2013/11/22 H. Peter Anvin <h...@zytor.com>: > On 11/21/2013 07:02 AM, Akinobu Mita wrote: >> >> I want to use DMA CMA for the feature that requires a single huge chunk >> (about 64MB) of host memory. > > That is tautological. "I want to use the Contiguous Memory Allocator to > allocate contiguous memory." > >> Please refer 3 of 24 in: >> http://www.jedec.org/sites/default/files/T_Fujisawa_MF_2013.pdf > > But yes, it makes sense that a device using host memory would not be > able to leverage swiotlb, because the software simply has no clue what > the DMA transactions are. In that sense it is fundamentally different > from a real IOTLB.
I'm currently working for integrating DMA CMA with intel-iommu which I can test. Regardless of which dma ops is actually used, I would like to allocate large contiguous memory with dma_alloc_coherent() and tell the base address to the device. We can use mem= or crashkernel= to reserve host memory, but it is complicated and unables to work with real IOMMU. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
