[PATCH] VAX: PIC: Use GOT for pointer for global function
Rin Okuyama
rokuyama.rk@gmail.com
Thu Nov 19 06:21:22 GMT 2026
Hi,
On 2026/11/17 3:52, Jan-Benedict Glaw wrote:
> Hi,
>
> On Sun, 2026-11-15 14:18:58 +0100, Jan-Benedict Glaw <jbglaw@lug-owl.de> wrote:
>> diff --git a/gas/config/tc-vax.c b/gas/config/tc-vax.c
>> index 3d21796..614cc8d 100644
>> --- a/gas/config/tc-vax.c
>> +++ b/gas/config/tc-vax.c
>> @@ -399,7 +399,13 @@ md_estimate_size_before_relax (fragS *fragP, segT segment)
>> || S_IS_EXTERNAL (fragP->fr_symbol)))
>> {
>> /* Indirect references cannot go through the GOT or PLT,
>> - let's hope they'll become local in the final link. */
>> + let's hope they'll become local in the final link.
>> + Only calls and jumps may go through the PLT. Any other
>> + reference wants the address of the symbol and must use the
>> + GOT, also for a function: a PLT reference is turned into an
>> + indirect one through the function's .got.plt slot, which
>> + holds the PLT entry until the lazy binding has happened, so
>> + the same function would have different addresses. */
>> if ((ELF_ST_VISIBILITY (S_GET_OTHER (fragP->fr_symbol))
>> != STV_DEFAULT)
>> || (p[0] & 0x10))
>> @@ -407,8 +413,7 @@ md_estimate_size_before_relax (fragS *fragP, segT segment)
>> else if (((unsigned char *) fragP->fr_opcode)[0] == VAX_CALLS
>> || ((unsigned char *) fragP->fr_opcode)[0] == VAX_CALLG
>> || ((unsigned char *) fragP->fr_opcode)[0] == VAX_JSB
>> - || ((unsigned char *) fragP->fr_opcode)[0] == VAX_JMP
>> - || S_IS_FUNCTION (fragP->fr_symbol))
>> + || ((unsigned char *) fragP->fr_opcode)[0] == VAX_JMP)
>> reloc_type = BFD_RELOC_32_PLT_PCREL;
>> else
>> reloc_type = BFD_RELOC_32_GOT_PCREL;
>> diff --git a/gas/testsuite/gas/vax/elf-got-func.d b/gas/testsuite/gas/vax/elf-got-func.d
>> new file mode 100644
>> index 0000000..f3d66d5
>> --- /dev/null
>> +++ b/gas/testsuite/gas/vax/elf-got-func.d
>> @@ -0,0 +1,14 @@
>> +#as: -k
>> +#objdump: -r
>> +#name: VAX ELF PIC address of a function uses the GOT
>> +
>> +.*: file format elf32-vax
>> +
>> +RELOCATION RECORDS FOR \[\.text\]:
>> +OFFSET +TYPE +VALUE
>> +00000004 R_VAX_GOT32 fn
>> +0000000b R_VAX_GOT32 fn
>> +00000012 R_VAX_PLT32 fn
>> +00000019 R_VAX_PLT32 fn
>> +0000001f R_VAX_PLT32 fn
>> +00000025 R_VAX_PLT32 fn
>> diff --git a/gas/testsuite/gas/vax/elf-got-func.s b/gas/testsuite/gas/vax/elf-got-func.s
>> new file mode 100644
>> index 0000000..c3fc335
>> --- /dev/null
>> +++ b/gas/testsuite/gas/vax/elf-got-func.s
>> @@ -0,0 +1,16 @@
>> +# Taking the address of a global function in PIC code must go through
>> +# the GOT even if the function is defined in this file; only calls and
>> +# jumps may use the PLT.
>> + .text
>> + .globl fn
>> + .type fn, @function
>> +fn:
>> + .word 0
>> + movab fn, %r0
>> + pushab fn
>> + calls $0, fn
>> + callg (%ap), fn
>> + jsb fn
>> + jmp fn
>> + ret
>> + .size fn, . - fn
>> diff --git a/gas/testsuite/gas/vax/vax.exp b/gas/testsuite/gas/vax/vax.exp
>> index 0c27be0..b521066 100644
>> --- a/gas/testsuite/gas/vax/vax.exp
>> +++ b/gas/testsuite/gas/vax/vax.exp
>> @@ -53,4 +53,5 @@ run_dump_test broken_word
>>
>> if { [istarget vax-*-*elf*] || [istarget vax-*-linux-*] } then {
>> run_dump_test elf-rel
>> + run_dump_test elf-got-func
>> }
>>
>>
>
> Committed unchanged with this commit message:
>
> VAX: In PIC code, use GOT for function pointers
>
> Use GOT references for function pointers in PIC code. Actual fix by
> Rin Okuyama, initially reported for NetBSD in https://gnats.netbsd.org/59326 .
>
> PR gas/32941
>
> gas/
> * config/tc-vax.c (md_estimate_size_before_relax): Use GOT instead of PLT.
> * testsuite/gas/vax/elf-got-func.s,
> * testsuite/gas/vax/elf-got-func.d: New test.
> * testsuite/gas/vax/vax.exp: Run it.
>
>
> MfG, JBG
>
Thank you very much for commit, as well as adding comments and
implementing the test case.
I will replace my local commit to NetBSD with yours soon.
BTW, I'm sorry Maciej I couldn't reply to your kind explanation.
Even today, I don't have enough bandwidth to fix VAX linker as
you proposed. Sorry, but let me take that into consideration for
the future.
Thanks,
rin
More information about the Binutils
mailing list