Received-SPF: pass (google.com: domain of linux-kernel+bounces-21448-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.199.223 as permitted sender) client-ip=147.75.199.223;
Sender: Tejun Heo <htejun@gmail.com>
Date: Tue, 9 Jan 2024 11:51:26 -1000
From: Tejun Heo <tj@kernel.org>
To: Andrea Righi <andrea.righi@canonical.com>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>,
	Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
	Alexei Starovoitov <ast@kernel.org>, linux-kernel@vger.kernel.org,
	Linux-Renesas <linux-renesas-soc@vger.kernel.org>
Subject: Re: [PATCH] kernfs: convert kernfs_idr_lock to an irq safe raw
 spinlock
Message-ID: <ZZ2_3n9rJnhcYmpk@slm.duckdns.org>
References: <20231229074916.53547-1-andrea.righi@canonical.com>
 <CAMuHMdV=AKt+mwY7svEq5gFPx41LoSQZ_USME5_MEdWQze13ww@mail.gmail.com>
 <ZZ18xVq4GtQsTC8Z@gpd>
 <ZZ2ZdoNyBhR7o83I@slm.duckdns.org>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <ZZ2ZdoNyBhR7o83I@slm.duckdns.org>

On Tue, Jan 09, 2024 at 09:07:34AM -1000, Tejun Heo wrote:
> On Tue, Jan 09, 2024 at 06:05:09PM +0100, Andrea Righi wrote:
> > On Tue, Jan 09, 2024 at 05:35:36PM +0100, Geert Uytterhoeven wrote:
> > > Reverting commit c312828c37a72fe2 fixes that.
> > > I have seen this issue on several Renesas arm32 and arm64 platforms.
> > > 
> > > Also, I am wondering if the issue fixed by commit c312828c37a72fe2
> > > can still be reproduced on v6.7-rc5 or later?
> > 
> > Yep, I can still reproduce it (this is with v6.7):
> ...
> > I'm wondering if using a regular spinlock instead of a raw spinlock
> > could be a reasonable compromise.
> 
> I don't think that'd work on RT as we can end up nesting mutex inside a raw
> spinlock.
> 
> > We have a GFP_ATOMIC allocation in __kernfs_new_node():
> > 
> > 	raw_spin_lock_irqsave(&kernfs_idr_lock, irqflags);
> > 	ret = idr_alloc_cyclic(&root->ino_idr, kn, 1, 0, GFP_ATOMIC);
> > 	...
> >         raw_spin_unlock_irqrestore(&kernfs_idr_lock, irqflags);
> > 
> > That should become valid using a
> > spin_lock_irqsave/spin_unlock_irqrestore(), right?
> 
> Yeah, this part should be fine. I think the right thing to do here is making
> the idr RCU safe so that lookup path doesn't depend on the lock.
> 
> Greg, can you please revert c312828c37a72fe2 for now?

I sent out a patchset to revert the commit and implement a different fix.

  http://lkml.kernel.org/r/20240109214828.252092-1-tj@kernel.org

Thanks.

-- 
tejun