This "newer" version allows using 3rd party PHYs that are already
supported in the upstream kernel. Some devices use non-QCA PHYs, and
thus must rely on the kernel PHY drivers instead.
Signed-off-by: Alexandru Gagniuc <mr.nuke.me@gmail.com>
Signed-off-by: Robert Marko <robimarko@gmail.com>
This reverts commit abdc8ff534, reversing
changes made to 76602b9c18.
It looks like this needs to be further refined as users are reporting
reduced performance and we also see spurious NAPI over-budget warnings.
Signed-off-by: Robert Marko <robimarko@gmail.com>
2e82c51 nss-dp: edma-v1: improve napi and irq handling for rx/tx
92aa9ba nss-dp: edma-v1: add support for threaded napi
75129b9 nss-dp: edma-v1: split tx/rx path to specific napi and irq
2c27cb4 nss-dp: edma-v1: convert rx/tx store to idr implementation
Utilize napi_gro_receive instead of plain netif_receive_skb on EDMA v1.
It provides significant performance improvements when testing with iperf3.
Signed-off-by: Robert Marko <robimarko@gmail.com>
A phy-handle is a more intuitive way to declare the PHY connection
when compared to the abomination that qualcomm implemented. In the
interest of maintaing the sanity of developers porting hardware,
shield them from the abomination wth the use of 'phy-handle.
Signed-off-by: Alexandru Gagniuc <mr.nuke.me@gmail.com>
Add patch to permit more modularization and disable
additional module not used. PPPOE is disabled by default.
Will be enabled later if PPPOE offload will be reintroduced.
Signed-off-by: Ansuel Smith <ansuelsmth@gmail.com>
Net core changed implementation upstream. This results in nss-dp
printing confusing info log to the user as it does print
the netdev name before netdev is registered. Move the old log to dbg
and add an additional info log after netdev_register to give the user
feedback of correct driver probe.
Signed-off-by: Ansuel Smith <ansuelsmth@gmail.com>
Drop use of dmac_{inv/clean}_range and use generic dma sync api.
This should permit us to compile nss-drv without adding additional
patch to the upstream kernel.
Signed-off-by: Ansuel Smith <ansuelsmth@gmail.com>
Without this, the "net.bridge.bridge" will append into /etc/sysctl.d/qca-nss-ecm.conf in every system reboot/startup.
Signed-off-by: Xiaoli Chan <2209553467@qq.com>
Codeaurora is now in archive-only mode as everything has been moved to the
Gitlab based git.codelinaro.org
So, update all of the source URL-s as well as update to the latest 11.4
commit if there is a newer one.
Refresh patches as well if any have offsets or fuzzes.
Signed-off-by: Robert Marko <robimarko@gmail.com>
Try and solve the roaming issue by trying to replicate what NSS bridge
module is doing, but by utilizing switchdev FDB notifiers instead of
adding new notifiers to the bridge code.
We register a new non-blocking switchdev notifier and simply wait for
notification, and then process the SWITCHDEV_FDB_DEL_TO_DEVICE
notifications.
Those tell us that a certain FDB entry should be removed, then a VSI ID
is fetched for the physical PPE port and using that VSI ID and the
notification provided MAC adress existing FDB entry gets removed.
Mark qca-nss-dp as conflicting with the old NSS bridge module so
both cant be installed at the same time.
Signed-off-by: Robert Marko <robimarko@gmail.com>
Mark the NSS-DP kmod as required for preinit, thus it gets loaded before
preinit.
This finally allows using failsafe as now the network interfaces are
available for preinit to configure them.
Its still required to set the correct interface in preinit script for
each board.
Signed-off-by: Robert Marko <robimarko@gmail.com>
Add nss-crypto from QSDK11.4.
It provides support for the EIP-197 crypto HW.
It requires the EIP-197 firmware binaries.
They are not freely distributed and have to be pulled
from the stock firmware.
Made for IPQ807x.
Signed-off-by: Robert Marko <robimarko@gmail.com>
This is just a helper package that helps ensure
that the EIP-197 firmware is installed by being
used as a dependency on nss-crypto.
Note that the package does not provide or ship
with firmware binaries itself.
Those need to be pulled from the stock firmware.
Binaries required are:
* ifpp.bin
* ipue.bin
* ofpp.bin
* opue.bin
They should be placed inside of the src folder.
Note that we add the src folder to the .gitignore to
prevent accidentally pushing the FW to a public repo.
Signed-off-by: Robert Marko <robimarko@gmail.com>
Since kernel 5.15 ethernet/PHY related IOCTL-s have been split from the
generic IOCTL netdev op.
So, implement the new op instead of the generic one which is considered
for private IOCTL-s only now for 5.15+.
Signed-off-by: Robert Marko <robimarko@gmail.com>
AX3600 requires the same LED quirk so that PHY LED-s will blink even
once Linux resets the PHY.
So, just check for its compatible.
Signed-off-by: Robert Marko <robimarko@gmail.com>
The AX9000 quirk patch got left in the NSS root folder as well,
drop it.
Fixes: c78f42fe59 ("qca807x: add a LED quirk for Xiaomi AX9000")
Signed-off-by: Robert Marko <robimarko@gmail.com>
Xiaomi AX9000 has a single LED for each of 4 gigabit ethernet ports that
are connected to QCA8075, and that LED is connected to the 100M LED pin.
So, by default it will only work when in 10 or 100Mbit mode, this is quite
annoying and makes no sense(If they have connected it to the 1000Mbit LED
pin then it would have worked for 10/100 by default as well).
So, to solve this add a check for system compatible as we cant parse if
from DTS in any other way and set the 100M LED to blink on 1000Base-T
as well.
Signed-off-by: Robert Marko <robimarko@gmail.com>
Add SSDK support for kernel 5.15, it just required the configuration check
and QCA8081 IRQ model update.
Signed-off-by: Robert Marko <robimarko@gmail.com>
There is no need for the mismatched 11.3 FW support on 11.4
SPF drivers, so drop the support.
This reverts commit f806309bbe.
Signed-off-by: Robert Marko <robimarko@gmail.com>
Qualcomm has finally made the firmware publicly
available, so lets update to it.
This will allow removing the compatibility patches.
Signed-off-by: Robert Marko <robimarko@gmail.com>
Recently OpenWrt backported the updated of_get_mac_address()
function which returns and error code instead.
So, patch the SSDK to use it and fix the compilation error.
Signed-off-by: Robert Marko <robimarko@gmail.com>
Recently OpenWrt backported the updated of_get_mac_address()
function which returns and error code instead.
So, patch the SSDK to use it and fix the compilation error.
Signed-off-by: Robert Marko <robimarko@gmail.com>
If using ECM, cpu load goes up (around 1.0) and stucks there.
This is due to using uninterruptible sleep function,
the patch changes this to interruptible sleep function.
Signed-off-by: Dirk Buchwalder buchwalder@posteo.de
Actual code is being built in single thread mode as
SSDK enforces it, this will just cause OpenWrt buildroot
to fail sometimes as the headers it tries to use are not
yet installed by the SSDK.
So, disable parallel build as the time saving was minor
and it caused issues with parallel building.
Signed-off-by: Robert Marko <robimarko@gmail.com>
This commit fixes a dependency issue that was not allowing
qca-nss-clients to show up in menuconfig unless kmod-ppp was selected
Signed-off-by: Tiago Gaspar <tiagogaspar8@gmail.com>
This commit fixes a dependency issue that was not allowing qca-nss-ecm
to show up in menuconfig unless kmod-ppp was selected
Signed-off-by: Tiago Gaspar <tiagogaspar8@gmail.com>
DMA functions used for NSS FW coredump are called
with NULL being passed instead of device pointer.
This will obviously cause a NULL pointer exception
and crash the kernel.
So, fix it by passing proper device structure pointer.
Signed-off-by: Robert Marko <robimarko@gmail.com>
By accident ECM depended twice on kmod-ipt-physdev
instead of depending on iptables-mod-physdev.
So fix this by replacing the duplicate dependency with
iptables-mod-physdev.
Signed-off-by: Robert Marko <robimarko@gmail.com>
