From mboxrd@z Thu Jan 1 00:00:00 1970 From: Arnaldo Carvalho de Melo Subject: Re: [PATCH perf 3/3] tools/perf: recognize and process RECORD_MMAP events for bpf progs Date: Thu, 20 Sep 2018 11:05:06 -0300 Message-ID: <20180920140506.GD19861@kernel.org> References: <20180919223935.999270-1-ast@kernel.org> <20180919223935.999270-4-ast@kernel.org> <20180920133617.GB19861@kernel.org> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: "David S . Miller" , daniel@iogearbox.net, peterz@infradead.org, netdev@vger.kernel.org, kernel-team@fb.com To: Alexei Starovoitov Return-path: Received: from mail.kernel.org ([198.145.29.99]:45306 "EHLO mail.kernel.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1731025AbeITTst (ORCPT ); Thu, 20 Sep 2018 15:48:49 -0400 Content-Disposition: inline In-Reply-To: <20180920133617.GB19861@kernel.org> Sender: netdev-owner@vger.kernel.org List-ID: Em Thu, Sep 20, 2018 at 10:36:17AM -0300, Arnaldo Carvalho de Melo escreveu: > Em Wed, Sep 19, 2018 at 03:39:35PM -0700, Alexei Starovoitov escreveu: > > Recognize JITed bpf prog load/unload events. > > Add/remove kernel symbols accordingly. > > > > Signed-off-by: Alexei Starovoitov > > --- > > tools/perf/util/machine.c | 27 +++++++++++++++++++++++++++ > > tools/perf/util/symbol.c | 13 +++++++++++++ > > tools/perf/util/symbol.h | 1 + > > 3 files changed, 41 insertions(+) > > > > diff --git a/tools/perf/util/machine.c b/tools/perf/util/machine.c > > index c4acd2001db0..ae4f8a0fdc7e 100644 > > --- a/tools/perf/util/machine.c > > +++ b/tools/perf/util/machine.c > > @@ -25,6 +25,7 @@ > > #include "sane_ctype.h" > > #include > > #include > > +#include > > > > static void __machine__remove_thread(struct machine *machine, struct thread *th, bool lock); > > > > @@ -1460,6 +1461,32 @@ static int machine__process_kernel_mmap_event(struct machine *machine, > > enum dso_kernel_type kernel_type; > > bool is_kernel_mmap; > > > > + /* process JITed bpf programs load/unload events */ > > + if (event->mmap.pid == ~0u && event->mmap.tid == BPF_FS_MAGIC) { > > > So, this would be in machine__process_kernel_munmap-event(machine), etc, > no check for BPF_FS_MAGIC would be needed with a PERF_RECORD_MUNMAP. > > > + struct symbol *sym; > > + u64 ip; > > + > > + map = map_groups__find(&machine->kmaps, event->mmap.start); > > + if (!map) { > > + pr_err("No kernel map for IP %lx\n", event->mmap.start); > > + goto out_problem; > > + } > > + ip = event->mmap.start - map->start + map->pgoff; > > + if (event->mmap.filename[0]) { > > + sym = symbol__new(ip, event->mmap.len, 0, 0, > > + event->mmap.filename); > > Humm, so the bpf program would be just one symbol... bpf-to-bpf calls > will be to a different bpf program, right? > > /me goes to read https://lwn.net/Articles/741773/ > "[PATCH bpf-next 00/13] bpf: introduce function calls" After reading it, yeah, I think we need some way to access a symbol table for a BPF program, and also its binary so that we can do annotation, static (perf annotate) and live (perf top), was this already considered? I think one can get the binary for a program giving sufficient perms somehow, right? One other thing I need to catch up 8-) - Arnaldo > > + dso__insert_symbol(map->dso, sym); > > + } else { > > + if (symbols__erase(&map->dso->symbols, ip)) { > > + pr_err("No bpf prog at IP %lx/%lx\n", > > + event->mmap.start, ip); > > + goto out_problem; > > + } > > + dso__reset_find_symbol_cache(map->dso); > > + } > > + return 0; > > + } > > + > > /* If we have maps from kcore then we do not need or want any others */ > > if (machine__uses_kcore(machine)) > > return 0; > > diff --git a/tools/perf/util/symbol.c b/tools/perf/util/symbol.c > > index d188b7588152..0653f313661d 100644 > > --- a/tools/perf/util/symbol.c > > +++ b/tools/perf/util/symbol.c > > @@ -353,6 +353,19 @@ static struct symbol *symbols__find(struct rb_root *symbols, u64 ip) > > return NULL; > > } > > > > +int symbols__erase(struct rb_root *symbols, u64 ip) > > +{ > > + struct symbol *s; > > + > > + s = symbols__find(symbols, ip); > > + if (!s) > > + return -ENOENT; > > + > > + rb_erase(&s->rb_node, symbols); > > + symbol__delete(s); > > + return 0; > > +} > > + > > static struct symbol *symbols__first(struct rb_root *symbols) > > { > > struct rb_node *n = rb_first(symbols); > > diff --git a/tools/perf/util/symbol.h b/tools/perf/util/symbol.h > > index f25fae4b5743..92ef31953d9a 100644 > > --- a/tools/perf/util/symbol.h > > +++ b/tools/perf/util/symbol.h > > @@ -310,6 +310,7 @@ char *dso__demangle_sym(struct dso *dso, int kmodule, const char *elf_name); > > > > void __symbols__insert(struct rb_root *symbols, struct symbol *sym, bool kernel); > > void symbols__insert(struct rb_root *symbols, struct symbol *sym); > > +int symbols__erase(struct rb_root *symbols, u64 ip); > > void symbols__fixup_duplicate(struct rb_root *symbols); > > void symbols__fixup_end(struct rb_root *symbols); > > void map_groups__fixup_end(struct map_groups *mg); > > -- > > 2.17.1