[1/2] Display ExceptionRecord for $_siginfo

  Uses the $_siginfo convenience variable to show the last exception.

The type looks like this:

(gdb) pt $_siginfo
type = struct EXCEPTION_RECORD {
    unsigned int ExceptionCode;
    unsigned int ExceptionFlags;
    struct EXCEPTION_RECORD *ExceptionRecord;
    void *ExceptionAddress;
    unsigned int NumberParameters;
    void *ExceptionInformation[15];
}

EXCEPTION_RECORD is documented at [1].

Example:

Program received signal SIGSEGV, Segmentation fault.
main () at crasher.c:4
4         *(int*)0x123 = 0;
(gdb) p $_siginfo
$1 = {
  ExceptionCode = 3221225477,
  ExceptionFlags = 0,
  ExceptionRecord = 0x0,
  ExceptionAddress = 0x401632 <main+18>,
  NumberParameters = 2,
  ExceptionInformation = {0x1, 0x123, 0x0 <repeats 13 times>}
}
(gdb) p/x $_siginfo.ExceptionCode
$2 = 0xc0000005

And 0xc0000005 is the value of EXCEPTION_ACCESS_VIOLATION.

[1] https://docs.microsoft.com/en-us/windows/win32/api/winnt/ns-winnt-exception_record

gdb/ChangeLog:

2020-01-17  Hannes Domani  <ssbssa@yahoo.de>

	* windows-nat.c (handle_exception): Set siginfo_er.
	(windows_nat_target::mourn_inferior): Reset siginfo_er.
	(windows_xfer_siginfo): New function.
	(windows_nat_target::xfer_partial): Call windows_xfer_siginfo.
	* windows-tdep.c (struct windows_gdbarch_data): New struct.
	(init_windows_gdbarch_data): New function.
	(get_windows_gdbarch_data): New function.
	(windows_get_siginfo_type): New function.
	(windows_init_abi): Register windows_get_siginfo_type.
	(_initialize_windows_tdep): Register init_windows_gdbarch_data.

gdb/gdbserver/ChangeLog:

2020-01-17  Hannes Domani  <ssbssa@yahoo.de>

	* win32-low.c (win32_clear_inferiors): Reset siginfo_er.
	(handle_exception): Set siginfo_er.
	(win32_xfer_siginfo): New function.
---
 gdb/gdbserver/win32-low.c | 28 +++++++++++++++++-
 gdb/windows-nat.c         | 29 ++++++++++++++++++
 gdb/windows-tdep.c        | 62 +++++++++++++++++++++++++++++++++++++++
 3 files changed, 118 insertions(+), 1 deletion(-)
  

On 2020-01-17 10:31 a.m., Hannes Domani via gdb-patches wrote:
> diff --git a/gdb/windows-nat.c b/gdb/windows-nat.c
> index 901e64263c..824ff4b322 100644
> --- a/gdb/windows-nat.c
> +++ b/gdb/windows-nat.c
> @@ -236,6 +236,7 @@ static DEBUG_EVENT current_event;	/* The current debug event from
>  					   WaitForDebugEvent */
>  static HANDLE current_process_handle;	/* Currently executing process */
>  static windows_thread_info *current_thread;	/* Info on currently selected thread */
> +static EXCEPTION_RECORD siginfo_er;	/* Contents of $_siginfo */

Huh... I was going to say that it shouldn't be a global variable, but a per-inferior
thing (or is it per-thread?), but pretty much all the state is already global...  so
I guess it's fine.  I gather that the windows-nat does not support debugging multiple
inferiors?  Same for win32-low in gdbserver?

