Simon Marchi [Thu, 30 Jul 2020 18:56:08 +0000 (14:56 -0400)]
gdb: handle non-const property types in ada_modulus (PR ada/26318)
PR 26318 shows that running `maint print symbols` on an Ada binary,
compiled with an Ada distribution that includes debug info for the
standard library, triggers this assertion:
The problem is in particular when printing type
`system__object_reader__decoded_ada_name__TTdecodedSP1___XDL_0`, which
has a dynamic high bound (PROP_LOCLIST kind). When printing a concrete
value of this type, this type gets resolved to a type with a constant
high bound, so ada_modulus can return this constant value.
However, when printing the dynamic range type on its own, such as with
`maint print symbols`, the high bound is still of kind PROP_LOCLIST.
When ada_modulus tries to access the property as a const value, the
assert triggers.
There's no sensible numerical value to return in this case. Ideally,
ada_modulus would return something to the caller indicating that the
value is dynamic and therefore can't be returned as an integer. The
callers would handle it, for example `maint print symbols` would say
that the high bound of the type is dynamic.
However, this patch implements the simpler fix of returning 0 in that
case. It kind of restores the previous behavior of before we validated
the dynamic property kind in the getters, where we would just return
whatever random integer value was in `const_val`. Except now it's
consistently 0.
This is what we had before we added dynamic property getters:
gdb/breakpoint: set the condition exp after parsing the condition successfully
In 'set_breakpoint_condition', GDB resets the condition expressions
before parsing the condition input by the user. This leads to the
problem of losing the condition expressions if the new condition
does not parse successfully and is thus rejected.
For instance:
$ gdb ./test
Reading symbols from ./test...
(gdb) start
Temporary breakpoint 1 at 0x114e: file test.c, line 4.
Starting program: test
Temporary breakpoint 1, main () at test.c:4
4 int a = 10;
(gdb) break 5
Breakpoint 2 at 0x555555555155: file test.c, line 5.
Now define a condition that would evaluate to false. Next, attempt
to overwrite that with an invalid condition:
(gdb) cond 2 a == 999
(gdb) cond 2 gibberish
No symbol "gibberish" in current context.
(gdb) info breakpoints
Num Type Disp Enb Address What
2 breakpoint keep y 0x0000555555555155 in main at test.c:5
stop only if a == 999
It appears as if the bad condition is successfully rejected. But if we
resume the program, we see that we hit the breakpoint although the condition
would evaluate to false.
(gdb) continue
Continuing.
Breakpoint 2, main () at test.c:5
5 a = a + 1; /* break-here */
Fix the problem by not resetting the condition expressions before
parsing the condition input.
Suppose the fix is applied. A similar problem could occur if the
condition is valid, but has "junk" at the end. In this case, parsing
succeeds, but an error is raised immediately after. It is too late,
though; the condition expression is already updated.
For instance:
$ gdb ./test
Reading symbols from ./test...
(gdb) start
Temporary breakpoint 1 at 0x114e: file test.c, line 4.
Starting program: test
Temporary breakpoint 1, main () at test.c:4
4 int a = 10;
(gdb) break 5
Breakpoint 2 at 0x555555555155: file test.c, line 5.
(gdb) cond 2 a == 999
(gdb) cond 2 a == 10 if
Junk at end of expression
(gdb) info breakpoints
Num Type Disp Enb Address What
2 breakpoint keep y 0x0000555555555155 in main at test.c:5
stop only if a == 999
(gdb) c
Continuing.
Breakpoint 2, main () at test.c:5
5 a = a + 1; /* break-here */
(gdb)
We should not have hit the breakpoint because the condition would
evaluate to false.
Fix this problem by updating the condition expression of the breakpoint
after parsing the input successfully and checking that there is no
remaining junk.
gdb/ChangeLog:
2020-07-30 Tankut Baris Aktemur <tankut.baris.aktemur@intel.com>
* breakpoint.c (set_breakpoint_condition): Update the condition
expressions after checking that the input condition string parses
successfully and does not contain junk at the end.
gdb/testsuite/ChangeLog:
2020-07-30 Tankut Baris Aktemur <tankut.baris.aktemur@intel.com>
* gdb.base/condbreak-bad.exp: Extend the test with scenarios
that attempt to overwrite an existing condition with a condition
that fails parsing and also with a condition that parses fine
but contains junk at the end.
gdb/breakpoint: do not update the condition string if parsing the condition fails
The condition of a breakpoint can be set with the 'cond' command. If
the condition has errors that make it problematic to evaluate, it
appears like GDB rejects the condition, but updates the breakpoint's
condition string, which causes incorrect/unintuitive behavior.
For instance:
$ gdb ./test
Reading symbols from ./test...
(gdb) break 5
Breakpoint 1 at 0x1155: file test.c, line 5.
(gdb) cond 1 gibberish
No symbol "gibberish" in current context.
At this point, it looks like the condition was rejected.
But "info breakpoints" shows the following:
(gdb) info breakpoints
Num Type Disp Enb Address What
1 breakpoint keep y 0x0000000000001155 in main at test.c:5
stop only if gibberish
Running the code gives the following behavior, where re-insertion of
the breakpoint causes failures.
(gdb) run
Starting program: test
warning: failed to reevaluate condition for breakpoint 1: No symbol "gibberish" in current context.
warning: failed to reevaluate condition for breakpoint 1: No symbol "gibberish" in current context.
warning: failed to reevaluate condition for breakpoint 1: No symbol "gibberish" in current context.
warning: failed to reevaluate condition for breakpoint 1: No symbol "gibberish" in current context.
warning: failed to reevaluate condition for breakpoint 1: No symbol "gibberish" in current context.
[Inferior 1 (process 19084) exited normally]
(gdb)
This broken behavior occurs because GDB updates the condition string
of the breakpoint *before* checking that it parses successfully.
When parsing fails, the update has already taken place.
Fix the problem by updating the condition string *after* parsing the
condition. We get the following behavior when this patch is applied:
$ gdb ./test
Reading symbols from ./test...
(gdb) break 5
Breakpoint 1 at 0x1155: file test.c, line 5.
(gdb) cond 1 gibberish
No symbol "gibberish" in current context.
(gdb) info breakpoints
Num Type Disp Enb Address What
1 breakpoint keep y 0x0000000000001155 in main at test.c:5
(gdb) run
Starting program: test
Breakpoint 1, main () at test.c:5
5 a = a + 1; /* break-here */
(gdb) c
Continuing.
[Inferior 1 (process 15574) exited normally]
(gdb)
A side note: The problem does not occur if the condition is given
at the time of breakpoint definition, as in "break 5 if gibberish",
because the parsing of the condition fails during symtab-and-line
creation, before the breakpoint is created.
Finally, the code included the following comment:
/* I don't know if it matters whether this is the string the user
typed in or the decompiled expression. */
This comment did not make sense to me because the condition string is
the user-typed input. The patch updates this comment, too.
gdb/ChangeLog:
2020-07-30 Tankut Baris Aktemur <tankut.baris.aktemur@intel.com>
* breakpoint.c (set_breakpoint_condition): Update the
condition string after parsing the new condition successfully.
gdb/testsuite/ChangeLog:
2020-07-30 Tankut Baris Aktemur <tankut.baris.aktemur@intel.com>
* gdb.base/condbreak-bad.c: New test.
* gdb.base/condbreak-bad.exp: New file.
Szabolcs Nagy [Wed, 29 Jul 2020 14:47:50 +0000 (15:47 +0100)]
aarch64: set sh_entsize of .plt to 0
On aarch64 the first PLT entry is 32 bytes, subsequent entries
are 16 bytes by default but can be 24 bytes with BTI or with
PAC-PLT.
sh_entsize of .plt was set to the PLT entry size, so in some
cases sh_size % sh_entsize != 0, which breaks some tools.
Note that PLT0 (and the TLSDESC stub code which is also in the
PLT) were historically not padded up to meet the sh_size
requirement, but to ensure that PLT stub code is aligned on
cache lines. Similar layout is present on other targets too
which just happens to make sh_size a multiple of sh_entsize and
it is not expected that sh_entsize of .plt is used for anything.
This patch sets sh_entsize of .plt to 0: the section does not
hold a table of fixed-size entries so other values are not
conforming in principle to the ELF spec.
bfd/ChangeLog:
PR ld/26312
* elfnn-aarch64.c (elfNN_aarch64_init_small_plt0_entry): Set sh_entsize
to 0.
(elfNN_aarch64_finish_dynamic_sections): Remove sh_entsize setting.
Fix this by allowing these module functions and variables to be missing.
Tested on x86_64-linux with gcc 4.8.5 and gcc 7.5.0.
gdb/testsuite/ChangeLog:
2020-07-30 Tom de Vries <tdevries@suse.de>
* lib/sym-info-cmds.exp (GDBInfoModuleSymbols::check_entry_1): Factor
out of ...
(GDBInfoModuleSymbols::check_entry): ... here.
(GDBInfoModuleSymbols::check_optional_entry): New proc.
* gdb.fortran/info-modules.exp: Use check_optional_entry for entries
related to __def_init_mod1_M1t1 / __vtype_mod1_M1t1 / __copy_mod1_M1t1.
In file included from /usr/include/sys/procfs.h:26,
from /vol/src/gnu/gdb/hg/master/dist/gdb/i386-sol2-nat.c:24:
/usr/include/sys/old_procfs.h:31:2: error: #error "Cannot use procfs in the large file compilation environment"
#error "Cannot use procfs in the large file compilation environment"
^~~~~
* On Solaris 11.3/x86 there are several more instances of this.
The interaction between procfs and large-file support historically has
been a royal mess on Solaris:
* There are two versions of the procfs interface:
** The old ioctl-based /proc, deprecated and not used any longer in
either gdb or binutils.
** The `new' (introduced in Solaris 2.6, 1997) structured /proc.
* There are two headers one can possibly include:
** <procfs.h> which only provides the structured /proc, definining
_STRUCTURED_PROC=1 and then including ...
** <sys/procfs.h> which defaults to _STRUCTURED_PROC=0, the ioctl-based
/proc, but provides structured /proc if _STRUCTURED_PROC == 1.
* procfs and the large-file environment didn't go well together:
** Until Solaris 11.3, <sys/procfs.h> would always #error in 32-bit
compilations when the large-file environment was active
(_FILE_OFFSET_BITS == 64).
** In both Solaris 11.4 and Illumos, this restriction was lifted for
structured /proc.
So one has to be careful always to define _STRUCTURED_PROC=1 when
testing for or using <sys/procfs.h> on Solaris. As the errors above
show, this isn't always the case in binutils-gdb right now.
Also one may need to disable large-file support for 32-bit compilations
on Solaris. config/largefile.m4 meant to do this by wrapping the
AC_SYS_LARGEFILE autoconf macro with appropriate checks, yielding
ACX_LARGEFILE. Unfortunately the macro doesn't always succeed because
it neglects the _STRUCTURED_PROC part.
To make things even worse, since GCC 9 g++ predefines
_FILE_OFFSET_BITS=64 on Solaris. So even if largefile.m4 deciced not to
enable large-file support, this has no effect, breaking the gdb build.
This patch addresses all this as follows:
* All tests for the <sys/procfs.h> header are made with
_STRUCTURED_PROC=1, the definition going into the various config.h
files instead of having to make them (and sometimes failing) in the
affected sources.
* To cope with the g++ predefine of _FILE_OFFSET_BITS=64,
-U_FILE_OFFSET_BITS is added to various *_CPPFLAGS variables. It had
been far easier to have just
#undef _FILE_OFFSET_BITS
in config.h, but unfortunately such a construct in config.in is
commented by config.status irrespective of indentation and whitespace
if large-file support is disabled. I found no way around this and
putting the #undef in several global headers for bfd, binutils, ld,
and gdb seemed way more invasive.
* Last, the applicability check in largefile.m4 was modified only to
disable largefile support if really needed. To do so, it checks if
<sys/procfs.h> compiles with _FILE_OFFSET_BITS=64 defined. If it
doesn't, the disabling only happens if gdb exists in-tree and isn't
disabled, otherwise (building binutils from a tarball), there's no
conflict.
What initially confused me was the check for $plugins here, which
originally caused the disabling not to take place. Since AC_PLUGINGS
does enable plugin support if <dlfcn.h> exists (which it does on
Solaris), the disabling never happened.
