[PATCH v3 6/6] x86emul: support USER_MSR instructions

[Jan Beulich <jbeulich@xxxxxxxx>]

While UWRMSR probably isn't of much use as long as we don't support
UINTR, URDMSR may well be useful to guests even without that (depending
on what OSes are willing to permit access to).

Since the two VEX encodings introduce a lonely opcode point in map 7,
for now don't bother introducing a full 256-entry table.

Signed-off-by: Jan Beulich <jbeulich@xxxxxxxx>
---
The retaining of (possible) #PF from the bitmap access is "speculative"
(the spec doesn't mention #PF as a possible exception; conceivably this
might also need converting to #GP).

I'm a little wary of the "MSRs Writeable by UWRMSR" table that the spec
has, and that our code thus also enforces: As new MSRs are added to that
table, we'll need piecemeal updates to that switch() statement.

The choice of using APERF in the test harness is connected to the also
pending RDPRU patch, where the register needs handling anyway.
---
v3: New.

--- a/tools/misc/xen-cpuid.c
+++ b/tools/misc/xen-cpuid.c
@@ -224,7 +224,7 @@ static const char *const str_7d1[32] =
 {
     [ 4] = "avx-vnni-int8",       [ 5] = "avx-ne-convert",
 
-    [14] = "prefetchi",
+    [14] = "prefetchi",           [15] = "user-msr",
 
     [18] = "cet-sss",
 };
--- a/tools/tests/x86_emulator/predicates.c
+++ b/tools/tests/x86_emulator/predicates.c
@@ -864,7 +864,9 @@ static const struct {
     { { 0xf6 }, { 2, 2 }, T, R, pfx_66 }, /* adcx */
     { { 0xf6 }, { 2, 2 }, T, R, pfx_f3 }, /* adox */
     { { 0xf8 }, { 2, 2 }, F, W, pfx_66 }, /* movdir64b */
+    { { 0xf8, 0xc0 }, { 0, 2 }, F, N, pfx_f3 }, /* uwrmsr */
     { { 0xf8 }, { 2, 2 }, F, W, pfx_f3 }, /* enqcmds */
+    { { 0xf8, 0xc0 }, { 0, 2 }, F, N, pfx_f2 }, /* urdmsr */
     { { 0xf8 }, { 2, 2 }, F, W, pfx_f2 }, /* enqcmd */
     { { 0xf9 }, { 2, 2 }, F, W }, /* movdiri */
 };
@@ -1502,6 +1504,9 @@ static const struct vex {
     { { 0xcf }, 3, T, R, pfx_66, W1, Ln }, /* vgf2p8affineinvqb */
     { { 0xdf }, 3, T, R, pfx_66, WIG, Ln }, /* vaeskeygenassist */
     { { 0xf0 }, 3, T, R, pfx_f2, Wn, L0 }, /* rorx */
+}, vex_map7[] = {
+    { { 0xf8, 0xc0 }, 6, F, N, pfx_f3, W0, L0 }, /* uwrmsr */
+    { { 0xf8, 0xc0 }, 6, F, N, pfx_f2, W0, L0 }, /* urdmsr */
 };
 
 static const struct {
@@ -1511,6 +1516,10 @@ static const struct {
     { vex_0f,   ARRAY_SIZE(vex_0f) },
     { vex_0f38, ARRAY_SIZE(vex_0f38) },
     { vex_0f3a, ARRAY_SIZE(vex_0f3a) },
+    { NULL,     0 }, /* map 4 */
+    { NULL,     0 }, /* map 5 */
+    { NULL,     0 }, /* map 6 */
+    { vex_map7, ARRAY_SIZE(vex_map7) },
 };
 
 static const struct xop {
@@ -2411,7 +2420,8 @@ void predicates_test(void *instr, struct
 
                 if ( vex[x].tbl[t].w == WIG || (vex[x].tbl[t].w & W0) )
                 {
-                    memcpy(ptr, vex[x].tbl[t].opc, vex[x].tbl[t].len);
+                    memcpy(ptr, vex[x].tbl[t].opc,
+                           MIN(vex[x].tbl[t].len, ARRAY_SIZE(vex->tbl->opc)));
 
                     if ( vex[x].tbl[t].l == LIG || (vex[x].tbl[t].l & L0) )
                         do_test(instr, vex[x].tbl[t].len + ((void *)ptr - 
instr),
@@ -2421,7 +2431,8 @@ void predicates_test(void *instr, struct
                     if ( vex[x].tbl[t].l == LIG || (vex[x].tbl[t].l & L1) )
                     {
                         ptr[-1] |= 4;
-                        memcpy(ptr, vex[x].tbl[t].opc, vex[x].tbl[t].len);
+                        memcpy(ptr, vex[x].tbl[t].opc,
+                               MIN(vex[x].tbl[t].len, 
ARRAY_SIZE(vex->tbl->opc)));
 
