On Thu, Mar 10, 2022 at 02:47:18PM +0100, Peter Zijlstra wrote: > On Thu, Mar 10, 2022 at 10:35:32AM +0100, Peter Zijlstra wrote: > > > > $ test_progs -t fentry > > > test_fentry_fexit:PASS:fentry_skel_load 0 nsec > > > test_fentry_fexit:PASS:fexit_skel_load 0 nsec > > > test_fentry_fexit:PASS:fentry_attach 0 nsec > > > test_fentry_fexit:FAIL:fexit_attach unexpected error: -1 (errno 19) > > > #54 fentry_fexit:FAIL > > This seems to cure the above. fexit_bpf2bpf is still failing, I'll dig > into that after lunch. > > diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c > index acb50fb7ed2d..2d86d3c09d64 100644 > --- a/kernel/trace/ftrace.c > +++ b/kernel/trace/ftrace.c > @@ -5354,6 +5381,11 @@ int modify_ftrace_direct(unsigned long ip, > mutex_lock(&direct_mutex); > > mutex_lock(&ftrace_lock); > + > + ip = ftrace_location(ip); > + if (!ip) > + goto out_unlock; > + > entry = find_direct_entry(&ip, &rec); > if (!entry) > goto out_unlock; This seems to cure most of the rest. I'm still seeing one failure: libbpf: prog 'connect_v4_prog': BPF program load failed: Invalid argument libbpf: failed to load program 'connect_v4_prog' libbpf: failed to load object './connect4_prog.o' test_fexit_bpf2bpf_common:FAIL:tgt_prog_load unexpected error: -22 (errno 22) #48/4 fexit_bpf2bpf/func_replace_verify:FAIL but the rest seems to be good again. diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c index 2b1e266ff95c..3fecfe8c4429 100644 --- a/arch/x86/net/bpf_jit_comp.c +++ b/arch/x86/net/bpf_jit_comp.c @@ -384,6 +395,13 @@ int bpf_arch_text_poke(void *ip, enum bpf_text_poke_type t, /* BPF poking in modules is not supported */ return -EINVAL; + /* + * See emit_prologue(), for IBT builds the trampoline hook is preceded + * with an ENDBR instruction. + */ + if (is_endbr(*(u32 *)ip)) + ip += ENDBR_INSN_SIZE; + return __bpf_arch_text_poke(ip, t, old_addr, new_addr, true); }
