RE: [PATCH] NTB: Address out of DMA descriptor issue with NTB

["Allen Hubbe" <Allen.Hubbe@emc.com>]

From: Dave Jiang <dave.jiang@intel.com>
> The transport right now does not handle the case where we run out of DMA
> descriptors. We just fail when we do not succeed. Adding code to retry
> for
> a bit attempting to use the DMA engine instead of instantly fail to CPU
> copy.

> @@ -1263,11 +1276,20 @@ static void ntb_async_rx(struct ntb_queue_entry
> *entry, void *offset)

> +	do {
> +		txd = device->device_prep_dma_memcpy(chan, unmap->addr[1],
> +						     unmap->addr[0], len,
> +						     DMA_PREP_INTERRUPT);
> +		if (!txd) {
> +			set_current_state(TASK_INTERRUPTIBLE);
> +			schedule_timeout(DMA_OUT_RESOURCE_TO);
> +		}
> +	} while (!txd && (++retries < DMA_RETRIES));

!txd is tested twice per loop.

How about:

for (retries = 0; retries < DMA_RETRIES; ++retries) {
    txd = prep_dma_memcpy();
    if (txd)
        break;
    set_current_state();
    schedule_timeout();
}

> @@ -1494,10 +1517,19 @@ static void ntb_async_tx(struct ntb_transport_qp

Also here.

Allen