[dpdk-dev] [PATCH 0/2] Fix Vhost PMD setup

[dpdk-dev] [PATCH 0/2] Fix Vhost PMD setup

[Maxime Coquelin]

This series fixes regression introduced in v20.02-rc3.
First patch fixes the error path, and second one fix
Vhost device reconfigure issue reported by Yinan.

Maxime Coquelin (2):
  net/vhost: fix Vhost setup error path
  net/vhost: prevent multiple setup on reconfig

 drivers/net/vhost/rte_eth_vhost.c | 26 +++++++++++++++++++-------
 1 file changed, 19 insertions(+), 7 deletions(-)

-- 
2.24.1

[dpdk-dev] [PATCH 1/2] net/vhost: fix Vhost setup error path

[Maxime Coquelin]

If for some reason vhost_driver_setup() fails, the list
element for the device may be freed without being removed
from the internal list of devices.

This patch fixes all the error paths, by unregistering the
device from Vhost library it has been registered, remove
the device from the list, reset device vring_state pointer
from the global table and only free vring state if it had
been allocated.

Fixes: 3d01b759d267 ("net/vhost: delay driver setup")

Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
---
 drivers/net/vhost/rte_eth_vhost.c | 21 ++++++++++++++-------
 1 file changed, 14 insertions(+), 7 deletions(-)

diff --git a/drivers/net/vhost/rte_eth_vhost.c b/drivers/net/vhost/rte_eth_vhost.c
index 90263ae77c..c0056bc8bf 100644
--- a/drivers/net/vhost/rte_eth_vhost.c
+++ b/drivers/net/vhost/rte_eth_vhost.c
@@ -878,12 +878,12 @@ vhost_driver_setup(struct rte_eth_dev *eth_dev)
 
 	list = rte_zmalloc_socket(name, sizeof(*list), 0, numa_node);
 	if (list == NULL)
-		goto error;
+		return -1;
 
 	vring_state = rte_zmalloc_socket(name, sizeof(*vring_state),
 					 0, numa_node);
 	if (vring_state == NULL)
-		goto error;
+		goto free_list;
 
 	list->eth_dev = eth_dev;
 	pthread_mutex_lock(&internal_list_lock);
@@ -894,30 +894,37 @@ vhost_driver_setup(struct rte_eth_dev *eth_dev)
 	vring_states[eth_dev->data->port_id] = vring_state;
 
 	if (rte_vhost_driver_register(internal->iface_name, internal->flags))
-		goto error;
+		goto list_remove;
 
 	if (internal->disable_flags) {
 		if (rte_vhost_driver_disable_features(internal->iface_name,
 						      internal->disable_flags))
-			goto error;
+			goto drv_unreg;
 	}
 
 	if (rte_vhost_driver_callback_register(internal->iface_name,
 					       &vhost_ops) < 0) {
 		VHOST_LOG(ERR, "Can't register callbacks\n");
-		goto error;
+		goto drv_unreg;
 	}
 
 	if (rte_vhost_driver_start(internal->iface_name) < 0) {
 		VHOST_LOG(ERR, "Failed to start driver for %s\n",
 			  internal->iface_name);
-		goto error;
+		goto drv_unreg;
 	}
 
 	return 0;
 
-error:
+drv_unreg:
+	rte_vhost_driver_unregister(internal->iface_name);
+list_remove:
+	pthread_mutex_lock(&internal_list_lock);
+	TAILQ_REMOVE(&internal_list, list, next);
+	pthread_mutex_unlock(&internal_list_lock);
+	vring_states[eth_dev->data->port_id] = NULL;
 	rte_free(vring_state);
+free_list:
 	rte_free(list);
 
 	return -1;
-- 
2.24.1

[dpdk-dev] [PATCH 2/2] net/vhost: prevent multiple setup on reconfig

[Maxime Coquelin]

Ethdev's .dev_configure callback can be called multiple
time during a device life-time, but Vhost makes the
wrong assumption that it is not the case and try to
setup again the device on reconfiguration.

