[Bug ld/22840] New: Several TLS tests FAIL on Solaris/SPARC

[ro at gcc dot gnu.org]

https://sourceware.org/bugzilla/show_bug.cgi?id=22840

            Bug ID: 22840
           Summary: Several TLS tests FAIL on Solaris/SPARC
           Product: binutils
           Version: 2.31 (HEAD)
            Status: NEW
          Severity: normal
          Priority: P2
         Component: ld
          Assignee: unassigned at sourceware dot org
          Reporter: ro at gcc dot gnu.org
  Target Milestone: ---
              Host: sparc*-sun-solaris2.*
            Target: sparc*-sun-solaris2.*
             Build: sparc*-sun-solaris2.*

Several of the ld TLS tests FAIL on Solaris/SPARC (both in a
sparc-sun-solaris2.11
and sparcv9-sun-solaris2.11 configuration):

FAIL: 32-bit: TLS -fpic -shared transitions
FAIL: 32-bit: TLS -fpic and -fno-pic exec transitions
FAIL: 32-bit: TLS -fno-pic -shared
FAIL: 32-bit: TLS -fpie
FAIL: 32-bit: GOTDATA relocations
FAIL: relaxed call WDISP30 relocation
FAIL: 64-bit: TLS -fpic -shared transitions
FAIL: 64-bit: TLS -fpic and -fno-pic exec transitions
FAIL: 64-bit: TLS -fno-pic -shared
FAIL: 64-bit: TLS -fpie
FAIL: 64-bit: GOTDATA relocations

There are several issues here:

* Initially, the failures are like

/var/gcc/binutils/sparcv7/obj/binutils/ld/ld-new: target elf32-sparc not found
/var/gcc/binutils/sparcv7/obj/binutils/ld/ld-new: target elf32-sparc not found

  or the elf64-sparc equivalent.  This happens because sparc.exp hardcodes
  -melf32_sparc resp. -melf64_sparc, while Solaris uses (and requires) the
  -melf32_sparc_sol2 resp. -melf64_sparc_sol2 emulations.

  Unfortunately, unlike the #ld: lines in the *.?d files which can be modified
  by setting options_regsub(ld) for perform the appropriate substitution, this
  is not currently possible for the arg to run_ld_link_tests.

* To make some progress, I've hardcoded the Solaris values for now, which leads
  me to the next issues:

  The file format lines in the *.?d files under test all hardcode elf32-sparc
  or elf64-sparc, but should allow for elf32-sparc-sol2 (resp. elf32-sparc.*).

* Even with that fixed, there are many failures due to the fact that the
Solaris
  ABI requires a couple of additional symbols (cf. emultempl/solaris2.em
  (elf_solaris2_before_allocation), and those aren't currently expected in
  the output.  I'm uncertain how best to handle that: allow for a
postprocessing
  step after readelf that removes those from the output or augment the expected
  output accordingly?  It seems regexp_diff currently has no support for
  optional or target-specific lines in the expected output.

  Rainer

-- 
You are receiving this mail because:
You are on the CC list for the bug.