Hook up the new VM exit codes and handle guest accesses, context switch,
and save/restore. At least for now don't allow the guest direct access
to the control MSR; this may need changing if guests were to frequently
access it (e.g. on their own context switch path).

While there also correct a one-off in union ldt_or_tr_instr_info's
comment.

Signed-off-by: Jan Beulich <jbeulich@suse.com>
---
Needing to change two places in hvm.c continues to be unhelpful; I
recall I already did forget to also adjust hvm_load_cpu_msrs() for XFD.
Considering that MSRs typically arrive in the order the table has it,
couldn't we incrementally look up the incoming MSR index there, falling
back to a full lookup only when the incremental lookup failed (and thus
not normally re-iterating through the initial part of the array)?

Said comment in union ldt_or_tr_instr_info is further odd (same for
union gdt_or_idt_instr_info's) in that Instruction Information is only a
32-bit field. Hence bits 32-63 aren't undefined, but simply don't exist.

RFC: The wee attempt to "deal" with nested is likely wrong, but I'm
     afraid I simply don't know where such enforcement would be done
     properly. Returning an error there is also commented out, for
     domain_cpu_policy_changed() returning void without "x86/xstate:
     re-size save area when CPUID policy changes" in place.
---
v5: Introduce user_msr_gpr().
v4: New.

--- a/xen/arch/x86/cpu-policy.c
+++ b/xen/arch/x86/cpu-policy.c
@@ -753,6 +753,12 @@ static void __init calculate_hvm_max_pol
          * situations until someone has cross-checked the behaviour for safety.
          */
         __clear_bit(X86_FEATURE_PKS, fs);
+
+        /*
+         * Don't expose USER_MSR until it is known how (if at all) it is
+         * virtualized on SVM.
+         */
+        __clear_bit(X86_FEATURE_USER_MSR, fs);
     }
 
     if ( !cpu_has_vmx_msrlist )
--- a/xen/arch/x86/domain.c
+++ b/xen/arch/x86/domain.c
@@ -447,6 +447,10 @@ void domain_cpu_policy_changed(struct do
         }
     }
 
+    /* Nested doesn't have the necessary processing, yet. */
+    if ( nestedhvm_enabled(d) && p->feat.user_msr )
+        return /* -EINVAL */;
+
     for_each_vcpu ( d, v )
     {
         cpu_policy_updated(v);
--- a/xen/arch/x86/hvm/hvm.c
+++ b/xen/arch/x86/hvm/hvm.c
@@ -1370,6 +1370,7 @@ static int cf_check hvm_load_cpu_xsave_s
 
 #define HVM_CPU_MSR_SIZE(cnt) offsetof(struct hvm_msr, msr[cnt])
 static const uint32_t msrs_to_send[] = {
+    MSR_USER_MSR_CTL,
     MSR_SPEC_CTRL,
     MSR_INTEL_MISC_FEATURES_ENABLES,
     MSR_PKRS,
@@ -1524,6 +1525,7 @@ static int cf_check hvm_load_cpu_msrs(st
         {
             int rc;
 
+        case MSR_USER_MSR_CTL:
         case MSR_SPEC_CTRL:
         case MSR_INTEL_MISC_FEATURES_ENABLES:
         case MSR_PKRS:
--- a/xen/arch/x86/hvm/vmx/vmx.c
+++ b/xen/arch/x86/hvm/vmx/vmx.c
@@ -676,13 +676,18 @@ static void cf_check vmx_vcpu_destroy(st
 }
 
 /*
- * To avoid MSR save/restore at every VM exit/entry time, we restore
- * the x86_64 specific MSRs at domain switch time. Since these MSRs
- * are not modified once set for para domains, we don't save them,
- * but simply reset them to values set in percpu_traps_init().
+ * To avoid MSR save/restore at every VM exit/entry time, we restore the
+ * x86_64 specific MSRs at vcpu switch time. Since these MSRs are not
+ * modified once set for para domains, we don't save them, but simply clear
+ * them or reset them to values set in percpu_traps_init().
  */
-static void vmx_restore_host_msrs(void)
+static void vmx_restore_host_msrs(const struct vcpu *v)
 {
+    const struct vcpu_msrs *msrs = v->arch.msrs;
+
+    if ( msrs->user_msr_ctl.enable )
+        wrmsrl(MSR_USER_MSR_CTL, 0);
+
     /* No PV guests?  No need to restore host SYSCALL infrastructure. */
     if ( !IS_ENABLED(CONFIG_PV) )
         return;
@@ -736,6 +741,9 @@ static void vmx_restore_guest_msrs(struc
 
