Not all targets support scheduler-locking. Add a check to see if the
taraget supports scheduler locking, and if it doesn't, don't run the
scheduler-locking tests that will otherwise fail.
There are actually a set of tests that try to use scheduler-locking
however, in most of these cases the test will not be run on smaller
targets (those that might not support threads and scheduler-locking)
due to the targets lack of support for threads, or some other larger
feature.
In the gdb.mi/mi-cmd-param-changed.exp test though, there's no
dependence on threads, or any other larger feature, and so, for the
small target I was using the test would otherwise try to run, only to
fail due to lack of support for scheduler-locking.
gdb/testsuite/ChangeLog:
* lib/gdb.exp (target_supports_scheduler_locking): New proc.
* gdb.mi/mi-cmd-param-changed.exp: Only run scheduler locking
tests if the target supports scheduler locking.
This fails for me on openSUSE leap 15.0:
...
FAIL: gdb.python/py-rbreak.exp: check number of returned breakpoints is 11
...
The rbreak "" command expects 11 breaks, but I see two extra for
__libc_csu_fini and __libc_csu_init:
...
Breakpoint 13 at 0x4005b0: file elf-init.c, line 106.^M
Breakpoint 14 at 0x400540: file elf-init.c, line 68.^M
...
This patch fixes the failing test by excluding functions starting with an
underscore.
Tested on x86_64-linux.
2018-08-04 Tom de Vries <tdevries@suse.de>
* gdb.python/py-rbreak.exp: Fix rbreak regexp.
PR symtab/16842 shows that gdb will crash when the user tries to
invoke "info address" of a template parameter.
The bug here is that dwarf2read.c does not set the symtab on the
template parameter symbols. This is pedantically correct, given that
the template symbols do not appear in a symtab. However, gdb
primarily uses the symtab backlink to find the symbol's objfile. So,
this patch simply sets the symtab on these symbols.
Tested by the buildbot.
gdb/ChangeLog
2018-08-02 Tom Tromey <tom@tromey.com>
PR symtab/16842.
* dwarf2read.c (read_func_scope): Set symtab on template parameter
symbols.
(process_structure_scope): Likewise.
gdb/testsuite/ChangeLog
2018-08-02 Tom Tromey <tom@tromey.com>
PR symtab/16842.
* gdb.cp/temargs.exp: Test "info address" of a template
parameter.
Invoking -var-info-path-expression on a dynamic varobj lead either in wrong
(nonsense) result or to a segmentation fault in cplus_describe_child().
This was caused by the fact that varobj_get_path_expr() called
cplus_path_expr_of_child() ignoring the fact the parent of the variable
is dynamic. Then, cplus_describe_child() accessed the underlaying C type
members by index, causing (i) either wrong (nonsense) expression being
returned (since dynamic child may be completely arbibtrary value)
or (ii) segmentation fault (in case the index higher than number of
underlaying C type members.
This fixes the problem by checking whether a varobj is a child of a dynamic
varobj and, if so, reporting an error as described in documentation.
gdb/ChangeLog:
* varobj.c (varobj_get_path_expr_parent): Report an error if
parent is a dynamic varobj.
gdb/testsuite/Changelog:
* gdb.python/py-mi-var-info-path-expression.c: New file.
* gdb.python/py-mi-var-info-path-expression.py: New file.
* gdb.python/py-mi-var-info-path-expression.exp: New file.
On commit:
commit 7f1f7e23939adc7d71036a17fc6081e3af7ca585
Author: Sergio Durigan Junior <sergiodj@redhat.com>
Date: Fri Jul 13 16:20:34 2018 -0400
Expect for another variant of error message when gdbserver cannot resolve hostname
I extended the regular expression being used to identify whether
gdbserver could not resolve a (host)name. This was needed because the
error message being printed had a different variation across some
systems. However, as it turns out, I've just noticed that the message
has yet another variation:
target remote tcp8:123:2353
tcp8:123:2353: cannot resolve name: System error
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
tcp8:123:2353: No such file or directory.
(gdb) FAIL: gdb.server/server-connect.exp: tcp8: connect to gdbserver using tcp8:123
which is causing FAILs on some systems (namely, Fedora-i686 on
BuildBot).
So instead of trying to predict everything that can be printed, I
decided to just match anything after the "cannot resolve name: " part.
This patch implements that.
Regression tested on the BuildBot.
gdb/testsuite/ChangeLog:
2018-07-30 Sergio Durigan Junior <sergiodj@redhat.com>
* lib/gdbserver-support.exp (gdbserver_start): Match any kind of
error after "cannot resolve name" string.
The gdb_load_shlib function will, on remote targets, try to run some
GDB commands. This obviously isn't going to work unless GDB is
running.
The gdb.trace/tspeed.exp test calls gdb_load_shlib before starting
GDB. Don't do that.
The failure that's triggered is actually DeJaGNU complaining that the
variable $use_gdb_stub doesn't exist, this is only created when GDB is
started. Something like this should trigger a failure:
make check-gdb \
RUNTESTFLAGS="--target_board=remote-gdbserver-on-localhost \
gdb.trace/tspeed.exp"
This commit also adds a check to gdb_load_shlib that GDB is running.
The check is always performed, so this should catch cases where a GDB
developer adds a use of gdb_load_shlib but doesn't test their code
with a remote target.
gdb/testsuite/ChangeLog:
* gdb.trace/tspeed.exp: Only call gdb_load_shlib after gdb has
started.
* lib/gdb.exp (gdb_load_shlib): Call perror if GDB is not running.
At -O3 -g -gstrict-dwarf, gcc generates for an optimized out vla 'a' a
DW_TAG_variable with type DW_TAG_array_type containing one
DW_TAG_subrange_type, but without DW_AT_upper_bound or DW_AT_count, which
makes the upper bound value 'unknown':
...
.uleb128 0x15 # (DIE (0x161) DW_TAG_variable)
.long 0xec # DW_AT_abstract_origin
.long 0x170 # DW_AT_type
...
.uleb128 0xa # (DIE (0x170) DW_TAG_array_type)
.long 0x110 # DW_AT_type
.long 0x17f # DW_AT_sibling
.uleb128 0x17 # (DIE (0x179) DW_TAG_subrange_type)
.long 0xc6 # DW_AT_type
.byte 0 # end of children of DIE 0x170
...
But gdb prints '0' for the size of 'a':
...
/gdb ./vla-1.exe -batch -ex "b f1" -ex "run" -ex "p sizeof(a)"
Breakpoint 1 at 0x4004c0: f1. (2 locations)
Breakpoint 1, f1 (i=<optimized out>) at vla-1.c:18
18 }
$1 = 0
...
while <optimized out> would be more appropriate.
This patch fixes that in evaluate_subexp_for_sizeof.
Build and reg-tested on x86_64-linux.
2018-07-28 Tom de Vries <tdevries@suse.de>
* eval.c (evaluate_subexp_for_sizeof): Interpret size of dynamic type
with undefined upper bound as <optimized out>.
* gdb.base/vla-optimized-out-o3-strict.exp: New file.
Add a maintenance command to disable the DWARF stack unwinders.
Normal users would not need this feature, but it is useful to allow
extended testing of fallback stack unwinding strategies, for example,
prologue scanners.
This is a partial implementation of the idea discussed in pr gdb/8434,
which talks about a generic ability to disable any frame unwinder.
Being able to arbitrarily disable any frame unwinder would be a more
complex patch, and I was unsure how useful such a feature would really
be, however, I can see (and have) a real need to disable DWARF
unwinders. That's why this patch only targets that specific set of
unwinders.
If in the future we find ourselves adding more switches to disable
different unwinders, then we should probably move to a more generic
solution, and remove this patch.
gdb/ChangeLog:
* dwarf2-frame-tailcall.c (tailcall_frame_sniffer): Exit early if
DWARF unwinders are disabled.
* dwarf2-frame.c: Add dwarf2read.h include.
(dwarf2_frame_sniffer): Exit early if DWARF unwinders are
disabled.
(dwarf2_frame_unwinders_enabled_p): Define.
(show_dwarf_unwinders_enabled_p): New function.
(_initialize_dwarf2_frame): Register switch to control DWARF
unwinder use.
* dwarf2-frame.h (dwarf2_frame_unwinders_enabled_p): Declare.
* dwarf2read.c (set_dwarf_cmdlist): Remove static keyword.
(show_dwarf_cmdlist): Remove static keyword.
* dwarf2read.h (set_dwarf_cmdlist): Declare.
(show_dwarf_cmdlist): Declare.
* NEWS: Document new feature.
gdb/doc/ChangeLog:
* gdb.texinfo (Maintenance Commands): Add description of
maintenance command to control dwarf unwinders.
gdb/testsuite/ChangeLog:
* gdb.base/maint.exp: Add check that dwarf unwinders control flag
is visible.
With the test-case contained in this patch and compiled for debug we run into
a segfault with trunk gdb:
...
$ gdb catch-follow-exec -batch -ex "catch exec" \
-ex "set follow-exec-mode new" -ex "run" -ex "info prog"
Catchpoint 1 (exec)
process xxx is executing new program: /usr/bin/ls
[New inferior 2 (process 0)]
[New process xxx]
Thread 2.1 "ls" hit Catchpoint 1 (exec'd /usr/bin/ls), in _start () from
/lib64/ld-linux-x86-64.so.2
Segmentation fault (core dumped)
...
The patch fixes the segfault by returning an error in info_program_command
if get_last_target_status returns minus_one_ptid.
The test-case is non-standard, because the standard approach runs into
PR23368, a problem with gdb going to the background.
Build and reg-tested on x86_64-linux.
2018-07-26 Tom de Vries <tdevries@suse.de>
PR breakpoints/23366
* infcmd.c (info_program_command): Handle ptid == minus_one_ptid.
* gdb.base/catch-follow-exec.c: New test.
* gdb.base/catch-follow-exec.exp: New file.
When compiling vla-optimized-out.c with -O3 and a recent gcc, and trying to
print the vla a in f1, we run into this gdb exception:
...
Cannot find matching parameter at DW_TAG_call_site 0x4003be at main
...
This is a regression introduced by 42dc7699a2 "[gdb/exp] Fix printing of type
of optimized out vla".
