Skip to contentSkip to conversation

x86: Merge cpuid and msr policy

RFC PATCH Superseded Andrew Cooper 3 participants 34 messages Mar 29, 2023 – Apr 3, 2023

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

cover x86: Merge cpuid and msr policy

tl;dr to add MSR_ARCH_CAPS features sensibly, cpu_{featureset<->policy}() need to not operate on objects of differing lifetimes, so structs {cpuid,msr}_policy need merging and cpu_policy is the obvious name.

But this does mean that we now have

cpu_policy->basic.$X
cpu_policy->feat.$Y
cpu_policy->arch_caps.$Z

and plenty of code now reads

d->arch.cpu_policy->feat.$Y

instead of

d->arch.cpuid->feat.$Y

The latter can be half-fixed with some union magic (see patch 9 commit message). The former can be fixed by putting cpuid/msr infixes in cpu_policy, which is doable but very invasive, and would make plenty of code read

d->arch.cpu_policy->cpuid.feat.$Y

and the two obviously shouldn't be done together.

So, RFC. Does this code layout look ok? If we want to make changes with naming, now is very much the right time to get them sorted.

Patches 1-8 are pretty ready to go. Patch 9 is the remainder to take out the temporary hacks, and I'm still in the process of merging the system policy derivation.

Andrew Cooper (9):
x86: Rename struct cpu_policy to struct old_cpuid_policy
x86: Rename {domctl,sysctl}.cpu_policy.{cpuid,msr_policy} fields
x86: Rename struct cpuid_policy to struct cpu_policy
x86: Merge struct msr_policy into struct cpu_policy
x86: Merge the system {cpuid,msr} policy objects
x86: Merge a domain's {cpuid,msr} policy objects
x86: Merge xc_cpu_policy's cpuid and msr objects
x86: Drop struct old_cpu_policy
RFC: Everything else

 tools/fuzz/cpu-policy/afl-policy-fuzzer.c     |  15 +-
 .../fuzz/x86_instruction_emulator/fuzz-emul.c |   2 +-
 tools/libs/guest/xg_cpuid_x86.c               |  48 +-
 tools/libs/guest/xg_private.h                 |   5 +-
 tools/tests/cpu-policy/test-cpu-policy.c      |  50 +-
 tools/tests/tsx/test-tsx.c                    |  58 +-
 tools/tests/x86_emulator/Makefile             |   2 +-
 tools/tests/x86_emulator/test_x86_emulator.c  |   2 +-
 tools/tests/x86_emulator/x86-emulate.c        |   2 +-
 tools/tests/x86_emulator/x86-emulate.h        |   2 +-
 xen/arch/x86/Makefile                         |   1 +
 xen/arch/x86/cpu-policy.c                     |  67 +++
 xen/arch/x86/cpu/common.c                     |   4 +-
 xen/arch/x86/cpu/mcheck/mce_intel.c           |   2 +-
 xen/arch/x86/cpu/vpmu_intel.c                 |   4 +-
 xen/arch/x86/cpuid.c                          | 101 ++--
 xen/arch/x86/domain.c                         |  18 +-
 xen/arch/x86/domctl.c                         |  51 +-
 xen/arch/x86/hvm/emulate.c                    |   2 +-
 xen/arch/x86/hvm/hvm.c                        |  38 +-
 xen/arch/x86/hvm/ioreq.c                      |   4 +-
 xen/arch/x86/hvm/mtrr.c                       |   2 +-
 xen/arch/x86/hvm/svm/svm.c                    |  18 +-
 xen/arch/x86/hvm/svm/svmdebug.c               |   2 +-
 xen/arch/x86/hvm/vlapic.c                     |   2 +-
 xen/arch/x86/hvm/vmx/vmx.c                    |  12 +-
 xen/arch/x86/hvm/vmx/vvmx.c                   |   2 +-
 xen/arch/x86/include/asm/cpu-policy.h         |  18 +
 xen/arch/x86/include/asm/cpuid.h              |  10 -
 xen/arch/x86/include/asm/domain.h             |   4 +-
 xen/arch/x86/include/asm/guest_pt.h           |   4 +-
 xen/arch/x86/include/asm/msr.h                |  13 +-
 xen/arch/x86/include/asm/paging.h             |   2 +-
 xen/arch/x86/mm/mem_sharing.c                 |   3 +-
 xen/arch/x86/mm/shadow/hvm.c                  |   2 +-
 xen/arch/x86/msr.c                            |  98 +---
 xen/arch/x86/pv/domain.c                      |   2 +-
 xen/arch/x86/pv/emul-priv-op.c                |   6 +-
 xen/arch/x86/pv/ro-page-fault.c               |   2 +-
 xen/arch/x86/sysctl.c                         |  77 +--
 xen/arch/x86/traps.c                          |   2 +-
 xen/arch/x86/x86_emulate.c                    |   2 +-
 xen/arch/x86/x86_emulate/x86_emulate.c        | 166 +++---
 xen/arch/x86/x86_emulate/x86_emulate.h        |   6 +-
 xen/arch/x86/xstate.c                         |   4 +-
 xen/include/public/domctl.h                   |   4 +-
 xen/include/public/sysctl.h                   |   4 +-
 xen/include/xen/lib/x86/cpu-policy.h          | 540 +++++++++++++++++-
 xen/include/xen/lib/x86/cpuid.h               | 475 ---------------
 xen/include/xen/lib/x86/msr.h                 | 104 ----
 xen/lib/x86/cpuid.c                           |  12 +-
 xen/lib/x86/msr.c                             |   6 +-
 xen/lib/x86/policy.c                          |   8 +-
 53 files changed, 986 insertions(+), 1104 deletions(-)
 create mode 100644 xen/arch/x86/cpu-policy.c
 create mode 100644 xen/arch/x86/include/asm/cpu-policy.h
 delete mode 100644 xen/include/xen/lib/x86/cpuid.h
 delete mode 100644 xen/include/xen/lib/x86/msr.h
Signature · 1 line
2.30.2
#1 lore raw

PATCH 1/9 x86: Rename struct cpu_policy to struct old_cpuid_policy

We want to merge struct cpuid_policy and struct msr_policy together, and the result wants to be called struct cpu_policy.

The current struct cpu_policy, being a pair of pointers, isn't terribly useful. Rename the type to struct old_cpu_policy, but it will disappear entirely once the merge is complete.

No functional change.

---
CC: Jan Beulich <[email protected]>
CC: Roger Pau Monné <[email protected]>
CC: Wei Liu <[email protected]>

---
 tools/libs/guest/xg_cpuid_x86.c          | 4 ++--
 tools/tests/cpu-policy/test-cpu-policy.c | 4 ++--
 xen/arch/x86/domctl.c                    | 4 ++--
 xen/arch/x86/include/asm/cpuid.h         | 2 +-
 xen/arch/x86/sysctl.c                    | 4 ++--
 xen/include/xen/lib/x86/cpu-policy.h     | 6 +++---
 xen/lib/x86/policy.c                     | 4 ++--
 7 files changed, 14 insertions(+), 14 deletions(-)
tools/libs/guest/xg_cpuid_x86.c +2 -2
index 4542878bbe88..1b02bc987af7 100644
@@ -868,8 +868,8 @@ bool xc_cpu_policy_is_compatible(xc_interface *xch, xc_cpu_policy_t *host,
                                  xc_cpu_policy_t *guest)
 {
     struct cpu_policy_errors err = INIT_CPU_POLICY_ERRORS;
-    struct cpu_policy h = { &host->cpuid, &host->msr };
-    struct cpu_policy g = { &guest->cpuid, &guest->msr };
+    struct old_cpu_policy h = { &host->cpuid, &host->msr };
+    struct old_cpu_policy g = { &guest->cpuid, &guest->msr };
     int rc = x86_cpu_policies_are_compatible(&h, &g, &err);
 
     if ( !rc )
tools/tests/cpu-policy/test-cpu-policy.c +2 -2
index d3f24fd6d274..909d6272f875 100644
@@ -602,7 +602,7 @@ static void test_is_compatible_success(void)
     for ( size_t i = 0; i < ARRAY_SIZE(tests); ++i )
     {
         struct test *t = &tests[i];
-        struct cpu_policy sys = {
+        struct old_cpu_policy sys = {
             &t->host_cpuid,
             &t->host_msr,
         }, new = {
@@ -654,7 +654,7 @@ static void test_is_compatible_failure(void)
     for ( size_t i = 0; i < ARRAY_SIZE(tests); ++i )
     {
         struct test *t = &tests[i];
-        struct cpu_policy sys = {
+        struct old_cpu_policy sys = {
             &t->host_cpuid,
             &t->host_msr,
         }, new = {
xen/arch/x86/domctl.c +2 -2
index 2118fcad5dfe..0b41b279507e 100644
@@ -40,8 +40,8 @@
 static int update_domain_cpu_policy(struct domain *d,
                                     xen_domctl_cpu_policy_t *xdpc)
 {
-    struct cpu_policy new = {};
-    const struct cpu_policy *sys = is_pv_domain(d)
+    struct old_cpu_policy new = {};
+    const struct old_cpu_policy *sys = is_pv_domain(d)
         ? &system_policies[XEN_SYSCTL_cpu_policy_pv_max]
         : &system_policies[XEN_SYSCTL_cpu_policy_hvm_max];
     struct cpu_policy_errors err = INIT_CPU_POLICY_ERRORS;
xen/arch/x86/include/asm/cpuid.h +1 -1
index 9c3637549a10..49b3128f06f9 100644
@@ -51,7 +51,7 @@ extern struct cpuid_policy raw_cpuid_policy, host_cpuid_policy,
     pv_max_cpuid_policy, pv_def_cpuid_policy,
     hvm_max_cpuid_policy, hvm_def_cpuid_policy;
 
-extern const struct cpu_policy system_policies[];
+extern const struct old_cpu_policy system_policies[];
 
 /* Check that all previously present features are still available. */
 bool recheck_cpu_features(unsigned int cpu);
xen/arch/x86/sysctl.c +2 -2
index 16625b57f01f..3f5b092df16a 100644
@@ -32,7 +32,7 @@
 #include <asm/psr.h>
 #include <asm/cpuid.h>
 
-const struct cpu_policy system_policies[6] = {
+const struct old_cpu_policy system_policies[6] = {
     [ XEN_SYSCTL_cpu_policy_raw ] = {
         &raw_cpuid_policy,
         &raw_msr_policy,
@@ -391,7 +391,7 @@ long arch_do_sysctl(
 
     case XEN_SYSCTL_get_cpu_policy:
     {
-        const struct cpu_policy *policy;
+        const struct old_cpu_policy *policy;
 
         /* Reserved field set, or bad policy index? */
         if ( sysctl->u.cpu_policy._rsvd ||
xen/include/xen/lib/x86/cpu-policy.h +3 -3
index 5a2c4c7b2d90..3a5300d1078c 100644
@@ -5,7 +5,7 @@
 #include <xen/lib/x86/cpuid.h>
 #include <xen/lib/x86/msr.h>
 
-struct cpu_policy
+struct old_cpu_policy
 {
     struct cpuid_policy *cpuid;
     struct msr_policy *msr;
@@ -33,8 +33,8 @@ struct cpu_policy_errors
  * incompatibility is detected, the optional err pointer may identify the
  * problematic leaf/subleaf and/or MSR.
  */
-int x86_cpu_policies_are_compatible(const struct cpu_policy *host,
-                                    const struct cpu_policy *guest,
+int x86_cpu_policies_are_compatible(const struct old_cpu_policy *host,
+                                    const struct old_cpu_policy *guest,
                                     struct cpu_policy_errors *err);
 
 #endif /* !XEN_LIB_X86_POLICIES_H */
xen/lib/x86/policy.c +2 -2
index f6cea4e2f9bd..2975711d7c6c 100644
@@ -2,8 +2,8 @@
 
 #include <xen/lib/x86/cpu-policy.h>
 
-int x86_cpu_policies_are_compatible(const struct cpu_policy *host,
-                                    const struct cpu_policy *guest,
+int x86_cpu_policies_are_compatible(const struct old_cpu_policy *host,
+                                    const struct old_cpu_policy *guest,
                                     struct cpu_policy_errors *err)
 {
     struct cpu_policy_errors e = INIT_CPU_POLICY_ERRORS;
Signature · 1 line
2.30.2
#2 lore raw

PATCH 2/9 x86: Rename {domctl,sysctl}.cpu_policy.{cpuid,msr_policy} fields

In reply to Andrew Cooper

These weren't great names to begin with, and using {leaves,msrs} matches up better with the existing nr_{leaves,msr} parameters anyway.

Furthermore,by renaming these fields we can get away with using some #define trickary to avoid the struct {cpuid,msr}_policy merge needing to happen in a single changeset.

No functional change.

---
CC: Jan Beulich <[email protected]>
CC: Roger Pau Monné <[email protected]>
CC: Wei Liu <[email protected]>

---
 tools/libs/guest/xg_cpuid_x86.c | 12 ++++++------
 xen/arch/x86/domctl.c           | 12 ++++++------
 xen/arch/x86/sysctl.c           |  8 ++++----
 xen/include/public/domctl.h     |  4 ++--
 xen/include/public/sysctl.h     |  4 ++--
 5 files changed, 20 insertions(+), 20 deletions(-)
tools/libs/guest/xg_cpuid_x86.c +6 -6
index 1b02bc987af7..5fae06e77804 100644
@@ -145,9 +145,9 @@ static int get_system_cpu_policy(xc_interface *xch, uint32_t index,
     sysctl.cmd = XEN_SYSCTL_get_cpu_policy;
     sysctl.u.cpu_policy.index = index;
     sysctl.u.cpu_policy.nr_leaves = *nr_leaves;
-    set_xen_guest_handle(sysctl.u.cpu_policy.cpuid_policy, leaves);
+    set_xen_guest_handle(sysctl.u.cpu_policy.leaves, leaves);
     sysctl.u.cpu_policy.nr_msrs = *nr_msrs;
-    set_xen_guest_handle(sysctl.u.cpu_policy.msr_policy, msrs);
+    set_xen_guest_handle(sysctl.u.cpu_policy.msrs, msrs);
 
     ret = do_sysctl(xch, &sysctl);
 
@@ -183,9 +183,9 @@ static int get_domain_cpu_policy(xc_interface *xch, uint32_t domid,
     domctl.cmd = XEN_DOMCTL_get_cpu_policy;
     domctl.domain = domid;
     domctl.u.cpu_policy.nr_leaves = *nr_leaves;
-    set_xen_guest_handle(domctl.u.cpu_policy.cpuid_policy, leaves);
+    set_xen_guest_handle(domctl.u.cpu_policy.leaves, leaves);
     domctl.u.cpu_policy.nr_msrs = *nr_msrs;
-    set_xen_guest_handle(domctl.u.cpu_policy.msr_policy, msrs);
+    set_xen_guest_handle(domctl.u.cpu_policy.msrs, msrs);
 
     ret = do_domctl(xch, &domctl);
 
@@ -232,9 +232,9 @@ int xc_set_domain_cpu_policy(xc_interface *xch, uint32_t domid,
     domctl.cmd = XEN_DOMCTL_set_cpu_policy;
     domctl.domain = domid;
     domctl.u.cpu_policy.nr_leaves = nr_leaves;
-    set_xen_guest_handle(domctl.u.cpu_policy.cpuid_policy, leaves);
+    set_xen_guest_handle(domctl.u.cpu_policy.leaves, leaves);
     domctl.u.cpu_policy.nr_msrs = nr_msrs;
-    set_xen_guest_handle(domctl.u.cpu_policy.msr_policy, msrs);
+    set_xen_guest_handle(domctl.u.cpu_policy.msrs, msrs);
     domctl.u.cpu_policy.err_leaf = -1;
     domctl.u.cpu_policy.err_subleaf = -1;
     domctl.u.cpu_policy.err_msr = -1;
xen/arch/x86/domctl.c +6 -6
index 0b41b279507e..944af63e68d0 100644
@@ -54,10 +54,10 @@ static int update_domain_cpu_policy(struct domain *d,
 
     /* Merge the toolstack provided data. */
     if ( (ret = x86_cpuid_copy_from_buffer(
-              new.cpuid, xdpc->cpuid_policy, xdpc->nr_leaves,
+              new.cpuid, xdpc->leaves, xdpc->nr_leaves,
               &err.leaf, &err.subleaf)) ||
          (ret = x86_msr_copy_from_buffer(
-              new.msr, xdpc->msr_policy, xdpc->nr_msrs, &err.msr)) )
+              new.msr, xdpc->msrs, xdpc->nr_msrs, &err.msr)) )
         goto out;
 
     /* Trim any newly-stale out-of-range leaves. */
@@ -1317,20 +1317,20 @@ long arch_do_domctl(
 
     case XEN_DOMCTL_get_cpu_policy:
         /* Process the CPUID leaves. */
-        if ( guest_handle_is_null(domctl->u.cpu_policy.cpuid_policy) )
+        if ( guest_handle_is_null(domctl->u.cpu_policy.leaves) )
             domctl->u.cpu_policy.nr_leaves = CPUID_MAX_SERIALISED_LEAVES;
         else if ( (ret = x86_cpuid_copy_to_buffer(
                        d->arch.cpuid,
-                       domctl->u.cpu_policy.cpuid_policy,
+                       domctl->u.cpu_policy.leaves,
                        &domctl->u.cpu_policy.nr_leaves)) )
             break;
 
         /* Process the MSR entries. */
-        if ( guest_handle_is_null(domctl->u.cpu_policy.msr_policy) )
+        if ( guest_handle_is_null(domctl->u.cpu_policy.msrs) )
             domctl->u.cpu_policy.nr_msrs = MSR_MAX_SERIALISED_ENTRIES;
         else if ( (ret = x86_msr_copy_to_buffer(
                        d->arch.msr,
-                       domctl->u.cpu_policy.msr_policy,
+                       domctl->u.cpu_policy.msrs,
                        &domctl->u.cpu_policy.nr_msrs)) )
             break;
 
xen/arch/x86/sysctl.c +4 -4
index 3f5b092df16a..3ed7c69f4315 100644
@@ -411,11 +411,11 @@ long arch_do_sysctl(
         }
 
         /* Process the CPUID leaves. */
-        if ( guest_handle_is_null(sysctl->u.cpu_policy.cpuid_policy) )
+        if ( guest_handle_is_null(sysctl->u.cpu_policy.leaves) )
             sysctl->u.cpu_policy.nr_leaves = CPUID_MAX_SERIALISED_LEAVES;
         else if ( (ret = x86_cpuid_copy_to_buffer(
                        policy->cpuid,
-                       sysctl->u.cpu_policy.cpuid_policy,
+                       sysctl->u.cpu_policy.leaves,
                        &sysctl->u.cpu_policy.nr_leaves)) )
             break;
 
@@ -427,11 +427,11 @@ long arch_do_sysctl(
         }
 
         /* Process the MSR entries. */
-        if ( guest_handle_is_null(sysctl->u.cpu_policy.msr_policy) )
+        if ( guest_handle_is_null(sysctl->u.cpu_policy.msrs) )
             sysctl->u.cpu_policy.nr_msrs = MSR_MAX_SERIALISED_ENTRIES;
         else if ( (ret = x86_msr_copy_to_buffer(
                        policy->msr,
-                       sysctl->u.cpu_policy.msr_policy,
+                       sysctl->u.cpu_policy.msrs,
                        &sysctl->u.cpu_policy.nr_msrs)) )
             break;
 
xen/include/public/domctl.h +2 -2
index 7280e9f96816..529801c89ba3 100644
@@ -683,8 +683,8 @@ struct xen_domctl_cpu_policy {
                          * 'cpuid_policy'. */
     uint32_t nr_msrs;   /* IN/OUT: Number of MSRs in/written to
                          * 'msr_policy' */
-    XEN_GUEST_HANDLE_64(xen_cpuid_leaf_t) cpuid_policy; /* IN/OUT */
-    XEN_GUEST_HANDLE_64(xen_msr_entry_t) msr_policy;    /* IN/OUT */
+    XEN_GUEST_HANDLE_64(xen_cpuid_leaf_t) leaves; /* IN/OUT */
+    XEN_GUEST_HANDLE_64(xen_msr_entry_t)  msrs;   /* IN/OUT */
 
     /*
      * OUT, set_policy only.  Written in some (but not all) error cases to
xen/include/public/sysctl.h +2 -2
index e8dded9fb94a..2b24d6bfd00e 100644
@@ -1050,8 +1050,8 @@ struct xen_sysctl_cpu_policy {
                            * 'msr_policy', or the maximum number of MSRs if
                            * the guest handle is NULL. */
     uint32_t _rsvd;       /* Must be zero. */
-    XEN_GUEST_HANDLE_64(xen_cpuid_leaf_t) cpuid_policy; /* OUT */
-    XEN_GUEST_HANDLE_64(xen_msr_entry_t) msr_policy;    /* OUT */
+    XEN_GUEST_HANDLE_64(xen_cpuid_leaf_t) leaves; /* OUT */
+    XEN_GUEST_HANDLE_64(xen_msr_entry_t)  msrs;   /* OUT */
 };
 typedef struct xen_sysctl_cpu_policy xen_sysctl_cpu_policy_t;
 DEFINE_XEN_GUEST_HANDLE(xen_sysctl_cpu_policy_t);
Signature · 1 line
2.30.2
#3 lore raw

PATCH 3/9 x86: Rename struct cpuid_policy to struct cpu_policy

In reply to Andrew Cooper

Also merge lib/x86/cpuid.h entirely into lib/x86/cpu-policy.h

Use a temporary define to make struct cpuid_policy still work.

There's one forward declaration of struct cpuid_policy in tools/tests/x86_emulator/x86-emulate.h that isn't covered by the define, and it's easier to rename that now than to rearrange the includes.

No functional change.