     if ( cp->feat.pks )
         wrpkrs(msrs->pkrs);
+
+    if ( msrs->user_msr_ctl.enable )
+        wrmsrl(MSR_USER_MSR_CTL, msrs->user_msr_ctl.raw);
 }
 
 void vmx_update_cpu_exec_control(struct vcpu *v)
@@ -1178,7 +1186,7 @@ static void cf_check vmx_ctxt_switch_fro
     if ( !v->arch.fully_eager_fpu )
         vmx_fpu_leave(v);
     vmx_save_guest_msrs(v);
-    vmx_restore_host_msrs();
+    vmx_restore_host_msrs(v);
     vmx_save_dr(v);
 
     if ( v->domain->arch.hvm.pi_ops.flags & PI_CSW_FROM )
@@ -4080,6 +4088,14 @@ static int vmx_handle_apic_write(void)
     return vlapic_apicv_write(current, exit_qualification & 0xfff);
 }
 
+static unsigned int user_msr_gpr(void)
+{
+    user_msr_instr_info_t info;
+
+    __vmread(VMX_INSTRUCTION_INFO, &info.raw);
+    return info.gpr;
+}
+
 static void undo_nmis_unblocked_by_iret(void)
 {
     unsigned long guest_info;
@@ -4580,6 +4596,41 @@ void asmlinkage vmx_vmexit_handler(struc
             hvm_inject_hw_exception(X86_EXC_GP, 0);
         break;
 
+    case EXIT_REASON_URDMSR:
+    {
+        uint64_t msr_content = 0;
+
+        __vmread(EXIT_QUALIFICATION, &exit_qualification);
+        switch ( hvm_msr_read_intercept(exit_qualification, &msr_content) )
+        {
+        case X86EMUL_OKAY:
+            *decode_gpr(regs, user_msr_gpr()) = msr_content;
+            update_guest_eip(); /* Safe: URDMSR */
+            break;
+
+        case X86EMUL_EXCEPTION:
+            hvm_inject_hw_exception(X86_EXC_GP, 0);
+            break;
+        }
+        break;
+    }
+
+    case EXIT_REASON_UWRMSR:
+        __vmread(EXIT_QUALIFICATION, &exit_qualification);
+        switch ( hvm_msr_write_intercept(exit_qualification,
+                                         *decode_gpr(regs, user_msr_gpr()),
+                                         true) )
+        {
+        case X86EMUL_OKAY:
+            update_guest_eip(); /* Safe: UWRMSR */
+            break;
+
+        case X86EMUL_EXCEPTION:
+            hvm_inject_hw_exception(X86_EXC_GP, 0);
+            break;
+        }
+        break;
+
     case EXIT_REASON_VMXOFF:
     case EXIT_REASON_VMXON:
     case EXIT_REASON_VMCLEAR:
--- a/xen/arch/x86/include/asm/hvm/vmx/vmx.h
+++ b/xen/arch/x86/include/asm/hvm/vmx/vmx.h
@@ -203,6 +203,8 @@ static inline void pi_clear_sn(struct pi
 #define EXIT_REASON_NOTIFY              75
 #define EXIT_REASON_RDMSRLIST           78
 #define EXIT_REASON_WRMSRLIST           79
+#define EXIT_REASON_URDMSR              80
+#define EXIT_REASON_UWRMSR              81
 /* Remember to also update VMX_PERF_EXIT_REASON_SIZE! */
 
 /*
@@ -674,8 +676,18 @@ typedef union ldt_or_tr_instr_info {
         base_reg_invalid        :1,  /* bit 27 - Base register invalid */
         instr_identity          :1,  /* bit 28 - 0:LDT, 1:TR */
         instr_write             :1,  /* bit 29 - 0:store, 1:load */
-                                :34; /* bits 31:63 - Undefined */
+                                :34; /* bits 30:63 - Undefined */
     };
 } ldt_or_tr_instr_info_t;
 
