Re: [RFC PATCH v2 1/3] getcpu_cache system call: cache CPU number of running thread

[Mathieu Desnoyers <mathieu.desnoyers@efficios.com>]

----- On Jan 28, 2016, at 6:12 AM, heiko carstens heiko.carstens@de.ibm.com wrote:

> On Wed, Jan 27, 2016 at 10:47:37PM +0000, Mathieu Desnoyers wrote:
>> ----- On Jan 27, 2016, at 5:11 PM, Josh Triplett josh@joshtriplett.org wrote:
>> 
>> > On Wed, Jan 27, 2016 at 09:34:35PM +0000, Mathieu Desnoyers wrote:
>> >> ----- On Jan 27, 2016, at 12:37 PM, Thomas Gleixner tglx@linutronix.de wrote:
>> >> 
>> >> > On Wed, 27 Jan 2016, Thomas Gleixner wrote:
>> >> > 
>> >> >> On Wed, 27 Jan 2016, Mathieu Desnoyers wrote:
>> >> >> > ----- On Jan 27, 2016, at 12:22 PM, Thomas Gleixner tglx@linutronix.de wrote:
>> >> >> > Sounds fair. What is the recommended typing for "ptr" then ?
>> >> >> > uint32_t ** or uint32_t * ?
>> >> >> > 
>> >> >> > It would be expected to pass a "uint32_t *" for the set
>> >> >> > operation, but the "get" operation requires a "uint32_t **".
>> >> >> 
>> >> >> Well, you can't change the types depending on the opcode, so you need to stick
>> >> >> with **.
>> >> > 
>> >> > Alternatively you make it:
>> >> > 
>> >> >  (opcode, *newptr, **oldptr, flags);
>> >> 
>> >> I'm tempted to stick to (opcode, **ptr, flags), because
>> >> other syscalls that have "*newptr", "**oldptr"
>> >> typically have them because they save the current state
>> >> into oldptr, and set the new state, which is really
>> >> not the case here. To eliminate any risk of confusion,
>> >> I am tempted to keep a single "**ptr".
>> >> 
>> >> Unless someone has a better idea...
>> > 
>> > Either that or you could define it as "void *" and interpret it based on
>> > flags, but that seems unfortunate; let's not imitate ioctl-style
>> > typeless parameters.  I'd stick with the double pointer and the current
>> > behavior.
>> 
>> Allright, will do! Thanks for the feedback :)
> 
> Please don't forget that you also need to implement compat handling since
> the size of the pointer that is being pointed to is only four bytes for
> compat tasks.

Oops, forgot about that. Will update my upcoming v3 to handle this properly.

Well spotted ! Thanks!

Mathieu


-- 
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com