This patch ensures the device hasn't been already setup
before proceeding.

Fixes: 3d01b759d267 ("net/vhost: delay driver setup")

Reported-by: Yinan Wang <yinan.wang@intel.com>
Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
---
 drivers/net/vhost/rte_eth_vhost.c | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/drivers/net/vhost/rte_eth_vhost.c b/drivers/net/vhost/rte_eth_vhost.c
index c0056bc8bf..c4643da120 100644
--- a/drivers/net/vhost/rte_eth_vhost.c
+++ b/drivers/net/vhost/rte_eth_vhost.c
@@ -876,6 +876,11 @@ vhost_driver_setup(struct rte_eth_dev *eth_dev)
 	unsigned int numa_node = eth_dev->device->numa_node;
 	const char *name = eth_dev->device->name;
 
+	/* Don't try to setup again if it has already been done. */
+	list = find_internal_resource(internal->iface_name);
+	if (list)
+		return 0;
+
 	list = rte_zmalloc_socket(name, sizeof(*list), 0, numa_node);
 	if (list == NULL)
 		return -1;
-- 
2.24.1

Re: [dpdk-dev] [PATCH 0/2] Fix Vhost PMD setup

[Wang, Yinan]

This patch set can fix Vhost device reconfigure issue, thanks!


> -----Original Message-----
> From: Maxime Coquelin <maxime.coquelin@redhat.com>
> Sent: 2020年2月18日 22:35
> To: dev@dpdk.org; oda@valinux.co.jp; Wang, Yinan <yinan.wang@intel.com>;
> Bie, Tiwei <tiwei.bie@intel.com>; amorenoz@redhat.com
> Cc: Maxime Coquelin <maxime.coquelin@redhat.com>
> Subject: [PATCH 0/2] Fix Vhost PMD setup
> 
> This series fixes regression introduced in v20.02-rc3.
> First patch fixes the error path, and second one fix Vhost device reconfigure
> issue reported by Yinan.
> 
> Maxime Coquelin (2):
>   net/vhost: fix Vhost setup error path
>   net/vhost: prevent multiple setup on reconfig
> 
>  drivers/net/vhost/rte_eth_vhost.c | 26 +++++++++++++++++++-------
>  1 file changed, 19 insertions(+), 7 deletions(-)
> 
> --
> 2.24.1

Re: [dpdk-dev] [PATCH 0/2] Fix Vhost PMD setup

[Maxime Coquelin]

On 2/18/20 4:24 PM, Wang, Yinan wrote:
> This patch set can fix Vhost device reconfigure issue, thanks!

Thanks for the quick test, I'll ask your Tested-by on patch 2.

Maxime

> 
>> -----Original Message-----
>> From: Maxime Coquelin <maxime.coquelin@redhat.com>
>> Sent: 2020年2月18日 22:35
>> To: dev@dpdk.org; oda@valinux.co.jp; Wang, Yinan <yinan.wang@intel.com>;
>> Bie, Tiwei <tiwei.bie@intel.com>; amorenoz@redhat.com
>> Cc: Maxime Coquelin <maxime.coquelin@redhat.com>
>> Subject: [PATCH 0/2] Fix Vhost PMD setup
>>
>> This series fixes regression introduced in v20.02-rc3.
>> First patch fixes the error path, and second one fix Vhost device reconfigure
>> issue reported by Yinan.
>>
>> Maxime Coquelin (2):
>>   net/vhost: fix Vhost setup error path
>>   net/vhost: prevent multiple setup on reconfig
>>
>>  drivers/net/vhost/rte_eth_vhost.c | 26 +++++++++++++++++++-------
>>  1 file changed, 19 insertions(+), 7 deletions(-)
>>
>> --
>> 2.24.1
> 

Re: [dpdk-dev] [PATCH 2/2] net/vhost: prevent multiple setup on reconfig

[Maxime Coquelin]

