This patch remedies the following DW_FORM_GNU_ref_alt related problem:
/* FIXME: Do we need to locate the correct CU, in a similar
fashion to the code in the DW_FORM_ref_addr case above ? */
Without the correct CU the wrong abbrevs are used, resulting in
errors and/or wrong file names.
There is scope for further work here. Parsing of CUs should be a two
step process, with the first stage just finding the bounds of the CU.
This would allow find_abstract_instance to quickly find the CU
referenced by DW_FORM_ref_addr or DW_FORM_GNU_ref_alt, then take the
second stage of CU parsing where abbrevs, ranges and suchlike consume
time and memory. As it is, we just process CUs from the start of
.debug_info until we find the one of interest. The testcase in the PR
takes 98G of virtual memory.
PR 25230
* dwarf2.c (struct dwarf2_debug_file): Add line_table and
abbrev_offsets.
(struct abbrev_offset_entry): New.
(hash_abbrev, eq_abbrev, del_abbrev): New functions.
(read_abbrevs): Check whether we have already read abbrevs at
given offset, and add new offset/abbrev to hash table.
(decode_line_info): Keep line table at offset zero in file struct.
Return this for a cu reusing the same dir/file list.
(find_abstract_instance): Find cu for DW_FORM_GNU_ref_alt.
(_bfd_dwarf2_slurp_debug_info): Create offset/abbrev hash tables.
(_bfd_dwarf2_cleanup_debug_info): Adjust deletion of lines and
abbrevs.
read_section does offset checking, reporting an error on out of
bounds. There's no need to duplicate the check in functions calling
read_section. Also, I spotted a place where a pointer difference
expression was being cast to unsigned int, possibly truncating
relevant bits on a 64-bit host.
* dwarf2.c (read_indirect_string): Don't duplicate offset check
done in read_section.
(read_indirect_line_string): Likewise.
(read_alt_indirect_string): Likewise.
(read_alt_indirect_ref): Likewise.
(read_abbrevs): Likewise. Free memory on all failure paths.
Use correct unsigned type for pointer difference comparison.
These are unused. Remove them. Also fix the wrong sort of 0 being
returned from read_alt_indirect_ref.
* dwarf2.c (struct dwarf2_debug): Update comments. Remove sec
and sec_info_ptr.
(_bfd_dwarf2_slurp_debug_info): Don't set sec or sec_info_ptr.
(stash_comp_unit): Likewise.
(read_alt_indirect_ref): Return NULL not FALSE.
There were a number of problems with the previous patch. Firstly,
_bfd_dwarf2_stash_syms didn't do anything when the original file had
dynamic symbols, and secondly, info found by the symbol search didn't
make it out of _bfd_elf_find_nearest_line except in the case of DWARF
functions without external linkage.
PR 23652
* dwarf2.c (_bfd_dwarf2_stash_syms): Break out of loop on finding
matching section.
(_bfd_dwarf2_find_nearest_line): Return an int, with value 2 when
returning info from the symbol table. Do the _bfd_elf_find_function
search also when !found. Call _bfd_dwarf2_stash_syms regardless of
symbols.
* elf64-alpha.c (elf64_alpha_find_nearest_line): Accept dwarf2
result of 1 only.
* elfxx-mips.c (_bfd_mips_elf_find_nearest_line): Likewise.
* libbfd-in.h (_bfd_dwarf2_find_nearest_line): Update prototype.
* libbfd.h: Regenerate.
Sometimes DWARF info for a function is incomplete, and the function
can be retrieved by examining symbols. However, when separate debug
files are used it may be that the original file is completely
stripped of symbols. This patch teaches BFD to look at symbols from
the debug file in that case.
The patch also removes arm_elf_find_function, instead implementing
elf_backend_maybe_function_sym. arm_elf_find_function was written
before the generic _bfd_elf_find_function called maybe_function_sym.
aarch64 copied arm, so that file gets the same treatment. There is
some chance this will speed up arm and aarch64 lookup of function/line.
PR 23652
* dwarf2.c (_bfd_dwarf2_stash_syms): New function.
