aboutsummaryrefslogtreecommitdiff
AgeCommit message (Collapse)AuthorFilesLines
2023-02-03Automatic date update in version.inGDB Administrator1-1/+1
2023-02-02gdb: initial support for ROCm platform (AMDGPU) debuggingSimon Marchi18-138/+5155
This patch adds the foundation for GDB to be able to debug programs offloaded to AMD GPUs using the AMD ROCm platform [1]. The latest public release of the ROCm release at the time of writing is 5.4, so this is what this patch targets. The ROCm platform allows host programs to schedule bits of code for execution on GPUs or similar accelerators. The programs running on GPUs are typically referred to as `kernels` (not related to operating system kernels). Programs offloaded with the AMD ROCm platform can be written in the HIP language [2], OpenCL and OpenMP, but we're going to focus on HIP here. The HIP language consists of a C++ Runtime API and kernel language. Here's an example of a very simple HIP program: #include "hip/hip_runtime.h" #include <cassert> __global__ void do_an_addition (int a, int b, int *out) { *out = a + b; } int main () { int *result_ptr, result; /* Allocate memory for the device to write the result to. */ hipError_t error = hipMalloc (&result_ptr, sizeof (int)); assert (error == hipSuccess); /* Run `do_an_addition` on one workgroup containing one work item. */ do_an_addition<<<dim3(1), dim3(1), 0, 0>>> (1, 2, result_ptr); /* Copy result from device to host. Note that this acts as a synchronization point, waiting for the kernel dispatch to complete. */ error = hipMemcpyDtoH (&result, result_ptr, sizeof (int)); assert (error == hipSuccess); printf ("result is %d\n", result); assert (result == 3); return 0; } This program can be compiled with: $ hipcc simple.cpp -g -O0 -o simple ... where `hipcc` is the HIP compiler, shipped with ROCm releases. This generates an ELF binary for the host architecture, containing another ELF binary with the device code. The ELF for the device can be inspected with: $ roc-obj-ls simple 1 host-x86_64-unknown-linux file://simple#offset=8192&size=0 1 hipv4-amdgcn-amd-amdhsa--gfx906 file://simple#offset=8192&size=34216 $ roc-obj-extract 'file://simple#offset=8192&size=34216' $ file simple-offset8192-size34216.co simple-offset8192-size34216.co: ELF 64-bit LSB shared object, *unknown arch 0xe0* version 1, dynamically linked, with debug_info, not stripped ^ amcgcn architecture that my `file` doesn't know about ----´ Running the program gives the very unimpressive result: $ ./simple result is 3 While running, this host program has copied the device program into the GPU's memory and spawned an execution thread on it. The goal of this GDB port is to let the user debug host threads and these GPU threads simultaneously. Here's a sample session using a GDB with this patch applied: $ ./gdb -q -nx --data-directory=data-directory ./simple Reading symbols from ./simple... (gdb) break do_an_addition Function "do_an_addition" not defined. Make breakpoint pending on future shared library load? (y or [n]) y Breakpoint 1 (do_an_addition) pending. (gdb) r Starting program: /home/smarchi/build/binutils-gdb-amdgpu/gdb/simple [Thread debugging using libthread_db enabled] Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1". [New Thread 0x7ffff5db7640 (LWP 1082911)] [New Thread 0x7ffef53ff640 (LWP 1082913)] [Thread 0x7ffef53ff640 (LWP 1082913) exited] [New Thread 0x7ffdecb53640 (LWP 1083185)] [New Thread 0x7ffff54bf640 (LWP 1083186)] [Thread 0x7ffdecb53640 (LWP 1083185) exited] [Switching to AMDGPU Wave 2:2:1:1 (0,0,0)/0] Thread 6 hit Breakpoint 1, do_an_addition (a=<error reading variable: DWARF-2 expression error: `DW_OP_regx' operations must be used either alone or in conjunction with DW_OP_piece or DW_OP_bit_piece.>, b=<error reading variable: DWARF-2 expression error: `DW_OP_regx' operations must be used either alone or in conjunction with DW_OP_piece or DW_OP_bit_piece.>, out=<error reading variable: DWARF-2 expression error: `DW_OP_regx' operations must be used either alone or in conjunction with DW_OP_piece or DW_OP_bit_piece.>) at simple.cpp:24 24 *out = a + b; (gdb) info inferiors Num Description Connection Executable * 1 process 1082907 1 (native) /home/smarchi/build/binutils-gdb-amdgpu/gdb/simple (gdb) info threads Id Target Id Frame 1 Thread 0x7ffff5dc9240 (LWP 1082907) "simple" 0x00007ffff5e9410b in ?? () from /opt/rocm-5.4.0/lib/libhsa-runtime64.so.1 2 Thread 0x7ffff5db7640 (LWP 1082911) "simple" __GI___ioctl (fd=3, request=3222817548) at ../sysdeps/unix/sysv/linux/ioctl.c:36 5 Thread 0x7ffff54bf640 (LWP 1083186) "simple" __GI___ioctl (fd=3, request=3222817548) at ../sysdeps/unix/sysv/linux/ioctl.c:36 * 6 AMDGPU Wave 2:2:1:1 (0,0,0)/0 do_an_addition ( a=<error reading variable: DWARF-2 expression error: `DW_OP_regx' operations must be used either alone or in conjunction with DW_OP_piece or DW_OP_bit_piece.>, b=<error reading variable: DWARF-2 expression error: `DW_OP_regx' operations must be used either alone or in conjunction with DW_OP_piece or DW_OP_bit_piece.>, out=<error reading variable: DWARF-2 expression error: `DW_OP_regx' operations must be used either alone or in conjunction with DW_OP_piece or DW_OP_bit_piece.>) at simple.cpp:24 (gdb) bt Python Exception <class 'gdb.error'>: Unhandled dwarf expression opcode 0xe1 #0 do_an_addition (a=<error reading variable: DWARF-2 expression error: `DW_OP_regx' operations must be used either alone or in conjunction with DW_OP_piece or DW_OP_bit_piece.