MFAR_EL3 is a read/write register, but was incorrectly marked as
read-only
[https://developer.arm.com/documentation/ddi0601/2021-09/AArch64-Registers/MFAR-EL3--PA-Fault-Address-Register?lang=en]
opcodes/
* aarch64-opc.c (aarch64_sys_regs): Mark mfar_el3 as read-write.
gas/
* testsuite/gas/aarch64/rme.s: Test writing to mfar_el3.
* testsuite/gas/aarch64/rme.d: Update accordingly.
* testsuite/gas/aarch64/rme-invalid.s: Delete.
* testsuite/gas/aarch64/rme-invalid.l: Likewise.
* testsuite/gas/aarch64/rme-invalid.d: Likewise.
We were incorrectly allowing writes to PMSIDR_EL1, which is
a read-only register.
[https://developer.arm.com/documentation/ddi0595/2021-09/AArch64-Registers/PMSIDR-EL1--Sampling-Profiling-ID-Register?lang=en]
opcodes/
* aarch64-opc.c (aarch64_sys_regs): Make pmsidr_el1 as F_REG_READ.
gas/
* testsuite/gas/aarch64/msr.s: Remove write to pmsidr_el1.
* testsuite/gas/aarch64/msr.d: Update accordingly.
* testsuite/gas/aarch64/illegal-sysreg-2.s,
* testsuite/gas/aarch64/illegal-sysreg-2.d,
* testsuite/gas/aarch64/illegal-sysreg-2.l: New test.
There is a lot of overlap between the ETM and ETE system registers,
so some registers were listed twice.
Already tested by etm.[sd] and ete.[sd].
opcodes/
* aarch64-opc.c (aarch64_sys_regs): Combine ETE and ETM blocks
and remove redundant entries.
gas/
* testsuite/gas/aarch64/etm.s: Remove duplicated test.
* testsuite/gas/aarch64/etm.d: Update accordingly.
Previously we would not accept:
A .req B
if A happened to be the name of an instruction. Adding new
instructions could therefore invalidate existing register aliases.
I noticed this with a test that used "zero" as a register alias
for "xzr", where "zero" is now also the name of an SME instruction.
I don't have any evidence that "real" code is doing this, but it
seems at least plausible.
This patch switches things so that we check for register aliases
first. It might slow down parsing slightly, but the difference
is unlikely to be noticeable.
Things like:
b .req + 0
still work, since create_register_alias checks for " .req ",
and with the input scrubber, we'll only keep whitespace after
.req if it's followed by another name. If there's some valid
expression that I haven't thought about that is scrubbed to
" .req ", users could avoid the ambiguity by wrapping .req
in parentheses.
The new test for invalid aliases already passed. I just wanted
something to exercise the !dot condition.
I can't find a way of exercising the (existing) p == base condition,
but I'm not brave enough to say that it can never happen. If it does
happen, get_mnemonic_name would return an empty string.
gas/
* config/tc-aarch64.c (opcode_lookup): Move mnemonic extraction
code to...
(md_assemble): ...here. Check for register aliases first.
* testsuite/gas/aarch64/register_aliases.d,
testsuite/gas/aarch64/register_aliases.s: Test for a register
alias called "zero".
* testsuite/gas/aarch64/register_aliases_invalid.d,
testsuite/gas/aarch64/register_aliases_invalid.l,
testsuite/gas/aarch64/register_aliases_invalid.s: New test.
When running the gdb.python/py-arch.exp tests on a GDB built
against Python 2 I ran into some errors. The problem is that this
test script exercises the gdb.Architecture.integer_type method, and
this method uses 'p' as an argument format specifier in a call to
gdb_PyArg_ParseTupleAndKeywords.
Unfortunately this specified was only added in Python 3.3, so will
cause an error for earlier versions of Python.
This commit switches to use the 'O' specifier to collect a PyObject,
and then uses PyObject_IsTrue to convert the object to a boolean.
