Still keep this function because directly using dev->dma_io_tlb_mem
will cause issues for memory allocation for existing devices. The pool
can't support atomic coherent allocation so we need to distinguish the
per device pool and the default pool in swiotlb_alloc.

Reply via email to