[PATCH 2/2] x86emul: use latched XCR0 in x86emul_get_fpu()

Avoid re-reading, and instead pass state into the function. For
get_fpu() to use "s" as new argument, we need a new local variable in
x86_emulate() though.

Signed-off-by: Jan Beulich <[email protected]>
---
That new local "s" can likely be leveraged to replace explicit uses of
"state" in the function (past the point where the alias #define is).
Long term we probably want to aim at consistently using "s" everywhere
where a struct x86_emulate_state * variable/parameter is needed.
x86emul_get_fpu() isn't using "s" right away just because that would end
up inconsistent with adjacent code (put_fpu() first and foremost). I
could certainly change that.

--- a/xen/arch/x86/x86_emulate/private.h
+++ b/xen/arch/x86/x86_emulate/private.h
@@ -743,12 +743,13 @@ int x86emul_get_cpl(struct x86_emulate_c
                     const struct x86_emulate_ops *ops);
 
 int x86emul_get_fpu(enum x86_emulate_fpu_type type,
+                    const struct x86_emulate_state *state,
                     struct x86_emulate_ctxt *ctxt,
                     const struct x86_emulate_ops *ops);
 
 #define get_fpu(type)                                           \
 do {                                                            \
-    rc = x86emul_get_fpu(fpu_type = (type), ctxt, ops);         \
+    rc = x86emul_get_fpu(fpu_type = (type), s, ctxt, ops);      \
     if ( rc ) goto done;                                        \
 } while (0)
 
--- a/xen/arch/x86/x86_emulate/x86_emulate.c
+++ b/xen/arch/x86/x86_emulate/x86_emulate.c
@@ -394,36 +394,30 @@ do {
 
 int x86emul_get_fpu(
     enum x86_emulate_fpu_type type,
+    const struct x86_emulate_state *state,
     struct x86_emulate_ctxt *ctxt,
     const struct x86_emulate_ops *ops)
 {
-    uint64_t xcr0;
     int rc;
 
     fail_if(!ops->get_fpu);
     ASSERT(type != X86EMUL_FPU_none);
 
-    if ( type < X86EMUL_FPU_ymm || !ops->read_xcr ||
-         ops->read_xcr(0, &xcr0, ctxt) != X86EMUL_OKAY )
-    {
-        ASSERT(!ctxt->event_pending);
-        xcr0 = 0;
-    }
-
     switch ( type )
     {
     case X86EMUL_FPU_zmm:
-        if ( !(xcr0 & X86_XCR0_ZMM) || !(xcr0 & X86_XCR0_HI_ZMM) ||
-             !(xcr0 & X86_XCR0_OPMASK) )
+        if ( !(state->xcr0 & X86_XCR0_ZMM) ||
+             !(state->xcr0 & X86_XCR0_HI_ZMM) ||
+             !(state->xcr0 & X86_XCR0_OPMASK) )
             return X86EMUL_UNHANDLEABLE;
         /* fall through */
     case X86EMUL_FPU_ymm:
-        if ( !(xcr0 & X86_XCR0_SSE) || !(xcr0 & X86_XCR0_YMM) )
+        if ( !(state->xcr0 & X86_XCR0_SSE) || !(state->xcr0 & X86_XCR0_YMM) )
             return X86EMUL_UNHANDLEABLE;
         break;
 
     case X86EMUL_FPU_opmask:
-        if ( !(xcr0 & X86_XCR0_SSE) || !(xcr0 & X86_XCR0_OPMASK) )
+        if ( !(state->xcr0 & X86_XCR0_SSE) || !(state->xcr0 & X86_XCR0_OPMASK) )
             return X86EMUL_UNHANDLEABLE;
         break;
 
@@ -1310,7 +1304,7 @@ x86_emulate(
     /* Shadow copy of register state. Committed on successful emulation. */
     struct cpu_user_regs _regs = *ctxt->regs;
     const struct cpu_policy *__maybe_unused cp = ctxt->cpu_policy;
-    struct x86_emulate_state state;
+    struct x86_emulate_state state, *s = &state;
     int rc;
     uint8_t b, d, *opc = NULL;
     unsigned int first_byte = 0, elem_bytes, insn_bytes = 0;
@@ -1439,7 +1433,7 @@ x86_emulate(
     /* With a memory operand, fetch the mask register in use (if any). */
     if ( ea.type == OP_MEM && evex.opmsk &&
          x86emul_get_fpu(fpu_type = X86EMUL_FPU_opmask,
-                         ctxt, ops) == X86EMUL_OKAY )
+                         s, ctxt, ops) == X86EMUL_OKAY )
     {
         uint8_t *stb = get_stub(stub);