When encountering a return for which we have not seen a corresponding call, GDB
starts a new back trace from level -1, i.e. from the level of the first function
in the trace.
In the presence of trace gaps, this may cause some rather big jump.
(gdb) record function-call-history /c 192, +8
192 sbrk
193 brk
194 __x86.get_pc_thunk.bx
195 brk
196 __kernel_vsyscall
197 [disabled]
198 __kernel_vsyscall
199 brk
200 sbrk
This doesn't help to make things more clear. Let's remain on the same level
instead.
(gdb) record function-call-history /c 192, +8
192 sbrk
193 brk
194 __x86.get_pc_thunk.bx
195 brk
196 __kernel_vsyscall
197 [disabled]
198 __kernel_vsyscall
199 brk
200 sbrk
In this case it will look like we were able to connect the trace parts across
the disabled gap. We were not. More work is required to achieve this.
In the general case, the function-call history for the two trace parts won't
match. They may be off by a few levels or they may be entirely different. All
this patch does is to preserve the indentation level of the record
function-call-history command.
The disabled gap is caused by a sysenter not returning to the next instruction.
(gdb) record function-call-history /i 196, +1
196 __kernel_vsyscall inst 66515,66519
(gdb) record instruction-history 66515
66515 0xb7fdcbf8 <__kernel_vsyscall+0>: push %ecx
66516 0xb7fdcbf9 <__kernel_vsyscall+1>: push %edx
66517 0xb7fdcbfa <__kernel_vsyscall+2>: push %ebp
66518 0xb7fdcbfb <__kernel_vsyscall+3>: mov %esp,%ebp
66519 0xb7fdcbfd <__kernel_vsyscall+5>: sysenter
[disabled]
66520 0xb7fdcc08 <__kernel_vsyscall+16>: pop %ebp
66521 0xb7fdcc09 <__kernel_vsyscall+17>: pop %edx
66522 0xb7fdcc0a <__kernel_vsyscall+18>: pop %ecx
66523 0xb7fdcc0b <__kernel_vsyscall+19>: ret
66524 0xb7e8e09e <brk+30>: xchg %ecx,%ebx
(gdb) disassemble 0xb7fdcbf8, 0xb7fdcc0c
Dump of assembler code from 0xb7fdcbf8 to 0xb7fdcc0c:
0xb7fdcbf8 <__kernel_vsyscall+0>: push %ecx
0xb7fdcbf9 <__kernel_vsyscall+1>: push %edx
0xb7fdcbfa <__kernel_vsyscall+2>: push %ebp
0xb7fdcbfb <__kernel_vsyscall+3>: mov %esp,%ebp
0xb7fdcbfd <__kernel_vsyscall+5>: sysenter
0xb7fdcbff <__kernel_vsyscall+7>: nop
0xb7fdcc00 <__kernel_vsyscall+8>: nop
0xb7fdcc01 <__kernel_vsyscall+9>: nop
0xb7fdcc02 <__kernel_vsyscall+10>: nop
0xb7fdcc03 <__kernel_vsyscall+11>: nop
0xb7fdcc04 <__kernel_vsyscall+12>: nop
0xb7fdcc05 <__kernel_vsyscall+13>: nop
0xb7fdcc06 <__kernel_vsyscall+14>: int $0x80
0xb7fdcc08 <__kernel_vsyscall+16>: pop %ebp
0xb7fdcc09 <__kernel_vsyscall+17>: pop %edx
0xb7fdcc0a <__kernel_vsyscall+18>: pop %ecx
0xb7fdcc0b <__kernel_vsyscall+19>: ret
End of assembler dump.
I've seen this on 32-bit Fedora 23. I have not investigated what causes this
and whether we can avoid the gap in the first place. Let's first try to make
GDB handle such gaps more gracefully.
gdb/
* btrace.c (ftrace_new_return): Start from the previous function's level
if we can't find a matching call for a return.
An unconditional jump to the start of a function typically indicates a tail
call.
If we can't determine the start of the function at the destination address, we
used to treat it as a tail call, as well. This results in lots of tail calls
for code for which we don't have symbol information.
Restrict the heuristic to only consider jumps as tail calls that switch
functions in the case where we can't determine the start of a function. This
effectively disables tail call detection for code without symbol information.
gdb/
* btrace.c (ftrace_update_function): Update tail call heuristic.
GDB ignores trace gaps from decode errors or overflows at the beginning of the
trace. There isn't really a gap in the trace; the trace just starts a bit
later than expected.
In cases where there is no trace at all or where the trace is smaller than
expected, this may hide the reason for the missing trace.
Allow leading trace gaps. They will be shown as decode warnings and by the
record function-call-history command.
(gdb) info record
Active record target: record-btrace
Recording format: Intel Processor Trace.
Buffer size: 16kB.
warning: Decode error (-6) at instruction 0 (offset = 0x58, pc = 0x0): unexpected packet context.
warning: Decode error (-6) at instruction 0 (offset = 0xb0, pc = 0x0): unexpected packet context.
warning: Decode error (-6) at instruction 0 (offset = 0x168, pc = 0x0): unexpected packet context.
warning: Decode error (-6) at instruction 54205 (offset = 0xe08, pc = 0x0): unexpected packet context.
warning: Decode error (-6) at instruction 54205 (offset = 0xe60, pc = 0x0): unexpected packet context.
warning: Decode error (-6) at instruction 54205 (offset = 0xed8, pc = 0x0): unexpected packet context.
Recorded 91582 instructions in 1111 functions (6 gaps) for thread 1 (process 15710).
(gdb) record function-call-history /c 1
1 [decode error (-6): unexpected packet context]
2 [decode error (-6): unexpected packet context]
3 [decode error (-6): unexpected packet context]
4 _dl_addr
5 ??
6 _dl_addr
7 ??
8 ??
9 ??
10 ??
Leading trace gaps will not be shown by the record instruction-history command
without further changes.
gdb/
* btrace.c (btrace_compute_ftrace_bts, ftrace_add_pt): Allow leading gaps.
* record-btrace.c (record_btrace_single_step_forward)
(record_btrace_single_step_backward): Jump back to last instruction if
step ends at a gap.
(record_btrace_goto_begin): Skip gaps.
Trace gaps due to overflows or non-contiguous trace are ignored in the 'info
record' command. Fix that.
