[dpdk-dev,v4,2/2] vhost: Add VHOST PMD
Commit Message
The patch introduces a new PMD. This PMD is implemented as thin wrapper
of librte_vhost. It means librte_vhost is also needed to compile the PMD.
The vhost messages will be handled only when a port is started. So start
a port first, then invoke QEMU.
The PMD has 2 parameters.
- iface: The parameter is used to specify a path to connect to a
virtio-net device.
- queues: The parameter is used to specify the number of the queues
virtio-net device has.
(Default: 1)
Here is an example.
$ ./testpmd -c f -n 4 --vdev 'eth_vhost0,iface=/tmp/sock0,queues=1' -- -i
To connect above testpmd, here is qemu command example.
$ qemu-system-x86_64 \
<snip>
-chardev socket,id=chr0,path=/tmp/sock0 \
-netdev vhost-user,id=net0,chardev=chr0,vhostforce,queues=1 \
-device virtio-net-pci,netdev=net0
Signed-off-by: Tetsuya Mukawa <mukawa@igel.co.jp>
---
config/common_linuxapp | 6 +
doc/guides/nics/index.rst | 1 +
doc/guides/rel_notes/release_2_2.rst | 2 +
drivers/net/Makefile | 4 +
drivers/net/vhost/Makefile | 62 +++
drivers/net/vhost/rte_eth_vhost.c | 783 ++++++++++++++++++++++++++++
drivers/net/vhost/rte_eth_vhost.h | 65 +++
drivers/net/vhost/rte_pmd_vhost_version.map | 8 +
mk/rte.app.mk | 8 +-
9 files changed, 938 insertions(+), 1 deletion(-)
create mode 100644 drivers/net/vhost/Makefile
create mode 100644 drivers/net/vhost/rte_eth_vhost.c
create mode 100644 drivers/net/vhost/rte_eth_vhost.h
create mode 100644 drivers/net/vhost/rte_pmd_vhost_version.map
Comments
A quick glimpse and the bug is gone now :)
Will have more test later on.
> -----Original Message-----
> From: Tetsuya Mukawa [mailto:mukawa@igel.co.jp]
> Sent: Friday, November 13, 2015 1:21 PM
> To: dev@dpdk.org; Wang, Zhihong <zhihong.wang@intel.com>; Liu, Yuanhan
> <yuanhan.liu@intel.com>
> Cc: Loftus, Ciara <ciara.loftus@intel.com>; pmatilai@redhat.com;
> ann.zhuangyanying@huawei.com; Richardson, Bruce
> <bruce.richardson@intel.com>; Xie, Huawei <huawei.xie@intel.com>;
> thomas.monjalon@6wind.com; stephen@networkplumber.org;
> rich.lane@bigswitch.com; Tetsuya Mukawa <mukawa@igel.co.jp>
> Subject: [PATCH v4 2/2] vhost: Add VHOST PMD
>
> The patch introduces a new PMD. This PMD is implemented as thin wrapper
> of librte_vhost. It means librte_vhost is also needed to compile the PMD.
> The vhost messages will be handled only when a port is started. So start
> a port first, then invoke QEMU.
>
> The PMD has 2 parameters.
> - iface: The parameter is used to specify a path to connect to a
> virtio-net device.
> - queues: The parameter is used to specify the number of the queues
> virtio-net device has.
> (Default: 1)
>
> Here is an example.
> $ ./testpmd -c f -n 4 --vdev 'eth_vhost0,iface=/tmp/sock0,queues=1' -- -i
>
> To connect above testpmd, here is qemu command example.
>
> $ qemu-system-x86_64 \
> <snip>
> -chardev socket,id=chr0,path=/tmp/sock0 \
> -netdev vhost-user,id=net0,chardev=chr0,vhostforce,queues=1 \
> -device virtio-net-pci,netdev=net0
>
> Signed-off-by: Tetsuya Mukawa <mukawa@igel.co.jp>
> ---
On Fri, Nov 13, 2015 at 02:20:31PM +0900, Tetsuya Mukawa wrote:
....
> +static pthread_mutex_t internal_list_lock = PTHREAD_MUTEX_INITIALIZER;
> +
> +static rte_atomic16_t nb_started_ports;
> +pthread_t session_th;
static?
> +
> +static struct rte_eth_link pmd_link = {
> + .link_speed = 10000,
> + .link_duplex = ETH_LINK_FULL_DUPLEX,
> + .link_status = 0
> +};
> +
> +static uint16_t
> +eth_vhost_rx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
> +{
> + struct vhost_queue *r = q;
> + uint16_t nb_rx = 0;
> +
> + if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
> + return 0;
> +
> + rte_atomic32_set(&r->while_queuing, 1);
> +
> + if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
> + goto out;
> +
> + /* Dequeue packets from guest TX queue */
> + nb_rx = rte_vhost_dequeue_burst(r->device,
> + r->virtqueue_id, r->mb_pool, bufs, nb_bufs);
> +
> + r->rx_pkts += nb_rx;
> +
> +out:
> + rte_atomic32_set(&r->while_queuing, 0);
> +
> + return nb_rx;
> +}
> +
> +static uint16_t
> +eth_vhost_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
> +{
> + struct vhost_queue *r = q;
> + uint16_t i, nb_tx = 0;
> +
> + if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
> + return 0;
> +
> + rte_atomic32_set(&r->while_queuing, 1);
> +
> + if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
> + goto out;
> +
> + /* Enqueue packets to guest RX queue */
> + nb_tx = rte_vhost_enqueue_burst(r->device,
> + r->virtqueue_id, bufs, nb_bufs);
> +
> + r->tx_pkts += nb_tx;
> + r->err_pkts += nb_bufs - nb_tx;
> +
> + for (i = 0; likely(i < nb_tx); i++)
> + rte_pktmbuf_free(bufs[i]);
We should free upto nb_bufs here, but not nb_tx, right?
> +
> +out:
> + rte_atomic32_set(&r->while_queuing, 0);
> +
> + return nb_tx;
> +}
> +
> +static int
> +eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
> +{
> + return 0;
> +}
> +
> +static inline struct pmd_internal *
> +find_internal_resource(char *ifname)
> +{
> + int found = 0;
> + struct pmd_internal *internal;
> +
> + if (ifname == NULL)
> + return NULL;
> +
> + pthread_mutex_lock(&internal_list_lock);
> +
> + TAILQ_FOREACH(internal, &internals_list, next) {
> + if (!strcmp(internal->iface_name, ifname)) {
> + found = 1;
> + break;
> + }
> + }
> +
> + pthread_mutex_unlock(&internal_list_lock);
> +
> + if (!found)
> + return NULL;
> +
> + return internal;
> +}
> +
...
> +static void *vhost_driver_session(void *param __rte_unused)
static void *
vhost_driver_session_start(..)
> +{
> + static struct virtio_net_device_ops *vhost_ops;
> +
> + vhost_ops = rte_zmalloc(NULL, sizeof(*vhost_ops), 0);
> + if (vhost_ops == NULL)
> + rte_panic("Can't allocate memory\n");
Why not making them static?
> +
> + /* set vhost arguments */
> + vhost_ops->new_device = new_device;
> + vhost_ops->destroy_device = destroy_device;
> + if (rte_vhost_driver_pmd_callback_register(vhost_ops) < 0)
> + rte_panic("Can't register callbacks\n");
> +
> + /* start event handling */
> + rte_vhost_driver_session_start();
> +
> + rte_free(vhost_ops);
> + pthread_exit(0);
> +}
> +
> +static void vhost_driver_session_start(void)
ditto.
> +{
> + int ret;
> +
> + ret = pthread_create(&session_th,
> + NULL, vhost_driver_session, NULL);
> + if (ret)
> + rte_panic("Can't create a thread\n");
> +}
> +
> +static void vhost_driver_session_stop(void)
Ditto.
> +{
> + int ret;
> +
> + ret = pthread_cancel(session_th);
> + if (ret)
> + rte_panic("Can't cancel the thread\n");
> +
> + ret = pthread_join(session_th, NULL);
> + if (ret)
> + rte_panic("Can't join the thread\n");
> +}
> +
> +static int
> +eth_dev_start(struct rte_eth_dev *dev)
...
