sfc: ensure recovery after allocation failures

After failing to allocate a receive buffer the driver may fail to ever
request additional allocations. EF10 NICs require new receive buffers to
be pushed in batches of eight or more. The test for whether a slow fill
should be scheduled failed to take account of this. There is little
downside to *always* requesting a slow fill if we failed to allocate a
buffer, so the condition has been removed completely. The timer that
triggers the request for a refill has also been shortened.

Signed-off-by: Robert Stonehouse <rstonehouse@solarflare.com>
Signed-off-by: Bert Kenward <bkenward@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This commit is contained in:
Robert Stonehouse 2019-02-14 17:27:43 +00:00 committed by David S. Miller
parent e772261b53
commit 50f444aa50
2 changed files with 2 additions and 3 deletions

View File

@ -915,7 +915,7 @@ rollback:
void efx_schedule_slow_fill(struct efx_rx_queue *rx_queue)
{
mod_timer(&rx_queue->slow_fill, jiffies + msecs_to_jiffies(100));
mod_timer(&rx_queue->slow_fill, jiffies + msecs_to_jiffies(10));
}
static bool efx_default_channel_want_txqs(struct efx_channel *channel)

View File

@ -360,8 +360,7 @@ void efx_fast_push_rx_descriptors(struct efx_rx_queue *rx_queue, bool atomic)
rc = efx_init_rx_buffers(rx_queue, atomic);
if (unlikely(rc)) {
/* Ensure that we don't leave the rx queue empty */
if (rx_queue->added_count == rx_queue->removed_count)
efx_schedule_slow_fill(rx_queue);
efx_schedule_slow_fill(rx_queue);
goto out;
}
} while ((space -= batch_size) >= batch_size);