On Fri, Mar 21, 2025 at 2:35 PM Akihiko Odaki <akihiko.od...@daynix.com> wrote:
>
> On 2025/03/21 9:35, Jason Wang wrote:
> > On Thu, Mar 20, 2025 at 1:36 PM Akihiko Odaki <akihiko.od...@daynix.com>
> > wrote:
> >>
> >> On 2025/03/20 10:50, Jason Wang wrote:
> >>> On Wed, Mar 19, 2025 at 12:48 PM Akihiko Odaki <akihiko.od...@daynix.com>
> >>> wrote:
> >>>>
> >>>> On 2025/03/19 10:43, Jason Wang wrote:
> >>>>> On Tue, Mar 18, 2025 at 5:57 PM Akihiko Odaki
> >>>>> <akihiko.od...@daynix.com> wrote:
> >>>>>>
> >>>>>> The new RSS configuration structures allow easily constructing data for
> >>>>>> VIRTIO_NET_CTRL_MQ_RSS_CONFIG as they strictly follow the order of data
> >>>>>> for the command.
> >>>>>>
> >>>>>> Signed-off-by: Akihiko Odaki <akihiko.od...@daynix.com>
> >>>>>> ---
> >>>>>> drivers/net/virtio_net.c | 117
> >>>>>> +++++++++++++++++------------------------------
> >>>>>> 1 file changed, 43 insertions(+), 74 deletions(-)
> >>>>>>
> >>>>>> diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> >>>>>> index d1ed544ba03a..4153a0a5f278 100644
> >>>>>> --- a/drivers/net/virtio_net.c
> >>>>>> +++ b/drivers/net/virtio_net.c
> >>>>>> @@ -360,24 +360,7 @@ struct receive_queue {
> >>>>>> struct xdp_buff **xsk_buffs;
> >>>>>> };
> >>>>>>
> >>>>>> -/* 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
> >>>>>> -struct virtio_net_ctrl_rss {
> >>>>>> - __le32 hash_types;
> >>>>>> - __le16 indirection_table_mask;
> >>>>>> - __le16 unclassified_queue;
> >>>>>> - __le16 hash_cfg_reserved; /* for HASH_CONFIG (see
> >>>>>> virtio_net_hash_config for details) */
> >>>>>> - __le16 max_tx_vq;
> >>>>>> - u8 hash_key_length;
> >>>>>> - u8 key[VIRTIO_NET_RSS_MAX_KEY_SIZE];
> >>>>>> -
> >>>>>> - __le16 *indirection_table;
> >>>>>> -};
> >>>>>>
> >>>>>> /* Control VQ buffers: protected by the rtnl lock */
> >>>>>> struct control_buf {
> >>>>>> @@ -421,7 +404,9 @@ struct virtnet_info {
> >>>>>> u16 rss_indir_table_size;
> >>>>>> u32 rss_hash_types_supported;
> >>>>>> u32 rss_hash_types_saved;
> >>>>>> - struct virtio_net_ctrl_rss rss;
> >>>>>> + struct virtio_net_rss_config_hdr *rss_hdr;
> >>>>>> + struct virtio_net_rss_config_trailer rss_trailer;
> >>>>>> + u8 rss_hash_key_data[VIRTIO_NET_RSS_MAX_KEY_SIZE];
> >>>>>>
> >>>>>> /* Has control virtqueue */
> >>>>>> bool has_cvq;
> >>>>>> @@ -523,23 +508,16 @@ enum virtnet_xmit_type {
> >>>>>> VIRTNET_XMIT_TYPE_XSK,
> >>>>>> };
> >>>>>>
> >>>>>> -static int rss_indirection_table_alloc(struct virtio_net_ctrl_rss
> >>>>>> *rss, u16 indir_table_size)
> >>>>>> +static size_t virtnet_rss_hdr_size(const struct virtnet_info *vi)
> >>>>>> {
> >>>>>> - if (!indir_table_size) {
> >>>>>> - rss->indirection_table = NULL;
> >>>>>> - return 0;
> >>>>>> - }
> >>>>>> + u16 indir_table_size = vi->has_rss ? vi->rss_indir_table_size
> >>>>>> : 1;
> >>>>>>
> >>>>>> - rss->indirection_table = kmalloc_array(indir_table_size,
> >>>>>> sizeof(u16), GFP_KERNEL);
