* [PATCH] cmake: workaround GCC 7.2 segfault on v?sn?printf
@ 2018-12-28 4:48 Alexander Turenko
2018-12-28 6:27 ` Alexander Turenko
2018-12-28 8:24 ` Vladimir Davydov
0 siblings, 2 replies; 4+ messages in thread
From: Alexander Turenko @ 2018-12-28 4:48 UTC (permalink / raw)
To: Vladimir Davydov; +Cc: Alexander Turenko, tarantool-patches
GCC 7.2.1 fails like so on our codebase:
/tarantool/src/lib/json/json.c: In function ‘json_token_snprint’:
/tarantool/src/lib/json/json.c:324:1: internal compiler error: Segmentation fault
json_token_snprint(char *buf, int size, const struct json_token *token,
^~~~~~~~~~~~~~~~~~
The backtrace of cc1 is the following (failed to install compatible
debuginfo on Fedora 26, so ??s):
#0 0x0000000000d0320c in get_source_location_for_substring(cpp_reader*, string_concat_db*, unsigned int, cpp_ttype, int, int, int, unsigned int*) ()
#1 0x00000000005c0e79 in c_get_substring_location(substring_loc const&, unsigned int*) ()
#2 0x00000000007f8418 in format_warning_va(substring_loc const&, source_range const*, char const*, int, char const*, __va_list_tag (*) [1]) ()
#3 0x00000000007f838b in format_warning_at_substring(substring_loc const&, source_range const*, char const*, int, char const*, ...) ()
#4 0x0000000000c72c2d in ?? ()
#5 0x0000000000f6f308 in execute_one_pass(opt_pass*) ()
#6 0x0000000000787da6 in ?? ()
#7 0x00000000012ac338 in symbol_table::compile() ()
#8 0x0000000000de650f in symbol_table::finalize_compilation_unit() ()
#9 0x00000000012ee8df in ?? ()
#10 0x0000000000d3ba84 in toplev::main(int, char**) ()
#11 0x0000000000d3d35b in main ()
The issue in the GCC bugtracker: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83448
The idea of the commit is to check builtin snprintf with the code from
GCC test suite and disable all related builtins if the compiler is GCC
and the case fails.
We also need to disable FORTIFY_SOURCES in the case, because it uses
v?sn?printf builtins directly.
---
no issue
https://github.com/tarantool/tarantool/tree/Totktonada/workaround-gcc-7.2-snprintf-segfault
Be attentive: I pushed DO NOT MERGE commit upward the commit I send here
to enable full testing on the bugfix branch: [1] (hope it will pass).
The topmost commit should NOT be pushed to 2.1.
[1]: https://travis-ci.org/tarantool/tarantool/builds/472881226
cmake/compiler.cmake | 43 +++++++++++++++++++++++++++++++++++++++++++
1 file changed, 43 insertions(+)
diff --git a/cmake/compiler.cmake b/cmake/compiler.cmake
index 4062d13ec..b471b0c38 100644
--- a/cmake/compiler.cmake
+++ b/cmake/compiler.cmake
@@ -333,6 +333,8 @@ if (NOT HAVE_BUILTIN_CTZ OR NOT HAVE_BUILTIN_CTZLL)
HAVE_FFSL)
check_c_source_compiles("#include <string.h>\n#include <strings.h>\nint main(void) { return ffsll(0UL); }"
HAVE_FFSLL)
+ set(CMAKE_REQUIRED_FLAGS "")
+ set(CMAKE_REQUIRED_DEFINITIONS "")
endif()
endif()
@@ -343,3 +345,44 @@ else()
set(CMAKE_HOST_C_COMPILER ${CMAKE_C_COMPILER})
set(CMAKE_HOST_CXX_COMPILER ${CMAKE_CXX_COMPILER})
endif()
+
+# Check for broken __builtin_v?sn?printf() in gcc-7.2.
+# https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83448
+
+# It is important for the test case to be compiled w/o any extra
+# flags and with -O2. Say, -Wall allows the case be compiled
+# successfully on gcc-7.2.1.
+set(PREVIOUS_CMAKE_C_FLAGS ${CMAKE_C_FLAGS})
+set(CMAKE_C_FLAGS "")
+set(CMAKE_REQUIRED_DEFINITIONS "")
+set(CMAKE_REQUIRED_INCLUDES "")
+set(CMAKE_REQUIRED_LIBRARIES "")
+set(CMAKE_REQUIRED_FLAGS "-O2")
+
+check_c_source_compiles("
+char *a;
+int b;
+
+int
+main(void)
+{
+ for (;;) {
+ if (b < 0)
+ main();
+ __builtin_snprintf(a, b, \"%*s\", b, \"\");
+ }
+ return 0;
+}" HAVE_BUILTIN_SNPRINTF)
+
+set(CMAKE_REQUIRED_FLAGS "")
+set(CMAKE_C_FLAGS ${PREVIOUS_CMAKE_C_FLAGS})
+
+# /usr/include/stdio.h (bits/stdio2.h to be exact) uses builtin
+# gcc v?sn?printf functions when fortify source is enabled.