On 2/18/20 3:35 PM, Maxime Coquelin wrote:
> Ethdev's .dev_configure callback can be called multiple
> time during a device life-time, but Vhost makes the
> wrong assumption that it is not the case and try to
> setup again the device on reconfiguration.
> 
> This patch ensures the device hasn't been already setup
> before proceeding.
> 
> Fixes: 3d01b759d267 ("net/vhost: delay driver setup")
> 
> Reported-by: Yinan Wang <yinan.wang@intel.com>
> Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
> ---
>  drivers/net/vhost/rte_eth_vhost.c | 5 +++++
>  1 file changed, 5 insertions(+)
> 
> diff --git a/drivers/net/vhost/rte_eth_vhost.c b/drivers/net/vhost/rte_eth_vhost.c
> index c0056bc8bf..c4643da120 100644
> --- a/drivers/net/vhost/rte_eth_vhost.c
> +++ b/drivers/net/vhost/rte_eth_vhost.c
> @@ -876,6 +876,11 @@ vhost_driver_setup(struct rte_eth_dev *eth_dev)
>  	unsigned int numa_node = eth_dev->device->numa_node;
>  	const char *name = eth_dev->device->name;
>  
> +	/* Don't try to setup again if it has already been done. */
> +	list = find_internal_resource(internal->iface_name);
> +	if (list)
> +		return 0;
> +
>  	list = rte_zmalloc_socket(name, sizeof(*list), 0, numa_node);
>  	if (list == NULL)
>  		return -1;
> 

Following Yinan's reply to cover letter:
Tested-by: Yinan Wang <yinan.wang@intel.com>

Re: [dpdk-dev] [PATCH 2/2] net/vhost: prevent multiple setup on reconfig

[Wang, Yinan]

Tested-by: Yinan Wang <yinan.wang@intel.com>

> -----Original Message-----
> From: Maxime Coquelin <maxime.coquelin@redhat.com>
> Sent: 2020年2月18日 22:35
> To: dev@dpdk.org; oda@valinux.co.jp; Wang, Yinan <yinan.wang@intel.com>;
> Bie, Tiwei <tiwei.bie@intel.com>; amorenoz@redhat.com
> Cc: Maxime Coquelin <maxime.coquelin@redhat.com>
> Subject: [PATCH 2/2] net/vhost: prevent multiple setup on reconfig
> 
> Ethdev's .dev_configure callback can be called multiple time during a device
> life-time, but Vhost makes the wrong assumption that it is not the case and try
> to setup again the device on reconfiguration.
> 
> This patch ensures the device hasn't been already setup before proceeding.
> 
> Fixes: 3d01b759d267 ("net/vhost: delay driver setup")
> 
> Reported-by: Yinan Wang <yinan.wang@intel.com>
> Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
> ---
>  drivers/net/vhost/rte_eth_vhost.c | 5 +++++
>  1 file changed, 5 insertions(+)
> 
> diff --git a/drivers/net/vhost/rte_eth_vhost.c
> b/drivers/net/vhost/rte_eth_vhost.c
> index c0056bc8bf..c4643da120 100644
> --- a/drivers/net/vhost/rte_eth_vhost.c
> +++ b/drivers/net/vhost/rte_eth_vhost.c
> @@ -876,6 +876,11 @@ vhost_driver_setup(struct rte_eth_dev *eth_dev)
>  	unsigned int numa_node = eth_dev->device->numa_node;
>  	const char *name = eth_dev->device->name;
> 
> +	/* Don't try to setup again if it has already been done. */
> +	list = find_internal_resource(internal->iface_name);
> +	if (list)
> +		return 0;
> +
>  	list = rte_zmalloc_socket(name, sizeof(*list), 0, numa_node);
>  	if (list == NULL)
>  		return -1;
> --
> 2.24.1

Re: [dpdk-dev] [PATCH 1/2] net/vhost: fix Vhost setup error path

[David Marchand]