> + internal->nb_rx_queues = queues;
> + internal->nb_tx_queues = queues;
> + internal->dev_name = strdup(name);
> + if (internal->dev_name == NULL)
> + goto error;
> + internal->iface_name = strdup(iface_name);
> + if (internal->iface_name == NULL) {
> + free(internal->dev_name);
> + goto error;
> + }
You still didn't resolve my comments from last email: if allocation
failed here, internal->dev_name is not freed.
> +
> + pthread_mutex_lock(&internal_list_lock);
> + TAILQ_INSERT_TAIL(&internals_list, internal, next);
> + pthread_mutex_unlock(&internal_list_lock);
> +
...
> +static struct rte_driver pmd_vhost_drv = {
> + .name = "eth_vhost",
> + .type = PMD_VDEV,
> + .init = rte_pmd_vhost_devinit,
> + .uninit = rte_pmd_vhost_devuninit,
> +};
> +
> +struct
> +virtio_net *rte_eth_vhost_portid2vdev(uint16_t port_id)
struct virtio_net *
rte_eth_vhost_portid2vdev()
BTW, why making a speical eth API for virtio? This doesn't make too much
sense to me.
Besides those minor nits, this patch looks good to me. Thanks for the
work!
--yliu
On Thu, Nov 12, 2015 at 9:20 PM, Tetsuya Mukawa <mukawa@igel.co.jp> wrote:
> +static uint16_t
> +eth_vhost_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
> +{
>
...
> +
> + /* Enqueue packets to guest RX queue */
> + nb_tx = rte_vhost_enqueue_burst(r->device,
> + r->virtqueue_id, bufs, nb_bufs);
> +
> + r->tx_pkts += nb_tx;
> + r->err_pkts += nb_bufs - nb_tx;
>
I don't think a full TX queue is counted as an error by physical NIC PMDs
like ixgbe and i40e. It is counted as an error by the af_packet, pcap, and
ring PMDs. I'd suggest not counting it as an error because it's a common
and expected condition, and the application might just retry the TX later.
Are the byte counts left out because it would be a performance hit? It
seems like it would be a minimal cost given how much we're already touching
each packet.
> +static int
> +new_device(struct virtio_net *dev)
> +{
>
...
> +
> + if ((dev->virt_qp_nb < internal->nb_rx_queues) ||
> + (dev->virt_qp_nb < internal->nb_tx_queues)) {
> + RTE_LOG(INFO, PMD, "Not enough queues\n");
> + return -1;
> + }
>
Would it make sense to take the minimum of the guest and host queuepairs
and use that below in place of nb_rx_queues/nb_tx_queues? That way the host
can support a large maximum number of queues and each guest can choose how
many it wants to use. The host application will receive vring_state_changed
callbacks for each queue the guest activates.
On 2015/11/20 20:43, Yuanhan Liu wrote:
> On Fri, Nov 13, 2015 at 02:20:31PM +0900, Tetsuya Mukawa wrote:
> ....
>> +static pthread_mutex_t internal_list_lock = PTHREAD_MUTEX_INITIALIZER;
>> +
>> +static rte_atomic16_t nb_started_ports;
>> +pthread_t session_th;
> static?
Hi Yuanhan,
I appreciate your carefully reviewing.
I will fix issues you commented, and submit it again.
I added below 2 comments.
Could you please check it?
>> +
>> +static uint16_t
>> +eth_vhost_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
>> +{
>> + struct vhost_queue *r = q;
>> + uint16_t i, nb_tx = 0;
>> +
>> + if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
>> + return 0;
>> +
>> + rte_atomic32_set(&r->while_queuing, 1);
>> +
>> + if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
>> + goto out;
>> +
>> + /* Enqueue packets to guest RX queue */
>> + nb_tx = rte_vhost_enqueue_burst(r->device,
>> + r->virtqueue_id, bufs, nb_bufs);
>> +
>> + r->tx_pkts += nb_tx;
>> + r->err_pkts += nb_bufs - nb_tx;
>> +
>> + for (i = 0; likely(i < nb_tx); i++)
>> + rte_pktmbuf_free(bufs[i]);
> We should free upto nb_bufs here, but not nb_tx, right?
I guess we don't need to free all packet buffers.
Could you please check l2fwd_send_burst() in l2fwd example?
It seems DPDK application frees packet buffers that failed to send.
>> +static struct rte_driver pmd_vhost_drv = {
>> + .name = "eth_vhost",
>> + .type = PMD_VDEV,
>> + .init = rte_pmd_vhost_devinit,
>> + .uninit = rte_pmd_vhost_devuninit,
>> +};
>> +
>> +struct
>> +virtio_net *rte_eth_vhost_portid2vdev(uint16_t port_id)
> struct virtio_net *
> rte_eth_vhost_portid2vdev()
>
> BTW, why making a speical eth API for virtio? This doesn't make too much
> sense to me.
This is a kind of helper function.
I assume that DPDK applications want to know relation between port_id
and virtio device structure.
But, in "new" callback handler that DPDK application registers,
application can receive virtio device structure, but it doesn't tell
which port is.
To know it, probably here are steps that DPDK application needs to do.
1. Store interface name that is specified when vhost pmd is invoked.
(For example, store information like /tmp/socket0 is for port0, and
/tmp/socket1 is for port1)
2. Compare above interface name and dev->ifname that is stored in virtio
device structure, then DPDK application can know which port is.
If DPDK application uses Port Hotplug, I guess above steps are easy.
But if they don't, interface name will be specified in "--vdev" EAL
command line option.
So probably it's not so easy to handle interface name in DPDK application.
This is why I added the function.
Thanks,
Tetsuya
On Tue, Nov 24, 2015 at 11:48:04AM +0900, Tetsuya Mukawa wrote:
> On 2015/11/20 20:43, Yuanhan Liu wrote:
> > On Fri, Nov 13, 2015 at 02:20:31PM +0900, Tetsuya Mukawa wrote:
> > ....
> >> +static pthread_mutex_t internal_list_lock = PTHREAD_MUTEX_INITIALIZER;
> >> +
> >> +static rte_atomic16_t nb_started_ports;
> >> +pthread_t session_th;
> > static?
>
> Hi Yuanhan,
>
> I appreciate your carefully reviewing.
> I will fix issues you commented, and submit it again.
>
> I added below 2 comments.
> Could you please check it?
>
> >> +
> >> +static uint16_t
> >> +eth_vhost_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
> >> +{
> >> + struct vhost_queue *r = q;
> >> + uint16_t i, nb_tx = 0;
> >> +
> >> + if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
> >> + return 0;
> >> +
> >> + rte_atomic32_set(&r->while_queuing, 1);
> >> +
> >> + if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
> >> + goto out;
> >> +
> >> + /* Enqueue packets to guest RX queue */
> >> + nb_tx = rte_vhost_enqueue_burst(r->device,
> >> + r->virtqueue_id, bufs, nb_bufs);
> >> +
> >> + r->tx_pkts += nb_tx;
> >> + r->err_pkts += nb_bufs - nb_tx;
> >> +
> >> + for (i = 0; likely(i < nb_tx); i++)
> >> + rte_pktmbuf_free(bufs[i]);
> > We should free upto nb_bufs here, but not nb_tx, right?
>
> I guess we don't need to free all packet buffers.
> Could you please check l2fwd_send_burst() in l2fwd example?
> It seems DPDK application frees packet buffers that failed to send.
Yes, you are right. I was thinking it's just a vhost app, and forgot
that this is for rte_eth_tx_burst, sigh ...
>
> >> +static struct rte_driver pmd_vhost_drv = {
> >> + .name = "eth_vhost",
> >> + .type = PMD_VDEV,
> >> + .init = rte_pmd_vhost_devinit,
> >> + .uninit = rte_pmd_vhost_devuninit,
> >> +};
> >> +
> >> +struct
> >> +virtio_net *rte_eth_vhost_portid2vdev(uint16_t port_id)
> > struct virtio_net *
> > rte_eth_vhost_portid2vdev()
> >
> > BTW, why making a speical eth API for virtio? This doesn't make too much
> > sense to me.
