[PATCH net-next v5 0/7] vmxnet3: upgrade to version 3

[PATCH net-next v5 0/7] vmxnet3: upgrade to version 3

[Shrikrishna Khare]

vmxnet3 emulation has recently added several new features which includes
support for new commands the driver can issue to emulation, change in
descriptor fields etc. This patch series extends the vmxnet3 driver to
leverage these new features.

Compatibility is maintained using existing vmxnet3 versioning mechanism as
follows:
 - new features added to vmxnet3 emulation are associated with new vmxnet3
   version viz. vmxnet3 version 3.
 - emulation advertises all the versions it supports to the driver.
 - during initialization, vmxnet3 driver picks the highest version number
 supported by both the emulation and the driver and configures emulation
 to run at that version.

In particular, following changes are introduced:

Patch 1:
  Some command definitions from previous vmxnet3 versions are
  missing. This patch adds those definitions before moving to vmxnet3
  version 3. It also fixes copyright info and maintained by.

Patch 2:
  This patch introduces generalized command interface which allows
  for easily adding new commands that vmxnet3 driver can issue to the
  emulation. Further patches in this series make use of this facility.

Patch 3:
  Transmit data ring buffer is used to copy packet headers or small
  packets. It is a fixed size buffer. This patch extends the driver to
  allow variable sized transmit data ring buffer.

Patch 4:
  This patch introduces receive data ring buffer - a set of small sized
  buffers that are always mapped by the emulation. This avoids memory
  mapping/unmapping overhead for small packets.

Patch 5:
  The vmxnet3 emulation supports a variety of coalescing modes. This patch
  extends vmxnet3 driver to allow querying and configuring these modes.

Patch 6:
  In vmxnet3 version 3, the emulation added support for the vmxnet3 driver
  to communicate information about the memory regions the driver will use
  for rx/tx buffers. This patch exposes related commands to the driver.

Patch 7:
  With all vmxnet3 version 3 changes incorporated in the vmxnet3 driver,
  with this patch, the driver can configure emulation to run at vmxnet3
  version 3.

Changes in v2:
 - v1 patch used special values of rx-usecs to differentiate between
 coalescing modes. v2 uses relevant fields in struct ethtool_coalesce
 to choose modes. Also, a new command VMXNET3_CMD_GET_COALESCE
 is introduced which allows driver to query the device for default
 coalescing configuration.

Changes in v3:
 - fix subject line to use vmxnet3: instead of Driver:Vmxnet3
 - resubmit when net-next is open

Changes in v4:
 - Address code review comments by Ben Hutchings: remove unnecessary memset
   from vmxnet3_get_coalesce.

Changes in v5:
 - Updated all the patches to add detailed commit messages.

---


Shrikrishna Khare (7):
  vmxnet3: prepare for version 3 changes
  vmxnet3: introduce generalized command interface to configure the
    device
  vmxnet3: allow variable length transmit data ring buffer
  vmxnet3: add receive data ring support
  vmxnet3: add support for get_coalesce, set_coalesce ethtool operations
  vmxnet3: introduce command to register memory region
  vmxnet3: update to version 3

 drivers/net/vmxnet3/Makefile          |   4 +-
 drivers/net/vmxnet3/upt1_defs.h       |   4 +-
 drivers/net/vmxnet3/vmxnet3_defs.h    | 105 ++++++++++++-
 drivers/net/vmxnet3/vmxnet3_drv.c     | 285 ++++++++++++++++++++++++++++------
 drivers/net/vmxnet3/vmxnet3_ethtool.c | 215 +++++++++++++++++++++++--
 drivers/net/vmxnet3/vmxnet3_int.h     |  54 ++++++-
 6 files changed, 585 insertions(+), 82 deletions(-)

-- 
2.8.2

[PATCH net-next v5 1/7] vmxnet3: prepare for version 3 changes

[Shrikrishna Khare]

vmxnet3 is currently at version 2, but some command definitions from
previous vmxnet3 versions are missing. Add those definitions before
moving to version 3.

Also, introduce utility macros for vmxnet3 version comparison and update
Copyright information and Maintained by.

Signed-off-by: Shrikrishna Khare <skhare@vmware.com>
---
 drivers/net/vmxnet3/Makefile          |  4 ++--
 drivers/net/vmxnet3/upt1_defs.h       |  4 ++--
 drivers/net/vmxnet3/vmxnet3_defs.h    |  9 ++++++---
 drivers/net/vmxnet3/vmxnet3_drv.c     | 22 +++++++++++++---------
 drivers/net/vmxnet3/vmxnet3_ethtool.c |  4 ++--
 drivers/net/vmxnet3/vmxnet3_int.h     | 13 +++++++++++--
 6 files changed, 36 insertions(+), 20 deletions(-)

diff --git a/drivers/net/vmxnet3/Makefile b/drivers/net/vmxnet3/Makefile
index 880f509..8cdbb63 100644
--- a/drivers/net/vmxnet3/Makefile
+++ b/drivers/net/vmxnet3/Makefile
@@ -2,7 +2,7 @@
 #
 # Linux driver for VMware's vmxnet3 ethernet NIC.
 #
-# Copyright (C) 2007-2009, VMware, Inc. All Rights Reserved.
+# Copyright (C) 2007-2016, VMware, Inc. All Rights Reserved.
 #
 # This program is free software; you can redistribute it and/or modify it
 # under the terms of the GNU General Public License as published by the
@@ -21,7 +21,7 @@
 # The full GNU General Public License is included in this distribution in
 # the file called "COPYING".
 #
-# Maintained by: Shreyas Bhatewara <pv-drivers@vmware.com>
+# Maintained by: pv-drivers@vmware.com
 #
 #
 ################################################################################
diff --git a/drivers/net/vmxnet3/upt1_defs.h b/drivers/net/vmxnet3/upt1_defs.h
index 969c751..db9f1fd 100644
--- a/drivers/net/vmxnet3/upt1_defs.h
+++ b/drivers/net/vmxnet3/upt1_defs.h
@@ -1,7 +1,7 @@
 /*
  * Linux driver for VMware's vmxnet3 ethernet NIC.
  *
- * Copyright (C) 2008-2009, VMware, Inc. All Rights Reserved.
+ * Copyright (C) 2008-2016, VMware, Inc. All Rights Reserved.
  *
  * This program is free software; you can redistribute it and/or modify it
  * under the terms of the GNU General Public License as published by the
@@ -20,7 +20,7 @@
  * The full GNU General Public License is included in this distribution in
  * the file called "COPYING".
  *
- * Maintained by: Shreyas Bhatewara <pv-drivers@vmware.com>
+ * Maintained by: pv-drivers@vmware.com
  *
  */
 
diff --git a/drivers/net/vmxnet3/vmxnet3_defs.h b/drivers/net/vmxnet3/vmxnet3_defs.h
index 72ba8ae..8345e0c 100644
--- a/drivers/net/vmxnet3/vmxnet3_defs.h
+++ b/drivers/net/vmxnet3/vmxnet3_defs.h
@@ -1,7 +1,7 @@
 /*
  * Linux driver for VMware's vmxnet3 ethernet NIC.
  *
- * Copyright (C) 2008-2015, VMware, Inc. All Rights Reserved.
+ * Copyright (C) 2008-2016, VMware, Inc. All Rights Reserved.
  *
  * This program is free software; you can redistribute it and/or modify it
  * under the terms of the GNU General Public License as published by the
@@ -20,7 +20,7 @@
  * The full GNU General Public License is included in this distribution in
  * the file called "COPYING".
  *
- * Maintained by: Shreyas Bhatewara <pv-drivers@vmware.com>
+ * Maintained by: pv-drivers@vmware.com
  *
  */
 
@@ -76,7 +76,9 @@ enum {
 	VMXNET3_CMD_UPDATE_IML,
 	VMXNET3_CMD_UPDATE_PMCFG,
 	VMXNET3_CMD_UPDATE_FEATURE,
+	VMXNET3_CMD_RESERVED1,
 	VMXNET3_CMD_LOAD_PLUGIN,
+	VMXNET3_CMD_RESERVED2,
 
 	VMXNET3_CMD_FIRST_GET = 0xF00D0000,
 	VMXNET3_CMD_GET_QUEUE_STATUS = VMXNET3_CMD_FIRST_GET,
@@ -87,7 +89,8 @@ enum {
 	VMXNET3_CMD_GET_DID_LO,
 	VMXNET3_CMD_GET_DID_HI,
 	VMXNET3_CMD_GET_DEV_EXTRA_INFO,
-	VMXNET3_CMD_GET_CONF_INTR
+	VMXNET3_CMD_GET_CONF_INTR,
+	VMXNET3_CMD_GET_RESERVED1,
 };
 
 /*
diff --git a/drivers/net/vmxnet3/vmxnet3_drv.c b/drivers/net/vmxnet3/vmxnet3_drv.c
index 08885bc..507c53d 100644
--- a/drivers/net/vmxnet3/vmxnet3_drv.c
+++ b/drivers/net/vmxnet3/vmxnet3_drv.c
@@ -1,7 +1,7 @@
 /*
  * Linux driver for VMware's vmxnet3 ethernet NIC.
  *
- * Copyright (C) 2008-2009, VMware, Inc. All Rights Reserved.
+ * Copyright (C) 2008-2016, VMware, Inc. All Rights Reserved.
  *
  * This program is free software; you can redistribute it and/or modify it
  * under the terms of the GNU General Public License as published by the
@@ -20,7 +20,7 @@
  * The full GNU General Public License is included in this distribution in
  * the file called "COPYING".
  *
- * Maintained by: Shreyas Bhatewara <pv-drivers@vmware.com>
+ * Maintained by: pv-drivers@vmware.com
  *
  */
 
@@ -1363,7 +1363,7 @@ vmxnet3_rq_rx_complete(struct vmxnet3_rx_queue *rq,
 			rbi->dma_addr = new_dma_addr;
 			rxd->addr = cpu_to_le64(rbi->dma_addr);
 			rxd->len = rbi->len;
-			if (adapter->version == 2 &&
+			if (VMXNET3_VERSION_GE_2(adapter) &&
 			    rcd->type == VMXNET3_CDTYPE_RXCOMP_LRO) {
 				struct Vmxnet3_RxCompDescExt *rcdlro;
 				rcdlro = (struct Vmxnet3_RxCompDescExt *)rcd;
@@ -3200,12 +3200,16 @@ vmxnet3_probe_device(struct pci_dev *pdev,
 		goto err_alloc_pci;
 