Also add a warning when decoding the trace and print the instruction number
preceding the trace gap in that warning message. It looks like this:
(gdb) info record
Active record target: record-btrace
Recording format: Intel Processor Trace.
Buffer size: 16kB.
warning: Decode error (-13) at instruction 101044 (offset = 0x29f0, pc = 0x7ffff728a642): no memory mapped at this address.
Recorded 101044 instructions in 2093 functions (1 gaps) for thread 1 (process 5360).
(gdb) record instruction-history 101044
101044 0x00007ffff728a640: pop %r13
[decode error (-13): no memory mapped at this address]
Remove the dead code that was supposed to print a gaps warning at the end of
trace decode. This isn't really needed since we now print a warning for each
gap.
gdb/
* btrace.c (ftrace_add_pt): Fix gap indication. Add warning for non-
contiguous trace and overflow. Rephrase trace decode warning and print
instruction number. Remove dead gaps warning.
(btrace_compute_ftrace_bts): Rephrase warnings and print instruction
number.
If the target can do software single step, it can do range
stepping.
gdb/gdbserver:
2016-10-27 Yao Qi <yao.qi@linaro.org>
* linux-low.c (linux_supports_agent): Return true if
can_software_single_step return true.
Nowadays, we select events to be reported to GDB in random, however
that is not enough when many GDBserver internal events (not reported
to GDB) are generated.
GDBserver pulls all events out of kernel via waitpid, and leave them
pending. When goes through threads which have pending events,
GDBserver uses find_inferior to find the first thread which has
pending event, and consumes it. Note that find_inferior always
iterate threads in a fixed order. If multiple threads keep hitting
GDBserver breakpoints, range stepping with single-step breakpoint for
example, threads in the head of the thread list are more likely to be
processed and threads in the tail are starved. This causes some timeout
fails in gdb.threads/non-stop-fair-events.exp when range stepping is
enabled on arm-linux.
This patch fixes this issue by randomly selecting pending events. It
adds a new function find_inferior_in_random, which iterates threads
which have pending events randomly.
gdb/gdbserver:
2016-10-27 Yao Qi <yao.qi@linaro.org>
* inferiors.c (find_inferior_in_random): New function.
* inferiors.h (find_inferior_in_random): Declare.
* linux-low.c (linux_wait_for_event_filtered): Call
find_inferior_in_random instead of find_inferior.
This patch removes single-step breakpoints if the event is only
GDBserver internal, IOW, isn't reported back to GDB.
gdb/gdbserver:
2016-10-27 Yao Qi <yao.qi@linaro.org>
* linux-low.c (linux_wait_1): If single-step breakpoints are
inserted, remove them.
In the ARC assembler, when a cpu type is specified using the .cpu
directive, we rely on the bfd list of arc machine types in order to
validate the cpu name passed in.
This validation is only used in order to check that the cpu type passed
to the .cpu directive matches any machine type selected earlier on the
command line. Once that initial check has passed a full check is
performed using the assemblers internal list of know cpu types.
The problem is that the assembler knows about more cpu types than bfd,
some cpu types known by the assembler are actually aliases for a base
cpu type plus a specific set of assembler extensions. One such example
is NPS400, though more could be added later.
This commit removes the need for the assembler to use the bfd list of
machine types for validation. Instead the error checking, to ensure
that any value passed to a '.cpu' directive matches any earlier command
line selection, is moved into the function arc_select_cpu.
I have taken the opportunity to bundle the 4 separate static globals
that describe the currently selected machine type into a single
structure (called selected_cpu).
gas/ChangeLog:
* config/tc-arc.c (arc_target): Delete.
(arc_target_name): Delete.
(arc_features): Delete.
(arc_mach_type): Delete.
(mach_type_specified_p): Delete.
(enum mach_selection_type): New enum.
(mach_selection_mode): New static global.
(selected_cpu): New static global.
(arc_eflag): Rename to ...
(arc_initial_eflag): ...this, and make const.
(arc_select_cpu): Update comment, new parameter, check how
previous machine type selection was made, and record this
selection. Use selected_cpu instead of old globals.
(arc_option): Remove use of arc_get_mach, instead use
arc_select_cpu to validate machine type selection. Use
selected_cpu over old globals.
(allocate_tok): Use selected_cpu over old globals.
(find_opcode_match): Likewise.
(assemble_tokens): Likewise.
(arc_cons_fix_new): Likewise.
(arc_extinsn): Likewise.
(arc_extcorereg): Likewise.
(md_begin): Update default machine type selection, use
selected_cpu over old globals.
(md_parse_option): Update machine type selection option handling,
use selected_cpu over old globals.
* testsuite/gas/arc/nps400-0.s: Add .cpu directive.
bfd/ChangeLog:
* cpu-arc.c (arc_get_mach): Delete.
The following testcases make GDB crash whenever an invalid sysroot is
provided, when GDB is unable to find a valid path to the symbol file:
gdb.base/catch-syscall.exp
gdb.base/execl-update-breakpoints.exp
gdb.base/foll-exec-mode.exp
gdb.base/foll-exec.exp
gdb.base/foll-vfork.exp
gdb.base/pie-execl.exp
gdb.multi/bkpt-multi-exec.exp
gdb.python/py-finish-breakpoint.exp
gdb.threads/execl.exp
gdb.threads/non-ldr-exc-1.exp
gdb.threads/non-ldr-exc-2.exp
gdb.threads/non-ldr-exc-3.exp
gdb.threads/non-ldr-exc-4.exp
gdb.threads/thread-execl.exp
The immediate cause of the segv is that follow_exec is passing a NULL
argument (the result of exec_file_find) to strlen.
However, the problem is deeper than that: follow_exec simply isn't
prepared for the case where sysroot translation fails to locate the
new executable. Actually all callers of exec_file_find have bugs due
to confusion between host and target pathnames. This commit attempts
to fix all that.
In terms of the testcases that were formerly segv'ing, GDB now prints
a warning but continues execution of the new program, so that the
tests now mostly FAIL instead. You could argue the FAILs are due to a
legitimate problem with the test environment setting up the sysroot
translation incorrectly.
A new representative test is added which exercises the ne wwarning
code path even with native testing.