+if (CMAKE_COMPILER_IS_GNUCC AND NOT HAVE_BUILTIN_SNPRINTF)
+ add_compile_flags("C;CXX" "-Wp,-U_FORTIFY_SOURCE")
+ add_compile_flags("C;CXX" "-fno-builtin-sprintf")
+ add_compile_flags("C;CXX" "-fno-builtin-snprintf")
+ add_compile_flags("C;CXX" "-fno-builtin-vsnprintf")
+ add_compile_flags("C;CXX" "-fno-builtin-vsprintf")
+endif()
--
2.20.1
^ permalink raw reply [flat|nested] 4+ messages in thread* Re: [PATCH] cmake: workaround GCC 7.2 segfault on v?sn?printf
2018-12-28 4:48 [PATCH] cmake: workaround GCC 7.2 segfault on v?sn?printf Alexander Turenko
@ 2018-12-28 6:27 ` Alexander Turenko
2018-12-28 8:24 ` Vladimir Davydov
1 sibling, 0 replies; 4+ messages in thread
From: Alexander Turenko @ 2018-12-28 6:27 UTC (permalink / raw)
To: Vladimir Davydov; +Cc: tarantool-patches
Fixed Ubuntu Artful:
```
diff --git a/cmake/compiler.cmake b/cmake/compiler.cmake
index b471b0c38..0d7bf6347 100644
--- a/cmake/compiler.cmake
+++ b/cmake/compiler.cmake
@@ -350,14 +350,14 @@ endif()
# https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83448
# It is important for the test case to be compiled w/o any extra
-# flags and with -O2. Say, -Wall allows the case be compiled
-# successfully on gcc-7.2.1.
+# flags and with -O2. Say, -Wformat allows the case be compiled
+# successfully on gcc-7.2.
set(PREVIOUS_CMAKE_C_FLAGS ${CMAKE_C_FLAGS})
set(CMAKE_C_FLAGS "")
set(CMAKE_REQUIRED_DEFINITIONS "")
set(CMAKE_REQUIRED_INCLUDES "")
set(CMAKE_REQUIRED_LIBRARIES "")
-set(CMAKE_REQUIRED_FLAGS "-O2")
+set(CMAKE_REQUIRED_FLAGS "-O2 -Wno-format")
check_c_source_compiles("
char *a;
```
https://travis-ci.org/tarantool/tarantool/builds/472898526
WBR, Alexander Turenko.
On Fri, Dec 28, 2018 at 07:48:54AM +0300, Alexander Turenko wrote:
> GCC 7.2.1 fails like so on our codebase:
>
> /tarantool/src/lib/json/json.c: In function ‘json_token_snprint’:
> /tarantool/src/lib/json/json.c:324:1: internal compiler error: Segmentation fault
> json_token_snprint(char *buf, int size, const struct json_token *token,
> ^~~~~~~~~~~~~~~~~~
>
> The backtrace of cc1 is the following (failed to install compatible
> debuginfo on Fedora 26, so ??s):
>
> #0 0x0000000000d0320c in get_source_location_for_substring(cpp_reader*, string_concat_db*, unsigned int, cpp_ttype, int, int, int, unsigned int*) ()
> #1 0x00000000005c0e79 in c_get_substring_location(substring_loc const&, unsigned int*) ()
> #2 0x00000000007f8418 in format_warning_va(substring_loc const&, source_range const*, char const*, int, char const*, __va_list_tag (*) [1]) ()
> #3 0x00000000007f838b in format_warning_at_substring(substring_loc const&, source_range const*, char const*, int, char const*, ...) ()
> #4 0x0000000000c72c2d in ?? ()
> #5 0x0000000000f6f308 in execute_one_pass(opt_pass*) ()
> #6 0x0000000000787da6 in ?? ()
> #7 0x00000000012ac338 in symbol_table::compile() ()
> #8 0x0000000000de650f in symbol_table::finalize_compilation_unit() ()
> #9 0x00000000012ee8df in ?? ()
> #10 0x0000000000d3ba84 in toplev::main(int, char**) ()
> #11 0x0000000000d3d35b in main ()
>
> The issue in the GCC bugtracker: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83448
>
> The idea of the commit is to check builtin snprintf with the code from
> GCC test suite and disable all related builtins if the compiler is GCC
> and the case fails.
>
> We also need to disable FORTIFY_SOURCES in the case, because it uses
> v?sn?printf builtins directly.
