RDMA/cxgb4: Use DSGLs for fastreg and adapter memory writes for T5.

It enables direct DMA by HW to memory region PBL arrays and fast register PBL
arrays from host memory, vs the T4 way of passing these arrays in the WR itself.
The result is lower latency for memory registration, and larger PBL array
support for fast register operations.

This patch also updates ULP_TX_MEM_WRITE command fields for T5. Ordering bit of
ULP_TX_MEM_WRITE is at bit position 22 in T5 and at 23 in T4.

Signed-off-by: Vipul Pandya <vipul@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/infiniband/hw/cxgb4/iw_cxgb4.h

@@ -369,7 +369,6 @@ struct c4iw_fr_page_list {
 	DEFINE_DMA_UNMAP_ADDR(mapping);
 	dma_addr_t dma_addr;
 	struct c4iw_dev *dev;
-	int size;
 };
 
 static inline struct c4iw_fr_page_list *to_c4iw_fr_page_list(
@@ -940,6 +939,7 @@ extern c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS];
 extern int c4iw_max_read_depth;
 extern int db_fc_threshold;
 extern int db_coalescing_threshold;
+extern int use_dsgl;
 
 
 #endif

drivers/infiniband/hw/cxgb4/mem.c

@@ -30,23 +30,89 @@
  * SOFTWARE.
  */
 
+#include <linux/module.h>
+#include <linux/moduleparam.h>
 #include <rdma/ib_umem.h>
 #include <linux/atomic.h>
 
 #include "iw_cxgb4.h"
 
+int use_dsgl = 1;
+module_param(use_dsgl, int, 0644);
+MODULE_PARM_DESC(use_dsgl, "Use DSGL for PBL/FastReg (default=1)");
+
 #define T4_ULPTX_MIN_IO 32
 #define C4IW_MAX_INLINE_SIZE 96
+#define T4_ULPTX_MAX_DMA 1024
+#define C4IW_INLINE_THRESHOLD 128
 
-static int write_adapter_mem(struct c4iw_rdev *rdev, u32 addr, u32 len,
-			     void *data)
+static int inline_threshold = C4IW_INLINE_THRESHOLD;
+module_param(inline_threshold, int, 0644);
+MODULE_PARM_DESC(inline_threshold, "inline vs dsgl threshold (default=128)");
+
+static int _c4iw_write_mem_dma_aligned(struct c4iw_rdev *rdev, u32 addr,
+				       u32 len, void *data, int wait)
+{
+	struct sk_buff *skb;
+	struct ulp_mem_io *req;
+	struct ulptx_sgl *sgl;
+	u8 wr_len;
+	int ret = 0;
+	struct c4iw_wr_wait wr_wait;
+
+	addr &= 0x7FFFFFF;
+
+	if (wait)
+		c4iw_init_wr_wait(&wr_wait);
+	wr_len = roundup(sizeof(*req) + sizeof(*sgl), 16);
+
+	skb = alloc_skb(wr_len, GFP_KERNEL | __GFP_NOFAIL);
+	if (!skb)
+		return -ENOMEM;
+	set_wr_txq(skb, CPL_PRIORITY_CONTROL, 0);
+
+	req = (struct ulp_mem_io *)__skb_put(skb, wr_len);
+	memset(req, 0, wr_len);
+	INIT_ULPTX_WR(req, wr_len, 0, 0);
+	req->wr.wr_hi = cpu_to_be32(FW_WR_OP(FW_ULPTX_WR) |
+			(wait ? FW_WR_COMPL(1) : 0));
+	req->wr.wr_lo = wait ? (__force __be64)&wr_wait : 0;
+	req->wr.wr_mid = cpu_to_be32(FW_WR_LEN16(DIV_ROUND_UP(wr_len, 16)));
+	req->cmd = cpu_to_be32(ULPTX_CMD(ULP_TX_MEM_WRITE));
+	req->cmd |= cpu_to_be32(V_T5_ULP_MEMIO_ORDER(1));
+	req->dlen = cpu_to_be32(ULP_MEMIO_DATA_LEN(len>>5));
+	req->len16 = cpu_to_be32(DIV_ROUND_UP(wr_len-sizeof(req->wr), 16));
+	req->lock_addr = cpu_to_be32(ULP_MEMIO_ADDR(addr));
+
+	sgl = (struct ulptx_sgl *)(req + 1);
+	sgl->cmd_nsge = cpu_to_be32(ULPTX_CMD(ULP_TX_SC_DSGL) |
+				    ULPTX_NSGE(1));
+	sgl->len0 = cpu_to_be32(len);
+	sgl->addr0 = cpu_to_be64(virt_to_phys(data));
+
+	ret = c4iw_ofld_send(rdev, skb);
+	if (ret)
+		return ret;
+	if (wait)
+		ret = c4iw_wait_for_reply(rdev, &wr_wait, 0, 0, __func__);
+	return ret;
+}
+
+static int _c4iw_write_mem_inline(struct c4iw_rdev *rdev, u32 addr, u32 len,
+				  void *data)
 {
 	struct sk_buff *skb;
 	struct ulp_mem_io *req;
 	struct ulptx_idata *sc;
 	u8 wr_len, *to_dp, *from_dp;
 	int copy_len, num_wqe, i, ret = 0;
 	struct c4iw_wr_wait wr_wait;
+	__be32 cmd = cpu_to_be32(ULPTX_CMD(ULP_TX_MEM_WRITE));
+
+	if (is_t4(rdev->lldi.adapter_type))
+		cmd |= cpu_to_be32(ULP_MEMIO_ORDER(1));
+	else
+		cmd |= cpu_to_be32(V_T5_ULP_MEMIO_IMM(1));
 
