On 8/26/24 10:31, Richard Biener wrote:
> On Mon, 26 Aug 2024, Bernd Edlinger wrote:
>
>> This recent change triggered various bootsteap-errors, mostly on
>> x86 targets because line info advance address entries were output
>> in the wrong section table.
>> The switch to the wrong line table happened in dwarfout_set_ignored_loc.
>> It must use the same section as the earlier called
>> dwarf2out_switch_text_section.
>>
>> But also ft32-elf was affected, because the assembler choked on
>> something simple as ".2byte .LM2-.LM1", but fortunately it is
>> able to use native location views, the configure test was just
>> not executed because the ft32 "nop" instruction was missing.
>
> OK for the configure part, I don't understand how using
> current_function_section is correct or how it even makes a
> differnce to function_section.
>
> It seems both would rely on the fact that fde->decl should be
> the same as cfun->decl and both eventually resort to how
> first_function_block_is_cold is set.
>
> Is this from final_scan_insn_1 where we seem to switch
> in_cold_section_p?
>
> The [current_]function_section API might be just confusing to me
> of course. I note that dwarf2out mixes both uses and
> current_function_section seems newer than function_section. Huh.
>
Well, this is how I debugged it:
I use the successfully bootstrapped x86_64-pc-linux-gnu-gcc as host compiler
and build this:
../gcc-trunk/configure --target=i386-linux-gnu CC="gcc -m32
-gno-as-loc-support" CXX="g++ -m32 -gno-as-loc-support"
make stops here:
g++ -m32 -gno-as-loc-support -fno-PIE -c -g -O2 -DIN_GCC
-DCROSS_DIRECTORY_STRUCTURE -fno-exceptions -fno-rtti
-fasynchronous-unwind-tables -W -Wall -Wno-narrowing -Wwrite-strings
-Wcast-qual -Wmissing-format-attribute -Wconditionally-supported
-Woverloaded-virtual -pedantic -Wno-long-long -Wno-variadic-macros
-Wno-overlength-strings -DHAVE_CONFIG_H -fno-PIE -I. -I. -I../../gcc-trunk/gcc
-I../../gcc-trunk/gcc/. -I../../gcc-trunk/gcc/../include
-I../../gcc-trunk/gcc/../libcpp/include -I../../gcc-trunk/gcc/../libcody
-I/home/ed/gnu/gcc-build-x/./gmp -I/home/ed/gnu/gcc-trunk/gmp
-I/home/ed/gnu/gcc-build-x/./mpfr/src -I/home/ed/gnu/gcc-trunk/mpfr/src
-I/home/ed/gnu/gcc-trunk/mpc/src -I../../gcc-trunk/gcc/../libdecnumber
-I../../gcc-trunk/gcc/../libdecnumber/bid -I../libdecnumber
-I../../gcc-trunk/gcc/../libbacktrace -I/home/ed/gnu/gcc-build-x/./isl/include
-I/home/ed/gnu/gcc-trunk/isl/include -o gtype-desc.o -MT gtype-desc.o -MMD -MP
-MF ./.deps/gtype-desc.TPo gtype-desc.cc
/tmp/ccB94xhL.s: Assembler messages:
/tmp/ccB94xhL.s:563836: Error: can't resolve .text.unlikely - .LM4229
/tmp/ccB94xhL.s:563841: Error: can't resolve .text - .LM4230
/tmp/ccB94xhL.s:564103: Error: can't resolve .text.unlikely - .LM4282
/tmp/ccB94xhL.s:564108: Error: can't resolve .text - .LM4283
/tmp/ccB94xhL.s:564115: Error: can't resolve .text.unlikely - .LM4284
make[2]: *** [Makefile:1194: gtype-desc.o] Error 1
I took the original g++ command, and replace "-c" with "-S" and
"-o gtype-desc.o" with "-o /proc/self/fd/1" and add "-wrapper gdb,--args":
$ g++ -m32 -gno-as-loc-support -fno-PIE -g -O2 -DIN_GCC
-DCROSS_DIRECTORY_STRUCTURE -fno-exceptions -fno-rtti
-fasynchronous-unwind-tables -W -Wall -Wno-narrowing -Wwrite-strings
-Wcast-qual -Wmissing-format-attribute -Wconditionally-supported
-Woverloaded-virtual -pedantic -Wno-long-long -Wno-variadic-macros
-Wno-overlength-strings -DHAVE_CONFIG_H -fno-PIE -I. -I. -I../../gcc-trunk/gcc
-I../../gcc-trunk/gcc/. -I../../gcc-trunk/gcc/../include
-I../../gcc-trunk/gcc/../libcpp/include -I../../gcc-trunk/gcc/../libcody
-I/home/ed/gnu/gcc-build-x/./gmp -I/home/ed/gnu/gcc-trunk/gmp
-I/home/ed/gnu/gcc-build-x/./mpfr/src -I/home/ed/gnu/gcc-trunk/mpfr/src
-I/home/ed/gnu/gcc-trunk/mpc/src -I../../gcc-trunk/gcc/../libdecnumber
-I../../gcc-trunk/gcc/../libdecnumber/bid -I../libdecnumber
-I../../gcc-trunk/gcc/../libbacktrace -I/home/ed/gnu/gcc-build-x/./isl/include
-I/home/ed/gnu/gcc-trunk/isl/include -MT gtype-desc.o -MMD -MP -MF
./.deps/gtype-desc.TPo gtype-desc.cc -S -o /proc/self/fd/1 -wrapper gdb,--args
(gdb) b dwarf2out_set_ignored_loc
(gdb) display in_cold_section_p
(gdb) display first_function_block_is_cold
(gdb) r
The first breakpoint is uninteresting:
[...]
.size _Z22gt_pch_p_11eh_region_dPvS_PFvS_S_S_ES_,
.-_Z22gt_pch_p_11eh_region_dPvS_PFvS_S_S_ES_
.p2align 4
.globl _Z21gt_pch_p_10eh_catch_dPvS_PFvS_S_S_ES_
.type _Z21gt_pch_p_10eh_catch_dPvS_PFvS_S_S_ES_, @function
_Z21gt_pch_p_10eh_catch_dPvS_PFvS_S_S_ES_:
.LFB8506:
.cfi_startproc
Breakpoint 1, dwarf2out_set_ignored_loc (line=12009, column=1,
filename=0x391bd90 "gtype-desc.cc")
at ../../gcc-trunk/gcc/dwarf2out.cc:28976
28976 dw_fde_ref fde = cfun->fde;
1: in_cold_section_p = false
2: first_function_block_is_cold = false
(gdb)
The next breakpoint is where the problem starts:
(gdb) c
[...]
.size
_Z27hashtab_entry_note_pointersIN8hash_mapIP6gimplei21simple_hashmap_traitsI19default_hash_traitsIS2_EiEE10hash_entryEEvPvS9_PFvS9_S9_S9_ES9_,
.-_Z27hashtab_entry_note_pointersIN8hash_mapIP6gimplei21simple_hashmap_traitsI19default_hash_traitsIS2_EiEE10hash_entryEEvPvS9_PFvS9_S9_S9_ES9_
.section .text.unlikely
.LCOLDB30:
.text
.LHOTB30:
.p2align 4
.globl
_Z49gt_pch_p_38function_summary_ipcp_transformation__PvS_PFvS_S_S_ES_
.type
_Z49gt_pch_p_38function_summary_ipcp_transformation__PvS_PFvS_S_S_ES_, @function
_Z49gt_pch_p_38function_summary_ipcp_transformation__PvS_PFvS_S_S_ES_:
.LFB8556:
.cfi_startproc
Breakpoint 1, dwarf2out_set_ignored_loc (line=12341, column=1,
filename=0x391bd90 "gtype-desc.cc")
at ../../gcc-trunk/gcc/dwarf2out.cc:28976
28976 dw_fde_ref fde = cfun->fde;
1: in_cold_section_p = false
2: first_function_block_is_cold = false
(gdb) c
Continuing.
.LM4229:
subl $12, %esp
.cfi_def_cfa_offset 16
movl 20(%esp), %eax
cmpl %eax, 16(%esp)
je .L1179
addl $12, %esp
.cfi_def_cfa_offset 4
ret
.cfi_endproc
.section .text.unlikely
.cfi_startproc
Breakpoint 1, dwarf2out_set_ignored_loc (line=12341, column=1,
filename=0x391bd90 "gtype-desc.cc")
at ../../gcc-trunk/gcc/dwarf2out.cc:28976
28976 dw_fde_ref fde = cfun->fde;
1: in_cold_section_p = true
2: first_function_block_is_cold = false
(gdb)
Remember, the .LM4229 was the first one complained about,
The nice thing, is I can repeat that to the point immediately before the error:
(gdb) r
[...]
(gdb) c
[...]
.size
_Z27hashtab_entry_note_pointersIN8hash_mapIP6gimplei21simple_hashmap_traitsI19default_hash_traitsIS2_EiEE10hash_entryEEvPvS9_PFvS9_S9_S9_ES9_,
.-_Z27hashtab_entry_note_pointersIN8hash_mapIP6gimplei21simple_hashmap_traitsI19default_hash_traitsIS2_EiEE10hash_entryEEvPvS9_PFvS9_S9_S9_ES9_
.section .text.unlikely
.LCOLDB30:
.text
.LHOTB30:
.p2align 4
.globl
_Z49gt_pch_p_38function_summary_ipcp_transformation__PvS_PFvS_S_S_ES_
.type
_Z49gt_pch_p_38function_summary_ipcp_transformation__PvS_PFvS_S_S_ES_, @function
_Z49gt_pch_p_38function_summary_ipcp_transformation__PvS_PFvS_S_S_ES_:
.LFB8556:
.cfi_startproc
Breakpoint 1, dwarf2out_set_ignored_loc (line=12341, column=1,
filename=0x391bd90 "gtype-desc.cc")
at ../../gcc-trunk/gcc/dwarf2out.cc:28976
28976 dw_fde_ref fde = cfun->fde;
28976 dw_fde_ref fde = cfun->fde;
1: in_cold_section_p = false
2: first_function_block_is_cold = false
(gdb) n
28978 fde->ignored_debug = false;
1: in_cold_section_p = false
2: first_function_block_is_cold = false
(gdb) n
28979 set_cur_line_info_table (function_section (fde->decl));
1: in_cold_section_p = false
2: first_function_block_is_cold = false
(gdb) n
28981 dwarf2out_source_line (line, column, filename, 0, true);
1: in_cold_section_p = false
2: first_function_block_is_cold = false
(gdb) b set_cur_line_info_table
Breakpoint 2 at 0xf25700: file ../../gcc-trunk/gcc/dwarf2out.cc, line 28594.
(gdb) c
Continuing.
.LM4229:
subl $12, %esp
.cfi_def_cfa_offset 16
movl 20(%esp), %eax
cmpl %eax, 16(%esp)
je .L1179
addl $12, %esp
.cfi_def_cfa_offset 4
ret
.cfi_endproc
.section .text.unlikely
.cfi_startproc
Breakpoint 2, set_cur_line_info_table (sec=sec@entry=0x7ffff3e643a0) at
../../gcc-trunk/gcc/dwarf2out.cc:28594
28594 if (sec == text_section)
1: in_cold_section_p = true
2: first_function_block_is_cold = false
Here, you see in_cold_section_p is truth.
