The existing DSU errata workarounds hijack the errata framework's inner
workings to register with it. However, that is undesirable as any change
to the framework may end up missing these workarounds. So convert the
checks and workarounds to macros and have them included with the
standard wrappers.
The only problem with this is the is_scu_present_in_dsu weak function.
Fortunately, it is only needed for 2 of the errata and only on 3 cores.
So drop it, assuming the default behaviour and have the callers handle
the exception.
Change-Id: Iefa36325804ea093e938f867b9a6f49a6984b8ae
Signed-off-by: Boyan Karatotev <boyan.karatotev@arm.com>
Most of the macros in build_macros.mk get lazily evaluated. That's
mostly fine, except for the fact that the `uppercase` macro needs to
spawn a subshell to get its output. And the target for every file
requires calling `uppercase` many, MANY, times, thrashing performance on
even the most trivial of make commands.
We can be a little clever and only call `uppercase` a handful of times
and then pass around the already uppercased strings.
The same is true about the verbosity augmentation variables. Simply
changing them to simply expanded variables allows for them to be
pre-processed and then used over and over again.
`make realclean` is a pretty good benchmark for this as it doesn't do
much else but must process all the rules, like every other make command.
On a clean checkout of TF-A on an Intel Xeon Gold 5218 (i.e. slow
single-core) workstation, that command used to take about 7 seconds.
With this patch it takes about 0.5.
Change-Id: I632236a12a40f169e834974ecbc73ff80aac3462
Signed-off-by: Boyan Karatotev <boyan.karatotev@arm.com>
Armv8.6 introduced the FEAT_LS64 extension, which provides a 64 *byte*
store instruction. A related instruction is ST64BV0, which will replace
the lowest 32 bits of the data with a value taken from the ACCDATA_EL1
system register (so that EL0 cannot alter them).
Using that ST64BV0 instruction and accessing the ACCDATA_EL1 system
register is guarded by two SCR_EL3 bits, which we should set to avoid a
trap into EL3, when lower ELs use one of those.
Add the required bits and pieces to make this feature usable:
- Add the ENABLE_FEAT_LS64_ACCDATA build option (defaulting to 0).
- Add the CPUID and SCR_EL3 bit definitions associated with FEAT_LS64.
- Add a feature check to check for the existing four variants of the
LS64 feature and detect future extensions.
- Add code to save and restore the ACCDATA_EL1 register on
secure/non-secure context switches.
- Enable the feature with runtime detection for FVP and Arm FPGA.
Please note that the *basic* FEAT_LS64 feature does not feature any trap
bits, it's only the addition of the ACCDATA_EL1 system register that
adds these traps and the SCR_EL3 bits.
Change-Id: Ie3e2ca2d9c4fbbd45c0cc6089accbb825579138a
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Newer cores implemented in the FPGAs used by Arm Ltd. support more
ARMv9 features.
Enable TCR2, MTE, MTE2, SME and SME2 as "enable if available" (:=2), so
any users of those features in lower ELs will not trigger a trap into
BL31.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Change-Id: Id99ecb7c5d6a25b77f7cc5fcad63f60027a4fd5a
The Arm FPGA platform builds a final AXF file, which is an ELF file
containing some required trampolines and binaries, like the DTB. This is
more a "container with load addresses" than an object or executable
file, but it's still built with the linker tool.
Commit acf0076ae2 ("build(fpga): correctly handle gcc as linker for
LTO") pulled in ${TF_LDFLAGS} when building this AXF file, which
includes "--gc-sections". That strips the kernel trampoline off that
file, making the board hang when the kernel is loaded at 0x80200000
(the recommended load address for "newer" kernels).
Drop the usage of TF_LDFLAGS altogether, since we need none of the
options specified there for our special linker step. Instead collect
the needed options (like -nostdlib) in a separate variable, and just
account for the slight syntax differences between GCC and clang.
"--strip-debug" turns out to be redundant, since "-s" already strips
more symbols, so remove that from the list.
Change-Id: I1349d58fa93973ba3add8cab2272259abdea84e0
Fixes: acf0076ae2 ("build(fpga): correctly handle gcc as linker for LTO")
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
This change introduces a few helper variables for dealing with verbose
and silent build modes: `silent`, `verbose`, `q` and `s`.
The `silent` and `verbose` variables are boolean values determining
whether the build system has been configured to run silently or
verbosely respectively (i.e. with `--silent` or `V=1`).
These two modes cannot be used together - if `silent` is truthy then
`verbose` is always falsy. As such:
make --silent V=1
... results in a silent build.
In addition to these boolean variables, we also introduce two new
variables - `s` and `q` - for use in rule recipes to conditionally
suppress the output of commands.
When building silently, `s` expands to a value which disables the
command that follows, and `q` expands to a value which supppresses
echoing of the command:
$(s)echo 'This command is neither echoed nor executed'
$(q)echo 'This command is executed but not echoed'
When building verbosely, `s` expands to a value which disables the
command that follows, and `q` expands to nothing:
$(s)echo 'This command is neither echoed nor executed'
$(q)echo 'This command is executed and echoed'
In all other cases, both `s` and `q` expand to a value which suppresses
echoing of the command that follows:
$(s)echo 'This command is executed but not echoed'
$(q)echo 'This command is executed but not echoed'
The `s` variable is predominantly useful for `echo` commands, where you
always want to suppress echoing of the command itself, whilst `q` is
more useful for all other commands.
Change-Id: I8d8ff6ed714d3cb401946c52955887ed7dca602b
Signed-off-by: Chris Kay <chris.kay@arm.com>
When LTO is enabled and gcc is used as a linker, then option for a
linker have to be provided with a -Wl prefix to gcc.
To build PLAT=arm_fpga with LTO enabled extra '-nostdlib' has to be
supplied to the linker at least, otherwise build fails with an error
about many undefined references in libc.
Since this option is defined as part of common TF_LDFLAGS already,
just use that variable with couple extra options.
Change-Id: Iaab72d894317c91af5b7d770652e4353b32aae88
Signed-off-by: Andrey Skvortsov <andrej.skvortzov@gmail.com>
This change migrates the values of `CC`, `CPP`, `AS` and other toolchain
variables to the new `$(toolchain)-$(tool)` variables, which were
introduced by the toolchain refactor patch. These variables should be
equivalent to the values that they're replacing.
Change-Id: I644fe4ce82ef1894bed129ddb4b6ab94fb04985d
Signed-off-by: Chris Kay <chris.kay@arm.com>
Disable FEAT_MPAM support for Aarch32 as it is not supported, following
[1]. ENABLE_FEAT_MPAM is set to 2 by default for Aarch64 in
arch_features.mk, eliminating the need for duplication in the platform
makefile.
[1] https://review.trustedfirmware.org/c/TF-A/trusted-firmware-a/+/23710
Change-Id: I1c8b6844254e00e6372900f1c87f995f292ae65c
Signed-off-by: Harrison Mutai <harrison.mutai@arm.com>
Restricts MPAM to only NS world and enables trap to EL3 for access of
MPAM registers from lower ELs of Secure and Realm world.
This patch removes MPAM enablement from global context and adds it to
EL3 State context which enables/disables MPAM during world switches.
Renamed ENABLE_MPAM_FOR_LOWER_ELS to ENABLE_FEAT_MPAM and
removed mpam_init_el3() as RESET behaviour is trapping.
Signed-off-by: Arvind Ram Prakash <arvind.ramprakash@arm.com>
Change-Id: I131f9dba5df236a71959b2d425ee11af7f3c38c4
Similar to the FVP and QEMU, the Arm FPGA systems come with different
CPU cores, and gain new features over time.
Add a list of ARMv9.2 features that require TF-A enablement to be usable
from non-secure world. Their existence will be detected at runtime, so
supporting all those features is not required for using the build.
