Skip to contentSkip to conversation

hyperv: Enable batched notification

PATCH Superseded K. Y. Srinivasan 2 participants 10 messages Mar 11, 2015 – Mar 12, 2015

Subsystems (3)
Networking Network drivers Hyper-V/Azure core and drivers

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore

K. Y. Srinivasan

author
#0 lore raw

v2 cover hyperv: Enable batched notification

Take into consideration the xmit_more flag in skb to decide if we should notify the host as we place packets in VMBUS.

The VMBUS API that would give us this control is already in Greg's tree, in this patch-set, that API is exported so it can be used in the netvsc driver.

Also included here is a siganlling related fix in the vmbus driver that is exposed by the netvsc driver. I would like to thank
Jason Wang <[email protected]> for pointing out this issue.

This version of the patch-set, adds an additional patch to fix the signalling issue.

K. Y. Srinivasan (3):
Drivers: hv: vmbus: Export the vmbus_sendpacket_pagebuffer_ctl()
Drivers: hv: vmbus: Fix a bug in the signalling logic with kick_q
hyperv: Support batched notification

 drivers/hv/channel.c              |   33 +++++++++++++++++++++++++++++++++
 drivers/net/hyperv/hyperv_net.h   |    2 +-
 drivers/net/hyperv/netvsc.c       |   14 +++++++++-----
 drivers/net/hyperv/netvsc_drv.c   |    3 ++-
 drivers/net/hyperv/rndis_filter.c |    2 +-
 5 files changed, 46 insertions(+), 8 deletions(-)
Signature · 1 line
1.7.4.1
author
#1 lore raw

PATCH v2 1/3 Drivers: hv: vmbus: Export the vmbus_sendpacket_pagebuffer_ctl()

Export the vmbus_sendpacket_pagebuffer_ctl() interface. This interface will be used in the netvsc driver to optimize signalling the host.

Signed-off-by K. Y. Srinivasan <[email protected]>
---
 drivers/hv/channel.c |    1 +
 1 files changed, 1 insertions(+), 0 deletions(-)
drivers/hv/channel.c +1 -0
index da53180..e58cdb7 100644
@@ -710,6 +710,7 @@ int vmbus_sendpacket_pagebuffer_ctl(struct vmbus_channel *channel,
 
 	return ret;
 }
+EXPORT_SYMBOL_GPL(vmbus_sendpacket_pagebuffer_ctl);
 
 /*
  * vmbus_sendpacket_pagebuffer - Send a range of single-page buffer
Signature · 1 line
1.7.4.1
author
#2 lore raw

PATCH v2 2/3 Drivers: hv: vmbus: Fix a bug in the signalling logic with kick_q

When the caller specifies that signalling should be deferred, we need to address the case where we are not able to place the current packet because the buffer is full. In this case, we will signal the host as some packets may have been placed on the ring buffer.
I would like to thank Jason Wang <[email protected]> for pointing out this issue.

Signed-off-by K. Y. Srinivasan <[email protected]>
---
 drivers/hv/channel.c |   32 ++++++++++++++++++++++++++++++++
 1 files changed, 32 insertions(+), 0 deletions(-)
drivers/hv/channel.c +32 -0
index e58cdb7..ae06ba9 100644
@@ -614,8 +614,24 @@ int vmbus_sendpacket_ctl(struct vmbus_channel *channel, void *buffer,
 
 	ret = hv_ringbuffer_write(&channel->outbound, bufferlist, 3, &signal);
 
+	/*
+	 * Here is the logic for signalling the host:
+	 * 1. If the host is already draining the ringbuffer,
+	 *    don't signal. This is indicated by the parameter
+	 *    "signal".
+	 *
+	 * 2. If we are not able to write, signal if kick_q is false.
+	 *    kick_q being false indicates that we may have placed zero or
+	 *    more packets with more packets to come. We will signal in
+	 *    this case even if potentially we may have not placed any
+	 *    packet. This is a rare enough condition that it should not
+	 *    matter.
+	 */
+
 	if ((ret == 0) && kick_q && signal)
 		vmbus_setevent(channel);
+	else if ((ret != 0) && !kick_q)
+		vmbus_setevent(channel);
 
 	return ret;
 }

@@ -705,8 +721,24 @@ int vmbus_sendpacket_pagebuffer_ctl(struct vmbus_channel *channel,
 
 	ret = hv_ringbuffer_write(&channel->outbound, bufferlist, 3, &signal);
 
+	/*
+	 * Here is the logic for signalling the host:
+	 * 1. If the host is already draining the ringbuffer,
+	 *    don't signal. This is indicated by the parameter
+	 *    "signal".
+	 *
+	 * 2. If we are not able to write, signal if kick_q is false.
+	 *    kick_q being false indicates that we may have placed zero or
+	 *    more packets with more packets to come. We will signal in
+	 *    this case even if potentially we may have not placed any
+	 *    packet. This is a rare enough condition that it should not
+	 *    matter.
+	 */
+
 	if ((ret == 0) && kick_q && signal)
 		vmbus_setevent(channel);
+	else if ((ret != 0) && !kick_q)
+		vmbus_setevent(channel);
 
 	return ret;
 }
Signature · 1 line
1.7.4.1
#3 lore raw

Re: [PATCH V2 2/3 net-next] Drivers: hv: vmbus: Fix a bug in the signalling logic with kick_q

Quoted 68 lines · K. Y. Srinivasan

When the caller specifies that signalling should be deferred, we need to address the case where we are not able to place the current packet because the buffer is full. In this case, we will signal the host as some packets may have been placed on the ring buffer.
I would like to thank Jason Wang <[email protected]> for pointing out this issue.

Signed-off-by K. Y. Srinivasan <[email protected]>
---
 drivers/hv/channel.c |   32 ++++++++++++++++++++++++++++++++
 1 files changed, 32 insertions(+), 0 deletions(-)
drivers/hv/channel.c +16 -0
index e58cdb7..ae06ba9 100644
@@ -614,8 +614,24 @@ int vmbus_sendpacket_ctl(struct vmbus_channel 
 *channel, void *buffer,
 
 	ret = hv_ringbuffer_write(&channel->outbound, bufferlist, 3, 
 &signal);
 
+	/*
+	 * Here is the logic for signalling the host:
+	 * 1. If the host is already draining the ringbuffer,
+	 *    don't signal. This is indicated by the parameter
+	 *    "signal".
+	 *
+	 * 2. If we are not able to write, signal if kick_q is false.
+	 *    kick_q being false indicates that we may have placed zero or
+	 *    more packets with more packets to come. We will signal in
+	 *    this case even if potentially we may have not placed any
+	 *    packet. This is a rare enough condition that it should not
+	 *    matter.
+	 */
+
 	if ((ret == 0) && kick_q && signal)
 		vmbus_setevent(channel);
+	else if ((ret != 0) && !kick_q)
+		vmbus_setevent(channel);
drivers/hv/channel.c +16 -0
 	return ret;
 }

@@ -705,8 +721,24 @@ int vmbus_sendpacket_pagebuffer_ctl(struct 
 vmbus_channel *channel,
 
 	ret = hv_ringbuffer_write(&channel->outbound, bufferlist, 3, 
 &signal);
 
+	/*
+	 * Here is the logic for signalling the host:
+	 * 1. If the host is already draining the ringbuffer,
+	 *    don't signal. This is indicated by the parameter
+	 *    "signal".
+	 *
+	 * 2. If we are not able to write, signal if kick_q is false.
+	 *    kick_q being false indicates that we may have placed zero or
+	 *    more packets with more packets to come. We will signal in
+	 *    this case even if potentially we may have not placed any
+	 *    packet. This is a rare enough condition that it should not
+	 *    matter.
+	 */
+
 	if ((ret == 0) && kick_q && signal)
 		vmbus_setevent(channel);
+	else if ((ret != 0) && !kick_q)
+		vmbus_setevent(channel);
Quoted 3 lines

return ret;
}

Looks like we need to kick unconditionally here. Consider we may get -EAGAIN when we want to send the last skb (kick_q is true) from the list. We need kick host in this case.

