In write_pieced_value, when checking whether the data can be transferred
byte-wise, the current logic verifies the source- and destination offsets
to be byte-aligned, but not the transfer size. This is fixed.
gdb/ChangeLog:
* dwarf2loc.c (write_pieced_value): Include transfer size in
byte-wise check.
In write_pieced_value, the number of bytes containing a portion of the
bit-field in a given piece is calculated with the wrong starting offset;
thus the result may be off by one. This bug was probably introduced when
copying this logic from read_pieced_value. Fix it.
gdb/ChangeLog:
* dwarf2loc.c (write_pieced_value): Fix copy/paste error in the
calculation of this_size.
In the case of targeting a bit-field, read_pieced_value and
write_pieced_value calculate the number of bits preceding the bit-field
without considering the relative offset of the value's parent. This is
relevant for a structure variable like this:
struct s {
uint64_t foo;
struct {
uint32_t bar;
uint32_t bf : 10; /* <-- target bit-field */
} baz;
} s;
In this scenario, if 'val' is a GDB value representing s.baz.bf,
val->parent represents the whole s.baz structure, and the following holds:
- value_offset (val) == sizeof s.baz.bar == 4
- value_offset (val->parent) == sizeof s.foo == 8
The current logic would only use value_offset(val), resulting in the wrong
offset into the target value. This is fixed.
gdb/ChangeLog:
* dwarf2loc.c (read_pieced_value): Respect parent value's offset
when targeting a bit-field.
(write_pieced_value): Likewise.
The test suite contains multiple instances of determining the target's
endianness with GDB's "show endian" command. This patch replaces these by
an invocation of a new convenience proc 'get_endianness'.
gdb/testsuite/ChangeLog:
* lib/gdb.exp (get_endianness): New proc.
* gdb.arch/aarch64-fp.exp: Use it.
* gdb.arch/altivec-regs.exp: Likewise.
* gdb.arch/e500-regs.exp: Likewise.
* gdb.arch/vsx-regs.exp: Likewise.
* gdb.base/dump.exp: Likewise.
* gdb.base/funcargs.exp: Likewise.
* gdb.base/gnu_vector.exp: Likewise.
* gdb.dwarf2/formdata16.exp: Likewise.
* gdb.dwarf2/implptrpiece.exp: Likewise.
* gdb.dwarf2/nonvar-access.exp: Likewise.
* gdb.python/py-inferior.exp: Likewise.
* gdb.trace/unavailable-dwarf-piece.exp: Likewise.
The addr_size field in the piece_closure data structure is a relic from
before introducing the typed DWARF stack. It is obsolete now. This patch
removes it.
gdb/ChangeLog:
* dwarf2loc.c (struct piece_closure) <addr_size>: Remove field.
(allocate_piece_closure): Drop addr_size parameter.
(dwarf2_evaluate_loc_desc_full): Adjust call to
allocate_piece_closure.
When taking a DW_OP_piece or DW_OP_bit_piece from a DW_OP_stack_value, the
existing logic always takes the piece from the lowest-addressed end, which
is wrong on big-endian targets. The DWARF standard states that the
"DW_OP_bit_piece operation describes a sequence of bits using the least
significant bits of that value", and this also matches the current logic
in GCC. For instance, the GCC guality test case pr54970.c fails on s390x
because of this.
This fix adjusts the piece accordingly on big-endian targets. It is
assumed that:
* DW_OP_piece shall take the piece from the LSB end as well;
* pieces reaching outside the stack value bits are considered undefined,
and a zero value can be used instead.
gdb/ChangeLog:
PR gdb/21226
* dwarf2loc.c (read_pieced_value): Anchor stack value pieces at
the LSB end, independent of endianness.
gdb/testsuite/ChangeLog:
PR gdb/21226
* gdb.dwarf2/nonvar-access.exp: Add checks for verifying that
stack value pieces are taken from the LSB end.
