readelf: allow build with LLVM/clang
Commit Message
Hi Tomasz,
On Mon, Jul 12, 2021 at 11:59:41AM +0200, Tomasz Pawe? Gajc via Elfutils-devel wrote:
> convert last nested function to allow build with
> LLVM/clang. Original patch comes from
> https://github.com/OpenMandrivaAssociation/elfutils/blob/master/elfutils-0.185-clang.patch
> Tested with OpenMandriva Lx cooker and LLVM/clang-12.0.1
> [...]
> --- a/configure.ac
> +++ b/configure.ac
> @@ -127,7 +127,7 @@ void baz (int n)
> ac_cv_c99=yes, ac_cv_c99=no)
> CFLAGS="$old_CFLAGS"])
> AS_IF([test "x$ac_cv_c99" != xyes],
> - AC_MSG_ERROR([gcc with GNU99 support required]))
> + AC_MSG_WARN([gcc with GNU99 support required]))
I don't think this is correct. If you believe the testcase is (now)
wrong because it tests for an unneeded feature please just adjust the
test.
> AC_CACHE_CHECK([whether gcc supports __attribute__((visibility()))],
> ac_cv_visibility, [dnl
> diff --git a/src/readelf.c b/src/readelf.c
> index 161d7e65..3d6f263e 100644
> --- a/src/readelf.c
> +++ b/src/readelf.c
> @@ -8763,13 +8763,17 @@ print_debug_line_section (Dwfl_Module *dwflmod, Ebl *ebl, GElf_Ehdr *ehdr,
> /* Apply the "operation advance" from a special opcode
> or DW_LNS_advance_pc (as per DWARF4 6.2.5.1). */
> unsigned int op_addr_advance;
> - inline void advance_pc (unsigned int op_advance)
> - {
> - op_addr_advance = minimum_instr_len * ((op_index + op_advance)
> - / max_ops_per_instr);
> - address += op_addr_advance;
> - op_index = (op_index + op_advance) % max_ops_per_instr;
> - }
> + bool show_op_index;
> + #define advance_pc(op_advance_arg) \
> + ( { \
> + unsigned int op_advance = op_advance_arg; \
> + op_addr_advance = minimum_instr_len * ((op_index + (op_advance)) \
> + / max_ops_per_instr); \
> + address += (op_advance); \
> + show_op_index = (op_index > 0 || \
> + (op_index + (op_advance)) % max_ops_per_instr > 0); \
> + op_index = (op_index + (op_advance)) % max_ops_per_instr; \
> + } )
This doesn't compile with gcc:
readelf.c: In function ?print_debug_line_section?:
readelf.c:8766:12: error: variable ?show_op_index? set but not used [-Werror=unused-but-set-variable]
8766 | bool show_op_index;
| ^~~~~~~~~~~~~
cc1: all warnings being treated as errors
Removing the unused show_op_index makes two testcases fail:
FAIL: run-readelf-line.sh
=========================
FAIL run-readelf-multi-noline.sh (exit status: 1)
Cheers,
Mark
Comments
On 14/07/2021 19:40, Mark Wielaard wrote:
>> AC_CACHE_CHECK([whether gcc supports __attribute__((visibility()))],
>> ac_cv_visibility, [dnl
>> diff --git a/src/readelf.c b/src/readelf.c
>> index 161d7e65..3d6f263e 100644
>> --- a/src/readelf.c
>> +++ b/src/readelf.c
>> @@ -8763,13 +8763,17 @@ print_debug_line_section (Dwfl_Module *dwflmod, Ebl *ebl, GElf_Ehdr *ehdr,
>> /* Apply the "operation advance" from a special opcode
>> or DW_LNS_advance_pc (as per DWARF4 6.2.5.1). */
>> unsigned int op_addr_advance;
>> - inline void advance_pc (unsigned int op_advance)
>> - {
>> - op_addr_advance = minimum_instr_len * ((op_index + op_advance)
>> - / max_ops_per_instr);
>> - address += op_addr_advance;
>> - op_index = (op_index + op_advance) % max_ops_per_instr;
>> - }
>> + bool show_op_index;
>> + #define advance_pc(op_advance_arg) \
>> + ( { \
>> + unsigned int op_advance = op_advance_arg; \
>> + op_addr_advance = minimum_instr_len * ((op_index + (op_advance)) \
>> + / max_ops_per_instr); \
>> + address += (op_advance); \
>> + show_op_index = (op_index > 0 || \
>> + (op_index + (op_advance)) % max_ops_per_instr > 0); \
>> + op_index = (op_index + (op_advance)) % max_ops_per_instr; \
>> + } )
>
> This doesn't compile with gcc:
>
> readelf.c: In function ?print_debug_line_section?:
> readelf.c:8766:12: error: variable ?show_op_index? set but not used [-Werror=unused-but-set-variable]
> 8766 | bool show_op_index;
> | ^~~~~~~~~~~~~
> cc1: all warnings being treated as errors
>
> Removing the unused show_op_index makes two testcases fail:
>
> FAIL: run-readelf-line.sh
> =========================
>
> --- readelf.out 2021-07-14 19:37:00.962372827 +0200
> +++ - 2021-07-14 19:37:00.971161054 +0200
> @@ -42,13 +42,13 @@
> [ 36] set prologue end flag
> [ 37] special opcode 19: address+0 = 0x100005a4 <main>, line+1 = 6
