Skip to contentSkip to conversation

clean up MSR save/restore code

PATCH Superseded Paul Durrant 4 participants 26 messages Jan 7, 2019 – Mar 11, 2019

v1 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

cover clean up MSR save/restore code

Patch #6 is not strictly related to save/restore. It's just clean-up of something I noticed along the way.

Paul Durrant (6):
x86: stop handling MSR_IA32_BNDCFGS save/restore in implementation
code
x86: save GUEST_BNDCFGS on context switch...
x86: move the saved value of MSR_IA32_XSS into struct vcpu_msrs
x86: stop handling MSR_IA32_XSS save/restore in implementation code
x86: remove defunct init/load/save_msr() hvm_funcs
x86: introduce dr_mask_idx() helper function...

 xen/arch/x86/domain.c          |   2 +-
 xen/arch/x86/hvm/hvm.c         | 102 ++++++++++++++++-----------------
 xen/arch/x86/hvm/vmx/vmx.c     |  83 +++------------------------
 xen/arch/x86/msr.c             |  48 +++++++++++-----
 xen/arch/x86/pv/emul-priv-op.c |   2 +-
 xen/include/asm-x86/hvm/hvm.h  |  17 ++----
 xen/include/asm-x86/hvm/vcpu.h |   1 -
 xen/include/asm-x86/msr.h      |  33 ++++++++++-
 8 files changed, 133 insertions(+), 155 deletions(-)

---
Cc: Andrew Cooper <[email protected]>
Cc: Jan Beulich <[email protected]>
Cc: Jun Nakajima <[email protected]>
Cc: Kevin Tian <[email protected]>
Cc: "Roger Pau Monné" <[email protected]>
Cc: Wei Liu <[email protected]>

Signature · 7 lines
2.20.1


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
#1 lore raw

PATCH 1/6 x86: stop handling MSR_IA32_BNDCFGS save/restore in implementation code

Saving and restoring the value of this MSR is currently handled by implementation-specific code despite it being architectural. This patch moves handling of accesses to this MSR from hvm.c into the msr.c, thus allowing the common MSR save/restore code to handle it.

This patch also changes hvm_get/set_guest_bndcfgs() to check CPUID policy for the appropriate feature, rather than hardware, and also re-works the get/set_guest_bndcfgs() hvm_funcs so they are no longer boolean. Uses of u64 are also converted to uint64_t.

NOTE: Because vmx_get/set_guest_bndcfgs() call vmx_vmcs_enter(), the
struct vcpu pointer passed in cannot be const.

---
Cc: Jan Beulich <[email protected]>
Cc: Andrew Cooper <[email protected]>
Cc: Wei Liu <[email protected]>
Cc: "Roger Pau Monné" <[email protected]>
Cc: Jun Nakajima <[email protected]>
Cc: Kevin Tian <[email protected]>

---
 xen/arch/x86/hvm/hvm.c         | 44 +++++++++++++++++++++-------------
 xen/arch/x86/hvm/vmx/vmx.c     | 31 +++++-------------------
 xen/arch/x86/msr.c             | 14 ++++++++++-
 xen/arch/x86/pv/emul-priv-op.c |  2 +-
 xen/include/asm-x86/hvm/hvm.h  | 13 ++++------
 xen/include/asm-x86/msr.h      |  2 +-
 6 files changed, 52 insertions(+), 54 deletions(-)
xen/arch/x86/hvm/hvm.c +27 -17
index 401c4a9312..5fd5478b7d 100644
@@ -308,11 +308,16 @@ int hvm_set_guest_pat(struct vcpu *v, u64 guest_pat)
     return 1;
 }
 
-bool hvm_set_guest_bndcfgs(struct vcpu *v, u64 val)
+bool hvm_set_guest_bndcfgs(struct vcpu *v, uint64_t val)
 {
-    if ( !hvm_funcs.set_guest_bndcfgs ||
-         !is_canonical_address(val) ||
-         (val & IA32_BNDCFGS_RESERVED) )
+    const struct cpuid_policy *cp = v->domain->arch.cpuid;
+
+    if ( !cp->feat.mpx )
+        return false;
+
+    ASSERT(hvm_funcs.set_guest_bndcfgs);
+
+    if ( !is_canonical_address(val) || (val & IA32_BNDCFGS_RESERVED) )
         return false;
 
     /*
@@ -342,7 +347,22 @@ bool hvm_set_guest_bndcfgs(struct vcpu *v, u64 val)
             /* nothing, best effort only */;
     }
 
-    return hvm_funcs.set_guest_bndcfgs(v, val);
+    hvm_funcs.set_guest_bndcfgs(v, val);
+
+    return true;
+}
+
+bool hvm_get_guest_bndcfgs(struct vcpu *v, uint64_t *val)
+{
+    const struct cpuid_policy *cp = v->domain->arch.cpuid;
+
+    if ( !cp->feat.mpx )
+        return false;
+
+    ASSERT(hvm_funcs.get_guest_bndcfgs);
+    *val = hvm_funcs.get_guest_bndcfgs(v);
+
+    return true;
 }
 
 /*
@@ -1312,6 +1332,7 @@ static int hvm_load_cpu_xsave_states(struct domain *d, hvm_domain_context_t *h)
 static const uint32_t msrs_to_send[] = {
     MSR_SPEC_CTRL,
     MSR_INTEL_MISC_FEATURES_ENABLES,
+    MSR_IA32_BNDCFGS,
     MSR_AMD64_DR0_ADDRESS_MASK,
     MSR_AMD64_DR1_ADDRESS_MASK,
     MSR_AMD64_DR2_ADDRESS_MASK,
@@ -1449,6 +1470,7 @@ static int hvm_load_cpu_msrs(struct domain *d, hvm_domain_context_t *h)
 
         case MSR_SPEC_CTRL:
         case MSR_INTEL_MISC_FEATURES_ENABLES:
+        case MSR_IA32_BNDCFGS:
         case MSR_AMD64_DR0_ADDRESS_MASK:
         case MSR_AMD64_DR1_ADDRESS_MASK ... MSR_AMD64_DR3_ADDRESS_MASK:
             rc = guest_wrmsr(v, ctxt->msr[i].index, ctxt->msr[i].val);
@@ -3472,12 +3494,6 @@ int hvm_msr_read_intercept(unsigned int msr, uint64_t *msr_content)
         *msr_content = v->arch.hvm.msr_xss;
         break;
 
-    case MSR_IA32_BNDCFGS:
-        if ( !d->arch.cpuid->feat.mpx ||
-             !hvm_get_guest_bndcfgs(v, msr_content) )
-            goto gp_fault;
-        break;
-
     case MSR_K8_ENABLE_C1E:
     case MSR_AMD64_NB_CFG:
          /*
@@ -3624,12 +3640,6 @@ int hvm_msr_write_intercept(unsigned int msr, uint64_t msr_content,
         v->arch.hvm.msr_xss = msr_content;
         break;
 
-    case MSR_IA32_BNDCFGS:
-        if ( !d->arch.cpuid->feat.mpx ||
-             !hvm_set_guest_bndcfgs(v, msr_content) )
-            goto gp_fault;
-        break;
-
     case MSR_AMD64_NB_CFG:
         /* ignore the write */
         break;
xen/arch/x86/hvm/vmx/vmx.c +6 -25
index 64af8bf943..4bfabe8d0e 100644
@@ -805,17 +805,6 @@ static unsigned int __init vmx_init_msr(void)
 
 static void vmx_save_msr(struct vcpu *v, struct hvm_msr *ctxt)
 {
-    vmx_vmcs_enter(v);
-
-    if ( cpu_has_mpx && cpu_has_vmx_mpx )
-    {
-        __vmread(GUEST_BNDCFGS, &ctxt->msr[ctxt->count].val);
-        if ( ctxt->msr[ctxt->count].val )
-            ctxt->msr[ctxt->count++].index = MSR_IA32_BNDCFGS;
-    }
-
-    vmx_vmcs_exit(v);
-
     if ( cpu_has_xsaves && cpu_has_vmx_xsaves )
     {
         ctxt->msr[ctxt->count].val = v->arch.hvm.msr_xss;
@@ -835,14 +824,6 @@ static int vmx_load_msr(struct vcpu *v, struct hvm_msr *ctxt)
     {
         switch ( ctxt->msr[i].index )
         {
-        case MSR_IA32_BNDCFGS:
-            if ( cpu_has_mpx && cpu_has_vmx_mpx &&
-                 is_canonical_address(ctxt->msr[i].val) &&
-                 !(ctxt->msr[i].val & IA32_BNDCFGS_RESERVED) )
-                __vmwrite(GUEST_BNDCFGS, ctxt->msr[i].val);
-            else if ( ctxt->msr[i].val )
-                err = -ENXIO;
-            break;
         case MSR_IA32_XSS:
             if ( cpu_has_xsaves && cpu_has_vmx_xsaves )
                 v->arch.hvm.msr_xss = ctxt->msr[i].val;
@@ -1204,26 +1185,26 @@ static int vmx_get_guest_pat(struct vcpu *v, u64 *gpat)
     return 1;
 }
 
-static bool vmx_set_guest_bndcfgs(struct vcpu *v, u64 val)
+static void vmx_set_guest_bndcfgs(struct vcpu *v, uint64_t val)
 {
     ASSERT(cpu_has_mpx && cpu_has_vmx_mpx);
 
     vmx_vmcs_enter(v);
     __vmwrite(GUEST_BNDCFGS, val);
     vmx_vmcs_exit(v);
-
-    return true;
 }
 