>
> This is a kind of helper function.
Yeah, I know that. I was thinking that an API prefixed with rte_eth_
should be a common interface for all eth drivers. Here this one is
for vhost PMD only, though.
I then had a quick check of DPDK code, and found a similar example,
bond, such as rte_eth_bond_create(). So, it might be okay to introduce
PMD specific eth APIs?
Anyway, I would suggest you to put it into another patch, so that
it can be reworked (or even dropped) if someone else doesn't like
it (or doesn't think it's necessary).
--yliu
>
> I assume that DPDK applications want to know relation between port_id
> and virtio device structure.
> But, in "new" callback handler that DPDK application registers,
> application can receive virtio device structure, but it doesn't tell
> which port is.
>
> To know it, probably here are steps that DPDK application needs to do.
>
> 1. Store interface name that is specified when vhost pmd is invoked.
> (For example, store information like /tmp/socket0 is for port0, and
> /tmp/socket1 is for port1)
> 2. Compare above interface name and dev->ifname that is stored in virtio
> device structure, then DPDK application can know which port is.
>
> If DPDK application uses Port Hotplug, I guess above steps are easy.
> But if they don't, interface name will be specified in "--vdev" EAL
> command line option.
> So probably it's not so easy to handle interface name in DPDK application.
> This is why I added the function.
>
> Thanks,
> Tetsuya
On 2015/11/24 12:40, Yuanhan Liu wrote:
> On Tue, Nov 24, 2015 at 11:48:04AM +0900, Tetsuya Mukawa wrote:
>> On 2015/11/20 20:43, Yuanhan Liu wrote:
>>> On Fri, Nov 13, 2015 at 02:20:31PM +0900, Tetsuya Mukawa wrote:
>>> ....
>>>> +static pthread_mutex_t internal_list_lock = PTHREAD_MUTEX_INITIALIZER;
>>>> +
>>>> +static rte_atomic16_t nb_started_ports;
>>>> +pthread_t session_th;
>>> static?
>> Hi Yuanhan,
>>
>> I appreciate your carefully reviewing.
>> I will fix issues you commented, and submit it again.
>>
>> I added below 2 comments.
>> Could you please check it?
>>
>>>> +
>>>> +static uint16_t
>>>> +eth_vhost_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
>>>> +{
>>>> + struct vhost_queue *r = q;
>>>> + uint16_t i, nb_tx = 0;
>>>> +
>>>> + if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
>>>> + return 0;
>>>> +
>>>> + rte_atomic32_set(&r->while_queuing, 1);
>>>> +
>>>> + if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
>>>> + goto out;
>>>> +
>>>> + /* Enqueue packets to guest RX queue */
>>>> + nb_tx = rte_vhost_enqueue_burst(r->device,
>>>> + r->virtqueue_id, bufs, nb_bufs);
>>>> +
>>>> + r->tx_pkts += nb_tx;
>>>> + r->err_pkts += nb_bufs - nb_tx;
>>>> +
>>>> + for (i = 0; likely(i < nb_tx); i++)
>>>> + rte_pktmbuf_free(bufs[i]);
>>> We should free upto nb_bufs here, but not nb_tx, right?
>> I guess we don't need to free all packet buffers.
>> Could you please check l2fwd_send_burst() in l2fwd example?
>> It seems DPDK application frees packet buffers that failed to send.
> Yes, you are right. I was thinking it's just a vhost app, and forgot
> that this is for rte_eth_tx_burst, sigh ...
>
>>>> +static struct rte_driver pmd_vhost_drv = {
>>>> + .name = "eth_vhost",
>>>> + .type = PMD_VDEV,
>>>> + .init = rte_pmd_vhost_devinit,
>>>> + .uninit = rte_pmd_vhost_devuninit,
>>>> +};
>>>> +
>>>> +struct
>>>> +virtio_net *rte_eth_vhost_portid2vdev(uint16_t port_id)
>>> struct virtio_net *
>>> rte_eth_vhost_portid2vdev()
>>>
>>> BTW, why making a speical eth API for virtio? This doesn't make too much
>>> sense to me.
>> This is a kind of helper function.
> Yeah, I know that. I was thinking that an API prefixed with rte_eth_
> should be a common interface for all eth drivers. Here this one is
> for vhost PMD only, though.
>
> I then had a quick check of DPDK code, and found a similar example,
> bond, such as rte_eth_bond_create(). So, it might be okay to introduce
> PMD specific eth APIs?
Yes, I guess so.
>
> Anyway, I would suggest you to put it into another patch, so that
> it can be reworked (or even dropped) if someone else doesn't like
> it (or doesn't think it's necessary).
Sure, it's nice idea.
I will split the patch.
Tetsuya
> --yliu
>
>> I assume that DPDK applications want to know relation between port_id
>> and virtio device structure.
>> But, in "new" callback handler that DPDK application registers,
>> application can receive virtio device structure, but it doesn't tell
>> which port is.
>>
>> To know it, probably here are steps that DPDK application needs to do.
>>
>> 1. Store interface name that is specified when vhost pmd is invoked.
>> (For example, store information like /tmp/socket0 is for port0, and
>> /tmp/socket1 is for port1)
>> 2. Compare above interface name and dev->ifname that is stored in virtio
>> device structure, then DPDK application can know which port is.
>>
>> If DPDK application uses Port Hotplug, I guess above steps are easy.
>> But if they don't, interface name will be specified in "--vdev" EAL
>> command line option.
>> So probably it's not so easy to handle interface name in DPDK application.
>> This is why I added the function.
>>
>> Thanks,
>> Tetsuya
On 2015/11/21 9:15, Rich Lane wrote:
> On Thu, Nov 12, 2015 at 9:20 PM, Tetsuya Mukawa <mukawa@igel.co.jp> wrote:
>
>> +static uint16_t
>> +eth_vhost_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
>> +{
>>
> ...
>
>> +
>> + /* Enqueue packets to guest RX queue */
>> + nb_tx = rte_vhost_enqueue_burst(r->device,
>> + r->virtqueue_id, bufs, nb_bufs);
>> +
>> + r->tx_pkts += nb_tx;
>> + r->err_pkts += nb_bufs - nb_tx;
>>
> I don't think a full TX queue is counted as an error by physical NIC PMDs
> like ixgbe and i40e. It is counted as an error by the af_packet, pcap, and
> ring PMDs. I'd suggest not counting it as an error because it's a common
> and expected condition, and the application might just retry the TX later.
Hi Rich,
Thanks for commenting.
I will count it as "imissed".
> Are the byte counts left out because it would be a performance hit? It
> seems like it would be a minimal cost given how much we're already touching
> each packet.
I just ignore it for performance.
But you are correct, I will add it.
>
>> +static int
>> +new_device(struct virtio_net *dev)
>> +{
>>
> ...
>
>> +
>> + if ((dev->virt_qp_nb < internal->nb_rx_queues) ||
>> + (dev->virt_qp_nb < internal->nb_tx_queues)) {
>> + RTE_LOG(INFO, PMD, "Not enough queues\n");
>> + return -1;
>> + }
>>
> Would it make sense to take the minimum of the guest and host queuepairs
> and use that below in place of nb_rx_queues/nb_tx_queues? That way the host
> can support a large maximum number of queues and each guest can choose how
> many it wants to use. The host application will receive vring_state_changed
> callbacks for each queue the guest activates.
>
Thanks for checking this.
I agree with you.
After reading your comment, here is my guess for this PMD behavior.
This PMD should assume that virtio-net device(QEMU) has same or more
queues than specified in vhost PMD option.
In a case that the assumption is break, application should handle
vring_state_change callback correctly.
(Then stop accessing to disabled queues not to spend CPU power.)
Anyway, I will just remove above if-condition, because of above PMD
assumption.
Thanks,
Tetsuya
The patch introduces a new PMD. This PMD is implemented as thin wrapper
of librte_vhost.
