[Intel-gfx] [BrownBag] drm/i915/gtt: Neuter the deferred unbind callback from gen6_ppgtt_cleanup

[Tvrtko Ursulin]

On 24/05/2019 09:17, Chris Wilson wrote:
> Quoting Tvrtko Ursulin (2019-05-24 09:13:14)
>>
>> On 24/05/2019 07:45, Chris Wilson wrote:
>>> Having deferred the vma destruction to a worker where we can acquire the
>>> struct_mutex, we have to avoid chasing back into the now destroyed
>>> ppgtt. The pd_vma is special in having a custom unbind function to scan
>>> for unused pages despite the VMA itself being notionally part of the
>>> GGTT. As such, we need to disable that callback to avoid a
>>> use-after-free.
>>>
>>> This unfortunately blew up so early during boot that CI declared the
>>> machine unreachable as opposed to being the major failure it was. Oops.
>>>
>>> Fixes: d3622099c76f ("drm/i915/gtt: Always acquire struct_mutex for gen6_ppgtt_cleanup")
>>> Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
>>> Cc: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
>>> Cc: Tomi Sarvela <tomi.p.sarvela at intel.com>
>>> ---
>>>    drivers/gpu/drm/i915/i915_gem_gtt.c | 28 ++++++++++++++++++++++++++++
>>>    1 file changed, 28 insertions(+)
>>>
>>> diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
>>> index 8d8a4b0ad4d9..266baa11df64 100644
>>> --- a/drivers/gpu/drm/i915/i915_gem_gtt.c
>>> +++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
>>> @@ -1847,6 +1847,33 @@ static void gen6_ppgtt_cleanup_work(struct work_struct *wrk)
>>>        kfree(work);
>>>    }
>>>    
>>> +static int nop_set_pages(struct i915_vma *vma)
>>> +{
>>> +     return -ENODEV;
>>> +}
>>> +
>>> +static void nop_clear_pages(struct i915_vma *vma)
>>> +{
>>> +}
>>> +
>>> +static int nop_bind(struct i915_vma *vma,
>>> +                 enum i915_cache_level cache_level,
>>> +                 u32 unused)
>>> +{
>>> +     return -ENODEV;
>>> +}
>>> +
>>> +static void nop_unbind(struct i915_vma *vma)
>>> +{
>>> +}
>>> +
>>> +static const struct i915_vma_ops nop_vma_ops = {
>>> +     .set_pages = nop_set_pages,
>>> +     .clear_pages = nop_clear_pages,
>>> +     .bind_vma = nop_bind,
>>> +     .unbind_vma = nop_unbind,
>>> +};
>>> +
>>>    static void gen6_ppgtt_cleanup(struct i915_address_space *vm)
>>>    {
>>>        struct gen6_hw_ppgtt *ppgtt = to_gen6_ppgtt(i915_vm_to_ppgtt(vm));
>>> @@ -1855,6 +1882,7 @@ static void gen6_ppgtt_cleanup(struct i915_address_space *vm)
>>>        /* FIXME remove the struct_mutex to bring the locking under control */
>>>        INIT_WORK(&work->base, gen6_ppgtt_cleanup_work);
>>>        work->vma = ppgtt->vma;
>>> +     work->vma->ops = &nop_vma_ops;
>>
>> Could we use some asserts before overriding the vma ops? Like
>> GEM_BUG_ON(vma->pages)? And something for still bound?
> 
> It technically still is bound as it is in the GGTT but currently
> unpinned -- that will be checked on destroy, it's just we also get an
> unbind callback. vma->pages doesn't exist for this (set to ERR_PTR).

If we are getting the unbind callback and we nop-ed it, who will 
actually do it's job?

Regards,

Tvrtko