Tested on x86_64 Fedora 23, native and gdbserver.
gdb/ChangeLog:
2016-10-25 Sandra Loosemore <sandra@codesourcery.com>
Luis Machado <lgustavo@codesourcery.com>
Pedro Alves <palves@redhat.com>
PR gdb/20569
* exceptions.c (exception_print_same): Moved here from exec.c.
* exceptions.h (exception_print_same): Declare.
* exec.h: Include "symfile-add-flags.h".
(try_open_exec_file): New declaration.
* exec.c (exception_print_same): Moved to exceptions.c.
(try_open_exec_file): New function.
(exec_file_locate_attach): Rename exec_file and full_exec_path
variables to avoid confusion between target and host pathnames.
Move pathname processing logic to exec_file_find. Do not return
early if pathname lookup fails; Call try_open_exec_file.
* infrun.c (follow_exec): Split and rename execd_pathname variable
to avoid confusion between target and host pathnames. Warn if
pathname lookup fails. Pass target pathname to
target_follow_exec, not hostpathname. Call try_open_exec_file.
* main.c (symbol_file_add_main_adapter): New function.
(captured_main_1): Use it.
* solib-svr4.c (open_symbol_file_object): Adjust to pass
symfile_add_flags to symbol_file_add_main.
* solib.c (exec_file_find): Incorporate fallback logic for relative
pathnames formerly in exec_file_locate_attach.
* symfile.c (symbol_file_add_main, symbol_file_add_main_1):
Replace 'from_tty' parameter with a symfile_add_file.
(symbol_file_command): Adjust to pass symfile_add_flags to
symbol_file_add_main.
* symfile.h (symbol_file_add_main): Replace 'from_tty' parameter
with a symfile_add_file.
gdb/testsuite/ChangeLog:
2016-10-25 Luis Machado <lgustavo@codesourcery.com>
* gdb.base/exec-invalid-sysroot.exp: New file.
This makes these flag types be "enum flag" types. The benefit is
making use of C++'s stronger typing -- mixing the flags types by
mistake errors at compile time.
This caught one old bug in symbol_file_add_main_1 already, fixed by
this patch as well:
@@ -1318,7 +1326,7 @@ symbol_file_add_main_1 (const char *args, int from_tty, int flags)
what is frameless. */
reinit_frame_cache ();
- if ((flags & SYMFILE_NO_READ) == 0)
+ if ((add_flags & SYMFILE_NO_READ) == 0)
set_initial_language ();
}
Above, "flags" are objfile flags, not symfile_add_flags. So that was
actually checking for "flag & OBJF_PSYMTABS_READ", which has the same
value as SYMFILE_NO_READ...
I moved the flags definitions to separate files to break circular
dependencies.
Built with --enable-targets=all and tested on x86-64 Fedora 23.
gdb/ChangeLog:
2016-10-26 Pedro Alves <palves@redhat.com>
* coffread.c (coff_symfile_read): Use symfile_add_flags.
* dbxread.c (dbx_symfile_read): Ditto.
* elfread.c (elf_symfile_read): Ditto.
* inferior.h: Include symfile-add-flags.h.
(struct inferior) <symfile_flags>: Now symfile_add_flags.
* machoread.c (macho_add_oso_symfile, macho_symfile_read_all_oso)
(macho_symfile_read, mipscoff_symfile_read): Use
symfile_add_flags.
* objfile-flags.h: New file.
* objfiles.c (allocate_objfile): Use objfile_flags.
* objfiles.h: Include objfile-flags.h.
(struct objfile) <flags>: Now an objfile_flags.
(OBJF_REORDERED, OBJF_SHARED, OBJF_READNOW, OBJF_USERLOADED)
(OBJF_PSYMTABS_READ, OBJF_MAINLINE, OBJF_NOT_FILENAME): Delete.
Converted to an enum-flags in objfile-flags.h.
(allocate_objfile): Use objfile_flags.
* python/py-objfile.c (objfpy_add_separate_debug_file): Remove
unnecessary local.
* solib.c (solib_read_symbols, solib_add)
(reload_shared_libraries_1): Use symfile_add_flags.
* solib.h: Include "symfile-add-flags.h".
(solib_read_symbols): Use symfile_add_flags.
* symfile-add-flags.h: New file.
* symfile-debug.c (debug_sym_read): Use symfile_add_flags.
* symfile-mem.c (symbol_file_add_from_memory): Use
symfile_add_flags.
* symfile.c (read_symbols, syms_from_objfile_1)
(syms_from_objfile, finish_new_objfile): Use symfile_add_flags.
(symbol_file_add_with_addrs): Use symfile_add_flags and
objfile_flags.
(symbol_file_add_separate): Use symfile_add_flags.
(symbol_file_add_from_bfd, symbol_file_add): Use symfile_add_flags
and objfile_flags.
(symbol_file_add_main_1): : Use objfile_flags. Fix add_flags vs
flags confusion.
(symbol_file_command): Use objfile_flags.
(add_symbol_file_command): Use symfile_add_flags and
objfile_flags.
(clear_symtab_users): Use symfile_add_flags.
* symfile.h: Include "symfile-add-flags.h" and "objfile-flags.h".
(struct sym_fns) <sym_read>: Use symfile_add_flags.
(clear_symtab_users): Use symfile_add_flags.
(enum symfile_add_flags): Delete, moved to symfile-add-flags.h and
converted to enum-flags.
(symbol_file_add, symbol_file_add_from_bfd)
(symbol_file_add_separate): Use symfile_add_flags.
* xcoffread.c (xcoff_initial_scan): Use symfile_add_flags.
Currently, with "maint set target-non-stop on", that is, when gdb
connects with the non-stop/asynchronous variant of the remote
protocol, even with "set non-stop off", GDB always sends one vCont
packet per thread resumed. This patch makes GDB aggregate and
coalesce vCont packets, so we send vCont packets like "vCont;s:p1.1;c"
in non-stop mode too.
Basically, this is done by:
- Adding a new target method target_commit_resume that is called
after calling target_resume one or more times. When resuming a
batch of threads, we'll only call target_commit_resume once after
calling target_resume for all threads.
- Making the remote target defer sending the actual vCont packet to
target_commit_resume.
Special care must be taken to avoid sending a vCont action with a
"wildcard" thread-id (all threads of process / all threads) when that
would resume threads/processes that should not be resumed. See
remote_commit_resume comments for details.