Btw, another method is let the driver to decide e.g exporting the vmbus_setevent() and call it in netvsc_start_xmit().

author
#4 lore raw

RE: [PATCH V2 2/3 net-next] Drivers: hv: vmbus: Fix a bug in the signalling logic with kick_q

Quoted 92 lines

-----Original Message-----
From: Jason Wang [mailto:[email protected]]
Sent: Wednesday, March 11, 2015 8:08 PM
To: KY Srinivasan
Cc: [email protected]; [email protected]; linux- [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; KY Srinivasan
Subject: Re: [PATCH V2 2/3 net-next] Drivers: hv: vmbus: Fix a bug in the signalling logic with kick_q

Quoted 68 lines · K. Y. Srinivasan

When the caller specifies that signalling should be deferred, we need to address the case where we are not able to place the current packet because the buffer is full. In this case, we will signal the host as some packets may have been placed on the ring buffer.
I would like to thank Jason Wang <[email protected]> for pointing out this issue.

Signed-off-by K. Y. Srinivasan <[email protected]>
---
 drivers/hv/channel.c |   32 ++++++++++++++++++++++++++++++++
 1 files changed, 32 insertions(+), 0 deletions(-)
drivers/hv/channel.c index renamed from drivers/hv/channel.c +0 -0

e58cdb7..ae06ba9 100644

drivers/hv/channel.c +16 -0
@@ -614,8 +614,24 @@ int vmbus_sendpacket_ctl(struct vmbus_channel
 *channel, void *buffer,
 
 	ret = hv_ringbuffer_write(&channel->outbound, bufferlist, 3,
 &signal);
 
+	/*
+	 * Here is the logic for signalling the host:
+	 * 1. If the host is already draining the ringbuffer,
+	 *    don't signal. This is indicated by the parameter
+	 *    "signal".
+	 *
+	 * 2. If we are not able to write, signal if kick_q is false.
+	 *    kick_q being false indicates that we may have placed zero or
+	 *    more packets with more packets to come. We will signal in
+	 *    this case even if potentially we may have not placed any
+	 *    packet. This is a rare enough condition that it should not
+	 *    matter.
+	 */
+
 	if ((ret == 0) && kick_q && signal)
 		vmbus_setevent(channel);
+	else if ((ret != 0) && !kick_q)
+		vmbus_setevent(channel);
drivers/hv/channel.c +16 -0
 	return ret;
 }

@@ -705,8 +721,24 @@ int vmbus_sendpacket_pagebuffer_ctl(struct
 vmbus_channel *channel,
 
 	ret = hv_ringbuffer_write(&channel->outbound, bufferlist, 3,
 &signal);
 
+	/*
+	 * Here is the logic for signalling the host:
+	 * 1. If the host is already draining the ringbuffer,
+	 *    don't signal. This is indicated by the parameter
+	 *    "signal".
+	 *
+	 * 2. If we are not able to write, signal if kick_q is false.
+	 *    kick_q being false indicates that we may have placed zero or
+	 *    more packets with more packets to come. We will signal in
+	 *    this case even if potentially we may have not placed any
+	 *    packet. This is a rare enough condition that it should not
+	 *    matter.
+	 */
+
 	if ((ret == 0) && kick_q && signal)
 		vmbus_setevent(channel);
+	else if ((ret != 0) && !kick_q)
+		vmbus_setevent(channel);
Quoted 3 lines

return ret;
}
--

Looks like we need to kick unconditionally here. Consider we may get - EAGAIN when we want to send the last skb (kick_q is true) from the list. We need kick host in this case.

Quoted 2 lines

Btw, another method is let the driver to decide e.g exporting the vmbus_setevent() and call it in netvsc_start_xmit().

There is other state that governs if the host needs to be signaled and I don't think we want to expose that to netvsc. In any case, netvsc will have to signal the host when EGAIN is received. We might as well do it in the vmbus driver.

Thank you Jason, I will respin and resubmit the series.

K. Y

Quoted 3 lines
author
#5 lore raw

RE: [PATCH V2 2/3 net-next] Drivers: hv: vmbus: Fix a bug in the signalling logic with kick_q

Quoted 109 lines

-----Original Message-----
From: devel [mailto:[email protected]] On Behalf Of KY Srinivasan
Sent: Wednesday, March 11, 2015 8:32 PM
To: Jason Wang
Cc: [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]
Subject: RE: [PATCH V2 2/3 net-next] Drivers: hv: vmbus: Fix a bug in the signalling logic with kick_q

Quoted 26 lines

-----Original Message-----
From: Jason Wang [mailto:[email protected]]
Sent: Wednesday, March 11, 2015 8:08 PM
To: KY Srinivasan
Cc: [email protected]; [email protected]; linux- [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; KY Srinivasan
Subject: Re: [PATCH V2 2/3 net-next] Drivers: hv: vmbus: Fix a bug in the signalling logic with kick_q

Quoted 12 lines · K. Y. Srinivasan

When the caller specifies that signalling should be deferred, we need to address the case where we are not able to place the current packet because the buffer is full. In this case, we will signal the host as some packets may have been placed on the ring buffer. I would like to thank Jason Wang <[email protected]> for pointing out this issue.

Signed-off-by K. Y. Srinivasan <[email protected]>
---
 drivers/hv/channel.c |   32 ++++++++++++++++++++++++++++++++
 1 files changed, 32 insertions(+), 0 deletions(-)
Quoted 5 lines

drivers/hv/channel.c int vmbus_sendpacket_ctl(struct

Quoted 5 lines
drivers/hv/channel.c index renamed from drivers/hv/channel.c +0 -0

e58cdb7..ae06ba9 100644

drivers/hv/channel.c +0 -0
@@ -614,8 +614,24 @@ int vmbus_sendpacket_ctl(struct

vmbus_channel

Quoted 59 lines
Quoted 51 lines
drivers/hv/channel.c +16 -0
 *channel, void *buffer,
 
 	ret = hv_ringbuffer_write(&channel->outbound, bufferlist, 3,
 &signal);
 
+	/*
+	 * Here is the logic for signalling the host:
+	 * 1. If the host is already draining the ringbuffer,
+	 *    don't signal. This is indicated by the parameter
+	 *    "signal".
+	 *
+	 * 2. If we are not able to write, signal if kick_q is false.
+	 *    kick_q being false indicates that we may have placed zero or
+	 *    more packets with more packets to come. We will signal in
+	 *    this case even if potentially we may have not placed any
+	 *    packet. This is a rare enough condition that it should not
+	 *    matter.
+	 */
+
 	if ((ret == 0) && kick_q && signal)
 		vmbus_setevent(channel);
+	else if ((ret != 0) && !kick_q)
+		vmbus_setevent(channel);
drivers/hv/channel.c +16 -0
 	return ret;
 }

@@ -705,8 +721,24 @@ int vmbus_sendpacket_pagebuffer_ctl(struct
 vmbus_channel *channel,
 
 	ret = hv_ringbuffer_write(&channel->outbound, bufferlist, 3,
 &signal);
 
+	/*
+	 * Here is the logic for signalling the host:
+	 * 1. If the host is already draining the ringbuffer,
+	 *    don't signal. This is indicated by the parameter
+	 *    "signal".
+	 *
+	 * 2. If we are not able to write, signal if kick_q is false.
+	 *    kick_q being false indicates that we may have placed zero or
+	 *    more packets with more packets to come. We will signal in
+	 *    this case even if potentially we may have not placed any
+	 *    packet. This is a rare enough condition that it should not
+	 *    matter.
+	 */
+
 	if ((ret == 0) && kick_q && signal)
 		vmbus_setevent(channel);
+	else if ((ret != 0) && !kick_q)
+		vmbus_setevent(channel);
Quoted 3 lines

return ret;
}
--

Looks like we need to kick unconditionally here. Consider we may get - EAGAIN when we want to send the last skb (kick_q is true) from the list. We need kick host in this case.

Quoted 2 lines

Btw, another method is let the driver to decide e.g exporting the vmbus_setevent() and call it in netvsc_start_xmit().

Quoted 6 lines

There is other state that governs if the host needs to be signaled and I don't think we want to expose that to netvsc. In any case, netvsc will have to signal the host when EGAIN is received.
We might as well do it in the vmbus driver.

Thank you Jason, I will respin and resubmit the series.

Looks like there may be a way to make the signaling more precise even in this case. I am going to experiment with it.

K. Y

Quoted 10 lines

K. Y

Quoted 3 lines

_______________________________________________
devel mailing list
[email protected]
http://driverdev.linuxdriverproject.org/mailman/listinfo/driverdev-devel

author
#6 lore raw

PATCH v2 3/3 hyperv: Support batched notification

In reply to K. Y. Srinivasan

Optimize notifying the host by deferring notification until there are no more packets to be sent. This will help in batching the requests on the host.

Signed-off-by K. Y. Srinivasan <[email protected]>
---
 drivers/net/hyperv/hyperv_net.h   |    2 +-
 drivers/net/hyperv/netvsc.c       |   14 +++++++++-----
 drivers/net/hyperv/netvsc_drv.c   |    3 ++-
 drivers/net/hyperv/rndis_filter.c |    2 +-
 4 files changed, 13 insertions(+), 8 deletions(-)
drivers/net/hyperv/hyperv_net.h +1 -1
index 4815843..3fd9896 100644
@@ -184,7 +184,7 @@ struct rndis_device {
 int netvsc_device_add(struct hv_device *device, void *additional_info);
 int netvsc_device_remove(struct hv_device *device);
 int netvsc_send(struct hv_device *device,
-		struct hv_netvsc_packet *packet);
+		struct hv_netvsc_packet *packet, bool kick_q);
 void netvsc_linkstatus_callback(struct hv_device *device_obj,
 				struct rndis_message *resp);
 int netvsc_recv_callback(struct hv_device *device_obj,
drivers/net/hyperv/netvsc.c +9 -5
index 208eb05..9003b94 100644
@@ -707,7 +707,7 @@ static u32 netvsc_copy_to_send_buf(struct netvsc_device *net_device,
 }
 
 int netvsc_send(struct hv_device *device,
-			struct hv_netvsc_packet *packet)
+			struct hv_netvsc_packet *packet, bool kick_q)
 {
 	struct netvsc_device *net_device;
 	int ret = 0;

@@ -719,6 +719,7 @@ int netvsc_send(struct hv_device *device,
 	u32 msg_size = 0;
 	struct sk_buff *skb = NULL;
 	u16 q_idx = packet->q_idx;
+	u32 vmbus_flags = VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED;
 
 
 	net_device = get_outbound_net_device(device);

@@ -768,18 +769,21 @@ int netvsc_send(struct hv_device *device,
 		return -ENODEV;
 
 	if (packet->page_buf_cnt) {
-		ret = vmbus_sendpacket_pagebuffer(out_channel,
+		ret = vmbus_sendpacket_pagebuffer_ctl(out_channel,
 						  packet->page_buf,
 						  packet->page_buf_cnt,
 						  &sendMessage,
 						  sizeof(struct nvsp_message),
-						  req_id);
+						  req_id,
+						  vmbus_flags,
+						  kick_q);
 	} else {
-		ret = vmbus_sendpacket(out_channel, &sendMessage,
+		ret = vmbus_sendpacket_ctl(out_channel, &sendMessage,
 				sizeof(struct nvsp_message),
 				req_id,
 				VM_PKT_DATA_INBAND,
-				VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
+				vmbus_flags,
+				kick_q);
 	}
 
 	if (ret == 0) {
drivers/net/hyperv/netvsc_drv.c +2 -1
index a06bd66..80b4b29 100644
@@ -384,6 +384,7 @@ static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
 	u32 net_trans_info;
 	u32 hash;
 	u32 skb_length = skb->len;
+	bool kick_q = !skb->xmit_more;
 
 
 	/* We will atmost need two pages to describe the rndis

@@ -556,7 +557,7 @@ do_send:
 	packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
 					skb, &packet->page_buf[0]);
 
-	ret = netvsc_send(net_device_ctx->device_ctx, packet);
+	ret = netvsc_send(net_device_ctx->device_ctx, packet, kick_q);
 
 drop:
 	if (ret == 0) {
drivers/net/hyperv/rndis_filter.c +1 -1
index ca81de0..05f3792 100644
@@ -238,7 +238,7 @@ static int rndis_filter_send_request(struct rndis_device *dev,
 
 	packet->send_completion = NULL;
 
-	ret = netvsc_send(dev->net_dev->dev, packet);
+	ret = netvsc_send(dev->net_dev->dev, packet, true);
 	return ret;
 }
Signature · 1 line
1.7.4.1
#7 lore raw

Re: [PATCH V2 3/3 net-next] hyperv: Support batched notification

Quoted 88 lines · K. Y. Srinivasan

Optimize notifying the host by deferring notification until there are no more packets to be sent. This will help in batching the requests on the host.

Signed-off-by K. Y. Srinivasan <[email protected]>
---
 drivers/net/hyperv/hyperv_net.h   |    2 +-
 drivers/net/hyperv/netvsc.c       |   14 +++++++++-----
 drivers/net/hyperv/netvsc_drv.c   |    3 ++-
 drivers/net/hyperv/rndis_filter.c |    2 +-
 4 files changed, 13 insertions(+), 8 deletions(-)
drivers/net/hyperv/hyperv_net.h +1 -1
index 4815843..3fd9896 100644
@@ -184,7 +184,7 @@ struct rndis_device {
 int netvsc_device_add(struct hv_device *device, void 
 *additional_info);
 int netvsc_device_remove(struct hv_device *device);
 int netvsc_send(struct hv_device *device,
-		struct hv_netvsc_packet *packet);
+		struct hv_netvsc_packet *packet, bool kick_q);
 void netvsc_linkstatus_callback(struct hv_device *device_obj,
 				struct rndis_message *resp);

int netvsc_recv_callback(struct hv_device *device_obj,

drivers/net/hyperv/netvsc.c +1 -1
index 208eb05..9003b94 100644
@@ -707,7 +707,7 @@ static u32 netvsc_copy_to_send_buf(struct 
 netvsc_device *net_device,
 }
 
 int netvsc_send(struct hv_device *device,
-			struct hv_netvsc_packet *packet)
+			struct hv_netvsc_packet *packet, bool kick_q)
 {
 	struct netvsc_device *net_device;
drivers/net/hyperv/netvsc.c +8 -4
 	int ret = 0;

@@ -719,6 +719,7 @@ int netvsc_send(struct hv_device *device,
 	u32 msg_size = 0;
 	struct sk_buff *skb = NULL;
 	u16 q_idx = packet->q_idx;
+	u32 vmbus_flags = VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED;
 
 
 	net_device = get_outbound_net_device(device);

@@ -768,18 +769,21 @@ int netvsc_send(struct hv_device *device,
 		return -ENODEV;
 
 	if (packet->page_buf_cnt) {
-		ret = vmbus_sendpacket_pagebuffer(out_channel,
+		ret = vmbus_sendpacket_pagebuffer_ctl(out_channel,
 						  packet->page_buf,
 						  packet->page_buf_cnt,
 						  &sendMessage,
 						  sizeof(struct nvsp_message),
-						  req_id);
+						  req_id,
+						  vmbus_flags,
+						  kick_q);
 	} else {
-		ret = vmbus_sendpacket(out_channel, &sendMessage,
+		ret = vmbus_sendpacket_ctl(out_channel, &sendMessage,
 				sizeof(struct nvsp_message),
 				req_id,
 				VM_PKT_DATA_INBAND,
-				VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
+				vmbus_flags,
+				kick_q);
 	}
 
 	if (ret == 0) {
drivers/net/hyperv/netvsc_drv.c +1 -0
index a06bd66..80b4b29 100644
@@ -384,6 +384,7 @@ static int netvsc_start_xmit(struct sk_buff *skb, 
 struct net_device *net)
 	u32 net_trans_info;
 	u32 hash;
 	u32 skb_length = skb->len;
+	bool kick_q = !skb->xmit_more;
drivers/net/hyperv/netvsc_drv.c +0 -0
 	/* We will atmost need two pages to describe the rndis

drivers/net/hyperv/netvsc_drv.c do_send: · line 560

Quoted 5 lines
drivers/net/hyperv/netvsc_drv.c +1 -1
@@ -556,7 +557,7 @@ do_send:
 	packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
 					skb, &packet->page_buf[0]);
 
-	ret = netvsc_send(net_device_ctx->device_ctx, packet);
+	ret = netvsc_send(net_device_ctx->device_ctx, packet, kick_q);

Maybe just a !skb->xmit_more here to save a local variable.

Quoted 25 lines
drivers/net/hyperv/netvsc_drv.c +0 -0
 drop:
 	if (ret == 0) {
drivers/net/hyperv/rndis_filter.c +1 -1
index ca81de0..05f3792 100644
@@ -238,7 +238,7 @@ static int rndis_filter_send_request(struct 
 rndis_device *dev,
 
 	packet->send_completion = NULL;
 
-	ret = netvsc_send(dev->net_dev->dev, packet);
+	ret = netvsc_send(dev->net_dev->dev, packet, true);
 	return ret;
 }
Signature · 7 lines
-- 
1.7.4.1

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
author
#8 lore raw

RE: [PATCH V2 3/3 net-next] hyperv: Support batched notification

Quoted 107 lines

-----Original Message-----
From: Jason Wang [mailto:[email protected]]
Sent: Wednesday, March 11, 2015 8:09 PM
To: KY Srinivasan
Cc: [email protected]; [email protected]; linux- [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; KY Srinivasan
Subject: Re: [PATCH V2 3/3 net-next] hyperv: Support batched notification

Quoted 39 lines · K. Y. Srinivasan

Optimize notifying the host by deferring notification until there are no more packets to be sent. This will help in batching the requests on the host.

Signed-off-by K. Y. Srinivasan <[email protected]>
---
 drivers/net/hyperv/hyperv_net.h   |    2 +-
 drivers/net/hyperv/netvsc.c       |   14 +++++++++-----
 drivers/net/hyperv/netvsc_drv.c   |    3 ++-
 drivers/net/hyperv/rndis_filter.c |    2 +-
 4 files changed, 13 insertions(+), 8 deletions(-)
(quoted diff) +0 -0

b/drivers/net/hyperv/hyperv_net.h index 4815843..3fd9896 100644

drivers/net/hyperv/hyperv_net.h +0 -0
@@ -184,7 +184,7 @@ struct rndis_device {  int
 netvsc_device_add(struct hv_device *device, void *additional_info);

int netvsc_device_remove(struct hv_device *device); int netvsc_send(struct hv_device *device,

drivers/net/hyperv/hyperv_net.h +1 -1
-		struct hv_netvsc_packet *packet);
+		struct hv_netvsc_packet *packet, bool kick_q);
 void netvsc_linkstatus_callback(struct hv_device *device_obj,
 				struct rndis_message *resp);
 int netvsc_recv_callback(struct hv_device *device_obj, diff --git

a/drivers/net/hyperv/netvsc.c b/drivers/net/hyperv/netvsc.c index 208eb05..9003b94 100644

drivers/net/hyperv/netvsc.c +1 -1
@@ -707,7 +707,7 @@ static u32 netvsc_copy_to_send_buf(struct
 netvsc_device *net_device,  }
 
 int netvsc_send(struct hv_device *device,
-			struct hv_netvsc_packet *packet)
+			struct hv_netvsc_packet *packet, bool kick_q)
 {
 	struct netvsc_device *net_device;
 	int ret = 0;

drivers/net/hyperv/netvsc.c int netvsc_send(struct hv_device *device, · line 722

Quoted 4 lines
drivers/net/hyperv/netvsc.c +1 -0
@@ -719,6 +719,7 @@ int netvsc_send(struct hv_device *device,
 	u32 msg_size = 0;
 	struct sk_buff *skb = NULL;
 	u16 q_idx = packet->q_idx;
+	u32 vmbus_flags =

VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED;

drivers/net/hyperv/netvsc.c int netvsc_send(struct hv_device *device, · line 723

Quoted 3 lines
drivers/net/hyperv/netvsc.c +0 -0
 	net_device = get_outbound_net_device(device); @@ -768,18

+769,21 @@

drivers/net/hyperv/netvsc.c int netvsc_send(struct hv_device *device,

Quoted 10 lines
drivers/net/hyperv/netvsc.c +0 -0
 int netvsc_send(struct hv_device *device,
 		return -ENODEV;
drivers/net/hyperv/netvsc.c +1 -1
 	if (packet->page_buf_cnt) {
-		ret = vmbus_sendpacket_pagebuffer(out_channel,
+		ret = vmbus_sendpacket_pagebuffer_ctl(out_channel,
 						  packet->page_buf,
 						  packet->page_buf_cnt,
 						  &sendMessage,
 						  sizeof(struct

nvsp_message),

drivers/net/hyperv/netvsc.c int netvsc_send(struct hv_device *device,

Quoted 11 lines
drivers/net/hyperv/netvsc.c +4 -3
-						  req_id);
+						  req_id,
+						  vmbus_flags,
+						  kick_q);
 	} else {
-		ret = vmbus_sendpacket(out_channel, &sendMessage,
+		ret = vmbus_sendpacket_ctl(out_channel, &sendMessage,
 				sizeof(struct nvsp_message),
 				req_id,
 				VM_PKT_DATA_INBAND,
-

	VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
Quoted 17 lines
drivers/net/hyperv/netvsc.c +2 -0
+				vmbus_flags,
+				kick_q);
 	}
 
 	if (ret == 0) {
(quoted diff) +0 -0

b/drivers/net/hyperv/netvsc_drv.c index a06bd66..80b4b29 100644

drivers/net/hyperv/netvsc_drv.c +1 -0
@@ -384,6 +384,7 @@ static int netvsc_start_xmit(struct sk_buff *skb,
 struct net_device *net)
 	u32 net_trans_info;
 	u32 hash;
 	u32 skb_length = skb->len;
+	bool kick_q = !skb->xmit_more;

drivers/net/hyperv/netvsc_drv.c

Quoted 3 lines
drivers/net/hyperv/netvsc_drv.c +1 -0
 	/* We will atmost need two pages to describe the rndis @@ -556,7
+557,7 @@ do_send:
 	packet->page_buf_cnt = init_page_array(rndis_msg,

rndis_msg_size,

Quoted 6 lines

drivers/net/hyperv/netvsc_drv.c do_send: · line 560

Quoted 4 lines
drivers/net/hyperv/netvsc_drv.c +1 -1
 					skb, &packet->page_buf[0]);
 
-	ret = netvsc_send(net_device_ctx->device_ctx, packet);
+	ret = netvsc_send(net_device_ctx->device_ctx, packet, kick_q);

Maybe just a !skb->xmit_more here to save a local variable.

Will fix it.

K. Y

Acked-by on v2 1/3
#9 lore raw

Re: [PATCH V2 1/3 net-next] Drivers: hv: vmbus: Export the vmbus_sendpacket_pagebuffer_ctl()

In reply to K. Y. Srinivasan
Quoted 20 lines · K. Y. Srinivasan

Export the vmbus_sendpacket_pagebuffer_ctl() interface. This interface will be used in the netvsc driver to optimize signalling the host.

Signed-off-by K. Y. Srinivasan <[email protected]>
---
 drivers/hv/channel.c |    1 +
 1 files changed, 1 insertions(+), 0 deletions(-)
drivers/hv/channel.c +1 -0
index da53180..e58cdb7 100644
@@ -710,6 +710,7 @@ int vmbus_sendpacket_pagebuffer_ctl(struct 
 vmbus_channel *channel,
 
 	return ret;
 }
+EXPORT_SYMBOL_GPL(vmbus_sendpacket_pagebuffer_ctl);
 
 /*
Quoted 3 lines

* vmbus_sendpacket_pagebuffer - Send a range of single-page buffer

Signature · 1 line
1.7.4.1
Acked-by Jason Wang <[email protected]>