Tom Tromey [Thu, 15 Apr 2021 16:14:11 +0000 (10:14 -0600)]
Avoid crash in Ada value printing with optimized-out array
The Ada value-printing code could crash when printing an array which
had been optimized out.
The crash is difficult to reproduce, but I did manage to write a test
that at least shows that the previous behavior was incorrect -- before
the patch, the array is printed as if it is valid and every value is 0.
gdb/ChangeLog
2021-04-15 Tom Tromey <tromey@adacore.com>
Tom Tromey [Thu, 15 Apr 2021 16:14:11 +0000 (10:14 -0600)]
Use common_val_print_checked in print_variable_and_value
GDB was crashing in "bt full" for a large Ada program. I tracked this
down to a combination of bugs.
The first bug, fixed here, is that print_variable_and_value calls
common_val_print. Normally only the value-printing implementation
should call this; from the top level, common_val_print_checked is
preferred, because it correctly handles values that are wholly
optimized out.
I wasn't able to write a test case for this.
gdb/ChangeLog
2021-04-15 Tom Tromey <tromey@adacore.com>
* printcmd.c (print_variable_and_value): Use
common_val_print_checked.
Tom Tromey [Thu, 15 Apr 2021 16:05:00 +0000 (10:05 -0600)]
Use block_symbol in var_value_operation
I noticed that var_value_operation takes a block and a symbol, and
most callers destructure a block_symbol to pass in. It seems better
for this class to simply hold a block_symbol instead.
Tested on x86-64 Fedora 32.
gdb/ChangeLog
2021-04-15 Tom Tromey <tromey@adacore.com>
Tom Tromey [Sun, 5 Jul 2020 19:02:40 +0000 (13:02 -0600)]
gdb: add "set startup-quietly" command
This adds a new command to change GDB to behave as though "-quiet"
were always given. This new command can be added to the gdbearlyinit
file to affect future GDB sessions.
gdb/ChangeLog:
* NEWS: Add entry.
* main.c (captured_main_1): Call check_quiet_mode.
* top.c (startup_quiet): New global.
(check_quiet_mode): New function.
(show_startup_quiet): New function.
(init_main): Register new command.
* top.h (check_quiet_mode): Declare.
Andrew Burgess [Fri, 25 Sep 2020 15:28:05 +0000 (16:28 +0100)]
gdb: process early initialization files and command line options
Adds the ability to process commands at a new phase during GDB's
startup. This phase is earlier than the current initialisation file
processing, before GDB has produced any output.
The number of commands that can be processed at this early stage will
be limited, and it is expected that the only commands that would be
processed at this stage will relate to some of the fundamentals of how
GDB starts up.
Currently the only commands that it makes sense to add to this early
initialization file are those like 'set style version ....' as the
version string is displayed during startup before the standard
initialization files are parsed. As such this commit fully resolved
bug cli/25956.
This commit adds a mechanism to execute these early initialization
files from a users HOME directory, as well as some corresponding
command line flags for GDB.
The early initialization files that GDB will currently check for are
~/.config/gdb/gdbearlyinit (on Linux like systems) or ~/.gdbearlyinit
if the former is not found.
The output of 'gdb --help' has been extended to include a list of the
early initialization files being processed.
gdb/ChangeLog:
PR cli/25956
* NEWS: Mention new early init files and command line options.
* config.in: Regenerate.
* configure: Regenerate.
* configure.ac: Define GDBEARLYINIT.
* main.c (get_earlyinit_files): New function.
(enum cmdarg_kind): Add CMDARG_EARLYINIT_FILE and
CMDARG_EARLYINIT_COMMAND.
(captured_main_1): Add support for new command line flags, and for
processing startup files.
(print_gdb_help): Include startup files in the output.
gdb/doc/ChangeLog:
PR cli/25956
* gdb.texinfo (File Options): Mention new command line options.
(Startup): Discuss when early init files are processed.
(Initialization Files): Add description of early init files.
(Output Styling): Update description of 'version' style.
(gdb man): Mention early init files.
gdb/testsuite/ChangeLog:
PR cli/25956
* gdb.base/early-init-file.c: New file.
* gdb.base/early-init-file.exp: New file.
* lib/gdb-utils.exp (style): Handle style 'none'.
Andrew Burgess [Thu, 14 Jan 2021 14:32:35 +0000 (14:32 +0000)]
gdb: refactor the initialization file lookup code
In preparation for the next patch, which adds startup files, this
commit refactors the code for looking up the initialization files so
that the code can be more easily reused in the next commit.
There should be no user visible changes after this commit.
gdb/ChangeLog:
* main.c (relocate_gdbinit_path_maybe_in_datadir): Rename to...
(relocate_file_path_maybe_in_datadir): ...this.
(class gdb_initfile_finder): New class.
(get_init_files): Now uses gdb_initfile_finder.
(print_gdb_help): Print 'None found' when there are no init files.
Alan Modra [Wed, 14 Apr 2021 15:06:42 +0000 (00:36 +0930)]
PR27725, better objcopy -p times
Nanosecond rather than second resolution.
PR 27725
* configure.ac: Check for sys/time.h and utimensat. Use standard
checks for mkstemp and mkdtemp. Whitespace. Check for nanosecond
members of struct stat.
* rename.c: Prefer sys/time.h for utimes over utime.h for utime.
(STAT_TIMESPEC, STAT_TIMESPEC_NS): Define
(get_stat_atime_ns, get_stat_mtime_ns): New inline functions.
(get_stat_atime, get_stat_mtime): Likewise.
(set_times): Choose first available of utimensat, utimes, utime.
Use above inline functions to set timespec and timeval values.
* configure: Regenerate.
* config.in: Regenerate.
* testsuite/binutils-all/objcopy.exp (objcopy_test): Add test of
file timestamp when --preserve-dates is used.
Alan Modra [Wed, 14 Apr 2021 16:03:13 +0000 (01:33 +0930)]
Make objcopy -p work when an output file is specified
More fallout from the PR27456 fixes.
PR 27456
* rename.c (smart_rename): When TO and FROM are equal, just set
file timestamp.
* objcopy.c (strip_main, copy_main): Always call smart_rename.
Nelson Chu [Wed, 14 Apr 2021 02:00:05 +0000 (10:00 +0800)]
RISC-V: PR27584, surpress local and empty name symbols for nm.
The local ".L0" symbol are usually generated to mark where the
auipc instruction is, and clang also uses ".LBB" to do the same
thing. Besides, clang also uses the empty name symbols for label
differences. Arm surpress their mapping symbols by regarding them
as special symbols, and MIPS surpress all local symbols. Therefore,
RISC-V probably need to surpress the local and empty name symbols
for nm and related tools. User can use "--special-syms" to show
these surpressed symbols.
bfd/
PR27584
* elfnn-riscv.c (riscv_elf_is_target_special_symbol): Added to
omit local and empty symbols since they usually generated to
mark the auipc instructions.
The special_function for reloc should be 'bfd_elf_generic_reloc',
reloc list:
R_CKCORE_PCREL_IMM16BY2,
R_CKCORE_PCREL_IMM16BY4,
R_CKCORE_PCREL_IMM10BY4,
R_CKCORE_PCREL_IMM18BY2,
R_CKCORE_PCREL_FLRW_IMM8BY4
bfd/
* elf32-csky.c (csky_elf_howto_table)
<R_CKCORE_PCREL_IMM16BY2>: Set special_function to
bfd_elf_generic_reloc.
<R_CKCORE_PCREL_IMM16BY4>: Likewise.
<R_CKCORE_PCREL_IMM10BY4>: Likewise.
<R_CKCORE_PCREL_IMM18BY2>: Likewise.
<R_CKCORE_PCREL_FLRW_IMM8BY4>: Likewise.
This causes GDB to fail matching the concrete lexical scope with the
corresponding abstract entry. Hence, the local vars of the abstract
function that are contained in the lexical scope are read separately
(and thus, in addition to) the local vars of the concrete scope.
Because the abstract definitions of the vars do not contain location
information, we see the extra 'value = <optimized out>' above.