>  
>  /* Counts of things.  */
>  static int exception_count = 0;
> @@ -1166,6 +1167,8 @@ handle_exception (struct target_waitstatus *ourstatus)
>    DWORD code = rec->ExceptionCode;
>    handle_exception_result result = HANDLE_EXCEPTION_HANDLED;
>  
> +  memcpy (&siginfo_er, rec, sizeof siginfo_er);
> +
>    ourstatus->kind = TARGET_WAITKIND_STOPPED;
>  
>    /* Record the context of the current thread.  */
> @@ -2862,6 +2865,7 @@ windows_nat_target::mourn_inferior ()
>        CHECK (CloseHandle (current_process_handle));
>        open_process_used = 0;
>      }
> +  siginfo_er.ExceptionCode = 0;
>    inf_child_target::mourn_inferior ();
>  }
>  
> @@ -2994,6 +2998,28 @@ windows_xfer_shared_libraries (struct target_ops *ops,
>    return len != 0 ? TARGET_XFER_OK : TARGET_XFER_EOF;
>  }
>  
> +static enum target_xfer_status
> +windows_xfer_siginfo (gdb_byte *readbuf, ULONGEST offset, ULONGEST len,
> +		      ULONGEST *xfered_len)
> +{
> +  if (!siginfo_er.ExceptionCode)

siginfo_er.ExceptionCode != 0

> +    return TARGET_XFER_E_IO;
> +
> +  if (!readbuf)

readbuf == nullptr

> +    return TARGET_XFER_E_IO;
> +
> +  if (offset > sizeof (siginfo_er))
> +    return TARGET_XFER_E_IO;
> +
> +  if (offset + len > sizeof (siginfo_er))
> +    len = sizeof (siginfo_er) - offset;
> +
> +  memcpy (readbuf, (char *) &siginfo_er + offset, len);
> +  *xfered_len = len;
> +
> +  return TARGET_XFER_OK;
> +}
> +
>  enum target_xfer_status
>  windows_nat_target::xfer_partial (enum target_object object,
>  				  const char *annex, gdb_byte *readbuf,
> @@ -3009,6 +3035,9 @@ windows_nat_target::xfer_partial (enum target_object object,
>        return windows_xfer_shared_libraries (this, object, annex, readbuf,
>  					    writebuf, offset, len, xfered_len);
>  
> +    case TARGET_OBJECT_SIGNAL_INFO:
> +      return windows_xfer_siginfo (readbuf, offset, len, xfered_len);
> +
>      default:
>        if (beneath () == NULL)
>  	{
> diff --git a/gdb/windows-tdep.c b/gdb/windows-tdep.c
> index 1fc2748581..50efa31709 100644
> --- a/gdb/windows-tdep.c
> +++ b/gdb/windows-tdep.c
> @@ -153,6 +153,26 @@ static const int FULL_TIB_SIZE = 0x1000;
>  
>  static bool maint_display_all_tib = false;
>  
> +static struct gdbarch_data *windows_gdbarch_data_handle;
> +
> +struct windows_gdbarch_data
> +  {
> +    struct type *siginfo_type;
> +  };

Unindent the curly braces and the field:

{
  struct type *siginfo_type;
}

I know there are some structures formatted this way, but the new ones we add are
aligned on column 0.

> +
> +static void *
> +init_windows_gdbarch_data (struct gdbarch *gdbarch)
> +{
> +  return GDBARCH_OBSTACK_ZALLOC (gdbarch, struct windows_gdbarch_data);
> +}
> +
> +static struct windows_gdbarch_data *
> +get_windows_gdbarch_data (struct gdbarch *gdbarch)
> +{
> +  return ((struct windows_gdbarch_data *)
> +	  gdbarch_data (gdbarch, windows_gdbarch_data_handle));
> +}
> +
>  /* Define Thread Local Base pointer type.  */
>  
>  static struct type *
> @@ -648,6 +668,43 @@ windows_gdb_signal_to_target (struct gdbarch *gdbarch, enum gdb_signal signal)
>    return -1;
>  }
>  
> +static struct type *
> +windows_get_siginfo_type (struct gdbarch *gdbarch)
> +{
> +  struct windows_gdbarch_data *windows_gdbarch_data;
> +  struct type *uint_type, *void_ptr_type;
> +  struct type *siginfo_ptr_type, *siginfo_type;
> +
> +  windows_gdbarch_data = get_windows_gdbarch_data (gdbarch);
> +  if (windows_gdbarch_data->siginfo_type != NULL)
> +    return windows_gdbarch_data->siginfo_type;
> +
> +  uint_type = arch_integer_type (gdbarch, gdbarch_int_bit (gdbarch),
> +				 1, "unsigned int");

You should be able to get this one from builtin_type (gdbarch)->builtin_unsigned_int.