This patch fixes the regression by wrapping the ctx.eval call in
dwarf2_locexpr_baton_eval in try/catch, similar to what is done in
dwarf2_evaluate_loc_desc_full.
Build and reg-tested on x86_64-linux.
2018-07-25 Tom de Vries <tdevries@suse.de>
* dwarf2loc.c (dwarf2_locexpr_baton_eval): Wrap ctx.eval call in
try/catch.
* gdb.base/vla-optimized-out-o3.exp: New file. Reuse
vla-optimized-out.c.
When a single breakpoint location enableness was modified by a CLI
command, observers were not notified about it. This issue is now fixed.
gdb/ChangeLog:
* breakpoint.c (enable_disable_bp_num_loc): Notify observers.
gdb/testsuite/ChangeLog:
* gdb.mi/mi-breakpoint-location-ena-dis.cc: New file.
* gdb.mi/mi-breakpoint-location-ena-dis.exp: New file.
Make the noclone attribute on f1 in vla-optimized-out.c conditional. This
makes the test-case by default identical to
gcc/testsuite/gcc.dg/guality/vla-1.c.
Tested on x86_64-linux.
2018-07-20 Tom de Vries <tdevries@suse.de>
* gdb.base/vla-optimized-out.c: Make noclone attribute conditional on
NOCLONE macro.
* gdb.base/vla-optimized-out.exp: Use additional_flags -DNOCLONE.
I ran into a gdb.mi/list-thread-groups-available.exp failure:
...
Running gdb/testsuite/gdb.mi/list-thread-groups-available.exp ...
FAIL: gdb.mi/list-thread-groups-available.exp:
list available thread groups (unexpected output)
PASS: gdb.mi/list-thread-groups-available.exp:
list available thread groups with filter
...
When doing an experiment of running it 100 times in a row, the failure
reproduced 3 times.
Analyzing the original failure led to insufficient quoting of square brackets
in a regexp. This patch fixes the regexp, which resulted in 0 failures in a
100-in-a-row run.
Tested on x86_64-linux.
2018-07-19 Tom de Vries <tdevries@suse.de>
* gdb.mi/list-thread-groups-available.exp (cores_re): Fix quoting in
regular expression.
Consider this snippet from gcc/testsuite/gcc.dg/guality/vla-1.c:
...
int __attribute__((noinline))
f1 (int i)
{
char a[i + 1];
a[0] = 5; /* { dg-final { gdb-test .+1 "i" "5" } } */
return a[0]; /* { dg-final { gdb-test . "sizeof (a)" "6" } } */
}
...
When we compile the test-case with -O1 -g, and query the size of optimized
out vla 'a', we get:
...
$ ./gdb -batch -ex "b f1" -ex "r" -ex "p sizeof (a)" vla-1.exe
Breakpoint 1 at 0x4004a8: file vla-1.c, line 17.
Breakpoint 1, f1 (i=i@entry=5) at vla-1.c:17
17 return a[0];
$1 = 0
...
while we expect a size of '6'.
The problem is that default_read_var_value does not resolve the dynamic type
of a variable if the variable is optimized out.
This patch fixes that, and consequently gdb prints '6', as expected.
Tested on x86_64-linux.
2018-07-18 Tom de Vries <tdevries@suse.de>
* findvar.c (default_read_var_value): Also resolve dynamic type for
LOC_OPTIMIZED_OUT vars.
* gdb.base/vla-optimized-out.c: New test.
* gdb.base/vla-optimized-out.exp: New file.
I've noticed that on a few hosts, when given an invalid hostname,
gdbserver fails with:
spawn /../../gdb/gdbserver/gdbserver --once tcp8:123:2353 /gdb/build/fedora-s390x/build/gdb/testsuite/outputs/gdb.server/server-connect/server-connect
tcp8:123:2353: cannot resolve name: No address associated with hostname
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Exiting
Unfortunately, this causes a fail on the new
gdb.server/server-connect.exp test (introduced by the IPv6 patch):
FAIL: gdb.server/server-connect.exp: tcp8: start gdbserver: gdbserver should fail but did not
This happens because we're expecting for another variant of this error
message:
cannot resolve name: Name or service not known
Therefore, this patch extends the helper function 'gdbserver_start' to
also recognize the "No address associated with hostname" message.
This "fixes" the testcase on the hosts that use this variant.
gdb/testsuite/ChangeLog:
2018-07-13 Sergio Durigan Junior <sergiodj@redhat.com>
* lib/gdbserver-support.exp (gdbserver_start): Expect for the
message "No address associated with hostname" when gdbserver
cannot resolve the hostname.
Also, add prefixes to make some non unique tests unique.
gdb/testsuite/ChangeLog
2018-07-12 Philippe Waroquiers <philippe.waroquiers@skynet.be>
* gdb.threads/pthreads.exp: Test qcs FLAG arguments.
Add some test prefixes to make tests unique.
gdb/testsuite/ChangeLog
2018-07-12 Philippe Waroquiers <philippe.waroquiers@skynet.be>
* gdb.base/frameapply.c: New file.
* gdb.base/frameapply.exp: New file.
Add helper functions parse_flags and parse_flags_qcs.
parse_flags helper function allows to look for a set of flags at
the start of a string.
A flag must be given individually.
parse_flags_qcs is a specialised helper function to handle
the flags -q, -c and -s, that are used in the new command 'frame apply'
and in the command 'thread apply.
Modify number_or_range_parser::get_number to differentiate a
- followed by digits from a - followed by an alpha (i.e. a flag or an option).
That is needed for the addition of the [FLAG]... arguments to
thread apply ID... [FLAG]... COMMAND
Remove bool number_or_range_parser::m_finished, rather
implement the 'finished' logic inside number_or_range_parser::finished.
The new logic properly detects the end of parsing even if not at
end of the string. This ensures that number_or_range_parser::cur_tok
really points past the last parsed token when parsing is finished.
Before, it was always pointing at the end of the string.
As parsing now is finished directly when not positioned on a number,
number_is_in_list must do an error check before the loop getting all
numbers.
The error message for 'thread apply -$unknownconvvar p 1'
is now the more clear:
Convenience variable must have integer value.
Invalid thread ID: -$unknownconvvar p 1
instead of previously:
negative value
gdb/ChangeLog
2018-07-12 Philippe Waroquiers <philippe.waroquiers@skynet.be>
* cli-utils.c (number_or_range_parser::get_number): Only handle
numbers or convenience var as numbers.
(parse_flags): New function.
(parse_flags_qcs): New function.
(number_or_range_parser::finished): Ensure parsing end is detected
before end of string.
* cli-utils.h (parse_flags): New function.
(parse_flags_qcs): New function.
(number_or_range_parser): Remove m_finished bool.
(number_or_range_parser::skip_range): Set m_in_range to false.
gdb/testsuite/ChangeLog
2018-07-12 Philippe Waroquiers <philippe.waroquiers@skynet.be>
* gdb.base/skip.exp: Update expected error message.
This patch implements IPv6 support for both GDB and gdbserver. Based
on my research, it is the fourth attempt to do that since 2006. Since
I used ideas from all of the previous patches, I also added their
authors's names on the ChangeLogs as a way to recognize their
efforts. For reference sake, you can find the previous attempts at:
https://sourceware.org/ml/gdb-patches/2006-09/msg00192.htmlhttps://sourceware.org/ml/gdb-patches/2014-02/msg00248.htmlhttps://sourceware.org/ml/gdb-patches/2016-02/msg00226.html
The basic idea behind the patch is to start using the new
'getaddrinfo'/'getnameinfo' calls, which are responsible for
translating names and addresses in a protocol-independent way. This
means that if we ever have a new version of the IP protocol, we won't
need to change the code again (or, at least, won't have to change the
majority of the code).
The function 'getaddrinfo' returns a linked list of possible addresses
to connect to. Dealing with multiple addresses proved to be a hard
task with the current TCP auto-retry mechanism implemented on
ser-tcp:net_open. For example, when gdbserver listened only on an
IPv4 socket:
$ ./gdbserver --once 127.0.0.1:1234 ./a.out
and GDB was instructed to try to connect to both IPv6 and IPv4
sockets:
$ ./gdb -ex 'target extended-remote localhost:1234' ./a.out
the user would notice a somewhat big delay before GDB was able to
connect to the IPv4 socket. This happened because GDB was trying to
connect to the IPv6 socket first, and had to wait until the connection
timed out before it tried to connect to the IPv4 socket.
For that reason, I had to rewrite the main loop and implement a new
method for handling multiple connections. After some discussion,
Pedro and I agreed on the following algorithm:
1) For each entry returned by 'getaddrinfo', we try to open a socket
and connect to it.
2.a) If we have a successful 'connect', we just use that connection.
2.b) If we don't have a successfull 'connect', but if we've got a
ECONNREFUSED (meaning the the connection was refused), we keep track
of this fact by using a flag.
2.c) If we don't have a successfull 'connect', but if we've got a
EINPROGRESS (meaning that the connection is in progress), we perform
a 'select' call on the socket until we have a result (either a
successful connection, or an error on the socket).
3) If tcp_auto_retry is true, and we haven't gotten a successful
connection, and at least one of our attempts failed with
ECONNREFUSED, then we wait a little bit (i.e., call
'wait_for_connect'), check to see if there was a
timeout/interruption (in which case we bail out), and then go back
to (1).
After multiple tests, I was able to connect without delay on the
scenario described above, and was also able to connect in all other
types of scenarios.
I also implemented some hostname parsing functions (along with their
corresponding unit tests) which are used to help GDB and gdbserver to
parse hostname strings provided by the user. These new functions are
living inside common/netstuff.[ch]. I've had to do that since IPv6
introduces a new URL scheme, which defines that square brackets can be
used to enclose the host part and differentiate it from the
port (e.g., "[::1]:1234" means "host ::1, port 1234"). I spent some
time thinking about a reasonable way to interpret what the user wants,
and I came up with the following:
- If the user has provided a prefix that doesn't specify the protocol
version (i.e., "tcp:" or "udp:"), or if the user has not provided
any prefix, don't make any assumptions (i.e., assume AF_UNSPEC when
dealing with 'getaddrinfo') *unless* the host starts with "[" (in
which case, assume it's an IPv6 host).
- If the user has provided a prefix that does specify the protocol
version (i.e., "tcp4:", "tcp6:", "udp4:" or "udp6:"), then respect
that.
This method doesn't follow strictly what RFC 2732 proposes (that
literal IPv6 addresses should be provided enclosed in "[" and "]")
because IPv6 addresses still can be provided without square brackets
in our case, but since we have prefixes to specify protocol versions I
think this is not an issue.
Another thing worth mentioning is the new 'GDB_TEST_SOCKETHOST'
testcase parameter, which makes it possible to specify the
hostname (without the port) to be used when testing GDB and
gdbserver. For example, to run IPv6 tests:
$ make check-gdb RUNTESTFLAGS='GDB_TEST_SOCKETHOST=tcp6:[::1]'
Or, to run IPv4 tests:
$ make check-gdb RUNTESTFLAGS='GDB_TEST_SOCKETHOST=tcp4:127.0.0.1'
This required a few changes on the gdbserver-base.exp, and also a
minimal adjustment on gdb.server/run-without-local-binary.exp.
Finally, I've implemented a new testcase,
gdb.server/server-connect.exp, which is supposed to run on the native
host and perform various "smoke tests" using different connection
methods.
This patch has been regression-tested on BuildBot and locally, and
also built using a x86_64-w64-mingw32 GCC, and no problems were found.
gdb/ChangeLog:
2018-07-11 Sergio Durigan Junior <sergiodj@redhat.com>
Jan Kratochvil <jan.kratochvil@redhat.com>
Paul Fertser <fercerpav@gmail.com>
Tsutomu Seki <sekiriki@gmail.com>
Pedro Alves <palves@redhat.com>
* Makefile.in (SUBDIR_UNITTESTS_SRCS): Add
'unittests/parse-connection-spec-selftests.c'.
(COMMON_SFILES): Add 'common/netstuff.c'.
(HFILES_NO_SRCDIR): Add 'common/netstuff.h'.
* NEWS (Changes since GDB 8.2): Mention IPv6 support.
* common/netstuff.c: New file.
* common/netstuff.h: New file.
* ser-tcp.c: Include 'netstuff.h' and 'wspiapi.h'.
(wait_for_connect): Update comment. New parameter
'gdb::optional<int> sock' instead of 'struct serial *scb'.
Use 'sock' directly instead of 'scb->fd'.
(try_connect): New function, with code from 'net_open'.
(net_open): Rewrite main loop to deal with multiple
sockets/addresses. Handle IPv6-style hostnames; implement
support for IPv6 connections.
* unittests/parse-connection-spec-selftests.c: New file.
gdb/gdbserver/ChangeLog:
2018-07-11 Sergio Durigan Junior <sergiodj@redhat.com>
Jan Kratochvil <jan.kratochvil@redhat.com>
Paul Fertser <fercerpav@gmail.com>
Tsutomu Seki <sekiriki@gmail.com>
* Makefile.in (SFILES): Add '$(srcdir)/common/netstuff.c'.
(OBS): Add 'common/netstuff.o'.
(GDBREPLAY_OBS): Likewise.
* gdbreplay.c: Include 'wspiapi.h' and 'netstuff.h'.
(remote_open): Implement support for IPv6
connections.
* remote-utils.c: Include 'netstuff.h', 'filestuff.h'
and 'wspiapi.h'.
(handle_accept_event): Accept connections from IPv6 sources.
(remote_prepare): Handle IPv6-style hostnames; implement
support for IPv6 connections.
(remote_open): Implement support for printing connections from
IPv6 sources.
gdb/testsuite/ChangeLog:
2018-07-11 Sergio Durigan Junior <sergiodj@redhat.com>
Jan Kratochvil <jan.kratochvil@redhat.com>
Paul Fertser <fercerpav@gmail.com>
Tsutomu Seki <sekiriki@gmail.com>
* README (Testsuite Parameters): Mention new 'GDB_TEST_SOCKETHOST'
parameter.
* boards/native-extended-gdbserver.exp: Do not set 'sockethost'
by default.
* boards/native-gdbserver.exp: Likewise.
* gdb.server/run-without-local-binary.exp: Improve regexp used
for detecting when a remote debugging connection succeeds.
* gdb.server/server-connect.exp: New file.
* lib/gdbserver-support.exp (gdbserver_default_get_comm_port):
Do not prefix the port number with ":".
(gdbserver_start): New global GDB_TEST_SOCKETHOST. Implement
support for detecting and using it. Add '$debughost_gdbserver'
to the list of arguments used to start gdbserver. Handle case
when gdbserver cannot resolve a network name.
gdb/doc/ChangeLog:
2018-07-11 Sergio Durigan Junior <sergiodj@redhat.com>
Jan Kratochvil <jan.kratochvil@redhat.com>
Paul Fertser <fercerpav@gmail.com>
Tsutomu Seki <sekiriki@gmail.com>
* gdb.texinfo (Remote Connection Commands): Add explanation
about new IPv6 support. Add new connection prefixes.
This patch fixes a failure that happens when a structure has a static
member whose type is the same as itself. From the bug report:
Example code:
struct A
{
static A Empty;
int a;
};
int main(void) {
A a;
return 0;
}
Output:
(gdb) ptype/o A
/* offset | size */ type = struct A {
static struct A {
static struct A {
static struct A {
static struct A {
static struct A {
static struct A {
... # infinite loop
The problem here is that GDB is not taking into account the fact that
static members inside a class/struct are not stored in the
class/struct, and therefore they should not be accounted for during
the display of the offsets/sizes. The fix is simple: we just check if
the field we're dealing with (on
c-typeprint.c:c_type_print_base_struct_union) is static, and if it is
then we don't iterate over it.
This patch also adds a new test for this case, and doesn't introduce
any regressions. I believe it is important enough to be included in
the 8.2 branch.
OK?
gdb/ChangeLog:
2018-07-11 Sergio Durigan Junior <sergiodj@redhat.com>
PR c++/23373
* c-typeprint.c (c_type_print_base_struct_union): Don't print
offsets/sizes for static members of a class/struct.
gdb/testsuite/ChangeLog:
2018-07-11 Sergio Durigan Junior <sergiodj@redhat.com>
PR c++/23373
* gdb.base/ptype-offsets.cc (struct static_member): New
struct.
(main) <stmember>: New variable.
* gdb.base/ptype-offsets.exp: Add test for printing a struct
with a static member in it.
Expand the gdb.base/watchpoint-reuse-slot.exp test to be run twice,
once with hardware watchpoints disabled (this is new) and then with
hardware watchpoints enabled (the old way).
Running with hardware watchpoints enabled is skipped if the board file
says that hardware watchpoints are not supported on this target.
gdb/testsuite/ChangeLog:
* gdb.base/watchpoint-reuse-slot.exp: Test with hardware
watchpoints enabled and disabled.
In the gdb.base/examine-backward.exp test script, we check to see if
address zero is readable, and then read memory first forward from
address zero, and then backward from address zero.
The problem is, being able to read address zero does not guarantee
that you'll be able to read from the other end of the address space,
and the test probably shouldn't assume that is the case.
This patch updates the test script so that even if address zero is
known non-readable, we still run the tests, the tests in question are
mostly about, can GDB calculate the correct address to read from, we
can confirm this even if the final read ultimately fails. We also no
longer assume that if address zero is readable, then the other end of
the address space will be readable.
One further change is that, when we examined the memory at address
zero, the regexp used to match the address expected that the zero
address would have two '0' digits as the least significant digits. As
GDB strips leading zeros from addresses this was causing the test to
fail. I've reduced the zero address to a single 0 digit.
gdb/testsuite/ChangeLog:
* gdb.base/examine-backward.exp: Still run tests around address
0x0, even if address 0x0 is not readable. Update the pattern for
matching address 0x0 in expected output.
The test associated with the source file
gdb.base/share-env-with-gdbserver.c relies on calling malloc and free
within the inferior from GDB. However, as the test source itself
makes no use of these functions, there's no requirement that they be
linked into the test executable.
This commit adds a dummy call to malloc and free to ensure they are
linked into the test executable.
gdb/testsuite/ChangeLog:
* gdb.base/share-env-with-gdbserver.c (main): Add call to
malloc/free.
The version check of the dwarf compilation unit header in
error_check_comp_unit_head is done too late, and consequently dwarf code with
an unsupported version in the compilation unit header is interpreted as dwarf5
code in read_comp_unit_head.
Fixed by moving the check earlier.
Build and reg-tested on x86_64-linux.
2018-07-04 Tom de Vries <tdevries@suse.de>
* dwarf2read.c (error_check_comp_unit_head): Move dwarf version
check ...
(read_comp_unit_head): ... here.
* gdb.dwarf2/dw2-error.exp: Update expected error message.
Implement MIPS target support for passing options to the disassembler,
complementing commit 65b48a81404c ("GDB: Add support for the new
set/show disassembler-options commands.").
This includes options that expect an argument, so adjust the generic
code and data structures used so as to handle such options. So as to
give backends syntax flexibility no specific delimiter has been defined
to separate options from their respective arguments, so it has to be
included as the last character of the option name. Completion code
however has not been adjusted and consequently option arguments cannot
be completed at this time.
Also the MIPS target has non-empty defaults for the options, so that ABI
names for the general-purpose registers respect our `set mips abi ...'
setting rather than always being determined from the ELF headers of the
binary file selected. Handle these defaults as implicit options, never
shown to the user and always prepended to the user-specified options, so
that the latters can override the defaults.
The resulting output for the MIPS target is as follows:
(gdb) show disassembler-options
The current disassembler options are ''
The following disassembler options are supported for use with the
'set disassembler-options <option>[,<option>...]' command:
no-aliases Use canonical instruction forms.
msa Recognize MSA instructions.
virt Recognize the virtualization ASE instructions.
xpa Recognize the eXtended Physical Address (XPA) ASE
instructions.
ginv Recognize the Global INValidate (GINV) ASE instructions.
gpr-names=ABI Print GPR names according to specified ABI.
Default: based on binary being disassembled.
fpr-names=ABI Print FPR names according to specified ABI.
Default: numeric.
cp0-names=ARCH Print CP0 register names according to specified architecture.
