[dpdk-dev] DPDK 2.2 build failing with vhost-kni

[dpdk-dev] DPDK 2.2 build failing with vhost-kni

[chintu hetam]

I am compiling DPDK 2.2 on Fedora 23 and i am seeing following build error
/home/vcr/devel/dpdk/dpdk-2.2.0/build/build/lib/librte_eal/linuxapp/kni/kni_vhost.c:
In function ‘kni_sock_poll’:
/home/vcr/devel/dpdk/dpdk-2.2.0/build/build/lib/librte_eal/linuxapp/kni/kni_vhost.c:254:25:
error: ‘SOCK_ASYNC_NOSPACE’ undeclared (first use in this function)
      (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock->flags) &&
                         ^
/home/vcr/devel/dpdk/dpdk-2.2.0/build/build/lib/librte_eal/linuxapp/kni/kni_vhost.c:254:25:
note: each undeclared identifier is reported only once for each function it
appears in
/home/vcr/devel/dpdk/dpdk-2.2.0/build/build/lib/librte_eal/linuxapp/kni/kni_vhost.c:
In function ‘kni_sk_write_space’:
/home/vcr/devel/dpdk/dpdk-2.2.0/build/build/lib/librte_eal/linuxapp/kni/kni_vhost.c:622:26:
error: ‘SOCK_ASYNC_NOSPACE’ undeclared (first use in this function)
      !test_and_clear_bit(SOCK_ASYNC_NOSPACE,
                          ^
/home/vcr/devel/dpdk/dpdk-2.2.0/build/build/lib/librte_eal/linuxapp/kni/kni_vhost.c:
In function ‘kni_vhost_backend_init’:
/home/vcr/devel/dpdk/dpdk-2.2.0/build/build/lib/librte_eal/linuxapp/kni/kni_vhost.c:669:38:
error: too few arguments to function ‘sk_alloc’
  if (!(q = (struct kni_vhost_queue *)sk_alloc(
                                      ^
In file included from
/home/vcr/devel/dpdk/dpdk-2.2.0/build/build/lib/librte_eal/linuxapp/kni/kni_vhost.c:27:0:
/usr/src/kernels/4.4.6-300.fc23.x86_64/include/net/sock.h:1511:14: note:
declared here
 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
              ^
/usr/src/kernels/4.4.6-300.fc23.x86_64/scripts/Makefile.build:258: recipe
for target
'/home/vcr/devel/dpdk/dpdk-2.2.0/build/build/lib/librte_eal/linuxapp/kni/kni_vhost.o'
failed
make[8]: ***
[/home/vcr/devel/dpdk/dpdk-2.2.0/build/build/lib/librte_eal/linuxapp/kni/kni_vhost.o]
Error 1
/usr/src/kernels/4.4.6-300.fc23.x86_64/Makefile:1388: recipe for target
'_module_/home/vcr/devel/dpdk/dpdk-2.2.0/build/build/lib/librte_eal/linuxapp/kni'
failed
make[7]: ***
[_module_/home/vcr/devel/dpdk/dpdk-2.2.0/build/build/lib/librte_eal/linuxapp/kni]
Error 2
Makefile:146: recipe for target 'sub-make' failed
make[6]: *** [sub-make] Error 2
/home/vcr/devel/dpdk/dpdk-2.2.0/mk/rte.module.mk:79: recipe for target
'rte_kni.ko' failed
make[5]: *** [rte_kni.ko] Error 2
/home/vcr/devel/dpdk/dpdk-2.2.0/mk/rte.subdir.mk:61: recipe for target
'kni' failed
make[4]: *** [kni] Error 2
/home/vcr/devel/dpdk/dpdk-2.2.0/mk/rte.subdir.mk:61: recipe for target
'linuxapp' failed
make[3]: *** [linuxapp] Error 2
/home/vcr/devel/dpdk/dpdk-2.2.0/mk/rte.subdir.mk:61: recipe for target
'librte_eal' failed
make[2]: *** [librte_eal] Error 2
/home/vcr/devel/dpdk/dpdk-2.2.0/mk/rte.sdkbuild.mk:77: recipe for target
'lib' failed
make[1]: *** [lib] Error 2
/home/vcr/devel/dpdk/dpdk-2.2.0/mk/rte.sdkroot.mk:123: recipe for target
'all' failed
make: *** [all] Error 2


it's confirmed it's failing after
1) CONFIG_RTE_KNI_VHOST=y a
2) make config T=x86_64-native-linuxapp-gcc
3) make