-static bool vmx_get_guest_bndcfgs(struct vcpu *v, u64 *val)
+static uint64_t vmx_get_guest_bndcfgs(struct vcpu *v)
 {
+    uint64_t val;
+
     ASSERT(cpu_has_mpx && cpu_has_vmx_mpx);
 
     vmx_vmcs_enter(v);
-    __vmread(GUEST_BNDCFGS, val);
+    __vmread(GUEST_BNDCFGS, &val);
     vmx_vmcs_exit(v);
 
-    return true;
+    return val;
 }
 
 static void vmx_handle_cd(struct vcpu *v, unsigned long value)
xen/arch/x86/msr.c +13 -1
index 9bb38b6d66..a3406c29a8 100644
@@ -115,7 +115,7 @@ int init_vcpu_msr_policy(struct vcpu *v)
     return 0;
 }
 
-int guest_rdmsr(const struct vcpu *v, uint32_t msr, uint64_t *val)
+int guest_rdmsr(struct vcpu *v, uint32_t msr, uint64_t *val)
 {
     const struct vcpu *curr = current;
     const struct domain *d = v->domain;
@@ -158,6 +158,12 @@ int guest_rdmsr(const struct vcpu *v, uint32_t msr, uint64_t *val)
         ret = guest_rdmsr_x2apic(v, msr, val);
         break;
 
+    case MSR_IA32_BNDCFGS:
+        if ( !is_hvm_domain(d) || !hvm_get_guest_bndcfgs(v, val) )
+            goto gp_fault;
+
+        break;
+
     case 0x40000000 ... 0x400001ff:
         if ( is_viridian_domain(d) )
         {
@@ -319,6 +325,12 @@ int guest_wrmsr(struct vcpu *v, uint32_t msr, uint64_t val)
         ret = guest_wrmsr_x2apic(v, msr, val);
         break;
 
+    case MSR_IA32_BNDCFGS:
+        if ( !is_hvm_domain(d) || !hvm_set_guest_bndcfgs(v, val) )
+            goto gp_fault;
+
+        break;
+
     case 0x40000000 ... 0x400001ff:
         if ( is_viridian_domain(d) )
         {
xen/arch/x86/pv/emul-priv-op.c +1 -1
index 942ece2ca0..678dad3792 100644
@@ -807,7 +807,7 @@ static inline bool is_cpufreq_controller(const struct domain *d)
 static int read_msr(unsigned int reg, uint64_t *val,
                     struct x86_emulate_ctxt *ctxt)
 {
-    const struct vcpu *curr = current;
+    struct vcpu *curr = current;
     const struct domain *currd = curr->domain;
     bool vpmu_msr = false;
     int ret;
xen/include/asm-x86/hvm/hvm.h +4 -9
index 0a10b51554..5c8237e087 100644
@@ -149,8 +149,8 @@ struct hvm_function_table {
     int  (*get_guest_pat)(struct vcpu *v, u64 *);
     int  (*set_guest_pat)(struct vcpu *v, u64);
 
-    bool (*get_guest_bndcfgs)(struct vcpu *v, u64 *);
-    bool (*set_guest_bndcfgs)(struct vcpu *v, u64);
+    uint64_t (*get_guest_bndcfgs)(struct vcpu *v);
+    void (*set_guest_bndcfgs)(struct vcpu *v, uint64_t);
 
     void (*set_tsc_offset)(struct vcpu *v, u64 offset, u64 at_tsc);
 
@@ -283,8 +283,6 @@ void hvm_get_segment_register(struct vcpu *v, enum x86_segment seg,
 void hvm_set_segment_register(struct vcpu *v, enum x86_segment seg,
                               struct segment_register *reg);
 
-bool hvm_set_guest_bndcfgs(struct vcpu *v, u64 val);
-
 bool hvm_check_cpuid_faulting(struct vcpu *v);
 void hvm_migrate_timers(struct vcpu *v);
 void hvm_do_resume(struct vcpu *v);
@@ -446,11 +444,8 @@ static inline unsigned long hvm_get_shadow_gs_base(struct vcpu *v)
     return hvm_funcs.get_shadow_gs_base(v);
 }
 
-static inline bool hvm_get_guest_bndcfgs(struct vcpu *v, u64 *val)
-{
-    return hvm_funcs.get_guest_bndcfgs &&
-           hvm_funcs.get_guest_bndcfgs(v, val);
-}
+bool hvm_get_guest_bndcfgs(struct vcpu *v, uint64_t *val);
+bool hvm_set_guest_bndcfgs(struct vcpu *v, uint64_t val);
 
 #define has_hvm_params(d) \
     ((d)->arch.hvm.params != NULL)
xen/include/asm-x86/msr.h +1 -1
index adfa2fa05b..ad8688a61f 100644
@@ -320,7 +320,7 @@ int init_vcpu_msr_policy(struct vcpu *v);
  * These functions are also used by the migration logic, so need to cope with
  * being used outside of v's context.
  */
-int guest_rdmsr(const struct vcpu *v, uint32_t msr, uint64_t *val);
+int guest_rdmsr(struct vcpu *v, uint32_t msr, uint64_t *val);
 int guest_wrmsr(struct vcpu *v, uint32_t msr, uint64_t val);
 
 #endif /* !__ASSEMBLY__ */
Signature · 7 lines
2.20.1


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Reviewed-by on v1 1/6
#2 lore raw

Re: [PATCH 1/6] x86: stop handling MSR_IA32_BNDCFGS save/restore in implementation code

Quoted 14 lines

From: Paul Durrant [mailto:[email protected]]
Sent: Monday, January 7, 2019 8:03 PM

Saving and restoring the value of this MSR is currently handled by implementation-specific code despite it being architectural. This patch moves handling of accesses to this MSR from hvm.c into the msr.c, thus allowing the common MSR save/restore code to handle it.

This patch also changes hvm_get/set_guest_bndcfgs() to check CPUID policy for the appropriate feature, rather than hardware, and also re-works the get/set_guest_bndcfgs() hvm_funcs so they are no longer boolean. Uses
of u64 are also converted to uint64_t.

Quoted 4 lines

NOTE: Because vmx_get/set_guest_bndcfgs() call vmx_vmcs_enter(), the
struct vcpu pointer passed in cannot be const.

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#3 lore raw

Re: [PATCH 1/6] x86: stop handling MSR_IA32_BNDCFGS save/restore in implementation code

In reply to Paul Durrant
Quoted 6 lines
Quoted 1 line
Quoted 1 line · On 07.01.19 at 13:02, <[email protected]> wrote:
xen/arch/x86/hvm/hvm.c +0 -0
@@ -308,11 +308,16 @@ int hvm_set_guest_pat(struct vcpu *v, u64 guest_pat)
     return 1;
 }

xen/arch/x86/hvm/hvm.c int hvm_set_guest_pat(struct vcpu *v, u64 guest_pat) · line 315

Quoted 10 lines
xen/arch/x86/hvm/hvm.c +4 -4
 
-bool hvm_set_guest_bndcfgs(struct vcpu *v, u64 val)
+bool hvm_set_guest_bndcfgs(struct vcpu *v, uint64_t val)
 {
-    if ( !hvm_funcs.set_guest_bndcfgs ||
-         !is_canonical_address(val) ||
-         (val & IA32_BNDCFGS_RESERVED) )
+    const struct cpuid_policy *cp = v->domain->arch.cpuid;
+
+    if ( !cp->feat.mpx )

Is the local variable really worth it?

xen/arch/x86/hvm/hvm.c int hvm_set_guest_pat(struct vcpu *v, u64 guest_pat) · line 318

Quoted 3 lines
xen/arch/x86/hvm/hvm.c +3 -0
+        return false;
+
+    ASSERT(hvm_funcs.set_guest_bndcfgs);

I think it would be better (more safe for release builds) if this was left as part of the if(). Otherwise, ...

xen/arch/x86/hvm/hvm.c bool hvm_set_guest_bndcfgs(struct vcpu *v, u64 val) · line 350

Quoted 6 lines
xen/arch/x86/hvm/hvm.c +1 -1
@@ -342,7 +347,22 @@ bool hvm_set_guest_bndcfgs(struct vcpu *v, u64 val)
             /* nothing, best effort only */;
     }
 
-    return hvm_funcs.set_guest_bndcfgs(v, val);
+    hvm_funcs.set_guest_bndcfgs(v, val);

... you risk calling NULL here.

xen/arch/x86/hvm/hvm.c int hvm_msr_read_intercept(unsigned int msr, uint64_t *msr_content) · line 3479

Quoted 9 lines
xen/arch/x86/hvm/hvm.c +0 -5
@@ -3472,12 +3494,6 @@ int hvm_msr_read_intercept(unsigned int msr, uint64_t *msr_content)
         *msr_content = v->arch.hvm.msr_xss;
         break;
 
-    case MSR_IA32_BNDCFGS:
-        if ( !d->arch.cpuid->feat.mpx ||
-             !hvm_get_guest_bndcfgs(v, msr_content) )
-            goto gp_fault;
-        break;

Note how this already checks the CPUID policy, somewhat contrary to what you say in the commit message. I think this wants to remain this way in guest_{rd,wr}msr() - see all the other CPUID policy checks there. That way hvm_get_guest_bndcfgs() can be made consistent with the hook pointer type again, and it can remain an inline function as before.

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#4 lore raw

Re: [PATCH 1/6] x86: stop handling MSR_IA32_BNDCFGS save/restore in implementation code

Quoted 16 lines

-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 08 March 2019 16:40
To: Paul Durrant <[email protected]>
Cc: Andrew Cooper <[email protected]>; Roger Pau Monne <[email protected]>; Wei Liu <[email protected]>; Jun Nakajima <[email protected]>; Kevin Tian <[email protected]>; xen- devel <[email protected]>
Subject: Re: [PATCH 1/6] x86: stop handling MSR_IA32_BNDCFGS save/restore in implementation code

Quoted 7 lines
Quoted 1 line
Quoted 1 line · On 07.01.19 at 13:02, <[email protected]> wrote:
xen/arch/x86/hvm/hvm.c +0 -0
@@ -308,11 +308,16 @@ int hvm_set_guest_pat(struct vcpu *v, u64 guest_pat)
     return 1;
 }
 
Quoted 11 lines

xen/arch/x86/hvm/hvm.c int hvm_set_guest_pat(struct vcpu *v, u64 guest_pat) · line 315

Quoted 9 lines
xen/arch/x86/hvm/hvm.c +4 -4
-bool hvm_set_guest_bndcfgs(struct vcpu *v, u64 val)
+bool hvm_set_guest_bndcfgs(struct vcpu *v, uint64_t val)
 {
-    if ( !hvm_funcs.set_guest_bndcfgs ||
-         !is_canonical_address(val) ||
-         (val & IA32_BNDCFGS_RESERVED) )
+    const struct cpuid_policy *cp = v->domain->arch.cpuid;
+
+    if ( !cp->feat.mpx )

Is the local variable really worth it?

Maybe not.

Quoted 8 lines

xen/arch/x86/hvm/hvm.c int hvm_set_guest_pat(struct vcpu *v, u64 guest_pat) · line 318

Quoted 4 lines
xen/arch/x86/hvm/hvm.c +3 -0
+        return false;
+
+    ASSERT(hvm_funcs.set_guest_bndcfgs);

I think it would be better (more safe for release builds) if this was left as part of the if(). Otherwise, ...

Quoted 8 lines

xen/arch/x86/hvm/hvm.c bool hvm_set_guest_bndcfgs(struct vcpu *v, u64 val) · line 350

Quoted 6 lines
xen/arch/x86/hvm/hvm.c +1 -1
@@ -342,7 +347,22 @@ bool hvm_set_guest_bndcfgs(struct vcpu *v, u64 val)
             /* nothing, best effort only */;
     }
 
-    return hvm_funcs.set_guest_bndcfgs(v, val);
+    hvm_funcs.set_guest_bndcfgs(v, val);

... you risk calling NULL here.

Ok, I'll return false instead.

Quoted 5 lines
Quoted 5 lines
xen/arch/x86/hvm/hvm.c +0 -0
@@ -3472,12 +3494,6 @@ int hvm_msr_read_intercept(unsigned int msr, uint64_t *msr_content)
         *msr_content = v->arch.hvm.msr_xss;
         break;
 
Quoted 11 lines

xen/arch/x86/hvm/hvm.c int hvm_msr_read_intercept(unsigned int msr, uint64_t *msr_content) · line 3479

Quoted 5 lines
xen/arch/x86/hvm/hvm.c +0 -5
-    case MSR_IA32_BNDCFGS:
-        if ( !d->arch.cpuid->feat.mpx ||
-             !hvm_get_guest_bndcfgs(v, msr_content) )
-            goto gp_fault;
-        break;

Note how this already checks the CPUID policy, somewhat contrary to what you say in the commit message. I think this wants to remain this way in guest_{rd,wr}msr() - see all the other CPUID policy checks there. That way hvm_get_guest_bndcfgs() can be made consistent with the hook pointer type again, and it can remain an inline function as before.

I thought it was neater to put the check inside the helper but I can pull it out into the switch statement if that's preferred.

Paul

Quoted 3 lines

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#5 lore raw

Re: [PATCH 1/6] x86: stop handling MSR_IA32_BNDCFGS save/restore in implementation code

Quoted 21 lines
Quoted 9 lines
Quoted 1 line · On 11.03.19 at 17:10, <[email protected]> wrote:

From: Jan Beulich [mailto:[email protected]]
Sent: 08 March 2019 16:40

Quoted 5 lines
Quoted 1 line
Quoted 1 line · On 07.01.19 at 13:02, <[email protected]> wrote:
xen/arch/x86/hvm/hvm.c +0 -0
@@ -3472,12 +3494,6 @@ int hvm_msr_read_intercept(unsigned int msr, uint64_t *msr_content)
         *msr_content = v->arch.hvm.msr_xss;
         break;
 
Quoted 11 lines

xen/arch/x86/hvm/hvm.c int hvm_msr_read_intercept(unsigned int msr, uint64_t *msr_content) · line 3479

Quoted 5 lines
xen/arch/x86/hvm/hvm.c +0 -5
-    case MSR_IA32_BNDCFGS:
-        if ( !d->arch.cpuid->feat.mpx ||
-             !hvm_get_guest_bndcfgs(v, msr_content) )
-            goto gp_fault;
-        break;

Note how this already checks the CPUID policy, somewhat contrary to what you say in the commit message. I think this wants to remain this way in guest_{rd,wr}msr() - see all the other CPUID policy checks there. That way hvm_get_guest_bndcfgs() can be made consistent with the hook pointer type again, and it can remain an inline function as before.

Quoted 2 lines

I thought it was neater to put the check inside the helper but I can pull it out into the switch statement if that's preferred.

Unless Andrew explicitly tells you otherwise, please do for the sake of consistency with other code in that switch().

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#6 lore raw

PATCH 2/6 x86: save GUEST_BNDCFGS on context switch...

In reply to Paul Durrant

...to avoid the need for a VMCS reload when the value of MSR_IA32_BNDCFGS is read by the tool-stack.

---
Cc: Jan Beulich <[email protected]>
Cc: Andrew Cooper <[email protected]>
Cc: Wei Liu <[email protected]>
Cc: "Roger Pau Monné" <[email protected]>
Cc: Jun Nakajima <[email protected]>
Cc: Kevin Tian <[email protected]>

---
 xen/arch/x86/hvm/hvm.c     | 18 +++++++++++++++---
 xen/arch/x86/hvm/vmx/vmx.c |  3 +++
 xen/include/asm-x86/msr.h  |  5 +++++
 3 files changed, 23 insertions(+), 3 deletions(-)
xen/arch/x86/hvm/hvm.c +15 -3
index 5fd5478b7d..b86aed7c24 100644
@@ -311,6 +311,7 @@ int hvm_set_guest_pat(struct vcpu *v, u64 guest_pat)
 bool hvm_set_guest_bndcfgs(struct vcpu *v, uint64_t val)
 {
     const struct cpuid_policy *cp = v->domain->arch.cpuid;
+    struct vcpu_msrs *msrs = v->arch.msrs;
 
     if ( !cp->feat.mpx )
         return false;
@@ -347,7 +348,8 @@ bool hvm_set_guest_bndcfgs(struct vcpu *v, uint64_t val)
             /* nothing, best effort only */;
     }
 
-    hvm_funcs.set_guest_bndcfgs(v, val);
+    msrs->bndcfgs.raw = val;
+    hvm_funcs.set_guest_bndcfgs(v, msrs->bndcfgs.raw);
 
     return true;
 }
@@ -355,12 +357,22 @@ bool hvm_set_guest_bndcfgs(struct vcpu *v, uint64_t val)
 bool hvm_get_guest_bndcfgs(struct vcpu *v, uint64_t *val)
 {
     const struct cpuid_policy *cp = v->domain->arch.cpuid;
+    struct vcpu_msrs *msrs = v->arch.msrs;
 
     if ( !cp->feat.mpx )
         return false;
 
-    ASSERT(hvm_funcs.get_guest_bndcfgs);
-    *val = hvm_funcs.get_guest_bndcfgs(v);
+    /*
+     * The value only need be read in current context as a context
+     * switch will save the value into msrs->bndcfgs.
+     */
+    if ( v == current )
+    {
+        ASSERT(hvm_funcs.get_guest_bndcfgs);
+        msrs->bndcfgs.raw = hvm_funcs.get_guest_bndcfgs(v);
+    }
+
+    *val = msrs->bndcfgs.raw;
 
     return true;
 }
xen/arch/x86/hvm/vmx/vmx.c +3 -0
index 4bfabe8d0e..7dba92da45 100644
@@ -497,6 +497,9 @@ static void vmx_restore_host_msrs(void)
 
 static void vmx_save_guest_msrs(struct vcpu *v)
 {
+    if ( cpu_has_mpx && cpu_has_vmx_mpx )
+        __vmread(GUEST_BNDCFGS, &v->arch.msrs->bndcfgs.raw);
+
     /*
      * We cannot cache SHADOW_GS_BASE while the VCPU runs, as it can
      * be updated at any time via SWAPGS, which we cannot trap.
xen/include/asm-x86/msr.h +5 -0
index ad8688a61f..c69ce56963 100644
@@ -305,6 +305,11 @@ struct vcpu_msrs
      * values here may be stale in current context.
      */
     uint32_t dr_mask[4];
+
+    /* 0x00000d90 - MSR_IA32_BNDCFGS */
+    struct {
+        uint64_t raw;
+    } bndcfgs;
 };
 
 void init_guest_msr_policy(void);
Signature · 7 lines
2.20.1


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
#7 lore raw

Re: [PATCH 2/6] x86: save GUEST_BNDCFGS on context switch...