PATCH v5 changes:
- Rebase on latest master.
- Fix RX/TX routine to count RX/TX bytes.
- Fix RX/TX routine not to count as error packets if enqueue/dequeue
cannot send all packets.
- Fix if-condition checking for multiqueues.
- Add "static" to pthread variable.
- Fix format.
- Change default behavior not to receive queueing event from driver.
- Split the patch to separate rte_eth_vhost_portid2vdev().
PATCH v4 changes:
- Rebase on latest DPDK tree.
- Fix cording style.
- Fix code not to invoke multiple messaging handling threads.
- Fix code to handle vdev parameters correctly.
- Remove needless cast.
- Remove needless if-condition before rt_free().
PATCH v3 changes:
- Rebase on latest matser
- Specify correct queue_id in RX/TX function.
PATCH v2 changes:
- Remove a below patch that fixes vhost library.
The patch was applied as a separate patch.
- vhost: fix crash with multiqueue enabled
- Fix typos.
(Thanks to Thomas, Monjalon)
- Rebase on latest tree with above bernard's patches.
PATCH v1 changes:
- Support vhost multiple queues.
- Rebase on "remove pci driver from vdevs".
- Optimize RX/TX functions.
- Fix resource leaks.
- Fix compile issue.
- Add patch to fix vhost library.
RFC PATCH v3 changes:
- Optimize performance.
In RX/TX functions, change code to access only per core data.
- Add below API to allow user to use vhost library APIs for a port managed
by vhost PMD. There are a few limitations. See "rte_eth_vhost.h".
- rte_eth_vhost_portid2vdev()
To support this functionality, vhost library is also changed.
Anyway, if users doesn't use vhost PMD, can fully use vhost library APIs.
- Add code to support vhost multiple queues.
Actually, multiple queues functionality is not enabled so far.
RFC PATCH v2 changes:
- Fix issues reported by checkpatch.pl
(Thanks to Stephen Hemminger)
Tetsuya Mukawa (3):
vhost: Add callback and private data for vhost PMD
vhost: Add VHOST PMD
vhost: Add helper function to convert port id to virtio device pointer
config/common_linuxapp | 6 +
doc/guides/nics/index.rst | 1 +
doc/guides/rel_notes/release_2_2.rst | 2 +
drivers/net/Makefile | 4 +
drivers/net/vhost/Makefile | 62 ++
drivers/net/vhost/rte_eth_vhost.c | 796 ++++++++++++++++++++++++++
drivers/net/vhost/rte_eth_vhost.h | 65 +++
drivers/net/vhost/rte_pmd_vhost_version.map | 8 +
lib/librte_vhost/rte_vhost_version.map | 6 +
lib/librte_vhost/rte_virtio_net.h | 3 +
lib/librte_vhost/vhost_user/virtio-net-user.c | 13 +-
lib/librte_vhost/virtio-net.c | 60 +-
lib/librte_vhost/virtio-net.h | 4 +-
mk/rte.app.mk | 8 +-
14 files changed, 1024 insertions(+), 14 deletions(-)
create mode 100644 drivers/net/vhost/Makefile
create mode 100644 drivers/net/vhost/rte_eth_vhost.c
create mode 100644 drivers/net/vhost/rte_eth_vhost.h
create mode 100644 drivers/net/vhost/rte_pmd_vhost_version.map
Hi Xie and Yuanhan,
Please let me make sure whether this patch is differed.
If it is differed, I guess I may need to add ABI breakage notice before
releasing DPDK-2.2, because the patches changes virtio_net structure.
Tetsuya,
On 2015/11/24 18:00, Tetsuya Mukawa wrote:
> The patch introduces a new PMD. This PMD is implemented as thin wrapper
> of librte_vhost.
>
> PATCH v5 changes:
> - Rebase on latest master.
> - Fix RX/TX routine to count RX/TX bytes.
> - Fix RX/TX routine not to count as error packets if enqueue/dequeue
> cannot send all packets.
> - Fix if-condition checking for multiqueues.
> - Add "static" to pthread variable.
> - Fix format.
> - Change default behavior not to receive queueing event from driver.
> - Split the patch to separate rte_eth_vhost_portid2vdev().
>
> PATCH v4 changes:
> - Rebase on latest DPDK tree.
> - Fix cording style.
> - Fix code not to invoke multiple messaging handling threads.
> - Fix code to handle vdev parameters correctly.
> - Remove needless cast.
> - Remove needless if-condition before rt_free().
>
> PATCH v3 changes:
> - Rebase on latest matser
> - Specify correct queue_id in RX/TX function.
>
> PATCH v2 changes:
> - Remove a below patch that fixes vhost library.
> The patch was applied as a separate patch.
> - vhost: fix crash with multiqueue enabled
> - Fix typos.
> (Thanks to Thomas, Monjalon)
> - Rebase on latest tree with above bernard's patches.
>
> PATCH v1 changes:
> - Support vhost multiple queues.
> - Rebase on "remove pci driver from vdevs".
> - Optimize RX/TX functions.
> - Fix resource leaks.
> - Fix compile issue.
> - Add patch to fix vhost library.
>
> RFC PATCH v3 changes:
> - Optimize performance.
> In RX/TX functions, change code to access only per core data.
> - Add below API to allow user to use vhost library APIs for a port managed
> by vhost PMD. There are a few limitations. See "rte_eth_vhost.h".
> - rte_eth_vhost_portid2vdev()
> To support this functionality, vhost library is also changed.
> Anyway, if users doesn't use vhost PMD, can fully use vhost library APIs.
> - Add code to support vhost multiple queues.
> Actually, multiple queues functionality is not enabled so far.
>
> RFC PATCH v2 changes:
> - Fix issues reported by checkpatch.pl
> (Thanks to Stephen Hemminger)
>
>
> Tetsuya Mukawa (3):
> vhost: Add callback and private data for vhost PMD
> vhost: Add VHOST PMD
> vhost: Add helper function to convert port id to virtio device pointer
>
> config/common_linuxapp | 6 +
> doc/guides/nics/index.rst | 1 +
> doc/guides/rel_notes/release_2_2.rst | 2 +
> drivers/net/Makefile | 4 +
> drivers/net/vhost/Makefile | 62 ++
> drivers/net/vhost/rte_eth_vhost.c | 796 ++++++++++++++++++++++++++
> drivers/net/vhost/rte_eth_vhost.h | 65 +++
> drivers/net/vhost/rte_pmd_vhost_version.map | 8 +
> lib/librte_vhost/rte_vhost_version.map | 6 +
> lib/librte_vhost/rte_virtio_net.h | 3 +
> lib/librte_vhost/vhost_user/virtio-net-user.c | 13 +-
> lib/librte_vhost/virtio-net.c | 60 +-
> lib/librte_vhost/virtio-net.h | 4 +-
> mk/rte.app.mk | 8 +-
> 14 files changed, 1024 insertions(+), 14 deletions(-)
> create mode 100644 drivers/net/vhost/Makefile
> create mode 100644 drivers/net/vhost/rte_eth_vhost.c
> create mode 100644 drivers/net/vhost/rte_eth_vhost.h
> create mode 100644 drivers/net/vhost/rte_pmd_vhost_version.map
>
On Tue, Dec 08, 2015 at 10:12:52AM +0900, Tetsuya Mukawa wrote:
> Hi Xie and Yuanhan,
>
> Please let me make sure whether this patch is differed.
> If it is differed, I guess I may need to add ABI breakage notice before
Tetsuya,
What do you mean by "differed"? Do you mean "delayed"?
Per my understanding, it's a bit late for v2.2 (even at few weeks
before). On the other hand, I'm still waiting for comments from
Huawei, for there are still one or two issues need more discussion.
> releasing DPDK-2.2, because the patches changes virtio_net structure.
I had sent a patch (which is just applied by Thomas) for reserving
some spaces for both virtio_net and vhost_virtqueue structure, so
it will not break anything if you simply add few more fields :)
--yliu
On 2015/12/08 11:03, Yuanhan Liu wrote:
> On Tue, Dec 08, 2015 at 10:12:52AM +0900, Tetsuya Mukawa wrote:
>> Hi Xie and Yuanhan,
>>
>> Please let me make sure whether this patch is differed.
