When building with 'make -j20 -C gdb/doc all-doc' I often see problems
caused from trying to build some dvi files in parallel with some pdf
files. The problem files are: gdb.dvi and gdb.pdf; stabs.dvi and
stabs.pdf; and annotate.dvi and annotate.pdf.
The problem is that building these files create temporary files in the
local directory. There's already a race here that two make threads
might try to create these files at the same time.
But it gets worse, to avoid issues where a failed build could leave
these temporary files in a corrupted state, and so prevent the next
build from succeeding, the recipe for each of these files deletes all
the temporary files first, this obviously causes problems if some
other thread has already started the build and is relying on these
temporary files.
To work around this problem I propose we start using the --build and
--build-dir options for texi2dvi (which is the same tool used to
create the pdf files). These options were added in texinfo 4.9 which
was released in June 2007. We already require using a version of
texinfo after 4.9 (I tried to build with 4.13 and the doc build failed
as some of the texinfo constructs were not understood), so this patch
has not changed the minimum required version at all.
The --build flag allows the temporary files to be placed into a
sub-directory, and the --build-dir option allows us to control the
name of that sub-directory.
What we do is create a unique sub-directory for each target that
invokes texi2dvi, all of the unique sub-directories are created within
a single directory texi2dvi_tmpdir, and so after a complete doc build,
we are left with a build tree like this:
build/gdb/doc/
'-- texi2dvi_tmpdir/
|-- annotate_dvi/
|-- annotate_pdf/
|-- gdb_dvi/
|-- gdb_pdf/
|-- stabs_dvi/
'-- stabs_pdf/
I've left out all the individual files that live within these
directories for simplicity.
To avoid corrupted temporary files preventing a future build to
complete, each recipe deletes its associated sub-directory from within
texi2dvi_tmpdir/ before it attempts a build, this ensures a fresh
start each time.
And the mostlyclean target deletes texi2dvi_tmpdir/ and all its
sub-directories, ensuring that everything is cleaned up.
For me, with this fix in place, I can now run 'make -j20 -C gdb/doc
all-doc' without seeing any build problems.
Approved-By: Pedro Alves <pedro@palves.net>
There are two problems we encounter when trying to build the refcard
related target in parallel, i.e.:
$ make -j20 -C gdb/doc/ refcard.dvi refcard.ps refcard.pdf
These problems are:
(1) The refcard.dvi and refcard.pdf targets both try and generate the
tmp.sed and sedref.tex files. If two make threads end up trying
to create these files at the same time then the result is these
files become corrupted.
I've fixed this by creating a new rule that creates sedref.tex,
both refcard.dvi and refcard.pdf now depend on this, and make will
build sedref.tex just once. The tmp.sed file is now generated as
refcard.sed, this is generated and deleted as a temporary file
within the sedref.tex recipe.
(2) Having created sedref.tex the recipes for refcard.dvi and
refcard.pdf both run various LaTeX based tools with sedref.tex as
the input file. The problem with this is that these tools all
rely on creating temporary files calls sedref.*.
If the refcard.dvi and refcard.pdf rules run at the same time then
these temporary files clash and overwrite each other causing the
build to fail.
We already copy the result file in order to rename it, our input
file is sedref.tex which results in an output file named
sedref.dvi or sedref.pdf, but we actually want refcard.dvi or
refcard.pdf. So within the recipe for refcard.dvi I copy the
input file from sedref.tex to sedref_dvi.tex. Now all the temp
files are named sedref_dvi.* and the output is sedref_dvi.dvi, I
then rename this new output file to refcard.dvi.
I've done the same thing for refcard.pdf, but I copy the input
to sedref_pdf.tex.
In this way the temp files no longer clash, and both recipes can
safely run in parallel.
After this commit I was able to reliably build all of the refcard
targets in parallel. There should be no change in the final file.
Approved-By: Tom Tromey <tom@tromey.com>
In gdb/doc/Makefile.in the TEXI2POD variable is used to invoke
texi2pod.pl, which process the .texinfo files. This also handles the
'include' directives within the .texinfo files.
Like the texi2dvi and texi2pdf tools, texi2pod.pl handles the -I flag
to add search directories for resolving 'include' directives within
.texinfo files.
When GDB runs TEXI2POD we include gdb-cfg.texi, which then includes
GDBvn.texi.
When building from a git checkout the gdb-cfg.texi files and
GDBvn.texi files will be created in the build directory, which is
where texi2pod.pl is invoked, so the files will be found just fine.
However, for a GDB release we ship gdb-cfg.texi and GDBvn.texi in the
source tree, along with the generated manual (.1 and .5) files.
So when building a release, what normally happens is that we spot that
the .1 and .5 man files are up to date, and don't run the recipe to
regenerate these files.
However, if we deliberately touch the *.texinfo files in a release
source tree, and then try to rebuild the man files, we'll get an error
like this:
make: Entering directory '/tmp/release-build/build/gdb/doc'
TEXI2POD gdb.1
cannot find GDBvn.texi at ../../../gdb-16.0.50.20240529/gdb/doc/../../etc/texi2pod.pl line 251, <GEN0> line 16.
make: *** [Makefile:664: gdb.1] Error 2
make: Leaving directory '/tmp/release-build/build/gdb/doc'
The problem is that texi2pod.pl doesn't know to look in the source
tree for the GDBvn.texi file.
If we compare this to the recipe for creating (for example) gdb.dvi,
which uses texi2dvi, this recipe adds '-I $(srcdir)' to the texi2dvi
command line, which allows texi2dvi to find GDBvn.texi in the source
tree.
In this commit I add a similar -I option to the texi2pod.pl command
line. After this, given a GDB release, it is possible to edit (or
just touch) the gdb.texinfo file and rebuild the man pages, the
GDBvn.texi will be picked up from the source tree.
