Re: [PATCH 1/1] aarch64: fix a crash during maintenance print cooked-registers

[Luis Machado <luis.machado@arm.com>]

On 2/3/25 14:38, Gerlicher, Klaus wrote:
> Hi Luis,
> 
> "info all-registers" ignores registers w/o name but "maint print cooked-registers" does not and I think that hasn't changed. I think maintenance commands aren't supposed to hide anything since they are used more in GDB development than by a GDB user.

That make sense. Probably an oversight on our end when we first enabled this.

> 
> Anyway, thanks for the quick feedback, I'll push this some time soon.

Sounds good. Thanks.

> 
> Thanks
> Klaus
> 
>> -----Original Message-----
>> From: Luis Machado <luis.machado@arm.com>
>> Sent: Monday, February 3, 2025 3:28 PM
>> To: Gerlicher, Klaus <klaus.gerlicher@intel.com>; gdb-
>> patches@sourceware.org
>> Subject: Re: [PATCH 1/1] aarch64: fix a crash during maintenance print
>> cooked-registers
>>
>> Hi,
>>
>> On 2/3/25 13:52, Klaus Gerlicher wrote:
>>> Hi all, Luis,
>>>
>>> I found this accidently while looking at Thiago's GDBserver improvements.
>> Luis
>>> I think added the pseudo register in question, so maybe he can comment
>>> on this issue. I used a aarch64 qemu 9.2.
>>
>> Thanks for catching this.
>>
>>>
>>> Thanks
>>> Klaus
>>>
>>> On aarch64 with pauth enabled I can see a crash when
>>> using "maintenance print cooked-registers".
>>>
>>> This happens because the register dump code tries to read
>>> a pseudo reg that is not handled here because it is supposedly
>>> only used in unwinding.
>>
>> Indeed. ra_sign_state is only used to track which frames have PAC enabled or
>> not.
>>
>> I was wondering why I had not caught this before, and I just noticed my VM
>> had pauth disabled. :-(
>>
>> Enabling it, now I see the crash. In any case...
>>
>>>
>>> Fix this by returning a zero value typed as a built-in uint64.
>>> ---
>>>  gdb/aarch64-tdep.c | 3 +++
>>>  1 file changed, 3 insertions(+)
>>>
>>> diff --git a/gdb/aarch64-tdep.c b/gdb/aarch64-tdep.c
>>> index 840f9877361..6e712b12b86 100644
>>> --- a/gdb/aarch64-tdep.c
>>> +++ b/gdb/aarch64-tdep.c
>>> @@ -3290,6 +3290,9 @@ aarch64_pseudo_read_value (gdbarch *gdbarch,
>> const frame_info_ptr &next_frame,
>>>      return aarch64_pseudo_read_value_1 (next_frame, pseudo_reg_num,
>>>  					pseudo_offset -
>> AARCH64_SVE_V0_REGNUM);
>>>
>>> +  if (tdep->has_pauth () && pseudo_reg_num == tdep-
>>> ra_sign_state_regnum)
>>> +    return value::zero (builtin_type (gdbarch)->builtin_uint64, lval_register);
>>> +
>>>    gdb_assert_not_reached ("regnum out of bound");
>>>  }
>>>
>>
>> ... the fix above looks OK to me. Though it makes me wonder why we starting
>> hitting this.
>>
>> Internally we don't assign a name to this register so it technically doesn't show
>> up in any
>> listing. Maybe something changed and we now list even pseudo registers
>> without a name?
>>
>> Approved-By: Luis Machado <luis.machado@arm.com>
> Intel Deutschland GmbH
> Registered Address: Am Campeon 10, 85579 Neubiberg, Germany
> Tel: +49 89 99 8853-0, www.intel.de
> Managing Directors: Sean Fennelly, Jeffrey Schneiderman, Tiffany Doon Silva
> Chairperson of the Supervisory Board: Nicole Lau
> Registered Office: Munich
> Commercial Register: Amtsgericht Muenchen HRB 186928