This patch adds alternate CPU names which adhere to the number of the
architecture document. So instead of having z196, zEC12, and z13 you
can use arch9, arch10, and arch11. The old cpu names stay valid and
should primarily be used.
The alternate names are supposed to improve compatibility with the IBM
XL compiler toolchain which uses the arch numbering.
opcodes/ChangeLog:
2016-09-12 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
* s390-mkopc.c (main): Support alternate arch strings.
gas/ChangeLog:
2016-09-12 Andreas Krebbel <krebbel@linux.vnet.ibm.com>
* config/tc-s390.c (s390_parse_cpu): Support alternate arch
strings.
* doc/as.texinfo: Document new arch strings.
* doc/c-s390.texi: Likewise.
GCC 6's ICF optimization pass is making the declaration of 'm1' and
'm2', on gdb.base/stap-probe.c, to be unified. However, this leads to
only one instance of the probe 'two' being created, which causes a
failure on the testsuite (which expects a multi-location breakpoint to
be inserted on the probe).
This patch fixes this failure by declaring a dummy variable on 'm1',
and using it as an argument to m1's version of probe 'two'. Since we
do not care about the contents of the functions nor about the
arguments of each probe 'two', this is OK.
gdb/testsuite/ChangeLog:
2016-09-11 Sergio Durigan Junior <sergiodj@redhat.com>
Jan Kratochvil <jan.kratochvil@redhat.com>
* gdb.base/stap-probe.c (m1): New variable 'dummy', necessary to
make m1's definition to be different from m2's. Use 'dummy' as an
argument for probe 'two'.
On various GNU Elf architectures, including AArch64, ARM, s390/s390x,
ppc32/64, and sparc32/64, the dynamic loader passes HWCAP as a parameter
to each ifunc resolver. Currently there is an open glibc Bugzilla that
requests this to be generalized to all architectures:
https://sourceware.org/bugzilla/show_bug.cgi?id=19766
And various ifunc resolvers already rely on receiving HWCAP. Currently
GDB always calls an ifunc resolver without any arguments; thus the
resolver may receive garbage, and based on that, the resolver may decide
to return a function that is not suited for the given platform.
This patch always passes HWCAP to ifunc resolvers, even on systems where
the dynamic loader currently behaves otherwise. The rationale is
that (1) the dynamic loader may get adjusted on those systems as well in
the future; (2) passing an unused argument should not cause a problem
with existing resolvers; and (3) the logic is much simpler without such
a distinction.
gdb/ChangeLog:
* elfread.c (auxv.h): New include.
(elf_gnu_ifunc_resolve_addr): Pass HWCAP to ifunc resolver.
gdb/testsuite/ChangeLog:
* gdb.base/gnu-ifunc-lib.c (resolver_hwcap): New external
variable declaration.
(gnu_ifunc): Add parameter hwcap. Store it in resolver_hwcap.
* gdb.base/gnu-ifunc.c (resolver_hwcap): New global variable.
* gdb.base/gnu-ifunc.exp: Add test to verify that the resolver
received HWCAP as its argument.
I happened to notice a few unneeded casts in remote.c. In some cases
these are no-ops, and in others these cast away const, but in a context
where this is not needed.
I'm checking this in under the obvious rule.
Tested by rebuilding on x86-64 Fedora 24.
2016-09-08 Tom Tromey <tom@tromey.com>
* remote.c (remote_notif_stop_ack, remote_wait_as)
(show_remote_cmd): Remove unneeded casts.
Originally only Pentium integer instructions are allowed for IAMCU.
This patch removes such a restriction. For example, 387 and SSE2
instructions can be enabled by passing "-march=iamcu+sse2+387" to
assembler.
gas/
* config/tc-i386.c (valid_iamcu_cpu_flags): Removed.
(set_cpu_arch): Updated.
(md_parse_option): Likewise.
* testsuite/gas/i386/i386.exp: Run iamcu-4 and iamcu-5. Remove
iamcu-inval-2 and iamcu-inval-3.
* testsuite/gas/i386/iamcu-4.d: New file.
* testsuite/gas/i386/iamcu-4.s: Likewise.
* testsuite/gas/i386/iamcu-5.d: Likewise.
* testsuite/gas/i386/iamcu-5.s: Likewise.
* testsuite/gas/i386/iamcu-inval-2.l: Removed.
* testsuite/gas/i386/iamcu-inval-2.s: Likewise.
