Re: [PATCH v3 2/4] drivers/net/virtio_net: Added basic RSS support.

From: Andrew Melnichenko
Date: Sun Feb 13 2022 - 12:02:03 EST


Hi all,

On Tue, Feb 8, 2022 at 10:37 PM Willem de Bruijn
<willemdebruijn.kernel@xxxxxxxxx> wrote:
>
> On Tue, Feb 8, 2022 at 1:19 PM Andrew Melnychenko <andrew@xxxxxxxxxx> wrote:
> >
> > Added features for RSS.
> > Added initialization, RXHASH feature and ethtool ops.
> > By default RSS/RXHASH is disabled.
> > Virtio RSS "IPv6 extensions" hashes disabled.
> > Added ethtools ops to set key and indirection table.
> >
> > Signed-off-by: Andrew Melnychenko <andrew@xxxxxxxxxx>
> > ---
> > drivers/net/virtio_net.c | 191 +++++++++++++++++++++++++++++++++++++--
> > 1 file changed, 185 insertions(+), 6 deletions(-)
> >
> > diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> > index 1404e683a2fd..495aed524e33 100644
> > --- a/drivers/net/virtio_net.c
> > +++ b/drivers/net/virtio_net.c
> > @@ -169,6 +169,24 @@ struct receive_queue {
> > struct xdp_rxq_info xdp_rxq;
> > };
> >
> > +/* This structure can contain rss message with maximum settings for indirection table and keysize
> > + * Note, that default structure that describes RSS configuration virtio_net_rss_config
> > + * contains same info but can't handle table values.
> > + * In any case, structure would be passed to virtio hw through sg_buf split by parts
> > + * because table sizes may be differ according to the device configuration.
> > + */
> > +#define VIRTIO_NET_RSS_MAX_KEY_SIZE 40
>
> Future proof, may want to support larger sizes.
>
> netdevice.h defines NETDEV_RSS_KEY_LEN at 52.
>
> tools/testing/selftests/net/toeplitz.c supports up to 60

According to virtio specification, the length of the key is
40bytes(and an indirection table is 128 entries max).
So for now, we support a maximum of the spec regardless of what the
kernel is capable of.

>
> > +#define VIRTIO_NET_RSS_MAX_TABLE_LEN 128
> > +struct virtio_net_ctrl_rss {
> > + u32 hash_types;
>
> conversely, u32 is a bit extreme?

No, the structure virtio_net_ctrl_rss is specified by the specification.

>
> > + u16 indirection_table_mask;
> > + u16 unclassified_queue;
> > + u16 indirection_table[VIRTIO_NET_RSS_MAX_TABLE_LEN];
> > + u16 max_tx_vq;
> > + u8 hash_key_length;
> > + u8 key[VIRTIO_NET_RSS_MAX_KEY_SIZE];
> > +};
> > +
> > /* Control VQ buffers: protected by the rtnl lock */
> > struct control_buf {
> > struct virtio_net_ctrl_hdr hdr;
> > @@ -178,6 +196,7 @@ struct control_buf {
> > u8 allmulti;
> > __virtio16 vid;
> > __virtio64 offloads;
> > + struct virtio_net_ctrl_rss rss;
> > };
> >
> > struct virtnet_info {
> > @@ -206,6 +225,12 @@ struct virtnet_info {
> > /* Host will merge rx buffers for big packets (shake it! shake it!) */
> > bool mergeable_rx_bufs;
> >
> > + /* Host supports rss and/or hash report */
> > + bool has_rss;
> > + u8 rss_key_size;
> > + u16 rss_indir_table_size;
> > + u32 rss_hash_types_supported;
> > +
> > /* Has control virtqueue */
> > bool has_cvq;
> >
> > @@ -2184,6 +2209,56 @@ static void virtnet_get_ringparam(struct net_device *dev,
> > ring->tx_pending = ring->tx_max_pending;
> > }
> >
> > +static bool virtnet_commit_rss_command(struct virtnet_info *vi)
> > +{
> > + struct net_device *dev = vi->dev;
> > + struct scatterlist sgs[4];
> > + unsigned int sg_buf_size;
> > +
> > + /* prepare sgs */
> > + sg_init_table(sgs, 4);
> > +
> > + sg_buf_size = offsetof(struct virtio_net_ctrl_rss, indirection_table);
> > + sg_set_buf(&sgs[0], &vi->ctrl->rss, sg_buf_size);
> > +
> > + sg_buf_size = sizeof(uint16_t) * vi->rss_indir_table_size;
> > + sg_set_buf(&sgs[1], vi->ctrl->rss.indirection_table, sg_buf_size);
> > +
> > + sg_buf_size = offsetof(struct virtio_net_ctrl_rss, key)
> > + - offsetof(struct virtio_net_ctrl_rss, max_tx_vq);
> > + sg_set_buf(&sgs[2], &vi->ctrl->rss.max_tx_vq, sg_buf_size);
> > +
> > + sg_buf_size = vi->rss_key_size;
> > + sg_set_buf(&sgs[3], vi->ctrl->rss.key, sg_buf_size);
> > +
> > + if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ,
> > + VIRTIO_NET_CTRL_MQ_RSS_CONFIG, sgs)) {
> > + dev_warn(&dev->dev, "VIRTIONET issue with committing RSS sgs\n");
> > + return false;
> > + }
> > + return true;
> > +}
> > +
> > +static void virtnet_init_default_rss(struct virtnet_info *vi)
> > +{
> > + u32 indir_val = 0;
> > + int i = 0;
> > +
> > + vi->ctrl->rss.hash_types = vi->rss_hash_types_supported;
> > + vi->ctrl->rss.indirection_table_mask = vi->rss_indir_table_size - 1;
>
> Is table size always a power of two?

