Re: [PATCH] might_sleep() improvements

[Nick Piggin <piggin@cyberone.com.au>]

Mitchell Blank Jr wrote:

>Nick Piggin wrote:
>
>>>Andrew - I thought this might be appropriate for -mm kernels.
>>>
>>>This patch makes the following improvements to might_sleep():
>>>
>>>o Add a "might_sleep_if()" macro for when we might sleep only if some
>>> condition is met.  I think this is a bit better than the currently used
>>> "if (cond) might_sleep();" since it's clearer that the test won't be
>>> compiled in if spinlock sleep debugging is turned off.  (Obviously
>>> gcc is smart enough to omit simple conditions in that case)  It also
>>> looks cleaner, IMO.  Think of it as analogous to BUG()/BUG_ON().
>>>
>>>
>>I think these should be pushed down to where the sleeping
>>actually happens if possible.
>>
>
>No, you want to generate the warning as early as possible in case the
>sleeping case happens very infrequently.  For instance:
>
>	newskb = skb_unshare(skb, GFP_KERNEL);
>
>might not even need to do any allocation (much less a sleep) in 99.9% of
>cases, but it's still a bug if it's called in atomic context and we want
>spinlock sleep debugging to catch that for us.
>
>

Yeah well in this case I guess skb_unshare is as low as might_sleep
can be pushed. I guess I don't have a problem with might_sleep_if as
long as its used nicely.