* testsuite/gas/i386/iamcu-inval-3.l: Likewise.
* testsuite/gas/i386/iamcu-inval-3.s: Likewise.
opcodes/
* i386-gen.c (cpu_flag_init): Remove CPU_IAMCU_COMPAT_FLAGS.
* i386-init.h: Regenerated.
2016-09-07 Richard Earnshaw <rearnsha@arm.com>
* opcode/arm.h (ARM_ARCH_V8A_CRC): New architecture.
2016-09-07 Richard Earnshaw <rearnsha@arm.com>
* config/tc-arm.c ((arm_cpus): Use ARM_ARCH_V8A_CRC for all
ARMv8-A CPUs except xgene1.
I noticed that if input is already pending on the new-ui TTY, gdb
internal-errors.
E.g., create /dev/pts/2, and type anything there (even just <return>
is sufficient).
Now start GDB creating a new UI on that TTY, while at the same time,
running a synchronous execution command. Something like:
$ gdb program -ex "new-ui console /dev/pts/2" -ex "start"
Back on /dev/pts/2, we get:
(gdb) .../src/gdb/event-top.c:360: internal-error: double prompt
A problem internal to GDB has been detected,
further debugging may prove unreliable.
While the main UI was waiting for "start" to finish, gdb kepts pumping
events, including the input fd of the extra console. The problem is
that stdin_event_handler doesn't restore the current UI back to what
it was, assuming that it's only ever called from the top level event
loop. However, in this case, it's being called from the nested event
loop from within maybe_wait_sync_command_done.
When finally the "start" command is done, we reach the code that
prints the prompt in the main UI, just before starting the main event
loop. Since now the current UI is pointing at the extra console (by
mistake), we find ourselves printing a double prompt on the extra
console. This is caught by the assertion that fails, as shown above.
Since other event handlers also don't restore the UI (e.g., signal
event handlers), I think it's better if whatever is pumping events to
take care to restore the UI, if it cares. That's what this patch
does. New test included.
gdb/ChangeLog:
2016-09-06 Pedro Alves <palves@redhat.com>
* top.c (wait_sync_command_done): Don't assume current_ui doesn't
change across events. Restore the current UI before returning.
(gdb_readline_wrapper): Restore the current UI before returning.
gdb/testsuite/ChangeLog:
2016-09-06 Pedro Alves <palves@redhat.com>
* gdb.base/new-ui-pending-input.c: New file.
* gdb.base/new-ui-pending-input.exp: New file.
* gdb.exp (clear_gdb_spawn_id): New procedure.
(with_spawn_id): Check whether gdb_spawn_id exists before
referencing it. If gdb_spawn_id didn't exist on entry, clear it
on exit.
Just a tidy, no functional changes.
gdb/ChangeLog:
2016-09-06 Pedro Alves <palves@redhat.com>
* event-top.c (restore_ui_cleanup): Now static.
(make_cleanup_restore_current_ui): New function.
(switch_thru_all_uis_init): Use it.
* infcall.c (call_thread_fsm_should_stop): Use it.
* infrun.c (fetch_inferior_event): Use it.
* top.c (new_ui_command): Use it.
* top.h (restore_ui_cleanup): Delete declaration.
(make_cleanup_restore_current_ui): New declaration.
We can resolve size relocation against symbol which needs copy relocation
when building executable.
bfd/
PR ld/20550
* elf64-x86-64.c (elf_x86_64_relocate_section): Resolve size
relocation with copy relocation when building executable.
ld/
PR ld/20550
* testsuite/ld-x86-64/pr20550a.s: New file.
* testsuite/ld-x86-64/pr20550b.s: Likewise.
* testsuite/ld-x86-64/x86-64.exp (x86_64tests): Add tests for
PR ld/20550.
* Makefile.am (CFLAGS_FOR_TARGET): Define as a copy of CFLAGS but
without any sanitization options.
(CXXFLAGS_FOR_TARGET): Define as a copy of CXXFLAGS but without
any sanitization options.
(check-DEJAGNU): Pass CFLAGS_FOR_TARGET and CXXFLAGS_FOR_TARGET
as CFLAGS and CXXFLAGS respectively.
Now that all the prerequisites are in place, this commit finally adds support
for handling the __float128 type on Intel and Power, by providing appropriate
platform-specific versions of the floatformat_for_type callback.
