Right patch version this time, sorry!
* Patch 180-usb-xhci-add-support-for-performing-fake-doorbell.patch had to be adjusted slightly because of upstream adapted code.
* Refreshed patches.
Compile-tested: ramips/mt7621, x86/64
Run-tested: ramips/mt7621, x86/64
Signed-off-by: Stijn Segers <foss@volatilesystems.org>
Tested-by: Rosen Penev <rosenp@gmail.com>
Cake in kernel space now splits stats structure handling across netlink
messages to reduce stack usage issue flagged by upstream kernel checks.
Update user space (tc) qdisc handling to understand this new regime.
Cake also reports packet overheads & compensation in a different way so
add display code for this. e.g.
'tc -s qdisc show dev eth0' reports this extra detail:
min/max transport layer size: 28 / 1500
min/max overhead-adjusted size: 65 / 1550
average transport hdr offset: 14
Cake also supports output in JSON format.
Patch is bulkier than before because a (slightly out of date - see above
stats) man page is included for reference. Better than nothing!
Signed-off-by: Kevin Darbyshire-Bryant <ldir@darbyshire-bryant.me.uk>
Relevant changes:
0afc1be Fixes for kernel 4.16
d2d6780 Reinitialise overhead compensation stats when reconfiguring.
a3bab9d Export overhead compensation stats to userspace.
9cd2fa8 Split tin stats to its own structure to decrease size of tc_cake_xstats
71c7b44 Gather more statistics about packet length transformations.
0517357 Rework overhead compensation to use dynamic transport header offset instead of (inaccurate) static one.
c1a0c8e Refactor length handling code to better centralise overhead calculations
Signed-off-by: Kevin Darbyshire-Bryant <ldir@darbyshire-bryant.me.uk>
PKG_EXTMOD_SUBDIRS should be set to the sub directory where the kernel
module gets build in, for the ath10k-ct driver this changed in commit
3888e77c1c from ath10k to ath10k-4.13. Without this fix the depends
line of the ath10*.ko modules is empty and the kernel module load system
will not automatically load the depended modules like mac80211.
Fixes: 3888e77c1c ("ath10k-ct driver: use dma_alloc_coherent, 4.13 based driver")
Fixes: 23a388fe41 ("ath10k-ct: Force loading mac80211 and ath modules.")
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
The HTT-MGT variants transport management frames over the
normal HTT tx path, just like data frames. This saves
limitted WMI buffers which can become depleted if lots of
management frames become stuck in TX queues due to peer
that went away.
In addition, at least for the wave-1 firmware, htt-mgt is
required in order for 802.11r (fast roaming) authentication
to function properly.
The htt-mgt firmware requires the use of the ath10k-ct
driver. Normal non-htt-mgt ath10k-ct firmware should work
with stock drivers.
Signed-off-by: Ben Greear <greearb@candelatech.com>
This updates to latest ath10k-ct firmware. Hopefully we are
at the end of the development cycle for this firmware release,
so these should be stable.
wave-1 changes since last release:
Release 20
* Allow flushing peer when deleting. Hopefully this will allow the
peer delete command to happen in a reasonable amount of time even
if the RF environment is busy (or peer has died).
To enable this, set the high flag in the mac-addr second word in the
ath10k driver near end of the ath10k_wmi_op_gen_peer_delete method:
cmd->peer_macaddr.word1 |= __cpu_to_le32(0x80000000);
* Attempt to fix crash seen in resmgr-ocs, appearantly due to list corruption.
Use a temporary list instead of trying to rely on for-each-safe.
* Add flag to tx-descriptor to allow driver to request no-ack on data
frames. This is bit 15 on the flag1 field (previously un-used).
* Add option to support specifying the tx-rate-code and retry count on
a per-packet basis. Only a single series is supported at this time.
Useful mainly for radiotap monitor-tx type testing at this point.
* Fix crash on startup when chip is at -40 deg C and calibration fails. Instead
of asserting, just keep retrying calibration, which appears to start working
after a few minutes (when the chip warms up).
