
> > > +                                             print_reg(config_name(c), reg.id);
> > > +                                             putchar('\n');
> > > +                                             ++failed_reject;
> > > +                                     }
> >
> > Thinking about this some more, shouldn't we attempt the set ioctl for
> > riscv reject registers as well, but look for different error numbers?
> >
> 
> Yes, we can. Currently, 2 different errno(EOPNOTSUPP/EINVAL) would be
> reported for the rejected registers in risc-v.
> These 2 errnos can be handled specially like below:
> 
> diff --git a/tools/testing/selftests/kvm/get-reg-list.c
> b/tools/testing/selftests/kvm/get-reg-list.c
> index 73f40e0842b8..f3f2c4519318 100644
> --- a/tools/testing/selftests/kvm/get-reg-list.c
> +++ b/tools/testing/selftests/kvm/get-reg-list.c
> @@ -255,6 +255,15 @@ static void run_test(struct vcpu_reg_list *c)
>                                                 putchar('\n');
>                                                 ++failed_reject;
>                                         }
> +                } else {
> +                                       ret = __vcpu_ioctl(vcpu,
> KVM_SET_ONE_REG, &reg);
> +                                       if (ret != -1 || (errno !=
> EINVAL && errno != EOPNOTSUPP)) {
> +                                               printf("%s: Failed to
> reject (ret=%d, errno=%d) ",
> +
> config_name(c), ret, errno);
> +
> print_reg(config_name(c), reg.id);
> +                                               putchar('\n');
> +                                               ++failed_reject;
> +                                       }

Instead of duplicating the code Arm uses, we just need an errno check
function, preferably one that takes the register as an input, so we
can check for specific errnos for specific registers.

> 
> One possible issue for the above change is that when new registers
> that don't support sets were added, we need
> to add them to the reject registers list, or the test would fail.
> 
> Initially, in the v1 patch, the design was to just skip the EOPNOTSUPP
> errno in set operations for all registers
> since it's a known errno for registers that don't support sets. This
> change cover all the registers even for future
> new ones.
> 
> What's your opinion?

I think we should only do the get/set tests on present, blessed list
registers, since if it's a new register we don't know its capabilities.

So, instead of

  for_each_reg(i) {
     /* get/set tests */
  }

we do
  
  for_each_present_blessed_reg(i) {
     /* get/set tests */
  }

where we have

 #define for_each_present_blessed_reg(i) \
     for ((i) = 0; (i) < blessed_n; ++(i)) \
         if (find_reg(reg_list->reg, reg_list->n, blessed_reg[i]))


Changing run_test() to work this way should be a separate patch.

Thanks,
drew