On Tue, Feb 18, 2020 at 3:35 PM Maxime Coquelin
<maxime.coquelin@redhat.com> wrote:
>
> If for some reason vhost_driver_setup() fails, the list
> element for the device may be freed without being removed
> from the internal list of devices.
>
> This patch fixes all the error paths, by unregistering the
> device from Vhost library it has been registered, remove
> the device from the list, reset device vring_state pointer
> from the global table and only free vring state if it had
> been allocated.
>
> Fixes: 3d01b759d267 ("net/vhost: delay driver setup")
>
> Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
> ---
>  drivers/net/vhost/rte_eth_vhost.c | 21 ++++++++++++++-------
>  1 file changed, 14 insertions(+), 7 deletions(-)
>
> diff --git a/drivers/net/vhost/rte_eth_vhost.c b/drivers/net/vhost/rte_eth_vhost.c
> index 90263ae77c..c0056bc8bf 100644
> --- a/drivers/net/vhost/rte_eth_vhost.c
> +++ b/drivers/net/vhost/rte_eth_vhost.c
> @@ -878,12 +878,12 @@ vhost_driver_setup(struct rte_eth_dev *eth_dev)
>
>         list = rte_zmalloc_socket(name, sizeof(*list), 0, numa_node);
>         if (list == NULL)
> -               goto error;
> +               return -1;
>
>         vring_state = rte_zmalloc_socket(name, sizeof(*vring_state),
>                                          0, numa_node);
>         if (vring_state == NULL)
> -               goto error;
> +               goto free_list;
>
>         list->eth_dev = eth_dev;
>         pthread_mutex_lock(&internal_list_lock);
> @@ -894,30 +894,37 @@ vhost_driver_setup(struct rte_eth_dev *eth_dev)
>         vring_states[eth_dev->data->port_id] = vring_state;
>
>         if (rte_vhost_driver_register(internal->iface_name, internal->flags))
> -               goto error;
> +               goto list_remove;
>
>         if (internal->disable_flags) {
>                 if (rte_vhost_driver_disable_features(internal->iface_name,
>                                                       internal->disable_flags))
> -                       goto error;
> +                       goto drv_unreg;
>         }
>
>         if (rte_vhost_driver_callback_register(internal->iface_name,
>                                                &vhost_ops) < 0) {
>                 VHOST_LOG(ERR, "Can't register callbacks\n");
> -               goto error;
> +               goto drv_unreg;
>         }
>
>         if (rte_vhost_driver_start(internal->iface_name) < 0) {
>                 VHOST_LOG(ERR, "Failed to start driver for %s\n",
>                           internal->iface_name);
> -               goto error;
> +               goto drv_unreg;
>         }
>
>         return 0;
>
> -error:
> +drv_unreg:
> +       rte_vhost_driver_unregister(internal->iface_name);
> +list_remove:
> +       pthread_mutex_lock(&internal_list_lock);
> +       TAILQ_REMOVE(&internal_list, list, next);
> +       pthread_mutex_unlock(&internal_list_lock);
> +       vring_states[eth_dev->data->port_id] = NULL;

We allocate/store in vring_states after inserting list in &internal_list.
Probably nitpicking but I would expect the opposite order on cleanup.


>         rte_free(vring_state);
> +free_list:
>         rte_free(list);
>
>         return -1;
> --
> 2.24.1
>

In any case,
Reviewed-by: David Marchand <david.marchand@redhat.com>



--
David Marchand

Re: [dpdk-dev] [PATCH 2/2] net/vhost: prevent multiple setup on reconfig

[David Marchand]