Since at this point we do not yet have any indication in the debug info to
distinguish different floating-point formats of the same length, we simply
use the type name as hint. Types named "__float128" get the IEEE format.
In addition to handling "__float128" itself, we also recognize "_Float128"
and (on Power) "_Float64x", as well as the complex versions of those.
(As pointed out by Joseph Myers, starting with GCC 7, __float128 is just
a typedef for _Float128 -- but it's good to handle this anyway.)
A new test case does some simple verification that the format is decoded
correctly, using both __float128 and "long double" to make sure using both
in the same file still works. Another new test verifies handling of the
_FloatN and _FloatNx types supported by GCC 7, as well as the complex
versions of those types.
Note that this still only supports basic format decoding and encoding.
We do not yet support the GNU extension 'g' suffix for __float128 constants.
In addition, since all *arithmetic* on floating-point values is still
performed in native host "long double" arithmetic, if that format is not
able to encode all target __float128 values, we may get incorrect results.
(To fix this would require implementing fully synthetic target floating-
point arithmetic along the lines of GCC's real.c, presumably using MPFR.)
gdb/ChangeLog:
* i386-tdep.c (i386_floatformat_for_type): New function.
(i386_gdbarch_init): Install it.
* ppc-linux-tdep.c (ppc_floatformat_for_type): New function.
(ppc_linux_init_abi): Install it.
gdb/testsuite/ChangeLog:
* gdb.base/float128.c: New file.
* gdb.base/float128.exp: Likewise.
* gdb.base/floatn.c: Likewise.
* gdb.base/floatn.exp: Likewise.
Signed-off-by: Ulrich Weigand <ulrich.weigand@de.ibm.com>
At this point, all TYPE_CODE_FLT types carry their floating-point format,
except for those creating from reading DWARF or stabs debug info. Those
will be addressed by this commit.
The main issue here is that we actually have to determine which floating-
point format to use. Currently, we only have the type length as input
to this decision. In the future, we may hopefully get --at least in
DWARF-- additional information to help disambiguate multiple different
formats of the same length. For now, we can still look at the type name
as a hint.
This decision logic is encapsulated in a gdbarch callback to allow
platform-specific overrides. The default implementation use the same
logic (compare type length against the various gdbarch_..._bit sizes)
that is currently implemented in floatformat_from_length.
With this commit, all platforms still use the default logic, so there
should be no actual change in behavior. A follow-on commit will add
support for __float128 on Intel and Power.
Once dwarf2read.c and stabsread.c make use of the new callback to
determine floating-point formats, we're now sure every TYPE_CODE_FLT
type will always carry its format. The commit therefore adds asserts
to verify_floatformat to ensure new code will continue to always
provide formats, and removes the code in floatformat_from_type that
used to handle types with a NULL TYPE_FLOATFORMAT.
gdb/ChangeLog:
* gdbarch.sh (floatformat_for_type): New gdbarch callback.
* gdbarch.h, gdbarch.c: Re-generate.
* arch-utils.h (default_floatformat_for_type): New prototype.
* arch-utils.c (default_floatformat_for_type): New function.
* doublest.c (floatformat_from_length): Remove.
(floatformat_from_type): Assume TYPE_FLOATFORMAT is non-NULL.
* gdbtypes.c (verify_floatformat): Require non-NULL format.
* dwarf2read.c (dwarf2_init_float_type): New function.
(read_base_type): Use it.
* stabsread.c (dbx_init_float_type): New function.
(read_sun_floating_type): Use it.
(read_range_type): Likewise.
Signed-off-by: Ulrich Weigand <ulrich.weigand@de.ibm.com>
Many callers of init_float_type and arch_float_type still pass a NULL
floatformat. This commit changes those callers where the floatformat
that is supposed to be use is obvious. There are two categories where
this is the case:
- A number of built-in types are intended to match the platform ABI
floating-point types (i.e. types that use gdbarch_float_bit etc.).
Those places should use the platform ABI floating-point formats
defined via gdbarch_float_format etc.
- A number of language built-in types should simply use IEEE floating-
point formats, since the language actually defines that this is the
format that must be used to implement floating-point types for this
language. (This affects Java, Go, and Rust.) The same applies for
to the predefined "RS/6000" stabs floating-point built-in types.
gdb/ChangeLog:
* ada-lang.c (ada_language_arch_info): Use gdbarch-provided
platform ABI floating-point formats for built-in types.
* d-lang.c (build_d_types): Likewise.
* f-lang.c (build_fortran_types): Likewise.
* m2-lang.c (build_m2_types): Likewise.
* mdebugread.c (basic_type): Likewise.
* go-lang.c (build_go_types): Use IEEE floating-point formats
for language built-in types as mandanted by the language.
* jv-lang.c (build_java_types): Likewise.
* rust-lang.c (rust_language_arch_info): Likewise.
* stabsread.c (rs6000_builtin_type): Likewise.
Signed-off-by: Ulrich Weigand <ulrich.weigand@de.ibm.com>
init_type (and arch_integer_type) currently use a special hack to set the
TYPE_NOSIGN flag if the type name is exactly "char". This commit moves the
hack up to the callers of those routines.
The special case currently can hit only for types created from dwarf2read,
but read_base_type actually implements the "char" check itself, so it is
redundant to do it in init_type as well. (Note that stabsread.c and the
other type readers always pass NULL as name to init_type, so the special
case can never hit for those.)
A few other cases create pre-definded types with a hard-coded name of "char";
the commit simply moves setting the TYPE_NOSIGN flag to those places.
No functional change intended.
gdb/ChangeLog:
* gdbtypes.c (init_type): Remove "char" special case.
(arch_integer_type): Likewise.
(gdbtypes_post_init): Set TYPE_NOSIGN for "char" type.
(objfile_type): Likewise.
* mdebugread.c (basic_type): Likewise.
* stabsread.c (rs6000_builtin_type): Likewise.
Signed-off-by: Ulrich Weigand <ulrich.weigand@de.ibm.com>
Now that init_type no longer takes a FLAGS argument, there is no user of
the TYPE_FLAGS_... enum values left. This commit removes them (and all
references to them in comments as well).
This is mostly a no-op, except for a change to the Python type printer,
which attempted to use them before. (As best as I can tell, this wasn't
really needed anyway, since it was only used to pretty-print type
*instance* flags, which only use the instance flags.)
gdb/ChangeLog:
* gdbtypes.h (enum type_flag_value): Remove.
Remove references to TYPE_FLAG_... in comments throughout.
* gdbtypes.c (recursive_dump_type): Do not print TYPE_FLAG_...
flags, print the corresponding TYPE_... access macro names.
Remove references to TYPE_FLAG_... in comments throughout.
* infcall.c: Remove references to TYPE_FLAG_... in comments.
* valprint.c: Likewise.
* gdb-gdb.py (class TypeFlag): No longer consider TYPE_FLAG_...
values, only TYPE_INSTANCE_FLAG_... values.
(class TypeFlagsPrinter): Likewise.
gdb/testsuite/ChangeLog:
* gdb.cp/hang.exp: Remove reference to TYPE_FLAG_STUB in comment.
Signed-off-by: Ulrich Weigand <ulrich.weigand@de.ibm.com>
This adds a number of helper routines for creating objfile-owned types;
these correspond 1:1 to the already existing helper routines for creating
gdbarch-owned types, and are intended to be used instead of init_type.
A shared fragment of init_float_type and arch_float_type is extracted into
a separate subroutine verify_subroutine.
The commit also brings the interface of init_type in line with the one for
arch_type. In particular, this means removing the FLAGS argument; callers
now set the required flags directly. (Since most callers use the new
helper routines, very few callers actually need to set any additional
flags directly any more.)
Note that this means all the TYPE_FLAGS_... defined are no longer needed
anywhere; they will be removed by a follow-on commit.
All users of init_type are changed to use on of the new helpers where
possible. No functional change intended.
gdb/ChangeLog:
* gdbtypes.h (init_type): Remove FLAGS argument. Move OBJFILE
argument to first position.
(init_integer_type): New prototype.
(init_character_type): Likewise.
(init_boolean_type): Likewise.
(init_float_type): Likewise.
(init_decfloat_type): Likewise.
(init_complex_type): Likewise.
(init_pointer_type): Likewise.
* gdbtypes.c (verify_floatflormat): New function.
(init_type): Remove FLAGS argument and processing. Move OBJFILE
argument to first position.
(init_integer_type): New function.
(init_character_type): Likewise.
(init_boolean_type): Likewise.
(init_float_type): Likewise.
(init_decfloat_type): Likewise.
(init_complex_type): Likewise.
(init_pointer_type): Likewise.
(arch_float_type): Use verify_floatflormat.
