We really shouldn't be overwriting bat registers with translation enabled,
especially when we're executing code using one of them for translating
the current instruction stream. Instead, disable address translation
while doing the final BAT setup.
In order to do this, setup_bats has to move back to asm code, because we
require translation to be enabled to have a stack for C code. The yucky
thing about that is that the assembler doesn't like ULL so we have to
switch to using HIGH/LOW pairs for physical addresses that are > 32 bits
in length.
Signed-off-by: Becky Bruce <beckyb@kernel.crashing.org>
Acked-by: York Sun <yorksun@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
After determining how much DDR is actually in the system, set DBAT0 and
IBAT0 accordingly. This ensures that the CPU won't attempt to access
(via speculation) addresses outside of actual memory.
On 86xx systems, DBAT0 and IBAT0 (the BATs for DDR) are initialized to 2GB
and kept that way. If the system has less than 2GB of memory (typical for
an MPC8610 HPCD), the CPU may attempt to access this memory during
speculation. The zlib code is notorious for generating such memory reads,
and indeed on the MPC8610, uncompressing the Linux kernel causes a machine
check (without this patch).
Currently we are limited to power of two sized DDR since we only use a
single bat. If a non-power of two size is used that is less than
CONFIG_MAX_MEM_MAPPED u-boot will crash.
Signed-off-by: Timur Tabi <timur@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
The BAT fields are architected; there's no need for these to be in
cpu-specific files. Drop the duplication and move these to
include/asm-ppc/mmu.h. Also, remove the BL_xxx defines that were only
used by the alaska board, and switch to using the BATU_BL_xxx defines
used by all the other boards. The BL_ defines previously in use
had to be shifted into the proper position for use, which was inefficient.
Signed-off-by: Becky Bruce <beckyb@kernel.crashing.org>
Currently MPC85xx and MPC86xx boards just calculate the localbus frequency
and print it out, but don't save it.
This changes where its calculated and stored to be more consistent with the
CPU, CCB, TB, and DDR frequencies and the MPC83xx localbus clock.
The localbus frequency is added to sysinfo and calculated when sysinfo is
set up, in cpu/mpc8[56]xx/speed.c, the same as the other frequencies are.
get_clocks() copies the frequency into the global data, as the other
frequencies are, into a new field that is only enabled for MPC85xx and
MPC86xx.
checkcpu() in cpu/mpc8[56]xx/cpu.c will print out the local bus frequency
from sysinfo, like the other frequencies, instead of calculating it on the
spot.
Signed-off-by: Trent Piepho <tpiepho@freescale.com>
Acked-by: Kumar Gala <galak@kernel.crashing.org>
Acked-by: Jon Loeliger <jdl@freescale.com>
This patch moves Freescale Localbus defines out of mpc83xx.h, so we could
use it on MPC85xx and MPC86xx processors.
Signed-off-by: Anton Vorontsov <avorontsov@ru.mvista.com>
Acked-by: Andy Fleming <afleming@freescale.com>
Signed-off-by: Kim Phillips <kim.phillips@freescale.com>
