Re: [PATCH v4 2/9] x86: support kmap_atomic_pfn_t() for persistent memory

From: Dan Williams <dan.j.williams@intel.com>
To: Christoph Hellwig <hch@lst.de>
Cc: Matthew Wilcox <willy@linux.intel.com>,
	"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
	Jens Axboe <axboe@kernel.dk>, Boaz Harrosh <boaz@plexistor.com>,
	david <david@fromorbit.com>,
	linux-arch@vger.kernel.org, Arnd Bergmann <arnd@arndb.de>,
	Ross Zwisler <ross.zwisler@linux.intel.com>,
	"linux-nvdimm@lists.01.org" <linux-nvdimm@lists.01.org>,
	Benjamin Herrenschmidt <benh@kernel.crashing.org>,
	linux-fsdevel <linux-fsdevel@vger.kernel.org>,
	Heiko Carstens <heiko.carstens@de.ibm.com>,
	Tejun Heo <tj@kernel.org>, Paul Mackerras <paulus@samba.org>,
	"H. Peter Anvin" <hpa@zytor.com>,
	Martin Schwidefsky <schwidefsky@de.ibm.com>,
	Andrew Morton <akpm@linux-foundation.org>,
	"torvalds@linux-foundation.org" <torvalds@linux-foundation.org>,
	Ingo Molnar <mingo@kernel.org>
Subject: Re: [PATCH v4 2/9] x86: support kmap_atomic_pfn_t() for persistent memory
Date: Wed, 10 Jun 2015 08:36:48 -0700	[thread overview]
Message-ID: <CAPcyv4hnd1JRwLYapo_s=vXFYHUA6Cv-ZpJt+Y3_tTCxQsJUPg@mail.gmail.com> (raw)
In-Reply-To: <20150610151100.GA12757@lst.de>

On Wed, Jun 10, 2015 at 8:11 AM, Christoph Hellwig <hch@lst.de> wrote:
> On Wed, Jun 10, 2015 at 11:03:35AM -0400, Matthew Wilcox wrote:
>> On Wed, Jun 10, 2015 at 02:12:02PM +0200, Christoph Hellwig wrote:
>> > Btw, I don't think this actually is safe without refcounting your kmap
>> > structure.
>> >
>> > The driver model ->remove callback can be called at any time, which
>> > will ioremap the memory and remap the kmap structure.  But at this
>> > point a user might still be using it.
>>
>> Won't the device data structures be pinned by the refcount on the bdev?
>
> An open filesystem only keeps a reference on the request_queue.  The
> underlying driver model ->remove method will still be called on
> a surprise removal.

On surprise removal my expectation is that the driver keeps the the
ioremap mapping alive for at least a synchronize_rcu() period.  With
that in place the rcu_read_lock() in kmap_atomic_pfn_t() should keep
the mapping alive for the short duration, or otherwise prevent new
mappings.  However, I missed converting the driver to order its
iounmap() with respect to the pfn range registration via devres, will
fix.