(objfile_type): Use init_..._type helpers instead of calling
init_type directly.
* dwarf2read.c (fixup_go_packaging): Update to changed init_type
prototype.
(read_namespace_type): Likewise.
(read_module_type): Likewise.
(read_typedef): Likewise.
(read_unspecified_type): Likewise.
(build_error_marker_type): Likewise.
(read_base_type): Use init_..._type helpers.
* mdebugread.c (basic_type): Use init_..._type helpers.
(parse_type): Update to changed init_type prototype.
(cross_ref): Likewise.
* stabsread.c (rs6000_builtin_type): Use init_..._type helpers.
(read_sun_builtin_type): Likewise.
(read_sun_floating_type): Likewise.
(read_range_type): Likewise. Also update to changed init_type
prototype.
Signed-off-by: Ulrich Weigand <ulrich.weigand@de.ibm.com>
gdbtypes provides a number of helper routines that can be called instead of
using arch_type directly to create a type of a particular kind. This patch
adds two additional such routines that have been missing so far, to allow
creation of TYPE_CODE_DECFLOAT and TYPE_CODE_POINTER types.
The patch also changes a number of places to use the new helper routines
instead of calling arch_type directly. No functional change intended.
gdb/ChangeLog:
* gdbtypes.h (arch_decfloat_type): New prototype.
(arch_pointer_type): Likewise.
* gdbtypes.c (arch_decfloat_type): New function.
(arch_pointer_type): Likewise.
(gdbtypes_post_init): Use arch_decfloat_type.
* avr-tdep.c (avr_gdbarch_init): Use arch_pointer_type.
* ft32-tdep.c (ft32_gdbarch_init): Likewise.
* m32c-tdep.c (make_types): Likewise.
* rl78-tdep.c (rl78_gdbarch_init): Likewise.
Signed-off-by: Ulrich Weigand <ulrich.weigand@de.ibm.com>
A type's TYPE_SPECIFIC_FIELD is supposed to be initialized as appropriate
for the type code. This does happen if the type is created via init_type,
but not if it created via arch_type.
Fixed by extracting the initialization logic into a new set_type_code
routine, which is then called from both places.
gdb/ChangeLog:
* gdbtypes.c (set_type_code): New function.
(init_type, arch_type): Use it.
Signed-off-by: Ulrich Weigand <ulrich.weigand@de.ibm.com>
This fixes a bug introduced by a wrong replacement here:
https://sourceware.org/ml/gdb-patches/2007-06/msg00196.html
The Ada "long_long_float" type is supposed to correspond to the
platform ABI long double type, not double.
gdb/ChangeLog:
* ada-lang.c (ada_language_arch_info): Use gdbarch_long_double_bit
instead of gdbarch_double_bit for "long_long_float".
Signed-off-by: Ulrich Weigand <ulrich.weigand@de.ibm.com>
Prior to the patch, addends for relocs were being adjusted even if
they went beyond an alignment boundary. This is wrong - to
preserve alignment constraints, the relaxation logic adds as many padding
bytes at the alignment boundary as was deleted, so addends beyond the
boundary should not be adjusted. avr-prop-7.s reproduces this
scenario.
Also, prior to this patch, the relaxation logic assumed that the addr
parameter pointed to the middle of the instruction to be deleted, and
that addr - count would therefore be the shrinked instruction's
address. This is true when actually shrinking instructions.
The alignment constraints handling logic also invokes the same logic
though, with addr as the starting offset of padding bytes and
with count as the number of bytes to be deleted. Calculating the
shrinked insn's address as addr - count is obviously wrong in this
case - that offset would point to count bytes before the last
non-padded byte. avr-prop-8.s reproduces this scenario.
To fix scenario 1, the patch adds an additional check to ensure reloc addends
aren't adjusted if they cross a shrink boundary. The shrink boundary
is either the section size or an alignment boundary. Addends pointing
at an alignment boundary don't need to be adjusted, as padding would
occur and keep the boundary the same. Addends pointing at section size
need to be adjusted though, as no padding occurs and the section size
itself would get decremented. The patch records whether padding
occured (did_pad) and uses that to detect and handle this condition.
To fix scenario 2, the patch adds an additional parameter
(delete_shrinks_insn) to elf32_avr_relax_delete_bytes to distinguish
instruction bytes deletion from padding bytes deletion. It then uses that to
correctly set shrinked_insn_address.
bfd/ChangeLog:
2016-09-02 Senthil Kumar Selvaraj <senthil_kumar.selvaraj@atmel.com>
PR ld/20545
* elf32-avr.c (elf32_avr_relax_delete_bytes): Add parameter
delete_shrinks_insn. Modify computation of shrinked_insn_address.