Quoted 6 lines

From: Paul Durrant [mailto:[email protected]]
Sent: Monday, January 7, 2019 8:03 PM

...to avoid the need for a VMCS reload when the value of MSR_IA32_BNDCFGS is
read by the tool-stack.

the frequency of context switch is much higher than the one of reading by tool-stack (at least in general case), then is it right thing to add a vmread for every context switch to save a VMCS reload in tool stack path?

Quoted 96 lines

---
Cc: Jan Beulich <[email protected]>
Cc: Andrew Cooper <[email protected]>
Cc: Wei Liu <[email protected]>
Cc: "Roger Pau Monné" <[email protected]>
Cc: Jun Nakajima <[email protected]>
Cc: Kevin Tian <[email protected]>

---
 xen/arch/x86/hvm/hvm.c     | 18 +++++++++++++++---
 xen/arch/x86/hvm/vmx/vmx.c |  3 +++
 xen/include/asm-x86/msr.h  |  5 +++++
 3 files changed, 23 insertions(+), 3 deletions(-)
xen/arch/x86/hvm/hvm.c +1 -0
index 5fd5478b7d..b86aed7c24 100644
@@ -311,6 +311,7 @@ int hvm_set_guest_pat(struct vcpu *v, u64
 guest_pat)
 bool hvm_set_guest_bndcfgs(struct vcpu *v, uint64_t val)
 {
     const struct cpuid_policy *cp = v->domain->arch.cpuid;
+    struct vcpu_msrs *msrs = v->arch.msrs;
 
     if ( !cp->feat.mpx )
         return false;
xen/arch/x86/hvm/hvm.c +2 -1
@@ -347,7 +348,8 @@ bool hvm_set_guest_bndcfgs(struct vcpu *v,
 uint64_t val)
             /* nothing, best effort only */;
     }
 
-    hvm_funcs.set_guest_bndcfgs(v, val);
+    msrs->bndcfgs.raw = val;
+    hvm_funcs.set_guest_bndcfgs(v, msrs->bndcfgs.raw);
 
     return true;

}

xen/arch/x86/hvm/hvm.c +12 -2
@@ -355,12 +357,22 @@ bool hvm_set_guest_bndcfgs(struct vcpu *v,
 uint64_t val)
 bool hvm_get_guest_bndcfgs(struct vcpu *v, uint64_t *val)
 {
     const struct cpuid_policy *cp = v->domain->arch.cpuid;
+    struct vcpu_msrs *msrs = v->arch.msrs;
 
     if ( !cp->feat.mpx )
         return false;
 
-    ASSERT(hvm_funcs.get_guest_bndcfgs);
-    *val = hvm_funcs.get_guest_bndcfgs(v);
+    /*
+     * The value only need be read in current context as a context
+     * switch will save the value into msrs->bndcfgs.
+     */
+    if ( v == current )
+    {
+        ASSERT(hvm_funcs.get_guest_bndcfgs);
+        msrs->bndcfgs.raw = hvm_funcs.get_guest_bndcfgs(v);
+    }
+
+    *val = msrs->bndcfgs.raw;
 
     return true;

}

xen/arch/x86/hvm/vmx/vmx.c +3 -0
index 4bfabe8d0e..7dba92da45 100644
@@ -497,6 +497,9 @@ static void vmx_restore_host_msrs(void)
 
 static void vmx_save_guest_msrs(struct vcpu *v)
 {
+    if ( cpu_has_mpx && cpu_has_vmx_mpx )
+        __vmread(GUEST_BNDCFGS, &v->arch.msrs->bndcfgs.raw);
+
     /*
      * We cannot cache SHADOW_GS_BASE while the VCPU runs, as it can
      * be updated at any time via SWAPGS, which we cannot trap.
xen/include/asm-x86/msr.h +5 -1
index ad8688a61f..c69ce56963 100644
@@ -305,6 +305,11 @@ struct vcpu_msrs
      * values here may be stale in current context.
      */
     uint32_t dr_mask[4];
+
+    /* 0x00000d90 - MSR_IA32_BNDCFGS */
+    struct {
+        uint64_t raw;
+    } bndcfgs;
 };
 
 void init_guest_msr_policy(void);
--
Quoted 1 line

2.20.1

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#8 lore raw

Re: [PATCH 2/6] x86: save GUEST_BNDCFGS on context switch...

Quoted 12 lines

-----Original Message-----
From: Tian, Kevin [mailto:[email protected]]
Sent: 28 January 2019 02:34
To: Paul Durrant <[email protected]>; [email protected]
Cc: Jan Beulich <[email protected]>; Andrew Cooper <[email protected]>; Wei Liu <[email protected]>; Roger Pau Monne <[email protected]>; Nakajima, Jun <[email protected]>
Subject: RE: [PATCH 2/6] x86: save GUEST_BNDCFGS on context switch...

Quoted 3 lines

From: Paul Durrant [mailto:[email protected]]
Sent: Monday, January 7, 2019 8:03 PM

Quoted 8 lines
Quoted 3 lines

...to avoid the need for a VMCS reload when the value of MSR_IA32_BNDCFGS is
read by the tool-stack.

the frequency of context switch is much higher than the one of reading by tool-stack (at least in general case), then is it right thing to add a vmread for every context switch to save a VMCS reload in tool stack path?

AIUI the cost of the VMCS reload is sufficiently high (and the extra vmread sufficiently cheap) to make this desirable.

Andrew, do you have any comment?

Paul

Quoted 95 lines
Quoted 95 lines

---
Cc: Jan Beulich <[email protected]>
Cc: Andrew Cooper <[email protected]>
Cc: Wei Liu <[email protected]>
Cc: "Roger Pau Monné" <[email protected]>
Cc: Jun Nakajima <[email protected]>
Cc: Kevin Tian <[email protected]>

---
 xen/arch/x86/hvm/hvm.c     | 18 +++++++++++++++---
 xen/arch/x86/hvm/vmx/vmx.c |  3 +++
 xen/include/asm-x86/msr.h  |  5 +++++
 3 files changed, 23 insertions(+), 3 deletions(-)
xen/arch/x86/hvm/hvm.c +1 -0
index 5fd5478b7d..b86aed7c24 100644
@@ -311,6 +311,7 @@ int hvm_set_guest_pat(struct vcpu *v, u64
 guest_pat)
 bool hvm_set_guest_bndcfgs(struct vcpu *v, uint64_t val)
 {
     const struct cpuid_policy *cp = v->domain->arch.cpuid;
+    struct vcpu_msrs *msrs = v->arch.msrs;
 
     if ( !cp->feat.mpx )
         return false;
xen/arch/x86/hvm/hvm.c +2 -1
@@ -347,7 +348,8 @@ bool hvm_set_guest_bndcfgs(struct vcpu *v,
 uint64_t val)
             /* nothing, best effort only */;
     }
 
-    hvm_funcs.set_guest_bndcfgs(v, val);
+    msrs->bndcfgs.raw = val;
+    hvm_funcs.set_guest_bndcfgs(v, msrs->bndcfgs.raw);
 
     return true;

}

xen/arch/x86/hvm/hvm.c +12 -2
@@ -355,12 +357,22 @@ bool hvm_set_guest_bndcfgs(struct vcpu *v,
 uint64_t val)
 bool hvm_get_guest_bndcfgs(struct vcpu *v, uint64_t *val)
 {
     const struct cpuid_policy *cp = v->domain->arch.cpuid;
+    struct vcpu_msrs *msrs = v->arch.msrs;
 
     if ( !cp->feat.mpx )
         return false;
 
-    ASSERT(hvm_funcs.get_guest_bndcfgs);
-    *val = hvm_funcs.get_guest_bndcfgs(v);
+    /*
+     * The value only need be read in current context as a context
+     * switch will save the value into msrs->bndcfgs.
+     */
+    if ( v == current )
+    {
+        ASSERT(hvm_funcs.get_guest_bndcfgs);
+        msrs->bndcfgs.raw = hvm_funcs.get_guest_bndcfgs(v);
+    }
+
+    *val = msrs->bndcfgs.raw;
 
     return true;

}

xen/arch/x86/hvm/vmx/vmx.c +3 -0
index 4bfabe8d0e..7dba92da45 100644
@@ -497,6 +497,9 @@ static void vmx_restore_host_msrs(void)
 
 static void vmx_save_guest_msrs(struct vcpu *v)
 {
+    if ( cpu_has_mpx && cpu_has_vmx_mpx )
+        __vmread(GUEST_BNDCFGS, &v->arch.msrs->bndcfgs.raw);
+
     /*
      * We cannot cache SHADOW_GS_BASE while the VCPU runs, as it can
      * be updated at any time via SWAPGS, which we cannot trap.
xen/include/asm-x86/msr.h +5 -0
index ad8688a61f..c69ce56963 100644
@@ -305,6 +305,11 @@ struct vcpu_msrs
      * values here may be stale in current context.
      */
     uint32_t dr_mask[4];
+
+    /* 0x00000d90 - MSR_IA32_BNDCFGS */
+    struct {
+        uint64_t raw;
+    } bndcfgs;
 };
 
Quoted 3 lines
Quoted 3 lines
xen/include/asm-x86/msr.h +0 -1
 void init_guest_msr_policy(void);
--

2.20.1

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#9 lore raw

Re: [PATCH 2/6] x86: save GUEST_BNDCFGS on context switch...

