[bug report] drm/msm: Take lru lock once per submit_pin_objects()

[bug report] drm/msm: Take lru lock once per submit_pin_objects()

[Dan Carpenter]

Hello Rob Clark,

Commit fc896cf3d691 ("drm/msm: Take lru lock once per
submit_pin_objects()") from Aug 2, 2023 (linux-next), leads to the
following Smatch static checker warning:

	drivers/gpu/drm/msm/msm_gem.c:565 get_and_pin_iova_range_locked()
	warn: sleeping in atomic context

drivers/gpu/drm/msm/msm_gem.c
    546 static int get_and_pin_iova_range_locked(struct drm_gem_object *obj,
    547                                          struct drm_gpuvm *vm, uint64_t *iova,
    548                                          u64 range_start, u64 range_end)
    549 {
    550         struct drm_gpuva *vma;
    551         int ret;
    552 
    553         msm_gem_assert_locked(obj);
    554 
    555         if (to_msm_bo(obj)->flags & MSM_BO_NO_SHARE)
    556                 return -EINVAL;
    557 
    558         vma = get_vma_locked(obj, vm, range_start, range_end);
    559         if (IS_ERR(vma))
    560                 return PTR_ERR(vma);
    561 
    562         ret = msm_gem_pin_vma_locked(obj, vma);
    563         if (!ret) {
    564                 *iova = vma->va.addr;
--> 565                 pin_obj_locked(obj);
                        ^^^^^^^^^^^^^^^^^^^^
This takes a mutex but Smatch thinks we're already holding a spinlock.

    566         }
    567 
    568         return ret;
    569 }

The call tree is:

update_cursor() <- disables preempt
-> msm_gem_get_and_pin_iova()
   -> msm_gem_get_and_pin_iova_range()
      -> get_and_pin_iova_range_locked()

The update_cursor() function is holding
	spin_lock_irqsave(&mdp4_crtc->cursor.lock, flags);

regards,
dan carpenter