>, b=<error reading variable: DWARF-2 expression error: `DW_OP_regx' operations must be used either alone or in conjunction with DW_OP_piece or DW_OP_bit_piece.>, out=<error reading variable: DWARF-2 expression error: `DW_OP_regx' operations must be used either alone or in conjunction with DW_OP_piece or DW_OP_bit_piece.>) at simple.cpp:24 (gdb) continue Continuing. result is 3 warning: Temporarily disabling breakpoints for unloaded shared library "file:///home/smarchi/build/binutils-gdb-amdgpu/gdb/simple#offset=8192&size=67208" [Thread 0x7ffff54bf640 (LWP 1083186) exited] [Thread 0x7ffff5db7640 (LWP 1082911) exited] [Inferior 1 (process 1082907) exited normally] One thing to notice is the host and GPU threads appearing under the same inferior. This is a design goal for us, as programmers tend to think of the threads running on the GPU as part of the same program as the host threads, so showing them in the same inferior in GDB seems natural. Also, the host and GPU threads share a global memory space, which fits the inferior model. Another thing to notice is the error messages when trying to read variables or printing a backtrace. This is expected for the moment, since the AMD GPU compiler produces some DWARF that uses some non-standard extensions: https://llvm.org/docs/AMDGPUDwarfExtensionsForHeterogeneousDebugging.html There were already some patches posted by Zoran Zaric earlier to make GDB support these extensions: https://inbox.sourceware.org/gdb-patches/20211105113849.118800-1-zoran.zaric@amd.com/ We think it's better to get the basic support for AMD GPU in first, which will then give a better justification for GDB to support these extensions. GPU threads are named `AMDGPU Wave`: a wave is essentially a hardware thread using the SIMT (single-instruction, multiple-threads) [3] execution model. GDB uses the amd-dbgapi library [4], included in the ROCm platform, for a few things related to AMD GPU threads debugging. Different components talk to the library, as show on the following diagram: +---------------------------+ +-------------+ +------------------+ | GDB | amd-dbgapi target | <-> | AMD | | Linux kernel | | +-------------------+ | Debugger | +--------+ | | | amdgcn gdbarch | <-> | API | <=> | AMDGPU | | | +-------------------+ | | | driver | | | | solib-rocm | <-> | (dbgapi.so) | +--------+---------+ +---------------------------+ +-------------+ - The amd-dbgapi target is a target_ops implementation used to control execution of GPU threads. While the debugging of host threads works by using the ptrace / wait Linux kernel interface (as usual), control of GPU threads is done through a special interface (dubbed `kfd`) exposed by the `amdgpu` Linux kernel module. GDB doesn't interact directly with `kfd`, but instead goes through the amd-dbgapi library (AMD Debugger API on the diagram). Since it provides execution control, the amd-dbgapi target should normally be a process_stratum_target, not just a target_ops. More on that later. - The amdgcn gdbarch (describing the hardware architecture of the GPU execution units) offloads some requests to the amd-dbgapi library, so that knowledge about the various architectures doesn't need to be duplicated and baked in GDB. This is for example for things like the list of registers. - The solib-rocm component is an solib provider that fetches the list of code objects loaded on the device from the amd-dbgapi library, and makes GDB read their symbols. This is very similar to other solib providers that handle shared libraries, except that here the shared libraries are the pieces of code loaded on the device. Given that Linux host threads are managed by the linux-nat target, and the GPU threads are managed by the amd-dbgapi target, having all threads appear in the same inferior requires the two targets to be in that inferior's target stack. However, there can only be one process_stratum_target in a given target stack, since there can be only one target per slot. To achieve it, we therefore resort the hack^W solution of placing the amd-dbgapi target in the arch_stratum slot of the target stack, on top of the linux-nat target. Doing so allows the amd-dbgapi target to intercept target calls and handle them if they concern GPU threads, and offload to beneath otherwise. See amd_dbgapi_target::fetch_registers for a simple example: void amd_dbgapi_target::fetch_registers (struct regcache *regcache, int regno) { if (!ptid_is_gpu (regcache->ptid ())) { beneath ()->fetch_registers (regcache, regno); return; } // handle it } ptids of GPU threads are crafted with the following pattern: (pid, 1, wave id) Where pid is the inferior's pid and "wave id" is the wave handle handed to us by the amd-dbgapi library (in practice, a monotonically incrementing integer). The idea is that on Linux systems, the combination (pid != 1, lwp == 1) is not possible. lwp == 1 would always belong to the init process, which would also have pid == 1 (and it's improbable for the init process to offload work to the GPU and much less for the user to debug it). We can therefore differentiate GPU and non-GPU ptids this way. See ptid_is_gpu for more details. Note that we believe that this scheme could break down in the context of containers, where the initial process executed in a container has pid 1 (in its own pid namespace). For instance, if you were to execute a ROCm program in a container, then spawn a GDB in that container and attach to the process, it