The LLVM Target parser currently does not allow to specify the security
extension as part of -march (see also LLVM Bug 40186 [0]). When trying
to use Clang with LLVM's integrated assembler, this leads to build
errors such as this:
clang-8: error: the clang compiler does not support '-Wa,-march=armv7-a+sec'
Use ".arch_extension sec" to enable the security extension in a more
portable fasion. Also make sure to use ".arch armv7-a" in case a v6/v7
multi-platform kernel is being built.
Note that this is technically not exactly the same as the old code
checked for availabilty of the security extension by calling as-instr.
However, there are already other sites which use ".arch_extension sec"
unconditionally, hence de-facto we need an assembler capable of
".arch_extension sec" already today (arch/arm/mm/proc-v7.S). The
arch extension "sec" is available since binutils 2.21 according to
its documentation [1].
[0] https://bugs.llvm.org/show_bug.cgi?id=40186
[1] https://sourceware.org/binutils/docs-2.21/as/ARM-Options.html
Signed-off-by: Stefan Agner <stefan@agner.ch>
Acked-by: Mans Rullgard <mans@mansr.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Krzysztof Kozlowski <krzk@kernel.org>
Signed-off-by: Olof Johansson <olof@lixom.net>
Init data marked const should be annotated with __initconst for
correctness and not __initdata. In some cases the array gathering
references to that data has to be marked const as well. This fixes
LTO builds that otherwise fail with section mismatch errors.
Signed-off-by: Nicolas Pitre <nico@linaro.org>
Signed-off-by: Olof Johansson <olof@lixom.net>
The block of comments in bcm_kona_do_smc() are somewhat confusing.
This patch attempts to clarify what's going on.
Signed-off-by: Alex Elder <elder@linaro.org>
Reviewed-by: Tim Kryger <tim.kryger@linaro.org>
Reviewed-by: Markus Mayer <markus.mayer@linaro.org>
Reviewed-by: Matt Porter <mporter@linaro.org>
Signed-off-by: Matt Porter <mporter@linaro.org>
Move the code that implements the "smc" call into a C function that
uses inline assembly. This allows us to make that function private,
and enables us to get rid of "arch/arm/mach-bcm/bcm_kona_smc_asm.S".
Rename what had been the "buffer_addr" argument to be "buffer_phys"
so it's consistent with other usage in this file.
Since it's now easy to do, verify that r12 contains SEC_EXIT_NORMAL
upon completion of the SMC. There really isn't a good way to handle
the abnormal completion of a secure monitor request.
Since "bcm_kona_smc.h" is now only included from C files, eliminate
the #ifndef __ASSEMBLY__.
Signed-off-by: Alex Elder <elder@linaro.org>
Reviewed-by: Tim Kryger <tim.kryger@linaro.org>
Reviewed-by: Markus Mayer <markus.mayer@linaro.org>
Reviewed-by: Matt Porter <mporter@linaro.org>
Signed-off-by: Matt Porter <mporter@linaro.org>
There's logic in bcm_kona_smc() to ensure __bcm_kona_smc() gets
called on CPU 0; if already executing on CPU 0, that function is
called directly. The direct call is not protected from interrupts,
however, which is not safe.
Note that smp_call_function_single() is designed to handle the case
where the target cpu is the current one. It also gets a reference
to the CPU and disables IRQs across the call.
So we can simplify things and at the same time be protected against
interrupts by calling smp_call_function_single() unconditionally.
Signed-off-by: Alex Elder <elder@linaro.org>
Reviewed-by: Tim Kryger <tim.kryger@linaro.org>
Reviewed-by: Markus Mayer <markus.mayer@linaro.org>
Reviewed-by: Matt Porter <mporter@linaro.org>
Signed-off-by: Matt Porter <mporter@linaro.org>
Currently it is assumed that SEC_ROM_RET_OK is the only valid "good"
result of a secure monitor request. However the values that can be
returned by a secure monitor request are dependent on which service
id was provided.
We therefore should handle the result in a request-dependent way.
The most natural way to do that is to have the initiator of the
request--where bcm_kona_smc() is called--handle the result in a way
appropriate to the request.
An "smc" operation must be performed only on core 0, while the
request can be initiated from any core. To pass back the request
result, we add a new field to the bcm_kona_smc_data structure, and
have bcm_kona_smc() return that value rather than 0.
There's only one caller right now. Move the existing check of the
result out of __bcm_kona_smc() and into the kona_l2_cache_init()
where the SSAPI_ENABLE_L2_CACHE request is initiated.
Signed-off-by: Alex Elder <elder@linaro.org>
Reviewed-by: Tim Kryger <tim.kryger@linaro.org>
Reviewed-by: Markus Mayer <markus.mayer@linaro.org>
Reviewed-by: Matt Porter <mporter@linaro.org>
Signed-off-by: Matt Porter <mporter@linaro.org>
This patch just does some simple cleanup in "bcm_kona_smc.c":
- Get rid of the secure_bridge_data structure. Instead, just
define two globals that record the physical and virtual
addresses of the SMC arguments buffer. Use "buffer" instead
of "bounce" in their names. Drop of the erroneous __iomem
annotation for the physical address.
- Get rid of the initialized flag and just use a non-null buffer
address to indicate that.
- Get the size of the memory region when fetching the SMC
arguments buffer location from the device tree. Use it to
call ioremap() directly rather than requiring of_iomap() to
go look it up again.
- Do some additional validation on that memory region size.
- Flush caches unconditionally in __bcm_kona_smc(); nothing
supplies SSAPI_BRCM_START_VC_CORE as a service id.
- Drop a needless initialization of "rc" in __bcm_kona_smc().
It also deletes most of the content of "bcm_kona_smc.h" because it's
never actually used and is of questionable value anyway.
Signed-off-by: Alex Elder <elder@linaro.org>
Reviewed-by: Tim Kryger <tim.kryger@linaro.org>
Reviewed-by: Markus Mayer <markus.mayer@linaro.org>
Reviewed-by: Matt Porter <mporter@linaro.org>
Signed-off-by: Matt Porter <mporter@linaro.org>
Several conditions in bcm_kona_smc_init() are handled with BUG_ON().
That function is capable of returning an error, so do that instead.
Also, don't assume of_get_address() returns a valid pointer.
Signed-off-by: Alex Elder <elder@linaro.org>
Reviewed-by: Tim Kryger <tim.kryger@linaro.org>
Reviewed-by: Markus Mayer <markus.mayer@linaro.org>
Reviewed-by: Matt Porter <mporter@linaro.org>
Signed-off-by: Matt Porter <mporter@linaro.org>
The pointer used to pass parameters to an "smc" call is produced
through a call to ioremap(). As such, we should be using functions
like writel() to access it.
Signed-off-by: Alex Elder <elder@linaro.org>
Reviewed-by: Tim Kryger <tim.kryger@linaro.org>
Reviewed-by: Markus Mayer <markus.mayer@linaro.org>
Reviewed-by: Matt Porter <mporter@linaro.org>
Signed-off-by: Matt Porter <mporter@linaro.org>
The current bcm_kona_smc_init function throws a BUG_ON if there's no SMC
device node defined in the device tree.
Since secure API access is optional depending the chip configuration,
fix this by allowing the rest of the code to run even if there's no SMC
device node defined
Signed-off-by: Christian Daudt <csd@broadcom.com>
Reviewed-by: Markus Mayer <markus.mayer@linaro.org>
[ this is a follow-up to this discussion:
http://archive.arm.linux.org.uk/lurker/message/20130730.230827.a1ceb12a.en.html ]
This patchset renames all uses of "bcm," name bindings to
"brcm," as they were done prior to knowing that brcm had
already been standardized as Broadcom vendor prefix
(in Documentation/devicetree/bindings/vendor-prefixes.txt).
This will not cause any churn on devices because none of
these bindings have made it into production yet.
Signed-off-by: Christian Daudt <csd@broadcom.com>
Acked-by: Stephen Warren <swarren@nvidia.com>
The bcm_kona_smc_init function references the bcm_kona_smc_ids variable
that is marked __initconst, so the function itself has to be __init
to avoid this build error:
WARNING: arch/arm/mach-bcm/built-in.o(.text+0x12c): Section mismatch in reference from the function bcm_kona_smc_init() to the (unknown reference) .init.rodata:(unknown)
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Cc: Christian Daudt <csd@broadcom.com>
- Adds a module to provide calls into secure monitor mode
- Uses this module to make secure monitor calls to enable L2 cache.
Updates from V1:
- Split DT portion into separate patch
- replace #ifdef-ed L2 code with "if".
- move init call to board init
Signed-off-by: Christian Daudt <csd@broadcom.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Ben Dooks