I could find no explanation why the linker plugin needs large-file
support but thought it would be enough if gld and GCC's lto-plugin
agreed on the _FILE_OFFSET_BITS value. Unfortunately, that's not
enough: lto-plugin uses the simple-object interface from libiberty,
which includes off_t arguments. So to fully disable large-file
support would mean also disabling it in libiberty and its users: gcc
and libstdc++-v3. This seems highly undesirable, so I decided to
disable the linker plugin instead if large-file support won't work.
The patch allows binutils+gdb to build on i386-pc-solaris2.11 (both
Solaris 11.3 and 11.4, using GCC 9.3.0 which is the worst case due to
predefined _FILE_OFFSET_BITS=64). Also regtested on
amd64-pc-solaris2.11 (again on Solaris 11.3 and 11.4),
x86_64-pc-linux-gnu and i686-pc-linux-gnu.
config:
* largefile.m4 (ACX_LARGEFILE) <sparc-*-solaris*|i?86-*-solaris*>:
Check for <sys/procfs.h> incompatilibity with large-file support
on Solaris.
Only disable large-file support and perhaps plugins if needed.
Set, substitute LARGEFILE_CPPFLAGS if so.
H.J. Lu [Thu, 30 Jul 2020 10:41:30 +0000 (03:41 -0700)]
elf: Add sym_cache to elf_link_hash_table
Since many ELF backends have sym_cache to their link hash tables, add
sym_cache to elf_link_hash_table. Also use sdynbss and srelbss in
elf_link_hash_table.
Tom de Vries [Thu, 30 Jul 2020 10:35:57 +0000 (12:35 +0200)]
[gdb/testsuite] Fix gdb.fortran/ptype-on-functions.exp with gcc-4.8
When running test-case gdb.fortran/ptype-on-functions.exp with gfortran 4.8.5,
we run into:
...
(gdb) ptype some_module::get_number^M
type = integer(kind=4) (Type __class_some_module_Number)^M
(gdb) FAIL: gdb.fortran/ptype-on-functions.exp: ptype some_module::get_number
ptype some_module::set_number^M
type = void (Type __class_some_module_Number, integer(kind=4))^M
(gdb) FAIL: gdb.fortran/ptype-on-functions.exp: ptype some_module::set_number
...
The test-case pattern expects a "_t" suffix on "__class_some_module_Number".
The difference is caused by a gcc commit 073afca6884 'class.c
(gfc_build_class_symbol): Append "_t" to target class names to make the
generated type names unique' which has been present since gcc 4.9.0.
Fix the pattern by optionally matching the _t suffix.
Tested on x86_64-linux, with gfortran 4.8.5 and 7.5.0.
gdb/testsuite/ChangeLog:
2020-07-30 Tom de Vries <tdevries@suse.de>
* gdb.fortran/ptype-on-functions.exp: Make "_t" suffix on
"__class_some_module_Number_t" optional.
Tom de Vries [Thu, 30 Jul 2020 07:23:01 +0000 (09:23 +0200)]
[gdb/build] Fix Wmaybe-uninitialized in gdb/ui-style.h
When building CFLAGS/CXXFLAGS="-O2 -g -Wall" and gcc 4.8.5, we run into:
...
src/gdb/cli/cli-style.c:154:42: warning: '*((void*)&<anonymous> +8)' \
may be used uninitialized in this function [-Wmaybe-uninitialized]
src/gdb/cli/cli-style.c:154:42: warning: '*((void*)&<anonymous> +9)' \
may be used uninitialized in this function [-Wmaybe-uninitialized]
src/gdb/cli/cli-style.c:154:42: warning: '*((void*)&<anonymous> +10)' \
may be used uninitialized in this function [-Wmaybe-uninitialized]
...
The root cause is that the data members of class color, nested in struct
ui_file_style in gdb/ui-style.h:
...
bool m_simple;
int m_value;
uint8_t m_red, m_green, m_blue;
...
are only partially initialized by this constructor:
...
color (int c)
: m_simple (true),
m_value (c)
{
gdb_assert (c >= -1 && c <= 255);
}
...
but the default copy constructor will copy all the fields.
The member m_simple acts as a discriminant, to indicate which other members
are valid:
- m_value (with m_simple == true)
- m_red, m_green, m_blue (with m_simple == false)
So, we don't need storage for both m_value and m_red/m_green/m_blue at the
same time.
Fix this by wrapping the respective members in a union:
...
bool m_simple;
union
{
int m_value;
struct
{
uint8_t m_red, m_green, m_blue;
};
};
...
which also fixes the warning.
Build and reg-tested on x86_64-linux.
gdb/ChangeLog:
2020-07-30 Tom de Vries <tdevries@suse.de>
PR build/26320
* ui-style.h (struct ui_file_style::color): Wrap m_value and
m_red/m_green/m_blue in a union.
Simon Marchi [Wed, 29 Jul 2020 20:02:57 +0000 (16:02 -0400)]
Run `autoreconf -vf` throughout
I ran
for i in $(find . -name configure.ac); do pushd $(dirname $i); autoreconf -vf; popd; done
to re-generate all automake/autoconf files throughout the repo (with
upstream autoconf 2.69 and automake 1.15.1). These were the changes
that came out. I am pushing this as obvious.
MIPS: Make the IRIX naming of local section symbols consistent
Make the MIPS/IRIX naming of local section symbols consistent between
files produced by generic ELF code and ELF linker code, complementing
commit 174fd7f95561 ("New bfd elf hook: force naming of local section
symbols"), <https://sourceware.org/ml/binutils/2004-02/msg00072.html>.
Local section symbols have no names in the standard ELF gABI, however
the lack of a name causes problems with IRIX's MIPSpro linker. To work
around the issue we give them names, however we do that in generic ELF
code only, based on what the `elf_backend_name_local_section_symbols'
hook returns if present. That makes objects created by GAS or `objdump'
work correctly, however not ones created by `ld -r'. That would not
normally cause issues with IRIX systems using GAS and `objdump' only
with the MIPSpro linker, however if GNU LD was used for whatever reason
in producing objects later fed to IRIX's MIPSpro linker, then things
would break.
Modify ELF linker code accordingly then, using the same hook. Adjust
the `ld-elf/64ksec-r' test accordingly so that it also accepts a section
symbol with a name.
Also modify the hook itself so that only actual ET_REL objects have
names assigned to local section symbols. Other kinds of ELF files are
not ever supposed to be relocated with the MIPSpro linker, so we can
afford producing more standard output.
Add suitable GAS, LD and `objcopy' test cases to the relevant testsuites
to keep these tools consistently verified. This change also fixes:
FAIL: objcopy executable (pr25662)
across MIPS targets using the IRIX compatibility mode.
bfd/
* elflink.c (bfd_elf_final_link): Give local symbols a name if
so requested.
* elfxx-mips.c (_bfd_mips_elf_name_local_section_symbols): Only
return TRUE if making ET_REL output.
binutils/
* testsuite/binutils-all/mips/global-local-symtab-sort-o32.d:
New test.
* testsuite/binutils-all/mips/global-local-symtab-sort-o32t.d:
New test.
* testsuite/binutils-all/mips/global-local-symtab-sort-n32.d:
New test.
* testsuite/binutils-all/mips/global-local-symtab-sort-n32t.d:
New test.
* testsuite/binutils-all/mips/global-local-symtab-sort-n64.d:
New test.
* testsuite/binutils-all/mips/global-local-symtab-sort-n64t.d:
New test.
* testsuite/binutils-all/mips/global-local-symtab-final-o32.d:
New test.
* testsuite/binutils-all/mips/global-local-symtab-final-n32.d:
New test.
* testsuite/binutils-all/mips/global-local-symtab-final-n64.d:
New test.
* testsuite/binutils-all/mips/mips.exp: Run the new tests.
gas/
* testsuite/gas/mips/global-local-symtab-sort-o32.d: New test.
* testsuite/gas/mips/global-local-symtab-sort-o32t.d: New test.
* testsuite/gas/mips/global-local-symtab-sort-n32.d: New test.
* testsuite/gas/mips/global-local-symtab-sort-n32t.d: New test.
* testsuite/gas/mips/global-local-symtab-sort-n64.d: New test.
* testsuite/gas/mips/global-local-symtab-sort-n64t.d: New test.
* testsuite/gas/mips/mips.exp: Run the new tests.
ld/
* testsuite/ld-elf/sec64k.exp: Also accept a section symbol with
a name.
* testsuite/ld-mips-elf/global-local-symtab-sort-o32.d: New
test.
* testsuite/ld-mips-elf/global-local-symtab-sort-o32t.d: New
test.
* testsuite/ld-mips-elf/global-local-symtab-sort-n32.d: New
test.
* testsuite/ld-mips-elf/global-local-symtab-sort-n32t.d: New
test.
* testsuite/ld-mips-elf/global-local-symtab-sort-n64.d: New
test.
* testsuite/ld-mips-elf/global-local-symtab-sort-n64t.d: New
test.
* testsuite/ld-mips-elf/global-local-symtab-final-o32.d: New
test.
* testsuite/ld-mips-elf/global-local-symtab-final-n32.d: New
test.
* testsuite/ld-mips-elf/global-local-symtab-final-n64.d: New
test.
* testsuite/ld-mips-elf/mips-elf.exp: Run the new tests.
MIPS/LD: Set symtab's `sh_info' correctly for IRIX emulations
Correct ELF linker code so as to set the `sh_info' value of the static
symbol table section according to the section symbols vs other symbols
split where required by the selection of the IRIX compatibility mode for
MIPS target. Add a `elf_backend_elfsym_local_is_section' hook for that
purpose, returning TRUE if it is only STB_LOCAL/STT_SECTION symbols that
are to be considered local for the purpose of this split rather than all
STB_LOCAL symbols.
We do it already in generic ELF code, and have done it since 1993, with
the `elf_backend_sym_is_global' hook, affecting GAS and `objcopy', so
these tools produce correct ELF output in the IRIX compatibility mode,
however if such output is fed as input to `ld -r', then the linker's
output is no longer valid for that mode. The relevant changes to
generic ELF code are:
(split across two GIT commits likely due to repository conversion
peculiarities).
The `elf_backend_sym_is_global' hook however operates on BFD rather than
ELF symbols, making it unsuitable for the ELF linker as the linker does
not convert any symbol tables processed into the BFD format. Converting
the hook to operate on ELF symbols would in principle be possible, but
it would still require a considerable rewrite of `bfd_elf_final_link' to
adapt to the interface.
Therefore, especially given that no new use for the IRIX compatibility
mode is expected, minimize changes made to the ELF linker code and just
add an entirely new hook, and wire it in the o32 and n32 MIPS backends
accordingly; the n64 backend never uses the IRIX compatibility mode.
Since we have no coverage here at all add suitable GAS, LD and `objcopy'
test cases to the relevant testsuites to keep these tools consistently
verified.
bfd/
* elf-bfd.h (elf_backend_data): Add
`elf_backend_elfsym_local_is_section' member.
* elfxx-target.h (elf_backend_elfsym_local_is_section): New
macro.
(elfNN_bed): Add `elf_backend_elfsym_local_is_section' member.
* elflink.c (bfd_elf_final_link): Use it to determine whether
set the `.symtab' section's `sh_info' value to the index of the
first non-local or non-section symbol.
* elf32-mips.c (mips_elf32_elfsym_local_is_section): New
function.
(elf_backend_elfsym_local_is_section): New macro.
* elfn32-mips.c (mips_elf_n32_elfsym_local_is_section): New
function.
(elf_backend_elfsym_local_is_section): New macro.
binutils/
* testsuite/binutils-all/mips/global-local-symtab-o32.d: New
test.
* testsuite/binutils-all/mips/global-local-symtab-o32t.d: New
test.
* testsuite/binutils-all/mips/global-local-symtab-n32.d: New
test.
* testsuite/binutils-all/mips/global-local-symtab-n32t.d: New
test.
* testsuite/binutils-all/mips/global-local-symtab-n64.d: New
test.
* testsuite/binutils-all/mips/mips.exp: Run the new tests.
gas/
* testsuite/gas/mips/global-local-symtab-o32.d: New test.
* testsuite/gas/mips/global-local-symtab-o32t.d: New test.
* testsuite/gas/mips/global-local-symtab-n32.d: New test.
* testsuite/gas/mips/global-local-symtab-n32t.d: New test.