In reply to Paul Durrant
Quoted 3 lines
Quoted 1 line
Quoted 1 line · On 07.01.19 at 13:02, <[email protected]> wrote:

...to avoid the need for a VMCS reload when the value of MSR_IA32_BNDCFGS is read by the tool-stack.

Is this a good trade-off? I.e. are tool stack reads at least as frequent as context switches?

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#10 lore raw

Re: [PATCH 2/6] x86: save GUEST_BNDCFGS on context switch...

Quoted 9 lines

-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 08 March 2019 16:43
To: Paul Durrant <[email protected]>
Cc: Andrew Cooper <[email protected]>; Roger Pau Monne <[email protected]>; Wei Liu <[email protected]>; Jun Nakajima <[email protected]>; Kevin Tian <[email protected]>; xen- devel <[email protected]>
Subject: Re: [PATCH 2/6] x86: save GUEST_BNDCFGS on context switch...

Quoted 7 lines
Quoted 3 lines
Quoted 1 line
Quoted 1 line · On 07.01.19 at 13:02, <[email protected]> wrote:

...to avoid the need for a VMCS reload when the value of MSR_IA32_BNDCFGS is read by the tool-stack.

Is this a good trade-off? I.e. are tool stack reads at least as frequent as context switches?

In the absence of any comment from Andrew on relative cost, I'll drop this.

Quoted 2 lines

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#11 lore raw

PATCH 3/6 x86: move the saved value of MSR_IA32_XSS into struct vcpu_msrs

In reply to Paul Durrant

Currently the value is saved directly in struct hvm_vcpu. This patch simply co-locates it with other saved MSR values. No functional change.

---
Cc: Jan Beulich <[email protected]>
Cc: Andrew Cooper <[email protected]>
Cc: Wei Liu <[email protected]>
Cc: "Roger Pau Monné" <[email protected]>
Cc: Jun Nakajima <[email protected]>
Cc: Kevin Tian <[email protected]>

---
 xen/arch/x86/domain.c          | 2 +-
 xen/arch/x86/hvm/hvm.c         | 4 ++--
 xen/arch/x86/hvm/vmx/vmx.c     | 4 ++--
 xen/include/asm-x86/hvm/vcpu.h | 1 -
 xen/include/asm-x86/msr.h      | 5 +++++
 5 files changed, 10 insertions(+), 6 deletions(-)
xen/arch/x86/domain.c +1 -1
index 32dc4253ff..c5f4b6a38a 100644
@@ -1704,7 +1704,7 @@ static void __context_switch(void)
                 BUG();
 
             if ( cpu_has_xsaves && is_hvm_vcpu(n) )
-                set_msr_xss(n->arch.hvm.msr_xss);
+                set_msr_xss(n->arch.msrs->xss.raw);
         }
         vcpu_restore_fpu_nonlazy(n, false);
         nd->arch.ctxt_switch->to(n);
xen/arch/x86/hvm/hvm.c +2 -2
index b86aed7c24..b55bb8b081 100644
@@ -3503,7 +3503,7 @@ int hvm_msr_read_intercept(unsigned int msr, uint64_t *msr_content)
     case MSR_IA32_XSS:
         if ( !d->arch.cpuid->xstate.xsaves )
             goto gp_fault;
-        *msr_content = v->arch.hvm.msr_xss;
+        *msr_content = v->arch.msrs->xss.raw;
         break;
 
     case MSR_K8_ENABLE_C1E:
@@ -3649,7 +3649,7 @@ int hvm_msr_write_intercept(unsigned int msr, uint64_t msr_content,
         /* No XSS features currently supported for guests. */
         if ( !d->arch.cpuid->xstate.xsaves || msr_content != 0 )
             goto gp_fault;
-        v->arch.hvm.msr_xss = msr_content;
+        v->arch.msrs->xss.raw = msr_content;
         break;
 
     case MSR_AMD64_NB_CFG:
xen/arch/x86/hvm/vmx/vmx.c +2 -2
index 7dba92da45..ec87607ec7 100644
@@ -810,7 +810,7 @@ static void vmx_save_msr(struct vcpu *v, struct hvm_msr *ctxt)
 {
     if ( cpu_has_xsaves && cpu_has_vmx_xsaves )
     {
-        ctxt->msr[ctxt->count].val = v->arch.hvm.msr_xss;
+        ctxt->msr[ctxt->count].val = v->arch.msrs->xss.raw;
         if ( ctxt->msr[ctxt->count].val )
             ctxt->msr[ctxt->count++].index = MSR_IA32_XSS;
     }
@@ -829,7 +829,7 @@ static int vmx_load_msr(struct vcpu *v, struct hvm_msr *ctxt)
         {
         case MSR_IA32_XSS:
             if ( cpu_has_xsaves && cpu_has_vmx_xsaves )
-                v->arch.hvm.msr_xss = ctxt->msr[i].val;
+                v->arch.msrs->xss.raw = ctxt->msr[i].val;
             else
                 err = -ENXIO;
             break;
xen/include/asm-x86/hvm/vcpu.h +0 -1
index c8a40f6d55..a8cfd13987 100644
@@ -171,7 +171,6 @@ struct hvm_vcpu {
     struct hvm_vcpu_asid n1asid;
 
     u64                 msr_tsc_adjust;
-    u64                 msr_xss;
 
     union {
         struct vmx_vcpu vmx;
xen/include/asm-x86/msr.h +5 -0
index c69ce56963..e9c685613e 100644
@@ -310,6 +310,11 @@ struct vcpu_msrs
     struct {
         uint64_t raw;
     } bndcfgs;
+
+    /* 0x00000da0 - MSR_IA32_XSS */
+    struct {
+        uint64_t raw;
+    } xss;
 };
 
 void init_guest_msr_policy(void);
Signature · 7 lines
2.20.1


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Reviewed-by on v1 3/6
#12 lore raw

Re: [PATCH 3/6] x86: move the saved value of MSR_IA32_XSS into struct vcpu_msrs

Quoted 7 lines

From: Paul Durrant [mailto:[email protected]]
Sent: Monday, January 7, 2019 8:03 PM

Currently the value is saved directly in struct hvm_vcpu. This patch simply co-locates it with other saved MSR values. No functional change.

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

Reviewed-by on v1 3/6
#13 lore raw

Re: [PATCH 3/6] x86: move the saved value of MSR_IA32_XSS into struct vcpu_msrs

In reply to Paul Durrant
Quoted 5 lines
Quoted 1 line
Quoted 1 line · On 07.01.19 at 13:02, <[email protected]> wrote:

Currently the value is saved directly in struct hvm_vcpu. This patch simply co-locates it with other saved MSR values. No functional change.

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#14 lore raw

PATCH 4/6 x86: stop handling MSR_IA32_XSS save/restore in implementation code

In reply to Paul Durrant

Saving and restoring the value of this MSR is currently handled by implementation-specific code despite it being architectural. This patch moves handling of accesses to this MSR from hvm.c into the msr.c, thus allowing the common MSR save/restore code to handle it.

This patch also adds proper checks of CPUID policy in the new get/set code.

NOTE: MSR_IA32_XSS is the last MSR to be saved and restored by
implementation-specific code. This patch therefore removes the (VMX) definitions and of the init_msr(), save_msr() and load_msr() hvm_funcs, as they are no longer necessary. The declarations of and calls to those hvm_funcs will be cleaned up by a subsequent patch.

---
Cc: Jan Beulich <[email protected]>
Cc: Andrew Cooper <[email protected]>
Cc: Wei Liu <[email protected]>
Cc: "Roger Pau Monné" <[email protected]>
Cc: Jun Nakajima <[email protected]>
Cc: Kevin Tian <[email protected]>

---
 xen/arch/x86/hvm/hvm.c     | 15 ++----------
 xen/arch/x86/hvm/vmx/vmx.c | 49 --------------------------------------
 xen/arch/x86/msr.c         | 18 ++++++++++++++
 3 files changed, 20 insertions(+), 62 deletions(-)
xen/arch/x86/hvm/hvm.c +2 -13
index b55bb8b081..856dcf696b 100644
@@ -1345,6 +1345,7 @@ static const uint32_t msrs_to_send[] = {
     MSR_SPEC_CTRL,
     MSR_INTEL_MISC_FEATURES_ENABLES,
     MSR_IA32_BNDCFGS,
+    MSR_IA32_XSS,
     MSR_AMD64_DR0_ADDRESS_MASK,
     MSR_AMD64_DR1_ADDRESS_MASK,
     MSR_AMD64_DR2_ADDRESS_MASK,
@@ -1483,6 +1484,7 @@ static int hvm_load_cpu_msrs(struct domain *d, hvm_domain_context_t *h)
         case MSR_SPEC_CTRL:
         case MSR_INTEL_MISC_FEATURES_ENABLES:
         case MSR_IA32_BNDCFGS:
+        case MSR_IA32_XSS:
         case MSR_AMD64_DR0_ADDRESS_MASK:
         case MSR_AMD64_DR1_ADDRESS_MASK ... MSR_AMD64_DR3_ADDRESS_MASK:
             rc = guest_wrmsr(v, ctxt->msr[i].index, ctxt->msr[i].val);
@@ -3500,12 +3502,6 @@ int hvm_msr_read_intercept(unsigned int msr, uint64_t *msr_content)
         *msr_content = var_range_base[index];
         break;
 