                         do_test(instr, vex[x].tbl[t].len + ((void *)ptr - 
instr),
                                 vex[x].tbl[t].modrm ? (void *)ptr - instr + 1 
: 0,
@@ -2432,7 +2443,8 @@ void predicates_test(void *instr, struct
                 if ( vex[x].tbl[t].w == WIG || (vex[x].tbl[t].w & W1) )
                 {
                     ptr[-1] = 0xf8 | vex[x].tbl[t].pfx;
-                    memcpy(ptr, vex[x].tbl[t].opc, vex[x].tbl[t].len);
+                    memcpy(ptr, vex[x].tbl[t].opc,
+                           MIN(vex[x].tbl[t].len, ARRAY_SIZE(vex->tbl->opc)));
 
                     if ( vex[x].tbl[t].l == LIG || (vex[x].tbl[t].l & L0) )
                         do_test(instr, vex[x].tbl[t].len + ((void *)ptr - 
instr),
@@ -2442,7 +2454,8 @@ void predicates_test(void *instr, struct
                     if ( vex[x].tbl[t].l == LIG || (vex[x].tbl[t].l & L1) )
                     {
                         ptr[-1] |= 4;
-                        memcpy(ptr, vex[x].tbl[t].opc, vex[x].tbl[t].len);
+                        memcpy(ptr, vex[x].tbl[t].opc,
+                               MIN(vex[x].tbl[t].len, 
ARRAY_SIZE(vex->tbl->opc)));
 
                         do_test(instr, vex[x].tbl[t].len + ((void *)ptr - 
instr),
                                 vex[x].tbl[t].modrm ? (void *)ptr - instr + 1 
: 0,
--- a/tools/tests/x86_emulator/test_x86_emulator.c
+++ b/tools/tests/x86_emulator/test_x86_emulator.c
@@ -684,6 +684,7 @@ static int blk(
 
 #ifdef __x86_64__
 static unsigned long gs_base, gs_base_shadow;
+static unsigned long uintr_timer;
 #endif
 
 static int read_segment(
@@ -718,6 +719,15 @@ static int write_segment(
 
     return X86EMUL_OKAY;
 }
+
+static const uint8_t __attribute__((aligned(0x1000))) umsr_bitmap[0x1000] = {
+#define RD(msr) [(msr) >> 3] = 1 << ((msr) & 7)
+#define WR(msr) [0x800 + ((msr) >> 3)] = 1 << ((msr) & 7)
+    RD(0x000000e8), /* APERF */
+    WR(0x00001b00), /* UINTR_TIMER */
+#undef WR
+#undef RD
+};
 #endif
 
 static int read_msr(
@@ -727,10 +737,22 @@ static int read_msr(
 {
     switch ( reg )
     {
+#ifdef __x86_64__
+    case 0x0000001c: /* USER_MSR_CTL */
+        *val = (unsigned long)umsr_bitmap | 1;
+        return X86EMUL_OKAY;
+#endif
+
     case 0x0000002f: /* BARRIER */
         *val = 0;
         return X86EMUL_OKAY;
 