>> If it is differed, I guess I may need to add ABI breakage notice before
> Tetsuya,
>
> What do you mean by "differed"? Do you mean "delayed"?
Hi Yuanhan,
I just guess the patch will not be merged in DPDK-2.2.
> Per my understanding, it's a bit late for v2.2 (even at few weeks
> before). On the other hand, I'm still waiting for comments from
> Huawei, for there are still one or two issues need more discussion.
Yes, I agree with you.
>> releasing DPDK-2.2, because the patches changes virtio_net structure.
> I had sent a patch (which is just applied by Thomas) for reserving
> some spaces for both virtio_net and vhost_virtqueue structure, so
> it will not break anything if you simply add few more fields :)
Sounds great idea!
Thanks for handling virtio things.
Tetsuya,
@@ -461,6 +461,12 @@ CONFIG_RTE_LIBRTE_VHOST_NUMA=n
CONFIG_RTE_LIBRTE_VHOST_DEBUG=n
#
+# Compile vhost PMD
+# To compile, CONFIG_RTE_LIBRTE_VHOST should be enabled.
+#
+CONFIG_RTE_LIBRTE_PMD_VHOST=y
+
+#
#Compile Xen domain0 support
#
CONFIG_RTE_LIBRTE_XEN_DOM0=n
@@ -47,6 +47,7 @@ Network Interface Controller Drivers
mlx4
mlx5
virtio
+ vhost
vmxnet3
pcap_ring
@@ -94,6 +94,8 @@ New Features
* **Added vhost-user multiple queue support.**
+* **Added vhost PMD.**
+
* **Added port hotplug support to vmxnet3.**
* **Added port hotplug support to xenvirt.**
@@ -50,5 +50,9 @@ DIRS-$(CONFIG_RTE_LIBRTE_VIRTIO_PMD) += virtio
DIRS-$(CONFIG_RTE_LIBRTE_VMXNET3_PMD) += vmxnet3
DIRS-$(CONFIG_RTE_LIBRTE_PMD_XENVIRT) += xenvirt
+ifeq ($(CONFIG_RTE_LIBRTE_VHOST),y)
+DIRS-$(CONFIG_RTE_LIBRTE_PMD_VHOST) += vhost
+endif # $(CONFIG_RTE_LIBRTE_VHOST)
+
include $(RTE_SDK)/mk/rte.sharelib.mk
include $(RTE_SDK)/mk/rte.subdir.mk
new file mode 100644
@@ -0,0 +1,62 @@
+# BSD LICENSE
+#
+# Copyright (c) 2010-2015 Intel Corporation.
+# All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions
+# are met:
+#
+# * Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+# * Redistributions in binary form must reproduce the above copyright
+# notice, this list of conditions and the following disclaimer in
+# the documentation and/or other materials provided with the
+# distribution.
+# * Neither the name of Intel corporation nor the names of its
+# contributors may be used to endorse or promote products derived
+# from this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+include $(RTE_SDK)/mk/rte.vars.mk
+
+#
+# library name
+#
+LIB = librte_pmd_vhost.a
+
+CFLAGS += -O3
+CFLAGS += $(WERROR_FLAGS)
+
+EXPORT_MAP := rte_pmd_vhost_version.map
+
+LIBABIVER := 1
+
+#
+# all source are stored in SRCS-y
+#
+SRCS-$(CONFIG_RTE_LIBRTE_PMD_VHOST) += rte_eth_vhost.c
+
+#
+# Export include files
+#
+SYMLINK-y-include += rte_eth_vhost.h
+
+# this lib depends upon:
+DEPDIRS-$(CONFIG_RTE_LIBRTE_PMD_VHOST) += lib/librte_mbuf
+DEPDIRS-$(CONFIG_RTE_LIBRTE_PMD_VHOST) += lib/librte_ether
+DEPDIRS-$(CONFIG_RTE_LIBRTE_PMD_VHOST) += lib/librte_kvargs
+DEPDIRS-$(CONFIG_RTE_LIBRTE_PMD_VHOST) += lib/librte_vhost
+
+include $(RTE_SDK)/mk/rte.lib.mk
new file mode 100644
@@ -0,0 +1,783 @@
+/*-
+ * BSD LICENSE
+ *
+ * Copyright (c) 2015 IGEL Co., Ltd.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *
+ * * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ * * Redistributions in binary form must reproduce the above copyright
+ * notice, this list of conditions and the following disclaimer in
+ * the documentation and/or other materials provided with the
+ * distribution.
+ * * Neither the name of IGEL Co.,Ltd. nor the names of its
+ * contributors may be used to endorse or promote products derived
+ * from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+#include <unistd.h>
+#include <pthread.h>
+
+#include <rte_mbuf.h>
+#include <rte_ethdev.h>
+#include <rte_malloc.h>
+#include <rte_memcpy.h>
+#include <rte_dev.h>
+#include <rte_kvargs.h>
+#include <rte_virtio_net.h>
+
+#include "rte_eth_vhost.h"
+
+#define ETH_VHOST_IFACE_ARG "iface"
+#define ETH_VHOST_QUEUES_ARG "queues"
+
+static const char *drivername = "VHOST PMD";
+
+static const char *valid_arguments[] = {
+ ETH_VHOST_IFACE_ARG,
+ ETH_VHOST_QUEUES_ARG,
+ NULL
+};
+
+static struct ether_addr base_eth_addr = {
+ .addr_bytes = {
+ 0x56 /* V */,
+ 0x48 /* H */,
+ 0x4F /* O */,
+ 0x53 /* S */,
+ 0x54 /* T */,
+ 0x00
+ }
+};
+
+struct vhost_queue {
+ rte_atomic32_t allow_queuing;
+ rte_atomic32_t while_queuing;
+ struct virtio_net *device;
+ struct pmd_internal *internal;
+ struct rte_mempool *mb_pool;
+ uint16_t virtqueue_id;
+ uint64_t rx_pkts;
+ uint64_t tx_pkts;
+ uint64_t err_pkts;
+};
+
+struct pmd_internal {
+ TAILQ_ENTRY(pmd_internal) next;
+ char *dev_name;
+ char *iface_name;
+ unsigned nb_rx_queues;
+ unsigned nb_tx_queues;
+
+ struct vhost_queue *rx_vhost_queues[RTE_MAX_QUEUES_PER_PORT];
+ struct vhost_queue *tx_vhost_queues[RTE_MAX_QUEUES_PER_PORT];
+
+ volatile uint16_t once;
+};
+
+TAILQ_HEAD(pmd_internal_head, pmd_internal);
+static struct pmd_internal_head internals_list =
+ TAILQ_HEAD_INITIALIZER(internals_list);
+
+static pthread_mutex_t internal_list_lock = PTHREAD_MUTEX_INITIALIZER;
+
+static rte_atomic16_t nb_started_ports;
+pthread_t session_th;
+
+static struct rte_eth_link pmd_link = {
+ .link_speed = 10000,
+ .link_duplex = ETH_LINK_FULL_DUPLEX,
+ .link_status = 0
+};
+
+static uint16_t
+eth_vhost_rx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
+{
+ struct vhost_queue *r = q;
+ uint16_t nb_rx = 0;
+
+ if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
+ return 0;
+
+ rte_atomic32_set(&r->while_queuing, 1);
+
+ if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
+ goto out;
+
+ /* Dequeue packets from guest TX queue */
+ nb_rx = rte_vhost_dequeue_burst(r->device,
+ r->virtqueue_id, r->mb_pool, bufs, nb_bufs);
+
+ r->rx_pkts += nb_rx;
+
+out:
+ rte_atomic32_set(&r->while_queuing, 0);
+
+ return nb_rx;
+}
+
+static uint16_t
+eth_vhost_tx(void *q, struct rte_mbuf **bufs, uint16_t nb_bufs)
+{
+ struct vhost_queue *r = q;
+ uint16_t i, nb_tx = 0;
+
+ if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
+ return 0;
+
+ rte_atomic32_set(&r->while_queuing, 1);
+
+ if (unlikely(rte_atomic32_read(&r->allow_queuing) == 0))
+ goto out;
+
+ /* Enqueue packets to guest RX queue */
+ nb_tx = rte_vhost_enqueue_burst(r->device,
+ r->virtqueue_id, bufs, nb_bufs);
+
+ r->tx_pkts += nb_tx;
+ r->err_pkts += nb_bufs - nb_tx;
+
+ for (i = 0; likely(i < nb_tx); i++)
+ rte_pktmbuf_free(bufs[i]);
+
+out:
+ rte_atomic32_set(&r->while_queuing, 0);
+
+ return nb_tx;
+}
+
+static int
+eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
+{
+ return 0;
+}
+
+static inline struct pmd_internal *
+find_internal_resource(char *ifname)
+{
+ int found = 0;
+ struct pmd_internal *internal;
+
+ if (ifname == NULL)
+ return NULL;
+
+ pthread_mutex_lock(&internal_list_lock);
+
+ TAILQ_FOREACH(internal, &internals_list, next) {
+ if (!strcmp(internal->iface_name, ifname)) {
+ found = 1;
+ break;
+ }
+ }
+
+ pthread_mutex_unlock(&internal_list_lock);
+
+ if (!found)
+ return NULL;
+
+ return internal;
+}
+
+static int
+new_device(struct virtio_net *dev)
+{
+ struct rte_eth_dev *eth_dev;
+ struct pmd_internal *internal;
+ struct vhost_queue *vq;
+ unsigned i;
+
+ if (dev == NULL) {
+ RTE_LOG(INFO, PMD, "Invalid argument\n");
+ return -1;
+ }
+
+ internal = find_internal_resource(dev->ifname);
+ if (internal == NULL) {
+ RTE_LOG(INFO, PMD, "Invalid device name\n");
+ return -1;
+ }
+
+ if ((dev->virt_qp_nb < internal->nb_rx_queues) ||
+ (dev->virt_qp_nb < internal->nb_tx_queues)) {
+ RTE_LOG(INFO, PMD, "Not enough queues\n");
+ return -1;
+ }
+
+ eth_dev = rte_eth_dev_allocated(internal->dev_name);
+ if (eth_dev == NULL) {
+ RTE_LOG(INFO, PMD, "Failed to find a ethdev\n");
+ return -1;
+ }
+
+ for (i = 0; i < internal->nb_rx_queues; i++) {
+ vq = internal->rx_vhost_queues[i];
+ if (vq == NULL)
+ continue;
+ vq->device = dev;
+ vq->internal = internal;
+ }
+ for (i = 0; i < internal->nb_tx_queues; i++) {
+ vq = internal->tx_vhost_queues[i];
+ if (vq == NULL)
+ continue;
+ vq->device = dev;
+ vq->internal = internal;
+ }
+
+ dev->flags |= VIRTIO_DEV_RUNNING;
+ dev->pmd_priv = eth_dev;
+ eth_dev->data->dev_link.link_status = 1;
+
+ for (i = 0; i < internal->nb_rx_queues; i++) {
+ vq = internal->rx_vhost_queues[i];
+ if (vq == NULL)
+ continue;
+ rte_atomic32_set(&vq->allow_queuing, 1);
+ }
+ for (i = 0; i < internal->nb_tx_queues; i++) {
+ vq = internal->tx_vhost_queues[i];
+ if (vq == NULL)
+ continue;
+ rte_atomic32_set(&vq->allow_queuing, 1);
+ }
+ RTE_LOG(INFO, PMD, "New connection established\n");
+
+ return 0;
+}
+
+static void
+destroy_device(volatile struct virtio_net *dev)
+{
+ struct rte_eth_dev *eth_dev;
+ struct pmd_internal *internal;
+ struct vhost_queue *vq;
+ unsigned i;
+
+ if (dev == NULL) {
+ RTE_LOG(INFO, PMD, "Invalid argument\n");
+ return;
+ }
+
+ eth_dev = (struct rte_eth_dev *)dev->pmd_priv;
+ if (eth_dev == NULL) {
+ RTE_LOG(INFO, PMD, "Failed to find a ethdev\n");
+ return;
+ }
+
+ internal = eth_dev->data->dev_private;
+
+ /* Wait until rx/tx_pkt_burst stops accessing vhost device */
+ for (i = 0; i < internal->nb_rx_queues; i++) {
+ vq = internal->rx_vhost_queues[i];
+ if (vq == NULL)
+ continue;
+ rte_atomic32_set(&vq->allow_queuing, 0);
+ while (rte_atomic32_read(&vq->while_queuing))
+ rte_pause();
+ }
+ for (i = 0; i < internal->nb_tx_queues; i++) {
+ vq = internal->tx_vhost_queues[i];
+ if (vq == NULL)
+ continue;
+ rte_atomic32_set(&vq->allow_queuing, 0);
+ while (rte_atomic32_read(&vq->while_queuing))
+ rte_pause();
+ }
+
+ eth_dev->data->dev_link.link_status = 0;
+
+ dev->pmd_priv = NULL;
+ dev->flags &= ~VIRTIO_DEV_RUNNING;
+
+ for (i = 0; i < internal->nb_rx_queues; i++) {
+ vq = internal->rx_vhost_queues[i];
+ if (vq == NULL)
+ continue;
+ vq->device = NULL;
+ }
+ for (i = 0; i < internal->nb_tx_queues; i++) {
+ vq = internal->tx_vhost_queues[i];
+ if (vq == NULL)
+ continue;
+ vq->device = NULL;
+ }
+
+ RTE_LOG(INFO, PMD, "Connection closed\n");
+}
+
+static void *vhost_driver_session(void *param __rte_unused)
+{
+ static struct virtio_net_device_ops *vhost_ops;
+
+ vhost_ops = rte_zmalloc(NULL, sizeof(*vhost_ops), 0);
+ if (vhost_ops == NULL)
+ rte_panic("Can't allocate memory\n");
+
+ /* set vhost arguments */
+ vhost_ops->new_device = new_device;
+ vhost_ops->destroy_device = destroy_device;
+ if (rte_vhost_driver_pmd_callback_register(vhost_ops) < 0)
+ rte_panic("Can't register callbacks\n");
+
+ /* start event handling */
+ rte_vhost_driver_session_start();
+
+ rte_free(vhost_ops);
+ pthread_exit(0);
+}
+
+static void vhost_driver_session_start(void)
+{
+ int ret;
+
+ ret = pthread_create(&session_th,
+ NULL, vhost_driver_session, NULL);
+ if (ret)
+ rte_panic("Can't create a thread\n");
+}
+
+static void vhost_driver_session_stop(void)
+{
+ int ret;
+
+ ret = pthread_cancel(session_th);
+ if (ret)
+ rte_panic("Can't cancel the thread\n");
+
+ ret = pthread_join(session_th, NULL);
+ if (ret)
+ rte_panic("Can't join the thread\n");
+}
+
+static int
+eth_dev_start(struct rte_eth_dev *dev)
+{
+ int ret;
+ struct pmd_internal *internal = dev->data->dev_private;
+
+ if (rte_atomic16_cmpset(&internal->once, 0, 1)) {
+ ret = rte_vhost_driver_register(internal->iface_name);
+ if (ret)
+ return ret;
+ }
+
+ /* We need only one message handling thread */
+ if (rte_atomic16_add_return(&nb_started_ports, 1) == 1)
+ vhost_driver_session_start();
+
+ return 0;
+}
+
+static void
+eth_dev_stop(struct rte_eth_dev *dev)
+{
+ struct pmd_internal *internal = dev->data->dev_private;
+
+ if (rte_atomic16_cmpset(&internal->once, 1, 0))
+ rte_vhost_driver_unregister(internal->iface_name);
+
+ if (rte_atomic16_sub_return(&nb_started_ports, 1) == 0)
+ vhost_driver_session_stop();
+}
+
+static int
+eth_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
+ uint16_t nb_rx_desc __rte_unused,
+ unsigned int socket_id,