If however a dependency for GDBvn.texi is changed in a release tree
then GDBvn.texi will be regenerated into the build directory and this
will be picked up in preference to the GDBvn.texi in the source tree,
just as you would want.
Approved-By: Tom Tromey <tom@tromey.com>
In a git checkout of the source code we don't have a version.subst
file in the gdb/doc directory. When building the GDB docs the
version.subst file is generated on demand (we have a recipe for that).
However, in a release tar file we do include a copy of the
version.subst file in the source tree, as a result the version.subst
recipe will not be run.
If, in a release build, we force the running of any recipe that
depends on version.subst then we run into a problem. For example,
slightly confusingly, if we 'touch gdb/doc/version.subst' within the
unpacked source tree of a release, then 'make -C gdb/doc GDBvn.texi'
in the build tree, we'll see:
make: Entering directory '/tmp/build/build/gdb/doc'
GEN GDBvn.texi
sed: can't read version.subst: No such file or directory
make: Leaving directory '/tmp/build/build/gdb/doc'
The problem is that every reference to version.subst in GDB's Makefile
assumes that the version.subst file will always be in the build
directory.
Handily version.subst is always the first dependency in every recipe
that uses that file. As such we can replace references to
version.subst with $<, make will expand this to the location where the
dependency was found.
In the case of the man page generation, the reference to version.subst
is hidden inside POD2MAN. It seemed a little confusing adding a use
of $< within POD2MAN, so I've moved the use into the recipe, which I
think is clearer.
I've also added comments for the two rules that I've modified to
explain our use of $<.
After this change it is possible to rebuild the man pages even when
version.subst is located in the source tree.
Approved-By: Tom Tromey <tom@tromey.com>
We have two rules, one each for building the .1 and .5 man pages. The
only actual difference is that one rule passes --section=1 and the
other passes --section=5 (see the definitions of POD2MAN1 and POD2MAN5
respectively.
I figure by using the suffix from the target of the rule we can
combine these two rules into one.
I use:
$(subst .,,$(suffix $@))
This gets the suffix from the target, either '.1' or '.5', and the
'subst' removes the '.' leaving '1' or '5'.
Now that I'm not using a static pattern rule for building the man
pages, the advice in the 'make' documentation is to not use $*, so
I've moved away from that to instead use $(basename $@), e.g. for
'gdbinit.5' this gives 'gdbinit', which is what we want.
There should be no difference in what is created after this change.
Approved-By: Tom Tromey <tom@tromey.com>
The build recipe for gdb.dvi and gdb.pdf contains instructions for
copying the GDBvn.texi file from the source tree into the build
directory if the GDBvn.texi file doesn't already exist in the build
directory.
The gdb.dvi and gdb.pdf targets also have a dependency on GDBvn.texi,
and we have a recipe for building GDBvn.texi.
What's happening here is this:
- In a git checkout of the source tree there is no GDBvn.texi in the
source tree, the GDBvn.texi dependency will trigger a rebuild of
GDBvn.texi, which is then used to build gdb.dvi and/or gdb.pdf.
- In a release tar file we do include a copy of GDBvn.texi. This
file will appear to be up to date, and so no copy of GDBvn.texi is
created within the build directory. Now when building gdb.dvi
and/or gdb.pdf we copy (or symlink) the version of GDBvn.texi from
the source tree into the build directory.
However, copying GDBvn.texi from the source directory is completely
unnecessary. The gdb.dvi/gdb.pdf recipes both invoke texi2dvi and
pass '-I $(srcdir)' as an argument, this means that texi2dvi will look
in the $(srcdir) to find included files, including GDBvn.texi.
As such I believe we can remove the code that copies GDBvn.texi from
the source tree into the build tree.
I've tested with a release build; creating a release with:
./src-release gdb
Then in an empty directory, unpacking the resulting .tar file,
creating a parallel build directory and doing the usual configure,
make, and 'make install'.
Having done this I can then 'touch gdb/doc/*.texinfo' in the unpacked
source tree, and do 'make -C gdb/doc pdf dvi' to rebuild all the pdf
and dvi files, this works fine without having to either build or copy
GDBvn.texi into the build directory.
Approved-By: Tom Tromey <tom@tromey.com>
After the x86 target description changes that I committed recently,
the first commit in the series being:
commit 8a29222b85
Date: Sat Jan 27 10:40:35 2024 +0000
gdb/gdbserver: share I386_LINUX_XSAVE_XCR0_OFFSET definition
and the last commit in the series being:
commit 646d754d14
Author: Andrew Burgess <aburgess@redhat.com>
Date: Tue Jan 30 15:37:23 2024 +0000
gdb/gdbserver: share x86/linux tdesc caching
The sourceware buildbot highlighted a regression on i386. On the GDB
side we'd see this:
Remote debugging using :54321
warning: Architecture rejected target-supplied description
Remote connection closed
(gdb)
while on the gdbserver side we'd see this:
$ ./gdbserver/gdbserver --once :54321 ~/empty
Process /srv/aburgess/empty created; pid = 31406
Listening on port 54321
Remote debugging from host ::1, port 39488
../../src/gdbserver/regcache.cc:272: A problem internal to GDBserver has been detected.
Unknown register st0 requested
Aborted (core dumped)
When I tried to reproduce this regression on my local i386 VM the
issue would not reproduce.
I eventually tracked the problem down to x86_linux_tdesc_for_tid in
gdb/nat/x86-linux-tdesc.c. In this function we have this line:
/* Check if PTRACE_GETREGSET works. */
if (ptrace (PTRACE_GETREGSET, tid,
(unsigned int) NT_X86_XSTATE, &iov) < 0)
{
... handle failure ...
}
else
{
... handle success ...
}
The problem is that on my VM the PTRACE_GETREGSET feature is
supported, while on sourceware's buildbot machine this feature is not
supported.
I did a quick search and it seems like the 'xsave' feature in
/proc/cpuinfo might be the indicator for whether PTRACE_GETREGSET is
supported or not, and indeed my machine has the 'xsave' feature while
the sourceware machine does not.
The point of divergence then is this ptrace call, on my machine the
call succeeds and we extract the xcr0 value from the iov vector, while
on the sourceware machine the ptrace call fails and we use a default
xcr0 value of 0.
This xcr0 value is then passed to i386_linux_read_description at the
end of x86_linux_tdesc_for_tid.
In gdb/arch/i386-linux-tdesc.c we find i386_linux_read_description
which does some caching but calls i386_create_target_description to
actually create the target descriptions when needed. The xcr0 value
is masked to only the bits that are interesting, but given a value of
0 we'll just pass 0 through to i386_create_target_description.
In gdb/arch/i386.c we find i386_create_target_description which checks
the xcr0 bits and builds the target description. What we can see is
that if no bits are set in the xcr0 value then no features will be
added to the created target description. This featureless target
description is then transmitted back to GDB, which is then rejected
due to lack of essential core registers.
So, how did things work prior to the above commit series? There are
three places of interest, on the GDB side there is
x86_linux_nat_target::read_description and
i386_linux_core_read_description. Then on the gdbserver side there is
x86_linux_read_description.
All of these locations have a call to i386_linux_read_description
followed by a check if the return value was nullptr. If we do get
back nullptr then we perform another call to
i386_linux_read_description with a default xcr0 value.
Looking in i386_linux_read_description we see a specific check for
xcr0 being 0 in which case we return nullptr.
And so, prior to the above series, if xcr0 was 0 due to
PTRACE_GETREGSET being unavailable we'd use a default xcr0 value.
After the above series this is no longer the case, the 'xcr0 == 0'
check has been removed from i386_linux_read_description and the
calling code is streamlined to remove the use of default xcr0 values.
The fix I propose here is to setup the default xcr0 value at the point
where we find that PTRACE_GETREGSET is unavailable. The default value
used is X86_XSTATE_SSE_MASK. This is the default used in
x86_linux_nat_target::read_description (for GDB) and in
x86_linux_read_description (for gdbserver). The above commit series
already fixed i386_linux_core_read_description to ensure that the
correct default xcr0 value was used, this case is a little special in
that it uses different defaults depending on which sections are
present in the core file, so that case always needed to be handled
differently.
The choice of X86_XSTATE_SSE_MASK corresponds to the default used for
i386 before the above series was committed. This mask includes the
X87 and SSE bits only, neither of these bits are checked for on amd64
or x32, so this default doesn't change the behaviour on these targets.
By setting the default xcr0 value at this early stage we ensure that
the cached xcr0 value on the gdbserver side is correct. This is
critical as this cached xcr0 value is passed through to the in process
agent (IPA). If we leave the cached xcr0 value as 0 and apply the
defaults later in the series we also have to encode the knowledge of
the default into the IPA, this just means we have the default encoded
in multiple locations, which seems like a bad idea. The approach used
in this patch means the default is present in just one location.
This commit should fix the i386 regressions seen on the sourceware
buildbot.
In addition to the fix in nat/x86-linux-tdesc.c I've also fixed the
layout of the declaration of x86_linux_tdesc_for_tid in the header
file.
Approved-By: Felix Willgerodt <felix.willgerodt@intel.com>
- Add type annotations
- Use a raw string in one spot (where we call re.sub), to avoid an
"invalid escape sequence" warning.
- Remove unused "os" import.
Change-Id: I0149cbb73ad2b05431f032fa9d9530282cb01e90
Reviewed-By: Guinevere Larsen <blarsen@redhat.com>
On fedora rawhide, I ran into:
...
(gdb) continue^M
Continuing.^M
^M
Catchpoint 2 (call to syscall clone3), 0x000000000042097d in __clone3 ()^M
(gdb) FAIL: gdb.threads/stepi-over-clone.exp: continue
...
Fix this by updating a regexp to also recognize __clone3.
Tested on x86_64-linux.
Tested-By: Guinevere Larsen <blarsen@redhat.com>
When running test-case gdb.base/watchpoint-running on ppc64le-linux (and
similar on arm-linux), we get:
...
(gdb) watch global_var^M
warning: Error when detecting the debug register interface. \
Debug registers will be unavailable.^M
Watchpoint 2: global_var^M
(gdb) FAIL: $exp: all-stop: hardware: watch global_var
FAIL: $exp: all-stop: hardware: watchpoint hit (timeout)
...
The problem is that ppc_linux_dreg_interface::detect fails to detect the
hardware watchpoint interface, because the calls to ptrace return with errno
set to ESRCH.
This is a feature of ptrace: if a call is done while the tracee is not
ptrace-stopped, it returns ESRCH.
Indeed, in the test-case "watch global_var" is executed while the inferior is
running, and that triggers the first call to ppc_linux_dreg_interface::detect.
And because the detection failure is cached, subsequent attempts at setting
hardware watchpoints will also fail, even if the tracee is ptrace-stopped.
The way to fix this is to make sure that ppc_linux_dreg_interface::detect is
called when we know that the thread is ptrace-stopped, which in the current
setup is best addressed by using target-specific post_attach and
post_startup_inferior overrides. However, as we can see in
aarch64_linux_nat_target, that causes code duplication.
Fix this by:
- defining a new target hook low_init_process, called from
linux_init_ptrace_procfs, which is called from both
linux_nat_target::post_attach and linux_nat_target::post_startup_inferior,
- adding implementations for ppc_linux_nat_target and arm_linux_nat_target
that detect the hardware watchpoint interface,
- replacing the aarch64_linux_nat_target implementations of post_attach and
post_startup_inferior with a low_init_process implementation.
