| Message ID | 1558607223-19630-1-git-send-email-andrew.cooper3@citrix.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | libx86: Introduce wrappers for extracting XCR0/XSS from a cpuid policy | expand |
>>> On 23.05.19 at 12:27, <andrew.cooper3@citrix.com> wrote: > --- a/xen/arch/x86/xstate.c > +++ b/xen/arch/x86/xstate.c > @@ -660,9 +660,7 @@ static bool valid_xcr0(u64 xcr0) > int validate_xstate(const struct domain *d, uint64_t xcr0, uint64_t xcr0_accum, > const struct xsave_hdr *hdr) > { > - const struct cpuid_policy *cp = d->arch.cpuid; > - uint64_t xcr0_max = > - ((uint64_t)cp->xstate.xcr0_high << 32) | cp->xstate.xcr0_low; > + uint64_t xcr0_max = cpuid_policy_xcr0(d->arch.cpuid); > unsigned int i; > > if ( (hdr->xstate_bv & ~xcr0_accum) || > @@ -686,9 +684,7 @@ int validate_xstate(const struct domain *d, uint64_t xcr0, uint64_t xcr0_accum, > int handle_xsetbv(u32 index, u64 new_bv) > { > struct vcpu *curr = current; > - const struct cpuid_policy *cp = curr->domain->arch.cpuid; > - uint64_t xcr0_max = > - ((uint64_t)cp->xstate.xcr0_high << 32) | cp->xstate.xcr0_low; > + uint64_t xcr0_max = cpuid_policy_xcr0(curr->domain->arch.cpuid); In both cases the variables are more appropriately named than the new helper. While I agree it's slightly more typing, did you consider calling it cpuid_policy_xcr0_max()? > --- a/xen/include/xen/lib/x86/cpuid.h > +++ b/xen/include/xen/lib/x86/cpuid.h > @@ -308,6 +308,18 @@ static inline void cpuid_featureset_to_policy( > p->feat._7a1 = fs[FEATURESET_7a1]; > } > > +static inline uint64_t cpuid_policy_xcr0(const struct cpuid_policy *p) > +{ > + return ((uint64_t)p->xstate.xcr0_high << 32) | p->xstate.xcr0_low; > +} > + > +static inline uint64_t cpuid_policy_xstates(const struct cpuid_policy *p) > +{ > + uint64_t val = p->xstate.xcr0_high | p->xstate.xss_high; > + > + return (val << 32) | p->xstate.xcr0_low | p->xstate.xss_low; > +} How about also having cpuid_policy_xss() (or cpuid_policy_xss_max()) and then simply making cpuid_policy_xstates() combine the two results? Anyway, as I can also live with things as they are, with or without either of the suggested changes Reviewed-by: Jan Beulich <jbeulich@suse.com> Jan
On 23/05/2019 12:52, Jan Beulich wrote: >>>> On 23.05.19 at 12:27, <andrew.cooper3@citrix.com> wrote: >> --- a/xen/arch/x86/xstate.c >> +++ b/xen/arch/x86/xstate.c >> @@ -660,9 +660,7 @@ static bool valid_xcr0(u64 xcr0) >> int validate_xstate(const struct domain *d, uint64_t xcr0, uint64_t xcr0_accum, >> const struct xsave_hdr *hdr) >> { >> - const struct cpuid_policy *cp = d->arch.cpuid; >> - uint64_t xcr0_max = >> - ((uint64_t)cp->xstate.xcr0_high << 32) | cp->xstate.xcr0_low; >> + uint64_t xcr0_max = cpuid_policy_xcr0(d->arch.cpuid); >> unsigned int i; >> >> if ( (hdr->xstate_bv & ~xcr0_accum) || >> @@ -686,9 +684,7 @@ int validate_xstate(const struct domain *d, uint64_t xcr0, uint64_t xcr0_accum, >> int handle_xsetbv(u32 index, u64 new_bv) >> { >> struct vcpu *curr = current; >> - const struct cpuid_policy *cp = curr->domain->arch.cpuid; >> - uint64_t xcr0_max = >> - ((uint64_t)cp->xstate.xcr0_high << 32) | cp->xstate.xcr0_low; >> + uint64_t xcr0_max = cpuid_policy_xcr0(curr->domain->arch.cpuid); > In both cases the variables are more appropriately named than > the new helper. While I agree it's slightly more typing, did you > consider calling it cpuid_policy_xcr0_max()? Fine. > >> --- a/xen/include/xen/lib/x86/cpuid.h >> +++ b/xen/include/xen/lib/x86/cpuid.h >> @@ -308,6 +308,18 @@ static inline void cpuid_featureset_to_policy( >> p->feat._7a1 = fs[FEATURESET_7a1]; >> } >> >> +static inline uint64_t cpuid_policy_xcr0(const struct cpuid_policy *p) >> +{ >> + return ((uint64_t)p->xstate.xcr0_high << 32) | p->xstate.xcr0_low; >> +} >> + >> +static inline uint64_t cpuid_policy_xstates(const struct cpuid_policy *p) >> +{ >> + uint64_t val = p->xstate.xcr0_high | p->xstate.xss_high; >> + >> + return (val << 32) | p->xstate.xcr0_low | p->xstate.xss_low; >> +} > How about also having cpuid_policy_xss() (or cpuid_policy_xss_max()) > and then simply making cpuid_policy_xstates() combine the two > results? I started with that, but the resulting code was a little awkward to read, and the asm generation was a little worse due to promoting everything first. I don't think we need cpuid_policy_xss{,_max}() until we actually implement something for guests (most likely CET at this rate). > > Anyway, as I can also live with things as they are, with or without > either of the suggested changes > Reviewed-by: Jan Beulich <jbeulich@suse.com> Thanks (although I'm still happy to play around with naming). ~Andrew
>>> On 23.05.19 at 13:59, <andrew.cooper3@citrix.com> wrote: > On 23/05/2019 12:52, Jan Beulich wrote: >>>>> On 23.05.19 at 12:27, <andrew.cooper3@citrix.com> wrote: >>> --- a/xen/include/xen/lib/x86/cpuid.h >>> +++ b/xen/include/xen/lib/x86/cpuid.h >>> @@ -308,6 +308,18 @@ static inline void cpuid_featureset_to_policy( >>> p->feat._7a1 = fs[FEATURESET_7a1]; >>> } >>> >>> +static inline uint64_t cpuid_policy_xcr0(const struct cpuid_policy *p) >>> +{ >>> + return ((uint64_t)p->xstate.xcr0_high << 32) | p->xstate.xcr0_low; >>> +} >>> + >>> +static inline uint64_t cpuid_policy_xstates(const struct cpuid_policy *p) >>> +{ >>> + uint64_t val = p->xstate.xcr0_high | p->xstate.xss_high; >>> + >>> + return (val << 32) | p->xstate.xcr0_low | p->xstate.xss_low; >>> +} >> How about also having cpuid_policy_xss() (or cpuid_policy_xss_max()) >> and then simply making cpuid_policy_xstates() combine the two >> results? > > I started with that, but the resulting code was a little awkward to > read, and the asm generation was a little worse due to promoting > everything first. > > I don't think we need cpuid_policy_xss{,_max}() until we actually > implement something for guests (most likely CET at this rate). Well, let's stick to what you have then. Jan
diff --git a/xen/arch/x86/xstate.c b/xen/arch/x86/xstate.c index 3da609a..04da569 100644 --- a/xen/arch/x86/xstate.c +++ b/xen/arch/x86/xstate.c @@ -660,9 +660,7 @@ static bool valid_xcr0(u64 xcr0) int validate_xstate(const struct domain *d, uint64_t xcr0, uint64_t xcr0_accum, const struct xsave_hdr *hdr) { - const struct cpuid_policy *cp = d->arch.cpuid; - uint64_t xcr0_max = - ((uint64_t)cp->xstate.xcr0_high << 32) | cp->xstate.xcr0_low; + uint64_t xcr0_max = cpuid_policy_xcr0(d->arch.cpuid); unsigned int i; if ( (hdr->xstate_bv & ~xcr0_accum) || @@ -686,9 +684,7 @@ int validate_xstate(const struct domain *d, uint64_t xcr0, uint64_t xcr0_accum, int handle_xsetbv(u32 index, u64 new_bv) { struct vcpu *curr = current; - const struct cpuid_policy *cp = curr->domain->arch.cpuid; - uint64_t xcr0_max = - ((uint64_t)cp->xstate.xcr0_high << 32) | cp->xstate.xcr0_low; + uint64_t xcr0_max = cpuid_policy_xcr0(curr->domain->arch.cpuid); u64 mask; if ( index != XCR_XFEATURE_ENABLED_MASK ) diff --git a/xen/include/xen/lib/x86/cpuid.h b/xen/include/xen/lib/x86/cpuid.h index 252d2c9..ea4db5b 100644 --- a/xen/include/xen/lib/x86/cpuid.h +++ b/xen/include/xen/lib/x86/cpuid.h @@ -308,6 +308,18 @@ static inline void cpuid_featureset_to_policy( p->feat._7a1 = fs[FEATURESET_7a1]; } +static inline uint64_t cpuid_policy_xcr0(const struct cpuid_policy *p) +{ + return ((uint64_t)p->xstate.xcr0_high << 32) | p->xstate.xcr0_low; +} + +static inline uint64_t cpuid_policy_xstates(const struct cpuid_policy *p) +{ + uint64_t val = p->xstate.xcr0_high | p->xstate.xss_high; + + return (val << 32) | p->xstate.xcr0_low | p->xstate.xss_low; +} + const uint32_t *x86_cpuid_lookup_deep_deps(uint32_t feature); /** diff --git a/xen/lib/x86/cpuid.c b/xen/lib/x86/cpuid.c index 23619c7..74c5b18 100644 --- a/xen/lib/x86/cpuid.c +++ b/xen/lib/x86/cpuid.c @@ -144,8 +144,7 @@ void x86_cpuid_policy_fill_native(struct cpuid_policy *p) cpuid_count_leaf(0xd, 0, &p->xstate.raw[0]); cpuid_count_leaf(0xd, 1, &p->xstate.raw[1]); - xstates = ((uint64_t)(p->xstate.xcr0_high | p->xstate.xss_high) << 32); - xstates |= (p->xstate.xcr0_low | p->xstate.xss_low); + xstates = cpuid_policy_xstates(p); for ( i = 2; i < min_t(unsigned int, 63, ARRAY_SIZE(p->xstate.raw)); ++i )
This avoids opencoding the slightly-awkward logic. More uses of these wrappers will be introduced shortly. Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com> --- CC: Jan Beulich <JBeulich@suse.com> CC: Wei Liu <wei.liu2@citrix.com> CC: Roger Pau Monné <roger.pau@citrix.com> I've decided to introduce this patch ahead of "[PATCH] libx86: Elide more empty CPUID leaves when serialising a policy" (which simplifies the xstate hunk a little) as I've found yet more cases where I need to use cpuid_policy_xstates(), and opencoding them all seemed very silly. --- xen/arch/x86/xstate.c | 8 ++------ xen/include/xen/lib/x86/cpuid.h | 12 ++++++++++++ xen/lib/x86/cpuid.c | 3 +-- 3 files changed, 15 insertions(+), 8 deletions(-)