login
Search
We provide an zero-copy method which driver side may get external buffers to DMA. Here external means driver don't use kernel space to allocate skb buffers. Currently the external buffer can be from guest virtio-net driver. The idea is simple, just to pin the guest VM user space and then let host NIC driver has the chance to directly DMA to it. The patches are based on vhost-net backend driver. We add a device which provides proto_ops as sendmsg/recvmsg to vhost-net to send/recv directly to/from the NIC driver. KVM guest who use the vhost-net backend may bind any ethX interface in the host side to get copyless data transfer thru guest virtio-net frontend. patch 01-11: net core and kernel changes. patch 12-14: new device as interface to mantpulate external buffers. patch 15: for vhost-net. patch 16: An example on modifying NIC driver to using napi_gro_frags(). patch 17: An example how to get guest buffers based on driver who using napi_gro_frags(). The guest virtio-net driver submits multiple requests thru vhost-net backend driver to the kernel. And the requests are queued and then completed after corresponding actions in h/w are done. For read, user space buffers are dispensed to NIC driver for rx when a page constructor API is invoked. Means NICs can allocate user buffers from a page constructor. We add a hook in netif_receive_skb() function to intercept the incoming packets, and notify the zero-copy device. For write, the zero-copy deivce may allocates a new host skb and puts payload on the skb_shinfo(skb)->frags, and copied the header to skb->data. The request remains pending until the skb is transmitted by h/w. We provide multiple submits and asynchronous notifiicaton to vhost-net too. Our goal is to improve the bandwidth and reduce the CPU usage. Exact performance data will be provided later. What we have not done yet: Performance tuning what we have done in v1: polish the RCU usage deal with write logging in asynchroush mode in vhost add ...
From: Xin Xiaohui <xiaohui.xin@intel.com>
Currently, it can get external buffers from mp device.
Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
Signed-off-by: Zhao Yu <yzhao81new@gmail.com>
Reviewed-by: Jeff Dike <jdike@linux.intel.com>
---
net/core/skbuff.c | 27 +++++++++++++++++++++++++++
1 files changed, 27 insertions(+), 0 deletions(-)
diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index 68e197e..a1018bd 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -261,11 +261,38 @@ struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
}
EXPORT_SYMBOL(__netdev_alloc_skb);
+struct page *netdev_alloc_ext_pages(struct net_device *dev, int npages)
+{
+ struct mp_port *port;
+ struct skb_ext_page *ext_page = NULL;
+
+ port = dev->mp_port;
+ if (!port)
+ goto out;
+ ext_page = port->ctor(port, NULL, npages);
+ if (ext_page)
+ return ext_page->page;
+out:
+ return NULL;
+
+}
+EXPORT_SYMBOL(netdev_alloc_ext_pages);
+
+struct page *netdev_alloc_ext_page(struct net_device *dev)
+{
+ return netdev_alloc_ext_pages(dev, 1);
+
+}
+EXPORT_SYMBOL(netdev_alloc_ext_page);
+
struct page *__netdev_alloc_page(struct net_device *dev, gfp_t gfp_mask)
{
int node = dev->dev.parent ? dev_to_node(dev->dev.parent) : -1;
struct page *page;
+ if (dev_is_mpassthru(dev))
+ return netdev_alloc_ext_page(dev);
+
page = alloc_pages_node(node, gfp_mask, 0);
return page;
}
--
1.7.3
--
From: Xin Xiaohui <xiaohui.xin@intel.com>
If device is in this zero-copy mode first, we cannot handle this,
so fail it. This patch is for this.
If bonding is created first, and one of the device will be in zero-copy
mode, this will be handled by mp device. It will first check if all the
slaves have the zero-copy capability. If no, fail too. Otherwise make
all the slaves in zero-copy mode.
Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
---
drivers/net/bonding/bond_main.c | 4 ++++
1 files changed, 4 insertions(+), 0 deletions(-)
diff --git a/drivers/net/bonding/bond_main.c b/drivers/net/bonding/bond_main.c
index 3b16f62..dfb6a2c 100644
--- a/drivers/net/bonding/bond_main.c
+++ b/drivers/net/bonding/bond_main.c
@@ -1428,6 +1428,10 @@ int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
bond_dev->name);
}
+ /* if the device is in zero-copy mode before bonding, fail it. */
+ if (dev_is_mpassthru(slave_dev))
+ return -EBUSY;
+
/* already enslaved */
if (slave_dev->flags & IFF_SLAVE) {
pr_debug("Error, Device was already enslaved\n");
--
1.7.3
--
From: Xin Xiaohui <xiaohui.xin@intel.com>
Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
Signed-off-by: Zhao Yu <yzhao81new@gmail.com>
Reviewed-by: Jeff Dike <jdike@linux.intel.com>
---
include/linux/mpassthru.h | 133 +++++++++++++++++++++++++++++++++++++++++++++
1 files changed, 133 insertions(+), 0 deletions(-)
create mode 100644 include/linux/mpassthru.h
diff --git a/include/linux/mpassthru.h b/include/linux/mpassthru.h
new file mode 100644
index 0000000..1115f55
--- /dev/null
+++ b/include/linux/mpassthru.h
@@ -0,0 +1,133 @@
+#ifndef __MPASSTHRU_H
+#define __MPASSTHRU_H
+
+#include <linux/types.h>
+#include <linux/if_ether.h>
+#include <linux/ioctl.h>
+
+/* ioctl defines */
+#define MPASSTHRU_BINDDEV _IOW('M', 213, int)
+#define MPASSTHRU_UNBINDDEV _IO('M', 214)
+#define MPASSTHRU_SET_MEM_LOCKED _IOW('M', 215, unsigned long)
+#define MPASSTHRU_GET_MEM_LOCKED_NEED _IOR('M', 216, unsigned long)
+
+#define COPY_THRESHOLD (L1_CACHE_BYTES * 4)
+#define COPY_HDR_LEN (L1_CACHE_BYTES < 64 ? 64 : L1_CACHE_BYTES)
+
+#define DEFAULT_NEED ((8192*2*2)*4096)
+
+struct frag {
+ u16 offset;
+ u16 size;
+};
+
+#define HASH_BUCKETS (8192*2)
+struct page_info {
+ struct list_head list;
+ struct page_info *next;
+ struct page_info *prev;
+ struct page *pages[MAX_SKB_FRAGS];
+ struct sk_buff *skb;
+ struct page_pool *pool;
+
+ /* The pointer relayed to skb, to indicate
+ * it's a external allocated skb or kernel
+ */
+ struct skb_ext_page ext_page;
+ /* flag to indicate read or write */
+#define INFO_READ 0
+#define INFO_WRITE 1
+ unsigned flags;
+ /* exact number of locked pages */
+ unsigned pnum;
+
+ /* The fields after that is for backend
+ * driver, now for vhost-net.
+ */
+ /* the kiocb structure related to */
+ struct kiocb *iocb;
+ /* the ring descriptor index */
+ unsigned int ...From: Xin Xiaohui <xiaohui.xin@intel.com> Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com> Reviewed-by: Jeff Dike <jdike@linux.intel.com> --- drivers/vhost/Kconfig | 10 ++++++++++ drivers/vhost/Makefile | 2 ++ 2 files changed, 12 insertions(+), 0 deletions(-) diff --git a/drivers/vhost/Kconfig b/drivers/vhost/Kconfig index e4e2fd1..a6b8cbf 100644 --- a/drivers/vhost/Kconfig +++ b/drivers/vhost/Kconfig @@ -9,3 +9,13 @@ config VHOST_NET To compile this driver as a module, choose M here: the module will be called vhost_net. +config MEDIATE_PASSTHRU + tristate "mediate passthru network driver (EXPERIMENTAL)" + depends on VHOST_NET + ---help--- + zerocopy network I/O support, we call it as mediate passthru to + be distiguish with hardare passthru. + + To compile this driver as a module, choose M here: the module will + be called mpassthru. + diff --git a/drivers/vhost/Makefile b/drivers/vhost/Makefile index 72dd020..c18b9fc 100644 --- a/drivers/vhost/Makefile +++ b/drivers/vhost/Makefile @@ -1,2 +1,4 @@ obj-$(CONFIG_VHOST_NET) += vhost_net.o vhost_net-y := vhost.o net.o + +obj-$(CONFIG_MEDIATE_PASSTHRU) += mpassthru.o -- 1.7.3 --
From: Xin Xiaohui <xiaohui.xin@intel.com> The patch add mp(mediate passthru) device, which now based on vhost-net backend driver and provides proto_ops to send/receive guest buffers data from/to guest vitio-net driver. It also exports async functions which can be used by other drivers like macvtap to utilize zero-copy too. Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com> Signed-off-by: Zhao Yu <yzhao81new@gmail.com> Reviewed-by: Jeff Dike <jdike@linux.intel.com> --- drivers/vhost/mpassthru.c | 1515 +++++++++++++++++++++++++++++++++++++++++++++ 1 files changed, 1515 insertions(+), 0 deletions(-) create mode 100644 drivers/vhost/mpassthru.c diff --git a/drivers/vhost/mpassthru.c b/drivers/vhost/mpassthru.c new file mode 100644 index 0000000..492430c --- /dev/null +++ b/drivers/vhost/mpassthru.c @@ -0,0 +1,1515 @@ +/* + * MPASSTHRU - Mediate passthrough device. + * Copyright (C) 2009 ZhaoYu, XinXiaohui, Dike, Jeffery G + * + * 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 Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + */ + +#define DRV_NAME "mpassthru" +#define DRV_DESCRIPTION "Mediate passthru device driver" +#define DRV_COPYRIGHT "(C) 2009 ZhaoYu, XinXiaohui, Dike, Jeffery G" + +#include <linux/compat.h> +#include <linux/module.h> +#include <linux/errno.h> +#include <linux/kernel.h> +#include <linux/major.h> +#include <linux/slab.h> +#include <linux/smp_lock.h> +#include <linux/poll.h> +#include <linux/fcntl.h> +#include <linux/init.h> +#include <linux/aio.h> + +#include <linux/skbuff.h> +#include ...
From: Xin Xiaohui <xiaohui.xin@intel.com>
This example is made on ixgbe driver which using napi_gro_frags().
It can get buffers from guest side directly using netdev_alloc_page()
and release guest buffers using netdev_free_page().
Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
---
drivers/net/ixgbe/ixgbe_main.c | 37 +++++++++++++++++++++++++++++++++----
1 files changed, 33 insertions(+), 4 deletions(-)
diff --git a/drivers/net/ixgbe/ixgbe_main.c b/drivers/net/ixgbe/ixgbe_main.c
index a4a5263..9f5598b 100644
--- a/drivers/net/ixgbe/ixgbe_main.c
+++ b/drivers/net/ixgbe/ixgbe_main.c
@@ -1032,7 +1032,14 @@ static inline void ixgbe_release_rx_desc(struct ixgbe_hw *hw,
static bool is_rx_buffer_mapped_as_page(struct ixgbe_rx_buffer *bi,
struct net_device *dev)
{
- return true;
+ return dev_is_mpassthru(dev);
+}
+
+static u32 get_page_skb_offset(struct net_device *dev)
+{
+ if (!dev_is_mpassthru(dev))
+ return 0;
+ return dev->mp_port->vnet_hlen;
}
/**
@@ -1105,7 +1112,8 @@ static void ixgbe_alloc_rx_buffers(struct ixgbe_adapter *adapter,
adapter->alloc_rx_page_failed++;
goto no_buffers;
}
- bi->page_skb_offset = 0;
+ bi->page_skb_offset =
+ get_page_skb_offset(adapter->netdev);
bi->dma = dma_map_page(&pdev->dev, bi->page_skb,
bi->page_skb_offset,
(PAGE_SIZE / 2),
@@ -1242,8 +1250,10 @@ static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
len = le16_to_cpu(rx_desc->wb.upper.length);
}
- if (is_no_buffer(rx_buffer_info))
+ if (is_no_buffer(rx_buffer_info)) {
+ printk("no buffers\n");
break;
+ }
cleaned = true;
if (!rx_buffer_info->mapped_as_page) {
@@ -1299,6 +1309,11 @@ static bool ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
rx_buffer_info->page_skb,
rx_buffer_info->page_skb_offset,
len);
+ if (dev_is_mpassthru(netdev) &&
+ netdev->mp_port->hash)
+ skb_shinfo(skb)->destructor_arg ...From: Xin Xiaohui <xiaohui.xin@intel.com>
This example is made on ixgbe driver.
It provides API is_rx_buffer_mapped_as_page() to indicate
if the driver use napi_gro_frags() interface or not.
The example allocates 2 pages for DMA for one ring descriptor
using netdev_alloc_page(). When packets is coming, using
napi_gro_frags() to allocate skb and to receive the packets.
Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
---
drivers/net/ixgbe/ixgbe.h | 3 +
drivers/net/ixgbe/ixgbe_main.c | 163 +++++++++++++++++++++++++++++++---------
2 files changed, 131 insertions(+), 35 deletions(-)
diff --git a/drivers/net/ixgbe/ixgbe.h b/drivers/net/ixgbe/ixgbe.h
index 9e15eb9..89367ca 100644
--- a/drivers/net/ixgbe/ixgbe.h
+++ b/drivers/net/ixgbe/ixgbe.h
@@ -131,6 +131,9 @@ struct ixgbe_rx_buffer {
struct page *page;
dma_addr_t page_dma;
unsigned int page_offset;
+ u16 mapped_as_page;
+ struct page *page_skb;
+ unsigned int page_skb_offset;
};
struct ixgbe_queue_stats {
diff --git a/drivers/net/ixgbe/ixgbe_main.c b/drivers/net/ixgbe/ixgbe_main.c
index e32af43..a4a5263 100644
--- a/drivers/net/ixgbe/ixgbe_main.c
+++ b/drivers/net/ixgbe/ixgbe_main.c
@@ -1029,6 +1029,12 @@ static inline void ixgbe_release_rx_desc(struct ixgbe_hw *hw,
IXGBE_WRITE_REG(hw, IXGBE_RDT(rx_ring->reg_idx), val);
}
+static bool is_rx_buffer_mapped_as_page(struct ixgbe_rx_buffer *bi,
+ struct net_device *dev)
+{
+ return true;
+}
+
/**
* ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
* @adapter: address of board private structure
@@ -1045,13 +1051,17 @@ static void ixgbe_alloc_rx_buffers(struct ixgbe_adapter *adapter,
i = rx_ring->next_to_use;
bi = &rx_ring->rx_buffer_info[i];
+
while (cleaned_count--) {
rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
+ bi->mapped_as_page =
+ is_rx_buffer_mapped_as_page(bi, adapter->netdev);
+
if (!bi->page_dma &&
(rx_ring->flags & IXGBE_RING_RX_PS_ENABLED)) {
if (!bi->page) ...From: Xin Xiaohui <xiaohui.xin@intel.com>
The vhost-net backend now only supports synchronous send/recv
operations. The patch provides multiple submits and asynchronous
notifications. This is needed for zero-copy case.
Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
---
drivers/vhost/net.c | 355 +++++++++++++++++++++++++++++++++++++++++++++----
drivers/vhost/vhost.c | 78 +++++++++++
drivers/vhost/vhost.h | 15 ++-
3 files changed, 423 insertions(+), 25 deletions(-)
diff --git a/drivers/vhost/net.c b/drivers/vhost/net.c
index 7c80082..17c599a 100644
--- a/drivers/vhost/net.c
+++ b/drivers/vhost/net.c
@@ -24,6 +24,8 @@
#include <linux/if_arp.h>
#include <linux/if_tun.h>
#include <linux/if_macvlan.h>
+#include <linux/mpassthru.h>
+#include <linux/aio.h>
#include <net/sock.h>
@@ -32,6 +34,7 @@
/* Max number of bytes transferred before requeueing the job.
* Using this limit prevents one virtqueue from starving others. */
#define VHOST_NET_WEIGHT 0x80000
+static struct kmem_cache *notify_cache;
enum {
VHOST_NET_VQ_RX = 0,
@@ -49,6 +52,7 @@ struct vhost_net {
struct vhost_dev dev;
struct vhost_virtqueue vqs[VHOST_NET_VQ_MAX];
struct vhost_poll poll[VHOST_NET_VQ_MAX];
+ struct kmem_cache *cache;
/* Tells us whether we are polling a socket for TX.
* We only do this when socket buffer fills up.
