> On Feb 3, 2022, at 11:41 PM, Song Liu <song@xxxxxxxxxx> wrote: > > On Mon, Jan 31, 2022 at 10:31 PM Song Liu <song@xxxxxxxxxx> wrote: >> >> Using prog->jited_len is simpler and more accurate than current >> estimation (header + header->size). >> >> Signed-off-by: Song Liu <song@xxxxxxxxxx> > > Hmm... CI [1] reports error on test_progs 159/tailcalls, and bisect points to > this one. However, I couldn't figure out why this breaks tail call. > round_up(PAGE_SIZE) does fix it though. But that won't be accurate, right? > > Any suggestions on what could be the reason for these failures? > > Thanks, > Song > > [1] https://github.com/kernel-patches/bpf/runs/5060194776?check_suite_focus=true I guess this is the missing piece: diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c index 1ae41d0cf96c..bbef86cb4e72 100644 --- a/kernel/bpf/verifier.c +++ b/kernel/bpf/verifier.c @@ -13067,6 +13067,7 @@ static int jit_subprogs(struct bpf_verifier_env *env) prog->jited = 1; prog->bpf_func = func[0]->bpf_func; + prog->jited_len = func[0]->jited_len; prog->aux->func = func; prog->aux->func_cnt = env->subprog_cnt; bpf_prog_jit_attempt_done(prog); Will send v9 with this. > >> --- >> kernel/bpf/core.c | 5 +---- >> 1 file changed, 1 insertion(+), 4 deletions(-) >> >> diff --git a/kernel/bpf/core.c b/kernel/bpf/core.c >> index 14199228a6f0..e3fe53df0a71 100644 >> --- a/kernel/bpf/core.c >> +++ b/kernel/bpf/core.c >> @@ -537,13 +537,10 @@ long bpf_jit_limit_max __read_mostly; >> static void >> bpf_prog_ksym_set_addr(struct bpf_prog *prog) >> { >> - const struct bpf_binary_header *hdr = bpf_jit_binary_hdr(prog); >> - unsigned long addr = (unsigned long)hdr; >> - >> WARN_ON_ONCE(!bpf_prog_ebpf_jited(prog)); >> >> prog->aux->ksym.start = (unsigned long) prog->bpf_func; >> - prog->aux->ksym.end = addr + hdr->size; >> + prog->aux->ksym.end = prog->aux->ksym.start + prog->jited_len; >> } >> >> static void >> -- >> 2.30.2 >>
