We have observed that GDB would randomly trip the following
assertion failure when debugging on Windows. When allowing
the program to run until the inferior exits, we occasionally see:
(gdb) cont
Continuing.
[Thread 48192.0xd100 exited with code 1]
[Thread 48192.0x10ad8 exited with code 1]
[Thread 48192.0x36e28 exited with code 0]
[Thread 48192.0x52be4 exited with code 0]
[Thread 48192.0x5aa40 exited with code 0]
../../src/gdb/thread.c:453: internal-error: void delete_thread_1(thread_inf
o*, bool): Assertion `thr != nullptr' failed.
Running the same scenario with some additional traces enabled...
(gdb) set verbose
(gdb) set debugevents
... allows us to understand what the issue is. To understand, we need
to first look at the events received when starting the program, and
in particular which threads got created how. First, we get a
CREATE_PROCESS_DEBUG_EVENT for tid=0x442a8:
gdb: kernel event for pid=317536 tid=0x442a8 code=CREATE_PROCESS_DEBUG_EVENT)
Shortly after, we get some CREATE_THREAD_DEBUG_EVENT events,
one of them being for tid=0x4010c:
gdb: kernel event for pid=317536 tid=0x4010c code=CREATE_THREAD_DEBUG_EVENT)
Fast forward a bit of debugging, and we do a "cont" as above,
at which point the programs reaches the end, and the system reports
"exit" events. The first interesting one is the following:
gdb: kernel event for pid=317536 tid=0x442a8 code=EXIT_THREAD_DEBUG_EVENT)
This is reporting a thread-exit event for a thread whose tid
is the TID of what we call the "main thread". That's the thread
that was created when we received the CREATE_PROCESS_DEBUG_EVENT
notification, and whose TID is actually stored in a global variable
named main_thread_id. This is not something we expected, as
the assumption we made was that the main thread would exit last,
and we would be notified of it via an EXIT_PROCESS_DEBUG_EVENT.
But apparently, this is not always true, at least on Windows Server
2012 and 2016 where this issue has been observed happening randomly.
The consequence of the above notification is that we call
windows_delete_thread for that thread, which removes it from
our list of known threads.
And a little bit later, then we then get the EXIT_PROCESS_DEBUG_EVENT,
and we can see that the associated tid is not the main_thread_id,
but rather the tid of one of the threads that was created during
the lifetime of the program, in this case tid=0x4010c:
gdb: kernel event for pid=317536 tid=0x4010c code=EXIT_PROCESS_DEBUG_EVENT)
And the debug trace printed right after shows why we're crashing:
[Deleting Thread 317536.0x442a8]
We are trying to delete the thread whose tid=0x442a8, which is
the main_thread_id! As we have already deleted that thread before,
the search for it returns a nullptr, which then trips the assertion
check in delete_thread_1.
This commit fixes this issue. It ignores the open question of
what to do with the main_thread_id global, particularly after
that thread has been removed from our list of threads. This will
be dealt with as a separate patch, to allow cherry-picking
this patch into a release branch.
For now, we fix the code so as to avoid this crash.
gdb/ChangeLog:
* windows-nat.c (get_windows_debug_event) <EXIT_PROCESS_DEBUG_EVENT>:
Use current_event.dwThreadId instead of main_thread_id.
When an Ada program is dynamically linked against libgnat, and when
one of the standard exceptions is used, the exception object may be
referenced by the main executable using a copy relocation.
In this situation, a "catch exception" for those exceptions will not
manage to stop. This happens because, under the hood, "catch
exception" creates an expression object that examines the object
addresses -- but in this case, the address will be incorrect.
This patch fixes the problem by arranging for these filter expressions
to examine all the relevant minimal symbols. This way, the object
from libgnat will be found as well.
Tested on x86-64 Fedora 29.
gdb/ChangeLog
2019-04-30 Tom Tromey <tromey@adacore.com>
* ada-lang.c (ada_lookup_simple_minsyms): New function.
(create_excep_cond_exprs): Iterate over program spaces.
(ada_exception_catchpoint_cond_string): Examine all minimal
symbols for exception types.
gdb/testsuite/ChangeLog
2019-04-30 Tom Tromey <tromey@adacore.com>
* lib/ada.exp (find_ada_tool): New proc.
* lib/gdb.exp (gdb_compile_shlib): Allow .o files as inputs.
* gdb.ada/catch_ex_std.exp: New file.
* gdb.ada/catch_ex_std/foo.adb: New file.
* gdb.ada/catch_ex_std/some_package.adb: New file.
* gdb.ada/catch_ex_std/some_package.ads: New file.
PR c++/24470 concerns a crash in dwarf2read.c that occurs with a
particular test case.
The issue turns out to be that process_structure_scope will pass NULL
to symbol_symtab. This happens because new_symbol decided not to
create a symbol for the particular DIE.
This patch fixes the problem by finding another reasonably-appropriate
symtab to use instead; issuing a complaint if one cannot be found for
some reason.
As mentioned in the bug, I think there are other bugs here. For
example, when using "ptype" on the "l" object in the test case, I
think I would expect to see the template parameter. I didn't research
this too closely, since it seemed more important to fix the crash.
Tested on x86-64 Fedora 29.
I'd like to check this in to the 8.3 branch as well.
2019-04-30 Tom Tromey <tromey@adacore.com>
PR c++/24470:
* dwarf2read.c (process_structure_scope): Handle case where type
has template parameters but no symbol was created.
gdb/testsuite/ChangeLog
2019-04-30 Tom Tromey <tromey@adacore.com>
PR c++/24470:
* gdb.cp/temargs.cc: Add test code from PR.
Fixes the following error:
ld.pod around line 568: Expected '=item *'
POD document had syntax errors at /usr/bin/pod2man line 71.
* ld.texi (How GNU properties are merged): Avoid pod2man error.
Correct example.
Types in Fortran can have the 'allocatable' qualifier attached to
indicate that memory needs to be explicitly allocated by the user.
This patch extends GDB to show this qualifier when printing types.
Lots of tests results are then updated to include this new qualifier
in the expected results.
gdb/ChangeLog:
* f-typeprint.c (f_type_print_base): Print 'allocatable' type
qualifier.
* gdbtypes.h (TYPE_IS_ALLOCATABLE): Define.
gdb/testsuite/ChangeLog:
* gdb.fortran/vla-datatypes.exp: Update expected results.
* gdb.fortran/vla-ptype.exp: Likewise.
* gdb.fortran/vla-type.exp: Likewise.
* gdb.fortran/vla-value.exp: Likewise.
The whitespace produced as types are printed seems inconsistent. This
commit updates the rules in an attempt to make whitespace more
balanced and consistent. Expected results are updated.
gdb/ChangeLog:
* f-typeprint.c (f_print_type): Update rules for printing
whitespace.
(f_type_print_varspec_suffix): Likewise.
gdb/testsuite/ChangeLog:
* gdb.fortran/ptr-indentation.exp: Update expected results.
* gdb.fortran/ptype-on-functions.exp: Likewise.
* gdb.fortran/vla-ptr-info.exp: Likewise.
* gdb.fortran/vla-value.exp: Likewise.
Before this commit using ptype on a Fortran function will include
information about the functions return type, but not the expected
arguments as it would for C or C++. After this commit argument types
are included in the ptype output.
For example, before GDB prints:
(gdb) ptype fun1
type = integer(kind=4) ()
(gdb) ptype is_bigger
type = logical(kind=4) ()
and after GDB prints:
(gdb) ptype fun1
type = integer(kind=4) (integer(kind=4))
(gdb) ptype is_bigger
type = logical(kind=4) (integer(kind=4), integer(kind=4))
gdb/ChangeLog:
* f-typeprint.c (f_type_print_varspec_suffix): Handle printing
function arguments.
gdb/testsuite/ChangeLog:
* gdb.fortran/ptype-on-functions.exp: New file.
* gdb.fortran/ptype-on-functions.f90: New file.
For a program compiled with gfortran the base type names are written
as lower cases in the DWARF, and so GDB will display them as lower
case. Additionally, in most places where GDB supplies its own type
names (for example all of the types defined in f-lang.c in
`build_fortran_types`), the type names are all lower case.
An exception to this is where GDB prints the void type for Fortran.
In this case GDB uses upper case.
I'm not aware of any reason why this type should merit special
attention, and it looks our of place when printing types, so this
commit changes from 'VOID' to 'void' to match all the other types.
gdb/ChangeLog:
* f-lang.c (build_fortran_types): Change name of void type to
lower case.
* f-typeprint.c (f_type_print_base): Print the name of the void
type, rather than a fixed string.
* f-valprint.c (f_decorations): Use lower case void string.
gdb/testsuite/ChangeLog:
* gdb.fortran/exprs.exp (test_convenience_variables): Expect lower
case void string.
Currently when using $_creal and $_cimag to access the components of a
complex number the types of these components will have C type names
'float', 'double', etc. This is because the components of a complex
number are not given type names in DWARF, so GDB has to pick some
suitable names, and currently we always use the C names.
This commit changes the type names used based on the language, so for
Fortran we will now use the Fortran float types, and so will get the
Fortran float type names 'real', 'real*8', etc.
gdb/ChangeLog:
* dwarf2read.c (dwarf2_init_complex_target_type): Use different
types for Fortran.
gdb/testsuite/ChangeLog:
* gdb.fortran/complex.exp: Expand.
* gdb.fortran/complex.f: Renamed to...
* gdb.fortran/complex.f90: ...this, and extended to add more
complex values.
Future commits will add more Fortran specific expression operators.
In preparation for these new operators, this commit adds a new
fortran-operator.def file similar to how GDB already has
ada-operator.def.
I've moved UNOP_KIND the Fortran specific operator I introduced in
commit 4d00f5d8f6 into this file, and renamed it to make it clearer
that the operator is Fortran specific. I've then updated the Fortran
exp_descriptor table (exp_descriptor_f) to use entirely Fortran
specific functions that now handle UNOP_FORTRAN_KIND (the new name for
UNOP_KIND).
There should be no visible changes for standard users after this
commit, though for developers, the output when 'set debug expression
1' is now better, before:
(gdb) p kind (l1)
Dump of expression @ 0x2ccc7a0, before conversion to prefix form:
Language fortran, 5 elements, 16 bytes each.
Index Opcode Hex Value String Value
0 OP_VAR_VALUE 42 *...............
1 OP_NULL 47730176 .N..............
2 BINOP_INTDIV 47729184 J..............
3 OP_VAR_VALUE 42 *...............
4 UNOP_KIND 78 N...............
Dump of expression @ 0x2ccc7a0, after conversion to prefix form:
Expression: `Invalid expression
(gdb)
and after:
(gdb) p kind (l1)
Dump of expression @ 0x294d0b0, before conversion to prefix form:
Language fortran, 5 elements, 16 bytes each.
Index Opcode Hex Value String Value
0 OP_VAR_VALUE 40 (...............
1 unknown opcode: 224 44088544 ................
2 unknown opcode: 208 44087504 ................
3 OP_VAR_VALUE 40 (...............
4 UNOP_FORTRAN_KIND 119 w...............
Dump of expression @ 0x294d0b0, after conversion to prefix form:
Expression: `KIND(test::l1)'
Language fortran, 5 elements, 16 bytes each.
0 UNOP_FORTRAN_KIND
1 OP_VAR_VALUE Block @0x2a0bce0, symbol @0x2a0b8d0 (l1)
$1 = 1
(gdb)
gdb/ChangeLog:
* gdb/expprint.c (dump_subexp_body_standard): Remove use of
UNOP_KIND.
* gdb/expression.h (exp_opcode): Include 'fortran-operator.def'.
* gdb/f-exp.y (exp): Rename UNOP_KIND to UNOP_FORTRAN_KIND.
* gdb/f-lang.c (evaluate_subexp_f): Likewise.
(operator_length_f): New fuction.
(print_subexp_f): New function.
(op_name_f): New function.
(dump_subexp_body_f): New function.
(operator_check_f): New function.
(exp_descriptor_f): Replace standard expression handling functions
with new functions.
* gdb/fortran-operator.def: New file.
* gdb/parse.c (operator_length_standard): Remove use of UNOP_KIND.
* gdb/std-operator.def: Remove UNOP_KIND.
What appears to be a stray double quote character in std-operator.def
causes incorrect highlighting in my editor.
The quote was introduced in this commit:
commit 858be34c5a
Date: Mon Sep 4 20:21:15 2017 +0100
Handle "p S::method()::static_var" in the C++ parser
I can't see any reason why the quote should be there, so this commit
removes it.
gdb/ChangeLog:
* std-operator.def: Remove unbalanced, stray double quote
character.