+    case 0x000000e8: /* APERF */
+#define APERF_LO_VALUE 0xAEAEAEAE
+#define APERF_HI_VALUE 0xEAEAEAEA
+        *val = ((uint64_t)APERF_HI_VALUE << 32) | APERF_LO_VALUE;
+        return X86EMUL_OKAY;
+
     case 0xc0000080: /* EFER */
         *val = ctxt->addr_size > 32 ? 0x500 /* LME|LMA */ : 0;
         return X86EMUL_OKAY;
@@ -766,6 +788,12 @@ static int write_msr(
 {
     switch ( reg )
     {
+    case 0x00001b00: /* UINTR_TIMER */
+        if ( ctxt->addr_size < 64 )
+            break;
+        uintr_timer = val;
+        return X86EMUL_OKAY;
+
     case 0xc0000101: /* GS_BASE */
         if ( ctxt->addr_size < 64 || !is_canonical_address(val) )
             break;
@@ -1565,6 +1593,62 @@ int main(int argc, char **argv)
          (gs_base != 0x0000222244446666UL) ||
          (gs_base_shadow != 0x0000111122224444UL) )
         goto fail;
+    printf("okay\n");
+
+    printf("%-40s", "Testing urdmsr %edx,%rcx...");
+    instr[0] = 0xf2; instr[1] = 0x0f; instr[2] = 0x38; instr[3] = 0xf8; 
instr[4] = 0xd1;
+    regs.rip = (unsigned long)&instr[0];
+    regs.rdx = 0x12345678000000e8UL; /* APERF; high half to be ignored */
+    rc = x86_emulate(&ctxt, &emulops);
+    if ( (rc != X86EMUL_OKAY) ||
+         (regs.rip != (unsigned long)&instr[5]) ||
+         (regs.rcx != (((uint64_t)APERF_HI_VALUE << 32) | APERF_LO_VALUE)) )
+        goto fail;
+    printf("okay\n");
+
+    printf("%-40s", "Testing urdmsr $0xe8,%rdx...");
+    instr[0] = 0xc4; instr[1] = 0xe7; instr[2] = 0x7b; instr[3] = 0xf8; 
instr[4] = 0xc2;
+    instr[5] = 0xe8; instr[6] = 0; instr[7] = 0; instr[8] = 0;
+    regs.rip = (unsigned long)&instr[0];
+    rc = x86_emulate(&ctxt, &emulops);
+    if ( (rc != X86EMUL_OKAY) ||
+         (regs.rip != (unsigned long)&instr[9]) ||
+         (regs.rdx != (((uint64_t)APERF_HI_VALUE << 32) | APERF_LO_VALUE)) )
+        goto fail;
+    printf("okay\n");
+
+    printf("%-40s", "Testing uwrmsr %rdi,%esi...");
+    instr[0] = 0xf3; instr[1] = 0x0f; instr[2] = 0x38; instr[3] = 0xf8; 
instr[4] = 0xf7;
+    regs.rip = (unsigned long)&instr[0];
+    regs.rsi = 0x4433221100001b00UL; /* UINTR_TIMER; high half to be ignored */
+    regs.rdi = 0x0011223344556677UL;
+    rc = x86_emulate(&ctxt, &emulops);
+    if ( (rc != X86EMUL_OKAY) ||
+         (regs.rip != (unsigned long)&instr[5]) ||
+         (uintr_timer != 0x0011223344556677UL) )
+        goto fail;
+    printf("okay\n");
+
+    printf("%-40s", "Testing uwrmsr %rsi,$0x1b00...");
+    instr[0] = 0xc4; instr[1] = 0xe7; instr[2] = 0x7a; instr[3] = 0xf8; 
instr[4] = 0xc6;
+    instr[5] = 0x00; instr[6] = 0x1b; instr[7] = 0; instr[8] = 0;
+    regs.rip = (unsigned long)&instr[0];
+    regs.rsi = 0x8877665544332211UL;
+    rc = x86_emulate(&ctxt, &emulops);
+    if ( (rc != X86EMUL_OKAY) ||
+         (regs.rip != (unsigned long)&instr[9]) ||
+         (uintr_timer != 0x8877665544332211UL) )
+        goto fail;
+
+    printf("%-40s", "Testing uwrmsr %rsi,$0x1b01...");
+    instr[5] = 0x01; /* UARCH_MISC_CTL (derived from UINTR_TIMER) */
+    regs.rip = (unsigned long)&instr[0];
+    regs.rsi = 0;
+    rc = x86_emulate(&ctxt, &emulops);
+    if ( (rc != X86EMUL_EXCEPTION) ||
+         (regs.rip != (unsigned long)&instr[0]) ||
+         (uintr_timer != 0x8877665544332211UL) )
+        goto fail;
 
     emulops.write_msr     = NULL;
 #endif
--- a/xen/arch/x86/include/asm/msr-index.h
+++ b/xen/arch/x86/include/asm/msr-index.h
@@ -24,6 +24,10 @@
 #define  APIC_BASE_ENABLE                   (_AC(1, ULL) << 11)
 #define  APIC_BASE_ADDR_MASK                _AC(0x000ffffffffff000, ULL)
 
