Skip to contentSkip to conversation

x86: use / "support" UDB

PATCH In review Jan Beulich 3 participants 4 messages Apr 8 – Jun 3

lore
Share & tools
lore
#0 lore raw

PATCH v10 1/1 x86: use / "support" UDB

With opcode D6h now firmly reserved as another #UD-raising one in 64-bit mode, use that instead of the two-byte UD2 for bug frame marking.

While there also make a minor adjustment to the emulator.

---
One of the table entries in stub_selftest() uses UD1, yet in not quite an appropriate way: The 0x90 following it (presumably meant to be a NOP) really is a ModR/M byte, requiring a displacement to follow. Wouldn't we better adjust that (e.g. using 0xcc instead)?
---
v10: Shrink the actual emulator change to just a comment adjustment.
Extend .byte directive commentary.
v9: New.

xen/arch/x86/include/asm/bug.h +2 -2
@@ -21,7 +21,7 @@
 
 #ifndef __ASSEMBLER__
 
-#define BUG_INSTR       "ud2"
+#define BUG_INSTR       ".byte 0xd6" /* UDB, requiring gas 2.46 */
 #define BUG_ASM_CONST   "c"
 
 #else  /* !__ASSEMBLER__ */
@@ -37,7 +37,7 @@
         .error "Invalid BUGFRAME index"
     .endif
 
-    .L\@ud: ud2a
+    .L\@ud: .byte 0xd6 /* UDB, requiring gas 2.46 */
 
     .pushsection .rodata.str1, "aMS", @progbits, 1
          .L\@s1: .asciz "\file_str"
xen/arch/x86/traps.c +3 -3
@@ -1338,7 +1338,7 @@ void asmlinkage do_trap(struct cpu_user_
 
 void asmlinkage do_invalid_op(struct cpu_user_regs *regs)
 {
-    u8 bug_insn[2];
+    uint8_t bug_insn;
     const void *eip = (const void *)regs->rip;
     int id;
 
@@ -1350,8 +1350,8 @@ void asmlinkage do_invalid_op(struct cpu
     }
 
     if ( !is_active_kernel_text(regs->rip) ||
-         copy_from_unsafe(bug_insn, eip, sizeof(bug_insn)) ||
-         memcmp(bug_insn, "\xf\xb", sizeof(bug_insn)) )
+         copy_from_unsafe(&bug_insn, eip, sizeof(bug_insn)) ||
+         bug_insn != 0xd6 /* UDB */ )
         goto die;
 
     id = do_bug_frame(regs, regs->rip);
xen/arch/x86/x86_emulate/decode.c +1 -1
@@ -651,7 +651,7 @@ decode_onebyte(struct x86_emulate_state
     case 0xce: /* into */
     case 0xd4: /* aam */
     case 0xd5: /* aad */
-    case 0xd6: /* salc */
+    case 0xd6: /* salc / udb */
         s->not_64bit = true;
         break;
 
#1 lore raw
Quoted 6 lines · Jan Beulich

With opcode D6h now firmly reserved as another #UD-raising one in 64-bit mode, use that instead of the two-byte UD2 for bug frame marking.

While there also make a minor adjustment to the emulator.

For the patch itself, Acked-by: Andrew Cooper <[email protected]>

Quoted 5 lines

---
One of the table entries in stub_selftest() uses UD1, yet in not quite an appropriate way: The 0x90 following it (presumably meant to be a NOP) really is a ModR/M byte, requiring a displacement to follow. Wouldn't we better adjust that (e.g. using 0xcc instead)?

That looks too much like breakpoint padding as opposed to nop padding.

What about:

xen/arch/x86/extable.c +0 -0
index a9b6c6b904f5..24440ccd03e3 100644
@@ -157,7 +157,7 @@ int __init cf_check stub_selftest(void)

         union stub_exception_token res;
     } tests[] __initconst = {
 #define endbr64 0xf3, 0x0f, 0x1e, 0xfa
-        { .opc = { endbr64, 0x0f, 0xb9, 0x90 }, /* ud1 */
+        { .opc = { endbr64, 0x0f, 0xb9, 0x00 }, /* ud1 (%rax),%eax */ .res.fields.trapnr = X86_EXC_UD },
         { .opc = { endbr64, 0x90, 0x02, 0x00 }, /* nop; add (%rax),%al */ .rax = 0x0123456789abcdef,

which also brings it in line with the adjacent example?

~Andrew

#2 lore raw
Quoted 9 lines · Andrew Cooper
Quoted 6 lines · Jan Beulich

With opcode D6h now firmly reserved as another #UD-raising one in 64-bit mode, use that instead of the two-byte UD2 for bug frame marking.

While there also make a minor adjustment to the emulator.

For the patch itself, Acked-by: Andrew Cooper <[email protected]>

Thanks.

Quoted 24 lines
Quoted 5 lines

---
One of the table entries in stub_selftest() uses UD1, yet in not quite an appropriate way: The 0x90 following it (presumably meant to be a NOP) really is a ModR/M byte, requiring a displacement to follow. Wouldn't we better adjust that (e.g. using 0xcc instead)?

That looks too much like breakpoint padding as opposed to nop padding.

What about:

xen/arch/x86/extable.c +1 -1
index a9b6c6b904f5..24440ccd03e3 100644
@@ -157,7 +157,7 @@ int __init cf_check stub_selftest(void)
          union stub_exception_token res;
      } tests[] __initconst = {
  #define endbr64 0xf3, 0x0f, 0x1e, 0xfa
-        { .opc = { endbr64, 0x0f, 0xb9, 0x90 }, /* ud1 */
+        { .opc = { endbr64, 0x0f, 0xb9, 0x00 }, /* ud1 (%rax),%eax */
            .res.fields.trapnr = X86_EXC_UD },

         { .opc = { endbr64, 0x90, 0x02, 0x00 }, /* nop; add (%rax),%al */ .rax = 0x0123456789abcdef,

Quoted 1 line

which also brings it in line with the adjacent example?

That's also okay, sure.

Jan

#3 lore raw
In reply to Jan Beulich
Quoted 6 lines · Jan Beulich

With opcode D6h now firmly reserved as another #UD-raising one in 64-bit mode, use that instead of the two-byte UD2 for bug frame marking.

While there also make a minor adjustment to the emulator.

Just a heads up. I was testing Xen with Fedora 44 QEMU 10.2.2 TCG, and a BUG() did not fire as expected. Single stepping over UDB I did not see any registers change and execution continued.

QEMU 11.0 does emulate UDB properly - the change was:

commit a71706fc116d263c6661b5c7176150dfe0298d29

Author: Paolo Bonzini <[email protected]>
Date:   Wed Dec 10 09:42:01 2025 +0100

target/i386/tcg: mark more instructions that are invalid in 64-bit mode

I didn't check earlier version of QEMU.

Regards,
Jason