* Protected by tx vq lock. */
@@ -109,11 +113,184 @@ static void tx_poll_start(struct vhost_net *net, struct socket *sock)
net->tx_poll_state = VHOST_NET_POLL_STARTED;
}
+struct kiocb *notify_dequeue(struct vhost_virtqueue *vq)
+{
+ struct kiocb *iocb = NULL;
+ unsigned long flags;
+
+ spin_lock_irqsave(&vq->notify_lock, flags);
+ if (!list_empty(&vq->notifier)) {
+ iocb = list_first_entry(&vq->notifier,
+ struct kiocb, ki_list);
+ list_del(&iocb->ki_list);
+ }
+ spin_unlock_irqrestore(&vq->notify_lock, flags);
+ return iocb;
+}
+
+static void handle_iocb(struct kiocb *iocb)
+{
+ struct ...From: Xin Xiaohui <xiaohui.xin@intel.com>
The external buffer owner can use the functions to get
the capability of the underlying NIC driver.
Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
Signed-off-by: Zhao Yu <yzhaonew@gmail.com>
Reviewed-by: Jeff Dike <jdike@linux.intel.com>
---
include/linux/netdevice.h | 2 ++
net/core/dev.c | 41 +++++++++++++++++++++++++++++++++++++++++
2 files changed, 43 insertions(+), 0 deletions(-)
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 575777f..8dcf6de 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -1736,6 +1736,8 @@ extern gro_result_t napi_frags_finish(struct napi_struct *napi,
gro_result_t ret);
extern struct sk_buff * napi_frags_skb(struct napi_struct *napi);
extern gro_result_t napi_gro_frags(struct napi_struct *napi);
+extern int netdev_mp_port_prep(struct net_device *dev,
+ struct mp_port *port);
static inline void napi_free_frags(struct napi_struct *napi)
{
diff --git a/net/core/dev.c b/net/core/dev.c
index 660dd41..84fbb83 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -2942,6 +2942,47 @@ out:
return ret;
}
+/* To support meidate passthru(zero-copy) with NIC driver,
+ * we'd better query NIC driver for the capability it can
+ * provide, especially for packet split mode, now we only
+ * query for the header size, and the payload a descriptor
+ * may carry.
+ * Now, it's only called by mpassthru device.
+ */
+#if defined(CONFIG_MEDIATE_PASSTHRU) || defined(CONFIG_MEDIATE_PASSTHRU_MODULE)
+int netdev_mp_port_prep(struct net_device *dev,
+ struct mp_port *port)
+{
+ int rc;
+ int npages, data_len;
+ const struct net_device_ops *ops = dev->netdev_ops;
+
+ if (ops->ndo_mp_port_prep) {
+ rc = ops->ndo_mp_port_prep(dev, port);
+ if (rc)
+ return rc;
+ } else
+ return -EINVAL;
+
+ if (port->hdr_len <= 0)
+ goto err;
+
+ npages = port->npages;
+ data_len = port->data_len;
+ if (npages <= 0 || npages > ...From: Xin Xiaohui <xiaohui.xin@intel.com>
The hook is called in __netif_receive_skb().
Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
Signed-off-by: Zhao Yu <yzhao81new@gmail.com>
Reviewed-by: Jeff Dike <jdike@linux.intel.com>
---
net/core/dev.c | 40 ++++++++++++++++++++++++++++++++++++++++
1 files changed, 40 insertions(+), 0 deletions(-)
diff --git a/net/core/dev.c b/net/core/dev.c
index 84fbb83..bdad1c8 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -2814,6 +2814,40 @@ int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master)
}
EXPORT_SYMBOL(__skb_bond_should_drop);
+#if defined(CONFIG_MEDIATE_PASSTHRU) || defined(CONFIG_MEDIATE_PASSTHRU_MODULE)
+/* Add a hook to intercept mediate passthru(zero-copy) packets,
+ * and insert it to the socket queue owned by mp_port specially.