Introduce a new print setting max-depth which can be set with 'set
print max-depth DEPTH'. The default value of DEPTH is 20, but this
can also be set to unlimited.
When GDB is printing a value containing nested structures GDB will
stop descending at depth DEPTH. Here is a small example:
typedef struct s1 { int a; } s1;
typedef struct s2 { s1 b; } s2;
typedef struct s3 { s2 c; } s3;
typedef struct s4 { s3 d; } s4;
s4 var = { { { { 3 } } } };
The following table shows how various depth settings affect printing
of 'var':
| Depth Setting | Result of 'p var' |
|---------------+--------------------------------|
| Unlimited | $1 = {d = {c = {b = {a = 3}}}} |
| 4 | $1 = {d = {c = {b = {a = 3}}}} |
| 3 | $1 = {d = {c = {b = {...}}}} |
| 2 | $1 = {d = {c = {...}}} |
| 1 | $1 = {d = {...}} |
| 0 | $1 = {...} |
Only structures, unions, and arrays are replaced in this way, scalars
and strings are not replaced.
The replacement is counted from the level at which you print, not from
the top level of the structure. So, consider the above example and
this GDB session:
(gdb) set print max-depth 2
(gdb) p var
$1 = {d = {c = {...}}}
(gdb) p var.d
$2 = {c = {b = {...}}}
(gdb) p var.d.c
$3 = {b = {a = 3}}
Setting the max-depth to 2 doesn't prevent the user from exploring
deeper into 'var' by asking for specific sub-fields to be printed.
The motivation behind this feature is to try and give the user more
control over how much is printed when examining large, complex data
structures.
The default max-depth of 20 means that there is a change in GDB's
default behaviour. Someone printing a data structure with 20 levels
of nesting will now see '{...}' instead of their data, they would need
to adjust the max depth, or call print again naming a specific field
in order to dig deeper into their data structure. If this is
considered a problem then we could increase the default, or even make
the default unlimited.
This commit relies on the previous commit, which added a new field to
the language structure, this new field was a string that contained the
pattern that should be used when a structure/union/array is replaced
in the output, this allows languages to use a syntax that is more
appropriate, mostly this will be selecting the correct types of
bracket '(...)' or '{...}', both of which are currently in use.
This commit should have no impact on MI output, expressions are
printed through the MI using -var-create and then -var-list-children.
As each use of -var-list-children only ever displays a single level of
an expression then the max-depth setting will have no impact.
This commit also adds the max-depth mechanism to the scripting
language pretty printers following basically the same rules as for the
built in value printing.
One quirk is that when printing a value using the display hint 'map',
if the keys of the map are structs then GDB will hide the keys one
depth level after it hides the values, this ensures that GDB produces
output like this:
$1 = map_object = {[{key1}] = {...}, [{key2}] = {...}}
Instead of this less helpful output:
$1 = map_object = {[{...}] = {...}, [{...}] = {...}}
This is covered by the new tests in gdb.python/py-nested-maps.exp.
gdb/ChangeLog:
* cp-valprint.c (cp_print_value_fields): Allow an additional level
of depth when printing anonymous structs or unions.
* guile/scm-pretty-print.c (gdbscm_apply_val_pretty_printer):
Don't print either the top-level value, or the children if the
max-depth is exceeded.
(ppscm_print_children): When printing the key of a map, allow one
extra level of depth.
* python/py-prettyprint.c (gdbpy_apply_val_pretty_printer): Don't
print either the top-level value, or the children if the max-depth
is exceeded.
(print_children): When printing the key of a map, allow one extra
level of depth.
* python/py-value.c (valpy_format_string): Add max_depth keyword.
* valprint.c: (PRINT_MAX_DEPTH_DEFAULT): Define.
(user_print_options): Initialise max_depth field.
(val_print_scalar_or_string_type_p): New function.
(val_print): Check to see if the max depth has been reached.
(val_print_check_max_depth): Define new function.
(show_print_max_depth): New function.
(_initialize_valprint): Add 'print max-depth' option.
* valprint.h (struct value_print_options) <max_depth>: New field.
(val_print_check_max_depth): Declare new function.
* NEWS: Document new feature.
gdb/doc/ChangeLog:
* gdb.texinfo (Print Settings): Document 'print max-depth'.
* guile.texi (Guile Pretty Printing API): Document that 'print
max-depth' can effect the display of a values children.
* python.texi (Pretty Printing API): Likewise.
(Values From Inferior): Document max_depth keyword.
gdb/testsuite/ChangeLog:
* gdb.base/max-depth.c: New file.
* gdb.base/max-depth.exp: New file.
* gdb.python/py-nested-maps.c: New file.
* gdb.python/py-nested-maps.exp: New file.
* gdb.python/py-nested-maps.py: New file.
* gdb.python/py-format-string.exp (test_max_depth): New proc.
(test_all_common): Call test_max_depth.
* gdb.fortran/max-depth.exp: New file.
* gdb.fortran/max-depth.f90: New file.
* gdb.go/max-depth.exp: New file.
* gdb.go/max-depth.go: New file.
* gdb.modula2/max-depth.exp: New file.
* gdb.modula2/max-depth.c: New file.
* lib/gdb.exp (get_print_expr_at_depths): New proc.
This commit is preparation work for the next commit, and by itself
makes no user visible change to GDB. I've split this work into a
separate commit in order to make code review easier.
This commit adds a new field 'la_is_string_type_p' to the language
struct, this predicate will return true if a type is a string type for
the given language.
Some languages already have a "is this a string" predicate that I was
able to reuse, while for other languages I've had to add a new
predicate. In this case I took inspiration from the value printing
code for that language - what different conditions would result in
printing something as a string.
A default "is this a string" method has also been added that looks for
TYPE_CODE_STRING, this is the fallback I've used for a couple of
languages.
In this commit I add the new field and initialise it for each
language, however at this stage the new field is never used.
gdb/ChangeLog:
* ada-lang.c (ada_language_defn): Initialise new field.
* c-lang.c (c_is_string_type_p): New function.
(c_language_defn): Initialise new field.
(cplus_language_defn): Initialise new field.
(asm_language_defn): Initialise new field.
(minimal_language_defn): Initialise new field.
* c-lang.h (c_is_string_type_p): Declare new function.
* d-lang.c (d_language_defn): Initialise new field.
