fa731838c5 cleared the forced speed in the QSGMII PCS_ALL_CH_CTL
register during probe, but this is only correct for GMACs that are not
configured with fixed-link. This prevented GMACs configured with both
phy-mode = "sgmii" and fixed-link from working properly, as discussed at
https://github.com/openwrt/openwrt/pull/3954#issuecomment-834625090 and
the comments that follow. Notably, this prevented all communication
between gmac2 and the switch on the Netgear R7800.
The correct behavior is to set the QSGMII PCS_ALL_CH_CTL register by
considering the gmac's fixed-link child, setting the speed as directed by
fixed-link if present, and otherwise clearing it as was done previously.
Fixes: fa731838c5 ("ipq806x: dwmac: clear forced speed during probe")
Signed-off-by: Mark Mentovai <mark@moxienet.com>
Tested-by: Hannu Nyman <hannu.nyman@iki.fi>
Run-tested: ipq806x/ubnt,unifi-ac-hd, ipq806x/netgear,r7800
Cc: Petr Štetiar <ynezz@true.cz>
Cc: Ansuel Smith <ansuelsmth@gmail.com>
Tested-by: Ansuel Smith <ansuelsmth@gmail.com>
On a Ubiquiti UniFi AC HD (ubnt,unifi-ac-hd, UAP-AC-HD, UAP301), a
forced speed on gmac1 is set in the QSGMII PCS_ALL_CH_CTL register,
presumably by the bootloader (4.3.28), preventing the interface from
being usable. The QSDK NSS GMAC driver takes care to clear the forced
speed in nss_gmac_qsgmii_dev_init
(https://source.codeaurora.org/quic/qsdk/oss/lklm/nss-gmac/tree/ipq806x/nss_gmac_init.c?h=nss
at d5bb14925861).
gmac1 is connected to the port on the device labeled SECONDARY, and is
currently eth0 but will be switched to eth1 by a subsequent patch. By
clearing the QSGMII PCS forced speed during dwmac initialization when
SGMII is in use, this port becomes usable.
This patch is upstreamable, and will be sent upstream after successful
testing in OpenWrt.
Signed-off-by: Mark Mentovai <mark@moxienet.com>
Build-tested: ipq806x/ubnt,unifi-ac-hd
Run-tested: ipq806x/ubnt,unifi-ac-hd
