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:   Thu, 25 Jun 2020 14:10:55 +0100
From:   Matthew Wilcox <willy@infradead.org>
To:     Michal Hocko <mhocko@kernel.org>
Cc:     linux-kernel@vger.kernel.org, linux-mm@kvack.org,
        linux-xfs@vger.kernel.org, dm-devel@redhat.com,
        Mikulas Patocka <mpatocka@redhat.com>,
        Jens Axboe <axboe@kernel.dk>, NeilBrown <neilb@suse.de>
Subject: Re: [PATCH 6/6] mm: Add memalloc_nowait
Message-ID: <20200625131055.GC7703@casper.infradead.org>
References: <20200625113122.7540-1-willy@infradead.org>
 <20200625113122.7540-7-willy@infradead.org>
 <20200625124017.GL1320@dhcp22.suse.cz>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20200625124017.GL1320@dhcp22.suse.cz>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Thu, Jun 25, 2020 at 02:40:17PM +0200, Michal Hocko wrote:
> On Thu 25-06-20 12:31:22, Matthew Wilcox wrote:
> > Similar to memalloc_noio() and memalloc_nofs(), memalloc_nowait()
> > guarantees we will not sleep to reclaim memory.  Use it to simplify
> > dm-bufio's allocations.
> 
> memalloc_nowait is a good idea! I suspect the primary usecase would be
> vmalloc.

That's funny.  My use case is allocating page tables in an RCU protected
page fault handler.  Jens' use case is allocating page cache.  This one
is a vmalloc consumer (which is also indirectly page table allocation).

> > @@ -877,7 +857,9 @@ static struct dm_buffer *__alloc_buffer_wait_no_callback(struct dm_bufio_client
> >  	 */
> >  	while (1) {
> >  		if (dm_bufio_cache_size_latch != 1) {
> > -			b = alloc_buffer(c, GFP_NOWAIT | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN);
> > +			unsigned nowait_flag = memalloc_nowait_save();
> > +			b = alloc_buffer(c, GFP_KERNEL | __GFP_NOMEMALLOC | __GFP_NOWARN);
> > +			memalloc_nowait_restore(nowait_flag);
> 
> This looks confusing though. I am not familiar with alloc_buffer and
> there is quite some tweaking around __GFP_NORETRY in alloc_buffer_data
> which I do not follow but GFP_KERNEL just struck my eyes. So why cannot
> we have 
> 		alloc_buffer(GFP_NOWAIT | __GFP_NOMEMALLOC | __GFP_NOWARN);

Actually, I wanted to ask about the proliferation of __GFP_NOMEMALLOC
in the block layer.  Am I right in thinking it really has no effect
unless GFP_ATOMIC is set?  It seems like a magic flag that some driver
developers are sprinkling around randomly, so we probably need to clarify
the documentation on it.

What I was trying to do was just use the memalloc_nofoo API to control
what was going on and then the driver can just use GFP_KERNEL.  I should
probably have completed that thought before sending the patches out.