> +  void_ptr_type = lookup_pointer_type (builtin_type (gdbarch)->builtin_void);
> +
> +  siginfo_type = arch_composite_type (gdbarch, "EXCEPTION_RECORD",
> +				      TYPE_CODE_STRUCT);
> +  siginfo_ptr_type = arch_pointer_type (gdbarch, gdbarch_ptr_bit (gdbarch),
> +					NULL, siginfo_type);
> +
> +  append_composite_type_field (siginfo_type, "ExceptionCode", uint_type);
> +  append_composite_type_field (siginfo_type, "ExceptionFlags", uint_type);
> +  append_composite_type_field (siginfo_type, "ExceptionRecord",
> +			       siginfo_ptr_type);
> +  append_composite_type_field (siginfo_type, "ExceptionAddress",
> +			       void_ptr_type);
> +  append_composite_type_field (siginfo_type, "NumberParameters", uint_type);
> +  append_composite_type_field_aligned (siginfo_type, "ExceptionInformation",
> +				       lookup_array_range_type (void_ptr_type,
> +								0, 14),
> +				       TYPE_LENGTH (void_ptr_type));

Shouldn't you use "DWORD" and other types named like what is found in the
real structure, instead of plain "unsigned int"?  Like what is done in
windows_get_tlb_type?

As a user, I would expect that "ptype $_siginfo" shows me "DWORD" and not
"unsigned int", don't you think?

Simon
  

Btw, as mentioned here:

https://docs.microsoft.com/en-us/windows/win32/api/winnt/ns-winnt-exception_record#remarks

The layout of the structure is different depending on if the debugged process is 32 or 64 bits.

From what I understand, you code adapts to both, since it uses gdbarch_ptr_bit and uses proper
alignment for the ExceptionInformation field, but I wanted to point it out just to be sure.

Simon
  

Am Freitag, 7. Februar 2020, 22:56:18 MEZ hat Simon Marchi <simark@simark.ca> Folgendes geschrieben:

> On 2020-01-17 10:31 a.m., Hannes Domani via gdb-patches wrote:
> > diff --git a/gdb/windows-nat.c b/gdb/windows-nat.c
> > index 901e64263c..824ff4b322 100644
> > --- a/gdb/windows-nat.c
> > +++ b/gdb/windows-nat.c
> > @@ -236,6 +236,7 @@ static DEBUG_EVENT current_event;    /* The current debug event from
> >                        WaitForDebugEvent */
> >  static HANDLE current_process_handle;    /* Currently executing process */
> >  static windows_thread_info *current_thread;    /* Info on currently selected thread */
> > +static EXCEPTION_RECORD siginfo_er;    /* Contents of $_siginfo */
>
> Huh... I was going to say that it shouldn't be a global variable, but a per-inferior
> thing (or is it per-thread?), but pretty much all the state is already global...  so
> I guess it's fine.  I gather that the windows-nat does not support debugging multiple
> inferiors?  Same for win32-low in gdbserver?
>
> >
> >  /* Counts of things.  */
> >  static int exception_count = 0;
> > @@ -1166,6 +1167,8 @@ handle_exception (struct target_waitstatus *ourstatus)
> >    DWORD code = rec->ExceptionCode;
> >    handle_exception_result result = HANDLE_EXCEPTION_HANDLED;
> >
> > +  memcpy (&siginfo_er, rec, sizeof siginfo_er);
> > +
> >    ourstatus->kind = TARGET_WAITKIND_STOPPED;
> >
> >    /* Record the context of the current thread.  */
> > @@ -2862,6 +2865,7 @@ windows_nat_target::mourn_inferior ()
> >        CHECK (CloseHandle (current_process_handle));
> >        open_process_used = 0;
> >      }
> > +  siginfo_er.ExceptionCode = 0;
> >    inf_child_target::mourn_inferior ();
> >  }
> >
> > @@ -2994,6 +2998,28 @@ windows_xfer_shared_libraries (struct target_ops *ops,
> >    return len != 0 ? TARGET_XFER_OK : TARGET_XFER_EOF;
> >  }
> >
> > +static enum target_xfer_status
> > +windows_xfer_siginfo (gdb_byte *readbuf, ULONGEST offset, ULONGEST len,
> > +              ULONGEST *xfered_len)
> > +{
> > +  if (!siginfo_er.ExceptionCode)
>
> siginfo_er.ExceptionCode != 0
>
> > +    return TARGET_XFER_E_IO;
> > +
> > +  if (!readbuf)
>
> readbuf == nullptr
>
> > +    return TARGET_XFER_E_IO;
> > +
> > +  if (offset > sizeof (siginfo_er))
> > +    return TARGET_XFER_E_IO;
> > +
> > +  if (offset + len > sizeof (siginfo_er))
> > +    len = sizeof (siginfo_er) - offset;
> > +
> > +  memcpy (readbuf, (char *) &siginfo_er + offset, len);
> > +  *xfered_len = len;
> > +
> > +  return TARGET_XFER_OK;
> > +}
> > +
> >  enum target_xfer_status
> >  windows_nat_target::xfer_partial (enum target_object object,
> >                    const char *annex, gdb_byte *readbuf,
> > @@ -3009,6 +3035,9 @@ windows_nat_target::xfer_partial (enum target_object object,
> >        return windows_xfer_shared_libraries (this, object, annex, readbuf,
> >                          writebuf, offset, len, xfered_len);
> >
> > +    case TARGET_OBJECT_SIGNAL_INFO:
> > +      return windows_xfer_siginfo (readbuf, offset, len, xfered_len);
> > +
> >      default:
> >        if (beneath () == NULL)
> >      {
> > diff --git a/gdb/windows-tdep.c b/gdb/windows-tdep.c
> > index 1fc2748581..50efa31709 100644
> > --- a/gdb/windows-tdep.c
> > +++ b/gdb/windows-tdep.c
> > @@ -153,6 +153,26 @@ static const int FULL_TIB_SIZE = 0x1000;
> >
> >  static bool maint_display_all_tib = false;
> >
> > +static struct gdbarch_data *windows_gdbarch_data_handle;
> > +
> > +struct windows_gdbarch_data
> > +  {
> > +    struct type *siginfo_type;
> > +  };
>
> Unindent the curly braces and the field:
>
> {
>   struct type *siginfo_type;
> }
>
> I know there are some structures formatted this way, but the new ones we add are
> aligned on column 0.

OK, I fill fix all these coding style problems and try to remember them.


> > +
> > +static void *
> > +init_windows_gdbarch_data (struct gdbarch *gdbarch)
> > +{
> > +  return GDBARCH_OBSTACK_ZALLOC (gdbarch, struct windows_gdbarch_data);
> > +}
> > +
> > +static struct windows_gdbarch_data *
> > +get_windows_gdbarch_data (struct gdbarch *gdbarch)
> > +{
> > +  return ((struct windows_gdbarch_data *)
> > +      gdbarch_data (gdbarch, windows_gdbarch_data_handle));
> > +}
> > +
> >  /* Define Thread Local Base pointer type.  */
> >
> >  static struct type *
> > @@ -648,6 +668,43 @@ windows_gdb_signal_to_target (struct gdbarch *gdbarch, enum gdb_signal signal)
> >    return -1;
> >  }
> >
> > +static struct type *
> > +windows_get_siginfo_type (struct gdbarch *gdbarch)
> > +{
> > +  struct windows_gdbarch_data *windows_gdbarch_data;
> > +  struct type *uint_type, *void_ptr_type;
> > +  struct type *siginfo_ptr_type, *siginfo_type;
> > +
> > +  windows_gdbarch_data = get_windows_gdbarch_data (gdbarch);
> > +  if (windows_gdbarch_data->siginfo_type != NULL)
> > +    return windows_gdbarch_data->siginfo_type;
> > +
> > +  uint_type = arch_integer_type (gdbarch, gdbarch_int_bit (gdbarch),
> > +                1, "unsigned int");
>
> You should be able to get this one from builtin_type (gdbarch)->builtin_unsigned_int.

