Message ID | 20210125152235.2942-2-dnlplm@gmail.com (mailing list archive) |
---|---|
State | Superseded |
Headers | show |
Series | net: usb: qmi_wwan: new mux_id sysfs file | expand |
On Mon, Jan 25, 2021 at 4:23 PM Daniele Palmas <dnlplm@gmail.com> wrote: > > Add qmimux interface sysfs file qmap/mux_id to show qmap id set > during the interface creation, in order to provide a method for > userspace to associate QMI control channels to network interfaces. > > Signed-off-by: Daniele Palmas <dnlplm@gmail.com> Thanks for doing this! Acked-by: Aleksander Morgado <aleksander@aleksander.es>
On Mon, 25 Jan 2021 17:14:28 +0100 Bjørn Mork wrote: > Daniele Palmas <dnlplm@gmail.com> writes: > > > Add qmimux interface sysfs file qmap/mux_id to show qmap id set > > during the interface creation, in order to provide a method for > > userspace to associate QMI control channels to network interfaces. > > > > Signed-off-by: Daniele Palmas <dnlplm@gmail.com> > > Acked-by: Bjørn Mork <bjorn@mork.no> We got two patches adding new sysfs files for QMI in close succession - is there a sense of how much this interface will grow over time? It's no secret that we prefer netlink in networking land.
Jakub Kicinski <kuba@kernel.org> writes: > On Mon, 25 Jan 2021 17:14:28 +0100 Bjørn Mork wrote: >> Daniele Palmas <dnlplm@gmail.com> writes: >> >> > Add qmimux interface sysfs file qmap/mux_id to show qmap id set >> > during the interface creation, in order to provide a method for >> > userspace to associate QMI control channels to network interfaces. >> > >> > Signed-off-by: Daniele Palmas <dnlplm@gmail.com> >> >> Acked-by: Bjørn Mork <bjorn@mork.no> > > We got two patches adding new sysfs files for QMI in close succession - > is there a sense of how much this interface will grow over time? The honest answer is no. I do not expect this interface to grow at all. But then I didn't expect it to grow before the two recent additions either... Both are results of feedback from the userspace developers actually using this interface. If I try to look into the future, then I do believe the first addition, the "pass_through" flag, makes further changes unnecessary. It allows the "rmnet" driver to take over all the functionality related to qmap/qmimux. The rmnet driver has a proper netlink interface for management. This is how the design should have been from the start, and would have been if the "rmnet" driver had existed when we added qmap support to qmi_wwan. Or if I had been aware that someone was working on such a driver. So why do we still need this last addition discussed here? Well, there are users of the qmi_wwan internal qmimux interface. They should move to "rmnet", but this might take some time and we obviously can't remove the old interface in any case. But there is a design flaw in that interface, which makes it rather difficult to use. This last addition fixes that flaw. I'll definitely accept the judgement if you want to put your foot down and say that this has to stop here, and that we are better served without this last fix. > It's no secret that we prefer netlink in networking land. Yes. But given that we have the sysfs interface for managing this qmimux feature, I don't see netlink as an alternative to this patch. The same really applies to the previous sysfs attribute, adding another flag to a set which is already exposed as sysfs attributes. The good news is that it allowed further qmimux handling to be offloaded to "rmnet", which does have a netlink interface. Bjørn
On Mon, Jan 25, 2021 at 04:22:34PM +0100, Daniele Palmas wrote: > Add qmimux interface sysfs file qmap/mux_id to show qmap id set > during the interface creation, in order to provide a method for > userspace to associate QMI control channels to network interfaces. > > Signed-off-by: Daniele Palmas <dnlplm@gmail.com> > --- > drivers/net/usb/qmi_wwan.c | 27 +++++++++++++++++++++++++++ > 1 file changed, 27 insertions(+) > > diff --git a/drivers/net/usb/qmi_wwan.c b/drivers/net/usb/qmi_wwan.c > index 7ea113f51074..9b85e2ed4760 100644 > --- a/drivers/net/usb/qmi_wwan.c > +++ b/drivers/net/usb/qmi_wwan.c > @@ -218,6 +218,31 @@ static int qmimux_rx_fixup(struct usbnet *dev, struct sk_buff *skb) > return 1; > } > > +static ssize_t mux_id_show(struct device *d, struct device_attribute *attr, char *buf) > +{ > + struct net_device *dev = to_net_dev(d); > + struct qmimux_priv *priv; > + ssize_t count = 0; > + > + priv = netdev_priv(dev); > + count += scnprintf(&buf[count], PAGE_SIZE - count, > + "0x%02x\n", priv->mux_id); Odd way to do this, please just use sysfs_emit(). It looks like you cut/pasted this from some other more complex logic. thanks, greg k-h
On Wed, 27 Jan 2021 08:26:13 +0100 Bjørn Mork wrote: > Jakub Kicinski <kuba@kernel.org> writes: > > We got two patches adding new sysfs files for QMI in close succession - > > is there a sense of how much this interface will grow over time? > > The honest answer is no. > > I do not expect this interface to grow at all. But then I didn't expect > it to grow before the two recent additions either... Both are results > of feedback from the userspace developers actually using this interface. > > If I try to look into the future, then I do believe the first addition, > the "pass_through" flag, makes further changes unnecessary. It allows > the "rmnet" driver to take over all the functionality related to > qmap/qmimux. The rmnet driver has a proper netlink interface for > management. This is how the design should have been from the start, and > would have been if the "rmnet" driver had existed when we added qmap > support to qmi_wwan. Or if I had been aware that someone was working on > such a driver. > > So why do we still need this last addition discussed here? Well, there > are users of the qmi_wwan internal qmimux interface. They should move > to "rmnet", but this might take some time and we obviously can't remove > the old interface in any case. But there is a design flaw in that > interface, which makes it rather difficult to use. This last addition > fixes that flaw. > > I'll definitely accept the judgement if you want to put your foot down > and say that this has to stop here, and that we are better served > without this last fix. > > > It's no secret that we prefer netlink in networking land. > > Yes. But given that we have the sysfs interface for managing this > qmimux feature, I don't see netlink as an alternative to this patch. > > The same really applies to the previous sysfs attribute, adding another > flag to a set which is already exposed as sysfs attributes. > > The good news is that it allowed further qmimux handling to be offloaded > to "rmnet", which does have a netlink interface. Thanks for the explanation. I'll trust you on this one :) I applied v2 and added the acks from v1.
diff --git a/drivers/net/usb/qmi_wwan.c b/drivers/net/usb/qmi_wwan.c index 7ea113f51074..9b85e2ed4760 100644 --- a/drivers/net/usb/qmi_wwan.c +++ b/drivers/net/usb/qmi_wwan.c @@ -218,6 +218,31 @@ static int qmimux_rx_fixup(struct usbnet *dev, struct sk_buff *skb) return 1; } +static ssize_t mux_id_show(struct device *d, struct device_attribute *attr, char *buf) +{ + struct net_device *dev = to_net_dev(d); + struct qmimux_priv *priv; + ssize_t count = 0; + + priv = netdev_priv(dev); + count += scnprintf(&buf[count], PAGE_SIZE - count, + "0x%02x\n", priv->mux_id); + + return count; +} + +static DEVICE_ATTR_RO(mux_id); + +static struct attribute *qmi_wwan_sysfs_qmimux_attrs[] = { + &dev_attr_mux_id.attr, + NULL, +}; + +static struct attribute_group qmi_wwan_sysfs_qmimux_attr_group = { + .name = "qmap", + .attrs = qmi_wwan_sysfs_qmimux_attrs, +}; + static int qmimux_register_device(struct net_device *real_dev, u8 mux_id) { struct net_device *new_dev; @@ -240,6 +265,8 @@ static int qmimux_register_device(struct net_device *real_dev, u8 mux_id) goto out_free_newdev; } + new_dev->sysfs_groups[0] = &qmi_wwan_sysfs_qmimux_attr_group; + err = register_netdevice(new_dev); if (err < 0) goto out_free_newdev;
Add qmimux interface sysfs file qmap/mux_id to show qmap id set during the interface creation, in order to provide a method for userspace to associate QMI control channels to network interfaces. Signed-off-by: Daniele Palmas <dnlplm@gmail.com> --- drivers/net/usb/qmi_wwan.c | 27 +++++++++++++++++++++++++++ 1 file changed, 27 insertions(+)