Tested on ppc64le-linux, arm-linux, aarch64-linux and x86_64-linux.
Co-Authored-By: Tom de Vries <tdevries@suse.de>
Approved-By: Luis Machado <luis.machado@arm.com>
PR tdep/31834
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=31834
PR tdep/31705
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=31705
When I changed the Rust parser to handle 128-bit ints, this
inadvertently broke some other gdb commands. For example, "info
symbol 0xffffffffffffffff" now fails, because the resulting value is
128 bits, but this is rejected by extract_integer.
This patch fixes the problem by changing extract_integer to allow
over-long integers as long as the high bytes are either 0, or (for
signed types) 0xff.
Regression tested on x86-64 Fedora 38.
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=31565
Approved-By: Andrew Burgess <aburgess@redhat.com>
When running test-case gdb.python/py-format-address.exp on arm-linux, I get:
...
(gdb) python print("Got: " + gdb.format_address(0x103dd))^M
Got: 0x103dd <main at py-format-address.c:30>^M
(gdb) FAIL: $exp: symbol_filename=on: gdb.format_address, \
result should have an offset
...
What is expected here is:
...
Got: 0x103dd <main+1 at py-format-address.c:30>^M
...
Main starts at main_addr:
...
(gdb) print /x &main^M
$1 = 0x103dc^M
...
and we obtained next_addr 0x103dd by adding 1 to it:
...
set next_addr [format 0x%x [expr $main_addr + 1]]
...
Adding 1 to $main_addr results in an address for a thumb function starting at
address 0x103dc, which is incorrect because main is an arm function (because
I'm running with target board unix/-marm).
At some point during the call to format_addr, arm_addr_bits_remove removes
the thumb bit, which causes the +1 offset to be dropped, causing the FAIL.
Fix this by using the address of the breakpoint on main, provided it's not at
the very start of main.
Tested on arm-linux.
PR testsuite/31452
Bug: https://www.sourceware.org/bugzilla/show_bug.cgi?id=31452
With test-case gdb.opt/inline-cmds.exp on ppc64le-linux, I ran into:
...
PASS: gdb.opt/inline-cmds.exp: finish from marker
...
PASS: gdb.opt/inline-cmds.exp: finish from marker
DUPLICATE: gdb.opt/inline-cmds.exp: finish from marker
...
Fix this by issuing less passes.
Tested on ppc64le-linux.
In commit 764af87825 ("[gdb/python] Add typesafe wrapper around
PyObject_CallMethod") I added poisoning of PyObject_CallMethod:
...
/* Poison PyObject_CallMethod. The typesafe wrapper gdbpy_call_method should be
used instead. */
template<typename... Args>
PyObject *
PyObject_CallMethod (Args...);
...
The idea was that subsequent code would be forced to use gdbpy_call_method
instead of PyObject_CallMethod.
However, that caused build issues with gcc 14 and python 3.13:
...
/usr/bin/ld: python/py-disasm.o: in function `gdb::ref_ptr<_object, gdbpy_ref_policy<_object> > gdbpy_call_method<unsigned int, long long>(_object*, char const*, unsigned int, long long)':
/data/vries/gdb/src/gdb/python/python-internal.h:207:(.text+0x384f): undefined reference to `_object* PyObject_CallMethod<_object*, char*, char*, unsigned int, long long>(_object*, char*, char*, unsigned int, long long)'
/usr/bin/ld: python/py-tui.o: in function `gdb::ref_ptr<_object, gdbpy_ref_policy<_object> > gdbpy_call_method<int>(_object*, char const*, int)':
/data/vries/gdb/src/gdb/python/python-internal.h:207:(.text+0x1235): undefined reference to `_object* PyObject_CallMethod<_object*, char*, char*, int>(_object*, char*, char*, int)'
/usr/bin/ld: python/py-tui.o: in function `gdb::ref_ptr<_object, gdbpy_ref_policy<_object> > gdbpy_call_method<int, int, int>(_object*, char const*, int, int, int)':
/data/vries/gdb/src/gdb/python/python-internal.h:207:(.text+0x12b0): undefined reference to `_object* PyObject_CallMethod<_object*, char*, char*, int, int, int>(_object*, char*, char*, int, int, int)'
collect2: error: ld returned 1 exit status
...
Fix this by poisoning without using templates.
Tested on x86_64-linux.
When running test-case gdb.base/complex-parts.exp on arm-linux, I get:
...
(gdb) p $_cimag (z3)^M
$6 = 6.5^M
(gdb) PASS: gdb.base/complex-parts.exp: long double imaginary: p $_cimag (z3)
ptype $^M
type = double^M
(gdb) FAIL: gdb.base/complex-parts.exp: long double imaginary: ptype $
...
Given that z3 is a complex long double, the test-case expects the type of the
imaginary part of z3 to be long double, but it's double instead.
This is due to the fact that the dwarf info doesn't specify an explicit target
type:
...
<5b> DW_AT_name : z3
<60> DW_AT_type : <0xa4>
...
<1><a4>: Abbrev Number: 2 (DW_TAG_base_type)
<a5> DW_AT_byte_size : 16
<a6> DW_AT_encoding : 3 (complex float)
<a7> DW_AT_name : complex long double
...
and consequently we're guessing in dwarf2_init_complex_target_type based on
the size:
...
case 64:
tt = builtin_type (gdbarch)->builtin_double;
break;
case 96: /* The x86-32 ABI specifies 96-bit long double. */
case 128:
tt = builtin_type (gdbarch)->builtin_long_double;
break;
...
For arm-linux, complex long double is 16 bytes, so the target type is assumed
to be 8 bytes, which is handled by the "case 64", which gets us double
instead of long double.
Fix this by searching for "long" in the name_hint parameter, and using long
double instead.
Note that base types in dwarf are not allowed to contain references to other
types, and the complex types are base types, so the missing explicit target
type is standard-conformant.
