On 12/17/2015 06:01 AM, Chris Wilson wrote: > On Wed, Dec 16, 2015 at 08:25:36PM -0200, Tiago Vignatti wrote: >> This function is meant to be used with dma-buf mmap, when finishing the CPU >> access of the mapped pointer. >> >> +static void i915_gem_end_cpu_access(struct dma_buf *dma_buf, enum >> dma_data_direction direction) >> +{ >> + struct drm_i915_gem_object *obj = dma_buf_to_obj(dma_buf); >> + struct drm_device *dev = obj->base.dev; >> + struct drm_i915_private *dev_priv = to_i915(dev); >> + bool was_interruptible, write = (direction == DMA_BIDIRECTIONAL || >> direction == DMA_TO_DEVICE); >> + int ret; >> + >> + mutex_lock(&dev->struct_mutex); >> + was_interruptible = dev_priv->mm.interruptible; >> + dev_priv->mm.interruptible = false; >> + >> + ret = i915_gem_object_set_to_gtt_domain(obj, write); > > This only needs to pass .write=false. The dma-buf direction is > only for the period of the user access, and we are now flushing the > caches. This is equivalent to the sw-finish ioctl and ideally we just > want the i915_gem_object_flush_cpu_write_domain().
in fact the only usage so far I found for end_cpu_access is when the pinned buffer is scanout out. Should I pretty much copy sw-finish in end_cpu_access then? Thanks, Tiago