Re: [PATCH 4/4] vdpa_sim_net: vendor satistics

[Jason Wang <jasowang@redhat.com>]

在 2022/12/21 21:52, Stefano Garzarella 写道:
> On Wed, Dec 21, 2022 at 02:16:52PM +0800, Jason Wang wrote:
>> This patch adds support for basic vendor stats that include counters
>> for tx, rx and cvq.
>>
>> Signed-off-by: Jason Wang <jasowang@redhat.com>
>> ---
>> drivers/vdpa/vdpa_sim/vdpa_sim.c     |   2 +
>> drivers/vdpa/vdpa_sim/vdpa_sim_net.c | 215 ++++++++++++++++++++++++++-
>> 2 files changed, 211 insertions(+), 6 deletions(-)
>>
>> diff --git a/drivers/vdpa/vdpa_sim/vdpa_sim.c 
>> b/drivers/vdpa/vdpa_sim/vdpa_sim.c
>> index 02e892f819e7..595d9d5a372f 100644
>> --- a/drivers/vdpa/vdpa_sim/vdpa_sim.c
>> +++ b/drivers/vdpa/vdpa_sim/vdpa_sim.c
>> @@ -755,8 +755,10 @@ static const struct vdpa_config_ops 
>> vdpasim_batch_config_ops = {
>>     .set_vq_cb              = vdpasim_set_vq_cb,
>>     .set_vq_ready           = vdpasim_set_vq_ready,
>>     .get_vq_ready           = vdpasim_get_vq_ready,
>> +    .get_vendor_vq_stats    = vdpasim_get_vq_stats,
>>     .set_vq_state           = vdpasim_set_vq_state,
>>     .get_vq_state           = vdpasim_get_vq_state,
>> +    .get_vendor_vq_stats    = vdpasim_get_vq_stats,
>>     .get_vq_align           = vdpasim_get_vq_align,
>>     .get_vq_group           = vdpasim_get_vq_group,
>>     .get_device_features    = vdpasim_get_device_features,
>
> This should go with the previous commit and defining it once.


Right, fixed.


>
>> diff --git a/drivers/vdpa/vdpa_sim/vdpa_sim_net.c 
>> b/drivers/vdpa/vdpa_sim/vdpa_sim_net.c
>> index 20cd5cdff919..3c05e932d90d 100644
>> --- a/drivers/vdpa/vdpa_sim/vdpa_sim_net.c
>> +++ b/drivers/vdpa/vdpa_sim/vdpa_sim_net.c
>> @@ -15,6 +15,7 @@
>> #include <linux/etherdevice.h>
>> #include <linux/vringh.h>
>> #include <linux/vdpa.h>
>> +#include <net/netlink.h>
>> #include <uapi/linux/virtio_net.h>
>> #include <uapi/linux/vdpa.h>
>>
>> @@ -36,6 +37,34 @@
>> #define VDPASIM_NET_AS_NUM    2
>> #define VDPASIM_NET_GROUP_NUM    2
>>
>> +struct vdpasim_dataq_stats {
>> +    struct u64_stats_sync syncp;
>> +    u64 pkts;
>> +    u64 bytes;
>> +    u64 drops;
>> +    u64 errors;
>> +    u64 overruns;
>> +};
>> +
>> +struct vdpasim_cq_stats {
>> +    struct u64_stats_sync syncp;
>> +    u64 requests;
>> +    u64 successes;
>> +    u64 errors;
>> +};
>> +
>> +struct vdpasim_net{
>> +    struct vdpasim vdpasim;
>> +    struct vdpasim_dataq_stats tx_stats;
>> +    struct vdpasim_dataq_stats rx_stats;
>> +    struct vdpasim_cq_stats cq_stats;
>> +};
>> +
>> +static struct vdpasim_net *sim_to_net(struct vdpasim *vdpasim)
>> +{
>> +    return container_of(vdpasim, struct vdpasim_net, vdpasim);
>> +}
>> +
>> static void vdpasim_net_complete(struct vdpasim_virtqueue *vq, size_t 
>> len)
>> {
>>     /* Make sure data is wrote before advancing index */
>> @@ -93,9 +122,11 @@ static virtio_net_ctrl_ack 
>> vdpasim_handle_ctrl_mac(struct vdpasim *vdpasim,
>> static void vdpasim_handle_cvq(struct vdpasim *vdpasim)
>> {
>>     struct vdpasim_virtqueue *cvq = &vdpasim->vqs[2];
>> +    struct vdpasim_net *net = sim_to_net(vdpasim);
>>     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
>>     struct virtio_net_ctrl_hdr ctrl;
>>     size_t read, write;
>> +    u64 requests = 0, errors = 0;
>>     int err;
>>
>>     if (!(vdpasim->features & (1ULL << VIRTIO_NET_F_CTRL_VQ)))
>> @@ -113,8 +144,12 @@ static void vdpasim_handle_cvq(struct vdpasim 
>> *vdpasim)
>>
>>         read = vringh_iov_pull_iotlb(&cvq->vring, &cvq->in_iov, &ctrl,
>>                          sizeof(ctrl));
>> -        if (read != sizeof(ctrl))
>> +        if (read != sizeof(ctrl)) {
>> +            ++errors;
>>             break;
>> +        }
>> +
>> +        ++requests;
>>
>>         switch (ctrl.class) {
>>         case VIRTIO_NET_CTRL_MAC:
>> @@ -141,6 +176,12 @@ static void vdpasim_handle_cvq(struct vdpasim 
>> *vdpasim)
>>             cvq->cb(cvq->private);
>>         local_bh_enable();
>>     }
>> +
>> +    u64_stats_update_begin(&net->cq_stats.syncp);
>> +    net->cq_stats.requests += requests;
>
> I don't know if I understand the meaning of the fields correctly, but 
> should cq_stats.requests count both requests completed successfully 
> and with error?
>
> I mean:
>         net->cq_stats.requests += requests + errors;
>
> If it is the case I would rename the local "requests" variable in 
> "successes".


