aboutsummaryrefslogtreecommitdiff
path: root/bfd/elf64.c
diff options
context:
space:
mode:
authorJose E. Marchesi <jose.marchesi@oracle.com>2016-07-27 06:59:16 -0700
committerJose E. Marchesi <jose.marchesi@oracle.com>2016-07-29 00:17:04 -0700
commitc88960d081f0b37ec03c66a13115e2a68e40d1ad (patch)
tree7e85bf4592f3da65142956460fdddd2c2f96a8f3 /bfd/elf64.c
parent76539d80f708c535133c4a2a1eef61da305a3e80 (diff)
downloadgdb-c88960d081f0b37ec03c66a13115e2a68e40d1ad.zip
gdb-c88960d081f0b37ec03c66a13115e2a68e40d1ad.tar.gz
gdb-c88960d081f0b37ec03c66a13115e2a68e40d1ad.tar.bz2
gas: avoid spurious failures in non-ELF targets in the SPARC testsuite.
Many of the existing sparc tests fail in non-ELF targets (coff and a.out) due to spurious differences in the expected results: - Unlike ELF, a.out text sections are aligned to 2**3 and padded accordingly. The padding instruction is a `nop' (01 00 00 00). - Likewise, coff text sections are also aligned to 2**3 and padded accordingly. However, the padding instruction in these targets is an `illtrap 0' (00 00 00 00). - Unlike ELF, a.out and coff binaries don't contain hardware capabilities bits that could be used by BFD to determine the opcodes architecture corresponding to the instructions encoded in the objects (v9, v9a, v9b, v9c, etc). Consequently, in both a.out and coff tests we would need to pass proper `-m sparc:vXXX' options when invoking objdump before comparing results. In order to fix these issues, the most obvious solution would be to have three variants of .d files per impacted test. For example, for save.d we would have: save-elf.d, save-aout.d and save-coff.d. Using the `#source' directive, a single save.s file would provide the input for all of them. However, this approach has the following problems: - The #target and #notarget .d directives are very limited: they use globs instead of regular expressions, and thus it is not possible (or too messy) to use them to discriminate between elf, coff and a.out sparc targets. - It adds little or no value to have variants of all these tests for all the target types, and it would be a burden to maintain them. Actually the features tested in the spuriously failing tests (relatively modern sparc instructions, registers and asis) are not really found in running coff or a.out sparc systems. This patch changes sparc.exp so it will run these tests only in ELF-targets, using the more standard `is_elf_format' from binutils-common.exp instead of the ad-hoc (and less convenient, as it must be called before _every_ single elf-only test) sparc_elf_setup. Incidentally, the patch also fixes the #name entry for save-args.d. Tested in sparc*-*-linux-gnu, sparc-aout and sparc-coff targets. gas/ChangeLog: 2016-07-27 Jose E. Marchesi <jose.marchesi@oracle.com> * testsuite/gas/sparc/sparc.exp: Use is_elf_format to discriminate ELF targets. Run natural, natural-32, pr4587, ticc-imm-reg, v8-movwr-imm, pause, save-args, cbcond, cfr, crypto edge, flush, hpcvis3, ima, ld_st_fsr, ldtw_sttw, ldd_std, ldx_stx, ldx_efsr, mwait, mcdper, sparc5vis4, xcrypto, v9branch1 and imm-plus-rreg only in ELF targets. (sparc_elf_setup): Delete. * testsuite/gas/sparc/save-args.d: Fix a copy-paste typo in the test's #name entry.
Diffstat (limited to 'bfd/elf64.c')
0 files changed, 0 insertions, 0 deletions