> >>>>>> - if (!rss->indirection_table)
> >>>>>> - return -ENOMEM;
> >>>>>> -
> >>>>>> - return 0;
> >>>>>> + return struct_size(vi->rss_hdr, indirection_table,
> >>>>>> indir_table_size);
> >>>>>> }
> >>>>>>
> >>>>>> -static void rss_indirection_table_free(struct virtio_net_ctrl_rss
> >>>>>> *rss)
> >>>>>> +static size_t virtnet_rss_trailer_size(const struct virtnet_info *vi)
> >>>>>> {
> >>>>>> - kfree(rss->indirection_table);
> >>>>>> + return struct_size(&vi->rss_trailer, hash_key_data,
> >>>>>> vi->rss_key_size);
> >>>>>> }
> >>>>>>
> >>>>>> /* We use the last two bits of the pointer to distinguish the xmit
> >>>>>> type. */
> >>>>>> @@ -3576,15 +3554,16 @@ static void
> >>>>>> virtnet_rss_update_by_qpairs(struct virtnet_info *vi, u16 queue_pair
> >>>>>>
> >>>>>> for (; i < vi->rss_indir_table_size; ++i) {
> >>>>>> indir_val = ethtool_rxfh_indir_default(i,
> >>>>>> queue_pairs);
> >>>>>> - vi->rss.indirection_table[i] = cpu_to_le16(indir_val);
> >>>>>> + vi->rss_hdr->indirection_table[i] =
> >>>>>> cpu_to_le16(indir_val);
> >>>>>> }
> >>>>>> - vi->rss.max_tx_vq = cpu_to_le16(queue_pairs);
> >>>>>> + vi->rss_trailer.max_tx_vq = cpu_to_le16(queue_pairs);
> >>>>>> }
> >>>>>>
> >>>>>> static int virtnet_set_queues(struct virtnet_info *vi, u16
> >>>>>> queue_pairs)
> >>>>>> {
> >>>>>> struct virtio_net_ctrl_mq *mq __free(kfree) = NULL;
> >>>>>> - struct virtio_net_ctrl_rss old_rss;
> >>>>>> + struct virtio_net_rss_config_hdr *old_rss_hdr;
> >>>>>> + struct virtio_net_rss_config_trailer old_rss_trailer;
> >>>>>> struct net_device *dev = vi->dev;
> >>>>>> struct scatterlist sg;
> >>>>>>
> >>>>>> @@ -3599,24 +3578,28 @@ static int virtnet_set_queues(struct
> >>>>>> virtnet_info *vi, u16 queue_pairs)
> >>>>>> * update (VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET below) and
> >>>>>> return directly.
> >>>>>> */
> >>>>>> if (vi->has_rss && !netif_is_rxfh_configured(dev)) {
> >>>>>> - memcpy(&old_rss, &vi->rss, sizeof(old_rss));
> >>>>>> - if (rss_indirection_table_alloc(&vi->rss,
> >>>>>> vi->rss_indir_table_size)) {
> >>>>>> - vi->rss.indirection_table =
> >>>>>> old_rss.indirection_table;
> >>>>>> + old_rss_hdr = vi->rss_hdr;
> >>>>>> + old_rss_trailer = vi->rss_trailer;
> >>>>>> + vi->rss_hdr = kmalloc(virtnet_rss_hdr_size(vi),
> >>>>>> GFP_KERNEL);
> >>>>>> + if (!vi->rss_hdr) {
> >>>>>> + vi->rss_hdr = old_rss_hdr;
> >>>>>> return -ENOMEM;
> >>>>>> }
> >>>>>>
> >>>>>> + *vi->rss_hdr = *old_rss_hdr;
> >>>>>> virtnet_rss_update_by_qpairs(vi, queue_pairs);
> >>>>>>
> >>>>>> if (!virtnet_commit_rss_command(vi)) {
> >>>>>> /* restore ctrl_rss if commit_rss_command
> >>>>>> failed */
> >>>>>> - rss_indirection_table_free(&vi->rss);
> >>>>>> - memcpy(&vi->rss, &old_rss, sizeof(old_rss));
> >>>>>> + kfree(vi->rss_hdr);
> >>>>>> + vi->rss_hdr = old_rss_hdr;
> >>>>>> + vi->rss_trailer = old_rss_trailer;
> >>>>>>
> >>>>>> dev_warn(&dev->dev, "Fail to set num of