* f-lang.c (f_is_string_type_p): New function.
(f_language_defn): Initialise new field.
* go-lang.c (go_is_string_type_p): New function.
(go_language_defn): Initialise new field.
* language.c (default_is_string_type_p): New function.
(unknown_language_defn): Initialise new field.
(auto_language_defn): Initialise new field.
* language.h (struct language_defn) <la_is_string_type_p>: New
member variable.
(default_is_string_type_p): Declare new function.
* m2-lang.c (m2_language_defn): Initialise new field.
* objc-lang.c (objc_language_defn): Initialise new field.
* opencl-lang.c (opencl_language_defn): Initialise new field.
* p-lang.c (pascal_is_string_type_p): New function.
(pascal_language_defn): Initialise new field.
* rust-lang.c (rust_is_string_type_p): New function.
(rust_language_defn): Initialise new field.
This commit is preparation work for a later commit, and by itself
makes no user visible change to GDB. I've split this work into a
separate commit in order to make code review easier.
This commit adds a new field 'la_struct_too_deep_ellipsis' to the
language struct, this string will be used in the next commit to print
a language specific string from within the generic value printing
code.
In this commit I add the new field and initialise it for each
language, however at this stage the new field is never used.
gdb/ChangeLog:
* language.h (struct language_defn) <la_struct_too_deep_ellipsis>:
New field.
* ada-lang.c (ada_language_defn): Initialise new field.
* c-lang.c (c_language_defn): Likewise.
(cplus_language_defn): Likewise.
(asm_language_defn): Likewise.
(minimal_language_defn): Likewise.
* d-lang.c (d_language_defn): Likewise.
* f-lang.c (f_language_defn): Likewise.
* go-lang.c (go_language_defn): Likewise.
* language.c (unknown_language_defn): Likewise.
(auto_language_defn): Likewise.
* m2-lang.c (m2_language_defn): Likewise.
* objc-lang.c (objc_language_defn): Likewise.
* opencl-lang.c (opencl_language_defn): Likewise.
* p-lang.c (pascal_language_defn): Likewise.
* rust-lang.c (rust_language_defn): Likewise.
A later commit would like to make use of a pointer to the function
ada_is_string_type, however, this will require the function to return
a bool (so the signature matches).
As the ada_is_string_type is a predicate function, and its return
value is only ever used as either true or false, then this commit
updates the function to return a bool.
As a consequence ada_is_character_type needs to change too.
There should be no user visible changes after this commit.
gdb/ChangeLog:
* ada-lang.c (ada_is_character_type): Change return type to bool.
(ada_is_string_type): Likewise.
* ada-lang.h (ada_is_character_type): Update declaration
(ada_is_string_type): Likewise.
When running gdb-caching-proc.exp, if skip_opencl_tests fails like this:
...
(gdb) run
Starting program: \
build/gdb/testsuite/outputs/gdb.base/gdb-caching-proc/opencltest13530.x
CHK_ERR (clGetPlatformIDs (1, &platform, NULL), -1001)
src/gdb/testsuite/lib/opencl_hostapp.c:73 error: Unknown
[Inferior 1 (process 13600) exited with code 01]
(gdb)
skip_opencl_tests: OpenCL support not detected
...
then this regexp in skip_opencl_tests fails to match:
...
-re ".*$inferior_exited_re code.*${gdb_prompt} $" {
...
so instead we hit the default clause after a 30 seconds timeout. With the
iteration count set at 10, we end up taking 6 minutes to run this test-case.
Fix this by adding the missing "with" in the regexp, bring back the runtime to
half a minute.
Tested on x86_64-linux.
gdb/testsuite/ChangeLog:
2019-04-29 Tom de Vries <tdevries@suse.de>
* lib/opencl.exp (skip_opencl_tests): Add missing "with" in regexp.
'thread|frame apply CMD' launches CMD so that CMD output goes to a string_file.
This patch ensures that string_file for such CMD output contains
style escape sequences that 'thread|frame apply' will later on
output on the real terminal, so as to have CMD output properly styled.
The idea is to have the class ui_file having overridable methods
to indicate that the output to this ui_file should be done using
'terminal' behaviour such as styling.
Then these methods are overriden in string_file so that a specially
constructed string_file will get output with style escape sequences.
After this patch, the output of CMD by thread|frame apply CMD is styled
similarly as when CMD is launched directly.
Note that string_file (term_out true) could also support wrapping,
but this is not done (yet?).
Tested on debian/amd64.
gdb/ChangeLog
2019-04-27 Philippe Waroquiers <philippe.waroquiers@skynet.be>
Support style in 'frame|thread apply'
* gdbcmd.h (execute_command_to_string): New term_out parameter.
* record.c (record_start, record_stop): Update callers of
execute_command_to_string with false.
* ui-file.h (class ui_file): New term_out and can_emit_style_escape
methods.
(class string_file): New constructor with term_out parameter.
Override methods term_out and can_emit_style_escape. New member
term_out.
(class stdio_file): Override can_emit_style_escape.
(class tee_file): Override term_out and can_emit_style_escape.
* utils.h (can_emit_style_escape): Remove.
* utils.c (can_emit_style_escape): Likewise.
Update all callers of can_emit_style_escape (SOMESTREAM) to
SOMESTREAM->can_emit_style_escape.
* source-cache.c (source_cache::get_source_lines): Likewise.
* stack.c (frame_apply_command_count): Call execute_command_to_string
passing the term_out characteristic of the current gdb_stdout.
* thread.c (thr_try_catch_cmd): Likewise.
* top.c (execute_command_to_string): pass term_out parameter
to construct the string_file for the command output.
* ui-file.c (term_cli_styling): New function (most code moved
from utils.c can_emit_style_escape).
(string_file::string_file, string_file::can_emit_style_escape,
stdio_file::can_emit_style_escape, tee_file::term_out,
tee_file::can_emit_style_escape): New functions.
Inferior function calls are powerful but might lead to undesired
results such as crashes when calling nested functions (frequently
used in particular in Ada).
This implements a GDB setting to disable calling inferior functions.
Note: the idea is that if/when the 'slash command' patch is pushed,
that this setting can be changed e.g. by using the shortcut /c.
This is version 2 of the patch. It handles all the received comments,
mostly replace 'can-call' by 'may-call', and avoid using
'inferior function call' in factor of 'calling function in the program'.
2019-04-26 Philippe Waroquiers <philippe.waroquiers@skynet.be>
gdb/ChangeLog
* NEWS: Mention the new set|show may-call-functions.
* infcall.c (may_call_functions_p): New variable.
(show_may_call_functions_p): New function.
(call_function_by_hand_dummy): Throws an error if not
may-call-functions.
(_initialize_infcall): Call add_setshow_boolean_cmd for
may-call-functions.
gdb/testsuite/ChangeLog
* gdb.base/callexit.exp: Test may-call-functions off.
gdb/doc/ChangeLog
* gdb.texinfo (Calling): Document the new
set|show may-call-functions.
Add several baseline MIPS32R6[1] and MIPS64R6[2] instructions
that were omitted from the initial spec. These instructions
are optional in implementations but not associated with any
ASE or pseudo-ASE. Their presence is indicated by the XNP bit
in the Config5 register.
[1] "MIPS Architecture for Programmers Volume II-A: The MIPS32
Instruction Set Manual", Imagination Technologies Ltd., Document
Number: MD00086, Revision 6.06, December 15, 2016, Section 3.2
"Alphabetical List of Instructions", pp. 228-229, pp. 354-357.
[2] "MIPS Architecture for Programmers Volume II-A: The MIPS64
Instruction Set Manual", Imagination Technologies Ltd., Document
Number: MD00087, Revision 6.06, December 15, 2016, Section 3.2
"Alphabetical List of Instructions", pp. 289-290 and pp. 458-460.
gas/
* config/tc-mips.c (macro) <M_LLWP_AB, M_LLDP_AB, M_SCWP_AB,
M_SCDP_AB>: New cases and expansions for paired instructions.
* testsuite/gas/mips/llpscp-32.s: New test source.
* testsuite/gas/mips/llpscp-64.s: Likewise.
* testsuite/gas/mips/llpscp-32.d: New test.
* testsuite/gas/mips/llpscp-64.d: Likewise.
* testsuite/gas/mips/mips.exp: Run the new tests.
* testsuite/gas/mips/r6.s: Add new instructions to test source.
* testsuite/gas/mips/r6-64.s: Likewise.
* testsuite/gas/mips/r6-64-n32.d: Check new instructions.
* testsuite/gas/mips/r6-64-n64.d: Likewise.
* testsuite/gas/mips/r6-n32.d: Likewise.
* testsuite/gas/mips/r6-n64.d: Likwwise.
* testsuite/gas/mips/r6.d: Likewise.
include/
* opcode/mips.h (M_LLWP_AB, M_LLDP_AB): New enum values.
(M_SCWP_AB, M_SCDP_AB): Likewise.
opcodes/
* mips-opc.c (mips_builtin_opcodes): Add llwp, lldp, scwp, scdp.
The .code16gcc directive supports 16bit mode with 32-bit address. Since
IRET (opcode 0xcf) in 16bit mode returns from an interrupt in 16bit mode,
we shouldn't add 0x66 prefix for IRET.
PR gas/24485
* config/tc-i386.c (process_suffix): Don't add DATA_PREFIX_OPCODE
to IRET for .code16gcc.
* testsuite/gas/i386/jump16.s: Add IRET tests.
* testsuite/gas/i386/jump16.d: Updated.
When undefined non-weak references in IR objects are optimized out
by LTO, we can have weak dynamic referencs to symbols marked with
bfd_link_hash_undefined. We shouldn't complain such undefined weak
dynamic references.
bfd/
PR ld/24486
* elflink.c (elf_link_output_extsym): Don't complain undefined
weak dynamic reference.
ld/
PR ld/24486
* testsuite/ld-plugin/lto.exp: Run PR ld/24486 tests.
* testsuite/ld-plugin/pr24486a.c: New file.
* testsuite/ld-plugin/pr24486b.c: Likewise.
* testsuite/ld-plugin/pr24486c.c: Likewise.
This bug finds another usage where we end up segfaulting while
normalizing user input. inspect_type and replace_type recurse,
attempting to substitute the "real" symbol name for the typedef name.
However, since the both these names are the same, they keep calling
each other until the stack overflows.
A simple reproducer for it is given by
typedef struct foo foo;
int qux (foo *f) { return 0; }
(gdb) b qux(foo*)
Segmentation fault
inspect_type already contains some special handling to prevent a
similar situation from occurring with namespaces. I wonder, however,
whether we need be so pedantic about the exact nature of the substitution.
This patch implements this rather more aggressive assumption that these
substitutions should be avoided whenever the replacement symbol's name is
exactly the same as the one we're trying to substitute. [In the above
example, we're trying to substitute the tyepdef named "foo" with the symbol
named "foo" (a struct).]
gdb/ChangeLog:
PR c++/24367
* cp-support.c (inspect_type): Don't attempt substitutions
of symbol with the same name.
gdb/testsuite/ChangeLog:
PR c++/24367
* gdb.cp/meth-typedefs.cc (incomplete_struct)
(another_incomplete_struct, test_incomplete): New definitions.
(main): Use new definitions.
* gdb.cp/meth-typedefs.exp: Add new tests for `test_incomplete'
functions.
PR gdb/24475 concerns a memory leak coming from gdb's exception
handling code.
The leak occurs because throw_exception_sjlj does not arrange to
destroy the exception object it is passed. However, because
gdb_exception has a destructor, it's undefined to longjmp in this
situation.
This patch fixes the problem by avoiding the need to run any
destructors in gdb_rl_callback_handler, by making the gdb_exception
"static".
gdb/ChangeLog
2019-04-25 Tom Tromey <tromey@adacore.com>
PR gdb/24475:
* event-top.c (gdb_rl_callback_handler): Make "gdb_rl_expt"
static.
This makes exception handling more efficient in a few spots, through
the use of const- and rvalue-references.
I wrote this patch by commenting out the gdb_exception copy
constructor and then examining the resulting error messages one by
one, introducing the use of std::move where appropriate.
gdb/ChangeLog
2019-04-25 Tom Tromey <tromey@adacore.com>
* xml-support.c (struct gdb_xml_parser) <set_error>: Take an
rvalue reference.
(gdb_xml_start_element_wrapper, gdb_xml_end_element_wrapper)
(gdb_xml_parser::parse): Use std::move.
* python/python-internal.h (gdbpy_convert_exception): Take a const
reference.
* python/py-value.c (valpy_getitem, valpy_nonzero): Use
std::move.
* python/py-utils.c (gdbpy_convert_exception): Take a const
reference.
* python/py-inferior.c (infpy_write_memory, infpy_search_memory):
Use std::move.
* python/py-breakpoint.c (bppy_set_condition, bppy_set_commands):
Use std::move.
* mi/mi-main.c (mi_print_exception): Take a const reference.
* main.c (handle_command_errors): Take a const reference.
* linespec.c (parse_linespec): Use std::move.
* infcall.c (run_inferior_call): Use std::move.
(call_function_by_hand_dummy): Use std::move.
* exec.c (try_open_exec_file): Use std::move.
* exceptions.h (exception_print, exception_fprintf)
(exception_print_same): Update.
* exceptions.c (print_exception, exception_print)
(exception_fprintf, exception_print_same): Change parameters to
const reference.
* event-top.c (gdb_rl_callback_read_char_wrapper): Update.
* common/new-op.c: Use std::move.
* common/common-exceptions.h (struct gdb_exception): Add move
constructor.
(struct gdb_exception_error, struct gdb_exception_quit, struct
gdb_quit_bad_alloc): Change constructor to move constructor.
(throw_exception): Change parameter to rvalue reference.
* common/common-exceptions.c (throw_exception): Take rvalue
reference.
* cli/cli-interp.c (safe_execute_command): Use std::move.
* breakpoint.c (insert_bp_location, location_to_sals): Use
std::move.
The Guile code will longjmp (via scm_throw) when an object requiring
destruction is on the stack. This is undefined behavior.
This changes this code to run any destructors in inner scopes, and to
pass a POD to gdbscm_throw_gdb_exception.
gdb/ChangeLog
2019-04-25 Tom Tromey <tromey@adacore.com>
* guile/scm-exception.c (gdbscm_scm_from_gdb_exception)
(gdbscm_throw_gdb_exception): Take a gdbscm_gdb_exception.
* guile/scm-block.c, guile/scm-breakpoint.c, guile/scm-cmd.c,
guile/scm-disasm.c, guile/scm-frame.c, guile/scm-lazy-string.c,
guile/scm-math.c, guile/scm-param.c, guile/scm-ports.c,
guile/scm-symbol.c, guile/scm-symtab.c, guile/scm-type.c,
guile/scm-value.c: Use unpack.
* guile/guile-internal.h (gdbscm_scm_from_gdb_exception): Take a
gdbscm_gdb_exception.
(gdbscm_throw_gdb_exception): Likewise.
(struct gdbscm_gdb_exception): New.
(unpack): New function.
(gdbscm_wrap): Use unpack.
This changes the SJLJ exception handling code to be a bit more
efficient, by using rvalue references and move assignment when
possible.
Tested by the buildbot.
gdb/ChangeLog
2019-04-25 Tom Tromey <tromey@adacore.com>
* event-top.c (gdb_rl_callback_read_char_wrapper_noexcept)
(gdb_rl_callback_handler): Use std::move.
* common/common-exceptions.h (struct gdb_exception): Add move
assignment operator.
(throw_exception_sjlj): Change "exception" to const reference.
* common/common-exceptions.c (exceptions_state_mc_catch): Update.
(throw_exception_sjlj): Change "exception" to const reference.
This patch has a long story, but it all started back in 2015, with
commit df8411da08 ("Implement support
for checking /proc/PID/coredump_filter"). The purpose of that commit
was to bring GDB's corefile generation closer to what the Linux kernel
does. However, back then, I did not implement the full support for
the dumping of memory mappings containing ELF headers (like mappings
of DSOs or executables). These mappings were being dumped most of
time, though, because the default value of /proc/PID/coredump_filter
is 0x33, which would cause anonymous private mappings (DSOs/executable
code mappings have this type) to be dumped. Well, until something
happened on binutils...
A while ago, I noticed something strange was happening with one of our
local testcases on Fedora GDB: it was failing due to some strange
build-id problem. On Fedora GDB, we (unfortunately) carry a bunch of
"local" patches, and some of these patches actually extend upstream's
build-id support in order to generate more useful information for the
user of a Fedora system (for example, when the user loads a corefile
into GDB, we detect whether the executable that generated that
corefile is present, and if it's not we issue a warning suggesting
that it should be installed, while also providing the build-id of the
executable). A while ago, Fedora GDB stopped printing those warnings.
I wanted to investigate this right away, and spent some time trying to
determine what was going on, but other things happened and I got
sidetracked. Meanwhile, the bug started to be noticed by some of our
users, and its priority started changing. Then, someone on IRC also
mentioned the problem, and when I tried helping him, I noticed he
wasn't running Fedora. Hm... So maybe the bug was *also* present
upstream.
After "some" time investigating, and with a lot of help from Keith and
others, I was finally able to determine that yes, the bug is also
present upstream, and that even though it started with a change in ld,
it is indeed a GDB issue.
So, as I said, the problem started with binutils, more specifically
after the following commit was pushed:
commit f6aec96dce
Author: H.J. Lu <hjl.tools@gmail.com>
Date: Tue Feb 27 11:34:20 2018 -0800
ld: Add --enable-separate-code
This commit makes ld use "-z separate-code" by default on x86-64
machines. What this means is that code pages and data pages are now
separated in the binary, which is confusing GDB when it tries to decide
what to dump.
BTW, Fedora 28 binutils doesn't have this code, which means that
Fedora 28 GDB doesn't have the problem. From Fedora 29 on, binutils
was rebased and incorporated the commit above, which started causing
Fedora GDB to fail.
Anyway, the first thing I tried was to pass "-z max-page-size" and
specify a bigger page size (I saw a patch that did this and was
proposed to Linux, so I thought it might help). Obviously, this
didn't work, because the real "problem" is that ld will always use
separate pages for code and data. So I decided to look into how GDB
dumped the pages, and that's where I found the real issue.