+ const struct rte_eth_rxconf *rx_conf __rte_unused,
+ struct rte_mempool *mb_pool)
+{
+ struct pmd_internal *internal = dev->data->dev_private;
+ struct vhost_queue *vq;
+
+ rte_free(internal->rx_vhost_queues[rx_queue_id]);
+
+ vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
+ RTE_CACHE_LINE_SIZE, socket_id);
+ if (vq == NULL) {
+ RTE_LOG(ERR, PMD, "Failed to allocate memory for rx queue\n");
+ return -ENOMEM;
+ }
+
+ vq->mb_pool = mb_pool;
+ vq->virtqueue_id = rx_queue_id * VIRTIO_QNUM + VIRTIO_TXQ;
+ internal->rx_vhost_queues[rx_queue_id] = vq;
+ dev->data->rx_queues[rx_queue_id] = vq;
+ return 0;
+}
+
+static int
+eth_tx_queue_setup(struct rte_eth_dev *dev, uint16_t tx_queue_id,
+ uint16_t nb_tx_desc __rte_unused,
+ unsigned int socket_id,
+ const struct rte_eth_txconf *tx_conf __rte_unused)
+{
+ struct pmd_internal *internal = dev->data->dev_private;
+ struct vhost_queue *vq;
+
+ rte_free(internal->tx_vhost_queues[tx_queue_id]);
+
+ vq = rte_zmalloc_socket(NULL, sizeof(struct vhost_queue),
+ RTE_CACHE_LINE_SIZE, socket_id);
+ if (vq == NULL) {
+ RTE_LOG(ERR, PMD, "Failed to allocate memory for tx queue\n");
+ return -ENOMEM;
+ }
+
+ vq->virtqueue_id = tx_queue_id * VIRTIO_QNUM + VIRTIO_RXQ;
+ internal->tx_vhost_queues[tx_queue_id] = vq;
+ dev->data->tx_queues[tx_queue_id] = vq;
+ return 0;
+}
+
+
+static void
+eth_dev_info(struct rte_eth_dev *dev,
+ struct rte_eth_dev_info *dev_info)
+{
+ struct pmd_internal *internal = dev->data->dev_private;
+
+ dev_info->driver_name = drivername;
+ dev_info->max_mac_addrs = 1;
+ dev_info->max_rx_pktlen = (uint32_t)-1;
+ dev_info->max_rx_queues = (uint16_t)internal->nb_rx_queues;
+ dev_info->max_tx_queues = (uint16_t)internal->nb_tx_queues;
+ dev_info->min_rx_bufsize = 0;
+}
+
+static void
+eth_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *igb_stats)
+{
+ unsigned i;
+ unsigned long rx_total = 0, tx_total = 0, tx_err_total = 0;
+ const struct pmd_internal *internal = dev->data->dev_private;
+
+ for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
+ i < internal->nb_rx_queues; i++) {
+ if (internal->rx_vhost_queues[i] == NULL)
+ continue;
+ igb_stats->q_ipackets[i] = internal->rx_vhost_queues[i]->rx_pkts;
+ rx_total += igb_stats->q_ipackets[i];
+ }
+
+ for (i = 0; i < RTE_ETHDEV_QUEUE_STAT_CNTRS &&
+ i < internal->nb_tx_queues; i++) {
+ if (internal->tx_vhost_queues[i] == NULL)
+ continue;
+ igb_stats->q_opackets[i] = internal->tx_vhost_queues[i]->tx_pkts;
+ igb_stats->q_errors[i] = internal->tx_vhost_queues[i]->err_pkts;
+ tx_total += igb_stats->q_opackets[i];
+ tx_err_total += igb_stats->q_errors[i];
+ }
+
+ igb_stats->ipackets = rx_total;
+ igb_stats->opackets = tx_total;
+ igb_stats->oerrors = tx_err_total;
+}
+
+static void
+eth_stats_reset(struct rte_eth_dev *dev)
+{
+ unsigned i;
+ struct pmd_internal *internal = dev->data->dev_private;
+
+ for (i = 0; i < internal->nb_rx_queues; i++) {
+ if (internal->rx_vhost_queues[i] == NULL)
+ continue;
+ internal->rx_vhost_queues[i]->rx_pkts = 0;
+ }
+ for (i = 0; i < internal->nb_tx_queues; i++) {
+ if (internal->tx_vhost_queues[i] == NULL)
+ continue;
+ internal->tx_vhost_queues[i]->tx_pkts = 0;
+ internal->tx_vhost_queues[i]->err_pkts = 0;
+ }
+}
+
+static void
+eth_queue_release(void *q __rte_unused)
+{
+ return;
+}
+
+static int
+eth_link_update(struct rte_eth_dev *dev __rte_unused,
+ int wait_to_complete __rte_unused)
+{
+ return 0;
+}
+
+static const struct eth_dev_ops ops = {
+ .dev_start = eth_dev_start,
+ .dev_stop = eth_dev_stop,
+ .dev_configure = eth_dev_configure,
+ .dev_infos_get = eth_dev_info,
+ .rx_queue_setup = eth_rx_queue_setup,
+ .tx_queue_setup = eth_tx_queue_setup,
+ .rx_queue_release = eth_queue_release,
+ .tx_queue_release = eth_queue_release,
+ .link_update = eth_link_update,
+ .stats_get = eth_stats_get,
+ .stats_reset = eth_stats_reset,
+};
+
+static int
+eth_dev_vhost_create(const char *name, int index,
+ char *iface_name,
+ int16_t queues,
+ const unsigned numa_node)
+{
+ struct rte_eth_dev_data *data = NULL;
+ struct pmd_internal *internal = NULL;
+ struct rte_eth_dev *eth_dev = NULL;
+ struct ether_addr *eth_addr = NULL;
+
+ RTE_LOG(INFO, PMD, "Creating VHOST-USER backend on numa socket %u\n",
+ numa_node);
+
+ /* now do all data allocation - for eth_dev structure, dummy pci driver
+ * and internal (private) data
+ */
+ data = rte_zmalloc_socket(name, sizeof(*data), 0, numa_node);
+ if (data == NULL)
+ goto error;
+
+ internal = rte_zmalloc_socket(name, sizeof(*internal), 0, numa_node);
+ if (internal == NULL)
+ goto error;
+
+ eth_addr = rte_zmalloc_socket(name, sizeof(*eth_addr), 0, numa_node);
+ if (eth_addr == NULL)
+ goto error;
+ *eth_addr = base_eth_addr;
+ eth_addr->addr_bytes[5] = index;
+
+ /* reserve an ethdev entry */
+ eth_dev = rte_eth_dev_allocate(name, RTE_ETH_DEV_VIRTUAL);
+ if (eth_dev == NULL)
+ goto error;
+
+ /* now put it all together
+ * - store queue data in internal,
+ * - store numa_node info in ethdev data
+ * - point eth_dev_data to internals
+ * - and point eth_dev structure to new eth_dev_data structure
+ */
+ internal->nb_rx_queues = queues;
+ internal->nb_tx_queues = queues;
+ internal->dev_name = strdup(name);
+ if (internal->dev_name == NULL)
+ goto error;
+ internal->iface_name = strdup(iface_name);
+ if (internal->iface_name == NULL) {
+ free(internal->dev_name);
+ goto error;
+ }
+
+ pthread_mutex_lock(&internal_list_lock);
+ TAILQ_INSERT_TAIL(&internals_list, internal, next);
+ pthread_mutex_unlock(&internal_list_lock);
+
+ data->dev_private = internal;
+ data->port_id = eth_dev->data->port_id;
+ memmove(data->name, eth_dev->data->name, sizeof(data->name));
+ data->nb_rx_queues = queues;
+ data->nb_tx_queues = queues;
+ data->dev_link = pmd_link;
+ data->mac_addrs = eth_addr;
+
+ /* We'll replace the 'data' originally allocated by eth_dev. So the
+ * vhost PMD resources won't be shared between multi processes.