On Tue, Feb 18, 2020 at 3:35 PM Maxime Coquelin
<maxime.coquelin@redhat.com> wrote:
>
> Ethdev's .dev_configure callback can be called multiple
> time during a device life-time, but Vhost makes the
> wrong assumption that it is not the case and try to
> setup again the device on reconfiguration.
>
> This patch ensures the device hasn't been already setup
> before proceeding.
>
> Fixes: 3d01b759d267 ("net/vhost: delay driver setup")
>
> Reported-by: Yinan Wang <yinan.wang@intel.com>
> Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
> ---
>  drivers/net/vhost/rte_eth_vhost.c | 5 +++++
>  1 file changed, 5 insertions(+)
>
> diff --git a/drivers/net/vhost/rte_eth_vhost.c b/drivers/net/vhost/rte_eth_vhost.c
> index c0056bc8bf..c4643da120 100644
> --- a/drivers/net/vhost/rte_eth_vhost.c
> +++ b/drivers/net/vhost/rte_eth_vhost.c
> @@ -876,6 +876,11 @@ vhost_driver_setup(struct rte_eth_dev *eth_dev)
>         unsigned int numa_node = eth_dev->device->numa_node;
>         const char *name = eth_dev->device->name;
>
> +       /* Don't try to setup again if it has already been done. */
> +       list = find_internal_resource(internal->iface_name);
> +       if (list)
> +               return 0;
> +
>         list = rte_zmalloc_socket(name, sizeof(*list), 0, numa_node);
>         if (list == NULL)
>                 return -1;
> --
> 2.24.1
>

Reviewed-by: David Marchand <david.marchand@redhat.com>


-- 
David Marchand

Re: [dpdk-dev] [PATCH 1/2] net/vhost: fix Vhost setup error path

[Maxime Coquelin]

On 2/18/20 5:15 PM, David Marchand wrote:
> On Tue, Feb 18, 2020 at 3:35 PM Maxime Coquelin
> <maxime.coquelin@redhat.com> wrote:
>>
>> If for some reason vhost_driver_setup() fails, the list
>> element for the device may be freed without being removed
>> from the internal list of devices.
>>
>> This patch fixes all the error paths, by unregistering the
>> device from Vhost library it has been registered, remove
>> the device from the list, reset device vring_state pointer
>> from the global table and only free vring state if it had
>> been allocated.
>>
>> Fixes: 3d01b759d267 ("net/vhost: delay driver setup")
>>
>> Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
>> ---
>>  drivers/net/vhost/rte_eth_vhost.c | 21 ++++++++++++++-------
>>  1 file changed, 14 insertions(+), 7 deletions(-)
>>
>> diff --git a/drivers/net/vhost/rte_eth_vhost.c b/drivers/net/vhost/rte_eth_vhost.c
>> index 90263ae77c..c0056bc8bf 100644
>> --- a/drivers/net/vhost/rte_eth_vhost.c
>> +++ b/drivers/net/vhost/rte_eth_vhost.c
>> @@ -878,12 +878,12 @@ vhost_driver_setup(struct rte_eth_dev *eth_dev)
>>
>>         list = rte_zmalloc_socket(name, sizeof(*list), 0, numa_node);
>>         if (list == NULL)
>> -               goto error;
>> +               return -1;
>>
>>         vring_state = rte_zmalloc_socket(name, sizeof(*vring_state),
>>                                          0, numa_node);
>>         if (vring_state == NULL)
>> -               goto error;
>> +               goto free_list;
>>
>>         list->eth_dev = eth_dev;
>>         pthread_mutex_lock(&internal_list_lock);
>> @@ -894,30 +894,37 @@ vhost_driver_setup(struct rte_eth_dev *eth_dev)
>>         vring_states[eth_dev->data->port_id] = vring_state;
>>
>>         if (rte_vhost_driver_register(internal->iface_name, internal->flags))
>> -               goto error;
>> +               goto list_remove;
>>
>>         if (internal->disable_flags) {
>>                 if (rte_vhost_driver_disable_features(internal->iface_name,
>>                                                       internal->disable_flags))
>> -                       goto error;
>> +                       goto drv_unreg;
>>         }
>>
>>         if (rte_vhost_driver_callback_register(internal->iface_name,
>>                                                &vhost_ops) < 0) {
>>                 VHOST_LOG(ERR, "Can't register callbacks\n");
>> -               goto error;
>> +               goto drv_unreg;
>>         }
>>
>>         if (rte_vhost_driver_start(internal->iface_name) < 0) {
>>                 VHOST_LOG(ERR, "Failed to start driver for %s\n",
>>                           internal->iface_name);
>> -               goto error;
>> +               goto drv_unreg;
>>         }
>>
>>         return 0;
>>
>> -error:
>> +drv_unreg:
>> +       rte_vhost_driver_unregister(internal->iface_name);
>> +list_remove:
>> +       pthread_mutex_lock(&internal_list_lock);
>> +       TAILQ_REMOVE(&internal_list, list, next);
>> +       pthread_mutex_unlock(&internal_list_lock);
>> +       vring_states[eth_dev->data->port_id] = NULL;
> 
> We allocate/store in vring_states after inserting list in &internal_list.
> Probably nitpicking but I would expect the opposite order on cleanup.

No nitpicking, it is a good practice to cleanup in opposite order.
I will post a v2.

> 
>>         rte_free(vring_state);
>> +free_list:
>>         rte_free(list);
>>
>>         return -1;
>> --
>> 2.24.1
>>
> 
> In any case,
> Reviewed-by: David Marchand <david.marchand@redhat.com>
> 

Thanks!
Maxime

> 
> --
> David Marchand
> 

Re: [dpdk-dev] [PATCH 2/2] net/vhost: prevent multiple setup on reconfig

[Itsuro ODA]

Hi Maxime, Tiewi,

On Wed, 19 Feb 2020 09:17:41 +0100
Maxime Coquelin <maxime.coquelin@redhat.com> wrote:

> Hi Tiwei,
> 
> On 2/19/20 4:44 AM, Tiwei Bie wrote:
> > On Tue, Feb 18, 2020 at 06:22:40PM +0100, Maxime Coquelin wrote:
> >> Ethdev's .dev_configure callback can be called multiple
> >> time during a device life-time, but Vhost makes the
> >> wrong assumption that it is not the case and try to
> >> setup again the device on reconfiguration.
> >>
> >> This patch ensures the device hasn't been already setup
> >> before proceeding.
> >>
> >> Fixes: 3d01b759d267 ("net/vhost: delay driver setup")
> >>
> >> Reported-by: Yinan Wang <yinan.wang@intel.com>
> >> Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
> >> Tested-by: Yinan Wang <yinan.wang@intel.com>
> >> Reviewed-by: David Marchand <david.marchand@redhat.com>
> >> ---
> >>  drivers/net/vhost/rte_eth_vhost.c | 5 +++++
> >>  1 file changed, 5 insertions(+)
> >>
> >> diff --git a/drivers/net/vhost/rte_eth_vhost.c b/drivers/net/vhost/rte_eth_vhost.c
> >> index 4a7b1b608c..458ed58f5f 100644
> >> --- a/drivers/net/vhost/rte_eth_vhost.c
> >> +++ b/drivers/net/vhost/rte_eth_vhost.c
> >> @@ -876,6 +876,11 @@ vhost_driver_setup(struct rte_eth_dev *eth_dev)
> >>  	unsigned int numa_node = eth_dev->device->numa_node;
> >>  	const char *name = eth_dev->device->name;
> >>  
> >> +	/* Don't try to setup again if it has already been done. */
> >> +	list = find_internal_resource(internal->iface_name);
> >> +	if (list)
> >> +		return 0;
> >> +
> >>  	list = rte_zmalloc_socket(name, sizeof(*list), 0, numa_node);
> >>  	if (list == NULL)
> >>  		return -1;
> >> -- 
> >> 2.24.1
> > 
> > Thanks Maxime for the fix!
> > 
> > Reviewed-by: Tiwei Bie <tiwei.bie@intel.com>
> > 
> > Not related to this fix, it seems there is a potential leak after
> > delaying the driver setup to _configure, as the list won't be
> > registered until users configure the device. So internal->iface_name
> > allocated in _create will leak if the device is released without
> > doing _configure first.
> > 
> > https://github.com/DPDK/dpdk/blob/e6c78e736d77/drivers/net/vhost/rte_eth_vhost.c#L1058
> > https://github.com/DPDK/dpdk/blob/e6c78e736d77/drivers/net/vhost/rte_eth_vhost.c#L1075
> > 
> > It's not a common case and it's quite late in this release,
> > probably it's fine to fix it later.
> 
> That's a valid point, and I also agree there is no urgency for v20.02.
> Itsuro, would you take care of fixing it for v20.05?

