[Bug gdb/34561] New: Problem reading OS generated corefiles on systems with Intel AMX support

https://sourceware.org/bugzilla/show_bug.cgi?id=34561

            Bug ID: 34561
           Summary: Problem reading OS generated corefiles on systems with
                    Intel AMX support
           Product: gdb
           Version: HEAD
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: gdb
          Assignee: unassigned at sourceware dot org
          Reporter: christina.joos at intel dot com
  Target Milestone: ---

On Systems with Intel AMX support the xsave size is 11008. This xsave size is
not
handled by gdb/i387-tdep.c:i387_guess_xsave_layout and is causing problems for
OS
generated corefiles.

Consider the following simple test program to generate a corefile:
~~~
$ cat main.c
int main ()
{
  int *pointer = 0;
  *pointer = 3;

  return 0;
}
~~~

Steps to reproduce:
~~~
$ ./main
Segmentation fault (core dumped)
[...]
$ gdb-up
GNU gdb (GDB) 19.0.50.20260814-git
Copyright (C) 2026 Free Software Foundation, Inc.
[...]
(gdb) core core_main_SEGV
[New LWP 2162880 (id 1)]
[...]
Core was generated by `./main'.
Program terminated with signal SIGSEGV, Segmentation fault.
#0  0x0000593156bd713d in main () at main.c:4
4         *pointer = 3;
(gdb) p $ymm0
$1 = void
~~~

The issue is that we should be able to print the register $ymm0
(or any other register belonging to a feature higher than SSE). 
For a live debug session we can print it:
~~~
Reading symbols from main...
(gdb) start
Temporary breakpoint 1 at 0x1131: file main.c, line 3.
Starting program: /tmp/main
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".

Temporary breakpoint 1, main () at main.c:3
3         int *pointer = 0;
(gdb) p $ymm0
$1 = {v16_bfloat16 = {0, 2.342e-38, 0 <repeats 14 times>}, v16_half = {0,
1.5199e-05, 0 <repeats 14 times>}, v8_float = {
    2.34180515e-38, 0, 0, 0, 0, 0, 0, 0}, v4_double = {8.256666972292243e-317,
0, 0, 0}, v32_int8 = {0, 0, -1,
    0 <repeats 29 times>}, v16_int16 = {0, 255, 0 <repeats 14 times>}, v8_int32
= {16711680, 0, 0, 0, 0, 0, 0, 0}, v4_int64 = {
    16711680, 0, 0, 0}, v2_int128 = {16711680, 0}}
[...]
~~~

This is should be reproducible on any CPU which supports AMX.
My machine has the following setup:
~~~
$ gcc --version
gcc (Ubuntu 13.3.0-6ubuntu2~24.04.1) 13.3.0
$ lsb_release -a
No LSB modules are available.
Distributor ID: Ubuntu
Description:    Ubuntu 24.04.4 LTS
~~~

Note that this is not reproducible for corefiles generated by the gcore
command, since gdb/i387-tdep.c:i387_fallback_xsave_layout will configure
the xsave_layout based on xcr0.  xcr0 is created based on the target
description from the corefile in i386_gdbarch_init.

i387_fallback_xsave_layout (i386_xcr0_from_tdesc (info.target_desc));

For OS generated corefiles the GDB target description is not available in
the corefile.

~~~
Reading symbols from main...
(gdb) start
Temporary breakpoint 1 at 0x1131: file main.c, line 3.
Starting program: /tmp/main
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".

Temporary breakpoint 1, main () at main.c:3
3         int *pointer = 0;
(gdb) gcore core_main_gcore
⚠️ warning: Memory read failed for corefile section, 4096 bytes at
0xffffffffff600000.
Saved corefile core_main_gcore
[...]
(gdb) core core_main_gcore
[New LWP 2163444 (id 1)]
Reading symbols from /tmp/main...
⚠️ warning: Unexpected size of section `.reg-xstate/2163444' in core file.
[...]
Core was generated by `/tmp/main'.
Program terminated with signal SIGTRAP, Trace/breakpoint trap.
⚠️ warning: Unexpected size of section `.reg-xstate/2163444' in core file.
#0  main () at main.c:3
3         int *pointer = 0;
(gdb) p $ymm0
$1 = {v16_bfloat16 = {0, 2.342e-38, 0 <repeats 14 times>}, v16_half = {0,
1.5199e-05, 0 <repeats 14 times>}, v8_float = {
    2.34180515e-38, 0, 0, 0, 0, 0, 0, 0}, v4_double = {8.256666972292243e-317,
0, 0, 0}, v32_int8 = {0, 0, -1,
    0 <repeats 29 times>}, v16_int16 = {0, 255, 0 <repeats 14 times>}, v8_int32
= {16711680, 0, 0, 0, 0, 0, 0, 0}, v4_int64 = {
    16711680, 0, 0, 0}, v2_int128 = {16711680, 0}}
~~~

Even though the xsave_size is configured to 2696 in i387_fallback_xsave_layout
which is not correct for this system, it does not seem to be a problem
(or at least, I couldn't find a problem so far).

-- 
You are receiving this mail because:
You are on the CC list for the bug.