[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <a10d6eda-d03f-42ea-af89-be0d798b5608@linux.alibaba.com>
Date: Wed, 6 Nov 2024 09:28:01 +0800
From: Philo Lu <lulie@...ux.alibaba.com>
To: Joe Damato <jdamato@...tly.com>, netdev@...r.kernel.org, mst@...hat.com,
jasowang@...hat.com, xuanzhuo@...ux.alibaba.com, eperezma@...hat.com,
andrew+netdev@...n.ch, davem@...emloft.net, edumazet@...gle.com,
kuba@...nel.org, pabeni@...hat.com, andrew@...nix.com,
virtualization@...ts.linux.dev, linux-kernel@...r.kernel.org
Subject: Re: [PATCH net 4/4] virtio_net: Update rss when set queue
On 2024/11/6 04:31, Joe Damato wrote:
> On Mon, Nov 04, 2024 at 04:57:06PM +0800, Philo Lu wrote:
>> RSS configuration should be updated with queue number. In particular, it
>> should be updated when (1) rss enabled and (2) default rss configuration
>> is used without user modification.
>>
>> During rss command processing, device updates queue_pairs using
>> rss.max_tx_vq. That is, the device updates queue_pairs together with
>> rss, so we can skip the sperate queue_pairs update
>> (VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET below) and return directly.
>>
>> Also remove the `vi->has_rss ?` check when setting vi->rss.max_tx_vq,
>> because this is not used in the other hash_report case.
>>
>> Fixes: c7114b1249fa ("drivers/net/virtio_net: Added basic RSS support.")
>> Signed-off-by: Philo Lu <lulie@...ux.alibaba.com>
>> Signed-off-by: Xuan Zhuo <xuanzhuo@...ux.alibaba.com>
>> ---
>> drivers/net/virtio_net.c | 65 +++++++++++++++++++++++++++++++---------
>> 1 file changed, 51 insertions(+), 14 deletions(-)
>>
>> diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
>> index 59d9fdf562e0..189afad3ffaa 100644
>> --- a/drivers/net/virtio_net.c
>> +++ b/drivers/net/virtio_net.c
>> @@ -3394,15 +3394,59 @@ static void virtnet_ack_link_announce(struct virtnet_info *vi)
>> dev_warn(&vi->dev->dev, "Failed to ack link announce.\n");
>> }
>>
>> +static bool virtnet_commit_rss_command(struct virtnet_info *vi);
>> +
>> +static void virtnet_rss_update_by_qpairs(struct virtnet_info *vi, u16 queue_pairs)
>> +{
>> + u32 indir_val = 0;
>> + int i = 0;
>> +
>> + for (; i < vi->rss_indir_table_size; ++i) {
>> + indir_val = ethtool_rxfh_indir_default(i, queue_pairs);
>> + vi->rss.indirection_table[i] = indir_val;
>> + }
>> + vi->rss.max_tx_vq = queue_pairs;
>> +}
>> +
>> static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs)
>> {
>> struct virtio_net_ctrl_mq *mq __free(kfree) = NULL;
>> - struct scatterlist sg;
>> + struct virtio_net_ctrl_rss old_rss;
>> struct net_device *dev = vi->dev;
>> + struct scatterlist sg;
>>
>> if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ))
>> return 0;
>>
>> + /* Firstly check if we need update rss. Do updating if both (1) rss enabled and
>> + * (2) no user configuration.
>> + *
>> + * During rss command processing, device updates queue_pairs using rss.max_tx_vq. That is,
>> + * the device updates queue_pairs together with rss, so we can skip the sperate queue_pairs
>> + * update (VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET below) and return directly.
>> + */
>> + if (vi->has_rss && !netif_is_rxfh_configured(dev)) {
>
> Does there need to be an error case when:
>
> vi->has_rss && netif_is_rxfh_configured(dev)
>
> to return EINVAL? I noted that other drivers don't let users adjust
> the queue count and return error in this case.
>
In fact, there are 2 possible cases if users have adjusted rss,
depending on the total queue pairs used in the indirection table (x),
and the requested new queue count (y).
Case A: If y < x, it's illegal and will be rejected by
ethtool_check_max_channel().
Case B: If x <= y, we only adjust the queue number without touching the
rss configuration set by users.
So I don't think it necessary to add the check (if the above processing
is agreed).
Thanks for your review, Joe.
--
Philo
Powered by blists - more mailing lists