A field f in a structure composed of DWARF pieces may be located in
multiple pieces, where the first and last of those may contain bits from
other fields as well. So when writing to f, the beginning of the first
and the end of the last of those pieces may have to be skipped. But the
logic in write_pieced_value for handling one of those pieces is flawed
when the first and last piece are the same, i.e., f is contained in a
single piece:
< - - - - - - - - - piece_size - - - - - - - - - ->
+-------------------------------------------------+
| skipped_bits | f_bits | / / / / / / / / / / |
+-------------------------------------------------+
The current logic determines the size of the sub-piece to operate on by
limiting the piece size to the bit size of f and then subtracting the
skipped bits:
min (piece_size, f_bits) - skipped_bits
Instead of:
min (piece_size - skipped_bits, f_bits)
So the resulting sub-piece size is corrupted, leading to wrong handling of
this piece in write_pieced_value.
Note that the same bug was already found in read_pieced_value and fixed
there (but not in write_pieced_value), see PR 15391.
This patch swaps the calculations, bringing them into the same (correct)
order as in read_pieced_value.
gdb/ChangeLog:
* dwarf2loc.c (write_pieced_value): Fix order of calculations for
size capping.
gdb/testsuite/ChangeLog:
* gdb.dwarf2/var-pieces.exp: Add test case for modifying a
variable at nonzero offset.
This adds a test for read/write access to variables with various types of
DWARF locations. It uses register- and memory locations and composite
locations with register- and memory pieces.
Since the new test calls gdb_test_no_output with commands that contain
braces, it is necessary for string_to_regexp to quote braces as well.
This was not done before.
gdb/testsuite/ChangeLog:
* gdb.dwarf2/var-access.c: New file.
* gdb.dwarf2/var-access.exp: New test.
* lib/gdb-utils.exp (string_to_regexp): Quote braces as well.
All target descriptions except mips initialization are called in -tdep.c,
instead of -nat.c, so this patch moves mips target descriptions
initialization to -tdep.c. Secondly, I want to change the target
descriptions from pre-generated to dynamical creation, so I want to test
that these pre-generated target descriptions equal to these dynamically
created ones. Move target descriptions initialization to -tdep.c files so
we can test them in any hosts (if they are still -nat.c, we can only test
them on mips-linux host.).
gdb:
2017-06-13 Yao Qi <yao.qi@linaro.org>
* mips-linux-nat.c: Move include features/mips*-linux.c to
mips-linux-tdep.c.
(_initialize_mips_linux_nat): Move initialize_tdesc_mips* calls
to mips-linux-tdep.c.
* mips-linux-tdep.c: Include features/mips*-linux.c
(_initialize_mips_linux_tdep): Call initialize_tdesc_mips*
functions.
* mips-linux-tdep.h (tdesc_mips_linux): Declare.
(tdesc_mips_dsp_linux, tdesc_mips64_linux): Declare.
(tdesc_mips64_dsp_linux): Declare.
Now that print_scalar_formatted is more capable, there's no need for
val_print_type_code_int. This patch removes it in favor of
val_print_scalar_formatted.
2017-06-12 Tom Tromey <tom@tromey.com>
* valprint.h (val_print_type_code_int): Remove.
* valprint.c (generic_val_print_int): Always call
val_print_scalar_formatted.
(val_print_type_code_int): Remove.
* printcmd.c (print_scalar_formatted): Handle options->format==0.
* f-valprint.c (f_val_print): Use val_print_scalar_formatted.
* c-valprint.c (c_val_print_int): Use val_print_scalar_formatted.
* ada-valprint.c (ada_val_print_num): Use
val_print_scalar_formatted.
This unifies the two switches in print_scalar_formatted, removing some
now-redundant code. Now scalar types are never converted to LONGEST,
instead printing is done using print_*_chars, operating on the byte
representation.
ChangeLog
2017-06-12 Tom Tromey <tom@tromey.com>
* printcmd.c (print_scalar_formatted): Unify the two switches.
Don't convert scalars to LONGEST.
2017-06-12 Tom Tromey <tom@tromey.com>
* gdb.arch/altivec-regs.exp: Expect decimal results for uint128.
mips_eabi_push_dummy_call is storing the address of a struct in a
buffer that must have the same of the confisued/set ABI register size.
Add a define for the maximum ABI size and use it to size the local
buffer. Also rename the 'regsize' local to 'abi_regsize' for clarity.
Tested that --enable-targets=all still builds.
gdb/ChangeLog:
2017-06-12 Pedro Alves <palves@redhat.com>
Alan Hayward <alan.hayward@arm.com>
* mips-tdep.c (MAX_MIPS_ABI_REGSIZE): New.
(mips_eabi_push_dummy_call): Rename local 'regsize' to
'abi_regsize'. Rename local array 'valbuf' to 'ref_valbuf', and
use MAX_MIPS_ABI_REGSIZE instead of MAX_REGISTER_SIZE to size it.
Assert that abi_regsize bytes fit in 'ref_valbuf'.
... instead of vector of pointers
There's no real reason for having mapped_symtab::data be a vector of
heap-allocated symtab_index_entries. symtab_index_entries is not that
large, it's movable, and it's cheap to move. Making the vector hold
values instead improves cache locality and eliminates many roundtrips
to the heap.
Using the same test as in the previous patch, against the same gdb
inferior, timing improves ~13% further:
~6.0s => ~5.2s (average of 5 runs).
Note that before the .gdb_index C++ifycation patch, we were at ~5.7s.
We're now consistenly better than before.
gdb/ChangeLog
2017-06-12 Pedro Alves <palves@redhat.com>
* dwarf2read.c (mapped_symtab::data): Now a vector of
symtab_index_entry instead of vector of
std::unique_ptr<symtab_index_entry>. All users adjusted to check
whether an element's name is NULL instead of checking whether the
element itself is NULL.
(find_slot): Change return type. Adjust.
(hash_expand, , add_index_entry, uniquify_cu_indices)
(write_hash_table): Adjust.
Using the same test as the previous patch, perf shows GDB spending
over 7% in "free". A substantial number of those calls comes from
insertions in the psyms_seen unordered_set causing lots of rehashing
and recreating buckets. Fix this by computing an estimate of the size
of the set upfront.
Using the same test as in the previous patch, against the same gdb
inferior, timing improves ~8% further:
~6.5s => ~6.0s (average of 5 runs).
gdb/ChangeLog:
2017-06-12 Pedro Alves <palves@redhat.com>
* dwarf2read.c (recursively_count_psymbols): New function.
(write_psymtabs_to_index): Call it to compute number of psyms and
pass estimate size of psyms_seen to unordered_set's ctor.
"perf" shows the unordered_map::emplace call in write_hash_table a bit
high up on profiles. Fix this using the find + insert idiom instead
of going straight to insert.
I tried doing the same to the other unordered_maps::emplace calls in
the file, but saw no performance improvement, so left them be.
With a '-g3 -O2' build of gdb, and:
$ cat save-index.cmd
set $i = 0
while $i < 100
save gdb-index .
set $i = $i + 1
end
$ time ./gdb -data-directory=data-directory -nx --batch -q -x save-index.cmd ./gdb.pristine
I get an improvement of ~7%:
~7.0s => ~6.5s (average of 5 runs).
gdb/ChangeLog:
2017-06-12 Pedro Alves <palves@redhat.com>
* dwarf2read.c (write_hash_table): Check if key already exists
before emplacing.
This avoids having to specify the integer size twice in the same line.
gdb/ChangeLog:
2017-06-12 Pedro Alves <palves@redhat.com>
* dwarf2read.c (data_buf::append_space): Rename to...
(data_buf::grow): ... this, and make private. Adjust all callers.
(data_buf::append_uint): New method.
(add_address_entry, write_one_signatured_type)
(write_psymtabs_to_index): Use it.
There's no real need for all this indirection.
gdb/ChangeLog:
2017-06-12 Pedro Alves <palves@redhat.com>
* dwarf2read.c (file_write(FILE *, const void *, size_t)): Delete.
(file_write (FILE *, const std::vector<Elem>&)): Delete.
(data_buf::file_write): Call ::fwrite directly.
gdb/ChangeLog
2017-06-12 Jan Kratochvil <jan.kratochvil@redhat.com>
Code cleanup: C++ify .gdb_index producer.
* dwarf2read.c: Include <unordered_set> and <unordered_map>.
(MAYBE_SWAP) [WORDS_BIGENDIAN]: Cast to offset_type.
(struct strtab_entry, hash_strtab_entry, eq_strtab_entry)
(create_strtab, add_string): Remove.
(file_write, data_buf): New.
(struct symtab_index_entry): Use std::vector for cu_indices.
(struct mapped_symtab): Use std::vector for data.
