Re: [RFC-PATCH 1/2] mm: Add __GFP_NO_LOCKS flag

[Paul E. McKenney]

On Sat, Aug 15, 2020 at 10:27:54AM +0200, Peter Zijlstra wrote:
> On Sat, Aug 15, 2020 at 01:14:53AM +0200, Thomas Gleixner wrote:
> > 
> > As a matter of fact I assume^Wdeclare that removing struct rcu_head which
> > provides a fallback is not an option at all. I know that you want to,
> > but it wont work ever. Dream on, but as we agreed on recently there is
> > this thing called reality which ruins everything.
> 
> It never was going to work, freeing memory can never hard rely on the
> success of allocating memory.

In neither case does the freeing of memory rely hard-rely on the success
of allocating memory.  This is because there is a fallback in both cases
should allocation fail.

Given an rcu_head structure, we use that, and accept the extra cache
misses at callback-invocation time.  Otherwise, without an rcu_head
structure, the allocation parameters are carefully chosen to avoid
indefinite sleeping, meaning that the allocation attempt either succeeds
or fails within a reasonable amount of time.  And upon failure we invoke
synchronize_rcu(), then immediately free.  Which is slow, but then again
life is like that under OOM conditions.

And yes, this means that the price of leaving the rcu_head structure out
of the structure to be freed is that you must call kvfree_free() from
a sleepable context.  If you don't like being restricted to sleepable
context, you can always supply the rcu_head structure.

							Thanx, Paul