> ---
>
> no issue
> https://github.com/tarantool/tarantool/tree/Totktonada/workaround-gcc-7.2-snprintf-segfault
>
> Be attentive: I pushed DO NOT MERGE commit upward the commit I send here
> to enable full testing on the bugfix branch: [1] (hope it will pass).
> The topmost commit should NOT be pushed to 2.1.
>
> [1]: https://travis-ci.org/tarantool/tarantool/builds/472881226
>
> cmake/compiler.cmake | 43 +++++++++++++++++++++++++++++++++++++++++++
> 1 file changed, 43 insertions(+)
>
> diff --git a/cmake/compiler.cmake b/cmake/compiler.cmake
> index 4062d13ec..b471b0c38 100644
> --- a/cmake/compiler.cmake
> +++ b/cmake/compiler.cmake
> @@ -333,6 +333,8 @@ if (NOT HAVE_BUILTIN_CTZ OR NOT HAVE_BUILTIN_CTZLL)
> HAVE_FFSL)
> check_c_source_compiles("#include <string.h>\n#include <strings.h>\nint main(void) { return ffsll(0UL); }"
> HAVE_FFSLL)
> + set(CMAKE_REQUIRED_FLAGS "")
> + set(CMAKE_REQUIRED_DEFINITIONS "")
> endif()
> endif()
>
> @@ -343,3 +345,44 @@ else()
> set(CMAKE_HOST_C_COMPILER ${CMAKE_C_COMPILER})
> set(CMAKE_HOST_CXX_COMPILER ${CMAKE_CXX_COMPILER})
> endif()
> +
> +# Check for broken __builtin_v?sn?printf() in gcc-7.2.
> +# https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83448
> +
> +# It is important for the test case to be compiled w/o any extra
> +# flags and with -O2. Say, -Wall allows the case be compiled
> +# successfully on gcc-7.2.1.
> +set(PREVIOUS_CMAKE_C_FLAGS ${CMAKE_C_FLAGS})
> +set(CMAKE_C_FLAGS "")
> +set(CMAKE_REQUIRED_DEFINITIONS "")
> +set(CMAKE_REQUIRED_INCLUDES "")
> +set(CMAKE_REQUIRED_LIBRARIES "")
> +set(CMAKE_REQUIRED_FLAGS "-O2")
> +
> +check_c_source_compiles("
> +char *a;
> +int b;
> +
> +int
> +main(void)
> +{
> + for (;;) {
> + if (b < 0)
> + main();
> + __builtin_snprintf(a, b, \"%*s\", b, \"\");
> + }
> + return 0;
> +}" HAVE_BUILTIN_SNPRINTF)
> +
> +set(CMAKE_REQUIRED_FLAGS "")
> +set(CMAKE_C_FLAGS ${PREVIOUS_CMAKE_C_FLAGS})
> +
> +# /usr/include/stdio.h (bits/stdio2.h to be exact) uses builtin
> +# gcc v?sn?printf functions when fortify source is enabled.
> +if (CMAKE_COMPILER_IS_GNUCC AND NOT HAVE_BUILTIN_SNPRINTF)
> + add_compile_flags("C;CXX" "-Wp,-U_FORTIFY_SOURCE")
> + add_compile_flags("C;CXX" "-fno-builtin-sprintf")
> + add_compile_flags("C;CXX" "-fno-builtin-snprintf")
> + add_compile_flags("C;CXX" "-fno-builtin-vsnprintf")
> + add_compile_flags("C;CXX" "-fno-builtin-vsprintf")
> +endif()
> --
> 2.20.1
>
^ permalink raw reply [flat|nested] 4+ messages in thread* Re: [PATCH] cmake: workaround GCC 7.2 segfault on v?sn?printf
2018-12-28 4:48 [PATCH] cmake: workaround GCC 7.2 segfault on v?sn?printf Alexander Turenko
2018-12-28 6:27 ` Alexander Turenko
@ 2018-12-28 8:24 ` Vladimir Davydov
2018-12-28 8:51 ` [tarantool-patches] " Kirill Yukhin
1 sibling, 1 reply; 4+ messages in thread
From: Vladimir Davydov @ 2018-12-28 8:24 UTC (permalink / raw)
To: Alexander Turenko; +Cc: tarantool-patches
On Fri, Dec 28, 2018 at 07:48:54AM +0300, Alexander Turenko wrote:
> GCC 7.2.1 fails like so on our codebase:
>
> /tarantool/src/lib/json/json.c: In function ‘json_token_snprint’:
> /tarantool/src/lib/json/json.c:324:1: internal compiler error: Segmentation fault
> json_token_snprint(char *buf, int size, const struct json_token *token,
> ^~~~~~~~~~~~~~~~~~
>
> The backtrace of cc1 is the following (failed to install compatible
> debuginfo on Fedora 26, so ??s):
>
> #0 0x0000000000d0320c in get_source_location_for_substring(cpp_reader*, string_concat_db*, unsigned int, cpp_ttype, int, int, int, unsigned int*) ()
> #1 0x00000000005c0e79 in c_get_substring_location(substring_loc const&, unsigned int*) ()
> #2 0x00000000007f8418 in format_warning_va(substring_loc const&, source_range const*, char const*, int, char const*, __va_list_tag (*) [1]) ()
> #3 0x00000000007f838b in format_warning_at_substring(substring_loc const&, source_range const*, char const*, int, char const*, ...) ()
> #4 0x0000000000c72c2d in ?? ()
> #5 0x0000000000f6f308 in execute_one_pass(opt_pass*) ()
> #6 0x0000000000787da6 in ?? ()
> #7 0x00000000012ac338 in symbol_table::compile() ()
> #8 0x0000000000de650f in symbol_table::finalize_compilation_unit() ()
> #9 0x00000000012ee8df in ?? ()
> #10 0x0000000000d3ba84 in toplev::main(int, char**) ()
> #11 0x0000000000d3d35b in main ()
>
> The issue in the GCC bugtracker: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83448
>
> The idea of the commit is to check builtin snprintf with the code from
> GCC test suite and disable all related builtins if the compiler is GCC
> and the case fails.
>
> We also need to disable FORTIFY_SOURCES in the case, because it uses
> v?sn?printf builtins directly.
Frankly, I don't think it's worth adding a workaround for a buggy
compiler version that's only used by distributions that passed EOL.
IMO we'd better disable compilation for those distributions.
^ permalink raw reply [flat|nested] 4+ messages in thread
* [tarantool-patches] Re: [PATCH] cmake: workaround GCC 7.2 segfault on v?sn?printf
2018-12-28 8:24 ` Vladimir Davydov
@ 2018-12-28 8:51 ` Kirill Yukhin
0 siblings, 0 replies; 4+ messages in thread
From: Kirill Yukhin @ 2018-12-28 8:51 UTC (permalink / raw)
To: tarantool-patches; +Cc: Alexander Turenko
On 28 Dec 11:24, Vladimir Davydov wrote:
> On Fri, Dec 28, 2018 at 07:48:54AM +0300, Alexander Turenko wrote:
> > GCC 7.2.1 fails like so on our codebase:
> >
> > /tarantool/src/lib/json/json.c: In function ‘json_token_snprint’:
> > /tarantool/src/lib/json/json.c:324:1: internal compiler error: Segmentation fault
> > json_token_snprint(char *buf, int size, const struct json_token *token,
> > ^~~~~~~~~~~~~~~~~~
> >
> > The backtrace of cc1 is the following (failed to install compatible
> > debuginfo on Fedora 26, so ??s):
> >
> > #0 0x0000000000d0320c in get_source_location_for_substring(cpp_reader*, string_concat_db*, unsigned int, cpp_ttype, int, int, int, unsigned int*) ()
> > #1 0x00000000005c0e79 in c_get_substring_location(substring_loc const&, unsigned int*) ()
> > #2 0x00000000007f8418 in format_warning_va(substring_loc const&, source_range const*, char const*, int, char const*, __va_list_tag (*) [1]) ()
> > #3 0x00000000007f838b in format_warning_at_substring(substring_loc const&, source_range const*, char const*, int, char const*, ...) ()
> > #4 0x0000000000c72c2d in ?? ()
> > #5 0x0000000000f6f308 in execute_one_pass(opt_pass*) ()
> > #6 0x0000000000787da6 in ?? ()
> > #7 0x00000000012ac338 in symbol_table::compile() ()
> > #8 0x0000000000de650f in symbol_table::finalize_compilation_unit() ()
> > #9 0x00000000012ee8df in ?? ()
> > #10 0x0000000000d3ba84 in toplev::main(int, char**) ()
> > #11 0x0000000000d3d35b in main ()
> >
> > The issue in the GCC bugtracker: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83448
> >
> > The idea of the commit is to check builtin snprintf with the code from
> > GCC test suite and disable all related builtins if the compiler is GCC
> > and the case fails.
> >
> > We also need to disable FORTIFY_SOURCES in the case, because it uses
> > v?sn?printf builtins directly.
>
> Frankly, I don't think it's worth adding a workaround for a buggy
> compiler version that's only used by distributions that passed EOL.
> IMO we'd better disable compilation for those distributions.
I agree. GCC is now 7.4 and problem was fixed in 7.3, so I'd put a note
into docs or even do nothing.
--
Regards, Kirill Yukhin
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2018-12-28 8:51 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2018-12-28 4:48 [PATCH] cmake: workaround GCC 7.2 segfault on v?sn?printf Alexander Turenko
2018-12-28 6:27 ` Alexander Turenko
2018-12-28 8:24 ` Vladimir Davydov
2018-12-28 8:51 ` [tarantool-patches] " Kirill Yukhin
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox