* [PATCH 0/2] x86emul: XCR0 handling
@ 2026-08-20 8:50 Jan Beulich
2026-08-20 8:52 ` [PATCH 1/2] x86emul: latch XCR0 early during decode Jan Beulich
2026-08-20 8:52 ` [PATCH 2/2] x86emul: use latched XCR0 in x86emul_get_fpu() Jan Beulich
0 siblings, 2 replies; 3+ messages in thread
From: Jan Beulich @ 2026-08-20 8:50 UTC (permalink / raw)
To: xen-devel@lists.xenproject.org
Cc: Andrew Cooper, Teddy Astie, Roger Pau Monné, Andrew Mbugua
Prompted by Andrew's report [1] I thought it may help if we accelerated two
of the patches I have pending as part of the (so far unposted) APX series.
We previously discussed latching some control register state once early, so
let's start with doing so for XCR0.
[1] https://lists.xen.org/archives/html/xen-devel/2026-08/msg01006.html
1: latch XCR0 early during decode
2: use latched XCR0 in x86emul_get_fpu()
Jan
^ permalink raw reply [flat|nested] 3+ messages in thread
* [PATCH 1/2] x86emul: latch XCR0 early during decode
2026-08-20 8:50 [PATCH 0/2] x86emul: XCR0 handling Jan Beulich
@ 2026-08-20 8:52 ` Jan Beulich
2026-08-20 8:52 ` [PATCH 2/2] x86emul: use latched XCR0 in x86emul_get_fpu() Jan Beulich
1 sibling, 0 replies; 3+ messages in thread
From: Jan Beulich @ 2026-08-20 8:52 UTC (permalink / raw)
To: xen-devel@lists.xenproject.org
Cc: Andrew Cooper, Teddy Astie, Roger Pau Monné, Andrew Mbugua
We'll need to consult it to decide whether to treat 0xD5 as a REX2
prefix, and whether to recognize extended EVEX encodings.
Utilize this in adjust_bnd() right away, but leave leveraging in
x86emul_get_fpu() for a separate change.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
---
While not directly relevant in this series, should we perhaps latch CR4
into the state structure as well right away, since we need to read it
here anyway?
--- a/xen/arch/x86/x86_emulate/decode.c
+++ b/xen/arch/x86/x86_emulate/decode.c
@@ -1036,6 +1036,24 @@ int x86emul_decode(struct x86_emulate_st
#endif
}
+ /* Latch XCR0, if available. */
+ if ( ops->read_cr && ops->read_xcr )
+ {
+ unsigned long cr4;
+
+ rc = ops->read_cr(4, &cr4, ctxt);
+ if ( rc == X86EMUL_OKAY && (cr4 & X86_CR4_OSXSAVE) &&
+ ops->read_xcr(0, &s->xcr0, ctxt) != X86EMUL_OKAY )
+ s->xcr0 = 0;
+
+ /*
+ * To ease consuming, strip 64-bit-only state right away for non-64-bit
+ * environments.
+ */
+ if ( !mode_64bit() )
+ s->xcr0 &= ~(X86_XCR0_TILE_CFG | X86_XCR0_TILE_DATA);
+ }
+
/* Prefix bytes. */
for ( ; ; )
{
--- a/xen/arch/x86/x86_emulate/private.h
+++ b/xen/arch/x86/x86_emulate/private.h
@@ -334,6 +334,8 @@ struct x86_emulate_state {
unsigned long ip;
+ uint64_t xcr0;
+
struct stub_exn *stub_exn;
#ifndef NDEBUG
--- a/xen/arch/x86/x86_emulate/x86_emulate.c
+++ b/xen/arch/x86/x86_emulate/x86_emulate.c
@@ -1234,17 +1234,17 @@ static bool is_branch_step(struct x86_em
return debugctl & IA32_DEBUGCTLMSR_BTF;
}
-static void adjust_bnd(struct x86_emulate_ctxt *ctxt,
+static void adjust_bnd(const struct x86_emulate_state *s,
+ struct x86_emulate_ctxt *ctxt,
const struct x86_emulate_ops *ops, enum vex_pfx pfx)
{
- uint64_t xcr0, bndcfg;
+ uint64_t bndcfg;
int rc;
