The following commit
d15d356887e7 ("perf/x86: Make perf callchains work without CONFIG_FRAME_POINTER")
was a step in the right direction, but it triggered some BPF selftest
failures. That commit exposed the fact that we currently can't unwind
through BPF code.
- Patch 1 (originally from Song Liu) fixes a bug in the above commit
(regs->ip getting skipped in the stack trace).
- Patch 2 fixes non-JIT BPF for the ORC unwinder.
- Patches 3-5 are preparatory improvements for patch 6.
- Patch 6 fixes JIT BPF for the FP unwinder.
- Patch 7 fixes JIT BPF for the ORC unwinder (building on patch 6).
- Patches 8-9 are some readability cleanups.
Josh Poimboeuf (8):
objtool: Fix ORC unwinding in non-JIT BPF generated code
x86/bpf: Move epilogue generation to a dedicated function
x86/bpf: Simplify prologue generation
x86/bpf: Support SIB byte generation
x86/bpf: Fix JIT frame pointer usage
x86/unwind/orc: Fall back to using frame pointers for generated code
x86/bpf: Convert asm comments to AT&T syntax
x86/bpf: Convert MOV function/macro argument ordering to AT&T syntax
Song Liu (1):
perf/x86: Always store regs->ip in perf_callchain_kernel()
arch/x86/events/core.c | 10 +-
arch/x86/kernel/unwind_orc.c | 26 ++-
arch/x86/net/bpf_jit_comp.c | 355 ++++++++++++++++++++---------------
kernel/bpf/core.c | 5 +-
tools/objtool/check.c | 16 +-
5 files changed, 246 insertions(+), 166 deletions(-)
--
2.20.1