Unlike all-stop's remote_resume implementation, this handles the case
of too many actions resulting in a too-big vCont packet, by flushing
the vCont packet and starting a new one.
E.g., imagining that the "c" action in:
vCont;s:1;c
overflows the packet buffer, we split the actions like:
vCont;s:1
vCont;c
Tested on x86_64 Fedora 20, with and without "maint set
target-non-stop on".
Also tested with a hack that makes remote_commit_resume flush the vCont
packet after every action appended (which caught a few bugs).
gdb/ChangeLog:
2016-10-26 Pedro Alves <palves@redhat.com>
* inferior.h (ALL_NON_EXITED_INFERIORS): New macro.
* infrun.c (do_target_resume): Call target_commit_resume.
(proceed): Defer target_commit_resume while looping over threads,
resuming them. Call target_commit_resume at the end.
* record-btrace.c (record_btrace_commit_resume): New function.
(init_record_btrace_ops): Install it as to_commit_resume method.
* record-full.c (record_full_commit_resume): New function.
(record_full_wait_1): Call the beneath target's to_commit_resume
method.
(init_record_full_ops): Install record_full_commit_resume as
to_commit_resume method.
* remote.c (struct private_thread_info) <last_resume_step,
last_resume_sig, vcont_resumed>: New fields.
(remote_add_thread): Set the new thread's vcont_resumed flag.
(demand_private_info): Delete.
(get_private_info_thread, get_private_info_ptid): New functions.
(remote_update_thread_list): Adjust.
(process_initial_stop_replies): Clear the thread's vcont_resumed
flag.
(remote_resume): If connected in non-stop mode, record the resume
request and return early.
(struct private_inferior): New.
(struct vcont_builder): New.
(vcont_builder_restart, vcont_builder_flush)
(vcont_builder_push_action): New functions.
(MAX_ACTION_SIZE): New macro.
(remote_commit_resume): New function.
(thread_pending_fork_status, is_pending_fork_parent_thread): New
functions.
(check_pending_event_prevents_wildcard_vcont_callback)
(check_pending_events_prevent_wildcard_vcont): New functions.
(process_stop_reply): Adjust. Clear the thread's vcont_resumed
flag.
(init_remote_ops): Install remote_commit_resume.
* target-delegates.c: Regenerate.
* target.c (defer_target_commit_resume): New global.
(target_commit_resume, make_cleanup_defer_target_commit_resume):
New functions.
* target.h (struct target_ops) <to_commit_resume>: New field.
(target_resume): Update comments.
(target_commit_resume): New declaration.
Currently GDB never sends more than one action per vCont packet, when
connected in non-stop mode. A follow up patch will change that, and
it exposed a gdbserver problem with the vCont handling.
For example, this in non-stop mode:
=> vCont;s:p1.1;c
<= OK
Should be equivalent to:
=> vCont;s:p1.1
<= OK
=> vCont;c
<= OK
But gdbserver currently doesn't handle this. In the latter case,
"vCont;c" makes gdbserver clobber the previous step request. This
patch fixes that.
Note the server side must ignore resume actions for the thread that
has a pending %Stopped notification (and any other threads with events
pending), until GDB acks the notification with vStopped. Otherwise,
e.g., the following case is mishandled:
#1 => g (or any other packet)
#2 <= [registers]
#3 <= %Stopped T05 thread:p1.2
#4 => vCont s:p1.1;c
#5 <= OK
Above, the server must not resume thread p1.2 when it processes the
vCont. GDB can't know that p1.2 stopped until it acks the %Stopped
notification. (Otherwise it wouldn't send a default "c" action.)
(The vCont documentation already specifies this.)
Finally, special care must also be given to handling fork/vfork
events. A (v)fork event actually tells us that two processes stopped
-- the parent and the child. Until we follow the fork, we must not
resume the child. Therefore, if we have a pending fork follow, we
must not send a global wildcard resume action (vCont;c). We can still
send process-wide wildcards though.
(The comments above will be added as code comments to gdb in a follow
up patch.)
gdb/gdbserver/ChangeLog:
2016-10-26 Pedro Alves <palves@redhat.com>
* linux-low.c (handle_extended_wait): Link parent/child fork
threads.
(linux_wait_1): Unlink them.
(linux_set_resume_request): Ignore resume requests for
already-resumed and unhandled fork child threads.
* linux-low.h (struct lwp_info) <fork_relative>: New field.
* server.c (in_queued_stop_replies_ptid, in_queued_stop_replies):
New functions.
(handle_v_requests) <vCont>: Don't call require_running.
* server.h (in_queued_stop_replies): New declaration.
Specifically, what happens with multiple actions that could match a
thread, and what happens when we get a vCont action that matches a
thread that was already running. E.g., what does:
"vCont;s:2"
"vCont;s:1;c"
mean for thread 2.
(Thread 2 continues stepping.)
gdb/doc/ChangeLog:
2016-10-26 Pedro Alves <palves@redhat.com>
* gdb.texinfo (Packets): Clarify vCont packets with multiple
actions that match a thread, and what happens when an action
matches a thread that is already running.
(Where "exits" includes being killed or detached.)
Nothing is clearing inferior->priv currently. This is a problem if we
change the inferior's process_stratum targets in a single debug
session. This field is currently only used by darwin-nat.c, but a
follow up patch will make remote.c use it too. Without the fix,
remote.c might end up mistaking the priv object allocated by
darwin-nat.c with its own.
(Found by inspection.)
gdb/ChangeLog:
2016-10-26 Pedro Alves <palves@redhat.com>
* inferior.c (exit_inferior_1): Free 'priv'.
Just some refactoring / TLC. Mainly split the old c/s/C/S packet
handling to a separate function.
gdb/ChangeLog:
2016-10-26 Pedro Alves <palves@redhat.com>
* remote.c (remote_resume_with_hc): New function, factored out
from ...
(remote_resume): ... this. Always try vCont first.
(remote_vcont_resume): Rename to ...
(remote_resume_with_vcont): ... this. Bail out if execution
direction is reverse.
This reverts commit 95e61695c1. People
still want to use older versions of bison, apparently.
Revert 2016-10-06 Alan Modra <amodra@gmail.com>
* config/rl78-parse.y: Do use old %name-prefix syntax.
* config/rx-parse.y: Likewise.
