During bootstrap, stage1/cc1 segfaulted. (The bug seems triggered because I am trying to build a C compiler with -pg, gprof feature. Not many people seem to use gprof feature of the C compiler, I am afraid, and thus it probably is not exercised much.) I hope this gets fixed soon. ======================================== THE BUG: ---------------------------------------- stage1/xgcc -Bstage1/ -B/usr/local/i686-pc-linux-gnu/bin/ -c -march=athlon-xp -mtune=athlon-xp -g -pg -O2 -DIN_GCC -W -Wall -Wwrite-strings -Wstrict-prototypes -Wmissing-prototypes -pedantic -Wno-long-long -Werror -fno-common -DHAVE_CONFIG_H -I. -I. -I/u2/tools/gcc-mainline-cvs/gcc/gcc -I/u2/tools/gcc-mainline-cvs/gcc/gcc/. -I/u2/tools/gcc-mainline-cvs/gcc/gcc/../include /u2/tools/gcc-mainline-cvs/gcc/gcc/cppexp.c -o cppexp.o /u2/tools/gcc-mainline-cvs/gcc/gcc/cppexp.c: In function `num_trim': /u2/tools/gcc-mainline-cvs/gcc/gcc/cppexp.c:1014: internal compiler error: Segmentation fault Please submit a full bug report, with preprocessed source if appropriate. See <URL:http://gcc.gnu.org/bugs.html> for instructions. make[2]: *** [cppexp.o] Error 1 make[2]: Leaving directory `/u2/tools/gcc-mainline-obj-dir/gcc' make[1]: *** [stage2_build] Error 2 make[1]: Leaving directory `/u2/tools/gcc-mainline-obj-dir/gcc' make: *** [bootstrap] Error 2 ======================================== Quote from GDB session: ---------------------------------------- I checked the value of svrtx using gdb, and found that it is 0 (NULL) and GET_CODE(svrtx) is called. ishikawa@duron$ stage1/cc1 -quiet -v -I. -I. -I/u2/tools/gcc-mainline-cvs/gcc/gcc -I/u2/tools/gcc-mainline-cvs/gcc/gcc/. -I/u2/tools/gcc-mainline-cvs/gcc/gcc/../include -iprefix /u2/tools/gcc-mainline-obj-dir/gcc/stage1/../lib/gcc/i686-pc-linux-gnu/3.4/ -isystem include -isystem stage1/include -isystem /usr/local/i686-pc-linux-gnu/bin/include -DIN_GCC -DHAVE_CONFIG_H /u2/tools/gcc-mainline-cvs/gcc/gcc/cppexp.c -quiet -dumpbase cppexp.c -march=athlon-xp -mtune=athlon-xp -auxbase-strip cppexp.o -g -O2 -W -Wall -Wwrite-strings -Wstrict-prototypes -Wmissing-prototypes -pedantic -Wno-long-long -Werror -version -p -fno-common -o /tmp/ccTCMJqs.s ignoring nonexistent directory "stage1/include" ignoring nonexistent directory "/usr/local/i686-pc-linux-gnu/bin/include" ignoring nonexistent directory "/u2/tools/gcc-mainline-obj-dir/gcc/stage1/../lib/gcc/i686-pc-linux-gnu/3.4/include" ignoring nonexistent directory "/u2/tools/gcc-mainline-obj-dir/gcc/stage1/../lib/gcc/i686-pc-linux-gnu/3.4/../../../../i686-pc-linux-gnu/include" ignoring nonexistent directory "NONE/include" ignoring nonexistent directory "/usr/local/lib/gcc/i686-pc-linux-gnu/3.4/include" ignoring nonexistent directory "/usr/local/lib/../i686-pc-linux-gnu/include" ignoring duplicate directory "." ignoring duplicate directory "/u2/tools/gcc-mainline-cvs/gcc/gcc/." #include "..." search starts here: #include <...