This bug is highly related to PR gdb/25695, but the root cause is not
exactly the same. In PR gdb/25695, GCC emits an extra
DW_TAG_lexical_block without an DW_AT_abstract_origin that wraps the
body of the inlined function. That is, the trees of the abstract DIE
for the function and its concrete instance are structurally not the
same. In the case of using Clang, the trees have the same structure.
To tackle the Clang case, when traversing the children of the concrete
instance root, keep a reference to the child of the abstract DIE that
corresponds to the concrete child, so that we can match the two DIEs
heuristically in case of missing DW_AT_abstract_origin attributes.
The updated gdb.opt/inline-locals.exp test has been checked with GCC
5-10 and Clang 5-11.
gdb/ChangeLog:
2021-04-14 Tankut Baris Aktemur <tankut.baris.aktemur@intel.com>
* dwarf2/read.c (inherit_abstract_dies): Keep a reference to the
corresponding child of the abstract DIE when iterating the
children of the concrete DIE.
gdb/testsuite/ChangeLog:
2021-04-14 Tankut Baris Aktemur <tankut.baris.aktemur@intel.com>
* gdb.opt/inline-locals.c (scoped): New function.
(main): Call 'scoped'.
* gdb.opt/inline-locals.exp: Update with "info locals" tests
for scoped variables.
* gdb.dwarf2/dw2-inline-with-lexical-scope.c: New file.
* gdb.dwarf2/dw2-inline-with-lexical-scope.exp: New file.
Andrew Burgess [Wed, 14 Apr 2021 12:14:27 +0000 (13:14 +0100)]
gdb/doc: add missing parentheses around prompt in some examples
While reading the manual for -info-os I noticed that the GDB prompt is
given as 'gdb' when it should really be '(gdb)'. This is because the
prompt is created with: '@value{GDBP}'.
The GDBP variable (the GDB program name) is intended for use as the
prompt string (though this is not used consistently throughout the
manual), however it is normally used like '(@value{GDBP})', but in a
couple of places the enclosing parentheses are missing.
In this commit I do the following:
- Change '@value{GDBP}' to '(@value{GDBP})' wherever the variable
represents a prompt string.
- Replaces '(gdb)' with '(@value{GDBP})' in one example where we are
already using '(@value{GDBP})', this makes that one example
consistent.
I have NOT:
- Changed all instances of '(gdb)' with '(@value{GDBP})', this would
be a huge change.
gdb/doc/ChangeLog:
* gdb.texinfo (GDB/MI Miscellaneous Commands): Add missing
parentheses to GDB prompt in example, and replace '(gdb)' with
'(@value{GDBP})' in one example where the latter was already in
use.
Alan Modra [Wed, 14 Apr 2021 03:12:27 +0000 (12:42 +0930)]
PR27716, build failure for msdosdjgpp: PATH_MAX undeclared
We shouldn't be using arbitrary limits like PATH_MAX in GNU programs.
This patch also fixes some memory leaks in readelf when processing
separate debug info.
PR 27716
binutils/
* objdump.c (show_line): Don't limit paths to PATH_MAX.
* readelf.c (struct filedata): Change program_interpreter from
a char array to a char pointer.
(process_program_headers): Sanity check PT_INTERP p_filesz.
Malloc program_interpreter using p_filesz and read directly from
file.
(process_dynamic_section): Check program_interpreter is non-NULL.
(free_filedata): New function, split out from..
(process_object): ..here.
(close_debug_file): Call free_filedata.
* sysdep.h: Don't include sys/param.h.
(PATH_MAX): Don't define.
* configure.ac: Don't check for sys/param.h.
* configure: Regenerate.
gprof/
* gprof.h (PATH_MAX): Don't define.
* corefile.c (core_create_line_syms): Don't use PATH_MAX for initial
file name size.
* source.c (annotate_source): Malloc file name buffer. Always
trim off "-ann" when dos 8.3 annotate file matches original.
* utils.c (print_name_only): Malloc file name buffer.
Alan Modra [Tue, 13 Apr 2021 23:53:12 +0000 (09:23 +0930)]
Re: PR27723, Internal error in select_cie_for_fde
Let's make sure what we allow in the CIE initial instructions and what
select_cie_for_fde compares for a match is always in sync. Also
correct the previous patch that allowed DW_CFA_GNU_window_save to be
part of the CIE initial instructions, which was likely a mistake.
Tom de Vries [Tue, 13 Apr 2021 15:38:53 +0000 (17:38 +0200)]
[gdb] Fix regoff_t incompatibility
I did an experiment with importing the regex module in gnulib, and trying to
build gdb.
The first problem I ran into was that:
- regoff_t was defined as long int, and
- the address of a regoff_t variable i in ui_file_style::parse was passed
as int * to function extended_color.
Fix this by changing the types of some function parameters of functions
read_semi_number and extended_color from int * to regoff_t *.
Tested on x86_64-linux.
gdb/ChangeLog:
2021-04-13 Tom de Vries <tdevries@suse.de>
* ui-style.c (read_semi_number, extended_color): Change idx parameter
type to regoff_t *.
Nelson Chu [Tue, 13 Apr 2021 04:32:16 +0000 (12:32 +0800)]
RISC-V: Don't report the mismatched version warning for the implicit extensions.
bfd/
* elfnn-riscv.c (riscv_version_mismatch): Do not report the warning
when the version of input or output is RISCV_UNKNOWN_VERSION, since
the extension is added implicitly.
* elfxx-riscv.c: Updated the obsolete comments.
(RISCV_UNKNOWN_VERSION): Moved to elfxx-riscv.h.
* elfxx-riscv.h (RISCV_UNKNOWN_VERSION): Added.
Will Schmidt [Mon, 12 Apr 2021 19:17:43 +0000 (14:17 -0500)]
[PATCH,rs6000] Fix vsx-regs.exp testcase failure
Hi,
This test exercise updates to the F* and VS* registers
and verifies updates to the same. Note that the registers
overlap; the doubleword[1] portion of any VS0-VS31
register contains the F0-F31 register contents, so any updates
to one can be measured in the other.
Per a brief investigation, we see that dl_main() currently
uses some VSX instructions, so the VS* values are not
going to be zero when this testcase reaches main, where these
tests begin. The test harness does not explicitly
initialize the full VS* values, so the first test loop
that updates the F* values means our VS* values are
uninitalized and will fail the first set of checks.
This update explicitly initializes the doubleword[0] portion
of the VS* registers, to allow this test to succeed.
2021-04-12 Will Schmidt <will_schmidt@vnet.ibm.com>
Will Schmidt [Mon, 12 Apr 2021 19:11:02 +0000 (14:11 -0500)]
[PATCH] gdb-power10-single-step
Hi,
This is based on a patch originally written by Alan Modra.
Powerpc / Power10 ISA 3.1 adds prefixed instructions, which
are 8 bytes in length. This is in contrast to powerpc previously
always having 4 byte instruction length. This patch implements
changes to allow GDB to better detect prefixed instructions, and
handle single stepping across the 8 byte instructions.
Added #defines to help test for PNOP and prefix instructions.
Update ppc_displaced_step_copy_insn() to handle pnop and prefixed
instructions whem R=0 (non-pc-relative).
Updated ppc_displaced_step_fixup() to properly handle the offset
value matching the current instruction size
Updated the for-loop within ppc_deal_with_atomic_sequence() to
count instructions properly in case we have a mix of 4-byte and
8-byte instructions within the atomic_sequence_length.
Added testcase and harness to exercise pc-relative load/store
instructions with R=0.
2021-04-12 Will Schmidt <will_schmidt@vnet.ibm.com>
gdb/ChangeLog:
* rs6000-tdep.c: Add support for single-stepping of
prefixed instructions.
Will Schmidt [Mon, 12 Apr 2021 18:35:54 +0000 (13:35 -0500)]
[PATCH, rs6000, v3][PR gdb/27525] displaced stepping across addpcis/lnia.
This addresses PR gdb/27525. The lnia and other variations
of the addpcis instruction write the value of the NIA into a target register.
If we are single-stepping across a breakpoint, the instruction is executed
from a displaced location, and thusly the written value of the PC/NIA
will be incorrect. The changes here will measure the displacement
offset, and adjust the target register value to compensate.
