Erratum A004510 says that under certain load conditions, modified
cache lines can be discarded, causing data corruption.
To work around this, several CCSR and DCSR register updates need to be
made in a careful manner, so that there is no other transaction in
corenet when the update is made.
The update is made from a locked cacheline, with a delay before to flush
any previous activity, and a delay after to flush the CCSR/DCSR update.
We can't use a readback because that would be another corenet
transaction, which is not allowed.
We lock the subsequent cacheline to prevent it from being fetched while
we're executing the previous cacheline. It is filled with nops so that a
branch doesn't cause us to fetch another cacheline.
Ordinarily we are running in a cache-inhibited mapping at this point, so
we temporarily change that. We make it guarded so that we should never
see a speculative load, and we never do an explicit load. Thus, only the
I-cache should ever fill from this mapping, and we flush/unlock it
afterward. Thus we should avoid problems from any potential cache
aliasing between inhibited and non-inhibited mappings.
NOTE that if PAMU is used with this patch, it will need to use a
dedicated LAW as described in the erratum. This is the responsibility
of the OS that sets up PAMU.
Signed-off-by: Scott Wood <scottwood@freescale.com>
Signed-off-by: Andy Fleming <afleming@freescale.com>
PowerPC mandates SP to be 16 bytes aligned.
Furthermore, a stack frame is added, pointing to the reset vector
which may in the way when gdb is walking the stack because
the reset vector may not accessible depending on emulator settings.
Also use a temp register so gdb doesn't pick up intermediate values.
Signed-off-by: Joakim Tjernlund <Joakim.Tjernlund@transmode.se>
Acked-by: Kumar Gala <galak@kernel.crashing.org>
Signed-off-by: Andy Fleming <afleming@freescale.com>
Update NAND code base to ovecome e500 and e500v2's second limitation i.e. IVPR
+ IVOR15 should be valid fetchable OP code address.
As NAND SPL does not compile vector table so making sure IVOR + IVOR15 points to
any fetchable valid data
Signed-off-by: Radu Lazarescu <radu.lazarescu@freescale.com>
Signed-off-by: Marius Grigoras <marius.grigoras@freescale.com>
Signed-off-by: Prabhakar Kushwaha <prabhakar@freescale.com>
Debugging of e500 and e500v1 processer requires debug exception vecter (IVPR +
IVOR15) to have valid and fetchable OP code.
1) While executing in translated space (AS=1), whenever a debug exception is
generated, the MSR[DS/IS] gets cleared i.e. AS=0 and the processor tries to
fetch an instruction from the debug exception vector (IVPR + IVOR15); since now
we are in AS=0, the application needs to ensure the proper TLB configuration to
have (IVOR + IVOR15) accessible from AS=0 also.
Create a temporary TLB in AS0 to make sure debug exception verctor is
accessible on debug exception.
2) Just after relocation in DDR, Make sure IVPR + IVOR15 points to valid opcode
Signed-off-by: Radu Lazarescu <radu.lazarescu@freescale.com>
Signed-off-by: Marius Grigoras <marius.grigoras@freescale.com>
Signed-off-by: Prabhakar Kushwaha <prabhakar@freescale.com>
Debugging of e500 and e500v1 processer requires MSR[DE] bit to be set always.
Where MSR = Machine State register
Make sure of MSR[DE] bit is set uniformaly across the different execution
address space i.e. AS0 and AS1.
Signed-off-by: Radu Lazarescu <radu.lazarescu@freescale.com>
Signed-off-by: Catalin Udma <catalin.udma@freescale.com>
Signed-off-by: Marius Grigoras <marius.grigoras@freescale.com>
Signed-off-by: Prabhakar Kushwaha <prabhakar@freescale.com>
The CCSR relocation code in start.S writes to MAS7 on all e500 parts, but
that register does not exist on e500v1.
Signed-off-by: Timur Tabi <timur@freescale.com>
NAND SPL code never compile the vector table.
So no need to setup interrupt vector table for NAND SPL.
Signed-off-by: Prabhakar Kushwaha <prabhakar@freescale.com>
Signed-off-by: Andy Fleming <afleming@freescale.com>
After relocation of vector table in SDRAM's lower address, IVORs value should
be updated with new handler addresses.
As vector tables are relocated to 0x100,0x200... 0xf00 address in DDR.IVORs
are updated with 0x100, 0x200,....f00 hard-coded values.
Signed-off-by: Prabhakar Kushwaha <prabhakar@freescale.com>
Signed-off-by: Andy Fleming <afleming@freescale.com>
For e500 and e500v2 architecturees processor IVPR address should be alinged on
64K boundary.
in start.S, CONFIG_SYS_MONITOR_BASE is stored blindly in IVPR assuming it to be
64K aligned. It may not be true always. If it is not aligned, IVPR + IVORs may
not point to an exception handler.
Signed-off-by: Prabhakar Kushwaha <prabhakar@freescale.com>
Signed-off-by: Andy Fleming <afleming@freescale.com>
Erratum A-003999: Running Floating Point instructions requires special
initialization.
Impact:
Floating point arithmetic operations may result in an incorrect value.
Workaround:
Perform a read modify write to set bit 7 to a 1 in SPR 977 before
executing any floating point arithmetic operation. This bit can be set
when setting MSR[FP], and can be cleared when clearing MSR[FP].
Alternatively, the bit can be set once at boot time, and never cleared.
There will be no performance degradation due to setting this bit.
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
On some Freescale systems (e.g. those booted from the on-chip ROM), the
TLB that covers the boot page can also cover CCSR, which breaks the CCSR
relocation code. To fix this, we resize the boot page TLB so that it only
covers the 4KB boot page.
Signed-off-by: Timur Tabi <timur@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
Verify that CCSR is actually located where it is supposed to be before
we relocate it. This is useful in detecting U-Boot configurations that
are broken (e.g. an incorrect value for CONFIG_SYS_CCSRBAR_DEFAULT).
If the current value is wrong, we enter an infinite loop, which is handy
for debuggers.
Signed-off-by: Timur Tabi <timur@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
Calls to tlbwe and tlbsx should be preceded with an isync/msync pair.
Signed-off-by: Timur Tabi <timur@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
Pre u-boot Flow:
1. User loads the u-boot image in flash
2. PBL/Configuration word is used to create LAW for Flash at 0xc0000000
(Please note that ISBC expects all these addresses, images to be
validated, entry point etc within 0 - 3.5G range)
3. ISBC validates the u-boot image, and passes control to u-boot
at 0xcffffffc.
Changes in u-boot:
1. Temporarily map CONFIG_SYS_MONITOR_BASE to the 1M
CONFIG_SYS_PBI_FLASH_WINDOW in AS=1.
(The CONFIG_SYS_PBI_FLASH_WINDOW is the address map for the flash
created by PBL/configuration word within 0 - 3.5G memory range. The
u-boot image at this address has been validated by ISBC code)
2. Remove TLB entries for 0 - 3.5G created by ISBC code
3. Remove the LAW entry for the CONFIG_SYS_PBI_FLASH_WINDOW created by
PBL/configuration word after switch to AS = 1
Signed-off-by: Ruchika Gupta <ruchika.gupta@freescale.com>
Signed-off-by: Kuldip Giroh <kuldip.giroh@freescale.com>
Acked-by: Wood Scott-B07421 <B07421@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
Before main memory (DDR) is initialized, the on-chip L1 cache is used as a
memory area for the stack and the global data (gd_t) structure. This is
called the initial RAM area, or initram. The L1 cache is locked and the TLBs
point to a non-existent address (so that there's no chance it will overlap
main memory or any device). The L1 cache is also configured not to write
out to memory or the L2 cache, so everything stays in the L1 cache.
One of the things we might do while running out of initram is relocate CCSR.
On reset, CCSR is typically located at some high 32-bit address, like
0xfe000000, and this may not be the best place for CCSR. For example, on
36-bit systems, CCSR is relocated to 0xffe000000, near the top of 36-bit
memory space.
On some future Freescale SOCs, the L1 cache will be forced to write to the
backing store, so we can no longer have the TLBs point to non-existent address.
Instead, we will point the TLBs to an unused area in CCSR. In order for this
technique to work, CCSR needs to be relocated before the initram memory is
enabled.
Unlike the original CCSR relocation code in cpu_init_early_f(), the TLBs
we create now for relocating CCSR are deleted after the relocation is finished.
cpu_init_early_f() will still need to create a TLB for CCSR (at the new
location) for normal U-Boot purposes. This is done to keep the impact to
existing U-Boot code minimal and to better isolate the CCSR relocation code.
