Re: [PATCH v4 6/6] arm64: cpufeature: Detect BBML3 based on ID_AA64MMFR2_EL1.BBM

On Wed, Aug 05, 2026 at 02:50:55PM +0530, Linu Cherian wrote:
> On Tue, Aug 04, 2026 at 02:51:28PM +0100, Will Deacon wrote:
> > On Mon, Aug 03, 2026 at 09:22:02AM +0530, Linu Cherian wrote:
> > > On Fri, Jul 31, 2026 at 04:13:20PM +0100, Will Deacon wrote:
> > > > On Thu, Jul 23, 2026 at 10:10:33AM +0530, Linu Cherian wrote:
> > > > > Add ID_AA64MMFR2_EL1.BBM based BBML3 feature detection in
> > > > > cpu_supports_bbml3() so that cpus with the feature would
> > > > > not have to be added into MIDR based supports_bbml3_list.
> > > > > 
> > > > > Reviewed-by: Gavin Shan <[email protected]>
> > > > > Reviewed-by: Anshuman Khandual <[email protected]>
> > > > > Signed-off-by: Linu Cherian <[email protected]>
> > > > > ---
> > > > >  arch/arm64/kernel/cpufeature.c | 17 +++++++++--------
> > > > >  1 file changed, 9 insertions(+), 8 deletions(-)
> > > > > 
> > > > > diff --git a/arch/arm64/kernel/cpufeature.c b/arch/arm64/kernel/cpufeature.c
> > > > > index 896bafdb00b1..dbd7d187520c 100644
> > > > > --- a/arch/arm64/kernel/cpufeature.c
> > > > > +++ b/arch/arm64/kernel/cpufeature.c
> > > > > @@ -2133,6 +2133,12 @@ static bool hvhe_possible(const struct arm64_cpu_capabilities *entry,
> > > > >  
> > > > >  bool cpu_supports_bbml3(void)
> > > > >  {
> > > > > +	u64 mmfr2;
> > > > > +
> > > > > +	mmfr2 = __read_sysreg_by_encoding(SYS_ID_AA64MMFR2_EL1);
> > > > > +	if (SYS_FIELD_GET(ID_AA64MMFR2_EL1, BBM, mmfr2) >= ID_AA64MMFR2_EL1_BBM_3)
> > > > > +		return true;
> > > > 
> > > > This is a bit of a nit, but I think it would be more consistent to use
> > > > has_cpuid_feature() here instead of __read_sysreg_by_encoding(), similarly
> > > > to how we handle kpti in unmap_kernel_at_el0() (which also has both an
> > > > ID register field and a list of MIDRs).
> > > 
> > > force_pte_mapping required by map_mem(during early boot) needs
> > > cpu_supports_bbml3 check and cpu features/capabilities are not 
> > > initialized by that time. Should i add a comment there to clarify this ?
> > 
> > It looks to me like has_cpuid_feature() will call
> > __read_sysreg_by_encoding() under the hood for SCOPE_LOCAL_CPU.
> > 
> > What am I missing?
> 
> Below is my understanding. Correct me if i am wrong.
> has_cpuid_feature depends on struct arm64_cpu_capabilities *entry.
> Inorder to derive *entry from a capability ID, cpucap_ptrs should be
> in initialized state and that wont be the case when force_pte_mapping
> invokes cpu_supports_bbml3. 

Ah, I see what you mean. We do have a 'struct arm64_cpu_capabilities'
entry for this in arm64_features[] but fishing that out is grotty at
the setup_arch() stage.

Will