What happens is that, because of "-z separate-code", the first two pages
of the ELF binary are (from /proc/PID/smaps):
00400000-00401000 r--p 00000000 fc:01 799548 /file
Size: 4 kB
KernelPageSize: 4 kB
MMUPageSize: 4 kB
Rss: 4 kB
Pss: 4 kB
Shared_Clean: 0 kB
Shared_Dirty: 0 kB
Private_Clean: 4 kB
Private_Dirty: 0 kB
Referenced: 4 kB
Anonymous: 0 kB
LazyFree: 0 kB
AnonHugePages: 0 kB
ShmemPmdMapped: 0 kB
Shared_Hugetlb: 0 kB
Private_Hugetlb: 0 kB
Swap: 0 kB
SwapPss: 0 kB
Locked: 0 kB
THPeligible: 0
VmFlags: rd mr mw me dw sd
00401000-00402000 r-xp 00001000 fc:01 799548 /file
Size: 4 kB
KernelPageSize: 4 kB
MMUPageSize: 4 kB
Rss: 4 kB
Pss: 4 kB
Shared_Clean: 0 kB
Shared_Dirty: 0 kB
Private_Clean: 0 kB
Private_Dirty: 4 kB
Referenced: 4 kB
Anonymous: 4 kB
LazyFree: 0 kB
AnonHugePages: 0 kB
ShmemPmdMapped: 0 kB
Shared_Hugetlb: 0 kB
Private_Hugetlb: 0 kB
Swap: 0 kB
SwapPss: 0 kB
Locked: 0 kB
THPeligible: 0
VmFlags: rd ex mr mw me dw sd
Whereas before, we had only one:
00400000-00401000 r-xp 00000000 fc:01 798593 /file
Size: 4 kB
KernelPageSize: 4 kB
MMUPageSize: 4 kB
Rss: 4 kB
Pss: 4 kB
Shared_Clean: 0 kB
Shared_Dirty: 0 kB
Private_Clean: 0 kB
Private_Dirty: 4 kB
Referenced: 4 kB
Anonymous: 4 kB
LazyFree: 0 kB
AnonHugePages: 0 kB
ShmemPmdMapped: 0 kB
Shared_Hugetlb: 0 kB
Private_Hugetlb: 0 kB
Swap: 0 kB
SwapPss: 0 kB
Locked: 0 kB
THPeligible: 0
VmFlags: rd ex mr mw me dw sd
Notice how we have "Anonymous" data mapped into the page. This will be
important.
So, the way GDB decides which pages it should dump has been revamped
by my patch in 2015, and now it takes the contents of
/proc/PID/coredump_filter into account. The default value for Linux
is 0x33, which means:
Dump anonymous private, anonymous shared, ELF headers and HugeTLB
private pages.
Or:
filter_flags filterflags = (COREFILTER_ANON_PRIVATE
| COREFILTER_ANON_SHARED
| COREFILTER_ELF_HEADERS
| COREFILTER_HUGETLB_PRIVATE);
Now, it is important to keep in mind that GDB doesn't always have *all*
of the necessary information to exactly determine the type of a page, so
the whole algorithm is based on heuristics (you can take a look at
linux-tdep.c:dump_mapping_p and
linux-tdep.c:linux_find_memory_regions_full for more info).
Before the patch to make ld use "-z separate-code", the (single) page
containing data and code was being flagged as an anonymous (due to the
non-zero "Anonymous:" field) private (due to the "r-xp" permission),
which means that it was being dumped into the corefile. That's why it
was working fine.
Now, as you can imagine, when "-z separate-code" is used, the *data*
page (which is where the ELF notes are, including the build-id one) now
doesn't have any "Anonymous:" mapping, so the heuristic is flagging it
as file-backed private, which is *not* dumped by default.
The next question I had to answer was: how come a corefile generated by
the Linux kernel was correct? Well, the answer is that GDB, unlike
Linux, doesn't actually implement the COREFILTER_ELF_HEADERS support.
On Linux, even though the data page is also treated as a file-backed
private mapping, it is also checked to see if there are any ELF headers
in the page, and then, because we *do* have ELF headers there, it is
dumped.
So, after more time trying to think of ways to fix this, I was able to
implement an algorithm that reads the first few bytes of the memory
mapping being processed, and checks to see if the ELF magic code is
present. This is basically what Linux does as well, except that, if
it finds the ELF magic code, it just dumps one page to the corefile,
whereas GDB will dump the whole mapping. But I don't think that's a
big issue, to be honest.
It's also important to explain that we *only* perform the ELF magic
code check if:
- The algorithm has decided *not* to dump the mapping so far, and;
- The mapping is private, and;
- The mapping's offset is zero, and;
- The user has requested us to dump mappings with ELF headers.
IOW, we're not going to blindly check every mapping.
As for the testcase, I struggled even more trying to write it. Since
our build-id support on upstream GDB is not very extensive, it's not
really possible to determine whether a corefile contains build-id
information or not just by using GDB. So, after thinking a lot about
the problem, I decided to rely on an external tool, eu-unstrip, in
order to verify whether the dump was successful. I verified the test
here on my machine, and everything seems to work as expected (i.e., it
fails without the patch, and works with the patch applied). We are
working hard to upstream our "local" Fedora GDB patches, and we intend
to submit our build-id extension patches "soon", so hopefully we'll be
able to use GDB itself to perform this verification.
I built and regtested this on the BuildBot, and no problems were
found.
gdb/ChangeLog:
2019-04-25 Sergio Durigan Junior <sergiodj@redhat.com>
PR corefiles/11608
PR corefiles/18187
* linux-tdep.c (dump_mapping_p): Add new parameters ADDR and
OFFSET. Verify if current mapping contains an ELF header.
(linux_find_memory_regions_full): Adjust call to
dump_mapping_p.
gdb/testsuite/ChangeLog:
2019-04-25 Sergio Durigan Junior <sergiodj@redhat.com>
PR corefiles/11608
PR corefiles/18187
* gdb.base/coredump-filter-build-id.exp: New file.
Add both board option and environment variable which enables gdbserver
debug and sends it to the file gdbserver.debug, located in the output
directory for the current test. Document this.
Add support for the environment variable in the Makefile.
The testsuite can be run with gdbserver debug enabled in the following way:
make check GDBSERVER_DEBUG=all
Disable tspeed.exp when debugging to prevent the log file filling
many gigabytes then timing out.
gdb/testsuite/ChangeLog:
* Makefile.in: Pass through GDBSERVER_DEBUG.
* README (Testsuite Parameters): Add GDBSERVER_DEBUG.
(gdbserver,debug): Add board setting.
* gdb.trace/tspeed.exp: Skip when debugging.
* lib/gdb.exp (gdbserver_debug_enabled): New procedure.