YYYY-MM-DD Will Schmidt <will_schmidt@vnet.ibm.com>
gdb/ChangeLog:
* rs6000-tdep.c (ppc_displaced_step_fixup): Update to handle
the addpcis/lnia instruction.
gdb/testsuite/ChangeLog:
* gdb.arch/powerpc-addpcis.exp: Testcase harness to
exercise single-stepping over subpcis,lnia,addpcis instructions
with displacement.
* gdb.arch/powerpc-addpcis.s: Testcase with stream
of addpcis/lnia/subpcis instructions.
* gdb.arch/powerpc-lnia.exp: Testcase harness to exercise
single-stepping over lnia instructions with displacement.
* gdb.arch/powerpc-lnia.s: Testcase with stream of
lnia instructions.
Alan Modra [Mon, 12 Apr 2021 15:02:11 +0000 (00:32 +0930)]
m68hc11 gas testsuite wart
Writing to a potentially read-only source directory is not good.
* testsuite/gas/m68hc11/m68hc11.exp (gas_m68hc11_message): Don't
write to $srcdir. Use gas_host_run and read output file rather
than gas_start/gas_finish.
Carl Love [Fri, 26 Mar 2021 21:54:47 +0000 (16:54 -0500)]
RS6000 Add support to print vector register contents as float128
This patch adds a floating point 128-bit composite field to the vsx
register type. When printing the register with p/f the float128 field will
be printed as a 128-bit floating point value. A test case to verify the new
vsx register field is visible and correctly prints out the value of a 128-bit
floating point value is also added.
Simon Marchi [Mon, 12 Apr 2021 15:10:57 +0000 (11:10 -0400)]
gdb, gdbserver: remove WinCE support code
The support for WinCE was removed with commit 84b300de3666 ("gdbserver:
remove support for ARM/WinCE"). There is some leftover code for WinCE
support, guarded by the _WIN32_WCE macro, which I didn't know of at the
time.
I didn't remove the _WIN32_WCE references in the tests, because in
theory we still support the WinCE architecture in GDB (when debugging
remotely). So someone could run a test with that (although I'd be
really surprised).
gdb/ChangeLog:
* nat/windows-nat.c: Remove all code guarded by _WIN32_WCE.
* nat/windows-nat.h: Likewise.
gdbserver/ChangeLog:
* win32-low.cc: Remove all code guarded by _WIN32_WCE.
* win32-low.h: Likewise.
In gdb.btrace/reconnect.exp, we test that we can disconnect and reconnect
again to a GDB session that is recording with the btrace recording format.
It does not really matter what we are recording.
The test assumed that stepping from _start will bring us into an area
without debug information. This is not correct on all systems.
Relax the expected output to also support systems where we do have debug
information for that code.
Alan Modra [Mon, 12 Apr 2021 07:46:58 +0000 (17:16 +0930)]
elf_backend_archive_symbol_lookup
elf_backend_archive_symbol_lookup might be called when the linker hash
table has entries of type generic_link_hash_entry. This happens for
instance when running the mmix target linker testsuite where the
output is mmo but input is elf64-mmix.
* elf-bfd.h (struct elf_backend_data): Return bfd_link_hash_entry*
from elf_backend_archive_symbol_lookup.
(_bfd_elf_archive_symbol_lookup): Return bfd_link_hash_entry*.
* elf64-ppc.c (ppc64_elf_archive_symbol_lookup): Likewise. Check
we have a ppc_hash_table before accessing ppc_link_hash_entry
fields.
* elflink.c (_bfd_elf_archive_symbol_lookup): Return
bfd_link_hash_entry*.
(elf_link_add_archive_symbols): Adjust to suit.
Nelson Chu [Mon, 12 Apr 2021 08:58:47 +0000 (16:58 +0800)]
RISC-V: Add i-ext as the implicit extension when e-ext is set.
The linker does not care the default versions of the extensions, since
it does not have the default ISA spec setting. Therefore, linker won't
insert the implicit extensions for the input objects. But we used to
insert the i-ext as the explicit extension, even if the e-ext is set.
This causes linker to report "cannot find default versions of the ISA
extension `i'" errors when linking the input objects with e-ext.
This patch fixes the above linker problem, and also remove the confused
riscv_ext_dont_care_version function. Unless these "dont care" extensions
are set in the input architecture explicitly, otherwise we always insert
them as the implicit ones. Afterwards, let riscv_arch_str1 surpress them
not to output to the architecture string if their versions are
RISCV_UNKNOWN_VERSION.
bfd/
* elfxx-riscv.c (riscv_ext_dont_care_version): Removed.
(riscv_parse_add_subset): Always add the implicit extensions, even if
their versions are RISCV_UNKNOWN_VERSION.
(riscv_parse_std_ext): Delay to add i-ext as the implicit extension
in the riscv_parse_add_implicit_subsets. Besides, add g-ext as the
implicit extension after it has been expanded.
(riscv_parse_add_implicit_subsets): Updated.
Alan Modra [Sun, 11 Apr 2021 23:32:46 +0000 (09:02 +0930)]
Power10 bignum operands
When built on a 32-bit host without --enable-64-bit-bfd, powerpc-linux
and other 32-bit powerpc targeted binutils fail to assemble some
power10 prefixed instructions with 34-bit fields. A typical error
seen when running the testsuite is
.../gas/testsuite/gas/ppc/prefix-pcrel.s:10: Error: bignum invalid
In practice this doesn't matter for addresses: 32-bit programs don't
need or use the top 2 bits of a d34 field when calculating addresses.
However it may matter when loading or adding 64-bit constants with
paddi. A power10 processor in 32-bit mode still has 64-bit wide GPRs.
So this patch enables limited support for O_big PowerPC operands, and
corrects sign extension of 32-bit constants using X_extrabit.
* config/tc-ppc.c (insn_validate): Use uint64_t for operand values.
(md_assemble): Likewise. Handle bignum operands.
(ppc_elf_suffix): Handle O_big. Remove unnecessary input_line_pointer
check.
* expr.c: Delete unnecessary forward declarations.
(generic_bignum_to_int32): Return uint32_t.
(generic_bignum_to_int64): Return uint64_t. Compile always.
(operand): Twiddle X_extrabit for unary '~'. Set X_unsigned and
clear X_extrabit for unary '!'.
* expr.h (generic_bignum_to_int32): Declare.
(generic_bignum_to_int64): Declare.
* testsuite/gas/ppc/prefix-pcrel.s,
* testsuite/gas/ppc/prefix-pcrel.d: Add more instructions.
Alan Modra [Sun, 11 Apr 2021 14:04:21 +0000 (23:34 +0930)]
PR27719, lang_mark_undefineds trashes memory
It's not enough to test that the output is ELF before casting
bfd_link_hash_entry to elf_link_hash_entry. Some ELF targets (d30v,
dlx, pj, s12z, xgate) use the generic linker support in bfd/linker.c
and thus their symbols are of type generic_link_hash_entry.
Not all of the places this patch touches can result in wrong accesses,
but I thought it worth ensuring that all occurrences of
elf_link_hash_entry in ld/ were obviously correct.
PR 27719
* ldlang.c (lang_mark_undefineds, undef_start_stop): Test that
the symbol hash table is the correct type before accessing
elf_link_hash_entry symbols.
* plugin.c (is_visible_from_outside): Likewise.
* emultempl/armelf.em (ld${EMULATION_NAME}_finish): Likewise.
* emultempl/solaris2.em (elf_solaris2_before_allocation): Likewise.
Nelson Chu [Wed, 7 Apr 2021 04:50:19 +0000 (12:50 +0800)]
RISC-V: Support to parse the multi-letter prefix in the architecture string.
The original discussion is as follows,
https://github.com/riscv/riscv-isa-manual/issues/637
I never considered the prefixes may have multiple letters, like zxm.
But the ISA spec has been updated for a long time that I haven't noticed.
This patch rewrites the part of architecture parser to support parsing
the multi-letter prefixes. Besides, I also improve the parser to report
errors in details. One of the most obvious improvement is - Do not parse
the prefixed extensions according to the orders in the parse_config.
