Re: [Linaro-mm-sig] [PATCH v2 2/3] gpu: ion: carveout_heap page wised cache flush

On Tue, Sep 4, 2012 at 3:40 PM, zhangfei gao zhangfei.gao@gmail.com wrote:

...

...

@@ -88,25 +88,36 @@ struct sg_table *ion_carveout_heap_map_dma(struct ion_heap *heap, struct ion_buffer *buffer) { struct sg_table *table;

•   int ret;
  

•   struct scatterlist *sg;
  

•   int ret, i;
  

•   int nents = PAGE_ALIGN(buffer->size) / PAGE_SIZE;
  

•   struct page *page = phys_to_page(buffer->priv_phys);
  
    table = kzalloc(sizeof(struct sg_table), GFP_KERNEL);
    if (!table)
            return ERR_PTR(-ENOMEM);
  

•   ret = sg_alloc_table(table, 1, GFP_KERNEL);
  

•   ret = sg_alloc_table(table, nents, GFP_KERNEL);
    if (ret) {
            kfree(table);
            return ERR_PTR(ret);
    }
  

•   sg_set_page(table->sgl, phys_to_page(buffer->priv_phys), buffer->size,
  

•               0);
  

•   sg = table->sgl;
  

•   for (i = 0; i < nents; i++) {
  

•           sg_set_page(sg, page + i, PAGE_SIZE, 0);
  

•           sg = sg_next(sg);
  

•   }
  

•   return table;
  

}

It would need page-wise sg_table only if buffer is allocated as cached.

This refers to ion_system_heap.c, personally, I think it is fine.

AFAIU the purpose of making page-wise sg_table is for dirty page tracking by ion and so is needed only if ION_FLAG_CACHED is set for the buffer. If buffer->flags does not have ION_FLAG_CACHED, sg_table with single nents should be fine. Right?