(_bfd_dwarf2_find_nearest_line): Use it here, passing syms to
_bfd_elf_find_function. Call _bfd_elf_find_function in cases
where _bfd_elf_find_nearest_line would do so.
* elf.c (_bfd_elf_find_nearest_line): Omit _bfd_elf_find_function
for dwarf2.
* elfxx-mips.c (_bfd_mips_elf_find_nearest_line): Similarly. Tidy.
* elf32-arm.c (elf32_arm_maybe_function_sym): New function.
(elf_backend_maybe_function_sym): Define.
(arm_elf_find_function, elf32_arm_find_nearest_line): Delete.
(bfd_elf32_find_nearest_line): Don't define.
* elfnn-aarch64.c (elfNN_aarch64_maybe_function_sym): New function.
(elf_backend_maybe_function_sym): Define.
(aarch64_elf_find_function, elfNN_aarch64_find_nearest_line): Delete.
(bfd_elfNN_find_nearest_line): Don't define.
This patch ensures qsort stability in line and function sorting done
in dwarf2.c. For the line sequences we make use of an existing field
that isn't used until later, as a monotonic counter for the qsort.
* dwarf2.c (struct lookup_funcinfo): Add idx field.
(compare_lookup_funcinfos): Perform final sort on idx.
(build_lookup_funcinfo_table): Set idx.
(compare_sequences): Perform final sort on num_lines.
(build_line_info_table): Set num_lines and line_info_lookup earlier.
(sort_line_sequences): Set num_lines for sort.
stash_maybe_enable_info_hash_tables sets
stash->info_hash_status = STASH_INFO_HASH_ON;
regardless of the result of stash_maybe_update_info_hash_tables call. In
case it fails this results in repeated invocation of comp_unit_hash_info
for the same comp unit and assertion failure in this function.
Only set stash->info_hash_status = STASH_INFO_HASH_ON; when
stash_maybe_update_info_hash_tables is successful.
bfd/
2019-10-11 Max Filippov <jcmvbkbc@gmail.com>
* dwarf2.c (stash_maybe_enable_info_hash_tables): Only set
stash->info_hash_status = STASH_INFO_HASH_ON when
stash_maybe_update_info_hash_tables succeeds.
Evil testcase with two debug info sections, with sizes of 2aaaabac4ec1
and ffffd5555453b140 result in a total size of 1. Reading the first
section of course overflows the buffer and tramples on other memory.
PR 25070
* dwarf2.c (_bfd_dwarf2_slurp_debug_info): Catch overflow of
total_size calculation.
This fixes a small leak of debug_filename. bfd_openr copies the file
name since git commit 1be5090bca.
PR 11983
* dwarf2.c (_bfd_dwarf2_slurp_debug_info): Free debug_filename
on success. Tidy.
A customer reported a case where addr2line was very slow. We tracked
this down to some N^2 behavior in _bfd_dwarf2_find_symbol_bias in the
unusual case where no function can be found.
This patch fixes the bug, and reduces the runtime for a particular
request from 127 seconds to 1 second.
bfd/ChangeLog
2019-08-19 Tom Tromey <tromey@adacore.com>
* dwarf2.c (_bfd_dwarf2_find_symbol_bias): Create hash table
holding symbols.
A customer program had a DWARF CU that consisted of just a CU DIE,
without any children. In this situation, scan_unit_for_symbols will
try to read past the end of the current CU, and will take use the
first bytes of the next CU as an abbrev, printing an error message.
This patch fixes the bug by changing scan_unit_for_symbols to stop at
the end of the CU rather than the end of the .debug_info section.
bfd/ChangeLog
2019-08-15 Tom Tromey <tromey@adacore.com>
* dwarf2.c (scan_unit_for_symbols): Check for end of CU, not end
of section.
PR 24623
* dwarf2.c (stash_comp_unit): New function, extracted from..
(_bfd_dwarf2_find_nearest_line): ..here.
(find_abstract_instance): Parse comp units and decode line info
as needed.
This parameter might appear to be used to set up offset_size, but
since git commit 024b2372f5 offset_size is either set from the
debug_info data or is set to 4.
