tools/tests/cpu-policy/test-cpu-policy.c | 71 ++++++++++++++++++++++++++++++++ xen/lib/x86/cpuid.c | 24 ++++++++++- 2 files changed, 94 insertions(+), 1 deletion(-)
x86_cpuid_copy_to_buffer() currently serialises the full content of the
various subleaf unions. While leaves 4, 0xb and 0xd don't have a concrete
max_subleaf field, they do have well defined upper bounds.
Diffing the results of `xen-cpuid -p` shows the resutling saving:
@@ -1,5 +1,5 @@
Xen reports there are maximum 114 leaves and 1 MSRs
-Raw policy: 93 leaves, 1 MSRs
+Raw policy: 38 leaves, 1 MSRs
CPUID:
leaf subleaf -> eax ebx ecx edx
00000000:ffffffff -> 00000016:756e6547:6c65746e:49656e69
@@ -32,7 +32,7 @@ Raw policy: 93 leaves, 1 MSRs
MSRs:
index -> value
000000ce -> 0000000080000000
-Host policy: 93 leaves, 1 MSRs
+Host policy: 33 leaves, 1 MSRs
CPUID:
leaf subleaf -> eax ebx ecx edx
00000000:ffffffff -> 0000000d:756e6547:6c65746e:49656e69
which is mostly due to no longer writing out 64 leaves for xstate when (on
this CoffeeLake system) 8 will do.
Extend the unit tests to cover empty and partially filled subleaf unions.
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>
---
tools/tests/cpu-policy/test-cpu-policy.c | 71 ++++++++++++++++++++++++++++++++
xen/lib/x86/cpuid.c | 24 ++++++++++-
2 files changed, 94 insertions(+), 1 deletion(-)
diff --git a/tools/tests/cpu-policy/test-cpu-policy.c b/tools/tests/cpu-policy/test-cpu-policy.c
index beced5e..fd96c0b 100644
--- a/tools/tests/cpu-policy/test-cpu-policy.c
+++ b/tools/tests/cpu-policy/test-cpu-policy.c
@@ -65,6 +65,77 @@ static void test_cpuid_serialise_success(void)
.name = "empty policy",
.nr_leaves = 4,
},
+
+ /* Leaf 4 serialisation stops at the first subleaf with type 0. */
+ {
+ .name = "empty leaf 4",
+ .p = {
+ .basic.max_leaf = 4,
+ },
+ .nr_leaves = 4 + 4,
+ },
+ {
+ .name = "partial leaf 4",
+ .p = {
+ .basic.max_leaf = 4,
+ .cache.subleaf[0].type = 1,
+ },
+ .nr_leaves = 4 + 4 + 1,
+ },
+
+ /* Leaf 7 serialisation stops at max_subleaf. */
+ {
+ .name = "empty leaf 7",
+ .p = {
+ .basic.max_leaf = 7,
+ },
+ .nr_leaves = 4 + 7,
+ },
+ {
+ .name = "partial leaf 7",
+ .p = {
+ .basic.max_leaf = 7,
+ .feat.max_subleaf = 1,
+ },
+ .nr_leaves = 4 + 7 + 1,
+ },
+
+ /* Leaf 0xb serialisation stops at the first subleaf with type 0. */
+ {
+ .name = "empty leaf 0xb",
+ .p = {
+ .basic.max_leaf = 0xb,
+ },
+ .nr_leaves = 4 + 0xb,
+ },
+ {
+ .name = "partial leaf 0xb",
+ .p = {
+ .basic.max_leaf = 0xb,
+ .topo.subleaf[0].type = 1,
+ },
+ .nr_leaves = 4 + 0xb + 1,
+ },
+
+ /*
+ * Leaf 0xd serialisation automatically has two leaves, and stops the
+ * highest bit set in {xcr0,xss}_{high,low}.
+ */
+ {
+ .name = "empty leaf 0xd",
+ .p = {
+ .basic.max_leaf = 0xd,
+ },
+ .nr_leaves = 4 + 0xd + 1,
+ },
+ {
+ .name = "partial 0xd",
+ .p = {
+ .basic.max_leaf = 0xd,
+ .xstate.xcr0_low = 7,
+ },
+ .nr_leaves = 4 + 0xd + 1 + 1,
+ },
};
printf("Testing CPUID serialise success:\n");
diff --git a/xen/lib/x86/cpuid.c b/xen/lib/x86/cpuid.c
index 23619c7..dcab1e7 100644
--- a/xen/lib/x86/cpuid.c
+++ b/xen/lib/x86/cpuid.c
@@ -242,7 +242,12 @@ int x86_cpuid_copy_to_buffer(const struct cpuid_policy *p,
{
case 0x4:
for ( subleaf = 0; subleaf < ARRAY_SIZE(p->cache.raw); ++subleaf )
+ {
COPY_LEAF(leaf, subleaf, &p->cache.raw[subleaf]);
+
+ if ( p->cache.subleaf[subleaf].type == 0 )
+ break;
+ }
break;
case 0x7:
@@ -254,13 +259,30 @@ int x86_cpuid_copy_to_buffer(const struct cpuid_policy *p,
case 0xb:
for ( subleaf = 0; subleaf < ARRAY_SIZE(p->topo.raw); ++subleaf )
+ {
COPY_LEAF(leaf, subleaf, &p->topo.raw[subleaf]);
+
+ if ( p->topo.subleaf[subleaf].type == 0 )
+ break;
+ }
break;
case 0xd:
- for ( subleaf = 0; subleaf < ARRAY_SIZE(p->xstate.raw); ++subleaf )
+ {
+ uint64_t xstates;
+
+ COPY_LEAF(leaf, 0, &p->xstate.raw[0]);
+ COPY_LEAF(leaf, 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);
+
+ for ( xstates >>= 2, subleaf = 2;
+ xstates && subleaf < ARRAY_SIZE(p->xstate.raw);
+ xstates >>= 1, ++subleaf )
COPY_LEAF(leaf, subleaf, &p->xstate.raw[subleaf]);
break;
+ }
default:
COPY_LEAF(leaf, XEN_CPUID_NO_SUBLEAF, &p->basic.raw[leaf]);
--
2.1.4
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel
>>> On 22.05.19 at 17:50, <andrew.cooper3@citrix.com> wrote: > x86_cpuid_copy_to_buffer() currently serialises the full content of the > various subleaf unions. While leaves 4, 0xb and 0xd don't have a concrete > max_subleaf field, they do have well defined upper bounds. > > Diffing the results of `xen-cpuid -p` shows the resutling saving: > > @@ -1,5 +1,5 @@ > Xen reports there are maximum 114 leaves and 1 MSRs > -Raw policy: 93 leaves, 1 MSRs > +Raw policy: 38 leaves, 1 MSRs > CPUID: > leaf subleaf -> eax ebx ecx edx > 00000000:ffffffff -> 00000016:756e6547:6c65746e:49656e69 > @@ -32,7 +32,7 @@ Raw policy: 93 leaves, 1 MSRs > MSRs: > index -> value > 000000ce -> 0000000080000000 > -Host policy: 93 leaves, 1 MSRs > +Host policy: 33 leaves, 1 MSRs > CPUID: > leaf subleaf -> eax ebx ecx edx > 00000000:ffffffff -> 0000000d:756e6547:6c65746e:49656e69 > > which is mostly due to no longer writing out 64 leaves for xstate when (on > this CoffeeLake system) 8 will do. > > Extend the unit tests to cover empty and partially filled subleaf unions. > > Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com> For the lib/x86/ part Reviewed-by: Jan Beulich <jbeulich@suse.com> For the test harness part Acked-by: Jan Beulich <jbeulich@suse.com> No idea how else I should represent that I didn't look overly closely at the harness additions. Jan _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
On 23/05/2019 09:33, Jan Beulich wrote: >>>> On 22.05.19 at 17:50, <andrew.cooper3@citrix.com> wrote: >> x86_cpuid_copy_to_buffer() currently serialises the full content of the >> various subleaf unions. While leaves 4, 0xb and 0xd don't have a concrete >> max_subleaf field, they do have well defined upper bounds. >> >> Diffing the results of `xen-cpuid -p` shows the resutling saving: >> >> @@ -1,5 +1,5 @@ >> Xen reports there are maximum 114 leaves and 1 MSRs >> -Raw policy: 93 leaves, 1 MSRs >> +Raw policy: 38 leaves, 1 MSRs >> CPUID: >> leaf subleaf -> eax ebx ecx edx >> 00000000:ffffffff -> 00000016:756e6547:6c65746e:49656e69 >> @@ -32,7 +32,7 @@ Raw policy: 93 leaves, 1 MSRs >> MSRs: >> index -> value >> 000000ce -> 0000000080000000 >> -Host policy: 93 leaves, 1 MSRs >> +Host policy: 33 leaves, 1 MSRs >> CPUID: >> leaf subleaf -> eax ebx ecx edx >> 00000000:ffffffff -> 0000000d:756e6547:6c65746e:49656e69 >> >> which is mostly due to no longer writing out 64 leaves for xstate when (on >> this CoffeeLake system) 8 will do. >> >> Extend the unit tests to cover empty and partially filled subleaf unions. >> >> Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com> > For the lib/x86/ part > Reviewed-by: Jan Beulich <jbeulich@suse.com> Thanks. > For the test harness part > Acked-by: Jan Beulich <jbeulich@suse.com> > No idea how else I should represent that I didn't look overly closely > at the harness additions. Well - I can state that the additions to the test harness did find bugs. Overall, I think the content of tools/tests/ is of relatively little importance in the grand scheme of things. I certainly don't spend as much time reviewing the test_x86_emulator changes as the changes to x86_emulate() itself. ~Andrew _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
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(-)
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 )
--
2.1.4
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel
>>> 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 _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
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 _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
>>> 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 _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
© 2016 - 2024 Red Hat, Inc.