> search starts here: . /u2/tools/gcc-mainline-cvs/gcc/gcc /u2/tools/gcc-mainline-cvs/gcc/gcc/../include include /usr/local/include /usr/include End of search list. GNU C version 3.4 20030916 (experimental) (i686-pc-linux-gnu) compiled by GNU C version 3.3. GGC heuristics: --param ggc-min-expand=30 --param ggc-min-heapsize=4096 /u2/tools/gcc-mainline-cvs/gcc/gcc/cppexp.c: In function `num_trim': /u2/tools/gcc-mainline-cvs/gcc/gcc/cppexp.c:1014: internal compiler error: Segmentation fault Please submit a full bug report, with preprocessed source if appropriate. See <URL:http://gcc.gnu.org/bugs.html> for instructions. ishikawa@duron$ gdb stage1/cc1 GNU gdb 5.3-debian Copyright 2002 Free Software Foundation, Inc. GDB is free software, covered by the GNU General Public License, and you are welcome to change it and/or distribute copies of it under certain conditions. Type "show copying" to see the conditions. There is absolutely no warranty for GDB. Type "show warranty" for details. This GDB was configured as "i386-linux"... Breakpoint 1 at 0x814f12c: file /u2/tools/gcc-mainline-cvs/gcc/gcc/diagnostic.c, line 593. Breakpoint 2 at 0x8049804 Breakpoint 3 at 0x80495a4 (gdb) run -quiet -v -I. -I. -I/u2/tools/gcc-mainline-cvs/gcc/gcc -I/u2/tools/gcc-mainline-cvs/gcc/gcc/. -I/u2/tools/gcc-mainline-cvs/gcc/gcc/../include -iprefix /u2/tools/gcc-mainline-obj-dir/gcc/stage1/../lib/gcc/i686-pc-linux-gnu/3.4/ -isystem include -isystem stage1/include -isystem /usr/local/i686-pc-linux-gnu/bin/include -DIN_GCC -DHAVE_CONFIG_H /u2/tools/gcc-mainline-cvs/gcc/gcc/cppexp.c -quiet -dumpbase cppexp.c -march=athlon-xp -mtune=athlon-xp -auxbase-strip cppexp.o -g -O2 -W -Wall -Wwrite-strings -Wstrict-prototypes -Wmissing-prototypes -pedantic -Wno-long-long -Werror -version -p -fno-common -o /tmp/ccTCMJqs.s Starting program: /u2/tools/gcc-mainline-obj-dir/gcc/stage1/cc1 -quiet -v -I. -I. -I/u2/tools/gcc-mainline-cvs/gcc/gcc -I/u2/tools/gcc-mainline-cvs/gcc/gcc/. -I/u2/tools/gcc-mainline-cvs/gcc/gcc/../include -iprefix /u2/tools/gcc-mainline-obj-dir/gcc/stage1/../lib/gcc/i686-pc-linux-gnu/3.4/ -isystem include -isystem stage1/include -isystem /usr/local/i686-pc-linux-gnu/bin/include -DIN_GCC -DHAVE_CONFIG_H /u2/tools/gcc-mainline-cvs/gcc/gcc/cppexp.c -quiet -dumpbase cppexp.c -march=athlon-xp -mtune=athlon-xp -auxbase-strip cppexp.o -g -O2 -W -Wall -Wwrite-strings -Wstrict-prototypes -Wmissing-prototypes -pedantic -Wno-long-long -Werror -version -p -fno-common -o /tmp/ccTCMJqs.s Breakpoint 2 at 0x4004b81e Breakpoint 3 at 0x4004a8c0 ignoring nonexistent directory "stage1/include" ignoring nonexistent directory "/usr/local/i686-pc-linux-gnu/bin/include" ignoring nonexistent directory "/u2/tools/gcc-mainline-obj-dir/gcc/stage1/../lib/gcc/i686-pc-linux-gnu/3.4/include" ignoring nonexistent directory "/u2/tools/gcc-mainline-obj-dir/gcc/stage1/../lib/gcc/i686-pc-linux-gnu/3.4/../../../../i686-pc-linux-gnu/include" ignoring nonexistent directory "NONE/include" ignoring nonexistent directory "/usr/local/lib/gcc/i686-pc-linux-gnu/3.4/include" ignoring nonexistent directory "/usr/local/lib/../i686-pc-linux-gnu/include" ignoring duplicate directory "." ignoring duplicate directory "/u2/tools/gcc-mainline-cvs/gcc/gcc/." #include "..." search starts here: #include <...