Re: [PATCH 0 of 4] Generic AIO by scheduling stacks

[Linus Torvalds]

On Tue, 30 Jan 2007, Linus Torvalds wrote:
> 
> Does that mean that we might not have some cases where we'd need to make 
> sure we do things differently? Of course not. Something migt show up. But 
> this actually makes it very clear what the difference between "struct 
> thread_struct" and "struct task_struct" are. One is shared between 
> fibrils, the other isn't.

Btw, this is also something where we should just disallow certain system 
calls from being done through the asynchronous method. 

Notably, clone/fork(), execve() and exit() are all things that we probably 
simply shouldn't allow as "AIO" events.

The process handling ones are obvious: they are very much about the shared 
"struct task_struct", so they rather clearly should only done "natively".

More interesting is the question about "close()", though. Currently we 
have an optimization (fget/fput_light) that basically boils down to "we 
know we are the only owners". That optimization becomes more "interesting" 
with AIO - we need to disable it when fibrils are active (because other 
fibrils or the main thread can do it), but we can still keep it for the 
non-fibril case.

So we can certainly allow close() to happen in a fibril, but at the same 
time, this is an area where just the *existence* of fibrils means that 
certain other decisions that were thread-related may be modified to be 
aware of the micro-threads too.

I suspect there are rather few of those, though. The only one I can think 
of is literally the fget/fput_light() case, but there could be others.

			Linus
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/