[root@localhost dpdk-2.2.0]# uname -a
Linux localhost.localdomain 4.4.6-300.fc23.x86_64 #1 SMP Wed Mar 16
22:10:37 UTC 2016 x86_64 x86_64 x86_64 GNU/Linux

Re: [dpdk-dev] DPDK 2.2 build failing with vhost-kni

[Xie, Huawei]

On 4/10/2016 7:26 AM, chintu hetam wrote:
> I am compiling DPDK 2.2 on Fedora 23 and i am seeing following build error
> /home/vcr/devel/dpdk/dpdk-2.2.0/build/build/lib/librte_eal/linuxapp/kni/kni_vhost.c:
> In function ‘kni_sock_poll’:
> /home/vcr/devel/dpdk/dpdk-2.2.0/build/build/lib/librte_eal/linuxapp/kni/kni_vhost.c:254:25:
> error: ‘SOCK_ASYNC_NOSPACE’ undeclared (first use in this function)
>       (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock->flags) &&
>                          ^

Hi, besides the issue, now user space vhost is the preferred way to
switch packets with the virtual machine, and we have plans to remove kni
vhost support. Do you have any reason to use kni vhost?

Re: [dpdk-dev] DPDK 2.2 build failing with vhost-kni

[chintu hetam]

Thanks Xie.

I am trying to test FreeBSD-9.3-virtio as guest. Somewhere in the forum i
found virtio-kni combination reaching around 2.7 Gbps performance, which is
enough for my test, though i dint find equivalent driver performance
characterization for qemu-vhost user space combination.
Also, as per nics-2.2 user guide, "Linux kernel with KVM module; vhost
module loaded and ioeventfd supported. Qemu standard backend without vhost
support isn’t tested, and probably isn’t supported",which is bit different
from vhost user space support,but still..

Just to be sure vhost user space = qemu virtio backend- tap-linux-bridge
configuration, as per nics guide, right?

Thanking in advance, again.







On Mon, Apr 11, 2016 at 4:25 AM, Xie, Huawei <huawei.xie@intel.com> wrote:

> On 4/10/2016 7:26 AM, chintu hetam wrote:
> > I am compiling DPDK 2.2 on Fedora 23 and i am seeing following build
> error
> >
> /home/vcr/devel/dpdk/dpdk-2.2.0/build/build/lib/librte_eal/linuxapp/kni/kni_vhost.c:
> > In function ‘kni_sock_poll’:
> >
> /home/vcr/devel/dpdk/dpdk-2.2.0/build/build/lib/librte_eal/linuxapp/kni/kni_vhost.c:254:25:
> > error: ‘SOCK_ASYNC_NOSPACE’ undeclared (first use in this function)
> >       (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock->flags) &&
> >                          ^
>
> Hi, besides the issue, now user space vhost is the preferred way to
> switch packets with the virtual machine, and we have plans to remove kni
> vhost support. Do you have any reason to use kni vhost?
>

Re: [dpdk-dev] DPDK 2.2 build failing with vhost-kni

[chintu hetam]

I tried vhost-user mode, and during init of ./vhost-switch it fails
with
PMD: eth_ixgbe_dev_init(): port 0 vendorID=0x8086 deviceID=0x1528
pf queue num: 0, configured vmdq pool num: 64, each vmdq pool has 2 queues
EAL: Error - exiting with code: 1
  Cause: Cannot initialize network ports

it's failing in port_init function, and it seems from
http://dpdk.org/doc/guides-1.8/rel_notes/resolved_issues.html

this has to with ixgbe checking for eeprom pba parameters. is that correct,
coz rebuilding ixgbe 4.3.13 by disabling those lines dint help either?

i tried enabling configure
CONFIG_RTE_LIBRTE_ETHDEV_DEBUG=y
 but still i don't see the log messages under /var/log nor on stdout?

just to keep ball rolling i'd like to see at least kni compiling and
working.