If we do so, then we used to get "unexpected ISA string at end" errors,
but the message is a little bit hard to know what is happening. I Remove
the confused message, and let riscv_parse_prefixed_ext to report the details.
bfd/
* elfxx-riscv.c (riscv_std_z_ext_strtab): Moved forward.
(riscv_std_s_ext_strtab): Likewise.
(riscv_std_h_ext_strtab): Likewise.
(riscv_std_zxm_ext_strtab): Added for the zxm prefix.
(enum riscv_prefix_ext_class): Moved forward and renamed from
riscv_isa_ext_class. Reorder them according to the parsing order,
since the enum values are used to check the orders in the
riscv_compare_subsets.
(struct riscv_parse_prefix_config): Moved forward and renamed from
riscv_parse_config_t. Also removed the ext_valid_p field, the
related functions are replaced by riscv_valid_prefixed_ext.
(parse_config): Moved forward and updated. The more letters of the
prefix string, the more forward it must be defined. Otherwise, we
will get the wrong mapping when using strncmp in riscv_get_prefix_class.
(riscv_get_prefix_class): Moved forward. Support to parse the
multi-letter prefix, like zxm.
(riscv_known_prefixed_ext): New function, check if the prefixed
extension is supported according to the right riscv_std_*_ext_strtab.
(riscv_valid_prefixed_ext): New function, used to replace the
riscv_ext_*_valid_p functions.
(riscv_init_ext_order): Do not set the values for prefix keywords
since they may have multiple letters for now.
(riscv_compare_subsets): Set the order values of prefix keywords
to negative numbers according to the riscv_prefix_ext_class.
(riscv_parse_std_ext): Call riscv_get_prefix_class to see if we
have parsed the prefixed extensions.
(riscv_parse_prefixed_ext): Updated and removed the parameter config.
Report error when the prefix is unknown.
(riscv_parse_subset): Do not parse the prefixed extensions according
to the orders in the parse_config. Remove the confused message and
let riscv_parse_prefixed_ext to report the details.
* elfxx-riscv.h (enum riscv_isa_ext_class): Moved to elfxx-riscv.c.
(riscv_get_prefix_class): Removed to static.
gas/
* testsuite/gas/riscv/march-fail-order-x-std.d: Renamed from
march-fail-porder-x-std.d.
* testsuite/gas/riscv/march-fail-order-z-std.d: Renamed from
march-fail-porder-z-std.d.
* testsuite/gas/riscv/march-fail-order-x-z.d: Renamed from
march-fail-porder-x-z.d.
* testsuite/gas/riscv/march-fail-order-zx-std.l: Added to replace
march-fail-porder.l.
* testsuite/gas/riscv/march-fail-order-x-z.l: Likewise.
* testsuite/gas/riscv/march-fail-order-x.l: Updated.
* testsuite/gas/riscv/march-fail-order-z.l: Likewise.
* testsuite/gas/riscv/march-fail-single-prefix-h.d: Renamed from
march-fail-single-char-h.d.
* testsuite/gas/riscv/march-fail-single-prefix-s.d: Renamed from
march-fail-single-char-s.d.
* testsuite/gas/riscv/march-fail-single-prefix-x.d: Renamed from
march-fail-single-char-x.d.
* testsuite/gas/riscv/march-fail-single-prefix-z.d: Renamed from
march-fail-single-char-z.d.
* testsuite/gas/riscv/march-fail-single-prefix-zmx.d: Added.
* testsuite/gas/riscv/march-fail-single-prefix.l: Added to replace
march-fail-single-prefix.l.
* testsuite/gas/riscv/march-fail-unknown-zxm.d: Added.
* testsuite/gas/riscv/march-fail-unknown-std.l: Updated.
* testsuite/gas/riscv/march-fail-unknown.l: Likewise.
Eli Zaretskii [Sun, 11 Apr 2021 18:37:29 +0000 (21:37 +0300)]
Improve support for loading DLLs at run time in gdbserver.
This fixes win32-low.cc in the same way as a recent change in
windows-nat.c did for GDB: if the lpImageName member of the load-DLL
debug event doesn't allow us to find the file name of the DLL, then
loop over all the DLLs mapped into the inferior to find the one loaded
at the same base address as given by the lpBaseOfDll member of the
debug event.
gdbserver/ChangeLog:
2021-04-11 Eli Zaretskii <eliz@gnu.org>
* win32-low.cc (win32_add_dll): New function, with body almost
identical to what win32_add_all_dlls did. Accepts one argument;
if that is non-NULL, returns the file name of the DLL that is
loaded at the base address equal to that argument, or NULL if not
found. If the argument is NULL, add all the DLLs loaded by the
inferior to the list of solibs and return NULL.
(win32_add_all_dlls): Now a thin wrapper around win32_add_dll.
(windows_nat::handle_load_dll) [!_WIN32_WCE]: If get_image_name
failed to glean the file name of the DLL, call win32_add_dll to
try harder using the lpBaseOfDll member of the load-DLL event.
Eli Zaretskii [Sat, 10 Apr 2021 08:33:08 +0000 (11:33 +0300)]
Fix handling DLL loads at run time
This patch makes handling a DLL load at run time (using LoadLibrary)
much more reliable when its file name cannot be obtained using the
lpImageName pointer provided by the DLL load debug event. The
solution is to enumerate all the DLLs loaded by the inferior, looking
for the DLL that's loaded at base address provided by the lpBaseOfDll
pointer of the debug event. Correctly resolving the DLL file name is
important, because without that GDB doesn't record the DLL in the list
of solibs, and then later is unable to show functions in that DLL in
the backtraces, which produces corrupted and truncated backtraces.
See this thread for the problems that causes:
* windows-nat.c (windows_nat::handle_load_dll): Call
windows_add_dll if get_image_name failed to glean the name of the
DLL by using the lpImageName pointer.
(windows_add_all_dlls): Now a thin wrapper around windows_add_dll.
(windows_add_dll): Now does what windows_add_all_dlls did before,
but also accepts an argument LOAD_ADDR, which, if non-NULL,
specifies the address where the DLL was loaded into the inferior,
and looks for the single DLL loaded at that address.
AArch64: Fix Diagnostic messaging for LD/ST Exclusive.
A summary of what this patch set fixes:
For instructions
STXR w0,x2,[x0]
STLXR w0,x2,[x0]
The warning we emit currently is misleading:
Warning: unpredictable: identical transfer and status registers --`stlxr w0,x2,[x0]'
Warning: unpredictable: identical transfer and status registers --`stxr w0,x2,[x0]'
it ought to be:
Warning: unpredictable: identical base and status registers --`stlxr w0,x2,[x0]'
Warning: unpredictable: identical base and status registers --`stxr w0,x2,[x0]'
For instructions:
ldaxp x0,x0,[x0]
ldxp x0,x0,[x0]
The warning we emit is incorrect
Warning: unpredictable: identical transfer and status registers --`ldaxp x0,x0,[x0]'
Warning: unpredictable: identical transfer and status registers --`ldxp x0,x0,[x0]'
it ought to be:
Warning: unpredictable load of register pair -- `ldaxp x0,x0,[x0]'
Warning: unpredictable load of register pair -- `ldxp x0,x0,[x0]'
For instructions
stlxp w0, x2, x2, [x0]
stxp w0, x2, x2, [x0]
We don't emit any warning when it ought to be:
Warning: unpredictable: identical base and status registers --`stlxp w0,x2,x2,[x0]'
Warning: unpredictable: identical base and status registers --`stxp w0,x2,x2,[x0]'
gas/ChangeLog:
2021-04-09 Tejas Belagod <tejas.belagod@arm.com>
* config/tc-aarch64.c (warn_unpredictable_ldst): Clean-up diagnostic messages
for LD/ST Exclusive instructions.
* testsuite/gas/aarch64/diagnostic.s: Add a diagnostic test for STLXP.
* testsuite/gas/aarch64/diagnostic.l: Fix-up test after message clean-up.