-    case MSR_IA32_XSS:
-        if ( !d->arch.cpuid->xstate.xsaves )
-            goto gp_fault;
-        *msr_content = v->arch.msrs->xss.raw;
-        break;
-
     case MSR_K8_ENABLE_C1E:
     case MSR_AMD64_NB_CFG:
          /*
@@ -3645,13 +3641,6 @@ int hvm_msr_write_intercept(unsigned int msr, uint64_t msr_content,
             goto gp_fault;
         break;
 
-    case MSR_IA32_XSS:
-        /* No XSS features currently supported for guests. */
-        if ( !d->arch.cpuid->xstate.xsaves || msr_content != 0 )
-            goto gp_fault;
-        v->arch.msrs->xss.raw = msr_content;
-        break;
-
     case MSR_AMD64_NB_CFG:
         /* ignore the write */
         break;
xen/arch/x86/hvm/vmx/vmx.c +0 -49

61 lines not on this page · raw message

xen/arch/x86/msr.c +18 -0
index a3406c29a8..3aa79031cf 100644
@@ -164,6 +164,13 @@ int guest_rdmsr(struct vcpu *v, uint32_t msr, uint64_t *val)
 
         break;
 
+    case MSR_IA32_XSS:
+        if ( !is_hvm_domain(d) || !cp->xstate.xsaves )
+            goto gp_fault;
+
+        *val = msrs->xss.raw;
+        break;
+
     case 0x40000000 ... 0x400001ff:
         if ( is_viridian_domain(d) )
         {
@@ -331,6 +338,17 @@ int guest_wrmsr(struct vcpu *v, uint32_t msr, uint64_t val)
 
         break;
 
+    case MSR_IA32_XSS:
+        if ( !is_hvm_domain(d) || !cp->xstate.xsaves )
+            goto gp_fault;
+
+        /* No XSS features currently supported for guests */
+        if ( val != 0 )
+            goto gp_fault;
+
+        msrs->xss.raw = val;
+        break;
+
     case 0x40000000 ... 0x400001ff:
         if ( is_viridian_domain(d) )
         {
Signature · 7 lines
2.20.1


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Reviewed-by on v1 4/6
#15 lore raw

Re: [PATCH 4/6] x86: stop handling MSR_IA32_XSS save/restore in implementation code

Quoted 10 lines

From: Paul Durrant [mailto:[email protected]]
Sent: Monday, January 7, 2019 8:03 PM

Saving and restoring the value of this MSR is currently handled by implementation-specific code despite it being architectural. This patch moves handling of accesses to this MSR from hvm.c into the msr.c, thus allowing the common MSR save/restore code to handle it.

This patch also adds proper checks of CPUID policy in the new get/set code.

Quoted 8 lines

NOTE: MSR_IA32_XSS is the last MSR to be saved and restored by
implementation-specific code. This patch therefore removes the (VMX) definitions and of the init_msr(), save_msr() and load_msr() hvm_funcs, as they are no longer necessary. The declarations of and calls to those hvm_funcs will be cleaned up by a subsequent patch.

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

Reviewed-by on v1 4/6
#16 lore raw

Re: [PATCH 4/6] x86: stop handling MSR_IA32_XSS save/restore in implementation code

In reply to Paul Durrant

xen/arch/x86/msr.c int guest_rdmsr(struct vcpu *v, uint32_t msr, uint64_t *val) · line 168

Quoted 9 lines
Quoted 1 line
Quoted 1 line · On 07.01.19 at 13:02, <[email protected]> wrote:
xen/arch/x86/msr.c +2 -0
@@ -164,6 +164,13 @@ int guest_rdmsr(struct vcpu *v, uint32_t msr, uint64_t *val)
 
         break;
 
+    case MSR_IA32_XSS:
+        if ( !is_hvm_domain(d) || !cp->xstate.xsaves )

Why the is_hvm_domain() check here and below? The CPUID policy bit alone should properly reflect things, and there should be no change needed here if we decided to support XSAVES also for PV guests. With them dropped

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#17 lore raw

Re: [PATCH 4/6] x86: stop handling MSR_IA32_XSS save/restore in implementation code

Quoted 14 lines

-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 08 March 2019 16:47
To: Paul Durrant <[email protected]>
Cc: Andrew Cooper <[email protected]>; Roger Pau Monne <[email protected]>; Wei Liu <[email protected]>; Jun Nakajima <[email protected]>; Kevin Tian <[email protected]>; xen- devel <[email protected]>
Subject: Re: [PATCH 4/6] x86: stop handling MSR_IA32_XSS save/restore in implementation code

Quoted 5 lines
Quoted 1 line
Quoted 1 line · On 07.01.19 at 13:02, <[email protected]> wrote:
xen/arch/x86/msr.c +0 -0
@@ -164,6 +164,13 @@ int guest_rdmsr(struct vcpu *v, uint32_t msr, uint64_t *val)
 
Quoted 6 lines

xen/arch/x86/msr.c int guest_rdmsr(struct vcpu *v, uint32_t msr, uint64_t *val) · line 168

Quoted 4 lines
xen/arch/x86/msr.c +2 -0
         break;
 
+    case MSR_IA32_XSS:
+        if ( !is_hvm_domain(d) || !cp->xstate.xsaves )

Why the is_hvm_domain() check here and below?

Only there because the older code was HVM-specific.

Quoted 5 lines

The CPUID policy bit
alone should properly reflect things, and there should be no change needed here if we decided to support XSAVES also for PV guests. With them dropped

Ok. Thanks,

Paul

Quoted 3 lines

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#18 lore raw

PATCH 5/6 x86: remove defunct init/load/save_msr() hvm_funcs

In reply to Paul Durrant

These hvm_funcs are no longer required since no MSR values are saved or restored by implementation-specific code.

---
Cc: Jan Beulich <[email protected]>
Cc: Andrew Cooper <[email protected]>
Cc: Wei Liu <[email protected]>
Cc: "Roger Pau Monné" <[email protected]>

---
 xen/arch/x86/hvm/hvm.c        | 31 ++++++++++---------------------
 xen/include/asm-x86/hvm/hvm.h |  4 ----
 2 files changed, 10 insertions(+), 25 deletions(-)
xen/arch/x86/hvm/hvm.c +10 -21
index 856dcf696b..199aa0c148 100644
@@ -1351,7 +1351,6 @@ static const uint32_t msrs_to_send[] = {
     MSR_AMD64_DR2_ADDRESS_MASK,
     MSR_AMD64_DR3_ADDRESS_MASK,
 };
-static unsigned int __read_mostly msr_count_max = ARRAY_SIZE(msrs_to_send);
 
 static int hvm_save_cpu_msrs(struct vcpu *v, hvm_domain_context_t *h)
 {
@@ -1361,7 +1360,7 @@ static int hvm_save_cpu_msrs(struct vcpu *v, hvm_domain_context_t *h)
     int err;
 
     err = _hvm_init_entry(h, CPU_MSR_CODE, v->vcpu_id,
-                             HVM_CPU_MSR_SIZE(msr_count_max));
+                             HVM_CPU_MSR_SIZE(ARRAY_SIZE(msrs_to_send)));
     if ( err )
         return err;
     ctxt = (struct hvm_msr *)&h->data[h->cur];
@@ -1394,10 +1393,7 @@ static int hvm_save_cpu_msrs(struct vcpu *v, hvm_domain_context_t *h)
         ctxt->msr[ctxt->count++].val = val;
     }
 
-    if ( hvm_funcs.save_msr )
-        hvm_funcs.save_msr(v, ctxt);
-
-    ASSERT(ctxt->count <= msr_count_max);
+    ASSERT(ctxt->count <= ARRAY_SIZE(msrs_to_send));
 
     for ( i = 0; i < ctxt->count; ++i )
         ctxt->msr[i]._rsvd = 0;
@@ -1472,10 +1468,7 @@ static int hvm_load_cpu_msrs(struct domain *d, hvm_domain_context_t *h)
             return -EOPNOTSUPP;
     /* Checking finished */
 
-    if ( hvm_funcs.load_msr )
-        err = hvm_funcs.load_msr(v, ctxt);
-
-    for ( i = 0; !err && i < ctxt->count; ++i )
+    for ( i = 0; i < ctxt->count; ++i )
     {
         switch ( ctxt->msr[i].index )
         {
@@ -1516,17 +1509,13 @@ static int __init hvm_register_CPU_save_and_restore(void)
                             sizeof(struct hvm_save_descriptor),
                         HVMSR_PER_VCPU);
 
