GitLab

Xilinx is introducing Versal, an adaptive compute acceleration platform
(ACAP), built on 7nm FinFET process technology. Versal ACAPs combine Scalar
Processing Engines, Adaptable Hardware Engines, and Intelligent Engines with
leading-edge memory and interfacing technologies to deliver powerful
heterogeneous acceleration for any application. The Versal AI Core series has
five devices, offering 128 to 400 AI Engines. The series includes dual-core Arm
Cortex-A72 application processors, dual-core Arm Cortex-R5 real-time
processors, 256KB of on-chip memory with ECC, more than 1,900 DSP engines
optimized for high-precision floating point with low latency.

This patch adds Virtual QEMU platform support for
this SoC "versal_virt".
Signed-off-by: Siva Durga Prasad Paladugu <siva.durga.paladugu@xilinx.com>
Signed-off-by: Michal Simek <michal.simek@xilinx.com>

Add supprot for Marvell platforms based on Armada-37xx SoC.
This includes support for the official Armada-3720 modular
development board and EspressoBin community board.
The Armada-37xx SoC contains dual Cortex-A53 Application CPU,
single secure CPU (Cortex-M3) and the following interfaces:
- SATA 3.0
- USB 3.0 and USB 2.0
- PCIe
- SDIO (supports boot from eMMC)
- SPI
- UART
- I2c
- Gigabit Ethernet
Signed-off-by: Konstantin Porotchkin <kostap@marvell.com>

A recent patch[0] has made setting up page tables into generic
code, complete the conversion for TI platforms by removing the
use of plat_arm_get_mmap() and using the mmap table directly.

[0] 0916c38d

 ("Convert arm_setup_page_tables into a generic helper")
Signed-off-by: Andrew F. Davis <afd@ti.com>

include/plat/arm/common isn't needed by them, and is removed to avoid
dependency on Arm platform code.

Change-Id: Id9fccba33326fd075b3d1029bf1e4b012dfa0483
Signed-off-by: Antonio Nino Diaz <antonio.ninodiaz@arm.com>

plat/arm files should only be used by Arm platforms. If other platforms
use them, they create dependencies that can introduce problems when
updating Arm platforms.

This patch copies the needed code from Arm platforms so that poplar can
be independent from them.

Change-Id: I0b194f5bdb0377b8ccacbd400e021614c026c7fe
Tested-by: Shawn Guo <shawn.guo@linaro.org>
Signed-off-by: Antonio Nino Diaz <antonio.ninodiaz@arm.com>

When the SoC issues a command IRQ to SCP, SCP sets STMTOBEIRQ as ACK.
The SoC must wait for it before issuing the next command.

This commit makes sure to meet the requirement.
Signed-off-by: Dai Okamura <okamura.dai@socionext.com>
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>

Skipping SCP_BL2 image is just a temporary workaround. If on-chip SCP
needs to work, BL2 should load the SCP_BL2 image.
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>

No functional changes.

Change-Id: I9b9f8d3dfde08d57706ad5450de6ff858a55ac01
Signed-off-by: Antonio Nino Diaz <antonio.ninodiaz@arm.com>

This patch adds support for the N1SDP (NeoVerse N1 System Development
Platform). It is an initial port and additional features are expected
to be added later.

The port includes only BL31 support as the System Control Processor
(SCP) is expected to take the role of primary boatloader

Change-Id: Ife17d8215a7bfcc1420204a72205e7ef920d0c10
Signed-off-by: Deepak Pandey <Deepak.Pandey@arm.com>

This patch enabled PIE for FVP when RESET_TO_BL31=1. The references
to BL31_BASE are replaced by BL31_START as being a symbol exported by
the linker, will create a dynamic relocation entry in .rela.dyn and
hence will be fixed up by dynamic linker at runtime. Also, we disable
RECLAIM_INIT_CODE when PIE is enabled as the init section overlay
creates some static relocations which cannot be handled by the
dynamic linker currently.

Change-Id: I86df1b0a8b2a8bbbe7c3f3c0b9a08c86c2963ec0
Signed-off-by: Soby Mathew <soby.mathew@arm.com>

This patch makes use of mov_imm macro where possible to load
immediate values within ARM platform layer.

Change-Id: I02bc7fbc1fa334c9fccf76fbddf515952f9a1298
Signed-off-by: Soby Mathew <soby.mathew@arm.com>

This patch adds a utility function to return
the address of a symbol. By default, the compiler
generates adr/adrp instruction pair to return
the reference and this utility is used to override
this compiler generated to code and use `ldr`
instruction.

This is needed for Position Independent Executable
when it needs to reference a symbol which is constant
and does not depend on the execute address of the
binary.