> [ 38] set column to 8
> - [ 3a] special opcode 47: address+8 = 0x100005a6 <main+0x2>, line+1 = 7
> + [ 3a] special opcode 47: address+8 = 0x100005ac <main+0x8>, line+1 = 7
> [ 3b] set 'is_stmt' to 0
> [ 3c] advance line by constant -7 to 0
> - [ 3e] special opcode 32: address+4 = 0x100005a7 <main+0x3>, line+0 = 0
> + [ 3e] special opcode 32: address+4 = 0x100005b0 <main+0xc>, line+0 = 0
> [ 3f] set column to 3
> [ 41] set 'is_stmt' to 1
> - [ 42] special opcode 108: address+24 = 0x100005ad <main+0x9>, line+6 = 6
> - [ 43] special opcode 76: address+16 = 0x100005b1 <main+0xd>, line+2 = 8
> - [ 44] advance address by 32 to 0x100005b9 <main+0x15>
> + [ 42] special opcode 108: address+24 = 0x100005c8 <main+0x24>, line+6 = 6
> + [ 43] special opcode 76: address+16 = 0x100005d8 <main+0x34>, line+2 = 8
> + [ 44] advance address by 32 to 0x100005f8
> [ 46] extended opcode 1: end of sequence
> FAIL run-readelf-line.sh (exit status: 1)
>
> FAIL: run-readelf-multi-noline.sh
> =================================
>
> --- readelf.out 2021-07-14 19:37:10.054186557 +0200
> +++ - 2021-07-14 19:37:10.062074795 +0200
> @@ -112,6 +112,6 @@
> [ 6e] extended opcode 2: set address to +0x724 <main>
> [ 79] copy
> [ 7a] set column to 15
> - [ 7c] special opcode 32: address+4 = +0x725 <main+0x1>, line+0 = 1
> - [ 7d] advance address by 4 to +0x726 <main+0x2>
> + [ 7c] special opcode 32: address+4 = +0x728 <main+0x4>, line+0 = 1
> + [ 7d] advance address by 4 to +0x72c
> [ 7f] extended opcode 1: end of sequence
> FAIL run-readelf-multi-noline.sh (exit status: 1)
FWIW, there is a different version of this patch at
https://sourceware.org/pipermail/elfutils-devel/2021q1/003674.html
that doesn't have those problems as far as I remember.
Hi Timm,
On Thu, Jul 15, 2021 at 07:46:15AM +0200, Timm Baeder via Elfutils-devel wrote:
> FWIW, there is a different version of this patch at
> https://sourceware.org/pipermail/elfutils-devel/2021q1/003674.html
> that doesn't have those problems as far as I remember.
Indeed, that looks more clean. I still like the nested function
approach better since it is more concise. But I added a ChangeLog
entry and pushed the above as:
commit 779c57ea864d104bad88455535df9b26336349fd
Author: Timm B?der <tbaeder@redhat.com>
Date: Thu Mar 18 10:25:24 2021 +0100
readelf: Pull advance_pc() in file scope
Make advance_pc() a static function so we can get rid of another nested
function. Rename it to run_advance_pc() and use a local advance_pc()
macro to pass all the local variables. This is similar to what the
equivalent code in libdw/dwarf_getsrclines.c is doing.
Signed-off-by: Timm B?der <tbaeder@redhat.com>
Sorry I missed this patch earilier. If there are any other pending
patches that need review please ping them.
Cheers,
Mark
@@ -42,13 +42,13 @@
[ 36] set prologue end flag
[ 37] special opcode 19: address+0 = 0x100005a4 <main>, line+1 = 6
[ 38] set column to 8
- [ 3a] special opcode 47: address+8 = 0x100005a6 <main+0x2>, line+1 = 7
+ [ 3a] special opcode 47: address+8 = 0x100005ac <main+0x8>, line+1 = 7
[ 3b] set 'is_stmt' to 0
[ 3c] advance line by constant -7 to 0
- [ 3e] special opcode 32: address+4 = 0x100005a7 <main+0x3>, line+0 = 0
+ [ 3e] special opcode 32: address+4 = 0x100005b0 <main+0xc>, line+0 = 0
[ 3f] set column to 3
[ 41] set 'is_stmt' to 1
- [ 42] special opcode 108: address+24 = 0x100005ad <main+0x9>, line+6 = 6
- [ 43] special opcode 76: address+16 = 0x100005b1 <main+0xd>, line+2 = 8
- [ 44] advance address by 32 to 0x100005b9 <main+0x15>
+ [ 42] special opcode 108: address+24 = 0x100005c8 <main+0x24>, line+6 = 6
+ [ 43] special opcode 76: address+16 = 0x100005d8 <main+0x34>, line+2 = 8
+ [ 44] advance address by 32 to 0x100005f8
[ 46] extended opcode 1: end of sequence
FAIL run-readelf-line.sh (exit status: 1)
FAIL: run-readelf-multi-noline.sh
=================================
@@ -112,6 +112,6 @@
[ 6e] extended opcode 2: set address to +0x724 <main>
[ 79] copy
[ 7a] set column to 15
- [ 7c] special opcode 32: address+4 = +0x725 <main+0x1>, line+0 = 1
- [ 7d] advance address by 4 to +0x726 <main+0x2>
+ [ 7c] special opcode 32: address+4 = +0x728 <main+0x4>, line+0 = 1
+ [ 7d] advance address by 4 to +0x72c
[ 7f] extended opcode 1: end of sequence