> >>>>>> queue pairs to %d, because committing RSS failed\n",
> >>>>>> queue_pairs);
> >>>>>> return -EINVAL;
> >>>>>> }
> >>>>>> - rss_indirection_table_free(&old_rss);
> >>>>>> + kfree(old_rss_hdr);
> >>>>>> goto succ;
> >>>>>> }
> >>>>>>
> >>>>>> @@ -4059,28 +4042,12 @@ static int virtnet_set_ringparam(struct
> >>>>>> net_device *dev,
> >>>>>> 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;
> >>>>>> + struct scatterlist sgs[2];
> >>>>>>
> >>>>>> /* prepare sgs */
> >>>>>> - sg_init_table(sgs, 4);
> >>>>>> -
> >>>>>> - sg_buf_size = offsetof(struct virtio_net_ctrl_rss,
> >>>>>> hash_cfg_reserved);
> >>>>>> - sg_set_buf(&sgs[0], &vi->rss, sg_buf_size);
> >>>>>> -
> >>>>>> - if (vi->has_rss) {
> >>>>>> - sg_buf_size = sizeof(uint16_t) *
> >>>>>> vi->rss_indir_table_size;
> >>>>>> - sg_set_buf(&sgs[1], vi->rss.indirection_table,
> >>>>>> sg_buf_size);
> >>>>>> - } else {
> >>>>>> - sg_set_buf(&sgs[1], &vi->rss.hash_cfg_reserved,
> >>>>>> sizeof(uint16_t));
> >>>>>> - }
> >>>>>> -
> >>>>>> - 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->rss.max_tx_vq, sg_buf_size);
> >>>>>> -
> >>>>>> - sg_buf_size = vi->rss_key_size;
> >>>>>> - sg_set_buf(&sgs[3], vi->rss.key, sg_buf_size);
> >>>>>> + sg_init_table(sgs, 2);
> >>>>>> + sg_set_buf(&sgs[0], vi->rss_hdr, virtnet_rss_hdr_size(vi));
> >>>>>> + sg_set_buf(&sgs[1], &vi->rss_trailer,
> >>>>>> virtnet_rss_trailer_size(vi));
> >>>>>
> >>>>> So I still see this:
> >>>>>
> >>>>> if (vi->has_rss || vi->has_rss_hash_report) {
> >>>>> if (!virtnet_commit_rss_command(vi)) {
> >>>>>
> >>>>> Should we introduce a hash config helper instead?
> >>>>
> >>>> I think it's fine to use virtnet_commit_rss_command() for hash
> >>>> reporting. struct virtio_net_hash_config and struct
> >>>> virtio_net_rss_config are defined to have a common layout to allow
> >>>> sharing this kind of logic.
> >>>
> >>> Well, this trick won't work if the reserved field in hash_config is
> >>> used in the future.
> >>
> >> Right, but we can add a hash config helper when that happens. It will
> >> only result in a duplication of logic for now.
> >>
> >> Regards,
> >> Akihiko Odaki
> >
> > That's tricky as the cvq commands were designed to be used separately.
> > Let's use a separate helper and virtio_net_hash_config uAPIs now.
>
> It's not tricky but is explicitly stated in the spec. 5.1.6.5.6.4 "Hash
> calculation" says:
> > Field reserved MUST contain zeroes. It is defined to make the
> > structure to match the layout of virtio_net_rss_config structure,
> > defined in 5.1.6.5.7.
This is kind of not elegant, but it's too late to fix.
Thanks
>
> By the way, I found it says field reserved MUST contain zeros but we do
> nothing to ensure that. I'll write a fix for that.
>
> Regards,
> Akihiko Odaki
>
> >
> > Thanks
> >
> >>
> >>>
> >>> Thanks
> >>>
> >>>>
> >>>> Regards,
> >>>> Akihiko Odaki
> >>>>
> >>>>>
> >>>>> Thanks
> >>>>>
> >>>>
> >>>
> >>
> >
>