Yes.


>
>> +    net->cq_stats.errors += errors;
>> +    net->cq_stats.successes += requests;
>> +    u64_stats_update_end(&net->cq_stats.syncp);
>> }
>>
>> static void vdpasim_net_work(struct work_struct *work)
>> @@ -148,8 +189,10 @@ static void vdpasim_net_work(struct work_struct 
>> *work)
>>     struct vdpasim *vdpasim = container_of(work, struct vdpasim, work);
>>     struct vdpasim_virtqueue *txq = &vdpasim->vqs[1];
>>     struct vdpasim_virtqueue *rxq = &vdpasim->vqs[0];
>> +    struct vdpasim_net *net = sim_to_net(vdpasim);
>>     ssize_t read, write;
>> -    int pkts = 0;
>> +    u64 tx_pkts = 0, rx_pkts = 0, tx_bytes = 0, rx_bytes = 0;
>> +    u64 rx_drops = 0, rx_overruns = 0, rx_errors = 0, tx_errors = 0;
>>     int err;
>>
>>     spin_lock(&vdpasim->lock);
>> @@ -168,14 +211,21 @@ static void vdpasim_net_work(struct work_struct 
>> *work)
>>     while (true) {
>>         err = vringh_getdesc_iotlb(&txq->vring, &txq->out_iov, NULL,
>>                        &txq->head, GFP_ATOMIC);
>> -        if (err <= 0)
>> +        if (err <= 0) {
>> +            if (err)
>> +                ++tx_errors;
>>             break;
>> +        }
>>
>> +        ++tx_pkts;
>>         read = vringh_iov_pull_iotlb(&txq->vring, &txq->out_iov,
>>                          vdpasim->buffer,
>>                          PAGE_SIZE);
>>
>> +        tx_bytes += read;
>> +
>>         if (!receive_filter(vdpasim, read)) {
>> +            ++rx_drops;
>>             vdpasim_net_complete(txq, 0);
>>             continue;
>>         }
>> @@ -183,19 +233,25 @@ static void vdpasim_net_work(struct work_struct 
>> *work)
>>         err = vringh_getdesc_iotlb(&rxq->vring, NULL, &rxq->in_iov,
>>                        &rxq->head, GFP_ATOMIC);
>>         if (err <= 0) {
>> +            ++rx_overruns;
>>             vdpasim_net_complete(txq, 0);
>>             break;
>>         }
>>
>>         write = vringh_iov_push_iotlb(&rxq->vring, &rxq->in_iov,
>>                           vdpasim->buffer, read);
>> -        if (write <= 0)
>> +        if (write <= 0) {
>> +            ++rx_errors;
>>             break;
>> +        }
>> +
>> +        ++rx_pkts;
>> +        rx_bytes += write;
>>
>>         vdpasim_net_complete(txq, 0);
>>         vdpasim_net_complete(rxq, write);
>>
>> -        if (++pkts > 4) {
>> +        if (tx_pkts > 4) {
>>             schedule_work(&vdpasim->work);
>>             goto out;
>>         }
>> @@ -203,6 +259,145 @@ static void vdpasim_net_work(struct work_struct 
>> *work)
>>
>> out:
>>     spin_unlock(&vdpasim->lock);
>> +
>> +    u64_stats_update_begin(&net->tx_stats.syncp);
>> +    net->tx_stats.pkts += tx_pkts;
>> +    net->tx_stats.bytes += tx_bytes;
>> +    net->tx_stats.errors += tx_errors;
>> +    u64_stats_update_end(&net->tx_stats.syncp);
>> +
>> +    u64_stats_update_begin(&net->rx_stats.syncp);
>> +    net->rx_stats.pkts += rx_pkts;
>> +    net->rx_stats.bytes += rx_bytes;
>> +    net->rx_stats.drops += rx_drops;
>> +    net->rx_stats.errors += rx_errors;
>> +    net->rx_stats.overruns += rx_overruns;
>> +    u64_stats_update_end(&net->rx_stats.syncp);
>> +}
>> +
>> +static int vdpasim_net_get_stats(struct vdpasim *vdpasim, u16 idx,
>> +                 struct sk_buff *msg,
>> +                 struct netlink_ext_ack *extack)
>> +{
>> +    struct vdpasim_net *net = sim_to_net(vdpasim);
>> +    u64 rx_pkts, rx_bytes, rx_errors, rx_overruns, rx_drops;
>> +    u64 tx_pkts, tx_bytes, tx_errors, tx_drops;
>> +    u64 cq_requests, cq_successes, cq_errors;
>> +    unsigned int start;
>> +    int err = -EMSGSIZE;
>> +
>> +    switch(idx) {
>> +    case 0:
>> +        do {
>> +            start = u64_stats_fetch_begin(&net->rx_stats.syncp);
>> +            rx_pkts = net->rx_stats.pkts;
>> +            rx_bytes = net->rx_stats.bytes;
>> +            rx_errors = net->rx_stats.errors;
>> +            rx_overruns = net->rx_stats.overruns;
>> +            rx_drops = net->rx_stats.drops;
>> +        } while (u64_stats_fetch_retry(&net->rx_stats.syncp, start));
>> +
>> +        if (nla_put_string(msg, VDPA_ATTR_DEV_VENDOR_ATTR_NAME,
