On Mon, Mar 18, 2024 at 7:06 PM Xuan Zhuo <xuanzhuo@xxxxxxxxxxxxxxxxx> wrote: > > As the spec https://github.com/oasis-tcs/virtio-spec/commit/42f389989823039724f95bbbd243291ab0064f82 > > make virtio-net support getting the stats from the device by ethtool -S > <eth0>. > > Due to the numerous descriptors stats, an organization method is > required. For this purpose, I have introduced the "virtnet_stats_map". > Utilizing this array simplifies coding tasks such as generating field > names, calculating buffer sizes for requests and responses, and parsing > replies from the device. By iterating over the "virtnet_stats_map," > these operations become more streamlined and efficient. > > NIC statistics: > rx0_packets: 582951 > rx0_bytes: 155307077 > rx0_drops: 0 > rx0_xdp_packets: 0 > rx0_xdp_tx: 0 > rx0_xdp_redirects: 0 > rx0_xdp_drops: 0 > rx0_kicks: 17007 > rx0_hw_packets: 2179409 > rx0_hw_bytes: 510015040 > rx0_hw_notifications: 0 > rx0_hw_interrupts: 0 > rx0_hw_drops: 12964 > rx0_hw_drop_overruns: 0 > rx0_hw_csum_valid: 2179409 > rx0_hw_csum_none: 0 > rx0_hw_csum_bad: 0 > rx0_hw_needs_csum: 2179409 > rx0_hw_ratelimit_packets: 0 > rx0_hw_ratelimit_bytes: 0 > tx0_packets: 15361 > tx0_bytes: 1918970 > tx0_xdp_tx: 0 > tx0_xdp_tx_drops: 0 > tx0_kicks: 15361 > tx0_timeouts: 0 > tx0_hw_packets: 32272 > tx0_hw_bytes: 4311698 > tx0_hw_notifications: 0 > tx0_hw_interrupts: 0 > tx0_hw_drops: 0 > tx0_hw_drop_malformed: 0 > tx0_hw_csum_none: 0 > tx0_hw_needs_csum: 32272 > tx0_hw_ratelimit_packets: 0 > tx0_hw_ratelimit_bytes: 0 > > Signed-off-by: Xuan Zhuo <xuanzhuo@xxxxxxxxxxxxxxxxx> > --- > drivers/net/virtio_net.c | 401 ++++++++++++++++++++++++++++++++++++++- > 1 file changed, 397 insertions(+), 4 deletions(-) > > diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c > index 8cb5bdd7ad91..70c1d4e850e0 100644 > --- a/drivers/net/virtio_net.c > +++ b/drivers/net/virtio_net.c > @@ -128,6 +128,129 @@ static const struct virtnet_stat_desc virtnet_rq_stats_desc[] = { > #define VIRTNET_SQ_STATS_LEN ARRAY_SIZE(virtnet_sq_stats_desc) > #define VIRTNET_RQ_STATS_LEN ARRAY_SIZE(virtnet_rq_stats_desc) > > +#define VIRTNET_STATS_DESC_CQ(name) \ > + {#name, offsetof(struct virtio_net_stats_cvq, name)} > + > +#define VIRTNET_STATS_DESC_RX(class, name) \ > + {#name, offsetof(struct virtio_net_stats_rx_ ## class, rx_ ## name)} > + > +#define VIRTNET_STATS_DESC_TX(class, name) \ > + {#name, offsetof(struct virtio_net_stats_tx_ ## class, tx_ ## name)} > + > +static const struct virtnet_stat_desc virtnet_stats_cvq_desc[] = { > + VIRTNET_STATS_DESC_CQ(command_num), > + VIRTNET_STATS_DESC_CQ(ok_num), > +}; > + > +static const struct virtnet_stat_desc virtnet_stats_rx_basic_desc[] = { > + VIRTNET_STATS_DESC_RX(basic, packets), > + VIRTNET_STATS_DESC_RX(basic, bytes), > + > + VIRTNET_STATS_DESC_RX(basic, notifications), > + VIRTNET_STATS_DESC_RX(basic, interrupts), > + > + VIRTNET_STATS_DESC_RX(basic, drops), > + VIRTNET_STATS_DESC_RX(basic, drop_overruns), > +}; > + > +static