+#define MSR_USER_MSR_CTL                    0x0000001c
+#define  USER_MSR_ENABLE                    (_AC(1, ULL) <<  0)
+#define  USER_MSR_ADDR_MASK                 0xfffffffffffff000ULL
+
 #define MSR_BARRIER                         0x0000002f
 
 #define MSR_TEST_CTRL                       0x00000033
--- a/xen/arch/x86/x86_emulate/decode.c
+++ b/xen/arch/x86/x86_emulate/decode.c
@@ -901,7 +901,7 @@ decode_0f38(struct x86_emulate_state *s,
     {
     case 0x00 ... 0xef:
     case 0xf2 ... 0xf5:
-    case 0xf7 ... 0xf8:
+    case 0xf7:
     case 0xfa ... 0xff:
         s->op_bytes = 0;
         /* fall through */
@@ -946,6 +946,18 @@ decode_0f38(struct x86_emulate_state *s,
     case X86EMUL_OPC_VEX_F2(0, 0xf7): /* shrx */
         break;
 
+    case 0xf8:
+        if ( s->modrm_mod == 3 ) /* u{rd,wr}msr */
+        {
+            s->desc = DstMem | SrcReg | Mov;
+            s->op_bytes = 4; /* Describes the MSR index. */
+            s->simd_size = simd_none;
+        }
+        else /* movdir64b / enqcmd{,s} */
+            s->op_bytes = 0;
+        ctxt->opcode |= MASK_INSR(s->vex.pfx, X86EMUL_OPC_PFX_MASK);
+        break;
+
     default:
         s->op_bytes = 0;
         break;
@@ -1251,6 +1263,16 @@ int x86emul_decode(struct x86_emulate_st
                          */
                         d = twobyte_table[0x38].desc;
                         break;
+
+                    case vex_map7:
+                        opcode |= MASK_INSR(7, X86EMUL_OPC_EXT_MASK);
+                        /*
+                         * No table lookup here for now, as there's only a 
single
+                         * opcode point populated in map 7.
+                         */
+                        if ( b == 0xf8 )
+                            d = DstMem | SrcImm | ModRM | Mov;
+                        break;
                     }
                 }
                 else if ( s->ext < ext_8f08 + ARRAY_SIZE(xop_table) )
@@ -1619,6 +1641,7 @@ int x86emul_decode(struct x86_emulate_st
             s->simd_size = ext8f09_table[b].simd_size;
             break;
 
+        case ext_map7:
         case ext_8f08:
         case ext_8f0a:
             /*
@@ -1833,6 +1856,7 @@ int x86emul_decode(struct x86_emulate_st
 
     case ext_map5:
     case ext_map6:
+    case ext_map7:
     case ext_8f09:
     case ext_8f0a:
         break;
--- a/xen/arch/x86/x86_emulate/private.h
+++ b/xen/arch/x86/x86_emulate/private.h
@@ -185,6 +185,7 @@ enum vex_opcx {
     vex_0f3a,
     evex_map5 = 5,
     evex_map6,
+    vex_map7,
 };
 
 enum vex_pfx {
@@ -241,6 +242,7 @@ struct x86_emulate_state {
         ext_0f3a = vex_0f3a,
         ext_map5 = evex_map5,
         ext_map6 = evex_map6,
+        ext_map7 = vex_map7,
         /*
          * For XOP use values such that the respective instruction field
          * can be used without adjustment.
--- a/xen/arch/x86/x86_emulate/x86_emulate.c
+++ b/xen/arch/x86/x86_emulate/x86_emulate.c
@@ -7109,10 +7109,68 @@ x86_emulate(
         state->simd_size = simd_none;
         break;
 
-    case X86EMUL_OPC_F2(0x0f38, 0xf8): /* enqcmd r,m512 */
-    case X86EMUL_OPC_F3(0x0f38, 0xf8): /* enqcmds r,m512 */
+    case X86EMUL_OPC_F2(0x0f38, 0xf8): /* enqcmd r,m512 / urdmsr r32,r64 */
+    case X86EMUL_OPC_F3(0x0f38, 0xf8): /* enqcmds r,m512 / uwrmsr r64,r32 */
+        if ( ea.type == OP_MEM )
+            goto enqcmd;
+        imm1 = src.val;
+        /* fall through */
+    case X86EMUL_OPC_VEX_F2(7, 0xf8): /* urdmsr imm32,r64 */
+    case X86EMUL_OPC_VEX_F3(7, 0xf8): /* uwrmsr r64,imm32 */
+        generate_exception_if(!mode_64bit() || ea.type != OP_REG, X86_EXC_UD);
+        generate_exception_if(vex.l || vex.w, X86_EXC_UD);
+        generate_exception_if(vex.opcx && (modrm_reg || vex.reg != 0xf),
+                              X86_EXC_UD);
+        fail_if(!ops->read_msr);
+        if ( ops->read_msr(MSR_USER_MSR_CTL, &msr_val, ctxt) != X86EMUL_OKAY )
+        {
+            x86_emul_reset_event(ctxt);
+            msr_val = 0;
+        }
+        generate_exception_if(!(msr_val & USER_MSR_ENABLE), X86_EXC_UD);
+        generate_exception_if(imm1 & ~0x3fff, X86_EXC_GP, 0);
+
+        /* Check the corresponding bitmap. */
+        ea.mem.off = msr_val & ~0xfff;
+        if ( vex.pfx != vex_f2 )
+            ea.mem.off += 0x800;
+        ea.mem.off += imm1 >> 3;
+        if ( (rc = ops->read(x86_seg_sys, ea.mem.off, &b, 1,
+                             ctxt)) != X86EMUL_OKAY )
+            goto done;
+        generate_exception_if(!(b & (1 << (imm1 & 7))), X86_EXC_GP, 0);
+
+        /* Carry out the actual MSR access. */
+        if ( vex.pfx == vex_f2 )
+        {
+            /* urdmsr */
+            if ( (rc = ops->read_msr(imm1, &msr_val, ctxt)) != X86EMUL_OKAY )
+                goto done;
+            dst.val = msr_val;
+            ASSERT(dst.type == OP_REG);
+            dst.bytes = 8;
+        }
+        else
+        {
+            /* uwrmsr */
+            switch ( imm1 )
+            {
+            case 0x1b00: /* UINTR_TIMER */
+            case 0x1b01: /* UARCH_MISC_CTL */
+                break;
+            default:
+                generate_exception(X86_EXC_GP, 0);
+            }
+            fail_if(!ops->write_msr);
+            /* NB: Cannot use dst.val here, as op_bytes was set to 4. */
+            if ( (rc = ops->write_msr(imm1, *dst.reg, ctxt)) != X86EMUL_OKAY )
+                goto done;
+            dst.type = OP_NONE;
+        }
+        break;
+
+    enqcmd:
         host_and_vcpu_must_have(enqcmd);
-        generate_exception_if(ea.type != OP_MEM, X86_EXC_UD);
         generate_exception_if(vex.pfx != vex_f2 && !mode_ring0(), X86_EXC_GP, 
0);
         src.val = truncate_ea(*dst.reg);
         generate_exception_if(!is_aligned(x86_seg_es, src.val, 64, ctxt, ops),
--- a/xen/include/public/arch-x86/cpufeatureset.h
+++ b/xen/include/public/arch-x86/cpufeatureset.h
@@ -315,6 +315,7 @@ XEN_CPUFEATURE(MCDT_NO,            13*32
 XEN_CPUFEATURE(AVX_VNNI_INT8,      15*32+ 4) /*A  AVX-VNNI-INT8 Instructions */
 XEN_CPUFEATURE(AVX_NE_CONVERT,     15*32+ 5) /*A  AVX-NE-CONVERT 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(CET_SSS,            15*32+18) /*   CET Supervisor Shadow Stacks 
safe to use */
 
 /* Intel-defined CPU features, MSR_ARCH_CAPS 0x10a.eax, word 16 */
--- a/xen/tools/gen-cpuid.py
+++ b/xen/tools/gen-cpuid.py
@@ -275,7 +275,7 @@ def crunch_numbers(state):
         # NO_LMSL indicates the absense of Long Mode Segment Limits, which
         # have been dropped in hardware.
         LM: [CX16, PCID, LAHF_LM, PAGE1GB, PKU, NO_LMSL, LKGS, CMPCCXADD,
-             MSRLIST],
+             MSRLIST, USER_MSR],
 
         # AMD K6-2+ and K6-III processors shipped with 3DNow+, beyond the
         # standard 3DNow in the earlier K6 processors.