Fix following compilation issue when SYS_DCACHE_OFF is enable:
drivers/misc/scmi_agent.c:128: undefined reference to `mmu_set_region_dcache_behaviour'
when SYS_DCACHE_OFF is enable, mmu_set_region_dcache_behaviour() must be
defined.
Signed-off-by: Patrice Chotard <patrice.chotard@st.com>
Signed-off-by: Patrice Chotard <patrice.chotard@foss.st.com>
Reviewed-by: Patrick Delaunay <patrick.delaunay@foss.st.com>
Update the initial value of Domain Access Control Register (DACR)
and set by default the access permission to client (DACR_Dn_CLIENT = 1U)
for each of the 16 domains and no more to all-supervisor
(DACR_Dn_MANAGER = 3U).
This patch allows to activate the domain checking in MMU against the
permission bits in the translation tables and avoids prefetching issue
on ARMv7 [1].
Today it was already done for OMAP2 architecture
./arch/arm/mach-omap2/omap-cache.c::arm_init_domains
introduced by commit de63ac278c ("ARM: mmu: Set domain permissions
to client access") which fixes lot of speculative prefetch aborts seen
on OMAP5 secure devices.
[1] https://developer.arm.com/documentation/ddi0406/b/System-Level-Architecture/Virtual-Memory-System-Architecture--VMSA-/Memory-access-control/The-Execute-Never--XN--attribute-and-instruction-prefetching
Signed-off-by: Patrick Delaunay <patrick.delaunay@st.com>
Reported-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Patrick Delaunay <patrick.delaunay@foss.st.com>
Move this out of the common header and include it only where needed. In
a number of cases this requires adding "struct udevice;" to avoid adding
another large header or in other cases replacing / adding missing header
files that had been pulled in, very indirectly. Finally, we have a few
cases where we did not need to include <asm/global_data.h> at all, so
remove that include.
Signed-off-by: Simon Glass <sjg@chromium.org>
Signed-off-by: Tom Rini <trini@konsulko.com>
The Linux coding style guide (Documentation/process/coding-style.rst)
clearly says:
It's a **mistake** to use typedef for structures and pointers.
Besides, using typedef for structures is annoying when you try to make
headers self-contained.
Let's say you have the following function declaration in a header:
void foo(bd_t *bd);
This is not self-contained since bd_t is not defined.
To tell the compiler what 'bd_t' is, you need to include <asm/u-boot.h>
#include <asm/u-boot.h>
void foo(bd_t *bd);
Then, the include direcective pulls in more bloat needlessly.
If you use 'struct bd_info' instead, it is enough to put a forward
declaration as follows:
struct bd_info;
void foo(struct bd_info *bd);
Right, typedef'ing bd_t is a mistake.
I used coccinelle to generate this commit.
The semantic patch that makes this change is as follows:
<smpl>
@@
typedef bd_t;
@@
-bd_t
+struct bd_info
</smpl>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Provide function for setting arbitrary virtual-physical MMU mapping
and cache settings for the given region.
Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
Reviewed-by: Tom Rini <trini@konsulko.com>
Solved the overflow on phys_addr_t type for start + size in
mmu_set_region_dcache_behaviour() function.
This overflow is avoided by dividing start and end by 2 before addition,
and we only expecting that start and size are even.
This patch doesn't change the current function behavior if the
parameters (start or size) are not aligned on MMU_SECTION_SIZE.
For example, this overflow occurs on ARM32 with:
start = 0xC0000000 and size = 0x40000000
then start + size = 0x100000000 and end = 0x0.
For information the function behavior change with risk of regression,
if we just shift start and size before the addition.
Example with 2MB section size:
MMU_SECTION_SIZE 0x200000 and MMU_SECTION_SHIFT = 21
with start = 0x1000000, size = 0x1000000,
- with the proposed patch, start = 0 and end = 0x1 as previously
- with the more simple patch:
end = (start >> MMU_SECTION_SHIFT) + (size >> MMU_SECTION_SHIFT)
the value of end change:
start >> 21 = 0, size >> 21 = 0 and end = 0x0 !!!