* lib/gdbserver-support.exp: Likewise
When a wrapper symbol, __wrap_FOO, is defined in IR, its resolution
should be LDPR_PREVAILING_DEF, not PREVAILING_DEF_IRONLY, since LTO
doesn't know that __wrap_FOO provides definition of FOO. And resolution
of FOO should be LDPR_RESOLVED_IR since it is resolved by __wrap_FOO in
IR.
PR ld/24406
* ld.texi: Remove LTO warning from --wrap.
* plugin.c (get_symbols): Update resolution for wrapper and
wrapped symbols.
* testsuite/ld-plugin/lto.exp: Run ld/24406 tests.
* testsuite/ld-plugin/pr24406-1.c: New file.
* testsuite/ld-plugin/pr24406-2a.c: Likewise.
* testsuite/ld-plugin/pr24406-2b.c: Likewise.
GDB was failing to catch cases where a corrupt ELF or core file
contained an invalid length value in a Dwarf debug frame FDE header.
It was checking for buffer overflow but not cases where the length was
negative or caused pointer wrap-around.
In addition to the additional validity check, this patch cleans up the
multiple signed/unsigned conversions on the length field so that an
unsigned representation is used consistently throughout.
This patch fixes CVE-2017-9778 and PR gdb/21600.
2019-04-25 Sandra Loosemore <sandra@codesourcery.com>
Kang Li <kanglictf@gmail.com>
PR gdb/21600
* dwarf2-frame.c (read_initial_length): Be consistent about using
unsigned representation of length.
(decode_frame_entry_1): Likewise. Check for wraparound of
end pointer as well as buffer overflow.
This patch aims to improve the definitions of BTI and PAC based PLTs.
The following changes are made:
* PLT0 does not need PAC instructions since the PLTGOT[2] (and PLTGOT[1])
are readonly so they cannot be corrupted at runtime. Thus both PAC plt0
and BTI+PAC plt0 are removed and we can use basic plt0 and BTI plt0
instead, respectively.
* We can remove the extra padding nops when we add the new bti instructions.
BTI plt0 and BTI TLSDESC plt are updated.
* For better performance PLTn could be padded to 24bytes. Both BTI pltn and
PAC pltn are updated.
*** bfd/ChangeLog ***
2019-04-25 Sudakshina Das <sudi.das@arm.com>
* elfnn-aarch64.c (PLT_BTI_ENTRY_SIZE): Remove.
(PLT_BTI_TLSDESC_ENTRY_SIZE): Remove.
(PLT_PAC_ENTRY_SIZE, PLT_BTI_PAC_ENTRY_SIZE): Remove.
(PLT_BTI_SMALL_ENTRY_SIZE, PLT_PAC_SMALL_ENTRY_SIZE): Update.
(elfNN_aarch64_small_plt0_pac_entry): Remove.
(elfNN_aarch64_small_plt0_bti_pac_entry): Remove.
(elfNN_aarch64_small_plt0_bti_entry): Update.
(elfNN_aarch64_small_plt_bti_entry): Update.
(elfNN_aarch64_small_plt_pac_entry): Update.
(elfNN_aarch64_tlsdesc_small_plt_bti_entry): Update.
(setup_plt_values): Setup new entries.
(elfNN_aarch64_finish_dynamic_sections): Remove size change.
(elfNN_aarch64_plt_sym_val): Likewise.
*** ld/ChangeLog ***
2019-04-25 Sudakshina Das <sudi.das@arm.com>
* testsuite/ld-aarch64/bti-pac-plt-1.d: Update.
* testsuite/ld-aarch64/bti-pac-plt-2.d: Update.
* testsuite/ld-aarch64/bti-plt-1.d: Update.
* testsuite/ld-aarch64/bti-plt-3.d: Update.
* testsuite/ld-aarch64/bti-plt-5.d: Update.
* testsuite/ld-aarch64/pac-plt-1.d: Update.
* testsuite/ld-aarch64/pac-plt-2.d: Update.
Complement commit 1586d91e32 ("/ 0 should send SIGFPE not SIGTRAP..."),
<https://sourceware.org/ml/binutils/2004-07/msg00260.html>, and update a
stale comment referring the 20-bit code field of the BREAK and SDBBP
instructions, by making it explicit that where permitted by choosing the
MIPS32 or a later ISA the whole field can now be set with a single
operand for the SDBBP instruction only.
include/
* opcode/mips.h: Update comment for MIPS32 CODE20 operand.
Targets such as xtensa incur a much higher overhead to resolve
location view numbers than e.g. x86, because the expressions used to
compute view numbers cannot be resolved soon enough.
Each view number is computed by incrementing the previous view, if
they are both at the same address, or by resetting it to zero
otherwise. If PV is the previous view number, PL is its location, and
NL is the location of the next view, its number is computed by
evaluating NV = !(NL > PL) * (PV + 1).
set_or_check_view uses resolve_expression to decide whether portions
of this expression can be simplified to constants. The (NL > PL)
subexpression is one that can often be resolved to a constant,
breaking chains of view number computations at instructions of nonzero
length, but not after alignment that might be unnecessary.
Alas, when nearly every frag ends with a relaxable instruction,
frag_offset_fixed_p will correctly fail to determine a known offset
between two unresolved addresses in neighboring frags, so the
unresolved symbolic operation will be constructed and used in the
computation of most view numbers. This results in very deep
expressions.
As view numbers get referenced in location view lists, each operand in
the list goes through symbol_clone_if_forward_ref, which recurses on
every subexpression. If each view number were to be referenced, this
would exhibit O(n^2) behavior, where n is the depth of the view number
expressions, i.e., the length of view number sequences without an
early resolution that cuts the expression short.
This patch enables address compares used by view numbering to be
resolved even when exact offsets are not known, using new logic to
determine when the location either remained the same or changed for
sure, even with the possibility of relaxation. This enables most view
number expressions to be resolved with a small, reasonable depth.
PR gas/24444
* frags.c (frag_gtoffset_p): New.
* frags.h (frag_gtoffset_p): Declare it.
* expr.c (resolve_expression): Use it.
This changes the gdb test suite to omit -fno-stack-protector when
compiling Rust code. This makes Rust testing work again.
I think I saw this patch somewhere already, but I couldn't find it
again just now, so I'm checking this version in.
gdb/testsuite/ChangeLog
2019-04-24 Tom Tromey <tromey@adacore.com>
* lib/gdb.exp (gdb_compile): Don't add -fno-stack-protector for
Rust.