 	ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_VRRS);
-	if (ver & 2) {
-		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_VRRS, 2);
-		adapter->version = 2;
-	} else if (ver & 1) {
-		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_VRRS, 1);
-		adapter->version = 1;
+	if (ver & (1 << VMXNET3_REV_2)) {
+		VMXNET3_WRITE_BAR1_REG(adapter,
+				       VMXNET3_REG_VRRS,
+				       1 << VMXNET3_REV_2);
+		adapter->version = VMXNET3_REV_2 + 1;
+	} else if (ver & (1 << VMXNET3_REV_1)) {
+		VMXNET3_WRITE_BAR1_REG(adapter,
+				       VMXNET3_REG_VRRS,
+				       1 << VMXNET3_REV_1);
+		adapter->version = VMXNET3_REV_1 + 1;
 	} else {
 		dev_err(&pdev->dev,
 			"Incompatible h/w version (0x%x) for adapter\n", ver);
diff --git a/drivers/net/vmxnet3/vmxnet3_ethtool.c b/drivers/net/vmxnet3/vmxnet3_ethtool.c
index 9ba11d7..163e99c 100644
--- a/drivers/net/vmxnet3/vmxnet3_ethtool.c
+++ b/drivers/net/vmxnet3/vmxnet3_ethtool.c
@@ -1,7 +1,7 @@
 /*
  * Linux driver for VMware's vmxnet3 ethernet NIC.
  *
- * Copyright (C) 2008-2009, VMware, Inc. All Rights Reserved.
+ * Copyright (C) 2008-2016, VMware, Inc. All Rights Reserved.
  *
  * This program is free software; you can redistribute it and/or modify it
  * under the terms of the GNU General Public License as published by the
@@ -20,7 +20,7 @@
  * The full GNU General Public License is included in this distribution in
  * the file called "COPYING".
  *
- * Maintained by: Shreyas Bhatewara <pv-drivers@vmware.com>
+ * Maintained by: pv-drivers@vmware.com
  *
  */
 
diff --git a/drivers/net/vmxnet3/vmxnet3_int.h b/drivers/net/vmxnet3/vmxnet3_int.h
index 3d2b64e..de068e9 100644
--- a/drivers/net/vmxnet3/vmxnet3_int.h
+++ b/drivers/net/vmxnet3/vmxnet3_int.h
@@ -1,7 +1,7 @@
 /*
  * Linux driver for VMware's vmxnet3 ethernet NIC.
  *
- * Copyright (C) 2008-2009, VMware, Inc. All Rights Reserved.
+ * Copyright (C) 2008-2016, VMware, Inc. All Rights Reserved.
  *
  * This program is free software; you can redistribute it and/or modify it
  * under the terms of the GNU General Public License as published by the
@@ -20,7 +20,7 @@
  * The full GNU General Public License is included in this distribution in
  * the file called "COPYING".
  *
- * Maintained by: Shreyas Bhatewara <pv-drivers@vmware.com>
+ * Maintained by: pv-drivers@vmware.com
  *
  */
 
@@ -79,6 +79,10 @@
 	#define VMXNET3_RSS
 #endif
 
+#define VMXNET3_REV_3		2	/* Vmxnet3 Rev. 3 */
+#define VMXNET3_REV_2		1	/* Vmxnet3 Rev. 2 */
+#define VMXNET3_REV_1		0	/* Vmxnet3 Rev. 1 */
+
 /*
  * Capabilities
  */
@@ -387,6 +391,11 @@ struct vmxnet3_adapter {
 #define VMXNET3_GET_ADDR_LO(dma)   ((u32)(dma))
 #define VMXNET3_GET_ADDR_HI(dma)   ((u32)(((u64)(dma)) >> 32))
 
+#define VMXNET3_VERSION_GE_2(adapter) \
+	(adapter->version >= VMXNET3_REV_2 + 1)
+#define VMXNET3_VERSION_GE_3(adapter) \
+	(adapter->version >= VMXNET3_REV_3 + 1)
+
 /* must be a multiple of VMXNET3_RING_SIZE_ALIGN */
 #define VMXNET3_DEF_TX_RING_SIZE    512
 #define VMXNET3_DEF_RX_RING_SIZE    256
-- 
2.8.2

[PATCH net-next v5 2/7] vmxnet3: introduce generalized command interface to configure the device

[Shrikrishna Khare]

Shared memory is used to exchange information between the vmxnet3 driver
and the emulation. In order to request emulation to perform a task, the
driver first populates specific fields in this shared memory and then
issues corresponding command by writing to the command register(CMD). The
layout of the shared memory was defined by vmxnet3 version 1 and cannot
be extended for every new command without breaking backward compatibility.

To address this problem, in vmxnet3 version 3, the emulation repurposed
a reserved field in the shared memory to represent command information
instead. For new commands, the driver first populates the command
information field in the shared memory and then issues the command. The
emulation interprets the data written to the command information depending
on the type of the command. This patch exposes this capability to the driver.

Signed-off-by: Guolin Yang <gyang@vmware.com>
Signed-off-by: Shrikrishna Khare <skhare@vmware.com>
---
 drivers/net/vmxnet3/vmxnet3_defs.h | 22 +++++++++++++++++++++-
 1 file changed, 21 insertions(+), 1 deletion(-)

diff --git a/drivers/net/vmxnet3/vmxnet3_defs.h b/drivers/net/vmxnet3/vmxnet3_defs.h
index 8345e0c..a26a69d 100644
--- a/drivers/net/vmxnet3/vmxnet3_defs.h
+++ b/drivers/net/vmxnet3/vmxnet3_defs.h
@@ -79,6 +79,7 @@ enum {
 	VMXNET3_CMD_RESERVED1,
 	VMXNET3_CMD_LOAD_PLUGIN,
 	VMXNET3_CMD_RESERVED2,
+	VMXNET3_CMD_RESERVED3,
 
 	VMXNET3_CMD_FIRST_GET = 0xF00D0000,
 	VMXNET3_CMD_GET_QUEUE_STATUS = VMXNET3_CMD_FIRST_GET,
@@ -612,6 +613,18 @@ struct Vmxnet3_RxQueueDesc {
 	u8				      __pad[88]; /* 128 aligned */
 };
 
+struct Vmxnet3_SetPolling {
+	u8					enablePolling;
+};
+
+/* If the command data <= 16 bytes, use the shared memory directly.
+ * otherwise, use variable length configuration descriptor.
+ */
+union Vmxnet3_CmdInfo {
+	struct Vmxnet3_VariableLenConfDesc	varConf;
+	struct Vmxnet3_SetPolling		setPolling;
+	__le64					data[2];
+};
 
 struct Vmxnet3_DSDevRead {
 	/* read-only region for device, read by dev in response to a SET cmd */
@@ -630,7 +643,14 @@ struct Vmxnet3_DriverShared {
 	__le32				pad;
 	struct Vmxnet3_DSDevRead	devRead;
 	__le32				ecr;
-	__le32				reserved[5];
+	__le32				reserved;
+	union {
+		__le32			reserved1[4];
+		union Vmxnet3_CmdInfo	cmdInfo; /* only valid in the context of
+						  * executing the relevant
+						  * command
+						  */
+	} cu;
 };
 
 
-- 
2.8.2

[PATCH net-next v5 3/7] vmxnet3: allow variable length transmit data ring buffer

[Shrikrishna Khare]

vmxnet3 driver supports transmit data ring viz. a set of fixed size
buffers used by the driver to copy packet headers. Small packets that
fit these buffers are copied into these buffers entirely.

Currently this buffer size of fixed at 128 bytes. This patch extends
transmit data ring implementation to allow variable length transmit
data ring buffers. The length of the buffer is read from the emulation
during initialization.

Signed-off-by: Sriram Rangarajan <rangarajans@vmware.com>
Signed-off-by: Shrikrishna Khare <skhare@vmware.com>
---
 drivers/net/vmxnet3/vmxnet3_defs.h    | 12 +++++++-
 drivers/net/vmxnet3/vmxnet3_drv.c     | 55 ++++++++++++++++++++++++++---------
 drivers/net/vmxnet3/vmxnet3_ethtool.c |  9 +++---
 drivers/net/vmxnet3/vmxnet3_int.h     |  7 ++++-
 4 files changed, 64 insertions(+), 19 deletions(-)

diff --git a/drivers/net/vmxnet3/vmxnet3_defs.h b/drivers/net/vmxnet3/vmxnet3_defs.h
index a26a69d..701d989 100644
--- a/drivers/net/vmxnet3/vmxnet3_defs.h
+++ b/drivers/net/vmxnet3/vmxnet3_defs.h
@@ -92,6 +92,7 @@ enum {
 	VMXNET3_CMD_GET_DEV_EXTRA_INFO,
 	VMXNET3_CMD_GET_CONF_INTR,
 	VMXNET3_CMD_GET_RESERVED1,
+	VMXNET3_CMD_GET_TXDATA_DESC_SIZE
 };
 
 /*
@@ -377,6 +378,10 @@ union Vmxnet3_GenericDesc {
 #define VMXNET3_RING_SIZE_ALIGN 32
 #define VMXNET3_RING_SIZE_MASK  (VMXNET3_RING_SIZE_ALIGN - 1)
 
+/* Tx Data Ring buffer size must be a multiple of 64 */
+#define VMXNET3_TXDATA_DESC_SIZE_ALIGN 64
+#define VMXNET3_TXDATA_DESC_SIZE_MASK  (VMXNET3_TXDATA_DESC_SIZE_ALIGN - 1)
+
 /* Max ring size */
 #define VMXNET3_TX_RING_MAX_SIZE   4096
 #define VMXNET3_TC_RING_MAX_SIZE   4096
@@ -384,6 +389,9 @@ union Vmxnet3_GenericDesc {
 #define VMXNET3_RX_RING2_MAX_SIZE  4096
 #define VMXNET3_RC_RING_MAX_SIZE   8192
 
