Re: [PATCH v2] net/tap: fix to populate fds in secondary process

[Ferruh Yigit <ferruh.yigit@intel.com>]

On 1/24/2022 12:54 PM, kumaraparameshwaran rathinavel wrote:
> 
> 
> On Mon, Jan 24, 2022 at 6:05 PM Ferruh Yigit <ferruh.yigit@intel.com <mailto:ferruh.yigit@intel.com>> wrote:
> 
>     On 1/24/2022 12:12 PM, Kumara Parameshwaran wrote:
>      > From: Kumara Parameshwaran <kparameshwar@vmware.com <mailto:kparameshwar@vmware.com>>
>      >
>      > When a tap device is hotplugged to primary process which in turn
>      > adds the device to all secondary process, the secondary process
>      > does a tap_mp_attach_queues, but the fds are not populated in
>      > the primary during the probe they are populated during the queue_setup,
>      > added a fix to sync the queues during rte_eth_dev_start
>      >
>      > Fixes: 4852aa8f6e21 ("drivers/net: enable hotplug on secondary process")
>      > Cc: stable@dpdk.org <mailto:stable@dpdk.org>
>      >
>      > Signed-off-by: Kumara Parameshwaran <kparameshwar@vmware.com <mailto:kparameshwar@vmware.com>>
>      > ---
>      >
>      > v2:
>      > * Addressed review comments to move the function declaration and version
>      >    map
>      >
> 
>     Thanks for adding patch version.
> 
>      >   drivers/net/tap/rte_eth_tap.c | 196 +++++++++++++---------------------
>      >   lib/ethdev/ethdev_driver.h    |  17 +++
>      >   lib/ethdev/rte_ethdev.c       |  11 ++
>      >   lib/ethdev/version.map        |   2 +
>      >   4 files changed, 102 insertions(+), 124 deletions(-)
>      >
> 
>     Can you please separate etdev (API) changes to another patch, so this will be a patchset
>     with two patches,
>     first patch adds ethdev API
>     second patch is tap patch, that uses the API in the first patch 
> 
>      Sure, I will do it.
> 
> 
>      > diff --git a/drivers/net/tap/rte_eth_tap.c b/drivers/net/tap/rte_eth_tap.c
>      > index f1b48cae82..f6c25d7e21 100644
>      > --- a/drivers/net/tap/rte_eth_tap.c
>      > +++ b/drivers/net/tap/rte_eth_tap.c
>      > @@ -66,7 +66,7 @@
>      >       (TAP_GSO_MBUFS_PER_CORE * TAP_GSO_MBUF_CACHE_SIZE)
>      >
>      >   /* IPC key for queue fds sync */
>      > -#define TAP_MP_KEY "tap_mp_sync_queues"
>      > +#define TAP_MP_REQ_START_RXTX "tap_mp_req_start_rxtx"
>      >
> 
>     We said we can drop "tap_mp_sync_queues", but thinking twice,
>     will current implementation cover following usecase:
> 
>     - Primary applicaiton started with tap interface, all config, setup,
>         start done
>     - Secondary app started without any parameter
> 
>     Since primary already started, I think secondary fds will be wrong,
>     what do you think?
>     That is true, the fds would be unitialised, we should retain for the secondary attach case.
>      >   #define TAP_IOV_DEFAULT_MAX 1024
>      >
>      > @@ -880,11 +880,48 @@ tap_link_set_up(struct rte_eth_dev *dev)
>      >       return tap_ioctl(pmd, SIOCSIFFLAGS, &ifr, 1, LOCAL_AND_REMOTE);
>      >   }
>      >
>      > +static int tap_mp_req_on_rxtx(struct rte_eth_dev *dev)
> 
>     Can you please follow the coding convention:
>     sure, will do it.

There are more comments/questions below, ca you please check them too?

