Re: [patch RFC 00/15] mm/highmem: Provide a preemptible variant of kmap_atomic & friends

From: Thomas Gleixner
Date: Wed Sep 23 2020 - 16:55:59 EST

On Wed, Sep 23 2020 at 11:52, Steven Rostedt wrote:
> On Wed, 23 Sep 2020 10:40:32 +0200
> peterz@xxxxxxxxxxxxx wrote:
>> However, with migrate_disable() we can have each task preempted in a
>> migrate_disable() region, worse we can stack them all on the _same_ CPU
>> (super ridiculous odds, sure). And then we end up only able to run one
>> task, with the rest of the CPUs picking their nose.
> What if we just made migrate_disable() a local_lock() available for !RT?
> I mean make it a priority inheritance PER CPU lock.
> That is, no two tasks could do a migrate_disable() on the same CPU? If
> one task does a migrate_disable() and then gets preempted and the
> preempting task does a migrate_disable() on the same CPU, it will block
> and wait for the first task to do a migrate_enable().
> No two tasks on the same CPU could enter the migrate_disable() section
> simultaneously, just like no two tasks could enter a preempt_disable()
> section.
> In essence, we just allow local_lock() to be used for both RT and !RT.
> Perhaps make migrate_disable() an anonymous local_lock()?
> This should lower the SHC in theory, if you can't have stacked migrate
> disables on the same CPU.

I'm pretty sure this ends up in locking hell pretty fast and aside of
that it's not working for scenarios like:


-> #PF
-> schedule()

which brought us into that discussion in the first place. You would stop
any other migrate disable user from running until the page fault is