Warning: unpredictable: identical transfer and status registers --`stlxr w0,x2,[x0]'
Warning: unpredictable: identical transfer and status registers --`stxr w0,x2,[x0]'
it ought to be:
Warning: unpredictable: identical base and status registers --`stlxr w0,x2,[x0]'
Warning: unpredictable: identical base and status registers --`stxr w0,x2,[x0]'
For instructions:
ldaxp x0,x0,[x0]
ldxp x0,x0,[x0]
The warning we emit is incorrect
Warning: unpredictable: identical transfer and status registers --`ldaxp x0,x0,[x0]'
Warning: unpredictable: identical transfer and status registers --`ldxp x0,x0,[x0]'
it ought to be:
Warning: unpredictable load of register pair -- `ldaxp x0,x0,[x0]'
Warning: unpredictable load of register pair -- `ldxp x0,x0,[x0]'
For instructions
stlxp w0, x2, x2, [x0]
stxp w0, x2, x2, [x0]
We don't emit any warning when it ought to be:
Warning: unpredictable: identical base and status registers --`stlxp w0,x2,x2,[x0]'
Warning: unpredictable: identical base and status registers --`stxp w0,x2,x2,[x0]'
For instructions:
st64bv x0, x2, [x0]
st64bv x2, x0, [x0]
We incorrectly warn when its not necessary. This is because we classify them
incorrectly as ldstexcl when it should be lse_atomics in the opcode table.
The incorrect classification makes it pick up the warnings from warning on
exclusive load/stores.
Patch 2: Reclassify Armv8.7-A ST/LD64 Atomics.
This patch reclassifies ST64B{V,V0}, LD64B as lse_atomics rather than ldstexcl
according to their encoding class as specified in the architecture. This also
has the fortunate side-effect of spurious unpredictable warnings getting
eliminated.
For eg. For instruction:
st64bv x0, x2, [x0]
We incorrectly warn when its not necessary:
Warning: unpredictable: identical transfer and status registers --`st64bv x0,x2,[x0]'
This is because we classify them incorrectly as ldstexcl when it should be
lse_atomics in the opcode table. The incorrect classification makes it pick
up the warnings from warning on exclusive load/stores. This patch fixes it
by reclassifying it and no warnings are issued for this instruction.
opcodes/ChangeLog:
2021-04-09 Tejas Belagod <tejas.belagod@arm.com>
* aarch64-tbl.h (struct aarch64_opcode aarch64_opcode_table): Reclassify
LD64/ST64 instructions to lse_atomic instead of ldstexcl.
Alan Modra [Tue, 6 Apr 2021 09:33:35 +0000 (19:03 +0930)]
PowerPC disassembly of pcrel references
This adds some annotation to Power10 pcrel instructions, displaying
the target address (ie. pc + D34 field) plus a symbol if there is one
at exactly that target address. pld from the .got or .plt will also
look up the entry and display it, symbolically if there is a dynamic
relocation on the entry.
include/
* dis-asm.h (struct disassemble_info): Add dynrelbuf and dynrelcount.
binutils/
* objdump.c (struct objdump_disasm_info): Delete dynrelbuf and
dynrelcount.
(find_symbol_for_address): Adjust for dynrelbuf and dynrelcount move.
(disassemble_section, disassemble_data): Likewise.
opcodes/
* ppc-dis.c (struct dis_private): Add "special".
(POWERPC_DIALECT): Delete. Replace uses with..
(private_data): ..this. New inline function.
(disassemble_init_powerpc): Init "special" names.
(skip_optional_operands): Add is_pcrel arg, set when detecting R
field of prefix instructions.
(bsearch_reloc, print_got_plt): New functions.
(print_insn_powerpc): For pcrel instructions, print target address
and symbol if known, and decode plt and got loads too.
gas/
* testsuite/gas/ppc/prefix-pcrel.d: Update expected output.
* testsuite/gas/ppc/prefix-reloc.d: Likewise.
* gas/testsuite/gas/ppc/vsx_32byte.d: Likewise.
ld/
* testsuite/ld-powerpc/inlinepcrel-1.d: Update expected output.
* testsuite/ld-powerpc/inlinepcrel-2.d: Likewise.
* testsuite/ld-powerpc/notoc2.d: Likewise.
* testsuite/ld-powerpc/notoc3.d: Likewise.
* testsuite/ld-powerpc/pcrelopt.d: Likewise.
* testsuite/ld-powerpc/startstop.d: Likewise.
* testsuite/ld-powerpc/tlsget.d: Likewise.
* testsuite/ld-powerpc/tlsget2.d: Likewise.
* testsuite/ld-powerpc/tlsld.d: Likewise.
* testsuite/ld-powerpc/weak1.d: Likewise.
* testsuite/ld-powerpc/weak1so.d: Likewise.
Tom Tromey [Thu, 8 Apr 2021 21:15:59 +0000 (15:15 -0600)]
Avoid sequence point warning in h8300 sim
GCC gives a -Wsequence-point warning for this code in the h8300 sim.
The bug is that memory_size is both assigned and used in the same
expression. The fix is to assign after the print.
sim/h8300/ChangeLog
2021-04-08 Tom Tromey <tom@tromey.com>
* compile.c (init_pointers): Fix sequence point warning.
Tom Tromey [Thu, 8 Apr 2021 14:52:50 +0000 (08:52 -0600)]
Add system includes in sim
This updates various parts of the sim to include missing system
headers. I made the includes unconditional, because other parts of
the tree are already doing this.
2021-04-08 Tom Tromey <tom@tromey.com>
* traps.c: Include stdlib.h.
* cris-tmpl.c: Include stdlib.h.
sim/erc32/ChangeLog
2021-04-08 Tom Tromey <tom@tromey.com>
* func.c: Include sys/time.h.
sim/frv/ChangeLog
2021-04-08 Tom Tromey <tom@tromey.com>
* traps.c: Include stdlib.h.
* registers.c: Include stdlib.h.
* profile.c: Include stdlib.h.
* memory.c: Include stdlib.h.
* interrupts.c: Include stdlib.h.
* frv.c: Include stdlib.h.
* cache.c: Include stdlib.h.
sim/iq2000/ChangeLog
2021-04-08 Tom Tromey <tom@tromey.com>
* iq2000.c: Include stdlib.h.
sim/m32r/ChangeLog
2021-04-08 Tom Tromey <tom@tromey.com>
* traps.c: Include stdlib.h.
* m32r.c: Include stdlib.h.
sim/ppc/ChangeLog
2021-04-08 Tom Tromey <tom@tromey.com>
Luis Machado [Wed, 7 Apr 2021 19:20:44 +0000 (16:20 -0300)]
Fix DTB generation mechanism and build failure
I ran into a build failure with --enable-targets=all due to the fact that
the moxie sim expects to be able to use the dtc tool. If it isn't available,
the builds fails.
The following patch adds a prebuilt dtb file to the tree. That file is the one
that is used for installations.
The patch also enables (re-)generation of the dtb file through maintainer
mode, if it needs to be updated due to a change in the dts file.
Simon Marchi [Thu, 8 Apr 2021 13:49:30 +0000 (09:49 -0400)]
sim: set ASAN_OPTIONS=detect_leaks=0 when running igen and opc2c
The igen/dgen and opc2c tools leak their heap-allocated memory (on
purpose) at program exit, which makes AddressSanitizer fail the tool
execution. This breaks the build, as it makes the tool return a
non-zero exit code.
Fix that by disabling leak detection through the setting of that
environment variable.
I also changed the opc2c rules for m32c to go through a temporary file.
What happened is that the failing opc2c would produce an incomplete file
(probably because ASan exits the process before stdout is flushed).
This meant that further make attempts didn't try to re-create the file,
as it already existed. A "clean" was therefore necessary. This can
also happen in regular builds if the user interrupts the build (^C) in
the middle of the opc2c execution and tries to resume it. Going to a
temporary file avoids this issue.
sim/m32c/ChangeLog:
* Makefile.in: Set ASAN_OPTIONS when running opc2c.
sim/mips/ChangeLog:
* Makefile.in: Set ASAN_OPTIONS when running igen.
sim/mn10300/ChangeLog:
* Makefile.in: Set ASAN_OPTIONS when running igen.
sim/ppc/ChangeLog:
* Makefile.in: Set ASAN_OPTIONS when running igen.
sim/v850/ChangeLog:
* Makefile.in: Set ASAN_OPTIONS when running igen.
gdb: Allow prologue detection via symbols for Intel compilers.
The next-gen Intel Fortran compiler isn't flang-based, but emits
prologue_end in the same manner. As do the newer Intel C/C++ compilers.
This allows prologue detection based on dwarf for all newer Intel compilers.
The cut-off version was not chosen for any specific reason other than the
effort to test this.
gdb/Changelog:
2021-04-08 Felix Willgerodt <felix.willgerodt@intel.com>
* i386-tdep.c (i386_skip_prologue): Use symbol table to find the
prologue end for Intel compilers.
* amd64-tdep.c (amd64_skip_prologue): Likewise.
* producer.c (producer_is_icc_ge_19): New function.
* producer.h (producer_is_icc_ge_19): New declaration.
The main goal of this patch is to get rid of a warning for the new Fortran
compiler:
(gdb) b 9
warning: Could not recognize version of Intel Compiler in: "Intel(R) Fortran 21.0-2087b"
Breakpoint 1 at 0x4048cf: file comp.f90, line 9.
While trying to fix this I analyzed DW_AT_producer of all latest Intel
compilers for C, C++ and Fortran. They do no longer necessarily start with
"Intel (R)" nor do they follow the internal and external version number
scheme that the original patch for this check assumed. Some newer compilers
even contradict the "intermediate" digit in the old version scheme and have
the MINOR number as the second digit, even when having 3 or 4 digits overall.
Therefore I rewrote the check to consider the first MAJOR.MINOR string found
as the version number. This might not be 100% correct for some older
internal compilers, but the only current user of this function is only
checking for the major version anyway. Hence this should be reliable enough
and extendable enough going forward.
gdb/ChangeLog:
2021-04-08 Felix Willgerodt <felix.willgerodt@intel.com>
* producer.c: (producer_is_icc): Update for new version scheme.
(producer_parsing_tests): Update names and expected results.
* producer.h: (producer_is_icc): Update comment accordingly.
Mike Frysinger [Sun, 4 Apr 2021 00:58:14 +0000 (20:58 -0400)]
sim: testsuite: support exit 77 for unsupported tests
Exit status 77 is common (including the autotools world) to indicate
"skip this test". Add support for mapping that to "unsupported" as
that's the closest in the dejagnu world.
Mike Frysinger [Sun, 4 Apr 2021 13:05:21 +0000 (09:05 -0400)]
sim: testsuite: skip tests when the port is disabled
If the port hasn't been enabled, don't try to run its tests. Making
this dynamic simplifies the test harnesses and avoids duplicating a
bunch of target tuple checks.
Jim Wilson [Thu, 8 Apr 2021 01:51:52 +0000 (18:51 -0700)]
Aarch64 sim fix for gcc-10 miscompilation.
This fixes a problem that occurs when compiled by gcc-10, as the code
is relying on undefined overflow behavior. This is fixed by replacing
compares between 32-bit and 64-bit results with compares that just use
the 64-bit results with a cast.
PR sim/27483
* simulator.c (set_flags_for_add32): Compare uresult against
itself. Compare sresult against itself.
Note that this doesn't implement the ISA to the letter regarding
dcbtds (and dcbtstds), which says that the TH field may be zero. That
doesn't make sense because allowing TH=0 would mean you no long have a
dcbtds but rather a dcbtct instruction. I'm interpreting the ISA
wording about allowing TH=0 to mean that the TH field of dcbtds is
optional (in which case the TH value is 0b1000).
opcodes/
PR 27676
* ppc-opc.c (DCBT_EO): Move earlier.
(insert_thct, extract_thct, insert_thds, extract_thds): New functions.
(powerpc_operands): Add THCT and THDS entries.
(powerpc_opcodes): Add dcbtstct, dcbtstds, dcbna, dcbtct, dcbtds.
gas/
* testsuite/gas/ppc/pr27676.d,
* testsuite/gas/ppc/pr27676.s: New test.
* testsuite/gas/ppc/ppc.exp: Run it.
* testsuite/gas/ppc/dcbt.d: Update.
* testsuite/gas/ppc/power4_32.d: Update.
Weimin Pan [Wed, 7 Apr 2021 18:07:48 +0000 (14:07 -0400)]
CTF: handle forward reference type
Added function fetch_tid_type which calls get_tid_type and will set up
the type, associated with a tid, if it is not read in yet. Also implement
function read_forward_type which handles the CTF_K_FORWARD kind.
Expanded gdb.base/ctf-ptype.exp to add cases with forward references.
gdb/ChangeLog:
* ctfread.c (fetch_tid_type): New function, use throughout file.
(read_forward_type): New function.
(read_type_record): Call read_forward_type.
gdb/testsuite/ChangeLog:
* gdb.base/ctf-ptype.c: Add struct link containing a forward
reference type.
* gdb.base/ctf-ptype.exp: Add "ptype struct link".
The motivation behind the original patch can be seen in the new test,
which currently gives a GDB session like this:
(gdb) ptype var8
type = Type type6
PTR TO -> ( Type type2 :: ptr_1 )
PTR TO -> ( Type type2 :: ptr_2 )
End Type type6
(gdb) ptype var8%ptr_2
type = PTR TO -> ( Type type2
integer(kind=4) :: spacer
Type type1, allocatable :: t2_array(:) <------ Issue #1
End Type type2 )
(gdb) ptype var8%ptr_2%t2_array
Cannot access memory at address 0x38 <------ Issue #2
(gdb)
Issue #1: Here we see the abstract dynamic type, rather than the
resolved concrete type. Though in some cases the user might be
interested in the abstract dynamic type, I think that in most cases
showing the resolved concrete type will be of more use. Plus, the
user can always figure out the dynamic type (by source code inspection
if nothing else) given the concrete type, but it is much harder to
figure out the concrete type given only the dynamic type.
Issue #2: In this example, GDB evaluates the expression in
EVAL_AVOID_SIDE_EFFECTS mode (due to ptype). The value returned for
var8%ptr_2 will be a non-lazy, zero value of the correct dynamic
type. However, when GDB asks about the type of t2_array this requires
GDB to access the value of var8%ptr_2 in order to read the dynamic
properties. As this value was forced to zero (thanks to the use of
EVAL_AVOID_SIDE_EFFECTS) then GDB ends up accessing memory at a base
of zero plus some offset.
Both this patch, and my previous two attempts, have all tried to
resolve this problem by stopping EVAL_AVOID_SIDE_EFFECTS replacing the
result value with a zero value in some cases.
This new patch is influenced by how Ada handles its tagged typed.
There are plenty of examples in ada-lang.c, but one specific case is
ada_structop_operation::evaluate. When GDB spots that we are dealing
with a tagged (dynamic) type, and we're in EVAL_AVOID_SIDE_EFFECTS
mode, then GDB re-evaluates the child operation in EVAL_NORMAL mode.
This commit handles two cases like this specifically for Fortran, a
new fortran_structop_operation, and the already existing
fortran_undetermined, which is where we handle array accesses.
In these two locations we spot when we are dealing with a dynamic type
and re-evaluate the child operation in EVAL_NORMAL mode so that we
are able to access the dynamic properties of the type.
The rest of this commit message is my attempt to record why my
previous patches failed.
To understand my second patch, and why it failed lets consider two
expressions, this Fortran expression:
In expression <A> we should assume that t2_array is of dynamic type.
Nothing has dynamic type in expression <B>.
This is how GDB currently handles expression <A>, in all cases,
EVAL_AVOID_SIDE_EFFECTS or EVAL_NORMAL, an OP_VAR_VALUE operation
always returns the real value of the symbol, this is not forced to a
zero value even in EVAL_AVOID_SIDE_EFFECTS mode. This means that (3),
(5), and (8) will always return a real lazy value for the symbol.
However a STRUCTOP_STRUCT will always replace its result with a
non-lazy, zero value with the same type as its result. So (2) will
lookup the field ptr_2 and create a zero value with that type. In
this case the type is a pointer to a dynamic type.
Then, when we evaluate (1) to figure out the resolved type of
t2_array, we need to read the types dynamic properties. These
properties are stored in memory relative to the objects base address,
and the base address is in var8%ptr_2, which we already figured out
has the value zero. GDB then evaluates the DWARF expressions that
take the base address, add an offset and dereference. GDB then ends
up trying to access addresses like 0x16, 0x8, etc.
To fix this, I proposed changing STRUCTOP_STRUCT so that instead of
returning a zero value we instead returned the actual value
representing the structure's field in the target. My thinking was
that GDB would not try to access the value's contents unless it needed
it to resolve a dynamic type. This belief was incorrect.
Consider expression <B>. We already know that (5) and (8) will return
real values for the symbols being referenced. The BINOP_LOGICAL_AND,
operation (4) will evaluate both of its children in
EVAL_AVOID_SIDE_EFFECTS in order to get the types, this is required
for C++ operator lookup. This means that even if the value of (5)
would result in the BINOP_LOGICAL_AND returning false (say, ptr is
NULL), we still evaluate (6) in EVAL_AVOID_SIDE_EFFECTS mode.
Operation (6) will evaluate both children in EVAL_AVOID_SIDE_EFFECTS
mode, operation (9) is easy, it just returns a value with the constant
packed into it, but (7) is where the problem lies. Currently in GDB
this STRUCTOP_STRUCT will always return a non-lazy zero value of the
correct type.
When the results of (7) and (9) are back in the BINOP_LOGICAL_AND
operation (6), the two values are passed to value_equal which performs
the comparison and returns a result. Note, the two things compared
here are the immediate value (9), and a non-lazy zero value from (7).
However, with my proposed patch operation (7) no longer returns a zero
value, instead it returns a lazy value representing the actual value
in target memory. When we call value_equal in (6) this code causes
GDB to try and fetch the actual value from target memory. If `ptr` is
NULL then this will cause GDB to access some invalid address at an
offset from zero, this will most likely fail, and cause GDB to throw
an error instead of returning the expected type.
And so, we can now describe the problem that we're facing. The way
GDB's expression evaluator is currently written we assume, when in
EVAL_AVOID_SIDE_EFFECTS mode, that any value returned from a child
operation can safely have its content read without throwing an
error. If child operations start returning real values (instead of
the fake zero values), then this is simply not true.
If we wanted to work around this then we would need to rewrite almost
all operations (I would guess) so that EVAL_AVOID_SIDE_EFFECTS mode
does not cause evaluation of an operation to try and read the value of
a child operation. As an example, consider this current GDB code from
eval.c:
struct value *
eval_op_equal (struct type *expect_type, struct expression *exp,
enum noside noside, enum exp_opcode op,
struct value *arg1, struct value *arg2)
{
if (binop_user_defined_p (op, arg1, arg2))
{
return value_x_binop (arg1, arg2, op, OP_NULL, noside);
}
else
{
binop_promote (exp->language_defn, exp->gdbarch, &arg1, &arg2);
int tem = value_equal (arg1, arg2);
struct type *type = language_bool_type (exp->language_defn,
exp->gdbarch);
return value_from_longest (type, (LONGEST) tem);
}
}
We could change this function to be this:
struct value *
eval_op_equal (struct type *expect_type, struct expression *exp,
enum noside noside, enum exp_opcode op,
struct value *arg1, struct value *arg2)
{
if (binop_user_defined_p (op, arg1, arg2))
{
return value_x_binop (arg1, arg2, op, OP_NULL, noside);
}
else
{
struct type *type = language_bool_type (exp->language_defn,
exp->gdbarch);
if (noside == EVAL_AVOID_SIDE_EFFECTS)
return value_zero (type, VALUE_LVAL (arg1));
else
{
binop_promote (exp->language_defn, exp->gdbarch, &arg1, &arg2);
int tem = value_equal (arg1, arg2);
return value_from_longest (type, (LONGEST) tem);
}
}
}
Now we don't call value_equal unless we really need to. However, we
would need to make the same, or similar change to almost all
operations, which would be a big task, and might not be a direction we
wanted to take GDB in.
So, for now, I'm proposing we go with the more targeted, Fortran
specific solution, that does the minimal required in order to
correctly resolve the dynamic types.
gdb/ChangeLog:
* f-exp.h (class fortran_structop_operation): New class.
* f-exp.y (exp): Create fortran_structop_operation instead of the
generic structop_operation.
* f-lang.c (fortran_undetermined::evaluate): Re-evaluate
expression as EVAL_NORMAL if the result type was dynamic so we can
extract the actual array bounds.
(fortran_structop_operation::evaluate): New function.
gdb/testsuite/ChangeLog:
* gdb.fortran/dynamic-ptype-whatis.exp: New file.
* gdb.fortran/dynamic-ptype-whatis.f90: New file.
Andrew Burgess [Tue, 9 Mar 2021 11:11:14 +0000 (11:11 +0000)]
gdb: allow casting to rvalue reference in more cases
It is not currently possible to cast some values to an rvaule
reference. This happens when simple scalar values are cast to an
rvalue reference of the same type, e.g.:
int global_var;
Then in GDB:
(gdb) p static_cast<int&&> (global_var)
Attempt to take address of value not located in memory.
Which is clearly silly.
The problem is that as part of the cast an intermediate value is
created within GDB that becomes an lval_none rather than the original
lval_memory. The casting logic basically goes like this:
The call tree that leads to the error looks like this:
value_cast
value_cast
value_ref
value_addr
error
The first value_cast call is casting the value for 'global_var' to
type 'int&&'. GDB spots that the target type is a reference, and so
calls value_cast again, this time casting 'global_var' to type 'int'.
We then call value_ref to convert the result of the second value_cast
into a reference.
Unfortunately, the second cast results in the value (for global_var)
changing from an lval_memory to an lval_none. This is because int to
int casting calls extract_unsigned_integer and then
value_from_longest.
In theory value_cast has a check at its head that should help in this
case, the code is:
if (value_type (arg2) == type)
return arg2;
However, this only works in some cases. In our case
'value_type (arg2)' will be an objfile owned type, while the type from
the expression parser 'int&&' will be gdbarch owned. The pointers
will not be equal, but the meaning of the type will be equal.
I did consider making the int to int casting case smarter, but this
obviously is only one example. We must also consider things like
float to float, or pointer to pointer....
So, I instead decided to try and make the initial check smarter.
Instead of a straight pointer comparison, I now propose that we use
types_deeply_equal. If this is true then we are casting something
back to its current type, in which case we can preserve the lval
setting by using value_copy.
gdb/ChangeLog:
* valops.c (value_cast): Call value_deeply_equal before performing
any cast.
gdb/testsuite/ChangeLog:
* gdb.cp/rvalue-ref-params.cc (f3): New function.
(f4): New function.
(global_int): New global variable.
(global_float): Likeiwse.
(main): Call both new functions.
* gdb.cp/rvalue-ref-params.exp: Add new tests.
Andrew Burgess [Wed, 24 Mar 2021 17:48:27 +0000 (17:48 +0000)]
gdb: move cheap pointer equality check earlier in types_equal
I noticed that in types equal we start with a cheap pointer equality
check, then resolve typedefs, then do a series of (semi-)expensive
checks, including checking type names, before, finally performing
another pointer equality check.
We should hoist the second pointer equality check to immediately after
we have resolved typedefs. This would save performing the more
expensive checks.
This isn't going to give any noticable performance improvement, I just
spotted this in passing and figured I might as well commit the fix.
There should be no user visible changes after this commit.
gdb/ChangeLog:
* gdbtypes.c (types_equal): Move pointer equality check earlier in
the function.
Direct leak of 4 byte(s) in 1 object(s) allocated from:
#0 0x7ffff7677459 in __interceptor_malloc /build/gcc/src/gcc/libsanitizer/asan/asan_malloc_linux.cpp:145
#1 0x55555555b3df in main /home/simark/src/binutils-gdb/sim/m32c/opc2c.c:658
#2 0x7ffff741fb24 in __libc_start_main (/usr/lib/libc.so.6+0x27b24)
Fix the leak in main by removing the vlist variable, which seems unused.