> search starts here: . /u2/tools/gcc-mainline-cvs/gcc/gcc /u2/tools/gcc-mainline-cvs/gcc/gcc/../include include /usr/local/include /usr/include End of search list. GNU C version 3.4 20030916 (experimental) (i686-pc-linux-gnu) compiled by GNU C version 3.3. GGC heuristics: --param ggc-min-expand=30 --param ggc-min-heapsize=4096 Program received signal SIGSEGV, Segmentation fault. 0x081c67b5 in profile_function (file=0x86660a0) at /u2/tools/gcc-mainline-cvs/gcc/gcc/final.c:1440 1440 if (sval && GET_CODE (svrtx) == REG) (gdb) list 1435 } 1436 1437 function_section (current_function_decl); 1438 1439 #if defined(ASM_OUTPUT_REG_PUSH) 1440 if (sval && GET_CODE (svrtx) == REG) 1441 ASM_OUTPUT_REG_PUSH (file, REGNO (svrtx)); 1442 #endif 1443 1444 #if defined(STATIC_CHAIN_INCOMING_REGNUM) && defined(ASM_OUTPUT_REG_PUSH) (gdb) print sval $1 = 1 (gdb) print svrtx $2 = 0x0 (gdb) quit A debugging session is active. Do you still want to close the debugger?(y or n) y ishikawa@duron$ cf. Quote from "pr -n cppexp.c": 995 /* Clears the unused high order bits of the number pointed to by PNUM. */ 996 static cpp_num 997 num_trim (cpp_num num, size_t precision) 998 { 999 if (precision > PART_PRECISION) 1000 { 1001 precision -= PART_PRECISION; 1002 if (precision < PART_PRECISION) 1003 num.high &= ((cpp_num_part) 1 << precision) - 1; 1004 } 1005 else 1006 { 1007 if (precision < PART_PRECISION) 1008 num.low &= ((cpp_num_part) 1 << precision) - 1; 1009 num.high = 0; 1010 } 1011 1012 return num; 1013 } 1014 This is how I configured GCC for bootstrap. (I updated my local source CVS tree a few hours ago.) : # set -vx LANG=C LC_ALL=C export LANG export LC_ALL make distclean clean rm config.cache # #--enable-languages=c # # Use absolute pathnaes # env CC=gcc-cvs # env CC=gcc /u2/tools/gcc-mainline-cvs/gcc/configure \ --enable-languages=c,c++,objc \ --program-suffix=-mainline-cvs --enable-nls # #env CC=gcc-cvs make BOOT_CFLAGS="-g -pg -O2 -ftest-coverage -fprofile-arcs" bootstrap # doesn't work # env CC="gcc" make BOOT_CFLAGS="-march=athlon-xp -mtune=athlon-xp -g -pg -O2 " bootstrap # bootstrap-lean
I recall that the bootstrap compilation with the exactly same configuration aroudn August 22th. So whatever changed in between today and that date would be the natural candidates for the cause of the bug.
The bootstrap succeeded WITHOUT "-pg". So the bug is in the profile-related code.
profile_function is in the profiling code so it this definitely a regression but I have not reduced it yet.