will likely not work. This is a known limitation. A workaround for this is to have a dummy process (like a shell) fork and execute the program of interest. The amd-dbgapi target watches native inferiors, and "attaches" to them using amd_dbgapi_process_attach, which gives it a notifier fd that is registered in the event loop (see enable_amd_dbgapi). Note that this isn't the same "attach" as in PTRACE_ATTACH, but being ptrace-attached is a precondition for amd_dbgapi_process_attach to work. When the debugged process enables the ROCm runtime, the amd-dbgapi target gets notified through that fd, and pushes itself on the target stack of the inferior. The amd-dbgapi target is then able to intercept target_ops calls. If the debugged process disables the ROCm runtime, the amd-dbgapi target unpushes itself from the target stack. This way, the amd-dbgapi target's footprint stays minimal when debugging a process that doesn't use the AMD ROCm platform, it does not intercept target calls. The amd-dbgapi library is found using pkg-config. Since enabling support for the amdgpu architecture (amdgpu-tdep.c) depends on the amd-dbgapi library being present, we have the following logic for the interaction with --target and --enable-targets: - if the user explicitly asks for amdgcn support with --target=amdgcn-*-* or --enable-targets=amdgcn-*-*, we probe for the amd-dbgapi and fail if not found - if the user uses --enable-targets=all, we probe for amd-dbgapi, enable amdgcn support if found, disable amdgcn support if not found - if the user uses --enable-targets=all and --with-amd-dbgapi=yes, we probe for amd-dbgapi, enable amdgcn if found and fail if not found - if the user uses --enable-targets=all and --with-amd-dbgapi=no, we do not probe for amd-dbgapi, disable amdgcn support - otherwise, amd-dbgapi is not probed for and support for amdgcn is not enabled Finally, a simple test is included. It only tests hitting a breakpoint in device code and resuming execution, pretty much like the example shown above. [1] https://docs.amd.com/category/ROCm_v5.4 [2] https://docs.amd.com/bundle/HIP-Programming-Guide-v5.4 [3] https://en.wikipedia.org/wiki/Single_instruction,_multiple_threads [4] https://docs.amd.com/bundle/ROCDebugger-API-Guide-v5.4 Change-Id: I591edca98b8927b1e49e4b0abe4e304765fed9ee Co-Authored-By: Zoran Zaric <zoran.zaric@amd.com> Co-Authored-By: Laurent Morichetti <laurent.morichetti@amd.com> Co-Authored-By: Tony Tye <Tony.Tye@amd.com> Co-Authored-By: Lancelot SIX <lancelot.six@amd.com> Co-Authored-By: Pedro Alves <pedro@palves.net>
2023-02-02gdb: make gdb_printing_disassembler::stream publicSimon Marchi1-2/+2
In the ROCm port, we need to access the underlying stream of a gdb_printing_disassembler, so make it public. The reason we need to access it is to know whether it supports style escape code. We then pass that information to a temporary string_file we use while symbolizing addresses. Change-Id: Ib95755a4a45b8f6478787993e9f904df60dd8dc1 Approved-By: Andrew Burgess <aburgess@redhat.com>
2023-02-02gdb/solib-svr4: don't disable probes interface if probe not foundSimon Marchi1-5/+10
In ROCm-GDB, we install an solib provider for the GPU code objects on top of the svr4 provider for the host, in order to add solibs representing the GPU code objects to the solib list containing the host process' shared libraries. We override the target_so_ops::handle_event function pointer with our own, in which we call svr4_so_ops.handle_event (which contains svr4_handle_solib_event) manually. When the host (un)loads a library, the ROCm part of handle_event is a no-op. When the GPU (un)loads a code object, we want the host side (svr4) to be a no-op. The problem is that when handle_event is called because of a GPU event, svr4_handle_solib_event gets called while not stopped at an svr4 probe. It then assumes this means there's a problem with the probes interface and disables it through the following sequence of events: - solib_event_probe_at return nullptr - svr4_handle_solib_event returns early - the make_scope_exit callback calls disable_probes_interface We could fix that by making the ROCm handle_event callback check if an svr4 probe is that the stop address, and only call svr4_so_ops.handle_event if so. However, it doesn't feel right to include some svr4 implementation detail in the ROCm event handler. Instead, this patch changes svr4_handle_solib_event to not assume it is an error if called while not at an svr4 probe location, and therefore not disable the probes interface. That just means moving the make_scope_exit call below where we lookup the probe by pc. Change-Id: Ie8ddf5beffa2e92b8ebfdd016454546252519244 Co-Authored-By: Lancelot SIX <lancelot.six@amd.com>
2023-02-02gdb: add gdbarch_upSimon Marchi1-0/+8
Add a gdbarch_up unique pointer type, that calls gdbarch_free on deletion. This is used in the ROCm support patch at the end of this series. Change-Id: I4b808892d35d69a590ce83180f41afd91705b2c8 Approved-By: Andrew Burgess <aburgess@redhat.com>
2023-02-02gdb: add inferior_pre_detach observableSimon Marchi3-0/+6
Add an observable notified in target_detach just before calling the detach method on the inferior's target stack. This allows observer to do some work on the inferior while it's still ptrace-attached, in the case of a native Linux inferior. Specifically, the amd-dbgapi target will need it in order to call amd_dbgapi_process_detach before the process gets ptrace-detached. Change-Id: I28b6065e251012a4c2db8a600fe13ba31671e3c9 Approved-By: Andrew Burgess <aburgess@redhat.com>
