Pedro Alves [Fri, 10 May 2013 13:51:49 +0000 (13:51 +0000)]
PR remote/15455 - QTro remote packet broken
In the function remote_trace_set_readonly_regions in gdb/remote.c, the
local variable 'offset' does not account for "QTro" at the start of
the packet with the result that if there are any read-only regions,
the packet is sent -- but without the "QTro" -- causing the remote
stub to report that the packet is unsupported:
We don't see the problem when testing with GDBserver, as that supports
qXfer:trace-frame-info:read, meaning GDBserver never needs to read
from the read-only sections directly itself. This commit adds a test
that explicitly disables qXfer:trace-frame-info:read.
gdb/
2013-05-10 David Taylor <dtaylor@emc.com>
PR remote/15455
* remote.c (remote_trace_set_readonly_regions): Do not overwrite
"QTro" at start of packet.
gdb/testsuite/
2013-05-10 Pedro Alves <palves@redhat.com>
PR remote/15455
* gdb.trace/qtro.c: New file.
* gdb.trace/qtro.exp: New file.
Joel Brobecker [Fri, 10 May 2013 12:30:05 +0000 (12:30 +0000)]
move sparc-sol-thread.c back into sol-thread.c.
The routines in sparc-sol-thread used to be SPARC-specific (and
documented as such in the ptrace man page), and therefore hosting them
in a sparc-specific file made sense. However, newer versions of
Solaris now use those callbacks (Solaris 10 Update 10, apparently),
and thus the note about these callbacks being specific to SPARC
was removed.
So this patch deletes sparc-sol-thread.c and moves the code back
inside sol-thread.c.
PR breakpoints/15413:
* breakpoint.c (condition_completer): Simplify the code to
disconsider multiple locations of breakpoints when completing the
"condition" command.
PR breakpoints/15413:
* gdb.base/pending.exp: Add test for completion of the "condition"
command for pending breakpoints.
* gdb.linespec/linespec.ex: Add test for completion of the
"condition" command when dealing with multiple locations.
Andrew Haley found a bug on GDB running on ARM when using
--enable-64-bit-bfd. Basically the issue happens when dealing with "bl"
instructions: GDB does branch destination calculation and (wrongly)
sign-extends the PC. Here is a piece of his original message explaining
the problem:
> next_pc = arm_get_next_pc (frame, get_frame_pc (frame));
>
> /* The Linux kernel offers some user-mode helpers in a high page. We can
> not read this page (as of 2.6.23), and even if we could then we couldn't
> set breakpoints in it, and even if we could then the atomic operations
> would fail when interrupted. They are all called as functions and return
> to the address in LR, so step to there instead. */
> if (next_pc > 0xffff0000)
> next_pc = get_frame_register_unsigned (frame, ARM_LR_REGNUM);
>
> arm_insert_single_step_breakpoint (gdbarch, aspace, next_pc);
>
> Unfortunately, branch destination addresses are SIGN EXTENDED to 64
> bits. So,
>
> (top-gdb) p/x next_pc
> $14 = 0xffffffffb6df2864
>
> Which triggers the next_pc = get_frame_register_unsigned(), and we
> cannot step into any branches because the destination PC is wrong.
Anyway, the fix is simple and Andrew himself provided it for us. It
took a while for me to figure out how to trigger the bug (in order to
write a testcase for it), but I finally made it.
The attached patch fixes the problem (by casting to `unsigned long'
instead of just `long'), and also includes a testcase to reproduce the
issue.
gdb/ChangeLog:
2013-04-25 Andrew Haley <aph@redhat.com>
* arm-tdep.c (BranchDest): Cast result as "unsigned long",
instead of "long".
* breakpoint.c (dprintf_print_recreate): New.
(save_breakpoints): Let it not save dprintf commands.
(initialize_breakpoint_ops): Set dprintf_print_recreate.
Pedro Alves [Fri, 19 Apr 2013 14:15:51 +0000 (14:15 +0000)]
Fix the x87 FP register printout when issuing the “info float” command.
Consider the following simple program:
.globl _start
.text
_start:
fldt val
.data
val: .byte 0x00,0x00,0x45,0x07,0x11,0x19,0x22,0xe9,0xfe,0xbf
With current GDB on x86-64 GNU/Linux hosts, after the moment the fldt
command has been executed the register st(0) looks like this,
according to the “info regs” output (TOP=7):
which is clearly wrong (just count its length). The problem is due to
the printf statement (see patch) printing a promoted integer value of
a char argument "raw[i]", and, since char is signed on x86-64
GNU/Linux, the erroneous “ffffff” are printed for the first three
bytes which turn out to be "negative". The fix is to use gdb_byte
instead which is unsigned (and is the type of value_contents(), the
type to be used for raw target bytes anyway). After the fix the value
will be printed correctly:
* main.c (get_init_files): Use filename_ncmp instead of strncmp.
