sfc: ef100 representor RX top half

Representor RX uses a NAPI context driven by a 'fake interrupt': when
 the parent PF receives a packet destined for the representor, it adds
 it to an SKB list (efv->rx_list), and schedules NAPI if the 'fake
 interrupt' is primed.  The NAPI poll then pulls packets off this list
 and feeds them to the stack with netif_receive_skb_list().
This scheme allows us to decouple representor RX from the parent PF's
 RX fast-path.
This patch implements the 'top half', which builds an SKB, copies data
 into it from the RX buffer (which can then be released), adds it to
 the queue and fires the 'fake interrupt' if necessary.

Signed-off-by: Edward Cree <ecree.xilinx@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

1 files changed, 55 insertions, 0 deletions

diff --git a/drivers/net/ethernet/sfc/ef100_rep.c b/drivers/net/ethernet/sfc/ef100_rep.c
index fe45ae963391..e6c6e9e764b2 100644
--- a/drivers/net/ethernet/sfc/ef100_rep.c
+++ b/drivers/net/ethernet/sfc/ef100_rep.c
@@ -13,9 +13,12 @@
 #include "ef100_netdev.h"
 #include "ef100_nic.h"
 #include "mae.h"
+#include "rx_common.h"
 
 #define EFX_EF100_REP_DRIVER	"efx_ef100_rep"
 
+#define EFX_REP_DEFAULT_PSEUDO_RING_SIZE	64
+
 static int efx_ef100_rep_poll(struct napi_struct *napi, int weight);
 
 static int efx_ef100_rep_init_struct(struct efx_nic *efx, struct efx_rep *efv,
@@ -198,6 +201,7 @@ static int efx_ef100_configure_rep(struct efx_rep *efv)
 	u32 selector;
 	int rc;
 
+	efv->rx_pring_size = EFX_REP_DEFAULT_PSEUDO_RING_SIZE;
 	/* Construct mport selector for corresponding VF */
 	efx_mae_mport_vf(efx, efv->idx, &selector);
 	/* Look up actual mport ID */
@@ -320,3 +324,54 @@ static int efx_ef100_rep_poll(struct napi_struct *napi, int weight)
 		}
 	return spent;
 }
+
+void efx_ef100_rep_rx_packet(struct efx_rep *efv, struct efx_rx_buffer *rx_buf)
+{
+	u8 *eh = efx_rx_buf_va(rx_buf);
+	struct sk_buff *skb;
+	bool primed;
+
+	/* Don't allow too many queued SKBs to build up, as they consume
+	 * GFP_ATOMIC memory.  If we overrun, just start dropping.
+	 */
+	if (efv->write_index - READ_ONCE(efv->read_index) > efv->rx_pring_size) {
+		atomic64_inc(&efv->stats.rx_dropped);
+		if (net_ratelimit())
+			netif_dbg(efv->parent, rx_err, efv->net_dev,
+				  "nodesc-dropped packet of length %u\n",
+				  rx_buf->len);
+		return;
+	}
+
+	skb = netdev_alloc_skb(efv->net_dev, rx_buf->len);
+	if (!skb) {
+		atomic64_inc(&efv->stats.rx_dropped);
+		if (net_ratelimit())
+			netif_dbg(efv->parent, rx_err, efv->net_dev,
+				  "noskb-dropped packet of length %u\n",
+				  rx_buf->len);
+		return;
+	}
+	memcpy(skb->data, eh, rx_buf->len);
+	__skb_put(skb, rx_buf->len);
+
+	skb_record_rx_queue(skb, 0); /* rep is single-queue */
+
+	/* Move past the ethernet header */
+	skb->protocol = eth_type_trans(skb, efv->net_dev);
+
+	skb_checksum_none_assert(skb);
+
+	atomic64_inc(&efv->stats.rx_packets);
+	atomic64_add(rx_buf->len, &efv->stats.rx_bytes);
+
+	/* Add it to the rx list */
+	spin_lock_bh(&efv->rx_lock);
+	primed = efv->read_index == efv->write_index;
+	list_add_tail(&skb->list, &efv->rx_list);
+	efv->write_index++;
+	spin_unlock_bh(&efv->rx_lock);
+	/* Trigger rx work */
+	if (primed)
+		napi_schedule(&efv->napi);
+}