2023-02-02gdbsupport: add type definitions for pid, lwp and tidSimon Marchi1-7/+11
A following patch will want to declare variables of the same type as some ptid_t components. To make that easy (and avoid harcoding those types everywhere), define some type definitions in the ptid_t struct for each of them. Use them throughout ptid.h. I initially used pid_t, lwp_t and tid_t, but there is the risk of some system defining the pid_t type using a macro instead of a typedef, which would break things. So, use the _type suffix instead. Change-Id: I820b0bea9dafcb4914f1c9ba4bb96b5c666c8dec Approved-By: Andrew Burgess <aburgess@redhat.com>
2023-02-02gdb: make install_breakpoint return a non-owning referencePedro Alves2-4/+8
A following patch will want to install a breakpoint and then keep a non-owning reference to it. Make install_breakpoint return a non-owning reference, to make that easy. Co-Authored-By: Simon Marchi <simon.marchi@efficios.com> Change-Id: I2e8106a784021ff276ce251e24708cbdccc2d479 Approved-By: Andrew Burgess <aburgess@redhat.com>
2023-02-02gdb: add supports_arch_info callback to gdbarch_registerLancelot SIX2-3/+10
In the ROCm GDB port, there are some amdgcn architectures known by BFD that we don't actually support in GDB. We don't want gdbarch_printable_names to return these architectures. gdbarch_printable_names is used for a few things: - completion of the "set architecture" command - the gdb.architecture_names function in Python - foreach-arch selftests Add an optional callback to gdbarch_register that is a predicate indicating whether the gdbarch supports the given bfd_arch_info. by default, it is nullptr, meaning that the gdbarch accepts all "mach"s for that architecture known by BFD. Change-Id: I712f94351b0b34ed1f42e5cf7fc7ba051315d860 Co-Authored-By: Simon Marchi <simon.marchi@efficios.com> Approved-By: Andrew Burgess <aburgess@redhat.com>
2023-02-02[gas] Update .loc syntax comment in dwarf2dbg.cTom de Vries1-1/+1
I noticed that a comment in gas/dwarf2dbg.c describing .loc syntax was missing the "view VALUE" option. Fix this by adding the missing option.
2023-02-02gdb: remove gdb_indent.shEnze Li1-92/+0
GDB has been converted to a C++ program for many years[1], and the gdb_indent.sh will not be used any more. Therefore, remove the script as obvious. [1] https://sourceware.org/gdb/wiki/cxx-conversion Approved-By: Simon Marchi <simark@simark.ca>
2023-02-02ld/doc: use "stack trace" instead of "unwind" for SFrameIndu Bhagat1-1/+1
SFrame format is meant for generating stack traces only. ld/ * ld.texi: Replace the use of "unwind" with "stack trace".
2023-02-02bfd: use "stack trace" instead of "unwind" for SFrameIndu Bhagat5-12/+13
SFrame format is meant for generating stack traces only. bfd/ * elf-bfd.h: Replace the use of "unwind" with "stack trace". * elf-sframe.c: Likewise. * elf64-x86-64.c: Likewise. * elfxx-x86.c: Likewise. include/ * elf/common.h: Likewise.
2023-02-02gas: use "stack trace" instead of "unwind" for SFrameIndu Bhagat11-33/+35
SFrame format is meant for generating stack traces only. gas/ * as.c: Replace the use of "unwind" with "stack trace". * config/tc-aarch64.c: Likewise. * config/tc-aarch64.h: Likewise. * config/tc-i386.c: Likewise. * config/tc-i386.h: Likewise. * gen-sframe.c: Likewise. * gen-sframe.h: Likewise. * testsuite/gas/cfi-sframe/cfi-sframe-aarch64-2.s: Likewise. * testsuite/gas/cfi-sframe/cfi-sframe-common-8.s: Likewise. * testsuite/gas/cfi-sframe/common-empty-2.s: Likewise. * testsuite/gas/cfi-sframe/common-empty-3.s: Likewise.
2023-02-02sframe: use "stack trace" instead of "unwind" for SFrameIndu Bhagat1-14/+14
SFrame format is meant for generating stack traces only. include/ * sframe.h: Fix comments in the header file.
2023-02-02libsframe/doc: use "stack trace" instead of "unwind" for SFrameIndu Bhagat1-18/+19
SFrame format is meant for generating stack traces only. libsframe/ * doc/sframe-spec.texi: Use "stack trace" instead of "unwind".
2023-02-02ld-elf/merge test updateAlan Modra7-35/+27
The merge test fais on numerous targets because they don't support the necessary pc-relative relocs. This patch removes that part of the merge test, and makes references to the merged strings from .data rather than .text to better support targets that relax text by default.
2023-02-02Automatic date update in version.inGDB Administrator1-1/+1
2023-02-02obj-elf.h BYTES_IN_WORDAlan Modra1-1/+0
Don't define this. It is defined just before elf-bfd.h is included, but doesn't have any relevance there. Instead is for aout64.h where the default is 4 anyway.
2023-02-02gas obj_endAlan Modra8-28/+69
Provide a way for config/obj-* to clean up at end of assembly, and do so for ELF. * obj.h (struct format_ops): Add "end". * config/obj-aout.c (aout_format_ops): Init new field. * config/obj-coff.c (coff_format_ops): Likewise. * config/obj-ecoff.c (ecoff_format_ops): Likewise. * config/obj-elf.c (elf_format_ops): Likewise. (elf_begin): Move later in file. Clear some more variables. (comment_section): Make file scope. (free_section_idx): Rewrite. (elf_adjust_symtab): Expand str_htab_create call and use free_section_idx as delete function. (elf_frob_file_after_relocs): Don't clean up groups.indexes here. (elf_end): New function. * config/obj-elf.h (obj_end): Define. * config/obj-multi.h (obj_end): Define. * output-file.c (output_file_close): Call obj_end.
2023-02-01gdbserver: Add PID parameter to linux_get_auxv and linux_get_hwcapThiago Jung Bauermann11-37/+36
This patch doesn't change gdbserver behaviour, but after later changes are made it avoids a null pointer dereference when HWCAP needs to be obtained for a specific process while current_thread is nullptr. Fixing linux_read_auxv, linux_get_hwcap and linux_get_hwcap2 to take a PID parameter seems more correct than setting current_thread in one particular code path. Changes are propagated to allow passing the new parameter through the call chain. Approved-By: Simon Marchi <simon.marchi@efficios.com>
2023-02-01gdbserver: Add assert in find_register_by_numberThiago Jung Bauermann1-2/+3
It helped me during development, catching bugs closer to when they actually happened. Also remove the equivalent gdb_assert in regcache_raw_read_unsigned, since it's checking the same condition a few frames above. Suggested-By: Simon Marchi <simon.marchi@efficios.com> Approved-By: Simon Marchi <simon.marchi@efficios.com>
2023-02-01gdb/testsuite: fix fetch_src_and_symbols.exp with native-gdbserver boardAndrew Burgess1-9/+20
I noticed that the gdb.debuginfod/fetch_src_and_symbols.exp script doesn't work with the native-gdbserver board, I see this error: ERROR: tcl error sourcing /tmp/build/gdb/testsuite/../../../src/gdb/testsuite/gdb.debuginfod/fetch_src_and_symbols.exp. ERROR: gdbserver does not support run without extended-remote while executing "error "gdbserver does not support $command without extended-remote"" (procedure "gdb_test_multiple" line 51) invoked from within This was introduced with this commit: commit 7dd38e31d67c2548b52bea313ab18e40824c05da Date: Fri Jan 6 18:45:27 2023 -0500 gdb/linespec.c: Fix missing source file during breakpoint re-set The problem is that the above commit introduces a direct use of the "run" command, which doesn't work with 'target remote' targets, as exercised by the native-gdbserver board. To avoid this, in this commit I switch to using runto_main. However, calling runto_main will, by default, delete all the currently set breakpoints. As the point of the above commit was to check that a breakpoint set before stating an inferior would be correctly re-set, we need to avoid this breakpoint deleting behaviour. To do this I make use of with_override, and override the delete_breakpoints proc with a dummy proc which does nothing. By reverting the GDB changes in commit 7dd38e31d67c I have confirmed that even after my changes in this commit, the test still fails. But with the fixes in commit 7dd38e31d67c, this test now passed using the unix, native-gdbserver, and native-extended-gdbserver boards.
2023-02-01gdb: defer warnings when loading separate debug filesAlexandra Hájková12-42/+419
Currently, when GDB loads debug information from a separate debug file, there are a couple of warnings that could be produced if things go wrong. In find_separate_debug_file_by_buildid (build-id.c) GDB can give a warning if the separate debug file doesn't include any actual debug information, and in separate_debug_file_exists (symfile.c) we can warn if the CRC checksum in the separate debug file doesn't match the checksum in the original executable. The problem here is that, when looking up debug information, GDB will try several different approaches, lookup by build-id, lookup by debug-link, and then a lookup from debuginfod. GDB can potentially give a warning from an earlier attempt, and then succeed with a later attempt. In the cases I have run into this is primarily a warning about some out of date debug information on my machine, but then GDB finds the correct information using debuginfod. This can be confusing to a user, they will see warnings from GDB when really everything is working just fine. For example: warning: the debug information found in "/usr/lib/debug//lib64/ld-2.32.so.debug" \ does not match "/lib64/ld-linux-x86-64.so.2" (CRC mismatch). This diagnostic was printed on Fedora 33 even when the correct debuginfo was downloaded. In this patch I propose that we defer any warnings related to looking up debug information from a separate debug file. If any of the approaches are successful then GDB will not print any of the warnings. As far as the user is concerned, everything "just worked". Only if GDB completely fails to find any suitable debug information will the warnings be printed. The crc_mismatch test compiles two executables: crc_mismatch and crc_mismatch-2 and then strips them of debuginfo creating separate debug files. The test then replaces crc_mismatch-2.debug with crc_mismatch.debug to trigger "CRC mismatch" warning. A local debuginfod server is setup to supply the correct debug file, now when GDB looks up the debug info no warning is given. The build-id-no-debug-warning.exp is similar to the previous test. It triggers the "separate debug info file has no debug info" warning by replacing the build-id based .debug file with the stripped binary and then loading it to GDB. It then also sets up local debuginfod server with the correct debug file to download to make sure no warnings are emitted.
2023-02-01Fix compilation of the assembler with sanitization enabled.Nick Clifton2-2/+7
* dwarf2dbg.c (emit_inc_line_addr): Use unsigned constants when checking addr_delta.
2023-02-01Recursion in as_info_whereAlan Modra1-2/+0
This function has a gas_assert, ie. possible call to as_abort, which calls as_report_context, which calls as_info_where. * messages.c (as_info_where): Don't gas_assert.