FYI, i am using X540-AT2

cheers

On Mon, Apr 11, 2016 at 10:09 AM, chintu hetam <rometoroam@gmail.com> wrote:

> Thanks Xie.
>
> I am trying to test FreeBSD-9.3-virtio as guest. Somewhere in the forum i
> found virtio-kni combination reaching around 2.7 Gbps performance, which is
> enough for my test, though i dint find equivalent driver performance
> characterization for qemu-vhost user space combination.
> Also, as per nics-2.2 user guide, "Linux kernel with KVM module; vhost
> module loaded and ioeventfd supported. Qemu standard backend without vhost
> support isn’t tested, and probably isn’t supported",which is bit different
> from vhost user space support,but still..
>
> Just to be sure vhost user space = qemu virtio backend- tap-linux-bridge
> configuration, as per nics guide, right?
>
> Thanking in advance, again.
>
>
>
>
>
>
>
> On Mon, Apr 11, 2016 at 4:25 AM, Xie, Huawei <huawei.xie@intel.com> wrote:
>
>> On 4/10/2016 7:26 AM, chintu hetam wrote:
>> > I am compiling DPDK 2.2 on Fedora 23 and i am seeing following build
>> error
>> >
>> /home/vcr/devel/dpdk/dpdk-2.2.0/build/build/lib/librte_eal/linuxapp/kni/kni_vhost.c:
>> > In function ‘kni_sock_poll’:
>> >
>> /home/vcr/devel/dpdk/dpdk-2.2.0/build/build/lib/librte_eal/linuxapp/kni/kni_vhost.c:254:25:
>> > error: ‘SOCK_ASYNC_NOSPACE’ undeclared (first use in this function)
>> >       (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock->flags) &&
>> >                          ^
>>
>> Hi, besides the issue, now user space vhost is the preferred way to
>> switch packets with the virtual machine, and we have plans to remove kni
>> vhost support. Do you have any reason to use kni vhost?
>>
>
>

Re: [dpdk-dev] DPDK 2.2 build failing with vhost-kni

[chintu hetam]

thanks ferruh yigit!!

cheers

On Mon, Apr 11, 2016 at 2:57 PM, chintu hetam <rometoroam@gmail.com> wrote:

> I tried vhost-user mode, and during init of ./vhost-switch it fails
> with
> PMD: eth_ixgbe_dev_init(): port 0 vendorID=0x8086 deviceID=0x1528
> pf queue num: 0, configured vmdq pool num: 64, each vmdq pool has 2 queues
> EAL: Error - exiting with code: 1
>   Cause: Cannot initialize network ports
>
> it's failing in port_init function, and it seems from
> http://dpdk.org/doc/guides-1.8/rel_notes/resolved_issues.html
>
> this has to with ixgbe checking for eeprom pba parameters. is that
> correct, coz rebuilding ixgbe 4.3.13 by disabling those lines dint help
> either?
>
> i tried enabling configure
> CONFIG_RTE_LIBRTE_ETHDEV_DEBUG=y
>  but still i don't see the log messages under /var/log nor on stdout?
>
> just to keep ball rolling i'd like to see at least kni compiling and
> working.
>
> FYI, i am using X540-AT2
>
> cheers
>
> On Mon, Apr 11, 2016 at 10:09 AM, chintu hetam <rometoroam@gmail.com>
> wrote:
>
>> Thanks Xie.
>>
>> I am trying to test FreeBSD-9.3-virtio as guest. Somewhere in the forum i
>> found virtio-kni combination reaching around 2.7 Gbps performance, which is
>> enough for my test, though i dint find equivalent driver performance
>> characterization for qemu-vhost user space combination.
>> Also, as per nics-2.2 user guide, "Linux kernel with KVM module; vhost
>> module loaded and ioeventfd supported. Qemu standard backend without vhost
>> support isn’t tested, and probably isn’t supported",which is bit different
>> from vhost user space support,but still..
>>
>> Just to be sure vhost user space = qemu virtio backend- tap-linux-bridge
>> configuration, as per nics guide, right?
>>
>> Thanking in advance, again.
>>
>>
>>
>>
>>
>>
>>
>> On Mon, Apr 11, 2016 at 4:25 AM, Xie, Huawei <huawei.xie@intel.com>
>> wrote:
>>
>>> On 4/10/2016 7:26 AM, chintu hetam wrote:
>>> > I am compiling DPDK 2.2 on Fedora 23 and i am seeing following build
>>> error
>>> >
>>> /home/vcr/devel/dpdk/dpdk-2.2.0/build/build/lib/librte_eal/linuxapp/kni/kni_vhost.c:
>>> > In function ‘kni_sock_poll’:
>>> >
>>> /home/vcr/devel/dpdk/dpdk-2.2.0/build/build/lib/librte_eal/linuxapp/kni/kni_vhost.c:254:25:
>>> > error: ‘SOCK_ASYNC_NOSPACE’ undeclared (first use in this function)
>>> >       (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock->flags) &&
>>> >                          ^
>>>
>>> Hi, besides the issue, now user space vhost is the preferred way to
>>> switch packets with the virtual machine, and we have plans to remove kni
>>> vhost support. Do you have any reason to use kni vhost?
>>>
>>
>>
>