For example, on the FVP, the GICv3 register context is
stored in a secure carveout (arm_el3_tzc_dram) within
DDR and does not relocate with the BL image. Now if
BL31 is executing at a different address other than
the compiled address, using adrp/adr instructions to
reference this memory will not work as they generate an
address that is PC relative. The way to get around this
problem is to reference it as non-PC relative (i.e
non-relocatable location) via `ldr` instruction.

Change-Id: I5008a951b007144258121690afb68dc8e12ee6f7
Signed-off-by: Soby Mathew <soby.mathew@arm.com>

This function is not related to Arm platforms and can be reused by other
platforms if needed.

Change-Id: Ia9c328ce57ce7e917b825a9e09a42b0abb1a53e8
Co-authored-by: Antonio Nino Diaz <antonio.ninodiaz@arm.com>
Signed-off-by: Antonio Nino Diaz <antonio.ninodiaz@arm.com>

There are several platforms using arm_setup_page_tables(), which is
supposed to be Arm platform only. This creates several dependency
problems between platforms.

This patch adds the definition XLAT_TABLES_LIB_V2 to the xlat tables lib
v2 makefile. This way it is possible to detect from C code which version
is being used and include the correct header.

The file arm_xlat_tables.h has been renamed to xlat_tables_compat.h and
moved to a common folder. This way, when in doubt, this header can be
used to guarantee compatibility, as it includes the correct header based
on XLAT_TABLES_LIB_V2.

This patch also removes the usage of ARM_XLAT_TABLES_V1 from QEMU (so
that is now locked in xlat lib v2) and ZynqMP (where it was added as a
workaround).

Change-Id: Ie1e22a23b44c549603d1402a237a70d0120d3e04
Signed-off-by: Antonio Nino Diaz <antonio.ninodiaz@arm.com>

There seems to be a problem where SCP can't turn CPU0 off. Instead of
returning PSCI_E_DENIED or crashing make CPU0 wait in a WFE loop. This
way all CPUs have a consistent behaviour from the point of view of the
caller.

Change-Id: I5c8c266ca3b69c9e7a4f5ae70afeea5dd36a0825
Signed-off-by: Antonio Nino Diaz <antonio.ninodiaz@arm.com>

This works fine for CPU1-3, but it fails for CPU0, where it is simply
ignored and leaves CPU0 in a WFI loop.

Change-Id: I7d73683fdd894f2021d6a5bc2cce6cd03e18e633
Signed-off-by: Antonio Nino Diaz <antonio.ninodiaz@arm.com>

The Amlogic Meson S905 is a SoC with a quad core Arm Cortex-A53 running
at 1.5Ghz. It also contains a Cortex-M3 used as SCP.

This port is a minimal implementation of BL31 capable of booting
mainline U-Boot and Linux:

- Partial SCPI support.
- Basic PSCI support (CPU_ON, SYSTEM_RESET, SYSTEM_OFF).
- GICv2 driver set up.
- Basic SIP services (read efuse data, enable/disable JTAG).

This port has been tested in an ODROID-C2.

Change-Id: Ia4bc82d7aca42a69d6b118b947279f82b3f6c6da
Tested-by: Amit Singh Tomar <amittomer25@gmail.com>
Signed-off-by: Antonio Nino Diaz <antonio.ninodiaz@arm.com>

SGI-Clark.Ares platform is a variant of the SGI-Clark platform. It has
two clusters of four Ares CPUs each. Though very similar to the SGI575
platform, there are subtle differences. HW_CONFIG and TB_FW_CONFIG dts
files have been added.

Change-Id: I740a33cbd1c3b1f1984cb56243b46ad379bab3e6
Signed-off-by: Chandni Cherukuri <chandni.cherukuri@arm.com>

SGI-Clark platform is the next version in the Arm's SGI platform
series. One of the primary difference between the SGI-575 platform and
the SGI-Clark platform is the MHU version (MHUv2 in case of SGI-Clark).
Add the required base support for SGI-Clark platform.

Change-Id: If396e5279fdf801d586662dad0b55195e81371e4
Signed-off-by: Chandni Cherukuri <chandni.cherukuri@arm.com>

On SGI platforms that include Ares CPUs, the 'CORE_PWRDN_EN' bit of
'CPUPWRCTLR_EL1' register requires an explicit write to clear it to
enable hotplug and idle to function correctly.

The reset value of the CORE_PWRDN_EN bit is zero but it still requires
this explicit clear to zero. This indicates that this could be a model
related issue but for now this issue can be fixed be clearing the
CORE_PWRDN_EN bit in the platform specific reset handler function.

Change-Id: I8b9884ae27a2986d789bfec2e9ae792ef930944e
Signed-off-by: Chandni Cherukuri <chandni.cherukuri@arm.com>

The default behaviour of the plat_crash_console_xxx functions isn't
obvious to someone that hasn't read all the documentation. As they are
not mandatory, it is unlikely that the code will be checked when doing a
platform port, which may mean that some platforms may not have crash
console support at all.

