[Intel-gfx] [PATCH 1/2] drm/i915: Don't deballoon unused ggtt drm_mm_node in linux guest

[Chris Wilson]

Quoting Zhenyu Wang (2019-08-21 04:35:56)
> On 2019.08.20 13:46:17 +0800, Xiong Zhang wrote:
> > The following call trace may exist in linux guest dmesg when guest i915
> > driver is unloaded.
> > [   90.776610] [drm:vgt_deballoon_space.isra.0 [i915]] deballoon space: range [0x0 - 0x0] 0 KiB.
> > [   90.776621] BUG: unable to handle kernel NULL pointer dereference at 00000000000000c0
> > [   90.776691] IP: drm_mm_remove_node+0x4d/0x320 [drm]
> > [   90.776718] PGD 800000012c7d0067 P4D 800000012c7d0067 PUD 138e4c067 PMD 0
> > [   90.777091] task: ffff9adab60f2f00 task.stack: ffffaf39c0fe0000
> > [   90.777142] RIP: 0010:drm_mm_remove_node+0x4d/0x320 [drm]
> > [   90.777573] Call Trace:
> > [   90.777653]  intel_vgt_deballoon+0x4c/0x60 [i915]
> > [   90.777729]  i915_ggtt_cleanup_hw+0x121/0x190 [i915]
> > [   90.777792]  i915_driver_unload+0x145/0x180 [i915]
> > [   90.777856]  i915_pci_remove+0x15/0x20 [i915]
> > [   90.777890]  pci_device_remove+0x3b/0xc0
> > [   90.777916]  device_release_driver_internal+0x157/0x220
> > [   90.777945]  driver_detach+0x39/0x70
> > [   90.777967]  bus_remove_driver+0x51/0xd0
> > [   90.777990]  pci_unregister_driver+0x23/0x90
> > [   90.778019]  SyS_delete_module+0x1da/0x240
> > [   90.778045]  entry_SYSCALL_64_fastpath+0x24/0x87
> > [   90.778072] RIP: 0033:0x7f34312af067
> > [   90.778092] RSP: 002b:00007ffdea3da0d8 EFLAGS: 00000206
> > [   90.778297] RIP: drm_mm_remove_node+0x4d/0x320 [drm] RSP: ffffaf39c0fe3dc0
> > [   90.778344] ---[ end trace f4b1bc8305fc59dd ]---
> > 
> > Four drm_mm_node are used to reserve guest ggtt space, but some of them
> > may be skipped and not initialised due to space constraints in
> > intel_vgt_balloon(). If drm_mm_remove_node() is called with
> > uninitialized drm_mm_node, the above call trace occurs.
> > 
> > This patch check drm_mm_node's validity before calling
> > drm_mm_remove_node().
> > 
> > Fixes: ff8f797557c7("drm/i915: return the correct usable aperture size under gvt environment")
> > Cc: stable at vger.kernel.org
> > Signed-off-by: Xiong Zhang <xiong.y.zhang at intel.com>
> > ---
> >  drivers/gpu/drm/i915/i915_vgpu.c | 3 +++
> >  1 file changed, 3 insertions(+)
> > 
> > diff --git a/drivers/gpu/drm/i915/i915_vgpu.c b/drivers/gpu/drm/i915/i915_vgpu.c
> > index bf2b837..d2fd66f 100644
> > --- a/drivers/gpu/drm/i915/i915_vgpu.c
> > +++ b/drivers/gpu/drm/i915/i915_vgpu.c
> > @@ -119,6 +119,9 @@ static struct _balloon_info_ bl_info;
> >  static void vgt_deballoon_space(struct i915_ggtt *ggtt,
> >                               struct drm_mm_node *node)
> >  {
> > +     if (!node->allocated)
> > +             return;
> > +
> >       DRM_DEBUG_DRIVER("deballoon space: range [0x%llx - 0x%llx] %llu KiB.\n",
> >                        node->start,
> >                        node->start + node->size,
> 
> Searching shows this is pretty old one and also with r-b from Chris,
> but be ignored that nobody picked this up..
> 
> I think I hit this once too and tried to fix it another way,
> but this looks simpler to me.
> 
> Acked-by: Zhenyu Wang <zhenyuw at linux.intel.com>

Better late than never, I guess. Thanks for the patch and checking it
over, pushed.
-Chris