Re: qemu-arm: zram: mkfs.ext4 : Unable to handle kernel NULL pointer dereference at virtual address 00000140

From: Minchan Kim <minchan@kernel.org>
To: Sergey Senozhatsky <senozhatsky@chromium.org>
Cc: Naresh Kamboju <naresh.kamboju@linaro.org>,
	open list <linux-kernel@vger.kernel.org>,
	linux-fsdevel@vger.kernel.org,
	Linux ARM <linux-arm-kernel@lists.infradead.org>,
	"open list:KERNEL SELFTEST FRAMEWORK"
	<linux-kselftest@vger.kernel.org>,
	linux-block <linux-block@vger.kernel.org>,
	regressions@lists.linux.dev, Jens Axboe <axboe@kernel.dk>,
	Nitin Gupta <ngupta@vflare.org>
Subject: Re: qemu-arm: zram: mkfs.ext4 : Unable to handle kernel NULL pointer dereference at virtual address 00000140
Date: Wed, 8 Jun 2022 13:45:30 -0700	[thread overview]
Message-ID: <YqEKapKLBgKEXGBg@google.com> (raw)
In-Reply-To: <YqAMmTiwcyS3Ttla@google.com>

On Wed, Jun 08, 2022 at 11:42:33AM +0900, Sergey Senozhatsky wrote:
> On (22/06/08 11:39), Sergey Senozhatsky wrote:
> > On (22/06/07 16:52), Minchan Kim wrote:
> > > > rootfs: https://oebuilds.tuxbuild.com/29zhlbEc3EWq2wod9Uy964Bp27q/images/am57xx-evm/rpb-console-image-lkft-am57xx-evm-20220601222434.rootfs.ext4.gz
> > > > kernel: https://builds.tuxbuild.com/29zhqJJizU2Y7Ka7ArhryUOrNDC/zImage
> > > > 
> > > > Boot command,
> > > >  /usr/bin/qemu-system-aarch64 -cpu host,aarch64=off -machine
> > > > virt-2.10,accel=kvm -nographic -net
> > > > nic,model=virtio,maaacaddr=BA:DD:AD:CC:09:04 -net tap -m 2048 -monitor
> > > > none -kernel kernel/zImage --append "console=ttyAMA0 root=/dev/vda rw"
> > > > -hda rootfs/rpb-console-image-lkft-am57xx-evm-20220601222434.rootfs.ext4
> > > > -m 4096 -smp 2
> > > > 
> > > > # cd /opt/kselftests/default-in-kernel/zram
> > > > # ./zram.sh
> > > > 
> > > > Allow me sometime I will try to bisect this problem.
> > > 
> > > Thanks for sharing the info. 
> > > 
> > > I managed to work your rootfs with my local arm build
> > > based on the problematic git tip. 
> > > However, I couldn't suceed to reproduce it.
> > > 
> > > I needed to build zsmalloc/zram built-in instead of modules
> > > Is it related? Hmm,
> > > 
> > > Yeah, It would be very helpful if you could help to bisect it.
> > 
> > This looks like a NULL lock->name dereference in lockdep. I suspect
> > that somehow local_lock doesn't get .dep_map initialized. Maybe running
> > the kernel with CONFIG_DEBUG_LOCK_ALLOC would help us? Naresh, can you
> > help us with this?
> 
> Hmm, actually, hold on. mapping_area is per-CPU, so what happens if CPU
> get offlined and onlined again? I don't see us re-initializing mapping_area
> local_lock_init(&zs_map_area.lock) and so on.

I am trying to understand the problem. AFAIK, the mapping_area was
static allocation per cpu so in zs_cpu_down, we never free the
mapping_area itself. Then, why do we need to reinitialize the local
lock again?