Re: [PATCH: PR gas/11323] Add R_ARM_GOT_PREL relocation to arm assembler

[Carrot Wei <carrot at google dot com>]

ping

On Mon, Mar 15, 2010 at 10:14 PM, Carrot Wei <carrot@google.com> wrote:
> Now this new patch can process the following relocation expression
>
> ? ? ?.word ? i(GOT_PREL) + (. - (.LPIC0+4))
>
>
> bfd/
> 2010-03-15 ?Wei Guozhi ?<carrot@google.com>
>
> ? ? ? PR gas/11323
> ? ? ? * bfd-in2.h (enum bfd_reloc_code_real): New BFD_RELOC_GOT_PREL type.
> ? ? ? * elf32-arm.c (elf32_arm_reloc_map): BFD_RELOC_GOT_PREL to
> ? ? ? R_ARM_GOT_PREL map.
> ? ? ? * libbfd.h (bfd_reloc_code_real_names): BFD_RELOC_GOT_PREL name.
> ? ? ? * reloc.c (comments): Document the new relocation.
>
> gas/
> 2010-03-15 ?Wei Guozhi ?<carrot@google.com>
>
> ? ? ? PR gas/11323
> ? ? ? * config/tc-arm.c (reloc_names): New relocation names.
> ? ? ? (md_apply_fix): New case for BFD_RELOC_ARM_GOT_PREL.
> ? ? ? (tc_gen_reloc): New case for BFD_RELOC_ARM_GOT_PREL.
> ? ? ? * doc/c-arm.texi (ARM-Relocations): Document the new relocation.
>
> gas/testsuite
> 2010-03-15 ?Wei Guozhi ?<carrot@google.com>
>
> ? ? ? PR gas/11323
> ? ? ? * gas/arm/got_prel.s: New test case.
> ? ? ? * gas/arm/got_prel.d: Likewise.
>
>
> Index: bfd-in2.h
> ===================================================================
> RCS file: /cvs/src/src/bfd/bfd-in2.h,v
> retrieving revision 1.507
> diff -u -r1.507 bfd-in2.h
> --- bfd-in2.h ? 8 Feb 2010 20:28:43 -0000 ? ? ? 1.507
> +++ bfd-in2.h ? 15 Mar 2010 14:03:08 -0000
> @@ -3114,6 +3114,7 @@
> ? BFD_RELOC_ARM_RELATIVE,
> ? BFD_RELOC_ARM_GOTOFF,
> ? BFD_RELOC_ARM_GOTPC,
> + ?BFD_RELOC_ARM_GOT_PREL,
>
> ?/* ARM thread-local storage relocations. ?*/
> ? BFD_RELOC_ARM_TLS_GD32,
>
> Index: elf32-arm.c
> ===================================================================
> RCS file: /cvs/src/src/bfd/elf32-arm.c,v
> retrieving revision 1.227
> diff -u -r1.227 elf32-arm.c
> --- elf32-arm.c 4 Mar 2010 17:16:08 -0000 ? ? ? 1.227
> +++ elf32-arm.c 15 Mar 2010 14:03:10 -0000
> @@ -1721,6 +1721,7 @@
> ? ? {BFD_RELOC_ARM_RELATIVE, ? ? ? ? R_ARM_RELATIVE},
> ? ? {BFD_RELOC_ARM_GOTOFF, ? ? ? ? ? R_ARM_GOTOFF32},
> ? ? {BFD_RELOC_ARM_GOTPC, ? ? ? ? ? ?R_ARM_GOTPC},
> + ? ?{BFD_RELOC_ARM_GOT_PREL, ? ? ? ? R_ARM_GOT_PREL},
> ? ? {BFD_RELOC_ARM_GOT32, ? ? ? ? ? ?R_ARM_GOT32},
> ? ? {BFD_RELOC_ARM_PLT32, ? ? ? ? ? ?R_ARM_PLT32},
> ? ? {BFD_RELOC_ARM_TARGET1, ? ? ? ? R_ARM_TARGET1},
>
> Index: libbfd.h
> ===================================================================
> RCS file: /cvs/src/src/bfd/libbfd.h,v
> retrieving revision 1.227
> diff -u -r1.227 libbfd.h
> --- libbfd.h ? ?8 Feb 2010 20:28:43 -0000 ? ? ? 1.227
> +++ libbfd.h ? ?15 Mar 2010 14:03:10 -0000
> @@ -1330,6 +1330,7 @@
> ? "BFD_RELOC_ARM_RELATIVE",
> ? "BFD_RELOC_ARM_GOTOFF",
> ? "BFD_RELOC_ARM_GOTPC",
> + ?"BFD_RELOC_ARM_GOT_PREL",
> ? "BFD_RELOC_ARM_TLS_GD32",
> ? "BFD_RELOC_ARM_TLS_LDO32",
> ? "BFD_RELOC_ARM_TLS_LDM32",
>
> Index: reloc.c
> ===================================================================
> RCS file: /cvs/src/src/bfd/reloc.c,v
> retrieving revision 1.198
> diff -u -r1.198 reloc.c
> --- reloc.c ? ? 21 Feb 2010 22:24:03 -0000 ? ? ?1.198
> +++ reloc.c ? ? 15 Mar 2010 14:03:11 -0000
> @@ -2940,6 +2940,8 @@
> ? BFD_RELOC_ARM_GOTOFF
> ?ENUMX
> ? BFD_RELOC_ARM_GOTPC
> +ENUMX
> + ?BFD_RELOC_ARM_GOT_PREL
> ?ENUMDOC
> ? Relocations for setting up GOTs and PLTs for shared libraries.
>
>
> Index: config/tc-arm.c
> ===================================================================
> RCS file: /cvs/src/src/gas/config/tc-arm.c,v
> retrieving revision 1.437
> diff -u -r1.437 tc-arm.c
> --- config/tc-arm.c ? ? 5 Mar 2010 10:41:04 -0000 ? ? ? 1.437
> +++ config/tc-arm.c ? ? 15 Mar 2010 14:09:16 -0000
> @@ -16191,7 +16191,8 @@
> ? { "tlsldm", ?BFD_RELOC_ARM_TLS_LDM32}, { "TLSLDM", ?BFD_RELOC_ARM_TLS_LDM32},
> ? { "tlsldo", ?BFD_RELOC_ARM_TLS_LDO32}, { "TLSLDO", ?BFD_RELOC_ARM_TLS_LDO32},
> ? { "gottpoff",BFD_RELOC_ARM_TLS_IE32}, ?{ "GOTTPOFF",BFD_RELOC_ARM_TLS_IE32},
> - ?{ "tpoff", ? BFD_RELOC_ARM_TLS_LE32}, ?{ "TPOFF", ? BFD_RELOC_ARM_TLS_LE32}
> + ?{ "tpoff", ? BFD_RELOC_ARM_TLS_LE32}, ?{ "TPOFF", ? BFD_RELOC_ARM_TLS_LE32},
> + ?{ "got_prel", ?BFD_RELOC_ARM_GOT_PREL},{ "GOT_PREL", ?BFD_RELOC_ARM_GOT_PREL}
> ?};
> ?#endif
>
> @@ -20477,7 +20478,12 @@
> ? ? ? if (fixP->fx_done || !seg->use_rela_p)
> ? ? ? ?md_number_to_chars (buf, 0, 4);
> ? ? ? break;
> -
> +
> + ? ?case BFD_RELOC_ARM_GOT_PREL:
> + ? ? ?if (fixP->fx_done || !seg->use_rela_p)
> + ? ? ? ?md_number_to_chars (buf, value, 4);
> + ? ? ?break;
> +
> ? ? case BFD_RELOC_ARM_TARGET2:
> ? ? ? /* TARGET2 is not partial-inplace, so we need to write the
> ? ? ? ? ?addend here for REL targets, because it won't be written out
> @@ -21062,6 +21068,7 @@
> ?#ifdef OBJ_ELF
> ? ? case BFD_RELOC_ARM_GOT32:
> ? ? case BFD_RELOC_ARM_GOTOFF:
> + ? ?case BFD_RELOC_ARM_GOT_PREL:
> ? ? case BFD_RELOC_ARM_PLT32:
> ? ? case BFD_RELOC_ARM_TARGET1:
> ? ? case BFD_RELOC_ARM_ROSEGREL32:
>
> Index: doc/c-arm.texi
> ===================================================================
> RCS file: /cvs/src/src/gas/doc/c-arm.texi,v
> retrieving revision 1.68
> diff -u -r1.68 c-arm.texi
> --- doc/c-arm.texi ? ? ?18 Feb 2010 10:56:27 -0000 ? ? ?1.68
> +++ doc/c-arm.texi ? ? ?15 Mar 2010 14:09:16 -0000
> @@ -452,7 +452,8 @@
> ?@code{TLSGD},
> ?@code{TLSLDM},
> ?@code{TLSLDO},
> -@code{GOTTPOFF}
> +@code{GOTTPOFF},
> +@code{GOT_PREL}
> ?and
> ?@code{TPOFF}.
