Re: [PATCH 3/5] fbsd-nat: Pass an optional register base to the register set helpers.

[John Baldwin <jhb@FreeBSD.org>]

On 11/22/22 12:38 PM, Simon Marchi wrote:
> On 7/7/22 20:58, John Baldwin wrote:
>> This is needed to permit using the helpers for register sets with a
>> variable base.  In particular regnum needs to be converted into a
>> relative register number before passed to regcache_map_supplies.
>> ---
>>   gdb/fbsd-nat.c | 34 +++++++++++++++++++---------------
>>   gdb/fbsd-nat.h | 49 ++++++++++++++++++++++++++++---------------------
>>   2 files changed, 47 insertions(+), 36 deletions(-)
>>
>> diff --git a/gdb/fbsd-nat.c b/gdb/fbsd-nat.c
>> index 9d7383ac0ab..edcd3fc3ed1 100644
>> --- a/gdb/fbsd-nat.c
>> +++ b/gdb/fbsd-nat.c
>> @@ -1732,14 +1732,15 @@ fbsd_nat_target::supports_disable_randomization ()
>>   bool
>>   fbsd_nat_target::fetch_register_set (struct regcache *regcache, int regnum,
>>   				     int fetch_op, const struct regset *regset,
>> -				     void *regs, size_t size)
>> +				     int regbase, void *regs, size_t size)
>>   {
>>     const struct regcache_map_entry *map
>>       = (const struct regcache_map_entry *) regset->regmap;
>>     pid_t pid = get_ptrace_pid (regcache->ptid ());
>>   
>> -  if (regnum == -1 || regcache_map_supplies (map, regnum, regcache->arch(),
>> -					     size))
>> +  if (regnum == -1
>> +      || (regnum >= regbase && regcache_map_supplies (map, regnum - regbase,
>> +						      regcache->arch(), size)))
> 
> My understanding is that it would be an internal error if the caller
> passed a regnum that is < regbase.  In order words, should you do:
> 
>    gdb_assert (regnum >= regbase);
> 
> ?
> 
> Actually, after reading the following patches, I think the answer is
> no.  When requesting a single register, we will still go through all the
> fetch functions and we expect them to do nothing if the register is not
> theirs.  Leaving the question there in case others wonder about the same
> thing.
> 
> I was also wondering if it would be better to do the absolute ->
> relative conversion in the caller.  All this function knows is the
> regset-relative view of things.  In the end it doesn't matter much, it
> just moves a condition from here to the callers.  I'm fine with whatever
> you prefer.

If you do the adjustment to an absolute value in the caller you have to still
special case for regnum == -1 when fetching all registers.  The current approach
here means that the special case for that is handled down in
regcache:transfer_regset which already has a special case for -1.

-- 
John Baldwin