Re: [commit] Allow cached cooked reads

[Andrew Cagney <ac131313@redhat.com>]

> On Jun 16,  9:22am, Andrew Cagney wrote:
> 
> 
>> Just stumbled across this.  When trying to save cooked registers the 
>> regcache was triggering an assertion failure instead of ignoring a bogus 
>> request.
>> 
>> 2003-06-16  Andrew Cagney  <cagney@redhat.com>
>> 
>> 	* regcache.c (do_cooked_read): Do not use register_valid_p.
>> 
>> Index: regcache.c
>> ===================================================================
>> RCS file: /cvs/src/src/gdb/regcache.c,v
>> retrieving revision 1.87
>> diff -u -r1.87 regcache.c
>> --- regcache.c	9 Jun 2003 01:02:06 -0000	1.87
>> +++ regcache.c	16 Jun 2003 13:18:46 -0000
>> @@ -423,8 +423,7 @@
>>  do_cooked_read (void *src, int regnum, void *buf)
>>  {
>>    struct regcache *regcache = src;
>> -  if (!regcache_valid_p (regcache, regnum)
>> -      && regcache->readonly_p)
>> +  if (!regcache->register_valid_p[regnum] && regcache->readonly_p)
>>      /* Don't even think about fetching a register from a read-only
>>         cache when the register isn't yet valid.  There isn't a target
>>         from which the register value can be fetched.  */
> 
> 
> Which assertion was failing?  The check for regcache != NULL or the
> bounds check?

The bounds check.  If regcache were NULL, GDB would be sunk.

> I'm wondering if the new code above should include some bounds checks. 
> Alternately, go back to using regcache_valid_p() and weaken the
> assertions in regcache_valid_p() somewhat.  E.g, perhaps rewrite
> regcache_valid_p() from:



> int
> regcache_valid_p (struct regcache *regcache, int regnum)
> {
>   gdb_assert (regcache != NULL);
>   gdb_assert (regnum >= 0 && regnum < regcache->descr->nr_raw_registers);
>   return regcache->register_valid_p[regnum];
> }
> 
> to:
> 
> int
> regcache_valid_p (struct regcache *regcache, int regnum)
> {
>   gdb_assert (regcache != NULL);
>   gdb_assert (regnum >= 0);
>   return regnum < regcache->descr->nr_raw_registers)
>          && regcache->register_valid_p[regnum];
> }

The intent of the external function is to indicate to external code if a 
raw register is valid (as in fetched).  So external code trying to call 
this with regnum >= NUM_REGS is broken.

Andrew