Good to know.


> > +  void_ptr_type = lookup_pointer_type (builtin_type (gdbarch)->builtin_void);
> > +
> > +  siginfo_type = arch_composite_type (gdbarch, "EXCEPTION_RECORD",
> > +                      TYPE_CODE_STRUCT);
> > +  siginfo_ptr_type = arch_pointer_type (gdbarch, gdbarch_ptr_bit (gdbarch),
> > +                    NULL, siginfo_type);
> > +
> > +  append_composite_type_field (siginfo_type, "ExceptionCode", uint_type);
> > +  append_composite_type_field (siginfo_type, "ExceptionFlags", uint_type);
> > +  append_composite_type_field (siginfo_type, "ExceptionRecord",
> > +                  siginfo_ptr_type);
> > +  append_composite_type_field (siginfo_type, "ExceptionAddress",
> > +                  void_ptr_type);
> > +  append_composite_type_field (siginfo_type, "NumberParameters", uint_type);
> > +  append_composite_type_field_aligned (siginfo_type, "ExceptionInformation",
> > +                      lookup_array_range_type (void_ptr_type,
> > +                                0, 14),
> > +                      TYPE_LENGTH (void_ptr_type));
>
>
> Shouldn't you use "DWORD" and other types named like what is found in the
> real structure, instead of plain "unsigned int"?  Like what is done in
> windows_get_tlb_type?
>
> As a user, I would expect that "ptype $_siginfo" shows me "DWORD" and not
> "unsigned int", don't you think?

Personally I didn't really care what the name of the type is,
but you are right, it probably is what users expect, so I will change the
type names accordingly.


Regards
Hannes Domani
  

Am Freitag, 7. Februar 2020, 23:07:08 MEZ hat Simon Marchi <simark@simark.ca> Folgendes geschrieben:

> Btw, as mentioned here:
>
> https://docs.microsoft.com/en-us/windows/win32/api/winnt/ns-winnt-exception_record#remarks
>
> The layout of the structure is different depending on if the debugged process is 32 or 64 bits.
>
> From what I understand, you code adapts to both, since it uses gdbarch_ptr_bit and uses proper
> alignment for the ExceptionInformation field, but I wanted to point it out just to be sure.

Yes, that's right.
I will add some comments to make this more clear.


Regards
Hannes Domani
  

>>>>> "Simon" == Simon Marchi <simark@simark.ca> writes:

Simon> Huh... I was going to say that it shouldn't be a global variable,
Simon> but a per-inferior thing (or is it per-thread?), but pretty much
Simon> all the state is already global...  so I guess it's fine.  I
Simon> gather that the windows-nat does not support debugging multiple
Simon> inferiors?  Same for win32-low in gdbserver?

Yes, that's correct.  I think it could be implemented, just nobody has
done it yet.

Tom
  

Am Dienstag, 11. Februar 2020, 16:28:37 MEZ hat Tom Tromey <tom@tromey.com> Folgendes geschrieben:

> >>>>> "Simon" == Simon Marchi <simark@simark.ca> writes:
>
> Simon> Huh... I was going to say that it shouldn't be a global variable,
> Simon> but a per-inferior thing (or is it per-thread?), but pretty much
> Simon> all the state is already global...  so I guess it's fine.  I
> Simon> gather that the windows-nat does not support debugging multiple
> Simon> inferiors?  Same for win32-low in gdbserver?
>
> Yes, that's correct.  I think it could be implemented, just nobody has
> done it yet.

How does debugging multiple inferiors work exactly?
Is this only possible with async mode (which doesn't work on Windows)?


Regards
Hannes Domani