+#define VMXNET3_TXDATA_DESC_MIN_SIZE 128
+#define VMXNET3_TXDATA_DESC_MAX_SIZE 2048
+
 /* a list of reasons for queue stop */
 
 enum {
@@ -470,7 +478,9 @@ struct Vmxnet3_TxQueueConf {
 	__le32		compRingSize; /* # of comp desc */
 	__le32		ddLen;        /* size of driver data */
 	u8		intrIdx;
-	u8		_pad[7];
+	u8		_pad1[1];
+	__le16		txDataRingDescSize;
+	u8		_pad2[4];
 };
 
 
diff --git a/drivers/net/vmxnet3/vmxnet3_drv.c b/drivers/net/vmxnet3/vmxnet3_drv.c
index 507c53d..4e42eb0 100644
--- a/drivers/net/vmxnet3/vmxnet3_drv.c
+++ b/drivers/net/vmxnet3/vmxnet3_drv.c
@@ -435,8 +435,8 @@ vmxnet3_tq_destroy(struct vmxnet3_tx_queue *tq,
 		tq->tx_ring.base = NULL;
 	}
 	if (tq->data_ring.base) {
-		dma_free_coherent(&adapter->pdev->dev, tq->data_ring.size *
-				  sizeof(struct Vmxnet3_TxDataDesc),
+		dma_free_coherent(&adapter->pdev->dev,
+				  tq->data_ring.size * tq->txdata_desc_size,
 				  tq->data_ring.base, tq->data_ring.basePA);
 		tq->data_ring.base = NULL;
 	}
@@ -478,8 +478,8 @@ vmxnet3_tq_init(struct vmxnet3_tx_queue *tq,
 	tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
 	tq->tx_ring.gen = VMXNET3_INIT_GEN;
 
-	memset(tq->data_ring.base, 0, tq->data_ring.size *
-	       sizeof(struct Vmxnet3_TxDataDesc));
+	memset(tq->data_ring.base, 0,
+	       tq->data_ring.size * tq->txdata_desc_size);
 
 	/* reset the tx comp ring contents to 0 and reset comp ring states */
 	memset(tq->comp_ring.base, 0, tq->comp_ring.size *
@@ -514,10 +514,10 @@ vmxnet3_tq_create(struct vmxnet3_tx_queue *tq,
 	}
 
 	tq->data_ring.base = dma_alloc_coherent(&adapter->pdev->dev,
-			tq->data_ring.size * sizeof(struct Vmxnet3_TxDataDesc),
+			tq->data_ring.size * tq->txdata_desc_size,
 			&tq->data_ring.basePA, GFP_KERNEL);
 	if (!tq->data_ring.base) {
-		netdev_err(adapter->netdev, "failed to allocate data ring\n");
+		netdev_err(adapter->netdev, "failed to allocate tx data ring\n");
 		goto err;
 	}
 
@@ -689,7 +689,7 @@ vmxnet3_map_pkt(struct sk_buff *skb, struct vmxnet3_tx_ctx *ctx,
 	if (ctx->copy_size) {
 		ctx->sop_txd->txd.addr = cpu_to_le64(tq->data_ring.basePA +
 					tq->tx_ring.next2fill *
-					sizeof(struct Vmxnet3_TxDataDesc));
+					tq->txdata_desc_size);
 		ctx->sop_txd->dword[2] = cpu_to_le32(dw2 | ctx->copy_size);
 		ctx->sop_txd->dword[3] = 0;
 
@@ -873,8 +873,9 @@ vmxnet3_parse_hdr(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
 			ctx->eth_ip_hdr_size = 0;
 			ctx->l4_hdr_size = 0;
 			/* copy as much as allowed */
-			ctx->copy_size = min((unsigned int)VMXNET3_HDR_COPY_SIZE
-					     , skb_headlen(skb));
+			ctx->copy_size = min_t(unsigned int,
+					       tq->txdata_desc_size,
+					       skb_headlen(skb));
 		}
 
 		if (skb->len <= VMXNET3_HDR_COPY_SIZE)
@@ -885,7 +886,7 @@ vmxnet3_parse_hdr(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
 			goto err;
 	}
 
-	if (unlikely(ctx->copy_size > VMXNET3_HDR_COPY_SIZE)) {
+	if (unlikely(ctx->copy_size > tq->txdata_desc_size)) {
 		tq->stats.oversized_hdr++;
 		ctx->copy_size = 0;
 		return 0;
@@ -2336,6 +2337,7 @@ vmxnet3_setup_driver_shared(struct vmxnet3_adapter *adapter)
 		tqc->ddPA           = cpu_to_le64(tq->buf_info_pa);
 		tqc->txRingSize     = cpu_to_le32(tq->tx_ring.size);
 		tqc->dataRingSize   = cpu_to_le32(tq->data_ring.size);
+		tqc->txDataRingDescSize = cpu_to_le32(tq->txdata_desc_size);
 		tqc->compRingSize   = cpu_to_le32(tq->comp_ring.size);
 		tqc->ddLen          = cpu_to_le32(
 					sizeof(struct vmxnet3_tx_buf_info) *
@@ -2689,7 +2691,8 @@ vmxnet3_adjust_rx_ring_size(struct vmxnet3_adapter *adapter)
 
 int
 vmxnet3_create_queues(struct vmxnet3_adapter *adapter, u32 tx_ring_size,
-		      u32 rx_ring_size, u32 rx_ring2_size)
+		      u32 rx_ring_size, u32 rx_ring2_size,
+		      u16 txdata_desc_size)
 {
 	int err = 0, i;
 
@@ -2698,6 +2701,7 @@ vmxnet3_create_queues(struct vmxnet3_adapter *adapter, u32 tx_ring_size,
 		tq->tx_ring.size   = tx_ring_size;
 		tq->data_ring.size = tx_ring_size;
 		tq->comp_ring.size = tx_ring_size;
+		tq->txdata_desc_size = txdata_desc_size;
 		tq->shared = &adapter->tqd_start[i].ctrl;
 		tq->stopped = true;
 		tq->adapter = adapter;
@@ -2754,9 +2758,34 @@ vmxnet3_open(struct net_device *netdev)
 	for (i = 0; i < adapter->num_tx_queues; i++)
 		spin_lock_init(&adapter->tx_queue[i].tx_lock);
 
-	err = vmxnet3_create_queues(adapter, adapter->tx_ring_size,
+	if (VMXNET3_VERSION_GE_3(adapter)) {
+		unsigned long flags;
+		u16 txdata_desc_size;
+
+		spin_lock_irqsave(&adapter->cmd_lock, flags);
+		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
+				       VMXNET3_CMD_GET_TXDATA_DESC_SIZE);
+		txdata_desc_size = VMXNET3_READ_BAR1_REG(adapter,
+							 VMXNET3_REG_CMD);
+		spin_unlock_irqrestore(&adapter->cmd_lock, flags);
+
+		if ((txdata_desc_size < VMXNET3_TXDATA_DESC_MIN_SIZE) ||
+		    (txdata_desc_size > VMXNET3_TXDATA_DESC_MAX_SIZE) ||
+		    (txdata_desc_size & VMXNET3_TXDATA_DESC_SIZE_MASK)) {
+			adapter->txdata_desc_size =
+				sizeof(struct Vmxnet3_TxDataDesc);
+		} else {
+			adapter->txdata_desc_size = txdata_desc_size;
+		}
+	} else {
+		adapter->txdata_desc_size = sizeof(struct Vmxnet3_TxDataDesc);
+	}
+
+	err = vmxnet3_create_queues(adapter,
+				    adapter->tx_ring_size,
 				    adapter->rx_ring_size,
-				    adapter->rx_ring2_size);
+				    adapter->rx_ring2_size,
+				    adapter->txdata_desc_size);
 	if (err)
 		goto queue_err;
 
diff --git a/drivers/net/vmxnet3/vmxnet3_ethtool.c b/drivers/net/vmxnet3/vmxnet3_ethtool.c
index 163e99c..3b70cfe 100644
--- a/drivers/net/vmxnet3/vmxnet3_ethtool.c
+++ b/drivers/net/vmxnet3/vmxnet3_ethtool.c
@@ -396,8 +396,7 @@ vmxnet3_get_regs(struct net_device *netdev, struct ethtool_regs *regs, void *p)
 		buf[j++] = VMXNET3_GET_ADDR_LO(tq->data_ring.basePA);
 		buf[j++] = VMXNET3_GET_ADDR_HI(tq->data_ring.basePA);
 		buf[j++] = tq->data_ring.size;
-		/* transmit data ring buffer size */
-		buf[j++] = VMXNET3_HDR_COPY_SIZE;
+		buf[j++] = tq->txdata_desc_size;
 
 		buf[j++] = VMXNET3_GET_ADDR_LO(tq->comp_ring.basePA);
 		buf[j++] = VMXNET3_GET_ADDR_HI(tq->comp_ring.basePA);
@@ -591,7 +590,8 @@ vmxnet3_set_ringparam(struct net_device *netdev,
 		vmxnet3_rq_destroy_all(adapter);
 
 		err = vmxnet3_create_queues(adapter, new_tx_ring_size,
-			new_rx_ring_size, new_rx_ring2_size);
+					    new_rx_ring_size, new_rx_ring2_size,
+					    adapter->txdata_desc_size);
 
 		if (err) {
 			/* failed, most likely because of OOM, try default
@@ -604,7 +604,8 @@ vmxnet3_set_ringparam(struct net_device *netdev,
 			err = vmxnet3_create_queues(adapter,
 						    new_tx_ring_size,
 						    new_rx_ring_size,
-						    new_rx_ring2_size);
+						    new_rx_ring2_size,
+						    adapter->txdata_desc_size);
 			if (err) {
 				netdev_err(netdev, "failed to create queues "
 					   "with default sizes. Closing it\n");
diff --git a/drivers/net/vmxnet3/vmxnet3_int.h b/drivers/net/vmxnet3/vmxnet3_int.h
index de068e9..94010de 100644
--- a/drivers/net/vmxnet3/vmxnet3_int.h
+++ b/drivers/net/vmxnet3/vmxnet3_int.h
@@ -241,6 +241,7 @@ struct vmxnet3_tx_queue {
 	int                             num_stop;  /* # of times the queue is
 						    * stopped */
 	int				qid;
+	u16				txdata_desc_size;
 } __attribute__((__aligned__(SMP_CACHE_BYTES)));
 
 enum vmxnet3_rx_buf_type {
@@ -363,6 +364,9 @@ struct vmxnet3_adapter {
 	u32 rx_ring_size;
 	u32 rx_ring2_size;
 
+	/* Size of buffer in the data ring */
+	u16 txdata_desc_size;
+
 	struct work_struct work;
 
 	unsigned long  state;    /* VMXNET3_STATE_BIT_xxx */
@@ -427,7 +431,8 @@ vmxnet3_set_features(struct net_device *netdev, netdev_features_t features);
 
 int
 vmxnet3_create_queues(struct vmxnet3_adapter *adapter,
-		      u32 tx_ring_size, u32 rx_ring_size, u32 rx_ring2_size);
+		      u32 tx_ring_size, u32 rx_ring_size, u32 rx_ring2_size,
+		      u16 txdata_desc_size);
 
 void vmxnet3_set_ethtool_ops(struct net_device *netdev);
 
-- 
2.8.2

[PATCH net-next v5 4/7] vmxnet3: add receive data ring support

[Shrikrishna Khare]

vmxnet3 driver preallocates buffers for receiving packets and posts the
buffers to the emulation. In order to deliver a received packet to the
guest, the emulation must map buffer(s) and copy the packet into it.

To avoid this memory mapping overhead, this patch introduces the receive
data ring - a set of small sized buffers that are always mapped by
the emulation. If a packet fits into the receive data ring buffer, the
emulation delivers the packet via the receive data ring (which must be
copied by the guest driver), or else the usual receive path is used.

Receive Data Ring buffer length is configurable via ethtool -G ethX rx-mini

Signed-off-by: Shrikrishna Khare <skhare@vmware.com>
---
 drivers/net/vmxnet3/vmxnet3_defs.h    |  14 +++-
 drivers/net/vmxnet3/vmxnet3_drv.c     | 153 +++++++++++++++++++++++++++-------
 drivers/net/vmxnet3/vmxnet3_ethtool.c |  48 ++++++++---
 drivers/net/vmxnet3/vmxnet3_int.h     |  23 ++++-
 4 files changed, 193 insertions(+), 45 deletions(-)

diff --git a/drivers/net/vmxnet3/vmxnet3_defs.h b/drivers/net/vmxnet3/vmxnet3_defs.h
index 701d989..f3b31c2 100644
--- a/drivers/net/vmxnet3/vmxnet3_defs.h
+++ b/drivers/net/vmxnet3/vmxnet3_defs.h
@@ -174,6 +174,8 @@ struct Vmxnet3_TxDataDesc {
 	u8		data[VMXNET3_HDR_COPY_SIZE];
 };
 
+typedef u8 Vmxnet3_RxDataDesc;
+
 #define VMXNET3_TCD_GEN_SHIFT	31
 #define VMXNET3_TCD_GEN_SIZE	1
 #define VMXNET3_TCD_TXIDX_SHIFT	0
@@ -382,6 +384,10 @@ union Vmxnet3_GenericDesc {
 #define VMXNET3_TXDATA_DESC_SIZE_ALIGN 64
 #define VMXNET3_TXDATA_DESC_SIZE_MASK  (VMXNET3_TXDATA_DESC_SIZE_ALIGN - 1)
 
+/* Rx Data Ring buffer size must be a multiple of 64 */
+#define VMXNET3_RXDATA_DESC_SIZE_ALIGN 64
+#define VMXNET3_RXDATA_DESC_SIZE_MASK  (VMXNET3_RXDATA_DESC_SIZE_ALIGN - 1)
+
 /* Max ring size */
 #define VMXNET3_TX_RING_MAX_SIZE   4096
 #define VMXNET3_TC_RING_MAX_SIZE   4096
@@ -392,6 +398,8 @@ union Vmxnet3_GenericDesc {
 #define VMXNET3_TXDATA_DESC_MIN_SIZE 128
 #define VMXNET3_TXDATA_DESC_MAX_SIZE 2048
 
+#define VMXNET3_RXDATA_DESC_MAX_SIZE 2048
+
 /* a list of reasons for queue stop */
 
 enum {
@@ -488,12 +496,14 @@ struct Vmxnet3_RxQueueConf {
 	__le64		rxRingBasePA[2];
 	__le64		compRingBasePA;
 	__le64		ddPA;            /* driver data */
-	__le64		reserved;
+	__le64		rxDataRingBasePA;
 	__le32		rxRingSize[2];   /* # of rx desc */
 	__le32		compRingSize;    /* # of rx comp desc */
 	__le32		ddLen;           /* size of driver data */
 	u8		intrIdx;
-	u8		_pad[7];
+	u8		_pad1[1];
+	__le16		rxDataRingDescSize;  /* size of rx data ring buffer */
+	u8		_pad2[4];
 };
 
 
diff --git a/drivers/net/vmxnet3/vmxnet3_drv.c b/drivers/net/vmxnet3/vmxnet3_drv.c
index 4e42eb0..6449d2e 100644
--- a/drivers/net/vmxnet3/vmxnet3_drv.c
+++ b/drivers/net/vmxnet3/vmxnet3_drv.c
@@ -1284,9 +1284,10 @@ vmxnet3_rq_rx_complete(struct vmxnet3_rx_queue *rq,
 			 */
 			break;
 		}
-		BUG_ON(rcd->rqID != rq->qid && rcd->rqID != rq->qid2);
+		BUG_ON(rcd->rqID != rq->qid && rcd->rqID != rq->qid2 &&
+		       rcd->rqID != rq->dataRingQid);
 		idx = rcd->rxdIdx;
-		ring_idx = rcd->rqID < adapter->num_rx_queues ? 0 : 1;
+		ring_idx = VMXNET3_GET_RING_IDX(adapter, rcd->rqID);
 		ring = rq->rx_ring + ring_idx;
 		vmxnet3_getRxDesc(rxd, &rq->rx_ring[ring_idx].base[idx].rxd,
 				  &rxCmdDesc);
@@ -1301,8 +1302,12 @@ vmxnet3_rq_rx_complete(struct vmxnet3_rx_queue *rq,
 		}
 
 		if (rcd->sop) { /* first buf of the pkt */
+			bool rxDataRingUsed;
+			u16 len;
+
 			BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_HEAD ||
-			       rcd->rqID != rq->qid);
+			       (rcd->rqID != rq->qid &&
+				rcd->rqID != rq->dataRingQid));
 
 			BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_SKB);
 			BUG_ON(ctx->skb != NULL || rbi->skb == NULL);
@@ -1318,8 +1323,12 @@ vmxnet3_rq_rx_complete(struct vmxnet3_rx_queue *rq,
 
 			skip_page_frags = false;
 			ctx->skb = rbi->skb;
+
+			rxDataRingUsed =
+				VMXNET3_RX_DATA_RING(adapter, rcd->rqID);
+			len = rxDataRingUsed ? rcd->len : rbi->len;
 			new_skb = netdev_alloc_skb_ip_align(adapter->netdev,
-							    rbi->len);
+							    len);
 			if (new_skb == NULL) {
 				/* Skb allocation failed, do not handover this
 				 * skb to stack. Reuse it. Drop the existing pkt
@@ -1330,25 +1339,48 @@ vmxnet3_rq_rx_complete(struct vmxnet3_rx_queue *rq,
 				skip_page_frags = true;
 				goto rcd_done;
 			}
-			new_dma_addr = dma_map_single(&adapter->pdev->dev,
-						      new_skb->data, rbi->len,
-						      PCI_DMA_FROMDEVICE);
-			if (dma_mapping_error(&adapter->pdev->dev,
-					      new_dma_addr)) {
-				dev_kfree_skb(new_skb);
-				/* Skb allocation failed, do not handover this
-				 * skb to stack. Reuse it. Drop the existing pkt
-				 */
-				rq->stats.rx_buf_alloc_failure++;
-				ctx->skb = NULL;
-				rq->stats.drop_total++;
-				skip_page_frags = true;
-				goto rcd_done;
-			}
 
-			dma_unmap_single(&adapter->pdev->dev, rbi->dma_addr,
-					 rbi->len,
-					 PCI_DMA_FROMDEVICE);
+			if (rxDataRingUsed) {
+				size_t sz;
+
+				BUG_ON(rcd->len > rq->data_ring.desc_size);
+
+				ctx->skb = new_skb;
+				sz = rcd->rxdIdx * rq->data_ring.desc_size;
+				memcpy(new_skb->data,
+				       &rq->data_ring.base[sz], rcd->len);
+			} else {
+				ctx->skb = rbi->skb;
+
+				new_dma_addr =
+					dma_map_single(&adapter->pdev->dev,
+						       new_skb->data, rbi->len,
+						       PCI_DMA_FROMDEVICE);
+				if (dma_mapping_error(&adapter->pdev->dev,
+						      new_dma_addr)) {
+					dev_kfree_skb(new_skb);
+					/* Skb allocation failed, do not
+					 * handover this skb to stack. Reuse
+					 * it. Drop the existing pkt.
+					 */
+					rq->stats.rx_buf_alloc_failure++;
+					ctx->skb = NULL;
+					rq->stats.drop_total++;
+					skip_page_frags = true;
+					goto rcd_done;
+				}
+
+				dma_unmap_single(&adapter->pdev->dev,
+						 rbi->dma_addr,
+						 rbi->len,
+						 PCI_DMA_FROMDEVICE);
+
+				/* Immediate refill */
+				rbi->skb = new_skb;
+				rbi->dma_addr = new_dma_addr;
+				rxd->addr = cpu_to_le64(rbi->dma_addr);
+				rxd->len = rbi->len;
+			}
 
 #ifdef VMXNET3_RSS
 			if (rcd->rssType != VMXNET3_RCD_RSS_TYPE_NONE &&
@@ -1359,11 +1391,6 @@ vmxnet3_rq_rx_complete(struct vmxnet3_rx_queue *rq,
 #endif
 			skb_put(ctx->skb, rcd->len);
 
-			/* Immediate refill */
-			rbi->skb = new_skb;
-			rbi->dma_addr = new_dma_addr;
-			rxd->addr = cpu_to_le64(rbi->dma_addr);
-			rxd->len = rbi->len;
 			if (VMXNET3_VERSION_GE_2(adapter) &&
 			    rcd->type == VMXNET3_CDTYPE_RXCOMP_LRO) {
 				struct Vmxnet3_RxCompDescExt *rcdlro;
@@ -1590,6 +1617,13 @@ static void vmxnet3_rq_destroy(struct vmxnet3_rx_queue *rq,
 		rq->buf_info[i] = NULL;
 	}
 
+	if (rq->data_ring.base) {
+		dma_free_coherent(&adapter->pdev->dev,
+				  rq->rx_ring[0].size * rq->data_ring.desc_size,
+				  rq->data_ring.base, rq->data_ring.basePA);
+		rq->data_ring.base = NULL;
+	}
+
 	if (rq->comp_ring.base) {
 		dma_free_coherent(&adapter->pdev->dev, rq->comp_ring.size
 				  * sizeof(struct Vmxnet3_RxCompDesc),
@@ -1605,6 +1639,25 @@ static void vmxnet3_rq_destroy(struct vmxnet3_rx_queue *rq,
 	}
 }
 
+void
+vmxnet3_rq_destroy_all_rxdataring(struct vmxnet3_adapter *adapter)
+{
+	int i;
+
+	for (i = 0; i < adapter->num_rx_queues; i++) {
+		struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
+
+		if (rq->data_ring.base) {
+			dma_free_coherent(&adapter->pdev->dev,
+					  (rq->rx_ring[0].size *
+					  rq->data_ring.desc_size),
+					  rq->data_ring.base,
+					  rq->data_ring.basePA);
+			rq->data_ring.base = NULL;
+			rq->data_ring.desc_size = 0;
+		}
+	}
+}
 
 static int
 vmxnet3_rq_init(struct vmxnet3_rx_queue *rq,
@@ -1698,6 +1751,22 @@ vmxnet3_rq_create(struct vmxnet3_rx_queue *rq, struct vmxnet3_adapter *adapter)
 		}
 	}
 
+	if ((adapter->rxdataring_enabled) && (rq->data_ring.desc_size != 0)) {
+		sz = rq->rx_ring[0].size * rq->data_ring.desc_size;
+		rq->data_ring.base =
+			dma_alloc_coherent(&adapter->pdev->dev, sz,
+					   &rq->data_ring.basePA,
+					   GFP_KERNEL);
+		if (!rq->data_ring.base) {
+			netdev_err(adapter->netdev,
+				   "rx data ring will be disabled\n");
+			adapter->rxdataring_enabled = false;
+		}
+	} else {
+		rq->data_ring.base = NULL;
+		rq->data_ring.desc_size = 0;
+	}
+
 	sz = rq->comp_ring.size * sizeof(struct Vmxnet3_RxCompDesc);
 	rq->comp_ring.base = dma_alloc_coherent(&adapter->pdev->dev, sz,
 						&rq->comp_ring.basePA,
@@ -1730,6 +1799,8 @@ vmxnet3_rq_create_all(struct vmxnet3_adapter *adapter)
 {
 	int i, err = 0;
 
+	adapter->rxdataring_enabled = VMXNET3_VERSION_GE_3(adapter);
+
 	for (i = 0; i < adapter->num_rx_queues; i++) {
 		err = vmxnet3_rq_create(&adapter->rx_queue[i], adapter);
 		if (unlikely(err)) {
@@ -1739,6 +1810,10 @@ vmxnet3_rq_create_all(struct vmxnet3_adapter *adapter)
 			goto err_out;
 		}
 	}
+
+	if (!adapter->rxdataring_enabled)
+		vmxnet3_rq_destroy_all_rxdataring(adapter);
+
 	return err;
 err_out:
 	vmxnet3_rq_destroy_all(adapter);
@@ -2046,10 +2121,9 @@ vmxnet3_request_irqs(struct vmxnet3_adapter *adapter)
 			struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
 			rq->qid = i;
 			rq->qid2 = i + adapter->num_rx_queues;
+			rq->dataRingQid = i + 2 * adapter->num_rx_queues;
 		}
 
-
-
 		/* init our intr settings */
 		for (i = 0; i < intr->num_intrs; i++)
 			intr->mod_levels[i] = UPT1_IML_ADAPTIVE;
@@ -2362,6 +2436,12 @@ vmxnet3_setup_driver_shared(struct vmxnet3_adapter *adapter)
 					(rqc->rxRingSize[0] +
 					 rqc->rxRingSize[1]));
 		rqc->intrIdx         = rq->comp_ring.intr_idx;
+		if (VMXNET3_VERSION_GE_3(adapter)) {
+			rqc->rxDataRingBasePA =
+				cpu_to_le64(rq->data_ring.basePA);
+			rqc->rxDataRingDescSize =
+				cpu_to_le16(rq->data_ring.desc_size);
+		}
 	}
 
 #ifdef VMXNET3_RSS
@@ -2692,7 +2772,7 @@ vmxnet3_adjust_rx_ring_size(struct vmxnet3_adapter *adapter)
 int
 vmxnet3_create_queues(struct vmxnet3_adapter *adapter, u32 tx_ring_size,
 		      u32 rx_ring_size, u32 rx_ring2_size,
-		      u16 txdata_desc_size)
+		      u16 txdata_desc_size, u16 rxdata_desc_size)
 {
 	int err = 0, i;
 
@@ -2718,12 +2798,15 @@ vmxnet3_create_queues(struct vmxnet3_adapter *adapter, u32 tx_ring_size,
 	adapter->rx_queue[0].rx_ring[0].size = rx_ring_size;
 	adapter->rx_queue[0].rx_ring[1].size = rx_ring2_size;
 	vmxnet3_adjust_rx_ring_size(adapter);
+
+	adapter->rxdataring_enabled = VMXNET3_VERSION_GE_3(adapter);
 	for (i = 0; i < adapter->num_rx_queues; i++) {
 		struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
 		/* qid and qid2 for rx queues will be assigned later when num
 		 * of rx queues is finalized after allocating intrs */
 		rq->shared = &adapter->rqd_start[i].ctrl;
 		rq->adapter = adapter;
+		rq->data_ring.desc_size = rxdata_desc_size;
 		err = vmxnet3_rq_create(rq, adapter);
 		if (err) {
 			if (i == 0) {
@@ -2741,6 +2824,10 @@ vmxnet3_create_queues(struct vmxnet3_adapter *adapter, u32 tx_ring_size,
 			}
 		}
 	}
+
+	if (!adapter->rxdataring_enabled)
+		vmxnet3_rq_destroy_all_rxdataring(adapter);
+
 	return err;
 queue_err:
 	vmxnet3_tq_destroy_all(adapter);
@@ -2785,7 +2872,8 @@ vmxnet3_open(struct net_device *netdev)
 				    adapter->tx_ring_size,
 				    adapter->rx_ring_size,
 				    adapter->rx_ring2_size,
-				    adapter->txdata_desc_size);
+				    adapter->txdata_desc_size,
+				    adapter->rxdata_desc_size);
 	if (err)
 		goto queue_err;
 
@@ -3260,6 +3348,9 @@ vmxnet3_probe_device(struct pci_dev *pdev,
 	SET_NETDEV_DEV(netdev, &pdev->dev);
 	vmxnet3_declare_features(adapter, dma64);
 
+	adapter->rxdata_desc_size = VMXNET3_VERSION_GE_3(adapter) ?
+		VMXNET3_DEF_RXDATA_DESC_SIZE : 0;
+
 	if (adapter->num_tx_queues == adapter->num_rx_queues)
 		adapter->share_intr = VMXNET3_INTR_BUDDYSHARE;
 	else
diff --git a/drivers/net/vmxnet3/vmxnet3_ethtool.c b/drivers/net/vmxnet3/vmxnet3_ethtool.c
index 3b70cfe..38f7c79 100644
--- a/drivers/net/vmxnet3/vmxnet3_ethtool.c
+++ b/drivers/net/vmxnet3/vmxnet3_ethtool.c
@@ -430,11 +430,10 @@ vmxnet3_get_regs(struct net_device *netdev, struct ethtool_regs *regs, void *p)
 		buf[j++] = rq->rx_ring[1].next2comp;
 		buf[j++] = rq->rx_ring[1].gen;
 
-		/* receive data ring */
-		buf[j++] = 0;
-		buf[j++] = 0;
-		buf[j++] = 0;
-		buf[j++] = 0;
+		buf[j++] = VMXNET3_GET_ADDR_LO(rq->data_ring.basePA);
+		buf[j++] = VMXNET3_GET_ADDR_HI(rq->data_ring.basePA);
+		buf[j++] = rq->rx_ring[0].size;
+		buf[j++] = rq->data_ring.desc_size;
 
 		buf[j++] = VMXNET3_GET_ADDR_LO(rq->comp_ring.basePA);
 		buf[j++] = VMXNET3_GET_ADDR_HI(rq->comp_ring.basePA);
@@ -503,12 +502,14 @@ vmxnet3_get_ringparam(struct net_device *netdev,
 
 	param->rx_max_pending = VMXNET3_RX_RING_MAX_SIZE;
 	param->tx_max_pending = VMXNET3_TX_RING_MAX_SIZE;
-	param->rx_mini_max_pending = 0;
+	param->rx_mini_max_pending = VMXNET3_VERSION_GE_3(adapter) ?
+		VMXNET3_RXDATA_DESC_MAX_SIZE : 0;
 	param->rx_jumbo_max_pending = VMXNET3_RX_RING2_MAX_SIZE;
 
 	param->rx_pending = adapter->rx_ring_size;
 	param->tx_pending = adapter->tx_ring_size;
-	param->rx_mini_pending = 0;
+	param->rx_mini_pending = VMXNET3_VERSION_GE_3(adapter) ?
+		adapter->rxdata_desc_size : 0;
 	param->rx_jumbo_pending = adapter->rx_ring2_size;
 }
 
@@ -519,6 +520,7 @@ vmxnet3_set_ringparam(struct net_device *netdev,
 {
 	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
 	u32 new_tx_ring_size, new_rx_ring_size, new_rx_ring2_size;
+	u16 new_rxdata_desc_size;
 	u32 sz;
 	int err = 0;
 
@@ -541,6 +543,15 @@ vmxnet3_set_ringparam(struct net_device *netdev,
 		return -EOPNOTSUPP;
 	}
 
+	if (VMXNET3_VERSION_GE_3(adapter)) {
+		if (param->rx_mini_pending < 0 ||
+		    param->rx_mini_pending > VMXNET3_RXDATA_DESC_MAX_SIZE) {
+			return -EINVAL;
+		}
+	} else if (param->rx_mini_pending != 0) {
+		return -EINVAL;
+	}
+
 	/* round it up to a multiple of VMXNET3_RING_SIZE_ALIGN */
 	new_tx_ring_size = (param->tx_pending + VMXNET3_RING_SIZE_MASK) &
 							~VMXNET3_RING_SIZE_MASK;
@@ -567,9 +578,19 @@ vmxnet3_set_ringparam(struct net_device *netdev,
 	new_rx_ring2_size = min_t(u32, new_rx_ring2_size,
 				  VMXNET3_RX_RING2_MAX_SIZE);
 
+	/* rx data ring buffer size has to be a multiple of
+	 * VMXNET3_RXDATA_DESC_SIZE_ALIGN
+	 */
+	new_rxdata_desc_size =
+		(param->rx_mini_pending + VMXNET3_RXDATA_DESC_SIZE_MASK) &
+		~VMXNET3_RXDATA_DESC_SIZE_MASK;
+	new_rxdata_desc_size = min_t(u16, new_rxdata_desc_size,
+				     VMXNET3_RXDATA_DESC_MAX_SIZE);
+
 	if (new_tx_ring_size == adapter->tx_ring_size &&
 	    new_rx_ring_size == adapter->rx_ring_size &&
-	    new_rx_ring2_size == adapter->rx_ring2_size) {
+	    new_rx_ring2_size == adapter->rx_ring2_size &&
+	    new_rxdata_desc_size == adapter->rxdata_desc_size) {
 		return 0;
 	}
 
@@ -591,8 +612,8 @@ vmxnet3_set_ringparam(struct net_device *netdev,
 
 		err = vmxnet3_create_queues(adapter, new_tx_ring_size,
 					    new_rx_ring_size, new_rx_ring2_size,
-					    adapter->txdata_desc_size);
-
+					    adapter->txdata_desc_size,
+					    new_rxdata_desc_size);
 		if (err) {
 			/* failed, most likely because of OOM, try default
 			 * size */
@@ -601,11 +622,15 @@ vmxnet3_set_ringparam(struct net_device *netdev,
 			new_rx_ring_size = VMXNET3_DEF_RX_RING_SIZE;
 			new_rx_ring2_size = VMXNET3_DEF_RX_RING2_SIZE;
 			new_tx_ring_size = VMXNET3_DEF_TX_RING_SIZE;
+			new_rxdata_desc_size = VMXNET3_VERSION_GE_3(adapter) ?
+				VMXNET3_DEF_RXDATA_DESC_SIZE : 0;
+
 			err = vmxnet3_create_queues(adapter,
 						    new_tx_ring_size,
 						    new_rx_ring_size,
 						    new_rx_ring2_size,
-						    adapter->txdata_desc_size);
+						    adapter->txdata_desc_size,
+						    new_rxdata_desc_size);
 			if (err) {
 				netdev_err(netdev, "failed to create queues "
 					   "with default sizes. Closing it\n");
@@ -621,6 +646,7 @@ vmxnet3_set_ringparam(struct net_device *netdev,
 	adapter->tx_ring_size = new_tx_ring_size;
 	adapter->rx_ring_size = new_rx_ring_size;
 	adapter->rx_ring2_size = new_rx_ring2_size;
+	adapter->rxdata_desc_size = new_rxdata_desc_size;
 
 out:
 	clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
diff --git a/drivers/net/vmxnet3/vmxnet3_int.h b/drivers/net/vmxnet3/vmxnet3_int.h
index 94010de..c46bf09 100644
--- a/drivers/net/vmxnet3/vmxnet3_int.h
+++ b/drivers/net/vmxnet3/vmxnet3_int.h
@@ -272,15 +272,23 @@ struct vmxnet3_rq_driver_stats {
 	u64 rx_buf_alloc_failure;
 };
 
+struct vmxnet3_rx_data_ring {
+	Vmxnet3_RxDataDesc *base;
+	dma_addr_t basePA;
+	u16 desc_size;
+};
+
 struct vmxnet3_rx_queue {
 	char			name[IFNAMSIZ + 8]; /* To identify interrupt */
 	struct vmxnet3_adapter	  *adapter;
 	struct napi_struct        napi;
 	struct vmxnet3_cmd_ring   rx_ring[2];
+	struct vmxnet3_rx_data_ring data_ring;
 	struct vmxnet3_comp_ring  comp_ring;
 	struct vmxnet3_rx_ctx     rx_ctx;
 	u32 qid;            /* rqID in RCD for buffer from 1st ring */
 	u32 qid2;           /* rqID in RCD for buffer from 2nd ring */
+	u32 dataRingQid;    /* rqID in RCD for buffer from data ring */
 	struct vmxnet3_rx_buf_info     *buf_info[2];
 	dma_addr_t                      buf_info_pa;
 	struct Vmxnet3_RxQueueCtrl            *shared;
@@ -366,6 +374,9 @@ struct vmxnet3_adapter {
 
 	/* Size of buffer in the data ring */
 	u16 txdata_desc_size;
+	u16 rxdata_desc_size;
+
+	bool rxdataring_enabled;
 
 	struct work_struct work;
 
@@ -405,9 +416,19 @@ struct vmxnet3_adapter {
 #define VMXNET3_DEF_RX_RING_SIZE    256
 #define VMXNET3_DEF_RX_RING2_SIZE   128
 
+#define VMXNET3_DEF_RXDATA_DESC_SIZE 128
+
 #define VMXNET3_MAX_ETH_HDR_SIZE    22
 #define VMXNET3_MAX_SKB_BUF_SIZE    (3*1024)
 
+#define VMXNET3_GET_RING_IDX(adapter, rqID)		\
+	((rqID >= adapter->num_rx_queues &&		\
+	 rqID < 2 * adapter->num_rx_queues) ? 1 : 0)	\
+
+#define VMXNET3_RX_DATA_RING(adapter, rqID)		\
+	(rqID >= 2 * adapter->num_rx_queues &&		\
+	rqID < 3 * adapter->num_rx_queues)		\
+
 int
 vmxnet3_quiesce_dev(struct vmxnet3_adapter *adapter);
 
@@ -432,7 +453,7 @@ vmxnet3_set_features(struct net_device *netdev, netdev_features_t features);
 int
 vmxnet3_create_queues(struct vmxnet3_adapter *adapter,
 		      u32 tx_ring_size, u32 rx_ring_size, u32 rx_ring2_size,
-		      u16 txdata_desc_size);
+		      u16 txdata_desc_size, u16 rxdata_desc_size);
 
 void vmxnet3_set_ethtool_ops(struct net_device *netdev);
 
-- 
2.8.2

[PATCH net-next v5 5/7] vmxnet3: add support for get_coalesce, set_coalesce ethtool operations

[Shrikrishna Khare]

The emulation supports a variety of coalescing modes viz. disabled
(no coalescing), adaptive, static (number of packets to batch before
raising an interrupt), rate based (number of interrupts per second).

This patch implements get_coalesce and set_coalesce methods to allow
querying and configuring different coalescing modes.

Signed-off-by: Keyong Sun <sunk@vmware.com>
Signed-off-by: Manoj Tammali <tammalim@vmware.com>
Signed-off-by: Shrikrishna Khare <skhare@vmware.com>
---
 drivers/net/vmxnet3/vmxnet3_defs.h    |  33 ++++++-
 drivers/net/vmxnet3/vmxnet3_drv.c     |  54 ++++++++++++
 drivers/net/vmxnet3/vmxnet3_ethtool.c | 158 ++++++++++++++++++++++++++++++++++
 drivers/net/vmxnet3/vmxnet3_int.h     |   9 ++
 4 files changed, 253 insertions(+), 1 deletion(-)

diff --git a/drivers/net/vmxnet3/vmxnet3_defs.h b/drivers/net/vmxnet3/vmxnet3_defs.h
index f3b31c2..274e145 100644
--- a/drivers/net/vmxnet3/vmxnet3_defs.h
+++ b/drivers/net/vmxnet3/vmxnet3_defs.h
@@ -80,6 +80,7 @@ enum {
 	VMXNET3_CMD_LOAD_PLUGIN,
 	VMXNET3_CMD_RESERVED2,
 	VMXNET3_CMD_RESERVED3,
+	VMXNET3_CMD_SET_COALESCE,
 
 	VMXNET3_CMD_FIRST_GET = 0xF00D0000,
 	VMXNET3_CMD_GET_QUEUE_STATUS = VMXNET3_CMD_FIRST_GET,
@@ -92,7 +93,8 @@ enum {
 	VMXNET3_CMD_GET_DEV_EXTRA_INFO,
 	VMXNET3_CMD_GET_CONF_INTR,
 	VMXNET3_CMD_GET_RESERVED1,
-	VMXNET3_CMD_GET_TXDATA_DESC_SIZE
+	VMXNET3_CMD_GET_TXDATA_DESC_SIZE,
+	VMXNET3_CMD_GET_COALESCE,
 };
 
 /*
@@ -637,6 +639,35 @@ struct Vmxnet3_SetPolling {
 	u8					enablePolling;
 };
 
+#define VMXNET3_COAL_STATIC_MAX_DEPTH		128
+#define VMXNET3_COAL_RBC_MIN_RATE		100
+#define VMXNET3_COAL_RBC_MAX_RATE		100000
+
+enum Vmxnet3_CoalesceMode {
+	VMXNET3_COALESCE_DISABLED   = 0,
+	VMXNET3_COALESCE_ADAPT      = 1,
+	VMXNET3_COALESCE_STATIC     = 2,
+	VMXNET3_COALESCE_RBC        = 3
+};
+
+struct Vmxnet3_CoalesceRbc {
+	u32					rbc_rate;
+};
+
+struct Vmxnet3_CoalesceStatic {
+	u32					tx_depth;
+	u32					tx_comp_depth;
+	u32					rx_depth;
+};
+
+struct Vmxnet3_CoalesceScheme {
+	enum Vmxnet3_CoalesceMode		coalMode;
+	union {
+		struct Vmxnet3_CoalesceRbc	coalRbc;
+		struct Vmxnet3_CoalesceStatic	coalStatic;
+	} coalPara;
+};
+
 /* If the command data <= 16 bytes, use the shared memory directly.
  * otherwise, use variable length configuration descriptor.
  */
diff --git a/drivers/net/vmxnet3/vmxnet3_drv.c b/drivers/net/vmxnet3/vmxnet3_drv.c
index 6449d2e..d0bcc1d9 100644
--- a/drivers/net/vmxnet3/vmxnet3_drv.c
+++ b/drivers/net/vmxnet3/vmxnet3_drv.c
@@ -2491,6 +2491,32 @@ vmxnet3_setup_driver_shared(struct vmxnet3_adapter *adapter)
 	/* the rest are already zeroed */
 }
 
+static void
+vmxnet3_init_coalesce(struct vmxnet3_adapter *adapter)
+{
+	struct Vmxnet3_DriverShared *shared = adapter->shared;
+	union Vmxnet3_CmdInfo *cmdInfo = &shared->cu.cmdInfo;
+	unsigned long flags;
+
+	if (!VMXNET3_VERSION_GE_3(adapter))
+		return;
+
+	spin_lock_irqsave(&adapter->cmd_lock, flags);
+	cmdInfo->varConf.confVer = 1;
+	cmdInfo->varConf.confLen =
+		cpu_to_le32(sizeof(*adapter->coal_conf));
+	cmdInfo->varConf.confPA  = cpu_to_le64(adapter->coal_conf_pa);
+
+	if (adapter->default_coal_mode) {
+		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
+				       VMXNET3_CMD_GET_COALESCE);
+	} else {
+		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
+				       VMXNET3_CMD_SET_COALESCE);
+	}
+
+	spin_unlock_irqrestore(&adapter->cmd_lock, flags);
+}
 
 int
 vmxnet3_activate_dev(struct vmxnet3_adapter *adapter)
@@ -2540,6 +2566,8 @@ vmxnet3_activate_dev(struct vmxnet3_adapter *adapter)
 		goto activate_err;
 	}
 
+	vmxnet3_init_coalesce(adapter);
+
 	for (i = 0; i < adapter->num_rx_queues; i++) {
 		VMXNET3_WRITE_BAR0_REG(adapter,
 				VMXNET3_REG_RXPROD + i * VMXNET3_REG_ALIGN,
@@ -3345,6 +3373,22 @@ vmxnet3_probe_device(struct pci_dev *pdev,
 		goto err_ver;
 	}
 
+	if (VMXNET3_VERSION_GE_3(adapter)) {
+		adapter->coal_conf =
+			dma_alloc_coherent(&adapter->pdev->dev,
+					   sizeof(struct Vmxnet3_CoalesceScheme)
+					   ,
+					   &adapter->coal_conf_pa,
+					   GFP_KERNEL);
+		if (!adapter->coal_conf) {
+			err = -ENOMEM;
+			goto err_ver;
+		}
+		memset(adapter->coal_conf, 0, sizeof(*adapter->coal_conf));
+		adapter->coal_conf->coalMode = VMXNET3_COALESCE_DISABLED;
+		adapter->default_coal_mode = true;
+	}
+
 	SET_NETDEV_DEV(netdev, &pdev->dev);
 	vmxnet3_declare_features(adapter, dma64);
 
@@ -3407,6 +3451,11 @@ vmxnet3_probe_device(struct pci_dev *pdev,
 	return 0;
 
 err_register:
+	if (VMXNET3_VERSION_GE_3(adapter)) {
+		dma_free_coherent(&adapter->pdev->dev,
+				  sizeof(struct Vmxnet3_CoalesceScheme),
+				  adapter->coal_conf, adapter->coal_conf_pa);
+	}
 	vmxnet3_free_intr_resources(adapter);
 err_ver:
 	vmxnet3_free_pci_resources(adapter);
@@ -3457,6 +3506,11 @@ vmxnet3_remove_device(struct pci_dev *pdev)
 
 	vmxnet3_free_intr_resources(adapter);
 	vmxnet3_free_pci_resources(adapter);
+	if (VMXNET3_VERSION_GE_3(adapter)) {
+		dma_free_coherent(&adapter->pdev->dev,
+				  sizeof(struct Vmxnet3_CoalesceScheme),
+				  adapter->coal_conf, adapter->coal_conf_pa);
+	}
 #ifdef VMXNET3_RSS
 	dma_free_coherent(&adapter->pdev->dev, sizeof(struct UPT1_RSSConf),
 			  adapter->rss_conf, adapter->rss_conf_pa);
diff --git a/drivers/net/vmxnet3/vmxnet3_ethtool.c b/drivers/net/vmxnet3/vmxnet3_ethtool.c
index 38f7c79..aabc6ef 100644
--- a/drivers/net/vmxnet3/vmxnet3_ethtool.c
+++ b/drivers/net/vmxnet3/vmxnet3_ethtool.c
@@ -725,6 +725,162 @@ vmxnet3_set_rss(struct net_device *netdev, const u32 *p, const u8 *key,
 }
 #endif
 
+static int
+vmxnet3_get_coalesce(struct net_device *netdev, struct ethtool_coalesce *ec)
+{
+	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
+
+	if (!VMXNET3_VERSION_GE_3(adapter))
+		return -EOPNOTSUPP;
+
+	switch (adapter->coal_conf->coalMode) {
+	case VMXNET3_COALESCE_DISABLED:
+		/* struct ethtool_coalesce is already initialized to 0 */
+		break;
+	case VMXNET3_COALESCE_ADAPT:
+		ec->use_adaptive_rx_coalesce = true;
+		break;
+	case VMXNET3_COALESCE_STATIC:
+		ec->tx_max_coalesced_frames =
+			adapter->coal_conf->coalPara.coalStatic.tx_comp_depth;
+		ec->rx_max_coalesced_frames =
+			adapter->coal_conf->coalPara.coalStatic.rx_depth;
+		break;
+	case VMXNET3_COALESCE_RBC: {
+		u32 rbc_rate;
+
+		rbc_rate = adapter->coal_conf->coalPara.coalRbc.rbc_rate;
+		ec->rx_coalesce_usecs = VMXNET3_COAL_RBC_USECS(rbc_rate);
+	}
+		break;
+	default:
+		return -EOPNOTSUPP;
+	}
+
+	return 0;
+}
+
+static int
+vmxnet3_set_coalesce(struct net_device *netdev, struct ethtool_coalesce *ec)
+{
+	struct vmxnet3_adapter *adapter = netdev_priv(netdev);
+	struct Vmxnet3_DriverShared *shared = adapter->shared;
+	union Vmxnet3_CmdInfo *cmdInfo = &shared->cu.cmdInfo;
+	unsigned long flags;
+
+	if (!VMXNET3_VERSION_GE_3(adapter))
+		return -EOPNOTSUPP;
+
+	if (ec->rx_coalesce_usecs_irq ||
+	    ec->rx_max_coalesced_frames_irq ||
+	    ec->tx_coalesce_usecs ||
+	    ec->tx_coalesce_usecs_irq ||
+	    ec->tx_max_coalesced_frames_irq ||
+	    ec->stats_block_coalesce_usecs ||
+	    ec->use_adaptive_tx_coalesce ||
+	    ec->pkt_rate_low ||
+	    ec->rx_coalesce_usecs_low ||
+	    ec->rx_max_coalesced_frames_low ||
+	    ec->tx_coalesce_usecs_low ||
+	    ec->tx_max_coalesced_frames_low ||
+	    ec->pkt_rate_high ||
+	    ec->rx_coalesce_usecs_high ||
+	    ec->rx_max_coalesced_frames_high ||
+	    ec->tx_coalesce_usecs_high ||
+	    ec->tx_max_coalesced_frames_high ||
+	    ec->rate_sample_interval) {
+		return -EINVAL;
+	}
+
+	if ((ec->rx_coalesce_usecs == 0) &&
+	    (ec->use_adaptive_rx_coalesce == 0) &&
+	    (ec->tx_max_coalesced_frames == 0) &&
+	    (ec->rx_max_coalesced_frames == 0)) {
+		memset(adapter->coal_conf, 0, sizeof(*adapter->coal_conf));
+		adapter->coal_conf->coalMode = VMXNET3_COALESCE_DISABLED;
+		goto done;
+	}
+
+	if (ec->rx_coalesce_usecs != 0) {
+		u32 rbc_rate;
+
+		if ((ec->use_adaptive_rx_coalesce != 0) ||
+		    (ec->tx_max_coalesced_frames != 0) ||
+		    (ec->rx_max_coalesced_frames != 0)) {
+			return -EINVAL;
+		}
+
+		rbc_rate = VMXNET3_COAL_RBC_RATE(ec->rx_coalesce_usecs);
+		if (rbc_rate < VMXNET3_COAL_RBC_MIN_RATE ||
+		    rbc_rate > VMXNET3_COAL_RBC_MAX_RATE) {
+			return -EINVAL;
+		}
+
+		memset(adapter->coal_conf, 0, sizeof(*adapter->coal_conf));
+		adapter->coal_conf->coalMode = VMXNET3_COALESCE_RBC;
+		adapter->coal_conf->coalPara.coalRbc.rbc_rate = rbc_rate;
+		goto done;
+	}
+
+	if (ec->use_adaptive_rx_coalesce != 0) {
+		if ((ec->rx_coalesce_usecs != 0) ||
+		    (ec->tx_max_coalesced_frames != 0) ||
+		    (ec->rx_max_coalesced_frames != 0)) {
+			return -EINVAL;
+		}
+		memset(adapter->coal_conf, 0, sizeof(*adapter->coal_conf));
+		adapter->coal_conf->coalMode = VMXNET3_COALESCE_ADAPT;
+		goto done;
+	}
+
+	if ((ec->tx_max_coalesced_frames != 0) ||
+	    (ec->rx_max_coalesced_frames != 0)) {
+		if ((ec->rx_coalesce_usecs != 0) ||
+		    (ec->use_adaptive_rx_coalesce != 0)) {
+			return -EINVAL;
+		}
+
+		if ((ec->tx_max_coalesced_frames >
+		    VMXNET3_COAL_STATIC_MAX_DEPTH) ||
+		    (ec->rx_max_coalesced_frames >
+		     VMXNET3_COAL_STATIC_MAX_DEPTH)) {
+			return -EINVAL;
+		}
+
+		memset(adapter->coal_conf, 0, sizeof(*adapter->coal_conf));
+		adapter->coal_conf->coalMode = VMXNET3_COALESCE_STATIC;
+
+		adapter->coal_conf->coalPara.coalStatic.tx_comp_depth =
+			(ec->tx_max_coalesced_frames ?
+			 ec->tx_max_coalesced_frames :
+			 VMXNET3_COAL_STATIC_DEFAULT_DEPTH);
+
+		adapter->coal_conf->coalPara.coalStatic.rx_depth =
+			(ec->rx_max_coalesced_frames ?
+			 ec->rx_max_coalesced_frames :
+			 VMXNET3_COAL_STATIC_DEFAULT_DEPTH);
+
+		adapter->coal_conf->coalPara.coalStatic.tx_depth =
+			 VMXNET3_COAL_STATIC_DEFAULT_DEPTH;
+		goto done;
+	}
+
+done:
+	adapter->default_coal_mode = false;
+	if (netif_running(netdev)) {
+		spin_lock_irqsave(&adapter->cmd_lock, flags);
+		cmdInfo->varConf.confVer = 1;
+		cmdInfo->varConf.confLen =
+			cpu_to_le32(sizeof(*adapter->coal_conf));
+		cmdInfo->varConf.confPA  = cpu_to_le64(adapter->coal_conf_pa);
+		VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
+				       VMXNET3_CMD_SET_COALESCE);
+		spin_unlock_irqrestore(&adapter->cmd_lock, flags);
+	}
+
+	return 0;
+}
+
 static const struct ethtool_ops vmxnet3_ethtool_ops = {
 	.get_settings      = vmxnet3_get_settings,
 	.get_drvinfo       = vmxnet3_get_drvinfo,
@@ -733,6 +889,8 @@ static const struct ethtool_ops vmxnet3_ethtool_ops = {
 	.get_wol           = vmxnet3_get_wol,
 	.set_wol           = vmxnet3_set_wol,
 	.get_link          = ethtool_op_get_link,
+	.get_coalesce      = vmxnet3_get_coalesce,
+	.set_coalesce      = vmxnet3_set_coalesce,
 	.get_strings       = vmxnet3_get_strings,
 	.get_sset_count	   = vmxnet3_get_sset_count,
 	.get_ethtool_stats = vmxnet3_get_ethtool_stats,
diff --git a/drivers/net/vmxnet3/vmxnet3_int.h b/drivers/net/vmxnet3/vmxnet3_int.h
index c46bf09..63df4f2 100644
--- a/drivers/net/vmxnet3/vmxnet3_int.h
+++ b/drivers/net/vmxnet3/vmxnet3_int.h
@@ -358,6 +358,7 @@ struct vmxnet3_adapter {
 	int		rx_buf_per_pkt;  /* only apply to the 1st ring */
 	dma_addr_t			shared_pa;
 	dma_addr_t queue_desc_pa;
+	dma_addr_t coal_conf_pa;
 
 	/* Wake-on-LAN */
 	u32     wol;
@@ -384,6 +385,9 @@ struct vmxnet3_adapter {
 
 	int share_intr;
 
+	struct Vmxnet3_CoalesceScheme *coal_conf;
+	bool   default_coal_mode;
+
 	dma_addr_t adapter_pa;
 	dma_addr_t pm_conf_pa;
 	dma_addr_t rss_conf_pa;
@@ -429,6 +433,11 @@ struct vmxnet3_adapter {
 	(rqID >= 2 * adapter->num_rx_queues &&		\
 	rqID < 3 * adapter->num_rx_queues)		\
 
+#define VMXNET3_COAL_STATIC_DEFAULT_DEPTH	64
+
+#define VMXNET3_COAL_RBC_RATE(usecs) (1000000 / usecs)
+#define VMXNET3_COAL_RBC_USECS(rbc_rate) (1000000 / rbc_rate)
+
 int
 vmxnet3_quiesce_dev(struct vmxnet3_adapter *adapter);
 
-- 
2.8.2

[PATCH net-next v5 6/7] vmxnet3: introduce command to register memory region

[Shrikrishna Khare]

In vmxnet3 version 3, the emulation added support for the vmxnet3 driver
to communicate information about the memory regions the driver will use
for rx/tx buffers. The driver can also indicate which rx/tx queue the
memory region is applicable for. If this information is communicated
to the emulation, the emulation will always keep these memory regions
mapped, thereby avoiding the mapping/unmapping overhead for every packet.

Currently, Linux vmxnet3 driver does not leverage this capability. The
feasibility of using this approach for the Linux vmxnet3 driver will be
investigated independently and if possible, will be part of a different
patch. This patch only exposes the emulation capability to the driver
(vmxnet3_defs.h is identical between the driver and the emulation).

Signed-off-by: Guolin Yang <gyang@vmware.com>
Signed-off-by: Shrikrishna Khare <skhare@vmware.com>
---
 drivers/net/vmxnet3/vmxnet3_defs.h | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)

diff --git a/drivers/net/vmxnet3/vmxnet3_defs.h b/drivers/net/vmxnet3/vmxnet3_defs.h
index 274e145..c3a3164 100644
--- a/drivers/net/vmxnet3/vmxnet3_defs.h
+++ b/drivers/net/vmxnet3/vmxnet3_defs.h
@@ -81,6 +81,7 @@ enum {
 	VMXNET3_CMD_RESERVED2,
 	VMXNET3_CMD_RESERVED3,
 	VMXNET3_CMD_SET_COALESCE,
+	VMXNET3_CMD_REGISTER_MEMREGS,
 
 	VMXNET3_CMD_FIRST_GET = 0xF00D0000,
 	VMXNET3_CMD_GET_QUEUE_STATUS = VMXNET3_CMD_FIRST_GET,
@@ -668,6 +669,22 @@ struct Vmxnet3_CoalesceScheme {
 	} coalPara;
 };
 
+struct Vmxnet3_MemoryRegion {
+	__le64					startPA;
+	__le32					length;
+	__le16					txQueueBits;
+	__le16					rxQueueBits;
+};
+
+#define MAX_MEMORY_REGION_PER_QUEUE 16
+#define MAX_MEMORY_REGION_PER_DEVICE 256
+
+struct Vmxnet3_MemRegs {
+	__le16					numRegs;
+	__le16					pad[3];
+	struct Vmxnet3_MemoryRegion		memRegs[1];
+};
+
 /* If the command data <= 16 bytes, use the shared memory directly.
  * otherwise, use variable length configuration descriptor.
  */
-- 
2.8.2

[PATCH net-next v5 7/7] vmxnet3: update to version 3

[Shrikrishna Khare]

With all vmxnet3 version 3 changes incorporated in the vmxnet3 driver,
the driver can configure emulation to run at vmxnet3 version 3, provided
the emulation advertises support for version 3.

Signed-off-by: Shrikrishna Khare <skhare@vmware.com>
---
 drivers/net/vmxnet3/vmxnet3_drv.c | 7 ++++++-
 drivers/net/vmxnet3/vmxnet3_int.h | 4 ++--
 2 files changed, 8 insertions(+), 3 deletions(-)

diff --git a/drivers/net/vmxnet3/vmxnet3_drv.c b/drivers/net/vmxnet3/vmxnet3_drv.c
index d0bcc1d9..c68fe49 100644
--- a/drivers/net/vmxnet3/vmxnet3_drv.c
+++ b/drivers/net/vmxnet3/vmxnet3_drv.c
@@ -3345,7 +3345,12 @@ vmxnet3_probe_device(struct pci_dev *pdev,
 		goto err_alloc_pci;
 
 	ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_VRRS);
-	if (ver & (1 << VMXNET3_REV_2)) {
+	if (ver & (1 << VMXNET3_REV_3)) {
+		VMXNET3_WRITE_BAR1_REG(adapter,
+				       VMXNET3_REG_VRRS,
+				       1 << VMXNET3_REV_3);
+		adapter->version = VMXNET3_REV_3 + 1;
+	} else if (ver & (1 << VMXNET3_REV_2)) {
 		VMXNET3_WRITE_BAR1_REG(adapter,
 				       VMXNET3_REG_VRRS,
 				       1 << VMXNET3_REV_2);
diff --git a/drivers/net/vmxnet3/vmxnet3_int.h b/drivers/net/vmxnet3/vmxnet3_int.h
index 63df4f2..74fc030 100644
--- a/drivers/net/vmxnet3/vmxnet3_int.h
+++ b/drivers/net/vmxnet3/vmxnet3_int.h
@@ -69,10 +69,10 @@
 /*
  * Version numbers
  */
-#define VMXNET3_DRIVER_VERSION_STRING   "1.4.8.0-k"
+#define VMXNET3_DRIVER_VERSION_STRING   "1.4.9.0-k"
 
 /* a 32-bit int, each byte encode a verion number in VMXNET3_DRIVER_VERSION */
-#define VMXNET3_DRIVER_VERSION_NUM      0x01040800
+#define VMXNET3_DRIVER_VERSION_NUM      0x01040900
 
 #if defined(CONFIG_PCI_MSI)
 	/* RSS only makes sense if MSI-X is supported. */
-- 
2.8.2

Re: [PATCH net-next v5 0/7] vmxnet3: upgrade to version 3

[David Miller]

From: Shrikrishna Khare <skhare@vmware.com>
Date: Thu, 16 Jun 2016 10:51:52 -0700

> vmxnet3 emulation has recently added several new features which includes
> support for new commands the driver can issue to emulation, change in
> descriptor fields etc. This patch series extends the vmxnet3 driver to
> leverage these new features.
> 
> Compatibility is maintained using existing vmxnet3 versioning mechanism as
> follows:
>  - new features added to vmxnet3 emulation are associated with new vmxnet3
>    version viz. vmxnet3 version 3.
>  - emulation advertises all the versions it supports to the driver.
>  - during initialization, vmxnet3 driver picks the highest version number
>  supported by both the emulation and the driver and configures emulation
>  to run at that version.
> 
> In particular, following changes are introduced:
 ... 

Series applied, thanks.