Mirror of the gdb-patches mailing list
 help / color / mirror / Atom feed
* [PATCH v2 1/1] gdb: Enable OS generated corefiles on systems with Intel AMX support.
@ 2026-09-11 16:30 Christina Joos
  2026-09-15 15:06 ` Andrew Burgess
  0 siblings, 1 reply; 2+ messages in thread
From: Christina Joos @ 2026-09-11 16:30 UTC (permalink / raw)
  To: gdb-patches; +Cc: aburgess, stephan.rohr

This patch addresses the issue described in PR gdb/34561.  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 corefiles generated by the linux kernel.  The problem is
visible regardless of the AMX enablement state in the dumping process,
so all corefiles on newer systems with AMX are broken.

~~~
$ gdb
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.
4         *pointer = 3;
(gdb) p $ymm0
$1 = void
~~~

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}}
[...]
~~~

Note that this is not reproducible for corefiles generated by the gcore
command, since in such corefiles the GDB target description is
available and gdb/i387-tdep.c:i387_fallback_xsave_layout configures
the xsave_layout based on xcr0 derived from the target description.

This patch fixes this issue by handling the new xsave size in
i387_guess_xsave_layout.  It is necessary even though GDB does not
support AMX yet, since we still pass the full xsave_size (11008)
extracted from the corefile to i387_guess_xsave_layout.
The patch does not fix i387_fallback_xsave_layout to configure the new
xsave size.  Configuring sizeof_xsave to 2696 is fine at this point,
since truncating to 2696 drops only the parts GDB doesn't model yet.

As a quick solution for the upcoming release and CPUs supporting
Intel AMX this patch should be sufficient.

Testing:

I noticed that we don't have tests for corefiles for the features AVX,
AVX512 and PKRU in the GDB testsuite.  So this commit adds corefile tests
for AVX, AVX512 and PKRU registers.

As a result of this patch, the warning:

  warning: Unexpected size of section `.reg-xstate/1373786' in core file.

which was causing failure in gdb.base/coredump-filter.exp and
gdb.arch/i386-tls-regs.exp on Intel AMX systems disappears, these tests now
pass.

Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=34561
---
 gdb/i387-tdep.c                        |  9 +++-
 gdb/testsuite/gdb.arch/i386-avx.c      |  9 ++++
 gdb/testsuite/gdb.arch/i386-avx.exp    | 68 +++++++++++++++++++++++---
 gdb/testsuite/gdb.arch/i386-avx512.c   |  9 ++++
 gdb/testsuite/gdb.arch/i386-avx512.exp | 62 +++++++++++++++++++++++
 gdb/testsuite/gdb.arch/i386-pkru.c     |  9 ++++
 gdb/testsuite/gdb.arch/i386-pkru.exp   | 45 +++++++++++++++++
 7 files changed, 201 insertions(+), 10 deletions(-)

diff --git a/gdb/i387-tdep.c b/gdb/i387-tdep.c
index f22a5e02bbb..495c463ed41 100644
--- a/gdb/i387-tdep.c
+++ b/gdb/i387-tdep.c
@@ -918,9 +918,14 @@ bool
 i387_guess_xsave_layout (uint64_t xcr0, size_t xsave_size,
 			 x86_xsave_layout &layout)
 {
-  if (HAS_PKRU (xcr0) && xsave_size == 2696)
+  if (HAS_PKRU (xcr0) && (xsave_size == 2696 || xsave_size == 11008))
     {
-      /* Intel CPUs supporting PKRU.  */
+      /* Intel CPUs supporting PKRU.
+	 In this case two possible XSAVE_SIZE values have to be handled.
