On Sun, Jul 22, 2018 at 11:18:34AM +0000, Shahaf Shuler wrote:
> Hi Nelio,
>
> Friday, July 13, 2018 4:54 PM, Nelio Laranjeiro:
> > Subject: [PATCH] net/mlx5: fix representors detection
> >
> > On systems where the required Netlink commands are not supported but
> > Mellanox OFED is installed, representors information must be retrieved
> > through sysfs.
> >
> > Fixes: 4c10141488fe ("net/mlx5: add port representor awareness")
> >
> > Signed-off-by: Nelio Laranjeiro <nelio.laranje...@6wind.com>
> > Acked-by: Adrien Mazarguil <adrien.mazarg...@6wind.com>
> > ---
> > drivers/net/mlx5/mlx5.c | 7 +++--
> > drivers/net/mlx5/mlx5.h | 2 ++
> > drivers/net/mlx5/mlx5_ethdev.c | 53
> > ++++++++++++++++++++++++++++++++++
> > 3 files changed, 60 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/net/mlx5/mlx5.c b/drivers/net/mlx5/mlx5.c index
> > df7f39844..8174947f3 100644
> > --- a/drivers/net/mlx5/mlx5.c
> > +++ b/drivers/net/mlx5/mlx5.c
> > @@ -1306,7 +1306,8 @@ mlx5_pci_probe(struct rte_pci_driver *pci_drv
> > __rte_unused,
> > * Netlink calls assuming kernel drivers are recent enough to
> > * support them.
> > *
> > - * In the event of identification failure through Netlink, either:
> > + * In the event of identification failure through Netlink, try again
> > + * through sysfs, then either:
> > *
> > * 1. No device matches (n == 0), complain and bail out.
> > * 2. A single IB device matches (n == 1) and is not a representor, @@
> > -1325,7 +1326,9 @@ mlx5_pci_probe(struct rte_pci_driver *pci_drv
> > __rte_unused,
> > if (nl_route < 0 ||
> > !list[i].ifindex ||
> > mlx5_nl_switch_info(nl_route, list[i].ifindex,
> > - &list[i].info)) {
> > + &list[i].info) ||
> > + ((!list[i].info.representor && !list[i].info.master) &&
> > + mlx5_sysfs_switch_info(list[i].ifindex, &list[i].info))) {
>
> I find this logic a bit complex to be under a single if statement. Can
> you simplify it w/ local variables?
It won't be so easy has the second step depends on the result of the
mlx5_nl_switch_info() call. To split this if statement into two, it
will be a full copy of the first block by replacing only the call to
mlx5_nl_switch_info().
> > list[i].ifindex = 0;
> > memset(&list[i].info, 0, sizeof(list[i].info));
> > continue;
> > diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h index
> > cc01310e0..fff88b1d6 100644
> > --- a/drivers/net/mlx5/mlx5.h
> > +++ b/drivers/net/mlx5/mlx5.h
> > @@ -254,6 +254,8 @@ eth_rx_burst_t mlx5_select_rx_function(struct
> > rte_eth_dev *dev); unsigned int mlx5_dev_to_port_id(const struct
> > rte_device *dev,
> > uint16_t *port_list,
> > unsigned int port_list_n);
> > +int mlx5_sysfs_switch_info(unsigned int ifindex,
> > + struct mlx5_switch_info *info);
> >
> > /* mlx5_mac.c */
> >
> > diff --git a/drivers/net/mlx5/mlx5_ethdev.c
> > b/drivers/net/mlx5/mlx5_ethdev.c index 05f66f7b6..101be4791 100644
> > --- a/drivers/net/mlx5/mlx5_ethdev.c
> > +++ b/drivers/net/mlx5/mlx5_ethdev.c
> > @@ -1296,3 +1296,56 @@ mlx5_dev_to_port_id(const struct rte_device
> > *dev, uint16_t *port_list,
> > }
> > return n;
> > }
> > +
> > +/**
> > + * Get switch information associated with network interface.
> > + *
> > + * @param ifindex
> > + * Network interface index.
> > + * @param[out] info
> > + * Switch information object, populated in case of success.
> > + *
> > + * @return
> > + * 0 on success, a negative errno value otherwise and rte_errno is set.
> > + */
> > +int
> > +mlx5_sysfs_switch_info(unsigned int ifindex, struct mlx5_switch_info
> > +*info) {
> > + char ifname[IF_NAMESIZE];
> > + FILE *file;
> > + struct mlx5_switch_info data = { .master = 0, };
> > + bool port_name_set = false;
> > + bool port_switch_id_set = false;
> > + char c;
> > +
> > + if (!if_indextoname(ifindex, ifname)) {
> > + rte_errno = errno;
> > + return -rte_errno;
> > + }
> > +
> > + MKSTR(phys_port_name, "/sys/class/net/%s/phys_port_name",
> > + ifname);
> > + MKSTR(phys_switch_id, "/sys/class/net/%s/phys_switch_id",
> > + ifname);
> > +
> > + file = fopen(phys_port_name, "rb");
> > + if (file == NULL)
> > + goto error;
>
> This goto is not correct.
> Representors will indeed be able to read, but PFs will not be able to
> open the phys_port_name (as there is none).
>From my testing, this is wrong, the PF can open the file, but it cannot
read (verified with strace).
> The phys_switch_id is accessible form both.
>
> I pasted a suggestion below[1], let me know what you think.
>
> > + port_name_set = fscanf(file, "%d%c", &data.port_name, &c) == 2
> > &&
> > + c == '\n';
> > + fclose(file);
> > + file = fopen(phys_switch_id, "rb");
> > + if (file == NULL)
> > + goto error;
> > + port_switch_id_set =
> > + fscanf(file, "%" SCNx64 "%c", &data.switch_id, &c) == 2 &&
> > + c == '\n';
> > + fclose(file);
> > + data.master = port_switch_id_set && !port_name_set;
> > + data.representor = port_switch_id_set && port_name_set;
> > + *info = data;
> > + return 0;
> > +error:
> > + rte_errno = errno;
> > + return -rte_errno;
> > +}
> > --
> > 2.18.0
>
> [1]
> diff --git a/drivers/net/mlx5/mlx5_ethdev.c b/drivers/net/mlx5/mlx5_ethdev.c
> index 6cfe02d..b64eb20 100644
> --- a/drivers/net/mlx5/mlx5_ethdev.c
> +++ b/drivers/net/mlx5/mlx5_ethdev.c
> @@ -1328,23 +1328,20 @@ struct ethtool_link_settings {
> ifname);
>
> file = fopen(phys_port_name, "rb");
> - if (file == NULL)
> - goto error;
> - port_name_set = fscanf(file, "%d%c", &data.port_name, &c) == 2 &&
> - c == '\n';
> - fclose(file);
> + if (file) {
> + port_name_set = fscanf(file, "%d%c", &data.port_name, &c) ==
> 2 &&
> + c == '\n';
> + fclose(file);
> + }
> file = fopen(phys_switch_id, "rb");
> - if (file == NULL)
> - goto error;
> - port_switch_id_set =
> - fscanf(file, "%" SCNx64 "%c", &data.switch_id, &c) == 2 &&
> - c == '\n';
> - fclose(file);
> + if (file) {
> + port_switch_id_set =
> + fscanf(file, "%" SCNx64 "%c", &data.switch_id, &c) ==
> 2 &&
> + c == '\n';
> + fclose(file);
> + }
> data.master = port_switch_id_set && !port_name_set;
> data.representor = port_switch_id_set && port_name_set;
Regards,
--
Nélio Laranjeiro
6WIND