Hi, The way I see it, the bug seems triggered when a function that returns a struct value is profiled. (I think that either the register that holds the hidden pointer to the area reserved for the return value [ passed from the caller], or a register that holds the small struct value that fits in a register [to be returned thus in a register] is forced out of the way for profiling code when this ASM_OUTPUT_REG_PUSH is called. However, I wonder what happens if ASM_OUTPUT_REG_PUSH is NOT defined. What happens on such architecture WITHOUT ASM_OUTPUT_REG_PUSH if we specify profiling?) The above observation might be obvious to you already, though. BTW, a simple file below, which includes functions that return struct values, causes stage1/cc1 to segfault in a very similar manner as observed during bootstrap. Looking at the generated asm output (by not adding -pg ) of this test file might reveal some insight. Test code: gcc-problem.c Attached. segfault example: stage1/cc1 -quiet -v -D__GNUC__=3 -D__GNUC_MINOR__=4 -D__GNUC_PATCHLEVEL__=0 /home/ishikawa/BUG-DIR/GCC-BUG/gcc-problem.c -quiet -dumpbase gcc-problem.c -auxbase gcc-problem -version -p -o /tmp/ccrvPYue.s ignoring nonexistent directory "NONE/include" ignoring nonexistent directory "/usr/local/lib/gcc/i686-pc-linux-gnu/3.4/include" ignoring nonexistent directory "/usr/local/lib/../i686-pc-linux-gnu/include" #include "..." search starts here: #include <...> search starts here: /usr/local/include /usr/include End of search list. GNU C version 3.4 20030919 (experimental) (i686-pc-linux-gnu) compiled by GNU C version 3.3. GGC heuristics: --param ggc-min-expand=30 --param ggc-min-heapsize=4096 /home/ishikawa/BUG-DIR/GCC-BUG/gcc-problem.c: In function `testfuncwordsimple2': /home/ishikawa/BUG-DIR/GCC-BUG/gcc-problem.c:24: internal compiler error: Segmentation fault Please submit a full bug report, with preprocessed source if appropriate. See <URL:http://gcc.gnu.org/bugs.html> for instructions. ishikawa@duron$ ishikawa@duron$ gdb stage1/cc1 GNU gdb 5.3-debian Copyright 2002 Free Software Foundation, Inc. GDB is free software, covered by the GNU General Public License, and you are welcome to change it and/or distribute copies of it under certain conditions. Type "show copying" to see the conditions. There is absolutely no warranty for GDB. Type "show warranty" for details. This GDB was configured as "i386-linux"... Breakpoint 1 at 0x814f5c4: file /u2/tools/gcc-mainline-cvs/gcc/gcc/diagnostic.c, line 593. Breakpoint 2 at 0x8049804 Breakpoint 3 at 0x80495a4 (gdb) run -quiet -v -D__GNUC__=3 -D__GNUC_MINOR__=4 -D__GNUC_PATCHLEVEL__=0 /home/ishikawa/BUG-DIR/GCC-BUG/gcc-problem.c -quiet -dumpbase gcc-problem.c -auxbase gcc-problem -version -p -o /tmp/ccrvPYue.s Starting program: /u2/tools/gcc-mainline-obj-dir/gcc/stage1/cc1 -quiet -v -D__GNUC__=3 -D__GNUC_MINOR__=4 -D__GNUC_PATCHLEVEL__=0 /home/ishikawa/BUG-DIR/GCC-BUG/gcc-problem.c -quiet -dumpbase gcc-problem.c -auxbase gcc-problem -version -p -o /tmp/ccrvPYue.s Breakpoint 2 at 0x4004b81e Breakpoint 3 at 0x4004a8c0 ignoring nonexistent directory "NONE/include" ignoring nonexistent directory "/usr/local/lib/gcc/i686-pc-linux-gnu/3.4/include" ignoring nonexistent directory "/usr/local/lib/../i686-pc-linux-gnu/include" #include "..." search starts here: #include <...> search starts here: /usr/local/include /usr/include End of search list. GNU C version 3.4 20030919 (experimental) (i686-pc-linux-gnu) compiled by GNU C version 3.3. GGC heuristics: --param ggc-min-expand=30 --param ggc-min-heapsize=4096 Program received signal SIGSEGV, Segmentation fault. 0x081c6675 in profile_function (file=0x86592f0) at /u2/tools/gcc-mainline-cvs/gcc/gcc/final.c:1440 1440 if (sval && GET_CODE (svrtx) == REG) (gdb) list 1435 } 1436 1437 function_section (current_function_decl); 1438 1439 #if defined(ASM_OUTPUT_REG_PUSH) 1440 if (sval && GET_CODE (svrtx) == REG) 1441 ASM_OUTPUT_REG_PUSH (file, REGNO (svrtx)); 1442 #endif 1443 1444 #if defined(STATIC_CHAIN_INCOMING_REGNUM) && defined(ASM_OUTPUT_REG_PUSH) (gdb) print sval $1 = 1 (gdb) print svrtx $2 = 0x0 (gdb) print current_function_decl $3 = 0x40198dec (gdb) print file $4 = (FILE *) 0x86592f0 (gdb) quit A debu The file gcc-problem.c is attached as the next post.
