bna: Enable Multi Buffer RX

The CT2 HW supports multi-buffer Rx. This patch provides the necessary changes
for bnad to use multi-buffer Rx feature. For BNAD, multi-buffer Rx is by
default enabled when MTU is > 4096. For >4096 MTU, q0 data/large buffers are of
2048 size. As the resource requirements of multi-buffer Rx are different new Rx
needs to be created to use this feature. ASIC posts multiple completions if
frame exceeds buffer size. The last completion is marked with EOP flag.
 - Separate HQ and DQ enums for resource allocations and configurations.
 - rx_config and rxq structure changes to pass the correct info from bnad.
 - DQ depth need not be same as HQ depth. So CQ depth is adjusted accordingly.
 - Rx CFG frame size is taken from configured MTU.
 - Rx q0 buffer size is configured from bnad s rx_config when multi-buffer is
   enabled.
 - Poll for entire frame completion.
 - Once EOP completion is received gather the number of vectors used by the
   frame to submit it to the stack.
 - Changed MTU to frame size wherever necessary.
Signed-off-by: Rasesh Mody <rmody@brocade.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/ethernet/brocade/bna/bfi_enet.h

@@ -472,7 +472,8 @@ enum bfi_enet_hds_type {
 
 struct bfi_enet_rx_cfg {
 	u8		rxq_type;
-	u8		rsvd[3];
+	u8		rsvd[1];
+	u16		frame_size;
 
 	struct {
 		u8			max_header_size;

drivers/net/ethernet/brocade/bna/bna_hw_defs.h

@@ -322,6 +322,10 @@ do {									\
 #define	BNA_CQ_EF_REMOTE	(1 << 19)
 
 #define	BNA_CQ_EF_LOCAL		(1 << 20)
+/* CAT2 ASIC does not use bit 21 as per the SPEC.
+ * Bit 31 is set in every end of frame completion
+ */
+#define BNA_CQ_EF_EOP		(1 << 31)
 
 /* Data structures */
 

drivers/net/ethernet/brocade/bna/bna_tx_rx.c

@@ -1811,6 +1811,7 @@ bna_bfi_rx_enet_start(struct bna_rx *rx)
 	cfg_req->mh.num_entries = htons(
 		bfi_msgq_num_cmd_entries(sizeof(struct bfi_enet_rx_cfg_req)));
 
+	cfg_req->rx_cfg.frame_size = bna_enet_mtu_get(&rx->bna->enet);
 	cfg_req->num_queue_sets = rx->num_paths;
 	for (i = 0, rxp_qe = bfa_q_first(&rx->rxp_q);
 		i < rx->num_paths;
@@ -1832,8 +1833,17 @@ bna_bfi_rx_enet_start(struct bna_rx *rx)
 			/* Large/Single RxQ */
 			bfi_enet_datapath_q_init(&cfg_req->q_cfg[i].ql.q,
 						&q0->qpt);
-			q0->buffer_size =
-				bna_enet_mtu_get(&rx->bna->enet);
+			if (q0->multi_buffer)
+				/* multi-buffer is enabled by allocating
+				 * a new rx with new set of resources.
+				 * q0->buffer_size should be initialized to
+				 * fragment size.
+				 */
+				cfg_req->rx_cfg.multi_buffer =
+					BNA_STATUS_T_ENABLED;
+			else
+				q0->buffer_size =
+					bna_enet_mtu_get(&rx->bna->enet);
 			cfg_req->q_cfg[i].ql.rx_buffer_size =
 				htons((u16)q0->buffer_size);
 			break;
@@ -2383,8 +2393,8 @@ bna_rx_res_req(struct bna_rx_config *q_cfg, struct bna_res_info *res_info)
 	u32 hq_depth;
 	u32 dq_depth;
 
-	dq_depth = q_cfg->q_depth;
-	hq_depth = ((q_cfg->rxp_type == BNA_RXP_SINGLE) ? 0 : q_cfg->q_depth);
+	dq_depth = q_cfg->q0_depth;
+	hq_depth = ((q_cfg->rxp_type == BNA_RXP_SINGLE) ? 0 : q_cfg->q1_depth);
 	cq_depth = dq_depth + hq_depth;
 
