Re: win32-arm-low.c regptr 96 bits stored in 32 bit variable

Surely nobody is interested in the long-dead FPA architecture these
days.  I'm not sure why GDB still supports it.

R.

On 28/10/2018 16:47, Simon Marchi wrote:
> On 2018-10-26 18:40, Bill Morgan wrote:
>> Should this static variable ULONG zero be at least 96 bits?
>>
>> static char *
>> regptr (CONTEXT* c, int r)
>> {
>>   if (mappings[r] < 0)
>>   {
>>     static ULONG zero;
>>     /* Always force value to zero, in case the user tried to write
>>        to this register before.  */
>>     zero = 0;
>>     return (char *) &zero;
>>   }
>>   else
>>     return (char *) c + mappings[r];
>> }
>>
>> reg-arm.dat shows 96 bits for the ones that have mappings[r] == -1
>>
>> name:arm
>> xmlarch:arm
>> expedite:r11,sp,pc
>> 32:r0
>> 32:r1
>> 32:r2
>> 32:r3
>> 32:r4
>> 32:r5
>> 32:r6
>> 32:r7
>> 32:r8
>> 32:r9
>> 32:r10
>> 32:r11
>> 32:r12
>> 32:sp
>> 32:lr
>> 32:pc
>> 96:f0
>> 96:f1
>> 96:f2
>> 96:f3
>> 96:f4
>> 96:f5
>> 96:f6
>> 96:f7
>> 32:fps
>> 32:cpsr
> 
> Hi Bill,
> 
> By inspection, it does seem like a mistake, and that we would need to
> return a pointer to a buffer at least as big as register r.  But I have
> no idea how to build/run/test gdbserver on win32/arm.  If you are able
> to confirm that there is a problem and test a fix, could you please
> provide a patch?
> 
> To avoid this kind of problem again, we could return a pointer to a
> dynamically-sized buffer adjusted to the size of the register. 
> Something like this:
> 
> static char *
> regptr (CONTEXT* c, struct regcache *regcache, int r)
> {
>   if (mappings[r] < 0)
>   {
>     static gdb::byte_vector zero;
>     /* Always force value to zero, in case the user tried to write
>        to this register before.  */
>     zero.assign (regcache_register_size (regcache, r), 0);
>     return (char *) zero.data ();
>   }
>   else
>     return (char *) c + mappings[r];
> }
> 
> Simon