Created attachment 4789 [details] test case of functions that return struct values: cc1 segfaults with -p This is a short test case file that contains functions that return struct values. They cause stage1/cc1 to segfault if we specify profiling option.
*** Bug 12421 has been marked as a duplicate of this bug. ***
I can confirm this on the mainline (20030926).
*** Bug 12748 has been marked as a duplicate of this bug. ***
The same happens with C++ (see e.g. PR 12748). A minimal testcase to illustrate this: ============================================ struct A {}; A foo() { return A(); } ============================================ Just compile with g++ -p to get a segfault.
Btw, according to Phil's regression hunter, the regression was introduced between 2003-09-03-trunk (#390) and 2003-09-04-trunk (#391).
I'm afraid DJ's (really big) patch http://gcc.gnu.org/ml/gcc-cvs/2003-09/msg00074.html seems to be responsible for the regression. DJ, could you please have a look?
Here for reference the patch as sent to gcc-patches: <http://gcc.gnu.org/ml/gcc-patches/2003 -09/msg00184.html>.
*** Bug 12809 has been marked as a duplicate of this bug. ***
Subject: Re: [3.4 Regression] ICE using -p with functions returning structs > I'm afraid DJ's (really big) patch > > http://gcc.gnu.org/ml/gcc-cvs/2003-09/msg00074.html > > seems to be responsible for the regression. > > DJ, could you please have a look? Try this. 2003-10-28 DJ Delorie <dj@redhat.com> * final.c (profile_function): Handle the case where svrtx is NULL. Index: final.c =================================================================== RCS file: /cvs/gcc/gcc/gcc/final.c,v retrieving revision 1.293 diff -p -2 -r1.293 final.c *** final.c 18 Oct 2003 19:13:04 -0000 1.293 --- final.c 28 Oct 2003 22:41:18 -0000 *************** profile_function (FILE *file ATTRIBUTE_U *** 1438,1442 **** #if defined(ASM_OUTPUT_REG_PUSH) ! if (sval && GET_CODE (svrtx) == REG) ASM_OUTPUT_REG_PUSH (file, REGNO (svrtx)); #endif --- 1438,1442 ---- #if defined(ASM_OUTPUT_REG_PUSH) ! if (sval && svrtx && GET_CODE (svrtx) == REG) ASM_OUTPUT_REG_PUSH (file, REGNO (svrtx)); #endif *************** profile_function (FILE *file ATTRIBUTE_U *** 1469,1473 **** #if defined(ASM_OUTPUT_REG_PUSH) ! if (sval && GET_CODE (svrtx) == REG) ASM_OUTPUT_REG_POP (file, REGNO (svrtx)); #endif --- 1469,1473 ---- #if defined(ASM_OUTPUT_REG_PUSH) ! if (sval && svrtx && GET_CODE (svrtx) == REG) ASM_OUTPUT_REG_POP (file, REGNO (svrtx)); #endif
There were two patches posted to gcc-patches today, both were the same: <http://gcc.gnu.org/ ml/gcc-patches/2003-10/msg02500.html> and <http://gcc.gnu.org/ml/gcc-patches/2003-10/ msg02503.html>.
*** Bug 12836 has been marked as a duplicate of this bug. ***
DJ, your patch is known to work, see Alan Modra's patch which really the same patch.
Subject: Bug 12315 CVSROOT: /cvs/gcc Module name: gcc Changes by: amodra@gcc.gnu.org 2003-11-01 00:48:51 Modified files: gcc : ChangeLog final.c Log message: PR 12315 * final.c (profile_function): Allow for NULL svrtx. Patches: http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&r1=2.1630&r2=2.1631 http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/final.c.diff?cvsroot=gcc&r1=1.293&r2=1.294
fixed