* testsuite/gas/mips/global-local-symtab-n64.d: New test.
* testsuite/gas/mips/global-local-symtab.s: New test source.
* testsuite/gas/mips/mips.exp: Run the new tests.
ld/
* testsuite/ld-mips-elf/global-local-symtab-o32.d: New test.
* testsuite/ld-mips-elf/global-local-symtab-o32t.d: New test.
* testsuite/ld-mips-elf/global-local-symtab-n32.d: New test.
* testsuite/ld-mips-elf/global-local-symtab-n32t.d: New test.
* testsuite/ld-mips-elf/global-local-symtab-n64.d: New test.
* testsuite/ld-mips-elf/global-local-symtab.ld: New test linker
script.
* testsuite/ld-mips-elf/mips-elf.exp: Run the new tests.
Andreas Arnez [Wed, 29 Jul 2020 17:46:44 +0000 (19:46 +0200)]
PR26279 Work around maybe-uninitialized warning in s390-mkopc.c
In s390-mkopc.c, the function insertExpandedMnemonic() searches for the
first occurrence of '*' or '$' in the given mnemonic, and, if a match is
found, chooses an extension table using a switch() on that character. The
switch statement contains a default case that prints an error message and
does not set the extension table. Although this case cannot occur, some
GCC versions obviously conclude that the extension table might have been
left uninitialized after the switch statement and consequently emit
maybe-uninitialized warnings for the variables 'ext_table' and
'ext_table_length'.
Circumvent the warning by handling the unreachable default case with
abort().
opcodes/
* s390-mkopc.c (insertExpandedMnemonic): Handle unreachable
default case with abort() instead of printing an error message and
continuing, to avoid a maybe-uninitialized warning.
Tom de Vries [Wed, 29 Jul 2020 16:16:26 +0000 (18:16 +0200)]
[gdb/testsuite] Fix captured_command_loop breakpoint in selftests
When building gcc with CFLAGS/CXXFLAGS="-O2 -g", and running the regression
tests, I run into the following FAILs:
...
FAIL: gdb.gdb/complaints.exp: breakpoint in captured_command_loop
FAIL: gdb.gdb/python-interrupts.exp: breakpoint in captured_command_loop
FAIL: gdb.gdb/python-selftest.exp: breakpoint in captured_command_loop
...
The problem is that when setting the breakpoint at captured_command_loop:
...
(gdb) break captured_command_loop^M
Breakpoint 1 at 0x4230ed: captured_command_loop. (2 locations)^M
(gdb) FAIL: gdb.gdb/complaints.exp: breakpoint in captured_command_loop
...
there are two breakpoint locations instead of one. This is due to
PR26096 - "gdb sets breakpoint at cold clone":
...
$ nm gdb | grep captured_command_loop| c++filt 0000000000659f20 t captured_command_loop() 00000000004230ed t captured_command_loop() [clone .cold]
...
Work around this by allowing multiple breakpoint locations for
captured_command_loop.
Reg-tested on x86_64-linux.
gdb/testsuite/ChangeLog:
2020-07-29 Tom de Vries <tdevries@suse.de>
* lib/selftest-support.exp (selftest_setup): Allow breakpoint at
multiple locations.
For DWARF v5 Dwarf Package Files (.dwp files), the section identifier encodings have changed. This patch updates dwarf2.h to contain the new encodings. (see http://dwarfstd.org/doc/DWARF5.pdf, section 7.3.5).
* dwarf2.h (enum dwarf_sect_v5): A new enum section for the
sections in a DWARF 5 DWP file (DWP version 5).
Tom de Vries [Wed, 29 Jul 2020 07:03:20 +0000 (09:03 +0200)]
[tdep/s390] Fix Wmaybe-uninitialized in s390_displaced_step_fixup
When building gdb with CFLAGS/CXXFLAGS="-O2 -g -Wall", I see:
...
src/gdb/s390-tdep.c: In function 'void s390_displaced_step_fixup(gdbarch*, \
displaced_step_closure*, CORE_ADDR, CORE_ADDR, regcache*)':
src/gdb/s390-tdep.c:528:30: warning: 'r2' may be used uninitialized in this \
function [-Wmaybe-uninitialized]
528 | if (insn[0] == op_basr && r2 == 0)
| ~~~~~~~~~~~~~~~~~~~^~~~~~~~~~
...
The problem is that the compiler is unaware that
'is_rr (insn, op_basr, &r1, &r2) == 1' ensures that 'insn[0] == op_basr':
...
if (is_rr (insn, op_basr, &r1, &r2)
|| is_rx (insn, op_bas, &r1, &d2, &x2, &b2))
{
/* Recompute saved return address in R1. */
regcache_cooked_write_unsigned (regs, S390_R0_REGNUM + r1,
amode | (from + insnlen));
/* Update PC iff the instruction doesn't actually branch. */
if (insn[0] == op_basr && r2 == 0)
regcache_write_pc (regs, from + insnlen);
}
...
Fix this by storing the result of the call, and using it instead of
'insn[0] ==op_basr'.
Build on x86_64-linux with --enable-targets=s390-suse-linux,s390x-suse-linux.
The disassembler no longer demangles function names in its output. So
we see things like this:
(gdb) disassemble tree_insert
Dump of assembler code for function _Z11tree_insertP4nodei:
....
Instead of this:
(gdb) disassemble tree_insert
Dump of assembler code for function tree_insert(node*, int):
....
This is because find_pc_partial_function now returns the linkage name
rather than the demangled name.
================
This patch fixes the problem by introducing a new "overload" of
find_pc_partial_function, which returns the general_symbol_info rather
than simply the name. This lets the disassemble command choose which
name to show.
Regression tested on x86-64 Fedora 32.
gdb/ChangeLog
2020-07-28 Tom Tromey <tromey@adacore.com>
PR symtab/26270:
* symtab.h (find_pc_partial_function_sym): Declare.
* cli/cli-cmds.c (disassemble_command): Use
find_pc_partial_function_sym. Check asm_demangle.
* blockframe.c (cache_pc_function_sym): New global.
(cache_pc_function_name): Remove.
(clear_pc_function_cache): Update.
(find_pc_partial_function_sym): New function, from
find_pc_partial_function.
(find_pc_partial_function): Rewrite using
find_pc_partial_function_sym.
gdb/testsuite/ChangeLog
2020-07-28 Andrew Burgess <andrew.burgess@embecosm.com>
PR symtab/26270:
* gdb.cp/disasm-func-name.cc: New file.
* gdb.cp/disasm-func-name.exp: New file.
Tom Tromey [Tue, 28 Jul 2020 16:55:43 +0000 (10:55 -0600)]
Fix bug in DW_OP_GNU_variable_value evaluation
A modified version of the gnat compiler (TBH I don't know if the
modifications are relevant to this bug or not, but I figured I'd
mention it) can generate a DWARF location expression like:
The intended behavior is that the debug info consumer computes the
value of that referenced variable at the current PC, and if it can
compute it and pushes the value as a generic type integer into the
DWARF stack
Instead, gdb is using the variable's type -- but this fails with some
operations, like DW_OP_and, which expect the types to match.
I believe what was intended was for the value to be cast to the DWARF
"untyped" type, which is what this patch implements. This patch also
updates varval.exp to exhibit the bug.
gdb/ChangeLog
2020-07-28 Tom Tromey <tromey@adacore.com>
* dwarf2/expr.c (dwarf_expr_context::execute_stack_op)
<DW_OP_GNU_variable_value>: Cast to address type.
gdb/testsuite/ChangeLog
2020-07-28 Tom Tromey <tromey@adacore.com>
* gdb.dwarf2/varval.exp (setup_exec): Add 'or' instruction to
'varval' location.
Implement xfer_partial TARGET_OBJECT_SIGNAL_INFO for NetBSD
NetBSD implements reading and overwriting siginfo_t received by the
tracee. With TARGET_OBJECT_SIGNAL_INFO signal information can be
examined and modified through the special variable $_siginfo.
Implement the "get_siginfo_type" gdbarch method for NetBSD architectures.
As with Linux architectures, cache the created type in the gdbarch when it
is first created. Currently NetBSD uses an identical siginfo type on
all architectures, so there is no support for architecture-specific fields.
gdb/ChangeLog:
* nbsd-nat.h (nbsd_nat_target::xfer_partial): New declaration.
* nbsd-nat.c (nbsd_nat_target::xfer_partial): New function.
* nbsd-tdep.c (nbsd_gdbarch_data_handle, struct nbsd_gdbarch_data)
(init_nbsd_gdbarch_data, get_nbsd_gdbarch_data)
(nbsd_get_siginfo_type): New.
(nbsd_init_abi): Install gdbarch "get_siginfo_type" method.
(_initialize_nbsd_tdep): New
Tom de Vries [Tue, 28 Jul 2020 13:07:44 +0000 (15:07 +0200)]
[gdb/build] Fix Wmaybe-uninitialized in gdb_optional.h
When building with CFLAGS/CXXFLAGS="-O2 -g -Wall", we run into:
...
In file included from src/gdb/exceptions.h:23,
from src/gdb/utils.h:24,
from src/gdb/defs.h:630,
from src/gdb/record-btrace.c:22:
src/gdb/ui-out.h: In function 'void btrace_insn_history(ui_out*, \
const btrace_thread_info*, const btrace_insn_iterator*, \
const btrace_insn_iterator*, gdb_disassembly_flags)':
src/gdb/ui-out.h:352:18: warning: \
'asm_list.ui_out_emit_type<ui_out_type_list>::m_uiout' may be used \
uninitialized in this function [-Wmaybe-uninitialized]
352 | m_uiout->end (Type);
| ~~~~~~~~~~~~~^~~~~~
src/gdb/record-btrace.c:795:35: note: \
'asm_list.ui_out_emit_type<ui_out_type_list>::m_uiout' was declared here
795 | gdb::optional<ui_out_emit_list> asm_list;
| ^~~~~~~~
...
This is reported as PR gcc/80635 - "[8/9/10/11 regression] std::optional and
bogus -Wmaybe-uninitialized warning".
Silence the warning by using the workaround suggested here (
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80635#c53 ):
...
union
{
struct { } m_dummy;
T m_item;
+ volatile char dont_use; // Silences -Wmaybe-uninitialized warning.
};
...
Build on x86_64-linux.
gdbsupport/ChangeLog:
2020-07-28 Tom de Vries <tdevries@suse.de>
PR build/26281
* gdb_optional.h (class optional): Add volatile member to union
contaning m_dummy and m_item.
Andrew Burgess [Wed, 22 Jul 2020 11:13:11 +0000 (12:13 +0100)]
gdb/python: make more use of RegisterDescriptors
This commit unifies all of the Python register lookup code (used by
Frame.read_register, PendingFrame.read_register, and
gdb.UnwindInfo.add_saved_register), and adds support for using a
gdb.RegisterDescriptor for register lookup.
Currently the register unwind code (PendingFrame and UnwindInfo) allow
registers to be looked up either by name, or by GDB's internal
number. I suspect the number was added for performance reasons, when
unwinding we don't want to repeatedly map from name to number for
every unwind. However, this kind-of sucks, it means Python scripts
could include GDB's internal register numbers, and if we ever change
this numbering in the future users scripts will break in unexpected
ways.
Meanwhile, the Frame.read_register method only supports accessing
registers using a string, the register name.
This commit unifies all of the register to register-number lookup code
in our Python bindings, and adds a third choice into the mix, the use
of gdb.RegisterDescriptor.
The register descriptors can be looked up by name, but once looked up,
they contain GDB's register number, and so provide all of the
performance benefits of using a register number directly. However, as
they are looked up by name we are no longer tightly binding the Python
API to GDB's internal numbering scheme.
As we may already have scripts in the wild that are using the register
numbers directly I have kept support for this in the API, but I have
listed this method last in the manual, and I have tried to stress that
this is NOT a good method to use and that users should use either a
string or register descriptor approach.
After this commit all existing Python code should function as before,
but users now have new options for how to identify registers.
gdb/ChangeLog:
* python/py-frame.c: Remove 'user-regs.h' include.
(frapy_read_register): Rewrite to make use of
gdbpy_parse_register_id.
* python/py-registers.c (gdbpy_parse_register_id): New function,
moved here from python/py-unwind.c. Updated the return type, and
also accepts register descriptor objects.