Sure, I will do it.

Thanks.
Itsuro Oda

> Thanks,
> Maxime
> > Thanks,
> > Tiwei
> > 

-- 
Itsuro ODA <oda@valinux.co.jp>

Re: [dpdk-dev] [PATCH 2/2] net/vhost: prevent multiple setup on reconfig

[Maxime Coquelin]

Hi Tiwei,

On 2/19/20 4:44 AM, Tiwei Bie wrote:
> On Tue, Feb 18, 2020 at 06:22:40PM +0100, Maxime Coquelin wrote:
>> Ethdev's .dev_configure callback can be called multiple
>> time during a device life-time, but Vhost makes the
>> wrong assumption that it is not the case and try to
>> setup again the device on reconfiguration.
>>
>> This patch ensures the device hasn't been already setup
>> before proceeding.
>>
>> Fixes: 3d01b759d267 ("net/vhost: delay driver setup")
>>
>> Reported-by: Yinan Wang <yinan.wang@intel.com>
>> Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
>> Tested-by: Yinan Wang <yinan.wang@intel.com>
>> Reviewed-by: David Marchand <david.marchand@redhat.com>
>> ---
>>  drivers/net/vhost/rte_eth_vhost.c | 5 +++++
>>  1 file changed, 5 insertions(+)
>>
>> diff --git a/drivers/net/vhost/rte_eth_vhost.c b/drivers/net/vhost/rte_eth_vhost.c
>> index 4a7b1b608c..458ed58f5f 100644
>> --- a/drivers/net/vhost/rte_eth_vhost.c
>> +++ b/drivers/net/vhost/rte_eth_vhost.c
>> @@ -876,6 +876,11 @@ vhost_driver_setup(struct rte_eth_dev *eth_dev)
>>  	unsigned int numa_node = eth_dev->device->numa_node;
>>  	const char *name = eth_dev->device->name;
>>  
>> +	/* Don't try to setup again if it has already been done. */
>> +	list = find_internal_resource(internal->iface_name);
>> +	if (list)
>> +		return 0;
>> +
>>  	list = rte_zmalloc_socket(name, sizeof(*list), 0, numa_node);
>>  	if (list == NULL)
>>  		return -1;
>> -- 
>> 2.24.1
> 
> Thanks Maxime for the fix!
> 
> Reviewed-by: Tiwei Bie <tiwei.bie@intel.com>
> 
> Not related to this fix, it seems there is a potential leak after
> delaying the driver setup to _configure, as the list won't be
> registered until users configure the device. So internal->iface_name
> allocated in _create will leak if the device is released without
> doing _configure first.
> 
> https://github.com/DPDK/dpdk/blob/e6c78e736d77/drivers/net/vhost/rte_eth_vhost.c#L1058
> https://github.com/DPDK/dpdk/blob/e6c78e736d77/drivers/net/vhost/rte_eth_vhost.c#L1075
> 
> It's not a common case and it's quite late in this release,
> probably it's fine to fix it later.

That's a valid point, and I also agree there is no urgency for v20.02.
Itsuro, would you take care of fixing it for v20.05?