Signed-off-by: Patrick Delaunay <patrick.delaunay@st.com>
Add the new flags DCACHE_DEFAULT_OPTION to define the default
option to use according the compilation flags
CONFIG_SYS_ARM_CACHE_*.
This new compilation flag allows to simplify dram_bank_mmu_setup()
and can be used as third parameter (option=dcache option to select)
of mmu_set_region_dcache_behaviour function.
Signed-off-by: Patrick Delaunay <patrick.delaunay@st.com>
Add protection in dram_bank_mmu_setup() to avoid access to bd->bi_dram
before relocation.
This patch allow to use the generic weak function dram_bank_mmu_setup
to activate the MMU and the data cache in SPL or in U-Boot before
relocation, when bd->bi_dram is not yet initialized.
In this cases, the MMU must be initialized explicitly with
mmu_set_region_dcache_behaviour function.
Signed-off-by: Patrick Delaunay <patrick.delaunay@st.com>
These functions belong in cpu_func.h. Another option would be cache.h
but that code uses driver model and we have not moved these cache
functions to use driver model. Since they are CPU-related it seems
reasonable to put them here.
Move them over.
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Tom Rini <trini@konsulko.com>
As part of disabling caches MMU as well gets disabled. But MMU is not
available on all armv7 cores like R5F. So disable MMU only if it is
available.
Signed-off-by: Lokesh Vutla <lokeshvutla@ti.com>
While converting CONFIG_SYS_[DI]CACHE_OFF to Kconfig, there are instances
where these configuration items are conditional on SPL. This commit adds SPL
variants of these configuration items, uses CONFIG_IS_ENABLED(), and updates
the configurations as required.
Acked-by: Alexey Brodkin <abrodkin@synopsys.com>
Signed-off-by: Trevor Woerner <trevor@toganlabs.com>
[trini: Make the default depend on the setting for full U-Boot, update
more zynq hardware]
Signed-off-by: Tom Rini <trini@konsulko.com>
Cache maintenance procedure is same for v7A and v7R
processors. So re-use cache-cp15.c file except for
mmu parts.
Tested-by: Michal Simek <michal.simek@xilinx.com>
Signed-off-by: Lokesh Vutla <lokeshvutla@ti.com>
Reviewed-by: Tom Rini <trini@konsulko.com>
Currently CPU_V7 kconfig symbol supports only ARMv7A architectures under
armv7 folder. This led to a misconception of creating separate folders
for armv7m and armv7r. There is no reason to create separate folder for
other armv7 based architectures when it can co-exist with few Kconfig
symbols.
As a first step towards a common folder, rename CPU_V7 as CPUV7A. Later
separate Kconfig symbols can be added for CPU_V7R and CPU_V7M and
can co exist in the same folder.
Reviewed-by: Tom Rini <trini@konsulko.com>
Tested-by: Michal Simek <michal.simek@xilinx.com>
Suggested-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Lokesh Vutla <lokeshvutla@ti.com>
When U-Boot started using SPDX tags we were among the early adopters and
there weren't a lot of other examples to borrow from. So we picked the
area of the file that usually had a full license text and replaced it
with an appropriate SPDX-License-Identifier: entry. Since then, the
Linux Kernel has adopted SPDX tags and they place it as the very first
line in a file (except where shebangs are used, then it's second line)
and with slightly different comment styles than us.
In part due to community overlap, in part due to better tag visibility
and in part for other minor reasons, switch over to that style.
This commit changes all instances where we have a single declared
license in the tag as both the before and after are identical in tag
contents. There's also a few places where I found we did not have a tag
and have introduced one.
Signed-off-by: Tom Rini <trini@konsulko.com>
The cp_delay() function was introduced because of a missing 'volatile'
attribute to the 'asm' statement in get_cr() which led to the 'mrc'
instruction in get_cr() being optimised out eventually.
This has been fixed in commit 53fd4b8c22 ("arm: mmu: Add missing volatile for reading SCTLR register")
but the bogus cp_delay() function which was introduced as a workaround
for the malfunctioning get_cr() was never removed.
Signed-off-by: Lothar Waßmann <LW@KARO-electronics.de>
If CONFIG_ARMV7_LPAE is not defined we should make sure that the feature
is disabled. This can happen if U-Boot is chain-loaded from another boot
loader which does enable LPAE.
