Re: [PATCH v4 1/2] serial: sifive: lock port in startup()/shutdown() callbacks

From: Petr Mladek <pmladek@suse.com>
To: Greg KH <gregkh@linuxfoundation.org>
Cc: Vlastimil Babka <vbabka@suse.cz>,
	Ryo Takakura <ryotkkr98@gmail.com>,
	alex@ghiti.fr, aou@eecs.berkeley.edu, bigeasy@linutronix.de,
	conor.dooley@microchip.com, jirislaby@kernel.org,
	john.ogness@linutronix.de, palmer@dabbelt.com,
	paul.walmsley@sifive.com, samuel.holland@sifive.com,
	u.kleine-koenig@baylibre.com, linux-kernel@vger.kernel.org,
	linux-riscv@lists.infradead.org, linux-serial@vger.kernel.org,
	stable@vger.kernel.org
Subject: Re: [PATCH v4 1/2] serial: sifive: lock port in startup()/shutdown() callbacks
Date: Fri, 25 Apr 2025 11:02:11 +0200	[thread overview]
Message-ID: <aAtPk6vxmIBdp1NT@pathway.suse.cz> (raw)
In-Reply-To: <2025042252-geology-causation-a455@gregkh>

On Tue 2025-04-22 15:16:21, Greg KH wrote:
> On Tue, Apr 22, 2025 at 03:07:54PM +0200, Vlastimil Babka wrote:
> > On 4/22/25 12:50, Greg KH wrote:
> > > On Tue, Apr 22, 2025 at 12:20:42PM +0200, Vlastimil Babka wrote:
> > >> 
> > >> I admit it's surprising to see such a request as AFAIK it's normally done to
> > >> mix stable fixes and new features in the same series (especially when the
> > >> patches depend on each other), and ordering the fixes first and marking only
> > >> them as stable should be sufficient. We do that all the time in -mm. I
> > >> thought that stable works with stable marked commits primarily, not series?
> > > 
> > > Yes, but when picking which "branch" to apply a series to, what would
> > > you do if you have some "fix some bugs, then add some new features" in a
> > > single patch series?  The one to go to -final or the one for the next
> > > -rc1?
> > 
> > As a maintainer I could split it myself.
> 
> You must not have that many patches to review, remember, some of us get
> a few more than others ;)
> 
> > > I see a lot of bugfixes delayed until -rc1 because of this issue, and
> > > it's really not a good idea at all.
> > 
> > In my experience, most of the time these fixes are created because a dev:
> > 
> > - works on the code to implement the feature part
> > - while working at the code, spots an existing bug
> > - the bug can be old (Fixes: commit a number of releases ago)
> > - wants to be helpful so isolates the fix separately as an early patch of
> > the series and marks stable because the bug can be serious enough in theory
> > - at the same time there are no known reports of the bug being hit in the wild
> > 
> > In that case I don't see the urgency to fix it ASAP (unless it's e.g.
> > something obviously dangerously exploitable) so it might not be such a bad
> > idea just to put everything towards next rc1.
> 
> Yes, but then look at the huge number of "bugfixes" that land in -rc1.
> Is that ok or not?  I don't know...
>
> > This very thread seems to be a good example of the above. I see the later
> > version added a
> > Fixes: 45c054d0815b ("tty: serial: add driver for the SiFive UART")
> > which is a v5.2 commit.
> 
> Agreed, but delaying a known-fix for weeks/months feels bad to me.

I personally push rc1 regression fixes ASAP. But it has a cost.
I do extra careful review, testing, and still I am nervous of causing
a regression which might leak to a stable release.

IMHO, it is perfectly fine to delay fixes for bugs which
were there for months or years. For example, this patch fixes
a bug which has been in the driver since the beginning (2019).

I think that the root of the problem is in the view of the stable release
process. I am pretty conservative. My experience is that some problems
are caught only in the -rc phase when the kernel gets more testing.
I am not sure if stable -rc kernels get the same amount of testing.

Best Regards,
Petr

_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv