[PATCH 1/3] x86-64: improve handling of branches to absolute addresses
H.J. Lu
hjl.tools@gmail.com
Thu Jun 30 17:50:32 GMT 2022
On Thu, Jun 30, 2022 at 5:08 AM Jan Beulich <jbeulich@suse.com> wrote:
>
> There are two related problems here: The use of "addr32" on a direct
> branch would, besides causing a warning, result in operands to be
> permitted which mistakenly are refused without "addr32". Plus at some
> point not too long ago I'm afraid it may have been me who regressed the
> relocation addends emitted for such branches. Correct both problems,
> adding a testcase to guard against regressing this again.
> ---
> In principle things like "JECXZ <absolute>" should be permitted as well,
> I think. Whether the destination is within reach can only be determined
> by the linker. But that likely being a more intrusive change, I guess we
> can leave this as is until someone really needs it to work.
>
> If "ELF: emit symbol table when there are relocations" (submitted
> earlier) goes in before this change, I'd be inclined to drop the label
> again from the new testcase. The original lack of a label there was how
> I noticed that other issue, so the testcase here could at once serve to
> test that changed behavior as well.
>
> --- a/gas/config/tc-i386.c
> +++ b/gas/config/tc-i386.c
> @@ -4975,7 +4975,9 @@ md_assemble (char *line)
> if (i.imm_operands)
> optimize_imm ();
>
> - if (i.disp_operands && !want_disp32 (current_templates->start))
> + if (i.disp_operands && !want_disp32 (current_templates->start)
> + && (!current_templates->start->opcode_modifier.jump
> + || i.jumpabsolute || i.types[0].bitfield.baseindex))
> {
> for (j = 0; j < i.operands; ++j)
> {
> @@ -5985,7 +5987,9 @@ optimize_disp (void)
> /* Optimize 64-bit displacement to 32-bit for 64-bit BFD. */
> if ((i.types[op].bitfield.disp32
> || (flag_code == CODE_64BIT
> - && want_disp32 (current_templates->start)))
> + && want_disp32 (current_templates->start)
> + && (!current_templates->start->opcode_modifier.jump
> + || i.jumpabsolute || i.types[op].bitfield.baseindex)))
> && fits_in_unsigned_long (op_disp))
> {
> /* If this operand is at most 32 bits, convert
> --- a/gas/testsuite/gas/i386/i386.exp
> +++ b/gas/testsuite/gas/i386/i386.exp
> @@ -1314,6 +1314,7 @@ if [gas_64_check] then {
> run_dump_test "x86-64-branch-3"
> run_list_test "x86-64-branch-4" "-al -mintel64"
> run_list_test "x86-64-branch-5" "-al"
> + run_dump_test "x86-64-branch-6"
>
> run_dump_test "x86-64-rip-2"
>
> --- /dev/null
> +++ b/gas/testsuite/gas/i386/x86-64-branch-6.d
> @@ -0,0 +1,21 @@
> +#objdump: -r
> +#name: x86-64 branch 6
> +#warning_output: x86-64-branch-6.e
> +
> +.*: +file format .*
> +
> +RELOCATION RECORDS FOR \[\.text\]:
> +OFFSET +TYPE +VALUE *
> +0+01 R_X86_64_PC32 +\*ABS\*\+0x000000008765431d
> +0+11 R_X86_64_PC32 +\*ABS\*\+0x000000087654320c
> +0+21 R_X86_64_PC32 +\*ABS\*\+0x000000008765431d
> +0+31 R_X86_64_PC32 +\*ABS\*\+0x000000087654320c
> +0+07 R_X86_64_PC32 +\*ABS\*\+0x000000008765431d
> +0+0c R_X86_64_PC32 +\*ABS\*\+0x000000008765431d
> +0+17 R_X86_64_PC32 +\*ABS\*\+0x000000087654320c
> +0+1c R_X86_64_PC32 +\*ABS\*\+0x000000087654320c
> +0+27 R_X86_64_PC32 +\*ABS\*\+0x000000008765431d
> +0+2c R_X86_64_PC32 +\*ABS\*\+0x000000008765431d
> +0+37 R_X86_64_PC32 +\*ABS\*\+0x000000087654320c
> +0+3c R_X86_64_PC32 +\*ABS\*\+0x000000087654320c
> +#pass
> --- /dev/null
> +++ b/gas/testsuite/gas/i386/x86-64-branch-6.s
> @@ -0,0 +1,18 @@
> + .text
> +
> +branch_6:
> + call 0x87654321
> + je 0x87654321
> + jmp 0x87654321
> +
> + call 0x876543210
> + je 0x876543210
> + jmp 0x876543210
> +
> + addr32 call 0x87654321
> + addr32 je 0x87654321
> + addr32 jmp 0x87654321
> +
> + addr32 call 0x876543210
> + addr32 je 0x876543210
> + addr32 jmp 0x876543210
> --- /dev/null
> +++ b/gas/testsuite/gas/i386/x86-64-branch-6.e
> @@ -0,0 +1,7 @@
> +.*: Assembler messages:
> +.*:12: Warning: skipping prefixes on `call'
> +.*:13: Warning: skipping prefixes on `je'
> +.*:14: Warning: skipping prefixes on `jmp'
> +.*:16: Warning: skipping prefixes on `call'
> +.*:17: Warning: skipping prefixes on `je'
> +.*:18: Warning: skipping prefixes on `jmp'
>
OK.
Thanks.
--
H.J.
More information about the Binutils
mailing list