 	addr &= 0x7FFFFFF;
 	PDBG("%s addr 0x%x len %u\n", __func__, addr, len);
@@ -77,7 +143,7 @@ static int write_adapter_mem(struct c4iw_rdev *rdev, u32 addr, u32 len,
 		req->wr.wr_mid = cpu_to_be32(
 				       FW_WR_LEN16(DIV_ROUND_UP(wr_len, 16)));
 
-		req->cmd = cpu_to_be32(ULPTX_CMD(ULP_TX_MEM_WRITE) | (1<<23));
+		req->cmd = cmd;
 		req->dlen = cpu_to_be32(ULP_MEMIO_DATA_LEN(
 				DIV_ROUND_UP(copy_len, T4_ULPTX_MIN_IO)));
 		req->len16 = cpu_to_be32(DIV_ROUND_UP(wr_len-sizeof(req->wr),
@@ -107,6 +173,50 @@ static int write_adapter_mem(struct c4iw_rdev *rdev, u32 addr, u32 len,
 	return ret;
 }
 
+int _c4iw_write_mem_dma(struct c4iw_rdev *rdev, u32 addr, u32 len, void *data)
+{
+	u32 remain = len;
+	u32 dmalen;
+	int ret = 0;
+
+	while (remain > inline_threshold) {
+		if (remain < T4_ULPTX_MAX_DMA) {
+			if (remain & ~T4_ULPTX_MIN_IO)
+				dmalen = remain & ~(T4_ULPTX_MIN_IO-1);
+			else
+				dmalen = remain;
+		} else
+			dmalen = T4_ULPTX_MAX_DMA;
+		remain -= dmalen;
+		ret = _c4iw_write_mem_dma_aligned(rdev, addr, dmalen, data,
+						 !remain);
+		if (ret)
+			goto out;
+		addr += dmalen >> 5;
+		data += dmalen;
+	}
+	if (remain)
+		ret = _c4iw_write_mem_inline(rdev, addr, remain, data);
+out:
+	return ret;
+}
+
+/*
+ * write len bytes of data into addr (32B aligned address)
+ * If data is NULL, clear len byte of memory to zero.
+ */
+static int write_adapter_mem(struct c4iw_rdev *rdev, u32 addr, u32 len,
+			     void *data)
+{
+	if (is_t5(rdev->lldi.adapter_type) && use_dsgl) {
+		if (len > inline_threshold)
+			return _c4iw_write_mem_dma(rdev, addr, len, data);
+		else
+			return _c4iw_write_mem_inline(rdev, addr, len, data);
+	} else
+		return _c4iw_write_mem_inline(rdev, addr, len, data);
+}
+
 /*
  * Build and write a TPT entry.
  * IN: stag key, pdid, perm, bind_enabled, zbva, to, len, page_size,
@@ -760,19 +870,23 @@ struct ib_fast_reg_page_list *c4iw_alloc_fastreg_pbl(struct ib_device *device,
 	struct c4iw_fr_page_list *c4pl;
 	struct c4iw_dev *dev = to_c4iw_dev(device);
 	dma_addr_t dma_addr;
-	int size = sizeof *c4pl + page_list_len * sizeof(u64);
+	int pll_len = roundup(page_list_len * sizeof(u64), 32);
 
