[PATCH RFC 09/18] staging: vchiq_core: do not initialize semaphores twice

Stefan Wahren stefan.wahren at i2se.com
Tue Nov 6 16:06:31 UTC 2018


Am 06.11.18 um 16:41 schrieb Nicolas Saenz Julienne:
> Hi Stefan,
> thanks for spending the time reviewing the code. I took note of the
> rest of comments.
>
> On Sun, 2018-10-28 at 21:45 +0100, Stefan Wahren wrote:
>> Hi Nicolas,
>>
>>> Nicolas Saenz Julienne <nsaenzjulienne at suse.de> hat am 26. Oktober
>>> 2018 um 15:48 geschrieben:
>>>
>>>
>>> vchiq_init_state() initialises a series of semaphores to then call
>>> remote_event_create() on the same semaphores, which initializes
>>> them
>>> again.
>> i would prefer to have all init stuff at one place in
>> vchiq_init_state() and drop this ugliness from remote_event_create()
>> instead. Is this possible?
> As I'm sure you're aware of, REMOTE_EVENT_T is shared between the CPU
> and VC4, which can't be expanded. And since storing a pointer is out of
> question because of arm64, I can only think of storing an index to an
> array of completions in the shared structure instead of the pointer
> magic implemented right now. It would be a little more explicit. Then
> we could completely decouple both initializations. I'm not sure if it's
> similar to what you had in mind. 

I don't think so, this was my intention:

 static inline void
 remote_event_create(VCHIQ_STATE_T *state, REMOTE_EVENT_T *event)
 {
    event->armed = 0;
    /* Don't clear the 'fired' flag because it may already have been set
    ** by the other side. */
-    sema_init((struct semaphore *)((char *)state + event->event), 0);
 }


>
> On a semi-related topic, I'm curious to know why these shared
> structures aren't set with the "__packed" preprocessor macro. Any
> ideas? As fas as I've been told, in general, the compiler may reorder
> or add unexpected padding to any structure. Which would be very bad in
> this case.

This would be better, but i assume the firmware side uses the same
source code. So using __packed only on ARM side could also break :-(

>
> Regards,
> Nicolas


More information about the devel mailing list