[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <50F640B7-AB54-4571-9988-EAD5FD5933A4@zytor.com>
Date: Sun, 21 Jan 2024 18:31:53 -0800
From: "H. Peter Anvin" <hpa@...or.com>
To: Steven Rostedt <rostedt@...dmis.org>
CC: Thorsten Glaser <tg@...ian.org>, Peter Zijlstra <peterz@...radead.org>,
x86@...nel.org, torvalds@...uxfoundation.org,
linux-kernel@...r.kernel.org, linux-toolchains@...r.kernel.org,
jpoimboe@...hat.com, alexei.starovoitov@...il.com, mhiramat@...nel.org
Subject: Re: [PATCH 1/2] x86: Remove dynamic NOP selection
On January 21, 2024 6:22:36 PM PST, Steven Rostedt <rostedt@...dmis.org> wrote:
>On Sun, 21 Jan 2024 18:15:39 -0800
>"H. Peter Anvin" <hpa@...or.com> wrote:
>
>> On 1/21/24 16:56, Steven Rostedt wrote:
>> >>
>> >> Yes, but it is a matter of where we optimize for performance as opposed to correctness.
>> >
>> > There is no such thing as "optimize for correctness", it is either
>> > correct or it is not. Correctness should always come before performance
>> > (at least that is what Thomas has pounded into me ;-)
>> >
>> > If a kernel use to work on a machine but a newer version no longer
>> > works, I call that a regression.
>> >
>>
>> There absolutely is such a thing as "optimize for correctness." It means
>> to keep the code clean, easily testable, and with a minimal number of
>> distinct code paths so that regressions and *especially* uncaught
>> regressions get caught quickly.
>
>I call that maintainability, not correctness. It is either correct and
>works, or is incorrect and does not work.
>
>You can change code to be more maintainable and still make it incorrect.
>
>-- Steve
Yes, of course. That's called failure :)
Powered by blists - more mailing lists