Signed-off-by: Simon Glass <sjg@chromium.org>
At present if CONFIG_ARMV7_LPAE is defined then mmu_setup() will use
instructions which are invalid on ARMv4T. This happens on Tegra since it
has an ARMv4T boot CPU. Add a check for the architecture version to allow
the code to be built. It will not actually be executed by the boot CPU,
but needs to compile.
Signed-off-by: Simon Glass <sjg@chromium.org>
While we setup the mmu initially we mark set_section_dcache with
DCACHE_OFF flag. In case of non-LPAE mode the DCACHE_OFF macro
is rightly defined with TTB_SECT_XN_MASK set so as to mark all the
4GB XN. In case of LPAE mode XN(Execute-never) bit is not set with
DCACHE_OFF. Hence XN bit is not set by default for DCACHE_OFF which
keeps all the regions execute okay and this leads to random speculative
fetches in random memory regions which was eventually caught by kernel
omap-l3-noc driver.
Fix this to mark the regions as XN by default.
Signed-off-by: Keerthy <j-keerthy@ti.com>
Reviewed-by: Alexander Graf <agraf@suse.de>
Reviewed-by: Tom Rini <trini@konsulko.com>
The page table is maintained by the CPU, hence it is safe to always
align cache flush to a whole cache line size. This allows to use
mmu_page_table_flush for a single page table, e.g. when configure
only small regions through mmu_set_region_dcache_behaviour.
Signed-off-by: Stefan Agner <stefan.agner@toradex.com>
Tested-by: Fabio Estevam <fabio.estevam@nxp.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Heiko Schocher <hs@denx.de>
Add LPAE support for mmu_set_region_dcache_behaviour. The function
is in use in some LPAE capable board such TI DRA7xx or NXP i.MX 7.
Signed-off-by: Stefan Agner <stefan.agner@toradex.com>
We currently always modify the SVC versions of registers and only support
the short descriptor PTE format.
Some boards however (like the RPi2) run in HYP mode. There, we need to modify
the HYP version of system registers and HYP mode only supports the long
descriptor PTE format.
So this patch introduces support for both long descriptor PTEs and HYP mode
registers.
Signed-off-by: Alexander Graf <agraf@suse.de>
The arch/arm/lib/cache-cp15.c checks for CONFIG_ARMV7 and if this macro is
set, it configures TTBR0 register. This register must be configured for the
cache on ARMv7 to operate correctly.
The problem is that noone actually sets the CONFIG_ARMV7 macro and thus the
TTBR0 is not configured at all. On SoCFPGA, this produces all sorts of minor
issues which are hard to replicate, for example certain USB sticks are not
detected or QSPI NOR sometimes fails to write pages completely.
The solution is to replace CONFIG_ARMV7 test with CONFIG_CPU_V7 one. This is
correct because the code which added the test(s) for CONFIG_ARMV7 was added
shortly after CONFIG_ARMV7 was replaced by CONFIG_CPU_V7 and this code was
not adjusted correctly to reflect that change.
Signed-off-by: Marek Vasut <marex@denx.de>
Cc: Tom Rini <trini@konsulko.com>
Cc: Albert Aribaud <albert.u.boot@aribaud.net>
Cc: Simon Glass <sjg@chromium.org>
The TTBR0 register and Table Descriptors of the ARMv7 TLB weren't being
properly set to allow for the configuration specified caching modes to
be active over DRAM. This commit fixes those issues.
Signed-off-by: Bryan Brinsko <bryan.brinsko@rockwellcollins.com>
size_t is the canonical type to represent variables that contain a size.
Use it instead of signed integer. Physical addresses can be larger than
32-bit, so use a more appropriate type for them as well. phys_addr_t is
a type that is 32-bit on systems that use 32-bit addresses and 64-bit if
the system is 64-bit or uses a form of physical address extension to use
a larger address space on 32-bit systems. Using these types the same API
can be implemented on a wider range of systems.
Signed-off-by: Thierry Reding <treding@nvidia.com>
Add configuration for the write-allocate mode of L1 D-Cache on ARM.
This is needed for D-Cache operation on Cortex-A9 on the SoCFPGA .