(hash_symtab_entry, eq_symtab_entry, delete_symtab_entry)
(create_symbol_hash_table, create_mapped_symtab, cleanup_mapped_symtab):
Remove.
(find_slot): Change return type. Update it to the new data structures.
(hash_expand, add_index_entry): Update it to the new data structures.
(offset_type_compare): Remove.
(uniquify_cu_indices): Update it to the new data structures.
(c_str_view, c_str_view_hasher, vector_hasher): New.
(add_indices_to_cpool): Remove.
(write_hash_table): Update it to the new data structures.
(struct psymtab_cu_index_map, hash_psymtab_cu_index)
(eq_psymtab_cu_index): Remove.
(psym_index_map): New typedef.
(struct addrmap_index_data): Change addr_obstack pointer to data_buf
reference and std::unordered_map for cu_index_htab.
(add_address_entry, add_address_entry_worker, write_address_map)
(write_psymbols): Update it to the new data structures.
(write_obstack): Remove.
(struct signatured_type_index_data): Change types_list to a data_buf
reference and psyms_seen to a std::unordered_set reference.
(write_one_signatured_type, recursively_write_psymbols)
(write_psymtabs_to_index): Update it to the new data structures.
Pass struct bfd_link_info * to merge_gnu_properties to give backend
control of how GNU properties are merged based on linker command line
options.
* elf-bfd.h (elf_backend_data): Add struct bfd_link_info *
to merge_gnu_properties.
* elf-properties.c (elf_merge_gnu_properties): Add struct
bfd_link_info * and pass it to merge_gnu_properties.
(elf_merge_gnu_property_list): Add struct bfd_link_info *
and pass it to elf_merge_gnu_properties.
(_bfd_elf_link_setup_gnu_properties): Pass info to
elf_merge_gnu_property_list.
* elf32-i386.c (elf_i386_merge_gnu_properties): Add struct
bfd_link_info *.
* elf64-x86-64.c (elf_x86_64_merge_gnu_properties): Likewise.
bfd/ChangeLog:
2017-06-12 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
* elf32-s390.c (elf_s390_relocate_section): Return false in case
the rewriting fails.
* elf64-s390.c (elf_s390_relocate_section): Likewise.
I helped someone figure out why their separate debug info (debug
link-based) was not found by gdb. It turns out that the debug file was
not named properly. It made me realize that it is quite difficult to
diagnose this kind of problems. This patch adds some debug output to
show where GDB looks for those files, so that it should be (more)
obvious to find what's wrong.
Here's an example of the result, first with an example of unsuccessful lookup,
and then a successful one.
(gdb) set debug separate-debug-file on
(gdb) file /usr/bin/gnome-calculator
Reading symbols from /usr/bin/gnome-calculator...
Looking for separate debug info (build-id) for /usr/bin/gnome-calculator
Trying /usr/local/lib/debug/.build-id/0d/5c5e8c86dbe4f4f95f7a13de04f91d377f3c6a.debug
Looking for separate debug info (debug link) for /usr/bin/gnome-calculator
Trying /usr/bin/5c5e8c86dbe4f4f95f7a13de04f91d377f3c6a.debug
Trying /usr/bin/.debug/5c5e8c86dbe4f4f95f7a13de04f91d377f3c6a.debug
Trying /usr/local/lib/debug//usr/bin/5c5e8c86dbe4f4f95f7a13de04f91d377f3c6a.debug
(no debugging symbols found)...done.
(gdb) set debug-file-directory /usr/lib/debug
(gdb) file /usr/bin/gnome-calculator
Reading symbols from /usr/bin/gnome-calculator...
Looking for separate debug info by build-id for /usr/bin/gnome-calculator
Trying /usr/lib/debug/.build-id/0d/5c5e8c86dbe4f4f95f7a13de04f91d377f3c6a.debug
Reading symbols from /usr/lib/debug/.build-id/0d/5c5e8c86dbe4f4f95f7a13de04f91d377f3c6a.debug...done.
done.
Note: here, the debug link happens to be named like the build-id, but it
doesn't have to be this way. It puzzled me for a minute.
gdb/ChangeLog:
* NEWS (Changes since GDB 8.0): Announce {set,show} debug
separate-debug-file commands.
* symfile.h (separate_debug_file_debug): New global.
* symfile.c (separate_debug_file_debug): New global.