>> +                    "rx packets"))
>> +            break;
>> +        if (nla_put_u64_64bit(msg, VDPA_ATTR_DEV_VENDOR_ATTR_VALUE,
>> +                      rx_pkts, VDPA_ATTR_PAD))
>> +            break;
>> +        if (nla_put_string(msg, VDPA_ATTR_DEV_VENDOR_ATTR_NAME,
>> +                  "rx bytes"))
>> +            break;
>> +        if (nla_put_u64_64bit(msg, VDPA_ATTR_DEV_VENDOR_ATTR_VALUE,
>> +                      rx_bytes, VDPA_ATTR_PAD))
>> +            break;
>> +        if (nla_put_string(msg, VDPA_ATTR_DEV_VENDOR_ATTR_NAME,
>> +                  "rx errors"))
>> +            break;
>> +        if (nla_put_u64_64bit(msg, VDPA_ATTR_DEV_VENDOR_ATTR_VALUE,
>> +                      rx_errors, VDPA_ATTR_PAD))
>> +            break;
>> +        if (nla_put_string(msg, VDPA_ATTR_DEV_VENDOR_ATTR_NAME,
>> +                  "rx overrunss"))
>> +            break;
>> +        if (nla_put_u64_64bit(msg, VDPA_ATTR_DEV_VENDOR_ATTR_VALUE,
>> +                      rx_overruns, VDPA_ATTR_PAD))
>> +            break;
>> +        if (nla_put_string(msg, VDPA_ATTR_DEV_VENDOR_ATTR_NAME,
>> +                  "rx drops"))
>> +            break;
>> +        if (nla_put_u64_64bit(msg, VDPA_ATTR_DEV_VENDOR_ATTR_VALUE,
>> +                      rx_drops, VDPA_ATTR_PAD))
>> +            break;
>> +        err = 0;
>> +        break;
>> +    case 1:
>> +        do {
>> +            start = u64_stats_fetch_begin(&net->tx_stats.syncp);
>> +            tx_pkts = net->tx_stats.pkts;
>> +            tx_bytes = net->tx_stats.bytes;
>> +            tx_errors = net->tx_stats.errors;
>> +            tx_drops = net->tx_stats.drops;
>> +        } while (u64_stats_fetch_retry(&net->tx_stats.syncp, start));
>> +
>> +        if (nla_put_string(msg, VDPA_ATTR_DEV_VENDOR_ATTR_NAME,
>> +                  "tx packets"))
>> +            break;
>> +        if (nla_put_u64_64bit(msg, VDPA_ATTR_DEV_VENDOR_ATTR_VALUE,
>> +                      tx_pkts, VDPA_ATTR_PAD))
>> +            break;
>> +        if (nla_put_string(msg, VDPA_ATTR_DEV_VENDOR_ATTR_NAME,
>> +                  "tx bytes"))
>> +            break;
>> +        if (nla_put_u64_64bit(msg, VDPA_ATTR_DEV_VENDOR_ATTR_VALUE,
>> +                      tx_bytes, VDPA_ATTR_PAD))
>> +            break;
>> +        if (nla_put_string(msg, VDPA_ATTR_DEV_VENDOR_ATTR_NAME,
>> +                  "tx errors"))
>> +            break;
>> +        if (nla_put_u64_64bit(msg, VDPA_ATTR_DEV_VENDOR_ATTR_VALUE,
>> +                      tx_errors, VDPA_ATTR_PAD))
>> +            break;
>> +        if (nla_put_string(msg, VDPA_ATTR_DEV_VENDOR_ATTR_NAME,
>> +                  "tx drops"))
>> +            break;
>> +        if (nla_put_u64_64bit(msg, VDPA_ATTR_DEV_VENDOR_ATTR_VALUE,
>> +                      tx_drops, VDPA_ATTR_PAD))
>> +            break;
>> +        err = 0;
>> +        break;
>> +    case 2:
>> +        do {
>> +            start = u64_stats_fetch_begin(&net->cq_stats.syncp);
>> +            cq_requests = net->cq_stats.requests;
>> +            cq_successes = net->cq_stats.successes;
>> +            cq_errors = net->cq_stats.errors;
>> +        } while (u64_stats_fetch_retry(&net->cq_stats.syncp, start));
>> +
>> +        if (nla_put_string(msg, VDPA_ATTR_DEV_VENDOR_ATTR_NAME,
>> +                  "cvq requests"))
>> +            break;
>> +        if (nla_put_u64_64bit(msg, VDPA_ATTR_DEV_VENDOR_ATTR_VALUE,
>> +                      cq_requests, VDPA_ATTR_PAD))
>> +            break;
>> +        if (nla_put_string(msg, VDPA_ATTR_DEV_VENDOR_ATTR_NAME,
>> +                  "cvq successes"))
>> +            break;
>> +        if (nla_put_u64_64bit(msg, VDPA_ATTR_DEV_VENDOR_ATTR_VALUE,
>> +                      cq_successes, VDPA_ATTR_PAD))
>> +            break;
>> +        if (nla_put_string(msg, VDPA_ATTR_DEV_VENDOR_ATTR_NAME,
>> +                  "cvq errors"))
>
> Would it be better to define macros for all these fields (e.g. "tx 
> packets", etc.) so we can avoid typos for the various drivers that 
> will support them?