A gcc PR was filed to add this as a dwarf extension (
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115272 ).
Tested on arm-linux.
The following recent change introduced a regression when building using
clang++:
commit 764af87825
Date: Wed Jun 12 18:58:49 2024 +0200
[gdb/python] Add typesafe wrapper around PyObject_CallMethod
The error message is:
../../gdb/python/python-internal.h:151:16: error: default initialization of an object of const type 'const char'
constexpr char gdbpy_method_format;
^
= '\0'
CXX python/py-block.o
1 error generated.
make[2]: *** [Makefile:1959: python/py-arch.o] Error 1
make[2]: *** Waiting for unfinished jobs....
In file included from ../../gdb/python/py-auto-load.c:25:
../../gdb/python/python-internal.h:151:16: error: default initialization of an object of const type 'const char'
constexpr char gdbpy_method_format;
^
= '\0'
1 error generated.
make[2]: *** [Makefile:1959: python/py-auto-load.o] Error 1
In file included from ../../gdb/python/py-block.c:23:
../../gdb/python/python-internal.h:151:16: error: default initialization of an object of const type 'const char'
constexpr char gdbpy_method_format;
^
= '\0'
1 error generated.
This patch fixes this by changing gdbpy_method_format to be a templated
struct, and only have its specializations define the static constexpr
member "format". This way, we avoid having an uninitialized constexpr
expression, regardless of it being instantiated or not.
Reviewed-By: Tom de Vries <tdevries@suse.de>
Change-Id: I5bec241144f13500ef78daea30f00d01e373692d
On s390x-linux, I run into:
...
(gdb) p (short []) s1^M
$3 = {0, 1, 0, <optimized out>}^M
(gdb) FAIL: gdb.dwarf2/shortpiece.exp: p (short []) s1
...
while this is expected:
...
(gdb) p (short []) s1^M
$3 = {1, 0, 0, <optimized out>}^M
(gdb) PASS: gdb.dwarf2/shortpiece.exp: p (short []) s1
...
The type of s1 is:
...
(gdb) ptype s1
type = struct S {
myint a;
myushort b;
}
...
so the difference is due the fact that viewing an int as two shorts gives
different results depending on the endianness.
Fix this by allowing both results.
Tested on x86_64-linux and s390x-linux.
Approved-By: Tom Tromey <tom@tromey.com>
I noticed that we started using "string cat", which has been available since
tcl version 8.6.2.
Add a local implementation for use with older tcl versions.
Tested on x86_64-linux.
Approved-By: Andrew Burgess <aburgess@redhat.com>
In commit 1a7d840a21 ("[gdb/tdep] Fix ARM_LINUX_JB_PC_EABI"), in absense of
osabi settings for newlib and uclibc for arm, I chose a best-effort approach
using ifdefs.
Post-commit review [1] pointed out that this may be causing more problems than
it's worth.
Fix this by removing the ifdefs and simply defining ARM_LINUX_JB_PC_EABI to 1.
Rebuild on x86_64-linux with --enable-targets=all.
Fixes: 1a7d840a21 ("[gdb/tdep] Fix ARM_LINUX_JB_PC_EABI")
[1] https://sourceware.org/pipermail/gdb-patches/2024-June/209779.html
Commit 97033da507 ("[gdb/build] Cleanup gdb/features/feature_to_c.sh")
factored out new file gdb/features/feature_to_c.awk out of
gdb/features/feature_to_c.sh, but failed to add the GPL header comment, so add
this now.
Tested on x86_64-linux.
The xsave_ymm_avx512_offset data structure contains the xsave
offset to the upper 128 bits of a ymm register. Similarly, for zmm this
offset is described by xsave_avx512_zmm_h_offset, h indicating the
high bits. This commit renames the xsave_ymm_avx512_offset to
xsave_ymm_h_avx512_offset - as well as the associated define from
XSAVE_YMM_AVX512_ADDR to XSAVE_YMM_H_AVX512_ADDR - to make this
more consistent.
Note, that the regnum defines already included the 'h' for ymm, like
I387_YMM16H_REGNUM and I387_YMMH_AVX512_END_REGNUM.
Co-authored-by: Nils-Christian Kempke <nils-christian.kempke@intel.com>
Approved-By: Andrew Burgess <aburgess@redhat.com>
When running test-case gdb.fortran/array-indices.exp on a system without
fortran compiler, I run into a duplicate:
...
Running /home/vries/gdb/src/gdb/testsuite/gdb.fortran/array-indices.exp ...
gdb compile failed, default_target_compile: Can't find gfortran.
UNTESTED: gdb.fortran/array-indices.exp: array-indices.exp
gdb compile failed, default_target_compile: Can't find gfortran.
UNTESTED: gdb.fortran/array-indices.exp: array-indices.exp
DUPLICATE: gdb.fortran/array-indices.exp: array-indices.exp
...
Fix this by adding a with_test_prefix at the toplevel.
Likewise in gdb.fortran/array-repeat.exp.
Tested on x86_64-linux.
Reviewed-By: Alexandra Petlanova Hajkova <ahajkova@redhat.com>
Clean up script gdb/features/feature_to_c.sh by:
- fixing shellcheck warnings,
- moving an embedded awk script out of the file, reducing the amount of
escaping in the awk script, making it more readable and maintainable, and
- adding emacs / vi settings for local tab size 2 (copied from ./ltmain.sh).
Tested on x86_64-linux.
Approved-by: Kevin Buettner <kevinb@redhat.com>
In emacs, on gdb/contrib/cc-with-tweaks.sh, do:
- M-x whitespace-cleanup,
- M-x mark-whole-buffer and M-x indent-region, and
- and undo the unwanted changes in the header comment.
Only whitespace changes.
Tested on x86_64-linux.
Approved-by: Kevin Buettner <kevinb@redhat.com>
Clean up script gdb/contrib/expect-read1.sh by:
- fixing shellcheck warnings,
- using mktemp (which takes TMPDIR into account) instead of a hardcoded
"/tmp/expect-read1.$$.so",
- adding comments, and
- adding emacs / vi settings for local tab size 2 (copied from ./ltmain.sh).
Tested on x86_64-linux.
Approved-by: Kevin Buettner <kevinb@redhat.com>
This introduces the new method language_defn::lookup_symbol_local, and
then changes lookup_symbol_local to use it. This removes an explicit
language check from this function, and makes it easier for other
languages to hook into this code.
This simplifies lookup_local_symbol a little, by having it check
whether the current block is the static or global block, instead of
first searching for the static block.
This changes lookup_local_symbol to directly examine any attached
template symbols, rather than gating this lookup on the use of C++ or
Fortran. As mentioned in an earlier patch, these objects are not
necessarily C++-specific, and doing the search generically seems
better.
This also renames cp_lookup_symbol_imports_or_template now that the
"template" part has been removed.
This patch renames is_cplus_template_function to is_template_function.
There is nothing C++-specific about this code, and the code in the
DWARF reader that creates these objects is not C++-specific. In fact
this may already be used by Rust (though I didn't check).
Since this commit:
commit 0ee6b1c511
Date: Wed May 18 13:32:04 2022 -0700
Use aarch64_features to describe register features in target descriptions.
There has been an issue with how aarch64 target descriptions are
cached within gdbserver, and specifically, how this caching impacts
the in process agent (IPA).
The function initialize_tracepoint_ftlib (gdbserver/tracepoint.cc) is
part of the IPA, this function is a constructor function, i.e. is
called as part of the global initialisation process. We can't
guarantee the ordering of when this function is called vs when other
global state is initialised.
Now initialize_tracepoint_ftlib calls initialize_tracepoint, which
calls initialize_low_tracepoint, which for aarch64 calls
aarch64_linux_read_description.
The aarch64_linux_read_description function lives in
linux-aarch64-tdesc.cc and after the above commit, depends on a
std::unordered_map having been initialized.
Prior to the above commit aarch64_linux_read_description used a global
C style array, which obviously requires no runtime initialization.
The consequence of the above is that any inferior linked with the IPA
(for aarch64) will experience undefined behaviour (access to an
uninitialized std::unordered_map) during startup, which for me
manifests as a segfault.
I propose fixing this by moving the std::unordered_map into the
function body, but leaving it static. The map will now be initialized
the first time the function is called, which removes the undefiend
behaviour.
The same problem exists for the expedited_registers global, however
this global can just be made into a function local instead. The
expedited_registers variable is used to build a pointer list which is
then passed to init_target_desc, however init_target_desc copies the
values it is given so expedited_registers does not need to live longer
than its containing function.
On most of the AArch64 machines I have access too tracing is not
supported, and so the gdb.trace/*.exp tests that use the IPA just exit
early reporting unsupported. I've added a test which links an
inferior with the IPA and just starts the inferior. No tracing is
performed. This exposes the current issue even on hosts that don't
support tracing. After this patch the test passes.
This commit builds on the previous series of commits to share the
target description caching code between GDB and gdbserver for
x86/Linux targets.
The objective of this commit is to move the four functions (2 each of)
i386_linux_read_description and amd64_linux_read_description into the
gdb/arch/ directory and combine them so we have just a single copy of
each. Then GDB, gdbserver, and the in-process-agent (IPA) will link
against these shared functions.
One curiosity with this patch is the function
x86_linux_post_init_tdesc. On the gdbserver side the two functions
amd64_linux_read_description and i386_linux_read_description have some
functionality that is not present on the GDB side, there is some
additional configuration that is performed as each target description
is created, to setup the expedited registers.
To support this I've added the function x86_linux_post_init_tdesc.
This function is called from the two *_linux_read_description
functions, but is implemented separately for GDB and gdbserver.
An alternative approach that avoids adding x86_linux_post_init_tdesc
would be to have x86_linux_tdesc_for_tid return a non-const target
description, then in x86_target::low_arch_setup we could inspect the
target description to figure out if it is 64-bit or not, and modify
the target description as needed. In the end I think that adding the
x86_linux_post_init_tdesc function is the simpler solution.
The contents of gdbserver/linux-x86-low.cc have moved to
gdb/arch/x86-linux-tdesc-features.c, and gdbserver/linux-x86-tdesc.h
has moved to gdb/arch/x86-linux-tdesc-features.h, this change leads to
some updates in the #includes in the gdbserver/ directory.
This commit also changes how target descriptions are cached.
Previously both GDB and gdbserver used static C-style arrays to act as
the tdesc cache. This was fine, except for two problems. Either the
C-style arrays would need to be placed in x86-linux-tdesc-features.c,
which would allow us to use the x86_linux_*_tdesc_count_1() functions
to size the arrays for us, or we'd need to hard code the array sizes
using separate #defines, which we'd then have to keep in sync with the
rest of the code in x86-linux-tdesc-features.c.
Given both of these problems I decided a better solution would be to
just switch to using a std::unordered_map to act as the cache. This
will resize automatically, and we can use the xcr0 value as the key.
At first inspection, using xcr0 might seem to be a problem; after all
the {i386,amd64}_create_target_description functions take more than
just the xcr0 value. However, this patch is only for x86/Linux
targets, and for x86/Linux all of the other flags passed to the tdesc
creation functions have constant values and so are irrelevant when we
consider tdesc caching.
For testing I've done the following:
- Built on x86-64 GNU/Linux for all targets, and just for the native
target,
- Build on i386 GNU/Linux for all targets, and just for the native
target,
- Build on a 64-bit, non-x86 GNU/Linux for all targets, just for the
native target, and for targets x86_64-*-linux and i386-*-linux.
Approved-By: Felix Willgerodt <felix.willgerodt@intel.com>
This commit is part of a series to share more of the x86 target
description creation code between GDB and gdbserver.
Unlike previous commits which were mostly refactoring, this commit is
the first that makes a real change, though that change should mostly
be for gdbserver; I've largely adopted the "GDB" way of doing things
for gdbserver, and this fixes a real gdbserver bug.
On a x86-64 Linux target, running the test:
gdb.server/connect-with-no-symbol-file.exp
results in two core files being created. Both of these core files are
from the inferior process, created after gdbserver has detached.
In this test a gdbserver process is started and then, after gdbserver
has started, but before GDB attaches, we either delete the inferior
executable, or change its permissions so it can't be read. Only after
doing this do we attempt to connect with GDB.
As GDB connects to gdbserver, gdbserver attempts to figure out the
target description so that it can send the description to GDB, this
involves a call to x86_linux_read_description.
In x86_linux_read_description one of the first things we do is try to
figure out if the process is 32-bit or 64-bit. To do this we look up
the executable via the thread-id, and then attempt to read the
architecture size from the executable. This isn't going to work if
the executable has been deleted, or is no longer readable.
And so, as we can't read the executable, we default to an i386 target
and use an i386 target description.
A consequence of using an i386 target description is that addresses
are assumed to be 32-bits. Here's an example session that shows the
problems this causes. This is run on an x86-64 machine, and the test
binary (xx.x) is a standard 64-bit x86-64 binary:
shell_1$ gdbserver --once localhost :54321 /tmp/xx.x
shell_2$ gdb -q
(gdb) set sysroot
(gdb) shell chmod 000 /tmp/xx.x
(gdb) target remote :54321
Remote debugging using :54321
warning: /tmp/xx.x: Permission denied.
0xf7fd3110 in ?? ()
(gdb) show architecture
The target architecture is set to "auto" (currently "i386").
(gdb) p/x $pc
$1 = 0xf7fd3110
(gdb) info proc mappings
process 2412639
Mapped address spaces:
Start Addr End Addr Size Offset Perms objfile
0x400000 0x401000 0x1000 0x0 r--p /tmp/xx.x
0x401000 0x402000 0x1000 0x1000 r-xp /tmp/xx.x
0x402000 0x403000 0x1000 0x2000 r--p /tmp/xx.x
0x403000 0x405000 0x2000 0x2000 rw-p /tmp/xx.x
0xf7fcb000 0xf7fcf000 0x4000 0x0 r--p [vvar]
0xf7fcf000 0xf7fd1000 0x2000 0x0 r-xp [vdso]
0xf7fd1000 0xf7fd3000 0x2000 0x0 r--p /usr/lib64/ld-2.30.so
0xf7fd3000 0xf7ff3000 0x20000 0x2000 r-xp /usr/lib64/ld-2.30.so
0xf7ff3000 0xf7ffb000 0x8000 0x22000 r--p /usr/lib64/ld-2.30.so
0xf7ffc000 0xf7ffe000 0x2000 0x2a000 rw-p /usr/lib64/ld-2.30.so
0xf7ffe000 0xf7fff000 0x1000 0x0 rw-p
0xfffda000 0xfffff000 0x25000 0x0 rw-p [stack]
0xff600000 0xff601000 0x1000 0x0 r-xp [vsyscall]
(gdb) info inferiors
Num Description Connection Executable
* 1 process 2412639 1 (remote :54321)
(gdb) shell cat /proc/2412639/maps
00400000-00401000 r--p 00000000 fd:03 45907133 /tmp/xx.x
00401000-00402000 r-xp 00001000 fd:03 45907133 /tmp/xx.x
00402000-00403000 r--p 00002000 fd:03 45907133 /tmp/xx.x
00403000-00405000 rw-p 00002000 fd:03 45907133 /tmp/xx.x
7ffff7fcb000-7ffff7fcf000 r--p 00000000 00:00 0 [vvar]
7ffff7fcf000-7ffff7fd1000 r-xp 00000000 00:00 0 [vdso]
7ffff7fd1000-7ffff7fd3000 r--p 00000000 fd:00 143904 /usr/lib64/ld-2.30.so
7ffff7fd3000-7ffff7ff3000 r-xp 00002000 fd:00 143904 /usr/lib64/ld-2.30.so
7ffff7ff3000-7ffff7ffb000 r--p 00022000 fd:00 143904 /usr/lib64/ld-2.30.so
7ffff7ffc000-7ffff7ffe000 rw-p 0002a000 fd:00 143904 /usr/lib64/ld-2.30.so
7ffff7ffe000-7ffff7fff000 rw-p 00000000 00:00 0
7ffffffda000-7ffffffff000 rw-p 00000000 00:00 0 [stack]
ffffffffff600000-ffffffffff601000 r-xp 00000000 00:00 0 [vsyscall]
(gdb)
Notice the difference between the mappings reported via GDB and those
reported directly from the kernel via /proc/PID/maps, the addresses of
every mapping is clamped to 32-bits for GDB, while the kernel reports
real 64-bit addresses.
Notice also that the $pc value is a 32-bit value. It appears to be
within one of the mappings reported by GDB, but is outside any of the
mappings reported from the kernel.
And this is where the problem arises. When gdbserver detaches from
the inferior we pass the inferior the address from which it should
resume. Due to the 32/64 bit confusion we tell the inferior to resume
from the 32-bit $pc value, which is not within any valid mapping, and
so, as soon as the inferior resumes, it segfaults.
If we look at how GDB (not gdbserver) figures out its target
description then we see an interesting difference. GDB doesn't try to
read the executable. Instead GDB uses ptrace to query the thread's
state, and uses this to figure out the if the thread is 32 or 64 bit.