-    if ( hvm_funcs.init_msr )
-        msr_count_max += hvm_funcs.init_msr();
-
-    if ( msr_count_max )
-        hvm_register_savevm(CPU_MSR_CODE,
-                            "CPU_MSR",
-                            hvm_save_cpu_msrs,
-                            hvm_load_cpu_msrs,
-                            HVM_CPU_MSR_SIZE(msr_count_max) +
-                                sizeof(struct hvm_save_descriptor),
-                            HVMSR_PER_VCPU);
+    hvm_register_savevm(CPU_MSR_CODE,
+                        "CPU_MSR",
+                        hvm_save_cpu_msrs,
+                        hvm_load_cpu_msrs,
+                        HVM_CPU_MSR_SIZE(ARRAY_SIZE(msrs_to_send)) +
+                            sizeof(struct hvm_save_descriptor),
+                        HVMSR_PER_VCPU);
 
     return 0;
 }
xen/include/asm-x86/hvm/hvm.h +0 -4
index 5c8237e087..a7b08d1db3 100644
@@ -115,10 +115,6 @@ struct hvm_function_table {
     void (*save_cpu_ctxt)(struct vcpu *v, struct hvm_hw_cpu *ctxt);
     int (*load_cpu_ctxt)(struct vcpu *v, struct hvm_hw_cpu *ctxt);
 
-    unsigned int (*init_msr)(void);
-    void (*save_msr)(struct vcpu *, struct hvm_msr *);
-    int (*load_msr)(struct vcpu *, struct hvm_msr *);
-
     /* Examine specifics of the guest state. */
     unsigned int (*get_interrupt_shadow)(struct vcpu *v);
     void (*set_interrupt_shadow)(struct vcpu *v, unsigned int intr_shadow);
Signature · 7 lines
2.20.1


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
#19 lore raw

Re: [PATCH 5/6] x86: remove defunct init/load/save_msr() hvm_funcs

xen/arch/x86/hvm/hvm.c static int hvm_load_cpu_msrs(struct domain *d, hvm_domain_context_t *h) · line 1474

Quoted 13 lines
Quoted 1 line
Quoted 1 line · On 07.01.19 at 13:02, <[email protected]> wrote:
xen/arch/x86/hvm/hvm.c +1 -4
@@ -1472,10 +1468,7 @@ static int hvm_load_cpu_msrs(struct domain *d, hvm_domain_context_t *h)
             return -EOPNOTSUPP;
     /* Checking finished */
 
-    if ( hvm_funcs.load_msr )
-        err = hvm_funcs.load_msr(v, ctxt);
-
-    for ( i = 0; !err && i < ctxt->count; ++i )
+    for ( i = 0; i < ctxt->count; ++i )
     {
         switch ( ctxt->msr[i].index )
         {

err can become set to non-zero in the loop body further down from here, so I don't think you should alter the loop header.

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#20 lore raw

Re: [PATCH 5/6] x86: remove defunct init/load/save_msr() hvm_funcs

Quoted 13 lines

-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 08 March 2019 16:50
To: Paul Durrant <[email protected]>
Cc: Andrew Cooper <[email protected]>; Roger Pau Monne <[email protected]>; Wei Liu <[email protected]>; xen-devel <[email protected]>
Subject: Re: [PATCH 5/6] x86: remove defunct init/load/save_msr() hvm_funcs

Quoted 5 lines
Quoted 1 line
Quoted 1 line · On 07.01.19 at 13:02, <[email protected]> wrote:
xen/arch/x86/hvm/hvm.c +0 -0
@@ -1472,10 +1468,7 @@ static int hvm_load_cpu_msrs(struct domain *d, hvm_domain_context_t *h)
             return -EOPNOTSUPP;
     /* Checking finished */
 
Quoted 11 lines

xen/arch/x86/hvm/hvm.c static int hvm_load_cpu_msrs(struct domain *d, hvm_domain_context_t *h) · line 1474

Quoted 8 lines
xen/arch/x86/hvm/hvm.c +1 -4
-    if ( hvm_funcs.load_msr )
-        err = hvm_funcs.load_msr(v, ctxt);
-
-    for ( i = 0; !err && i < ctxt->count; ++i )
+    for ( i = 0; i < ctxt->count; ++i )
     {
         switch ( ctxt->msr[i].index )
         {

err can become set to non-zero in the loop body further down from here, so I don't think you should alter the loop header.

True. Bailing early is a change of semantic, so I'll drop the check.

Paul

Quoted 3 lines

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#21 lore raw

Re: [PATCH 5/6] x86: remove defunct init/load/save_msr() hvm_funcs

Quoted 33 lines

-----Original Message-----
From: Paul Durrant
Sent: 11 March 2019 17:46
To: 'Jan Beulich' <[email protected]>
Cc: Andrew Cooper <[email protected]>; Roger Pau Monne <[email protected]>; Wei Liu <[email protected]>; xen-devel <[email protected]>
Subject: RE: [PATCH 5/6] x86: remove defunct init/load/save_msr() hvm_funcs

Quoted 13 lines

-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 08 March 2019 16:50
To: Paul Durrant <[email protected]>
Cc: Andrew Cooper <[email protected]>; Roger Pau Monne <[email protected]>; Wei Liu <[email protected]>; xen-devel <[email protected]>
Subject: Re: [PATCH 5/6] x86: remove defunct init/load/save_msr() hvm_funcs

Quoted 5 lines
Quoted 1 line
Quoted 1 line · On 07.01.19 at 13:02, <[email protected]> wrote:
xen/arch/x86/hvm/hvm.c +0 -0
@@ -1472,10 +1468,7 @@ static int hvm_load_cpu_msrs(struct domain *d, hvm_domain_context_t *h)
             return -EOPNOTSUPP;
     /* Checking finished */
 
Quoted 11 lines

xen/arch/x86/hvm/hvm.c static int hvm_load_cpu_msrs(struct domain *d, hvm_domain_context_t *h) · line 1474

Quoted 8 lines
xen/arch/x86/hvm/hvm.c +1 -4
-    if ( hvm_funcs.load_msr )
-        err = hvm_funcs.load_msr(v, ctxt);
-
-    for ( i = 0; !err && i < ctxt->count; ++i )
+    for ( i = 0; i < ctxt->count; ++i )
     {
         switch ( ctxt->msr[i].index )
         {

err can become set to non-zero in the loop body further down from here, so I don't think you should alter the loop header.

Quoted 1 line

True. Bailing early is a change of semantic, so I'll drop the check.

^ Not

Quoted 6 lines

Paul

Quoted 3 lines

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#22 lore raw

PATCH 6/6 x86: introduce dr_mask_idx() helper function...

In reply to Paul Durrant

...to avoid repeated open-coding.

Unfortunately the mapping from MSR index MSR_AMD64_DR<idx>_ADDRESS_MASK to the 'idx' value is non-trivial since the MSR index corresponding to idx value 0 is non-consecutive with the MSR indices corresponding to idx values 1-3. This mapping is currently dealt with by near-identical open coding in guest_rdmsr() and guest_wrmsr().

This patch adds a helper function, dr_mask_idx(), to handle the mapping and then uses this in guest_rdmsr() and guest_wrmsr() instead, thus making the code somewhat neater.

---
Cc: Jan Beulich <[email protected]>
Cc: Andrew Cooper <[email protected]>
Cc: Wei Liu <[email protected]>
Cc: "Roger Pau Monné" <[email protected]>

---
 xen/arch/x86/msr.c        | 18 +++++-------------
 xen/include/asm-x86/msr.h | 21 +++++++++++++++++++++
 2 files changed, 26 insertions(+), 13 deletions(-)
xen/arch/x86/msr.c +5 -13
index 3aa79031cf..4e4c9c805a 100644
@@ -21,7 +21,6 @@
 
 #include <xen/init.h>
 #include <xen/lib.h>
-#include <xen/nospec.h>
 #include <xen/sched.h>
 
 #include <asm/debugreg.h>
@@ -121,7 +120,7 @@ int guest_rdmsr(struct vcpu *v, uint32_t msr, uint64_t *val)
     const struct domain *d = v->domain;
     const struct cpuid_policy *cp = d->arch.cpuid;
     const struct msr_policy *mp = d->arch.msr;
-    const struct vcpu_msrs *msrs = v->arch.msrs;
+    struct vcpu_msrs *msrs = v->arch.msrs;
     int ret = X86EMUL_OKAY;
 
     switch ( msr )
@@ -202,13 +201,10 @@ int guest_rdmsr(struct vcpu *v, uint32_t msr, uint64_t *val)
          */
 #ifdef CONFIG_HVM
         if ( v == current && is_hvm_domain(d) && v->arch.hvm.flag_dr_dirty )
-            rdmsrl(msr, *val);
-        else
+            rdmsrl(msr, msrs->dr_mask[dr_mask_idx(msr)]);
 #endif
-            *val = msrs->dr_mask[
-                array_index_nospec((msr == MSR_AMD64_DR0_ADDRESS_MASK)
-                                   ? 0 : (msr - MSR_AMD64_DR1_ADDRESS_MASK + 1),
-                                   ARRAY_SIZE(msrs->dr_mask))];
+
+        *val = msrs->dr_mask[dr_mask_idx(msr)];
         break;
 
     default:
@@ -377,11 +373,7 @@ int guest_wrmsr(struct vcpu *v, uint32_t msr, uint64_t val)
         if ( !cp->extd.dbext || val != (uint32_t)val )
             goto gp_fault;
 
-        msrs->dr_mask[
-            array_index_nospec((msr == MSR_AMD64_DR0_ADDRESS_MASK)
-                               ? 0 : (msr - MSR_AMD64_DR1_ADDRESS_MASK + 1),
-                               ARRAY_SIZE(msrs->dr_mask))] = val;
-
+        msrs->dr_mask[dr_mask_idx(msr)] = val;
         if ( v == curr && (curr->arch.dr7 & DR7_ACTIVE_MASK) )
             wrmsrl(msr, val);
         break;
xen/include/asm-x86/msr.h +21 -0
index e9c685613e..5eb5e6dace 100644
@@ -5,6 +5,7 @@
 
