1. Patchew
  2. linux
  3. [PATCH v3 0/2] Properly initialize speed/duplex and remove vDPA config updates
  4. View patch

[PATCH v3 1/2] vdpa/mlx5: Set speed and duplex of vDPA devices to UNKNOWN

Carlos Bilbao posted 2 patches 1 year, 3 months ago
[PATCH v3 1/2] vdpa/mlx5: Set speed and duplex of vDPA devices to UNKNOWN
Posted by Carlos Bilbao 1 year, 3 months ago 
From: Carlos Bilbao <cbilbao@digitalocean.com>

Initialize the speed and duplex fields in virtio_net_config to UNKNOWN.
This is needed because mlx5_vdpa vDPA devices currently do not support the
VIRTIO_NET_F_SPEED_DUPLEX feature which reports speed and duplex. Add
needed helper cpu_to_mlx5vdpa32() to convert endianness of speed.

Signed-off-by: Carlos Bilbao <cbilbao@digitalocean.com>
Reviewed-by: Dragos Tatulea <dtatulea@nvidia.com>
---
 drivers/vdpa/mlx5/net/mlx5_vnet.c | 12 ++++++++++++
 1 file changed, 12 insertions(+)

diff --git a/drivers/vdpa/mlx5/net/mlx5_vnet.c b/drivers/vdpa/mlx5/net/mlx5_vnet.c
index b56aae3f7be3..41ca268d43ff 100644
--- a/drivers/vdpa/mlx5/net/mlx5_vnet.c
+++ b/drivers/vdpa/mlx5/net/mlx5_vnet.c
@@ -173,6 +173,11 @@ static __virtio16 cpu_to_mlx5vdpa16(struct mlx5_vdpa_dev *mvdev, u16 val)
 	return __cpu_to_virtio16(mlx5_vdpa_is_little_endian(mvdev), val);
 }
 
+static __virtio32 cpu_to_mlx5vdpa32(struct mlx5_vdpa_dev *mvdev, u32 val)
+{
+	return __cpu_to_virtio32(mlx5_vdpa_is_little_endian(mvdev), val);
+}
+
 static u16 ctrl_vq_idx(struct mlx5_vdpa_dev *mvdev)
 {
 	if (!(mvdev->actual_features & BIT_ULL(VIRTIO_NET_F_MQ)))
@@ -3433,6 +3438,13 @@ static int mlx5_vdpa_dev_add(struct vdpa_mgmt_dev *v_mdev, const char *name,
 	init_rwsem(&ndev->reslock);
 	config = &ndev->config;
 
+	/*
+	 * mlx5_vdpa vDPA devices currently don't support reporting or
+	 * setting the speed or duplex.
+	 */
+	config->speed  = cpu_to_mlx5vdpa32(mvdev, SPEED_UNKNOWN);
+	config->duplex = DUPLEX_UNKNOWN;
+
 	if (add_config->mask & BIT_ULL(VDPA_ATTR_DEV_NET_CFG_MTU)) {
 		err = config_func_mtu(mdev, add_config->net.mtu);
 		if (err)
-- 
2.34.1
Re: [PATCH v3 1/2] vdpa/mlx5: Set speed and duplex of vDPA devices to UNKNOWN
Posted by Michael S. Tsirkin 1 year, 1 month ago 
On Wed, Sep 04, 2024 at 10:11:14AM -0500, Carlos Bilbao wrote:
> From: Carlos Bilbao <cbilbao@digitalocean.com>
> 
> Initialize the speed and duplex fields in virtio_net_config to UNKNOWN.
> This is needed because mlx5_vdpa vDPA devices currently do not support the
> VIRTIO_NET_F_SPEED_DUPLEX feature which reports speed and duplex.

I see no logic here. Without this feature bit, guests will not read
this field, why do we suddenly need to initialize it?

> Add
> needed helper cpu_to_mlx5vdpa32() to convert endianness of speed.
> 
> Signed-off-by: Carlos Bilbao <cbilbao@digitalocean.com>
> Reviewed-by: Dragos Tatulea <dtatulea@nvidia.com>
> ---
>  drivers/vdpa/mlx5/net/mlx5_vnet.c | 12 ++++++++++++
>  1 file changed, 12 insertions(+)
> 
> diff --git a/drivers/vdpa/mlx5/net/mlx5_vnet.c b/drivers/vdpa/mlx5/net/mlx5_vnet.c
> index b56aae3f7be3..41ca268d43ff 100644
> --- a/drivers/vdpa/mlx5/net/mlx5_vnet.c
> +++ b/drivers/vdpa/mlx5/net/mlx5_vnet.c
> @@ -173,6 +173,11 @@ static __virtio16 cpu_to_mlx5vdpa16(struct mlx5_vdpa_dev *mvdev, u16 val)
>  	return __cpu_to_virtio16(mlx5_vdpa_is_little_endian(mvdev), val);
>  }
>  
> +static __virtio32 cpu_to_mlx5vdpa32(struct mlx5_vdpa_dev *mvdev, u32 val)
> +{
> +	return __cpu_to_virtio32(mlx5_vdpa_is_little_endian(mvdev), val);
> +}
> +
>  static u16 ctrl_vq_idx(struct mlx5_vdpa_dev *mvdev)
>  {
>  	if (!(mvdev->actual_features & BIT_ULL(VIRTIO_NET_F_MQ)))
> @@ -3433,6 +3438,13 @@ static int mlx5_vdpa_dev_add(struct vdpa_mgmt_dev *v_mdev, const char *name,
>  	init_rwsem(&ndev->reslock);
>  	config = &ndev->config;
>  
> +	/*
> +	 * mlx5_vdpa vDPA devices currently don't support reporting or
> +	 * setting the speed or duplex.
> +	 */
> +	config->speed  = cpu_to_mlx5vdpa32(mvdev, SPEED_UNKNOWN);
> +	config->duplex = DUPLEX_UNKNOWN;
> +
>  	if (add_config->mask & BIT_ULL(VDPA_ATTR_DEV_NET_CFG_MTU)) {
>  		err = config_func_mtu(mdev, add_config->net.mtu);
>  		if (err)
> -- 
> 2.34.1
Re: [PATCH v3 1/2] vdpa/mlx5: Set speed and duplex of vDPA devices to UNKNOWN
Posted by Dragos Tatulea 1 year, 1 month ago 

On 07.11.24 22:50, Michael S. Tsirkin wrote:
> On Wed, Sep 04, 2024 at 10:11:14AM -0500, Carlos Bilbao wrote:
>> From: Carlos Bilbao <cbilbao@digitalocean.com>
>>
>> Initialize the speed and duplex fields in virtio_net_config to UNKNOWN.
>> This is needed because mlx5_vdpa vDPA devices currently do not support the
>> VIRTIO_NET_F_SPEED_DUPLEX feature which reports speed and duplex.
> 
> I see no logic here. Without this feature bit, guests will not read
> this field, why do we suddenly need to initialize it?
> 
IIRC, Carlos was reading data via ioctl VHOST_VDPA_GET_CONFIG which calls
.get_config() directly, always exposing the speed and duplex config fields [0].
Carlos, was this the case?

[0] https://lore.kernel.org/lkml/afcbf041-7613-48e6-8088-9d52edd907ff@nvidia.com/T/

Thanks,
Dragos

>> Add
>> needed helper cpu_to_mlx5vdpa32() to convert endianness of speed.
>>
>> Signed-off-by: Carlos Bilbao <cbilbao@digitalocean.com>
>> Reviewed-by: Dragos Tatulea <dtatulea@nvidia.com>
>> ---
>>  drivers/vdpa/mlx5/net/mlx5_vnet.c | 12 ++++++++++++
>>  1 file changed, 12 insertions(+)
>>
>> diff --git a/drivers/vdpa/mlx5/net/mlx5_vnet.c b/drivers/vdpa/mlx5/net/mlx5_vnet.c
>> index b56aae3f7be3..41ca268d43ff 100644
>> --- a/drivers/vdpa/mlx5/net/mlx5_vnet.c
>> +++ b/drivers/vdpa/mlx5/net/mlx5_vnet.c
>> @@ -173,6 +173,11 @@ static __virtio16 cpu_to_mlx5vdpa16(struct mlx5_vdpa_dev *mvdev, u16 val)
>>  	return __cpu_to_virtio16(mlx5_vdpa_is_little_endian(mvdev), val);
>>  }
>>  
>> +static __virtio32 cpu_to_mlx5vdpa32(struct mlx5_vdpa_dev *mvdev, u32 val)
>> +{
>> +	return __cpu_to_virtio32(mlx5_vdpa_is_little_endian(mvdev), val);
>> +}
>> +
>>  static u16 ctrl_vq_idx(struct mlx5_vdpa_dev *mvdev)
>>  {
>>  	if (!(mvdev->actual_features & BIT_ULL(VIRTIO_NET_F_MQ)))
>> @@ -3433,6 +3438,13 @@ static int mlx5_vdpa_dev_add(struct vdpa_mgmt_dev *v_mdev, const char *name,
>>  	init_rwsem(&ndev->reslock);
>>  	config = &ndev->config;
>>  
>> +	/*
>> +	 * mlx5_vdpa vDPA devices currently don't support reporting or
>> +	 * setting the speed or duplex.
>> +	 */
>> +	config->speed  = cpu_to_mlx5vdpa32(mvdev, SPEED_UNKNOWN);
>> +	config->duplex = DUPLEX_UNKNOWN;
>> +
>>  	if (add_config->mask & BIT_ULL(VDPA_ATTR_DEV_NET_CFG_MTU)) {
>>  		err = config_func_mtu(mdev, add_config->net.mtu);
>>  		if (err)
>> -- 
>> 2.34.1
>
Re: [PATCH v3 1/2] vdpa/mlx5: Set speed and duplex of vDPA devices to UNKNOWN
Posted by Michael S. Tsirkin 1 year, 1 month ago 
On Fri, Nov 08, 2024 at 10:31:58AM +0100, Dragos Tatulea wrote:
> 
> 
> On 07.11.24 22:50, Michael S. Tsirkin wrote:
> > On Wed, Sep 04, 2024 at 10:11:14AM -0500, Carlos Bilbao wrote:
> >> From: Carlos Bilbao <cbilbao@digitalocean.com>
> >>
> >> Initialize the speed and duplex fields in virtio_net_config to UNKNOWN.
> >> This is needed because mlx5_vdpa vDPA devices currently do not support the
> >> VIRTIO_NET_F_SPEED_DUPLEX feature which reports speed and duplex.
> > 
> > I see no logic here. Without this feature bit, guests will not read
> > this field, why do we suddenly need to initialize it?
> > 
> IIRC, Carlos was reading data via ioctl VHOST_VDPA_GET_CONFIG which calls
> .get_config() directly, always exposing the speed and duplex config fields [0].
> Carlos, was this the case?
> 
> [0] https://lore.kernel.org/lkml/afcbf041-7613-48e6-8088-9d52edd907ff@nvidia.com/T/
> 
> Thanks,
> Dragos

Basically, driver should ignore these if feature is not set.


> >> Add
> >> needed helper cpu_to_mlx5vdpa32() to convert endianness of speed.
> >>
> >> Signed-off-by: Carlos Bilbao <cbilbao@digitalocean.com>
> >> Reviewed-by: Dragos Tatulea <dtatulea@nvidia.com>
> >> ---
> >>  drivers/vdpa/mlx5/net/mlx5_vnet.c | 12 ++++++++++++
> >>  1 file changed, 12 insertions(+)
> >>
> >> diff --git a/drivers/vdpa/mlx5/net/mlx5_vnet.c b/drivers/vdpa/mlx5/net/mlx5_vnet.c
> >> index b56aae3f7be3..41ca268d43ff 100644
> >> --- a/drivers/vdpa/mlx5/net/mlx5_vnet.c
> >> +++ b/drivers/vdpa/mlx5/net/mlx5_vnet.c
> >> @@ -173,6 +173,11 @@ static __virtio16 cpu_to_mlx5vdpa16(struct mlx5_vdpa_dev *mvdev, u16 val)
> >>  	return __cpu_to_virtio16(mlx5_vdpa_is_little_endian(mvdev), val);
> >>  }
> >>  
> >> +static __virtio32 cpu_to_mlx5vdpa32(struct mlx5_vdpa_dev *mvdev, u32 val)
> >> +{
> >> +	return __cpu_to_virtio32(mlx5_vdpa_is_little_endian(mvdev), val);
> >> +}
> >> +
> >>  static u16 ctrl_vq_idx(struct mlx5_vdpa_dev *mvdev)
> >>  {
> >>  	if (!(mvdev->actual_features & BIT_ULL(VIRTIO_NET_F_MQ)))
> >> @@ -3433,6 +3438,13 @@ static int mlx5_vdpa_dev_add(struct vdpa_mgmt_dev *v_mdev, const char *name,
> >>  	init_rwsem(&ndev->reslock);
> >>  	config = &ndev->config;
> >>  
> >> +	/*
> >> +	 * mlx5_vdpa vDPA devices currently don't support reporting or
> >> +	 * setting the speed or duplex.
> >> +	 */
> >> +	config->speed  = cpu_to_mlx5vdpa32(mvdev, SPEED_UNKNOWN);
> >> +	config->duplex = DUPLEX_UNKNOWN;
> >> +
> >>  	if (add_config->mask & BIT_ULL(VDPA_ATTR_DEV_NET_CFG_MTU)) {
> >>  		err = config_func_mtu(mdev, add_config->net.mtu);
> >>  		if (err)
> >> -- 
> >> 2.34.1
> >
Re: [PATCH v3 1/2] vdpa/mlx5: Set speed and duplex of vDPA devices to UNKNOWN
Posted by Carlos Bilbao 1 year, 1 month ago 
Hello,

On 11/8/24 5:51 AM, Michael S. Tsirkin wrote:
> On Fri, Nov 08, 2024 at 10:31:58AM +0100, Dragos Tatulea wrote:
>>
>> On 07.11.24 22:50, Michael S. Tsirkin wrote:
>>> On Wed, Sep 04, 2024 at 10:11:14AM -0500, Carlos Bilbao wrote:
>>>> From: Carlos Bilbao <cbilbao@digitalocean.com>
>>>>
>>>> Initialize the speed and duplex fields in virtio_net_config to UNKNOWN.
>>>> This is needed because mlx5_vdpa vDPA devices currently do not support the
>>>> VIRTIO_NET_F_SPEED_DUPLEX feature which reports speed and duplex.
>>> I see no logic here. Without this feature bit, guests will not read
>>> this field, why do we suddenly need to initialize it?
>>>
>> IIRC, Carlos was reading data via ioctl VHOST_VDPA_GET_CONFIG which calls
>> .get_config() directly, always exposing the speed and duplex config fields [0].
>> Carlos, was this the case?
>>
>> [0] https://lore.kernel.org/lkml/afcbf041-7613-48e6-8088-9d52edd907ff@nvidia.com/T/
>>
>> Thanks,
>> Dragos
> Basically, driver should ignore these if feature is not set.


There _is_ a chance the guest could read this field; As Dragos mentioned, I
was using the VHOST_VDPA_GET_CONFIG ioctl from userspace, and the incorrect
field initialization led me to believe I was in half-duplex mode --
something people told me they hadn't seen since the 80s. But as Andrew
(CCed) mentioned, "If the speed is 0, does duplex even matter?".

I also tried to remove the pointless ioctl call altogether to avoid further
confusion [0] but that wasn't viable due to spec compliance.

Initializing the fields seems the simple solution here (and, IMHO, more
generally, it's as a good programming practice)

Thanks for looking into this!

Best,
Carlos

[0] https://lore.kernel.org/lkml/CACGkMEvfdUYLjx-Z+oB11XW-54ErJsQMKcnu2p=dsj5N_BiEKw@mail.gmail.com/


