Re: [PATCH 1/7] atomic: Export fetch_or()

From: Chris Metcalf <cmetcalf@ezchip.com>
To: Frederic Weisbecker <fweisbec@gmail.com>
Cc: LKML <linux-kernel@vger.kernel.org>,
	Peter Zijlstra <peterz@infradead.org>,
	Thomas Gleixner <tglx@linutronix.de>,
	Luiz Capitulino <lcapitulino@redhat.com>,
	Christoph Lameter <cl@linux.com>, Ingo Molnar <mingo@kernel.org>,
	Viresh Kumar <viresh.kumar@linaro.org>,
	Rik van Riel <riel@redhat.com>
Subject: Re: [PATCH 1/7] atomic: Export fetch_or()
Date: Mon, 30 Nov 2015 13:17:14 -0500	[thread overview]
Message-ID: <565C92AA.4060109@ezchip.com> (raw)
In-Reply-To: <20151130173619.GA9444@lerouge>

(Resending as plain text.  Not sure what Thunderbird was smoking to make
this message multipart/alternative originally...)

On 11/30/2015 12:36 PM, Frederic Weisbecker wrote:
> On Tue, Nov 24, 2015 at 04:48:35PM -0500, Chris Metcalf wrote:
>> >Unless you want to take a big diversion into atomics, I'd be tempted
>> >to leave Peter's macro alone and just write it off as necessary evil
>> >to handle the fact that thread_info.flags is all kinds of different sizes
>> >and types on different platforms, and definitely never an atomic_t.
>> >Instead just create an inline function atomic_return_or(), or
>> >whatever name you prefer, that operates on an atomic_t, and use
>> >the atomic_t type for your structure field.  It's clearly a win to mark
>> >the data types as being atomic to the extent we can do so, I think.
> I agree that cmpxchg, test_and_set_bit, fetch_or... functions with loose
> namespaces aren't the best layout.
>
> But casting thread_info to atomic_t really worries me, I'm not sure the ending
> result would be correct at all. I prefer to sacrify correctness over namespace
> sanity:-)

Just to clear, I wasn't suggesting that that necessarily was the way for you to go.
The last four lines of my message quoted above are what I think might be the
best way forward, and don't involve messing with thread_info at all.

-- 
Chris Metcalf, EZChip Semiconductor
http://www.ezchip.com