Use IS_DIR_SEPARATOR instead of looking for a character inside
SLASH_STRING. Include filenames.h.
(captured_main) [__MINGW32__]: Make argv[0] absolute, so that
relocate_gdb_directory works when passed gdb_program_name.
Jan Kratochvil [Fri, 5 Apr 2013 19:17:27 +0000 (19:17 +0000)]
gdb/
Fix compatibility with Linux kernel 3.8.3.
* linux-tdep.c (linux_find_memory_regions_full): Move variable number
to more inner block. Remove parsing of NUMBER from outer block.
Parse NUMBER only if KEYWORD has been identified.
Pedro Alves [Thu, 4 Apr 2013 19:15:55 +0000 (19:15 +0000)]
tracepoint->step_count fixes
If a tracepoint's actions list includes a while-stepping action, and
then the actions are changed to a list without any while-stepping
action, the tracepoint's step_count will be left with a stale value.
For example:
(gdb) trace subr
Tracepoint 1 at 0x4004d9: file ../../../src/gdb/testsuite//actions-changed.c, line 31.
(gdb) actions
Enter actions for tracepoint 1, one per line.
End with a line saying just "end".
>collect $reg
>end
(gdb) set debug remote 1
(gdb) tstart
Sending packet: $QTinit#59...Packet received: OK
Sending packet: $QTDP:1:00000000004004d9:E:0:0-#a3...Packet received: OK
Sending packet: $QTDP:-1:00000000004004d9:R03FFFFFFFFFFFFFFFFFF#2b...Packet received: OK
(gdb) tstop
Sending packet: $QTStop#4b...Packet received: OK
Sending packet: $QTNotes:#e8...Packet received: OK
(gdb) actions
Enter actions for tracepoint 1, one per line.
End with a line saying just "end".
>collect $reg
>while-stepping 1
>collect $reg
>end
>end
(gdb) tstart
Sending packet: $QTinit#59...Packet received: OK
Sending packet: $QTDP:1:00000000004004d9:E:1:0-#a4...Packet received: OK
Sending packet: $QTDP:-1:00000000004004d9:R03FFFFFFFFFFFFFFFFFF-#58...Packet received: OK
Sending packet: $QTDP:-1:00000000004004d9:SR03FFFFFFFFFFFFFFFFFF#7e...Packet received: OK
(gdb) tstop
Sending packet: $QTStop#4b...Packet received: OK
Sending packet: $QTNotes:#e8...Packet received: OK
(gdb) actions
Enter actions for tracepoint 1, one per line.
End with a line saying just "end".
>collect $regs
>end
(gdb) tstart
Sending packet: $QTinit#59...Packet received: OK
Sending packet: $QTDP:1:00000000004004d9:E:1:0-#a4...Packet received: OK
Sending packet: $QTDP:-1:00000000004004d9:R03FFFFFFFFFFFFFFFFFF#2b...Packet received: OK
A related issue is that the "commands" command actually supports
setting commands to a range of breakpoints/tracepoints at once. But,
hacking "maint info breakpoints" to print t->step_count, reveals:
(gdb) trace main
Tracepoint 5 at 0x45a2ab: file ../../src/gdb/gdb.c, line 29.
(gdb) trace main
Note: breakpoint 5 also set at pc 0x45a2ab.
Tracepoint 6 at 0x45a2ab: file ../../src/gdb/gdb.c, line 29.
(gdb) commands 5-6
Type commands for breakpoint(s) 5-6, one per line.
End with a line saying just "end".
> while-stepping 5
>end
> end
(gdb) maint info breakpoints 5
Num Type Disp Enb Address What
5 tracepoint keep y 0x000000000045a2ab in main at ../../src/gdb/gdb.c:29 inf 1
step_count=5
^^^^^^^^^^^^
while-stepping 5
end
not installed on target
(gdb) maint info breakpoints 6
Num Type Disp Enb Address What
6 tracepoint keep y 0x000000000045a2ab in main at ../../src/gdb/gdb.c:29 inf 1
step_count=0
^^^^^^^^^^^^
while-stepping 5
end
not installed on target
(gdb)
that tracepoint 6 doesn't end up with the correct step_count.
validate_actionline is never called for tracepoints other than the
first (the copy_command_lines path). Right below, we have:
/* If a breakpoint was on the list more than once, we don't need to
do anything. */
if (b->commands != info->cmd)
{
validate_commands_for_breakpoint (b, info->cmd->commands);
incref_counted_command_line (info->cmd);
decref_counted_command_line (&b->commands);
b->commands = info->cmd;
observer_notify_breakpoint_modified (b);
}
And validate_commands_for_breakpoint looks like the right place to put
a call; if we reset step_count there too, we have a nice central fix
for the first issue as well, because trace_actions_command calls
breakpoint_set_commands that also calls
validate_commands_for_breakpoint.
