Re: [PATCH 1/2] drm/msm: Fix refcnt underflow in error path

[Stephan Gerhold <stephan.gerhold@linaro.org>]

On Mon, Sep 22, 2025 at 09:41:07AM -0700, Rob Clark wrote:
> On Mon, Sep 22, 2025 at 9:33 AM Stephan Gerhold
> <stephan.gerhold@linaro.org> wrote:
> > On Wed, Jul 23, 2025 at 12:08:49PM -0700, Rob Clark wrote:
> > > If we hit an error path in GEM obj creation before msm_gem_new_handle()
> > > updates obj->resv to point to the gpuvm resv object, then obj->resv
> > > still points to &obj->_resv.  In this case we don't want to decrement
> > > the refcount of the object being freed (since the refcnt is already
> > > zero).  This fixes the following splat:
> > >
> > >    ------------[ cut here ]------------
> > >    refcount_t: underflow; use-after-free.
> > >    WARNING: CPU: 9 PID: 7013 at lib/refcount.c:28 refcount_warn_saturate+0xf4/0x148
> > >    Modules linked in: uinput snd_seq_dummy snd_hrtimer aes_ce_ccm snd_soc_wsa884x regmap_sdw q6prm_clocks q6apm_lpass_da>
> > >     qcom_pil_info i2c_hid drm_kms_helper qcom_common qcom_q6v5 phy_snps_eusb2 qcom_geni_serial drm qcom_sysmon pinctrl_s>
> > >    CPU: 9 UID: 1000 PID: 7013 Comm: deqp-vk Not tainted 6.16.0-rc4-debug+ #25 PREEMPT(voluntary)
> > >    Hardware name: LENOVO 83ED/LNVNB161216, BIOS NHCN53WW 08/02/2024
> > >    pstate: 61400005 (nZCv daif +PAN -UAO -TCO +DIT -SSBS BTYPE=--)
> > >    pc : refcount_warn_saturate+0xf4/0x148
> > >    lr : refcount_warn_saturate+0xf4/0x148
> > >    sp : ffff8000a2073920
> > >    x29: ffff8000a2073920 x28: 0000000000000010 x27: 0000000000000010
> > >    x26: 0000000000000042 x25: ffff000810e09800 x24: 0000000000000010
> > >    x23: ffff8000a2073b94 x22: ffff000ddb22de00 x21: ffff000ddb22dc00
> > >    x20: ffff000ddb22ddf8 x19: ffff0008024934e0 x18: 000000000000000a
> > >    x17: 0000000000000000 x16: ffff9f8c67d77340 x15: 0000000000000000
> > >    x14: 00000000ffffffff x13: 2e656572662d7265 x12: 7466612d65737520
> > >    x11: 3b776f6c66726564 x10: 00000000ffff7fff x9 : ffff9f8c67506c70
> > >    x8 : ffff9f8c69fa26f0 x7 : 00000000000bffe8 x6 : c0000000ffff7fff
> > >    x5 : ffff000f53e14548 x4 : ffff6082ea2b2000 x3 : ffff0008b86ab080
> > >    x2 : 0000000000000000 x1 : 0000000000000000 x0 : ffff0008b86ab080
> > >    Call trace:
> > >     refcount_warn_saturate+0xf4/0x148 (P)
> > >     msm_gem_free_object+0x248/0x260 [msm]
> > >     drm_gem_object_free+0x24/0x40 [drm]
> > >     msm_gem_new+0x1c4/0x1e0 [msm]
> > >     msm_gem_new_handle+0x3c/0x1a0 [msm]
> > >     msm_ioctl_gem_new+0x38/0x70 [msm]
> > >     drm_ioctl_kernel+0xc8/0x138 [drm]
> > >     drm_ioctl+0x2c8/0x618 [drm]
> > >     __arm64_sys_ioctl+0xac/0x108
> > >     invoke_syscall.constprop.0+0x64/0xe8
> > >     el0_svc_common.constprop.0+0x40/0xe8
> > >     do_el0_svc+0x24/0x38
> > >     el0_svc+0x54/0x1d8
> > >     el0t_64_sync_handler+0x10c/0x138
> > >     el0t_64_sync+0x19c/0x1a0
> > >    irq event stamp: 3698694
> > >    hardirqs last  enabled at (3698693): [<ffff9f8c675021dc>] __up_console_sem+0x74/0x90
> > >    hardirqs last disabled at (3698694): [<ffff9f8c68ce8164>] el1_dbg+0x24/0x90
> > >    softirqs last  enabled at (3697578): [<ffff9f8c6744ec5c>] handle_softirqs+0x454/0x4b0
> > >    softirqs last disabled at (3697567): [<ffff9f8c67360244>] __do_softirq+0x1c/0x28
> > >    ---[ end trace 0000000000000000 ]---
> > >
> > > Fixes: b58e12a66e47 ("drm/msm: Add _NO_SHARE flag")
> > > Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
> > > ---
> > >  drivers/gpu/drm/msm/msm_gem.c | 4 +++-
> > >  1 file changed, 3 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/drivers/gpu/drm/msm/msm_gem.c b/drivers/gpu/drm/msm/msm_gem.c
> > > index 33d3354c6102..958bac4e2768 100644
> > > --- a/drivers/gpu/drm/msm/msm_gem.c
> > > +++ b/drivers/gpu/drm/msm/msm_gem.c
> > > @@ -1114,10 +1114,12 @@ static void msm_gem_free_object(struct drm_gem_object *obj)
> > >               put_pages(obj);
> > >       }
> > >
> > > -     if (msm_obj->flags & MSM_BO_NO_SHARE) {
> > > +     if (obj->resv != &obj->_resv) {
> > >               struct drm_gem_object *r_obj =
> > >                       container_of(obj->resv, struct drm_gem_object, _resv);
> > >
> > > +             WARN_ON(!(msm_obj->flags & MSM_BO_NO_SHARE));
> > > +
> > >               /* Drop reference we hold to shared resv obj: */
> > >               drm_gem_object_put(r_obj);
> > >       }
> >
> > This patch seems to break something for direct IRIS/video playback using
> > dmabuf. I use a simple GStreamer test pipeline for testing IRIS on X1E
> > (on GNOME, in case that matters):
> >
> >  $ gst-launch-1.0 filesrc location=bbb_sunflower_2160p_60fps_normal.mp4 \
> >    ! qtdemux name=d d.video_0 ! h264parse ! v4l2h264dec \
> >    ! capture-io-mode=dmabuf ! waylandsink
> >
> > The video plays fine, but if I try to exit (CTRL+C) the display hangs
> > for a few seconds and then the console is spammed with pretty much
> > exactly the messages that you tried to fix here. If I revert this patch,
> > everything is fine again. It feels like your patch does exactly the
> > opposite for this use case. :-)
> >
> > It seems to run into the WARN_ON you added.
> 
> Hmm, are we allocating from drm and importing into v4l2, or the other direction?
> 

Is there an easy way to check?

I would need to study the code to be sure, you probably know more about
this than I do. I just run this command and it always worked so far
somehow. :-)

Thanks,
Stephan