Default: based on binary being disassembled.
hwr-names=ARCH Print HWR names according to specified architecture.
Default: based on binary being disassembled.
reg-names=ABI Print GPR and FPR names according to specified ABI.
reg-names=ARCH Print CP0 register and HWR names according to specified
architecture.
For the options above, the following values are supported for "ABI":
numeric 32 n32 64
For the options above, the following values are supported for "ARCH":
numeric r3000 r3900 r4000 r4010 vr4100 vr4111 vr4120 r4300 r4400 r4600
r4650 r5000 vr5400 vr5500 r5900 r6000 rm7000 rm9000 r8000 r10000 r12000
r14000 r16000 mips5 mips32 mips32r2 mips32r3 mips32r5 mips32r6 mips64
mips64r2 mips64r3 mips64r5 mips64r6 interaptiv-mr2 sb1 loongson2e
loongson2f loongson3a octeon octeon+ octeon2 octeon3 xlr xlp
(gdb)
which corresponds to what `objdump --help' used to print for the MIPS
target, with minor formatting changes, most notably option argument
lists being wrapped, but also the amount of white space separating
options from the respective descriptions. The relevant part the new
code is now also used by `objdump --help', which means these formatting
changes apply to both outputs, except for argument list wrapping, which
is GDB-specific.
This also adds a separating new line between the heading and option
lists where descriptions are provided, hence:
(gdb) set architecture s390:31-bit
(gdb) show disassembler-options
The current disassembler options are ''
The following disassembler options are supported for use with the
'set disassembler-options <option>[,<option>...]' command:
esa Disassemble in ESA architecture mode
zarch Disassemble in z/Architecture mode
insnlength Print unknown instructions according to length from first two bits
(gdb)
but:
(gdb) set architecture powerpc:common
(gdb) show disassembler-options
The current disassembler options are ''
The following disassembler options are supported for use with the
'set disassembler-options <option>[,<option>...]' command:
403, 405, 440, 464, 476, 601, 603, 604, 620, 7400, 7410, 7450, 7455, 750cl,
821, 850, 860, a2, altivec, any, booke, booke32, cell, com, e200z4, e300,
e500, e500mc, e500mc64, e5500, e6500, e500x2, efs, efs2, power4, power5,
power6, power7, power8, power9, ppc, ppc32, 32, ppc64, 64, ppc64bridge,
ppcps, pwr, pwr2, pwr4, pwr5, pwr5x, pwr6, pwr7, pwr8, pwr9, pwrx, raw, spe,
spe2, titan, vle, vsx
(gdb)
Existing affected target backends have been adjusted accordingly.
This has been verified manually with:
(gdb) set architecture arm
(gdb) set architecture powerpc:common
(gdb) set architecture s390:31-bit
to cause no issues with the `show disassembler-options' and `set
disassembler-options' commands. A test case for the MIPS target has
also been provided, covering the default settings with ABI overrides as
well as disassembler option overrides.
2018-07-02 Maciej W. Rozycki <macro@mips.com>
Simon Marchi <simon.marchi@polymtl.ca>
include/
PR tdep/8282
* dis-asm.h (disasm_option_arg_t): New typedef.
(disasm_options_and_args_t): Likewise.
(disasm_options_t): Add `arg' member, document members.
(disassembler_options_mips): New prototype.
(disassembler_options_arm, disassembler_options_powerpc)
(disassembler_options_s390): Update prototypes.
opcodes/
PR tdep/8282
* mips-dis.c (mips_option_arg_t): New enumeration.
(mips_options): New variable.
(disassembler_options_mips): New function.
(print_mips_disassembler_options): Reimplement in terms of
`disassembler_options_mips'.
* arm-dis.c (disassembler_options_arm): Adapt to using the
`disasm_options_and_args_t' structure.
* ppc-dis.c (disassembler_options_powerpc): Likewise.
* s390-dis.c (disassembler_options_s390): Likewise.
gdb/
PR tdep/8282
* disasm.h (gdb_disassembler): Add
`m_disassembler_options_holder'. member
* disasm.c (get_all_disassembler_options): New function.
(gdb_disassembler::gdb_disassembler): Use it.
(gdb_buffered_insn_length_init_dis): Likewise.
(gdb_buffered_insn_length): Adjust accordingly.
(set_disassembler_options): Handle options with arguments.
(show_disassembler_options_sfunc): Likewise. Add a leading new
line if showing options with descriptions.
(disassembler_options_completer): Adapt to using the
`disasm_options_and_args_t' structure.
* mips-tdep.c (mips_disassembler_options): New variable.
(mips_disassembler_options_o32): Likewise.
(mips_disassembler_options_n32): Likewise.
(mips_disassembler_options_n64): Likewise.
(gdb_print_insn_mips): Don't set `disassembler_options'.
(gdb_print_insn_mips_n32, gdb_print_insn_mips_n64): Remove
functions.
(mips_gdbarch_init): Always set `gdbarch_print_insn' to
`gdb_print_insn_mips'. Set `gdbarch_disassembler_options',
`gdbarch_disassembler_options_implicit' and
`gdbarch_valid_disassembler_options'.
* arm-tdep.c (_initialize_arm_tdep): Adapt to using the
`disasm_options_and_args_t' structure.
* gdbarch.sh (disassembler_options_implicit): New `gdbarch'
method.
(valid_disassembler_options): Switch from `disasm_options_t' to
the `disasm_options_and_args_t' structure.
* NEWS: Document `set disassembler-options' support for the MIPS
target.
* gdbarch.h: Regenerate.
* gdbarch.c: Regenerate.
gdb/doc/
PR tdep/8282
* gdb.texinfo (Source and Machine Code): Document `set
disassembler-options' support for the MIPS target.
gdb/testsuite/
PR tdep/8282
* gdb.arch/mips-disassembler-options.exp: New test.
* gdb.arch/mips-disassembler-options.s: New test source.
It's long annoyed me that "info threads"'s columns are misaligned.
Particularly the "Target Id" column's content is usually longer than
the specified column width, so the table ends up with the "Frame"
column misaligned. For example, currently we get this:
(gdb) info threads
Id Target Id Frame
1 Thread 0x7ffff7fb5740 (LWP 9056) "threads" 0x00007ffff7bc28ad in __pthread_join (threadid=140737345763072, thread_return=0x7fffffffd3e8) at pthread_join.c:90
2 Thread 0x7ffff7803700 (LWP 9060) "function0" thread_function0 (arg=0x0) at threads.c:90
* 3 Thread 0x7ffff7002700 (LWP 9061) "threads" thread_function1 (arg=0x1) at threads.c:106
The fact that the "Frame" heading is in a weird spot is particularly
annoying.
This commit turns the above into into this:
(gdb) info threads
Id Target Id Frame
1 Thread 0x7ffff7fb5740 (LWP 7548) "threads" 0x00007ffff7bc28ad in __pthread_join (threadid=140737345763072, thread_return=0x7fffffffd3e8) at pthread_join.c:90
2 Thread 0x7ffff7803700 (LWP 7555) "function0" thread_function0 (arg=0x0) at threads.c:91
* 3 Thread 0x7ffff7002700 (LWP 7557) "threads" thread_function1 (arg=0x1) at threads.c:104
It does that by computing the max width of the "Target Id" column and
using that as column width when creating the table.
This results in calling target_pid_to_str / target_extra_thread_info /
target_thread_name twice for each thread, but I think that it doesn't
matter in practice performance-wise, because the remote target caches
the info, and with native targets it shouldn't be noticeable. It
could matter if we have many threads (say, thousands), but then "info
threads" is practically useless in such a scenario anyway -- better
thread filtering and aggregation would be necessary.
(Note: I have an old branch somewhere where I attempted at making
gdb's "info threads"-like tables follow a model/view design, so that a
general framework took care of issues like these, but it's incomplete
and a much bigger change. This patch doesn't prevent going in that
direction in the future, of course.)
gdb/ChangeLog:
2018-06-29 Pedro Alves <palves@redhat.com>
* thread.c (thread_target_id_str): New, factored out from ...
(print_thread_info_1): ... here. Use it to compute the max
"Target Id" column width.
gdb/testsuite/ChangeLog:
2018-06-29 Pedro Alves <palves@redhat.com>
* gdb.threads/names.exp: Adjust expected "info threads" output.
While experimenting with the previous patch, I noticed this inconsistency
in GDB's output:
(gdb) b 32
Breakpoint 1 at 0x40062f: file inline-break.c, line 32. (1)
(gdb) r
....
Breakpoint 1, func1 (x=1) at inline-break.c:32 (2)
32 return x * 23; /* break here */
(gdb) info breakpoints
Num Type Disp Enb Address What
1 breakpoint keep y 0x40062f in main at inline-break.c:32 (3)
breakpoint already hit 1 time
(gdb)
Notice that when the breakpoint as set, GDB showed "inline-break.c,
line 32" (1), the same line number that was specified in the command.
When we run to the breakpoint, we present the stop at the same line
number, and correctly show "func1" as the function name (2).
But in "info break" output (3), notice that we say "in main", not "in
func1".
The same thing happens if you set a breakpoint by address. I.e.:
(gdb) b *0x40062f
Breakpoint 2 at 0x40062f: file inline-break.c, line 32.
(gdb) info breakpoints
Num Type Disp Enb Address What
2 breakpoint keep y 0x000000000040062f in main at inline-break.c:32
(gdb) r
....
Breakpoint 2, func1 (x=1) at inline-break.c:32
32 return x * 23; /* break here */
The problem is that the breakpoints were set at an inline function,
but when we set such a breakpoint by line number or address, we don't
record the functions symbol in the sal, and as consequence the
breakpoint location does not have an associated symbol either.
Then, in print_breakpoint_location, if the location does not have a
symbol, we call find_pc_sect_function to find one, and this is what
finds "main", because find_pc_sect_function uses
block_linkage_function:
/* Return the symbol for the function which contains a specified
lexical block, described by a struct block BL. The return value
will not be an inlined function; the containing function will be
returned instead. */
struct symbol *
block_linkage_function (const struct block *bl)
To fix this, this commit adds an alternative to find_pc_sect_function
that uses block_containing_function instead:
/* Return the symbol for the function which contains a specified
block, described by a struct block BL. The return value will be
the closest enclosing function, which might be an inline
function. */
struct symbol *
block_containing_function (const struct block *bl)
(It seems odd to me that block_linkage_function says "the CONTAINING
function will be returned", and then block_containing_function says it
returns "the closest enclosing function". Something seems reversed
here. Still, I've kept the same nomenclature and copied the comments,
so that at least there's consistency. Maybe we should fix that up
somehow.)
