[Prism54-devel] Re: CVS Update

[Aurelien Alleaume]

After reading related code, I think the current code is OK. We could 
change "if(!priv->mib) return;" to "BUG_ON(!priv->mib);".
But if we want the driver safe, we need to ensure nobody accesses  the 
priv struct after netdev is unregistered:
* I think that unregistering netdevice syncs and disable ioctl (not sure 
but it 'should' be true).
* We must be sure that no wireless-stats update remains scheduled. (use 
  "flush_scheduled_work();")
Then it is safe to call "mgt_clean();"

The only thing missing is thus a "flush_scheduled_work();" in 
prism54_remove, right after "unregister_netdev()".

	...And justice for all
	Aurelien

Margit Schubert-While wrote:
> At 13:37 06.08.2004 +0200, Aurelien scribeth:
> 
>> Hi Margit,
>> the (!priv->mib) check in mgt_get_request should be a debugging test. 
>> It should never happen. At least  if the rest is well coded (which 
>> isn't the  case right now)... To solve the races with alloc/free of 
>> mib memory, we should probably use the mib_sem. ie : take a read lock 
>> before  accessing this memory and a write lock before allocating/freeing.
>> There sould not be any need to recode all mgt_get_request calls in 
>> isl_ioctl. Changing the locking in some function of oid_mgt.c should  
>> be enough.
>> I'm gonna take a deeper look at this right now to see if all this is 
>> possible.
> 
> 
> Yes, I agree.
> Go for it !
> I just want to get the kernel driver in good shape.
> 
> Thanks
> 
> Cheers
> 
> Margit
> 
> 
> _______________________________________________
> Prism54-devel mailing list
> Prism54-devel at prism54.org
> http://prism54.org/mailman/listinfo/prism54-devel
> 
>