>     static int
>     tap_mp_req_on_rxtx(struct rte_eth_dev *dev)
>     {
> 
>      > +{
>      > +     struct rte_mp_msg msg;
>      > +     struct ipc_queues *request_param = (struct ipc_queues *)msg.param;
>      > +     int err;
>      > +     int fd_iterator = 0;
>      > +     struct pmd_process_private *process_private = dev->process_private;
>      > +     int i;
>      > +
>      > +     memset(&msg, 0, sizeof(msg));
>      > +     strlcpy(msg.name <http://msg.name>, TAP_MP_REQ_START_RXTX, sizeof(msg.name <http://msg.name>));
>      > +     strlcpy(request_param->port_name, dev->data->name, sizeof(request_param->port_name));
>      > +     msg.len_param = sizeof(*request_param);
>      > +     for (i = 0; i < dev->data->nb_tx_queues; i++) {
>      > +             msg.fds[fd_iterator++] = process_private->txq_fds[i];
>      > +             msg.num_fds++;
>      > +             request_param->txq_count++;
>      > +     }
>      > +     for (i = 0; i < dev->data->nb_rx_queues; i++) {
>      > +             msg.fds[fd_iterator++] = process_private->rxq_fds[i];
>      > +             msg.num_fds++;
>      > +             request_param->rxq_count++;
>      > +     }
>      > +
>      > +     err = rte_mp_sendmsg(&msg);
>      > +     if (err < 0) {
>      > +             TAP_LOG(ERR, "Failed to send start req to secondary %d",
>      > +                     rte_errno);
>      > +             return -1;
>      > +     }
>      > +
>      > +     return 0;
>      > +}
>      > +
>      >   static int
>      >   tap_dev_start(struct rte_eth_dev *dev)
>      >   {
>      >       int err, i;
>      >
>      > +     if (rte_eal_process_type() == RTE_PROC_PRIMARY)
>      > +             tap_mp_req_on_rxtx(dev);
>      > +
>      >       err = tap_intr_handle_set(dev, 1);
>      >       if (err)
>      >               return err;
>      > @@ -901,6 +938,34 @@ tap_dev_start(struct rte_eth_dev *dev)
>      >       return err;
>      >   }
>      >
>      > +static int
>      > +tap_mp_req_start_rxtx(const struct rte_mp_msg *request, __rte_unused const void *peer)
>      > +{
> 
>     I asked last time but I don't remember the response,
>     what should be in the 'peer' variable?
> 
>      > +     struct rte_eth_dev *dev;
>      > +     const struct ipc_queues *request_param =
>      > +             (const struct ipc_queues *)request->param;
>      > +     int fd_iterator;
>      > +     int queue;
>      > +     struct pmd_process_private *process_private;
>      > +
>      > +     dev = rte_get_eth_dev_by_name(request_param->port_name);
>      > +     if (!dev) {
>      > +             TAP_LOG(ERR, "Failed to get dev for %s",
>      > +                     request_param->port_name);
>      > +             return -1;
>      > +     }
>      > +     process_private = dev->process_private;
>      > +     fd_iterator = 0;
>      > +     TAP_LOG(DEBUG, "tap_attach rx_q:%d tx_q:%d\n", request_param->rxq_count,
>      > +             request_param->txq_count);
>      > +     for (queue = 0; queue < request_param->txq_count; queue++)
>      > +             process_private->txq_fds[queue] = request->fds[fd_iterator++];
>      > +     for (queue = 0; queue < request_param->rxq_count; queue++)
>      > +             process_private->rxq_fds[queue] = request->fds[fd_iterator++];
>      > +
>      > +     return 0;
>      > +}
>      > +
>      >   /* This function gets called when the current port gets stopped.
>      >    */
>      >   static int
>      > @@ -1084,6 +1149,7 @@ tap_dev_close(struct rte_eth_dev *dev)
>      >
>      >       if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
>      >               rte_free(dev->process_private);
>      > +             rte_mp_action_unregister(TAP_MP_REQ_START_RXTX);
>      >               return 0;
>      >       }
>      >
>      > @@ -1140,8 +1206,6 @@ tap_dev_close(struct rte_eth_dev *dev)
>      >               internals->ioctl_sock = -1;
>      >       }
>      >       rte_free(dev->process_private);
>      > -     if (tap_devices_count == 1)
>      > -             rte_mp_action_unregister(TAP_MP_KEY);
>      >       tap_devices_count--;
>      >       /*
>      >        * Since TUN device has no more opened file descriptors
>      > @@ -2292,113 +2356,6 @@ rte_pmd_tun_probe(struct rte_vdev_device *dev)
>      >       return ret;
>      >   }
>      >
>      > -/* Request queue file descriptors from secondary to primary. */
>      > -static int
>      > -tap_mp_attach_queues(const char *port_name, struct rte_eth_dev *dev)
>      > -{
>      > -     int ret;
>      > -     struct timespec timeout = {.tv_sec = 1, .tv_nsec = 0};
>      > -     struct rte_mp_msg request, *reply;
>      > -     struct rte_mp_reply replies;
>      > -     struct ipc_queues *request_param = (struct ipc_queues *)request.param;
>      > -     struct ipc_queues *reply_param;
>      > -     struct pmd_process_private *process_private = dev->process_private;
>      > -     int queue, fd_iterator;
>      > -
>      > -     /* Prepare the request */
>      > -     memset(&request, 0, sizeof(request));
>      > -     strlcpy(request.name <http://request.name>, TAP_MP_KEY, sizeof(request.name <http://request.name>));
>      > -     strlcpy(request_param->port_name, port_name,
>      > -             sizeof(request_param->port_name));
>      > -     request.len_param = sizeof(*request_param);
>      > -     /* Send request and receive reply */
>      > -     ret = rte_mp_request_sync(&request, &replies, &timeout);
>      > -     if (ret < 0 || replies.nb_received != 1) {
>      > -             TAP_LOG(ERR, "Failed to request queues from primary: %d",
>      > -                     rte_errno);
>      > -             return -1;
>      > -     }
>      > -     reply = &replies.msgs[0];
>      > -     reply_param = (struct ipc_queues *)reply->param;
>      > -     TAP_LOG(DEBUG, "Received IPC reply for %s", reply_param->port_name);
>      > -
>      > -     /* Attach the queues from received file descriptors */
>      > -     if (reply_param->rxq_count + reply_param->txq_count != reply->num_fds) {
>      > -             TAP_LOG(ERR, "Unexpected number of fds received");
>      > -             return -1;
>      > -     }
>      > -
>      > -     dev->data->nb_rx_queues = reply_param->rxq_count;
>      > -     dev->data->nb_tx_queues = reply_param->txq_count;
>      > -     fd_iterator = 0;
>      > -     for (queue = 0; queue < reply_param->rxq_count; queue++)
>      > -             process_private->rxq_fds[queue] = reply->fds[fd_iterator++];
>      > -     for (queue = 0; queue < reply_param->txq_count; queue++)
>      > -             process_private->txq_fds[queue] = reply->fds[fd_iterator++];
>      > -     free(reply);
>      > -     return 0;
>      > -}
>      > -
>      > -/* Send the queue file descriptors from the primary process to secondary. */
>      > -static int
>      > -tap_mp_sync_queues(const struct rte_mp_msg *request, const void *peer)
>      > -{
>      > -     struct rte_eth_dev *dev;
>      > -     struct pmd_process_private *process_private;
>      > -     struct rte_mp_msg reply;
>      > -     const struct ipc_queues *request_param =
>      > -             (const struct ipc_queues *)request->param;
>      > -     struct ipc_queues *reply_param =
>      > -             (struct ipc_queues *)reply.param;
>      > -     uint16_t port_id;
>      > -     int queue;
>      > -     int ret;
>      > -
>      > -     /* Get requested port */
>      > -     TAP_LOG(DEBUG, "Received IPC request for %s", request_param->port_name);
>      > -     ret = rte_eth_dev_get_port_by_name(request_param->port_name, &port_id);
>      > -     if (ret) {
>      > -             TAP_LOG(ERR, "Failed to get port id for %s",
>      > -                     request_param->port_name);
>      > -             return -1;
