Message-ID: <763df3b2-dc7e-d68f-a4ea-b7d0d45dc111@kernel.dk>
Date:   Thu, 5 Oct 2017 16:07:10 -0600
From:   Jens Axboe <axboe@...nel.dk>
To:     Thomas Gleixner <tglx@...utronix.de>
Cc:     Kees Cook <keescook@...omium.org>,
        LKML <linux-kernel@...r.kernel.org>,
        Michal Hocko <mhocko@...e.com>,
        Andrew Morton <akpm@...ux-foundation.org>,
        Jan Kara <jack@...e.cz>, Johannes Weiner <hannes@...xchg.org>,
        Nicholas Piggin <npiggin@...il.com>,
        Vladimir Davydov <vdavydov.dev@...il.com>,
        Matthew Wilcox <mawilcox@...rosoft.com>,
        Jeff Layton <jlayton@...hat.com>, linux-block@...r.kernel.org,
        Linux-MM <linux-mm@...ck.org>
Subject: Re: [PATCH] block/laptop_mode: Convert timers to use timer_setup()

On 10/05/2017 03:53 PM, Thomas Gleixner wrote:
> Jens,
> 
> On Thu, 5 Oct 2017, Jens Axboe wrote:
>> On 10/05/2017 01:23 PM, Thomas Gleixner wrote:
>>> Come on. You know very well that a prerequisite for global changes which is
>>> not yet used in Linus tree can get merged post merge windew in order to
>>> avoid massive inter maintainer tree dependencies. We've done that before.
>>
>> My point is that doing it this late makes things harder than they should
>> have been. If this was in for -rc1, it would have made things a lot
>> easier. Or even -rc2. I try and wait to fork off the block tree for as
>> long as I can, -rc2 is generally where that happens.
> 
> Well, yes. I know it's about habits. There is no real technical reason not
> to merge -rc3 or later into your devel/next branch. I actually do that for
> various reasons, one being that I prefer to have halfways testable
> branches, which is often not the case when they are based of rc1, which is
> especially true in this 4.14 cycle. The other is to pick up stuff which
> went into Linus tree via a urgent branch or even got applied from mail
> directly.

Yes, it's not impossible, I just usually prefer not to. For this case, I
just setup a for-4.15/timer, that is the current block branch with -rc3
pulled in. I applied the two patches for floppy and amiflop, I'm
assuming Kees will respin the writeback/laptop version and I can shove
that in there too.

>> I'm not judging this based on whether I find it interesting or not, but
>> rather if it's something that's generally important to get in. Maybe it
>> is, but I don't see any justification for that at all. So just looking
>> at the isolated change, it does not strike me as something that's
>> important enough to warrant special treatment (and the pain associated
>> with that). I don't care about the core change, it's obviously trivial.
>> Expecting maintainers to pick up this dependency asap mid cycle is what
>> sucks.
> 
> I'm really not getting the 'pain' point.
> 
> 'git merge linus' is not really a pain and it does not break workflows
> assumed that you do that on a branch which has immutable state. If you want
> to keep your branches open for rebasing due to some wreckage in the middle
> of it, that's a different story.

I try never to rebase the development branches. It'll happen very
rarely, but only for cases where the screwup has been short and I'm
assuming/hoping no one pulled it yet.

>> Please stop accusing me of being hypocritical. I'm questionning the
>> timing of the change, that should be possible without someone resorting
>> to ad hominem attacks.
> 
> Well, it seemed hypocritical to me for a hopefully understandable reason. I
> didn't want to attack or offend you in any way.
> 
> I just know from repeated experience how painful it is to do full tree
> overhauls and sit on large patch queues for a long time. There is some
> point where you need to get things going and I really appreciate the work
> of people doing that. Refactoring the kernel to get rid of legacy burdens
> and in this case to address a popular attack vector is definitely useful
> for everybody and should be supported. We tried to make it easy by pushing
> this to Linus and I really did not expect that merging Linus -rc3 into a
> devel/next branch is a painful work to do.
> 
> As Kees said already, we can set that particular patch aside and push it
> along with the rest of ignored ones around 15-rc1 time so we can remove the
> old interfaces. Though we hopefully wont end up with a gazillion of ignored
> or considered too painful ones.

No worries, we'll get over it. The new branch is setup, so as soon as
the patches are funneled in, hopefully it can be ignored/forgotten until
the merge window opens.

-- 
Jens Axboe

Powered by blists - more mailing lists