2023-01-31gdb/dwarf: rename cooked_index_vector to cooked_indexSimon Marchi5-38/+35
See previous patch's commit message for rationale. Change-Id: I6b8cdc045dffccc1c01ed690ff258af09f6ff076 Approved-By: Tom Tromey <tom@tromey.com>
2023-01-31gdb/dwarf: rename cooked_index to cooked_index_shardSimon Marchi3-24/+24
I propose to rename cooked_index_vector and cooked_index such that the "main" object, that is the entry point to the index, is called cooked_index. The fact that the cooked index is implemented as a vector of smaller indexes is an implementation detail. This patch renames cooked_index to cooked_index_shard. The following patch renames cooked_index_vector to cooked_index. Change-Id: Id650f97dcb23c48f8409fa0974cd093ca0b75177 Approved-By: Tom Tromey <tom@tromey.com>
2023-02-01[gas] Emit v2 .debug_line for -gdwarf-2Tom de Vries4-24/+26
Currently, when using -gdwarf-2, gas emits a v3 .debug_line contribution. Fix this by emitting a v2 .debug_line contribution instead. gas/ChangeLog: 2023-01-31 Tom de Vries <tdevries@suse.de> PR 23941 * dwarf2dbg.c (DWARF2_LINE_VERSION): Set to 2 for -gdwarf-2. (DWARF2_LINE_OPCODE_BASE): Handle DWARF2_LINE_VERSION == 2. (dwarf2_directive_loc): Bump dwarf_level when encountering v3 .loc options. (out_debug_line): Don't output v3 standard opcodes for v2. * testsuite/gas/i386/debug1.d: Update. * testsuite/gas/i386/dwarf2-line-1.d: Update. * testsuite/gas/i386/dwarf2-line-4.d: Update.
2023-02-01Automatic date update in version.inGDB Administrator1-1/+1
2023-01-31gdb: add nullptr check to cooked_index_functions::dumpSimon Marchi1-4/+5
Since commit 7d82b08e9e0a ("gdb/dwarf: dump cooked index contents in cooked_index_functions::dump"), we see: maint print objfiles /home/smarchi/build/binutils-gdb/gdb/testsuite/outputs/gdb.dwarf2/dw2-error/dw2-error^M ^M Object file /home/smarchi/build/binutils-gdb/gdb/testsuite/outputs/gdb.dwarf2/dw2-error/dw2-error: Objfile at 0x614000005040, bfd at 0x6120000e08c0, 15 minsyms^M ^M Cooked index in use:^M ^M /home/smarchi/src/binutils-gdb/gdb/../gdbsupport/gdb-checked-static-cast.h:58: internal-error: checked_static_cast: Assertion `result != nullptr' failed.^M A problem internal to GDB has been detected,^M further debugging may prove unreliable.^M ----- Backtrace -----^M FAIL: gdb.dwarf2/dw2-error.exp: maint print objfiles /home/smarchi/build/binutils-gdb/gdb/testsuite/outputs/gdb.dwarf2/dw2-error/dw2-error (GDB internal error) The problem is that when cooked_index_functions fails to build an index, per_objfile->index_table is nullptr. Therefore, add a nullptr check, like other methods of cooked_index_functions already do. Print the "Cooked index in use" message after the nullptr check, such that if the cooked index failed to build, that message is not printed. Change-Id: Id67aef592e76c41b1e3bde9838a4e36cef873253
2023-01-31gdbsupport: allow passing nullptr to checked_static_castSimon Marchi1-0/+3
Both static_cast and dynamic_cast handle nullptr (they return nullptr), so I think checked_static_cast should too. This will allow doing a null check after a checked_static_cast: cooked_index_vector *table = (gdb::checked_static_cast<cooked_index_vector *> (per_bfd->index_table.get ())); if (table != nullptr) return; Change-Id: If5c3134e63696f8e417c87b5f3901240c9f7ea97
2023-01-31gdb/testsuite: adjust ensure_gdb_index to cooked_index_functions::dump changesSimon Marchi1-1/+1
Following 7d82b08e9e0a ("gdb/dwarf: dump cooked index contents in cooked_index_functions::dump"), I see some failures like: (gdb) mt print objfiles with-mf^M ^M Object file /home/smarchi/build/binutils-gdb/gdb/testsuite/outputs/gdb.base/with-mf/with-mf: Objfile at 0x614000005040, bfd at 0x6120000e08c0, 18 minsyms ^M ^M Cooked index in use:^M ^M ... (gdb) FAIL: gdb.base/with-mf.exp: check if index present This is because the format of the "Cooked index in use" line changed slightly. Adjust ensure_gdb_index to expect the trailing colon. Change-Id: If0a87575c02d8a0bc0d4b8ead540c234c62760f8
2023-01-31gdb/testsuite: fix xfail in gdb.ada/ptype_tagged_param.expSimon Marchi1-1/+1
I see: ERROR: wrong # args: should be "xfail message" while executing "xfail "no debug info" $gdb_test_name" ("uplevel" body line 3) invoked from within "uplevel { if {!$has_runtime_debug_info} { xfail "no debug info" $gdb_test_name } else { fail $gdb_test_name } }" This is because the xfail takes only one argument, fix that. Change-Id: I2e304d4fd3aa61067c04b5dac2be2ed34dab3190
2023-01-31Updated Swedish translation for the binutils sub-directoryNick Clifton2-762/+314
2023-01-31Re: Another fix for EFI generation with LTO enabledAlan Modra2-16/+2
Revert 1c66b8a03989 and instead fix the broken list pointer. PR 29998 * pe-dll.c (build_filler_bfd): Revert last change. * ldlang.c (lang_process): When rescanning archives for lto, fix file_chain.tail pointer if the insert point happens to be at the end of the list.
