| Message ID | 20210603120609.58932-2-chris.chiu@canonical.com (mailing list archive) |
|---|---|
| State | Changes Requested |
| Delegated to: | Kalle Valo |
| Headers | show |
| Series | Make ampdu tx work correctly | expand |
Hi Chris, > Since AMPDU_AGGREGATION is set so packets will be handed to the > driver with a flag indicating A-MPDU aggregation and device should > be responsible for setting up and starting the TX aggregation with > the AMPDU_TX_START action. The TX aggregation is usually started by > the rate control algorithm so the HAS_RATE_CONTROL has to be unset > for the mac80211 to start BA session by ieee80211_start_tx_ba_session. > > The realtek chips tx rate will still be handled by the rate adaptive > mechanism in the underlying firmware which is controlled by the > rate mask H2C command in the driver. Unset HAS_RATE_CONTROL cause > no change for the tx rate control and the TX BA session can be started > by the mac80211 default rate control mechanism. This seems ... strange, to say the least? You want to run the full minstrel algorithm just to have it start aggregation sessions at the beginning? I really don't think this makes sense, and it's super confusing. It may also result in things like reporting a TX rate to userspace/other components that *minstrel* thinks is the best rate, rather than your driver's implementation, etc. I suggest you instead just call ieee80211_start_tx_ba_session() at some appropriate time, maybe copying parts of the logic of minstrel_aggr_check(). johannes
On Fri, Jun 11, 2021 at 4:18 AM Johannes Berg <johannes@sipsolutions.net> wrote: > > Hi Chris, > > > Since AMPDU_AGGREGATION is set so packets will be handed to the > > driver with a flag indicating A-MPDU aggregation and device should > > be responsible for setting up and starting the TX aggregation with > > the AMPDU_TX_START action. The TX aggregation is usually started by > > the rate control algorithm so the HAS_RATE_CONTROL has to be unset > > for the mac80211 to start BA session by ieee80211_start_tx_ba_session. > > > > The realtek chips tx rate will still be handled by the rate adaptive > > mechanism in the underlying firmware which is controlled by the > > rate mask H2C command in the driver. Unset HAS_RATE_CONTROL cause > > no change for the tx rate control and the TX BA session can be started > > by the mac80211 default rate control mechanism. > > This seems ... strange, to say the least? You want to run the full > minstrel algorithm just to have it start aggregation sessions at the > beginning? > > I really don't think this makes sense, and it's super confusing. It may > also result in things like reporting a TX rate to userspace/other > components that *minstrel* thinks is the best rate, rather than your > driver's implementation, etc. > > I suggest you instead just call ieee80211_start_tx_ba_session() at some > appropriate time, maybe copying parts of the logic of > minstrel_aggr_check(). > > johannes > > Based on the description in https://github.com/torvalds/linux/blob/master/net/mac80211/agg-tx.c#L32 to L36, if we set HAS_RATE_CONTROL, which means we don't want the software rate control (default minstrel), then we will have to deal with both the rate control and the TX aggregation in the driver, and the .ampdu_action is not really required. Since the rtl8xxxu driver doesn't handle the TX aggregation, and the minstrel is the default rate control (can't even be disabled), that's the reason why I want to unset the HAS_RATE_CONTROL to make use of the existing mac80211 aggregation handling. And the minstrel doesn't really take effect for rate selection in HT mode because most drivers don't provide HT/VHT rates in .bitrates of the ieee80211_supported_band data structure which is required for hw->wiphy->bands. The mac80211 API ieee80211_get_tx_rate() will return 0 when the IEEE80211_TX_RC_MCS is set in rate flags. The tx rate which is filled in the tx descriptor makes no difference because the underlying rate selection will be actually controlled by the controller which we can set rate mask via H2C command. Unless we force the fixed rate in the TX descriptor, we don't really have to fill the tx rate. Reporting TX rate of each packet will not depend on the rate from the minstrel, drivers have to handle it by itself. I'll also try to address that in my next PATCH series. Chris
On Fri, 2021-06-11 at 22:47 +0800, Chris Chiu wrote: > > Based on the description in > https://github.com/torvalds/linux/blob/master/net/mac80211/agg-tx.c#L32 > to L36, if we set HAS_RATE_CONTROL, which means we don't want the > software rate control (default minstrel), then we will have to deal > with both the rate control and the TX aggregation in the driver, and > the .ampdu_action is not really required. > I don't think this is true. You'll probably still want to use the A-MPDU state machine in mac80211, etc. What you *don't* get without rate control in mac80211 is any decision on whether or not to enable A-MPDU, but that's something you can easily do elsewhere and just call ieee80211_start_tx_ba_session() at an appropriate point in time. johannes
diff --git a/drivers/net/wireless/realtek/rtl8xxxu/rtl8xxxu_core.c b/drivers/net/wireless/realtek/rtl8xxxu/rtl8xxxu_core.c index 9ff09cf7eb62..4cf13d2f86b1 100644 --- a/drivers/net/wireless/realtek/rtl8xxxu/rtl8xxxu_core.c +++ b/drivers/net/wireless/realtek/rtl8xxxu/rtl8xxxu_core.c @@ -6678,7 +6678,6 @@ static int rtl8xxxu_probe(struct usb_interface *interface, /* * The firmware handles rate control */ - ieee80211_hw_set(hw, HAS_RATE_CONTROL); ieee80211_hw_set(hw, AMPDU_AGGREGATION); wiphy_ext_feature_set(hw->wiphy, NL80211_EXT_FEATURE_CQM_RSSI_LIST);