Then I wondered, why make print_breakpoint_location look up the symbol
every time it is called, instead of just always storing the symbol
when the location is created, since the location already stores the
symbol in some cases. So to find which cases might be missing setting
the symbol in the sal which is used to create the breakpoint location,
I added an assertion to print_breakpoint_location, and ran the
testsuite. That caught a few places, unsurprisingly:
- setting a breakpoint by line number
- setting a breapoint by address
- ifunc resolving
Those are all fixed by this commit. I decided not to add the
assertion to block_linkage_function and leave the existing "if (sym)"
check in place, because it's plausible that we have symtabs with line
info but no symbols. I.e., that would not be a GDB bug, but
a peculiarity of debug info input.
gdb/ChangeLog:
2018-06-29 Pedro Alves <palves@redhat.com>
* blockframe.c (find_pc_sect_containing_function): New function.
* breakpoint.c (print_breakpoint_location): Don't call
find_pc_sect_function.
* linespec.c (create_sals_line_offset): Record the location's
symbol in the sal.
* linespec.c (convert_address_location_to_sals): Fill in sal's
symbol with find_pc_sect_containing_function.
* symtab.c (find_function_start_sal): Rename to ...
(find_function_start_sal_1): ... this.
(find_function_start_sal): Reimplement as wrapper around
find_function_start_sal_1, and use
find_pc_sect_containing_function to fill in the sal's symbol.
(find_function_start_sal(symbol*, bool)): Adjust.
* symtab.h (find_pc_function, find_pc_sect_function): Adjust
comments.
(find_pc_sect_containing_function): Declare.
gdb/testsuite/ChangeLog:
2018-06-29 Pedro Alves <palves@redhat.com>
* gdb.opt/inline-break.exp (line number, address): Add "info
break" tests.
Commit 61b04dd04ac2 ("Change inline frame breakpoint skipping logic
(fix gdb.gdb/selftest.exp)") caused a GDB crash when you set a
breakpoint by line number in an inline function, and then run to the
breakpoint:
$ gdb -q test Reading symbols from test...done.
(gdb) b inline-break.c:32
Breakpoint 1 at 0x40062f: file inline-break.c, line 32.
(gdb) run
Starting program: /[...]/test
[1] 75618 segmentation fault /[...]/gdb -q test
The problem occurs because we assume that a bp_location's symbol is
not NULL, which is not true when we set the breakpoint with a linespec
location:
Program received signal SIGSEGV, Segmentation fault.
0x00000000006f42bb in stopped_by_user_bp_inline_frame (
stop_chain=<optimized out>, frame_block=<optimized out>)
at gdb/inline-frame.c:305
305 && frame_block == SYMBOL_BLOCK_VALUE (loc->symbol))
(gdb) p loc->symbol
$1 = (const symbol *) 0x0
The same thing happens if you run to a breakpoint set in an inline
function by address:
(gdb) b *0x40062f
Breakpoint 3 at 0x40062f: file inline-break.c, line 32.
To fix this, add a null pointer check, to avoid the crash, and make it
so that if there's no symbol for the location, then we present the
stop at the inline function. This preserves the previous behavior
when e.g., setting a breakpoint by address, with "b *ADDRESS".
gdb/ChangeLog:
2018-06-29 Pedro Alves <palves@redhat.com>
* inline-frame.c (stopped_by_user_bp_inline_frame): Return
true if the the location has no symbol.
gdb/testsuite/ChangeLog:
2018-06-29 Pedro Alves <palves@redhat.com>
* gdb.opt/inline-break.c (func1): Add "break here" marker.
* gdb.opt/inline-break.exp: Test setting breakpoints by line
number and address and running to them.
Fedora rpmbuild has been complaining:
*** WARNING: ./usr/src/debug/gdb-8.1.50.20180618-24.fc28.x86_64/gdb/gdbserver/x86-tdesc.h is executable but has empty or no shebang, removing executable bit
gdb/gdbserver/ChangeLog
2018-06-28 Jan Kratochvil <jan.kratochvil@redhat.com>
* x86-tdesc.h: Remove executable permission flag.
gdb/testsuite/ChangeLog
2018-06-28 Jan Kratochvil <jan.kratochvil@redhat.com>
* lib/compiler.c: Remove executable permission flag.
watchpoint-hw-attach.exp was noticed to fail on some machines.
Thanks to the input from sergiodj and palves on the IRC channel,
it was concluded that the test case incorrectly assumed that on
attach it was landed in the top-most frame of the inferior. This
was fixed by running to a break point in main by explicitly
defining the source file name before continuing with the test.
Tested on the following architectures x86_64, aarch64 and ppc64le.
gdb/testsuite/ChangeLog:
* gdb.base/watchpoint-hw-attach.c (main): Remove unneeded
code.
* gdb.base/watchpoint-hw-attach.exp: Break in outermost frame.
If all sections of a symbol file are loaded with a fixed offset, it
is easier to specify that offset than listing all sections
explicitly. There is also a similar option for "symbol-file".
gdb/ChangeLog:
2018-06-28 Petr Tesarik <ptesarik@suse.cz>
* symfile.c (add_symbol_file_command, _initialize_symfile): Add
option "-o" to add-symbol-file-load to add an offset to each
section's load address.
* symfile.c (set_objfile_default_section_offset): New function.
gdb/doc/ChangeLog:
2018-06-28 Petr Tesarik <ptesarik@suse.cz>
* gdb.texinfo (Files): Document "add-symbol-file -o offset".
gdb/testsuite/ChangeLog:
2018-06-28 Petr Tesarik <ptesarik@suse.cz>
* gdb.base/relocate.exp: Add test for "add-symbol-file -o ".
The (first) .text section must be always specified as the second
non-option argument. The documentation states that GDB cannot
figure out this address by itself. This is true if the object file
was indeed relocated, but it is also confusing, because all other
sections can be omitted and will use the address provided by BFD.
gdb/ChangeLog:
2018-06-28 Petr Tesarik <ptesarik@suse.cz>
* symfile.c (add_symbol_file_command, _initialize_symfile): Do not
require the second argument. If omitted, load sections at the
addresses specified in the file.
gdb/doc/ChangeLog:
2018-06-28 Petr Tesarik <ptesarik@suse.cz>
* gdb.texinfo (Files): The address argument for "add-symbol-file"
is no longer mandatory.
gdb/testsuite/ChangeLog:
2018-06-28 Petr Tesarik <ptesarik@suse.cz>
* gdb.base/relocate.exp: Test add-symbol-file behavior when the
address argument is omitted.
If the main file is relocated at runtime, all symbols are offset by
a fixed amount. Let the user specify this offset when loading a
symbol file.
gdb/ChangeLog:
2018-06-28 Petr Tesarik <ptesarik@suse.cz>
* symfile.c (symbol_file_command, symbol_file_add_main_1)
(_initialize_symfile): Add option "-o" to symbol-file to add an
offset to each section of the symbol file.
gdb/doc/ChangeLog:
2018-06-28 Petr Tesarik <ptesarik@suse.cz>
* gdb.texinfo (Files): Document "symbol-file -o offset".
gdb/testsuite/ChangeLog:
2018-06-28 Petr Tesarik <ptesarik@suse.cz>
* gdb.base/relocate.exp: Add test for "symbol-file -o ".
This patch fixes a unique condition where GDB fails to provide line
information of symbol at address zero when code is compiled with text
address zero but loaded at an offset > 0.
For example lets compile following code snippet:
int main() {
return 0;
}
gcc -O0 -g3 -nostdlib -emain -Wl,-Ttext=0x00 -o file.out file.c
Start gdb and run:
add-symbol-file file.out 0xffff0000
info line main
GDB will return error saying no line info is available for the symbol.
This is a direct consequence of the fix for PR 12528 where GDB tries to ignore
line table for a function which has been garbage collected by the linker.
As the garbage collected symbols are sent to address zero GDB assumes a symbol
actually placed at address zero as garbage collected.
This was fixed with an additional check address < lowpc. But when symbol is
loaded at an offset lowpc becomes lowpc + offset while no offset is added to
address rather final symbol address is calculated based on baseaddr and address
added together. So in case where symbols are loaded at an offset the condition
address < lowpc will always return true.
This patch fixes this by comparing address against a non offset lowpc.
This patch also adds a GDB test case to replicate this behavior.
gdb:
2018-06-27 Omair Javaid <omair.javaid@linaro.org>
PR gdb/21695
* dwarf2read.c (lnp_state_machine::check_line_address): Update declaration.
(dwarf_decode_lines_1): Adjust.
gdb/testsuite:
2018-06-27 Omair Javaid <omair.javaid@linaro.org>
PR gdb/21695
* gdb.base/infoline-reloc-main-from-zero.exp: New test.
* gdb.base/infoline-reloc-main-from-zero.c: New file.
This adds support for ptype/o to the Rust language code.
By default, the Rust compiler reorders fields to reduce padding. So,
the Rust language code sorts the fields by offset before printing.
This may yield somewhat odd-looking results, but it is faithful to
"what really happens", and might be useful when doing lower-level
debugging.
The reordering can be disabled using #[repr(c)]; ptype/o might be more
useful in this case.
gdb/ChangeLog
2018-06-26 Tom Tromey <tom@tromey.com>
PR rust/22574:
* typeprint.c (whatis_exp): Allow ptype/o for Rust.
* rust-lang.c (rust_print_struct_def): Add podata parameter.
Update.
(rust_internal_print_type): Add podata parameter.
(rust_print_type): Update.
gdb/testsuite/ChangeLog
2018-06-26 Tom Tromey <tom@tromey.com>
PR rust/22574:
* gdb.rust/simple.exp (test_one_slice): Add ptype/o tests.
* gdb.rust/simple.rs (struct SimpleLayout): New.
