[resend PATCH] drm/ttm: Fix a deadlock if the target BO is not idle during swap

[Daniel Vetter]

On Tue, Sep 07, 2021 at 11:28:23AM +0200, Christian König wrote:
> Am 07.09.21 um 11:05 schrieb Daniel Vetter:
> > On Tue, Sep 07, 2021 at 08:22:20AM +0200, Christian König wrote:
> > > Added a Fixes tag and pushed this to drm-misc-fixes.
> > We're in the merge window, this should have been drm-misc-next-fixes. I'll
> > poke misc maintainers so it's not lost.
> 
> Hui? It's a fix for a problem in stable and not in drm-misc-next.

Ah the flow chart is confusing. There is no current -rc, so it's always
-next-fixes. Or you're running the risk that it's lost until after -rc1.
Maybe we should clarify that "is the bug in current -rc?" only applies if
there is a current -rc.

Anyway Thomas sent out a pr, so it's all good.
-Daniel

> 
> Christian.
> 
> > -Daniel
> > 
> > > It will take a while until it cycles back into the development branches, so
> > > feel free to push some version to amd-staging-drm-next as well. Just ping
> > > Alex when you do this.
> > > 
> > > Thanks,
> > > Christian.
> > > 
> > > Am 07.09.21 um 06:08 schrieb xinhui pan:
> > > > The ret value might be -EBUSY, caller will think lru lock is still
> > > > locked but actually NOT. So return -ENOSPC instead. Otherwise we hit
> > > > list corruption.
> > > > 
> > > > ttm_bo_cleanup_refs might fail too if BO is not idle. If we return 0,
> > > > caller(ttm_tt_populate -> ttm_global_swapout ->ttm_device_swapout) will
> > > > be stuck as we actually did not free any BO memory. This usually happens
> > > > when the fence is not signaled for a long time.
> > > > 
> > > > Signed-off-by: xinhui pan <xinhui.pan at amd.com>
> > > > Reviewed-by: Christian König <christian.koenig at amd.com>
> > > > ---
> > > >    drivers/gpu/drm/ttm/ttm_bo.c | 6 +++---
> > > >    1 file changed, 3 insertions(+), 3 deletions(-)
> > > > 
> > > > diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
> > > > index 8d7fd65ccced..23f906941ac9 100644
> > > > --- a/drivers/gpu/drm/ttm/ttm_bo.c
> > > > +++ b/drivers/gpu/drm/ttm/ttm_bo.c
> > > > @@ -1152,9 +1152,9 @@ int ttm_bo_swapout(struct ttm_buffer_object *bo, struct ttm_operation_ctx *ctx,
> > > >    	}
> > > >    	if (bo->deleted) {
> > > > -		ttm_bo_cleanup_refs(bo, false, false, locked);
> > > > +		ret = ttm_bo_cleanup_refs(bo, false, false, locked);
> > > >    		ttm_bo_put(bo);
> > > > -		return 0;
> > > > +		return ret == -EBUSY ? -ENOSPC : ret;
> > > >    	}
> > > >    	ttm_bo_del_from_lru(bo);
> > > > @@ -1208,7 +1208,7 @@ int ttm_bo_swapout(struct ttm_buffer_object *bo, struct ttm_operation_ctx *ctx,
> > > >    	if (locked)
> > > >    		dma_resv_unlock(bo->base.resv);
> > > >    	ttm_bo_put(bo);
> > > > -	return ret;
> > > > +	return ret == -EBUSY ? -ENOSPC : ret;
> > > >    }
> > > >    void ttm_bo_tt_destroy(struct ttm_buffer_object *bo)
> 

-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch