Re: [PATCH v2 2/2] mm: prevent gup_fast from racing with COW during fork

From: John Hubbard <jhubbard@nvidia.com>
To: "Ahmed S. Darwish" <a.darwish@linutronix.de>
Cc: Jason Gunthorpe <jgg@nvidia.com>, Peter Xu <peterx@redhat.com>,
	<linux-kernel@vger.kernel.org>,
	Linus Torvalds <torvalds@linux-foundation.org>,
	Andrea Arcangeli <aarcange@redhat.com>,
	Andrew Morton <akpm@linux-foundation.org>,
	Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>,
	Christoph Hellwig <hch@lst.de>, Hugh Dickins <hughd@google.com>,
	Jan Kara <jack@suse.cz>, Jann Horn <jannh@google.com>,
	Kirill Shutemov <kirill@shutemov.name>,
	Kirill Tkhai <ktkhai@virtuozzo.com>,
	Leon Romanovsky <leonro@nvidia.com>,
	Linux-MM <linux-mm@kvack.org>, "Michal Hocko" <mhocko@suse.com>,
	Oleg Nesterov <oleg@redhat.com>,
	Peter Zijlstra <peterz@infradead.org>,
	Ingo Molnar <mingo@redhat.com>, Will Deacon <will@kernel.org>,
	Thomas Gleixner <tglx@linutronix.de>,
	Sebastian Siewior <bigeasy@linutronix.de>
Subject: Re: [PATCH v2 2/2] mm: prevent gup_fast from racing with COW during fork
Date: Mon, 2 Nov 2020 23:05:24 -0800	[thread overview]
Message-ID: <869a07ad-97ba-a965-76ef-3b9663491a21@nvidia.com> (raw)
In-Reply-To: <20201103065225.GA63301@lx-t490>

On 11/2/20 10:52 PM, Ahmed S. Darwish wrote:
> On Mon, Nov 02, 2020 at 06:20:45PM -0800, John Hubbard wrote:
>> On 11/2/20 4:41 PM, Ahmed S. Darwish wrote:
>>> On Mon, Nov 02, 2020 at 08:25:32PM -0400, Jason Gunthorpe wrote:
>>>> On Tue, Nov 03, 2020 at 01:17:12AM +0100, Ahmed S. Darwish wrote:
>>>>
>>>>> Please stick with the official exported API: raw_write_seqcount_begin().
>>>>
>>>> How did you know this was 'offical exported API' ??
>>>>
>>>
>>> All the official exported seqlock.h APIs are marked with verbose
>>> kernel-doc annotations on top. The rest are internal...
>>>
>>
>> OK, but no one here was able to deduce that, probably because there is not
>> enough consistency throughout the kernel to be able to assume such things--even
>> though your seqlock project is internally consistent. It's just not *quite*
>> enough communication.
>>
>> I think if we added the following it would be very nice:
>>
> 
> The problem is, I've already documented seqlock.h to death.... There are
> more comments than code in there, and there is "seqlock.rst" under
> Documentation/ to further describe the big picture.
> 
> There comes a point where you decide what level of documentation to add,
> and what level to skip.
> 
> Because in the end, you don't want to confuse "Joe, the general driver
> developer" with too much details that's not relevant to their task at
> hand.  (I work in the Embedded domain, and I've seen so much ugly code
> from embedded drivers/SoC developers already, sorry)
> 
> See for example my reply to Linus, where any talk about the lockdep-free
> and barrier-free parts of the API was explicitly not mentioned in
> seqlock.rst. This was done on purpose: 1) you want to keep the generic
> case simple, but the special case do-able, 2) you want to encourage
> people to use the standard entry/exit points as much as possible.
> 

Well, OK, I'll leave it alone. This was not a normal case for the documentation,
anyway. We are using seqcount-ing in a little bit different way than usual,
so I guess an email thread like this is actually the proper way to work it out.

Just wanted to be sure that we didn't miss an opportunity to clarify anything
that needed clarifying it. Thanks for the detailed responses throughout!

thanks,
-- 
John Hubbard
NVIDIA