Commit
e813d34 ("Align natural-format register values to the same column")
changed the output of "info registers" (tabs to spaces), but didn't
update gdb.base/jit-reader.exp. Update the regexes to expect spaces
instead.
gdb/testsuite/ChangeLog:
* gdb.base/jit-reader.exp (jit_reader_test): Expect spaces in
"info registers" output.
This commit fixes a bug whereby hardware watchpoints are not used on
aarch64 when attaching to a target. The fix adds an aarch64 specialization
of post_attach which records the number of available hardware debug registers
using aarch64_linux_get_debug_reg_capacity. This implementation mirrors that
of aarch64_linux_child_post_startup_inferior which successfully enables the
use of hardware watchpoints when launching the target under the debugger.
gdb/ChangeLog:
* aarch64-linux-nat.c (post_attach): New.
(aarch64_linux_nat_target::post_attach): Override post_attach to
record the number of hardware debug registers.
gdb/testsuite/ChangeLog:
* gdb.base/watchpoint-hw-attach.c: New test.
* gdb.base/watchpoint-hw-attach.exp: New file.
The cc-with-tweaks.sh script needs to be executed with bash. When
trying to run this:
make check RUNTESTFLAGS="--target_board=dwarf4-gdb-index" TESTS="gdb.base/return.exp"
I get:
gdb compile failed, /home/emaisin/src/binutils-gdb/gdb/contrib/cc-with-tweaks.sh: 174: /home/emaisin/src/binutils-gdb/gdb/contrib/cc-with-tweaks.sh: Bad substitution
The reason is that the board files execute cc-with-tweaks.sh using
/bin/sh, which points to dash on my machine. Remove the /bin/sh part
and let the shebang choose the right interpreter.
gdb/testsuite/ChangeLog:
* boards/cc-with-tweaks.exp: Don't call cc-with-tweaks.sh
through /bin/sh.
* boards/dwarf4-gdb-index.exp: Likewise.
* boards/fission-dwp.exp: Likewise.
If the "print large128" sub-test fails in a manner that typically
indicates internal overflow due to GDB being built without MPFR
support, explicitly state this in the failure message.
gdb/testsuite/ChangeLog:
2018-06-20 Ulrich Weigand <uweigand@de.ibm.com>
* gdb.base/float128.exp: Add comment and improved fail message
to the failure case of "print large128" test.
When trying to run the update-gnulib.sh script in gdb, I get this:
Error: Wrong automake version (Unescaped left brace in regex is deprecated, passed through in regex; marked by <-- HERE in m/\${ <-- HERE ([^ =:+{}]+)}/ at /opt/automake/1.11.1/bin/automake line 4113.), we need 1.11.1.
Aborting.
Apparently, it's an issue with a regex in automake that triggers a
warning starting with Perl 5.22. It has been fixed in automake 1.15.1.
So I think it's a good excuse to bump the versions of autoconf and
automake used in the gnulib import. And to avoid requiring multiple
builds of autoconf/automake, it was suggested that we bump the required
version of those tools for all binutils-gdb.
For autoconf, the 2.69 version is universally available, so it's an easy
choice. For automake, different distros and distro versions have
different automake versions. But 1.15.1 seems to be the most readily
available as a package. In any case, it's easy to build it from source.
I removed the version checks from AUTOMAKE_OPTIONS and AC_PREREQ,
because I don't think they are useful in our case. They only specify a
lower bound for the acceptable version of automake/autoconf. That's
useful if you let the user choose the version of the tool they want to
use, but want to set a minimum version (because you use a feature that
was introduced in that version). In our case, we force people to use a
specific version anyway. For the autoconf version, we have the check in
config/override.m4 that enforces the version we want. It will be one
less thing to update next time we change autotools version.
I hit a few categories of problems that required some changes. They are
described below along with the chosen solutions.
Problem 1:
configure.ac:17: warning: AM_INIT_AUTOMAKE: two- and three-arguments forms are deprecated. For more info, see:
configure.ac:17: http://www.gnu.org/software/automake/manual/automake.html#Modernize-AM_005fINIT_005fAUTOMAKE-invocation
Solution 1:
Adjust the code based on the example at that URL.
Problem 2 (in zlib/):
Makefile.am: error: required file './INSTALL' not found
Makefile.am: 'automake --add-missing' can install 'INSTALL'
Makefile.am: error: required file './NEWS' not found
Makefile.am: error: required file './AUTHORS' not found
Makefile.am: error: required file './COPYING' not found
Makefile.am: 'automake --add-missing' can install 'COPYING'
Solution 2:
Add the foreign option to AUTOMAKE_OPTIONS.
Problem 3:
doc/Makefile.am:20: error: support for Cygnus-style trees has been removed
Solution 3:
Remove the cygnus options.
Problem 4:
Makefile.am:656: warning: 'INCLUDES' is the old name for 'AM_CPPFLAGS' (or '*_CPPFLAGS')
Solution 4:
Rename "INCLUDES = " to "AM_CPPFLAGS += " (because AM_CPPFLAGS is
already defined earlier).
Problem 5:
doc/Makefile.am:71: warning: suffix '.texinfo' for Texinfo files is discouraged; use '.texi' instead
doc/Makefile.am: warning: Oops!
doc/Makefile.am: It appears this file (or files included by it) are triggering
doc/Makefile.am: an undocumented, soon-to-be-removed automake hack.
doc/Makefile.am: Future automake versions will no longer place in the builddir
doc/Makefile.am: (rather than in the srcdir) the generated '.info' files that
doc/Makefile.am: appear to be cleaned, by e.g. being listed in CLEANFILES or
doc/Makefile.am: DISTCLEANFILES.
doc/Makefile.am: If you want your '.info' files to be placed in the builddir
doc/Makefile.am: rather than in the srcdir, you have to use the shiny new
doc/Makefile.am: 'info-in-builddir' automake option.
Solution 5:
Rename .texinfo files to .texi.
Problem 6:
doc/Makefile.am: warning: Oops!
doc/Makefile.am: It appears this file (or files included by it) are triggering
doc/Makefile.am: an undocumented, soon-to-be-removed automake hack.
doc/Makefile.am: Future automake versions will no longer place in the builddir
doc/Makefile.am: (rather than in the srcdir) the generated '.info' files that
doc/Makefile.am: appear to be cleaned, by e.g. being listed in CLEANFILES or
doc/Makefile.am: DISTCLEANFILES.
doc/Makefile.am: If you want your '.info' files to be placed in the builddir
doc/Makefile.am: rather than in the srcdir, you have to use the shiny new
doc/Makefile.am: 'info-in-builddir' automake option.
Solution 6:
Remove the hack at the bottom of doc/Makefile.am and use
the info-in-builddir automake option.
Problem 7:
doc/Makefile.am:35: error: required file '../texinfo.tex' not found
doc/Makefile.am:35: 'automake --add-missing' can install 'texinfo.tex'
Solution 7:
Use the no-texinfo.tex automake option. We also have one in
texinfo/texinfo.tex, not sure if we should point to that, or move it
(or a newer version of it added with automake --add-missing) to
top-level.
Problem 8:
Makefile.am:131: warning: source file 'config/tc-aarch64.c' is in a subdirectory,
Makefile.am:131: but option 'subdir-objects' is disabled
automake: warning: possible forward-incompatibility.
automake: At least a source file is in a subdirectory, but the 'subdir-objects'
automake: automake option hasn't been enabled. For now, the corresponding output
automake: object file(s) will be placed in the top-level directory. However,
automake: this behaviour will change in future Automake versions: they will
automake: unconditionally cause object files to be placed in the same subdirectory
automake: of the corresponding sources.
automake: You are advised to start using 'subdir-objects' option throughout your
automake: project, to avoid future incompatibilities.
Solution 8:
Use subdir-objects, that means adjusting references to some .o that will now
be in config/.
Problem 9:
configure.ac:375: warning: AC_LANG_CONFTEST: no AC_LANG_SOURCE call detected in body
../../lib/autoconf/lang.m4:193: AC_LANG_CONFTEST is expanded from...
../../lib/autoconf/general.m4:2601: _AC_COMPILE_IFELSE is expanded from...
../../lib/autoconf/general.m4:2617: AC_COMPILE_IFELSE is expanded from...
../../lib/m4sugar/m4sh.m4:639: AS_IF is expanded from...
../../lib/autoconf/general.m4:2042: AC_CACHE_VAL is expanded from...
../../lib/autoconf/general.m4:2063: AC_CACHE_CHECK is expanded from...
configure.ac:375: the top level
Solution 9:
Use AC_LANG_SOURCE, or use proper quoting.
Problem 10 (in intl/):
configure.ac:7: warning: AC_COMPILE_IFELSE was called before AC_USE_SYSTEM_EXTENSIONS
/usr/share/aclocal/threadlib.m4:36: gl_THREADLIB_EARLY_BODY is expanded from...
/usr/share/aclocal/threadlib.m4:29: gl_THREADLIB_EARLY is expanded from...
/usr/share/aclocal/threadlib.m4:318: gl_THREADLIB is expanded from...
/usr/share/aclocal/lock.m4:9: gl_LOCK is expanded from...
/usr/share/aclocal/intl.m4:211: gt_INTL_SUBDIR_CORE is expanded from...
/usr/share/aclocal/intl.m4:25: AM_INTL_SUBDIR is expanded from...
/usr/share/aclocal/gettext.m4:57: AM_GNU_GETTEXT is expanded from...
configure.ac:7: the top level
Solution 10:
Add AC_USE_SYSTEM_EXTENSIONS in configure.ac.
ChangeLog:
* libtool.m4: Use AC_LANG_SOURCE.
* configure.ac: Remove AC_PREREQ, use AC_LANG_SOURCE.
* README-maintainer-mode: Update version requirements.
* ar-lib: New file.
* test-driver: New file.
* configure: Re-generate.
bfd/ChangeLog:
* Makefile.am (AUTOMAKE_OPTIONS): Remove 1.11.
(INCLUDES): Rename to ...
(AM_CPPFLAGS): ... this.
* configure.ac: Remove AC_PREREQ.
* doc/Makefile.am (AUTOMAKE_OPTIONS): Remove 1.9, cygnus, add
info-in-builddir no-texinfo.tex.
(info_TEXINFOS): Rename bfd.texinfo to bfd.texi.