(separate_debug_file_exists, find_separate_debug_file): Add
debug output.
(_initialize_symfile): Add "set debug separate-debug-file"
command.
* build-id.c (build_id_to_debug_bfd,
find_separate_debug_file_by_buildid): Add debug output.
gdb/doc/ChangeLog:
* gdb.texinfo (Optional Messages about Internal Happenings):
Document {set,show} debug separate-debug-file commands.
The displaced_step_free_closure gdbarch hook allows architectures to
free data they might have allocated to complete a displaced step.
However, all architectures using that hook use the
simple_displaced_step_free_closure provided in arch-utils.{c,h}, which
does a simple xfree. We can remove it and do an xfree directly instead
of calling the hook.
gdb/ChangeLog:
* gdbarch.sh (displaced_step_free_closure): Remove.
* gdbarch.h, gdbarch.c: Re-generate.
* aarch64-linux-tdep.c (aarch64_linux_init_abi): Don't set
displaced_step_free_closure.
* amd64-linux-tdep.c (amd64_linux_init_abi_common): Likewise.
* arm-linux-tdep.c (arm_linux_init_abi): Likewise.
* i386-linux-tdep.c (i386_linux_init_abi): Likewise.
* rs6000-aix-tdep.c (rs6000_aix_init_osabi): Likewise.
* rs6000-tdep.c (rs6000_gdbarch_init): Likewise.
* s390-linux-tdep.c (s390_gdbarch_init): Likewise.
* arch-utils.h (simple_displaced_step_free_closure): Remove.
* arch-utils.c (simple_displaced_step_free_closure): Remove.
* infrun.c (displaced_step_clear): Call xfree instead of
gdbarch_displaced_step_free_closure.
Hi,
This is another obvious patch that fixes a thinko from my previous
startup-with-shell series. We should conditionally include <signal.h>
on gdb/gdbserver/fork-child.c because gdbserver will be putting the
inferior's terminal on the correct mode after the call to
fork_inferior, and for that it needs to ignore SIGTTOU.
This patch fixes a bunch of regressions happening on AArch64 that were
reported by Yao.
gdb/gdbserver/ChangeLog:
2017-06-09 Sergio Durigan Junior <sergiodj@redhat.com>
* fork-child.c: Conditionally include <signal.h>.
This reverts commit bc327528fd.
This patch can only be committed after PC-relative relocation types
support on copy relocation elimination is also completed.
Hi,
This bug is related to:
<https://sourceware.org/ml/gdb-patches/2017-06/msg00216.html>
On stringify_argv, we have to check if args[0] is not NULL before
stringifying anything, otherwise we might do the wrong thing when
trimming the "ret" string in the end. args[0] will be NULL when no
arguments are passed to the inferior that will be started.
Checked in as obvious.
gdb/ChangeLog:
2017-06-08 Sergio Durigan Junior <sergiodj@redhat.com>
* common/common-utils.c (stringify_argv): Check for "arg[0] !=
NULL".
Fixed conditions to create the dynamic sections.
Previously there would be times where the dynamic sections would not be created
although they were actually required for linking to work.
Issue found through OpenADK build, more precisely the ublicb testsuite package.
bfd/ChangeLog:
Cupertino Miranda <cmiranda@synopsys.com>
elf32-arc.c (elf_arc_check_relocs): Fixed conditions to generate
dynamic sections.
TEXTREL was being generated even when relocatable .o files had the .rela.text
section. Now it is limitted only to dynamic object files that still have them.
Nevertheless, our target aborts in those cases due to architecture limitations
where icache is not coherent with dcache, and to force this coherence expensive
kernel level support would be needed.
bfd/ChangeLog:
Cupertino Miranda <cmiranda@synopsys.com>
* elf32-arc.c (elf_arc_size_dynamic_sections): Changed condition to
require TEXTREL.
In the case of static relocation, the GOT entries are fixed at link time
and are set by the linker.
In order to compute the right TLS offset it is necessary to add TCB_SIZE
to the offset, just in case the dynamic linker is not expected to be
executed (static linked case).
This problem does appear in dynamic linked applications, as the dynamic
linker is adding this TCB_SIZE by operating the TCB block structure.