Not sure, but I'd leave the code as-is since it is specific to a device 
type and we haven't had any type specific before.

Maybe we can do that after we have other type specific headers.

Thanks


>
> Thanks,
> Stefano
>
>> +            break;
>> +        if (nla_put_u64_64bit(msg, VDPA_ATTR_DEV_VENDOR_ATTR_VALUE,
>> +                      cq_errors, VDPA_ATTR_PAD))
>> +            break;
>> +        err = 0;
>> +        break;
>> +    default:
>> +        err = -EINVAL;
>> +        break;
>> +    }
>> +
>> +    return err;
>> }
>>
>> static void vdpasim_net_get_config(struct vdpasim *vdpasim, void 
>> *config)
>> @@ -239,6 +434,7 @@ static int vdpasim_net_dev_add(struct 
>> vdpa_mgmt_dev *mdev, const char *name,
>>                    const struct vdpa_dev_set_config *config)
>> {
>>     struct vdpasim_dev_attr dev_attr = {};
>> +    struct vdpasim_net *net;
>>     struct vdpasim *simdev;
>>     int ret;
>>
>> @@ -249,10 +445,11 @@ static int vdpasim_net_dev_add(struct 
>> vdpa_mgmt_dev *mdev, const char *name,
>>     dev_attr.nvqs = VDPASIM_NET_VQ_NUM;
>>     dev_attr.ngroups = VDPASIM_NET_GROUP_NUM;
>>     dev_attr.nas = VDPASIM_NET_AS_NUM;
>> -    dev_attr.alloc_size = sizeof(struct vdpasim);
>> +    dev_attr.alloc_size = sizeof(struct vdpasim_net);
>>     dev_attr.config_size = sizeof(struct virtio_net_config);
>>     dev_attr.get_config = vdpasim_net_get_config;
>>     dev_attr.work_fn = vdpasim_net_work;
>> +    dev_attr.get_stats = vdpasim_net_get_stats;
>>     dev_attr.buffer_size = PAGE_SIZE;
>>
>>     simdev = vdpasim_create(&dev_attr, config);
>> @@ -265,6 +462,12 @@ static int vdpasim_net_dev_add(struct 
>> vdpa_mgmt_dev *mdev, const char *name,
>>     if (ret)
>>         goto reg_err;
>>
>> +    net = sim_to_net(simdev);
>> +
>> +    u64_stats_init(&net->tx_stats.syncp);
>> +    u64_stats_init(&net->rx_stats.syncp);
>> +    u64_stats_init(&net->cq_stats.syncp);
>> +
>>     return 0;
>>
>> reg_err:
>> -- 
>> 2.25.1
>>
>