* doc/bfd.texinfo: Rename to ...
* doc/bfd.texi: ... this.
* Makefile.in: Re-generate.
* aclocal.m4: Re-generate.
* config.in: Re-generate.
* configure: Re-generate.
* doc/Makefile.in: Re-generate.
binutils/ChangeLog:
* configure.ac: Remove AC_PREREQ.
* doc/Makefile.am (AUTOMAKE_OPTIONS): Remove cygnus, add
info-in-builddir no-texinfo.tex.
* Makefile.in: Re-generate.
* aclocal.m4: Re-generate.
* config.in: Re-generate.
* configure: Re-generate.
* doc/Makefile.in: Re-generate.
config/ChangeLog:
* override.m4 (_GCC_AUTOCONF_VERSION): Bump from 2.64 to 2.69.
etc/ChangeLog:
* configure.in: Remove AC_PREREQ.
* configure: Re-generate.
gas/ChangeLog:
* Makefile.am (AUTOMAKE_OPTIONS): Remove 1.11, add subdir-objects.
(TARG_CPU_O, OBJ_FORMAT_O, ATOF_TARG_O): Add config/ prefix.
* configure.ac (TARG_CPU_O, OBJ_FORMAT_O, ATOF_TARG_O, emfiles,
extra_objects): Add config/ prefix.
* doc/as.texinfo: Rename to...
* doc/as.texi: ... this.
* doc/Makefile.am: Rename as.texinfo to as.texi throughout.
Remove DISTCLEANFILES hack.
(AUTOMAKE_OPTIONS): Remove 1.8, cygnus, add no-texinfo.tex and
info-in-builddir.
* Makefile.in: Re-generate.
* aclocal.m4: Re-generate.
* config.in: Re-generate.
* configure: Re-generate.
* doc/Makefile.in: Re-generate.
gdb/ChangeLog:
* common/common-defs.h (PACKAGE_NAME, PACKAGE_VERSION,
PACKAGE_STRING, PACKAGE_TARNAME): Undefine.
* configure.ac: Remove AC_PREREQ, add missing quoting.
* gnulib/configure.ac: Modernize usage of
AC_INIT/AM_INIT_AUTOMAKE. Remove AC_PREREQ.
* gnulib/update-gnulib.sh (AUTOCONF_VERSION): Bump to 2.69.
(AUTOMAKE_VERSION): Bump to 1.15.1.
* configure: Re-generate.
* config.in: Re-generate.
* aclocal.m4: Re-generate.
* gnulib/aclocal.m4: Re-generate.
* gnulib/config.in: Re-generate.
* gnulib/configure: Re-generate.
* gnulib/import/Makefile.in: Re-generate.
gdb/gdbserver/ChangeLog:
* configure.ac: Remove AC_PREREQ, add missing quoting.
* configure: Re-generate.
* config.in: Re-generate.
* aclocal.m4: Re-generate.
gdb/testsuite/ChangeLog:
* configure.ac: Remove AC_PREREQ.
* configure: Re-generate.
gold/ChangeLog:
* configure.ac: Remove AC_PREREQ, add missing quoting and usage
of AC_LANG_SOURCE.
* Makefile.in: Re-generate.
* aclocal.m4: Re-generate.
* configure: Re-generate.
* testsuite/Makefile.in: Re-generate.
gprof/ChangeLog:
* configure.ac: Remove AC_PREREQ.
* Makefile.am: Remove DISTCLEANFILES hack.
(AUTOMAKE_OPTIONS): Remove 1.11, add info-in-builddir.
* Makefile.in: Re-generate.
* aclocal.m4: Re-generate.
* configure: Re-generate.
* gconfig.in: Re-generate.
intl/ChangeLog:
* configure.ac: Add AC_USE_SYSTEM_EXTENSIONS, remove AC_PREREQ.
* configure: Re-generate.
* config.h.in: Re-generate.
* aclocal.m4: Re-generate.
ld/ChangeLog:
* configure.ac: Remove AC_PREREQ.
* Makefile.am: Remove DISTCLEANFILES hack, rename ld.texinfo to
ld.texi, ldint.texinfo to ldint.texi throughout.
(AUTOMAKE_OPTIONS): Add info-in-builddir.
* README: Rename ld.texinfo to ld.texi, ldint.texinfo to
ldint.texi throughout.
* gen-doc.texi: Likewise.
* h8-doc.texi: Likewise.
* ld.texinfo: Rename to ...
* ld.texi: ... this.
* ldint.texinfo: Rename to ...
* ldint.texi: ... this.
* Makefile.in: Re-generate.
* aclocal.m4: Re-generate.
* config.in: Re-generate.
* configure: Re-generate.
libdecnumber/ChangeLog:
* configure.ac: Remove AC_PREREQ.
* configure: Re-generate.
* aclocal.m4.
libiberty/ChangeLog:
* configure.ac: Remove AC_PREREQ.
* configure: Re-generate.
* config.in: Re-generate.
opcodes/ChangeLog:
* Makefile.am (AUTOMAKE_OPTIONS): Remove 1.11.
* configure.ac: Remove AC_PREREQ.
* Makefile.in: Re-generate.
* aclocal.m4: Re-generate.
* configure: Re-generate.
readline/ChangeLog.gdb:
* configure: Re-generate.
* examples/rlfe/configure: Re-generate.
sim/ChangeLog:
* All configure.ac: Remove AC_PREREQ.
* All configure: Re-generate.
zlib/ChangeLog.bin-gdb:
* configure.ac: Modernize AC_INIT call, remove AC_PREREQ.
* Makefile.am (AUTOMAKE_OPTIONS): Remove 1.8, cygnus, add
foreign.
* Makefile.in: Re-generate.
* aclocal.m4: Re-generate.
* configure: Re-generate.
Currently, gdb.gdb/selftest.exp fails if you build GDB with
optimization (-O2, etc.).
The reason is that after setting a breakpoint in captured_main, we
stop at:
...
Breakpoint 1, captured_main_1 (context=<optimized out>) at src/gdb/main.c:492
...
while selftest_setup expects a stop at captured_main.
Here, captured_main_1 has been inlined into captured_main, and
captured_main has been inlined into gdb_main:
...
$ nm ./build/gdb/gdb | egrep ' [tT] .*captured_main|gdb_main' | c++filt
000000000061b950 T gdb_main(captured_main_args*)
...
Indeed, the two inlined functions show up in the backtrace:
...
(gdb) bt
#0 captured_main_1 (context=<optimized out>) at main.c:492
#1 captured_main (data=<optimized out>) at main.c:1147
#2 gdb_main (args=args@entry=0x7fffffffdb80) at main.c:1173
#3 0x000000000040fea5 in main (argc=<optimized out>, argv=<optimized out>)
at gdb.c:32
...
We're now stopping at captured_main_1 because commit ddfe970e6bec
("Don't elide all inlined frames") makes GDB present a stop at the
innermost inlined frame if the program stopped by a user breakpoint.
Now, the selftest.exp testcase explicitly asks to stop at
"captured_main", not "captured_main_1", so I'm thinking that it's
GDB'S behavior that should be improved. That is what this commit
does, by only showing a stop at an inline frame if the user breakpoint
was set in that frame's block.
Before this commit:
(top-gdb) b captured_main
Breakpoint 1 at 0x792f99: file src/gdb/main.c, line 492.
(top-gdb) r
Starting program: build/gdb/gdb
Breakpoint 1, captured_main_1 (context=<optimized out>) at src/gdb/main.c:492
492 lim_at_start = (char *) sbrk (0);
(top-gdb)
After this commit, we now instead get:
(top-gdb) b captured_main
Breakpoint 1 at 0x791339: file src/gdb/main.c, line 492.
(top-gdb) r
Starting program: build/gdb/gdb
Breakpoint 1, captured_main (data=<optimized out>) at src/gdb/main.c:1147
1147 captured_main_1 (context);
(top-gdb)
and:
(top-gdb) b captured_main_1
Breakpoint 2 at 0x791339: file src/gdb/main.c, line 492.
(top-gdb) r
Starting program: build/gdb/gdb
Breakpoint 2, captured_main_1 (context=<optimized out>) at src/gdb/main.c:492
492 lim_at_start = (char *) sbrk (0);
(top-gdb)
Note that both captured_main and captured_main_1 resolved to the same
address, 0x791339. That is necessary to trigger the issue in
question. The gdb.base/inline-break.exp testcase currently does not
exercise that, but the new test added by this commit does. That new
test fails without the GDB fix and passes with the fix. No
regressions on x86-64 GNU/Linux.
While at it, the THIS_PC comparison in stopped_by_user_bp_inline_frame
is basically a nop, so just remove it -- if a software or hardware
breakpoint explains the stop, then it must be that it was installed at
the current PC.
gdb/ChangeLog:
2018-06-19 Pedro Alves <palves@redhat.com>
* inline-frame.c (stopped_by_user_bp_inline_frame): Replace PC
parameter with a block parameter. Compare location's block symbol
with the frame's block instead of addresses.
(skip_inline_frames): Pass the current block instead of the
frame's address. Break out as soon as we determine the frame
should not be skipped.
gdb/testsuite/ChangeLog:
2018-06-19 Pedro Alves <palves@redhat.com>
* gdb.opt/inline-break.c (func_inline_callee, func_inline_caller)
(func_extern_caller): New.
(main): Call func_extern_caller.
* gdb.opt/inline-break.exp: Add tests for inline frame skipping
logic change.
The test case below demonstrates the problem, as described in this PR's Comment 5:
typedef struct {
int x;
} A;
struct C : A {
int y;
};
int main()
{
C c;
return 55;
}
$ gdb a.out
(gdb) ptype C::x
Internal error: non-aggregate type to value_struct_elt_for_reference
In value_struct_elt_for_reference(), need to call check_typedef() on
the aggregate type to handle the case of *curtype being ptr->typedef.
Tested on x86_64-linux. No regressions.
At the moment, bp_inlined_func.exp passes for a combined current gcc and
gdb-binutils repos build but fails for a build with system gcc (7.3.1) and
ld (2.29.1).