The idea of this patch is to force platform maintainers to decide how
the crash console has to behave so that the final behaviour isn't
unexpected.

Change-Id: I40b2a7b56c5530c1dcd63eace5bd37ae6335056e
Signed-off-by: Antonio Nino Diaz <antonio.ninodiaz@arm.com>

This platform depends on weak functions defined in
``plat/common/aarch64/platform_helpers.S`` that are going to be removed.

Change-Id: I5104d091c32271d77ed9690e9dc257c061289def
Signed-off-by: Antonio Nino Diaz <antonio.ninodiaz@arm.com>

Platforms that wish to use the sample functions have to add the file to
their Makefile. It is not included by default.

Change-Id: I713617bb58dc218967199248f68da86241d7ec40
Signed-off-by: Antonio Nino Diaz <antonio.ninodiaz@arm.com>

In Arm platforms the crash console doesn't print anything if the crash
happens early enough. This happens in all images, not only BL1. The
reason is that they the files ``plat/common/aarch64/platform_helpers.S``
and ``plat/arm/common/aarch64/arm_helpers.S``, and the crash console
functions are defined as weak in both files. In practice, the linker
can pick the one in ``plat/common``, which simply switches the multi
console to crash mode when it wants to initialize the crash console.
In the case of Arm platforms, there are no console drivers registered
at that point, so nothing is printed.

This patch makes the functions in plat/arm strong so that they override
the weak functions in plat/common.

Change-Id: Id358db7d2567d7df0951790a695636cf6c9ac57f
Signed-off-by: Antonio Nino Diaz <antonio.ninodiaz@arm.com>

Even though at this point plat_crash_console_flush is optional, it will
stop being optional in a following patch.

The console driver of warp7 doesn't support flush, so the implementation
is a placeholder.

TI had ``plat_crash_console_init`` and ``plat_crash_console_putc``, but
they weren't global so they weren't actually used. Also, they were
calling the wrong functions.

imx8_helpers.S only has placeholders for all of the functions.

Change-Id: I8d17bbf37c7dad74e134c61ceb92acb9af497718
Signed-off-by: Antonio Nino Diaz <antonio.ninodiaz@arm.com>

Non-Arm platforms shouldn't use Arm platform code. This patch copies the
implementation of the functions in arm_helpers.S to zynqmp_helpers.S to
remove this dependency of zynqmp on Arm platforms.

Change-Id: Ia85f303c4c63bcf0ffa57c7f3ef9d88376729b6b
Signed-off-by: Antonio Nino Diaz <antonio.ninodiaz@arm.com>

When a device tree blob is present at a known address, instead of, for
example, relying on the user modifying the Linux command line to warn
about the memory reserved for the Trusted Firmware, pass it on the DTB.

The current code deletes the memory reserved for the default bootstrap
of the Raspberry Pi and adds the region used by the Trusted Firmware.

This system replaces the previous one consisting on adding
``memmap=16M$256M`` to the Linux command line. It's also meant to be
used by U-Boot and any other bootloader that understands DTB files.

Change-Id: I13ee528475fb043d6e8d9e9f24228e37ac3ac436
Signed-off-by: Antonio Nino Diaz <antonio.ninodiaz@arm.com>

Version 1.4.7 of libfdt is bigger than the current one (1.4.2) and the
current reserved space for BL2 in Juno isn't enough to fit the Trusted
Firmware when compiling with clang or armclang.

Change-Id: I7b73394ca60d17f417773f56dd5b3d54495a45a8
Signed-off-by: Antonio Nino Diaz <antonio.ninodiaz@arm.com>

The definitions FAIL_CONTROL_*_SHIFT were incorrect, they have been
fixed.

The types tzc_region_attributes_t and tzc_action_t have been removed and
replaced by unsigned int because it is not allowed to do logical
operations on enums.

Also, fix some address definitions in arm_def.h.

Change-Id: Id37941d76883f9fe5045a5f0a4224c133c504d8b
Signed-off-by: Antonio Nino Diaz <antonio.ninodiaz@arm.com>

Move doimage utility from toos/doimage to tools/marvell/doimage.
This is done for supporting mode Marvell tools in the future.
Signed-off-by: Konstantin Porotchkin <kostap@marvell.com>

PSCI requires a core to turn itself off, which we can't do properly by
just executing an algorithm on that very core. As a consequence we just
put a core into WFI on CPU_OFF right now.
To fix this let's task the "arisc" management processor (an OpenRISC
core) with that task of asserting reset and turning off the core's power
domain. We use a handcrafted sequence of OpenRISC instructions to
achieve this, and hand this data over to the new sunxi_execute_arisc_code()
routine.
The commented source code for this routine is provided in a separate file,
but the ATF code contains the already encoded instructions as data.
The H6 uses the same algorithm, but differs in the MMIO addresses, so
provide a SoC (family) specific copy of that code.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>

The more recent Allwinner SoCs contain an OpenRISC management
controller (called arisc or CPUS), which shares the bus with the ARM cores,
but runs on a separate power domain. This is meant to handle power
management with the ARM cores off.
There are efforts to run sophisticated firmware on that core
(communicating via SCPI with the ARM world), but for now can use it for
the rather simple task of helping to turn the ARM cores off. As this
cannot be done by ARM code itself (because execution stops at the
first of the three required steps), we can offload some instructions to
this management processor.
This introduces a helper function to hand over a bunch of instructions
and triggers execution. We introduce a bakery lock to avoid two cores
trying to use that (single) arisc core. The arisc code is expected to
put itself into reset after is has finished execution.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>

There are reports that activating the DC1SW before certain other
regulators leads to the PMIC overheating and consequently shutting down.
To avoid this situation, delay the activation of the DC1SW line until
the very end, so those other lines are always activated earlier.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>

Based on the just introduced PMIC FDT framework, we check the DT for more
voltage rails that need to be setup early:
- DCDC1 is typically the main board power rail, used for I/O pins, for
instance. The PMIC's default is 3.0V, but 3.3V is what most boards use,
so this needs to be adjusted as soon as possible.
- DCDC5 is supposed to be connected to the DRAM. The AXP has some
configurable reset voltage, but some boards get that wrong, so we better
set up this here to avoid over- or under-volting.
- DLDO1,2,3 and FLDO1 mostly drive some graphics related IP, some boards
need this to be up to enable HDMI or the LCD screen, so we get screen
output in U-Boot.

To get the right setup, but still being flexible, we query the DT for
the required voltage and whether that regulator is actually used. That
gives us some robust default setup U-Boot is happy with.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>

Now that we have a pointer to the device tree blob, let's use that to
do some initial setup of the PMIC:
- We scan the DT for the compatible string to find the PMIC node.
- We switch the N_VBUSEN pin if the DT property tells us so.
- We scan over all regulator subnodes, and switch DC1SW if there is at
least one other node referencing it (judging by the existence of a
phandle property in that subnode).
This is just the first part of the setup, a follow up patch will setup
voltages.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>

For Allwinner boards we now use some heuritistics to find a preloaded
.dtb file.

Pass this address on to the PMIC setup routine, so that it can use the
information contained therein to setup some initial power rails.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>

The initial PMIC setup for the Allwinner platform is quite board
specific, and used to be guarded by reading the .dtb stub *name* from the
SPL image in the legacy ATF port. This doesn't scale particularly well,
and requires constant maintainance.
Instead having the actual .dtb available would be much better, as the PMIC
setup requirements could be read from there directly.
The only available BL33 for Allwinner platforms so far is U-Boot, and
fortunately U-Boot comes with the full featured .dtb, appended to the
end of the U-Boot image.

Introduce some code that scans the beginning of the BL33 image to look
for the load address, which is followed by the image size. Adding those
two values together gives us the end of the image and thus the .dtb
address. Verify that this heuristic is valid by sanitising some values
and checking the DTB magic.

Print out the DTB address and the model name, if specified in the root
node.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>

Boards with the Allwinner A64 SoC are mostly paired with an AXP803 PMIC,
which allows to programmatically power down the board.

Use the newly introduced RSB driver to detect and program the PMIC on
boot, then later to turn off the main voltage rails when receiving a
PSCI SYSTEM_POWER_OFF command.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>

In the H6 platform code there is a routine to do the platform
initialisation of the R_I2C controller. We will need a very similar
setup routine to initialise the RSB controller on the A64.

Move this code to sunxi_common.c and generalise it to support all SoCs
and also to cover the related RSB bus.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>

Allwinner produces reference board designs, which apparently most board
vendors copy from. So every H5 board I checked uses regulators which are
controlled by the same PortL GPIO pins to power the ARM CPU cores, the
DRAM and the I/O ports.
Add a SoC specific power down routine, which turns those regulators off
when ATF detects running on an H5 SoC and the rich OS triggers a
SYSTEM_POWEROFF PSCI call.

NOTE: It sounds very tempting to turn the CPU power off, but this is not
working as expected, instead the system is rebooting. Most probably this
is due to VCC-SYS also being controlled by the same GPIO line, and
turning this off requires an elaborate and not fully understood setup.
Apparently not even Allwinner reference code is turning this regulator
off. So for now we refrain to pulling down PL8, the power consumption is
quite low anyway, so we are as close to poweroff as reasonably possible.
Many thanks to Samuel for doing some research on that topic.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>