2023-01-31gas/ppc: Additional tests for DFP instructionsAndrew Burgess3-0/+186
I noticed that some of the Power6 DFP instructions were not covered by the assembler tests. I've added a new test file which I believe covers all the DFP Power6 instructions. The existing gas/testsuite/gas/ppc/power6.d test is called: POWER6 tests (includes DFP and Altivec) And does cover some of the DFP instructions. But, given the number of additional instructions I'm adding I opted to add a whole new test file. I've left the original power6.d unchanged, so there is now some overlap, but I don't think that should hurt much.
2023-01-31RISC-V: make C-extension JAL available again for (32-bit) assemblyJan Beulich5-1/+54
Along with the normal JAL alias, the C-extension one should have been moved as well by 839189bc932e ("RISC-V: re-arrange opcode table for consistent alias handling"), for the assembler to actually be able to use it where/when possible. Since neither this nor any other compressed branch insn was being tested so far, take the opportunity and introduce a new testcase covering those.
2023-01-31Silence ubsan warning about 1<<31Alan Modra1-1/+1
* merge.c (hash_blob): Write 1u << 31.
2023-01-31PR 30060, ASAN error in bfd_cache_closeAlan Modra2-8/+3
After bfd_close nothing should access bfd memory. Now that bfd_close always tidies up even after an error, attempting to tidy the cached bfd list by calling bfd_cache_close is wrong and not needed. PR 30060 * ar.c (remove_output): Don't call bfd_cache_close. (output_bfd): Delete. * arsup.c (ar_end): Call bfd_close_all_done, not bfd_cache_close.
2023-01-31testsuite XPASSesAlan Modra6-19/+12
This adjusts the testsuite to get rid of a number of XPASSes that have appeared. Someone might like to look into a better patch for the s390 change. aarch64-pe XPASS: weak symbols arm-nacl XPASS: rgn-over8 mcore-pe XPASS: ld-scripts/provide-8 mips64-linux-gnuabi64 XPASS: vers4 mips64-linux-gnuabi64 XPASS: vers4b mips-linux-gnu XPASS: vers4 mips-linux-gnu XPASS: vers4b s390-linux-gnu XPASS: undefined line sh4-linux-gnu XPASS: --gc-sections with __start_SECTIONNAME sh-coff XPASS: objcopy object (simple copy) sh-coff XPASS: objcopy executable (pr25662) binutils/ * testsuite/binutils-all/objcopy.exp: Don't xfail "simple copy" and "pr25662" on sh-*-coff. Remove all non-ELF xfails on "ELF unknown section type" test. ld/ * testsuite/ld-elfvers/vers.exp (vers4, vers4b): Don't xfail all mips, just xfail mips irix. * testsuite/ld-gc/pr19161.d: Don't xfail sh. * testsuite/ld-scripts/rgn-over8-ok.d: Don't xfail nacl. * testsuite/ld-scripts/weak.exp: Don't xfail aarch64-pe. * testsuite/ld-undefined/undefined.exp: Conditionally xfail "undefined line" depending on gcc version for s390.
2023-01-31Automatic date update in version.inGDB Administrator1-1/+1
2023-01-30Remove value_next declarationTom Tromey1-7/+0
value_next is declared but not defined. It's long obsolete. This patch removes the stray declaration.
2023-01-30gdb: fix dwarf2/cooked-index.c compilation on 32-bit systemsSimon Marchi1-1/+1
The i386 builder shows: ../../binutils-gdb/gdb/dwarf2/cooked-index.c: In member function ‘void cooked_index_vector::dump(gdbarch*) const’: ../../binutils-gdb/gdb/dwarf2/cooked-index.c:492:40: error: format ‘%lx’ expects argument of type ‘long unsigned int’, but argument 2 has type ‘std::__underlying_type_impl<sect_offset, true>::type’ {aka ‘long long unsigned int’} [-Werror=format=] 492 | gdb_printf (" DIE offset: 0x%lx\n", | ~~^ | | | long unsigned int | %llx 493 | to_underlying (entry->die_offset)); | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ | | | std::__underlying_type_impl<sect_offset, true>::type {aka long long unsigned int} The die_offset's underlying type is uint64, so use PRIx64 in the format string. Change-Id: Ibdde4c624ed1bb50eced9a514a4e37aec70a1323
2023-01-30gdb: Replace memcpy with std::copy to avoid some g++ warnings on sparcMark Wielaard1-1/+1
For some reason g++ 12.2.1 on sparc produces spurious warnings for stringop-overread and restrict in fbsd-tdep.c for a memcpy call. Use std::copy to avoid the warnings: In function ‘void* memcpy(void*, const void*, size_t)’, inlined from ‘gdb::optional<std::vector<unsigned char, gdb::default_init_allocator<unsigned char, std::allocator<unsigned char> > > > fbsd_make_note_desc(target_object, uint32_t)’ at ../../binutils-gdb/gdb/fbsd-tdep.c:666:10: /usr/include/bits/string_fortified.h:29:33: error: ‘void* __builtin_memcpy(void*, const void*, long unsigned int)’ specified bound 18446744073709551612 exceeds maximum object size 9223372036854775807 [-Werror=stringop-overflow=] In function ‘void* memcpy(void*, const void*, size_t)’, inlined from ‘gdb::optional<std::vector<unsigned char, gdb::default_init_allocator<unsigned char, std::allocator<unsigned char> > > > fbsd_make_note_desc(target_object, uint32_t)’ at ../../binutils-gdb/gdb/fbsd-tdep.c:673:10: /usr/include/bits/string_fortified.h:29:33: error: ‘void* __builtin_memcpy(void*, const void*, long unsigned int)’ accessing 18446744073709551612 bytes at offsets 0 and 0 overlaps 9223372036854775801 bytes at offset -9223372036854775805 [-Werror=restrict] gdb/ChangeLog: * fbsd-tdep.c (fbsd_make_note_desc): Use std::copy instead of memcpy.
