Merge branch 'gve-implement-xdp-hw-rx-timestamping-support-for-dq'

Tim Hostetler says:

====================
gve: Implement XDP HW RX Timestamping support for DQ

From: Tim Hostetler <[e-mail]>

This patch series adds support for bpf_xdp_metadata_rx_timestamp from an
XDP program loaded into the driver on its own or bound to an XSK. This
is only supported for DQ.
====================

Link: https://patch.msgid.link/[e-mail]
Signed-off-by: Paolo Abeni <[e-mail]>

drivers/net/ethernet/google/gve/gve.h

···
 	s16 next;
 };
 
+/* Wrapper for XDP Rx metadata */
+struct gve_xdp_buff {
+	struct xdp_buff xdp;
+	struct gve_priv *gve;
+	const struct gve_rx_compl_desc_dqo *compl_desc;
+};
+
 /* `head` and `tail` are indices into an array, or -1 if empty. */
 struct gve_index_list {
 	s16 head;

drivers/net/ethernet/google/gve/gve_dqo.h

···
 netdev_features_t gve_features_check_dqo(struct sk_buff *skb,
 					 struct net_device *dev,
 					 netdev_features_t features);
+int gve_xdp_rx_timestamp(const struct xdp_md *_ctx, u64 *timestamp);
 bool gve_tx_poll_dqo(struct gve_notify_block *block, bool do_clean);
 bool gve_xdp_poll_dqo(struct gve_notify_block *block);
 bool gve_xsk_tx_poll_dqo(struct gve_notify_block *block, int budget);

drivers/net/ethernet/google/gve/gve_main.c

···
 		}
 
 		kernel_config->rx_filter = HWTSTAMP_FILTER_ALL;
-		gve_clock_nic_ts_read(priv);
-		ptp_schedule_worker(priv->ptp->clock, 0);
-	} else {
-		ptp_cancel_worker_sync(priv->ptp->clock);
 	}
 
 	priv->ts_config.rx_filter = kernel_config->rx_filter;
···
 	xdp_set_features_flag_locked(priv->dev, xdp_features);
 }
 
+static const struct xdp_metadata_ops gve_xdp_metadata_ops = {
+	.xmo_rx_timestamp	= gve_xdp_rx_timestamp,
+};
+
 static int gve_init_priv(struct gve_priv *priv, bool skip_describe_device)
 {
 	int num_ntfy;
···
 	}
 
 	gve_set_netdev_xdp_features(priv);
+	if (!gve_is_gqi(priv))
+		priv->dev->xdp_metadata_ops = &gve_xdp_metadata_ops;
+
 	err = gve_setup_device_resources(priv);
 	if (err)
 		goto err_free_xsk_bitmap;

drivers/net/ethernet/google/gve/gve_ptp.c

···
 		err = -ENOMEM;
 		goto release_ptp;
 	}
+	err = gve_clock_nic_ts_read(priv);
+	if (err) {
+		dev_err(&priv->pdev->dev, "failed to read NIC clock %d\n", err);
+		goto release_nic_ts_report;
+	}
+	ptp_schedule_worker(priv->ptp->clock,
+			    msecs_to_jiffies(GVE_NIC_TS_SYNC_INTERVAL_MS));
 
 	return 0;
 
+release_nic_ts_report:
+	dma_free_coherent(&priv->pdev->dev,
+			  sizeof(struct gve_nic_ts_report),
+			  priv->nic_ts_report, priv->nic_ts_report_bus);
+	priv->nic_ts_report = NULL;
 release_ptp:
 	gve_ptp_release(priv);
 	return err;

drivers/net/ethernet/google/gve/gve_rx_dqo.c

···
 		rx->rx_headroom = 0;
 	}
 
+	/* struct gve_xdp_buff is overlaid on struct xdp_buff_xsk and utilizes
+	 * the 24 byte field cb to store gve specific data.
+	 */
+	XSK_CHECK_PRIV_TYPE(struct gve_xdp_buff);
+
 	rx->dqo.num_buf_states = cfg->raw_addressing ? buffer_queue_slots :
 		gve_get_rx_pages_per_qpl_dqo(cfg->ring_size);
 	rx->dqo.buf_states = kvcalloc_node(rx->dqo.num_buf_states,
···
  * Note that this means if the time delta between packet reception and the last
  * clock read is greater than ~2 seconds, this will provide invalid results.
  */
+static ktime_t gve_rx_get_hwtstamp(struct gve_priv *gve, u32 hwts)
+{
+	u64 last_read = READ_ONCE(gve->last_sync_nic_counter);
+	u32 low = (u32)last_read;
+	s32 diff = hwts - low;
+
+	return ns_to_ktime(last_read + diff);
+}
+
 static void gve_rx_skb_hwtstamp(struct gve_rx_ring *rx,
 				const struct gve_rx_compl_desc_dqo *desc)
 {
-	u64 last_read = READ_ONCE(rx->gve->last_sync_nic_counter);
 	struct sk_buff *skb = rx->ctx.skb_head;
-	u32 ts, low;
-	s32 diff;
 
-	if (desc->ts_sub_nsecs_low & GVE_DQO_RX_HWTSTAMP_VALID) {
-		ts = le32_to_cpu(desc->ts);
-		low = (u32)last_read;
-		diff = ts - low;
-		skb_hwtstamps(skb)->hwtstamp = ns_to_ktime(last_read + diff);
-	}
+	if (desc->ts_sub_nsecs_low & GVE_DQO_RX_HWTSTAMP_VALID)
+		skb_hwtstamps(skb)->hwtstamp =
+			gve_rx_get_hwtstamp(rx->gve, le32_to_cpu(desc->ts));
+}
+
+int gve_xdp_rx_timestamp(const struct xdp_md *_ctx, u64 *timestamp)
+{
+	const struct gve_xdp_buff *ctx = (void *)_ctx;
+
+	if (!ctx->gve->nic_ts_report)
+		return -ENODATA;
+
+	if (!(ctx->compl_desc->ts_sub_nsecs_low & GVE_DQO_RX_HWTSTAMP_VALID))
+		return -ENODATA;
+
+	*timestamp = gve_rx_get_hwtstamp(ctx->gve,
+					 le32_to_cpu(ctx->compl_desc->ts));
+	return 0;
 }
 
 static void gve_rx_free_skb(struct napi_struct *napi, struct gve_rx_ring *rx)
···
 }
 
 static int gve_rx_xsk_dqo(struct napi_struct *napi, struct gve_rx_ring *rx,
-			  struct gve_rx_buf_state_dqo *buf_state, int buf_len,
+			  const struct gve_rx_compl_desc_dqo *compl_desc,
+			  struct gve_rx_buf_state_dqo *buf_state,
 			  struct bpf_prog *xprog)
 {
 	struct xdp_buff *xdp = buf_state->xsk_buff;
+	int buf_len = compl_desc->packet_len;
 	struct gve_priv *priv = rx->gve;
+	struct gve_xdp_buff *gve_xdp;
 	int xdp_act;
 
 	xdp->data_end = xdp->data + buf_len;
 	xsk_buff_dma_sync_for_cpu(xdp);
+
+	gve_xdp = (void *)xdp;
+	gve_xdp->gve = priv;
+	gve_xdp->compl_desc = compl_desc;
 
 	if (xprog) {
 		xdp_act = bpf_prog_run_xdp(xprog, xdp);
···
 
 	xprog = READ_ONCE(priv->xdp_prog);
 	if (buf_state->xsk_buff)
-		return gve_rx_xsk_dqo(napi, rx, buf_state, buf_len, xprog);
+		return gve_rx_xsk_dqo(napi, rx, compl_desc, buf_state, xprog);
 
 	/* Page might have not been used for awhile and was likely last written
 	 * by a different thread.
···
 	}
 
 	if (xprog) {
-		struct xdp_buff xdp;
+		struct gve_xdp_buff gve_xdp;
 		void *old_data;
 		int xdp_act;
 
-		xdp_init_buff(&xdp, buf_state->page_info.buf_size,
+		xdp_init_buff(&gve_xdp.xdp, buf_state->page_info.buf_size,
 			      &rx->xdp_rxq);
-		xdp_prepare_buff(&xdp,
+		xdp_prepare_buff(&gve_xdp.xdp,
 				 buf_state->page_info.page_address +
 				 buf_state->page_info.page_offset,
 				 buf_state->page_info.pad,
 				 buf_len, false);
-		old_data = xdp.data;
-		xdp_act = bpf_prog_run_xdp(xprog, &xdp);
-		buf_state->page_info.pad += xdp.data - old_data;
-		buf_len = xdp.data_end - xdp.data;
+		gve_xdp.gve = priv;
+		gve_xdp.compl_desc = compl_desc;
+
+		old_data = gve_xdp.xdp.data;
+		xdp_act = bpf_prog_run_xdp(xprog, &gve_xdp.xdp);
+		buf_state->page_info.pad += gve_xdp.xdp.data - old_data;
+		buf_len = gve_xdp.xdp.data_end - gve_xdp.xdp.data;
 		if (xdp_act != XDP_PASS) {
-			gve_xdp_done_dqo(priv, rx, &xdp, xprog, xdp_act,
+			gve_xdp_done_dqo(priv, rx, &gve_xdp.xdp, xprog, xdp_act,
 					 buf_state);
 			return 0;
 		}