Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Sender: Guenter Roeck <groeck7@gmail.com>
Message-ID: <cdb13c6c-9cef-905b-0ecd-94649ca075d0@roeck-us.net>
Date:   Tue, 27 Jun 2023 16:23:38 -0700
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.11.0
Content-Language: en-US
To:     Linus Torvalds <torvalds@linux-foundation.org>,
        Dinh Nguyen <dinguyen@kernel.org>,
        "Paul E. McKenney" <paulmck@kernel.org>,
        Shuah Khan <shuah@kernel.org>
Cc:     linux-kernel@vger.kernel.org, vishal.moola@gmail.com,
        akpm@linux-foundation.org, sfr@canb.auug.org.au
References: <20230627221430.464073-1-dinguyen@kernel.org>
 <CAHk-=wgh1CDUTJD+F_utKnbW6d=wApBMC4D8hMfOs15bAn23Bg@mail.gmail.com>
From:   Guenter Roeck <linux@roeck-us.net>
Subject: Re: [PATCH] Revert "nios2: Convert __pte_free_tlb() to use ptdescs"
In-Reply-To: <CAHk-=wgh1CDUTJD+F_utKnbW6d=wApBMC4D8hMfOs15bAn23Bg@mail.gmail.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 6/27/23 15:35, Linus Torvalds wrote:
> On Tue, 27 Jun 2023 at 15:14, Dinh Nguyen <dinguyen@kernel.org> wrote:
>>
>> This reverts commit 6ebe94baa2b9ddf3ccbb7f94df6ab26234532734.
>>
>> The patch "nios2: Convert __pte_free_tlb() to use ptdescs" was supposed
>> to go together with a patchset that Vishal Moola had planned taking it
>> through the mm tree. By just having this patch, all NIOS2 builds are
>> broken.
> 
> This is now at least the third time just this merge window where some
> base tree was broken, and people thought that linux-next is some kind
> of testing ground for it all.
> 
> NO!
> 
> Linux-next is indeed for testing, and for finding situations where
> there are interactions between different trees.
> 
> But linux-next is *not* a replacement for "this tree has to work on
> its own". THAT testing needs to be done independently, and *before* a
> tree hits linux-next.
> 
> It is *NOT* ok to say "this will work in combination with that other
> tree". EVERY SINGLE TREE needs to work on its own, because otherwise
> you cannot bisect the end result sanely.
> 
> We apparently had the NIOS2 tree being broken. And the RCU tree was
> broken. And the KUnit tree was broken.
> 

Actually, this one is broken in linux-next as well because it was pulled
into it, but the context patches needed to make it work (compile) are not
there. It is also broken in next/pending-fixes for the same reason.

Only this happened so quick that by the time I noticed and reported
and argued that, no, I did not try to apply this patch on its own,
the pull request into mainline was already sent and applied.

Problem with linux-next is that it is so badly broken that it would take
a full-time position to track down all its failures. Then there are those
last-minute patches added in the week (or days) before the commit window
opens which break it again. This is one example, but there is at least
one more in linux-next (and pending-fixes); see
https://kerneltests.org/builders/next-sh-pending-fixes/builds/822/steps/buildcommand/logs/stdio

Guenter

> In all those cases, the base tree did not compile properly on its own,
> and linux-next "magically fixed" it by either having Stephen Rothwell
> literally fix the build breakage by hand, or by having some other tree
> hide the problem.
> 
> This is very much not ok.
> 
> I'm not sure why it happened so much this release, but this needs to
> stop. People need to realize that you can't just throw shit at the
> wall and see if it sticks. You need to test your own trees *first*,
> and *independently* of other peoples trees.
> 
> Then, if  you have done basic testing, you can then have it in
> linux-next and that hopefully then finds any issues with bad
> interactions with other trees, and maybe also ends up getting more
> coverage testing on odd architectures and with odd configurations.
> 
> But linux-next must not in *any* way be a replacement for doing basic
> testing on your own tree first.
> 
>                    Linus