Yes, it should be.

>
> > + vi->ctrl->rss.unclassified_queue = 0;
> > +
> > + for (; i < vi->rss_indir_table_size; ++i) {
> > + indir_val = ethtool_rxfh_indir_default(i, vi->curr_queue_pairs);
> > + vi->ctrl->rss.indirection_table[i] = indir_val;
> > + }
> > +
> > + vi->ctrl->rss.max_tx_vq = vi->curr_queue_pairs;
> > + vi->ctrl->rss.hash_key_length = vi->rss_key_size;
> > +
> > + netdev_rss_key_fill(vi->ctrl->rss.key, vi->rss_key_size);
> > +}
> > +
> >
> > static void virtnet_get_drvinfo(struct net_device *dev,
> > struct ethtool_drvinfo *info)
> > @@ -2412,6 +2487,71 @@ static void virtnet_update_settings(struct virtnet_info *vi)
> > vi->duplex = duplex;
> > }
> >
> > +static u32 virtnet_get_rxfh_key_size(struct net_device *dev)
> > +{
> > + return ((struct virtnet_info *)netdev_priv(dev))->rss_key_size;
> > +}
> > +
> > +static u32 virtnet_get_rxfh_indir_size(struct net_device *dev)
> > +{
> > + return ((struct virtnet_info *)netdev_priv(dev))->rss_indir_table_size;
> > +}
> > +
> > +static int virtnet_get_rxfh(struct net_device *dev, u32 *indir, u8 *key, u8 *hfunc)
> > +{
> > + struct virtnet_info *vi = netdev_priv(dev);
> > + int i;
> > +
> > + if (indir) {
> > + for (i = 0; i < vi->rss_indir_table_size; ++i)
> > + indir[i] = vi->ctrl->rss.indirection_table[i];
> > + }
> > +
> > + if (key)
> > + memcpy(key, vi->ctrl->rss.key, vi->rss_key_size);
> > +
> > + if (hfunc)
> > + *hfunc = ETH_RSS_HASH_TOP;
> > +
> > + return 0;
> > +}
> > +
> > +static int virtnet_set_rxfh(struct net_device *dev, const u32 *indir, const u8 *key, const u8 hfunc)
> > +{
> > + struct virtnet_info *vi = netdev_priv(dev);
> > + int i;
> > +
> > + if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
> > + return -EOPNOTSUPP;
> > +
> > + if (indir) {
> > + for (i = 0; i < vi->rss_indir_table_size; ++i)
> > + vi->ctrl->rss.indirection_table[i] = indir[i];
> > + }
> > + if (key)
> > + memcpy(vi->ctrl->rss.key, key, vi->rss_key_size);
> > +
> > + virtnet_commit_rss_command(vi);
> > +
> > + return 0;
> > +}
> > +
> > +static int virtnet_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info, u32 *rule_locs)
> > +{
> > + struct virtnet_info *vi = netdev_priv(dev);
> > + int rc = 0;
> > +
> > + switch (info->cmd) {
> > + case ETHTOOL_GRXRINGS:
> > + info->data = vi->curr_queue_pairs;
> > + break;
> > + default:
> > + rc = -EOPNOTSUPP;
> > + }
> > +
> > + return rc;
> > +}
> > +
> > static const struct ethtool_ops virtnet_ethtool_ops = {
> > .supported_coalesce_params = ETHTOOL_COALESCE_MAX_FRAMES,
> > .get_drvinfo = virtnet_get_drvinfo,
> > @@ -2427,6 +2567,11 @@ static const struct ethtool_ops virtnet_ethtool_ops = {
> > .set_link_ksettings = virtnet_set_link_ksettings,
> > .set_coalesce = virtnet_set_coalesce,
> > .get_coalesce = virtnet_get_coalesce,
> > + .get_rxfh_key_size = virtnet_get_rxfh_key_size,
> > + .get_rxfh_indir_size = virtnet_get_rxfh_indir_size,
> > + .get_rxfh = virtnet_get_rxfh,
> > + .set_rxfh = virtnet_set_rxfh,
> > + .get_rxnfc = virtnet_get_rxnfc,
> > };
> >
> > static void virtnet_freeze_down(struct virtio_device *vdev)
> > @@ -2679,6 +2824,16 @@ static int virtnet_set_features(struct net_device *dev,
> > vi->guest_offloads = offloads;
> > }
> >
> > + if ((dev->features ^ features) & NETIF_F_RXHASH) {
> > + if (features & NETIF_F_RXHASH)
> > + vi->ctrl->rss.hash_types = vi->rss_hash_types_supported;
> > + else
> > + vi->ctrl->rss.hash_types = VIRTIO_NET_HASH_REPORT_NONE;
> > +
> > + if (!virtnet_commit_rss_command(vi))
> > + return -EINVAL;
> > + }
> > +
> > return 0;
> > }
> >
> > @@ -3073,6 +3228,8 @@ static bool virtnet_validate_features(struct virtio_device *vdev)
> > "VIRTIO_NET_F_CTRL_VQ") ||
> > VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_MQ, "VIRTIO_NET_F_CTRL_VQ") ||
> > VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR,
> > + "VIRTIO_NET_F_CTRL_VQ") ||
> > + VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_RSS,
> > "VIRTIO_NET_F_CTRL_VQ"))) {
> > return false;
> > }
> > @@ -3113,13 +3270,14 @@ static int virtnet_probe(struct virtio_device *vdev)
> > u16 max_queue_pairs;
> > int mtu;
> >
> > - /* Find if host supports multiqueue virtio_net device */
> > - err = virtio_cread_feature(vdev, VIRTIO_NET_F_MQ,
> > - struct virtio_net_config,
> > - max_virtqueue_pairs, &max_queue_pairs);
> > + /* Find if host supports multiqueue/rss virtio_net device */
> > + max_queue_pairs = 1;
> > + if (virtio_has_feature(vdev, VIRTIO_NET_F_MQ) || virtio_has_feature(vdev, VIRTIO_NET_F_RSS))
> > + max_queue_pairs =
> > + virtio_cread16(vdev, offsetof(struct virtio_net_config, max_virtqueue_pairs));
>
> Instead of testing either feature and treating them as somewhat equal,
> shouldn't RSS be dependent on MQ?

