This fixes some problems with the 6345 support and adds a macro for CPU
identification that is easier on the eyes. The first thing it does is to not
initialize MPI on the 6345 as it does not have PCI. The second thing it does is
to use a static value for the CPU frequency of the 6345 chip to provide an
accurate timer.
Signed-off-by: Axel Gembe <ago@bastart.eu.org>
git-svn-id: svn://svn.openwrt.org/openwrt/trunk@11183 3c298f89-4303-0410-b956-a3cf2f4a3e73
The CFE detection failed to account for zero termination.
Signed-off-by: Axel Gembe <ago@bastart.eu.org>
git-svn-id: svn://svn.openwrt.org/openwrt/trunk@11182 3c298f89-4303-0410-b956-a3cf2f4a3e73
This basically selects the new generic MIPS timer code for BCM963xx and
simplifies the timer setup code.
Signed-off-by: Axel Gembe <ago@bastart.eu.org>
git-svn-id: svn://svn.openwrt.org/openwrt/trunk@11181 3c298f89-4303-0410-b956-a3cf2f4a3e73
This removes some code that has been deprecated in kernels >= 2.6.24.
Signed-off-by: Axel Gembe <ago@bastart.eu.org>
git-svn-id: svn://svn.openwrt.org/openwrt/trunk@11180 3c298f89-4303-0410-b956-a3cf2f4a3e73
The ISR ended up in an endless loop because the TX ISR never got used or masked.
This patch basically makes the TX ISR mask the the TX interrupt when it
encounters it, because it doesn't even use the TX interrupt.
Signed-off-by: Axel Gembe <ago@bastart.eu.org>
git-svn-id: svn://svn.openwrt.org/openwrt/trunk@11179 3c298f89-4303-0410-b956-a3cf2f4a3e73
This patch adds interrupt handling as on AR7. The old code was very messy and
didn't work too well. It also removes the unused file int-handler.S.
Signed-off-by: Axel Gembe <ago@bastart.eu.org>
git-svn-id: svn://svn.openwrt.org/openwrt/trunk@11178 3c298f89-4303-0410-b956-a3cf2f4a3e73
The load address for BCM963xx is 0x80010000, not 0xf8001000 as in the current
sources. I think this is just a typo, so this patch fixes it (tested on 96345).
Signed-off-by: Axel Gembe <ago@bastart.eu.org>
git-svn-id: svn://svn.openwrt.org/openwrt/trunk@11177 3c298f89-4303-0410-b956-a3cf2f4a3e73
This adds CFE tagged image generation in the BCM963xx target image Makefile.
It has to generate its own LZMA compressed kernel because CFE is a LZMA nazi.
It also seems that the Broadcom image tagger as well as CFE use AdvanceCOMP,
which is based on a slightly older LZMA SDK. Anyways, some of the code is
GPL, some of it is LGPL, so it might be that Broadcom owes some sources.
Also, LZMA has a bug which causes it to generate different output when you
are using stdin / stdout compared to plain files. I've just worked around
the issue by using plain files.
Signed-off-by: Axel Gembe <ago@bastart.eu.org>
git-svn-id: svn://svn.openwrt.org/openwrt/trunk@11171 3c298f89-4303-0410-b956-a3cf2f4a3e73
This replaces the proprietary bcmImageBuilder program from the Broadcom source
drops. It basically adds a 256 bytes header in front of the kernel + rootfs
which contains a few text signatures, the locations of the data as well as the
checksums of the data and the tag. It also adds a 12 bytes header in front of
the LZMA kernel which contains the load address, kernel entry and the size of
the compressed LZMA data.
Signed-off-by: Axel Gembe <ago@bastart.eu.org>
git-svn-id: svn://svn.openwrt.org/openwrt/trunk@11170 3c298f89-4303-0410-b956-a3cf2f4a3e73