+ */
+static inline struct sk_buff *handle_mpassthru(struct sk_buff *skb,
+ struct packet_type **pt_prev,
+ int *ret,
+ struct net_device *orig_dev)
+{
+ struct mp_port *mp_port = NULL;
+ struct sock *sk = NULL;
+
+ if (!dev_is_mpassthru(skb->dev) && !dev_is_mpassthru(orig_dev))
+ return skb;
+ if (dev_is_mpassthru(skb->dev))
+ mp_port = skb->dev->mp_port;
+ else if (orig_dev->master == skb->dev && dev_is_mpassthru(orig_dev))
+ mp_port = orig_dev->mp_port;
+
+ if (*pt_prev) {
+ *ret = deliver_skb(skb, *pt_prev, orig_dev);
+ *pt_prev = NULL;
+ }
+
+ sk = mp_port->sock->sk;
+ skb_queue_tail(&sk->sk_receive_queue, skb);
+ sk->sk_state_change(sk);
+
+ return NULL;
+}
+#else
+#define handle_mpassthru(skb, pt_prev, ret, orig_dev) (skb)
+#endif
+
static int __netif_receive_skb(struct sk_buff *skb)
{
struct packet_type *ptype, *pt_prev;
@@ -2891,6 +2925,11 @@ static int __netif_receive_skb(struct sk_buff *skb)
ncls:
#endif
+ /* To intercept mediate passthru(zero-copy) packets here */
+ skb = handle_mpassthru(skb, &pt_prev, &ret, orig_dev);
+ if (!skb)
+ goto out;
+
/* Handle special case of ...From: Xin Xiaohui <xiaohui.xin@intel.com> Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com> Signed-off-by: Zhao Yu <yzhao81new@gmail.com> Reviewed-by: Jeff Dike <jdike@linux.intel.com> --- net/core/skbuff.c | 6 ++++++ 1 files changed, 6 insertions(+), 0 deletions(-) diff --git a/net/core/skbuff.c b/net/core/skbuff.c index 3d81113..075f4c5 100644 --- a/net/core/skbuff.c +++ b/net/core/skbuff.c @@ -557,6 +557,12 @@ bool skb_recycle_check(struct sk_buff *skb, int skb_size) if (skb_shared(skb) || skb_cloned(skb)) return false; + /* if the device wants to do mediate passthru, the skb may + * get external buffer, so don't recycle + */ + if (dev_is_mpassthru(skb->dev)) + return 0; + skb_release_head_state(skb); shinfo = skb_shinfo(skb); -- 1.7.3 --
From: Xin Xiaohui <xiaohui.xin@intel.com>
Add a structure in structure net_device, the new field is
named as mp_port. It's for mediate passthru (zero-copy).
It contains the capability for the net device driver,
a socket, and an external buffer creator, external means
skb buffer belongs to the device may not be allocated from
kernel space.
Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
Signed-off-by: Zhao Yu <yzhao81new@gmail.com>
Reviewed-by: Jeff Dike <jdike@linux.intel.com>
---
include/linux/netdevice.h | 25 ++++++++++++++++++++++++-
1 files changed, 24 insertions(+), 1 deletions(-)
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 46c36ff..f6b1870 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -325,6 +325,28 @@ enum netdev_state_t {
__LINK_STATE_DORMANT,
};
+/*The structure for mediate passthru(zero-copy). */
+struct mp_port {
+ /* the header len */
+ int hdr_len;
+ /* the max payload len for one descriptor */
+ int data_len;
+ /* the pages for DMA in one time */
+ int npages;
+ /* the socket bind to */
+ struct socket *sock;
+ /* the header len for virtio-net */
+ int vnet_hlen;
+ /* the external buffer page creator */
+ struct skb_ext_page *(*ctor)(struct mp_port *,
+ struct sk_buff *, int);
+ /* the hash function attached to find according
+ * backend ring descriptor info for one external
+ * buffer page.
+ */
+ struct skb_ext_page *(*hash)(struct net_device *,
+ struct page *);
+};
/*
* This structure holds at boot time configured netdevice settings. They
@@ -1045,7 +1067,8 @@ struct net_device {
/* GARP */
struct garp_port *garp_port;
-
+ /* mpassthru */
+ struct mp_port *mp_port;
/* class/net/name entry */
struct device dev;
/* space for optional device, statistics, and wireless sysfs groups */
--
1.7.3
--
From: Xin Xiaohui <xiaohui.xin@intel.com>
Currently, it can get external buffers from mp device.
Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
Signed-off-by: Zhao Yu <yzhao81new@gmail.com>
Reviewed-by: Jeff Dike <jdike@linux.intel.com>
---
include/linux/skbuff.h | 4 +++-
net/core/skbuff.c | 24 ++++++++++++++++++++++++
2 files changed, 27 insertions(+), 1 deletions(-)
diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index 6e1e991..6309ce6 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -1586,9 +1586,11 @@ static inline struct page *netdev_alloc_page(struct net_device *dev)
return __netdev_alloc_page(dev, GFP_ATOMIC);
}
+extern void __netdev_free_page(struct net_device *dev, struct page *page);
+
static inline void netdev_free_page(struct net_device *dev, struct page *page)
{
- __free_page(page);
+ __netdev_free_page(dev, page);
}
/**
diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index a1018bd..3d81113 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -298,6 +298,30 @@ struct page *__netdev_alloc_page(struct net_device *dev, gfp_t gfp_mask)
}
EXPORT_SYMBOL(__netdev_alloc_page);
+void netdev_free_ext_page(struct net_device *dev, struct page *page)
+{
+ struct skb_ext_page *ext_page = NULL;
+ if (dev_is_mpassthru(dev) && dev->mp_port->hash) {
+ ext_page = dev->mp_port->hash(dev, page);
+ if (ext_page)
+ ext_page->dtor(ext_page);
+ else
+ __free_page(page);
+ }
+}
+EXPORT_SYMBOL(netdev_free_ext_page);
+
+void __netdev_free_page(struct net_device *dev, struct page *page)
+{
+ if (dev_is_mpassthru(dev)) {
+ netdev_free_ext_page(dev, page);
+ return;
+ }
+
+ __free_page(page);
+}
+EXPORT_SYMBOL(__netdev_free_page);
+
void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
int size)
{
--
1.7.3
--
From: Xin Xiaohui <xiaohui.xin@intel.com>
If buffer is external, then use the callback to destruct
buffers.
Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
Signed-off-by: Zhao Yu <yzhao81new@gmail.com>
Reviewed-by: Jeff Dike <jdike@linux.intel.com>
---
include/linux/skbuff.h | 7 ++++---
net/core/skbuff.c | 7 +++++++
2 files changed, 11 insertions(+), 3 deletions(-)
diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index 696e690..6e1e991 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -199,14 +199,15 @@ struct skb_shared_info {
struct sk_buff *frag_list;
struct skb_shared_hwtstamps hwtstamps;
+ /* Intermediate layers must ensure that destructor_arg
+ * remains valid until skb destructor */
+ void * destructor_arg;
+
/*
* Warning : all fields before dataref are cleared in __alloc_skb()
*/
atomic_t dataref;
- /* Intermediate layers must ensure that destructor_arg
- * remains valid until skb destructor */
- void * destructor_arg;
/* must be last field, see pskb_expand_head() */
skb_frag_t frags[MAX_SKB_FRAGS];
};
diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index c83b421..68e197e 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -343,6 +343,13 @@ static void skb_release_data(struct sk_buff *skb)
if (skb_has_frags(skb))
skb_drop_fraglist(skb);
+ if (skb->dev && dev_is_mpassthru(skb->dev)) {
+ struct skb_ext_page *ext_page =
+ skb_shinfo(skb)->destructor_arg;
+ if (ext_page && ext_page->dtor)
+ ext_page->dtor(ext_page);
+ }
+
kfree(skb->head);
}
}
--
1.7.3
--
From: Xin Xiaohui <xiaohui.xin@intel.com>
Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
Signed-off-by: Zhao Yu <yzhao81new@gmail.com>
Reviewed-by: Jeff Dike <jdike@linux.intel.com>
---
include/linux/netdevice.h | 5 +++++
1 files changed, 5 insertions(+), 0 deletions(-)
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 8dcf6de..f91d9bb 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -1739,6 +1739,11 @@ extern gro_result_t napi_gro_frags(struct napi_struct *napi);
extern int netdev_mp_port_prep(struct net_device *dev,
struct mp_port *port);
+static inline bool dev_is_mpassthru(struct net_device *dev)
+{
+ return dev && dev->mp_port;
+}
+
static inline void napi_free_frags(struct napi_struct *napi)
{
kfree_skb(napi->skb);
--
1.7.3
--
From: Xin Xiaohui <xiaohui.xin@intel.com>
Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
Signed-off-by: Zhao Yu <yzhao81new@gmail.com>
Reviewed-by: Jeff Dike <jdike@linux.intel.com>
---
include/linux/skbuff.h | 9 +++++++++
1 files changed, 9 insertions(+), 0 deletions(-)
diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index 77eb60d..696e690 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -211,6 +211,15 @@ struct skb_shared_info {
skb_frag_t frags[MAX_SKB_FRAGS];
};
+/* The structure is for a skb which pages may point to
+ * an external buffer, which is not allocated from kernel space.
+ * It also contains a destructor for itself.
+ */
+struct skb_ext_page {
+ struct page *page;
+ void (*dtor)(struct skb_ext_page *);
+};
+
/* We divide dataref into two halves. The higher 16 bits hold references
* to the payload part of skb->data. The lower 16 bits hold references to
* the entire skb->data. A clone of a headerless skb holds the length of
--
1.7.3
--
From: Xin Xiaohui <xiaohui.xin@intel.com>
If the driver want to allocate external buffers,
then it can export it's capability, as the skb
buffer header length, the page length can be DMA, etc.
The external buffers owner may utilize this.
Signed-off-by: Xin Xiaohui <xiaohui.xin@intel.com>
Signed-off-by: Zhao Yu <yzhao81new@gmail.com>
Reviewed-by: Jeff Dike <jdike@linux.intel.com>
---
include/linux/netdevice.h | 10 ++++++++++
1 files changed, 10 insertions(+), 0 deletions(-)
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index f6b1870..575777f 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -723,6 +723,12 @@ struct netdev_rx_queue {
* int (*ndo_set_vf_port)(struct net_device *dev, int vf,
* struct nlattr *port[]);
* int (*ndo_get_vf_port)(struct net_device *dev, int vf, struct sk_buff *skb);
+ *
+ * int (*ndo_mp_port_prep)(struct net_device *dev, struct mp_port *port);
+ * If the driver want to allocate external buffers,
+ * then it can export it's capability, as the skb
+ * buffer header length, the page length can be DMA, etc.
+ * The external buffers owner may utilize this.
*/
#define HAVE_NET_DEVICE_OPS
struct net_device_ops {
@@ -795,6 +801,10 @@ struct net_device_ops {
int (*ndo_fcoe_get_wwn)(struct net_device *dev,
u64 *wwn, int type);
#endif
+#if defined(CONFIG_MEDIATE_PASSTHRU) || defined(CONFIG_MEDIATE_PASSTHRU_MODULE)
+ int (*ndo_mp_port_prep)(struct net_device *dev,
+ struct mp_port *port);
+#endif
};
/*
--
1.7.3
--
From: xiaohui.xin@intel.com 1) Please use kernel-doc style comments to document structure members. 2) The idea to key off of skb->dev in skb_release_data() is fundamentally flawed since many actions can change skb->dev on you, which will end up causing a leak of your external data areas. --
How about this one? If the destructor_arg is not a good candidate,
then I have to add an apparent field in shinfo.
Thanks
Xiaohui
diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index 10ba67d..ad4636e 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -199,14 +199,15 @@ struct skb_shared_info {
struct sk_buff *frag_list;
struct skb_shared_hwtstamps hwtstamps;
+ /* Intermediate layers must ensure that destructor_arg
+ * remains valid until skb destructor */
+ void * destructor_arg;
+
/*
* Warning : all fields before dataref are cleared in __alloc_skb()
*/
atomic_t dataref;
- /* Intermediate layers must ensure that destructor_arg
- * remains valid until skb destructor */
- void * destructor_arg;
/* must be last field, see pskb_expand_head() */
skb_frag_t frags[MAX_SKB_FRAGS];
};
diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index c83b421..eb040f4 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -343,6 +343,13 @@ static void skb_release_data(struct sk_buff *skb)
if (skb_has_frags(skb))
skb_drop_fraglist(skb);
+ if (skb_shinfo(skb)->destructor_arg) {
+ struct skb_ext_page *ext_page =
+ skb_shinfo(skb)->destructor_arg;
+ if (ext_page->dtor)
+ ext_page->dtor(ext_page);
+ }
+
kfree(skb->head);
}
}
--
1.7.3
--
From: xiaohui.xin@intel.com If destructor_arg is actually a net_device pointer or similar, you will need to take a reference count on it or similar. Which means --> good bye performance especially on SMP. You're going to be adding new serialization points and at least two new atomics per packet. --
Do you mean destructor_arg will be consumed by other user? If that case, may I add a new structure member in shinfo? --
How about this? It really needs somewhere to track the external data area, and if something wrong with it, we can also release the data area. We think skb_release_data() is the right place to deal with it. If I understood right, that destructor_arg will be used by other else that why reference count is needed, then how about add a new structure member in shinfo? Thanks --
