From mboxrd@z Thu Jan 1 00:00:00 1970 From: david.gilbert@linaro.org (Dr. David Alan Gilbert) Date: Thu, 13 Jan 2011 13:00:45 +0000 Subject: Possible patch; fix perf Annotation of Thumb code Message-ID: <20110113130037.GB5241@davesworkthinkpad> To: linux-arm-kernel@lists.infradead.org List-Id: linux-arm-kernel.lists.infradead.org Hi, I'm finding that if I annotate an ARM Thumb function with perf that it's mis-disassembling it; below is a patch that fixes it, but I'm not sure if it's the best place for the fix. The problem is that on Thumb, the bottom bit of the symbol address is set for thumb functions, and thus perf starts disassembling at address+1, and since all thumb instructions are 2 bytes aligned to 2 bytes you end up disassembling across the middle of pairs of instructions. The patch removes that bottom bit during symbol loading. Questions: 1) Is this the right place to do it - does some other bit of Perf need the raw symbol value? My alternative is to mask it just before the objdump, but then my worry is if it also needs masking somewhere else. 2) Should the check be more selective - i.e. only some symbol types? This is against the Linaro 2.6.37 tree; I'm happy to rebase it against the clean 2.6.37 if people agree the patch is doing the right thing. Dave (For reference this corresponds to this bug https://bugs.launchpad.net/linux-linaro/+bug/677547 ) Signed-off-by: Dr. David Alan Gilbert --- diff --git a/tools/perf/util/symbol.c b/tools/perf/util/symbol.c index 439ab94..3d77b5e 100644 --- a/tools/perf/util/symbol.c +++ b/tools/perf/util/symbol.c @@ -1129,6 +1129,11 @@ static int dso__load_sym(struct dso *self, struct map *map, const char *name, section_name = elf_sec__name(&shdr, secstrs); + /* On ARM, symbols for thumb functions have 1 added to + the symbol address as a flag - remove it */ + if ((ehdr.e_machine == EM_ARM) && (sym.st_value & 1)) + sym.st_value-=1; + if (self->kernel != DSO_TYPE_USER || kmodule) { char dso_name[PATH_MAX];