We end up calling validate_actionline twice for the first tracepoint,
since read_command_lines calls it too, through
check_tracepoint_command, but that should be harmless.
2013-04-04 Pedro Alves <palves@redhat.com>
Hui Zhu <hui@codesourcery.com>
* breakpoint.c (validate_commands_for_breakpoint): If validating a
tracepoint, reset its STEP_COUNT and call validate_actionline.
2013-04-04 Stan Shebs <stan@codesourcery.com>
Pedro Alves <palves@redhat.com>
* gdb.trace/Makefile.in (PROGS): Add actions-changed.
* gdb.trace/actions-changed.c: New file.
* gdb.trace/actions-changed.exp: New file.
* lib/trace-support.exp (gdb_trace_setactions): Rename to ...
(gdb_trace_setactions_command): ... this. Add "actions_command"
parameter, and handle it.
(gdb_trace_setactions, gdb_trace_setcommands): New procedures.
Pedro Alves [Tue, 2 Apr 2013 18:01:06 +0000 (18:01 +0000)]
NEWS/manual: Missing documentation for new commands in 7.6.
I hacked "apropos" to dump the whole set of commands (just make it
accept the entry string as regex), and then diffed the output of 7.5
vs 7.6, --enable-targets=all builds. That allowed then checking
whether some commands had not been mentioned in NEWS or the manual.
These are what I found missing. We've been a bit negligent in
requiring documentation bits for debug commands.
Pedro Alves [Tue, 2 Apr 2013 13:54:21 +0000 (13:54 +0000)]
unpush the remote target if serial_write fails.
PR gdb/15275 notes that when debugging with a remote connection over a
serial link and the link is disconnected, say by disconnecting USB
serial port, the GDB quit command no longer works:
This is not reproducible with TCP connections, as with that, sending
doesn't error out, but instead the error is detected on the subsequent
readchar. When that read fails, we unpush the remote target, and
throw TARGET_CLOSE_ERROR. To address PR gdb/15275, instead of
catching the error in remote_get_trace_status as presently done (which
leaves this same issue latent for another packet to trip on), or of
making ser-unix.c fake success too on failed writes, so we'd get to
readchar detecting the error on serial ports too, better let the error
propagate out of serial_write, and catch it at the remote.c level,
throwing away the target if writing fails too, instead of delaying
that until the next read.
gdb/
2013-04-02 Pedro Alves <palves@redhat.com>
PR gdb/15275
* remote.c (send_interrupt_sequence): Use remote_serial_write.
(remote_serial_write): New function.
(putpkt_binary, getpkt_or_notif_sane_1): Use remote_serial_write.
Mike Frysinger [Sun, 31 Mar 2013 00:43:14 +0000 (00:43 +0000)]
sim: frv/m32r: back out hard failure when dv-sockser is not available
These sims have optional support for the dv-sockser model, so do not make
them hard failures. The Makefile made it seem like they didn't actually
support things dynamically, but a further code dive into the source and
the Makefile shows that things work out.
Pedro Alves [Thu, 28 Mar 2013 11:50:00 +0000 (11:50 +0000)]
Fix PR gdb/15294: list with unlimited listsize broken
Currently, "set listsize -1" is supposed to mean "unlimited" source
lines, but, alas, it doesn't actually work:
(gdb) set listsize -1
(gdb) show listsize
Number of source lines gdb will list by default is unlimited.
(gdb) list 1
(gdb) list 1
(gdb) list 1
(gdb) set listsize 10
(gdb) list 1
1 /* Main function for CLI gdb.
2 Copyright (C) 2002-2013 Free Software Foundation, Inc.
3
4 This file is part of GDB.
5
6 This program is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 3 of the License, or
9 (at your option) any later version.
10
was applied, the "set listsize" command was a var_integer command, and
"unlimited" was set with 0. Internally, var_integer maps 0 to INT_MAX
case var_integer:
{
...
if (val == 0 && c->var_type == var_integer)
val = INT_MAX;
The change in that patch to zuinteger_unlimited command, meant that -1
is left as -1 in the command's control variable (lines_to_list), and
the code in source.c isn't expecting that -- it only expects positive
numbers.