(gdb) bt
#0 set_cur_line_info_table (sec=sec@entry=0x7ffff3e643a0) at
../../gcc-trunk/gcc/dwarf2out.cc:28594
#1 0x0000000000f3fa3d in dwarf2out_switch_text_section () at
../../gcc-trunk/gcc/dwarf2out.cc:1339
#2 0x0000000000fd9297 in final_scan_insn_1 (insn=0x7fffed8a1e70,
file=0x38ee590, seen=0x7fffffffd06c, nopeepholes=0,
optimize_p=2) at ../../gcc-trunk/gcc/final.cc:2182
#3 0x0000000000fda26c in final_scan_insn (insn=insn@entry=0x7fffed8a1e70,
file=file@entry=0x38ee590,
optimize_p=optimize_p@entry=2, nopeepholes=nopeepholes@entry=0,
seen=<optimized out>, seen@entry=0x7fffffffd06c)
at ../../gcc-trunk/gcc/final.cc:2886
#4 0x0000000000fda493 in final_1 (first=0x7fffed8a18c0, file=0x38ee590,
seen=<optimized out>, optimize_p=2)
at ../../gcc-trunk/gcc/final.cc:1977
#5 0x0000000000fdaf80 in rest_of_handle_final () at
../../gcc-trunk/gcc/final.cc:4239
#6 (anonymous namespace)::pass_final::execute (this=<optimized out>) at
../../gcc-trunk/gcc/final.cc:4317
#7 0x00000000012cef3f in execute_one_pass (pass=pass@entry=0x38e95c0) at
../../gcc-trunk/gcc/passes.cc:2647
#8 0x00000000012cf870 in execute_pass_list_1 (pass=0x38e95c0) at
../../gcc-trunk/gcc/passes.cc:2756
#9 0x00000000012cf882 in execute_pass_list_1 (pass=0x38e9010) at
../../gcc-trunk/gcc/passes.cc:2757
#10 0x00000000012cf882 in execute_pass_list_1 (pass=0x38e7130) at
../../gcc-trunk/gcc/passes.cc:2757
#11 0x00000000012cf8a9 in execute_pass_list (fn=0x7fffed8b4b40, pass=<optimized
out>)
at ../../gcc-trunk/gcc/passes.cc:2767
#12 0x0000000000ecb9d7 in cgraph_node::expand (this=0x7fffee381ee0) at
../../gcc-trunk/gcc/cgraphunit.cc:1845
#13 cgraph_node::expand (this=0x7fffee381ee0) at
../../gcc-trunk/gcc/cgraphunit.cc:1798
#14 0x0000000000ecd7fb in expand_all_functions () at
../../gcc-trunk/gcc/cgraphunit.cc:2028
#15 symbol_table::compile (this=0x7ffff77e5000) at
../../gcc-trunk/gcc/cgraphunit.cc:2404
#16 0x0000000000ecfeb8 in symbol_table::compile (this=0x7ffff77e5000) at
../../gcc-trunk/gcc/cgraphunit.cc:2315
#17 symbol_table::finalize_compilation_unit (this=0x7ffff77e5000) at
../../gcc-trunk/gcc/cgraphunit.cc:2589
#18 0x00000000013fe602 in compile_file () at ../../gcc-trunk/gcc/toplev.cc:478
#19 0x0000000000a9bf90 in do_compile () at ../../gcc-trunk/gcc/toplev.cc:2209
#20 toplev::main (this=this@entry=0x7fffffffd34e, argc=<optimized out>,
argc@entry=88, argv=<optimized out>,
argv@entry=0x7fffffffd478) at ../../gcc-trunk/gcc/toplev.cc:2369
#21 0x0000000000a9d7ee in main (argc=88, argv=0x7fffffffd478) at
../../gcc-trunk/gcc/main.cc:39
(gdb) c
Continuing.
