[PATCH net-next 09/10] netvsc: allow controlling send/recv buffer size

[Stephen Hemminger <stephen@networkplumber.org>]

Control the size of the buffer areas via ethtool ring settings.
They aren't really traditional hardware rings, but host API breaks
receive and send buffer into chunks. The final size of the chunks are
controlled by the host.

The default value of send and receive buffer area for host DMA
is much larger than it needs to be. Experimentation shows that
4M receive and 1M send is sufficient.

Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
---
 drivers/net/hyperv/hyperv_net.h |   9 ++--
 drivers/net/hyperv/netvsc.c     |  70 +++++++++++++-----------
 drivers/net/hyperv/netvsc_drv.c | 117 ++++++++++++++++++++++++++++++++++++++--
 3 files changed, 157 insertions(+), 39 deletions(-)

diff --git a/drivers/net/hyperv/hyperv_net.h b/drivers/net/hyperv/hyperv_net.h
index f620c90307ed..4ccfc335a356 100644
--- a/drivers/net/hyperv/hyperv_net.h
+++ b/drivers/net/hyperv/hyperv_net.h
@@ -148,6 +148,8 @@ struct netvsc_device_info {
 	unsigned char mac_adr[ETH_ALEN];
 	int  ring_size;
 	u32  num_chn;
+	u32  send_sections;
+	u32  recv_sections;
 };
 
 enum rndis_device_state {
@@ -634,12 +636,12 @@ struct nvsp_message {
 #define NETVSC_SEND_BUFFER_SIZE			(1024 * 1024 * 15)   /* 15MB */
 #define NETVSC_INVALID_INDEX			-1
 
+#define NETVSC_SEND_SECTION_SIZE		6144
+#define NETVSC_RECV_SECTION_SIZE		1728
 
 #define NETVSC_RECEIVE_BUFFER_ID		0xcafe
 #define NETVSC_SEND_BUFFER_ID			0
 
-#define NETVSC_PACKET_SIZE                      4096
-
 #define VRSS_SEND_TAB_SIZE 16  /* must be power of 2 */
 #define VRSS_CHANNEL_MAX 64
 #define VRSS_CHANNEL_DEFAULT 8
@@ -754,14 +756,13 @@ struct netvsc_device {
 
 	/* Receive buffer allocated by us but manages by NetVSP */
 	void *recv_buf;
-	u32 recv_buf_size;
 	u32 recv_buf_gpadl_handle;
 	u32 recv_section_cnt;
+	u32 recv_section_size;
 	u32 recv_completion_cnt;
 
 	/* Send buffer allocated by us */
 	void *send_buf;
-	u32 send_buf_size;
 	u32 send_buf_gpadl_handle;
 	u32 send_section_cnt;
 	u32 send_section_size;
diff --git a/drivers/net/hyperv/netvsc.c b/drivers/net/hyperv/netvsc.c
index 91b0674cfdb0..5b3a2b8fa77e 100644
--- a/drivers/net/hyperv/netvsc.c
+++ b/drivers/net/hyperv/netvsc.c
@@ -75,6 +75,10 @@ static struct netvsc_device *alloc_net_device(void)
 	atomic_set(&net_device->open_cnt, 0);
 	net_device->max_pkt = RNDIS_MAX_PKT_DEFAULT;
 	net_device->pkt_align = RNDIS_PKT_ALIGN_DEFAULT;
+
+	net_device->recv_section_size = NETVSC_RECV_SECTION_SIZE;
+	net_device->send_section_size = NETVSC_SEND_SECTION_SIZE;
+
 	init_completion(&net_device->channel_init_wait);
 
 	return net_device;
@@ -142,6 +146,7 @@ static void netvsc_destroy_buf(struct hv_device *device)
 				"revoke receive buffer to netvsp\n");
 			return;
 		}
+		net_device->recv_section_cnt = 0;
 	}
 
 	/* Teardown the gpadl on the vsp end */
@@ -172,7 +177,7 @@ static void netvsc_destroy_buf(struct hv_device *device)
 	 * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need
 	 * to send a revoke msg here
 	 */
-	if (net_device->send_section_size) {
+	if (net_device->send_section_cnt) {
 		/* Send the revoke receive buffer */
 		revoke_packet = &net_device->revoke_packet;
 		memset(revoke_packet, 0, sizeof(struct nvsp_message));
@@ -204,6 +209,7 @@ static void netvsc_destroy_buf(struct hv_device *device)
 				   "revoke send buffer to netvsp\n");
 			return;
 		}
+		net_device->send_section_cnt = 0;
 	}
 	/* Teardown the gpadl on the vsp end */
 	if (net_device->send_buf_gpadl_handle) {
@@ -243,18 +249,25 @@ int netvsc_alloc_recv_comp_ring(struct netvsc_device *net_device, u32 q_idx)
 }
 
 static int netvsc_init_buf(struct hv_device *device,
-			   struct netvsc_device *net_device)
+			   struct netvsc_device *net_device,
+			   const struct netvsc_device_info *device_info)
 {
 	struct nvsp_1_message_send_receive_buffer_complete *resp;
 	struct net_device *ndev = hv_get_drvdata(device);
 	struct nvsp_message *init_packet;
+	unsigned int buf_size;
 	size_t map_words;
 	int ret = 0;
 
-	net_device->recv_buf = vzalloc(net_device->recv_buf_size);
+	/* Get receive buffer area. */
+	buf_size = device_info->recv_sections * net_device->recv_section_size;
+	buf_size = roundup(buf_size, PAGE_SIZE);
+
+	net_device->recv_buf = vzalloc(buf_size);
 	if (!net_device->recv_buf) {
-		netdev_err(ndev, "unable to allocate receive "
-			"buffer of size %d\n", net_device->recv_buf_size);
+		netdev_err(ndev,
+			   "unable to allocate receive buffer of size %u\n",
+			   buf_size);
 		ret = -ENOMEM;
 		goto cleanup;
 	}
@@ -265,7 +278,7 @@ static int netvsc_init_buf(struct hv_device *device,
 	 * than the channel to establish the gpadl handle.
 	 */
 	ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
-				    net_device->recv_buf_size,
+				    buf_size,
 				    &net_device->recv_buf_gpadl_handle);
 	if (ret != 0) {
 		netdev_err(ndev,
@@ -311,31 +324,31 @@ static int netvsc_init_buf(struct hv_device *device,
 		   resp->num_sections, resp->sections[0].sub_alloc_size,
 		   resp->sections[0].num_sub_allocs);
 
-	net_device->recv_section_cnt = resp->num_sections;
-
-	/*
-	 * For 1st release, there should only be 1 section that represents the
-	 * entire receive buffer
-	 */
-	if (net_device->recv_section_cnt != 1 ||
-	    resp->sections[0].offset != 0) {
+	/* There should only be one section for the entire receive buffer */
+	if (resp->num_sections != 1 || resp->sections[0].offset != 0) {
 		ret = -EINVAL;
 		goto cleanup;
 	}
 
+	net_device->recv_section_size = resp->sections[0].sub_alloc_size;
+	net_device->recv_section_cnt = resp->sections[0].num_sub_allocs;
+
 	/* Setup receive completion ring */
 	net_device->recv_completion_cnt
-		= round_up(resp->sections[0].num_sub_allocs + 1,
+		= round_up(net_device->recv_section_cnt + 1,
 			   PAGE_SIZE / sizeof(u64));
 	ret = netvsc_alloc_recv_comp_ring(net_device, 0);
 	if (ret)
 		goto cleanup;
 
 	/* Now setup the send buffer. */
-	net_device->send_buf = vzalloc(net_device->send_buf_size);
+	buf_size = device_info->send_sections * net_device->send_section_size;
+	buf_size = round_up(buf_size, PAGE_SIZE);
+
+	net_device->send_buf = vzalloc(buf_size);
 	if (!net_device->send_buf) {
-		netdev_err(ndev, "unable to allocate send "
-			   "buffer of size %d\n", net_device->send_buf_size);
+		netdev_err(ndev, "unable to allocate send buffer of size %u\n",
+			   buf_size);
 		ret = -ENOMEM;
 		goto cleanup;
 	}
@@ -345,7 +358,7 @@ static int netvsc_init_buf(struct hv_device *device,
 	 * than the channel to establish the gpadl handle.
 	 */
 	ret = vmbus_establish_gpadl(device->channel, net_device->send_buf,
-				    net_device->send_buf_size,
+				    buf_size,
 				    &net_device->send_buf_gpadl_handle);
 	if (ret != 0) {
 		netdev_err(ndev,
@@ -390,10 +403,8 @@ static int netvsc_init_buf(struct hv_device *device,
 	net_device->send_section_size = init_packet->msg.
 				v1_msg.send_send_buf_complete.section_size;
 
-	/* Section count is simply the size divided by the section size.
-	 */
-	net_device->send_section_cnt =
-		net_device->send_buf_size / net_device->send_section_size;
+	/* Section count is simply the size divided by the section size. */
+	net_device->send_section_cnt = buf_size / net_device->send_section_size;
 
 	netdev_dbg(ndev, "Send section size: %d, Section count:%d\n",
 		   net_device->send_section_size, net_device->send_section_cnt);
@@ -471,7 +482,8 @@ static int negotiate_nvsp_ver(struct hv_device *device,
 }
 
 static int netvsc_connect_vsp(struct hv_device *device,
-			      struct netvsc_device *net_device)
+			      struct netvsc_device *net_device,
+			      const struct netvsc_device_info *device_info)
 {
 	const u32 ver_list[] = {
 		NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
@@ -521,14 +533,8 @@ static int netvsc_connect_vsp(struct hv_device *device,
 	if (ret != 0)
 		goto cleanup;
 
-	/* Post the big receive buffer to NetVSP */
-	if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
-		net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
-	else
-		net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
-	net_device->send_buf_size = NETVSC_SEND_BUFFER_SIZE;
 
-	ret = netvsc_init_buf(device, net_device);
+	ret = netvsc_init_buf(device, net_device, device_info);
 
 cleanup:
 	return ret;
@@ -1284,7 +1290,7 @@ struct netvsc_device *netvsc_device_add(struct hv_device *device,
 	rcu_assign_pointer(net_device_ctx->nvdev, net_device);
 
 	/* Connect with the NetVsp */
-	ret = netvsc_connect_vsp(device, net_device);
+	ret = netvsc_connect_vsp(device, net_device, device_info);
 	if (ret != 0) {
 		netdev_err(ndev,
 			"unable to connect to NetVSP - %d\n", ret);
diff --git a/drivers/net/hyperv/netvsc_drv.c b/drivers/net/hyperv/netvsc_drv.c
index 7b465e40869b..873c83a66cc2 100644
--- a/drivers/net/hyperv/netvsc_drv.c
+++ b/drivers/net/hyperv/netvsc_drv.c
@@ -45,7 +45,12 @@
 
 #include "hyperv_net.h"
 
-#define RING_SIZE_MIN 64
+#define RING_SIZE_MIN		64
+#define NETVSC_MIN_TX_SECTIONS	10
+#define NETVSC_DEFAULT_TX	192	/* ~1M */
+#define NETVSC_MIN_RX_SECTIONS	10	/* ~64K */
+#define NETVSC_DEFAULT_RX	2048	/* ~4M */
+
 #define LINKCHANGE_INT (2 * HZ)
 #define VF_TAKEOVER_INT (HZ / 10)
 
@@ -831,11 +836,13 @@ static int netvsc_set_channels(struct net_device *net,
 	if (was_opened)
 		rndis_filter_close(nvdev);
 
-	rndis_filter_device_remove(dev, nvdev);
-
 	memset(&device_info, 0, sizeof(device_info));
 	device_info.num_chn = count;
 	device_info.ring_size = ring_size;
+	device_info.send_sections = nvdev->send_section_cnt;
+	device_info.recv_sections = nvdev->recv_section_cnt;
+
+	rndis_filter_device_remove(dev, nvdev);
 
 	nvdev = rndis_filter_device_add(dev, &device_info);
 	if (!IS_ERR(nvdev)) {
@@ -947,6 +954,8 @@ static int netvsc_change_mtu(struct net_device *ndev, int mtu)
 	memset(&device_info, 0, sizeof(device_info));
 	device_info.ring_size = ring_size;
 	device_info.num_chn = nvdev->num_chn;
+	device_info.send_sections = nvdev->send_section_cnt;
+	device_info.recv_sections = nvdev->recv_section_cnt;
 
 	rndis_filter_device_remove(hdev, nvdev);
 
@@ -1351,6 +1360,104 @@ static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
 	return rndis_filter_set_rss_param(rndis_dev, key, ndev->num_chn);
 }
 
+/* Hyper-V RNDIS protocol does not have ring in the HW sense.
+ * It does have pre-allocated receive area which is divided into sections.
+ */
+static void __netvsc_get_ringparam(struct netvsc_device *nvdev,
+				   struct ethtool_ringparam *ring)
+{
+	u32 max_buf_size;
+
+	ring->rx_pending = nvdev->recv_section_cnt;
+	ring->tx_pending = nvdev->send_section_cnt;
+
+	if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
+		max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
+	else
+		max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
+
+	ring->rx_max_pending = max_buf_size / nvdev->recv_section_size;
+	ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE
+		/ nvdev->send_section_size;
+}
+
+static void netvsc_get_ringparam(struct net_device *ndev,
+				 struct ethtool_ringparam *ring)
+{
+	struct net_device_context *ndevctx = netdev_priv(ndev);
+	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
+
+	if (!nvdev)
+		return;
+
+	__netvsc_get_ringparam(nvdev, ring);
+}
+
+static int netvsc_set_ringparam(struct net_device *ndev,
+				struct ethtool_ringparam *ring)
+{
+	struct net_device_context *ndevctx = netdev_priv(ndev);
+	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
+	struct hv_device *hdev = ndevctx->device_ctx;
+	struct netvsc_device_info device_info;
+	struct ethtool_ringparam orig;
+	u32 new_tx, new_rx;
+	bool was_opened;
+	int ret = 0;
+
+	if (!nvdev || nvdev->destroy)
+		return -ENODEV;
+
+	memset(&orig, 0, sizeof(orig));
+	__netvsc_get_ringparam(nvdev, &orig);
+
+	new_tx = clamp_t(u32, ring->tx_pending,
+			 NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending);
+	new_rx = clamp_t(u32, ring->rx_pending,
+			 NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending);
+
+	if (new_tx == orig.tx_pending &&
+	    new_rx == orig.rx_pending)
+		return 0;	 /* no change */
+
+	memset(&device_info, 0, sizeof(device_info));
+	device_info.num_chn = nvdev->num_chn;
+	device_info.ring_size = ring_size;
+	device_info.send_sections = new_tx;
+	device_info.recv_sections = new_rx;
+
+	netif_device_detach(ndev);
+	was_opened = rndis_filter_opened(nvdev);
+	if (was_opened)
+		rndis_filter_close(nvdev);
+
+	rndis_filter_device_remove(hdev, nvdev);
+
+	nvdev = rndis_filter_device_add(hdev, &device_info);
+	if (IS_ERR(nvdev)) {
+		ret = PTR_ERR(nvdev);
+
+		device_info.send_sections = orig.tx_pending;
+		device_info.recv_sections = orig.rx_pending;
+		nvdev = rndis_filter_device_add(hdev, &device_info);
+		if (IS_ERR(nvdev)) {
+			netdev_err(ndev, "restoring ringparam failed: %ld\n",
+				   PTR_ERR(nvdev));
+			return ret;
+		}
+	}
+
+	if (was_opened)
+		rndis_filter_open(nvdev);
+	netif_device_attach(ndev);
+
+	/* We may have missed link change notifications */
+	ndevctx->last_reconfig = 0;
+	schedule_delayed_work(&ndevctx->dwork, 0);
+
+	return ret;
+}
+
 static const struct ethtool_ops ethtool_ops = {
 	.get_drvinfo	= netvsc_get_drvinfo,
 	.get_link	= ethtool_op_get_link,
@@ -1367,6 +1474,8 @@ static const struct ethtool_ops ethtool_ops = {
 	.set_rxfh	= netvsc_set_rxfh,
 	.get_link_ksettings = netvsc_get_link_ksettings,
 	.set_link_ksettings = netvsc_set_link_ksettings,
+	.get_ringparam	= netvsc_get_ringparam,
+	.set_ringparam	= netvsc_set_ringparam,
 };
 
 static const struct net_device_ops device_ops = {
@@ -1782,6 +1891,8 @@ static int netvsc_probe(struct hv_device *dev,
 	memset(&device_info, 0, sizeof(device_info));
 	device_info.ring_size = ring_size;
 	device_info.num_chn = VRSS_CHANNEL_DEFAULT;
+	device_info.send_sections = NETVSC_DEFAULT_TX;
+	device_info.recv_sections = NETVSC_DEFAULT_RX;
 
 	nvdev = rndis_filter_device_add(dev, &device_info);
 	if (IS_ERR(nvdev)) {
-- 
2.11.0