Is A Reserved Mem Func Well as Dma_create When we use sra Test Nwrite for T serial Soc.

[liguofa]

Hello Dears，

We uses T2080 for test rapidio.

Then about RIWAR register .

Cache coherency is maintained when "snoop local processor" option for WTT is selected in the corresponding RIWAR register.

what is the difference of 

Enhanced read, snoop local processor

 and 

Read,snoop local processor

We found sra demo used default is Read ,can tell me the diff of 0x5 and 0xd

Thanks,

By the way if my app not used dma alloc memory from usdpaa_mem but use reserved memory。

Our system does not use dma_create func, but the reserved address. There is no problem with function sending and receiving  nwrite now,but the problem is the speed is too low ,when we copy used memcpy from reserved memory to user space memory. we used mmap to translate phy to virt.

[yipingwang]

Hello Li guofa ,

Please refer to "Init DMA pool" section in usdpaa-apps source code src/sra/sra.c.

static int dma_usmem_init(struct dma_pool *pool)
{
        int err;

        dma_mem_generic = dma_mem_create(DMA_MAP_FLAG_ALLOC,
                                        NULL, SRIO_POOL_SIZE);
        if (!dma_mem_generic) {
                err = -EINVAL;
                error(0, -err, "%s(): dma_mem_create()", __func__);
                return err;
        }

        pool->dma_virt_base = __dma_mem_memalign(64, SRIO_POOL_PORT_OFFSET);
        if (!pool->dma_virt_base) {
                err = -EINVAL;
                error(0, -err, "%s(): __dma_mem_memalign()", __func__);
                return err;
        }
        pool->dma_phys_base = __dma_mem_vtop(pool->dma_virt_base);

        return 0;
}

Thanks,

Yiping