Thanks,
Maxime
> Thanks,
> Tiwei
> 

Re: [dpdk-dev] [PATCH 2/2] net/vhost: prevent multiple setup on reconfig

[Tiwei Bie]

On Tue, Feb 18, 2020 at 06:22:40PM +0100, Maxime Coquelin wrote:
> Ethdev's .dev_configure callback can be called multiple
> time during a device life-time, but Vhost makes the
> wrong assumption that it is not the case and try to
> setup again the device on reconfiguration.
> 
> This patch ensures the device hasn't been already setup
> before proceeding.
> 
> Fixes: 3d01b759d267 ("net/vhost: delay driver setup")
> 
> Reported-by: Yinan Wang <yinan.wang@intel.com>
> Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
> Tested-by: Yinan Wang <yinan.wang@intel.com>
> Reviewed-by: David Marchand <david.marchand@redhat.com>
> ---
>  drivers/net/vhost/rte_eth_vhost.c | 5 +++++
>  1 file changed, 5 insertions(+)
> 
> diff --git a/drivers/net/vhost/rte_eth_vhost.c b/drivers/net/vhost/rte_eth_vhost.c
> index 4a7b1b608c..458ed58f5f 100644
> --- a/drivers/net/vhost/rte_eth_vhost.c
> +++ b/drivers/net/vhost/rte_eth_vhost.c
> @@ -876,6 +876,11 @@ vhost_driver_setup(struct rte_eth_dev *eth_dev)
>  	unsigned int numa_node = eth_dev->device->numa_node;
>  	const char *name = eth_dev->device->name;
>  
> +	/* Don't try to setup again if it has already been done. */
> +	list = find_internal_resource(internal->iface_name);
> +	if (list)
> +		return 0;
> +
>  	list = rte_zmalloc_socket(name, sizeof(*list), 0, numa_node);
>  	if (list == NULL)
>  		return -1;
> -- 
> 2.24.1

Thanks Maxime for the fix!

Reviewed-by: Tiwei Bie <tiwei.bie@intel.com>

Not related to this fix, it seems there is a potential leak after
delaying the driver setup to _configure, as the list won't be
registered until users configure the device. So internal->iface_name
allocated in _create will leak if the device is released without
doing _configure first.

https://github.com/DPDK/dpdk/blob/e6c78e736d77/drivers/net/vhost/rte_eth_vhost.c#L1058
https://github.com/DPDK/dpdk/blob/e6c78e736d77/drivers/net/vhost/rte_eth_vhost.c#L1075

It's not a common case and it's quite late in this release,
probably it's fine to fix it later.

Thanks,
Tiwei

[dpdk-dev] [PATCH 2/2] net/vhost: prevent multiple setup on reconfig

[Maxime Coquelin]

Ethdev's .dev_configure callback can be called multiple
time during a device life-time, but Vhost makes the
wrong assumption that it is not the case and try to
setup again the device on reconfiguration.

This patch ensures the device hasn't been already setup
before proceeding.

Fixes: 3d01b759d267 ("net/vhost: delay driver setup")

Reported-by: Yinan Wang <yinan.wang@intel.com>
Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Tested-by: Yinan Wang <yinan.wang@intel.com>
Reviewed-by: David Marchand <david.marchand@redhat.com>
---
 drivers/net/vhost/rte_eth_vhost.c | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/drivers/net/vhost/rte_eth_vhost.c b/drivers/net/vhost/rte_eth_vhost.c
index 4a7b1b608c..458ed58f5f 100644
--- a/drivers/net/vhost/rte_eth_vhost.c
+++ b/drivers/net/vhost/rte_eth_vhost.c
@@ -876,6 +876,11 @@ vhost_driver_setup(struct rte_eth_dev *eth_dev)
 	unsigned int numa_node = eth_dev->device->numa_node;
 	const char *name = eth_dev->device->name;
 
+	/* Don't try to setup again if it has already been done. */
+	list = find_internal_resource(internal->iface_name);
+	if (list)
+		return 0;
+
 	list = rte_zmalloc_socket(name, sizeof(*list), 0, numa_node);
 	if (list == NULL)
 		return -1;
-- 
2.24.1