An earlier version of this patch incorrectly switched from using 'p'
to use 'i', however, it was pointed out during review that this would
cause some changes in behaviour, for example both of these will work
with 'p', but not with 'i':
gdb.selected_inferior().architecture().integer_type(32, None)
gdb.selected_inferior().architecture().integer_type(32, "foo")
The new approach of using 'O' works fine with these cases. I've added
some new tests to cover both of the above.
There should be no user visible changes after this commit.
When running test-case gdb.base/style.exp with a gdb build using
stub-termcap.c, we run into:
...
(gdb) PASS: gdb.base/style.exp: all styles enabled: frame when width=20
^M<et width 30^M
(gdb) FAIL: gdb.base/style.exp: all styles enabled: set width 30
...
The problem is that we're trying to issue the command "set width 30" while
width is set to 20, which causes horizontal scrolling.
Fix this by resetting the width to 0 before issuing the "set width 30"
command.
Tested on x86_64-linux.
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=24582
* dwarf.c (find_debug_info_for_offset): Use dwarf_vma type for
offsets, sizes and ranges.
(display_loc_list): Likewise. Also use print_dwarf_vma to print
the offset.
(display_loclists_list): Likewise.
(display_loc_list_dwo): Likewise.
(display_debug_str): Likewise.
(display_debug_aranges): Likewise.
(display_debug_ranges_list): Likewise.
(display_debug_rnglists_list): Likewise.
(display_debug_ranges): Likewise.
The gdb.python/py-inferior-leak.exp test makes use of the tracemalloc
module. When running the Python tests with a GDB built against Python
2 I ran into a test failure due to the tracemalloc module not being
available.
This commit adds a new helper function to lib/gdb-python.exp that
checks if a named module is available. Using this we can then skip
the py-inferior-leak.exp test when the tracemalloc module is not
available.
After this commit:
commit 76b43c9b5c
Date: Tue Oct 5 15:10:12 2021 +0100
gdb: improve error reporting from the disassembler
We started seeing FAILs in the gdb.base/all-architectures*.exp tests,
when running on a 32-bit ARM target, though I suspect running on any
target that compiles such that bfd_vma is 32-bits would also trigger
the failures.
The problem is that the test is expected GDB's disassembler to print
an error like this:
Cannot access memory at address 0x0
However, after the above commit we see an error like:
unknown disassembler error (error = -1)
The reason for this is this code in opcodes/i386-dis.c (in the
print_insn function):
if (address_mode == mode_64bit && sizeof (bfd_vma) < 8)
{
(*info->fprintf_func) (info->stream,
_("64-bit address is disabled"));
return -1;
}
This code effectively disallows us from ever disassembling 64-bit x86
code if we compiled GDB with a 32-bit bfd_vma. Notice we return
-1 (indicating a failure to disassemble), but never call the
memory_error_func callback.
Prior to the above commit GDB, when it received the -1 return value
would assume that a memory error had occurred and just print whatever
value happened to be in the memory error address variable, the default
value of 0 just happened to be fine because the test had asked GDB to
do this 'disassemble 0x0,+4'.
If we instead change the test to do 'disassemble 0x100,+4' then GDB
would (previously) have still reported:
Cannot access memory at address 0x0
which makes far less sense.
In this commit I propose to fix this issue by changing the test to
accept either the "Cannot access memory ..." string, or the newer
"unknown disassembler error ..." string. With this change done the
test now passes.
However, there is one weakness with this strategy; if GDB broke such
that we _always_ reported "unknown disassembler error ..." we would
never notice. This clearly would be bad. To avoid this issue I have
adjusted the all-architectures*.exp tests so that, when we disassemble
for the default architecture (the one selected by "auto") we _only_
expect to get the "Cannot access memory ..." error string.
[ Note: In an ideal world we should be able to disassemble any
architecture at all times. There's no reason why the 64-bit x86
disassembler requires a 64-bit bfd_vma, other than the code happens
to be written that way. We could rewrite the disassemble to not
have this requirement, but, I don't plan to do that any time soon. ]
Further, I have changed the all-architectures*.exp test so that we now
disassemble at address 0x100, this should avoid us being able to pass
by printing a default address of 0x0. I did originally change the
address we disassembled at to 0x4, however, some architectures,
e.g. ia64, have a default instruction alignment that is greater than
4, so would still round down to 0x0. I could have just picked 0x8 as
an address, but I figured that 0x100 was likely to satisfy most
architectures alignment requirements.
