From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S932718AbcFOTNo (ORCPT ); Wed, 15 Jun 2016 15:13:44 -0400 Received: from mail.efficios.com ([78.47.125.74]:43078 "EHLO mail.efficios.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753285AbcFOTNm (ORCPT ); Wed, 15 Jun 2016 15:13:42 -0400 Date: Wed, 15 Jun 2016 19:13:39 +0000 (UTC) From: Mathieu Desnoyers To: Josh Poimboeuf Cc: lttng-dev , Linux Kernel Mailing List Message-ID: <734452688.38065.1466018019445.JavaMail.zimbra@efficios.com> In-Reply-To: <20160615181817.vvnf66z6cpwcraq7@treble> References: <1074038231.37468.1466009716273.JavaMail.zimbra@efficios.com> <20160615181817.vvnf66z6cpwcraq7@treble> Subject: Re: stack validation warning on lttng-modules bytecode interpreter MIME-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: 7bit X-Originating-IP: [78.47.125.74] X-Mailer: Zimbra 8.6.0_GA_1178 (ZimbraWebClient - FF45 (Linux)/8.6.0_GA_1178) Thread-Topic: stack validation warning on lttng-modules bytecode interpreter Thread-Index: pt8yi993PB2G1Sn3RRKEQLgE53v6nA== Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org ----- On Jun 15, 2016, at 2:18 PM, Josh Poimboeuf jpoimboe@redhat.com wrote: > On Wed, Jun 15, 2016 at 04:55:16PM +0000, Mathieu Desnoyers wrote: >> Hi Josh, >> >> I notice that with gcc 6.1.1, kernel 4.6, with >> CONFIG_STACK_VALIDATION=y, building lttng-modules master >> at commit 6c09dd94 gives this warning: >> >> lttng-modules/lttng-filter-interpreter.o: warning: objtool: >> lttng_filter_interpret_bytecode()+0x58: sibling call from callable instruction >> with changed frame pointer >> >> this object implements a bytecode interpreter using an explicit >> jump table (see >> https://github.com/lttng/lttng-modules/blob/master/lttng-filter-interpreter.c) >> >> If I define "INTERPRETER_USE_SWITCH" at the top of the file, >> thus using the switch-case fallback implementation, the >> warning vanishes. >> >> We use an explicit jump table rather than a switch case whenever >> possible for performance reasons. >> >> I notice that tools/objtool/builtin-check.c needs to be aware of >> switch-cases transformed into jump tables by the compiler. Are >> explicit jump tables supported by the stack validator ? Do we >> need to add annotation to our code ? > > Hi Mathieu, > > Unfortunately objtool doesn't know how to validate this type of jump > table. So to avoid the warning you'll need to add an annotation to tell > objtool to ignore it: > > STACK_FRAME_NON_STANDARD(lttng_filter_interpret_bytecode); > > We had to annotate __bpf_prog_run() in the kernel for the same reason. Thanks for the tip! Unfortunately it does not seem to work. objdump -t lttng/lttng-filter-interpreter.o output gives: 0000000000000000 l d __func_stack_frame_non_standard 0000000000000000 __func_stack_frame_non_standard 0000000000000000 l O __func_stack_frame_non_standard 0000000000000008 __func_stack_frame_non_standard_lttng_filter_interpret_bytecode Running objtool check (built in O0) in gdb on lttng-filter-interpreter.o built with the STACK_FRAME_NON_STANDARD define, it appears that the following function: static bool ignore_func(struct objtool_file *file, struct symbol *func) { struct rela *rela; struct instruction *insn; /* check for STACK_FRAME_NON_STANDARD */ if (file->whitelist && file->whitelist->rela) list_for_each_entry(rela, &file->whitelist->rela->rela_list, list) if (rela->sym->sec == func->sec && rela->addend == func->offset) return true; /* check if it has a context switching instruction */ func_for_each_insn(file, func, insn) if (insn->type == INSN_CONTEXT_SWITCH) return true; return false; } For lttng_filter_interpret_bytecode, while in the first list iteration: (gdb) print rela->sym->sec $18 = (struct section *) 0x7ffff7e20010 (gdb) print func->sec $19 = (struct section *) 0x7ffff7e20010 But (gdb) print rela->addend $20 = 0 (gdb) print func->offset $21 = 928 So for some reason it never match the ignore_func. This happens both when I build lttng-modules as a kernel module, and when I build it into the kernel image. Any idea why ? Thanks, Mathieu > > -- > Josh -- Mathieu Desnoyers EfficiOS Inc. http://www.efficios.com