 	BNA_TO_POWER_OF_2_HIGH(cq_depth);
@@ -2501,10 +2511,10 @@ bna_rx_create(struct bna *bna, struct bnad *bnad,
 	struct bna_rxq *q0;
 	struct bna_rxq *q1;
 	struct bna_intr_info *intr_info;
-	u32 page_count;
+	struct bna_mem_descr *hqunmap_mem;
+	struct bna_mem_descr *dqunmap_mem;
 	struct bna_mem_descr *ccb_mem;
 	struct bna_mem_descr *rcb_mem;
-	struct bna_mem_descr *unmapq_mem;
 	struct bna_mem_descr *cqpt_mem;
 	struct bna_mem_descr *cswqpt_mem;
 	struct bna_mem_descr *cpage_mem;
@@ -2514,8 +2524,10 @@ bna_rx_create(struct bna *bna, struct bnad *bnad,
 	struct bna_mem_descr *dsqpt_mem;
 	struct bna_mem_descr *hpage_mem;
 	struct bna_mem_descr *dpage_mem;
-	int i;
-	int dpage_count, hpage_count, rcb_idx;
+	u32 dpage_count, hpage_count;
+	u32 hq_idx, dq_idx, rcb_idx;
+	u32 cq_depth, i;
+	u32 page_count;
 
 	if (!bna_rx_res_check(rx_mod, rx_cfg))
 		return NULL;
@@ -2523,7 +2535,8 @@ bna_rx_create(struct bna *bna, struct bnad *bnad,
 	intr_info = &res_info[BNA_RX_RES_T_INTR].res_u.intr_info;
 	ccb_mem = &res_info[BNA_RX_RES_MEM_T_CCB].res_u.mem_info.mdl[0];
 	rcb_mem = &res_info[BNA_RX_RES_MEM_T_RCB].res_u.mem_info.mdl[0];
-	unmapq_mem = &res_info[BNA_RX_RES_MEM_T_UNMAPQ].res_u.mem_info.mdl[0];
+	dqunmap_mem = &res_info[BNA_RX_RES_MEM_T_UNMAPDQ].res_u.mem_info.mdl[0];
+	hqunmap_mem = &res_info[BNA_RX_RES_MEM_T_UNMAPHQ].res_u.mem_info.mdl[0];
 	cqpt_mem = &res_info[BNA_RX_RES_MEM_T_CQPT].res_u.mem_info.mdl[0];
 	cswqpt_mem = &res_info[BNA_RX_RES_MEM_T_CSWQPT].res_u.mem_info.mdl[0];
 	cpage_mem = &res_info[BNA_RX_RES_MEM_T_CQPT_PAGE].res_u.mem_info.mdl[0];
@@ -2575,7 +2588,8 @@ bna_rx_create(struct bna *bna, struct bnad *bnad,
 	}
 
 	rx->num_paths = rx_cfg->num_paths;
-	for (i = 0, rcb_idx = 0; i < rx->num_paths; i++) {
+	for (i = 0, hq_idx = 0, dq_idx = 0, rcb_idx = 0;
+			i < rx->num_paths; i++) {
 		rxp = bna_rxp_get(rx_mod);
 		list_add_tail(&rxp->qe, &rx->rxp_q);
 		rxp->type = rx_cfg->rxp_type;
@@ -2618,9 +2632,13 @@ bna_rx_create(struct bna *bna, struct bnad *bnad,
 		q0->rxp = rxp;
 
 		q0->rcb = (struct bna_rcb *) rcb_mem[rcb_idx].kva;
-		q0->rcb->unmap_q = (void *)unmapq_mem[rcb_idx].kva;
-		rcb_idx++;
-		q0->rcb->q_depth = rx_cfg->q_depth;
+		q0->rcb->unmap_q = (void *)dqunmap_mem[dq_idx].kva;
+		rcb_idx++; dq_idx++;
+		q0->rcb->q_depth = rx_cfg->q0_depth;
+		q0->q_depth = rx_cfg->q0_depth;
+		q0->multi_buffer = rx_cfg->q0_multi_buf;
+		q0->buffer_size = rx_cfg->q0_buf_size;
+		q0->num_vecs = rx_cfg->q0_num_vecs;
 		q0->rcb->rxq = q0;
 		q0->rcb->bnad = bna->bnad;
 		q0->rcb->id = 0;
@@ -2640,15 +2658,18 @@ bna_rx_create(struct bna *bna, struct bnad *bnad,
 			q1->rxp = rxp;
 
 			q1->rcb = (struct bna_rcb *) rcb_mem[rcb_idx].kva;
-			q1->rcb->unmap_q = (void *)unmapq_mem[rcb_idx].kva;
-			rcb_idx++;
-			q1->rcb->q_depth = rx_cfg->q_depth;
+			q1->rcb->unmap_q = (void *)hqunmap_mem[hq_idx].kva;
+			rcb_idx++; hq_idx++;
+			q1->rcb->q_depth = rx_cfg->q1_depth;
+			q1->q_depth = rx_cfg->q1_depth;
+			q1->multi_buffer = BNA_STATUS_T_DISABLED;
+			q1->num_vecs = 1;
 			q1->rcb->rxq = q1;
 			q1->rcb->bnad = bna->bnad;
 			q1->rcb->id = 1;
 			q1->buffer_size = (rx_cfg->rxp_type == BNA_RXP_HDS) ?
 					rx_cfg->hds_config.forced_offset
-					: rx_cfg->small_buff_size;
+					: rx_cfg->q1_buf_size;
 			q1->rx_packets = q1->rx_bytes = 0;
 			q1->rx_packets_with_error = q1->rxbuf_alloc_failed = 0;
 
@@ -2663,9 +2684,14 @@ bna_rx_create(struct bna *bna, struct bnad *bnad,
 		/* Setup CQ */
 
 		rxp->cq.ccb = (struct bna_ccb *) ccb_mem[i].kva;
-		rxp->cq.ccb->q_depth =	rx_cfg->q_depth +
-					((rx_cfg->rxp_type == BNA_RXP_SINGLE) ?
-					0 : rx_cfg->q_depth);
+		cq_depth = rx_cfg->q0_depth +
+			((rx_cfg->rxp_type == BNA_RXP_SINGLE) ?
+			 0 : rx_cfg->q1_depth);
+		/* if multi-buffer is enabled sum of q0_depth
+		 * and q1_depth need not be a power of 2
+		 */
+		BNA_TO_POWER_OF_2_HIGH(cq_depth);
+		rxp->cq.ccb->q_depth = cq_depth;
 		rxp->cq.ccb->cq = &rxp->cq;
 		rxp->cq.ccb->rcb[0] = q0->rcb;
 		q0->rcb->ccb = rxp->cq.ccb;

drivers/net/ethernet/brocade/bna/bna_types.h

@@ -109,20 +109,21 @@ enum bna_tx_res_req_type {
 enum bna_rx_mem_type {
 	BNA_RX_RES_MEM_T_CCB		= 0,	/* CQ context */
 	BNA_RX_RES_MEM_T_RCB		= 1,	/* CQ context */
-	BNA_RX_RES_MEM_T_UNMAPQ		= 2,	/* UnmapQ for RxQs */
-	BNA_RX_RES_MEM_T_CQPT		= 3,	/* CQ QPT */
-	BNA_RX_RES_MEM_T_CSWQPT		= 4,	/* S/W QPT */
-	BNA_RX_RES_MEM_T_CQPT_PAGE	= 5,	/* CQPT page */
-	BNA_RX_RES_MEM_T_HQPT		= 6,	/* RX QPT */
-	BNA_RX_RES_MEM_T_DQPT		= 7,	/* RX QPT */
-	BNA_RX_RES_MEM_T_HSWQPT		= 8,	/* RX s/w QPT */
-	BNA_RX_RES_MEM_T_DSWQPT		= 9,	/* RX s/w QPT */
-	BNA_RX_RES_MEM_T_DPAGE		= 10,	/* RX s/w QPT */
-	BNA_RX_RES_MEM_T_HPAGE		= 11,	/* RX s/w QPT */
-	BNA_RX_RES_MEM_T_IBIDX		= 12,
-	BNA_RX_RES_MEM_T_RIT		= 13,
-	BNA_RX_RES_T_INTR		= 14,	/* Rx interrupts */
-	BNA_RX_RES_T_MAX		= 15
+	BNA_RX_RES_MEM_T_UNMAPHQ	= 2,
+	BNA_RX_RES_MEM_T_UNMAPDQ	= 3,
+	BNA_RX_RES_MEM_T_CQPT		= 4,
+	BNA_RX_RES_MEM_T_CSWQPT		= 5,
+	BNA_RX_RES_MEM_T_CQPT_PAGE	= 6,
+	BNA_RX_RES_MEM_T_HQPT		= 7,
+	BNA_RX_RES_MEM_T_DQPT		= 8,
+	BNA_RX_RES_MEM_T_HSWQPT		= 9,
+	BNA_RX_RES_MEM_T_DSWQPT		= 10,
+	BNA_RX_RES_MEM_T_DPAGE		= 11,
+	BNA_RX_RES_MEM_T_HPAGE		= 12,
+	BNA_RX_RES_MEM_T_IBIDX		= 13,
+	BNA_RX_RES_MEM_T_RIT		= 14,
+	BNA_RX_RES_T_INTR		= 15,
+	BNA_RX_RES_T_MAX		= 16
 };
 
 enum bna_tx_type {
@@ -583,6 +584,8 @@ struct bna_rxq {
 
 	int			buffer_size;
 	int			q_depth;
+	u32			num_vecs;
+	enum bna_status		multi_buffer;
 
 	struct bna_qpt qpt;
 	struct bna_rcb *rcb;
@@ -632,6 +635,8 @@ struct bna_ccb {
 	struct bna_rcb *rcb[2];
 	void			*ctrl; /* For bnad */
 	struct bna_pkt_rate pkt_rate;
+	u32			pkts_una;
+	u32			bytes_per_intr;
 
 	/* Control path */
 	struct bna_cq *cq;
@@ -671,14 +676,22 @@ struct bna_rx_config {
 	int			num_paths;
 	enum bna_rxp_type rxp_type;
 	int			paused;
-	int			q_depth;
 	int			coalescing_timeo;
 	/*
 	 * Small/Large (or Header/Data) buffer size to be configured
-	 * for SLR and HDS queue type. Large buffer size comes from
-	 * enet->mtu.
+	 * for SLR and HDS queue type.
 	 */
-	int			small_buff_size;
+	u32			frame_size;
+
+	/* header or small queue */
+	u32			q1_depth;
+	u32			q1_buf_size;
+
+	/* data or large queue */
+	u32			q0_depth;
+	u32			q0_buf_size;
+	u32			q0_num_vecs;
+	enum bna_status		q0_multi_buf;
 
 	enum bna_status rss_status;
 	struct bna_rss_config rss_config;

drivers/net/ethernet/brocade/bna/bnad.h

@@ -105,6 +105,9 @@ struct bnad_rx_ctrl {
 #define BNAD_NUM_TXQ			(bnad->num_tx * bnad->num_txq_per_tx)
 #define BNAD_NUM_RXP			(bnad->num_rx * bnad->num_rxp_per_rx)
 
+#define BNAD_FRAME_SIZE(_mtu) \
+	(ETH_HLEN + VLAN_HLEN + (_mtu) + ETH_FCS_LEN)
+
 /*
  * DATA STRUCTURES
  */
@@ -241,12 +244,13 @@ struct bnad_rx_unmap {
 
 enum bnad_rxbuf_type {
 	BNAD_RXBUF_NONE		= 0,
-	BNAD_RXBUF_SKB		= 1,
+	BNAD_RXBUF_SK_BUFF	= 1,
 	BNAD_RXBUF_PAGE		= 2,
-	BNAD_RXBUF_MULTI	= 3
+	BNAD_RXBUF_MULTI_BUFF	= 3
 };
 
-#define BNAD_RXBUF_IS_PAGE(_type)	((_type) == BNAD_RXBUF_PAGE)
+#define BNAD_RXBUF_IS_SK_BUFF(_type)	((_type) == BNAD_RXBUF_SK_BUFF)
+#define BNAD_RXBUF_IS_MULTI_BUFF(_type)	((_type) == BNAD_RXBUF_MULTI_BUFF)
 
 struct bnad_rx_unmap_q {
 	int			reuse_pi;
@@ -256,6 +260,9 @@ struct bnad_rx_unmap_q {
 	struct bnad_rx_unmap	unmap[0];
 };
 
+#define BNAD_PCI_DEV_IS_CAT2(_bnad) \
+	((_bnad)->pcidev->device == BFA_PCI_DEVICE_ID_CT2)
+
 /* Bit mask values for bnad->cfg_flags */
 #define	BNAD_CF_DIM_ENABLED		0x01	/* DIM */
 #define	BNAD_CF_PROMISC			0x02