This commits adds a new sub-class of gdb.TargetConnection,
gdb.RemoteTargetConnection. This sub-class is created for all
'remote' and 'extended-remote' targets.
This new sub-class has one additional method over its base class,
'send_packet'. This new method is equivalent to the 'maint
packet' CLI command, it allows a custom packet to be sent to a remote
target.
The outgoing packet can either be a bytes object, or a Unicode string,
so long as the Unicode string contains only ASCII characters.
The result of calling RemoteTargetConnection.send_packet is a bytes
object containing the reply that came from the remote.
In a later commit I will add a Python API to access the 'maint packet'
functionality, that is, sending a user specified packet to the target.
To make implementing this easier, this commit refactors how this
command is currently implemented so that the packet_command function
is now global.
The new global send_remote_packet function takes an object that is an
implementation of an abstract interface. Two functions within this
interface are then called, one just before a packet is sent to the
remote target, and one when the reply has been received from the
remote target. Using an interface object in this way allows (1) for
the error checking to be done before the first callback is made, this
means we only print out what packet it being sent once we know we are
going to actually send it, and (2) we don't need to make a copy of the
reply if all we want to do is print it.
One user visible changes after this commit are the error
messages, which I've changed to be less 'maint packet' command
focused, this will make them (I hope) better for when
send_remote_packet can be called from Python code.
So: "command can only be used with remote target"
Becomes: "packets can only be sent to a remote target"
And: "remote-packet command requires packet text as argument"
Becomes: "a remote packet must not be empty"
Additionally, in this commit, I've added support for packet replies
that contain binary data. Before this commit, the code that printed
the reply treated the reply as a C string, it assumed that the string
only contained printable characters, and had a null character only at
the end.
One way to show the problem with this is if we try to read the auxv
data from a remote target, the auxv data is binary, so, before this
commit:
(gdb) target remote :54321
...
(gdb) maint packet qXfer:auxv:read::0,1000
sending: "qXfer:auxv:read::0,1000"
received: "l!"
(gdb)
And after this commit:
(gdb) target remote :54321
...
(gdb) maint packet qXfer:auxv:read::0,1000
sending: "qXfer:auxv:read::0,1000"
received: "l!\x00\x00\x00\x00\x00\x00\x00\x00\xf0\xfc\xf7\xff\x7f\x00\x00\x10\x00\x00\x00\x00\x00\x00\x00\xff\xf>
(gdb)
The binary contents of the reply are now printed as escaped hex.
This commit adds a new object type gdb.TargetConnection. This new
type represents a connection within GDB (a connection as displayed by
'info connections').
There's three ways to find a gdb.TargetConnection, there's a new
'gdb.connections()' function, which returns a list of all currently
active connections.
Or you can read the new 'connection' property on the gdb.Inferior
object type, this contains the connection for that inferior (or None
if the inferior has no connection, for example, it is exited).
Finally, there's a new gdb.events.connection_removed event registry,
this emits a new gdb.ConnectionEvent whenever a connection is removed
from GDB (this can happen when all inferiors using a connection exit,
though this is not always the case, depending on the connection type).
The gdb.ConnectionEvent has a 'connection' property, which is the
gdb.TargetConnection being removed from GDB.
The gdb.TargetConnection has an 'is_valid()' method. A connection
object becomes invalid when the underlying connection is removed from
GDB (as discussed above, this might be when all inferiors using a
connection exit, or it might be when the user explicitly replaces a
connection in GDB by issuing another 'target' command).
The gdb.TargetConnection has the following read-only properties:
'num': The number for this connection,
'type': e.g. 'native', 'remote', 'sim', etc
'description': The longer description as seen in the 'info
connections' command output.