No, RSS is dependent on CTRL_VQ. Technically RSS and MQ are similar features.

>
> >
> > /* We need at least 2 queue's */
> > - if (err || max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
> > + if (max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
> > max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
> > !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
> > max_queue_pairs = 1;
> > @@ -3207,6 +3365,23 @@ static int virtnet_probe(struct virtio_device *vdev)
> > if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
> > vi->mergeable_rx_bufs = true;
> >
> > + if (virtio_has_feature(vdev, VIRTIO_NET_F_RSS)) {
> > + vi->has_rss = true;
> > + vi->rss_indir_table_size =
> > + virtio_cread16(vdev, offsetof(struct virtio_net_config,
> > + rss_max_indirection_table_length));
> > + vi->rss_key_size =
> > + virtio_cread8(vdev, offsetof(struct virtio_net_config, rss_max_key_size));
> > +
> > + vi->rss_hash_types_supported =
> > + virtio_cread32(vdev, offsetof(struct virtio_net_config, supported_hash_types));
> > + vi->rss_hash_types_supported &=
> > + ~(VIRTIO_NET_RSS_HASH_TYPE_IP_EX |
> > + VIRTIO_NET_RSS_HASH_TYPE_TCP_EX |
> > + VIRTIO_NET_RSS_HASH_TYPE_UDP_EX);
> > +
> > + dev->hw_features |= NETIF_F_RXHASH;
>
> Only make the feature visible when the hash is actually reported in
> the skb, patch 3.

VirtioNET has two features: RSS(steering only) and hash(hash report in
vnet header)
Both features may be enabled/disabled separately:
1. rss on and hash off - packets steered to the corresponding vqs
2. rss off and hash on - packets steered by tap(like mq) but headers
have properly calculated hash.
3. rss on and hash on - packets steered to corresponding vqs and hash
is present in the header.

RXHASH feature allows the user to enable/disable the rss/hash(any combination).
I think it's a good idea to leave RXHASH in patch 2/4 to give the user
ability to manipulate the rss only feature.
But, if you think that it requires to move it to the 3/4, I'll do it.

>
> Also, clearly separate the feature patches (2) rss, (3) rxhash, (4)
> rxhash config.

Currently:
Patch 2/4 - adds VirtioNet rss feature.
Patch 3/4 - adds VirtioNet hash report feature.
Patch 4/4 - adds the ability to manipulate supported hash types.

Can you provide more detailed suggestions on how to move hunks?