Bugfix from linux head - don't delay acks from ECN congestion
experienced in some situations.
git-svn-id: svn://svn.openwrt.org/openwrt/trunk@33686 3c298f89-4303-0410-b956-a3cf2f4a3e73
This allows sysupgrade for devices such as the Gateworks Avila/Cambria
product families based on the ixp4xx using the redboot bootloader with
combined FIS directory and RedBoot config partitions on larger FLASH
devices with larger eraseblocks.
This second iteration of this patch addresses previous issues:
- whitespace breakage fixed
- unlock in all scenarios
- simplification and fix logic bug
Signed-off-by: Tim Harvey <tharvey@gateworks.com>
git-svn-id: svn://svn.openwrt.org/openwrt/trunk@33681 3c298f89-4303-0410-b956-a3cf2f4a3e73
This feature was experimental in old kernels but
that flag has been removed in 2.6.36. Additionally,
the option is enabled by default since then.
See: http://www.spinics.net/lists/linux-usb/msg31993.html
git-svn-id: svn://svn.openwrt.org/openwrt/trunk@33574 3c298f89-4303-0410-b956-a3cf2f4a3e73
Fixes this section mismatch warning:
WARNING: vmlinux.o(.meminit.text+0xb14): Section mismatch in reference
from the function memblock_insert_region() to the function
.init.text:crashlog_init_memblock()
The function __meminit memblock_insert_region() references
a function __init crashlog_init_memblock().
If crashlog_init_memblock is only used by memblock_insert_region then
annotate crashlog_init_memblock with a matching annotation.
Also, remove the __init annotations from 'include/linux/crashlog.h'
git-svn-id: svn://svn.openwrt.org/openwrt/trunk@33137 3c298f89-4303-0410-b956-a3cf2f4a3e73
One condition before codel_Newton_step() was not good if
we never left the dropping state for a flow. As a result
rec_inv_sqrt was 0, instead of the ~0 initial value.
codel control law was then set to a very aggressive mode, dropping
many packets before reaching 'target' and recovering from this problem.
Brought over from 3.5-stable
Signed-off-by: Dave Taht <dave.taht@bufferbloat.net>
git-svn-id: svn://svn.openwrt.org/openwrt/trunk@32950 3c298f89-4303-0410-b956-a3cf2f4a3e73
It will be used to start/stop the switch if that is
supported by the given board.
git-svn-id: svn://svn.openwrt.org/openwrt/trunk@32943 3c298f89-4303-0410-b956-a3cf2f4a3e73
It is currently badly broken on ARM, until fixed, disable it
git-svn-id: svn://svn.openwrt.org/openwrt/trunk@32926 3c298f89-4303-0410-b956-a3cf2f4a3e73
I did not add the last patch, because it add some stubs for the flash drivers and our patches have to be changed for that, if the flash support in bcma is able to do something I will integrate it.
This also contains 3 additional patches fixing some bugs in bcma.
git-svn-id: svn://svn.openwrt.org/openwrt/trunk@32880 3c298f89-4303-0410-b956-a3cf2f4a3e73
Despite Westwood's theoretical advantages, in nearly
every benchmark we ran last year, TCP cubic won, whether it be
on correct RTT estimates, amount of buffering, responsiveness,
etc. on current hardware and software designs.
(both need timestamps on to work well, besides)
TCP cubic is better maintained and understood than westwood,
also.
While a scenario where westwood would win possibly exists,
there is too much buffering in the wifi stack in particular
at present, to see any improvement.
If you wish to exercise various TCPs under contention,
the current svn head of netperf (2.6) has options to switch
congestion control agorithms on the fly, as does iperf.
git-svn-id: svn://svn.openwrt.org/openwrt/trunk@32514 3c298f89-4303-0410-b956-a3cf2f4a3e73
The dcache bug that it works around is a generic issue, not a brcm47xx cache quirk
git-svn-id: svn://svn.openwrt.org/openwrt/trunk@32395 3c298f89-4303-0410-b956-a3cf2f4a3e73
