Introduce RT6352 instead of matching against RF7620.
Clean up channel setting rfvals.
Port bandwidth filter calibration.
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
So here is another round of improvements for MT7620 WiFi.
This commit fixes a few significant issues related to TX_PWR_CFG_x and
TX_ALC and also makes the code more readable by adding register
descriptions for things added for MT7620 and use the usual bit-field
access macros and the now defined macros instead of plain bit-ops and
magic numbers.
Properly describe EEPROM_TARGET_POWER at word 0x68 (== byte 0xD0) and
thereby fix internal TXALC which would otherwise just read
out-of-bounds of the EEPROM map.
Split-out tx-power/ALC related stuff into an additional function.
Fix VCO calibration, it was carried out properly in the channel
switching but incomplete in the actual VCO calibration function.
Also there is no need to trigger VCO calibration in channel switching,
the VCO calibration function is already being called at this point.
Remove it from channel switching function to avoid redundant code.
The TX power calibration differs significantly from all other
Mediatek/Ralink chips: They finally allow 0.5dB steps stored as 8-bit
values for (almost) each bitrate -- and promptly ran out of space and
for some reason didn't want to change the EEPROM layout. The hence
opted for a scheme of sharing values for some adjecent bitrates and
a highly over-complicated (or obfuscated?) way to populate the
TX_PWR_CFG_x registers with the values stored in the EEPROM.
The code here now looks much less complicated than what you see in the
vendor's driver, however, it does the exact same thing:
bGpwrdeltaMinus is a constant and always TRUE, hence half of the
code was dead. Gpwrdelta is always 0 (rather than using the value read
from the EEPROM). What remains is some very grotesque effort to avoid
0x20, probably some hardware bug related to some misunderstanding of
what a singed 8-bit value is (imagine: if it was a signed 6-bit value
then someone could believe that 0x20 == 0x0). And then they didn't
clean it up once they later on anandonned that whole story of having a
constant offset for 40 MHz channels and just set the offset to be
constant 0 -- there is no effort for avoiding 0x20 for the 20 MHz
values stored in the EEPROM, hence that's probably just a forbidden
value in the EEPROM specs and won't appear anyway...
Anyway, the whole thing felt like solving some college math test
where in the end everything cancels out and the result equals 0 ;)
To make sure that channel bandwidth power compensation really doesn't
need to be taken care of, output a warning when the corresponding
value stored in the EEPROM is non-zero.
Also there is no apparent reason to refrain from initializing RFCSR
register 13, it doesn't fail what-so-ever.
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
Effects of the bugs could include memory corruption, tx hangs, kernel
crahes, possibly other things as well
Signed-off-by: Felix Fietkau <nbd@nbd.name>
This commit combines all the changes I've made on my staging tree
into a single commit fixing many issues with our patch for MT7620.
First of all, checkpatch.pl revealed numerous code style issues with
the patch, so fix all the white-space and commets. Also use
usleep_range instead of legacy timing and relax timing for VCO
calibration just like the vendor driver does.
Several line programming registers were commented out in the patch.
Originally this came from the features present but disabled by default
in the vendor's driver (RTMP_TEMPERATURE_CALIBRATION and
ADJUST_POWER_CONSUMPTION_SUPPORT). Remove the dead code for now, it can
easily be re-added if we actually intend to support those features.
Move values from mt7620_freqconfig type into the existing rf_channel
struct, this shouldn't be a new typedef and it is possible to use the
existing struct because rf_channel got 4 32-bit fields, so two of the
8-bit values from mt7620_freqconfig can easily be stored in the same
32-bit field.
Map values such that
Rdiv -> rf1
N -> rf2
K -> rf3[0:7]
D -> rf3[8:15]
Ksd -> rf4
This makes the channel switching logic already look a bit more like
what we are used to in rt2x00... Probably many of the read-modify-write
calls could still be replaced by macros intended for that.
iq calibration seems to be identical to RT5592, so just enable it.
Test shows that this improves things quite a lot, datarates went up
by a couple of megabits when running iperf, signal quality seems jumpy
in the first few seconds once a station connencts, the stabelizes on a
value significantly better than what it was before.
Add description to the patch and reference the original OpenWrt commit
by which it was added.
The patch now passes checkpatch.pl and can thus be discussed with the
upstream authors of the rt2x00 driver.
Funded-by: https://www.kickstarter.com/projects/1327597961/better-support-for-mt7620a-n-in-openwrt-lede/
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
I needed a moment to figure out relation between this patchset and the
nl80211: fix validation of scheduled scan info for wowlan netdetect
It appears nl80211 commit will go on top of brcmfmac changes so it's
safe to backport these patches.
One patch that was excluded is commit 2a2a5d1835b6 ("brcmfmac: add
.update_connect_params() callback") as it depends on missing commit
088e8df82f91 ("cfg80211: Add support to update connection parameters").
Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
The radio would stop communicating completely. This issue was easiest to
trigger on AR913x devices, e.g. the TP-Link TL-WR1043ND, but other
hardware was occasionally affected as well.
The most critical issue was a race condition in disabling/enabling IRQs
between the IRQ handler and the IRQ processing tasklet
Signed-off-by: Felix Fietkau <nbd@nbd.name>
There was a bug in brcmfmac patch that could result in treating random
memory as source of country codes.
Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
This reverts commit c296ba834d.
According to several reports, the issues with the airtime fairness
changes are gone in current versions.
It's time to re-apply the patch now.
Signed-off-by: Felix Fietkau <nbd@nbd.name>
This allows some basic region switching on Netgear R8000. More devices &
codes may be added. Ideally it should be converted into DT info & patch.
Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
This makes use of cfg80211 feature backported & described in
188626f17c ("mac80211: backport cfg80211 support for
ieee80211-freq-limit DT property").
Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
Most mac80211 drivers leave the SMPS field in the HT capabilities
uninitialized (unfortunately defaults to static SMPS), which leads to
some devices limiting themselves to single-stream rates in some modes
(mostly mesh and IBSS).
Signed-off-by: Felix Fietkau <nbd@nbd.name>
This property allows specifying extra limits for wireless device in DT.
For a full documentation see upstream commit b330b25eaabd ("dt-bindings:
document common IEEE 802.11 frequency limit property").
Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
This reverts commit 528f46d082.
After this commit, several users reported stability issues. Revert it
now so it doesn't cause issues for the upcoming release
Signed-off-by: Felix Fietkau <nbd@nbd.name>
An external reset patch for AR955x accidentally led to external reset
being issued twice on AR913x, once before the RTC reset and once after.
This may be causing some stability issues.
Signed-off-by: Felix Fietkau <nbd@nbd.name>
This commit was added to improve reset time on old SoC devices that run
into chip hangs more frequently. However with the more recent addition
of full WMAC reset on these chips, it could be problematic.
Drop this patch to ensure that DMA activity is really stopped before the
chip reset is issued
Signed-off-by: Felix Fietkau <nbd@nbd.name>
This adds a patch that introduces airtime fairness scheduling to ath9k,
which can significantly improve network efficiency in mixed-rate
environments.
Signed-off-by: Toke Høiland-Jørgensen <toke@toke.dk>
Signed-off-by: Felix Fietkau <nbd@nbd.name>
This adds a patch that introduces airtime fairness scheduling to ath9k,
which can significantly improve network efficiency in mixed-rate
environments.
Signed-off-by: Toke Høiland-Jørgensen <toke@toke.dk>
Signed-off-by: Felix Fietkau <nbd@nbd.name>
The patch commit states:
"It's possible to make scanning consume almost arbitrary amounts
of memory, e.g. by sending beacon frames with random BSSIDs at
high rates while somebody is scanning.
Limit the number of BSS table entries we're willing to cache to
1000, limiting maximum memory usage to maybe 4-5MB, but lower
in practice - that would be the case for having both full-sized
beacon and probe response frames for each entry; this seems not
possible in practice, so a limit of 1000 entries will likely be
closer to 0.5 MB."
Signed-off-by: Johannes Berg <johannes.berg@intel.com>"
This patch was added in 4.4.36. But because LEDE backports
cfg80211, mac80211 and the wifi drivers separately, it needs
to be added manually for now. It can be dropped later as it
will be part of the next mac80211 refresh.
Signed-off-by: Christian Lamparter <chunkeey@googlemail.com>
