Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Fri, 26 Feb 2021 12:21:43 -0400
From:   Jason Gunthorpe <jgg@ziepe.ca>
To:     Matthew Wilcox <willy@infradead.org>
Cc:     linux-mm@kvack.org, linuxppc-dev@lists.ozlabs.org,
        linux-s390@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: Freeing page tables through RCU
Message-ID: <20210226162143.GW2643399@ziepe.ca>
References: <20210225205820.GC2858050@casper.infradead.org>
 <20210226144200.GV2643399@ziepe.ca>
 <20210226160354.GE2723601@casper.infradead.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20210226160354.GE2723601@casper.infradead.org>
Precedence: bulk

On Fri, Feb 26, 2021 at 04:03:54PM +0000, Matthew Wilcox wrote:
> On Fri, Feb 26, 2021 at 10:42:00AM -0400, Jason Gunthorpe wrote:
> > On Thu, Feb 25, 2021 at 08:58:20PM +0000, Matthew Wilcox wrote:
> > 
> > > I'd like to hear better ideas than this.
> > 
> > You didn't like my suggestion to put a sleepable lock around the
> > freeing of page tables during flushing?
> > 
> > I still don't see how you convert the sleepable page walkers to use
> > rcu??
> 
> I don't want to convert the sleepable ones to use RCU ... I want to
> convert the non-sleeping ones to use RCU.  

Why? Convert them to use the normal page table spinlocks?

It makes everything much more understandable if it is locked properly.

The lockless walks should be reserved for special places because they
are actually hard to do properly.

> A page_table_free_lock might work, but it might have its own
> problems later (eg a sleeping lock can't be acquired under RCU or
> spinlock, and it can't be a spinlock because it'd have to be held
> while we wait for IPIs).

The mmap_sem today is serving the function of the page_table_free_lock
idea, so I think a sleepable lock is already proven to work?

Jason