'details': A string or None. Extra details for the connection, for
example, a remote connection's details might be
'hostname:port'.
According the rvv spec,
https://github.com/riscv/riscv-v-spec/blob/master/vtype-format.adoc
The bits of vtype immediate from 8 to (xlen - 1) should be reserved.
Therefore, we should also dump the vtype immediate as numbers, when
they are set over 8-bits. I think this is a bug that we used to support
vediv extension and use the bit 8 and 9 of vtype, but forgot to update
the behavior when removing the vediv.
Consider the testcases,
vsetvli a0, a1, 0x700 # the reserved bit 10, 9 and 8 are used.
vsetvli a0, a1, 0x400 # the reserved bit 10 is used.
vsetvli a0, a1, 0x300 # the reserved bit 9 and 8 are used.
vsetvli a0, a1, 0x100 # the reserved bit 8 is used.
vsetivli a0, 0xb, 0x300 # the reserved bit 9 and 8 are used.
vsetivli a0, 0xb, 0x100 # the reserved bit 8 is used.
The original objdump shows the following result,
0000000000000000 <.text>:
0: 7005f557 vsetvli a0,a1,1792
4: 4005f557 vsetvli a0,a1,1024
8: 3005f557 vsetvli a0,a1,e8,m1,tu,mu
c: 1005f557 vsetvli a0,a1,e8,m1,tu,mu
10: f005f557 vsetivli a0,11,e8,m1,tu,mu
14: d005f557 vsetivli a0,11,e8,m1,tu,mu
But in fact the correct result should be,
0000000000000000 <.text>:
0: 7005f557 vsetvli a0,a1,1792
4: 4005f557 vsetvli a0,a1,1024
8: 3005f557 vsetvli a0,a1,768
c: 1005f557 vsetvli a0,a1,256
10: f005f557 vsetivli a0,11,768
14: d005f557 vsetivli a0,11,256
gas/
* testsuite/gas/riscv/vector-insns.d: Added testcases to
test the reserved bit 8 to (xlen-1) of vtype.
* testsuite/gas/riscv/vector-insns.s: Likewise.
include/
* opcode/riscv.h: Removed OP_MASK_VTYPE_RES and OP_SH_VTYPE_RES,
since they are different for operand Vc and Vb.
opcodes/
* riscv-dis.c (print_insn_args): Updated imm_vtype_res to
extract the reserved immediate of vtype correctly.
Consider the following case,
vsetvli a0, a1, 0x4 # unrecognized vlmul
vsetvli a0, a1, 0x20 # unrecognized vsew
vsetivli a0, 0xb, 0x4 # unrecognized vlmul
vsetivli a0, 0xb, 0x20 # unrecognized vsew
For the current dis-assembler, we get the result,
0000000000000000 <.text>:
0: 0045f557 vsetvli a0,a1,e8,(null),tu,mu
4: 0205f557 vsetvli a0,a1,e128,m1,tu,mu
8: c045f557 vsetivli a0,11,e8,(null),tu,mu
c: c205f557 vsetivli a0,11,e128,m1,tu,mu
The vsew e128 and vlmul (null) are preserved according to the spec,
so dump these fields looks wrong. Consider that we are used to dump
the unrecognized csr as csr numbers directly, we should also dump
the whole vset[i]vli immediates as numbers, once the vsew or vlmul
is reserved. Therefore, following is what I expected,
0000000000000000 <.text>:
0: 0045f557 vsetvli a0,a1,4
4: 0205f557 vsetvli a0,a1,32
8: c045f557 vsetivli a0,11,4
c: c205f557 vsetivli a0,11,32
gas/
* testsuite/gas/riscv/vector-insns.d: Rewrite the vset[i]vli
testcases since we should dump the immediate as numbers once
the vsew or vlmul is reserved.
* testsuite/gas/riscv/vector-insns.s: Likewise.
opcodes/
* riscv-dis.c (print_insn_args): The reserved vsew and vlmul
are NULL string in the riscv_vsew and riscv_vlmul, so dump the
whole imm as numbers once one of them is NULL.