[dpdk-dev] [PATCH] kni: fix vhost-kni compile errors

[Ferruh Yigit]

fix vhost-kni compile errors because of Linux kernel API changes

- SOCK_ASYNC_WAITDATA renamed to SOCKWQ_ASYNC_WAITDATA
  Linux commit id: 9cd3e072
  Updated in Linux kernel 4.4

- sk_alloc() gets new parameter
  Linux commit id: 11aa9c28b
  Updated in Linux kernel 4.2
  New parameter is: "@kern: is this to be a kernel socket?"

Reported-by: chintu hetam <rometoroam@gmail.com>
Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>

---

Patch not tested, just fixed compiler errors.
---
 lib/librte_eal/linuxapp/kni/kni_vhost.c | 21 +++++++++++++++++----
 1 file changed, 17 insertions(+), 4 deletions(-)

diff --git a/lib/librte_eal/linuxapp/kni/kni_vhost.c b/lib/librte_eal/linuxapp/kni/kni_vhost.c
index 6f2a961..482b384 100644
--- a/lib/librte_eal/linuxapp/kni/kni_vhost.c
+++ b/lib/librte_eal/linuxapp/kni/kni_vhost.c
@@ -251,7 +251,11 @@ kni_sock_poll(struct file *file, struct socket *sock, poll_table * wait)
 		mask |= POLLIN | POLLRDNORM;
 
 	if (sock_writeable(&q->sk) ||
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(4,4,0)
+	    (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &q->sock->flags) &&
+#else
 	    (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock->flags) &&
+#endif
 	     sock_writeable(&q->sk)))
 		mask |= POLLOUT | POLLWRNORM;
 
@@ -619,8 +623,11 @@ kni_sk_write_space(struct sock *sk)
 	wait_queue_head_t *wqueue;
 
 	if (!sock_writeable(sk) ||
-	    !test_and_clear_bit(SOCK_ASYNC_NOSPACE,
-				&sk->sk_socket->flags))
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(4,4,0)
+	    !test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
+#else
+	    !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
+#endif
 		return;
 	wqueue = sk_sleep(sk);
 	if (wqueue && waitqueue_active(wqueue))
@@ -666,8 +673,14 @@ kni_vhost_backend_init(struct kni_dev *kni)
 	if (kni->vhost_queue != NULL)
 		return -1;
 
-	if (!(q = (struct kni_vhost_queue *)sk_alloc(
-		      net, AF_UNSPEC, GFP_KERNEL, &kni_raw_proto)))
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(4,2,0)
+	q = (struct kni_vhost_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
+			&kni_raw_proto, 0);
+#else
+	q = (struct kni_vhost_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
+			&kni_raw_proto);
+#endif
+	if (!q)
 		return -ENOMEM;
 
 	err = sock_create_lite(AF_UNSPEC, SOCK_RAW, IPPROTO_RAW, &q->sock);
-- 
2.5.5

Re: [dpdk-dev] [PATCH] kni: fix vhost-kni compile errors

[Thomas Monjalon]

2016-04-11 17:54, Ferruh Yigit:
> +#if LINUX_VERSION_CODE >= KERNEL_VERSION(4,4,0)
> +	    (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &q->sock->flags) &&
> +#else
>  	    (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock->flags) &&
> +#endif

You could avoid some issues with backported feature by using
#ifdef SOCKWQ_ASYNC_NOSPACE

Re: [dpdk-dev] [PATCH] kni: fix vhost-kni compile errors

[Ferruh Yigit]

On 4/11/2016 6:04 PM, Thomas Monjalon wrote:
> 2016-04-11 17:54, Ferruh Yigit:
>> +#if LINUX_VERSION_CODE >= KERNEL_VERSION(4,4,0)
>> +	    (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &q->sock->flags) &&
>> +#else
>>  	    (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock->flags) &&
>> +#endif
> 
> You could avoid some issues with backported feature by using
> #ifdef SOCKWQ_ASYNC_NOSPACE
> 
Right, I am sending a new version.