>      > -     }
>      > -     dev = &rte_eth_devices[port_id];
>      > -     process_private = dev->process_private;
>      > -
>      > -     /* Fill file descriptors for all queues */
>      > -     reply.num_fds = 0;
>      > -     reply_param->rxq_count = 0;
>      > -     if (dev->data->nb_rx_queues + dev->data->nb_tx_queues >
>      > -                     RTE_MP_MAX_FD_NUM){
>      > -             TAP_LOG(ERR, "Number of rx/tx queues exceeds max number of fds");
>      > -             return -1;
>      > -     }
>      > -
>      > -     for (queue = 0; queue < dev->data->nb_rx_queues; queue++) {
>      > -             reply.fds[reply.num_fds++] = process_private->rxq_fds[queue];
>      > -             reply_param->rxq_count++;
>      > -     }
>      > -     RTE_ASSERT(reply_param->rxq_count == dev->data->nb_rx_queues);
>      > -
>      > -     reply_param->txq_count = 0;
>      > -     for (queue = 0; queue < dev->data->nb_tx_queues; queue++) {
>      > -             reply.fds[reply.num_fds++] = process_private->txq_fds[queue];
>      > -             reply_param->txq_count++;
>      > -     }
>      > -     RTE_ASSERT(reply_param->txq_count == dev->data->nb_tx_queues);
>      > -
>      > -     /* Send reply */
>      > -     strlcpy(reply.name <http://reply.name>, request->name, sizeof(reply.name <http://reply.name>));
>      > -     strlcpy(reply_param->port_name, request_param->port_name,
>      > -             sizeof(reply_param->port_name));
>      > -     reply.len_param = sizeof(*reply_param);
>      > -     if (rte_mp_reply(&reply, peer) < 0) {
>      > -             TAP_LOG(ERR, "Failed to reply an IPC request to sync queues");
>      > -             return -1;
>      > -     }
>      > -     return 0;
>      > -}
>      > -
>      >   /* Open a TAP interface device.
>      >    */
>      >   static int
>      > @@ -2442,9 +2399,11 @@ rte_pmd_tap_probe(struct rte_vdev_device *dev)
>      >                       return -1;
>      >               }
>      >
>      > -             ret = tap_mp_attach_queues(name, eth_dev);
>      > -             if (ret != 0)
>      > -                     return -1;
>      > +             ret = rte_mp_action_register(TAP_MP_REQ_START_RXTX, tap_mp_req_start_rxtx);
>      > +             if (ret < 0 && rte_errno != ENOTSUP)
>      > +                     TAP_LOG(ERR, "tap: Failed to register IPC callback: %s",
>      > +                             strerror(rte_errno));
>      > +
>      >               rte_eth_dev_probing_finish(eth_dev);
>      >               return 0;
>      >       }
>      > @@ -2492,15 +2451,6 @@ rte_pmd_tap_probe(struct rte_vdev_device *dev)
>      >
>      >       TAP_LOG(DEBUG, "Initializing pmd_tap for %s", name);
>      >
>      > -     /* Register IPC feed callback */
>      > -     if (!tap_devices_count) {
>      > -             ret = rte_mp_action_register(TAP_MP_KEY, tap_mp_sync_queues);
>      > -             if (ret < 0 && rte_errno != ENOTSUP) {
>      > -                     TAP_LOG(ERR, "tap: Failed to register IPC callback: %s",
>      > -                             strerror(rte_errno));
>      > -                     goto leave;
>      > -             }
>      > -     }
>      >       tap_devices_count++;
>      >       tap_devices_count_increased = 1;
>      >       ret = eth_dev_tap_create(dev, tap_name, remote_iface, &user_mac,
>      > @@ -2511,8 +2461,6 @@ rte_pmd_tap_probe(struct rte_vdev_device *dev)
>      >               TAP_LOG(ERR, "Failed to create pmd for %s as %s",
>      >                       name, tap_name);
>      >               if (tap_devices_count_increased == 1) {
>      > -                     if (tap_devices_count == 1)
>      > -                             rte_mp_action_unregister(TAP_MP_KEY);
>      >                       tap_devices_count--;
>      >               }
>      >       }
>      > diff --git a/lib/ethdev/ethdev_driver.h b/lib/ethdev/ethdev_driver.h
>      > index d95605a355..a08991bcdf 100644
>      > --- a/lib/ethdev/ethdev_driver.h
>      > +++ b/lib/ethdev/ethdev_driver.h
>      > @@ -1629,6 +1629,23 @@ rte_eth_hairpin_queue_peer_bind(uint16_t cur_port, uint16_t cur_queue,
>      >                               struct rte_hairpin_peer_info *peer_info,
>      >                               uint32_t direction);
>      >
>      > +/**
> 
>     Please add '@internal' tag into doxygen comment.
