[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-Id: <20221208054045.3600-7-gerhard@engleder-embedded.com>
Date: Thu, 8 Dec 2022 06:40:45 +0100
From: Gerhard Engleder <gerhard@...leder-embedded.com>
To: netdev@...r.kernel.org, bpf@...r.kernel.org
Cc: davem@...emloft.net, kuba@...nel.org, edumazet@...gle.com,
pabeni@...hat.com, ast@...nel.org, daniel@...earbox.net,
hawk@...nel.org, john.fastabend@...il.com,
Gerhard Engleder <gerhard@...leder-embedded.com>
Subject: [PATCH net-next v2 6/6] tsnep: Add XDP RX support
If BPF program is set up, then run BPF program for every received frame
and execute the selected action.
Test results with A53 1.2GHz:
XDP_DROP (samples/bpf/xdp1)
proto 17: 883878 pkt/s
XDP_TX (samples/bpf/xdp2)
proto 17: 255693 pkt/s
XDP_REDIRECT (samples/bpf/xdpsock)
sock0@...2:0 rxdrop xdp-drv
pps pkts 1.00
rx 855,582 5,404,523
tx 0 0
XDP_REDIRECT (samples/bpf/xdp_redirect)
eth2->eth1 613,267 rx/s 0 err,drop/s 613,272 xmit/s
Signed-off-by: Gerhard Engleder <gerhard@...leder-embedded.com>
---
drivers/net/ethernet/engleder/tsnep_main.c | 126 +++++++++++++++++++++
1 file changed, 126 insertions(+)
diff --git a/drivers/net/ethernet/engleder/tsnep_main.c b/drivers/net/ethernet/engleder/tsnep_main.c
index 2b662a98b62a..d59cb714c8cd 100644
--- a/drivers/net/ethernet/engleder/tsnep_main.c
+++ b/drivers/net/ethernet/engleder/tsnep_main.c
@@ -27,6 +27,7 @@
#include <linux/phy.h>
#include <linux/iopoll.h>
#include <linux/bpf.h>
+#include <linux/bpf_trace.h>
#define TSNEP_SKB_PAD (NET_SKB_PAD + NET_IP_ALIGN)
#define TSNEP_HEADROOM ALIGN(max(TSNEP_SKB_PAD, XDP_PACKET_HEADROOM), 4)
@@ -44,6 +45,9 @@
#define TSNEP_COALESCE_USECS_MAX ((ECM_INT_DELAY_MASK >> ECM_INT_DELAY_SHIFT) * \
ECM_INT_DELAY_BASE_US + ECM_INT_DELAY_BASE_US - 1)
+#define TSNEP_XDP_TX BIT(0)
+#define TSNEP_XDP_REDIRECT BIT(1)
+
enum {
__TSNEP_DOWN,
};
@@ -626,6 +630,33 @@ static void tsnep_xdp_xmit_flush(struct tsnep_tx *tx)
iowrite32(TSNEP_CONTROL_TX_ENABLE, tx->addr + TSNEP_CONTROL);
}
+static int tsnep_xdp_xmit_back(struct tsnep_adapter *adapter,
+ struct xdp_buff *xdp)
+{
+ struct xdp_frame *xdpf = xdp_convert_buff_to_frame(xdp);
+ int cpu = smp_processor_id();
+ int queue;
+ struct netdev_queue *nq;
+ int retval;
+
+ if (unlikely(!xdpf))
+ return -EFAULT;
+
+ queue = cpu % adapter->num_tx_queues;
+ nq = netdev_get_tx_queue(adapter->netdev, queue);
+
+ __netif_tx_lock(nq, cpu);
+
+ /* Avoid transmit queue timeout since we share it with the slow path */
+ txq_trans_cond_update(nq);
+
+ retval = tsnep_xdp_xmit_frame_ring(xdpf, &adapter->tx[queue], false);
+
+ __netif_tx_unlock(nq);
+
+ return retval;
+}
+
static bool tsnep_tx_poll(struct tsnep_tx *tx, int napi_budget)
{
unsigned long flags;
@@ -792,6 +823,11 @@ static unsigned int tsnep_rx_offset(struct tsnep_rx *rx)
return TSNEP_SKB_PAD;
}
+static unsigned int tsnep_rx_offset_xdp(void)
+{
+ return XDP_PACKET_HEADROOM;
+}
+
static void tsnep_rx_ring_cleanup(struct tsnep_rx *rx)
{
struct device *dmadev = rx->adapter->dmadev;
@@ -997,6 +1033,67 @@ static int tsnep_rx_refill(struct tsnep_rx *rx, int count, bool reuse)
return i;
}
+static bool tsnep_xdp_run_prog(struct tsnep_rx *rx, struct bpf_prog *prog,
+ struct xdp_buff *xdp, int *status)
+{
+ unsigned int length;
+ unsigned int sync;
+ u32 act;
+
+ length = xdp->data_end - xdp->data_hard_start - tsnep_rx_offset_xdp();
+
+ act = bpf_prog_run_xdp(prog, xdp);
+
+ /* Due xdp_adjust_tail: DMA sync for_device cover max len CPU touch */
+ sync = xdp->data_end - xdp->data_hard_start - tsnep_rx_offset_xdp();
+ sync = max(sync, length);
+
+ switch (act) {
+ case XDP_PASS:
+ return false;
+ case XDP_TX:
+ if (tsnep_xdp_xmit_back(rx->adapter, xdp) < 0)
+ goto out_failure;
+ *status |= TSNEP_XDP_TX;
+ return true;
+ case XDP_REDIRECT:
+ if (xdp_do_redirect(rx->adapter->netdev, xdp, prog) < 0)
+ goto out_failure;
+ *status |= TSNEP_XDP_REDIRECT;
+ return true;
+ default:
+ bpf_warn_invalid_xdp_action(rx->adapter->netdev, prog, act);
+ fallthrough;
+ case XDP_ABORTED:
+out_failure:
+ trace_xdp_exception(rx->adapter->netdev, prog, act);
+ fallthrough;
+ case XDP_DROP:
+ page_pool_put_page(rx->page_pool, virt_to_head_page(xdp->data),
+ sync, true);
+ return true;
+ }
+}
+
+static void tsnep_finalize_xdp(struct tsnep_adapter *adapter, int status)
+{
+ int cpu = smp_processor_id();
+ int queue;
+ struct netdev_queue *nq;
+
+ if (status & TSNEP_XDP_TX) {
+ queue = cpu % adapter->num_tx_queues;
+ nq = netdev_get_tx_queue(adapter->netdev, queue);
+
+ __netif_tx_lock(nq, cpu);
+ tsnep_xdp_xmit_flush(&adapter->tx[queue]);
+ __netif_tx_unlock(nq);
+ }
+
+ if (status & TSNEP_XDP_REDIRECT)
+ xdp_do_flush();
+}
+
static struct sk_buff *tsnep_build_skb(struct tsnep_rx *rx, struct page *page,
int length)
{
@@ -1035,12 +1132,16 @@ static int tsnep_rx_poll(struct tsnep_rx *rx, struct napi_struct *napi,
int desc_available;
int done = 0;
enum dma_data_direction dma_dir;
+ struct bpf_prog *prog;
struct tsnep_rx_entry *entry;
+ struct xdp_buff xdp;
+ int xdp_status = 0;
struct sk_buff *skb;
int length;
desc_available = tsnep_rx_desc_available(rx);
dma_dir = page_pool_get_dma_dir(rx->page_pool);
+ prog = READ_ONCE(rx->adapter->xdp_prog);
while (likely(done < budget) && (rx->read != rx->write)) {
entry = &rx->entry[rx->read];
@@ -1084,6 +1185,28 @@ static int tsnep_rx_poll(struct tsnep_rx *rx, struct napi_struct *napi,
rx->read = (rx->read + 1) % TSNEP_RING_SIZE;
desc_available++;
+ if (prog) {
+ bool consume;
+
+ xdp_init_buff(&xdp, PAGE_SIZE, &rx->xdp_rxq);
+ xdp_prepare_buff(&xdp, page_address(entry->page),
+ tsnep_rx_offset_xdp() + TSNEP_RX_INLINE_METADATA_SIZE,
+ length - TSNEP_RX_INLINE_METADATA_SIZE,
+ false);
+
+ consume = tsnep_xdp_run_prog(rx, prog, &xdp,
+ &xdp_status);
+ if (consume) {
+ rx->packets++;
+ rx->bytes +=
+ length - TSNEP_RX_INLINE_METADATA_SIZE;
+
+ entry->page = NULL;
+
+ continue;
+ }
+ }
+
skb = tsnep_build_skb(rx, entry->page, length);
if (skb) {
page_pool_release_page(rx->page_pool, entry->page);
@@ -1102,6 +1225,9 @@ static int tsnep_rx_poll(struct tsnep_rx *rx, struct napi_struct *napi,
entry->page = NULL;
}
+ if (xdp_status)
+ tsnep_finalize_xdp(rx->adapter, xdp_status);
+
if (desc_available)
tsnep_rx_refill(rx, desc_available, false);
--
2.30.2
Powered by blists - more mailing lists