Re: [Qemu-devel] [PATCH 08/10] coroutine: process the coroutines woken by child when child yield

[Kevin Wolf]

Am 03.04.2012 10:38, schrieb Lai Jiangshan:
> If the child wake up someone, process them.
> 
> It would the child complete its works if woken coroutine release the locks
> that the child needs.
> 
> It may help for the cache, if the child wake up some someone, they are
> probably accessing the same data.
> 
> Signed-off-by: Lai Jiangshan <address@hidden>

Are you very sure that this is correct, particularly with respect to
reentrancy of nested coroutines? This is why this has become a bottom
half in the first place (instead of directly reentering the unlocked
coroutine in qemu_co_mutex_unlock), and it's not completely clear to me
if the problem doesn't exist when you do it after a coroutine has exited
instead of in an BH.

I'd be very careful with such changes.

Kevin