-	c4pl = dma_alloc_coherent(&dev->rdev.lldi.pdev->dev, size,
-				  &dma_addr, GFP_KERNEL);
+	c4pl = kmalloc(sizeof(*c4pl), GFP_KERNEL);
 	if (!c4pl)
 		return ERR_PTR(-ENOMEM);
 
+	c4pl->ibpl.page_list = dma_alloc_coherent(&dev->rdev.lldi.pdev->dev,
+						  pll_len, &dma_addr,
+						  GFP_KERNEL);
+	if (!c4pl->ibpl.page_list) {
+		kfree(c4pl);
+		return ERR_PTR(-ENOMEM);
+	}
 	dma_unmap_addr_set(c4pl, mapping, dma_addr);
 	c4pl->dma_addr = dma_addr;
 	c4pl->dev = dev;
-	c4pl->size = size;
-	c4pl->ibpl.page_list = (u64 *)(c4pl + 1);
-	c4pl->ibpl.max_page_list_len = page_list_len;
+	c4pl->ibpl.max_page_list_len = pll_len;
 
 	return &c4pl->ibpl;
 }
@@ -781,8 +895,10 @@ void c4iw_free_fastreg_pbl(struct ib_fast_reg_page_list *ibpl)
 {
 	struct c4iw_fr_page_list *c4pl = to_c4iw_fr_page_list(ibpl);
 
-	dma_free_coherent(&c4pl->dev->rdev.lldi.pdev->dev, c4pl->size,
-			  c4pl, dma_unmap_addr(c4pl, mapping));
+	dma_free_coherent(&c4pl->dev->rdev.lldi.pdev->dev,
+			  c4pl->ibpl.max_page_list_len,
+			  c4pl->ibpl.page_list, dma_unmap_addr(c4pl, mapping));
+	kfree(c4pl);
 }
 
 int c4iw_dereg_mr(struct ib_mr *ib_mr)

drivers/infiniband/hw/cxgb4/qp.c

@@ -54,6 +54,10 @@ MODULE_PARM_DESC(db_coalescing_threshold,
 		 "QP count/threshold that triggers"
 		 " disabling db coalescing (default = 0)");
 
+static int max_fr_immd = T4_MAX_FR_IMMD;
+module_param(max_fr_immd, int, 0644);
+MODULE_PARM_DESC(max_fr_immd, "fastreg threshold for using DSGL instead of immedate");
+
 static void set_state(struct c4iw_qp *qhp, enum c4iw_qp_state state)
 {
 	unsigned long flag;
@@ -539,7 +543,7 @@ static int build_rdma_recv(struct c4iw_qp *qhp, union t4_recv_wr *wqe,
 }
 
 static int build_fastreg(struct t4_sq *sq, union t4_wr *wqe,
-			 struct ib_send_wr *wr, u8 *len16)
+			 struct ib_send_wr *wr, u8 *len16, u8 t5dev)
 {
 
 	struct fw_ri_immd *imdp;
@@ -561,28 +565,51 @@ static int build_fastreg(struct t4_sq *sq, union t4_wr *wqe,
 	wqe->fr.va_hi = cpu_to_be32(wr->wr.fast_reg.iova_start >> 32);
 	wqe->fr.va_lo_fbo = cpu_to_be32(wr->wr.fast_reg.iova_start &
 					0xffffffff);
-	WARN_ON(pbllen > T4_MAX_FR_IMMD);
-	imdp = (struct fw_ri_immd *)(&wqe->fr + 1);
-	imdp->op = FW_RI_DATA_IMMD;
-	imdp->r1 = 0;
-	imdp->r2 = 0;
-	imdp->immdlen = cpu_to_be32(pbllen);
-	p = (__be64 *)(imdp + 1);
-	rem = pbllen;
-	for (i = 0; i < wr->wr.fast_reg.page_list_len; i++) {
-		*p = cpu_to_be64((u64)wr->wr.fast_reg.page_list->page_list[i]);
-		rem -= sizeof *p;
-		if (++p == (__be64 *)&sq->queue[sq->size])
-			p = (__be64 *)sq->queue;
-	}
-	BUG_ON(rem < 0);
-	while (rem) {
-		*p = 0;
-		rem -= sizeof *p;
-		if (++p == (__be64 *)&sq->queue[sq->size])
-			p = (__be64 *)sq->queue;
+
+	if (t5dev && use_dsgl && (pbllen > max_fr_immd)) {
+		struct c4iw_fr_page_list *c4pl =
+			to_c4iw_fr_page_list(wr->wr.fast_reg.page_list);
+		struct fw_ri_dsgl *sglp;
+
+		for (i = 0; i < wr->wr.fast_reg.page_list_len; i++) {
+			wr->wr.fast_reg.page_list->page_list[i] = (__force u64)
+				cpu_to_be64((u64)
+				wr->wr.fast_reg.page_list->page_list[i]);
+		}
+
+		sglp = (struct fw_ri_dsgl *)(&wqe->fr + 1);
+		sglp->op = FW_RI_DATA_DSGL;
+		sglp->r1 = 0;
+		sglp->nsge = cpu_to_be16(1);
+		sglp->addr0 = cpu_to_be64(c4pl->dma_addr);
+		sglp->len0 = cpu_to_be32(pbllen);
+
+		*len16 = DIV_ROUND_UP(sizeof(wqe->fr) + sizeof(*sglp), 16);
+	} else {
+		imdp = (struct fw_ri_immd *)(&wqe->fr + 1);
+		imdp->op = FW_RI_DATA_IMMD;
+		imdp->r1 = 0;
+		imdp->r2 = 0;
+		imdp->immdlen = cpu_to_be32(pbllen);
+		p = (__be64 *)(imdp + 1);
+		rem = pbllen;
+		for (i = 0; i < wr->wr.fast_reg.page_list_len; i++) {
+			*p = cpu_to_be64(
+				(u64)wr->wr.fast_reg.page_list->page_list[i]);
+			rem -= sizeof(*p);
+			if (++p == (__be64 *)&sq->queue[sq->size])
+				p = (__be64 *)sq->queue;
+		}
+		BUG_ON(rem < 0);
+		while (rem) {
+			*p = 0;
+			rem -= sizeof(*p);
+			if (++p == (__be64 *)&sq->queue[sq->size])
+				p = (__be64 *)sq->queue;
+		}
+		*len16 = DIV_ROUND_UP(sizeof(wqe->fr) + sizeof(*imdp)
+				      + pbllen, 16);
 	}
-	*len16 = DIV_ROUND_UP(sizeof wqe->fr + sizeof *imdp + pbllen, 16);
 	return 0;
 }
 
@@ -683,7 +710,10 @@ int c4iw_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 		case IB_WR_FAST_REG_MR:
 			fw_opcode = FW_RI_FR_NSMR_WR;
 			swsqe->opcode = FW_RI_FAST_REGISTER;
-			err = build_fastreg(&qhp->wq.sq, wqe, wr, &len16);
+			err = build_fastreg(&qhp->wq.sq, wqe, wr, &len16,
+					    is_t5(
+					    qhp->rhp->rdev.lldi.adapter_type) ?
+					    1 : 0);
 			break;
 		case IB_WR_LOCAL_INV:
 			if (wr->send_flags & IB_SEND_FENCE)

drivers/infiniband/hw/cxgb4/t4.h

@@ -84,7 +84,7 @@ struct t4_status_page {
 			sizeof(struct fw_ri_isgl)) / sizeof(struct fw_ri_sge))
 #define T4_MAX_FR_IMMD ((T4_SQ_NUM_BYTES - sizeof(struct fw_ri_fr_nsmr_wr) - \
 			sizeof(struct fw_ri_immd)) & ~31UL)
-#define T4_MAX_FR_DEPTH (T4_MAX_FR_IMMD / sizeof(u64))
+#define T4_MAX_FR_DEPTH (1024 / sizeof(u64))
 
 #define T4_RQ_NUM_SLOTS 2
 #define T4_RQ_NUM_BYTES (T4_EQ_ENTRY_SIZE * T4_RQ_NUM_SLOTS)

drivers/net/ethernet/chelsio/cxgb4/t4_msg.h

@@ -787,4 +787,12 @@ struct ulp_mem_io {
 #define ULP_MEMIO_LOCK(x) ((x) << 31)
 };
 
+#define S_T5_ULP_MEMIO_IMM    23
+#define V_T5_ULP_MEMIO_IMM(x) ((x) << S_T5_ULP_MEMIO_IMM)
+#define F_T5_ULP_MEMIO_IMM    V_T5_ULP_MEMIO_IMM(1U)
+
+#define S_T5_ULP_MEMIO_ORDER    22
+#define V_T5_ULP_MEMIO_ORDER(x) ((x) << S_T5_ULP_MEMIO_ORDER)
+#define F_T5_ULP_MEMIO_ORDER    V_T5_ULP_MEMIO_ORDER(1U)
+
 #endif  /* __T4_MSG_H */