* riscv-opc.c (riscv_vsew): Set the reserved vsew to NULL.
(riscv_vlmul): Set the reserved vlmul to NULL.
The Rust compiler plans to change the encoding of a Rust 'char' type
to use DW_ATE_UTF. You can see the discussion here:
https://github.com/rust-lang/rust/pull/89887
However, this fails in gdb. I looked into this, and it turns out that
the handling of DW_ATE_UTF is currently fairly specific to C++. In
particular, the code here assumes the C++ type names, and it creates
an integer type.
This comes from commit 53e710acd ("GDB thinks char16_t and char32_t
are signed in C++"). The message says:
Both places need fixing. But since I couldn't tell why dwarf2read.c
needs to create a new type, I've made it use the per-arch built-in
types instead, so that the types are only created once per arch
instead of once per objfile. That seems to work fine.
... which is fine, but it seems to me that it's also correct to make a
new character type; and this approach is better because it preserves
the type name as well. This does use more memory, but first we
shouldn't be too concerned about the memory use of types coming from
debuginfo; and second, if we are, we should implement type interning
anyway.
Changing this code to use a character type revealed a couple of
oddities in the C/C++ handling of TYPE_CODE_CHAR. This patch fixes
these as well.
I filed PR rust/28637 for this issue, so that this patch can be
backported to the gdb 11 branch.
During debuginfod downloads, ctrl-c should result in the download
being cancelled and skipped. However in some cases, ctrl-c fails to
get delivered to gdb during downloading. This can result in downloads
being unskippable.
Fix this by ensuring that target_terminal::ours is in effect for the
duration of each download.
Co-authored-by: Tom de Vries <tdevries@suse.de>
https://sourceware.org/bugzilla/show_bug.cgi?id=27026#c3
PR28539 describes a segfault in lambda function search_one_symtab due to
psymbol_functions::expand_symtabs_matching calling expansion_notify with a
nullptr symtab:
...
struct compunit_symtab *symtab =
psymtab_to_symtab (objfile, ps);
if (expansion_notify != NULL)
if (!expansion_notify (symtab))
return false;
...
This happens as follows. The partial symtab ps is a dwarf2_include_psymtab
for some header file:
...
(gdb) p ps.filename
$5 = 0x64fcf80 "/usr/include/c++/11/bits/stl_construct.h"
...
The includer of ps is a shared symtab for a partial unit, with as user:
...
(gdb) p ps.includer().user.filename
$11 = 0x64fc9f0 \
"/usr/src/debug/llvm13-13.0.0-1.2.x86_64/tools/clang/lib/AST/Decl.cpp"
...
The call to psymtab_to_symtab expands the Decl.cpp symtab (and consequently
the shared symtab), but returns nullptr because:
...
struct dwarf2_include_psymtab : public partial_symtab
{
...
compunit_symtab *get_compunit_symtab (struct objfile *objfile) const override
{
return nullptr;
}
...
Fix this by returning the Decl.cpp symtab instead, which fixes the segfault
in the PR.
Tested on x86_64-linux.
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=28539
Proc lines contains a typo:
...
string_form { set $_line_string_form $value }
...
Remove the incorrect '$' in '$_line_string_form'.
Tested on x86_64-linux.
While debugging a problem in gdb.dwarf2/dw2-lines.exp, I realized that the
test-case generates all executables and associated temporary files using the
same filenames.
Fix this by adding a new proc prefix_id in lib/gdb.exp, and using it in the
test-case.
Tested on x86_64-linux.
When running test-case gdb.dwarf2/dw2-lines.exp with target board -unix/-m32,
we run into another instance of PR28383, where the dwarf assembler generates
64-bit relocations which are not supported by the 32-bit assembler:
...
dw2-lines-dw.S: Assembler messages:^M
outputs/gdb.dwarf2/dw2-lines/dw2-lines-dw.S:76: Error: \
cannot represent relocation type BFD_RELOC_64^M
...
Fix this by using _op_offset in _line_finalize_header.
Tested on x86_64-linux.