---
CC: Jan Beulich <[email protected]>
CC: Roger Pau Monné <[email protected]>
CC: Wei Liu <[email protected]>

---
 tools/fuzz/cpu-policy/afl-policy-fuzzer.c |   2 +-
 tools/tests/x86_emulator/Makefile         |   2 +-
 tools/tests/x86_emulator/x86-emulate.h    |   2 +-
 xen/arch/x86/include/asm/cpuid.h          |   1 -
 xen/arch/x86/x86_emulate/x86_emulate.h    |   2 +-
 xen/include/xen/lib/x86/cpu-policy.h      | 463 ++++++++++++++++++++-
 xen/include/xen/lib/x86/cpuid.h           | 475 ----------------------
 xen/lib/x86/cpuid.c                       |   2 +-
 8 files changed, 467 insertions(+), 482 deletions(-)
 delete mode 100644 xen/include/xen/lib/x86/cpuid.h
tools/fuzz/cpu-policy/afl-policy-fuzzer.c +1 -1
index 7d0f274c6cdd..79e42e8bfd04 100644
@@ -9,7 +9,7 @@
 #include <getopt.h>
 
 #include <xen-tools/common-macros.h>
-#include <xen/lib/x86/cpuid.h>
+#include <xen/lib/x86/cpu-policy.h>
 #include <xen/lib/x86/msr.h>
 #include <xen/domctl.h>
 
tools/tests/x86_emulator/Makefile +1 -1
index 7b07c31bbde4..2f7c58040004 100644
@@ -286,7 +286,7 @@ HOSTCFLAGS += $(CFLAGS_xeninclude) -I. $(HOSTCFLAGS-$(XEN_COMPILE_ARCH))
 x86.h := $(addprefix $(XEN_ROOT)/tools/include/xen/asm/,\
                      x86-vendors.h x86-defns.h msr-index.h) \
          $(addprefix $(XEN_ROOT)/tools/include/xen/lib/x86/, \
-                     cpuid.h cpuid-autogen.h)
+                     cpu-policy.h)
 x86_emulate.h := x86-emulate.h x86_emulate/x86_emulate.h $(x86.h)
 
 x86-emulate.o cpuid.o test_x86_emulator.o evex-disp8.o predicates.o wrappers.o: %.o: %.c $(x86_emulate.h)
tools/tests/x86_emulator/x86-emulate.h +1 -1
index 1af986f78d16..19544b0b3a8a 100644
@@ -65,7 +65,7 @@
 #define is_canonical_address(x) (((int64_t)(x) >> 47) == ((int64_t)(x) >> 63))
 
 extern uint32_t mxcsr_mask;
-extern struct cpuid_policy cp;
+extern struct cpu_policy cp;
 
 #define MMAP_SZ 16384
 bool emul_test_init(void);
xen/arch/x86/include/asm/cpuid.h +0 -1
index 49b3128f06f9..d418e8100dde 100644
@@ -9,7 +9,6 @@
 #include <xen/percpu.h>
 
 #include <xen/lib/x86/cpu-policy.h>
-#include <xen/lib/x86/cpuid.h>
 
 #include <public/sysctl.h>
 
xen/arch/x86/x86_emulate/x86_emulate.h +1 -1
index bb7af967ffee..75015104fbdb 100644
@@ -23,7 +23,7 @@
 #ifndef __X86_EMULATE_H__
 #define __X86_EMULATE_H__
 
-#include <xen/lib/x86/cpuid.h>
+#include <xen/lib/x86/cpu-policy.h>
 
 #define MAX_INST_LEN 15
 
xen/include/xen/lib/x86/cpu-policy.h +462 -1

477 lines not on this page · raw message

xen/include/xen/lib/x86/cpuid.h deleted +0 -475

475 lines not on this page · raw message

xen/lib/x86/cpuid.c +1 -1
index 8eb88314f53c..e81f76c779c0 100644
@@ -1,6 +1,6 @@
 #include "private.h"
 
-#include <xen/lib/x86/cpuid.h>
+#include <xen/lib/x86/cpu-policy.h>
 
 static void zero_leaves(struct cpuid_leaf *l,
                         unsigned int first, unsigned int last)
Signature · 1 line
2.30.2
#4 lore raw

PATCH 4/9 x86: Merge struct msr_policy into struct cpu_policy

In reply to Andrew Cooper

As with the cpuid side, use a temporary define to make struct msr_policy still work.

Note, this means that domains now have two separate struct cpu_policy allocations with disjoint information, and system policies are in a similar position. Both will be deduplicated in the following patches.

---
CC: Jan Beulich <[email protected]>
CC: Roger Pau Monné <[email protected]>
CC: Wei Liu <[email protected]>

---
 tools/fuzz/cpu-policy/afl-policy-fuzzer.c |   1 -
 xen/arch/x86/include/asm/msr.h            |   3 +-
 xen/include/xen/lib/x86/cpu-policy.h      |  81 ++++++++++++++++-
 xen/include/xen/lib/x86/msr.h             | 104 ----------------------
 xen/lib/x86/msr.c                         |   2 +-
 5 files changed, 83 insertions(+), 108 deletions(-)
 delete mode 100644 xen/include/xen/lib/x86/msr.h
tools/fuzz/cpu-policy/afl-policy-fuzzer.c +0 -1
index 79e42e8bfd04..0ce3d8e16626 100644
@@ -10,7 +10,6 @@
 
 #include <xen-tools/common-macros.h>
 #include <xen/lib/x86/cpu-policy.h>
-#include <xen/lib/x86/msr.h>
 #include <xen/domctl.h>
 
 static bool debug;
xen/arch/x86/include/asm/msr.h +2 -1
index 7946b6b24c11..02eddd919c27 100644
@@ -6,8 +6,9 @@
 #include <xen/types.h>
 #include <xen/percpu.h>
 #include <xen/errno.h>
+#include <xen/kernel.h>
 
-#include <xen/lib/x86/msr.h>
+#include <xen/lib/x86/cpu-policy.h>
 
 #include <asm/asm_defns.h>
 #include <asm/cpufeature.h>
xen/include/xen/lib/x86/cpu-policy.h +80 -1

120 lines not on this page · raw message

xen/include/xen/lib/x86/msr.h deleted +0 -104

104 lines not on this page · raw message

xen/lib/x86/msr.c +1 -1
index 7d71e92a380a..c4d885e7b568 100644
@@ -1,6 +1,6 @@
 #include "private.h"
 
-#include <xen/lib/x86/msr.h>
+#include <xen/lib/x86/cpu-policy.h>
 
 /*
  * Copy a single MSR into the provided msr_entry_buffer_t buffer, performing a
Signature · 1 line
2.30.2
#5 lore raw

Re: [PATCH 4/9] x86: Merge struct msr_policy into struct cpu_policy

Quoted 11 lines · Andrew Cooper
xen/include/xen/lib/x86/cpu-policy.h +0 -1
@@ -3,7 +3,6 @@
 #define XEN_LIB_X86_POLICIES_H
 
 #include <xen/lib/x86/cpuid-autogen.h>
-#include <xen/lib/x86/msr.h>
 
 #define FEATURESET_1d     0 /* 0x00000001.edx      */
 #define FEATURESET_1c     1 /* 0x00000001.ecx      */

xen/include/xen/lib/x86/cpu-policy.h const char *x86_cpuid_vendor_to_str(unsigned int vendor); · line 110

Quoted 5 lines
xen/include/xen/lib/x86/cpu-policy.h +2 -0
@@ -107,6 +106,9 @@ const char *x86_cpuid_vendor_to_str(unsigned int vendor);
      CPUID_GUEST_NR_XSTATE - !!CPUID_GUEST_NR_XSTATE +  \
      CPUID_GUEST_NR_EXTD + 2 /* hv_limit and hv2_limit */ )
 
+/* Maximum number of MSRs written when serialising msr_policy. */
+#define MSR_MAX_SERIALISED_ENTRIES 2

The comment better wouldn't refer to msr_policy anymore, I think. I also wonder whether the comment wouldn't better move ...

xen/include/xen/lib/x86/cpu-policy.h struct cpu_policy · line 330

Quoted 6 lines
xen/include/xen/lib/x86/cpu-policy.h +2 -0
@@ -324,6 +326,44 @@ struct cpu_policy
         };
     } extd;
 
+    /*
+     * 0x000000ce - MSR_INTEL_PLATFORM_INFO

... e.g. above here, to increase the chance of it being spotted that it needs updating if another MSR is added here.

Jan

#6 lore raw

Re: [PATCH 4/9] x86: Merge struct msr_policy into struct cpu_policy

Quoted 15 lines · Jan Beulich
Quoted 14 lines · Andrew Cooper
xen/include/xen/lib/x86/cpu-policy.h +0 -1
@@ -3,7 +3,6 @@
 #define XEN_LIB_X86_POLICIES_H
 
 #include <xen/lib/x86/cpuid-autogen.h>
-#include <xen/lib/x86/msr.h>
 
 #define FEATURESET_1d     0 /* 0x00000001.edx      */
 #define FEATURESET_1c     1 /* 0x00000001.ecx      */
@@ -107,6 +106,9 @@ const char *x86_cpuid_vendor_to_str(unsigned int vendor);
      CPUID_GUEST_NR_XSTATE - !!CPUID_GUEST_NR_XSTATE +  \
      CPUID_GUEST_NR_EXTD + 2 /* hv_limit and hv2_limit */ )
 
Quoted 3 lines

xen/include/xen/lib/x86/cpu-policy.h const char *x86_cpuid_vendor_to_str(unsigned int vendor); · line 110

Quoted 2 lines
xen/include/xen/lib/x86/cpu-policy.h +2 -0
+/* Maximum number of MSRs written when serialising msr_policy. */
+#define MSR_MAX_SERIALISED_ENTRIES 2

The comment better wouldn't refer to msr_policy anymore, I think.

Ah yes.  (There's so much comment and library cleanup still to do)

Quoted 11 lines

I also
wonder whether the comment wouldn't better move ...

xen/include/xen/lib/x86/cpu-policy.h struct cpu_policy · line 330

Quoted 6 lines
xen/include/xen/lib/x86/cpu-policy.h +2 -0
@@ -324,6 +326,44 @@ struct cpu_policy
         };
     } extd;
 
+    /*
+     * 0x000000ce - MSR_INTEL_PLATFORM_INFO

... e.g. above here, to increase the chance of it being spotted that it needs updating if another MSR is added here.

I'm not sure about that.  In its current position, it's next to it's CPUID partner.

The unit tests in test-cpu-policy cross-check that we never get -ENOBUFS for a MSR_MAX_* sized destination, so I'm not worried about it actually getting stale.

But, with the new merged policies, I need to change the serialising behaviour anyway.

Right now we serialise everything unconditionally because that's the only way that merging incremental deltas could be made to work, but now the MSR enumeration bits are in the same structure we can use those instead, along with the various "clear" passes we already have.

~Andrew

#7 lore raw

Re: [PATCH 4/9] x86: Merge struct msr_policy into struct cpu_policy

Quoted 34 lines · Andrew Cooper
Quoted 15 lines · Jan Beulich
Quoted 14 lines · Andrew Cooper
xen/include/xen/lib/x86/cpu-policy.h +0 -1
@@ -3,7 +3,6 @@
 #define XEN_LIB_X86_POLICIES_H
 
 #include <xen/lib/x86/cpuid-autogen.h>
-#include <xen/lib/x86/msr.h>
 
 #define FEATURESET_1d     0 /* 0x00000001.edx      */
 #define FEATURESET_1c     1 /* 0x00000001.ecx      */
@@ -107,6 +106,9 @@ const char *x86_cpuid_vendor_to_str(unsigned int vendor);
      CPUID_GUEST_NR_XSTATE - !!CPUID_GUEST_NR_XSTATE +  \
      CPUID_GUEST_NR_EXTD + 2 /* hv_limit and hv2_limit */ )
 
Quoted 3 lines

xen/include/xen/lib/x86/cpu-policy.h const char *x86_cpuid_vendor_to_str(unsigned int vendor); · line 110

Quoted 2 lines
xen/include/xen/lib/x86/cpu-policy.h +2 -0
+/* Maximum number of MSRs written when serialising msr_policy. */
+#define MSR_MAX_SERIALISED_ENTRIES 2

The comment better wouldn't refer to msr_policy anymore, I think.

Ah yes.  (There's so much comment and library cleanup still to do)

Quoted 11 lines

I also
wonder whether the comment wouldn't better move ...

xen/include/xen/lib/x86/cpu-policy.h struct cpu_policy · line 330

Quoted 6 lines
xen/include/xen/lib/x86/cpu-policy.h +2 -0
@@ -324,6 +326,44 @@ struct cpu_policy
         };
     } extd;
 
+    /*
+     * 0x000000ce - MSR_INTEL_PLATFORM_INFO

... e.g. above here, to increase the chance of it being spotted that it needs updating if another MSR is added here.

Quoted 6 lines

I'm not sure about that.  In its current position, it's next to it's CPUID partner.

The unit tests in test-cpu-policy cross-check that we never get -ENOBUFS for a MSR_MAX_* sized destination, so I'm not worried about it actually getting stale.

Well, I wasn't worried so much about ending up with a bad commit, but rather that something you could spot on the first pass while writing new code might cause you a build or test failure later on, resulting in extra hunting. But this was merely a thought, nothing I would insist one (the more that your CPUID sibling argument is of course also relevant).

Jan

#8 lore raw

PATCH 5/9 x86: Merge the system {cpuid,msr} policy objects

In reply to Andrew Cooper

Right now, they're the same underlying type, containing disjoint information.

Introduce a new cpu-policy.{h,c} to be the new location for all policy handling logic. Place the combined objects in __ro_after_init, which has appeared since the original logic was written.

As we're trying to phase out the use of struct old_cpu_policy entirely, rework update_domain_cpu_policy() to not pointer-chase through system_policies[].

This in turn allows system_policies[] in sysctl.c to become static and reduced in scope to XEN_SYSCTL_get_cpu_policy.

No practical change, but we do half the amount of compile time space required for the system policies, which saves 6x almost-2k at the moment.

---
CC: Jan Beulich <[email protected]>
CC: Roger Pau Monné <[email protected]>
CC: Wei Liu <[email protected]>

---
 xen/arch/x86/Makefile                 |  1 +
 xen/arch/x86/cpu-policy.c             | 18 +++++++
 xen/arch/x86/cpu/common.c             |  4 +-
 xen/arch/x86/cpuid.c                  | 66 +++++++++++---------------
 xen/arch/x86/domctl.c                 | 17 +++++--
 xen/arch/x86/include/asm/cpu-policy.h | 14 ++++++
 xen/arch/x86/include/asm/cpuid.h      |  6 ---
 xen/arch/x86/include/asm/msr.h        |  7 ---
 xen/arch/x86/msr.c                    | 38 ++++++---------
 xen/arch/x86/sysctl.c                 | 67 ++++++++++-----------------
 10 files changed, 114 insertions(+), 124 deletions(-)
 create mode 100644 xen/arch/x86/cpu-policy.c
 create mode 100644 xen/arch/x86/include/asm/cpu-policy.h
xen/arch/x86/Makefile +1 -0
index 5accbe4c6746..f213a6b56a4d 100644
@@ -18,6 +18,7 @@ obj-y += bitops.o
 obj-bin-y += bzimage.init.o
 obj-bin-y += clear_page.o
 obj-bin-y += copy_page.o
+obj-y += cpu-policy.o
 obj-y += cpuid.o
 obj-$(CONFIG_PV) += compat.o
 obj-$(CONFIG_PV32) += x86_64/compat.o
xen/arch/x86/cpu-policy.c new file +18 -0
new file mode 100644
index 000000000000..663e9a084c53
@@ -0,0 +1,18 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+#include <xen/cache.h>
+#include <xen/kernel.h>
+
+#include <xen/lib/x86/cpu-policy.h>
+
+#include <asm/cpu-policy.h>
+
+struct cpu_policy __ro_after_init     raw_cpu_policy;
+struct cpu_policy __ro_after_init    host_cpu_policy;
+#ifdef CONFIG_PV
+struct cpu_policy __ro_after_init  pv_max_cpu_policy;
+struct cpu_policy __ro_after_init  pv_def_cpu_policy;
+#endif
+#ifdef CONFIG_HVM
+struct cpu_policy __ro_after_init hvm_max_cpu_policy;
+struct cpu_policy __ro_after_init hvm_def_cpu_policy;
+#endif
xen/arch/x86/cpu/common.c +3 -1
index 5ad347534a22..f11dcda57a69 100644
@@ -3,6 +3,8 @@
 #include <xen/delay.h>
 #include <xen/param.h>
 #include <xen/smp.h>
+
+#include <asm/cpu-policy.h>
 #include <asm/current.h>
 #include <asm/debugreg.h>
 #include <asm/processor.h>
@@ -141,7 +143,7 @@ bool __init probe_cpuid_faulting(void)
 		return false;
 
 	if ((rc = rdmsr_safe(MSR_INTEL_PLATFORM_INFO, val)) == 0)
-		raw_msr_policy.platform_info.cpuid_faulting =
+		raw_cpu_policy.platform_info.cpuid_faulting =
 			val & MSR_PLATFORM_INFO_CPUID_FAULTING;
 
 	if (rc ||
xen/arch/x86/cpuid.c +28 -38

171 lines not on this page · raw message

xen/arch/x86/domctl.c +12 -5
index 944af63e68d0..5800bb10bc4a 100644
@@ -35,18 +35,25 @@
 #include <asm/mem_sharing.h>
 #include <asm/xstate.h>
 #include <asm/psr.h>
-#include <asm/cpuid.h>
+#include <asm/cpu-policy.h>
 
 static int update_domain_cpu_policy(struct domain *d,
                                     xen_domctl_cpu_policy_t *xdpc)
 {
     struct old_cpu_policy new = {};
-    const struct old_cpu_policy *sys = is_pv_domain(d)
-        ? &system_policies[XEN_SYSCTL_cpu_policy_pv_max]
-        : &system_policies[XEN_SYSCTL_cpu_policy_hvm_max];
+    struct cpu_policy *sys = is_pv_domain(d)
+        ? (IS_ENABLED(CONFIG_PV)  ?  &pv_max_cpu_policy : NULL)
+        : (IS_ENABLED(CONFIG_HVM) ? &hvm_max_cpu_policy : NULL);
+    struct old_cpu_policy old_sys = { sys, sys };
     struct cpu_policy_errors err = INIT_CPU_POLICY_ERRORS;
     int ret = -ENOMEM;
 
+    if ( !sys )
+    {
+        ASSERT_UNREACHABLE();
+        return -EOPNOTSUPP;
+    }
+
     /* Start by copying the domain's existing policies. */
     if ( !(new.cpuid = xmemdup(d->arch.cpuid)) ||
          !(new.msr   = xmemdup(d->arch.msr)) )
@@ -64,7 +71,7 @@ static int update_domain_cpu_policy(struct domain *d,
     x86_cpuid_policy_clear_out_of_range_leaves(new.cpuid);
 
     /* Audit the combined dataset. */
-    ret = x86_cpu_policies_are_compatible(sys, &new, &err);
+    ret = x86_cpu_policies_are_compatible(&old_sys, &new, &err);
     if ( ret )
         goto out;
 
xen/arch/x86/include/asm/cpu-policy.h new file +14 -0
new file mode 100644
index 000000000000..eef14bb4267e
@@ -0,0 +1,14 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+#ifndef X86_CPU_POLICY_H
+#define X86_CPU_POLICY_H
+
+struct cpu_policy;
+
+extern struct cpu_policy     raw_cpu_policy;
+extern struct cpu_policy    host_cpu_policy;
+extern struct cpu_policy  pv_max_cpu_policy;
+extern struct cpu_policy  pv_def_cpu_policy;
+extern struct cpu_policy hvm_max_cpu_policy;
+extern struct cpu_policy hvm_def_cpu_policy;
+
+#endif /* X86_CPU_POLICY_H */
xen/arch/x86/include/asm/cpuid.h +0 -6
index d418e8100dde..ea0586277331 100644
@@ -46,12 +46,6 @@ DECLARE_PER_CPU(struct cpuidmasks, cpuidmasks);
 /* Default masking MSR values, calculated at boot. */
 extern struct cpuidmasks cpuidmask_defaults;
 
-extern struct cpuid_policy raw_cpuid_policy, host_cpuid_policy,
-    pv_max_cpuid_policy, pv_def_cpuid_policy,
-    hvm_max_cpuid_policy, hvm_def_cpuid_policy;
-
-extern const struct old_cpu_policy system_policies[];
-
 /* Check that all previously present features are still available. */
 bool recheck_cpu_features(unsigned int cpu);
 
xen/arch/x86/include/asm/msr.h +0 -7
index 02eddd919c27..022230acc0af 100644
@@ -292,13 +292,6 @@ static inline void wrmsr_tsc_aux(uint32_t val)
 
 uint64_t msr_spec_ctrl_valid_bits(const struct cpuid_policy *cp);
 
-extern struct msr_policy     raw_msr_policy,
-                            host_msr_policy,
-                          pv_max_msr_policy,
-                          pv_def_msr_policy,
-                         hvm_max_msr_policy,
-                         hvm_def_msr_policy;
-
 /* Container object for per-vCPU MSRs */
 struct vcpu_msrs
 {
xen/arch/x86/msr.c +14 -24
index 7ddf0078c3a2..bff26bc4e2b5 100644
@@ -25,6 +25,7 @@
 #include <xen/sched.h>
 
 #include <asm/amd.h>
+#include <asm/cpu-policy.h>
 #include <asm/debugreg.h>
 #include <asm/hvm/nestedhvm.h>
 #include <asm/hvm/viridian.h>
@@ -37,20 +38,9 @@
 
 DEFINE_PER_CPU(uint32_t, tsc_aux);
 
-struct msr_policy __read_mostly     raw_msr_policy,
-                  __read_mostly    host_msr_policy;
-#ifdef CONFIG_PV
-struct msr_policy __read_mostly  pv_max_msr_policy;
-struct msr_policy __read_mostly  pv_def_msr_policy;
-#endif
-#ifdef CONFIG_HVM
-struct msr_policy __read_mostly hvm_max_msr_policy;
-struct msr_policy __read_mostly hvm_def_msr_policy;
-#endif
-
 static void __init calculate_raw_policy(void)
 {
-    struct msr_policy *mp = &raw_msr_policy;
+    struct msr_policy *mp = &raw_cpu_policy;
 
     /* 0x000000ce  MSR_INTEL_PLATFORM_INFO */
     /* Was already added by probe_cpuid_faulting() */
@@ -61,9 +51,9 @@ static void __init calculate_raw_policy(void)
 
 static void __init calculate_host_policy(void)
 {
-    struct msr_policy *mp = &host_msr_policy;
+    struct msr_policy *mp = &host_cpu_policy;
 
-    *mp = raw_msr_policy;
+    *mp = raw_cpu_policy;
 
     /* 0x000000ce  MSR_INTEL_PLATFORM_INFO */
     /* probe_cpuid_faulting() sanity checks presence of MISC_FEATURES_ENABLES */
@@ -81,25 +71,25 @@ static void __init calculate_host_policy(void)
 
 static void __init calculate_pv_max_policy(void)
 {
-    struct msr_policy *mp = &pv_max_msr_policy;
+    struct msr_policy *mp = &pv_max_cpu_policy;
 
-    *mp = host_msr_policy;
+    *mp = host_cpu_policy;
 
     mp->arch_caps.raw = 0; /* Not supported yet. */
 }
 
 static void __init calculate_pv_def_policy(void)
 {
-    struct msr_policy *mp = &pv_def_msr_policy;
+    struct msr_policy *mp = &pv_def_cpu_policy;
 
-    *mp = pv_max_msr_policy;
+    *mp = pv_max_cpu_policy;
 }
 
 static void __init calculate_hvm_max_policy(void)
 {
-    struct msr_policy *mp = &hvm_max_msr_policy;
+    struct msr_policy *mp = &hvm_max_cpu_policy;
 
-    *mp = host_msr_policy;
+    *mp = host_cpu_policy;
 
     /* It's always possible to emulate CPUID faulting for HVM guests */
     mp->platform_info.cpuid_faulting = true;
@@ -109,9 +99,9 @@ static void __init calculate_hvm_max_policy(void)
 
 static void __init calculate_hvm_def_policy(void)
 {
-    struct msr_policy *mp = &hvm_def_msr_policy;
+    struct msr_policy *mp = &hvm_def_cpu_policy;
 
-    *mp = hvm_max_msr_policy;
+    *mp = hvm_max_cpu_policy;
 }
 
 void __init init_guest_msr_policy(void)
@@ -135,8 +125,8 @@ void __init init_guest_msr_policy(void)
 int init_domain_msr_policy(struct domain *d)
 {
     struct msr_policy *mp = is_pv_domain(d)
-        ? (IS_ENABLED(CONFIG_PV)  ?  &pv_def_msr_policy : NULL)
-        : (IS_ENABLED(CONFIG_HVM) ? &hvm_def_msr_policy : NULL);
+        ? (IS_ENABLED(CONFIG_PV)  ?  &pv_def_cpu_policy : NULL)
+        : (IS_ENABLED(CONFIG_HVM) ? &hvm_def_cpu_policy : NULL);
 
     if ( !mp )
     {
xen/arch/x86/sysctl.c +24 -43

109 lines not on this page · raw message

Signature · 1 line
2.30.2
#9 lore raw

Re: [PATCH 5/9] x86: Merge the system {cpuid,msr} policy objects

Quoted 6 lines · Andrew Cooper

Right now, they're the same underlying type, containing disjoint information.