+	 As GDB does not yet support Intel AMX and CPUs that support this
+	 feature will have a different XSAVE_SIZE (11008).  If we don't
+	 handle this size then we will not be able to handle any xsave
+	 registers from OS generated core files on AMX systems.  */
       layout.avx_offset = 576;
       layout.k_offset = 1088;
       layout.zmm_h_offset = 1152;
diff --git a/gdb/testsuite/gdb.arch/i386-avx.c b/gdb/testsuite/gdb.arch/i386-avx.c
index b6cd89e5bea..369262857ef 100644
--- a/gdb/testsuite/gdb.arch/i386-avx.c
+++ b/gdb/testsuite/gdb.arch/i386-avx.c
@@ -23,6 +23,8 @@
 /* Align sufficient to be able to use vmovaps.  */
 #define ALIGN 32
 
+volatile int should_dump_core_p = 1;
+
 typedef struct {
   _Alignas (ALIGN) float f[8];
 } v8sf_t;
@@ -87,6 +89,13 @@ main (int argc, char **argv)
 
   asm ("nop"); /* first breakpoint here */
 
+  /* Crash for OS corefile.  */
+  if (should_dump_core_p)
+    {
+      /* Generate SIGSEGV to crash.  */
+      *(volatile int *) 0;
+    };
+
   asm (
        "vmovaps %%ymm0, 0(%0)\n\t"
        "vmovaps %%ymm1, 32(%0)\n\t"
diff --git a/gdb/testsuite/gdb.arch/i386-avx.exp b/gdb/testsuite/gdb.arch/i386-avx.exp
index 1a786ee8a51..ed79a8535c6 100644
--- a/gdb/testsuite/gdb.arch/i386-avx.exp
+++ b/gdb/testsuite/gdb.arch/i386-avx.exp
@@ -42,12 +42,27 @@ if { [gdb_compile "${srcdir}/${subdir}/${srcfile}" "${binfile}" executable \
     return
 }
 
+# Check reading registers after the first breakpoint for ymm int8 and
+# float types.
+proc test_regs_first_bp {} {
+    for { set r 0 } { $r < $::nr_regs } { incr r } {
+	gdb_test "print \$ymm$r.v8_float" \
+	    ".. = \\{$r, $r.125, $r.25, $r.375, $r.5, $r.625, $r.75, $r.875\\}.*" \
+	    "check float contents of %ymm$r"
+	gdb_test "print \$ymm$r.v32_int8" \
+	    ".. = \\{(-?${::decimal}, ){31}-?${::decimal}\\}.*" \
+	    "check int8 contents of %ymm$r"
+    }
+}
+
 clean_restart ${::testfile}
 
 if {![runto_main]} {
     return
 }
 
+gdb_test_no_output "set should_dump_core_p=0"
+
 gdb_test "break [gdb_get_line_number "first breakpoint here"]" \
 	 "Breakpoint .* at .*i386-avx.c.*" \
 	 "set first breakpoint in main"
@@ -59,14 +74,7 @@ if {[is_amd64_regs_target]} {
     set nr_regs 8
 }
 
-for { set r 0 } { $r < $nr_regs } { incr r } {
-    gdb_test "print \$ymm$r.v8_float" \
-	".. = \\{$r, $r.125, $r.25, $r.375, $r.5, $r.625, $r.75, $r.875\\}.*" \
-	"check float contents of %ymm$r"
-    gdb_test "print \$ymm$r.v32_int8" \
-	".. = \\{(-?${::decimal}, ){31}-?${::decimal}\\}.*" \
-	"check int8 contents of %ymm$r"
-}
+test_regs_first_bp
 
 for { set r 0 } { $r < $nr_regs } { incr r } {
     gdb_test "set var \$ymm$r.v8_float\[0\] = $r + 10" "" "set %ymm$r"
@@ -100,9 +108,53 @@ with_test_prefix "force-disable xml descriptions" {
 	    return
 	}
 
+	gdb_test_no_output "set should_dump_core_p=0"
+
 	# With qXfer:features:read disabled, we won't know anything
 	# about YMM registers.
 	gdb_test "print \$ymm0" " = void"
 	gdb_test "print \$xmm0" "v4_float.*"
     }
 }
+
+# Restart gdb, load the corefile generated by gcore or the OS
+# and check reading registers from the corefile.
+proc test_corefile {core_filename} {
+    clean_restart $::testfile
+
+    gdb_test "core $core_filename" "Core was generated by .*" \
+	"load corefile"
+
+    test_regs_first_bp
+}
+
+with_test_prefix "OS generated corefile" {
+    # This test relies on reading address zero triggering a SIGSEGV.
+    # If address zero is readable skip this test.
+    if { [is_address_zero_readable] } {
+	unsupported "address zero is readable"
+    } else {
+	set corefile [core_find $binfile]
+	if { $corefile eq "" } {
+	    unsupported "unable to generate core file"
+	} else {
+	    test_corefile $corefile
+	}
+    }
+}
+
+with_test_prefix "gcore corefile" {
+    clean_restart ${::testfile}
+    if { ![runto_main] } {
+	return
+    }
+
+    set line [gdb_get_line_number "first breakpoint here"]
+    gdb_breakpoint $line
+    gdb_continue_to_breakpoint "first breakpoint here" ".*$srcfile:$line.*"
+
+    set gcorefile $binfile.gcore
+    if { [gdb_gcore_cmd $gcorefile "save a corefile"] } {
+	test_corefile $gcorefile
+    }
+}
diff --git a/gdb/testsuite/gdb.arch/i386-avx512.c b/gdb/testsuite/gdb.arch/i386-avx512.c
index b1e023ce4d4..d3edf780914 100644
--- a/gdb/testsuite/gdb.arch/i386-avx512.c
+++ b/gdb/testsuite/gdb.arch/i386-avx512.c
@@ -19,6 +19,8 @@
 
 #include "x86-cpuid.h"
 
+volatile int should_dump_core_p = 1;
+
 typedef struct
 {
   double f[8];
@@ -233,6 +235,13 @@ main (int argc, char **argv)
       move_zmm_data_to_reg ();
       asm ("nop"); /* third breakpoint here  */
 
+      /* Crash for OS corefile.  */
+      if (should_dump_core_p)
+	{
+	  /* Generate SIGSEGV to crash.  */
+	  *(volatile int *) 0;
+	};
+
       /* Test script incremented values,
 	 move back to array and check values.  */
       move_zmm_data_to_memory ();
diff --git a/gdb/testsuite/gdb.arch/i386-avx512.exp b/gdb/testsuite/gdb.arch/i386-avx512.exp
index c79563366dc..b6834152e8a 100644
--- a/gdb/testsuite/gdb.arch/i386-avx512.exp
+++ b/gdb/testsuite/gdb.arch/i386-avx512.exp
@@ -35,6 +35,8 @@ if {![runto_main]} {
     return
 }
 
+gdb_test_no_output "set should_dump_core_p=0"
+
 set supports_avx512 0
 
 set test "probe AVX512 support"
@@ -181,3 +183,63 @@ gdb_test "print \$zmm0.v16_int32" "= {-1, -1, -1, -1, 0 <repeats 12 times>}"
 if { $nr_regs >= 16 } {
     gdb_test "print \$zmm16.v16_int32" "= {-1 <repeats 16 times>}"
 }
+
+# Test reading k and zmm registers and for double and int16 types.
+proc test_regs_corefile {} {
+    for { set r 1 } { $r < 8 } { incr r } {
+	gdb_test "print/x \$k$r" \
+	    ".. = 0x[format %x $r]2[format %x $r]1" \
+	    "check contents of %k$r"
+    }
+
+    for { set r 0 } { $r < $::nr_regs } { incr r } {
+	gdb_test "print \$zmm$r.v8_double" \
+	    ".. = \\{$r, $r.125, $r.25, $r.375, $r.5, $r.625, $r.75, $r.875\\}.*" \
+	    "check double contents of %zmm$r"
+	gdb_test "print \$zmm$r.v32_int16" \
+	    ".. = \\{(-?${::decimal}, ){31}-?${::decimal}\\}.*" \
+	    "check int16 contents of %zmm$r"
+    }
+}
+
+# Restart gdb, load the corefile generated by gcore or the OS
+# and check reading registers from the corefile.
+proc test_corefile {core_filename} {
+    clean_restart $::testfile
+
+    gdb_test "core $core_filename" "Core was generated by .*" \
+	"load corefile"
+
+    test_regs_corefile
+}
+
+with_test_prefix "OS generated corefile" {
+    # This test relies on reading address zero triggering a SIGSEGV.
+    # If address zero is readable skip this test.
+    if { [is_address_zero_readable] } {
+	unsupported "address zero is readable"
+    } else {
+	set corefile [core_find $binfile]
+	if { $corefile eq "" } {
+	    unsupported "unable to generate core file"
+	} else {
+	    test_corefile $corefile
+	}
+    }
+}
+
+with_test_prefix "gcore corefile" {
+    clean_restart ${::testfile}
+    if { ![runto_main] } {
+	return
+    }
+
+    set line [gdb_get_line_number "third breakpoint here"]
+    gdb_breakpoint $line
+    gdb_continue_to_breakpoint "third breakpoint here" ".*$srcfile:$line.*"
+
+    set gcorefile $binfile.gcore
+    if { [gdb_gcore_cmd $gcorefile "save a corefile"] } {
+	test_corefile $gcorefile
+    }
+}
diff --git a/gdb/testsuite/gdb.arch/i386-pkru.c b/gdb/testsuite/gdb.arch/i386-pkru.c
index 9561d1e923a..159acdea219 100644
--- a/gdb/testsuite/gdb.arch/i386-pkru.c
+++ b/gdb/testsuite/gdb.arch/i386-pkru.c
@@ -24,6 +24,8 @@
 #define NOINLINE __attribute__ ((noinline))
 #endif
 
+volatile int should_dump_core_p = 1;
+
 unsigned int have_pkru (void) NOINLINE;
 
 static inline unsigned long
@@ -83,6 +85,13 @@ main (int argc, char **argv)
       wrpkru (wr_value);
       asm ("nop\n\t");	/* break here 1.  */
 
+      /* Crash for OS corefile.  */
+      if (should_dump_core_p)
+	{
+	  /* Generate SIGSEGV to crash.  */
+	  *(volatile int *) 0;
+	}
+
       rd_value = rdpkru ();
       asm ("nop\n\t");	/* break here 2.  */
     }
diff --git a/gdb/testsuite/gdb.arch/i386-pkru.exp b/gdb/testsuite/gdb.arch/i386-pkru.exp
index b9b0b99b0de..ae90a2bea1a 100644
--- a/gdb/testsuite/gdb.arch/i386-pkru.exp
+++ b/gdb/testsuite/gdb.arch/i386-pkru.exp
@@ -38,6 +38,8 @@ if {![runto_main]} {
     return
 }
 
+gdb_test_no_output "set should_dump_core_p=0"
+
 set supports_pkru 0
 set test "probe PKRU support"
 gdb_test_multiple "print have_pkru()" $test {
@@ -108,3 +110,46 @@ gdb_test_multiple "print /x rd_value" "variable after reading pkru" {
 	}
     }
 }
+
+# Restart gdb, load the corefile generated by gcore or the OS
+# and check reading the pkru register from the corefile.
+proc test_corefile {core_filename} {
+    clean_restart $::testfile
+
+    gdb_test "core $core_filename" "Core was generated by .*" \
+	"load corefile"
+
+    gdb_test "info register pkru" ".*pkru.*$::val1.*" \
+	"read pkru register"
+}
+
+with_test_prefix "OS generated corefile" {
+    # This test relies on reading address zero triggering a SIGSEGV.
+    # If address zero is readable skip this test.
+    if { [is_address_zero_readable] } {
+	unsupported "address zero is readable"
+    } else {
+	set corefile [core_find $binfile]
+	if { $corefile eq "" } {
+	    unsupported "unable to generate core file"
+	} else {
+	    test_corefile $corefile
+	}
+    }
+}
+
+with_test_prefix "gcore corefile" {
+    clean_restart ${::testfile}
+    if { ![runto_main] } {
+	return
+    }
+
+    set line [gdb_get_line_number "break here 1"]
+    gdb_breakpoint $line
+    gdb_continue_to_breakpoint "break here 1" ".*$srcfile:$line.*"
+
+    set gcorefile $binfile.gcore
+    if { [gdb_gcore_cmd $gcorefile "save a corefile"] } {
+	test_corefile $gcorefile
+    }
+}
-- 
2.53.0

________________________________________
Intel Deutschland GmbH 

Registered Address: Dornacher Strasse 1, 85622 Feldkirchen, Germany 

Tel: +49 (89) 99143-0 

www.intel.de 

Managing Directors: Candice Moore, Jeffrey Schneiderman, Ramachandran Sitaraman

Chairperson of the Supervisory Board: Sonja Pierer

Registered Seat: Munich Commercial Register B: Amtsgericht Munich HRB 186928

This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.


^ permalink raw reply	[flat|nested] 2+ messages in thread
end of thread, other threads:[~2026-09-15 15:06 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2026-09-11 16:30 [PATCH v2 1/1] gdb: Enable OS generated corefiles on systems with Intel AMX support Christina Joos
2026-09-15 15:06 ` Andrew Burgess

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox