Re: lockdep oddity

[Daniel Walker]

On Wed, 2006-09-06 at 10:40 +0200, Ingo Molnar wrote:
> * Hua Zhong <hzhong@xxxxxxxxx> wrote:
> 
> > We are just trading accuracy for speed here.
> 
> no, we are trading _both_ accuracy and speed here! a global 'likeliness' 
> pointer for commonly executed codepaths is causing global cacheline 
> ping-pongs - which is as bad as it gets.

Up stream or no, would be better for it to again be light weight.

> the right approach, which incidentally would also be perfectly accurate, 
> is to store an alloc_percpu()-ed pointer at the call site, not the 
> counter itself.

I don't think it could be done via the macro. If it were called during
run time it would have to be special alloc_percpu() that didn't call
back into the profiling code (which almost everything does do).

> the current code needs more work before it can go upstream i think.

It was never really planned to go upstream. It's ultimately a debugging
feature that's really only needed in -mm .. 

Daniel

-
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/