Ref: https://sourceware.org/ml/gdb-patches/2016-10/msg00662.html
$ make WERROR_CFLAGS="-Wnon-virtual-dtor" dwarf2expr.o
...
In file included from .../src/gdb/dwarf2expr.c:28:0:
.../src/gdb/dwarf2expr.h:68:8: warning: ‘struct dwarf_expr_context’ has virtual functions and accessible non-virtual destructor [-Wnon-virtual-dtor]
struct dwarf_expr_context
^~~~~~~~~~~~~~~~~~
Happens to not be a problem in practice currently because concrete
subclasses are allocated on the stack. I.e., we don't ever delete
objects of types that derive from dwarf_expr_context through pointers
to dwarf_expr_context.
gdb/ChangeLog:
2016-10-25 Pedro Alves <palves@redhat.com>
* dwarf2expr.h (struct dwarf_expr_context) <~dwarf_expr_context>:
Make virtual.
gdb 7.12 doesn't compile as C++ (tried with g++ 4.9) on Solaris (tried
10 and 12, sparc and x86). The following patch (relative to the 7.12
release, though I expect most if not all issues to be present on trunk,
too) fixes this.
Only a few of the changes bear explanation:
* Initially, compilation failed whereever defs.h. was included:
In file included from /vol/src/gnu/gdb/gdb-7.12/gdb/gdb.c:19:0:
/vol/src/gnu/gdb/gdb-7.12/gdb/defs.h:630:33: error: 'double atof(const char*)' conflicts with a previous declaration
extern double atof (const char *); /* X3.159-1989 4.10.1.1 */
^
In file included from /usr/include/stdlib.h:17:0,
from build-gnulib/import/stdlib.h:36,
from /vol/src/gnu/gdb/gdb-7.12/gdb/common/common-defs.h:32,
from /vol/src/gnu/gdb/gdb-7.12/gdb/defs.h:28,
from /vol/src/gnu/gdb/gdb-7.12/gdb/gdb.c:19:
/vol/gcc-4.9/lib/gcc/i386-pc-solaris2.10/4.9.0/include-fixed/iso/stdlib_iso.h:119:15: note: previous declaration 'double std::atof(const char*)'
extern double atof(const char *);
^
This is due to this gem in gdb/defs.h which seems to have been present
like forever:
#ifndef atof
extern double atof (const char *); /* X3.159-1989 4.10.1.1 */
#endif
In the Solaris headers, the appropriate functions are in namespace std,
thus the conflict. I've wrapped the defs.h declaration in !__cplusplus
to avoid this; perhaps it can go completely instead.
* All the casts are necessary to appease g++ and should be pretty
obvious.
* The sol-thread.c changes are here to handle
/vol/src/gnu/gdb/gdb-7.12/gdb/sol-thread.c: In function 'void _initialize_sol_thread()':
/vol/src/gnu/gdb/gdb-7.12/gdb/sol-thread.c:1252:36: error: invalid conversion from 'void*' to 'void (*)(int)' [-fpermissive]
if (!(p_##X = dlsym (dlhandle, #X))) \
^
/vol/src/gnu/gdb/gdb-7.12/gdb/sol-thread.c:1255:3: note: in expansion of macro 'resolve'
resolve (td_log);
^
and are modeled after linux-thread-db.c (try_thread_db_load_1).
The patch allowed both 32 and 64-bit C++ builds on sparc-sun-solaris2.10
and i386-pc-solaris2.10 to complete. The resulting binary hasn't seen
more than a smoke test (invoke it on itself, b main, run) yet.
When investigating the failure to detect -static-libstdc++
support (more below), I found two more issues which only show up with
-Werror:
/vol/src/gnu/gdb/gdb/local/gdb/procfs.c: In function 'ssd* proc_get_LDT_entry(procinfo*, int)':
/vol/src/gnu/gdb/gdb/local/gdb/procfs.c:2487:19: error: variable 'old_chain' set but not used [-Werror=unused-but-set-variable]
struct cleanup *old_chain = NULL;
^
Unless I'm mistaken, you need to run do_cleanups on every return from
the function.
Afterwards, I ran a 32-bit compilation, which (after adding
--disable-largefile to avoid
In file included from /usr/include/sys/procfs.h:28:0,
from /vol/src/gnu/gdb/gdb/local/gdb/i386-sol2-nat.c:23:
/usr/include/sys/old_procfs.h:39:2: error: #error "Cannot use procfs in the large file compilation environment"
#error "Cannot use procfs in the large file compilation environment"
^
and two more instances) revealed
/vol/src/gnu/gdb/gdb/local/gdb/top.c: In function 'void gdb_safe_append_history()':
/vol/src/gnu/gdb/gdb/local/gdb/top.c:1170:59: error: format '%d' expects argument of type 'int', but argument 3 has type 'pid_t {aka long int}' [-Werror=format=]
= xstrprintf ("%s-gdb%d~", history_filename, getpid ());
^
Fixed by casting pid_t to long and printing it as such.
Revert commit f6abaf7a40 (gdb: no longer define
__STDC_CONSTANT_MACROS/__STDC_LIMIT_MACROS), with the tweak suggested
in that commit's log: the macros are now defined before any system
header is included.
This should fix AIX:
https://sourceware.org/ml/gdb-patches/2016-10/msg00682.html
gdb/ChangeLog:
2016-10-25 Pedro Alves <palves@redhat.com>
* common/common-defs.h (__STDC_CONSTANT_MACROS)
(__STDC_LIMIT_MACROS): Define.
Nowadays, if we build GDB with -fsanitize=address, we can get the asan
error below,
(gdb) quit
=================================================================
==9723==ERROR: AddressSanitizer: alloc-dealloc-mismatch (malloc vs operator delete) on 0x60200003bf70
#0 0x7f88f3837527 in operator delete(void*) (/usr/lib/x86_64-linux-gnu/libasan.so.1+0x55527)
#1 0xac8e13 in __gnu_cxx::new_allocator<void (*)()>::deallocate(void (**)(), unsigned long) /usr/include/c++/4.9/ext/new_allocator.h:110
#2 0xac8cc2 in __gnu_cxx::__alloc_traits<std::allocator<void (*)()> >::deallocate(std::allocator<void (*)()>&, void (**)(), unsigned long) /usr/include/c++/4.9/ext/alloc_traits.h:185
....