Compute shrink_boundary and adjust addend only if
addend_within_shrink_boundary.
(elf32_avr_relax_section): Modify calls to
elf32_avr_relax_delete_bytes to pass extra parameter.
ld/ChangeLog:
2016-09-02 Senthil Kumar Selvaraj <senthil_kumar.selvaraj@atmel.com>
PR ld/20545
* testsuite/ld-avr/avr-prop-7.d: New test.
* testsuite/ld-avr/avr-prop-7.s: New test.
* testsuite/ld-avr/avr-prop-8.d: New test.
* testsuite/ld-avr/avr-prop-8.s: New test.
This removes all support for building gdb & gdbserver with a C
compiler from gdb & gdbserver's build machinery.
gdb/ChangeLog:
2016-09-05 Pedro Alves <palves@redhat.com>
* NEWS: Mention that a C++ compiler is now required.
* Makefile.in (COMPILER, COMPILER_CFLAGS): Remove.
(COMPILE.pre, CC_LD): Use CXX directly.
(INTERNAL_CFLAGS_BASE): Use CXXFLAGS directly.
* acinclude.m4: Don't include build-with-cxx.m4.
* build-with-cxx.m4: Delete file.
* configure.ac: Remove GDB_AC_BUILD_WITH_CXX call.
* warning.m4: Assume $enable_build_with_cxx is yes.
* configure: Regenerate.
gdb/gdbserver/ChangeLog:
2016-09-05 Pedro Alves <palves@redhat.com>
* Makefile.in (COMPILER, COMPILER_CFLAGS): Remove.
(COMPILE.pre, CC_LD): Use CXX directly.
(INTERNAL_CFLAGS_BASE): Use CXXFLAGS directly.
* acinclude.m4: Don't include build-with-cxx.m4.
* configure.ac: Remove GDB_AC_BUILD_WITH_CXX call.
* configure: Regenerate.
This fixes the problem exercised by Kevin's test at:
https://sourceware.org/ml/gdb-patches/2016-08/msg00216.html
This was originally exposed by the OpenJDK Python-based unwinder.
If an unwinder attempts to call parse_and_eval from within its
sniffing method, GDB's unwinding machinery enters infinite recursion.
However, parse_and_eval is a pretty reasonable thing to call, because
Python/Scheme-based unwinders will often need to read globals out of
inferior memory. The recursion happens because:
- get_current_frame() is called soon after the target stops.
- current_frame is NULL, and so we unwind it from the sentinel frame
(which is special and has level == -1).
- We reach get_prev_frame_if_no_cycle, which does cycle detection
based on frame id, and thus tries to compute the frame id of the new
frame.
- Frame id computation requires an unwinder, so we go through all
unwinder sniffers trying to see if one accepts the new frame (the
current frame).
- the unwinder's sniffer calls parse_and_eval().
- parse_and_eval depends on the selected frame/block, and if not set
yet, the selected frame is set to the current frame.
- get_current_frame () is called again. current_frame is still NULL,
so ...
- recurse forever.
In Kevin's test at:
https://sourceware.org/ml/gdb-patches/2016-08/msg00216.html
gdb doesn't recurse forever simply because the Python unwinder
contains code to detect and stop the recursion itself. However, GDB
goes downhill from here, e.g., by showing the sentinel frame as
current frame (note the -1):
Breakpoint 1, ccc (arg=<unavailable>) at py-recurse-unwind.c:23
23 }
(gdb) bt
#-1 ccc (arg=<unavailable>) at py-recurse-unwind.c:23
Backtrace stopped: previous frame identical to this frame (corrupt stack?)
That "-1" frame level comes from this:
if (catch_exceptions (current_uiout, unwind_to_current_frame,
sentinel_frame, RETURN_MASK_ERROR) != 0)
{
/* Oops! Fake a current frame? Is this useful? It has a PC
of zero, for instance. */
current_frame = sentinel_frame;
}
which is bogus. It's never correct to set the current frame to the
sentinel frame. The only reason this has survived so long is that
getting here normally indicates something wrong has already happened
before and we fix that. And this case is no exception -- it doesn't
really matter how precisely we managed to get to that bogus code (it
has to do with the the stash), because anything after recursion
happens is going to be invalid.
So the fix is to avoid the recursion in the first place.
Observations:
#1 - The recursion happens because we try to do cycle detection from
within get_prev_frame_if_no_cycle. That requires computing the
frame id of the frame being unwound, and that itself requires
calling into the unwinders.
#2 - But, the first time we're unwinding from the sentinel frame,
when we reach get_prev_frame_if_no_cycle, there's no frame chain
at all yet:
- current_frame is NULL.
- the frame stash is empty.
Thus, there's really no need to do cycle detection the first time we
reach get_prev_frame_if_no_cycle, when building the current frame.
So we can break the recursion by making get_current_frame call a
simplified version of get_prev_frame_if_no_cycle that results in
setting the current_frame global _before_ computing the current
frame's id.
But, we can go a little bit further. As there's really no reason
anymore to compute the current frame's frame id immediately, we can
defer computing it to when some caller of get_current_frame might need
it. This was actually how the frame id was computed for all frames
before the stash-based cycle detection was added. So in a way, this
patch reintroduces the lazy frame id computation, but unlike before,
only for the case of the current frame, which turns out to be special.
This lazyness, however, requires adjusting
gdb.python/py-unwind-maint.exp, because that assumes unwinders are
immediately called as side effect of some commands. I didn't see a
need to preserve the behavior expected by that test (all it would take
is call get_frame_id inside get_current_frame), so I adjusted the
test.
gdb/ChangeLog:
2016-09-05 Pedro Alves <palves@redhat.com>
PR backtrace/19927
* frame.c (get_frame_id): Compute the frame id if not computed
yet.
(unwind_to_current_frame): Delete.
(get_current_frame): Use get_prev_frame_always_1 to get the
current frame and assert that that always succeeds.
(get_prev_frame_if_no_cycle): Skip cycle detection if returning
the current frame.
gdb/testsuite/ChangeLog:
2016-09-05 Pedro Alves <palves@redhat.com>
PR backtrace/19927
* gdb.python/py-unwind-maint.exp: Adjust tests to not expect that
unwinders are immediately called as side effect of "source" or
"disable unwinder" commands.
* gdb.python/py-recurse-unwind.exp: Remove setup_kfail calls.
Currently gdb supports DW_OP_GNU_push_tls_address, but not
DW_OP_form_tls_address. I think it would be better if the toolchain
as a whole moved to using the standard opcode, and the prerequisite to
this is getting gdb to recognize it.
GCC can sometimes emit DW_OP_form_tls_address for emultls targets. As
far as I know, nobody has ever tried this with gdb (since it wouldn't
work at all).
I don't think there's a major drawback to using a single opcode for
all targets, because computing the location of a thread-local is
already target specific.
This is PR gdb/11616.
I don't know how to write a test case for this; though it's worth
noting that there aren't explicit tests for DW_OP_GNU_push_tls_address
either -- and if I change GCC, these paths will be tested to the same
extent they are now.
2016-09-02 Tom Tromey <tom@tromey.com>
PR gdb/11616:
* dwarf2read.c (decode_locdesc): Handle DW_OP_form_tls_address.
* dwarf2loc.c (dwarf2_compile_expr_to_ax): Handle
DW_OP_form_tls_address.
(locexpr_describe_location_piece): Likewise.
* dwarf2expr.h (struct dwarf_expr_context_funcs): Update comment.
* dwarf2expr.c (execute_stack_op): Handle DW_OP_form_tls_address.
(ctx_no_get_tls_address): Mention DW_OP_form_tls_address.
* compile/compile-loc2c.c (struct insn_info): Update comment.
(compute_stack_depth_worker): Handle DW_OP_form_tls_address.
reinsert_breakpoint is used for software single step, so it is more
clear to rename it to single_step_breakpoint. This was pointed out in
the review https://sourceware.org/ml/gdb-patches/2016-05/msg00429.html
I don't rename "other_breakpoint" in this patch.
gdb/gdbserver:
2016-09-02 Yao Qi <yao.qi@linaro.org>
* linux-low.c: Replace "reinsert_breakpoints" with
"single_step_breakpoints". Replace "reinsert breakpoints"
with "single-step breakpoints".
* mem-break.c: Likewise.