>
> gas/testsuite/gas/arm/got_prel.s
> =============================================================
> ? ? ? ?.code ? 16
> ? ? ? ?.text
> .Ltext0:
> ? ? ? ?.section ? ? ? ?.text.foo,"ax",%progbits
> ? ? ? ?.align ?2
> ? ? ? ?.global foo
> ? ? ? ?.code ? 16
> ? ? ? ?.thumb_func
> ? ? ? ?.type ? foo, %function
> foo:
> ? ? ? ?.fnstart
> ? ? ? ?ldr ? ? r3, .L3
> .LPIC0:
> ? ? ? ?add ? ? r3, pc
> ? ? ? ?ldr ? ? r3, [r3]
> ? ? ? ?ldr ? ? r2, [r3]
> ? ? ? ?str ? ? r0, [r3]
> ? ? ? ?mov ? ? r0, r2
> ? ? ? ?bx ? ? ?lr
> ? ? ? ?.align ?2
> .L3:
> ? ? ? ?.word ? i(GOT_PREL) + (. - (.LPIC0+4))
> ? ? ? ?.fnend
>
> gas/testsuite/gas/arm/got_prel.d
> =============================================================
> # name: R_ARM_GOT_PREL relocation
> # source: got_prel.s
> # as: -march=armv5te -meabi=5
> # readelf: -x 4 -r
>
> Relocation section '.rel.text.foo' at offset 0x3f0 contains 1 entries:
> ?Offset ? ? Info ? ?Type ? ? ? ? ? ?Sym.Value ?Sym. Name
> 00000010 ?00000c60 R_ARM_GOT_PREL ? ?00000000 ? i
>
> Relocation section '.rel.ARM.exidx.text.foo' at offset 0x3f8 contains 2 entries:
> ?Offset ? ? Info ? ?Type ? ? ? ? ? ?Sym.Value ?Sym. Name
> 00000000 ?0000042a R_ARM_PREL31 ? ? ?00000000 ? .text.foo
> 00000000 ?00000d00 R_ARM_NONE ? ? ? ?00000000 ? __aeabi_unwind_cpp_pr0
>
> Hex dump of section '.text.foo':
> ?NOTE: This section has relocations against it, but these have NOT
> been applied to this dump.
> ?0x00000000 034b7b44 1b681a68 1860101c 7047c046 .K{D.h.h.`..pG.F
> ?0x00000010 0a000000 ? ? ? ? ? ? ? ? ? ? ? ? ? ?....
>
>
> On Mon, Mar 15, 2010 at 4:51 PM, Richard Earnshaw <rearnsha@arm.com> wrote:
>>
>> On Mon, 2010-03-15 at 16:37 +0800, Carrot Wei wrote:
>>> Hi
>>>
>>> ? As described in http://sourceware.org/bugzilla/show_bug.cgi?id=11323, this
>>> patch processes following expression
>>>
>>> ? ? ? .word ? var(GOTABS)-(.LPIC0+4)
>>>
>>> And produces a relocation of type R_ARM_GOT_PREL.
>>>
>>> This patch has been tested by "make check-gas".
>>>
>>
>> I understand why you've done it this way, but I don't like the fact that
>> the (GOTABS) annotation leads to a GOT_PREL relocation. ?I think the
>> correct way of expressing this is to have a GOT_PREL annotation itself,
>> and then to factor compensation for the addend into the remainder of the
>> expression:
>>
>> So, mathematically
>> ? ? ? ?var(GOT_PREL) = &GOT_SLOT(var) - .
>>
>> and the expression that should then be used in the assembly should
>> compensate for this:
>>
>> ? ? ? ?.word var(GOT_PREL) - (.LPIC0+4 - .)
>>
>> The advantage of this way of expressing things is that we've now got two
>> parts to the expression: a var (with relocation type) and and addend
>> (that's constant). ?Further, if the addend is zero, then the expression
>> collapses naturally to '.word var(GOT_PREL)' and we still get exactly
>> the right output.
>>
>