0x60200003bf70 is located 0 bytes inside of 8-byte region [0x60200003bf70,0x60200003bf78)
allocated by thread T0 here:
#0 0x7f88f38367ef in __interceptor_malloc (/usr/lib/x86_64-linux-gnu/libasan.so.1+0x547ef)
#1 0xbd2762 in operator new(unsigned long) /home/yao/SourceCode/gnu/gdb/git/gdb/common/new-op.c:42
#2 0xac8edc in __gnu_cxx::new_allocator<void (*)()>::allocate(unsigned long, void const*) /usr/include/c++/4.9/ext/new_allocator.h:104
#3 0xac8d81 in __gnu_cxx::__alloc_traits<std::allocator<void (*)()> >::allocate(std::allocator<void (*)()>&, unsigned long) /usr/include/c++/4.9/ext/alloc_traits.h:182
The reason for this is that we override operator new but don't override
operator delete. This patch does the override if the code is NOT
compiled with asan.
gdb:
2016-10-25 Yao Qi <yao.qi@linaro.org>
PR gdb/20716
* common/new-op.c (__has_feature): New macro.
Don't override operator new if asan is used.
This patch addresses a potential NULL pointer dereference when we try to
duplicate a string. The input pointer can be NULL and that may lead to
crashes. We simply add a check for that case.
gdb/ChangeLog:
2016-10-24 Luis Machado <lgustavo@codesourcery.com>
* exec.c (exec_file_locate_attach): Prevent NULL pointer dereference
when duplicating a string.
This patch fixes a gotcha when comparing exception's messages in
exception_print_same. It should've used the statically-allocated
string versions msg1 and msg2 instead.
As is, it could lead to crashes.
gdb/ChangeLog:
2016-10-24 Luis Machado <lgustavo@codesourcery.com>
* exec.c (exception_print_same): Fix string comparison to use
statically-allocated ones.
Since
[commit] [testsuite patch] Fix gcc_compiled for gcc 6 & 7
https://sourceware.org/ml/gdb-patches/2016-10/msg00620.html
there has started running again
Running gdb/testsuite/gdb.base/morestack.exp ...
+FAIL: gdb.base/morestack.exp: continue
+PASS: gdb.base/morestack.exp: up 3000
but as you can see it FAILs now - on Fedora 24 x86_64 (although for example it
still PASSes on CentOS-7.2 x86_64).
Program received signal SIGSEGV, Segmentation fault.
0x00007ffff787c7bb in malloc_consolidate (av=av@entry=0x7ffff7bbcb00 <main_arena>) at malloc.c:4181
4181 unlink(av, nextchunk, bck, fwd);
(gdb) bt
[...]
[...]
This apparently is due to - man gcc - -fsplit-stack:
When code compiled with -fsplit-stack calls code compiled without
-fsplit-stack, there may not be much stack space available for the
latter code to run. If compiling all code, including library code,
with -fsplit-stack is not an option, then the linker can fix up these
calls so that the code compiled without -fsplit-stack always has
a large stack. Support for this is implemented in the gold linker in
GNU binutils release 2.21 and later.
gdb/testsuite/ChangeLog
2016-10-24 Jan Kratochvil <jan.kratochvil@redhat.com>
* gdb.base/morestack.exp: Try to build it using -fuse-ld=gold first.
I got the following warning if I build GDBserver for aarch64_be-linux-gnu,
git/gdb/gdbserver/linux-aarch64-low.c:1539:39: error: invalid conversion from 'void*' to 'uint32_t* {aka unsigned int*}' [-fpermissive]
uint32_t *le_buf = xmalloc (byte_len);
^
The patch is to fix the warning.
gdb/gdbserver:
2016-10-24 Yao Qi <yao.qi@linaro.org>
PR server/20733
* linux-aarch64-low.c (append_insns): Cast the return value to
'uint32_t *'.
This patch changes some dwarf_expr_context to be pure virtual, as
mentioned during the discussion of an earlier patch in this series.
2016-10-21 Tom Tromey <tom@tromey.com>
* dwarf2expr.h (class dwarf_expr_context)
<get_frame_base, get_frame_cfa, get_tls_address, dwarf_call,
push_dwarf_block_entry_value, get_addr_index, get_object_address>:
Now pure virtual.
* dwarf2-frame.c (class dwarf_expr_executor)
<get_frame_base, get_frame_cfa, get_tls_address, dwarf_call,
push_dwarf_block_entry_value, get_addr_index, get_object_address>:
New methods.
<invalid>: New method.
This changes an "int" to a "bool" in the signature for
minimal_symbol_reader::record_full, and then fixes the callers.
2016-10-21 Tom Tromey <tom@tromey.com>
* minsyms.h (minimal_symbol_reader::record_full): "copy_name" now
a bool.
(record, record_with_info): Update.
* minsyms.c (record): Fix indentation.
(record_full): Fix indentation. Update for type change.
* elfread.c (record_minimal_symbol): "copy_name" now a bool.
(elf_symtab_read): "copy_names" now a bool.
(elf_rel_plt_read, elf_read_minimal_symbols): Update.
An earlier patch split captured_main into two parts. This patch
removes the last remaining cleanup from captured_main_1, and also
replaces a second hand-rolled VEC-like array with a std::vector.
2016-10-21 Tom Tromey <tom@tromey.com>
* main.c: Include <vector>.
(cmdarg_s): Remove typedef. Don't define VEC.
(captured_main_1): Use vector, not VEC. Remove cleanups.
This patch converts the function pointers in dwarf_expr_context_funcs
into methods on dwarf_expr_context, and then updates the various
implementations and callers to follow.
NB this patch uses "override" (which caught a couple of renaming bugs
during development) -- but this is C++11, so this patch at least has
to wait for Pedro's patch that adds the OVERRIDE macro.
After this patch it would be possible to do one more, that makes
various members of dwarf_expr_context "protected"; but I haven't done
this.
2016-10-21 Tom Tromey <tom@tromey.com>
* dwarf2loc.c (struct dwarf_expr_context_funcs): Don't declare.
(dwarf_expr_read_addr_from_reg, dwarf_expr_get_reg_value)
(dwarf_expr_read_mem, dwarf_expr_frame_base): Rename; turn into
methods.
(get_frame_pc_for_per_cu_dwarf_call): New function.