Signed-off-by: Timur Tabi <timur@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
This provides a function that will override the weak function
flush_icache to let 85xx boards to flush the icache
cc: Kumar Gala <kumar.gala@freescale.com>
Signed-off-by: Matthew McClintock <msm@freescale.com>
This patch removes the architecture specific implementation of
version_string where possible. Some architectures use a special place
and therefore we provide U_BOOT_VERSION_STRING definition and a common
weak symbol version_string.
Signed-off-by: Andreas Biemann <andreas.devel@googlemail.com>
CC: Mike Frysinger <vapier@gentoo.org>
CC: Peter Pan <pppeterpppan@gmail.com>
Acked-by: Mike Frysinger <vapier@gentoo.org>
-msingle-pic-base is a new gcc option for ppc and
it reduces the size of my u-boot with 6-8 KB.
While at it, add -fno-jump-tables too to save a
few more bytes.
-msingle-pic-base will be in gcc 4.6, however
backported patches are available at
http://bugs.gentoo.org/show_bug.cgi?id=347281
Signed-off-by: Joakim Tjernlund <Joakim.Tjernlund@transmode.se>
Currently, _end is used for end of BSS section. We want _end to mean
end of u-boot image, so we rename _end to __bss_end__ first.
Signed-off-by: Po-Yu Chuang <ratbert@faraday-tech.com>
We set the L1 dache register with a bogus register value. Need to be
using 'r3' instead of 'r0'.
Reported-by: John Traill <john.traill@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
The fixup procedure just stored a constant value in the
fixup table rather than just adjusting the table.
Although that doesn't seem to do any harm, it prevents
relocation more that once.
Signed-off-by: Joakim Tjernlund <Joakim.Tjernlund@transmode.se>
Use CONFIG_SYS_MONITOR_BASE instead of CONFIG_SYS_TEXT_BASE in early
init code so we can share the same code with NAND or NOR boot and not
have additional ifdefs in here.
Signed-off-by: Haiying Wang <Haiying.Wang@freescale.com>
Signed-off-by: Kumar Gala <galak@kernel.crashing.org>
CONFIG_SYS_GBL_DATA_SIZE has always been just a bad workarond for not
being able to use "sizeof(struct global_data)" in assembler files.
Recent experience has shown that manual synchronization is not
reliable enough. This patch renames CONFIG_SYS_GBL_DATA_SIZE into
GENERATED_GBL_DATA_SIZE which gets automatically generated by the
asm-offsets tool. In the result, all definitions of this value can be
deleted from the board config files. We have to make sure that all
files that reference such data include the new <asm-offsets.h> file.
No other changes have been done yet, but it is obvious that similar
changes / simplifications can be done for other, related macro
definitions as well.
Signed-off-by: Wolfgang Denk <wd@denx.de>
Acked-by: Kumar Gala <galak@kernel.crashing.org>
The fixup routine must not fixup NULL pointers.
Problem can be seen by
char *testfun(void) __attribute__((weak));
char *(*myfun)(void) = testfun;
Then add
printf("myfun:%p, &myfun:%p\n", myfun, &myfun);
before relocation and after relocation.
myfun should be NULL in both cases but it is not.
Signed-off-by: Joakim Tjernlund <Joakim.Tjernlund@transmode.se>
The change is currently needed to be able to remove the board
configuration scripting from the top level Makefile and replace it by
a simple, table driven script.
Moving this configuration setting into the "CONFIG_*" name space is
also desirable because it is needed if we ever should move forward to
a Kconfig driven configuration system.
Signed-off-by: Wolfgang Denk <wd@denx.de>
If 36-bit physical address is used, move the INIT_RAM_ADDR to higher
address. This frees the low 4GB address space for better use.
Signed-off-by: York Sun <yorksun@freescale.com>
As discussed on the list, move "arch/ppc" to "arch/powerpc" to
better match the Linux directory structure.
Please note that this patch also changes the "ppc" target in
MAKEALL to "powerpc" to match this new infrastructure. But "ppc"
is kept as an alias for now, to not break compatibility with
scripts using this name.
Signed-off-by: Stefan Roese <sr@denx.de>
Acked-by: Wolfgang Denk <wd@denx.de>
Acked-by: Detlev Zundel <dzu@denx.de>
Acked-by: Kim Phillips <kim.phillips@freescale.com>
Cc: Peter Tyser <ptyser@xes-inc.com>
Cc: Anatolij Gustschin <agust@denx.de>
