[PATCH] BFD linker support for x86 nacl targets
H.J. Lu
hjl.tools@gmail.com
Sat Mar 31 14:52:00 GMT 2012
On Fri, Mar 30, 2012 at 5:17 PM, Roland McGrath <mcgrathr@google.com> wrote:
> This pretty much completes the support for i?86-nacl and x86_64-nacl targets.
>
> This touches generic code (excluding test suites) only in making elf.sc
> grok the new SEPARATE_CODE parameter. I verified with an
> --enable-targets=all build that ldscripts/* all comes out identical
> before and after the change.
>
> This touches the test suites quite a lot, but in ways that are simple
> conceptually. Basically, I enabled for *-nacl all the test cases that
> are enabled for *-linux*, and then tweaked things to make them pass for
> the nacl targets' output. The targets are similar enough that nearly
> all the same features can be tested, yet subtly different enough that
> it's actually worthwhile to test the nacl variant backends too so as to
> ensure that some strange interaction doesn't cause them to regress vs
> the linux-gnu targets.
>
> I've regression-tested x86_64-linux-gnu native and i686-linux-gnu target.
>
> The simplest changes are just to change target constraints to match the
> nacl targets. The expected output often differs in irrelevant ways such
> as the exact addresses chosen, because the address space layout is quite
> different. So the pervasive changes that make the patch so large are to
> loosen the regexps so they match both targets' output details. I
> believe I've only loosened things that don't really matter to each
> individual test, and that they still thoroughly test the things they
> were intended to test.
>
> In numerous cases, the disassembly matching had been regexpified so it
>
> disassembled instructions, but it was still matching literal hex bytes
> in the middle column, so the looseness of the disassembly text regexp
> didn't actually make the test insensitive to those exact values as was
> clearly the intent. In other cases, there was exact matching on
> addresses that really didn't seem relevant to the test, so I regexpified
> those too.
>
> I avoided as much as possible adding separate nacl variants of the
> expected output, when the existing files could be regexpified enough to
> match both variants without becoming so loose that they stopped testing
> what they're there to test. The section/segment layout and the PLT
> generation are sufficiently different for nacl that those expectation
> files had to have nacl variants. I wanted basic tests for PLT
> generation for the nacl variant, and there weren't any existing ones, so
> I added those for vanilla i?86 and x86_64 as well as the nacl variants.
>
> Note this is a -b diff for readability; rest assured my actual changes
> have all the indentation correct.
>
> Ok for trunk?
>
>
> Thanks,
> Roland
>
>
> bfd/
> 2012-03-30 Roland McGrath <mcgrathr@google.com>
>
> * elf-nacl.c: New file.
> * elf-nacl.h: New file.
> * elf32-i386.c (elf_backend_modify_segment_map): Define for
> bfd_elf32_i386_nacl_vec.
> (elf_backend_modify_program_headers): Likewise.
> * elf64-x86-64.c (elf_backend_modify_segment_map): Define for
> bfd_elf64_x86_64_nacl_vec and bfd_elf32_x86_64_nacl_vec.
> (elf_backend_modify_program_headers): Likewise.
> * Makefile.am (BFD32_BACKENDS, BFD64_BACKENDS): Add elf-nacl.lo here.
> (BFD32_BACKENDS_CFILES, BFD64_BACKENDS_CFILES): Add elf-nacl.c here.
> * Makefile.in: Regenerated.
> * configure.in (bfd_elf64_x86_64_nacl_vec): Add elf-nacl.o to tb here.
> (bfd_elf32_x86_64_nacl_vec): Likewise.
> (bfd_elf64_x86_64_vec, bfd_elf32_x86_64_vec): Likewise.
> (bfd_elf64_x86_64_freebsd_vec, bfd_elf64_x86_64_sol2_vec): Likewise.
> (bfd_elf64_l1om_vec, bfd_elf64_l1om_freebsd_vec): Likewise.
> (bfd_elf64_k1om_vec, bfd_elf64_k1om_freebsd_vec): Likewise.
> (bfd_elf32_i386_nacl_vec): Likewise.
> (bfd_elf32_i386_sol2_vec, bfd_elf32_i386_freebsd_vec): Likewise.
> (bfd_elf32_i386_vxworks_vec, bfd_elf32_i386_vec): Likewise.
> * configure: Regenerated.
>
> binutils/testsuite/
> 2012-03-30 Roland McGrath <mcgrathr@google.com>
>
> * binutils-all/i386/i386.exp: Accept nacl targets too.
