Re: [PATCH v7 03/12] KVM: arm64: PMU: Clear PM{C,I}NTEN{SET,CLR} and PMOVS{SET,CLR} on vCPU reset

[Raghavendra Rao Ananta]

On Wed, Oct 18, 2023 at 2:16 PM Raghavendra Rao Ananta
<rananta@xxxxxxxxxx> wrote:
>
> On Tue, Oct 17, 2023 at 9:59 AM Raghavendra Rao Ananta
> <rananta@xxxxxxxxxx> wrote:
> >
> > Hi Eric,
> > On Tue, Oct 17, 2023 at 2:23 AM Eric Auger <eauger@xxxxxxxxxx> wrote:
> > >
> > > Hi,
> > > On 10/16/23 23:28, Raghavendra Rao Ananta wrote:
> > > > On Mon, Oct 16, 2023 at 12:45 PM Eric Auger <eauger@xxxxxxxxxx> wrote:
> > > >>
> > > >> Hi Raghavendra,
> > > >>
> > > >> On 10/10/23 01:08, Raghavendra Rao Ananta wrote:
> > > >>> From: Reiji Watanabe <reijiw@xxxxxxxxxx>
> > > >>>
> > > >>> On vCPU reset, PMCNTEN{SET,CLR}_EL0, PMINTEN{SET,CLR}_EL1, and
> > > >>> PMOVS{SET,CLR}_EL1 for a vCPU are reset by reset_pmu_reg().
> > > >> PMOVS{SET,CLR}_EL0?
> > > > Ah, yes. It should be PMOVS{SET,CLR}_EL0.
> > > >
> > > >>> This function clears RAZ bits of those registers corresponding
> > > >>> to unimplemented event counters on the vCPU, and sets bits
> > > >>> corresponding to implemented event counters to a predefined
> > > >>> pseudo UNKNOWN value (some bits are set to 1).
> > > >>>
> > > >>> The function identifies (un)implemented event counters on the
> > > >>> vCPU based on the PMCR_EL0.N value on the host. Using the host
> > > >>> value for this would be problematic when KVM supports letting
> > > >>> userspace set PMCR_EL0.N to a value different from the host value
> > > >>> (some of the RAZ bits of those registers could end up being set to 1).
> > > >>>
> > > >>> Fix this by clearing the registers so that it can ensure
> > > >>> that all the RAZ bits are cleared even when the PMCR_EL0.N value
> > > >>> for the vCPU is different from the host value. Use reset_val() to
> > > >>> do this instead of fixing reset_pmu_reg(), and remove
> > > >>> reset_pmu_reg(), as it is no longer used.
> > > >> do you intend to restore the 'unknown' behavior at some point?
> > > >>
> > > > I believe Reiji's (original author) intention was to keep them
> > > > cleared, which would still imply an 'unknown' behavior. Do you think
> > > > there's an issue with this?
> > > Then why do we bother using reset_unknown in the other places if
> > > clearing the bits is enough here?
> > >
> > Hmm. Good point. I can bring back reset_unknown to keep the original behavior.
> >
> I had a brief discussion about this with Oliver, and it looks like we
> might need a couple of additional changes for these register accesses:
> - For the userspace accesses, we have to implement explicit get_user
> and set_user callbacks that to filter out the unimplemented counters
> using kvm_pmu_valid_counter_mask().
Re-thinking the first case: Since these registers go through a reset
(reset_pmu_reg()) during initialization, where the valid counter mask
is applied, and since we are sanitizing the registers with the mask
before running the guest (below case), will implementing the
{get,set}_user() add any value, apart from just keeping userspace in
sync with every update of PMCR.N?
> - For the guest accesses to be correct, we might have to apply the
> same mask while serving KVM_REQ_RELOAD_PMU.
>
> Thank you.
> Raghavendra
>
> > Thank you.
> > Raghavendra
> > > Thanks
> > >
> > > Eric
> > > >
> > > > Thank you.
> > > > Raghavendra
> > > >> Thanks
> > > >>
> > > >> Eric
> > > >>>
> > > >>> Signed-off-by: Reiji Watanabe <reijiw@xxxxxxxxxx>
> > > >>> Signed-off-by: Raghavendra Rao Ananta <rananta@xxxxxxxxxx>
> > > >>> ---
> > > >>>  arch/arm64/kvm/sys_regs.c | 21 +--------------------
> > > >>>  1 file changed, 1 insertion(+), 20 deletions(-)
> > > >>>
> > > >>> diff --git a/arch/arm64/kvm/sys_regs.c b/arch/arm64/kvm/sys_regs.c
> > > >>> index 818a52e257ed..3dbb7d276b0e 100644
> > > >>> --- a/arch/arm64/kvm/sys_regs.c
> > > >>> +++ b/arch/arm64/kvm/sys_regs.c
> > > >>> @@ -717,25 +717,6 @@ static unsigned int pmu_visibility(const struct kvm_vcpu *vcpu,
> > > >>>       return REG_HIDDEN;
> > > >>>  }
> > > >>>
> > > >>> -static u64 reset_pmu_reg(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
> > > >>> -{
> > > >>> -     u64 n, mask = BIT(ARMV8_PMU_CYCLE_IDX);
> > > >>> -
> > > >>> -     /* No PMU available, any PMU reg may UNDEF... */
> > > >>> -     if (!kvm_arm_support_pmu_v3())
> > > >>> -             return 0;
> > > >>> -
> > > >>> -     n = read_sysreg(pmcr_el0) >> ARMV8_PMU_PMCR_N_SHIFT;
> > > >>> -     n &= ARMV8_PMU_PMCR_N_MASK;
> > > >>> -     if (n)
> > > >>> -             mask |= GENMASK(n - 1, 0);
> > > >>> -
> > > >>> -     reset_unknown(vcpu, r);
> > > >>> -     __vcpu_sys_reg(vcpu, r->reg) &= mask;
> > > >>> -
> > > >>> -     return __vcpu_sys_reg(vcpu, r->reg);
> > > >>> -}
> > > >>> -
> > > >>>  static u64 reset_pmevcntr(struct kvm_vcpu *vcpu, const struct sys_reg_desc *r)
> > > >>>  {
> > > >>>       reset_unknown(vcpu, r);
> > > >>> @@ -1115,7 +1096,7 @@ static bool access_pmuserenr(struct kvm_vcpu *vcpu, struct sys_reg_params *p,
> > > >>>         trap_wcr, reset_wcr, 0, 0,  get_wcr, set_wcr }
> > > >>>
> > > >>>  #define PMU_SYS_REG(name)                                            \
> > > >>> -     SYS_DESC(SYS_##name), .reset = reset_pmu_reg,                   \
> > > >>> +     SYS_DESC(SYS_##name), .reset = reset_val,                       \
> > > >>>       .visibility = pmu_visibility
> > > >>>
> > > >>>  /* Macro to expand the PMEVCNTRn_EL0 register */
> > > >>
> > > >
> > >