On 2019-01-25 13:41, Stanislaw Gruszka wrote:
> On Fri, Jan 25, 2019 at 11:25:46AM +0100, Lorenzo Bianconi wrote:
>> > On Fri, Jan 25, 2019 at 10:02:38AM +0100, Lorenzo Bianconi wrote:
>> > > > On Thu, Jan 24, 2019 at 11:20:42PM +0100, Lorenzo Bianconi wrote:
>> > > > > > > >
>> > > > > > > > I guess this does not work if you add 2 vifs and then you remove the first one
>> > > > > > > > (you will end up with a wrong configuration in MT_MAC_ADDR_DW{0,1}). I guess
>> > > > > > > > the hw will not work well if MT_MAC_ADDR_DW{0,1} is not properly configured
>> > > > > >
>> > > > > > Maybe I am missing something, but let's assume you add the interface vif0 with address
>> > > > > > 00:11:22:33:44:55 (MT_MAC_ADDR_DW{0,1} will be set to 00:11:22:33:44:55) and
>> > > > > > then you add vif1 with address 00:aa:bb:cc:dd:ee. If at some point you remove
>> > > > > > vif0 MT_MAC_ADDR_DW{0,1} will not be properly reconfigured. The problem will
>> > > > > > be more complex if you have more interfaces
>> > > >
>> > > > Ok, so in remove_interface extra code can be added to implement that.
>> >
>> > Something like this should address the issue you raised:
>> >
>> > diff --git a/drivers/net/wireless/mediatek/mt76/mt76x02_util.c b/drivers/net/wireless/mediatek/mt76/mt76x02_util.c
>> > index 3880caa0c64a..44b4af928a4e 100644
>> > --- a/drivers/net/wireless/mediatek/mt76/mt76x02_util.c
>> > +++ b/drivers/net/wireless/mediatek/mt76/mt76x02_util.c
>> > @@ -320,6 +320,15 @@ void mt76x02_sta_remove(struct mt76_dev *mdev, struct ieee80211_vif *vif,
>> > }
>> > EXPORT_SYMBOL_GPL(mt76x02_add_interface);
>> >
>> > +static void mt76x02_setaddr_iterator(void *data, u8 *mac,
>> > + struct ieee80211_vif *vif)
>> > +{
>> > + struct mt76x02_dev *dev = data;
>> > +
>> > + if (!ether_addr_equal(dev->mt76.macaddr, vif->addr))
>> > + mt76x02_mac_setaddr(dev, vif->addr);
>>
>> This will end-up with multiple configurations if we have more than two
>> interfaces, right?
>
> No. It is called only when number of vifs change from 2 to 1 . But there
> should be additional check 'dev->vif_mask & BIT(mvif->idx)' since we can
> have removing interface still present in mac80211.
>
>> > void mt76x02_remove_interface(struct ieee80211_hw *hw,
>> > struct ieee80211_vif *vif)
>> > {
>> > @@ -328,6 +337,10 @@ void mt76x02_remove_interface(struct ieee80211_hw *hw,
>> >
>> > mt76_txq_remove(&dev->mt76, vif->txq);
>> > dev->vif_mask &= ~BIT(mvif->idx);
>> > +
>> > + if (hweight16(dev->vif_mask) == 1)
>> > + ieee80211_iterate_interfaces(hw, 0, mt76x02_setaddr_iterator,
>> > + dev);
>>
>> I guess we have the same issue if we have more than two interfaces and we
>> remove the first one that has been configured.
>
> We can not configure more than one MAC address without using MT_MAC_ADDR_EXT
> registers. If there are more than one vif, there is no point to set MT_MAC_ADDR.
>
>> Moreover I am a little worried about tpt regressions with this patch.
>> Are you sure that if you use complete different mac addresses on a multivif scenario
>> you can get the same tpt on all the interfaces? Could you please provide some
>> tpt results?
>
> How exactly posted patch can cause tpt regression ?
>
> Posted patch just add possibility to configure HW MAC address
> by this:
>
> iw dev wlan0 del
> iw phy phy0 interface add wlan0 type managed addr 00:11:22:33:44:55
>
> what is feature of mt76x2u. Patch just extend that possibly to other
> mt76x02 devices and allow to remove custom mt76x2u add_interfacea
> callback.
The main part that could cause issues is that you're changing the way
that the vif index is calculated. Without the patch, it's calculated
from the MAC address in a way consistent with what the hardware expects.
With the patch, it's just allocated from a mask.
The vif index ends up being passed down to the hardware as a BSS index
WCID attribute in mt76x02_mac_wcid_setup.
We would have to run some tests with multiple AP interfaces, bringing up
secondary interfaces in a different order to see if there are any
regressions there if the BSS index no longer matches the MAC address
based index.
- Felix
