[PATCH v4 01/16] x86emul: support SIMD MOVRS

[Jan Beulich <jbeulich@xxxxxxxx>]

As we ignore cachability aspects of insns, they're treated like simple
VMOVs.

Signed-off-by: Jan Beulich <jbeulich@xxxxxxxx>
---
SDE: -dmr
---
v4: Correct main switch case labels to reference Map5. Switch to using
    fallthrough pseudo-keyword. Move ahead in series, switching feature
    dependency to plain AVX10.
v3: New.

--- a/tools/tests/x86_emulator/evex-disp8.c
+++ b/tools/tests/x86_emulator/evex-disp8.c
@@ -720,6 +720,13 @@ static const struct test vpclmulqdq_all[
     INSN(pclmulqdq, 66, 0f3a, 44, vl, q_nb, vl)
 };
 
+static const struct test movrs_all[] = {
+    INSN(movrsb, f2, map5, 6f, vl,    b, vl),
+    INSN(movrsd, f3, map5, 6f, vl, d_nb, vl),
+    INSN(movrsq, f3, map5, 6f, vl, q_nb, vl),
+    INSN(movrsw, f2, map5, 6f, vl,    w, vl),
+};
+
 static const unsigned char vl_all[] = { VL_512, VL_128, VL_256 };
 static const unsigned char vl_128[] = { VL_128 };
 static const unsigned char vl_no128[] = { VL_512, VL_256 };
@@ -1071,8 +1078,8 @@ void evex_disp8_test(void *instr, struct
     emulops.read = read;
     emulops.write = write;
 
-#define RUN(feat, vl) do { \
-    if ( cpu_has_##feat ) \
+#define run(cond, feat, vl) do { \
+    if ( cond ) \
     { \
         printf("%-40s", "Testing " #feat "/" #vl " disp8 handling..."); \
         test_group(feat ## _ ## vl, ARRAY_SIZE(feat ## _ ## vl), \
@@ -1081,6 +1088,8 @@ void evex_disp8_test(void *instr, struct
     } \
 } while ( false )
 
+#define RUN(feat, vl) run(cpu_has_ ## feat, feat, vl)
+
     RUN(avx512f, all);
     RUN(avx512f, 128);
     RUN(avx512f, no128);
@@ -1111,4 +1120,9 @@ void evex_disp8_test(void *instr, struct
         RUN(vaes, all);
         RUN(vpclmulqdq, all);
     }
+
+    if ( cpu_has_avx10 )
+    {
+        run(ctxt->addr_size == 64 && cpu_has_movrs, movrs, all);
+    }
 }
--- a/tools/tests/x86_emulator/predicates.c
+++ b/tools/tests/x86_emulator/predicates.c
@@ -2124,6 +2124,8 @@ static const struct evex {
     { { 0x5f }, 2, T, R, pfx_no, W0, Ln }, /* vmaxph */
     { { 0x5f }, 2, T, R, pfx_f3, W0, LIG }, /* vmaxsh */
     { { 0x6e }, 2, T, R, pfx_66, WIG, L0 }, /* vmovw */
+    { { 0x6f }, 2, T, R, pfx_f3, Wn, Ln }, /* vmovrs{d,q} */
+    { { 0x6f }, 2, T, R, pfx_f2, Wn, Ln }, /* vmovrs{b,w} */
     { { 0x78 }, 2, T, R, pfx_no, W0, Ln }, /* vcvttph2udq */
     { { 0x78 }, 2, T, R, pfx_66, W0, Ln }, /* vcvttph2uqq */
     { { 0x78 }, 2, T, R, pfx_f3, Wn, LIG }, /* vcvttsh2usi */
--- a/tools/tests/x86_emulator/x86-emulate.h
+++ b/tools/tests/x86_emulator/x86-emulate.h
@@ -209,12 +209,14 @@ void wrpkru(unsigned int val);
                                      xcr0_mask(0xe6))
 #define cpu_has_cmpccxadd            cpu_policy.feat.cmpccxadd
 #define cpu_has_avx_ifma            (cpu_policy.feat.avx_ifma && xcr0_mask(6))
+#define cpu_has_movrs                cpu_policy.feat.movrs
 #define cpu_has_avx_vnni_int8       (cpu_policy.feat.avx_vnni_int8 && \
                                      xcr0_mask(6))
 #define cpu_has_avx_ne_convert      (cpu_policy.feat.avx_ne_convert && \
                                      xcr0_mask(6))
 #define cpu_has_avx_vnni_int16      (cpu_policy.feat.avx_vnni_int16 && \
                                      xcr0_mask(6))
+#define cpu_has_avx10               (cpu_policy.feat.avx10 && xcr0_mask(0xe6))
 
 #define cpu_has_xgetbv1             (cpu_has_xsave && 
cpu_policy.xstate.xgetbv1)
 
--- a/xen/arch/x86/x86_emulate/private.h
+++ b/xen/arch/x86/x86_emulate/private.h
@@ -622,6 +622,8 @@ amd_like(const struct x86_emulate_ctxt *
 #define vcpu_has_avx_vnni_int16() (ctxt->cpuid->feat.avx_vnni_int16)
 #define vcpu_has_user_msr()    (ctxt->cpuid->feat.user_msr)
 
+#define vcpu_has_avx10()       (ctxt->cpuid->feat.avx10)
+
 #define vcpu_must_have(feat) \
     generate_exception_if(!vcpu_has_##feat(), X86_EXC_UD)
 
--- a/xen/arch/x86/x86_emulate/x86_emulate.c
+++ b/xen/arch/x86/x86_emulate/x86_emulate.c
@@ -7880,6 +7880,17 @@ x86_emulate(
         op_bytes = 8 << evex.lr;
         goto simd_zmm;
 
+    case X86EMUL_OPC_EVEX_F2(5, 0x6f): /* vmovrs{b,w} mem,[xyz]mm{k} */
+        elem_bytes = 1 << evex.w;
+        fallthrough;
+    case X86EMUL_OPC_EVEX_F3(5, 0x6f): /* vmovrs{d,q} mem,[xyz]mm{k} */
+        generate_exception_if(ea.type != OP_MEM || evex.brs, X86_EXC_UD);
+        vcpu_must_have(avx10);
+        vcpu_must_have(movrs);
+        avx512_vlen_check(false);
+        op_bytes = 16 << evex.lr;
+        goto simd_zmm;
+
     case X86EMUL_OPC_EVEX_66(5, 0x78): /* vcvttph2uqq xmm/mem,[xyz]mm{k} */
     case X86EMUL_OPC_EVEX_66(5, 0x79): /* vcvtph2uqq xmm/mem,[xyz]mm{k} */
     case X86EMUL_OPC_EVEX_66(5, 0x7a): /* vcvttph2qq xmm/mem,[xyz]mm{k} */