I previously suggested fixing the code and keeping the new behavior,
but I found that "set listsize 0" is currently used in the wild, and
we do have a bunch of other commands where "0" means unlimited, so I'm
thinking that changing this command alone in isolation is not a good
idea.
So I now strongly prefer reverting back the behavior in 7.6 to the
same behavior the command has had since 2006 (0==unlimited, -1=error).
Before that, set listsize -1 would be accepted as unlimited as well.
After 7.6 is out, in mainline, we can get back to reconsidering
changing this command's behavior, if there's a real need for being
able to suppress output. For now, let's play it safe.
The "list line 1 with unlimited listsize" test in list.exp was
originally written years and years ago expecting 0 to mean "no
output", but GDB never actually worked that way, even when the tests
were written, so the tests had been xfailed then. This patch now
adjusts the test to the new behavior, so that the test actually
passes, and the xfail is removed.
gdb/
2013-03-28 Pedro Alves <palves@redhat.com>
PR gdb/15294
* source.c (_initialize_source): Change back "set listsize" to an
integer command.
gdb/testsuite/
2013-03-28 Pedro Alves <palves@redhat.com>
PR gdb/15294
* gdb.base/list.exp (set_listsize): Adjust to accept $arg == 0 to
mean unlimited instead of $arg < 0.
(test_listsize): Remove "listsize of 0 suppresses output" test.
Test that "set listsize 0" ends up with an unlimited listsize.
gdb/doc/
2013-03-28 Pedro Alves <palves@redhat.com>
PR gdb/15294
* gdb.texinfo (List) <set listsize>: Adjust to document that
listsize 0 means no limit, and remove mention of -1.
Pedro Alves [Thu, 28 Mar 2013 11:46:26 +0000 (11:46 +0000)]
list.exp: Avoid hardcoding line numbers.
The previous patch actually wasn't the first time I had to update line
numbers in this file.
This avoids hard coding line numbers, hopefully making the next time a
little easier.
Tested on x86_64 Fedora 17.
gdb/testsuite/
2013-03-28 Pedro Alves <palves@redhat.com>
* gdb.base/list.exp (last_line): New global.
(last_line_re): New global.
(test_listsize, test_list_function, test_list_forward)
(test_repeat_list_command, test_list_range)
(test_list_filename_and_function): Use them.
* gdb.base/list0.c: Comment the last line of the file with "last
line".
Pedro Alves [Thu, 28 Mar 2013 11:45:28 +0000 (11:45 +0000)]
list.exp: Adjust "set listsize -1" to current test source's real line count.
The "set listsize -1" test in list.exp can't work correct anymore
nowadays, because the test's source files grew over time, and this
particular test was never updated.
This fixes it in the obvious way.
gdb/testsuite/
2013-03-28 Pedro Alves <palves@redhat.com>
* gdb.base/list.exp (test_listsize): Adjust test to make sure we
list the whole file.
Before the changes starting at
http://sourceware.org/ml/gdb-patches/2012-08/msg00020.html, the 'set
listsize' command only accepted "0" as special value, meaning
"unlimited". The testsuite actually tried "set listsize -1" and
expected that to mean unlimited too.
If you tried testing list.exp at the time of that patch above,
you'd get:
(gdb) PASS: gdb.base/list.exp: list line 10 with listsize 100
set listsize 0
(gdb) PASS: gdb.base/list.exp: setting listsize to 0 #6
show listsize
Number of source lines gdb will list by default is unlimited.
(gdb) PASS: gdb.base/list.exp: show listsize unlimited #6
list 1
1 #include "list0.h"
2
...
42 /* Not used for anything */
43 }
(gdb) PASS: gdb.base/list.exp: listsize of 0 suppresses output
set listsize -1
integer 4294967295 out of range
(gdb) PASS: gdb.base/list.exp: setting listsize to -1 #7
show listsize
Number of source lines gdb will list by default is unlimited.
(gdb) PASS: gdb.base/list.exp: show listsize unlimited #7
list 1
1 #include "list0.h"
Notice that "set listsize -1" actually failed with "integer 4294967295
out of range", but we issued a PASS anyway.
(and notice how the "listsize of 0 suppresses output" test passes bogusly too.)
This patch fixes that testsuite problem in the obvious way.
gdb/testsuite/
2013-03-28 Pedro Alves <palves@redhat.com>
* gdb.base/list.exp (set_listsize): Use gdb_test_no_output for
"set listsize".
Pedro Alves [Wed, 27 Mar 2013 12:22:24 +0000 (12:22 +0000)]
Forbid set history size (INT_MAX..UINT_MAX)
The whole readline interface is signed, and works with the 0..INT_MAX
range.
We don't allow setting the size to UINT_MAX directly. The documented
user visible interface is "use 0 for unlimited". The UINT_MAX
representation is an implementation detail we could change, e.g., by
keeping a separate flag for "unlimited", which is actually what the
readline interface does (stifled vs non stifled). Generically
speaking, exposing this detail to clients of the interface may make
our lives complicated when we find the need to extend the range of
some command in the future, and it's better if users
(frontends/scripts) aren't relying on anything but what we tell them
to use for "unlimited". Making values other than 0 error out is the
way to prevent users from using those ranges inappropriately. Quite
related, note:
(gdb) set history size 0xffffffff
integer 4294967295 out of range
But,
(gdb) set history size 0xfffffffe
(gdb) show history size
The size of the command history is unlimited.
(gdb) set history size 0x100000000
integer 4294967296 out of range
If values over INT_MAX are accepted as unlimited, then there's no good
argument for only accepting [INT_MAX..UINT_MAX) as valid "unlimited"
magic numbers, while not accepting [UINT_MAX..inf).
Making the setting's control variable of different type (unsigned int)
of the rest of the related code (int) adds the need to recall that one
variable among all these is unsigned, and that one need to think about
whether these comparisons are signed or unsigned, along with the
promotion/conversion rules. Since this is an easy to forget detail,
this patch renames the variable to at least make it more obvious that
this variable is not one of GNU history's public int variables, which
are all signed. We don't actually need the only code that presently
is affected by this, though, the code that is computing the current
history's length. We can just use GNU history's history_length
instead:
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Variable: int history_length
The number of entries currently stored in the history list.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/* Return the history entry which is logically at OFFSET in the history array.
OFFSET is relative to history_base. */
HIST_ENTRY *
history_get (offset)
int offset;
{
int local_index;
At the time this code was added (gdb 4.13 ~1994), 'history_length' was
extern, but not documented in readline's GNU history documents, so I
guess it wasn't considered public then and the loop was the
workaround.
One of the warts of GDB choosing 0 to mean unlimited is that "set
history size 0" behaves differently from 'HISTSIZE=0 gdb'. The latter
leaves GDB with no history, while the former means "unlimited"...
$ HISTSIZE=0 ./gdb
...
(gdb) show history size
The size of the command history is 0.
We shouldn't really change what HISTSIZE=0 means, as bash, etc. also
handle 0 as real zero, and zero it's what really makes sense.
gdb/
2013-03-27 Pedro Alves <palves@redhat.com>
* top.c (history_size): Rename to ...
(history_size_setshow_var): ... this. Add comment.
(show_commands): Use readline's 'history_length' instead of
computing the history length by calling history_get in a loop.
(set_history_size_command): Error out for sizes over INT_MAX.
Restore previous history size on invalid size.
(init_history): If HISTSIZE is negative, leave the history size as
zero. Add comments.
(init_main): Adjust.
Pedro Alves [Wed, 27 Mar 2013 09:57:50 +0000 (09:57 +0000)]
Make "set/show debug coff_pe_read" a zuinteger instead of uinteger.
Being a uinteger means you can't disable debug output after enabling it...
(gdb) show debug coff_pe_read
Coff PE read debugging is 0.
(gdb) set debug coff_pe_read 0
(gdb) show debug coff_pe_read
Coff PE read debugging is unlimited.
(gdb)
gdb/
2013-03-27 Pedro Alves <palves@redhat.com>
* coff-pe-read.c (_initialize_coff_pe_read): Make the command
zuinteger instead of uinteger.
Mike Frysinger [Tue, 26 Mar 2013 18:01:53 +0000 (18:01 +0000)]
sim: rewrite SIM_AC_OPTION_HARDWARE a bit to simplify things
There's no need to put the majority of the logic into the 3rd arg of the
AC_ARG_ENABLE. Coupled with the lack of indentation, it makes it hard to
follow, error prone to update, and duplicates code (with the 4th arg).
So pull the logic out of the 3rd arg and outside of the AC_ARG_ENABLE
macro. This allows us to gut the 4th arg entirely, merge with the code
that followed the macro, and fix bugs related to the new dv-sockser in
the process.
Hopefully building the various sims with the default sim-hardware
settings, as well as with explicit --{dis,en}able-sim-hardware flags,
should all just work now.
projects / thirdparty / binutils-gdb.git / log
