perf tools: Fix strict alias issue for find_first_bit
RFC 1/1 perf tools: Fix strict alias issue for find_first_bit
hi,
got issue below when compiling perf tool on i686 with gcc 4.4, but not sure the patch is correct workaround here.
thanks for comments,
jirka
---
When compiling perf tool code with gcc 4.4.7 I'm getting following error:
CC util/session.o
cc1: warnings being treated as errorsutil/session.c: In function ‘perf_session_deliver_event’:/root/linux/tools/perf/util/include/linux/bitops.h:109: error: dereferencing pointer ‘p’ does break strict-aliasing rules /root/linux/tools/perf/util/include/linux/bitops.h:101: error: dereferencing pointer ‘p’ does break strict-aliasing rules util/session.c:697: note: initialized from here /root/linux/tools/perf/util/include/linux/bitops.h:101: note: initialized from here
make[1]: *** [util/session.o] Error 1
make: *** [util/session.o] Error 2
The aliased types here are u64 and unsigned long pointers, which is safe for the find_first_bit processing.
This error shows up for me only for gcc 4.4 on 32bit x86, even for -Wstrict-aliasing=3, while newer gcc are quiet and scream here for -Wstrict-aliasing={2,1}. Looks like newer gcc changed the rules for strict alias warnings.
The gcc documentation offers workaround for valid
aliasing by using __may_alias__ attribute:
http://gcc.gnu.org/onlinedocs/gcc-4.4.0/gcc/Type-Attributes.html
Using this workaround for the find_first_bit function.
--- tools/perf/util/include/linux/bitops.h | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
tools/perf/util/include/linux/bitops.h +3 -1
@@ -87,13 +87,15 @@ static __always_inline unsigned long __ffs(unsigned long word)return num; } +typedef const unsigned long __attribute__((__may_alias__)) long_alias_t; +/* * Find the first set bit in a memory region. */ static inline unsigned long find_first_bit (const unsigned long *addr,unsigned long size) { -const unsigned long *p = addr; +long_alias_t *p = (long_alias_t *) addr;unsigned long result =0 ;unsigned long tmp;
Signature · 1 line
1.8.3.1
Peter Zijlstra
maintainerQuoted 25 lines · Jiri Olsa
hi,
got issue below when compiling perf tool on i686 with gcc 4.4, but not sure the patch is correct workaround here.thanks for comments,
jirka---
When compiling perf tool code with gcc 4.4.7 I'm getting following error:CC
util/session.o
cc1: warnings being treated as errorsutil/session.c: In function ‘perf_session_deliver_event’:/root/linux/tools/perf/util/include/linux/bitops.h:109: error: dereferencing pointer ‘p’ does break strict-aliasing rules/root/linux/tools/perf/util/include/linux/bitops.h:101: error: dereferencing pointer ‘p’ does break strict-aliasing rulesutil/session.c:697: note: initialized from here/root/linux/tools/perf/util/include/linux/bitops.h:101: note: initialized from here
make[1]: *** [util/session.o] Error 1
make: *** [util/session.o] Error 2The aliased types here are u64 and unsigned long pointers, which is safe for the
find_first_bitprocessing.
Quoted 9 lines
This error shows up for me only for gcc 4.4 on 32bit x86, even for -Wstrict-aliasing=3, while newer gcc are quiet and scream here for -Wstrict-aliasing={2,1}. Looks like newer gcc changed the rules for strict alias warnings.
The gcc documentation offers workaround for valid
aliasing by using__may_alias__attribute:
http://gcc.gnu.org/onlinedocs/gcc-4.4.0/gcc/Type-Attributes.html
The kernel builds with -fno-strict-aliasing because the C aliasing rules are a bunch of monkey poo. Perf tool should probably do the same.
PATCH 1/1 perf tools: Do not compile with -fstrict-aliasing
Quoted 31 lines · Peter Zijlstra
Quoted 30 lines · Jiri Olsa
hi,
got issue below when compiling perf tool on i686 with gcc 4.4, but not sure the patch is correct workaround here.thanks for comments,
jirka---
When compiling perf tool code with gcc 4.4.7 I'm getting following error:CC
util/session.o
cc1: warnings being treated as errorsutil/session.c: In function ‘perf_session_deliver_event’:/root/linux/tools/perf/util/include/linux/bitops.h:109: error: dereferencing pointer ‘p’ does break strict-aliasing rules/root/linux/tools/perf/util/include/linux/bitops.h:101: error: dereferencing pointer ‘p’ does break strict-aliasing rulesutil/session.c:697: note: initialized from here/root/linux/tools/perf/util/include/linux/bitops.h:101: note: initialized from here
make[1]: *** [util/session.o] Error 1
make: *** [util/session.o] Error 2The aliased types here are u64 and unsigned long pointers, which is safe for the
find_first_bitprocessing.This error shows up for me only for gcc 4.4 on 32bit x86, even for -Wstrict-aliasing=3, while newer gcc are quiet and scream here for -Wstrict-aliasing={2,1}. Looks like newer gcc changed the rules for strict alias warnings.