* mem-break.h: Likewise.
return-nodebug.exp does the test for various types, but we shouldn't
test with floating point type if gdb_skip_float_test returns true.
gdb/testsuite:
2016-09-02 Yao Qi <yao.qi@linaro.org>
* gdb.base/return-nodebug.exp: Skip the test if skip_float_test
is true and $type is "float" or "double".
We recently found a ARM kernel ptrace bug
http://lists.infradead.org/pipermail/linux-arm-kernel/2016-May/431962.html
Details can be found in the comment in gdb_skip_float_test. We can
skip floating point tests if the kernel bug is detected.
This patch adds more code in gdb_skip_float_test to detect the broken
ptrace on arm-linux. Such detection should be done at the beginning
of the test, because it starts a fresh GDB, so change the test cases
to invoke gdb_skip_float_test at the beginning of test, and use its
return value afterwards.
Since gdb_skip_float_test becomes a gdb_caching_proc, so it can't
have an argument, this patch also removes argument "msg", which isn't
useful.
gdb/testsuite:
2016-09-02 Yao Qi <yao.qi@linaro.org>
* gdb.arch/arm-neon.exp: Skip it if gdb_skip_float_test returns
true.
* gdb.base/call-ar-st.exp: Invoke gdb_skip_float_test.
* gdb.base/call-rt-st.exp: Likewise.
* gdb.base/call-sc.exp: Invoke gdb_skip_float_test and use its
return value instead of gdb,skip_float_test.
* gdb.base/callfuncs.exp: Invoke gdb_skip_float_test.
(do_function_calls): Use its return value instead of
gdb,skip_float_test.
* gdb.base/finish.exp: Likewise.
* gdb.base/funcargs.exp: Likewise.
* gdb.base/return.exp: Likewise.
* gdb.base/return2.exp: Likewise.
* gdb.base/varargs.exp: Likewise.
* lib/gdb.exp (gdb_skip_float_test): Change it to
gdb_caching_proc. Detect the broken ptrace on arm-linux.
This commit moves the target_wait prototype from the GDB-specific
target.h header to the common target/target.h header. Then, it
creates a compatible implementation of target_wait on gdbserver using
the_target->wait, and adjusts the (only) caller (mywait function).
Pretty straightforward, no regressions introduced.
gdb/gdbserver/ChangeLog:
2016-09-01 Sergio Durigan Junior <sergiodj@redhat.com>
* target.c (mywait): Call target_wait instead of
the_target->wait.
(target_wait): New function.
gdb/ChangeLog:
2016-09-01 Sergio Durigan Junior <sergiodj@redhat.com>
* target.c (target_wait): Mention that the function's prototype
can be found at target/target.h.
* target.h (target_wait): Move prototype from here...
* target/target.h (target_wait): ... to here.
This commit implements a new function, target_continue, on top of the
target_resume function. Then, it replaces all calls to target_resume
by calls to target_continue or to the already existing
target_continue_no_signal.
This is one of the (many) necessary steps needed to consolidate the
target interface between GDB and gdbserver. In particular, I am
interested in the impact this change will have on the unification of
the fork_inferior function (which I have been working on).
Tested on the BuildBot, no regressions introduced.
gdb/gdbserver/ChangeLog:
2016-09-31 Sergio Durigan Junior <sergiodj@redhat.com>
* server.c (start_inferior): New variable 'ptid'. Replace calls
to the_target->resume by target_continue{,_no_signal}, depending
on the case.
* target.c (target_stop_and_wait): Call target_continue_no_signal
instead of the_target->resume.
(target_continue): New function.
gdb/ChangeLog:
2016-09-31 Sergio Durigan Junior <sergiodj@redhat.com>
* fork-child.c (startup_inferior): Replace calls to target_resume
by target_continue{,_no_signal}, depending on the case.
* linux-nat.c (cleanup_target_stop): Call
target_continue_no_signal instead of target_resume.
* procfs.c (procfs_wait): Likewise.
* target.c (target_continue): New function.
* target/target.h (target_continue): New prototype.
bfd/
* elf32-arm.c (cmse_entry_fct_p): Store instruction encoding in an
array of bytes and use bfd_get_16 to interpret its encoding according
to endianness of target.
Fixes a gdb segfault if a section named .opd is found in ELFv2 binaries.
* elf64-ppc.c (synthetic_opd): New static var.
(compare_symbols): Don't treat symbols in .opd specially for ELFv2.
(ppc64_elf_get_synthetic_symtab): Likewise. Comment.