Introduce a new cpu-policy.{h,c} to be the new location for all policy handling logic. Place the combined objects in __ro_after_init, which has appeared since the original logic was written.

Quoted 8 lines

As we're trying to phase out the use of struct old_cpu_policy entirely, rework update_domain_cpu_policy() to not pointer-chase through system_policies[].

This in turn allows system_policies[] in sysctl.c to become static and reduced in scope to XEN_SYSCTL_get_cpu_policy.

No practical change, but we do half the amount of compile time space required for the system policies, which saves 6x almost-2k at the moment.

But what you save here is only what was introduced earlier in the series, if I'm not mistaken (just like the "containing disjoint" further up)? Also do you mean "halve" in place of "half"?

xen/arch/x86/sysctl.c long arch_do_sysctl( · line 363

Quoted 6 lines
xen/arch/x86/sysctl.c +1 -1
@@ -391,7 +360,19 @@ long arch_do_sysctl(
 
     case XEN_SYSCTL_get_cpu_policy:
     {
-        const struct old_cpu_policy *policy;
+        static const struct cpu_policy *system_policies[6] = {

Another const wanted here after *?

Jan

#10 lore raw

Re: [PATCH 5/9] x86: Merge the system {cpuid,msr} policy objects

Quoted 10 lines · Jan Beulich
Quoted 9 lines · Andrew Cooper

Right now, they're the same underlying type, containing disjoint information.

Introduce a new cpu-policy.{h,c} to be the new location for all policy handling logic. Place the combined objects in __ro_after_init, which has appeared since the original logic was written.

As we're trying to phase out the use of struct old_cpu_policy entirely, rework update_domain_cpu_policy() to not pointer-chase through system_policies[].

Quoted 7 lines
Quoted 5 lines

This in turn allows system_policies[] in sysctl.c to become static and reduced in scope to XEN_SYSCTL_get_cpu_policy.

No practical change, but we do half the amount of compile time space required for the system policies, which saves 6x almost-2k at the moment.

But what you save here is only what was introduced earlier in the series, if I'm not mistaken (just like the "containing disjoint" further up)?

Hmm yeah - I should reword to make it clearer that this is returning to how it was before the series.

Quoted 1 line

Also do you mean "halve" in place of "half"?

Yes.

Quoted 8 lines

xen/arch/x86/sysctl.c long arch_do_sysctl( · line 363

Quoted 7 lines
xen/arch/x86/sysctl.c +1 -1
@@ -391,7 +360,19 @@ long arch_do_sysctl(
 
     case XEN_SYSCTL_get_cpu_policy:
     {
-        const struct old_cpu_policy *policy;
+        static const struct cpu_policy *system_policies[6] = {

Another const wanted here after *?

Will do.

~Andrew

#11 lore raw

PATCH 6/9 x86: Merge a domain's {cpuid,msr} policy objects

In reply to Andrew Cooper

Right now, they're the same underlying type, containing disjoint information.

Drop the d->arch.msr pointer, and union d->arch.cpuid to give it a second name of cpu_policy in the interim.

Merge init_domain_{cpuid,msr}_policy() into a single init_domain_cpu_policy(), moving the implementation into cpu-policy.c

No practical change, but this does half the memory allocated for a policy information per domain.

---
CC: Jan Beulich <[email protected]>
CC: Roger Pau Monné <[email protected]>
CC: Wei Liu <[email protected]>

---
 xen/arch/x86/cpu-policy.c             | 49 +++++++++++++++++++++++++
 xen/arch/x86/cpuid.c                  | 23 ------------
 xen/arch/x86/domain.c                 | 16 +++------
 xen/arch/x86/domctl.c                 | 35 +++++++++---------
 xen/arch/x86/include/asm/cpu-policy.h |  4 +++
 xen/arch/x86/include/asm/cpuid.h      |  3 --
 xen/arch/x86/include/asm/domain.h     |  7 ++--
 xen/arch/x86/include/asm/msr.h        |  1 -
 xen/arch/x86/mm/mem_sharing.c         |  3 +-
 xen/arch/x86/msr.c                    | 52 ++-------------------------
 10 files changed, 83 insertions(+), 110 deletions(-)
xen/arch/x86/cpu-policy.c +49 -0
index 663e9a084c53..e9ac1269c35a 100644
@@ -1,10 +1,13 @@
 /* SPDX-License-Identifier: GPL-2.0-or-later */
 #include <xen/cache.h>
 #include <xen/kernel.h>
+#include <xen/sched.h>
 
 #include <xen/lib/x86/cpu-policy.h>
 
 #include <asm/cpu-policy.h>
+#include <asm/msr-index.h>
+#include <asm/setup.h>
 
 struct cpu_policy __ro_after_init     raw_cpu_policy;
 struct cpu_policy __ro_after_init    host_cpu_policy;
@@ -16,3 +19,49 @@ struct cpu_policy __ro_after_init  pv_def_cpu_policy;
 struct cpu_policy __ro_after_init hvm_max_cpu_policy;
 struct cpu_policy __ro_after_init hvm_def_cpu_policy;
 #endif
+
+int init_domain_cpu_policy(struct domain *d)
+{
+    struct cpu_policy *p = is_pv_domain(d)
+        ? (IS_ENABLED(CONFIG_PV)  ?  &pv_def_cpu_policy : NULL)
+        : (IS_ENABLED(CONFIG_HVM) ? &hvm_def_cpu_policy : NULL);
+
+    if ( !p )
+    {
+        ASSERT_UNREACHABLE();
+        return -EOPNOTSUPP;
+    }
+
+    p = xmemdup(p);
+    if ( !p )
+        return -ENOMEM;
+
+    /* See comment in ctxt_switch_levelling() */
+    if ( !opt_dom0_cpuid_faulting && is_control_domain(d) && is_pv_domain(d) )
+        p->platform_info.cpuid_faulting = false;
+
+    /*
+     * Expose the "hardware speculation behaviour" bits of ARCH_CAPS to dom0,
+     * so dom0 can turn off workarounds as appropriate.  Temporary, until the
+     * domain policy logic gains a better understanding of MSRs.
+     */
+    if ( is_hardware_domain(d) && cpu_has_arch_caps )
+    {
+        uint64_t val;
+
+        rdmsrl(MSR_ARCH_CAPABILITIES, val);
+
+        p->arch_caps.raw = val &
+            (ARCH_CAPS_RDCL_NO | ARCH_CAPS_IBRS_ALL | ARCH_CAPS_RSBA |
+             ARCH_CAPS_SSB_NO | ARCH_CAPS_MDS_NO | ARCH_CAPS_IF_PSCHANGE_MC_NO |
+             ARCH_CAPS_TAA_NO | ARCH_CAPS_SBDR_SSDP_NO | ARCH_CAPS_FBSDP_NO |
+             ARCH_CAPS_PSDP_NO | ARCH_CAPS_FB_CLEAR | ARCH_CAPS_RRSBA |
+             ARCH_CAPS_BHI_NO | ARCH_CAPS_PBRSB_NO);
+    }
+
+    d->arch.cpu_policy = p;
+
+    recalculate_cpuid_policy(d);
+
+    return 0;
+}
xen/arch/x86/cpuid.c +0 -23
index 0916bfe175c8..df3e503ced9d 100644
@@ -784,29 +784,6 @@ void recalculate_cpuid_policy(struct domain *d)
         p->extd.raw[0x19] = EMPTY_LEAF;
 }
 
-int init_domain_cpuid_policy(struct domain *d)
-{
-    struct cpuid_policy *p = is_pv_domain(d)
-        ? (IS_ENABLED(CONFIG_PV)  ?  &pv_def_cpu_policy : NULL)
-        : (IS_ENABLED(CONFIG_HVM) ? &hvm_def_cpu_policy : NULL);
-
-    if ( !p )
-    {
-        ASSERT_UNREACHABLE();
-        return -EOPNOTSUPP;
-    }
-
-    p = xmemdup(p);
-    if ( !p )
-        return -ENOMEM;
-
-    d->arch.cpuid = p;
-
-    recalculate_cpuid_policy(d);
-
-    return 0;
-}
-
 void __init init_dom0_cpuid_policy(struct domain *d)
 {
     struct cpuid_policy *p = d->arch.cpuid;
xen/arch/x86/domain.c +5 -11
index d5847f70f890..7cf66aee042c 100644
@@ -66,6 +66,7 @@
 #ifdef CONFIG_COMPAT
 #include <compat/vcpu.h>
 #endif
+#include <asm/cpu-policy.h>
 #include <asm/psr.h>
 #include <asm/pv/domain.h>
 #include <asm/pv/mm.h>
@@ -573,7 +574,6 @@ int arch_vcpu_create(struct vcpu *v)
         /* Idle domain */
         v->arch.cr3 = __pa(idle_pg_table);
         rc = 0;
-        v->arch.msrs = ZERO_BLOCK_PTR; /* Catch stray misuses */
     }
 
     if ( rc )
@@ -743,8 +743,7 @@ int arch_domain_create(struct domain *d,
 
         d->arch.ctxt_switch = &idle_csw;
 
-        d->arch.cpuid = ZERO_BLOCK_PTR; /* Catch stray misuses. */
-        d->arch.msr = ZERO_BLOCK_PTR;
+        d->arch.cpu_policy = ZERO_BLOCK_PTR; /* Catch stray misuses. */
 
         return 0;
     }
@@ -799,10 +798,7 @@ int arch_domain_create(struct domain *d,
         goto fail;
     paging_initialised = true;
 
-    if ( (rc = init_domain_cpuid_policy(d)) )
-        goto fail;
-
-    if ( (rc = init_domain_msr_policy(d)) )
+    if ( (rc = init_domain_cpu_policy(d)) )
         goto fail;
 
     d->arch.ioport_caps =
@@ -873,8 +869,7 @@ int arch_domain_create(struct domain *d,
     iommu_domain_destroy(d);
     cleanup_domain_irq_mapping(d);
     free_xenheap_page(d->shared_info);
-    xfree(d->arch.cpuid);
-    xfree(d->arch.msr);
+    XFREE(d->arch.cpu_policy);
     if ( paging_initialised )
         paging_final_teardown(d);
     free_perdomain_mappings(d);
@@ -888,8 +883,7 @@ void arch_domain_destroy(struct domain *d)
         hvm_domain_destroy(d);
 
     xfree(d->arch.e820);
-    xfree(d->arch.cpuid);
-    xfree(d->arch.msr);
+    XFREE(d->arch.cpu_policy);
 
     free_domain_pirqs(d);
     if ( !is_idle_domain(d) )
xen/arch/x86/domctl.c +17 -18
index 5800bb10bc4a..81be25c67731 100644
@@ -40,11 +40,11 @@
 static int update_domain_cpu_policy(struct domain *d,
                                     xen_domctl_cpu_policy_t *xdpc)
 {
-    struct old_cpu_policy new = {};
+    struct cpu_policy *new;
     struct cpu_policy *sys = is_pv_domain(d)
         ? (IS_ENABLED(CONFIG_PV)  ?  &pv_max_cpu_policy : NULL)
         : (IS_ENABLED(CONFIG_HVM) ? &hvm_max_cpu_policy : NULL);
-    struct old_cpu_policy old_sys = { sys, sys };
+    struct old_cpu_policy old_sys = { sys, sys }, old_new;
     struct cpu_policy_errors err = INIT_CPU_POLICY_ERRORS;
     int ret = -ENOMEM;
 
@@ -54,33 +54,33 @@ static int update_domain_cpu_policy(struct domain *d,
         return -EOPNOTSUPP;
     }
 
-    /* Start by copying the domain's existing policies. */
-    if ( !(new.cpuid = xmemdup(d->arch.cpuid)) ||
-         !(new.msr   = xmemdup(d->arch.msr)) )
+    /* Start by copying the domain's existing policy. */
+    if ( !(new = xmemdup(d->arch.cpu_policy)) )
         goto out;
 
+    old_new = (struct old_cpu_policy){ new, new };
+
     /* Merge the toolstack provided data. */
     if ( (ret = x86_cpuid_copy_from_buffer(
-              new.cpuid, xdpc->leaves, xdpc->nr_leaves,
+              new, xdpc->leaves, xdpc->nr_leaves,
               &err.leaf, &err.subleaf)) ||
          (ret = x86_msr_copy_from_buffer(
-              new.msr, xdpc->msrs, xdpc->nr_msrs, &err.msr)) )
+              new, xdpc->msrs, xdpc->nr_msrs, &err.msr)) )
         goto out;
 
     /* Trim any newly-stale out-of-range leaves. */
-    x86_cpuid_policy_clear_out_of_range_leaves(new.cpuid);
+    x86_cpuid_policy_clear_out_of_range_leaves(new);
 
     /* Audit the combined dataset. */
-    ret = x86_cpu_policies_are_compatible(&old_sys, &new, &err);
+    ret = x86_cpu_policies_are_compatible(&old_sys, &old_new, &err);
     if ( ret )
         goto out;
 
     /*
-     * Audit was successful.  Replace existing policies, leaving the old
-     * policies to be freed.
+     * Audit was successful.  Replace the existing policy, leaving the old one
+     * to be freed.
      */
-    SWAP(new.cpuid, d->arch.cpuid);
-    SWAP(new.msr,   d->arch.msr);
+    SWAP(new, d->arch.cpu_policy);
 
     /* TODO: Drop when x86_cpu_policies_are_compatible() is completed. */
     recalculate_cpuid_policy(d);
@@ -89,9 +89,8 @@ static int update_domain_cpu_policy(struct domain *d,
     domain_cpu_policy_changed(d);
 
  out:
-    /* Free whichever cpuid/msr structs are not installed in struct domain. */
-    xfree(new.cpuid);
-    xfree(new.msr);
+    /* Free whichever struct is not installed in struct domain. */
+    xfree(new);
 
     if ( ret )
     {
@@ -1327,7 +1326,7 @@ long arch_do_domctl(
         if ( guest_handle_is_null(domctl->u.cpu_policy.leaves) )
             domctl->u.cpu_policy.nr_leaves = CPUID_MAX_SERIALISED_LEAVES;
         else if ( (ret = x86_cpuid_copy_to_buffer(
-                       d->arch.cpuid,
+                       d->arch.cpu_policy,
                        domctl->u.cpu_policy.leaves,
                        &domctl->u.cpu_policy.nr_leaves)) )
             break;
@@ -1336,7 +1335,7 @@ long arch_do_domctl(
         if ( guest_handle_is_null(domctl->u.cpu_policy.msrs) )
             domctl->u.cpu_policy.nr_msrs = MSR_MAX_SERIALISED_ENTRIES;
         else if ( (ret = x86_msr_copy_to_buffer(
-                       d->arch.msr,
+                       d->arch.cpu_policy,
                        domctl->u.cpu_policy.msrs,
                        &domctl->u.cpu_policy.nr_msrs)) )
             break;
xen/arch/x86/include/asm/cpu-policy.h +4 -0
index eef14bb4267e..9ba34bbf5ea1 100644
@@ -3,6 +3,7 @@
 #define X86_CPU_POLICY_H
 
 struct cpu_policy;
+struct domain;
 
 extern struct cpu_policy     raw_cpu_policy;
 extern struct cpu_policy    host_cpu_policy;
@@ -11,4 +12,7 @@ extern struct cpu_policy  pv_def_cpu_policy;
 extern struct cpu_policy hvm_max_cpu_policy;
 extern struct cpu_policy hvm_def_cpu_policy;
 
+/* Allocate and initialise a CPU policy suitable for the domain. */
+int init_domain_cpu_policy(struct domain *d);
+
 #endif /* X86_CPU_POLICY_H */
xen/arch/x86/include/asm/cpuid.h +0 -3
index ea0586277331..7f81b998ce01 100644
@@ -49,9 +49,6 @@ extern struct cpuidmasks cpuidmask_defaults;
 /* Check that all previously present features are still available. */
 bool recheck_cpu_features(unsigned int cpu);
 
-/* Allocate and initialise a CPUID policy suitable for the domain. */
-int init_domain_cpuid_policy(struct domain *d);
-
 /* Apply dom0-specific tweaks to the CPUID policy. */
 void init_dom0_cpuid_policy(struct domain *d);
 
xen/arch/x86/include/asm/domain.h +4 -3
index 17780ad9db2f..a1deab1d0dcc 100644
@@ -386,9 +386,10 @@ struct arch_domain
      */
     uint8_t x87_fip_width;
 
-    /* CPUID and MSR policy objects. */
-    struct cpuid_policy *cpuid;
-    struct msr_policy *msr;
+    union {
+        struct cpu_policy *cpuid; /* Temporary */
+        struct cpu_policy *cpu_policy;
+    };
 
     struct PITState vpit;
 
xen/arch/x86/include/asm/msr.h +0 -1
index 022230acc0af..b59a51d238a7 100644
@@ -419,7 +419,6 @@ struct vcpu_msrs
 };
 
 void init_guest_msr_policy(void);
-int init_domain_msr_policy(struct domain *d);
 int init_vcpu_msr_policy(struct vcpu *v);
 
 /*
xen/arch/x86/mm/mem_sharing.c +1 -2
index 649d93dc5444..5b3449db7a11 100644
@@ -1902,8 +1902,7 @@ static int fork(struct domain *cd, struct domain *d)
 
         domain_pause(d);
         cd->max_pages = d->max_pages;
-        *cd->arch.cpuid = *d->arch.cpuid;
-        *cd->arch.msr = *d->arch.msr;
+        *cd->arch.cpu_policy = *d->arch.cpu_policy;
         cd->vmtrace_size = d->vmtrace_size;
         cd->parent = d;
     }
xen/arch/x86/msr.c +3 -49

87 lines not on this page · raw message

Signature · 1 line
2.30.2
#12 lore raw

Re: [PATCH 6/9] x86: Merge a domain's {cpuid,msr} policy objects

xen/arch/x86/domain.c int arch_vcpu_create(struct vcpu *v) · line 577

Quoted 6 lines · Andrew Cooper
xen/arch/x86/domain.c +0 -1
@@ -573,7 +574,6 @@ int arch_vcpu_create(struct vcpu *v)
         /* Idle domain */
         v->arch.cr3 = __pa(idle_pg_table);
         rc = 0;
-        v->arch.msrs = ZERO_BLOCK_PTR; /* Catch stray misuses */
     }

Is this intentional? It's a vCPU pointer here, not a domain one.

xen/arch/x86/domctl.c line 47

Quoted 13 lines
xen/arch/x86/domctl.c +2 -2
@@ -40,11 +40,11 @@
 static int update_domain_cpu_policy(struct domain *d,
                                     xen_domctl_cpu_policy_t *xdpc)
 {
-    struct old_cpu_policy new = {};
+    struct cpu_policy *new;
     struct cpu_policy *sys = is_pv_domain(d)
         ? (IS_ENABLED(CONFIG_PV)  ?  &pv_max_cpu_policy : NULL)
         : (IS_ENABLED(CONFIG_HVM) ? &hvm_max_cpu_policy : NULL);
-    struct old_cpu_policy old_sys = { sys, sys };
+    struct old_cpu_policy old_sys = { sys, sys }, old_new;

Interesting name, but as long as it's transitional only, that's of course fine.

Jan

#13 lore raw

Re: [PATCH 6/9] x86: Merge a domain's {cpuid,msr} policy objects

Quoted 8 lines · Jan Beulich

xen/arch/x86/domain.c int arch_vcpu_create(struct vcpu *v) · line 577

Quoted 6 lines · Andrew Cooper
xen/arch/x86/domain.c +0 -1
@@ -573,7 +574,6 @@ int arch_vcpu_create(struct vcpu *v)
         /* Idle domain */
         v->arch.cr3 = __pa(idle_pg_table);
         rc = 0;
-        v->arch.msrs = ZERO_BLOCK_PTR; /* Catch stray misuses */
     }

Is this intentional? It's a vCPU pointer here, not a domain one.

Ah, no.  And it answers one of my TODO notes that I hadn't got to yet (of why MSRs was different to CPUID in this case).

It looks like it got caught in my arch.msrs rework.  I'll drop this hunk.

Quoted 16 lines
Quoted 4 lines

xen/arch/x86/domctl.c line 47

Quoted 10 lines
xen/arch/x86/domctl.c +2 -2
@@ -40,11 +40,11 @@
 static int update_domain_cpu_policy(struct domain *d,
                                     xen_domctl_cpu_policy_t *xdpc)
 {
-    struct old_cpu_policy new = {};
+    struct cpu_policy *new;
     struct cpu_policy *sys = is_pv_domain(d)
         ? (IS_ENABLED(CONFIG_PV)  ?  &pv_max_cpu_policy : NULL)
         : (IS_ENABLED(CONFIG_HVM) ? &hvm_max_cpu_policy : NULL);
-    struct old_cpu_policy old_sys = { sys, sys };
+    struct old_cpu_policy old_sys = { sys, sys }, old_new;

Interesting name, but as long as it's transitional only, that's of course fine.

Yeah... it was the best I could come up with.

It does get removed in patch 8.

~Andrew

#14 lore raw

PATCH 7/9 x86: Merge xc_cpu_policy's cpuid and msr objects

In reply to Andrew Cooper

Right now, they're the same underlying type, containing disjoint information.

Use a single object instead. Also take the opportunity to rename 'entries' to 'msrs' which is more descriptive, and more in line with nr_msrs being the count of MSR entries in the API.

test-tsx uses xg_private.h to access the internals of xc_cpu_policy, so needs updating at the same time.

No practical change, but it does reduce the size of struct xc_cpu_policy by ~2k.

---
CC: Jan Beulich <[email protected]>
CC: Roger Pau Monné <[email protected]>
CC: Wei Liu <[email protected]>

---
 tools/libs/guest/xg_cpuid_x86.c | 36 ++++++++++----------
 tools/libs/guest/xg_private.h   |  5 ++-
 tools/tests/tsx/test-tsx.c      | 58 ++++++++++++++++-----------------
 3 files changed, 48 insertions(+), 51 deletions(-)
tools/libs/guest/xg_cpuid_x86.c +18 -18
index 5fae06e77804..5061fe357767 100644
@@ -431,7 +431,7 @@ int xc_cpuid_apply_policy(xc_interface *xch, uint32_t domid, bool restore,
     xc_dominfo_t di;
     unsigned int i, nr_leaves, nr_msrs;
     xen_cpuid_leaf_t *leaves = NULL;
-    struct cpuid_policy *p = NULL;
+    struct cpu_policy *p = NULL;
     uint32_t err_leaf = -1, err_subleaf = -1, err_msr = -1;
     uint32_t host_featureset[FEATURESET_NR_ENTRIES] = {};
     uint32_t len = ARRAY_SIZE(host_featureset);
@@ -692,7 +692,7 @@ static int deserialize_policy(xc_interface *xch, xc_cpu_policy_t *policy,
     uint32_t err_leaf = -1, err_subleaf = -1, err_msr = -1;
     int rc;
 
-    rc = x86_cpuid_copy_from_buffer(&policy->cpuid, policy->leaves,
+    rc = x86_cpuid_copy_from_buffer(&policy->policy, policy->leaves,
                                     nr_leaves, &err_leaf, &err_subleaf);
     if ( rc )
     {
@@ -702,7 +702,7 @@ static int deserialize_policy(xc_interface *xch, xc_cpu_policy_t *policy,
         return rc;
     }
 
-    rc = x86_msr_copy_from_buffer(&policy->msr, policy->entries,
+    rc = x86_msr_copy_from_buffer(&policy->policy, policy->msrs,
                                   nr_entries, &err_msr);
     if ( rc )
     {
@@ -719,18 +719,18 @@ int xc_cpu_policy_get_system(xc_interface *xch, unsigned int policy_idx,
                              xc_cpu_policy_t *policy)
 {
     unsigned int nr_leaves = ARRAY_SIZE(policy->leaves);
-    unsigned int nr_entries = ARRAY_SIZE(policy->entries);
+    unsigned int nr_msrs = ARRAY_SIZE(policy->msrs);
     int rc;
 
     rc = get_system_cpu_policy(xch, policy_idx, &nr_leaves, policy->leaves,
-                               &nr_entries, policy->entries);
+                               &nr_msrs, policy->msrs);
     if ( rc )
     {
         PERROR("Failed to obtain %u policy", policy_idx);
         return rc;
     }
 
-    rc = deserialize_policy(xch, policy, nr_leaves, nr_entries);
+    rc = deserialize_policy(xch, policy, nr_leaves, nr_msrs);
     if ( rc )
     {
         errno = -rc;
@@ -744,18 +744,18 @@ int xc_cpu_policy_get_domain(xc_interface *xch, uint32_t domid,
                              xc_cpu_policy_t *policy)
 {
     unsigned int nr_leaves = ARRAY_SIZE(policy->leaves);
-    unsigned int nr_entries = ARRAY_SIZE(policy->entries);
+    unsigned int nr_msrs = ARRAY_SIZE(policy->msrs);
     int rc;
 
     rc = get_domain_cpu_policy(xch, domid, &nr_leaves, policy->leaves,
-                               &nr_entries, policy->entries);
+                               &nr_msrs, policy->msrs);
     if ( rc )
     {
         PERROR("Failed to obtain domain %u policy", domid);
         return rc;
     }
 
-    rc = deserialize_policy(xch, policy, nr_leaves, nr_entries);
+    rc = deserialize_policy(xch, policy, nr_leaves, nr_msrs);
     if ( rc )
     {
         errno = -rc;
@@ -770,16 +770,16 @@ int xc_cpu_policy_set_domain(xc_interface *xch, uint32_t domid,
 {
     uint32_t err_leaf = -1, err_subleaf = -1, err_msr = -1;
     unsigned int nr_leaves = ARRAY_SIZE(policy->leaves);
-    unsigned int nr_entries = ARRAY_SIZE(policy->entries);
+    unsigned int nr_msrs = ARRAY_SIZE(policy->msrs);
     int rc;
 
     rc = xc_cpu_policy_serialise(xch, policy, policy->leaves, &nr_leaves,
-                                 policy->entries, &nr_entries);
+                                 policy->msrs, &nr_msrs);
     if ( rc )
         return rc;
 
     rc = xc_set_domain_cpu_policy(xch, domid, nr_leaves, policy->leaves,
-                                  nr_entries, policy->entries,
+                                  nr_msrs, policy->msrs,
                                   &err_leaf, &err_subleaf, &err_msr);
     if ( rc )
     {
@@ -802,7 +802,7 @@ int xc_cpu_policy_serialise(xc_interface *xch, const xc_cpu_policy_t *p,
 
     if ( leaves )
     {
-        rc = x86_cpuid_copy_to_buffer(&p->cpuid, leaves, nr_leaves);
+        rc = x86_cpuid_copy_to_buffer(&p->policy, leaves, nr_leaves);
         if ( rc )
         {
             ERROR("Failed to serialize CPUID policy");
@@ -813,7 +813,7 @@ int xc_cpu_policy_serialise(xc_interface *xch, const xc_cpu_policy_t *p,
 
     if ( msrs )
     {
-        rc = x86_msr_copy_to_buffer(&p->msr, msrs, nr_msrs);
+        rc = x86_msr_copy_to_buffer(&p->policy, msrs, nr_msrs);
         if ( rc )
         {
             ERROR("Failed to serialize MSR policy");
@@ -831,7 +831,7 @@ int xc_cpu_policy_update_cpuid(xc_interface *xch, xc_cpu_policy_t *policy,
                                uint32_t nr)
 {
     unsigned int err_leaf = -1, err_subleaf = -1;
-    int rc = x86_cpuid_copy_from_buffer(&policy->cpuid, leaves, nr,
+    int rc = x86_cpuid_copy_from_buffer(&policy->policy, leaves, nr,
                                         &err_leaf, &err_subleaf);
 
     if ( rc )
@@ -850,7 +850,7 @@ int xc_cpu_policy_update_msrs(xc_interface *xch, xc_cpu_policy_t *policy,
                               const xen_msr_entry_t *msrs, uint32_t nr)
 {
     unsigned int err_msr = -1;
-    int rc = x86_msr_copy_from_buffer(&policy->msr, msrs, nr, &err_msr);
+    int rc = x86_msr_copy_from_buffer(&policy->policy, msrs, nr, &err_msr);
 
     if ( rc )
     {
@@ -868,8 +868,8 @@ bool xc_cpu_policy_is_compatible(xc_interface *xch, xc_cpu_policy_t *host,
                                  xc_cpu_policy_t *guest)
 {
     struct cpu_policy_errors err = INIT_CPU_POLICY_ERRORS;
-    struct old_cpu_policy h = { &host->cpuid, &host->msr };
-    struct old_cpu_policy g = { &guest->cpuid, &guest->msr };
+    struct old_cpu_policy h = { &host->policy, &host->policy };
+    struct old_cpu_policy g = { &guest->policy, &guest->policy };
     int rc = x86_cpu_policies_are_compatible(&h, &g, &err);
 
     if ( !rc )
tools/libs/guest/xg_private.h +2 -3
index 09e24f122760..e729a8106c3e 100644
@@ -173,10 +173,9 @@ int pin_table(xc_interface *xch, unsigned int type, unsigned long mfn,
 #include <xen/lib/x86/cpu-policy.h>
 
 struct xc_cpu_policy {
-    struct cpuid_policy cpuid;
-    struct msr_policy msr;
+    struct cpu_policy policy;
     xen_cpuid_leaf_t leaves[CPUID_MAX_SERIALISED_LEAVES];
-    xen_msr_entry_t entries[MSR_MAX_SERIALISED_ENTRIES];
+    xen_msr_entry_t msrs[MSR_MAX_SERIALISED_ENTRIES];
 };
 #endif /* x86 */
 
tools/tests/tsx/test-tsx.c +28 -30

138 lines not on this page · raw message

Signature · 1 line
2.30.2
#15 lore raw

Re: [PATCH 7/9] x86: Merge xc_cpu_policy's cpuid and msr objects

tools/tests/tsx/test-tsx.c static void dump_tsx_details(const struct xc_cpu_policy *p, const char *pref) · line 312

Quoted 10 lines · Andrew Cooper

/* Sanity test various invariants we expect in the default/max policies. */ static void test_guest_policies(const struct xc_cpu_policy *max,
const struct xc_cpu_policy *def)
{

tools/tests/tsx/test-tsx.c +2 -4
-    const struct cpuid_policy *cm = &max->cpuid;
-    const struct cpuid_policy *cd = &def->cpuid;
-    const struct msr_policy *mm = &max->msr;
-    const struct msr_policy *md = &def->msr;
+    const struct cpu_policy *m = &max->policy;
+    const struct cpu_policy *d = &def->policy;

Looks like you could reduce code churn by keeping "cm" and "cd" as the names, which would also be consistent with you continuing to use "cp" in hypervisor code.

Jan

Quoted 40 lines
     dump_tsx_details(max, "Max:");
     dump_tsx_details(def, "Def:");
tools/tests/tsx/test-tsx.c +6 -6
-    if ( ((cm->feat.raw[0].d | cd->feat.raw[0].d) &
+    if ( ((m->feat.raw[0].d | d->feat.raw[0].d) &
           (bitmaskof(X86_FEATURE_TSX_FORCE_ABORT) |
            bitmaskof(X86_FEATURE_RTM_ALWAYS_ABORT) |
            bitmaskof(X86_FEATURE_SRBDS_CTRL))) ||
-         ((mm->arch_caps.raw | md->arch_caps.raw) & ARCH_CAPS_TSX_CTRL) )
+         ((m->arch_caps.raw | d->arch_caps.raw) & ARCH_CAPS_TSX_CTRL) )
         fail("  Xen-only TSX controls offered to guest\n");
 
     switch ( rtm_behaviour )
     {
     case RTM_UD:
-        if ( (cm->feat.raw[0].b | cd->feat.raw[0].b) &
+        if ( (m->feat.raw[0].b | d->feat.raw[0].b) &
              (bitmaskof(X86_FEATURE_HLE) | bitmaskof(X86_FEATURE_RTM)) )
              fail("  HLE/RTM offered to guests despite not being available\n");
         break;
 
     case RTM_ABORT:
-        if ( cd->feat.raw[0].b &
+        if ( d->feat.raw[0].b &
              (bitmaskof(X86_FEATURE_HLE) | bitmaskof(X86_FEATURE_RTM)) )
              fail("  HLE/RTM offered to guests by default despite not being usable\n");
         break;
 
     case RTM_OK:
-        if ( !cm->feat.rtm || !cd->feat.rtm )
+        if ( !m->feat.rtm || !d->feat.rtm )
              fail("  RTM not offered to guests despite being available\n");
         break;
     }
 
-    if ( cd->feat.hle )
+    if ( d->feat.hle )
         fail("  Fail: HLE offered in default policy\n");
 }
#16 lore raw

Re: [PATCH 7/9] x86: Merge xc_cpu_policy's cpuid and msr objects

Quoted 11 lines · Jan Beulich

tools/tests/tsx/test-tsx.c static void dump_tsx_details(const struct xc_cpu_policy *p, const char *pref) · line 312

Quoted 10 lines · Andrew Cooper

/* Sanity test various invariants we expect in the default/max policies. */ static void test_guest_policies(const struct xc_cpu_policy *max,
const struct xc_cpu_policy *def)
{

tools/tests/tsx/test-tsx.c +2 -4
-    const struct cpuid_policy *cm = &max->cpuid;
-    const struct cpuid_policy *cd = &def->cpuid;
-    const struct msr_policy *mm = &max->msr;
-    const struct msr_policy *md = &def->msr;
+    const struct cpu_policy *m = &max->policy;
+    const struct cpu_policy *d = &def->policy;
Quoted 3 lines

Looks like you could reduce code churn by keeping "cm" and "cd" as the names, which would also be consistent with you continuing to use "cp" in hypervisor code.

It's a unit test - I'm not worried about churn; I'm more concerned about the end readability.

I did consider swapping {d,m} with {max,def} so the logic below reads:

    if ( ((max->feat.raw[0].d | def->feat.raw[0].d) &

but it occurs to me that because the policies are now merged, I can get rid of all of the xc_cpu_policy passing and use cpu_policy instead. (For inspection purposes, we don't care about the serialisation buffers on the tail of xc_cpu_policy).

It will be a slightly larger change, but the end result will be better IMO.

~Andrew

#17 lore raw

PATCH 8/9 x86: Drop struct old_cpu_policy

In reply to Andrew Cooper

With all the complicated callers of x86_cpu_policies_are_compatible() updated to use a single cpu_policy object, we can drop the final user of struct old_cpu_policy.

Update x86_cpu_policies_are_compatible() to take (new) cpu_policy pointers, reducing the amount of internal pointer chasing, and update all callers to pass their cpu_policy objects directly.

---
CC: Jan Beulich <[email protected]>
CC: Roger Pau Monné <[email protected]>
CC: Wei Liu <[email protected]>

---
 tools/libs/guest/xg_cpuid_x86.c          |  4 +--
 tools/tests/cpu-policy/test-cpu-policy.c | 40 ++++++------------------
 xen/arch/x86/domctl.c                    |  7 ++---
 xen/include/xen/lib/x86/cpu-policy.h     | 12 ++-----
 xen/lib/x86/policy.c                     | 12 +++----
 5 files changed, 22 insertions(+), 53 deletions(-)
tools/libs/guest/xg_cpuid_x86.c +1 -3
index 5061fe357767..14c00304c03c 100644
@@ -868,9 +868,7 @@ bool xc_cpu_policy_is_compatible(xc_interface *xch, xc_cpu_policy_t *host,
                                  xc_cpu_policy_t *guest)
 {
     struct cpu_policy_errors err = INIT_CPU_POLICY_ERRORS;
-    struct old_cpu_policy h = { &host->policy, &host->policy };
-    struct old_cpu_policy g = { &guest->policy, &guest->policy };
-    int rc = x86_cpu_policies_are_compatible(&h, &g, &err);
+    int rc = x86_cpu_policies_are_compatible(&host->policy, &host->policy, &err);
 
     if ( !rc )
         return true;
tools/tests/cpu-policy/test-cpu-policy.c +10 -30
index 909d6272f875..5b38702b1c62 100644
@@ -574,23 +574,20 @@ static void test_is_compatible_success(void)
 {
     static struct test {
         const char *name;
-        struct cpuid_policy host_cpuid;
-        struct cpuid_policy guest_cpuid;
-        struct msr_policy host_msr;
-        struct msr_policy guest_msr;
+        struct cpu_policy host, guest;
     } tests[] = {
         {
             .name = "Host CPUID faulting, Guest not",
-            .host_msr = {
+            .host = {
                 .platform_info.cpuid_faulting = true,
             },
         },
         {
             .name = "Host CPUID faulting, Guest wanted",
-            .host_msr = {
+            .host = {
                 .platform_info.cpuid_faulting = true,
             },
-            .guest_msr = {
+            .guest = {
                 .platform_info.cpuid_faulting = true,
             },
         },
@@ -602,15 +599,8 @@ static void test_is_compatible_success(void)
     for ( size_t i = 0; i < ARRAY_SIZE(tests); ++i )
     {
         struct test *t = &tests[i];
-        struct old_cpu_policy sys = {
-            &t->host_cpuid,
-            &t->host_msr,
-        }, new = {
-            &t->guest_cpuid,
-            &t->guest_msr,
-        };
         struct cpu_policy_errors e;
-        int res = x86_cpu_policies_are_compatible(&sys, &new, &e);
+        int res = x86_cpu_policies_are_compatible(&t->host, &t->guest, &e);
 
         /* Check the expected error output. */
         if ( res != 0 || memcmp(&no_errors, &e, sizeof(no_errors)) )
@@ -624,25 +614,22 @@ static void test_is_compatible_failure(void)
 {
     static struct test {
         const char *name;
-        struct cpuid_policy host_cpuid;
-        struct cpuid_policy guest_cpuid;
-        struct msr_policy host_msr;
-        struct msr_policy guest_msr;
+        struct cpu_policy host, guest;
         struct cpu_policy_errors e;
     } tests[] = {
         {
             .name = "Host basic.max_leaf out of range",
-            .guest_cpuid.basic.max_leaf = 1,
+            .guest.basic.max_leaf = 1,
             .e = { 0, -1, -1 },
         },
         {
             .name = "Host extd.max_leaf out of range",
-            .guest_cpuid.extd.max_leaf = 1,
+            .guest.extd.max_leaf = 1,
             .e = { 0x80000000, -1, -1 },
         },
         {
             .name = "Host no CPUID faulting, Guest wanted",
-            .guest_msr = {
+            .guest = {
                 .platform_info.cpuid_faulting = true,
             },
             .e = { -1, -1, 0xce },
@@ -654,15 +641,8 @@ static void test_is_compatible_failure(void)
     for ( size_t i = 0; i < ARRAY_SIZE(tests); ++i )
     {
         struct test *t = &tests[i];
-        struct old_cpu_policy sys = {
-            &t->host_cpuid,
-            &t->host_msr,
-        }, new = {
-            &t->guest_cpuid,
-            &t->guest_msr,
-        };
         struct cpu_policy_errors e;
-        int res = x86_cpu_policies_are_compatible(&sys, &new, &e);
+        int res = x86_cpu_policies_are_compatible(&t->host, &t->guest, &e);
 
         /* Check the expected error output. */
         if ( res == 0 || memcmp(&t->e, &e, sizeof(t->e)) )
xen/arch/x86/domctl.c +2 -5
index 81be25c67731..c02528594102 100644
@@ -41,10 +41,9 @@ static int update_domain_cpu_policy(struct domain *d,
                                     xen_domctl_cpu_policy_t *xdpc)
 {
     struct cpu_policy *new;
-    struct cpu_policy *sys = is_pv_domain(d)
+    const struct cpu_policy *sys = is_pv_domain(d)
         ? (IS_ENABLED(CONFIG_PV)  ?  &pv_max_cpu_policy : NULL)
         : (IS_ENABLED(CONFIG_HVM) ? &hvm_max_cpu_policy : NULL);
-    struct old_cpu_policy old_sys = { sys, sys }, old_new;
     struct cpu_policy_errors err = INIT_CPU_POLICY_ERRORS;
     int ret = -ENOMEM;
 
@@ -58,8 +57,6 @@ static int update_domain_cpu_policy(struct domain *d,
     if ( !(new = xmemdup(d->arch.cpu_policy)) )
         goto out;
 
-    old_new = (struct old_cpu_policy){ new, new };
-
     /* Merge the toolstack provided data. */
     if ( (ret = x86_cpuid_copy_from_buffer(
               new, xdpc->leaves, xdpc->nr_leaves,
@@ -72,7 +69,7 @@ static int update_domain_cpu_policy(struct domain *d,
     x86_cpuid_policy_clear_out_of_range_leaves(new);
 
     /* Audit the combined dataset. */
-    ret = x86_cpu_policies_are_compatible(&old_sys, &old_new, &err);
+    ret = x86_cpu_policies_are_compatible(sys, new, &err);
     if ( ret )
         goto out;
 
xen/include/xen/lib/x86/cpu-policy.h +3 -9
index 5af756a02da0..51f88f1e217e 100644
@@ -379,12 +379,6 @@ struct cpu_policy
 #define cpuid_policy cpu_policy
 #define msr_policy cpu_policy
 
-struct old_cpu_policy
-{
-    struct cpuid_policy *cpuid;
-    struct msr_policy *msr;
-};
-
 struct cpu_policy_errors
 {
     uint32_t leaf, subleaf;
@@ -559,7 +553,7 @@ int x86_msr_copy_from_buffer(struct msr_policy *policy,
                              const msr_entry_buffer_t msrs, uint32_t nr_entries,
                              uint32_t *err_msr);
 
-/*
+/**
  * Calculate whether two policies are compatible.
  *
  * i.e. Can a VM configured with @guest run on a CPU supporting @host.
@@ -573,8 +567,8 @@ int x86_msr_copy_from_buffer(struct msr_policy *policy,
  * incompatibility is detected, the optional err pointer may identify the
  * problematic leaf/subleaf and/or MSR.
  */
-int x86_cpu_policies_are_compatible(const struct old_cpu_policy *host,
-                                    const struct old_cpu_policy *guest,
+int x86_cpu_policies_are_compatible(const struct cpu_policy *host,
+                                    const struct cpu_policy *guest,
                                     struct cpu_policy_errors *err);
 
 #endif /* !XEN_LIB_X86_POLICIES_H */
xen/lib/x86/policy.c +6 -6
index 2975711d7c6c..a9c60000af9d 100644
@@ -2,8 +2,8 @@
 
 #include <xen/lib/x86/cpu-policy.h>
 
-int x86_cpu_policies_are_compatible(const struct old_cpu_policy *host,
-                                    const struct old_cpu_policy *guest,
+int x86_cpu_policies_are_compatible(const struct cpu_policy *host,
+                                    const struct cpu_policy *guest,
                                     struct cpu_policy_errors *err)
 {
     struct cpu_policy_errors e = INIT_CPU_POLICY_ERRORS;
@@ -15,18 +15,18 @@ int x86_cpu_policies_are_compatible(const struct old_cpu_policy *host,
 #define FAIL_MSR(m) \
     do { e.msr = (m); goto out; } while ( 0 )
 
-    if ( guest->cpuid->basic.max_leaf > host->cpuid->basic.max_leaf )
+    if ( guest->basic.max_leaf > host->basic.max_leaf )
         FAIL_CPUID(0, NA);
 
-    if ( guest->cpuid->feat.max_subleaf > host->cpuid->feat.max_subleaf )
+    if ( guest->feat.max_subleaf > host->feat.max_subleaf )
         FAIL_CPUID(7, 0);
 
-    if ( guest->cpuid->extd.max_leaf > host->cpuid->extd.max_leaf )
+    if ( guest->extd.max_leaf > host->extd.max_leaf )
         FAIL_CPUID(0x80000000, NA);
 
     /* TODO: Audit more CPUID data. */
 
-    if ( ~host->msr->platform_info.raw & guest->msr->platform_info.raw )
+    if ( ~host->platform_info.raw & guest->platform_info.raw )
         FAIL_MSR(MSR_INTEL_PLATFORM_INFO);
 
 #undef FAIL_MSR
Signature · 1 line
2.30.2
#18 lore raw

Re: [PATCH 8/9] x86: Drop struct old_cpu_policy

tools/libs/guest/xg_cpuid_x86.c bool xc_cpu_policy_is_compatible(xc_interface *xch, xc_cpu_policy_t *host, · line 871

Quoted 10 lines · Andrew Cooper
tools/libs/guest/xg_cpuid_x86.c +1 -3
@@ -868,9 +868,7 @@ bool xc_cpu_policy_is_compatible(xc_interface *xch, xc_cpu_policy_t *host,
                                  xc_cpu_policy_t *guest)
 {
     struct cpu_policy_errors err = INIT_CPU_POLICY_ERRORS;
-    struct old_cpu_policy h = { &host->policy, &host->policy };
-    struct old_cpu_policy g = { &guest->policy, &guest->policy };
-    int rc = x86_cpu_policies_are_compatible(&h, &g, &err);
+    int rc = x86_cpu_policies_are_compatible(&host->policy, &host->policy, &err);

One of the two surely wants to be &guest->policy?

