On 11/6/20 11:12 AM, Martin Kepplinger wrote:
hi,
Hi,
I start the driver with options rsi_sdio dev_oper_mode=13 and (try to)
connect to a usual wpa2 encrypted network:
The rsi driver is unusably slow for me still. It slowly (sometimes)
connects to a network, but even then it's unusable. To me it seems like
the drivers sends way too little; tx() that is registered with mac80211
is rarely called - just 4 times in the below logs (where other drivers
would send data for each received packet afaik).
Thanks a lot Marek for working on the driver! I already use
https://lore.kernel.org/netdev/20201015111616.429220-1-marex@xxxxxxx/ here.
Make sure to add also:
https://lore.kernel.org/netdev/20201103180941.443528-1-marex@xxxxxxx/
that one made a lot of difference on RX performance, not so much on TX
though.
I also have this locally (better use the module param than patch the
kernel), because it seems wifi/bt coex is broken.
diff --git a/drivers/net/wireless/rsi/rsi_91x_sdio.c
b/drivers/net/wireless/rsi/rsi_91x_sdio.c
index 1bebba4e8527..2b2a9485aae0 100644
--- a/drivers/net/wireless/rsi/rsi_91x_sdio.c
+++ b/drivers/net/wireless/rsi/rsi_91x_sdio.c
@@ -22,7 +22,7 @@
#include "rsi_hal.h"
/* Default operating mode is wlan STA + BT */
-static u16 dev_oper_mode = DEV_OPMODE_STA_BT_DUAL;
+static u16 dev_oper_mode = DEV_OPMODE_WIFI_ALONE;
module_param(dev_oper_mode, ushort, 0444);
MODULE_PARM_DESC(dev_oper_mode,
"1[Wi-Fi], 4[BT], 8[BT LE], 5[Wi-Fi STA + BT classic]\n"
There seem to be a huge amount of issues with those RSI WiFi devices,
thanks to Angus for keeping track of them here:
https://source.puri.sm/Librem5/linux-next/-/issues?scope=all&utf8=%E2%9C%93&state=opened&search=redpine
There is also the downstream driver here:
git://github.com/SiliconLabs/RS911X-nLink-OSD
unfortunatelly, there is no changelog nor separate patches, just one
huge code-dump. The driver is some sort of derivative of the upstream
driver, so you can do a diff, but the diff is some thousands of lines of
change.
You can however try compiling it and see whether this makes your wifi
work, if so, then try to figure out which change is the fix. That's how
I came to the TX EAPOL fix, I just spent an insane amount of time trying
to figure out what change made the wifi work against iwlwifi AP in the
vendor driver, removing parts of the diff until I came to the one line.
Does anyone actually use the driver or at least has seen similar
behaviour, or even know what could go wrong?
So I added messages at a few spots in the rx/tx path and this is while
connected to a network (without doing much):
[ 124.277947] rsi_91x: RX Packet Type: QOS DATA
[ 124.377107] rsi_91x: rsi_mgmt_pkt_recv: Msg Len: 333, Msg Type: 2
[ 124.377118] rsi_91x: rsi_mgmt_pkt_to_core: rssi=33 channel=5
[ 124.377123] rsi_91x: RX Dot11 Mgmt Pkt Type: BEACON
[ 124.377128] rsi_91x: RX Packet Type: BEACON
[ 124.377915] rsi_91x: RX Packet Type: QOS DATA
I haven't seen this one, no, sorry. How is your AP configured (which
wifi is in it, how is the hostap configured, ...)?