Breakpoint 1, dwarf2out_set_ignored_loc (line=12341, column=1,
filename=0x391bd90 "gtype-desc.cc")
at ../../gcc-trunk/gcc/dwarf2out.cc:28976
28976 dw_fde_ref fde = cfun->fde;
1: in_cold_section_p = true
2: first_function_block_is_cold = false
(gdb) bt
#0 dwarf2out_set_ignored_loc (line=12341, column=1, filename=0x391bd90
"gtype-desc.cc")
at ../../gcc-trunk/gcc/dwarf2out.cc:28976
#1 0x0000000000fd9f30 in final_scan_insn_1 (insn=0x7fffed8a1e70,
file=0x38ee590, seen=0x7fffffffd06c, nopeepholes=0,
optimize_p=2) at ../../gcc-trunk/gcc/final.cc:2191
#2 0x0000000000fda26c in final_scan_insn (insn=insn@entry=0x7fffed8a1e70,
file=file@entry=0x38ee590,
optimize_p=optimize_p@entry=2, nopeepholes=nopeepholes@entry=0,
seen=<optimized out>, seen@entry=0x7fffffffd06c)
at ../../gcc-trunk/gcc/final.cc:2886
#3 0x0000000000fda493 in final_1 (first=0x7fffed8a18c0, file=0x38ee590,
seen=<optimized out>, optimize_p=2)
at ../../gcc-trunk/gcc/final.cc:1977
#4 0x0000000000fdaf80 in rest_of_handle_final () at
../../gcc-trunk/gcc/final.cc:4239
#5 (anonymous namespace)::pass_final::execute (this=<optimized out>) at
../../gcc-trunk/gcc/final.cc:4317
#6 0x00000000012cef3f in execute_one_pass (pass=pass@entry=0x38e95c0) at
../../gcc-trunk/gcc/passes.cc:2647
#7 0x00000000012cf870 in execute_pass_list_1 (pass=0x38e95c0) at
../../gcc-trunk/gcc/passes.cc:2756
#8 0x00000000012cf882 in execute_pass_list_1 (pass=0x38e9010) at
../../gcc-trunk/gcc/passes.cc:2757
#9 0x00000000012cf882 in execute_pass_list_1 (pass=0x38e7130) at
../../gcc-trunk/gcc/passes.cc:2757
#10 0x00000000012cf8a9 in execute_pass_list (fn=0x7fffed8b4b40, pass=<optimized
out>)
at ../../gcc-trunk/gcc/passes.cc:2767
#11 0x0000000000ecb9d7 in cgraph_node::expand (this=0x7fffee381ee0) at
../../gcc-trunk/gcc/cgraphunit.cc:1845
#12 cgraph_node::expand (this=0x7fffee381ee0) at
../../gcc-trunk/gcc/cgraphunit.cc:1798
#13 0x0000000000ecd7fb in expand_all_functions () at
../../gcc-trunk/gcc/cgraphunit.cc:2028
#14 symbol_table::compile (this=0x7ffff77e5000) at
../../gcc-trunk/gcc/cgraphunit.cc:2404
#15 0x0000000000ecfeb8 in symbol_table::compile (this=0x7ffff77e5000) at
../../gcc-trunk/gcc/cgraphunit.cc:2315
#16 symbol_table::finalize_compilation_unit (this=0x7ffff77e5000) at
../../gcc-trunk/gcc/cgraphunit.cc:2589
#17 0x00000000013fe602 in compile_file () at ../../gcc-trunk/gcc/toplev.cc:478
#18 0x0000000000a9bf90 in do_compile () at ../../gcc-trunk/gcc/toplev.cc:2209
#19 toplev::main (this=this@entry=0x7fffffffd34e, argc=<optimized out>,
argc@entry=88, argv=<optimized out>,
argv@entry=0x7fffffffd478) at ../../gcc-trunk/gcc/toplev.cc:2369
#20 0x0000000000a9d7ee in main (argc=88, argv=0x7fffffffd478) at
../../gcc-trunk/gcc/main.cc:39
now since dwarf2out_set_ignored_loc() uses function_section()
and function section() uses first_function_block_is_cold which is false
this is now the wrong table, but there is no output of any assembler
code in between those two calls of set_cur_line_info_table()
however the function dwarf2out_switch_text_section() uses
current_function_section () which uses in_cold_section_p, and only here
they disagree, but I think current_function_section() is more appropriate for
determining the section where the code is currently being output to,
while function_section() is telling you in which section the function entry
point will be output to. Since lables are usually outout together with code,
the change I've propsed seems right.
WDYT?
Thanks
Bernd.