Problem revealed in GLIBC with static linking.
bfd/ChangeLog:
Cupertino Miranda <cmiranda@synopsys.com>
arc-got.h (relocate_fix_got_relocs_for_got_info): Added TCB_SIZE to
patched section contents for TLS IE reloc.
elf32-arc.c: Remove TCB_SIZE preprocessor macro.
Rebase to 0006
This miss was identified in the context of openssh building for ARC.
bfd/ChangeLog:
Cupertino Miranda <cmiranda@synopsys.com>
elf32-arc.c (elf_arc_relocate_section): Added "call" to
RELOC_FOR_GLOBAL_SYMBOL macro.
Fixed issue related to the generation of ARC_PC32 dynamic relocs when symbol
is dynamic but still defined in a non shared object.
bfd/ChangeLog:
Cupertino Miranda <cmiranda@synopsys.com>
elf32-arc.c (elf_arc_relocate_section): Small refactor and condition
changes.
Historically the arc abi demanded that a GOT[0] should be referencible as
[pc+_DYNAMIC@gotpc]. Hence we convert a _DYNAMIC@gotpc to a GOTPC reference to
_GLOBAL_OFFSET_TABLE_.
This is no longer the case and uClibc and upcomming GNU libc don't expect this
to happen.
gas/ChangeLog:
Vineet Gupta <vgupta@synopsys.com>
Cupertino Miranda <cmiranda@synopsys.com>
* config/tc-arc.c (md_undefined_symbol): Changed.
* config/tc-arc.h (DYNAMIC_STRUCT_NAME): Removed.
GOT information would not be reassign to symbol when it became a indect
symbol.
bfd/ChangeLog:
Cupertino Miranda <cmiranda@synopsys.com>
* elf32-arc.c (elf_arc_relocate_section): Fixed reassign of indirect
symbols.
R_ARC_32 and R_ARC_32_ME cannot be generated as dynamic relocs.
However, a warning message and check_relocs was aborting when this type of
reloc was being resolved to a local symbol.
This is wrong as local symbols are resolvable at link time.
bfd/ChangeLog:
Cupertino Miranda <cmiranda@synopsys.com>
* elf32-arc.c (elf_arc_check_relocs): Added condition to disable
warning and "Bad value" for local symbols ARC_32 or ARC_32_ME relocs.
It's generally a bad idea to use assertions to validate our idea of
what an input file looks like. We need to be as liberal as possible
in what we accept with respect to standards and conservative with what
we produce.
Currently, if gcc is used to produce an assembler file which contains
only data, but the FPU is set to fpv4-sp-d16 and mfloat-abi=hard, then
the following attributes will be set in the output:
.cpu arm7tdmi
.eabi_attribute 27, 1 @ Tag_ABI_HardFP_use
.eabi_attribute 28, 1 @ Tag_ABI_VFP_args
.eabi_attribute 20, 1 @ Tag_ABI_FP_denormal
.eabi_attribute 21, 1 @ Tag_ABI_FP_exceptions
.eabi_attribute 23, 3 @ Tag_ABI_FP_number_model
.eabi_attribute 24, 1 @ Tag_ABI_align8_needed
.eabi_attribute 25, 1 @ Tag_ABI_align8_preserved
.eabi_attribute 26, 2 @ Tag_ABI_enum_size
.eabi_attribute 30, 6 @ Tag_ABI_optimization_goals
.eabi_attribute 34, 0 @ Tag_CPU_unaligned_access
.eabi_attribute 18, 4 @ Tag_ABI_PCS_wchar_t
There is then no .fpu directive to cause Tag_FP_arch to be set,
because there are no functions containing code in the object file. If
this object file is assembled by hand, but without -mfpu on the
invocation of the assembler, then the build attributes produced will
trigger an assertion during linking.
Thinking about the build attributes, the combination of a
single-precision only implementation of no floating-point architecture
is still no floating-point architecture. Hence the assertion on the
input BFD in the linker makes no real sense.
We should, however, be more conservative in what we generate, so I've
left the assertion on the output bfd in place; I don't think we can
trigger it with this change since we never merge the problematic tags
from a perversely generated input file.
* elf32-arm.c (elf32_arm_merge_eabi_attributes): Remove assertion
that the input bfd has Tag_FP_ARCH non-zero if Tag_ABI_HardFP_use
is non-zero. Add clarifying comments.