const struct virtnet_stat_desc virtnet_stats_tx_basic_desc[] = { > + VIRTNET_STATS_DESC_TX(basic, packets), > + VIRTNET_STATS_DESC_TX(basic, bytes), > + > + VIRTNET_STATS_DESC_TX(basic, notifications), > + VIRTNET_STATS_DESC_TX(basic, interrupts), > + > + VIRTNET_STATS_DESC_TX(basic, drops), > + VIRTNET_STATS_DESC_TX(basic, drop_malformed), > +}; > + > +static const struct virtnet_stat_desc virtnet_stats_rx_csum_desc[] = { > + VIRTNET_STATS_DESC_RX(csum, csum_valid), > + VIRTNET_STATS_DESC_RX(csum, needs_csum), > + > + VIRTNET_STATS_DESC_RX(csum, csum_none), > + VIRTNET_STATS_DESC_RX(csum, csum_bad), > +}; > + > +static const struct virtnet_stat_desc virtnet_stats_tx_csum_desc[] = { > + VIRTNET_STATS_DESC_TX(csum, needs_csum), > + VIRTNET_STATS_DESC_TX(csum, csum_none), > +}; > + > +static const struct virtnet_stat_desc virtnet_stats_rx_gso_desc[] = { > + VIRTNET_STATS_DESC_RX(gso, gso_packets), > + VIRTNET_STATS_DESC_RX(gso, gso_bytes), > + VIRTNET_STATS_DESC_RX(gso, gso_packets_coalesced), > + VIRTNET_STATS_DESC_RX(gso, gso_bytes_coalesced), > +}; > + > +static const struct virtnet_stat_desc virtnet_stats_tx_gso_desc[] = { > + VIRTNET_STATS_DESC_TX(gso, gso_packets), > + VIRTNET_STATS_DESC_TX(gso, gso_bytes), > + VIRTNET_STATS_DESC_TX(gso, gso_segments), > + VIRTNET_STATS_DESC_TX(gso, gso_segments_bytes), > + VIRTNET_STATS_DESC_TX(gso, gso_packets_noseg), > + VIRTNET_STATS_DESC_TX(gso, gso_bytes_noseg), > +}; > + > +static const struct virtnet_stat_desc virtnet_stats_rx_speed_desc[] = { > + VIRTNET_STATS_DESC_RX(speed, ratelimit_packets), > + VIRTNET_STATS_DESC_RX(speed, ratelimit_bytes), > +}; > + > +static const struct virtnet_stat_desc virtnet_stats_tx_speed_desc[] = { > + VIRTNET_STATS_DESC_TX(speed, ratelimit_packets), > + VIRTNET_STATS_DESC_TX(speed, ratelimit_bytes), > +}; > + > +#define VIRTNET_Q_TYPE_RX 0 > +#define VIRTNET_Q_TYPE_TX 1 > +#define VIRTNET_Q_TYPE_CQ 2 > + > +struct virtnet_stats_map { > + /* The stat type in bitmap. */ > + u64 stat_type; > + > + /* The bytes of the response for the stat. */ > + u32 len; > + > + /* The num of the response fields for the stat. */ > + u32 num; > + > + /* The type of queue corresponding to the statistics. (cq, rq, sq) */ > + u32 queue_type; > + > + /* The reply type of the stat. */ > + u8 reply_type; > + > + /* Describe the name and the offset in the response. */ > + const struct virtnet_stat_desc *desc; > +}; > + > +#define VIRTNET_DEVICE_STATS_MAP_ITEM(TYPE, type, queue_type) \ > + { \ > + VIRTIO_NET_STATS_TYPE_##TYPE, \ > + sizeof(struct virtio_net_stats_ ## type), \ > + ARRAY_SIZE(virtnet_stats_ ## type ##_desc), \ > + VIRTNET_Q_TYPE_##queue_type, \ > + VIRTIO_NET_STATS_TYPE_REPLY_##TYPE, \ > + &virtnet_stats_##type##_desc[0] \ > + } > + > +static struct virtnet_stats_map virtio_net_stats_map[] = { > + VIRTNET_DEVICE_STATS_MAP_ITEM(CVQ, cvq, CQ), > + > + VIRTNET_DEVICE_STATS_MAP_ITEM(RX_BASIC, rx_basic, RX), > + VIRTNET_DEVICE_STATS_MAP_ITEM(RX_CSUM, rx_csum, RX), > + VIRTNET_DEVICE_STATS_MAP_ITEM(RX_GSO, rx_gso, RX), > + VIRTNET_DEVICE_STATS_MAP_ITEM(RX_SPEED, rx_speed, RX), > + > + VIRTNET_DEVICE_STATS_MAP_ITEM(TX_BASIC, tx_basic, TX), > + VIRTNET_DEVICE_STATS_MAP_ITEM(TX_CSUM, tx_csum, TX), > + VIRTNET_DEVICE_STATS_MAP_ITEM(TX_GSO, tx_gso, TX), > + VIRTNET_DEVICE_STATS_MAP_ITEM(TX_SPEED, tx_speed, TX), > +}; I think the reason you did this is to ease the future extensions but multiple levels of nested macros makes the code hard to review. Any way to eliminate this? > + > struct virtnet_interrupt_coalesce { > u32 max_packets; > u32 max_usecs; > @@ -244,6 +367,7 @@ struct control_buf { > struct virtio_net_ctrl_coal_tx coal_tx; > struct virtio_net_ctrl_coal_rx coal_rx; > struct virtio_net_ctrl_coal_vq coal_vq; > + struct virtio_net_stats_capabilities stats_cap; > }; > > struct virtnet_info { > @@ -329,6 +453,8 @@ struct virtnet_info { > > /* failover when STANDBY feature enabled */ > struct failover *failover; > + > + u64 device_stats_cap; > }; > > struct padded_vnet_hdr { > @@ -389,6 +515,17 @@ static int rxq2vq(int rxq) > return rxq * 2; > } > > +static int vq_type(struct virtnet_info *vi, int qid) > +{ > + if (qid == vi->max_queue_pairs * 2) > + return VIRTNET_Q_TYPE_CQ; > + > + if (qid % 2) > + return VIRTNET_Q_TYPE_TX; > + > + return VIRTNET_Q_TYPE_RX; > +} > + > static inline struct virtio_net_common_hdr * > skb_vnet_common_hdr(struct sk_buff *skb) > { > @@ -3263,6 +3400,223 @@ static int virtnet_set_channels(struct net_device *dev, > return err; > } > > +static void virtnet_get_hw_stats_string(struct virtnet_info *vi, int type, int qid, u8 **data) > +{ > + struct virtnet_stats_map *m; > + int i, j; > + u8 *p = *data; > + > + if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_DEVICE_STATS)) > + return; > + > + for (i = 0; i < ARRAY_SIZE(virtio_net_stats_map); ++i) { > + m = &virtio_net_stats_map[i]; > + > + if (m->queue_type != type) > + continue; > + > + if (!(vi->device_stats_cap & m->stat_type)) > + continue; > + > + for (j = 0; j < m->num; ++j) { > + switch (type) { > + case VIRTNET_Q_TYPE_RX: > + ethtool_sprintf(&p, "rx_queue_hw_%u_%s", qid, m->desc[j].desc); > + break; > + > + case VIRTNET_Q_TYPE_TX: > + ethtool_sprintf(&p, "tx_queue_hw_%u_%s", qid, m->desc[j].desc); > + break; > + > + case VIRTNET_Q_TYPE_CQ: > + ethtool_sprintf(&p, "cq_hw_%s", m->desc[j].desc); > + break; > + } > + } > + } > + > + *data = p; > +} > + > +struct virtnet_stats_ctx { > + u32 desc_num[3]; > + > + u32 bitmap[3]; > + > + u32 size[3]; > + > + u64 *data; > +}; Let's explain the meaning of each field here. > + > +static void virtnet_stats_ctx_init(struct virtnet_info *vi, > + struct virtnet_stats_ctx *ctx, > + u64 *data) > +{ > + struct virtnet_stats_map *m; > + int i; > + > + ctx->data = data; > + > + for (i = 0; i < ARRAY_SIZE(virtio_net_stats_map); ++i) { > + m = &virtio_net_stats_map[i]; > + > + if (!(vi->device_stats_cap & m->stat_type)) > + continue; > + > + ctx->bitmap[m->queue_type] |= m->stat_type; > + ctx->desc_num[m->queue_type] += m->num; > + ctx->size[m->queue_type] += m->len; > + } > +} > + > +/* virtnet_fill_stats - copy the stats to ethtool -S > + * The stats source is the device. > + * > + * @vi: virtio net info > + * @qid: the vq id > + * @ctx: stats ctx (initiated by virtnet_stats_ctx_init()) > + * @base: pointer to the device reply. > + * @type: the type of the device reply > + */ > +static void virtnet_fill_stats(struct virtnet_info *vi, u32 qid, > + struct virtnet_stats_ctx *ctx, > + const u8 *base, u8 type) > +{ > + u32 queue_type, num_rx, num_tx, num_cq; > + struct virtnet_stats_map *m; > + u64 offset, bitmap; > + const __le64 *v; > + int i, j; > + > + num_rx = VIRTNET_RQ_STATS_LEN + ctx->desc_num[VIRTNET_Q_TYPE_RX]; > + num_tx = VIRTNET_SQ_STATS_LEN + ctx->desc_num[VIRTNET_Q_TYPE_TX]; > + num_cq = ctx->desc_num[VIRTNET_Q_TYPE_CQ]; > + > + queue_type = vq_type(vi, qid); > + bitmap = ctx->bitmap[queue_type]; > + offset = 0; > + > + if (queue_type == VIRTNET_Q_TYPE_TX) { > + offset = num_cq + num_rx * vi->curr_queue_pairs + num_tx * (qid / 2); > + offset += VIRTNET_SQ_STATS_LEN; > + } else if (queue_type == VIRTNET_Q_TYPE_RX) { > + offset = num_cq + num_rx * (qid / 2) + VIRTNET_RQ_STATS_LEN; > + } > + > + for (i = 0; i < ARRAY_SIZE(virtio_net_stats_map); ++i) { > + m = &virtio_net_stats_map[i]; > + > + if (m->stat_type & bitmap) > + offset += m->num; > + > + if (type != m->reply_type) > + continue; > + > + for (j = 0; j < m->num; ++j) { > + v = (const __le64 *)(base + m->desc[j].offset); > + ctx->data[offset + j] = le64_to_cpu(*v); > + } > + > + break; > + } > +} > + > +static int __virtnet_get_hw_stats(struct virtnet_info *vi, > + struct virtnet_stats_ctx *ctx, > + struct virtio_net_ctrl_queue_stats *req, > + int req_size, void *reply, int res_size) > +{ > + struct virtio_net_stats_reply_hdr *hdr; > + struct scatterlist sgs_in, sgs_out; > + void *p; > + u32 qid; > + int ok; > + > + sg_init_one(&sgs_out, req, req_size); > + sg_init_one(&sgs_in, reply, res_size); > + > + ok = virtnet_send_command(vi, VIRTIO_NET_CTRL_STATS, > + VIRTIO_NET_CTRL_STATS_GET, > + &sgs_out, &sgs_in); > + kfree(req); I'd suggest letting the caller free this for simplicity. > + > + if (!ok) { > + kfree(reply); > + return ok; > + } > + > + for (p = reply; p - reply < res_size; p += le16_to_cpu(hdr->size)) { hdr->size is under the control of a device which might be malicious? Btw, I'd expect that type implies size but looks not, any reason for that? > + hdr = p; > + qid = le16_to_cpu(hdr->vq_index); > + virtnet_fill_stats(vi, qid, ctx, p, hdr->type); > + } > + > + kfree(reply); As caller has the logic for err handling like: reply = kmalloc(res_size, GFP_KERNEL); if (!reply) { kfree(req); return -ENOMEM; } So let's free it from the caller. > + return 0; > +} > + > +static void virtnet_make_stat_req(struct virtnet_info *vi, > + struct virtnet_stats_ctx *ctx, > + struct virtio_net_ctrl_queue_stats *req, > + int qid, int *idx) > +{ > + int qtype = vq_type(vi, qid); > + u64 bitmap = ctx->bitmap[qtype]; > + > + if (!bitmap) > + return; > + > + req->stats[*idx].vq_index = cpu_to_le16(qid); > + req->stats[*idx].types_bitmap[0] = cpu_to_le64(bitmap); > + *idx += 1; > +} > + > +static int virtnet_get_hw_stats(struct virtnet_info *vi, > + struct virtnet_stats_ctx *ctx) > +{ > + struct virtio_net_ctrl_queue_stats *req; > + int qnum, i, j, res_size, qtype, last_vq; > + void *reply; > + > + if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_DEVICE_STATS)) > + return 0; > + > + last_vq = vi->curr_queue_pairs * 2 - 1; > + > + qnum = 0; > + res_size = 0; > + for (i = 0; i <= last_vq ; ++i) { > + qtype = vq_type(vi, i); > + if (ctx->bitmap[qtype]) { > + ++qnum; > + res_size += ctx->size[qtype]; > + } > + } > + > + if (ctx->bitmap[VIRTNET_Q_TYPE_CQ]) { > + res_size += ctx->size[VIRTNET_Q_TYPE_CQ]; > + qnum += 1; > + } > + > + req = kcalloc(qnum, sizeof(*req), GFP_KERNEL); > + if (!req) > + return -ENOMEM; > + > + reply = kmalloc(res_size, GFP_KERNEL); > + if (!reply) { > + kfree(req); > + return -ENOMEM; > + } > + > + j = 0; > + for (i = 0; i <= last_vq ; ++i) > + virtnet_make_stat_req(vi, ctx, req, i, &j); > + > + virtnet_make_stat_req(vi, ctx, req, vi->max_queue_pairs * 2, &j); Instead of preparing those on the fly, could we prepare the request during the probe instead of here? As most of the field (except the data) won't be changed. > + > + return __virtnet_get_hw_stats(vi, ctx, req, sizeof(*req) * j, reply, res_size); > +} > + > static void virtnet_get_strings(struct net_device *dev, u32 stringset, u8 *data) > { > struct virtnet_info *vi = netdev_priv(dev); > @@ -3271,16 +3625,22 @@ static void virtnet_get_strings(struct net_device *dev, u32 stringset, u8 *data) > > switch (stringset) { > case ETH_SS_STATS: > + virtnet_get_hw_stats_string(vi, VIRTNET_Q_TYPE_CQ, 0, &p); > + > for (i = 0; i < vi->curr_queue_pairs; i++) { > for (j = 0; j < VIRTNET_RQ_STATS_LEN; j++) > ethtool_sprintf(&p, "rx%u_%s", i, > virtnet_rq_stats_desc[j].desc); > + > + virtnet_get_hw_stats_string(vi, VIRTNET_Q_TYPE_RX, i, &p); > } > > for (i = 0; i < vi->curr_queue_pairs; i++) { > for (j = 0; j < VIRTNET_SQ_STATS_LEN; j++) > ethtool_sprintf(&p, "tx%u_%s", i, > virtnet_sq_stats_desc[j].desc); > + > + virtnet_get_hw_stats_string(vi, VIRTNET_Q_TYPE_TX, i, &p); > } > break; > } > @@ -3289,11 +3649,35 @@ static