]> git.proxmox.com Git - mirror_ubuntu-kernels.git/commit
selftests/bpf: make multi-uprobe tests work in RELEASE=1 mode
authorAndrii Nakryiko <andrii@kernel.org>
Fri, 29 Mar 2024 19:04:10 +0000 (12:04 -0700)
committerAlexei Starovoitov <ast@kernel.org>
Sat, 30 Mar 2024 00:18:30 +0000 (17:18 -0700)
commit623bdd58be3727318d374f0052f9dfff1e87b854
tree4fa897fca4d3678d7fe97924daa9a2c7d1f3ff73
parent59f2f841179aa6a0899cb9cf53659149a35749b7
selftests/bpf: make multi-uprobe tests work in RELEASE=1 mode

When BPF selftests are built in RELEASE=1 mode with -O2 optimization
level, uprobe_multi binary, called from multi-uprobe tests is optimized
to the point that all the thousands of target uprobe_multi_func_XXX
functions are eliminated, breaking tests.

So ensure they are preserved by using weak attribute.

But, actually, compiling uprobe_multi binary with -O2 takes a really
long time, and is quite useless (it's not a benchmark). So in addition
to ensuring that uprobe_multi_func_XXX functions are preserved, opt-out
of -O2 explicitly in Makefile and stick to -O0. This saves a lot of
compilation time.

With -O2, just recompiling uprobe_multi:

  $ touch uprobe_multi.c
  $ time make RELEASE=1 -j90
  make RELEASE=1 -j90  291.66s user 2.54s system 99% cpu 4:55.52 total

With -O0:
  $ touch uprobe_multi.c
  $ time make RELEASE=1 -j90
  make RELEASE=1 -j90  22.40s user 1.91s system 99% cpu 24.355 total

5 minutes vs (still slow, but...) 24 seconds.

Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20240329190410.4191353-1-andrii@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
tools/testing/selftests/bpf/Makefile
tools/testing/selftests/bpf/uprobe_multi.c