* Allow reporting per-chain rssi for management frames. We pack the values into
empty space in the mgt-frame wmi header. This will only be enabled if the driver
requests it, since otherwise the driver is assumed to not understand the new API.
ath10k-ct drivers that support this feature will automatically enable it.
* A customer reports a case that appears to be the hardware not properly detecting
end of AMPDU, so frames were being mis-delivered to the wrong peer. Attempt to
work around this, and in doing so, clean up a bunch of void* abuse in the block-ack
reordering code (could not ever confirm there was a problem in this area).
* Re-work the rx-mem logic to be less complicated and to use less memory.
* Attempt to fix crash that appearanty happens because the driver can sometimes
delete a vdev in 'up' state.
* Attempt to fix hung scan state machine issues.
* Fix crash in tx path due to un-initialized memory.
wave-2 changes since last release:
Release 10
* Fix an assert related to tx scheduling. This hopefully fixes
what appears to be a regression that I added some time back.
* Enable CSI reporting for 9984, and maybe 9888/9886. Only in
non-trimmed builds.
* Other stability improvements, including regression fixes from
some tricky bugs introduced in earlier releases.
* Allow compiling for IPQ4019 chipset.
* Firmware will now send txbf frames to the host (driver) if the
TXBF (0xF00000001) set-special feature is enabled, or when the radio
is in monitor mode. But, if the frame is consumed by the txbf_cv
logic, then the pkt cannot be delivered to the host in this manner. Instead,
a WMI event will be sent and host can find the txbf_cv data in shared
memory. See ath10k_wmi_event_txbf_cv_mesg() in ath10k-ct driver.
* Support rx-all-mgt option. When enabled, the firmware will deliver all
management frames that it can to the host. No RX filters are changed
when this option is enabled.
* Fix at least some problems with sending tx-beamforming frames to SU-MIMO
peers. Looks like this was a regression in my code.
* Fix a crash in rate-ctrl due to nss mismatch. This was something I introduced
while trying to fix other bugs in rate-ctrl some time back.
* Attempt to fix a sw-peer-key object leak in IBSS mode. The peer key code
is very complex, and shares some pointers as union members. I think I fixed
at least some of the issues, but would not be surprised if more exist.
* Improve ath10k user guide to document CT firmware features:
https://www.candelatech.com/ath10k-ug.php
* Add ct-special option to configure the txbf sounding time. See ath10k-ug.php
* Fix and allow the driver to tell the firmware to send sounding frames. See ath10k-ug.php
In further testing, this seems to fail much of the time, and I am not sure why.
Disabling this in diet (trimmed) builds.
* Fix crashes related to deleting peers while they are in power-save mode. Reported
by LEDE user on r7800 with 9984 NIC.
* Make rate-ctrl txbf probe work better. If enabled, the rate-ctrl logic will periodically
send out probes at an NSS that can to txbf. Previously, txbf probes would not reliably happen
if both AP and peer had the same nss (ie, 2x2 talking to 2x2). To enable this feature, you
need to enable the fwtest-cmdid number 20.
* Report rx-timeout error counters. These were previously un-reported, though the
field existed in the wmi struct already.
* txbf: Ignore frames not destined for us. If NIC is in promisc mode, it
could acquire and process NDPA frames that were not destined for it. Check
the dest-MAC and ignore frames not for us (pass them up the stack for monitor
mode instead of save them in the peer's rate-ctrl logic.)
* Port ping-pong crash handling and othe related features to IPQ4019 target. It should
now act similar to 9984 in this regard.
* Fix a few asserts related to txbf and tx-seq logic.
* Add custom-stats support, for rx-reorder-stats. Similar to what I did for wave-1.
* Disable AMSDU for IBSS. This now matches what I did for peregrine. It seems to
work better this way, though I did not debug it in detail.
* Enable the set-special command to re-enable AMSDU for IBSS if user wants to experiment.
* Fix bug where dbglog did not disable IRQs, so if you made dbglog messages from the IRQ
handler, it could cause corruption that could crash the firmware and/or corrupt the log
message buffers.
* Don't assert if there are no buffer descriptors for RX of non-data frame.
* Retry any stuck block-ack sessions every 20 seconds instead of just disabling BA for
ever when we get too many failures.
* Fix SGI flag when reporting tx-rate info. The flag moved since wave-1 days, and
I did not notice that when I ported my changes forward to wave-2.
* Allow disabling special CCA handling for IBSS txqs. Earlier testing indicated this
might improve throughput in some testing on 9984 chips in IBSS mode, but subsequent
testing looks about the same without it. Since I do not really understand what this
setting exists for, leave it at upstream defaults. A new set-special API command (0x12)
can be used to enable this hack for testing. Setting 0x1 bit disables special CCA handling
for non-beacon IBSS txqs, setting 0x2 bit disables it for beacon queues as well.
* Add MCAST-BCAST feature flag. This tells driver we do not need a monitor interface
to do MESH.
* When calculating the rx-address filter (affects ACK & BLOCK-ACK, among other things),
to not add in monitor interfaces if other interfaces are up. There is no need for
a monitor device to ACK frames.
Signed-off-by: Ben Greear <greearb@candelatech.com>
Among other things, this will check for an htt-mgt variant of
ath10k-ct firmware before loading 'normal' firmware, and it disables
verbose printing of firmware DBGLOG messages by default.
Signed-off-by: Ben Greear <greearb@candelatech.com>
Update DEPENDS and PROVIDES so that ath10k-ct firmware
and drivers can be used to replace stock firmware
and drivers. The -htt firmware variant, which requires
ath10k-ct driver now selects ath10k-ct driver when the
firmware is selected.
Signed-off-by: Ben Greear <greearb@candelatech.com>
During upstreaming the intel phy driver, support for the vr9 v1.1
embedded phys got lost. Backport the upstream send patch adding support
for the vr9 v1.1 embbeded phys to the driver.
Signed-off-by: Mathias Kresin <dev@kresin.me>
cosmetic fixes
Signed-off-by: Mathias Kresin <dev@kresin.me>
WHR-G300N has 5 ethernet ports (lan: 4, wan: 1), but there was no
correct configuration in 02_network script and 6 ports was configured
on the switch.
Also, since the MAC address was not acquired from factory partition,
incorrect values was set to LAN and WAN interfaces.
This commit fixes these issues.
Signed-off-by: INAGAKI Hiroshi <musashino.open@gmail.com>
The APPSBL and anything after that it not allowed to touch some of the
memory regions which are used by other components. Still trying to write to
the memory can lead to sudden device restarts
(IPQ40xx) # mw 87e80000 0
data abort
pc : [<873149f8>] lr : [<87308578>]
sp : 86edfc28 ip : 86ef4412 fp : 00000000
r10: 00000000 r9 : 00000000 r8 : 86edff68
r7 : 00000003 r6 : 8737e624 r5 : 86ef4420 r4 : 8736c154
r3 : 00000000 r2 : 00000010 r1 : 00000000 r0 : 00000000
Flags: nZCv IRQs off FIQs off Mode SVC_32
Resetting CPU ...
The device manufacturer only provided a very rough list of regions:
* rsvd1: 0x87000000 0x500000
* wifi_dump: 0x87500000 0x600000
* rsvd2: 0x87b00000 0x500000
A more detailed list for devices using the AP.DK reference design memory
maps was provided by Roman Yeryomin <roman@advem.lv> in commit 56f2df879fd
("ipq806x: ipq4019: add ap-dk01.1-c1 board support"):
* apps_bl: 0x87000000 0x400000
* sbl: 0x87400000 0x100000
* cnss_debug: 0x87500000 0x600000
* cpu_context_dump: 0x87b00000 0x080000
* tz_apps: 0x87b80000 0x280000
* smem: 0x87e00000 0x080000
* tz: 0x87e80000 0x180000
The u-boot function ipq_fdt_mem_rsvd_fixup seems to suggest that only the
rsvd2 (tz_apps, smem, tz) should be protected. All other regions would have
been removed by it when CONFIG_QCA_APPSBL_DLOAD is not enabled. This allows
to reduce the 16MB reserved memory region to only 4.5MB.
Signed-off-by: Sven Eckelmann <sven.eckelmann@openmesh.com>
The APPSBL and anything after that it not allowed to touch some of the
memory regions which are used by other components. Still trying to write to
the memory can lead to sudden device restarts
(IPQ40xx) # mw 87e80000 0
data abort
pc : [<873149f8>] lr : [<87308578>]
sp : 86edfc28 ip : 86ef4412 fp : 00000000
r10: 00000000 r9 : 00000000 r8 : 86edff68
r7 : 00000003 r6 : 8737e624 r5 : 86ef4420 r4 : 8736c154
r3 : 00000000 r2 : 00000010 r1 : 00000000 r0 : 00000000
Flags: nZCv IRQs off FIQs off Mode SVC_32
Resetting CPU ...
The device manufacturer only provided a very rough list of regions:
* rsvd1: 0x87000000 0x500000
* wifi_dump: 0x87500000 0x600000
* rsvd2: 0x87b00000 0x500000
A more detailed list for devices using the AP.DK reference design memory
maps was provided by Roman Yeryomin <roman@advem.lv> in commit 56f2df879fd
("ipq806x: ipq4019: add ap-dk01.1-c1 board support"):
* apps_bl: 0x87000000 0x400000
* sbl: 0x87400000 0x100000
* cnss_debug: 0x87500000 0x600000
* cpu_context_dump: 0x87b00000 0x080000
* tz_apps: 0x87b80000 0x280000
* smem: 0x87e00000 0x080000
* tz: 0x87e80000 0x180000
The u-boot function ipq_fdt_mem_rsvd_fixup seems to suggest that only the
rsvd2 (tz_apps, smem, tz) should be protected. All other regions would have
been removed by it when CONFIG_QCA_APPSBL_DLOAD is not enabled. This allows
to reduce the 16MB reserved memory region to only 4.5MB.
Signed-off-by: Sven Eckelmann <sven.eckelmann@openmesh.com>
There is currently no code to read the phy reset gpios for the ethernet
PHY. It would also have been better to use the more common name
"phy-reset-gpios" for this property.
Signed-off-by: Sven Eckelmann <sven@narfation.org>
The AVM Fritz!Box 4040 uses an IPQ4018 as SoC and not an IPQ4019. The DTS
must be adjusted to reflect this.
Signed-off-by: Sven Eckelmann <sven@narfation.org>
The GPIO configuration in the DTS have as third parameter the active
low/high configuration. This parameter is not easy to parse by humans when
it is only set to 0/1. It is better to use the predefined constants
GPIO_ACTIVE_HIGH and GPIO_ACTIVE_LOW.
Signed-off-by: Sven Eckelmann <sven.eckelmann@openmesh.com>
The "Devicetree Specification, Release v0.2 - 2.3.4 status" [1] only allows
the "okay" value for an operational device. The "ok" value will be accepted
by the kernel but should be avoided.
Signed-off-by: Sven Eckelmann <sven.eckelmann@openmesh.com>
The OpenMesh A42 will use the default config entry in the FIT when no other
on is found but prefers the config@om.a42. This only becomes relevant when
a Multi-FIT image is prepared for this device.
Signed-off-by: Sven Eckelmann <sven.eckelmann@openmesh.com>
Drop providing the virtual package ip by busybox which was added in commit
1cec4d4ef0.
Letting busybox provide the virtual package ip is not optimal for the
following reasons :
- Applications depending on ip expect either the ip-full or
ip-tiny package to be enabled.
- Busybox ip applet cannot be added or removed at runtime
Signed-off-by: Hans Dedecker <dedeckeh@gmail.com>
The driver updates include:
ath10k driver backport to fix WPA 'pn' related security bugs
(4.13 based driver only currently),
a fix for off-channel TX for CT wave-1 firmware, a likely
fix for napi related crashes, and a backport of the firmware fetch
patch.
AHB is needed for the IPQ4019 platform radios.
Signed-off-by: Ben Greear <greearb@candelatech.com>
[use common subject format]
Signed-off-by: Jo-Philipp Wich <jo@mein.io>
We use the dtc from the kernel and that does not have all the options
which u-boot would like to use now. make these parameters optional.
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
[wigyori@uid0.hu: renamed to 221-compatible-old-dtc.patch from 220-]
Signed-off-by: Zoltan HERPAI <wigyori@uid0.hu>
This fixes:
drivers/net/phy/b53/b53_priv.h:325:2: error: enumeration value '<board>' not handled in switch [-Werror=switch]
errors.
Fixes: 0de2213eea ("kernel: b53: look for NVRAM's "robo_reset" entry on every platform")
Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
Since kernel 4.1 bcm47xx_nvram_gpio_pin() is now defined in a global
header and can be safely called even on non-Broadcom platforms.
This change makes b53 look for "robo_reset" on ARCH_BCM_5301X and
slightly simplifies the code.
Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
The chosen dts configuration linux,initrd-* gives an error message
on bootup of kirkwood-iconnect. Since initramfs/initrd is not used
remove these options from the dts.
Reported-by: Daniel Engberg <daniel.engberg.lists@pyret.net>
Tested-by: Daniel Engberg <daniel.engberg.lists@pyret.net>
Signed-off-by: Paul Wassi <p.wassi@gmx.at>
Add patches for 4.14, undoing upstream changes for Linksys devices
regarding DSA. Instead, the switchdev driver marvell,88e6171 is used.
Tested-by: Daniel Engberg <daniel.engberg.lists@pyret.net>
Tested-by: Alberto Bursi <alberto.bursi@outlook.it>
Signed-off-by: Paul Wassi <p.wassi@gmx.at>
U-Boot now requires GCC > 5
Catch up with upstream and move some configuration options from
the header files to the corresponding defconfig files.
Also move some options of patch 010 affecting the whole platform
to 010's device only.
Signed-off-by: Paul Wassi <p.wassi@gmx.at>
A regression seriously affecting the Linksys WRT1900ACv1 (Mamba) was
introduced some time between the OpenWrt/LEDE v4.4 and v4.9 kernels.
The root cause has not yet been identified, despite many attempts for
more than a year. Disabling the SoC specific CPU idle support should
mitigate this issue.
The symptoms on an affected system are unwanted reboots at a variable
frequency. In many cases almost immediately after boot, causing a
bootloop. This effectively disables support for Mamba on OpenWrt
with kernels > v4.4.
Signed-off-by: Felix Fietkau <nbd@nbd.name>
print_int used 'int' type internally, whereas print_uint used 'uint64_t'
These helper functions eventually call vfprintf(fp, fmt, args) which is
a variable argument list function and is dependent upon 'fmt' containing
correct information about the length of the passed arguments.
Unfortunately print_int v print_uint offered no clue to the programmer
that internally passed ints to print_uint were being promoted to 64bits,
thus the format passed in 'fmt' string vs the actual passed integer
could be different lengths. This is even more interesting on big endian
architectures where 'vfprintf' would be looking in the middle of an
int64 type. Symptoms of this included tc qdisc showing bizarre values
for a variety of fields across a variety of qdiscs (e.g. refcnt, flows,
quantum)
print_u/int now stick with native int size.
A similar patch has been sent upstream.
Fixes FS#1425
Signed-off-by: Kevin Darbyshire-Bryant <ldir@darbyshire-bryant.me.uk>
94b6878 Tidy crypto.c of old library compat. Now need libnettle 3.
8b96552 Fix compiler warning.
Signed-off-by: Kevin Darbyshire-Bryant <ldir@darbyshire-bryant.me.uk>
The latest bootloader versions load the firmware into memory and call
`chk_dniimg` (defined in Netgear GPL release), which expects to find
three consecutive block-aligned uImages. Add two fake uImage headers
after the kernel to fool this check.
This wastes up to 128k of space for alignment. The alternative would be
to put the rootfs in a second uImage, but this would limit the firmware
size to 0x710000 (the number of bytes loaded and verified by the
bootloader) instead of 0x7b0000 (the size of the firmware partition).
Signed-off-by: Thomas Nixon <tom@tomn.co.uk>