* python/py-unwind.c: Remove 'user-regs.h' include.
(pyuw_parse_register_id): Moved to python/py-registers.c.
(unwind_infopy_add_saved_register): Update to use
gdbpy_parse_register_id.
(pending_framepy_read_register): Likewise.
* python/python-internal.h (gdbpy_parse_register_id): Declare.
gdb/testsuite/ChangeLog:
* gdb.python/py-unwind.py: Update to make use of a register
descriptor.
gdb/doc/ChangeLog:
* python.texi (Unwinding Frames in Python): Update descriptions
for PendingFrame.read_register and
gdb.UnwindInfo.add_saved_register.
(Frames In Python): Update description of Frame.read_register.
Andrew Burgess [Wed, 22 Jul 2020 13:02:30 +0000 (14:02 +0100)]
gdb: Add a find method for RegisterDescriptorIterator
Adds a new method 'find' to the gdb.RegisterDescriptorIterator class,
this allows gdb.RegisterDescriptor objects to be looked up directly by
register name rather than having to iterate over all registers.
This will be of use for a later commit.
I've documented the new function in the manual, but I don't think a
NEWS entry is required here, as, since the last release, the whole
register descriptor mechanism is new, and is already mentioned in the
NEWS file.
gdb/ChangeLog:
* python/py-registers.c: Add 'user-regs.h' include.
(register_descriptor_iter_find): New function.
(register_descriptor_iterator_object_methods): New static global
methods array.
(register_descriptor_iterator_object_type): Add pointer to methods
array.
Alan Modra [Tue, 28 Jul 2020 02:14:16 +0000 (11:44 +0930)]
More just-syms changes
* ldlang.c (lang_check): Don't complain about relocs or merge
attributes from --just-symbols input.
* testsuite/ld-misc/just-symbols.exp: Just dump .data section.
Don't run test on a number of targets.
Alan Modra [Tue, 28 Jul 2020 01:13:34 +0000 (10:43 +0930)]
Re: Allow new just-symbols test to run on XCOFF and PE
This ensures we don't match random data *before* the line we want to
see, ie. that --just-symbols has excluded section contents from
just-symbols-0.o. Oops, missed the ChangeLog entry before too.
* testsuite/ld-misc/just-symbols-1.dd: Revert last change.
John Baldwin [Mon, 27 Jul 2020 15:58:48 +0000 (08:58 -0700)]
Use SIGTRAP si_code values for all FreeBSD architectures on 11.3 and later.
Fixes to set correct si_code values (such as TRAP_BRKPT) were made to
the remaining FreeBSD architectures (MIPS and sparc64) in the head
branch leading up to 12.0 and were merged back between the 11.2 and
11.3 releases.
gdb/ChangeLog:
* fbsd-nat.h: Include <osreldate.h>. Define USE_SIGTRAP_SIGINFO
for all architectures on FreeBSD 11.3 and later.
Alan Modra [Mon, 27 Jul 2020 09:18:57 +0000 (18:48 +0930)]
Allow new just-symbols test to run on XCOFF and PE
* testsuite/ld-misc/just-symbols.exp: Run for x86_64 PE too.
Set LDFLAGS for PE and XCOFF.
* testsuite/ld-misc/just-symbols.ld: Accept XCOFF mapped .data.
Alan Modra [Mon, 27 Jul 2020 08:03:59 +0000 (17:33 +0930)]
Accept --just-symbols symbols as absolute for xcoff
This patch is aimed at curing
just-symbols-1.o: loader reloc in unrecognized section `*ABS*'
for xcoff by treating symbols defined by --just-symbols objects as
absolute.
* xcofflink.c (xcoff_need_ldrel_p): Accept --just-symbols symbols and
similar as absolute.
(bfd_xcoff_import_symbol): Don't fuss over absolute symbol
redefinitions here.
* options.h (DEFINE_enum): Add optional_arg__ param, adjust
all uses.
(General_options): Add --power10-stubs and --no-power10-stubs.
* options.cc (General_options::finalize): Handle --power10-stubs.
* powerpc.cc (set_power10_stubs): Don't set when --power10-stubs=no.
(power10_stubs_auto): New.
(struct Plt_stub_ent): Add toc_ and tocoff_. Don't use a bitfield
for indx_.
(struct Branch_stub_ent): Add toc_and tocoff_. Use bitfields for
iter_, notoc_ and save_res_.
(add_plt_call_entry): Set toc_. Adjust resizing conditions for
--power10-stubs=auto.
(add_long_branch_entry): Set toc_.
(add_eh_frame, define_stub_syms): No longer use const_iterators
for plt and long branch stub iteration.
(build_tls_opt_head, build_tls_opt_tail): Change parameters and
return value. Move tests for __tls_get_addr to callers.
(plt_call_size): Handle --power10-stubs=auto.
(branch_stub_size): Likewise.
(Stub_table::do_write): Likewise.
(relocate): Likewise.
I'm running into a build breaker:
...
src/gdb/ser-tcp.c:65:13: error: conflicting declaration ‘typedef int
socklen_t’
65 | typedef int socklen_t;
| ^~~~~~~~~
In file included from ../gnulib/import/unistd.h:40,
from
/home/vries/gdb_versions/devel/src/gdb/../gnulib/import/pathmax.h:42,
from
/home/vries/gdb_versions/devel/src/gdb/../gdbsupport/common-defs.h:120,
from src/gdb/defs.h:28,
from src/gdb/ser-tcp.c:20:
/usr/include/unistd.h:277:21: note: previous declaration as ‘typedef
__socklen_t socklen_t’
277 | typedef __socklen_t socklen_t;
| ^~~~~~~~~
...
after commit 05a6b8c28b "Don't unnecessarily redefine 'socklen_t' type in
MinGW builds".
The root cause is a typo in gdb/configure.ac, using sys/sockets.h where
sys/socket.h was meant:
...
AC_CHECK_HEADERS([sys/sockets.h])
...
Eli Zaretskii [Sun, 26 Jul 2020 23:06:02 +0000 (16:06 -0700)]
libctf: compilation failure on MinGW due to missing errno values
This commit fixes a compilation failure in a couple of libctf files
due to the use of EOVERFLOW and ENOTSUP, which are not defined
when compiling on MinGW.
libctf/ChangeLog:
PR binutils/25155:
* ctf-create.c (EOVERFLOW): If not defined by system header,
redirect to ERANGE as a poor man's substitute.
* ctf-subr.c (ENOTSUP): If not defined, use ENOSYS instead.
Eli Zaretskii [Sun, 26 Jul 2020 16:35:48 +0000 (19:35 +0300)]
Don't unnecessarily redefine 'socklen_t' type in MinGW builds.
The original configure-time tests in gdb/ and gdbserver/ failed to
detect that 'socklen_t' is defined in MinGW headers because the test
program included only sys/socket.h, which is absent in MinGW system
headers. However on MS-Windows this data type is declared in another
header, ws2tcpip.h. The modified test programs try using ws2tcpip.h
if sys/socket.h is unavailable.
Thanks to Joel Brobecker who helped me regenerate the configure
scripts and the config.in files.
gdb/ChangeLog:
2020-07-26 Eli Zaretskii <eliz@gnu.org>
* configure.ac (AC_CHECK_HEADERS): Check for sys/socket.h and
ws2tcpip.h. When checking whether socklen_t type is defined, use
ws2tcpip.h if it is available and sys/socket.h isn't.
* configure: Regenerate.
* config.in: Regenerate.
gdbserver/ChangeLog:
2020-07-26 Eli Zaretskii <eliz@gnu.org>
* configure.ac (AC_CHECK_HEADERS): Add ws2tcpip.h.
When checking whether socklen_t type is defined, use ws2tcpip.h if
it is available and sys/socket.h isn't.
* configure: Regenerate.
* config.in: Regenerate.
MIPS/binutils/testsuite: Correct mips.exp test ABI/emul/endian arrangement
The binutils testsuite supports involving LD in processing test cases
and with the MIPS target that has the same issues the LD testsuite does.
So to support LD in the MIPS part of the binutils testsuite similarly
to commit 86b24e15c45b ("MIPS/LD/testsuite: Correct comm-data.exp test
ABI/emul/endian arrangement") update the mips.exp test script to:
- correctly select emulations for targets using non-traditional MIPS
emulations,
- correctly select ABIs for targets that do not support all of them,
- use the default endianness selection where possible to benefit targets
that support only one,
- simplify test invocation by providing ABI-specific `run_dump_test'
wrappers, specifically `run_dump_test_o32', `run_dump_test_n32' and
`run_dump_test_n64', which remove the need to use conditionals across
the Expect script or to repeat ABI-specific GAS and LD flags with each
invocation,
borrowing changes from commit 78da84f99405 ("MIPS/LD/testsuite: Correct
mips-elf.exp test ABI/emul/endian arrangement").
As a side effect this disables o32 ABI testing for targets that are not
supposed to support them and do not with LD, but still have such support
with BFD and GAS due to our inflexibility in configuration. Ultimately
we ought to support having o32 completely disabled.
binutils/
* testsuite/binutils-all/mips/mips.exp (run_dump_test_abi)
(run_dump_test_o32, run_dump_test_n32, run_dump_test_n64): New
procedures.
(has_newabi): Remove variable.
(has_abi, abi_asflags, abi_ldflags): New associative array
variables.
(irixemul): New variable.
Replace `run_dump_test' calls where applicable throughout with
`run_dump_test_o32', `run_dump_test_n32' and `run_dump_test_n64'
as appropriate. Use `noarch' for tests that require their own
architecture setting.
* testsuite/binutils-all/mips/mips-ase-1.d: Remove GAS flags.
* testsuite/binutils-all/mips/mips-ase-2.d: Likewise.
* testsuite/binutils-all/mips/mips-ase-3.d: Likewise.
* testsuite/binutils-all/mips/mips-note-2-n32.d: Likewise.
* testsuite/binutils-all/mips/mips-note-2-n64.d: Likewise.
* testsuite/binutils-all/mips/mips-note-2.d: Likewise.
* testsuite/binutils-all/mips/mips-note-2r-n32.d: Likewise.
* testsuite/binutils-all/mips/mips-note-2r-n64.d: Likewise.
* testsuite/binutils-all/mips/mips-note-2r.d: Likewise.
* testsuite/binutils-all/mips/mips-reginfo-n32.d: Likewise.
* testsuite/binutils-all/mips/mips-reginfo.d: Likewise.
* testsuite/binutils-all/mips/mips16-extend-noinsn.d: Likewise.
* testsuite/binutils-all/mips/mips16-pcrel.d: Likewise.
* testsuite/binutils-all/mips/mips16-alias.d: Remove `-32' from
GAS flags.
* testsuite/binutils-all/mips/mips16-extend-insn.d: Likewise.
* testsuite/binutils-all/mips/mips16-noalias.d: Likewise.
* testsuite/binutils-all/mips/mips16-undecoded.d: Likewise.
* testsuite/binutils-all/mips/mips16e2-extend-insn.d: Likewise.
* testsuite/binutils-all/mips/mips16e2-undecoded.d: Likewise.
* testsuite/binutils-all/mips/mixed-micromips.d: Likewise.
* testsuite/binutils-all/mips/mixed-mips16.d: Likewise.
Andrew Burgess [Thu, 9 Jul 2020 15:26:23 +0000 (16:26 +0100)]
gdb/fortran: resolve dynamic types when readjusting after an indirection
After dereferencing a pointer (in value_ind) or following a
reference (in coerce_ref) we call readjust_indirect_value_type to
"fixup" the type of the resulting value object.
This fixup handles cases relating to the type of the resulting object
being different (a sub-class) of the original pointers target type.
If we encounter a pointer to a dynamic type then after dereferencing a
pointer (in value_ind) the type of the object created will have had
its dynamic type resolved. However, in readjust_indirect_value_type,
we use the target type of the original pointer to "fixup" the type of
the resulting value. In this case, the target type will be a dynamic
type, so the resulting value object, once again has a dynamic type.
This then triggers an assertion later within GDB.
The solution I propose here is that we call resolve_dynamic_type on
the pointer's target type (within readjust_indirect_value_type) so
that the resulting value is not converted back to a dynamic type.
The test case is based on the original test in the bug report.
gdb/ChangeLog:
PR fortran/23051
PR fortran/26139
* valops.c (value_ind): Pass address to
readjust_indirect_value_type.