if ( pfx == vex_f2 || !cpu_has_mpx || !vcpu_has_mpx() )
return;
- if ( !ops->read_xcr || ops->read_xcr(0, &xcr0, ctxt) != X86EMUL_OKAY ||
- !(xcr0 & X86_XCR0_BNDREGS) || !(xcr0 & X86_XCR0_BNDCSR) )
+ if ( !(s->xcr0 & X86_XCR0_BNDREGS) || !(s->xcr0 & X86_XCR0_BNDCSR) )
{
ASSERT(!ctxt->event_pending);
return;
@@ -1952,7 +1952,7 @@ x86_emulate(
case 0x70 ... 0x7f: /* jcc (short) */
if ( test_cc(b, _regs.eflags) )
jmp_rel((int32_t)src.val);
- adjust_bnd(ctxt, ops, vex.pfx);
+ adjust_bnd(state, ctxt, ops, vex.pfx);
break;
case 0x80: case 0x81: case 0x82: case 0x83: /* Grp1 */
@@ -2363,7 +2363,7 @@ x86_emulate(
(rc = ops->insn_fetch(dst.val, NULL, 0, ctxt)) )
goto done;
_regs.r(ip) = dst.val;
- adjust_bnd(ctxt, ops, vex.pfx);
+ adjust_bnd(state, ctxt, ops, vex.pfx);
break;
case 0xc4: /* les */
@@ -2594,7 +2594,7 @@ x86_emulate(
op_bytes = ((op_bytes == 4) && mode_64bit()) ? 8 : op_bytes;
src.val = _regs.r(ip);
jmp_rel(rel);
- adjust_bnd(ctxt, ops, vex.pfx);
+ adjust_bnd(state, ctxt, ops, vex.pfx);
goto push;
}
@@ -2602,7 +2602,7 @@ x86_emulate(
case 0xeb: /* jmp (short) */
jmp_rel((int32_t)src.val);
if ( !(b & 2) )
- adjust_bnd(ctxt, ops, vex.pfx);
+ adjust_bnd(state, ctxt, ops, vex.pfx);
break;
case 0xea: /* jmp (far, absolute) */
@@ -2885,14 +2885,14 @@ x86_emulate(
goto done;
_regs.r(ip) = src.val;
src.val = dst.val;
- adjust_bnd(ctxt, ops, vex.pfx);
+ adjust_bnd(state, ctxt, ops, vex.pfx);
goto push;
case 4: /* jmp (near) */
if ( (rc = ops->insn_fetch(src.val, NULL, 0, ctxt)) )
goto done;
_regs.r(ip) = src.val;
dst.type = OP_NONE;
- adjust_bnd(ctxt, ops, vex.pfx);
+ adjust_bnd(state, ctxt, ops, vex.pfx);
break;
case 3: /* call (far, absolute indirect) */
case 5: /* jmp (far, absolute indirect) */
@@ -4981,7 +4981,7 @@ x86_emulate(
case X86EMUL_OPC(0x0f, 0x80) ... X86EMUL_OPC(0x0f, 0x8f): /* jcc (near) */
if ( test_cc(b, _regs.eflags) )
jmp_rel((int32_t)src.val);
- adjust_bnd(ctxt, ops, vex.pfx);
+ adjust_bnd(state, ctxt, ops, vex.pfx);
break;
case X86EMUL_OPC(0x0f, 0x90) ... X86EMUL_OPC(0x0f, 0x9f): /* setcc */
^ permalink raw reply [flat|nested] 3+ messages in thread
* [PATCH 2/2] x86emul: use latched XCR0 in x86emul_get_fpu()
2026-08-20 8:50 [PATCH 0/2] x86emul: XCR0 handling Jan Beulich
2026-08-20 8:52 ` [PATCH 1/2] x86emul: latch XCR0 early during decode Jan Beulich
@ 2026-08-20 8:52 ` Jan Beulich
1 sibling, 0 replies; 3+ messages in thread
From: Jan Beulich @ 2026-08-20 8:52 UTC (permalink / raw)
To: xen-devel@lists.xenproject.org
Cc: Andrew Cooper, Teddy Astie, Roger Pau Monné, Andrew Mbugua
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 <jbeulich@suse.com>
---
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);
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2026-08-20 8:52 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-08-20 8:50 [PATCH 0/2] x86emul: XCR0 handling Jan Beulich
2026-08-20 8:52 ` [PATCH 1/2] x86emul: latch XCR0 early during decode Jan Beulich
2026-08-20 8:52 ` [PATCH 2/2] x86emul: use latched XCR0 in x86emul_get_fpu() Jan Beulich
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.