>
>
>>>> Add
>>>> needed helper cpu_to_mlx5vdpa32() to convert endianness of speed.
>>>>
>>>> Signed-off-by: Carlos Bilbao <cbilbao@digitalocean.com>
>>>> Reviewed-by: Dragos Tatulea <dtatulea@nvidia.com>
>>>> ---
>>>>  drivers/vdpa/mlx5/net/mlx5_vnet.c | 12 ++++++++++++
>>>>  1 file changed, 12 insertions(+)
>>>>
>>>> diff --git a/drivers/vdpa/mlx5/net/mlx5_vnet.c b/drivers/vdpa/mlx5/net/mlx5_vnet.c
>>>> index b56aae3f7be3..41ca268d43ff 100644
>>>> --- a/drivers/vdpa/mlx5/net/mlx5_vnet.c
>>>> +++ b/drivers/vdpa/mlx5/net/mlx5_vnet.c
>>>> @@ -173,6 +173,11 @@ static __virtio16 cpu_to_mlx5vdpa16(struct mlx5_vdpa_dev *mvdev, u16 val)
>>>>  	return __cpu_to_virtio16(mlx5_vdpa_is_little_endian(mvdev), val);
>>>>  }
>>>>  
>>>> +static __virtio32 cpu_to_mlx5vdpa32(struct mlx5_vdpa_dev *mvdev, u32 val)
>>>> +{
>>>> +	return __cpu_to_virtio32(mlx5_vdpa_is_little_endian(mvdev), val);
>>>> +}
>>>> +
>>>>  static u16 ctrl_vq_idx(struct mlx5_vdpa_dev *mvdev)
>>>>  {
>>>>  	if (!(mvdev->actual_features & BIT_ULL(VIRTIO_NET_F_MQ)))
>>>> @@ -3433,6 +3438,13 @@ static int mlx5_vdpa_dev_add(struct vdpa_mgmt_dev *v_mdev, const char *name,
>>>>  	init_rwsem(&ndev->reslock);
>>>>  	config = &ndev->config;
>>>>  
>>>> +	/*
>>>> +	 * mlx5_vdpa vDPA devices currently don't support reporting or
>>>> +	 * setting the speed or duplex.
>>>> +	 */
>>>> +	config->speed  = cpu_to_mlx5vdpa32(mvdev, SPEED_UNKNOWN);
>>>> +	config->duplex = DUPLEX_UNKNOWN;
>>>> +
>>>>  	if (add_config->mask & BIT_ULL(VDPA_ATTR_DEV_NET_CFG_MTU)) {
>>>>  		err = config_func_mtu(mdev, add_config->net.mtu);
>>>>  		if (err)
>>>> -- 
>>>> 2.34.1
Re: [PATCH v3 1/2] vdpa/mlx5: Set speed and duplex of vDPA devices to UNKNOWN
Posted by Jason Wang 1 year, 3 months ago 
On Wed, Sep 4, 2024 at 11:18 PM Carlos Bilbao
<carlos.bilbao.osdev@gmail.com> wrote:
>
> From: Carlos Bilbao <cbilbao@digitalocean.com>
>
> Initialize the speed and duplex fields in virtio_net_config to UNKNOWN.
> This is needed because mlx5_vdpa vDPA devices currently do not support the
> VIRTIO_NET_F_SPEED_DUPLEX feature which reports speed and duplex. Add
> needed helper cpu_to_mlx5vdpa32() to convert endianness of speed.
>
> Signed-off-by: Carlos Bilbao <cbilbao@digitalocean.com>
> Reviewed-by: Dragos Tatulea <dtatulea@nvidia.com>

Acked-by: Jason Wang <jasowang@redhat.com>

Thanks

> ---
>  drivers/vdpa/mlx5/net/mlx5_vnet.c | 12 ++++++++++++
>  1 file changed, 12 insertions(+)
>
> diff --git a/drivers/vdpa/mlx5/net/mlx5_vnet.c b/drivers/vdpa/mlx5/net/mlx5_vnet.c
> index b56aae3f7be3..41ca268d43ff 100644
> --- a/drivers/vdpa/mlx5/net/mlx5_vnet.c
> +++ b/drivers/vdpa/mlx5/net/mlx5_vnet.c
> @@ -173,6 +173,11 @@ static __virtio16 cpu_to_mlx5vdpa16(struct mlx5_vdpa_dev *mvdev, u16 val)
>         return __cpu_to_virtio16(mlx5_vdpa_is_little_endian(mvdev), val);
>  }
>
> +static __virtio32 cpu_to_mlx5vdpa32(struct mlx5_vdpa_dev *mvdev, u32 val)
> +{
> +       return __cpu_to_virtio32(mlx5_vdpa_is_little_endian(mvdev), val);
> +}
> +
>  static u16 ctrl_vq_idx(struct mlx5_vdpa_dev *mvdev)
>  {
>         if (!(mvdev->actual_features & BIT_ULL(VIRTIO_NET_F_MQ)))
> @@ -3433,6 +3438,13 @@ static int mlx5_vdpa_dev_add(struct vdpa_mgmt_dev *v_mdev, const char *name,
>         init_rwsem(&ndev->reslock);
>         config = &ndev->config;
>
> +       /*
> +        * mlx5_vdpa vDPA devices currently don't support reporting or
> +        * setting the speed or duplex.
> +        */
> +       config->speed  = cpu_to_mlx5vdpa32(mvdev, SPEED_UNKNOWN);
> +       config->duplex = DUPLEX_UNKNOWN;
> +
>         if (add_config->mask & BIT_ULL(VDPA_ATTR_DEV_NET_CFG_MTU)) {
>                 err = config_func_mtu(mdev, add_config->net.mtu);
>                 if (err)
> --
> 2.34.1
>

Patchew 2.1-devel-b67e4c2

© 2016 - 2025 Red Hat, Inc.