Re: [RFC PATCH v1 01/15] x86/msr: Replace __wrmsr() with native_wrmsrl()

[Xin Li]

On 3/31/2025 1:32 PM, H. Peter Anvin wrote:
> On March 31, 2025 3:17:30 AM PDT, Ingo Molnar <mingo@xxxxxxxxxx> wrote:
> > * Xin Li (Intel) <xin@xxxxxxxxx> wrote:
> >
> > > -	__wrmsr      (MSR_AMD_DBG_EXTN_CFG, val | 3ULL << 3, val >> 32);
> > > +	native_wrmsrl(MSR_AMD_DBG_EXTN_CFG, val | 3ULL << 3);
> >
> > This is an improvement.
> >
> > > -	__wrmsr      (MSR_IA32_PQR_ASSOC, rmid_p, plr->closid);
> > > +	native_wrmsrl(MSR_IA32_PQR_ASSOC, (u64)plr->closid << 32 | rmid_p);
> >
> > > -	__wrmsr      (MSR_IA32_PQR_ASSOC, rmid_p, closid_p);
> > > +	native_wrmsrl(MSR_IA32_PQR_ASSOC, (u64)closid_p << 32 | rmid_p);
> >
> > This is not an improvement.
> >
> > Please provide a native_wrmsrl() API variant where natural [rmid_p, closid_p]
> > high/lo parameters can be used, without the shift-uglification...
> >
> > Thanks,
> >
> > 	Ingo
>
> Directing this question primarily to Ingo, who is more than anyone else the namespace consistency guardian:
>
> On the subject of msr function naming ... *msrl() has always been misleading. The -l suffix usually means 32 bits; sometimes it means the C type "long" (which in the kernel is used instead of size_t/uintptr_t, which might end up being "fun" when 128-bit architectures appear some time this century), but for a fixed 64-but type we normally use -q.
>
> Should we rename the *msrl() functions to *msrq() as part of this overhaul?

Per "struct msr" defined in arch/x86/include/asm/shared/msr.h:

struct msr {
        union {
                struct {
                        u32 l;
                        u32 h;
                };
                u64 q;
        };
};

Probably *msrq() is what we want?