2023-01-30gdb/dwarf: dump cooked index contents in cooked_index_functions::dumpSimon Marchi3-1/+109
As I am investigating a crash I see with the cooked index, I thought it would be useful to have a way to dump the index contents. For those not too familiar with it (that includes me), it can help get a feel of what it contains and how it is structured. The cooked_index_functions::dump function is called as part of the "maintenance print objfiles" command. I tried to make the output well structured and indented to help readability, as this prints a lot of text. The dump function first dumps all cooked index entries, like this: [25] ((cooked_index_entry *) 0x621000121220) name: __ioinit canonical: __ioinit DWARF tag: DW_TAG_variable flags: 0x2 [IS_STATIC] DIE offset: 0x21a4 parent: ((cooked_index_entry *) 0x6210000f9610) [std] Then the information about the main symbol: main: ((cooked_index_entry *) 0x621000123b40) [main] And finally the address map contents: [1] ((addrmap *) 0x6210000f7910) [0x0] ((dwarf2_per_cu_data *) 0) [0x118a] ((dwarf2_per_cu_data *) 0x60c000007f00) [0x1cc7] ((dwarf2_per_cu_data *) 0) [0x1cc8] ((dwarf2_per_cu_data *) 0x60c000007f00) [0x1cdf] ((dwarf2_per_cu_data *) 0) [0x1ce0] ((dwarf2_per_cu_data *) 0x60c000007f00) The display of address maps above could probably be improved, to show it more as ranges, but I think this is a reasonable start. Note that this patch depends on Pedro Alves' patch "enum_flags to_string" [1]. If my patch is to be merged before Pedro's series, I will cherry-pick this patch from his series and merge it before mine. [1] https://inbox.sourceware.org/gdb-patches/20221212203101.1034916-8-pedro@palves.net/ Change-Id: Ida13e479fd4c8d21102ddd732241778bc3b6904a
2023-01-30enum_flags to_stringPedro Alves2-6/+129
This commit introduces shared infrastructure that can be used to implement enum_flags -> to_string functions. With this, if we want to support converting a given enum_flags specialization to string, we just need to implement a function that provides the enumerator->string mapping, like so: enum some_flag { SOME_FLAG1 = 1 << 0, SOME_FLAG2 = 1 << 1, SOME_FLAG3 = 1 << 2, }; DEF_ENUM_FLAGS_TYPE (some_flag, some_flags); static std::string to_string (some_flags flags) { static constexpr some_flags::string_mapping mapping[] = { MAP_ENUM_FLAG (SOME_FLAG1), MAP_ENUM_FLAG (SOME_FLAG2), MAP_ENUM_FLAG (SOME_FLAG3), }; return flags.to_string (mapping); } .. and then to_string(SOME_FLAG2 | SOME_FLAG3) produces a string like "0x6 [SOME_FLAG2 SOME_FLAG3]". If we happen to forget to update the mapping array when we introduce a new enumerator, then the string representation will pretty-print the flags it knows about, and then the leftover flags in hex (one single number). For example, if we had missed mapping SOME_FLAG2 above, we'd end up with: to_string(SOME_FLAG2 | SOME_FLAG3) => "0x6 [SOME_FLAG2 0x4]"); Other than in the unit tests included, no actual usage of the functionality is added in this commit. Approved-By: Simon Marchi <simon.marchi@efficios.com> Change-Id: I835de43c33d13bc0c95132f42c3f97318b875779
2023-01-30Fix comparator bug in cooked indexTom Tromey2-85/+132
Simon pointed out that the cooked index template-matching patch introduced a failure in libstdc++ debug mode. In particular, the new code violates the assumption of std::lower_bound and std::upper_bound that the range is sorted with respect to the comparison. When I first debugged this, I thought the problem was unfixable as-is and that a second layer of filtering would have to be done. However, on irc, Simon pointed out that it could perhaps be solved if the comparison function were assured that one operand always came from the index, with the other always being the search string. This patch implements this idea. First, a new mode is introduced: a sorting mode for cooked_index_entry::compare. In this mode, strings are compared case-insensitively, but we're careful to always sort '<' before any other printable character. This way, two names like "func" and "func<param>" will be sorted next to each other -- i.e., "func1" will not be seen between them. This is important when searching. Second, the compare function is changed to work in a strcmp-like way. This makes it easier to test and (IMO) understand. Third, the compare function is modified so that in non-sorting modes, the index entry is always the first argument. This allows consistency in compares. I regression tested this in libstdc++ debug mode on x86-64 Fedora 36. It fixes the crash that Simon saw. This is v2. I believe it addresses the review comments, except for the 'enum class' change, as I mentioned in email on the list. Approved-By: Simon Marchi <simon.marchi@efficios.com>
2023-01-30Clean up lnp_state_machine constructorTom Tromey1-20/+13
This changes the lnp_state_machine constructor to initialize members directly; and changes lnp_state_machine itself to initialize members inline when possible. Reviewed-By: Lancelot Six <lancelot.six@amd.com>
2023-01-30Make addrmap const-correct in cooked indexTom Tromey3-9/+11
After the cooked index is created, the addrmaps should be const. Change-Id: I8234520ab346ced40a8dd6e478ba21fc438c2ba2
generated by cgit v1.1 at 2024-12-27 15:27:09 +0000