We'll rely on the toolchain probing to determine whether each arch's
tests can be run rather the current configure target. This allows
testing all of the ports in a multitarget configuration.
For now, we don't reformat the files entirely to make it easier to
review, and in case we need to make adjustments. Once this feels
like it's stable, we can flatten the code a bit by removing the if
statement entirely.
Break up the dejagnu logic so that we can parallelize the testsuite.
This takes a page from gcc & gdb where each .exp is run in isolation
instead of in serial.
For most targets, this doesn't make much of a difference as they only
have a single .exp. A few (like cris & frv) have multiple .exp though
and will see a bit of a speed up.
The real gain is when testing a multitarget build. This way we can
run all the targets in parallel and cut the execution time a bit.
On my system, it goes from ~155sec to ~100sec.
We can gain further speedups by splitting up some of the larger .exp
files into smaller groups. We'll do that in a followup though.
Leverage the new per-port toolchain settings to initialize the env
for eeach set of tests. This allows us to run all the tests in a
multitarget build if the user sets up the vars. If they don't, we
can still skip all the tests.
Gas does not support multitarget builds -- it still only supports
a single input & output format. ld is a bit better, but requires
manual flags to select the right output. This makes it impossible
to run the complete testsuite in a multitarget build.
To address this limitation, create a suite of FOR_TARGET variables
so these can be set to precompiled as & ld programs. It requires
a bit of setup ahead of time, but it's a one-time cost, and makes
running the full testsuite at once much easier.
Some of the core sim headers rely on the SIM_AC_OPTION_BITSIZE macro
which can change the size of core types. Since these haven't been
unified across ports, add checks to make sure they aren't accidentally
included when building for all ports. This caught the sim-load file
using poisoned headers that it didn't actually need.
Now that we've unified all the syscall tables, this file does not rely
on any port-specific settings, so move it up to building as part of the
common step so we only do it once in a multibuild.
Now that all ports have switched to target-newlib-* files, there's
no need for these files & generating things at build time. So punt
the logic and make target-newlib-syscall a hard requirement.
Use the new target-newlib-syscall module. This is needed to merge all
the architectures into a single build, and mcore has a custom syscall
table for its newlib/libgloss port.
Use the new target-newlib-syscall module. This is needed to merge all
the architectures into a single build, and riscv has a custom syscall
table for its newlib/libgloss port.
Use the new target-newlib-syscall module. This is needed to merge all
the architectures into a single build, and cr16 has a custom syscall
table for its newlib/libgloss port.
This allows cleaning up the syscall ifdef logic. We know these will
always exist now.
Use the new target-newlib-syscall module. This is needed to merge all
the architectures into a single build, and d10v has a custom syscall
table for its newlib/libgloss port.
This allows cleaning up the syscall ifdef logic. We know these will
always exist now.
Use the new target-newlib-syscall module. This is needed to merge all
the architectures into a single build, and sh has a custom syscall
table for its newlib/libgloss port.
Use the new target-newlib-syscall module. This is needed to merge all
the architectures into a single build, and v850 has a custom syscall
table for its newlib/libgloss port.
This allows cleaning up the syscall ifdef logic. We know these will
always exist now.
Use the new target-newlib-syscall.h to provide the target syscall
defines. These code paths are written specifically for the newlib
ABI rather than being generalized, so switching them to the defines
rather than trying to go through the dynamic callback conversion
seems like the best trade-off for now. Might have to reconsider
this in the future.
Like we just did for pulling out the errno map, pull out the syscall
maps into a dedicated common file. Most newlib ports are using the
same syscall map, but not all, which means we have to do a bit more
work to migrate.
This commit adds the maps and switches the ports using the common
default syscall table over to it. Ports using unique syscall tables
are still using the old targ-map.c logic.
Switching common ports over is easy by checking NL_TARGET, but the
ppc code needs a bit more cleanup here hence its larger diff.
Avoid use of TARGET_<syscall> defines and rely on the callback layers
to resolve these dynamically so we can support multiple syscall layers
instead of assuming the newlib/libgloss numbers all the time.