Quoted 7 lines
Quoted 4 lines
The gcc documentation offers workaround for valid
aliasing by using__may_alias__attribute:
http://gcc.gnu.org/onlinedocs/gcc-4.4.0/gcc/Type-Attributes.htmlThe kernel builds with -fno-strict-aliasing because the C aliasing rules are a bunch of monkey poo. Perf tool should probably do the same.
my pleasure ;-)
thanks,
jirka
---
Switching off the strict aliasing for perf tool compilation, because it causes compilation issues/errors for older gcc, and to quote Peter:
"C aliasing rules are a bunch of monkey poo"
We don't enable strict aliasing actively, it is forced by -Ox option. Using -fno-strict-aliasing option at the same time we use -O.
Removing -Wstrict-aliasing=3 option as well.
--- tools/perf/config/Makefile | 2 +- tools/scripts/Makefile.include | 1 - 2 files changed, 1 insertion(+), 2 deletions(-)
tools/perf/config/Makefile +1 -1
@@ -94,7 +94,7 @@ ifndef DEBUG endif ifeq ($(DEBUG),0) - CFLAGS += -O6 + CFLAGS += -O6 -fno-strict-aliasing endif ifdef PARSER_DEBUG
tools/scripts/Makefile.include +0 -1
@@ -32,7 +32,6 @@ EXTRA_WARNINGS += -Wold-style-definition EXTRA_WARNINGS += -Wpacked EXTRA_WARNINGS += -Wredundant-decls EXTRA_WARNINGS += -Wshadow -EXTRA_WARNINGS += -Wstrict-aliasing=3 EXTRA_WARNINGS += -Wstrict-prototypes EXTRA_WARNINGS += -Wswitch-default EXTRA_WARNINGS += -Wswitch-enum
Signature · 1 line
1.8.3.1
tip-bot for Jiri Olsa
merge botperf tools: Fix strict alias issue for find_first_bit
Commit-ID: b39c2a57a00a841f057a75b41df4c26173288b66 Gitweb: http://git.kernel.org/tip/b39c2a57a00a841f057a75b41df4c26173288b66 Author: Jiri Olsa <[email protected]> AuthorDate: Wed, 26 Feb 2014 18:14:26 +0100 Committer: Arnaldo Carvalho de Melo <[email protected]> CommitDate: Fri, 28 Feb 2014 10:39:40 -0300
perf tools: Fix strict alias issue for find_first_bit
When compiling perf tool code with gcc 4.4.7 I'm getting following error:
CC util/session.o
cc1: warnings being treated as errors
util/session.c: In function ‘perf_session_deliver_event’:
tools/perf/util/include/linux/bitops.h:109: error: dereferencing pointer ‘p’ does break strict-aliasing rules tools/perf/util/include/linux/bitops.h:101: error: dereferencing pointer ‘p’ does break strict-aliasing rules util/session.c:697: note: initialized from here
tools/perf/util/include/linux/bitops.h:101: note: initialized from here
make[1]: *** [util/session.o] Error 1
make: *** [util/session.o] Error 2
The aliased types here are u64 and unsigned long pointers, which is safe for the find_first_bit processing.
This error shows up for me only for gcc 4.4 on 32bit x86, even for -Wstrict-aliasing=3, while newer gcc are quiet and scream here for -Wstrict-aliasing={2,1}. Looks like newer gcc changed the rules for strict alias warnings.
The gcc documentation offers workaround for valid aliasing by using __may_alias__ attribute:
http://gcc.gnu.org/onlinedocs/gcc-4.4.0/gcc/Type-Attributes.html
Using this workaround for the find_first_bit function.
--- tools/perf/util/include/linux/bitops.h | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
tools/perf/util/include/linux/bitops.h +3 -1
@@ -87,13 +87,15 @@ static __always_inline unsigned long __ffs(unsigned long word)return num; } +typedef const unsigned long __attribute__((__may_alias__)) long_alias_t; +/* * Find the first set bit in a memory region. */ static inline unsigned long find_first_bit (const unsigned long *addr,unsigned long size) { -const unsigned long *p = addr; +long_alias_t *p = (long_alias_t *) addr;unsigned long result =0 ;unsigned long tmp;