It checks for 4 breakpoints on read_small:
...
gdb_test "break read_small" \
"Breakpoint $decimal at $hex: read_small\\. \\(4 locations\\)" \
"set breakpoint at read_small"
...
and fails because it gets 5 breakpoint locations instead:
...
(gdb) break read_small
Breakpoint 2 at 0x401f9a: read_small. (5 locations)
(gdb) FAIL: gdb.ada/bp_inlined_func.exp: set breakpoint at read_small
...
The 4 expected breakpoint locations are inlined versions of read_small, and
the 5th breakpoint location has this address:
...
(gdb) info breakpoint
Num Type Disp Enb Address What
1 breakpoint keep y <MULTIPLE>
1.1 y 0x0000000000401f9a in b.read_small
at bp_inlined_func/b.adb:20
...
which is the read_small function itself:
...
(gdb) x 0x0000000000401f9a
0x401f9a <b__read_small+4>: 0x22f8058b
...
This patch updates the test to allow 5 breakpoint locations.
Tested on the configurations mentioned above, on x86_64.
2018-06-18 Tom de Vries <tdevries@suse.de>
* gdb.ada/bp_inlined_func.exp: Allow 5 breakpoint locations.
This patch fixes an issue where GDB would sometimes hang when
attaching to a multi-threaded process. This issue was especially
likely to trigger if the machine (running the inferior) was under
load.
In summary, the problem is an imbalance between two functions in
linux-nat.c, stop_callback and stop_wait_callback. In stop_callback
we send SIGSTOP to a thread, but _only_ if the thread is not already
stopped, and if it is not signalled, which means it should stop soon.
In stop_wait_callback we wait for the SIGSTOP to arrive, however, we
are aware that the thread might have been signalled for some other
reason, and so if a signal other than SIGSTOP causes the thread to
stop then we stash that signal away so it can be reported back later.
If we get a SIGSTOP then this is discarded, after all, this signal was
sent from stop_callback. Except that this might not be the case, it
could be that SIGSTOP was sent to a thread from elsewhere in GDB, in
which case we would not have sent another SIGSTOP from stop_callback
and the SIGSTOP received in stop_wait_callback should not be ignored.
Below I've laid out the exact sequence of events that I saw that lead
me to track down the above diagnosis.
After attaching to the inferior GDB sends a SIGSTOP to all of the
threads and then returns to the event loop waiting for interesting
things to happen.
Eventually the first target event is detected (this will be the first
SIGSTOP arriving) and GDB calls inferior_event_handler which calls
fetch_inferior_event. Inside fetch_inferior_event GDB calls
do_target_wait which calls target_wait to find a thread with an event.
The target_wait call ends up in linux_nat_wait_1, which first checks
to see if any threads already have stashed stop events to report, and
if there are none then we enter a loop fetching as many events as
possible out of the kernel. This event fetching is non-blocking, and
we give up once the kernel has no more events ready to give us.
All of the events from the kernel are passed through
linux_nat_filter_event which stashes the wait status for all of the
threads that reported a SIGSTOP, these will be returned by future
calls to linux_nat_wait_1.
Lets assume for a moment that we've attached to a multi-threaded
inferior, and that all but one thread has reported its stop during the
initial wait call in linux_nat_wait_1. The other thread will be
reporting a SIGSTOP, but the kernel has not yet managed to deliver
that signal to GDB before GDB gave up waiting and continued handling
the events it already had. GDB selects one of the threads that has
reported a SIGSTOP and passes this thread ID back to
fetch_inferior_event.
To handle the thread's SIGSTOP, GDB calls handle_signal_stop, which
calls stop_all_threads, this calls wait_one, which in turn calls
target_wait.
The first call to target_wait at this point will result in a stashed
wait status being returned, at which point we call setup_inferior.
The call to setup_inferior leads to a call into try_thread_db_load_1
which results in a call to linux_stop_and_wait_all_lwps. This in turn
calls stop_callback on each thread followed by stop_wait_callback on
each thread.
We're now ready to make the mistake. In stop_callback we see that our
problem thread is not stopped, but is signalled, so it should stop
soon. As a result we don't send another SIGSTOP.
We then enter stop_wait_callback, eventually the problem thread stops
with SIGSTOP which we _incorrectly_ assume came from stop_callback,
and we discard.
Once stop_wait_callback has done its damage we return from
linux_stop_and_wait_all_lwps, finish in try_thread_db_load_1, and
eventually unwind back to the call to setup_inferior in
stop_all_threads. GDB now loops around, and performs another
target_wait to get the next event from the inferior.
The target_wait calls causes us to once again reach linux_nat_wait_1,
and we pass through some code that calls resume_stopped_resumed_lwps.
This allows GDB to resume threads that are physically stopped, but
which GDB doesn't see any good reason for the thread to remain
stopped. In our case, the problem thread which had its SIGSTOP
discarded is stopped, but doesn't have a stashed wait status to
report, and so GDB sets the thread going again.
We are now stuck waiting for an event on the problem thread that might
never arrive.
When considering how to write a test for this bug I struggled. The
issue was only spotted _randomly_ when a machine was heavily loaded
with many multi-threaded applications, and GDB was being attached (by
script) to all of these applications in parallel. In one reproducer I
required around 5 applications each of 5 threads per machine core in
order to reproduce the bug 2 out of 3 times.
What we really want to do though is simulate the kernel being slow to
report events through waitpid during the initial attach. The solution
I came up with was to write an LD_PRELOAD library that intercepts
(some) waitpid calls and rate limits them to one per-second. Any more
than that simply return 0 indicating there's no event available.
Obviously this can only be applied to waitpid calls that have the
WNOHANG flag set.
Unfortunately, once you ignore a waitpid call GDB can get a bit stuck.
Usually, once the kernel has made a child status available to waitpid
GDB will be sent a SIGCHLD signal. However, if the kernel makes 5
child statuses available but, due to the preload library we only
collect one of them, then the kernel will not send any further SIGCHLD
signals, and so, when GDB, thinking that the remaining statuses have
not yet arrived sits waiting for a SIGCHLD it will be disappointed.
The solution, implemented within the preload library, is that, when we
hold back a waitpid result from GDB we spawn a new thread. This
thread delays for a short period, and then sends GDB a SIGCHLD. This
causes GDB to retry the waitpid, at which point sufficient time has
passed and our library allows the waitpid call to complete.
gdb/ChangeLog:
* linux-nat.c (stop_wait_callback): Don't discard SIGSTOP if it
was requested by GDB.
gdb/testsuite/ChangeLog:
* gdb.threads/attach-slow-waitpid.c: New file.
* gdb.threads/attach-slow-waitpid.exp: New file.
* gdb.threads/slow-waitpid.c: New file.
Currently, the gdb.base/fork-running-state.exp testcase leaves a few
processes lingering until a 3 minutes alarm kills them:
pedro 28308 1 0 13:55 ? 00:00:00 /home/pedro/gdb/binutils-gdb/build/gdb/testsuite/outputs/gdb.base/fork-running-state/fork-running-state
pedro 28340 1 0 13:55 ? 00:00:00 /home/pedro/gdb/binutils-gdb/build/gdb/testsuite/outputs/gdb.base/fork-running-state/fork-running-state
pedro 28372 1 0 13:55 ? 00:00:00 /home/pedro/gdb/binutils-gdb/build/gdb/testsuite/outputs/gdb.base/fork-running-state/fork-running-state
pedro 28400 1 0 13:55 ? 00:00:00 /home/pedro/gdb/binutils-gdb/build/gdb/testsuite/outputs/gdb.base/fork-running-state/fork-running-state
pedro 28431 1 0 13:55 ? 00:00:00 /home/pedro/gdb/binutils-gdb/build/gdb/testsuite/outputs/gdb.base/fork-running-state/fork-running-state
pedro 28463 1 0 13:55 ? 00:00:00 /home/pedro/gdb/binutils-gdb/build/gdb/testsuite/outputs/gdb.base/fork-running-state/fork-running-state
Those processes used to kill themselves, but that was changed by
commit f50d8a2eaea0 ("Fix gdb.base/fork-running-state.exp race").
This commit restores the self-killing, but only in the cases gdb won't
try killing the processes, thus avoiding the old race.
(The restored code in fork_parent isn't exactly the same as it was.
In this version, we're exiting immediately when 'wait' returns
success, while in the old version we'd loop again and end up in the
perror call. The output from that perror call is not expected by the
"kill inferior" tests, and would result in a test FAIL.)
gdb/testsuite/ChangeLog:
2018-06-14 Pedro Alves <palves@redhat.com>
* gdb.base/fork-running-state.c: Include <errno.h>.
(exit_if_relative_exits): New.
(fork_child): If 'exit_if_relative_exits' is true, exit if the parent
exits.
(fork_parent): If 'exit_if_relative_exits' is true, exit if the
child exits.
Consider this testcase:
...
struct s {
int a;
int b;
};
struct s foo ()
{
struct s r;
r.a = 1;
r.b = 2;
return r;
}
int
main (void)
{
struct s v;
v = foo ();
return v.a + v.b;
}
...
When we compile it with -g, load the exec with gdb, and run till the end of foo,
we can print r:
...
(gdb) p r
$1 = {a = 1, b = 2}
...
and by setting pretty printing to on, we can get the fields of r printed each
on its own line:
...
(gdb) set print pretty
(gdb) p r
$2 = {
a = 1,
b = 2
}
...
However, when we finish foo, the printed function result value is not using
the pretty printing setting:
...
(gdb) finish
Run till exit from #0 foo () at test.c:11
0x00000000004004c1 in main () at test.c:18
18 v = foo ();
Value returned is $3 = {a = 1, b = 2}
...
This patch fixes that by using get_user_print_options instead of
get_no_prettyformat_print_options in print_return_value_1, which gives us:
...
(gdb) finish
Run till exit from #0 foo () at test.c:11
0x00000000004004c1 in main () at test.c:18
18 v = foo ();
Value returned is $2 = {
a = 1,
b = 2
}
...
Build & reg-tested on x86_64.
2018-06-14 Tom de Vries <tdevries@suse.de>
PR cli/22573
* infcmd.c (print_return_value_1): Use get_user_print_options instead of
get_no_prettyformat_print_options.
* gdb.base/finish-pretty.c: New test.
* gdb.base/finish-pretty.exp: New file.