* dwarf2.c (_bfd_dwarf2_find_nearest_line): Remove addr_size parameter.
* libbfd-in.h (_bfd_dwarf2_find_nearest_line): Update prototype.
* coffgen.c (coff_find_nearest_line_with_names): Adjust
_bfd_dwarf2_find_nearest_line calls.
* elf.c (_bfd_elf_find_nearest_line, _bfd_elf_find_line): Likewise.
* elf32-arm.c (elf32_arm_find_nearest_line): Likewise.
* elf64-alpha.c (elf64_alpha_find_nearest_line): Likewise.
* elfnn-aarch64.c (elfNN_aarch64_find_nearest_line): Likewise.
* elfxx-mips.c (_bfd_mips_elf_find_nearest_line): Likewise.
* mach-o.c (bfd_mach_o_find_nearest_line): Likewise.
* libbfd.h: Regenerate.
The XCOFF linker temporarily trims the output bfd section list,
without adjusting section_count to suit. This is a little rude, but
the dwarf line number code can easily cope with this situation. So
check for a NULL end of list as well as limiting the saved section
VMAs to the first section_count list entries.
Also fixes
-FAIL: Weak test 3 (main, static) (32-bit)
-FAIL: Weak test 3 (main, static) (64-bit)
PR 24596
* dwarf2.c (save_section_vma, section_vma_same): Check for NULL
end of section list as well as section_count.
* xcofflink.c (xcoff_link_add_symbols): Fix temporarily changed
section list before returning error.
PR 24334
* dwarf2.c (struct dwarf2_debug): Add sec_vma_count field.
(save_section_vma): Initialise field to the number of entries in
the sec_vma table.
(section_vma_same): Check that the number of entries in the
sec_vma table matches the number of sections in the bfd.
* dwarf2.c (_bfd_dwarf2_find_symbol_bias): Check for a NULL symbol
table pointer.
* coffgen.c (coff_find_nearest_line_with_names): Do not call
_bfd_dwarf2_find_symbol_bias if there is no symbol table available.
https://bugzilla.redhat.com/show_bug.cgi?id=1685727
Fixes the bugs exposed by the testcases in the PR, plus two more bugs
I noticed when looking at _bfd_stab_section_find_nearest_line.
PR 23686
* dwarf2.c (read_section): Error when attempting to malloc
"(bfd_size_type) -1".
* syms.c (_bfd_stab_section_find_nearest_line): Bounds check
function_name. Bounds check reloc address. Formatting. Ensure
.stabstr zero terminated.
dwarf2.c code reasonably assumes that debug info is local to a file,
an assumption now violated by gcc, resulting in "DWARF error: invalid
abstract instance DIE ref" or wrong details when attempting to print
linker error messages with file, function and line reported.
This is because find_abstract_instance is only prepared to handle
DW_FORM_ref_addr when the .debug_info section referenced is in the
current file. When that isn't the case, relocations to access another
file's .debug_info will typically be against a symbol defined at the
start of that .debug_info section, plus an addend. Since the dwarf2.c
code only considers the current file's debug info, that symbol will be
undefined, resolving to zero. In effect the ref_addr will wrongly
resolve to the current file's .debug_info.
This patch avoids the problem by treating relocations in debug
sections against undefined symbols in a similar manner to the way
relocations against symbols defined in discarded sections are
resolved. They result in a zero value (except in .debug_ranges)
regardless of the addend.
PR 23425
* reloc.c (bfd_generic_get_relocated_section_contents): Zero reloc
fields in debug sections when reloc is against an undefined symbol
and called from bfd_simple_get_relocated_section_contents or
similar.
* dwarf2.c (find_abstract_instance): Return true for zero offset
DW_FORM_ref_addr without returning values.
PR 22895
PR 22893
* dwarf2.c (read_n_bytes): Replace size parameter with dwarf_block
pointer. Drop unused abfd parameter. Check the size of the block
before initialising the data field. Return the end pointer if the
size is invalid.
(read_attribute_value): Adjust invocations of read_n_bytes.