Caroline Tice [Wed, 17 Mar 2021 22:54:22 +0000 (22:54 +0000)]
gdb: handle relative paths to DWO files
DWARF allows .dwo file paths to be relative rather than absolute.
When they are relative, DWARF uses DW_AT_comp_dir to find the .dwo
file. DW_AT_comp_dir can also be relative, making the entire search
patch for the .dwo file relative.
In this case, GDB currently searches relative to its current working
directory, i.e. the directory from which the debugger was launched,
but not relative to the directory containing the built binary. This
cannot be right, as the compiler, when generating the relative paths,
knows where it's building the binary but can have no idea where the
debugger will be launched.
The correct thing is to add the directory containing the binary to the
search paths used for resolving relative locations of dwo files. That
is what this patch does.
gdb/ChangeLog:
* dwarf2/read.c (try_open_dwop_file): Add path for the binary to
the search paths used resolve relative location of .dwo file.
gdb/testsuite/ChangeLog:
* gdb.dwarf2/fission-relative-dwo.c: New file.
* gdb.dwarf2/fission-relative-dwo.exp: New file.
Andrew Burgess [Fri, 26 Mar 2021 12:06:37 +0000 (12:06 +0000)]
gdb/testsuite: fix fission support in the Dwarf assembler
This commit fixes fission support in the Dwarf assembler. I added the
new test gdb.dwarf2/fission-absolute-dwo.exp which is a simple example
of using the fission support. I also rewrote the existing test
gdb.dwarf2/fission-multi-cu.exp to use the new functionality (instead
of using an x86-64 only assembler file).
To better support compiling the assembler files produced by the Dwarf
assembler I have added the new proc build_executable_and_dwo_files in
lib/dwarf.exp, this replaces build_executable_from_fission_assembler,
all the tests that used the old proc have been updated. Where the old
proc assumed a single .S source file which contained the entire test,
the new proc allows for multiple source files.
The Dwarf assembler already had some fission support, however, this
was not actually used in any tests, and when I tried using it there
were a few issues.
The biggest change is that we now generate DW_FORM_GNU_addr_index
instead of DW_FORM_addr for the low and high pc in
_handle_macro_at_range, support for the DW_FORM_GNU_addr_index is new
in this commit.
gdb/testsuite/ChangeLog:
* gdb.dwarf2/fission-absolute-dwo.c: New file.
* gdb.dwarf2/fission-absolute-dwo.exp: New file.
* gdb.dwarf2/fission-base.exp: Use build_executable_and_dwo_files
instead of build_executable_from_fission_assembler.
* gdb.dwarf2/fission-loclists-pie.exp: Likewise.
* gdb.dwarf2/fission-loclists.exp: Likewise.
Andrew Burgess [Mon, 22 Mar 2021 16:37:39 +0000 (16:37 +0000)]
gdb: Handle missing .debug_str section
While messing with the Dwarf assembler (gdb/testsuite/lib/dwarf.exp) I
managed to create an ELF which made use of DW_FORM_strp, but didn't
include a .debug_str section.
When I started GDB on this ELF, GDB crashed. I would have expected to
get an error instead.
I tracked this down to an unfortunate design choice in
dwarf2_section_info, a class which wraps around a bfd section, and is
used for reading in debug information. GBB creates many
dwarf2_section_info objects, one for each debug section that might
need to be read, then as we find the input bfd sections we associate
them with the corresponding dwarf2_section_info.
If no matching input bfd section is found then the dwarf2_section_info
is left in an unassociated state, its internal bfd section pointer is
null.
If later GDB tries to read content from the dwarf2_section_info, for
example, which trying to read the string associated with DW_FORM_strp,
we spot that there is no associated bfd section and issue an error
message.
To make the users life easier, the error message includes the section
name being looked for, and the bfd from which the section was
obtained.
However, we get the section name by calling bfd_section_name on the
associated section, and we get the bfd filename by calling
bfd_get_filename on the owner of the associated section.
Of course, if there is no associated section then both the calls
bfd_section_name and dwarf2_section_info::get_bfd_owner will result in
undefined behaviour (e.g. a crash).
The solution I propose in this patch is, I know, not ideal. I simply
spot the case where there is no associated section, and print a
simpler error message, leaving out the section name and filename.
A better solution would involve redesigning dwarf2_section_info, we
could associate each dwarf2_section_info with the initial bfd being
parsed. We would then display this filename if there's nothing better
to display (e.g. if we find a section in a dwo/dwp split dwarf file
then we would probably use that filename in preference).
Each dwarf2_section_info could also have the concept of the default
section name that would be read for that section, for example, string
data might appear in ".debug_str" or ".zdebug_str", but if neither is
found, then it would probably be OK to just say ".debug_str" is
missing.
Anyway, I didn't do any of that redesign, I just wanted to stop GDB
crashing for now, so instead we get this:
Dwarf Error: DW_FORM_strp used without required section
Which isn't the best, but in context, isn't too bad:
Reading symbols from /path/to/executable...
Dwarf Error: DW_FORM_strp used without required section
(No debugging symbols found in /path/to/executable)
I also added some asserts into dwarf2_section_info which should
trigger before GDB crashes in future, if we trigger any other bad
paths through this code.
And there's a test for the specific issue I hit.
gdb/ChangeLog:
* dwarf2/section.c (dwarf2_section_info::get_bfd_owner): Add an
assert.
(dwarf2_section_info::get_file_name): Add an assert.
(dwarf2_section_info::read_string): Display a minimal, sane error
when the dwarf2_section_info is not associated with a bfd section.
gdb/testsuite/ChangeLog:
* gdb.dwarf2/dw2-using-debug-str.exp: Add an additional test.
Andrew Burgess [Fri, 26 Mar 2021 17:14:26 +0000 (17:14 +0000)]
gdb/py: fix gdb.parameter('data-directory')
It was reported on IRC that using gdb.parameter('data-directory')
doesn't work correctly.
The problem is that the data directory is stored in 'gdb_datadir',
however the set/show command is associated with a temporary
'staged_gdb_datadir'.
When the user does 'set data-directory VALUE', the VALUE is stored in
'staged_gdb_datadir' by GDB, then set_gdb_datadir is called. This in
turn calls set_gdb_data_directory to copy the value from
staged_gdb_datadir into gdb_datadir.
However, set_gdb_data_directory will resolve relative paths, so the
value stored in gdb_datadir might not match the value in
staged_gdb_datadir.
The Python gdb.parameter API fetches the parameter values by accessing
the variable associated with the show command, so in this case
staged_gdb_datadir. This causes two problems:
1. Initially staged_gdb_datadir is NULL, and remains as such until the
user does 'set data-directory VALUE' (which might never happen), but
gdb_datadir starts with GDB's default data-directory value. So
initially from Python gdb.parameter('data-directory') will return the
empty string, even though at GDB's CLI 'show data-directory' prints a
real path.
2. If the user does 'set data-directory ./some/relative/path', GDB
will resolve the relative path, thus, 'show data-directory' at the CLI
will print an absolute path. However, the value is staged_gdb_datadir
will still be the relative path, and gdb.parameter('data-directory')
from Python will return the relative path.
In this commit I fix both of these issues by:
1. Initialising the value in staged_gdb_datadir based on the initial
value in gdb_datadir, and
2. In set_gdb_datadir, after calling set_gdb_data_directory, I copy
the value in gdb_datadir back into staged_gdb_datadir.
With these two changes in place the value in staged_gdb_datadir should
always match the value in gdb_datadir, and accessing data-directory
from Python should now work correctly.
gdb/ChangeLog:
* top.c (staged_gdb_datadir): Update comment.
(set_gdb_datadir): Copy the value of gdb_datadir back into
staged_datadir.
(init_main): Initialise staged_gdb_datadir.
gdb/testsuite/ChangeLog:
* gdb.python/py-parameter.exp: Add test for reading data-directory
using gdb.parameter API.