 #ifndef __ASSEMBLY__
 
+#include <xen/nospec.h>
 #include <xen/types.h>
 #include <xen/percpu.h>
 #include <xen/errno.h>
@@ -317,6 +318,26 @@ struct vcpu_msrs
     } xss;
 };
 
+static inline unsigned int dr_mask_idx(uint32_t msr)
+{
+    switch (msr)
+    {
+    default:
+        ASSERT_UNREACHABLE();
+        /* Fallthrough */
+    case MSR_AMD64_DR0_ADDRESS_MASK:
+        return 0;
+
+    case MSR_AMD64_DR1_ADDRESS_MASK ... MSR_AMD64_DR3_ADDRESS_MASK:
+    {
+        struct vcpu_msrs msrs; /* only used for ARRAY_SIZE() */
+
+        return array_index_nospec(msr - MSR_AMD64_DR1_ADDRESS_MASK + 1,
+                                  ARRAY_SIZE(msrs.dr_mask));
+    }
+    }
+}
+
 void init_guest_msr_policy(void);
 int init_domain_msr_policy(struct domain *d);
 int init_vcpu_msr_policy(struct vcpu *v);
Signature · 7 lines
2.20.1


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
#23 lore raw

Re: [PATCH 6/6] x86: introduce dr_mask_idx() helper function...

Quoted 6 lines
Quoted 1 line
Quoted 1 line · On 07.01.19 at 13:02, <[email protected]> wrote:
xen/arch/x86/msr.c +0 -1
@@ -202,13 +201,10 @@ int guest_rdmsr(struct vcpu *v, uint32_t msr, uint64_t *val)
          */
 #ifdef CONFIG_HVM
         if ( v == current && is_hvm_domain(d) && v->arch.hvm.flag_dr_dirty )
-            rdmsrl(msr, *val);

xen/arch/x86/msr.c int guest_rdmsr(struct vcpu *v, uint32_t msr, uint64_t *val) · line 208

Quoted 10 lines
xen/arch/x86/msr.c +3 -5
-        else
+            rdmsrl(msr, msrs->dr_mask[dr_mask_idx(msr)]);
 #endif
-            *val = msrs->dr_mask[
-                array_index_nospec((msr == MSR_AMD64_DR0_ADDRESS_MASK)
-                                   ? 0 : (msr - MSR_AMD64_DR1_ADDRESS_MASK + 1),
-                                   ARRAY_SIZE(msrs->dr_mask))];
+
+        *val = msrs->dr_mask[dr_mask_idx(msr)];
         break;

While I don't really mind this behavioral change (of updating *msrs), I'd like to get Andrew's opinion on this from a conceptual pov.

xen/include/asm-x86/msr.h struct vcpu_msrs · line 323

Quoted 7 lines
xen/include/asm-x86/msr.h +3 -0
@@ -317,6 +318,26 @@ struct vcpu_msrs
     } xss;
 };
 
+static inline unsigned int dr_mask_idx(uint32_t msr)
+{
+    switch (msr)

Missing blanks immediately inside the parentheses.

xen/include/asm-x86/msr.h struct vcpu_msrs · line 333

Quoted 10 lines
xen/include/asm-x86/msr.h +10 -0
+    {
+    default:
+        ASSERT_UNREACHABLE();
+        /* Fallthrough */
+    case MSR_AMD64_DR0_ADDRESS_MASK:
+        return 0;
+
+    case MSR_AMD64_DR1_ADDRESS_MASK ... MSR_AMD64_DR3_ADDRESS_MASK:
+    {
+        struct vcpu_msrs msrs; /* only used for ARRAY_SIZE() */

I don't think you need this - you can use e.g. ARRAY_SIZE(current->arch.vcpu.msrs.dr_mask), can't you?

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#24 lore raw

Re: [PATCH 6/6] x86: introduce dr_mask_idx() helper function...

Quoted 10 lines · Jan Beulich
Quoted 10 lines
Quoted 1 line
Quoted 1 line · On 07.01.19 at 13:02, <[email protected]> wrote:
xen/arch/x86/msr.c +1 -3
@@ -202,13 +201,10 @@ int guest_rdmsr(struct vcpu *v, uint32_t msr, uint64_t *val)
          */
 #ifdef CONFIG_HVM
         if ( v == current && is_hvm_domain(d) && v->arch.hvm.flag_dr_dirty )
-            rdmsrl(msr, *val);
-        else
+            rdmsrl(msr, msrs->dr_mask[dr_mask_idx(msr)]);
 #endif
-            *val = msrs->dr_mask[
Quoted 8 lines

xen/arch/x86/msr.c int guest_rdmsr(struct vcpu *v, uint32_t msr, uint64_t *val) · line 208

Quoted 6 lines
xen/arch/x86/msr.c +2 -3
-                array_index_nospec((msr == MSR_AMD64_DR0_ADDRESS_MASK)
-                                   ? 0 : (msr - MSR_AMD64_DR1_ADDRESS_MASK + 1),
-                                   ARRAY_SIZE(msrs->dr_mask))];
+
+        *val = msrs->dr_mask[dr_mask_idx(msr)];
         break;

While I don't really mind this behavioral change (of updating *msrs), I'd like to get Andrew's opinion on this from a conceptual pov.

So, considering...

Quoted 16 lines

xen/include/asm-x86/msr.h struct vcpu_msrs · line 323

Quoted 8 lines
xen/include/asm-x86/msr.h +3 -0
@@ -317,6 +318,26 @@ struct vcpu_msrs
     } xss;
 };
 
+static inline unsigned int dr_mask_idx(uint32_t msr)
+{
+    switch (msr)

Missing blanks immediately inside the parentheses.

Quoted 6 lines
xen/include/asm-x86/msr.h +6 -0
+    {
+    default:
+        ASSERT_UNREACHABLE();
+        /* Fallthrough */
+    case MSR_AMD64_DR0_ADDRESS_MASK:
+        return 0;

... this reintroduces a half-Spectre-v1 gadget, I'm -1 for the change.

I don't anticipate this translation being needed anywhere else, which is why I didn't introduce a helper the first time around.

I'd just drop the patch and leave the code as it is.

~Andrew

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#25 lore raw

Re: [PATCH 6/6] x86: introduce dr_mask_idx() helper function...

Quoted 47 lines

-----Original Message-----
From: Andrew Cooper
Sent: 08 March 2019 17:10
To: Jan Beulich <[email protected]>; Paul Durrant <[email protected]>
Cc: Roger Pau Monne <[email protected]>; Wei Liu <[email protected]>; xen-devel <xen- [email protected]>
Subject: Re: [PATCH 6/6] x86: introduce dr_mask_idx() helper function...

Quoted 10 lines · Jan Beulich
Quoted 10 lines
Quoted 1 line
Quoted 1 line · On 07.01.19 at 13:02, <[email protected]> wrote:
xen/arch/x86/msr.c +1 -3
@@ -202,13 +201,10 @@ int guest_rdmsr(struct vcpu *v, uint32_t msr, uint64_t *val)
          */
 #ifdef CONFIG_HVM
         if ( v == current && is_hvm_domain(d) && v->arch.hvm.flag_dr_dirty )
-            rdmsrl(msr, *val);
-        else
+            rdmsrl(msr, msrs->dr_mask[dr_mask_idx(msr)]);
 #endif
-            *val = msrs->dr_mask[
Quoted 8 lines

xen/arch/x86/msr.c int guest_rdmsr(struct vcpu *v, uint32_t msr, uint64_t *val) · line 208

Quoted 6 lines
xen/arch/x86/msr.c +2 -3
-                array_index_nospec((msr == MSR_AMD64_DR0_ADDRESS_MASK)
-                                   ? 0 : (msr - MSR_AMD64_DR1_ADDRESS_MASK + 1),
-                                   ARRAY_SIZE(msrs->dr_mask))];
+
+        *val = msrs->dr_mask[dr_mask_idx(msr)];
         break;

While I don't really mind this behavioral change (of updating *msrs), I'd like to get Andrew's opinion on this from a conceptual pov.

So, considering...

Quoted 16 lines

xen/include/asm-x86/msr.h struct vcpu_msrs · line 323

Quoted 8 lines
xen/include/asm-x86/msr.h +3 -0
@@ -317,6 +318,26 @@ struct vcpu_msrs
     } xss;
 };
 
+static inline unsigned int dr_mask_idx(uint32_t msr)
+{
+    switch (msr)

Missing blanks immediately inside the parentheses.

Quoted 6 lines
xen/include/asm-x86/msr.h +6 -0
+    {
+    default:
+        ASSERT_UNREACHABLE();
+        /* Fallthrough */
+    case MSR_AMD64_DR0_ADDRESS_MASK:
+        return 0;
Quoted 6 lines

... this reintroduces a half-Spectre-v1 gadget, I'm -1 for the change.

I don't anticipate this translation being needed anywhere else, which is why I didn't introduce a helper the first time around.

I'd just drop the patch and leave the code as it is.

Ok, I'll drop it.

Paul

Quoted 2 lines

~Andrew

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel