| Summary: | Assembler preprocessor loses track of \@ | ||
|---|---|---|---|
| Product: | binutils | Reporter: | Wolfgang Ocker <weo> |
| Component: | gas | Assignee: | Alan Modra <amodra> |
| Status: | RESOLVED FIXED | ||
| Severity: | normal | CC: | jeremip11 |
| Priority: | P2 | ||
| Version: | 2.29 | ||
| Target Milestone: | 2.31 | ||
| Host: | Target: | ||
| Build: | Last reconfirmed: | 2018-01-31 00:00:00 | |
| Attachments: | This archive contains the sample files demonstrating the problem | ||
The master branch has been updated by Alan Modra <amodra@sourceware.org>: https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;h=ab1fadc6b2f057b817e1fc093650b63d9f6dd6c5 commit ab1fadc6b2f057b817e1fc093650b63d9f6dd6c5 Author: Alan Modra <amodra@gmail.com> Date: Wed Jan 31 13:34:18 2018 +1030 PR22714, Assembler preprocessor loses track of \@ The PR22714 testcase is such that the input buffer processed by do_scrub_chars ends on this line 1: bug "Returning to usermode but unexpected PSR bits set?", \@ right at the backslash. (The line is part of a macro definition.) The next input buffer then starts with '@' which starts a comment on ARM, and the check for \@ fails due to to == tostart. Now it would be possible to simply access to[-1] in this particular case, but that's ugly, and to be absolutely safe from people deliberately trying to crash gas we'd need the read.c:read_a_source_file buffer passed to do_scrub_chars to have a single byte pad at the start. PR 22714 * app.c (last_char): New static var. (struct app_save): Add last_char field. (app_push, app_pop): Handle it. (do_scrub_chars): Use last_char in test for "\@". Set last_char. Fixed |
Created attachment 10750 [details] This archive contains the sample files demonstrating the problem OS: Fedora 27, all updates arm-linux-gnueabi-as, version 2.29.1, build from sources $ arm-linux-gnueabi-as -version GNU assembler (GNU Binutils) 2.29.1 Copyright (C) 2017 Free Software Foundation, Inc. This program is free software; you may redistribute it under the terms of the GNU General Public License version 3 or later. This program has absolutely no warranty. This assembler was configured for a target of `arm-linux-gnueabi'. The attached file "entry-armv-okay-1. s" can be translated, the file "entry-armv-fail-1. s" cannot. The error message is: /home/user/aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa/kernel/linux/arch/arm/kernel/entry-header.S: Assembler messages: /home/user/aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa/kernel/linux/arch/arm/kernel/entry-header.S:319: Warning: stray `\' /home/user/aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa/kernel/linux/arch/arm/kernel/entry-armv.S:773: Warning: stray `\' /home/user/aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa/kernel/linux/arch/arm/kernel/entry-armv.S:773: Error: bad expression Command line: arm-linux-gnueabi-as entry-armv-fail-1.s The files differ in this way: --- entry-armv-okay-1.s 2018-01-15 17:03:30.339301143 +0100 +++ entry-armv-fail-1.s 2018-01-15 17:00:43.578260650 +0100 @@ -7,7 +7,7 @@ # 1 "./include/generated/autoconf.h" 1 -# 16 "/home/user/aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa/kernel/linux/include/linux/kconfig.h" 2 +# 6 "/home/user/aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa/kernel/linux/include/linux/kconfig.h" 2 # 1 "<command-line>" 2 # 1 "/home/user/aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa/kernel/linux/arch/arm/include/asm/unified.h" 1 # 74 "/home/user/aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa/kernel/linux/arch/arm/include/asm/unified.h" The error also occurs in other variations. The long file paths in the comments, inserted by the preprocessor, are striking. I also observed the bug with the assembler v2.28 (stable Debian). The file comes from the Linux kernel: arch/arm/kernel/entry-armv.S (preprocessed)