The nm utility supports -l for using debug information to obtain file and line information for each symbol, if available.
We have a tool that consumes this information and displays it.
This identified a problem with the 'nm' utility.
When a source is compiled with -O2, functions can be inlined. The compiler also produces an uninlined copy of the function, normally for linking to other object files.
In the case of DWARF2 debug information, the compiler generates debug information to describe a function. If that function is inlined, the compiler then references that debug information from the inlined and uninlined copies of the routine through the use of the DW_AT_abstract_origin reference.
When nm is used on such a file, it is not able to find file and line information because that information is present in the common debug information and not at each actual implementation of the function.
The 'nm' utility only retrieves the name of the function from the abstract origin debug information and no more.
What I am proposing is to modify the find_abstract_instance_name() function (which I renamed to find_abstract_instance() ) to return the name of the function as well as any file and line information. The routine is already parsing all of the debug information in the abstract instance, so it is easy to pick up the file and line information at that time. If, for some reason, the file and line information is not present, the routine behaves as before.
For example, if I have a simple test case:
int foo(int j)
{
if (j < 15)
j += j << 2;
else
j += j << 6;
return j;
}
int main (int argc,char **argv)
{
int i = argc;
i += foo(i);
return i;
}
If that test case is compiled with -O2 and then 'nm -l' reads that executable, it currently produces this symbol output (ignoring a lot of library symbols):
8048400 T foo
080482e0 T main /scratch/pcarroll/its254/test/mytest.c:12
If I modify 'nm' to return file and line information for abstract instances, it produces the following output:
08048400 T foo /scratch/pcarroll/its254/test/mytest.c:1
080482e0 T main /scratch/pcarroll/its254/test/mytest.c:12
--------------------------------------------------------------------------
bfd * bfd/dwarf2.c (find_abstract_name): Modified to return file and
line information in addition to name, if they can be found.
Like the PR22230 fix, we can allocate a buffer with an extra byte
rather than letting bfd_simple_get_relocated_section_contents malloc
and return a buffer. Much better than allocating another buffer
afterwards.
PR 22047
* dwarf2.c (read_section): Allocate buffer with extra byte for
bfd_simple_get_relocated_section_contents rather than copying
afterwards.
Using read_attribute_value accomplishes two things: It checks for
unexpected formats, and ensures the buffer pointer always increments.
PR 22210
* dwarf2.c (read_formatted_entries): Use read_attribute_value to
read data.
This patch adds bounds checking for DW_FORM_ref_addr die refs, and
calculates them relative to the first .debug_info section. See the
big comment for why calculating relative to the current .debug_info
section was wrong for relocatable object files.
PR 22209
* dwarf2.c (struct comp_unit): Delete sec_info_ptr field.
(find_abstract_instance_name): Calculate DW_FORM_ref_addr relative
to stash->info_ptr_memory, and check die_ref is within that memory.
Set info_ptr_end correctly when another CU is refd. Check die_ref
for DW_FORM_ref4 etc. is within CU.
The PR22200 fuzzer testcase found one way to put NULLs into .debug_line
file tables. PR22205 finds another. This patch gives up on trying to
prevent NULL files in the file table and instead just copes with them.
Arguably, this is better than giving up and showing no info from
.debug_line. I've also fixed a case where the fairly recent DWARF5
support in handling broken DWARG could result in uninitialized memory
reads, and made a small tidy.
PR 22205
* dwarf2.c (concat_filename): Return "<unknown>" on NULL filename.
(read_formatted_entries): Init "fe".
(decode_line_info <DW_LNE_define_file>): Use line_info_add_file_name.
DW_AT_name ought to always have a string value.
PR 22201
* dwarf2.c (scan_unit_for_symbols): Ignore DW_AT_name unless it
has string form.
(parse_comp_unit): Likewise.
table->sequences is a linked list before it is replaced by a bfd_alloc
array in sort_line_sequences.
PR 22191
* dwarf2.c (decode_line_info): Properly free line sequences on error.
Alan Modra
Max Filippov
Tom Tromey
Nick Clifton
Millan Wolff
Paul Carroll