> * binutils-all/x86-64/x86-64.exp: Likewise.
>
> ld/
> 2012-03-30 Roland McGrath <mcgrathr@google.com>
>
> * configure.tgt (i[3-7]86-*-nacl*, x86_64-*-nacl*): Handle them.
> * emulparams/elf_nacl.sh: New file.
> * emulparams/elf_i386_nacl.sh: New file.
> * emulparams/elf32_x86_64_nacl.sh: New file.
> * emulparams/elf_x86_64_nacl.sh: New file.
> * Makefile.am (ALL_EMULATION_SOURCES): Add eelf_i386_nacl.c here.
> (ALL_64_EMULATION_SOURCES): Add eelf32_x86_64_nacl.c and
> eelf_x86_64_nacl.c here.
> (eelf_i386_nacl.c, eelf32_x86_64_nacl.c, eelf_x86_64_nacl.c):
> New targets.
> * Makefile.in: Regenerated.
>
> * scripttempl/elf.sc: Handle SEPARATE_CODE cases.
>
> ld/testsuite/
> 2012-03-30 Roland McGrath <mcgrathr@google.com>
>
> * ld-x86-64/ilp32-4-nacl.d: New file.
> * ld-x86-64/x86-64.exp: Run it.
>
> * ld-discard/discard.exp: Accept nacl targets too.
> * ld-elf/binutils.exp: Likewise.
> * ld-elf/comm-data.exp: Likewise.
> * ld-elf/elf.exp: Likewise.
> * ld-elf/tls_common.exp: Likewise.
> * ld-elfvers/vers.exp: Likewise.
> * ld-elfvsb/elfvsb.exp: Likewise.
> * ld-elfweak/elfweak.exp: Likewise.
> * ld-gc/gc.exp: Likewise.
> * ld-ifunc/binutils.exp: Likewise.
> * ld-ifunc/ifunc.exp: Likewise.
> * ld-linkonce/linkonce.exp:Likewise.
> * ld-pie/pie.exp: Likewise.
> * ld-shared/shared.exp: Likewise.
> * ld-undefined/weak-undef.exp: Likewise.
> * ld-unique/unique.exp: Likewise.
> * ld-x86-64/dwarfreloc.exp: Likewise.
> * ld-x86-64/line.exp: Likewise.
>
> * lib/ld-lib.exp (slurp_options): Support global array
> options_regsub to apply substitutions to the contents
> of options lines read from the file.
> * ld-i386/emit-relocs.d: Renamed to ...
> * ld-i386/emit-relocs.rd: ... this.
> * ld-i386/i386.exp: Accept nacl targets too.
> For them, use options_regsub to replace elf_i386 with
> elf_i386_nacl in run_dump_test cases; apply the same
> substitution in $i386tests; replace foo.rd expectations
> files with foo-nacl.rd in $i386tests.
> (i386tests): Change emit-relocs.d to emit-relocs.rd here.
> * ld-i386/emit-relocs-nacl.rd: New file.
> * ld-i386/plt-nacl.pd: New file.
> * ld-i386/plt-pic-nacl.pd: New file.
> * ld-i386/tlsbin-nacl.rd: New file.
> * ld-i386/tlsbindesc-nacl.rd: New file.
> * ld-i386/tlsdesc-nacl.rd: New file.
> * ld-i386/tlsgdesc-nacl.rd: New file.
> * ld-i386/tlsnopic-nacl.rd: New file.
> * ld-i386/tlspic-nacl.rd: New file.
> * ld-x86-64/x86-64.exp: Accept nacl targets too.
> For them, use options_regsub to replace elf_x86_64 with
> elf_x86_64_nacl in run_dump_test cases; apply the same
> substitution in $x86_64tests; replace foo.rd expectations
> files with foo-nacl.rd in $x86_64tests.
> Add explicit -melf_x86_64 to ld options in tests that need it,
> in case the default emulation is x32 (as it is for x86_64-nacl).
> * ld/testsuite/ld-x86-64/plt-nacl.pd: New file.
> * ld/testsuite/ld-x86-64/split-by-file-nacl.rd: New file.
> * ld/testsuite/ld-x86-64/tlsbin-nacl.rd: New file.
> * ld/testsuite/ld-x86-64/tlsbindesc-nacl.rd: New file.
> * ld/testsuite/ld-x86-64/tlsdesc-nacl.pd: New file.
> * ld/testsuite/ld-x86-64/tlsdesc-nacl.rd: New file.
> * ld/testsuite/ld-x86-64/tlsgdesc-nacl.rd: New file.
> * ld/testsuite/ld-x86-64/tlspic-nacl.rd: New file.
>
> * ld-i386/hidden2.d: Loosen regexps to match any file format variant,
> and not to depend on exact addresses, displacements, etc. where
> they are irrelevant.
> * ld-i386/pcrel16.d: Likewise.
> * ld-i386/pcrel16abs.d: Likewise.
> * ld-i386/pr12718.d: Likewise.
> * ld-i386/pr12921.d: Likewise.
> * ld-i386/reloc.d: Likewise.
> * ld-i386/tlsbin.dd: Likewise.
> * ld-i386/tlsbin.sd: Likewise.
> * ld-i386/tlsbin.td: Likewise.
> * ld-i386/tlsbindesc.dd: Likewise.
> * ld-i386/tlsbindesc.sd: Likewise.
> * ld-i386/tlsbindesc.td: Likewise.
> * ld-i386/tlsdesc.dd: Likewise.
> * ld-i386/tlsdesc.sd: Likewise.
> * ld-i386/tlsdesc.td: Likewise.
> * ld-i386/tlsg.sd: Likewise.
> * ld-i386/tlsgdesc.dd: Likewise.
> * ld-i386/tlsindntpoff.dd: Likewise.
> * ld-i386/tlsnopic.dd: Likewise.
> * ld-i386/tlsnopic.sd: Likewise.
> * ld-i386/tlspic.dd: Likewise.
> * ld-i386/tlspic.sd: Likewise.
> * ld-i386/tlspic.td: Likewise.
> * ld-i386/tlspie2.d: Likewise.
> * ld-x86-64/hidden2.d: Likewise.
> * ld-x86-64/pcrel16.d: Likewise.
> * ld-x86-64/pr12718.d: Likewise.
> * ld-x86-64/pr12921.d: Likewise.
> * ld-x86-64/protected3.d: Likewise.
> * ld-x86-64/tlsbin.dd: Likewise.
> * ld-x86-64/tlsbin.sd: Likewise.
> * ld-x86-64/tlsbin.td: Likewise.
> * ld-x86-64/tlsbindesc.dd: Likewise.
> * ld-x86-64/tlsbindesc.sd: Likewise.
> * ld-x86-64/tlsbindesc.td: Likewise.
> * ld-x86-64/tlsdesc.dd: Likewise.
> * ld-x86-64/tlsdesc.sd: Likewise.
> * ld-x86-64/tlsdesc.td: Likewise.
> * ld-x86-64/tlsg.sd: Likewise.
> * ld-x86-64/tlsgd5.dd: Likewise.
> * ld-x86-64/tlsgd6.dd: Likewise.
> * ld-x86-64/tlsgdesc.dd: Likewise.
> * ld-x86-64/tlspic.dd: Likewise.
> * ld-x86-64/tlspic.sd: Likewise.
> * ld-x86-64/tlspic.td: Likewise.
>
> * ld-x86-64/ilp32-8.d: Match any file format variant.
> Use a -Ttext and adjust expected results, to handle variant layouts.
> * ld-x86-64/ilp32-9.d: Likewise.
>
> * ld-i386/alloc.t: Remove superfluous OUTPUT_FORMAT statement.
> * ld-i386/pr12627.t: Likewise.
>
> * ld-x86-64/abs-l1om.d: Add target: constraint.
> * ld-x86-64/protected2-l1om.d: Likewise.
> * ld-x86-64/protected3-l1om.d: Likewise.
> * ld-x86-64/ilp32-4.d: Likewise.
>
> * ld-x86-64/plt.s: New file.
> * ld-x86-64/pltlib.s: New file.
> * ld-x86-64/plt.pd: New file.
> * ld-x86-64/x86-64.exp (x86_64tests): Add them.
>
> * ld-i386/plt.s: New file.
> * ld-i386/pltlib.s: New file.
> * ld-i386/plt.pd: New file.
> * ld-i386/plt-pic.s: New file.
> * ld-i386/plt-pic.pd: New file.
> * ld-i386/i386.exp (i386tests): Add them.
>
The i386 and x86-64 changes are OK.
Thanks.
H.J.
More information about the Binutils
mailing list