Jan

#19 lore raw

Re: [PATCH 8/9] x86: Drop struct old_cpu_policy

Quoted 12 lines · Jan Beulich

tools/libs/guest/xg_cpuid_x86.c bool xc_cpu_policy_is_compatible(xc_interface *xch, xc_cpu_policy_t *host, · line 871

Quoted 10 lines · Andrew Cooper
tools/libs/guest/xg_cpuid_x86.c +1 -3
@@ -868,9 +868,7 @@ bool xc_cpu_policy_is_compatible(xc_interface *xch, xc_cpu_policy_t *host,
                                  xc_cpu_policy_t *guest)
 {
     struct cpu_policy_errors err = INIT_CPU_POLICY_ERRORS;
-    struct old_cpu_policy h = { &host->policy, &host->policy };
-    struct old_cpu_policy g = { &guest->policy, &guest->policy };
-    int rc = x86_cpu_policies_are_compatible(&h, &g, &err);
+    int rc = x86_cpu_policies_are_compatible(&host->policy, &host->policy, &err);

One of the two surely wants to be &guest->policy?

Ah, yes it does.  The second, specifically.

(One of the problems of having an API that's not used by anything interesting yet.)

~Andrew

#20 lore raw

PATCH 9/9 RFC: Everything else

In reply to Andrew Cooper

Looking at this diff, I'm wondering whether keeping

    union {
        struct cpu_policy *cpuid;
        struct cpu_policy *cpu_policy;
    };

permentantly might be a good idea, because d->arch.cpuid->$X reads rather better than d->arch.cpu_policy->$X

Thoughts?

---
CC: Jan Beulich <[email protected]>
CC: Roger Pau Monné <[email protected]>
CC: Wei Liu <[email protected]>

---
 tools/fuzz/cpu-policy/afl-policy-fuzzer.c     |  12 +-
 .../fuzz/x86_instruction_emulator/fuzz-emul.c |   2 +-
 tools/tests/cpu-policy/test-cpu-policy.c      |  10 +-
 tools/tests/x86_emulator/test_x86_emulator.c  |   2 +-
 tools/tests/x86_emulator/x86-emulate.c        |   2 +-
 xen/arch/x86/cpu/mcheck/mce_intel.c           |   2 +-
 xen/arch/x86/cpu/vpmu_intel.c                 |   4 +-
 xen/arch/x86/cpuid.c                          |  26 +--
 xen/arch/x86/domain.c                         |   2 +-
 xen/arch/x86/hvm/emulate.c                    |   2 +-
 xen/arch/x86/hvm/hvm.c                        |  38 ++--
 xen/arch/x86/hvm/ioreq.c                      |   4 +-
 xen/arch/x86/hvm/mtrr.c                       |   2 +-
 xen/arch/x86/hvm/svm/svm.c                    |  18 +-
 xen/arch/x86/hvm/svm/svmdebug.c               |   2 +-
 xen/arch/x86/hvm/vlapic.c                     |   2 +-
 xen/arch/x86/hvm/vmx/vmx.c                    |  12 +-
 xen/arch/x86/hvm/vmx/vvmx.c                   |   2 +-
 xen/arch/x86/include/asm/domain.h             |   5 +-
 xen/arch/x86/include/asm/guest_pt.h           |   4 +-
 xen/arch/x86/include/asm/msr.h                |   2 +-
 xen/arch/x86/include/asm/paging.h             |   2 +-
 xen/arch/x86/mm/shadow/hvm.c                  |   2 +-
 xen/arch/x86/msr.c                            |  24 +--
 xen/arch/x86/pv/domain.c                      |   2 +-
 xen/arch/x86/pv/emul-priv-op.c                |   6 +-
 xen/arch/x86/pv/ro-page-fault.c               |   2 +-
 xen/arch/x86/sysctl.c                         |   4 +-
 xen/arch/x86/traps.c                          |   2 +-
 xen/arch/x86/x86_emulate.c                    |   2 +-
 xen/arch/x86/x86_emulate/x86_emulate.c        | 166 +++++++++---------
 xen/arch/x86/x86_emulate/x86_emulate.h        |   4 +-
 xen/arch/x86/xstate.c                         |   4 +-
 xen/include/xen/lib/x86/cpu-policy.h          |  28 ++-
 xen/lib/x86/cpuid.c                           |  10 +-
 xen/lib/x86/msr.c                             |   4 +-
 36 files changed, 205 insertions(+), 212 deletions(-)
tools/fuzz/cpu-policy/afl-policy-fuzzer.c +6 -6
index 0ce3d8e16626..58b4dbacf96b 100644
@@ -16,9 +16,9 @@ static bool debug;
 
 #define EMPTY_LEAF ((struct cpuid_leaf){})
 
-static void check_cpuid(struct cpuid_policy *cp)
+static void check_cpuid(struct cpu_policy *cp)
 {
-    struct cpuid_policy new = {};
+    struct cpu_policy new = {};
     size_t data_end;
     xen_cpuid_leaf_t *leaves = malloc(CPUID_MAX_SERIALISED_LEAVES *
                                       sizeof(xen_cpuid_leaf_t));
@@ -76,9 +76,9 @@ static void check_cpuid(struct cpuid_policy *cp)
     free(leaves);
 }
 
-static void check_msr(struct msr_policy *mp)
+static void check_msr(struct cpu_policy *mp)
 {
-    struct msr_policy new = {};
+    struct cpu_policy new = {};
     xen_msr_entry_t *msrs = malloc(MSR_MAX_SERIALISED_ENTRIES *
                                    sizeof(xen_msr_entry_t));
     unsigned int nr = MSR_MAX_SERIALISED_ENTRIES;
@@ -144,8 +144,8 @@ int main(int argc, char **argv)
     while ( __AFL_LOOP(1000) )
 #endif
     {
-        struct cpuid_policy *cp = NULL;
-        struct msr_policy *mp = NULL;
+        struct cpu_policy *cp = NULL;
+        struct cpu_policy *mp = NULL;
 
         if ( fp != stdin )
         {
tools/fuzz/x86_instruction_emulator/fuzz-emul.c +1 -1
index 966e46bee199..4885a68210d0 100644
@@ -893,7 +893,7 @@ int LLVMFuzzerTestOneInput(const uint8_t *data_p, size_t size)
     struct x86_emulate_ctxt ctxt = {
         .data = &state,
         .regs = &input.regs,
-        .cpuid = &cp,
+        .cpu_policy = &cp,
         .addr_size = 8 * sizeof(void *),
         .sp_size = 8 * sizeof(void *),
     };
tools/tests/cpu-policy/test-cpu-policy.c +5 -5
index 5b38702b1c62..a4ca07f33973 100644
@@ -98,7 +98,7 @@ static bool msrs_are_sorted(const xen_msr_entry_t *entries, unsigned int nr)
 
 static void test_cpuid_current(void)
 {
-    struct cpuid_policy p;
+    struct cpu_policy p;
     xen_cpuid_leaf_t leaves[CPUID_MAX_SERIALISED_LEAVES];
     unsigned int nr = ARRAY_SIZE(leaves);
     int rc;
@@ -118,7 +118,7 @@ static void test_cpuid_current(void)
 static void test_cpuid_serialise_success(void)
 {
     static const struct test {
-        struct cpuid_policy p;
+        struct cpu_policy p;
         const char *name;
         unsigned int nr_leaves;
     } tests[] = {
@@ -242,7 +242,7 @@ static void test_cpuid_serialise_success(void)
 static void test_msr_serialise_success(void)
 {
     static const struct test {
-        struct msr_policy p;
+        struct cpu_policy p;
         const char *name;
         unsigned int nr_msrs;
     } tests[] = {
@@ -430,7 +430,7 @@ static void test_cpuid_out_of_range_clearing(void)
     static const struct test {
         const char *name;
         unsigned int nr_markers;
-        struct cpuid_policy p;
+        struct cpu_policy p;
     } tests[] = {
         {
             .name = "basic",
@@ -550,7 +550,7 @@ static void test_cpuid_out_of_range_clearing(void)
     for ( size_t i = 0; i < ARRAY_SIZE(tests); ++i )
     {
         const struct test *t = &tests[i];
-        struct cpuid_policy *p = memdup(&t->p);
+        struct cpu_policy *p = memdup(&t->p);
         void *ptr;
         unsigned int nr_markers;
 
tools/tests/x86_emulator/test_x86_emulator.c +1 -1
index 31586f805726..7b7fbaaf45ec 100644
@@ -909,7 +909,7 @@ int main(int argc, char **argv)
 
     ctxt.regs = &regs;
     ctxt.force_writeback = 0;
-    ctxt.cpuid     = &cp;
+    ctxt.cpu_policy = &cp;
     ctxt.lma       = sizeof(void *) == 8;
     ctxt.addr_size = 8 * sizeof(void *);
     ctxt.sp_size   = 8 * sizeof(void *);
tools/tests/x86_emulator/x86-emulate.c +1 -1
index ea286d6ad87b..5ad282b57545 100644
@@ -38,7 +38,7 @@
 #define put_stub(stb) ((stb).addr = 0)
 
 uint32_t mxcsr_mask = 0x0000ffbf;
-struct cpuid_policy cp;
+struct cpu_policy cp;
 
 static char fpu_save_area[0x4000] __attribute__((__aligned__((64))));
 static bool use_xsave;
xen/arch/x86/cpu/mcheck/mce_intel.c +1 -1
index 301533722d1a..2f23f02923d2 100644
@@ -1008,7 +1008,7 @@ int vmce_intel_wrmsr(struct vcpu *v, uint32_t msr, uint64_t val)
 
 int vmce_intel_rdmsr(const struct vcpu *v, uint32_t msr, uint64_t *val)
 {
-    const struct cpuid_policy *cp = v->domain->arch.cpuid;
+    const struct cpu_policy *cp = v->domain->arch.cpu_policy;
     unsigned int bank = msr - MSR_IA32_MC0_CTL2;
 
     switch ( msr )
xen/arch/x86/cpu/vpmu_intel.c +2 -2
index bcfa187a14b6..045f40dd582e 100644
@@ -839,8 +839,8 @@ static int cf_check core2_vpmu_initialise(struct vcpu *v)
     u64 msr_content;
     static bool_t ds_warned;
 
-    if ( v->domain->arch.cpuid->basic.pmu_version <= 1 ||
-         v->domain->arch.cpuid->basic.pmu_version >= 6 )
+    if ( v->domain->arch.cpu_policy->basic.pmu_version <= 1 ||
+         v->domain->arch.cpu_policy->basic.pmu_version >= 6 )
         return -EINVAL;
 
     if ( (arch_pmc_cnt + fixed_pmc_cnt) == 0 )
xen/arch/x86/cpuid.c +13 -13
index df3e503ced9d..c7e268e14918 100644
@@ -178,7 +178,7 @@ static void sanitise_featureset(uint32_t *fs)
     }
 }
 
-static void recalculate_xstate(struct cpuid_policy *p)
+static void recalculate_xstate(struct cpu_policy *p)
 {
     uint64_t xstates = XSTATE_FP_SSE;
     uint32_t xstate_size = XSTATE_AREA_MIN_SIZE;
@@ -256,7 +256,7 @@ static void recalculate_xstate(struct cpuid_policy *p)
  * Misc adjustments to the policy.  Mostly clobbering reserved fields and
  * duplicating shared fields.  Intentionally hidden fields are annotated.
  */
-static void recalculate_misc(struct cpuid_policy *p)
+static void recalculate_misc(struct cpu_policy *p)
 {
     p->basic.raw_fms &= 0x0fff0fff; /* Clobber Processor Type on Intel. */
     p->basic.apic_id = 0; /* Dynamic. */
@@ -334,7 +334,7 @@ static void recalculate_misc(struct cpuid_policy *p)
 
 static void __init calculate_raw_policy(void)
 {
-    struct cpuid_policy *p = &raw_cpu_policy;
+    struct cpu_policy *p = &raw_cpu_policy;
 
     x86_cpuid_policy_fill_native(p);
 
@@ -344,7 +344,7 @@ static void __init calculate_raw_policy(void)
 
 static void __init calculate_host_policy(void)
 {
-    struct cpuid_policy *p = &host_cpu_policy;
+    struct cpu_policy *p = &host_cpu_policy;
     unsigned int max_extd_leaf;
 
     *p = raw_cpu_policy;
@@ -445,7 +445,7 @@ static void __init guest_common_feature_adjustments(uint32_t *fs)
 
 static void __init calculate_pv_max_policy(void)
 {
-    struct cpuid_policy *p = &pv_max_cpu_policy;
+    struct cpu_policy *p = &pv_max_cpu_policy;
     uint32_t pv_featureset[FSCAPINTS];
     unsigned int i;
 
@@ -476,7 +476,7 @@ static void __init calculate_pv_max_policy(void)
 
 static void __init calculate_pv_def_policy(void)
 {
-    struct cpuid_policy *p = &pv_def_cpu_policy;
+    struct cpu_policy *p = &pv_def_cpu_policy;
     uint32_t pv_featureset[FSCAPINTS];
     unsigned int i;
 
@@ -496,7 +496,7 @@ static void __init calculate_pv_def_policy(void)
 
 static void __init calculate_hvm_max_policy(void)
 {
-    struct cpuid_policy *p = &hvm_max_cpu_policy;
+    struct cpu_policy *p = &hvm_max_cpu_policy;
     uint32_t hvm_featureset[FSCAPINTS];
     unsigned int i;
     const uint32_t *hvm_featuremask;
@@ -587,7 +587,7 @@ static void __init calculate_hvm_max_policy(void)
 
 static void __init calculate_hvm_def_policy(void)
 {
-    struct cpuid_policy *p = &hvm_def_cpu_policy;
+    struct cpu_policy *p = &hvm_def_cpu_policy;
     uint32_t hvm_featureset[FSCAPINTS];
     unsigned int i;
     const uint32_t *hvm_featuremask;
@@ -658,8 +658,8 @@ bool recheck_cpu_features(unsigned int cpu)
 
 void recalculate_cpuid_policy(struct domain *d)
 {
-    struct cpuid_policy *p = d->arch.cpuid;
-    const struct cpuid_policy *max = is_pv_domain(d)
+    struct cpu_policy *p = d->arch.cpu_policy;
+    const struct cpu_policy *max = is_pv_domain(d)
         ? (IS_ENABLED(CONFIG_PV)  ?  &pv_max_cpu_policy : NULL)
         : (IS_ENABLED(CONFIG_HVM) ? &hvm_max_cpu_policy : NULL);
     uint32_t fs[FSCAPINTS], max_fs[FSCAPINTS];
@@ -786,7 +786,7 @@ void recalculate_cpuid_policy(struct domain *d)
 
 void __init init_dom0_cpuid_policy(struct domain *d)
 {
-    struct cpuid_policy *p = d->arch.cpuid;
+    struct cpu_policy *p = d->arch.cpu_policy;
 
     /* dom0 can't migrate.  Give it ITSC if available. */
     if ( cpu_has_itsc )
@@ -824,7 +824,7 @@ void guest_cpuid(const struct vcpu *v, uint32_t leaf,
                  uint32_t subleaf, struct cpuid_leaf *res)
 {
     const struct domain *d = v->domain;
-    const struct cpuid_policy *p = d->arch.cpuid;
+    const struct cpu_policy *p = d->arch.cpu_policy;
 
     *res = EMPTY_LEAF;
 
@@ -1199,7 +1199,7 @@ static void __init __maybe_unused build_assertions(void)
     BUILD_BUG_ON(ARRAY_SIZE(deep_features) != FSCAPINTS);
 
     /* Find some more clever allocation scheme if this trips. */
-    BUILD_BUG_ON(sizeof(struct cpuid_policy) > PAGE_SIZE);
+    BUILD_BUG_ON(sizeof(struct cpu_policy) > PAGE_SIZE);
 
     BUILD_BUG_ON(sizeof(raw_cpu_policy.basic) !=
                  sizeof(raw_cpu_policy.basic.raw));
xen/arch/x86/domain.c +1 -1
index 7cf66aee042c..c628619ead2c 100644
@@ -283,7 +283,7 @@ void update_guest_memory_policy(struct vcpu *v,
 
 void domain_cpu_policy_changed(struct domain *d)
 {
-    const struct cpuid_policy *p = d->arch.cpuid;
+    const struct cpu_policy *p = d->arch.cpu_policy;
     struct vcpu *v;
 
     if ( is_pv_domain(d) )
xen/arch/x86/hvm/emulate.c +1 -1
index 95364deb1996..53f58c805185 100644
@@ -2846,7 +2846,7 @@ void hvm_emulate_init_once(
 
     hvmemul_ctxt->validate = validate;
     hvmemul_ctxt->ctxt.regs = regs;
-    hvmemul_ctxt->ctxt.cpuid = curr->domain->arch.cpuid;
+    hvmemul_ctxt->ctxt.cpu_policy = curr->domain->arch.cpu_policy;
     hvmemul_ctxt->ctxt.force_writeback = true;
 }
 
xen/arch/x86/hvm/hvm.c +19 -19
index d326fa1c0136..0cf1b483dbbe 100644
@@ -925,7 +925,7 @@ const char *hvm_efer_valid(const struct vcpu *v, uint64_t value,
                            signed int cr0_pg)
 {
     const struct domain *d = v->domain;
-    const struct cpuid_policy *p = d->arch.cpuid;
+    const struct cpu_policy *p = d->arch.cpu_policy;
 
     if ( value & ~EFER_KNOWN_MASK )
         return "Unknown bits set";
@@ -962,7 +962,7 @@ const char *hvm_efer_valid(const struct vcpu *v, uint64_t value,
 /* These bits in CR4 can be set by the guest. */
 unsigned long hvm_cr4_guest_valid_bits(const struct domain *d)
 {
-    const struct cpuid_policy *p = d->arch.cpuid;
+    const struct cpu_policy *p = d->arch.cpu_policy;
     bool mce, vmxe, cet;
 
     /* Logic broken out simply to aid readability below. */
@@ -1040,7 +1040,7 @@ static int cf_check hvm_load_cpu_ctxt(struct domain *d, hvm_domain_context_t *h)
         return -EINVAL;
     }
 
-    if ( ctxt.cr3 >> d->arch.cpuid->extd.maxphysaddr )
+    if ( ctxt.cr3 >> d->arch.cpu_policy->extd.maxphysaddr )
     {
         printk(XENLOG_G_ERR "HVM%d restore: bad CR3 %#" PRIx64 "\n",
                d->domain_id, ctxt.cr3);
@@ -1418,7 +1418,7 @@ static int cf_check hvm_save_cpu_msrs(struct vcpu *v, hvm_domain_context_t *h)
         switch ( msr )
         {
         case MSR_SPEC_CTRL:
-            val &= msr_spec_ctrl_valid_bits(d->arch.cpuid);
+            val &= msr_spec_ctrl_valid_bits(d->arch.cpu_policy);
             break;
         }
 
@@ -2397,7 +2397,7 @@ int hvm_set_cr3(unsigned long value, bool noflush, bool may_defer)
     struct vcpu *curr = current;
     struct domain *currd = curr->domain;
 
-    if ( value >> currd->arch.cpuid->extd.maxphysaddr )
+    if ( value >> currd->arch.cpu_policy->extd.maxphysaddr )
     {
         HVM_DBG_LOG(DBG_LEVEL_1,
                     "Attempt to set reserved CR3 bit(s): %lx", value);
@@ -2666,7 +2666,7 @@ bool hvm_vcpu_virtual_to_linear(
         else if ( last_byte > reg->limit )
             goto out; /* last byte is beyond limit */
         else if ( last_byte < offset &&
-                  v->domain->arch.cpuid->x86_vendor == X86_VENDOR_AMD )
+                  v->domain->arch.cpu_policy->x86_vendor == X86_VENDOR_AMD )
             goto out; /* access wraps */
     }
 