(dwarf_expr_frame_cfa, dwarf_expr_frame_pc)
(dwarf_expr_tls_address): Rename; turn into methods.
(per_cu_dwarf_call): Remove arguments. Use
get_frame_pc_for_per_cu_dwarf_call.
(dwarf_evaluate_loc_desc): New class.
(dwarf_expr_dwarf_call, dwarf_expr_context)
(dwarf_expr_push_dwarf_reg_entry_value)
(dwarf_expr_get_addr_index, dwarf_expr_get_obj_addr): Rename; turn
into methods.
(dwarf_expr_ctx_funcs): Remove.
(dwarf2_evaluate_loc_desc_full): Update.
(dwarf2_locexpr_baton_eval): Update.
(symbol_needs_eval_context): New class.
(symbol_needs_read_addr_from_reg, symbol_needs_get_reg_value)
(symbol_needs_read_mem, symbol_needs_frame_base)
(symbol_needs_frame_cfa, symbol_needs_tls_address)
(symbol_needs_dwarf_call, needs_dwarf_reg_entry_value): Rename;
turn into methods.
(needs_get_addr_index, needs_get_obj_addr): Remove; turn into
methods.
(symbol_needs_ctx_funcs): Remove.
(dwarf2_loc_desc_get_symbol_read_needs): Update.
* dwarf2expr.h (struct dwarf_expr_context_funcs): Remove; turn
contents into methods.
(struct dwarf_expr_context) <baton, funcs>: Remove.
<read_addr_from_reg, get_reg_value, read_mem, get_frame_base,
get_frame_cfa, get_frame_pc, get_tls_address, dwarf_call,
impl_get_base_type, push_dwarf_block_entry_value, get_addr_index,
get_object_address>: Declare new methods.
(ctx_no_get_frame_base, ctx_no_get_frame_cfa)
(ctx_no_get_frame_pc, ctx_no_get_tls_address, ctx_no_dwarf_call)
(ctx_no_get_base_type, ctx_no_push_dwarf_reg_entry_value)
(ctx_no_get_addr_index): Don't declare.
* dwarf2expr.c (get_base_type): Use impl_get_base_type.
(execute_stack_op): Update.
(ctx_no_get_frame_base, ctx_no_get_frame_cfa)
(ctx_no_get_frame_pc, ctx_no_get_tls_address, ctx_no_dwarf_call)
(ctx_no_get_base_type, ctx_no_push_dwarf_reg_entry_value)
(ctx_no_get_addr_index): Remove; now methods on
dwarf_expr_context.
* dwarf2-frame.c (read_addr_from_reg): Take a frame_info, not a
baton.
(class dwarf_expr_executor): New class.
(get_reg_value, read_mem): Rename, turn into methods.
(execute_stack_op): Use dwarf_expr_executor.
This converts various DWARF expr functions to be members on
dwarf_expr_context, then fixes up the various users. This results in
somewhat less wordy code and sets the stage for the next patch.
2016-10-21 Tom Tromey <tom@tromey.com>
* dwarf2loc.c (per_cu_dwarf_call)
(dwarf_expr_push_dwarf_reg_entry_value)
(dwarf2_evaluate_loc_desc_full, dwarf2_locexpr_baton_eval)
(needs_dwarf_reg_entry_value)
(dwarf2_loc_desc_get_symbol_read_needs): Update.
* dwarf2expr.h (dwarf_expr_context) <push_address, eval, fetch,
fetch_address, fetch_in_stack_memory, address_type, grow_stack,
push, stack_empty_p, add_piece, get_base_type, execute_stack_op,
pop>: New method declarations.
(dwarf_expr_push_address, dwarf_expr_eval, dwarf_expr_fetch)
(dwarf_expr_fetch_address, dwarf_expr_fetch_in_stack_memory):
Don't declare.
* dwarf2expr.c (address_type, grow_stack, push, push_address)
(pop, fetch, fetch_address, fetch_in_stack_memory)
(stack_empty_p, add_piece, eval, get_base_type)
(execute_stack_op): Rename. Turn into methods.
* dwarf2-frame.c (execute_stack_op): Update.
This is the first step in the conversion of dwarf_expr_ctx to a C++
class. This conversion is done in steps to make the patches, and the
reviews, a bit simpler. This patch changes dwarf_expr_ctx to be
stack-allocated and removes the associated cleanup.
2016-10-21 Tom Tromey <tom@tromey.com>
* dwarf2loc.c (dwarf2_evaluate_loc_desc_full): Stack-allocate
dwarf_expr_context. Remove cleanups.
(dwarf2_locexpr_baton_eval)
(dwarf2_loc_desc_get_symbol_read_needs): Likewise.
* dwarf2expr.h (dwarf_expr_context, ~dwarf_expr_context): Add
constructors and destructors.
(new_dwarf_expr_context, free_dwarf_expr_context)
(make_cleanup_free_dwarf_expr_context): Don't declare.
* dwarf2-frame.c (execute_stack_op): Stack-allocate
dwarf_expr_context. Remove cleanups.
(dwarf_expr_context): Rename from new_dwarf_expr_context. Turn
into constructor.
(free_dwarf_expr_context, free_dwarf_expr_context_cleanup):
Remove.
(~dwarf_expr_context): Rename from
make_cleanup_free_dwarf_expr_context. Turn into destructor.
This removes some cleanups and manual allocation handling in
dwarf2loc.c with std::vector. Note that this patch has a case where
the vector would normally fall into the "use gdb::unique_ptr"
guidelines -- but here because the vector is immediately initialized,
I moved the initialization into the constructor call, for further
code savings.
2016-10-21 Tom Tromey <tom@tromey.com>
* dwarf2loc.c: Include <vector>.
(read_pieced_value, write_pieced_value)
(dwarf2_compile_expr_to_ax): Use std::vector.
This removes make_cleanup_restore_current_uiout in favor of an
RAII-based class.
2016-10-21 Tom Tromey <tom@tromey.com>
* stack.c (print_stack_frame_to_uiout): Use scoped_restore.
* ui-out.c (make_cleanup_restore_current_uiout)
(restore_current_uiout_cleanup): Remove.
* infrun.c (print_stop_event): Use scoped_restore.
* ui-out.h (make_cleanup_restore_current_uiout): Don't declare.