> 
>      > +* Get rte_eth_dev from device name. The device name should be specified
>      > +* as below:
>      > +* - PCIe address (Domain:Bus:Device.Function), for example- 0000:2:00.0
>      > +* - SoC device name, for example- fsl-gmac0
>      > +* - vdev dpdk name, for example- net_[pcap0|null0|tap0]
>      > +*
>      > +* @param name
>      > +*  pci address or name of the device
>      > +* @return
>      > +*   - rte_eth_dev if successful
>      > +*   - NULL on failure
>      > +*/
>      > +__rte_internal
>      > +struct rte_eth_dev*
>      > +rte_get_eth_dev_by_name(const char *name);
> 
>     As the API name, better to start with 'rte_eth_' prefix to be consistent with
>     rest of the APIs.
>     I suggest 'rte_eth_dev_get_by_name' but feel free to chose better one.
> 
>      > +
>      >   /**
>      >    * @internal
>      >    * Reset the current queue state and configuration to disconnect (unbind) it
>      > diff --git a/lib/ethdev/rte_ethdev.c b/lib/ethdev/rte_ethdev.c
>      > index a1d475a292..9192b0d664 100644
>      > --- a/lib/ethdev/rte_ethdev.c
>      > +++ b/lib/ethdev/rte_ethdev.c
>      > @@ -894,6 +894,17 @@ rte_eth_dev_get_port_by_name(const char *name, uint16_t *port_id)
>      >       return -ENODEV;
>      >   }
>      >
>      > +struct rte_eth_dev *
>      > +rte_get_eth_dev_by_name(const char *name)
>      > +{
>      > +     uint16_t pid;
>      > +
>      > +     if (rte_eth_dev_get_port_by_name(name, &pid))
>      > +             return NULL;
>      > +
>      > +     return &rte_eth_devices[pid];
>      > +}
>      > +
>      >   static int
>      >   eth_err(uint16_t port_id, int ret)
>      >   {
>      > diff --git a/lib/ethdev/version.map b/lib/ethdev/version.map
>      > index c2fb0669a4..7e3797189b 100644
>      > --- a/lib/ethdev/version.map
>      > +++ b/lib/ethdev/version.map
>      > @@ -256,6 +256,7 @@ EXPERIMENTAL {
>      >       rte_flow_flex_item_create;
>      >       rte_flow_flex_item_release;
>      >       rte_flow_pick_transfer_proxy;
>      > +
> 
>     This is unintendent change.
> 
>      >   };
>      >
>      >   INTERNAL {
>      > @@ -282,4 +283,5 @@ INTERNAL {
>      >       rte_eth_representor_id_get;
>      >       rte_eth_switch_domain_alloc;
>      >       rte_eth_switch_domain_free;
>      > +     rte_get_eth_dev_by_name;
> 
>     Please add in a sorted way.
> 
>      >   };
>