+/* VM-Exit instruction info for URDMSR and UWRMSR */
+typedef union user_msr_instr_info {
+    unsigned long raw;
+    struct {
+        unsigned int            :3,  /* Bits 0:2 - Undefined */
+        gpr                     :4,  /* Bits 3:6 - Source/Destination register */
+                                :25; /* bits 7:31 - Undefined */
+    };
+} user_msr_instr_info_t;
+
 #endif /* __ASM_X86_HVM_VMX_VMX_H__ */
--- a/xen/arch/x86/include/asm/msr.h
+++ b/xen/arch/x86/include/asm/msr.h
@@ -296,6 +296,20 @@ uint64_t msr_spec_ctrl_valid_bits(const
 struct vcpu_msrs
 {
     /*
+     * 0x0000001c - MSR_USER_MSR_CTL
+     *
+     * Value is guest chosen, and always loaded in vcpu context.
+     */
+    union {
+        uint64_t raw;
+        struct {
+            bool enable:1;
+            unsigned int :11;
+            unsigned long bitmap:52;
+        };
+    } user_msr_ctl;
+
+    /*
      * 0x00000048 - MSR_SPEC_CTRL
      * 0xc001011f - MSR_VIRT_SPEC_CTRL (if X86_FEATURE_AMD_SSBD)
      *
--- a/xen/arch/x86/include/asm/perfc_defn.h
+++ b/xen/arch/x86/include/asm/perfc_defn.h
@@ -6,7 +6,7 @@ PERFCOUNTER_ARRAY(exceptions,
 
 #ifdef CONFIG_HVM
 
-#define VMX_PERF_EXIT_REASON_SIZE 80
+#define VMX_PERF_EXIT_REASON_SIZE 82
 #define VMEXIT_NPF_PERFC 143
 #define SVM_PERF_EXIT_REASON_SIZE (VMEXIT_NPF_PERFC + 1)
 PERFCOUNTER_ARRAY(vmexits,              "vmexits",
--- a/xen/arch/x86/msr.c
+++ b/xen/arch/x86/msr.c
@@ -206,6 +206,12 @@ int guest_rdmsr(struct vcpu *v, uint32_t
         *val = msrs->xss.raw;
         break;
 
+    case MSR_USER_MSR_CTL:
+        if ( !cp->feat.user_msr )
+            goto gp_fault;
+        *val = msrs->user_msr_ctl.raw;
+        break;
+
     case 0x40000000 ... 0x400001ff:
         if ( is_viridian_domain(d) )
         {
@@ -536,6 +542,19 @@ int guest_wrmsr(struct vcpu *v, uint32_t
         msrs->xss.raw = val;
         break;
 
+    case MSR_USER_MSR_CTL:
+        if ( !cp->feat.user_msr )
+            goto gp_fault;
+
+        if ( (val & ~(USER_MSR_ENABLE | USER_MSR_ADDR_MASK)) ||
+             !is_canonical_address(val) )
+            goto gp_fault;
+
+        msrs->user_msr_ctl.raw = val;
+        if ( v == curr )
+            wrmsrl(MSR_USER_MSR_CTL, val);
+        break;
+
     case 0x40000000 ... 0x400001ff:
         if ( is_viridian_domain(d) )
         {
--- a/xen/include/public/arch-x86/cpufeatureset.h
+++ b/xen/include/public/arch-x86/cpufeatureset.h
@@ -350,7 +350,7 @@ XEN_CPUFEATURE(AVX_NE_CONVERT,     15*32
 XEN_CPUFEATURE(AMX_COMPLEX,        15*32+ 8) /*   AMX Complex Instructions */
 XEN_CPUFEATURE(AVX_VNNI_INT16,     15*32+10) /*A  AVX-VNNI-INT16 Instructions */
 XEN_CPUFEATURE(PREFETCHI,          15*32+14) /*A  PREFETCHIT{0,1} Instructions */
-XEN_CPUFEATURE(USER_MSR,           15*32+15) /*   U{RD,WR}MSR Instructions */
+XEN_CPUFEATURE(USER_MSR,           15*32+15) /*s  U{RD,WR}MSR Instructions */
 XEN_CPUFEATURE(CET_SSS,            15*32+18) /*   CET Supervisor Shadow Stacks safe to use */
 
 /* Intel-defined CPU features, MSR_ARCH_CAPS 0x10a.eax, word 16 */