Regards,
ferruh

[dpdk-dev] [PATCH v2] kni: fix vhost-kni compile errors

[Ferruh Yigit]

fix vhost-kni compile errors because of Linux kernel API changes

- SOCK_ASYNC_WAITDATA renamed to SOCKWQ_ASYNC_WAITDATA
  Linux commit id: 9cd3e072
  Updated in Linux kernel 4.4

- sk_alloc() gets new parameter
  Linux commit id: 11aa9c28b
  Updated in Linux kernel 4.2
  New parameter is: "@kern: is this to be a kernel socket?"

Reported-by: chintu hetam <rometoroam@gmail.com>
Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>

---

Patch not tested, just fixed compiler errors.

v2:
* use "#ifdef SOCKWQ_ASYNC_WAITDATA" instead of version check
---
 lib/librte_eal/linuxapp/kni/kni_vhost.c | 21 +++++++++++++++++----
 1 file changed, 17 insertions(+), 4 deletions(-)

diff --git a/lib/librte_eal/linuxapp/kni/kni_vhost.c b/lib/librte_eal/linuxapp/kni/kni_vhost.c
index 6f2a961..a3ca849 100644
--- a/lib/librte_eal/linuxapp/kni/kni_vhost.c
+++ b/lib/librte_eal/linuxapp/kni/kni_vhost.c
@@ -251,7 +251,11 @@ kni_sock_poll(struct file *file, struct socket *sock, poll_table * wait)
 		mask |= POLLIN | POLLRDNORM;
 
 	if (sock_writeable(&q->sk) ||
+#ifdef SOCKWQ_ASYNC_NOSPACE
+	    (!test_and_set_bit(SOCKWQ_ASYNC_NOSPACE, &q->sock->flags) &&
+#else
 	    (!test_and_set_bit(SOCK_ASYNC_NOSPACE, &q->sock->flags) &&
+#endif
 	     sock_writeable(&q->sk)))
 		mask |= POLLOUT | POLLWRNORM;
 
@@ -619,8 +623,11 @@ kni_sk_write_space(struct sock *sk)
 	wait_queue_head_t *wqueue;
 
 	if (!sock_writeable(sk) ||
-	    !test_and_clear_bit(SOCK_ASYNC_NOSPACE,
-				&sk->sk_socket->flags))
+#ifdef SOCKWQ_ASYNC_NOSPACE
+	    !test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &sk->sk_socket->flags))
+#else
+	    !test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags))
+#endif
 		return;
 	wqueue = sk_sleep(sk);
 	if (wqueue && waitqueue_active(wqueue))
@@ -666,8 +673,14 @@ kni_vhost_backend_init(struct kni_dev *kni)
 	if (kni->vhost_queue != NULL)
 		return -1;
 
-	if (!(q = (struct kni_vhost_queue *)sk_alloc(
-		      net, AF_UNSPEC, GFP_KERNEL, &kni_raw_proto)))
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 2, 0)
+	q = (struct kni_vhost_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
+			&kni_raw_proto, 0);
+#else
+	q = (struct kni_vhost_queue *)sk_alloc(net, AF_UNSPEC, GFP_KERNEL,
+			&kni_raw_proto);
+#endif
+	if (!q)
 		return -ENOMEM;
 
 	err = sock_create_lite(AF_UNSPEC, SOCK_RAW, IPPROTO_RAW, &q->sock);
-- 
2.5.5

Re: [dpdk-dev] [PATCH v2] kni: fix vhost-kni compile errors

[Thomas Monjalon]

2016-04-11 19:30, Ferruh Yigit:
> fix vhost-kni compile errors because of Linux kernel API changes
> 
> - SOCK_ASYNC_WAITDATA renamed to SOCKWQ_ASYNC_WAITDATA
>   Linux commit id: 9cd3e072
>   Updated in Linux kernel 4.4
> 
> - sk_alloc() gets new parameter
>   Linux commit id: 11aa9c28b
>   Updated in Linux kernel 4.2
>   New parameter is: "@kern: is this to be a kernel socket?"
> 
> Reported-by: chintu hetam <rometoroam@gmail.com>
> Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>

Applied, thanks