Signed-off-by: Marek Vasut <marex@denx.de>
Cc: Chin Liang See <clsee@altera.com>
Cc: Dinh Nguyen <dinguyen@altera.com>
Cc: Albert Aribaud <albert.u.boot@aribaud.net>
Cc: Tom Rini <trini@ti.com>
Cc: Wolfgang Denk <wd@denx.de>
Cc: Pavel Machek <pavel@denx.de>
Acked-by: Pavel Machek <pavel@denx.de>
The patch fixes a corner case where adding size to DRAM start resulted
in a value (1 << 32), which in turn overflew the u32 computation, which
resulted in 0 and it therefore prevented correct setup of the MMU tables.
The addition of DRAM bank start and it's size can end up right at the end
of the address space in the special case of a machine with enough memory.
To prevent this overflow, shift the start and size separately and add them
only after they were shifted.
Hopefully, we only have systems in tree which have DRAM size aligned to
1MiB boundary. If not, this patch would break such systems. On the other
hand, such system would be broken by design anyway.
Signed-off-by: Marek Vasut <marex@denx.de>
Cc: Albert ARIBAUD <albert.u.boot@aribaud.net>
This is not only more readable but also prevents a warning
about a missing prototype. The prototypes which are actually
missing are added.
cc: Albert Aribaud <albert.u.boot@aribaud.net>
Signed-off-by: Jeroen Hofstee <jeroen@myspectrum.nl>
Reviewed-by: Tom Rini <trini@ti.com>
The 'XN' execute never bit is set in the pagetables. This will
prevent speculative prefetches to non executable regions. But the
domain permissions are set as master in the DACR register.
So the pagetable attribute for 'XN' is not effective. Change the
permissions to client.
This fixes lot of speculative prefetch aborts seen on OMAP5
secure devices.
Signed-off-by: R Sricharan <r.sricharan@ti.com>
Tested-by: Vincent Stehle <v-stehle@ti.com>
Cc: Vincent Stehle <v-stehle@ti.com>
Cc: Tom Rini <trini@ti.com>
Cc: Albert ARIBAUD <albert.u.boot@aribaud.net>
Introduce a weak version of dram_bank_setup function
to allow a platform specific function.
This is used in the subsequent patch to setup dram region
without 'XN' attribute in order to enable the region
under client permissions.
Signed-off-by: R Sricharan <r.sricharan@ti.com>
Cc: Vincent Stehle <v-stehle@ti.com>
Cc: Tom Rini <trini@ti.com>
Cc: Albert ARIBAUD <albert.u.boot@aribaud.net>
Move these fields into arch_global_data and tidy up.
Signed-off-by: Simon Glass <sjg@chromium.org>
[trini: Address tlb_size in this patch as well]
Signed-off-by: Tom Rini <trini@ti.com>
In Cortex-A15 architecture, when we run cache invalidate
the cache clean operation executes automatically.
So if there are any dirty cache lines before disabling the L2 cache
these will be synchronized with the main memory when
invalidate_dcache_all() runs in the last part of U-boot
The two functions after flush_dcache_all is using the stack. So this
data will be on the cache. After disable when invalidate is called the
data will be flushed from cache to memory. This corrupts the stack in
invalida_dcache_all. So this change is required to avoid the u-boot
hang.
So flush has to be done just before clearing CR_C bit
Signed-off-by: Arun Mankuzhi <arun.m@samsung.com>
Signed-off-by: Simon Glass <sjg@chromium.org>
Add support for adjusting the L1 cache behavior by updating the MMU
configuration. The mmu_set_region_dcache_behaviour() function allows
drivers to make these changes after the MMU is set up.
It is implemented only for ARMv7 at present.
This is needed for LCD support, where we want to make the LCD frame buffer
write-through (or off) rather than write-back.
Signed-off-by: Simon Glass <sjg@chromium.org>
Signed-off-by: Tom Warren <twarren@nvidia.com>
The following is the cleanup sequence in arch/arm/cpu/armv7/cpu.c
int cleanup_before_linux(void)
{
...