+ */
+ eth_dev->data = data;
+ eth_dev->dev_ops = &ops;
+ eth_dev->driver = NULL;
+ eth_dev->data->dev_flags = RTE_ETH_DEV_DETACHABLE;
+ eth_dev->data->kdrv = RTE_KDRV_NONE;
+ eth_dev->data->drv_name = internal->dev_name;
+ eth_dev->data->numa_node = numa_node;
+
+ /* finally assign rx and tx ops */
+ eth_dev->rx_pkt_burst = eth_vhost_rx;
+ eth_dev->tx_pkt_burst = eth_vhost_tx;
+
+ return data->port_id;
+
+error:
+ rte_free(data);
+ rte_free(internal);
+ rte_free(eth_addr);
+
+ return -1;
+}
+
+static inline int
+open_iface(const char *key __rte_unused, const char *value, void *extra_args)
+{
+ const char **iface_name = extra_args;
+
+ if (value == NULL)
+ return -1;
+
+ *iface_name = value;
+
+ return 0;
+}
+
+static inline int
+open_queues(const char *key __rte_unused, const char *value, void *extra_args)
+{
+ uint16_t *q = extra_args;
+
+ if ((value == NULL) || (extra_args == NULL))
+ return -EINVAL;
+
+ *q = (uint16_t)strtoul(value, NULL, 0);
+ if ((*q == USHRT_MAX) && (errno == ERANGE))
+ return -1;
+
+ if (*q > RTE_MAX_QUEUES_PER_PORT)
+ return -1;
+
+ return 0;
+}
+
+static int
+rte_pmd_vhost_devinit(const char *name, const char *params)
+{
+ struct rte_kvargs *kvlist = NULL;
+ int ret = 0;
+ int index;
+ char *iface_name;
+ uint16_t queues;
+
+ RTE_LOG(INFO, PMD, "Initializing pmd_vhost for %s\n", name);
+
+ if (strlen(name) < strlen("eth_vhost"))
+ return -1;
+
+ index = strtol(name + strlen("eth_vhost"), NULL, 0);
+ if (errno == ERANGE)
+ return -1;
+
+ kvlist = rte_kvargs_parse(params, valid_arguments);
+ if (kvlist == NULL)
+ return -1;
+
+ if (rte_kvargs_count(kvlist, ETH_VHOST_IFACE_ARG) == 1) {
+ ret = rte_kvargs_process(kvlist, ETH_VHOST_IFACE_ARG,
+ &open_iface, &iface_name);
+ if (ret < 0)
+ goto out_free;
+ } else {
+ ret = -1;
+ goto out_free;
+ }
+
+ if (rte_kvargs_count(kvlist, ETH_VHOST_QUEUES_ARG) == 1) {
+ ret = rte_kvargs_process(kvlist, ETH_VHOST_QUEUES_ARG,
+ &open_queues, &queues);
+ if (ret < 0)
+ goto out_free;
+
+ } else
+ queues = 1;
+
+ eth_dev_vhost_create(name, index,
+ iface_name, queues, rte_socket_id());
+
+out_free:
+ rte_kvargs_free(kvlist);
+ return ret;
+}
+
+static int
+rte_pmd_vhost_devuninit(const char *name)
+{
+ struct rte_eth_dev *eth_dev = NULL;
+ struct pmd_internal *internal;
+ unsigned int i;
+
+ RTE_LOG(INFO, PMD, "Un-Initializing pmd_vhost for %s\n", name);
+
+ if (name == NULL)
+ return -EINVAL;
+
+ /* find an ethdev entry */
+ eth_dev = rte_eth_dev_allocated(name);
+ if (eth_dev == NULL)
+ return -ENODEV;
+
+ internal = eth_dev->data->dev_private;
+
+ pthread_mutex_lock(&internal_list_lock);
+ TAILQ_REMOVE(&internals_list, internal, next);
+ pthread_mutex_unlock(&internal_list_lock);
+
+ eth_dev_stop(eth_dev);
+
+ if ((internal) && (internal->dev_name))
+ free(internal->dev_name);
+ if ((internal) && (internal->iface_name))
+ free(internal->iface_name);
+
+ rte_free(eth_dev->data->mac_addrs);
+ rte_free(eth_dev->data);
+
+ for (i = 0; i < internal->nb_rx_queues; i++)
+ rte_free(internal->rx_vhost_queues[i]);
+ for (i = 0; i < internal->nb_tx_queues; i++)
+ rte_free(internal->tx_vhost_queues[i]);
+ rte_free(internal);
+
+ rte_eth_dev_release_port(eth_dev);
+
+ return 0;
+}
+
+static struct rte_driver pmd_vhost_drv = {
+ .name = "eth_vhost",
+ .type = PMD_VDEV,
+ .init = rte_pmd_vhost_devinit,
+ .uninit = rte_pmd_vhost_devuninit,
+};
+
+struct
+virtio_net *rte_eth_vhost_portid2vdev(uint16_t port_id)
+{
+ struct rte_eth_dev *eth_dev;
+
+ if (rte_eth_dev_is_valid_port(port_id) == 0)
+ return NULL;
+
+ eth_dev = &rte_eth_devices[port_id];
+ if (strncmp("eth_vhost", eth_dev->data->drv_name,
+ strlen("eth_vhost")) == 0) {
+ struct pmd_internal *internal;
+ struct vhost_queue *vq;
+
+ internal = eth_dev->data->dev_private;
+ vq = internal->rx_vhost_queues[0];
+ if ((vq != NULL) && (vq->device != NULL))
+ return vq->device;
+ }
+
+ return NULL;
+}
+
+PMD_REGISTER_DRIVER(pmd_vhost_drv);
new file mode 100644
@@ -0,0 +1,65 @@
+/*-
+ * BSD LICENSE
+ *
+ * Copyright(c) 2015 IGEL Co., Ltd.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *
+ * * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ * * Redistributions in binary form must reproduce the above copyright
+ * notice, this list of conditions and the following disclaimer in
+ * the documentation and/or other materials provided with the
+ * distribution.
+ * * Neither the name of IGEL Co., Ltd. nor the names of its
+ * contributors may be used to endorse or promote products derived
+ * from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef _RTE_ETH_AF_PACKET_H_
+#define _RTE_ETH_AF_PACKET_H_
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include <rte_virtio_net.h>
+
+/**
+ * The function convert specified port_id to virtio device structure.
+ * The retured device can be used for vhost library APIs.
+ * To use vhost library APIs and vhost PMD parallely, below API should
+ * not be called, because the API will be called by vhost PMD.
+ * - rte_vhost_driver_session_start()
+ * Once a device is managed by vhost PMD, below API should not be called.
+ * - rte_vhost_driver_unregister()
+ * To unregister the device, call Port Hotplug APIs.
+ *
+ * @param port_id
+ * port number
+ * @return
+ * virtio net device structure corresponding to the specified port
+ * NULL will be returned in error cases.
+ */
+struct virtio_net *rte_eth_vhost_portid2vdev(uint16_t port_id);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif
new file mode 100644
@@ -0,0 +1,8 @@
+DPDK_2.2 {
+
+ global:
+
+ rte_eth_vhost_portid2vdev;
+
+ local: *;
+};
@@ -148,7 +148,13 @@ _LDLIBS-$(CONFIG_RTE_LIBRTE_PMD_PCAP) += -lrte_pmd_pcap
_LDLIBS-$(CONFIG_RTE_LIBRTE_PMD_AF_PACKET) += -lrte_pmd_af_packet
_LDLIBS-$(CONFIG_RTE_LIBRTE_PMD_NULL) += -lrte_pmd_null
-endif # ! $(CONFIG_RTE_BUILD_SHARED_LIB)
+ifeq ($(CONFIG_RTE_LIBRTE_VHOST),y)
+
+_LDLIBS-$(CONFIG_RTE_LIBRTE_PMD_VHOST) += -lrte_pmd_vhost
+
+endif # ! $(CONFIG_RTE_LIBRTE_VHOST)
+
+endif # $(CONFIG_RTE_BUILD_SHARED_LIB)
endif # ! CONFIG_RTE_BUILD_COMBINE_LIBS