[PATCH] drm/radeon: fix adding all VAs to the freed list on remove v2

[Michel Dänzer]

On 26.06.2015 19:31, Christian König wrote:
> From: Christian König <christian.koenig at amd.com>
> 
> We only should do so when the BO_VA was actually mapped.
> Otherwise we get a nice error message on the next CS.
> 
> v2: It actually doesn't matter if it was invalidated or not,
>     if it was mapped we need to clear the area where it was mapped.
> 
> Signed-off-by: Christian König <christian.koenig at amd.com>
> Tested-by: Michel Dänzer <michel.daenzer at amd.com> (v1)
> ---
>  drivers/gpu/drm/radeon/radeon_vm.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/gpu/drm/radeon/radeon_vm.c b/drivers/gpu/drm/radeon/radeon_vm.c
> index 3662157..ec10533 100644
> --- a/drivers/gpu/drm/radeon/radeon_vm.c
> +++ b/drivers/gpu/drm/radeon/radeon_vm.c
> @@ -1129,12 +1129,12 @@ void radeon_vm_bo_rmv(struct radeon_device *rdev,
>  		interval_tree_remove(&bo_va->it, &vm->va);
>  
>  	spin_lock(&vm->status_lock);
> -	if (list_empty(&bo_va->vm_status)) {
> +	list_del(&bo_va->vm_status);
> +	if (bo_va->it.start || bo_va->it.last) {
>  		bo_va->bo = radeon_bo_ref(bo_va->bo);
>  		list_add(&bo_va->vm_status, &vm->freed);
>  	} else {
>  		radeon_fence_unref(&bo_va->last_pt_update);
> -		list_del(&bo_va->vm_status);
>  		kfree(bo_va);
>  	}
>  	spin_unlock(&vm->status_lock);
> 

Even with this v2 patch, I was still running into these messages
sometimes, accompanied by Mesa complaining about a CS being rejected
and visual corruption:

 radeon 0000:00:01.0: bo ffff88021b433000 don't has a mapping in vm ffff8802355c1800

I tested the v1 patch again, same problem.

Reverting commit 161ab658a611df14fb0365b7b70a8c5fed3e4870 ("drm/radeon:
stop using addr to check for BO move") instead of applying this patch
fixes it.

Unfortunately, I don't have a very reliable way to reproduce it. On
this Kaveri laptop, it seemed to happen every time doing

	xinit =xterm -- :1 -retro

which resulted in a completely corrupted display instead of the
expected root weave and xterm window. However, I haven't seen it on my
desktop Kaveri machine with either the v1 or v2 patch.


-- 
Earthling Michel Dänzer               |               http://www.amd.com
Libre software enthusiast             |             Mesa and X developer