@@ -3554,12 +3554,12 @@ int hvm_msr_read_intercept(unsigned int msr, uint64_t *msr_content)
         break;
 
     case MSR_MTRRcap:
-        if ( !d->arch.cpuid->basic.mtrr )
+        if ( !d->arch.cpu_policy->basic.mtrr )
             goto gp_fault;
         *msr_content = v->arch.hvm.mtrr.mtrr_cap;
         break;
     case MSR_MTRRdefType:
-        if ( !d->arch.cpuid->basic.mtrr )
+        if ( !d->arch.cpu_policy->basic.mtrr )
             goto gp_fault;
         *msr_content = v->arch.hvm.mtrr.def_type |
                        MASK_INSR(v->arch.hvm.mtrr.enabled, MTRRdefType_E) |
@@ -3567,27 +3567,27 @@ int hvm_msr_read_intercept(unsigned int msr, uint64_t *msr_content)
                                  MTRRdefType_FE);
         break;
     case MSR_MTRRfix64K_00000:
-        if ( !d->arch.cpuid->basic.mtrr )
+        if ( !d->arch.cpu_policy->basic.mtrr )
             goto gp_fault;
         *msr_content = fixed_range_base[0];
         break;
     case MSR_MTRRfix16K_80000:
     case MSR_MTRRfix16K_A0000:
-        if ( !d->arch.cpuid->basic.mtrr )
+        if ( !d->arch.cpu_policy->basic.mtrr )
             goto gp_fault;
         index = msr - MSR_MTRRfix16K_80000;
         *msr_content = fixed_range_base[array_index_nospec(index + 1,
                                    ARRAY_SIZE(v->arch.hvm.mtrr.fixed_ranges))];
         break;
     case MSR_MTRRfix4K_C0000...MSR_MTRRfix4K_F8000:
-        if ( !d->arch.cpuid->basic.mtrr )
+        if ( !d->arch.cpu_policy->basic.mtrr )
             goto gp_fault;
         index = msr - MSR_MTRRfix4K_C0000;
         *msr_content = fixed_range_base[array_index_nospec(index + 3,
                                    ARRAY_SIZE(v->arch.hvm.mtrr.fixed_ranges))];
         break;
     case MSR_IA32_MTRR_PHYSBASE(0)...MSR_IA32_MTRR_PHYSMASK(MTRR_VCNT_MAX - 1):
-        if ( !d->arch.cpuid->basic.mtrr )
+        if ( !d->arch.cpu_policy->basic.mtrr )
             goto gp_fault;
         index = msr - MSR_IA32_MTRR_PHYSBASE(0);
         if ( (index / 2) >=
@@ -3693,14 +3693,14 @@ int hvm_msr_write_intercept(unsigned int msr, uint64_t msr_content,
         goto gp_fault;
 
     case MSR_MTRRdefType:
-        if ( !d->arch.cpuid->basic.mtrr )
+        if ( !d->arch.cpu_policy->basic.mtrr )
             goto gp_fault;
         if ( !mtrr_def_type_msr_set(v->domain, &v->arch.hvm.mtrr,
                                     msr_content) )
            goto gp_fault;
         break;
     case MSR_MTRRfix64K_00000:
-        if ( !d->arch.cpuid->basic.mtrr )
+        if ( !d->arch.cpu_policy->basic.mtrr )
             goto gp_fault;
         if ( !mtrr_fix_range_msr_set(v->domain, &v->arch.hvm.mtrr, 0,
                                      msr_content) )
@@ -3708,7 +3708,7 @@ int hvm_msr_write_intercept(unsigned int msr, uint64_t msr_content,
         break;
     case MSR_MTRRfix16K_80000:
     case MSR_MTRRfix16K_A0000:
-        if ( !d->arch.cpuid->basic.mtrr )
+        if ( !d->arch.cpu_policy->basic.mtrr )
             goto gp_fault;
         index = msr - MSR_MTRRfix16K_80000 + 1;
         if ( !mtrr_fix_range_msr_set(v->domain, &v->arch.hvm.mtrr,
@@ -3716,7 +3716,7 @@ int hvm_msr_write_intercept(unsigned int msr, uint64_t msr_content,
             goto gp_fault;
         break;
     case MSR_MTRRfix4K_C0000...MSR_MTRRfix4K_F8000:
-        if ( !d->arch.cpuid->basic.mtrr )
+        if ( !d->arch.cpu_policy->basic.mtrr )
             goto gp_fault;
         index = msr - MSR_MTRRfix4K_C0000 + 3;
         if ( !mtrr_fix_range_msr_set(v->domain, &v->arch.hvm.mtrr,
@@ -3724,7 +3724,7 @@ int hvm_msr_write_intercept(unsigned int msr, uint64_t msr_content,
             goto gp_fault;
         break;
     case MSR_IA32_MTRR_PHYSBASE(0)...MSR_IA32_MTRR_PHYSMASK(MTRR_VCNT_MAX - 1):
-        if ( !d->arch.cpuid->basic.mtrr )
+        if ( !d->arch.cpu_policy->basic.mtrr )
             goto gp_fault;
         index = msr - MSR_IA32_MTRR_PHYSBASE(0);
         if ( ((index / 2) >=
@@ -3830,7 +3830,7 @@ void hvm_ud_intercept(struct cpu_user_regs *regs)
 {
     struct vcpu *cur = current;
     bool should_emulate =
-        cur->domain->arch.cpuid->x86_vendor != boot_cpu_data.x86_vendor;
+        cur->domain->arch.cpu_policy->x86_vendor != boot_cpu_data.x86_vendor;
     struct hvm_emulate_ctxt ctxt;
 
     hvm_emulate_init_once(&ctxt, opt_hvm_fep ? NULL : is_cross_vendor, regs);
@@ -4737,7 +4737,7 @@ static int do_altp2m_op(
 
     case HVMOP_altp2m_set_suppress_ve_multi:
     {
-        uint64_t max_phys_addr = (1UL << d->arch.cpuid->extd.maxphysaddr) - 1;
+        uint64_t max_phys_addr = (1UL << d->arch.cpu_policy->extd.maxphysaddr) - 1;
 
         a.u.suppress_ve_multi.last_gfn = min(a.u.suppress_ve_multi.last_gfn,
                                              max_phys_addr);
xen/arch/x86/hvm/ioreq.c +2 -2
index 0bdcca1e1a5f..f397ce68cdb5 100644
@@ -295,9 +295,9 @@ bool arch_ioreq_server_get_type_addr(const struct domain *d,
         *addr = ((uint64_t)sbdf.sbdf << 32) | reg;
         /* AMD extended configuration space access? */
         if ( CF8_ADDR_HI(cf8) &&
-             d->arch.cpuid->x86_vendor == X86_VENDOR_AMD &&
+             d->arch.cpu_policy->x86_vendor == X86_VENDOR_AMD &&
              (x86_fam = get_cpu_family(
-                 d->arch.cpuid->basic.raw_fms, NULL, NULL)) >= 0x10 &&
+                 d->arch.cpu_policy->basic.raw_fms, NULL, NULL)) >= 0x10 &&
              x86_fam < 0x17 )
         {
             uint64_t msr_val;
xen/arch/x86/hvm/mtrr.c +1 -1
index f1a88d761635..805e6d4ff243 100644
@@ -460,7 +460,7 @@ bool_t mtrr_var_range_msr_set(
         return 0;
 
     if ( d == current->domain )
-        phys_addr = d->arch.cpuid->extd.maxphysaddr;
+        phys_addr = d->arch.cpu_policy->extd.maxphysaddr;
     else
         phys_addr = paddr_bits;
     msr_mask = ~((((uint64_t)1) << phys_addr) - 1);
xen/arch/x86/hvm/svm/svm.c +9 -9
index 02563e4b7027..5c2399e3eaba 100644
@@ -328,7 +328,7 @@ static void svm_save_dr(struct vcpu *v)
     v->arch.hvm.flag_dr_dirty = 0;
     vmcb_set_dr_intercepts(vmcb, ~0u);
 
-    if ( v->domain->arch.cpuid->extd.dbext )
+    if ( v->domain->arch.cpu_policy->extd.dbext )
     {
         svm_intercept_msr(v, MSR_AMD64_DR0_ADDRESS_MASK, MSR_INTERCEPT_RW);
         svm_intercept_msr(v, MSR_AMD64_DR1_ADDRESS_MASK, MSR_INTERCEPT_RW);
@@ -359,7 +359,7 @@ static void __restore_debug_registers(struct vmcb_struct *vmcb, struct vcpu *v)
 
     ASSERT(v == current);
 
-    if ( v->domain->arch.cpuid->extd.dbext )
+    if ( v->domain->arch.cpu_policy->extd.dbext )
     {
         svm_intercept_msr(v, MSR_AMD64_DR0_ADDRESS_MASK, MSR_INTERCEPT_NONE);
         svm_intercept_msr(v, MSR_AMD64_DR1_ADDRESS_MASK, MSR_INTERCEPT_NONE);
@@ -583,11 +583,11 @@ static void cf_check svm_cpuid_policy_changed(struct vcpu *v)
 {
     struct svm_vcpu *svm = &v->arch.hvm.svm;
     struct vmcb_struct *vmcb = svm->vmcb;
-    const struct cpuid_policy *cp = v->domain->arch.cpuid;
+    const struct cpu_policy *cp = v->domain->arch.cpu_policy;
     u32 bitmap = vmcb_get_exception_intercepts(vmcb);
 
     if ( opt_hvm_fep ||
-         (v->domain->arch.cpuid->x86_vendor != boot_cpu_data.x86_vendor) )
+         (v->domain->arch.cpu_policy->x86_vendor != boot_cpu_data.x86_vendor) )
         bitmap |= (1U << TRAP_invalid_op);
     else
         bitmap &= ~(1U << TRAP_invalid_op);
@@ -967,7 +967,7 @@ static void cf_check svm_ctxt_switch_from(struct vcpu *v)
      */
     if ( v->arch.msrs->virt_spec_ctrl.raw & SPEC_CTRL_SSBD )
     {
-        ASSERT(v->domain->arch.cpuid->extd.virt_ssbd);
+        ASSERT(v->domain->arch.cpu_policy->extd.virt_ssbd);
         amd_set_legacy_ssbd(false);
     }
 }
@@ -1001,7 +1001,7 @@ static void cf_check svm_ctxt_switch_to(struct vcpu *v)
     /* Load SSBD if set by the guest. */
     if ( v->arch.msrs->virt_spec_ctrl.raw & SPEC_CTRL_SSBD )
     {
-        ASSERT(v->domain->arch.cpuid->extd.virt_ssbd);
+        ASSERT(v->domain->arch.cpu_policy->extd.virt_ssbd);
         amd_set_legacy_ssbd(true);
     }
 }
@@ -1928,7 +1928,7 @@ static int cf_check svm_msr_read_intercept(
 
     case MSR_AMD_OSVW_ID_LENGTH:
     case MSR_AMD_OSVW_STATUS:
-        if ( !d->arch.cpuid->extd.osvw )
+        if ( !d->arch.cpu_policy->extd.osvw )
             goto gpf;
         *msr_content = d->arch.hvm.svm.osvw.raw[msr - MSR_AMD_OSVW_ID_LENGTH];
         break;
@@ -2121,7 +2121,7 @@ static int cf_check svm_msr_write_intercept(
 
     case MSR_AMD_OSVW_ID_LENGTH:
     case MSR_AMD_OSVW_STATUS:
-        if ( !d->arch.cpuid->extd.osvw )
+        if ( !d->arch.cpu_policy->extd.osvw )
             goto gpf;
         /* Write-discard */
         break;
@@ -2187,7 +2187,7 @@ static void svm_vmexit_do_rdtsc(struct cpu_user_regs *regs, bool rdtscp)
     const struct domain *currd = curr->domain;
     unsigned int inst_len;
 
-    if ( rdtscp && !currd->arch.cpuid->extd.rdtscp )
+    if ( rdtscp && !currd->arch.cpu_policy->extd.rdtscp )
     {
         hvm_inject_hw_exception(TRAP_invalid_op, X86_EVENT_NO_EC);
         return;
xen/arch/x86/hvm/svm/svmdebug.c +1 -1
index bce86f0ef706..05a2d119c744 100644
@@ -130,7 +130,7 @@ bool svm_vmcb_isvalid(const char *from, const struct vmcb_struct *vmcb,
          ((cr3 & 7) ||
           ((!(cr4 & X86_CR4_PAE) || (efer & EFER_LMA)) && (cr3 & 0xfe0)) ||
           ((efer & EFER_LMA) &&
-           (cr3 >> v->domain->arch.cpuid->extd.maxphysaddr))) )
+           (cr3 >> v->domain->arch.cpu_policy->extd.maxphysaddr))) )
         PRINTF("CR3: MBZ bits are set (%#"PRIx64")\n", cr3);
 
     valid = hvm_cr4_guest_valid_bits(v->domain);
xen/arch/x86/hvm/vlapic.c +1 -1
index dc93b5e930b1..f4f5ffc673e5 100644
@@ -1083,7 +1083,7 @@ static void set_x2apic_id(struct vlapic *vlapic)
 
 int guest_wrmsr_apic_base(struct vcpu *v, uint64_t value)
 {
-    const struct cpuid_policy *cp = v->domain->arch.cpuid;
+    const struct cpu_policy *cp = v->domain->arch.cpu_policy;
     struct vlapic *vlapic = vcpu_vlapic(v);
 
     if ( !has_vlapic(v->domain) )
xen/arch/x86/hvm/vmx/vmx.c +6 -6
index e05588505871..593400a8608f 100644
@@ -710,7 +710,7 @@ static void vmx_restore_host_msrs(void)
 
 static void vmx_save_guest_msrs(struct vcpu *v)
 {
-    const struct cpuid_policy *cp = v->domain->arch.cpuid;
+    const struct cpu_policy *cp = v->domain->arch.cpu_policy;
     struct vcpu_msrs *msrs = v->arch.msrs;
 
     /*
@@ -731,7 +731,7 @@ static void vmx_save_guest_msrs(struct vcpu *v)
 
 static void vmx_restore_guest_msrs(struct vcpu *v)
 {
-    const struct cpuid_policy *cp = v->domain->arch.cpuid;
+    const struct cpu_policy *cp = v->domain->arch.cpu_policy;
     const struct vcpu_msrs *msrs = v->arch.msrs;
 
     write_gs_shadow(v->arch.hvm.vmx.shadow_gs);
@@ -784,11 +784,11 @@ void vmx_update_exception_bitmap(struct vcpu *v)
 
 static void cf_check vmx_cpuid_policy_changed(struct vcpu *v)
 {
-    const struct cpuid_policy *cp = v->domain->arch.cpuid;
+    const struct cpu_policy *cp = v->domain->arch.cpu_policy;
     int rc = 0;
 
     if ( opt_hvm_fep ||
-         (v->domain->arch.cpuid->x86_vendor != boot_cpu_data.x86_vendor) )
+         (v->domain->arch.cpu_policy->x86_vendor != boot_cpu_data.x86_vendor) )
         v->arch.hvm.vmx.exception_bitmap |= (1U << TRAP_invalid_op);
     else
         v->arch.hvm.vmx.exception_bitmap &= ~(1U << TRAP_invalid_op);
@@ -3521,7 +3521,7 @@ static int cf_check vmx_msr_write_intercept(
     unsigned int msr, uint64_t msr_content)
 {
     struct vcpu *v = current;
-    const struct cpuid_policy *cp = v->domain->arch.cpuid;
+    const struct cpu_policy *cp = v->domain->arch.cpu_policy;
 
     HVM_DBG_LOG(DBG_LEVEL_MSR, "ecx=%#x, msr_value=%#"PRIx64, msr, msr_content);
 
@@ -4451,7 +4451,7 @@ void vmx_vmexit_handler(struct cpu_user_regs *regs)
         vmx_invlpg_intercept(exit_qualification);
         break;
     case EXIT_REASON_RDTSCP:
-        if ( !currd->arch.cpuid->extd.rdtscp )
+        if ( !currd->arch.cpu_policy->extd.rdtscp )
         {
             hvm_inject_hw_exception(TRAP_invalid_op, X86_EVENT_NO_EC);
             break;
xen/arch/x86/hvm/vmx/vvmx.c +1 -1
index 674cdabb0736..f78247d9122c 100644
@@ -2175,7 +2175,7 @@ int nvmx_msr_read_intercept(unsigned int msr, u64 *msr_content)
     int r = 1;
 
     /* VMX capablity MSRs are available only when guest supports VMX. */
-    if ( !nestedhvm_enabled(d) || !d->arch.cpuid->basic.vmx )
+    if ( !nestedhvm_enabled(d) || !d->arch.cpu_policy->basic.vmx )
         return 0;
 
     /*
xen/arch/x86/include/asm/domain.h +1 -4
index a1deab1d0dcc..6ab396c28589 100644
@@ -386,10 +386,7 @@ struct arch_domain
      */
     uint8_t x87_fip_width;
 
-    union {
-        struct cpu_policy *cpuid; /* Temporary */
-        struct cpu_policy *cpu_policy;
-    };
+    struct cpu_policy *cpu_policy;
 
     struct PITState vpit;
 
xen/arch/x86/include/asm/guest_pt.h +2 -2
index 6802db2a415a..bb5e5c4b9fba 100644
@@ -293,7 +293,7 @@ static always_inline bool guest_pks_enabled(const struct vcpu *v)
 static always_inline uint64_t guest_rsvd_bits(const struct vcpu *v)
 {
     return ((PADDR_MASK &
-             ~((1ul << v->domain->arch.cpuid->extd.maxphysaddr) - 1)) |
+             ~((1ul << v->domain->arch.cpu_policy->extd.maxphysaddr) - 1)) |
             (guest_nx_enabled(v) ? 0 : put_pte_flags(_PAGE_NX_BIT)));
 }
 
@@ -332,7 +332,7 @@ static always_inline bool guest_l4e_rsvd_bits(const struct vcpu *v,
                                               guest_l4e_t l4e)
 {
     return l4e.l4 & (guest_rsvd_bits(v) | GUEST_L4_PAGETABLE_RSVD |
-                     ((v->domain->arch.cpuid->x86_vendor == X86_VENDOR_AMD)
+                     ((v->domain->arch.cpu_policy->x86_vendor == X86_VENDOR_AMD)
                       ? _PAGE_GLOBAL : 0));
 }
 #endif /* GUEST_PAGING_LEVELS >= 4 */
xen/arch/x86/include/asm/msr.h +1 -1
index b59a51d238a7..3efa4b02449e 100644
@@ -290,7 +290,7 @@ static inline void wrmsr_tsc_aux(uint32_t val)
     }
 }
 
-uint64_t msr_spec_ctrl_valid_bits(const struct cpuid_policy *cp);
+uint64_t msr_spec_ctrl_valid_bits(const struct cpu_policy *cp);
 
 /* Container object for per-vCPU MSRs */
 struct vcpu_msrs
xen/arch/x86/include/asm/paging.h +1 -1
index 2647b95e67a7..5ae8745bb190 100644
@@ -346,7 +346,7 @@ int paging_set_allocation(struct domain *d, unsigned int pages,
 /* Is gfn within maxphysaddr for the domain? */
 static inline bool gfn_valid(const struct domain *d, gfn_t gfn)
 {
-    return !(gfn_x(gfn) >> (d->arch.cpuid->extd.maxphysaddr - PAGE_SHIFT));
+    return !(gfn_x(gfn) >> (d->arch.cpu_policy->extd.maxphysaddr - PAGE_SHIFT));
 }
 
 /* Maxphysaddr supportable by the paging infrastructure. */
xen/arch/x86/mm/shadow/hvm.c +1 -1
index e2ee1c77056f..cc84af01925a 100644
@@ -319,7 +319,7 @@ const struct x86_emulate_ops *shadow_init_emulation(
     memset(sh_ctxt, 0, sizeof(*sh_ctxt));
 
     sh_ctxt->ctxt.regs = regs;
-    sh_ctxt->ctxt.cpuid = curr->domain->arch.cpuid;
+    sh_ctxt->ctxt.cpu_policy = curr->domain->arch.cpu_policy;
     sh_ctxt->ctxt.lma = hvm_long_mode_active(curr);
 
     /* Segment cache initialisation. Primed with CS. */
xen/arch/x86/msr.c +12 -12
index 672961dd3ac1..bf6c8f5f36ec 100644
@@ -40,7 +40,7 @@ DEFINE_PER_CPU(uint32_t, tsc_aux);
 
 static void __init calculate_raw_policy(void)
 {
-    struct msr_policy *mp = &raw_cpu_policy;
+    struct cpu_policy *mp = &raw_cpu_policy;
 
     /* 0x000000ce  MSR_INTEL_PLATFORM_INFO */
     /* Was already added by probe_cpuid_faulting() */
@@ -51,7 +51,7 @@ static void __init calculate_raw_policy(void)
 
 static void __init calculate_host_policy(void)
 {
-    struct msr_policy *mp = &host_cpu_policy;
+    struct cpu_policy *mp = &host_cpu_policy;
 
     *mp = raw_cpu_policy;
 
@@ -71,7 +71,7 @@ static void __init calculate_host_policy(void)
 
 static void __init calculate_pv_max_policy(void)
 {
-    struct msr_policy *mp = &pv_max_cpu_policy;
+    struct cpu_policy *mp = &pv_max_cpu_policy;
 
     *mp = host_cpu_policy;
 
@@ -80,14 +80,14 @@ static void __init calculate_pv_max_policy(void)
 
 static void __init calculate_pv_def_policy(void)
 {
-    struct msr_policy *mp = &pv_def_cpu_policy;
+    struct cpu_policy *mp = &pv_def_cpu_policy;
 
     *mp = pv_max_cpu_policy;
 }
 
 static void __init calculate_hvm_max_policy(void)
 {
-    struct msr_policy *mp = &hvm_max_cpu_policy;
+    struct cpu_policy *mp = &hvm_max_cpu_policy;
 
     *mp = host_cpu_policy;
 
@@ -99,7 +99,7 @@ static void __init calculate_hvm_max_policy(void)
 
 static void __init calculate_hvm_def_policy(void)
 {
-    struct msr_policy *mp = &hvm_def_cpu_policy;
+    struct cpu_policy *mp = &hvm_def_cpu_policy;
 
     *mp = hvm_max_cpu_policy;
 }
@@ -138,7 +138,7 @@ int guest_rdmsr(struct vcpu *v, uint32_t msr, uint64_t *val)
 {
     const struct vcpu *curr = current;
     const struct domain *d = v->domain;
-    const struct cpuid_policy *cp = d->arch.cpuid;
+    const struct cpu_policy *cp = d->arch.cpu_policy;
     const struct vcpu_msrs *msrs = v->arch.msrs;
     int ret = X86EMUL_OKAY;
 
@@ -409,7 +409,7 @@ int guest_rdmsr(struct vcpu *v, uint32_t msr, uint64_t *val)
  * separate CPUID features for this functionality, but only set will be
  * active.
  */
-uint64_t msr_spec_ctrl_valid_bits(const struct cpuid_policy *cp)
+uint64_t msr_spec_ctrl_valid_bits(const struct cpu_policy *cp)
 {
     bool ssbd = cp->feat.ssbd || cp->extd.amd_ssbd;
     bool psfd = cp->feat.intel_psfd || cp->extd.psfd;
@@ -428,7 +428,7 @@ int guest_wrmsr(struct vcpu *v, uint32_t msr, uint64_t val)
 {
     const struct vcpu *curr = current;
     struct domain *d = v->domain;
-    const struct cpuid_policy *cp = d->arch.cpuid;
+    const struct cpu_policy *cp = d->arch.cpu_policy;
     struct vcpu_msrs *msrs = v->arch.msrs;
     int ret = X86EMUL_OKAY;
 
@@ -469,7 +469,7 @@ int guest_wrmsr(struct vcpu *v, uint32_t msr, uint64_t val)
          * for backwards compatiblity, the OS should write 0 to it before
          * trying to access the current microcode version.
          */
-        if ( d->arch.cpuid->x86_vendor != X86_VENDOR_INTEL || val != 0 )
+        if ( d->arch.cpu_policy->x86_vendor != X86_VENDOR_INTEL || val != 0 )
             goto gp_fault;
         break;
 
@@ -479,7 +479,7 @@ int guest_wrmsr(struct vcpu *v, uint32_t msr, uint64_t val)
          * to AMD CPUs as well (at least the architectural/CPUID part does).
          */
         if ( is_pv_domain(d) ||
-             d->arch.cpuid->x86_vendor != X86_VENDOR_AMD )
+             d->arch.cpu_policy->x86_vendor != X86_VENDOR_AMD )
             goto gp_fault;
         break;
 
@@ -491,7 +491,7 @@ int guest_wrmsr(struct vcpu *v, uint32_t msr, uint64_t val)
          * by any CPUID bit.
          */
         if ( is_pv_domain(d) ||
-             d->arch.cpuid->x86_vendor != X86_VENDOR_INTEL )
+             d->arch.cpu_policy->x86_vendor != X86_VENDOR_INTEL )
             goto gp_fault;
         break;
 
xen/arch/x86/pv/domain.c +1 -1
index f94f28c8e271..dbda57894543 100644
@@ -145,7 +145,7 @@ static void release_compat_l4(struct vcpu *v)
 
 unsigned long pv_fixup_guest_cr4(const struct vcpu *v, unsigned long cr4)
 {
-    const struct cpuid_policy *p = v->domain->arch.cpuid;
+    const struct cpu_policy *p = v->domain->arch.cpu_policy;
 
     /* Discard attempts to set guest controllable bits outside of the policy. */
     cr4 &= ~((p->basic.tsc     ? 0 : X86_CR4_TSD)      |
xen/arch/x86/pv/emul-priv-op.c +3 -3
index 5da00e24e4ff..04416f197951 100644
@@ -885,7 +885,7 @@ static int cf_check read_msr(
 {
     struct vcpu *curr = current;
     const struct domain *currd = curr->domain;
-    const struct cpuid_policy *cp = currd->arch.cpuid;
+    const struct cpu_policy *cp = currd->arch.cpu_policy;
     bool vpmu_msr = false, warn = false;
     uint64_t tmp;
     int ret;
@@ -1034,7 +1034,7 @@ static int cf_check write_msr(
 {
     struct vcpu *curr = current;
     const struct domain *currd = curr->domain;
-    const struct cpuid_policy *cp = currd->arch.cpuid;
+    const struct cpu_policy *cp = currd->arch.cpu_policy;
     bool vpmu_msr = false;
     int ret;
 
@@ -1327,7 +1327,7 @@ int pv_emulate_privileged_op(struct cpu_user_regs *regs)
     struct domain *currd = curr->domain;
     struct priv_op_ctxt ctxt = {
         .ctxt.regs = regs,
-        .ctxt.cpuid = currd->arch.cpuid,
+        .ctxt.cpu_policy = currd->arch.cpu_policy,
         .ctxt.lma = !is_pv_32bit_domain(currd),
     };
     int rc;
xen/arch/x86/pv/ro-page-fault.c +1 -1
index 5963f5ee2d51..0d02c7d2ab10 100644
@@ -356,7 +356,7 @@ int pv_ro_page_fault(unsigned long addr, struct cpu_user_regs *regs)
     unsigned int addr_size = is_pv_32bit_domain(currd) ? 32 : BITS_PER_LONG;
     struct x86_emulate_ctxt ctxt = {
         .regs      = regs,
-        .cpuid     = currd->arch.cpuid,
+        .cpu_policy = currd->arch.cpu_policy,
         .addr_size = addr_size,
         .sp_size   = addr_size,
         .lma       = addr_size > 32,
xen/arch/x86/sysctl.c +2 -2
index 3b2efecede2f..14d3dbca147b 100644
@@ -295,7 +295,7 @@ long arch_do_sysctl(
 
     case XEN_SYSCTL_get_cpu_featureset:
     {
-        static const struct cpuid_policy *const policy_table[6] = {
+        static const struct cpu_policy *policy_table[6] = {
             [XEN_SYSCTL_cpu_featureset_raw]     = &raw_cpu_policy,
             [XEN_SYSCTL_cpu_featureset_host]    = &host_cpu_policy,
 #ifdef CONFIG_PV
@@ -307,7 +307,7 @@ long arch_do_sysctl(
             [XEN_SYSCTL_cpu_featureset_hvm_max] = &hvm_max_cpu_policy,
 #endif
         };
-        const struct cpuid_policy *p = NULL;
+        const struct cpu_policy *p = NULL;
         uint32_t featureset[FSCAPINTS];
         unsigned int nr;
 
xen/arch/x86/traps.c +1 -1
index cade9e12f8fa..d12004b1c6fc 100644
@@ -1035,7 +1035,7 @@ void cpuid_hypervisor_leaves(const struct vcpu *v, uint32_t leaf,
                              uint32_t subleaf, struct cpuid_leaf *res)
 {
     const struct domain *d = v->domain;
-    const struct cpuid_policy *p = d->arch.cpuid;
+    const struct cpu_policy *p = d->arch.cpu_policy;
     uint32_t base = is_viridian_domain(d) ? 0x40000100 : 0x40000000;
     uint32_t idx  = leaf - base;
     unsigned int limit = is_viridian_domain(d) ? p->hv2_limit : p->hv_limit;
xen/arch/x86/x86_emulate.c +1 -1
index 8c7d18521807..ff94758e0ebe 100644
@@ -74,7 +74,7 @@ int cf_check x86emul_read_xcr(
         return X86EMUL_OKAY;
 
     case 1:
-        if ( current->domain->arch.cpuid->xstate.xgetbv1 )
+        if ( current->domain->arch.cpu_policy->xstate.xgetbv1 )
             break;
         /* fall through */
     default:
xen/arch/x86/x86_emulate/x86_emulate.c +83 -83

198 lines not on this page · raw message

xen/arch/x86/x86_emulate/x86_emulate.h +2 -2
index 75015104fbdb..8c98a398643a 100644
@@ -565,8 +565,8 @@ struct x86_emulate_ctxt
      * Input-only state:
      */
 
-    /* CPUID Policy for the domain. */
-    const struct cpuid_policy *cpuid;
+    /* CPU Policy for the domain. */
+    const struct cpu_policy *cpu_policy;
 
     /* Set this if writes may have side effects. */
     bool force_writeback;
xen/arch/x86/xstate.c +2 -2
index d481e1db3e7e..22038a251f6d 100644
@@ -684,7 +684,7 @@ void xstate_init(struct cpuinfo_x86 *c)
 int validate_xstate(const struct domain *d, uint64_t xcr0, uint64_t xcr0_accum,
                     const struct xsave_hdr *hdr)
 {
-    uint64_t xcr0_max = cpuid_policy_xcr0_max(d->arch.cpuid);
+    uint64_t xcr0_max = cpuid_policy_xcr0_max(d->arch.cpu_policy);
     unsigned int i;
 
     if ( (hdr->xstate_bv & ~xcr0_accum) ||
@@ -708,7 +708,7 @@ 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;
-    uint64_t xcr0_max = cpuid_policy_xcr0_max(curr->domain->arch.cpuid);
+    uint64_t xcr0_max = cpuid_policy_xcr0_max(curr->domain->arch.cpu_policy);
     u64 mask;
 
     if ( index != XCR_XFEATURE_ENABLED_MASK )
xen/include/xen/lib/x86/cpu-policy.h +12 -16
index 51f88f1e217e..82b4c1c0546f 100644
@@ -94,7 +94,7 @@ const char *x86_cpuid_vendor_to_str(unsigned int vendor);
                                       CPUID_GUEST_NR_EXTD_AMD)
 
 /*
- * Maximum number of leaves a struct cpuid_policy turns into when serialised
+ * Maximum number of leaves a struct cpu_policy turns into when serialised
  * for interaction with the toolstack.  (Sum of all leaves in each union, less
  * the entries in basic which sub-unions hang off of.)
  */
@@ -375,10 +375,6 @@ struct cpu_policy
     uint8_t x86_vendor;
 };
 
-/* Temporary */
-#define cpuid_policy cpu_policy
-#define msr_policy cpu_policy
-
 struct cpu_policy_errors
 {
     uint32_t leaf, subleaf;
@@ -389,7 +385,7 @@ struct cpu_policy_errors
 
 /* Fill in a featureset bitmap from a CPUID policy. */
 static inline void cpuid_policy_to_featureset(
-    const struct cpuid_policy *p, uint32_t fs[FEATURESET_NR_ENTRIES])
+    const struct cpu_policy *p, uint32_t fs[FEATURESET_NR_ENTRIES])
 {
     fs[FEATURESET_1d]  = p->basic._1d;
     fs[FEATURESET_1c]  = p->basic._1c;
@@ -411,7 +407,7 @@ static inline void cpuid_policy_to_featureset(
 
 /* Fill in a CPUID policy from a featureset bitmap. */
 static inline void cpuid_featureset_to_policy(
-    const uint32_t fs[FEATURESET_NR_ENTRIES], struct cpuid_policy *p)
+    const uint32_t fs[FEATURESET_NR_ENTRIES], struct cpu_policy *p)
 {
     p->basic._1d  = fs[FEATURESET_1d];
     p->basic._1c  = fs[FEATURESET_1c];
@@ -431,12 +427,12 @@ static inline void cpuid_featureset_to_policy(
     p->feat._7d1  = fs[FEATURESET_7d1];
 }
 
-static inline uint64_t cpuid_policy_xcr0_max(const struct cpuid_policy *p)
+static inline uint64_t cpuid_policy_xcr0_max(const struct cpu_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)
+static inline uint64_t cpuid_policy_xstates(const struct cpu_policy *p)
 {
     uint64_t val = p->xstate.xcr0_high | p->xstate.xss_high;
 
@@ -448,7 +444,7 @@ const uint32_t *x86_cpuid_lookup_deep_deps(uint32_t feature);
 /**
  * Recalculate the content in a CPUID policy which is derived from raw data.
  */
-void x86_cpuid_policy_recalc_synth(struct cpuid_policy *p);
+void x86_cpuid_policy_recalc_synth(struct cpu_policy *p);
 
 /**
  * Fill a CPUID policy using the native CPUID instruction.
@@ -457,7 +453,7 @@ void x86_cpuid_policy_recalc_synth(struct cpuid_policy *p);
  * Values may be influenced by a hypervisor or from masking/faulting
  * configuration.
  */
-void x86_cpuid_policy_fill_native(struct cpuid_policy *p);
+void x86_cpuid_policy_fill_native(struct cpu_policy *p);
 
 /**
  * Clear leaf data beyond the policies max leaf/subleaf settings.
@@ -468,7 +464,7 @@ void x86_cpuid_policy_fill_native(struct cpuid_policy *p);
  * with out-of-range leaves with stale content in them.  This helper clears
  * them.
  */
-void x86_cpuid_policy_clear_out_of_range_leaves(struct cpuid_policy *p);
+void x86_cpuid_policy_clear_out_of_range_leaves(struct cpu_policy *p);
 
 #ifdef __XEN__
 #include <public/arch-x86/xen.h>
@@ -492,7 +488,7 @@ typedef xen_msr_entry_t msr_entry_buffer_t[];
  * leaves array is too short.  On success, nr_entries is updated with the
  * actual number of leaves written.
  */
-int x86_cpuid_copy_to_buffer(const struct cpuid_policy *policy,
+int x86_cpuid_copy_to_buffer(const struct cpu_policy *policy,
                              cpuid_leaf_buffer_t leaves, uint32_t *nr_entries);
 
 /**
@@ -512,7 +508,7 @@ int x86_cpuid_copy_to_buffer(const struct cpuid_policy *policy,
  * No content validation of in-range leaves is performed.  Synthesised data is
  * recalculated.
  */
-int x86_cpuid_copy_from_buffer(struct cpuid_policy *policy,
+int x86_cpuid_copy_from_buffer(struct cpu_policy *policy,
                                const cpuid_leaf_buffer_t leaves,
                                uint32_t nr_entries, uint32_t *err_leaf,
                                uint32_t *err_subleaf);
@@ -529,7 +525,7 @@ int x86_cpuid_copy_from_buffer(struct cpuid_policy *policy,
  * buffer array is too short.  On success, nr_entries is updated with the
  * actual number of msrs written.
  */
-int x86_msr_copy_to_buffer(const struct msr_policy *policy,
+int x86_msr_copy_to_buffer(const struct cpu_policy *policy,
                            msr_entry_buffer_t msrs, uint32_t *nr_entries);
 
 /**
@@ -549,7 +545,7 @@ int x86_msr_copy_to_buffer(const struct msr_policy *policy,
  *
  * No content validation is performed on the data stored in the policy object.
  */
-int x86_msr_copy_from_buffer(struct msr_policy *policy,
+int x86_msr_copy_from_buffer(struct cpu_policy *policy,
                              const msr_entry_buffer_t msrs, uint32_t nr_entries,
                              uint32_t *err_msr);
 
xen/lib/x86/cpuid.c +5 -5
index e81f76c779c0..58a14163b1cb 100644
@@ -60,13 +60,13 @@ const char *x86_cpuid_vendor_to_str(unsigned int vendor)
     }
 }
 
-void x86_cpuid_policy_recalc_synth(struct cpuid_policy *p)
+void x86_cpuid_policy_recalc_synth(struct cpu_policy *p)
 {
     p->x86_vendor = x86_cpuid_lookup_vendor(
         p->basic.vendor_ebx, p->basic.vendor_ecx, p->basic.vendor_edx);
 }
 
-void x86_cpuid_policy_fill_native(struct cpuid_policy *p)
+void x86_cpuid_policy_fill_native(struct cpu_policy *p)
 {
     unsigned int i;
 
@@ -183,7 +183,7 @@ void x86_cpuid_policy_fill_native(struct cpuid_policy *p)
     x86_cpuid_policy_recalc_synth(p);
 }
 
-void x86_cpuid_policy_clear_out_of_range_leaves(struct cpuid_policy *p)
+void x86_cpuid_policy_clear_out_of_range_leaves(struct cpu_policy *p)
 {
     unsigned int i;
 
@@ -291,7 +291,7 @@ static int copy_leaf_to_buffer(uint32_t leaf, uint32_t subleaf,
     return 0;
 }
 
-int x86_cpuid_copy_to_buffer(const struct cpuid_policy *p,
+int x86_cpuid_copy_to_buffer(const struct cpu_policy *p,
                              cpuid_leaf_buffer_t leaves, uint32_t *nr_entries_p)
 {
     const uint32_t nr_entries = *nr_entries_p;
@@ -377,7 +377,7 @@ int x86_cpuid_copy_to_buffer(const struct cpuid_policy *p,
     return 0;
 }
 
-int x86_cpuid_copy_from_buffer(struct cpuid_policy *p,
+int x86_cpuid_copy_from_buffer(struct cpu_policy *p,
                                const cpuid_leaf_buffer_t leaves,
                                uint32_t nr_entries, uint32_t *err_leaf,
                                uint32_t *err_subleaf)
xen/lib/x86/msr.c +2 -2
index c4d885e7b568..e04b9ca01302 100644
@@ -23,7 +23,7 @@ static int copy_msr_to_buffer(uint32_t idx, uint64_t val,
     return 0;
 }
 
-int x86_msr_copy_to_buffer(const struct msr_policy *p,
+int x86_msr_copy_to_buffer(const struct cpu_policy *p,
                            msr_entry_buffer_t msrs, uint32_t *nr_entries_p)
 {
     const uint32_t nr_entries = *nr_entries_p;
@@ -48,7 +48,7 @@ int x86_msr_copy_to_buffer(const struct msr_policy *p,
     return 0;
 }
 
-int x86_msr_copy_from_buffer(struct msr_policy *p,
+int x86_msr_copy_from_buffer(struct cpu_policy *p,
                              const msr_entry_buffer_t msrs, uint32_t nr_entries,
                              uint32_t *err_msr)
 {
Signature · 1 line
2.30.2
#21 lore raw

Re: [PATCH 9/9] RFC: Everything else

Quoted 11 lines · Andrew Cooper

Looking at this diff, I'm wondering whether keeping

    union {
        struct cpu_policy *cpuid;
        struct cpu_policy *cpu_policy;
    };

permentantly might be a good idea, because d->arch.cpuid->$X reads rather better than d->arch.cpu_policy->$X

Thoughts?

I'm not overly fussed, but perhaps yes. Nevertheless e.g. ...

tools/fuzz/x86_instruction_emulator/fuzz-emul.c int LLVMFuzzerTestOneInput(const uint8_t *data_p, size_t size) · line 896

Quoted 8 lines
tools/fuzz/x86_instruction_emulator/fuzz-emul.c +1 -1
@@ -893,7 +893,7 @@ int LLVMFuzzerTestOneInput(const uint8_t *data_p, size_t size)
     struct x86_emulate_ctxt ctxt = {
         .data = &state,
         .regs = &input.regs,
-        .cpuid = &cp,
+        .cpu_policy = &cp,

... this and ...

tools/tests/x86_emulator/test_x86_emulator.c int main(int argc, char **argv) · line 912

Quoted 8 lines
tools/tests/x86_emulator/test_x86_emulator.c +1 -1
@@ -909,7 +909,7 @@ int main(int argc, char **argv)
 
     ctxt.regs = &regs;
     ctxt.force_writeback = 0;
-    ctxt.cpuid     = &cp;
+    ctxt.cpu_policy = &cp;

... this imo want keeping as you have it here.

Jan

#22 lore raw

Re: [PATCH 9/9] RFC: Everything else

Quoted 8 lines · Jan Beulich
Quoted 7 lines · Andrew Cooper

Looking at this diff, I'm wondering whether keeping

    union {
        struct cpu_policy *cpuid;
        struct cpu_policy *cpu_policy;
    };
Quoted 5 lines
Quoted 4 lines

permentantly might be a good idea, because d->arch.cpuid->$X reads rather better than d->arch.cpu_policy->$X

Thoughts?

I'm not overly fussed, but perhaps yes.

If we were to do this, we ought to keep d->arch.msr too for the same reason.

(Honestly - I'm still undecided on whether this is a good idea or not...)

Quoted 16 lines

Nevertheless e.g. ...

tools/fuzz/x86_instruction_emulator/fuzz-emul.c int LLVMFuzzerTestOneInput(const uint8_t *data_p, size_t size) · line 896

Quoted 8 lines
tools/fuzz/x86_instruction_emulator/fuzz-emul.c +1 -1
@@ -893,7 +893,7 @@ int LLVMFuzzerTestOneInput(const uint8_t *data_p, size_t size)
     struct x86_emulate_ctxt ctxt = {
         .data = &state,
         .regs = &input.regs,
-        .cpuid = &cp,
+        .cpu_policy = &cp,

... this and ...

Quoted 4 lines
tools/tests/x86_emulator/test_x86_emulator.c +0 -0
@@ -909,7 +909,7 @@ int main(int argc, char **argv)
 
Quoted 5 lines

tools/tests/x86_emulator/test_x86_emulator.c int main(int argc, char **argv) · line 912

Quoted 4 lines
tools/tests/x86_emulator/test_x86_emulator.c +1 -1
     ctxt.regs = &regs;
     ctxt.force_writeback = 0;
-    ctxt.cpuid     = &cp;
+    ctxt.cpu_policy = &cp;

... this imo want keeping as you have it here.

Ok, so that's a firm "switch to using cpu_policy for emul_ctxt" then.

Which is fine - in fact it's one I'd already started splitting out of this patch.

~Andrew

#23 lore raw

Re: [PATCH 9/9] RFC: Everything else

Quoted 9 lines · Andrew Cooper
Quoted 8 lines · Jan Beulich
Quoted 7 lines · Andrew Cooper

Looking at this diff, I'm wondering whether keeping

    union {
        struct cpu_policy *cpuid;
        struct cpu_policy *cpu_policy;
    };
Quoted 7 lines
Quoted 5 lines
Quoted 4 lines

permentantly might be a good idea, because d->arch.cpuid->$X reads rather better than d->arch.cpu_policy->$X

Thoughts?

I'm not overly fussed, but perhaps yes.

If we were to do this, we ought to keep d->arch.msr too for the same reason.

I'm not sure this is necessarily a consequence.

Quoted 24 lines

(Honestly - I'm still undecided on whether this is a good idea or not...)

Quoted 16 lines

Nevertheless e.g. ...

tools/fuzz/x86_instruction_emulator/fuzz-emul.c int LLVMFuzzerTestOneInput(const uint8_t *data_p, size_t size) · line 896

Quoted 8 lines
tools/fuzz/x86_instruction_emulator/fuzz-emul.c +1 -1
@@ -893,7 +893,7 @@ int LLVMFuzzerTestOneInput(const uint8_t *data_p, size_t size)
     struct x86_emulate_ctxt ctxt = {
         .data = &state,
         .regs = &input.regs,
-        .cpuid = &cp,
+        .cpu_policy = &cp,

... this and ...

Quoted 4 lines
tools/tests/x86_emulator/test_x86_emulator.c +0 -0
@@ -909,7 +909,7 @@ int main(int argc, char **argv)
 
Quoted 5 lines

tools/tests/x86_emulator/test_x86_emulator.c int main(int argc, char **argv) · line 912

Quoted 4 lines
tools/tests/x86_emulator/test_x86_emulator.c +1 -1
     ctxt.regs = &regs;
     ctxt.force_writeback = 0;
-    ctxt.cpuid     = &cp;
+    ctxt.cpu_policy = &cp;

... this imo want keeping as you have it here.

Quoted 4 lines

Ok, so that's a firm "switch to using cpu_policy for emul_ctxt" then.

Which is fine - in fact it's one I'd already started splitting out of this patch.

Hmm, wait - CPUID "basic" and "feat" and alike uses I still would prefer to see using "cpuid". It's really only such initializers which (imo even logically) want to use the name with the wider coverage.

Jan

#24 lore raw

Re: [PATCH 9/9] RFC: Everything else

Quoted 21 lines · Jan Beulich
Quoted 20 lines · Andrew Cooper
Quoted 19 lines · Jan Beulich

tools/fuzz/x86_instruction_emulator/fuzz-emul.c int LLVMFuzzerTestOneInput(const uint8_t *data_p, size_t size) · line 896

Quoted 8 lines
tools/fuzz/x86_instruction_emulator/fuzz-emul.c +1 -1
@@ -893,7 +893,7 @@ int LLVMFuzzerTestOneInput(const uint8_t *data_p, size_t size)
     struct x86_emulate_ctxt ctxt = {
         .data = &state,
         .regs = &input.regs,
-        .cpuid = &cp,
+        .cpu_policy = &cp,

... this and ...

tools/tests/x86_emulator/test_x86_emulator.c int main(int argc, char **argv) · line 912

Quoted 8 lines
tools/tests/x86_emulator/test_x86_emulator.c +1 -1
@@ -909,7 +909,7 @@ int main(int argc, char **argv)
 
     ctxt.regs = &regs;
     ctxt.force_writeback = 0;
-    ctxt.cpuid     = &cp;
+    ctxt.cpu_policy = &cp;

... this imo want keeping as you have it here.

Quoted 7 lines
Quoted 4 lines

Ok, so that's a firm "switch to using cpu_policy for emul_ctxt" then.

Which is fine - in fact it's one I'd already started splitting out of this patch.

Hmm, wait - CPUID "basic" and "feat" and alike uses I still would prefer to see using "cpuid". It's really only such initializers which (imo even logically) want to use the name with the wider coverage.

So its the other way around and you're saying you don't want the field name to change, and you don't want to see

-#define vcpu_has_fpu()         (ctxt->cpuid->basic.fpu) +#define vcpu_has_fpu()         (ctxt->cpu_policy->basic.fpu)

in the resulting diff ?

~Andrew

#25 lore raw

Re: [PATCH 9/9] RFC: Everything else

Quoted 36 lines · Andrew Cooper
Quoted 21 lines · Jan Beulich
Quoted 20 lines · Andrew Cooper
Quoted 19 lines · Jan Beulich

tools/fuzz/x86_instruction_emulator/fuzz-emul.c int LLVMFuzzerTestOneInput(const uint8_t *data_p, size_t size) · line 896

Quoted 8 lines
tools/fuzz/x86_instruction_emulator/fuzz-emul.c +1 -1
@@ -893,7 +893,7 @@ int LLVMFuzzerTestOneInput(const uint8_t *data_p, size_t size)
     struct x86_emulate_ctxt ctxt = {
         .data = &state,
         .regs = &input.regs,
-        .cpuid = &cp,
+        .cpu_policy = &cp,

... this and ...

tools/tests/x86_emulator/test_x86_emulator.c int main(int argc, char **argv) · line 912

Quoted 8 lines
tools/tests/x86_emulator/test_x86_emulator.c +1 -1
@@ -909,7 +909,7 @@ int main(int argc, char **argv)
 
     ctxt.regs = &regs;
     ctxt.force_writeback = 0;
-    ctxt.cpuid     = &cp;
+    ctxt.cpu_policy = &cp;

... this imo want keeping as you have it here.

Quoted 7 lines
Quoted 4 lines

Ok, so that's a firm "switch to using cpu_policy for emul_ctxt" then.

Which is fine - in fact it's one I'd already started splitting out of this patch.

Hmm, wait - CPUID "basic" and "feat" and alike uses I still would prefer to see using "cpuid". It's really only such initializers which (imo even logically) want to use the name with the wider coverage.

So its the other way around and you're saying you don't want the field name to change, and you don't want to see

(quoted diff) +1 -1
-#define vcpu_has_fpu()         (ctxt->cpuid->basic.fpu)
+#define vcpu_has_fpu()         (ctxt->cpu_policy->basic.fpu)
Quoted 1 line

in the resulting diff ?

Neither. I want the change as seen in the diff further up, but indeed I'd prefer if the two quoted diff lines in your most recent reply weren't there. This would again be by way of using a union, this time in struct x86_emulate_ctxt:

    /* CPU Policy for the domain. */
    union {
        const struct cpu_policy *cpuid;
        const struct cpu_policy *cpu_policy;
    };

Having said that: I realize that this is against what C mandates for the use of unions, but since we (ab)use unions in similar ways in many other places, I don't think we need to be concerned. Furthermore this specific use could, aiui, be "legalized" by making the declaration

    /* CPU Policy for the domain. */
    union {
        struct {
            const struct cpu_policy *cpuid;
        };
        struct {
            const struct cpu_policy *cpu_policy;
        };
    };

Jan

Reviewed-by on the v1 series
#26 lore raw
In reply to Andrew Cooper
Quoted 29 lines · Andrew Cooper

tl;dr to add MSR_ARCH_CAPS features sensibly, cpu_{featureset<->policy}() need to not operate on objects of differing lifetimes, so structs {cpuid,msr}_policy need merging and cpu_policy is the obvious name.

But this does mean that we now have

cpu_policy->basic.$X
cpu_policy->feat.$Y
cpu_policy->arch_caps.$Z

and plenty of code now reads

d->arch.cpu_policy->feat.$Y

instead of

d->arch.cpuid->feat.$Y

The latter can be half-fixed with some union magic (see patch 9 commit message). The former can be fixed by putting cpuid/msr infixes in cpu_policy, which is doable but very invasive, and would make plenty of code read

d->arch.cpu_policy->cpuid.feat.$Y

and the two obviously shouldn't be done together.

So, RFC. Does this code layout look ok? If we want to make changes with naming, now is very much the right time to get them sorted.

Quoted 7 lines

Patches 1-8 are pretty ready to go. Patch 9 is the remainder to take out the temporary hacks, and I'm still in the process of merging the system policy derivation.

Andrew Cooper (9):
x86: Rename struct cpu_policy to struct old_cpuid_policy
x86: Rename {domctl,sysctl}.cpu_policy.{cpuid,msr_policy} fields

Nit: I guess the last closing brace wants moving forward a little.

Quoted 6 lines

x86: Rename struct cpuid_policy to struct cpu_policy
x86: Merge struct msr_policy into struct cpu_policy
x86: Merge the system {cpuid,msr} policy objects
x86: Merge a domain's {cpuid,msr} policy objects
x86: Merge xc_cpu_policy's cpuid and msr objects
x86: Drop struct old_cpu_policy

With the small comments on individual patches taken care of one way or another, up to here:

Jan

Quoted 61 lines

RFC: Everything else

 tools/fuzz/cpu-policy/afl-policy-fuzzer.c     |  15 +-
 .../fuzz/x86_instruction_emulator/fuzz-emul.c |   2 +-
 tools/libs/guest/xg_cpuid_x86.c               |  48 +-
 tools/libs/guest/xg_private.h                 |   5 +-
 tools/tests/cpu-policy/test-cpu-policy.c      |  50 +-
 tools/tests/tsx/test-tsx.c                    |  58 +-
 tools/tests/x86_emulator/Makefile             |   2 +-
 tools/tests/x86_emulator/test_x86_emulator.c  |   2 +-
 tools/tests/x86_emulator/x86-emulate.c        |   2 +-
 tools/tests/x86_emulator/x86-emulate.h        |   2 +-
 xen/arch/x86/Makefile                         |   1 +
 xen/arch/x86/cpu-policy.c                     |  67 +++
 xen/arch/x86/cpu/common.c                     |   4 +-
 xen/arch/x86/cpu/mcheck/mce_intel.c           |   2 +-
 xen/arch/x86/cpu/vpmu_intel.c                 |   4 +-
 xen/arch/x86/cpuid.c                          | 101 ++--
 xen/arch/x86/domain.c                         |  18 +-
 xen/arch/x86/domctl.c                         |  51 +-
 xen/arch/x86/hvm/emulate.c                    |   2 +-
 xen/arch/x86/hvm/hvm.c                        |  38 +-
 xen/arch/x86/hvm/ioreq.c                      |   4 +-
 xen/arch/x86/hvm/mtrr.c                       |   2 +-
 xen/arch/x86/hvm/svm/svm.c                    |  18 +-
 xen/arch/x86/hvm/svm/svmdebug.c               |   2 +-
 xen/arch/x86/hvm/vlapic.c                     |   2 +-
 xen/arch/x86/hvm/vmx/vmx.c                    |  12 +-
 xen/arch/x86/hvm/vmx/vvmx.c                   |   2 +-
 xen/arch/x86/include/asm/cpu-policy.h         |  18 +
 xen/arch/x86/include/asm/cpuid.h              |  10 -
 xen/arch/x86/include/asm/domain.h             |   4 +-
 xen/arch/x86/include/asm/guest_pt.h           |   4 +-
 xen/arch/x86/include/asm/msr.h                |  13 +-
 xen/arch/x86/include/asm/paging.h             |   2 +-
 xen/arch/x86/mm/mem_sharing.c                 |   3 +-
 xen/arch/x86/mm/shadow/hvm.c                  |   2 +-
 xen/arch/x86/msr.c                            |  98 +---
 xen/arch/x86/pv/domain.c                      |   2 +-
 xen/arch/x86/pv/emul-priv-op.c                |   6 +-
 xen/arch/x86/pv/ro-page-fault.c               |   2 +-
 xen/arch/x86/sysctl.c                         |  77 +--
 xen/arch/x86/traps.c                          |   2 +-
 xen/arch/x86/x86_emulate.c                    |   2 +-
 xen/arch/x86/x86_emulate/x86_emulate.c        | 166 +++---
 xen/arch/x86/x86_emulate/x86_emulate.h        |   6 +-
 xen/arch/x86/xstate.c                         |   4 +-
 xen/include/public/domctl.h                   |   4 +-
 xen/include/public/sysctl.h                   |   4 +-
 xen/include/xen/lib/x86/cpu-policy.h          | 540 +++++++++++++++++-
 xen/include/xen/lib/x86/cpuid.h               | 475 ---------------
 xen/include/xen/lib/x86/msr.h                 | 104 ----
 xen/lib/x86/cpuid.c                           |  12 +-
 xen/lib/x86/msr.c                             |   6 +-
 xen/lib/x86/policy.c                          |   8 +-
 53 files changed, 986 insertions(+), 1104 deletions(-)
 create mode 100644 xen/arch/x86/cpu-policy.c
 create mode 100644 xen/arch/x86/include/asm/cpu-policy.h
 delete mode 100644 xen/include/xen/lib/x86/cpuid.h
 delete mode 100644 xen/include/xen/lib/x86/msr.h
#27 lore raw
Quoted 5 lines · Jan Beulich
Quoted 3 lines · Andrew Cooper

Andrew Cooper (9):
x86: Rename struct cpu_policy to struct old_cpuid_policy
x86: Rename {domctl,sysctl}.cpu_policy.{cpuid,msr_policy} fields

Nit: I guess the last closing brace wants moving forward a little.

Oops yes.

Quoted 10 lines
Quoted 6 lines

x86: Rename struct cpuid_policy to struct cpu_policy
x86: Merge struct msr_policy into struct cpu_policy
x86: Merge the system {cpuid,msr} policy objects
x86: Merge a domain's {cpuid,msr} policy objects
x86: Merge xc_cpu_policy's cpuid and msr objects
x86: Drop struct old_cpu_policy

With the small comments on individual patches taken care of one way or another, up to here:

Thanks.  I'll fold this in on these patches, but I don't intend to commit anything until I've got the full series together.

Right now there's still a lot of shuffling of minor things (comments) between patches, and not-so-minor things (deciding to split things differently).

~Andrew

#28 lore raw
In reply to Andrew Cooper
Quoted 3 lines · Andrew Cooper

tl;dr to add MSR_ARCH_CAPS features sensibly, cpu_{featureset<->policy}() need to not operate on objects of differing lifetimes, so structs {cpuid,msr}_policy need merging and cpu_policy is the obvious name.

So the problem is that there's a chance we might get a cpu_policy object that contains a valid (allocated) cpuid object, but not an msr one?

Or the problem is rather with the domain struct containing separate cpuid/msr fields not encapsulated in the cpu_policy struct?

I don't think the current set of cpu_policy operations permit you to operate on incomplete objects anyway.

I assume you are worried about the usage of x86_msr_copy_from_buffer() for example that could load data into the MSR_ARCH_CAPS field for the msr object without checking that the corresponding CPUID bit is set?

Quoted 5 lines

But this does mean that we now have

cpu_policy->basic.$X
cpu_policy->feat.$Y
cpu_policy->arch_caps.$Z

I'm not sure I like the fact that we now can't differentiate between policy fields related to MSRs or CPUID leafs.

Isn't there a chance we might in the future get some name space collision by us having decided to unify both?

Thanks, Roger.

#29 lore raw
Quoted 7 lines · Roger Pau Monné
Quoted 3 lines · Andrew Cooper

tl;dr to add MSR_ARCH_CAPS features sensibly, cpu_{featureset<->policy}() need to not operate on objects of differing lifetimes, so structs {cpuid,msr}_policy need merging and cpu_policy is the obvious name.

So the problem is that there's a chance we might get a cpu_policy object that contains a valid (allocated) cpuid object, but not an msr one?

No - not cpu_policy.  It is that we can get a cpuid_policy and an msr_policy that aren't at the same point in their lifecycle.

... which is exactly what happens right now for the raw/host msr right now if you featureset_to_policy() to include MSR data.

Merging the two together into cpu_policy causes there to be a single object lifecycle.

It's probably worth repeating the advise from the footnote in
https://lwn.net/Articles/193245/ again.  Get your datastructures right, and the code takes care of itself.  Don't get them right, and the code tends to be unmaintainable.

Quoted 10 lines
Quoted 5 lines

But this does mean that we now have

cpu_policy->basic.$X
cpu_policy->feat.$Y
cpu_policy->arch_caps.$Z

I'm not sure I like the fact that we now can't differentiate between policy fields related to MSRs or CPUID leafs.

Isn't there a chance we might in the future get some name space collision by us having decided to unify both?

The names are chosen by me so far, and the compiler will tell us if things actually collide.

And renaming the existing field is a perfectly acceptable way of resolving a conflict which arises in the future.

But yes - this was the whole point of asking the question.

~Andrew

#30 lore raw
Quoted 9 lines · Andrew Cooper
Quoted 7 lines · Roger Pau Monné
Quoted 3 lines · Andrew Cooper

tl;dr to add MSR_ARCH_CAPS features sensibly, cpu_{featureset<->policy}() need to not operate on objects of differing lifetimes, so structs {cpuid,msr}_policy need merging and cpu_policy is the obvious name.

So the problem is that there's a chance we might get a cpu_policy object that contains a valid (allocated) cpuid object, but not an msr one?

Quoted 5 lines

No - not cpu_policy.  It is that we can get a cpuid_policy and an msr_policy that aren't at the same point in their lifecycle.

... which is exactly what happens right now for the raw/host msr right now if you featureset_to_policy() to include MSR data.

I see, but that's mostly because we handle the featureset_to_policy() in two different places for CPUID vs MSR, those need to be unified into a single helper that does both at the same point.

I assume not having such pointers in side of cpu_policy makes it clearer that both msr and cpuid should be handled at the same time, but ultimately this would imply passing a cpu_policy object to featureset_to_policy() so that both CPUID and MSR sub-structs are filled from the same featureset.

Sorry, maybe I'm being a bit dull here, just would like to understand the motivation of the change.

Quoted 21 lines

Merging the two together into cpu_policy causes there to be a single object lifecycle.

It's probably worth repeating the advise from the footnote in
https://lwn.net/Articles/193245/ again.  Get your datastructures right, and the code takes care of itself.  Don't get them right, and the code tends to be unmaintainable.

Quoted 10 lines
Quoted 5 lines

But this does mean that we now have

cpu_policy->basic.$X
cpu_policy->feat.$Y
cpu_policy->arch_caps.$Z

I'm not sure I like the fact that we now can't differentiate between policy fields related to MSRs or CPUID leafs.

Isn't there a chance we might in the future get some name space collision by us having decided to unify both?

Quoted 7 lines

The names are chosen by me so far, and the compiler will tell us if things actually collide.

And renaming the existing field is a perfectly acceptable way of resolving a conflict which arises in the future.

But yes - this was the whole point of asking the question.

I think I would prefer to keep the cpu_policy->{cpuid,msr}. distinction if it doesn't interfere with further work.

Thanks, Roger.

#31 lore raw
Quoted 18 lines · Roger Pau Monné
Quoted 5 lines · Andrew Cooper
Quoted 4 lines · Roger Pau Monné
Quoted 3 lines · Andrew Cooper

tl;dr to add MSR_ARCH_CAPS features sensibly, cpu_{featureset<->policy}() need to not operate on objects of differing lifetimes, so structs {cpuid,msr}_policy need merging and cpu_policy is the obvious name.

Quoted 8 lines
Quoted 3 lines

So the problem is that there's a chance we might get a cpu_policy object that contains a valid (allocated) cpuid object, but not an msr one?

No - not cpu_policy.  It is that we can get a cpuid_policy and an msr_policy that aren't at the same point in their lifecycle.

... which is exactly what happens right now for the raw/host msr right now if you featureset_to_policy() to include MSR data.

I see, but that's mostly because we handle the featureset_to_policy() in two different places for CPUID vs MSR, those need to be unified into a single helper that does both at the same point.

Quoted 8 lines

I assume not having such pointers in side of cpu_policy makes it clearer that both msr and cpuid should be handled at the same time, but ultimately this would imply passing a cpu_policy object to featureset_to_policy() so that both CPUID and MSR sub-structs are filled from the same featureset.

Sorry, maybe I'm being a bit dull here, just would like to understand the motivation of the change.

That's pretty much it.  Forcing them to be one object removes a class of errors, and makes the resulting code easier to follow.

(Based on having tried to do the non-merged approach first, and deciding that it's not code I'm willing to try putting upstream...)

Quoted 23 lines
Quoted 23 lines

Merging the two together into cpu_policy causes there to be a single object lifecycle.

It's probably worth repeating the advise from the footnote in
https://lwn.net/Articles/193245/ again.  Get your datastructures right, and the code takes care of itself.  Don't get them right, and the code tends to be unmaintainable.

Quoted 10 lines
Quoted 5 lines

But this does mean that we now have

cpu_policy->basic.$X
cpu_policy->feat.$Y
cpu_policy->arch_caps.$Z

I'm not sure I like the fact that we now can't differentiate between policy fields related to MSRs or CPUID leafs.

Isn't there a chance we might in the future get some name space collision by us having decided to unify both?

The names are chosen by me so far, and the compiler will tell us if things actually collide.

Quoted 6 lines
Quoted 4 lines

And renaming the existing field is a perfectly acceptable way of resolving a conflict which arises in the future.

But yes - this was the whole point of asking the question.

I think I would prefer to keep the cpu_policy->{cpuid,msr}. distinction if it doesn't interfere with further work.

Unfortunately that's the opposite of what Jan asked for.  What I have done, based on the prior conversation is:

struct arch_domain {
    ...

    /*
     * The domain's CPU Policy.  "cpu_policy" is considered the canonical
     * pointer, but the "cpuid" and "msr" aliases exist so the most
     * appropriate one can be used for local code clarity.
     */
    union {
        struct cpu_policy *cpu_policy;
        struct cpu_policy *cpuid;
        struct cpu_policy *msr;
    };

So all the cases where you do have d->arch.cpuid.feat.$X, this continues to work.

The cases where you pull a cpu_policy out into a local variable, there will be no cpuid or msr infix, but these cases already have no cpuid/msr part to their naming.

~Andrew

#32 lore raw
Quoted 41 lines · Andrew Cooper
Quoted 16 lines · Roger Pau Monné
Quoted 15 lines · Andrew Cooper
Quoted 11 lines · Roger Pau Monné
Quoted 5 lines · Andrew Cooper

But this does mean that we now have

cpu_policy->basic.$X
cpu_policy->feat.$Y
cpu_policy->arch_caps.$Z

I'm not sure I like the fact that we now can't differentiate between policy fields related to MSRs or CPUID leafs.

Isn't there a chance we might in the future get some name space collision by us having decided to unify both?

The names are chosen by me so far, and the compiler will tell us if things actually collide.

Quoted 6 lines
Quoted 4 lines

And renaming the existing field is a perfectly acceptable way of resolving a conflict which arises in the future.

But yes - this was the whole point of asking the question.

I think I would prefer to keep the cpu_policy->{cpuid,msr}. distinction if it doesn't interfere with further work.

Unfortunately that's the opposite of what Jan asked for.  What I have done, based on the prior conversation is:

struct arch_domain {
    ...

    /*
     * The domain's CPU Policy.  "cpu_policy" is considered the canonical
     * pointer, but the "cpuid" and "msr" aliases exist so the most
     * appropriate one can be used for local code clarity.
     */
    union {
        struct cpu_policy *cpu_policy;
        struct cpu_policy *cpuid;
        struct cpu_policy *msr;
    };

Quoted 6 lines

So all the cases where you do have d->arch.cpuid.feat.$X, this continues to work.

The cases where you pull a cpu_policy out into a local variable, there will be no cpuid or msr infix, but these cases already have no cpuid/msr part to their naming.

I see. I'm fine with this. There's still the remote possibility of field name clash between cpuid and msr names, but we can likely sort this out if we ever get into this position.

Thanks, Roger.

#33 lore raw
Quoted 44 lines · Roger Pau Monné
Quoted 43 lines · Andrew Cooper
Quoted 16 lines · Roger Pau Monné
Quoted 15 lines · Andrew Cooper
Quoted 11 lines · Roger Pau Monné
Quoted 5 lines · Andrew Cooper

But this does mean that we now have

cpu_policy->basic.$X
cpu_policy->feat.$Y
cpu_policy->arch_caps.$Z

I'm not sure I like the fact that we now can't differentiate between policy fields related to MSRs or CPUID leafs.

Isn't there a chance we might in the future get some name space collision by us having decided to unify both?

The names are chosen by me so far, and the compiler will tell us if things actually collide.

Quoted 6 lines
Quoted 4 lines

And renaming the existing field is a perfectly acceptable way of resolving a conflict which arises in the future.

But yes - this was the whole point of asking the question.

I think I would prefer to keep the cpu_policy->{cpuid,msr}. distinction if it doesn't interfere with further work.

Unfortunately that's the opposite of what Jan asked for.  What I have done, based on the prior conversation is:

struct arch_domain {
    ...

    /*
     * The domain's CPU Policy.  "cpu_policy" is considered the canonical
     * pointer, but the "cpuid" and "msr" aliases exist so the most
     * appropriate one can be used for local code clarity.
     */
    union {
        struct cpu_policy *cpu_policy;
        struct cpu_policy *cpuid;
        struct cpu_policy *msr;
    };

So all the cases where you do have d->arch.cpuid.feat.$X, this continues to work.

Quoted 6 lines
Quoted 3 lines

The cases where you pull a cpu_policy out into a local variable, there will be no cpuid or msr infix, but these cases already have no cpuid/msr part to their naming.

I see. I'm fine with this. There's still the remote possibility of field name clash between cpuid and msr names, but we can likely sort this out if we ever get into this position.

Thanks.  Yeah, we can rename if things become problematic.

~Andrew