If we update gdbserver to do it the "GDB" way then the above problem
is resolved, gdbserver now sees the process as 64-bit, and when we
detach from the inferior we give it the correct 64-bit address, and
the inferior no longer segfaults.
Now, I could just update the gdbserver code, but better, I think, to
share one copy of the code between GDB and gdbserver in gdb/nat/.
That is what this commit does.
The cores of x86_linux_read_description from gdbserver and
x86_linux_nat_target::read_description from GDB are moved into a new
file gdb/nat/x86-linux-tdesc.c and combined into a single function
x86_linux_tdesc_for_tid which is called from each location.
This new function does things mostly the GDB way, some changes are
needed to allow for the sharing; we now take some pointers for where
the shared code can cache the xcr0 and xsave layout values.
Another thing to note about this commit is how the functions
i386_linux_read_description and amd64_linux_read_description are
handled. For now I've left these function as implemented separately
in GDB and gdbserver. I've moved the declarations of these functions
into gdb/arch/{i386,amd64}-linux-tdesc.h, but the implementations are
left where they are.
A later commit in this series will make these functions shared too,
but doing this is not trivial, so I've left that for a separate
commit. Merging the declarations as I've done here ensures that
everyone implements the function to the same API, and once these
functions are shared (in a later commit) we'll want a shared
declaration anyway.
Reviewed-By: Felix Willgerodt <felix.willgerodt@intel.com>
Acked-By: John Baldwin <jhb@FreeBSD.org>
Currently, in i386_linux_core_read_description, if GDB fails to
extract an xcr0 value from the core file, then we will have a default
zero value for the xcr0 variable, we still call the
i386_linux_read_description function, which checks for this zero value
and returns nullptr.
Back in i386_linux_core_read_description we spot the nullptr return
value from i386_linux_read_description and call
i386_linux_read_description again, but this time passing a default
value for xcr0.
In the next commit I plan to rework i386_linux_read_description, and
in so doing I will remove the check for xcr0 == 0, this is inline with
how the amd64 code is written.
However, this means that the 'xcr0 == 0' check needs to move up the
stack to i386_linux_core_read_description, again, this brings the i386
code into line with the amd64 code.
This is just a refactor in preparation for the next commit, there
should be no user visible changes after this commit.
Approved-By: Felix Willgerodt <felix.willgerodt@intel.com>
Approved-By: John Baldwin <jhb@FreeBSD.org>
This patch is part of a series that has the aim sharing the x86 Linux
target description creation code between GDB and gdbserver.
Within GDB part of this process involves reading the cs and ds state
from the 'struct user_regs_struct' using a ptrace call.
This isn't done by gdbserver, which is part of the motivation for this
whole series; the approach gdbserver takes is inferior to the approach
GDB takes (gdbserver relies on reading the file being debugged, and
extracting similar information from the file headers).
This commit moves the reading of cs and ds, which is used to figure
out if a thread is 32-bit or 64-bit (or in x32 mode), into the gdb/nat
directory so that the code can be shared with gdbserver, but at this
point I'm not actually using the code in gdbserver, that will come
later.
As such there should be no user visible changes after this commit, GDB
continues to do things as it did before (reading cs/ds), while
gdbserver continues to use its own approach (which doesn't require
reading cs/ds).
Approved-By: John Baldwin <jhb@FreeBSD.org>
Reviewed-By: Felix Willgerodt <felix.willgerodt@intel.com>
In a later commit I want to access have_ptrace_getregset from a .c
file in the nat/ directory. To achieve this I need access to the
declaration of have_ptrace_getregset.
Currently have_ptrace_getregset is declared (and defined) twice, once
in GDB and once in gdbserver.
This commit moves the declaration into nat/linux-nat.h, but leaves the
two definitions where they are. Now, in my later commit, I can pull
in the declaration from nat/linux-nat.h.
There should be no user visible changes after this commit.
Approved-By: Felix Willgerodt <felix.willgerodt@intel.com>
The have_ptrace_getfpxregs global tracks whether GDB or gdbserver is
running on a kernel that supports the GETFPXREGS ptrace request.
Currently this global is declared twice (once in GDB and once in
gdbserver), I think it makes sense to move this global into the nat/
directory, and have a single declaration and definition.
While moving this variable I have converted it to a tribool, as that
was what it really was, if even used the same numbering as the tribool
enum (-1, 0, 1). Where have_ptrace_getfpxregs was used I have updated
in the obvious way.
However, while making this change I noticed what I think is a bug in
x86_linux_nat_target::read_description and x86_linux_read_description,
both of these functions can be called multiple times, but in both
cases we only end up calling i386_linux_read_description the first
time through in the event that PTRACE_GETFPXREGS is not supported.
This is because initially have_ptrace_getfpxregs will be
TRIBOOL_UNKNOWN, but after the ptrace call fails we set
have_ptrace_getfpxregs to TRIBOOL_FALSE. The next time we attempt to
read the target description we'll skip the ptrace call, and so skip
the call to i386_linux_read_description.
I've not tried to address this preexisting bug in this commit, this is
purely a refactor, there should be no user visible changes after this
commit. In later commits I'll merge the gdbserver and GDB code
together into the nat/ directory, and after that I'll try to address
this bug.
Reviewed-By: Felix Willgerodt <felix.willgerodt@intel.com>
Share the definition of I386_LINUX_XSAVE_XCR0_OFFSET between GDB and
gdbserver.
This commit moves the definition into gdbsupport/x86-xstate.h, which
allows the #define to be shared.
There should be no user visible changes after this commit.
Approved-By: Felix Willgerodt <felix.willgerodt@intel.com>
This changes gdbpy_call_method to return a gdbpy_ref<>. This is
slightly safer because it makes it simpler to correctly handle
reference counts.
Reviewed-By: Tom de Vries <tdevries@suse.de>