void virtnet_get_strings(struct net_device *dev, u32 stringset, u8 *data) > static int virtnet_get_sset_count(struct net_device *dev, int sset) > { > struct virtnet_info *vi = netdev_priv(dev); > + struct virtnet_stats_ctx ctx = {0}; > + u32 pair_count; > > switch (sset) { > case ETH_SS_STATS: > - return vi->curr_queue_pairs * (VIRTNET_RQ_STATS_LEN + > - VIRTNET_SQ_STATS_LEN); > + if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_DEVICE_STATS) && > + !vi->device_stats_cap) { > + struct scatterlist sg; > + > + sg_init_one(&sg, &vi->ctrl->stats_cap, sizeof(vi->ctrl->stats_cap)); > + > + if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_STATS, > + VIRTIO_NET_CTRL_STATS_QUERY, > + NULL, &sg)) { > + dev_warn(&dev->dev, "Fail to get stats capability\n"); Should we fail here? > + } else { > + __le64 v; > + > + v = vi->ctrl->stats_cap.supported_stats_types[0]; > + vi->device_stats_cap = le64_to_cpu(v); > + } > + } > + > + virtnet_stats_ctx_init(vi, &ctx, NULL); > + > + pair_count = VIRTNET_RQ_STATS_LEN + VIRTNET_SQ_STATS_LEN; > + pair_count += ctx.desc_num[VIRTNET_Q_TYPE_RX] + ctx.desc_num[VIRTNET_Q_TYPE_TX]; > + > + return ctx.desc_num[VIRTNET_Q_TYPE_CQ] + vi->curr_queue_pairs * pair_count; I wonder why we don't do this during the probe? > default: > return -EOPNOTSUPP; > } > @@ -3303,11 +3687,18 @@ static void virtnet_get_ethtool_stats(struct net_device *dev, > struct ethtool_stats *stats, u64 *data) > { > struct virtnet_info *vi = netdev_priv(dev); > - unsigned int idx = 0, start, i, j; > + struct virtnet_stats_ctx ctx = {0}; > + unsigned int idx, start, i, j; > const u8 *stats_base; > const u64_stats_t *p; > size_t offset; > > + virtnet_stats_ctx_init(vi, &ctx, data); > + if (virtnet_get_hw_stats(vi, &ctx)) > + dev_warn(&vi->dev->dev, "Failed to get hw stats.\n"); > + > + idx = ctx.desc_num[VIRTNET_Q_TYPE_CQ]; > + > for (i = 0; i < vi->curr_queue_pairs; i++) { > struct receive_queue *rq = &vi->rq[i]; > > @@ -3321,6 +3712,7 @@ static void virtnet_get_ethtool_stats(struct net_device *dev, > } > } while (u64_stats_fetch_retry(&rq->stats.syncp, start)); > idx += VIRTNET_RQ_STATS_LEN; > + idx += ctx.desc_num[VIRTNET_Q_TYPE_RX]; > } > > for (i = 0; i < vi->curr_queue_pairs; i++) { > @@ -3336,6 +3728,7 @@ static void virtnet_get_ethtool_stats(struct net_device *dev, > } > } while (u64_stats_fetch_retry(&sq->stats.syncp, start)); > idx += VIRTNET_SQ_STATS_LEN; > + idx += ctx.desc_num[VIRTNET_Q_TYPE_TX]; > } > } > > @@ -4963,7 +5356,7 @@ static struct virtio_device_id id_table[] = { > VIRTIO_NET_F_SPEED_DUPLEX, VIRTIO_NET_F_STANDBY, \ > VIRTIO_NET_F_RSS, VIRTIO_NET_F_HASH_REPORT, VIRTIO_NET_F_NOTF_COAL, \ > VIRTIO_NET_F_VQ_NOTF_COAL, \ > - VIRTIO_NET_F_GUEST_HDRLEN > + VIRTIO_NET_F_GUEST_HDRLEN, VIRTIO_NET_F_DEVICE_STATS > > static unsigned int features[] = { > VIRTNET_FEATURES, > -- > 2.32.0.3.g01195cf9f > Thanks