* value.c (readjust_indirect_value_type): Make parameter
non-const, and add extra address parameter. Resolve original type
before using it.
* value.h (readjust_indirect_value_type): Update function
signature and comment.
gdb/testsuite/ChangeLog:
PR fortran/23051
PR fortran/26139
* gdb.fortran/class-allocatable-array.exp: New file.
* gdb.fortran/class-allocatable-array.f90: New file.
* gdb.fortran/pointer-to-pointer.exp: New file.
* gdb.fortran/pointer-to-pointer.f90: New file.
Tom de Vries [Fri, 24 Jul 2020 22:23:06 +0000 (00:23 +0200)]
[gdb/symtab] Ignore zero line table entries
The DWARF standard states for the line register in the line number information
state machine the following:
...
An unsigned integer indicating a source line number. Lines are numbered
beginning at 1. The compiler may emit the value 0 in cases where an
instruction cannot be attributed to any source line.
...
So, it's possible to have a zero line number in the DWARF line table.
This is currently not handled by GDB. The zero value is read in as any other
line number, but internally the zero value has a special meaning:
end-of-sequence, so the line table entry ends up having a different
interpretation than intended in some situations.
I've created a test-case where various aspects are tested, which has these 4
interesting tests.
1. Next-step through a zero-line instruction, is_stmt == 1
gdb.dwarf2/dw2-line-number-zero.exp: bar1, 2nd next
2. Next-step through a zero-line instruction, is_stmt == 0
gdb.dwarf2/dw2-line-number-zero.exp: bar2, 2nd next
3. Show source location at zero-line instruction, is_stmt == 1
gdb.dwarf2/dw2-line-number-zero.exp: continue to breakpoint: bar1_label_3
4. Show source location at zero-line instruction, is_stmt == 0
gdb.dwarf2/dw2-line-number-zero.exp: continue to breakpoint: bar2_label_3
And we have the following results:
8.3.1, 9.2:
...
FAIL: gdb.dwarf2/dw2-line-number-zero.exp: bar1, 2nd next
PASS: gdb.dwarf2/dw2-line-number-zero.exp: bar2, 2nd next
PASS: gdb.dwarf2/dw2-line-number-zero.exp: continue to breakpoint: bar1_label_3
FAIL: gdb.dwarf2/dw2-line-number-zero.exp: continue to breakpoint: bar2_label_3
...
commit 8c95582da8 "gdb: Add support for tracking the DWARF line table is-stmt
field":
...
PASS: gdb.dwarf2/dw2-line-number-zero.exp: bar1, 2nd next
PASS: gdb.dwarf2/dw2-line-number-zero.exp: bar2, 2nd next
FAIL: gdb.dwarf2/dw2-line-number-zero.exp: continue to breakpoint: bar1_label_3
FAIL: gdb.dwarf2/dw2-line-number-zero.exp: continue to breakpoint: bar2_label_3
...
commit d8cc8af6a1 "[gdb/symtab] Fix line-table end-of-sequence sorting",
master:
FAIL: gdb.dwarf2/dw2-line-number-zero.exp: bar1, 2nd next
FAIL: gdb.dwarf2/dw2-line-number-zero.exp: bar2, 2nd next
PASS: gdb.dwarf2/dw2-line-number-zero.exp: continue to breakpoint: bar1_label_3
PASS: gdb.dwarf2/dw2-line-number-zero.exp: continue to breakpoint: bar2_label_3
...
The regression in test 2 at commit d8cc8af6a1 was filed as PR symtab/26243,
where clang emits zero line numbers.
The way to fix all tests is to make sure line number zero internally doesn't
clash with special meaning values, and by handling it appropriately
everywhere. That however looks too intrusive for the GDB 10 release.
Instead, we decide to ensure defined behaviour for line number zero by
ignoring it. This gives us back the test results from before commit d8cc8af6a1, fixing PR26243.
We mark the FAILs for tests 3 and 4 as KFAILs. Test 4 was already failing for
the 9.2 release, and we consider the regression of test 3 from gdb 9.2 to gdb
10 the cost for having defined behaviour.
Build and reg-tested on x86_64-linux.
gdb/ChangeLog:
2020-07-25 Tom de Vries <tdevries@suse.de>
PR symtab/26243
* dwarf2/read.c (lnp_state_machine::record_line): Ignore zero line
entries.
gdb/testsuite/ChangeLog:
2020-07-25 Tom de Vries <tdevries@suse.de>
PR symtab/26243
* gdb.dwarf2/dw2-line-number-zero.c: New test.
* gdb.dwarf2/dw2-line-number-zero.exp: New file.
Tom de Vries [Fri, 24 Jul 2020 12:10:50 +0000 (14:10 +0200)]
[gdb/testsuite] Require gnatmake-8 for gdb.ada/mi_prot.exp
With gcc-7, I run into:
...
gcc -c -I./ -gnata -Isrc/gdb/testsuite/gdb.ada/mi_prot -g -lm -I- \
src/gdb/testsuite/gdb.ada/mi_prot/prot.adb^M
prot.adb:21:04: info: "Obj_Type" is frozen here, aspects evaluated at this \
point^M
prot.adb:23:09: visibility of aspect for "Obj_Type" changes after freeze \
point^M
gnatmake: "src/gdb/testsuite/gdb.ada/mi_prot/prot.adb" compilation error^M
compiler exited with status 1
...
FAIL: gdb.ada/mi_prot.exp: compilation prot.adb
...
Fix this by requiring gnatmake-8 for this test-case.
Tested on x86_64-linux, with gnatmake-7, gnatmake-8 and gnatmake-11.
PR ld/26288: Allow the use of `--just-symbols' with ET_EXEC input
Fix a regression from commit a87e1817a435 ("Have the linker fail if any
attempt to link in an executable is made.") and do not reject ET_EXEC
input supplied with the `--just-symbols' option. Such use is legitimate
as the file requested is not actually linked and only the symbols are
extracted. Furthermore it is often the most useful application, as
already observed in our documentation for the option, where it allows
"to refer symbolically to absolute locations of memory defined in other
programs."
Provide a set of tests for the use of ET_EXEC with `--just-symbols'.
These are excluded however for SH/PE targets because they complain if a
section's VMA is 0:
ld: zero vma section reloc detected: `.text' #0 f=32795
ld: zero vma section reloc detected: `.data' #1 f=291
and for x86_64/PE targets because they seem to hardwire the VMA:
ld/
PR ld/26288
* ldelf.c (ldelf_after_open): Do not reject ET_EXEC input
supplied with `--just-symbols'.
* testsuite/ld-misc/just-symbols.exp: New test script.
* testsuite/ld-misc/just-symbols-1.dd: New test dump.
* testsuite/ld-misc/just-symbols.ld: New test linker script.
* testsuite/ld-misc/just-symbols-0.s: New test source.
* testsuite/ld-misc/just-symbols-1.s: New test source.
PR ld/26288: Revert obsolete part of PR ld/26047 fix
Revert commit a3fc941881e4 ("Stop the linker from accepting executable
ELF files as inputs to other links."), which has been made obsolete by
commit a87e1817a435 ("Have the linker fail if any attempt to link in an
executable is made."). An earlier check triggers added with the latter
commit making the piece of code removed dead.
ld/
PR ld/26288
Revert:
PR 26047
* ldelf.c (ldelf_after_open): Fail if attempting to link one
executable into another.
- We create a new frame_info object in restore_selected_frame (by
calling find_relative_frame)
- The frame is allocated on the frame_cache_obstack
- In frame_unwind_try_unwinder, we try to find an unwinder for that
frame
- While trying unwinders, memory read fails because the remote target
closes, because of "monitor exit"
- That calls reinit_frame_cache (as shown above), which resets
frame_cache_obstack
- When handling the exception in frame_unwind_try_unwinder, we try to
set some things on the frame_info object (like *this_cache, which
in fact tries to write into frame_info::prologue_cache), but the
frame_info object is no more, it went away with the obstack.
Fix this by maintaining a frame cache generation counter. Then in
exception handling code paths, don't touch frame objects if the
generation is not the same as it was on entry.
This commit generalizes the gdb.server/server-kill.exp testcase and
reuses it to test the scenario in question. The new tests fail
without the GDB fix.
gdb/ChangeLog:
* frame-unwind.c (frame_unwind_try_unwinder): On exception, don't
touch THIS_CACHE/THIS_FRAME if the frame cache was cleared
meanwhile.
* frame.c (frame_cache_generation, get_frame_cache_generation):
New.
(reinit_frame_cache): Increment FRAME_CACHE_GENERATION.
(get_prev_frame_if_no_cycle): On exception, don't touch
PREV_FRAME/THIS_FRAME if the frame cache was cleared meanwhile.
* frame.h (get_frame_cache_generation): Declare.
gdb/testsuite/ChangeLog:
* gdb.server/server-kill.exp (prepare): New, factored out from the
top level.
(kill_server): New.
(test_tstatus, test_unwind_nosyms, test_unwind_syms): New.
(top level) : Call test_tstatus, test_unwind_nosyms, test_unwind_syms.
Tom de Vries [Thu, 23 Jul 2020 11:45:46 +0000 (13:45 +0200)]
[gdb/tui] Fix Wmaybe-uninitialized warning in tui-winsource.c
When compiling with CFLAGS/CXXFLAGS="-O0 -g -Wall" and using g++ 11.0.0, we
run into:
...
src/gdb/tui/tui-winsource.c: In function \
'void tui_update_all_breakpoint_info(breakpoint*)':
src/gdb/tui/tui-winsource.c:427:58: warning: '<unknown>' may be used \
uninitialized [-Wmaybe-uninitialized]
427 | for (tui_source_window_base *win : tui_source_windows ())
| ^
In file included from src/gdb/tui/tui-winsource.c:38:
src/gdb/tui/tui-winsource.h:236:30: note: by argument 1 of type \
'const tui_source_windows*' to 'tui_source_window_iterator \
tui_source_windows::begin() const' declared here
236 | tui_source_window_iterator begin () const
| ^~~~~
src/gdb/tui/tui-winsource.c:427:58: note: '<anonymous>' declared here
427 | for (tui_source_window_base *win : tui_source_windows ())
| ^
...
The warning doesn't make sense for an empty struct, PR gcc/96295 has been
filed about that.
For now, work around the warning by defining a default constructor.
Build on x86_64-linux.
gdb/ChangeLog:
2020-07-23 Tom de Vries <tdevries@suse.de>
PR tui/26282
* tui/tui-winsource.h (struct tui_source_windows::tui_source_windows):
New default constructor.
Andrew Burgess [Tue, 21 Jul 2020 10:21:50 +0000 (11:21 +0100)]
gdb/disassembly: Update to handle non-statement addresses
After the introduction of support for non-statement addresses in the
line table, the output for 'disassemble /m' can be broken in some
cases.
With the /m format disassembly GDB associates a set of addresses with
each line, these addresses are then sorted and printed for each line.
When the non-statement support was added GDB was incorrectly told to
ignore non-statement instructions, and not add these to the result
table. This means that these instructions are completely missing from
the output.
This commit removes the code that caused non-statement lines to be
ignored.
A result of this change is that GDB will now potentially include new
line numbers in the 'disassemble /m' output, lines that previously
were only in the line table as non-statement lines will now appear in
the disassembly output. This feels like an improvement though.
* lib/gdb.exp (gdb_wrapper_init): Reset
"gdb_wrapper_initialized" to 0 if "wrapper_file" does
not exist.
attempted to fix problems finding the gdb test wrapper gdb_tg.o in
some tests that cd to some non-default directory by rebuilding also
the test wrapper in that directory. This had the side-effect of
leaving these .o files in various places in the GDB source directory
tree.
Furthermore, while the tests that cd to some non-default directory
cannot run on remote host, the code that was added to probe for the
presence of the wrapper file was also specific to host == build.
This patch reverts the problematic parts of that commit and replaces
it with forcing use of an absolute (rather than relative) pathname to
the .o file for linking when host == build.
While debugging this patch, I also observed that use of the construct
"[info exists gdb_wrapper_file]" was not reliable for detecting when
that variable had been initialized by gdb_wrapper_init. I rewrote
that so that the variable is always initialized and has a value of an
empty string when no wrapper file is needed.
gdb/testsuite/
* lib/gdb.exp (gdb_wrapper_file, gdb_wrapper_flags):
Initialize to empty string at top level.
(gdb_wrapper_init): Revert check for file existence on build.
Build the wrapper in its default place, not a build-specific
location. When host == build, make the pathname absolute.
(gdb_compile): Delete leftover declaration of
gdb_wrapper_initialized. Check gdb_wrapper_file being an empty
string instead of uninitialized.
In a nutshell, my earlier patches could not handle the case in
which a read-only mapping created with mmap() was created at
an address used by other file-backed read-only memory in use by
the process.
This problem has been fixed (for Linux, anyway) by the commit "Use
NT_FILE note section for reading core target memory".
When I run this test without any of my recent corefile patches,
I see these failures:
The ones involving mbuf_ro will almost certainly fail when run on
non-Linux systems; I've used setup_xfail on those tests to prevent
them from outright FAILing when not run on Linux. For a time, I
had considered skipping these tests altogether when not run on
Linux, but I changed my mind due to this failure...
I think it *should* pass without my recent corefile patches. The fact
that it doesn't is likely due to a bug in GDB. The following
interaction with GDB demonstrates the problem:
That latter output was obtained with the use of all of my current
corefile patches. I see no failures on Linux when running this test
with my current set of corefile patches. I tested 3 architectures:
x86_64, s390x, and aarch64.
I also tested on FreeBSD 12.1-RELEASE. I see the following results
both with and without the current set of core file patches:
# of expected passes 26
# of expected failures 8
Of particular interest is that I did *not* see the problematic mbuf_rw
failure noted earlier (both with and without the core file patches).
I still don't have an explanation for why this failure occurred on
Linux. Prior to running the tests, I had hypothesized that I'd see
this failure on FreeBSD too, but testing shows that this is not the
case.
Also of importance is that we see no FAILs with this test on FreeBSD
which indicates that I XFAILed the correct tests.
This version runs the interesting tests twice, once with a kernel
created core file and another time with a gcore created core file.
It also does a very minimal test of the new command "maint print
core-file-backed-mappings".
gdb/testsuite/ChangeLog:
* gdb.base/corefile2.exp: New file.
* gdb.base/coremaker2.exp: New file.
Kevin Buettner [Sat, 4 Jul 2020 04:55:51 +0000 (21:55 -0700)]
Add new command "maint print core-file-backed-mappings"
I wrote a read_core_file_mappings method for FreeBSD and then registered
this gdbarch method. I saw some strange behavior while testing it and
wanted a way to make sure that mappings were being correctly loaded
into corelow.c, so I wrote the new command which is the topic of this
commit. I think it might be occasionally useful for debugging strange
corefile behavior.
With regard to FreeBSD, my work isn't ready yet. Unlike Linux,
FreeBSD puts all mappings into its core file note. And, unlike Linux,
it doesn't dump load segments which occupy no space in the file. So
my (perhaps naive) implementation of a FreeBSD read_core_file_mappings
didn't work all that well: I saw more failures in the corefile2.exp
tests than without it. I think it should be possible to make FreeBSD
work as well as Linux, but it will require doing something with all of
the mappings, not just the file based mappings that I was considering.
In the v4 series, Pedro asked the following:
I don't understand what this command provides that "info proc
mappings" doesn't? Can you give an example of when you'd use this
command over "info proc mappings" ?
On Linux, "info proc mappings" and "maint print core-file-backed-mappings"
will produce similar, possibly identical, output. This need not be
the case for other OSes. E.g. on FreeBSD, had I finished the
implementation, the output from these commands would have been very
different. The FreeBSD "info proc mappings" command would show
additional (non-file-backed) mappings in addition to at least one
additional field (memory permissions) for each mapping.
As noted earlier, I was seeing some unexpected behavior while working
on the FreeBSD implementation and wanted to be certain that the
mappings were being correctly loaded by corelow.c. "info proc
mappings" prints the core file mappings, but doesn't tell us anything
about whether they've been loaded by corelow.c This new maintenance
command directly interrogates the data structures and prints the
values found there.
gdb/ChangeLog:
* corelow.c (gdbcmd.h): Include.
(core_target::info_proc_mappings): New method.
(get_current_core_target): New function.
(maintenance_print_core_file_backed_mappings): New function.
(_initialize_corelow): Add core-file-backed-mappings to
"maint print" commands.
Kevin Buettner [Sat, 4 Jul 2020 03:10:22 +0000 (20:10 -0700)]
Adjust coredump-filter.exp to account for NT_FILE note handling
This commit makes adjustments to coredump-filter.exp to account
for the fact that NT_FILE file-backed mappings are now available
when a core file is loaded. Thus, a test which was expected
to PASS when a memory region was determined to be unavailable
(due to no file-backed mappings being available) will now FAIL
due to those mappings being available from having loaded the
NT_FILE note.
I had originally marked the test as XFAIL, but Mihails Strasuns
suggested a much better approach:
1) First test that it still works if file is accessible in the
filesystem.
2) Temporarily move / rename the file and test that disassembly
doesn't work anymore.
That's what this commit implements.
gdb/testsuite/ChangeLog:
* gdb.base/coredump-filter.exp: Add second
non-Private-Shared-Anon-File test.
(test_disasm): Rename binfile for test which is expected
to fail.
Kevin Buettner [Wed, 1 Jul 2020 13:34:50 +0000 (06:34 -0700)]
gcore command: Place all file-backed mappings in NT_FILE note
When making a core file with the GDB's gcore command on Linux,
the same criteria used for determining which mappings should be
dumped were also being used for determining which entries should
be placed in the NT_FILE note. This is wrong; we want to place
all file-backed mappings in this note.
The predicate function, dump_mapping_p, was used to determine whether
or not to dump a mapping from within linux_find_memory_regions_full.
This commit leaves this predicate in place, but adds a new parameter,
should_dump_mapping_p, to linux_find_memory_regions_full. It then
calls should_dump_mapping_p instead of dump_mapping_p. dump_mapping_p
is passed to linux_find_memory_regions_full at one call site; at the
other call site, dump_note_entry_p is passed instead.
gdb/ChangeLog:
* linux-tdep.c (dump_note_entry_p): New function.
(linux_dump_mapping_p_ftype): New typedef.
(linux_find_memory_regions_full): Add new parameter,
should_dump_mapping_p.
(linux_find_memory_regions): Adjust call to
linux_find_memory_regions_full.
(linux_make_mappings_core_file_notes): Use dump_note_entry_p in
call to linux_find_memory_regions_full.
Kevin Buettner [Tue, 16 Jun 2020 18:39:22 +0000 (11:39 -0700)]
Add test for accessing read-only mmapped data in a core file
This test passes when run using a GDB with my corefile patches. When
run against a GDB without my patches, I see the following failures,
the first of which is due to the test added by this commit:
FAIL: gdb.base/corefile.exp: accessing read-only mmapped data in core file (mapping address not found in core file)
FAIL: gdb.base/corefile.exp: accessing anonymous, unwritten-to mmap data
gdb/testsuite/ChangeLog:
* gdb.base/corefile.exp: Add test "accessing read-only mmapped
data in core file".
* gdb.base/coremaker.c (buf2ro): New global.
(mmapdata): Add a read-only mmap mapping.
Kevin Buettner [Fri, 12 Jun 2020 02:20:03 +0000 (19:20 -0700)]
Use NT_FILE note section for reading core target memory
In his reviews of my v1 and v2 corefile related patches, Pedro
identified two cases which weren't handled by those patches.
In https://sourceware.org/pipermail/gdb-patches/2020-May/168826.html,
Pedro showed that debugging a core file in which mmap() is used to
create a new mapping over an existing file-backed mapping will
produce incorrect results. I.e, for his example, GDB would
show:
(gdb) disassemble main
Dump of assembler code for function main:
0x00000000004004e6 <+0>: push %rbp
0x00000000004004e7 <+1>: mov %rsp,%rbp
=> 0x00000000004004ea <+4>: callq 0x4003f0 <abort@plt>
End of assembler dump.
This sort of looks like it might be correct, but is not due to the
fact that mmap(...MAP_FIXED...) was used to create a mapping (of all
zeros) on top of the .text section. So, the correct result should be:
(gdb) disassemble main
Dump of assembler code for function main:
0x00000000004004e6 <+0>: add %al,(%rax)
0x00000000004004e8 <+2>: add %al,(%rax)
=> 0x00000000004004ea <+4>: add %al,(%rax)
0x00000000004004ec <+6>: add %al,(%rax)
0x00000000004004ee <+8>: add %al,(%rax)
End of assembler dump.
The other case that Pedro found involved an attempted examination of a
particular section in the test case from gdb.base/corefile.exp. On
Fedora 27 or 28, the following behavior may be observed:
FYI, this section appears to be unrelocated vtable data. See
https://sourceware.org/pipermail/gdb-patches/2020-May/168331.html for
a detailed analysis.
The important thing here is that GDB should be able to access this
address since it should be backed by the shared library. I.e. it
should do this:
In a nutshell, this commit opens a "binary" target BFD for each of the
files that are mentioned in an NT_FILE / .note.linuxcore.file note
section. It then uses these mappings instead of the file stratum
mappings that GDB has used in the past.
If this note section doesn't exist or is mangled for some reason, then
GDB will use the file stratum as before. Should this happen, then
we can expect both of the above problems to again be present.
See the code comments in the commit for other details.
gdb/ChangeLog:
* corelow.c (solist.h, unordered_map): Include.
(class core_target): Add field m_core_file_mappings and
method build_file_mappings.
(core_target::core_target): Call build_file_mappings.
(core_target::~core_target): Free memory associated with
m_core_file_mappings.
(core_target::build_file_mappings): New method.
(core_target::xfer_partial): Use m_core_file_mappings
for memory transfers.
* linux-tdep.c (linux_read_core_file_mappings): New
function.
(linux_core_info_proc_mappings): Rewrite to use
linux_read_core_file_mappings.
(linux_init_abi): Register linux_read_core_file_mappings.
Kevin Buettner [Fri, 3 Jul 2020 20:32:08 +0000 (13:32 -0700)]
Add new gdbarch method, read_core_file_mappings
The new gdbarch method, read_core_file_mappings, will be used for
reading file-backed mappings from a core file. It'll be used
for two purposes: 1) to construct a table of file-backed mappings
in corelow.c, and 2) for display of core file mappings.
For Linux, I tried a different approach in which knowledge of the note
format was placed directly in corelow.c. This seemed okay at first;
it was only one note format and the note format was fairly simple.
After looking at FreeBSD's note/mapping reading code, I concluded
that it's best to leave architecture specific details for decoding
the note in (architecture specific) tdep files.
With regard to display of core file mappings, I experimented with
placing the mappings display code in corelow.c. It has access to the
file-backed mappings which were read in when the core file was loaded.
And, better, still common code could be used for all architectures.
But, again, the FreeBSD mapping code convinced me that this was not
the best approach since it has even more mapping info than Linux.
Display code which would work well for Linux will leave out mappings
as well as protection info for mappings.
So, for these reasons, I'm introducing a new gdbarch method for
reading core file mappings.
gdb/ChangeLog:
* arch-utils.c (default_read_core_file_mappings): New function.
* arch-utils.c (default_read_core_file_mappings): Declare.
* gdbarch.sh (read_core_file_mappings): New gdbarch method.
* gdbarch.h, gdbarch.c: Regenerate.
Kevin Buettner [Fri, 12 Jun 2020 01:58:49 +0000 (18:58 -0700)]
Update binary_get_section_contents to seek using section's file position
I have a patch for GDB which opens and reads from BFDs using the
"binary" target. However, for it to work, we need to be able to get a
section's contents based from the file position of that section.
At the moment, reading a section's contents will always read from the
start of the file regardless of where that section is located. While
this was fine for the original use of the "binary" target, it won't
work for my use case. This change shouldn't impact any existing
callers due to the fact that the single .data section is initialized
with a filepos of 0.
bfd/ChangeLog:
* binary.c (binary_get_section_contents): Seek using offset
from section's file position.
Now let's run it from GDB. I've already placed a breakpoint on the
line with the abort() call and have run to that breakpoint.
Breakpoint 1, main (argc=1, argv=0x7fffffffd678) at mkmmapcore.c:11
11 abort ();
(gdb) x/x buf
0x7ffff7fcb000: 0x00000000
Note that we can examine the memory allocated via the call to mmap().
Now let's try debugging a core file created by running this program.
Depending on your system, in order to make a core file, you may have to
run the following as root (or using sudo):
echo core > /proc/sys/kernel/core_pattern
It may also be necessary to do:
ulimit -c unlimited
I'm using Fedora 31. YMMV if you're using one of the BSDs or some other
(non-Linux) system.
This is what things look like when we debug the core file:
[kev@f31-1 tmp]$ gdb -q ./mkmmapcore core.304767
Reading symbols from ./mkmmapcore...
[New LWP 304767]
Core was generated by `/tmp/mkmmapcore'.
Program terminated with signal SIGABRT, Aborted.
#0 __GI_raise (sig=sig@entry=6) at ../sysdeps/unix/sysv/linux/raise.c:50
50 return ret;
(gdb) x/x buf
0x7ffff7fcb000: Cannot access memory at address 0x7ffff7fcb000
Note that we can no longer access the memory region allocated by mmap().
Back in 2007, a hack for GDB was added to _bfd_elf_make_section_from_phdr()
in bfd/elf.c:
/* Hack for gdb. Segments that have not been modified do
not have their contents written to a core file, on the
assumption that a debugger can find the contents in the
executable. We flag this case by setting the fake
section size to zero. Note that "real" bss sections will
always have their contents dumped to the core file. */
if (bfd_get_format (abfd) == bfd_core)
newsect->size = 0;
You can find the entire patch plus links to other discussion starting
here:
This hack sets the size of certain BFD sections to 0, which
effectively causes GDB to ignore them. I think it's likely that the
bug described above existed even before this hack was added, but I
have no easy way to test this now.
The output from objdump -h shows the result of this hack:
(This is a digression, but I think it's good that objdump will now show
the correct size.)
If we remove the hack from bfd/elf.c, but do nothing to GDB, we'll
see the following regression:
FAIL: gdb.base/corefile.exp: print coremaker_ro
The reason for this is that all sections which have the BFD flag
SEC_ALLOC set, but for which SEC_HAS_CONTENTS is not set no longer
have zero size. Some of these sections have data that can (and should)
be read from the executable. (Sections for which SEC_HAS_CONTENTS
is set should be read from the core file; sections which do not have
this flag set need to either be read from the executable or, failing
that, from the core file using whatever BFD decides is the best value
to present to the user - it uses zeros.)
At present, due to the way that the target strata are traversed when
attempting to access memory, the non-SEC_HAS_CONTENTS sections will be
read as zeroes from the process_stratum (which in this case is the
core file stratum) without first checking the file stratum, which is
where the data might actually be found.
What we should be doing is this:
- Attempt to access core file data for SEC_HAS_CONTENTS sections.
- Attempt to access executable file data if the above fails.
- Attempt to access core file data for non SEC_HAS_CONTENTS sections, if
both of the above fail.
This corresponds to the analysis of Daniel Jacobowitz back in 2007
when the hack was added to BFD:
The difference, observed by Pedro in his review of my v1 patches, is
that I'm using "the section flags as proxy for the p_filesz/p_memsz
checks."
gdb/ChangeLog:
PR corefiles/25631
* corelow.c (core_target:xfer_partial): Revise
TARGET_OBJECT_MEMORY case to consider non-SEC_HAS_CONTENTS
case after first checking the stratum beneath the core
target.
(has_all_memory): Return true.
* target.c (raw_memory_xfer_partial): Revise comment
regarding use of has_all_memory.
Kevin Buettner [Thu, 5 Mar 2020 00:42:41 +0000 (17:42 -0700)]
section_table_xfer_memory: Replace section name with callback predicate
This patch is motivated by the need to be able to select sections
that section_table_xfer_memory_partial should consider for memory
transfers. I'll use this facility in the next patch in this series.
section_table_xfer_memory_partial() can currently be passed a section
name which may be used to make name-based selections. This is similar
to what I want to do, except that I want to be able to consider
section flags instead of the name.
I'm replacing the section name parameter with a predicate that,
when passed a pointer to a target_section struct, will return
true if that section should be further considered, or false which
indicates that it shouldn't.
I've converted the one existing use where a non-NULL section
name is passed to section_table_xfer_memory_partial(). Instead
of passing the section name, it now looks like this:
Kevin Buettner [Wed, 13 May 2020 00:44:19 +0000 (17:44 -0700)]
Adjust corefile.exp test to show regression after bfd hack removal
In his review of my BZ 25631 patch series, Pedro was unable to
reproduce the regression which should occur after patch #1, "Remove
hack for GDB which sets the section size to 0", is applied.
Pedro was using an ld version older than 2.30. Version 2.30
introduced the linker option -z separate-code. Here's what the man
page has to say about it:
Create separate code "PT_LOAD" segment header in the object. This
specifies a memory segment that should contain only instructions
and must be in wholly disjoint pages from any other data.
In ld version 2.31, use of separate-code became the default for
Linux/x86. So, really, 2.31 or later is required in order to see the
regression that occurs in recent Linux distributions when only the
bfd hack removal patch is applied.
For the test case in question, use of the separate-code linker option
means that the global variable "coremaker_ro" ends up in a separate
load segment (though potentially with other read-only data). The
upshot of this is that when only patch #1 is applied, GDB won't be
able to correctly access coremaker_ro. The reason for this is due
to the fact that this section will now have a non-zero size, but
will not have contents from the core file to find this data.
So GDB will ask BFD for the contents and BFD will respond with
zeroes for anything from those sections. GDB should instead be
looking in the executable for this data. Failing that, it can
then ask BFD for a reasonable value. This is what a later patch
in this series does.
When using ld versions earlier than 2.31 (or 2.30 w/ the
-z separate-code option explicitly provided to the linker), there is
the possibility that coremaker_ro ends up being placed near other data
which is recorded in the core file. That means that the correct value
will end up in the core file, simply because it resides on a page that
the kernel chooses to put in the core file. This is why Pedro wasn't
able to reproduce the regression that should occur after fixing the
BFD hack.
This patch places a big chunk of memory, two pages worth on x86, in
front of "coremaker_ro" to attempt to force it onto another page
without requiring use of that new-fangled linker switch.
Speaking of which, I considered changing the test to use
-z separate-code, but this won't work because it didn't
exist prior to version 2.30. The linker would probably complain
of an unrecognized switch. Also, it likely won't be available in
other linkers not based on current binutils. I.e. it probably won't
work in FreeBSD, NetBSD, etc.
To make this more concrete, this is what *should* happen when
attempting to access coremaker_ro when only patch #1 is applied:
Core was generated by `/mesquite2/sourceware-git/f28-coresegs/bld/gdb/testsuite/outputs/gdb.base/coref'.
Program terminated with signal SIGABRT, Aborted.
#0 0x00007f68205deefb in raise () from /lib64/libc.so.6
(gdb) p coremaker_ro
$1 = 0
Note that this result is wrong; 201 should have been printed instead.
But that's the point of the rest of the patch series.
However, without this commit, or when using an old Linux distro with
a pre-2.31 ld, this is what you might see instead:
Core was generated by `/mesquite2/sourceware-git/f28-coresegs/bld/gdb/testsuite/outputs/gdb.base/coref'.
Program terminated with signal SIGABRT, Aborted.
#0 0x00007f63dd658efb in raise () from /lib64/libc.so.6
(gdb) p coremaker_ro
$1 = 201
I.e. it prints the right answer, which sort of makes it seem like the
rest of the series isn't required.
Now, back to the patch itself... what should be the size of the memory
chunk placed before coremaker_ro?
It needs to be at least as big as the page size (PAGE_SIZE) from
the kernel. For x86 and several other architectures this value is
4096. I used MAPSIZE which is defined to be 8192 in coremaker.c.
So it's twice as big as what's currently needed for most Linux
architectures. The constant PAGE_SIZE is available from <sys/user.h>,
but this isn't portable either. In the end, it seemed simpler to
just pick a value and hope that it's big enough. (Running a separate
program which finds the page size via sysconf(_SC_PAGESIZE) and then
passes it to the compilation via a -D switch seemed like overkill
for a case which is rendered moot by recent linker versions.)
That hack mostly allowed GDB's handling of core files to continue to
work without any changes to GDB.
The problem with setting the section size to zero is that GDB won't
know how big that section is/was. Often, this doesn't matter because
the data in question are found in the exec file. But it can happen
that the section describes memory that had been allocated, but never
written to. In this instance, the contents of that memory region are
not written to the core file. Also, since the region in question was
dynamically allocated, it won't appear in the exec file. We don't
want these regions to appear as inaccessible to GDB (since they *were*
accessible when the process was live), so it's important that GDB know
the size of the region.
I've made changes to GDB which correctly handles this case. When
attempting to access memory, GDB will first consider core file data
for which both SEC_ALLOC and SEC_HAS_CONTENTS is set. Next, if that
fails, GDB will attempt to find the data in the exec file. Finally,
if that also fails, GDB will attempt to access memory in the sections
which are flagged as SEC_ALLOC, but not SEC_HAS_CONTENTS.
bfd/ChangeLog:
* elf.c (_bfd_elf_make_section_from_phdr): Remove hack for GDB.
Tom Tromey [Wed, 22 Jul 2020 18:28:33 +0000 (12:28 -0600)]
Fix crash in -stack-list-arguments
-stack-list-arguments will crash when stopped in an Ada procedure that
has an argument with a certain name ("_objectO" -- which can only be
generated by the compiler). The bug occurs because lookup_symbol will
fail in this case.
This patch changes -stack-list-arguments to mirror what is done with
arguments elsewhere. (As an aside, I don't understand why this lookup
is even needed, but I assume it is some stabs thing?)
In the longer term I think it would be good to share this code between
MI and the CLI. However, due to the upcoming release, I preferred a
more local fix.
gdb/ChangeLog
2020-07-22 Tom Tromey <tromey@adacore.com>
* mi/mi-cmd-stack.c (list_args_or_locals): Use
lookup_symbol_search_name.
gdb/testsuite/ChangeLog
2020-07-22 Tom Tromey <tromey@adacore.com>
* gdb.ada/mi_prot.exp: New file.
* gdb.ada/mi_prot/pkg.adb: New file.
* gdb.ada/mi_prot/pkg.ads: New file.
* gdb.ada/mi_prot/prot.adb: New file.
Nick Alcock [Tue, 21 Jul 2020 14:38:08 +0000 (15:38 +0100)]
libctf: fixes for systems on which sizeof (void *) > sizeof (long)
Systems like mingw64 have pointers that can only be represented by 'long
long'. Consistently cast integers stored in pointers through uintptr_t
to cater for this.
The recent commit "libctf, binutils: support CTF archives like objdump"
broke opening of CTF archives on big-endian platforms.
This didn't affect anyone much before now because the linker never
emitted CTF archives because it wasn't detecting ambiguous types
properly: now it does, and this bug becomes obvious.
Fix trivial.
libctf/
* ctf-archive.c (ctf_arc_bufopen): Endian-swap the archive magic
number if needed.
Nick Alcock [Tue, 14 Jul 2020 17:33:11 +0000 (18:33 +0100)]
ld, testsuite: do not run CTF tests at all on non-ELF for now
Right now, the linker is not emitting CTF sections on (at least some)
non-ELF platforms, because work similar to that done for ELF needs to be
done to each platform in turn to emit linker-generated sections whose
contents are programmatically derived. (Or something better needs to be
done.)
So, for now, the CTF tests will fail on non-ELF for lack of a .ctf
section in the output: so skip the CTF tests there temporarily.
(This is not the same as the permanent skip of the diags tests, which is
done because the input for those is assembler that depends on the ELF
syntax of pseudos like .section: this is only a temporary skip, until
the linker grows support for CTF on more targets.)
ld/
* testsuite/ld-ctf/ctf.exp: Skip on non-ELF for now.
Nick Alcock [Thu, 11 Jun 2020 19:19:07 +0000 (20:19 +0100)]
ld: do not produce one empty output .ctf section for every input .ctf
The trick we use to prevent ld doing as it does for almost all other
sections and copying the input CTF section into the output has recently
broken, causing output to be produced with a valid CTF section followed
by massive numbers of CTF sections, one per .ctf in the input (minus
one, for the one that was filled out by ctf_link). Their size is being
forcibly set to zero, but they're still present, wasting space and
looking ridiculous.