This changes elf_read_minimal_symbols to use gdb::unique_ptr rather
than an explicit allocation. This removes a cleanup.
2016-10-21 Tom Tromey <tom@tromey.com>
* elfread.c (elf_read_minimal_symbols): Use gdb::unique_ptr.
This replaces a couple of uses of xmalloc with gdb::unique_ptr, also
removing a couple of cleanups.
2016-10-21 Tom Tromey <tom@tromey.com>
* cli/cli-dump.c (dump_memory_to_file): Use gdb::unique_ptr.
(restore_binary_file): Likewise.
This removes make_command_stats_cleanup in favor of an RAII class.
The patch is reasonably straightforward, but keeping the same
semantics without excessive reindentation required splitting
captured_main in two.
2016-10-21 Tom Tromey <tom@tromey.com>
* maint.h (scoped_command_stats): New class.
(make_command_stats_cleanup): Don't declare.
* maint.c (struct cmd_stats): Remove.
(~scoped_command_stats): Rename from report_command_stats. Now a
destructor.
(scoped_command_stats): Rename from make_command_stats_cleanup.
Now a constructor.
* main.c (captured_main_1): New function. Use
scoped_command_stats.
(captured_main): Call captured_main_1.
* event-top.c (command_handler): Use scoped_command_stats.
This patch removes a couple of cleanups from MI by using
gdb::unique_ptr.
2016-10-21 Tom Tromey <tom@tromey.com>
* mi/mi-main.c (mi_cmd_data_read_memory): Use gdb::unique_ptr.
Remove some cleanups.
This removes make_cleanup_restore_current_ui by converting the last
use. The last use was in a few functions used to iterate over all
UIs. This patch replaces these functions with a class, and arranges
for the class destructor to do the needed cleanup.
2016-10-21 Tom Tromey <tom@tromey.com>
* tui/tui-interp.c (tui_on_normal_stop, tui_on_signal_received)
(tui_on_end_stepping_range, tui_on_signal_exited, tui_on_exited)
(tui_on_no_history, tui_on_user_selected_context_changed):
Update.
* top.h (switch_thru_all_uis): New class.
(SWITCH_THRU_ALL_UIS): Rewrite.
(make_cleanup_restore_current_ui, switch_thru_all_uis_init)
(switch_thru_all_uis_cond, switch_thru_all_uis_next): Don't
declare.
* mi/mi-interp.c (mi_new_thread, mi_thread_exit)
(mi_record_changed, mi_inferior_added, mi_inferior_appeared)
(mi_inferior_exit, mi_inferior_removed, mi_on_signal_received)
(mi_on_end_stepping_range, mi_on_signal_exited, mi_on_exited)
(mi_on_no_history, mi_on_normal_stop, mi_traceframe_changed)
(mi_tsv_created, mi_tsv_deleted, mi_tsv_modified)
(mi_breakpoint_created, mi_breakpoint_deleted)
(mi_breakpoint_modified, mi_output_running_pid, mi_on_resume)
(mi_solib_loaded, mi_solib_unloaded, mi_command_param_changed)
(mi_memory_changed, mi_user_selected_context_changed): Update.
* infrun.c (all_uis_check_sync_execution_done)
(all_uis_on_sync_execution_starting, normal_stop): Update.
* event-top.c (restore_ui_cleanup)
(make_cleanup_restore_current_ui, switch_thru_all_uis_init)
(switch_thru_all_uis_cond, switch_thru_all_uis_next): Remove.
* cli/cli-interp.c (cli_on_normal_stop, cli_on_signal_received)
(cli_on_end_stepping_range, cli_on_signal_exited, cli_on_exited)
(cli_on_no_history, cli_on_user_selected_context_changed):
Update.
* breakpoint.c (watchpoint_check): Update.
This changes minimal_symbol_reader to require the objfile to be
passed to the constructor. The class now records the objfile and
automatically uses it later in "install".
This is a minor cleanup that will come in useful in the next patch.
It is separate from the first patch to keep that one a bit simpler to
understand.
2016-10-21 Tom Tromey <tom@tromey.com>
* xcoffread.c (xcoff_initial_scan): Update.
* mipsread.c (mipscoff_symfile_read): Update.
* minsyms.c (minimal_symbol_reader): Add obj argument.
Initialize member.
(install): Remove objfile argument. Update.
* mdebugread.c (elfmdebug_build_psymtabs): Update.
* machoread.c (macho_symfile_read): Update.
* elfread.c (elf_read_minimal_symbols): Update.
* dbxread.c (dbx_symfile_read): Update.
* coffread.c (coff_symfile_read): Update.
* minsyms.h (minimal_symbol_reader): Add m_objfile member.
(constructor): Add objfile argument.
(minimal_symbol_reader::install): Remove objfile argument.
This patch introduced minimal_symbol_reader, a RAII-based class which
replaces the current make_cleanup_discard_minimal_symbols.
2016-10-21 Tom Tromey <tom@tromey.com>
* xcoffread.c (xcoff_initial_scan): Use
minimal_symbol_reader.
* mipsread.c (mipscoff_symfile_read): Use
minimal_symbol_reader.
* minsyms.h (minimal_symbol_reader): New class.
(init_minimal_symbol_collection)
(make_cleanup_discard_minimal_symbols, install_minimal_symbols):
Don't declare.
* minsyms.c (minimal_symbol_reader): Renamed from
init_minimal_symbol_collection, turned into constructor.
(~minimal_symbol_reader): Renamed from
do_discard_minimal_symbols_cleanup, turned into destructor.
(make_cleanup_discard_minimal_symbols): Remove.
(minimal_symbol_reader::install): Rename form
install_minimal_symbols.
* mdebugread.c (elfmdebug_build_psymtabs): Use
minimal_symbol_reader.
* machoread.c (macho_symfile_read): Use
minimal_symbol_reader.
* elfread.c (elf_read_minimal_symbols): Use
minimal_symbol_reader.
* dbxread.c (dbx_symfile_read): Use minimal_symbol_reader.
* coffread.c (coff_symfile_read): Use
minimal_symbol_reader.
Markus Metzger
GDB Administrator
Yao Qi
Andrew Burgess
Sandra Loosemore
Pedro Alves
Alan Modra
Rainer Orth
Maciej W. Rozycki
Luis Machado
Jan Kratochvil
Tom Tromey