Expectations from IOMMU functions

static struct vcodec_iommu_ops drm_ops = {
	.create = vcodec_drm_create,
	.import = vcodec_drm_import,
	.free = vcodec_drm_free,
	.free_fd = vcodec_drm_free_fd,
	.map_kernel = vcodec_drm_map_kernel,
	.unmap_kernel = vcodec_drm_unmap_kernel,
	.map_iommu = vcodec_drm_map_iommu,
	.unmap_iommu = vcodec_drm_unmap_iommu,
	.destroy = vcodec_drm_destroy,
	.dump = vcdoec_drm_dump_info,
	.attach = vcodec_drm_attach,
	.detach = vcodec_drm_detach,
	.clear = vcodec_drm_clear_session,
};

Analysis of the VCODEC IOMMU ION implementation :

.map_iommu returns 0 on success, a negative value on failure, using :

ion_map_iommu(vpu_dev:     struct device *iommu_dev,
              ion_client:  struct ion_client *client,
              ion_handle:  struct ion_handle *handle,
              iova:        unsigned long *iova (Not setup),
              size:        unsigned long *size (Not setup))
→ __ion_iommu_map(handle->buffer: struct ion_buffer *buffer,
                  iommu_dev:      struct device *iommu_dev,
                  iova:           unsigned long *iova (Not setup))
 → map_iommu(buffer:        struct ion_buffer *buffer,
              iommu_dev:     struct device *iommu_dev,
              data:          struct ion_iommu_map *data,
              buffer->size:  unsigned long iova_length,
              buffer->flags: unsigned long flags)
  → rockchip_iovmm_map(iommu_dev:   struct device *dev,
                        table->sgl:  struct scatterlist *sg,
                        0:           off_t offset,
                        iova_length: size_t size);
   ← Returns (dma_addr_t) ( gen_pool_alloc(vmm->vmm_pool, size+IOMMU_REGION_GUARD) + offset_in_page(sg_phys(sg)) )
      vmm->vmm_pool = gen_pool_create(PAGE_SHIFT, -1);
      gen_pool_add(vmm->vmm_pool, IOVA_START, IOVM_SIZE, -1);
      #define IOVA_START 0x10000000
      #define IOVM_SIZE (SZ_1G - SZ_4K) /* last 4K is for error values */
  → Setup data->iova to the value returned by rockchip_iovmm_map

.map_kernel returns NULL (0) when using the drm_heap ION implementation and the cpu address of the ION buffer allocated using the coherent allocator of the Linux DMA API; when using the cma_heap ION implementation.