The fix is to exclude these unwanted input sections from being present
in the output. We tried this before and it broke things, because if you
exclude all the .ctf sections there isn't going to be one in the output
so there is nowhere to put the deduplicated CTF. The solution to that is
really simple: set SEC_EXCLUDE on *all but one* CTF section. We don't
care which one (they're all the same once their size has been zeroed),
so just pick the first we see.
ld/
* ldlang.c (ldlang_open_ctf): Set SEC_EXCLUDE on all but the
first input .ctf section.
Nick Alcock [Thu, 11 Jun 2020 14:44:48 +0000 (15:44 +0100)]
ld, testsuite: only run CTF tests when ld and GCC support CTF
The CTF testsuite runs GCC to generate CTF that it knows matches the
input .c files before doing a run_dump_test over it. So we need a GCC
capable of doing that, and we need to always avoid running those tests
if libctf was disabled because the linker will never be capable of it.
Uses the new cc option to run_dump_test to compile most tests from C
code, ensuring that the types in the C code accurately describe what the
.d file is testing.
(Some tests, mostly those testing malformed CTF, run directly from .s,
or include both .s and .c.)
ld/
* testsuite/ld-ctf/ctf.exp: New file.
* testsuite/ld-ctf/A-2.c: New file.
* testsuite/ld-ctf/A.c: New file.
* testsuite/ld-ctf/B-2.c: New file.
* testsuite/ld-ctf/B.c: New file.
* testsuite/ld-ctf/C-2.c: New file.
* testsuite/ld-ctf/C.c: New file.
* testsuite/ld-ctf/array-char.c: New file.
* testsuite/ld-ctf/array-int.c: New file.
* testsuite/ld-ctf/array.d: New file.
* testsuite/ld-ctf/child-float.c: New file.
* testsuite/ld-ctf/child-int.c: New file.
* testsuite/ld-ctf/conflicting-cycle-1.B-1.d: New file.
* testsuite/ld-ctf/conflicting-cycle-1.B-2.d: New file.
* testsuite/ld-ctf/conflicting-cycle-1.parent.d: New file.
* testsuite/ld-ctf/conflicting-cycle-2.A-1.d: New file.
* testsuite/ld-ctf/conflicting-cycle-2.A-2.d: New file.
* testsuite/ld-ctf/conflicting-cycle-2.parent.d: New file.
* testsuite/ld-ctf/conflicting-cycle-3.C-1.d: New file.
* testsuite/ld-ctf/conflicting-cycle-3.C-2.d: New file.
* testsuite/ld-ctf/conflicting-cycle-3.parent.d: New file.
* testsuite/ld-ctf/conflicting-enums.d: New file.
* testsuite/ld-ctf/conflicting-typedefs.d: New file.
* testsuite/ld-ctf/cross-tu-1.c: New file.
* testsuite/ld-ctf/cross-tu-2.c: New file.
* testsuite/ld-ctf/cross-tu-conflicting-2.c: New file.
* testsuite/ld-ctf/cross-tu-cyclic-1.c: New file.
* testsuite/ld-ctf/cross-tu-cyclic-2.c: New file.
* testsuite/ld-ctf/cross-tu-cyclic-3.c: New file.
* testsuite/ld-ctf/cross-tu-cyclic-4.c: New file.
* testsuite/ld-ctf/cross-tu-cyclic-conflicting.d: New file.
* testsuite/ld-ctf/cross-tu-cyclic-nonconflicting.d: New file.
* testsuite/ld-ctf/cross-tu-into-cycle.d: New file.
* testsuite/ld-ctf/cross-tu-noncyclic.d: New file.
* testsuite/ld-ctf/cycle-1.c: New file.
* testsuite/ld-ctf/cycle-1.d: New file.
* testsuite/ld-ctf/cycle-2.A.d: New file.
* testsuite/ld-ctf/cycle-2.B.d: New file.
* testsuite/ld-ctf/cycle-2.C.d: New file.
* testsuite/ld-ctf/diag-ctf-version-0.d: New file.
* testsuite/ld-ctf/diag-ctf-version-0.s: New file.
* testsuite/ld-ctf/diag-ctf-version-2-unsupported-feature.d: New file.
* testsuite/ld-ctf/diag-ctf-version-2-unsupported-feature.s: New file.
* testsuite/ld-ctf/diag-ctf-version-f.d: New file.
* testsuite/ld-ctf/diag-ctf-version-f.s: New file.
* testsuite/ld-ctf/diag-cttname-invalid.d: New file.
* testsuite/ld-ctf/diag-cttname-invalid.s: New file.
* testsuite/ld-ctf/diag-cttname-null.d: New file.
* testsuite/ld-ctf/diag-cttname-null.s: New file.
* testsuite/ld-ctf/diag-cuname.d: New file.
* testsuite/ld-ctf/diag-cuname.s: New file.
* testsuite/ld-ctf/diag-decompression-failure.d: New file.
* testsuite/ld-ctf/diag-decompression-failure.s: New file.
* testsuite/ld-ctf/diag-parlabel.d: New file.
* testsuite/ld-ctf/diag-parlabel.s: New file.
* testsuite/ld-ctf/diag-parname.d: New file.
* testsuite/ld-ctf/diag-parname.s: New file.
* testsuite/ld-ctf/diag-unsupported-flag.d: New file.
* testsuite/ld-ctf/diag-unsupported-flag.s: New file.
* testsuite/ld-ctf/diag-wrong-magic-number-mixed.d: New file.
* testsuite/ld-ctf/diag-wrong-magic-number.d: New file.
* testsuite/ld-ctf/diag-wrong-magic-number.s: New file.
* testsuite/ld-ctf/enum-2.c: New file.
* testsuite/ld-ctf/enum.c: New file.
* testsuite/ld-ctf/function.c: New file.
* testsuite/ld-ctf/function.d: New file.
* testsuite/ld-ctf/slice.c: New file.
* testsuite/ld-ctf/slice.d: New file.
* testsuite/ld-ctf/super-sub-cycles.c: New file.
* testsuite/ld-ctf/super-sub-cycles.d: New file.
* testsuite/ld-ctf/typedef-int.c: New file.
* testsuite/ld-ctf/typedef-long.c: New file.
* testsuite/ld-ctf/union-1.c: New file.
Nick Alcock [Thu, 11 Jun 2020 14:48:55 +0000 (15:48 +0100)]
binutils, testsuite: allow compilation before doing run_dump_test
The CTF assembler emitted by GCC has architecture-dependent pseudos in
it, and is (obviously) tightly tied to a particular set of C source
files with specific types in them. The CTF tests do run_dump_test on
some candidate input, link it using the run_dump_test ld machinery, and
compare objdump --ctf output. To avoid skew, we'd like to be able
to easily regenerate the .s being scanned so that the .c doesn't get
out of sync with it, but since GCC emits arch-dependent pseudos, we
are forced to hand-hack the output every time (quite severely on some
arches, like x86-32 and -64, where every single pseudo used is not only
arch-dependent but undocumented).
To avoid this, teach run_dump_test how to optionally compile things
given new, optional additional flags passed in in the cc option.
Only sources with the .c suffix are compiled, so there is no effect on
any existing tests. The .s files go into the tmpdir, from which
existing run_dump_test code picks them up as usual.
Nick Alcock [Fri, 5 Jun 2020 22:18:06 +0000 (23:18 +0100)]
ld: new options --ctf-variables and --ctf-share-types
libctf recently changed to make it possible to not emit the CTF
variables section. Make this the default for ld: the variables section
is a simple name -> type mapping, and the names can be quite voluminous.
Nothing in the variables section appears in the symbol table, by
definition, so GDB cannot make use of them: special-purpose projects
that implement their own analogues of symbol table lookup can do so, but
they'll need to tell the linker to emit the variables section after all.
The new --ctf-variables option does this.
The --ctf-share-types option (valid values "share-duplicated" and
"share-unconflicted") allow the caller to specify the CTF link mode.
Most users will want share-duplicated, since it allows for more
convenient debugging: but very large projects composed of many decoupled
components may want to use share-unconflicted mode, which places types
that appear in only one TU into per-TU dicts. (They may also want to
relink the CTF using the ctf_link API and cu-mapping, to make their
"components" larger than a single TU. Right now the linker does not
expose the CU-mapping machinery. Perhaps it should in future to make
this use case easier.)
For now, giving the linker the ability to emit share-duplicated CTF lets
us add testcases for that mode to the testsuite.
Nick Alcock [Fri, 5 Jun 2020 21:57:06 +0000 (22:57 +0100)]
libctf, link: tie in the deduplicating linker
This fairly intricate commit connects up the CTF linker machinery (which
operates in terms of ctf_archive_t's on ctf_link_inputs ->
ctf_link_outputs) to the deduplicator (which operates in terms of arrays
of ctf_file_t's, all the archives exploded).
The nondeduplicating linker is retained, but is not called unless the
CTF_LINK_NONDEDUP flag is passed in (which ld never does), or the
environment variable LD_NO_CTF_DEDUP is set. Eventually, once we have
confidence in the much-more-complex deduplicating linker, I hope the
nondeduplicating linker can be removed.
In brief, what this does is traverses each input archive in
ctf_link_inputs, opening every member (if not already open) and tying
child dicts to their parents, shoving them into an array and
constructing a corresponding parents array that tells the deduplicator
which dict is the parent of which child. We then call ctf_dedup and
ctf_dedup_emit with that array of inputs, taking the outputs that result
and putting them into ctf_link_outputs where the rest of the CTF linker
expects to find them, then linking in the variables just as is done by
the nondeduplicating linker.
It also implements much of the CU-mapping side of things. The problem
CU-mapping introduces is that if you map many input CUs into one output,
this is saying that you want many translation units to produce at most
one child dict if conflicting types are found in any of them. This
means you can suddenly have multiple distinct types with the same name
in the same dict, which libctf cannot really represent because it's not
something you can do with C translation units.
The deduplicator machinery already committed does as best it can with
these, hiding types with conflicting names rather than making child
dicts out of them: but we still need to call it. This is done similarly
to the main link, taking the inputs (one CU output at a time),
deduplicating them, taking the output and making it an input to the
final link. Two (significant) optimizations are done: we share atoms
tables between all these links and the final link (so e.g. all type hash
values are shared, all decorated type names, etc); and any CU-mapped
links with only one input (and no child dicts) doesn't need to do
anything other than renaming the CU: the CU-mapped link phase can be
skipped for it. Put together, large CU-mapped links can save 50% of
their memory usage and about as much time (and the memory usage for
CU-mapped links is significant, because all those output CUs have to
have all their types stored in memory all at once).
Nick Alcock [Fri, 5 Jun 2020 17:35:46 +0000 (18:35 +0100)]
libctf, dedup: add deduplicator
This adds the core deduplicator that the ctf_link machinery calls
(possibly repeatedly) to link the CTF sections: it takes an array
of input ctf_file_t's and another array that indicates which entries in
the input array are parents of which other entries, and returns an array
of outputs. The first output is always the ctf_file_t on which
ctf_link/ctf_dedup/etc was called: the other outputs are child dicts
that have the first output as their parent.
include/
* ctf-api.h (CTF_LINK_SHARE_DUPLICATED): No longer unimplemented.
libctf/
* ctf-impl.h (ctf_type_id_key): New, the key in the
cd_id_to_file_t.
(ctf_dedup): New, core deduplicator state.
(ctf_file_t) <ctf_dedup>: New.
<ctf_dedup_atoms>: New.
<ctf_dedup_atoms_alloc>: New.
(ctf_hash_type_id_key): New prototype.
(ctf_hash_eq_type_id_key): Likewise.
(ctf_dedup_atoms_init): Likewise.
* ctf-hash.c (ctf_hash_eq_type_id_key): New.
(ctf_dedup_atoms_init): Likewise.
* ctf-create.c (ctf_serialize): Adjusted.
(ctf_add_encoded): No longer static.
(ctf_add_reftype): Likewise.
* ctf-open.c (ctf_file_close): Destroy the
ctf_dedup_atoms_alloc.
* ctf-dedup.c: New file.
* ctf-decls.h [!HAVE_DECL_STPCPY]: Add prototype.
* configure.ac: Check for stpcpy.
* Makefile.am: Add it.
* Makefile.in: Regenerate.
* config.h.in: Regenerate.
* configure: Regenerate.
projects / thirdparty / binutils-gdb.git / log
