This series cleans up recordmcount and then makes it into
an objtool subcommand.
The series starts with 8 cleanup patches which make recordmcount
easier to review and integrate with objtool. The final 5 patches
show the beginning steps of converting recordmcount to use objtool's
ELF code rather than its own open-coded methods of accessing ELF
files.
---
v3:
Rebased on mainline. s/elf_open/elf_read/ in recordmcount.c
v2:
Fix whitespace before line continuation
Add ftrace/mcount/record.h to objtool_dep
Rename the Makefile variable BUILD_C_RECORDMCOUNT to
better reflect its purpose
Similar: rename recordmcount_source => recordmcount_dep
When using objtool we can just depend on the
binary rather than the source the binary is
built from. This should address Josh's feedback and
make the Makefile code a bit clearer
Add a comment to make reading the Makefile a little
easier
Rebased to latest mainline -rc
Collected some build time measurements
Build times measurements (measured for v2 posting) -- median of multiple
runs in a VM measured with "time":
mainline (5.2.0-rc4) build times (median of 3 runs):
real 2m58.379s
user 2m29.621s
sys 1m35.116s
Post recordmcount-cleanup build times (median of 5 runs):
real 2m51.973s
user 2m29.094s
sys 1m33.688s
objtool mcount build times (median of 7 runs):
real 2m57.92s
user 2m33.73s
sys 1m37.06s
Note: I saw some significant variation especially in the "real" time
measurements probably because it was in a VM on a machine with
various "idle" GUI tasks running. This is why I took the median
rather than the mean. Though I haven't run the statistics, my
sense is the numbers don't support concluding that things really
got any faster or slower.
Matt Helsley (13):
recordmcount: Remove redundant strcmp
recordmcount: Remove uread()
recordmcount: Remove unused fd from uwrite() and ulseek()
recordmcount: Rewrite error/success handling
recordmcount: Kernel style function signature formatting
recordmcount: Kernel style formatting
recordmcount: Remove redundant cleanup() calls
recordmcount: Clarify what cleanup() does
objtool: Prepare to merge recordmcount
objtool: Make recordmcount into an objtool subcmd
objtool: recordmcount: Start using objtool's elf wrapper
objtool: recordmcount: Search for __mcount_loc before walking the
sections
objtool: recordmcount: Convert do_func() relhdrs
Makefile | 6 +-
scripts/.gitignore | 1 -
scripts/Makefile | 1 -
scripts/Makefile.build | 25 +-
tools/objtool/.gitignore | 1 +
tools/objtool/Build | 1 +
tools/objtool/Makefile | 1 +
tools/objtool/builtin-mcount.c | 72 +++++
tools/objtool/builtin-mcount.h | 23 ++
tools/objtool/builtin.h | 1 +
tools/objtool/objtool.c | 1 +
{scripts => tools/objtool}/recordmcount.c | 350 ++++++++++-----------
{scripts => tools/objtool}/recordmcount.h | 197 +++++++-----
{scripts => tools/objtool}/recordmcount.pl | 0
14 files changed, 406 insertions(+), 274 deletions(-)
create mode 100644 tools/objtool/builtin-mcount.c
create mode 100644 tools/objtool/builtin-mcount.h
rename {scripts => tools/objtool}/recordmcount.c (78%)
rename {scripts => tools/objtool}/recordmcount.h (78%)
rename {scripts => tools/objtool}/recordmcount.pl (100%)
--
2.20.1
On Wed, Jul 24, 2019 at 02:04:54PM -0700, Matt Helsley wrote:
> This series cleans up recordmcount and then makes it into
> an objtool subcommand.
>
> The series starts with 8 cleanup patches which make recordmcount
> easier to review and integrate with objtool. The final 5 patches
> show the beginning steps of converting recordmcount to use objtool's
> ELF code rather than its own open-coded methods of accessing ELF
> files.
>
> ---
>
> v3:
> Rebased on mainline. s/elf_open/elf_read/ in recordmcount.c
So this will be the first time objtool will be used on non-x86 arches.
Have you tested the other HAVE_C_RECORDMCOUNT arches, both native and
cross compilation?
Other than my few minor comments, it looks good. The patches are nicely
split up. I'll try to be more prompt next time ;-)
--
Josh