The hardware queue scheduling is apparently configured with fixed
priorities, which creates a nasty fairness issue where traffic from one
CPU can starve traffic from all other CPUs.
Work around this issue by forcing all tx packets to go through one CPU,
until this issue is fixed properly.
Signed-off-by: Felix Fietkau <nbd@nbd.name>
Commit af79fdbe4a changed the code to use tx_nbits for dual SPI tx
transfers, however the SPI stack only allows this when the device mode
includes the relevant bit as well
Signed-off-by: Felix Fietkau <nbd@nbd.name>
This code was marked as incompatible to Linux 4.4 well over a year ago
and nobody cared, and now it's breaking builds.
Signed-off-by: Felix Fietkau <nbd@nbd.name>
This target has consistently failed builds for a long time and it does
not look like it has any users left, nor any maintainers.
Signed-off-by: Felix Fietkau <nbd@nbd.name>
Rework the bundle-libraries.sh implementation to use a more robust approach
for executing host binaries through the shipped ELF loader and libraries.
The previous approach relied on symlinks pointing to a wrapper script which
caused various issues, especially with multicall binaries as the original
argv[0] name was not preserved through the ld.so invocation. Another down-
side was the fact that the actual binaries got moved into another directory
which caused executables to fail looking up resources with paths relative
to the executable location.
The new library wrapper implements the following improvements:
- Instead of symlinks pointing to a common wrapper, each ELF executable
is now replaced by a unqiue shell script which retains the original
program name getting called
- Instead of letting ld.so invoke the ELF executable directly, launch
the final ELF binary through a helper program which fixes up the argv[0]
argument for the target program
- Support sharing a common location for the bundled libraries instead of
having one copy in each directory containing wrapped binaries
Finally modify the SDK build to wrap the staging_dir and toolchain binaries
which allows to use the SDK on systems with a different glibc version.
Signed-off-by: Jo-Philipp Wich <jo@mein.io>
Without this patch swconfig will only allow setting up a total of 16
VLANs, with VLAN ID range of 0-15.
Tested on ubnt-erx.
Signed-off-by: Antonis Kanouras <antonis@metadosis.eu>
Fixing pvs1 bin voltage as found in GPL tarball and Qualcomm SDK.
Also adjusting smb208_s2a/b minimum voltage to reflect lowest value in pvs table.
Current ipq8065 devices in LEDE choose (based on eFuse data) pvs bin n4, that has
correct values in DT, but in future a new device may choose n1 or n6.
Without this change n1 is semi-correct and n6 is not available, because OPP driver
disregards bins with voltage values that regulator cannot supply.
Signed-off-by: Pavel Kubelun <be.dissent@gmail.com>
This adds USB initialization fixes for Danube, Amazon SE and xrx300 and
should fix the clock on at least Danube which hopefully closes FS#351.
The xrx200 usb driver now uses more memory for the dwc2 fifos, this was
increased in the chip compared to ar9.
This is based in part on the vendor documentation and the vendor code
base as a reference.
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
The DMA controller used only local locks, which will not lock the
section against other CPUs, fix this by using spin locks.
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
Rafał Miłecki