[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Wed, 11 Nov 2020 10:34:05 -0700
From: Shuah Khan <skhan@...uxfoundation.org>
To: Peter Zijlstra <peterz@...radead.org>
Cc: corbet@....net, keescook@...omium.org, gregkh@...uxfoundation.org,
linux-doc@...r.kernel.org, linux-kernel@...r.kernel.org,
skhan@...uxfoundation.org
Subject: Re: [PATCH 01/13] seqnum_ops: Introduce Sequence Number Ops
On 11/11/20 9:04 AM, Peter Zijlstra wrote:
> On Wed, Nov 11, 2020 at 08:56:49AM -0700, Shuah Khan wrote:
>
>> Why would you say no to read and inc?
>
> Because they don't guarantee uniqueness (bar wrapping), which is the
> only reason to use an atomic to begin with.
>
Thanks for the explanation. I see what you are saying.
Not sure what to make of the 6080 atomic_read()s and 3413
atomic_inc()s, some of which might be assuming uniqueness
guarantee.
As far as the sequence number api is concerned, I am with you on
not exposing read() and inc().
inc()s can just map to inc_return().
For read():
In the context of up counters, there is a definitely a need for get
current value type interface that guarantees uniqueness - similar to
inc_return without actually incrementing.
I will work on v2 based on the discussion.
thanks,
-- Shuah
Powered by blists - more mailing lists