idpf: use libeth Rx buffer management for payload buffer

idpf uses Page Pool for data buffers with hardcoded buffer lengths of
4k for "classic" buffers and 2k for "short" ones. This is not flexible
and does not ensure optimal memory usage. Why would you need 4k buffers
when the MTU is 1500?
Use libeth for the data buffers and don't hardcode any buffer sizes. Let
them be calculated from the MTU for "classics" and then divide the
truesize by 2 for "short" ones. The memory usage is now greatly reduced
and 2 buffer queues starts make sense: on frames <= 1024, you'll recycle
(and resync) a page only after 4 HW writes rather than two.

Signed-off-by: Alexander Lobakin <[email protected]>
Signed-off-by: Tony Nguyen <[email protected]>

Author: alobakin

drivers/net/ethernet/intel/idpf/Kconfig

@@ -6,7 +6,6 @@ config IDPF
 	depends on PCI_MSI
 	select DIMLIB
 	select LIBETH
-	select PAGE_POOL
 	help
 	  This driver supports Intel(R) Infrastructure Data Path Function
 	  devices.

drivers/net/ethernet/intel/idpf/idpf.h

@@ -264,7 +264,6 @@ struct idpf_port_stats {
  *		    the worst case.
  * @num_bufqs_per_qgrp: Buffer queues per RX queue in a given grouping
  * @bufq_desc_count: Buffer queue descriptor count
- * @bufq_size: Size of buffers in ring (e.g. 2K, 4K, etc)
  * @num_rxq_grp: Number of RX queues in a group
  * @rxq_grps: Total number of RX groups. Number of groups * number of RX per
  *	      group will yield total number of RX queues.
@@ -308,7 +307,6 @@ struct idpf_vport {
 	u32 rxq_desc_count;
 	u8 num_bufqs_per_qgrp;
 	u32 bufq_desc_count[IDPF_MAX_BUFQS_PER_RXQ_GRP];
-	u32 bufq_size[IDPF_MAX_BUFQS_PER_RXQ_GRP];
 	u16 num_rxq_grp;
 	struct idpf_rxq_group *rxq_grps;
 	u32 rxq_model;

drivers/net/ethernet/intel/idpf/idpf_singleq_txrx.c

@@ -857,20 +857,24 @@ bool idpf_rx_singleq_buf_hw_alloc_all(struct idpf_rx_queue *rx_q,
 				      u16 cleaned_count)
 {
 	struct virtchnl2_singleq_rx_buf_desc *desc;
+	const struct libeth_fq_fp fq = {
+		.pp		= rx_q->pp,
+		.fqes		= rx_q->rx_buf,
+		.truesize	= rx_q->truesize,
+		.count		= rx_q->desc_count,
+	};
 	u16 nta = rx_q->next_to_alloc;
-	struct idpf_rx_buf *buf;
 
 	if (!cleaned_count)
 		return false;
 
 	desc = &rx_q->single_buf[nta];
-	buf = &rx_q->rx_buf[nta];
 
 	do {
 		dma_addr_t addr;
 
-		addr = idpf_alloc_page(rx_q->pp, buf, rx_q->rx_buf_size);
-		if (unlikely(addr == DMA_MAPPING_ERROR))
+		addr = libeth_rx_alloc(&fq, nta);
+		if (addr == DMA_MAPPING_ERROR)
 			break;
 
 		/* Refresh the desc even if buffer_addrs didn't change
@@ -880,11 +884,9 @@ bool idpf_rx_singleq_buf_hw_alloc_all(struct idpf_rx_queue *rx_q,
 		desc->hdr_addr = 0;
 		desc++;
 
-		buf++;
 		nta++;
 		if (unlikely(nta == rx_q->desc_count)) {
 			desc = &rx_q->single_buf[0];
-			buf = rx_q->rx_buf;
 			nta = 0;
 		}
 
@@ -1004,28 +1006,26 @@ static int idpf_rx_singleq_clean(struct idpf_rx_queue *rx_q, int budget)
 		idpf_rx_singleq_extract_fields(rx_q, rx_desc, &fields);
 
 		rx_buf = &rx_q->rx_buf[ntc];
-		if (!fields.size) {
-			idpf_rx_put_page(rx_buf);
+		if (!libeth_rx_sync_for_cpu(rx_buf, fields.size))
 			goto skip_data;
-		}
 
-		idpf_rx_sync_for_cpu(rx_buf, fields.size);
 		if (skb)
 			idpf_rx_add_frag(rx_buf, skb, fields.size);
 		else
-			skb = idpf_rx_construct_skb(rx_q, rx_buf, fields.size);
+			skb = idpf_rx_build_skb(rx_buf, fields.size);
 
 		/* exit if we failed to retrieve a buffer */
 		if (!skb)
 			break;
 
 skip_data:
-		IDPF_SINGLEQ_BUMP_RING_IDX(rx_q, ntc);
+		rx_buf->page = NULL;
 
+		IDPF_SINGLEQ_BUMP_RING_IDX(rx_q, ntc);
 		cleaned_count++;
 
 		/* skip if it is non EOP desc */
-		if (idpf_rx_singleq_is_non_eop(rx_desc))
+		if (idpf_rx_singleq_is_non_eop(rx_desc) || unlikely(!skb))
 			continue;
 
 #define IDPF_RXD_ERR_S FIELD_PREP(VIRTCHNL2_RX_BASE_DESC_QW1_ERROR_M, \
@@ -1062,6 +1062,7 @@ static int idpf_rx_singleq_clean(struct idpf_rx_queue *rx_q, int budget)
 
 	rx_q->next_to_clean = ntc;
 
+	page_pool_nid_changed(rx_q->pp, numa_mem_id());
 	if (cleaned_count)
 		failure = idpf_rx_singleq_buf_hw_alloc_all(rx_q, cleaned_count);