...
dcache_disable();
v7_outer_cache_disable();
invalidate_dcache_all();
}
1) invalidate_dcache_all call expects that all the caches has been
flushed, invalidated and there are no dirty entries prior to its
execution. In the above sequence dcache_disable() flushes, invalidates
the caches and turns off the mmu. But after it cleanups the cache
and before the mmu is disabled there is a cp_delay() function which
has STR instruction. On certain cores like the cortex-a15, cache hit
and a write can happen to a cache line even when the dcache is
disabled. So the above mentioned STR instruction creates a dirty entry
after cleaning. The mmu gets disabled after this.
2) invalidate_dcache_all invalidates the cache lines. Again on
cores like cortex-a15, invalidate instruction flushes the dirty
line as well. So some times the dirty line from sequence 1
can corrupt the memory resulting in a crash.
Fixing this by moving the get_cr() and cp_delay() calls before
cleaning up the cache, thus avoiding the dirty entry.
Signed-off-by: R Sricharan <r.sricharan@ti.com>
1. make sure that page table setup is not done multiple times
2. flush_dcache_all() is more appropriate while disabling cache
than a range flush on the entire memory(flush_cache())
Provide a default implementation for flush_dcache_all()
for backward compatibility and to avoid build issues.
Signed-off-by: Aneesh V <aneesh@ti.com>
- Enable I-cache on bootup
- Enable MMU and D-cache immediately after relocation
- Do necessary initialization before enabling d-cache and MMU
- Changes to cleanup_before_linux()
- Make changes according to the new framework
Signed-off-by: Aneesh V <aneesh@ti.com>
Replace the cache related CONFIG flags with more meaningful
names. Following are the changes:
CONFIG_L2_OFF -> CONFIG_SYS_L2CACHE_OFF
CONFIG_SYS_NO_ICACHE -> CONFIG_SYS_ICACHE_OFF
CONFIG_SYS_NO_DCACHE -> CONFIG_SYS_DCACHE_OFF
Signed-off-by: Aneesh V <aneesh@ti.com>
V2:
* Changed CONFIG_L2_OFF -> CONFIG_SYS_NO_L2CACHE
V4:
* Changed all three flags to the final names suggested as above
and accordingly changed the commit message
When this define was introduced, the idea was to provide a soft
migration path for ARM boards to get adapted to the new relocation
support. However, other recent changes led to a different
implementation (ELF relocation), where this no longer works. By now
CONFIG_SYS_ARM_WITHOUT_RELOC does not only not help any more, but it
actually hurts because it obfuscates the actual code by sprinkling it
with lots of dead and non-working debris.
So let's make a clean cut and drop CONFIG_SYS_ARM_WITHOUT_RELOC.
Signed-off-by: Wolfgang Denk <wd@denx.de>
Tested-by: Heiko Schocher <hs@denx.de>
Tested-by: Reinhard Meyer <u-boot@emk-elektronik.de>
!! This breaks support for all arm boards !!
To compile in old style, you must define
CONFIG_SYS_ARM_WITHOUT_RELOC or you can compile
with "CONFIG_SYS_ARM_WITHOUT_RELOC=1 ./MAKEALL board"
!! This define will be removed soon, so convert your
board to use relocation support
Portions of this work were supported by funding from
the CE Linux Forum.
Signed-off-by: Heiko Schocher <hs@denx.de>
Fix boot from NAND for non-ARM systems
Signed-off-by: Wolfgang Denk <wd@denx.de>
This has been tested on at91sam9263 and STN8815.
Again, I didn't check if it has bad effects
on non-arm926 cores.
Initially I had a "done" bit to only set up page tables
at the beginning. However, since the aligmnent requirement
was for the whole object file, this extra integer tool 16kB
in BSS, so I chose to remove it.
Also, note not all boards use PHYS_SDRAM, but it looks like
it's the most used name (more than CONFIG_SYS_DRAM_BASE for
example).
Portions of this work were supported by funding from
the CE Linux Forum.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>
Signed-off-by: Heiko Schocher <hs@denx.de>
Also move lib_$ARCH/config.mk to arch/$ARCH/config.mk
This change is intended to clean up the top-level directory structure
and more closely mimic Linux's directory organization.
Signed-off-by: Peter Tyser <ptyser@xes-inc.com>