This fixes the Linux kernel booting on a ARMv9.2 FPGA core.
Change-Id: Ie93c32b13ce4f9968081bf38296cd45edad0a928
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Currently we have a large series of errata_refactor patches pending
and they are all failing on arm_fpga build when we add errata_framework.
Errata framework can cause the size to grow and thus causing build
failure on bl31 size. This as of today is blocking us from
merging most of our changes as it will introduce a CI failure.
As an workaround we try to just reduce the arm_fpga build by a710
platform, we have a715 and a720 which should be ok I think.
Once everyone are available for further discussion we could revert this
change back and discuss further whats the right approach.
Signed-off-by: Govindraj Raja <govindraj.raja@arm.com>
Change-Id: I96a821e10aaecf04db7407fb2df38012839bfb94
So far we have the ENABLE_AMU build option to include AMU register
handling code for enabling and context switch. There is also an
ENABLE_FEAT_AMUv1 option, solely to protect the HAFGRTR_EL2 system
register handling. The latter needs some alignment with the new feature
scheme, but it conceptually overlaps with the ENABLE_AMU option.
Since there is no real need for two separate options, unify both into a
new ENABLE_FEAT_AMU name in a first step. This is mostly just renaming at
this point, a subsequent patch will make use of the new feature handling
scheme.
Change-Id: I97d8a55bdee2ed1e1509fa9f2b09fd0bdd82736e
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
* ASM files are renamed to have public IP names in their filename.
* updated other files to include ASM filename changes.
Signed-off-by: Rupinderjit Singh <rupinderjit.singh@arm.com>
Change-Id: Ie899c512b11fd7c4312e3a808bb6b9d2376cdb8c
This has been introduced to simplify dependencies on the FDT wrappers.
We generally want to avoid pulling in components on a file-by-file
basis, particularly as we are trying to draw conceptual boxes around
components in preparation for transitioning the build system to CMake,
where dependencies are modelled on libraries rather than files.
Signed-off-by: Chris Kay <chris.kay@arm.com>
Change-Id: Idb7ee05a9b54a8caa3e07f36e608867e20b6dcd5
Commit 434d0491c5 ("refactor(makefile): remove BL prefixes in build
macros") changed the MAKE_S macro to expect "bl31" instead of just "31".
Adjust our calls to MAKE_S and MAKE_LD to fix the build for arm_fpga.
Change-Id: I2743e421c10eaecb39bfa4515ea049a1b8d18fcb
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
For convenience we let the build system generate an ELF file (named
bl31.axf), containing all the trampolines, BL31 code and the DTB in one
file. This can be fed directly into the FPGA payload tool, and it will
load the bits at the right addresses.
Since this ELF file is more used as a "container with load addresses",
there is no need for normal ELF features like alignment or a symbol
table.
Remove unnecessary sections from that output file, by doing a static
"link", dropping the NOBITS stacks section, and by adding "-n" to the
linker command line (to avoid page alignment). This trims the generated
.axf file, and makes it smaller.
Change-Id: I5768543101d667fb4a3b70e60b08cfe970d2a2b6
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
The arm64 Linux kernel needed to be loaded at a certain offset within any
2MB aligned region; this value was configured at compile time and stored
in the Linux kernel image header. The default value was always 512KiB,
so this is the value we use in the TF-A build system for the kernel
load address.
However the whole scheme around the TEXT_OFFSET changed in Linux v5.8:
Linux kernels became fully relocatable, so this value is largely ignored
now, and its default value changed to 0. The only remainder is a warning
message at boot time in case there is a mismatch:
[Firmware Bug]: Kernel image misaligned at boot, please fix your bootloader!
To avoid this warning, and to make TF-A Linux kernel boot protocol
compliant, we should load newer kernels to offset 0 of a 2 MB
region. This can be done by the user at FPGA boot time, but BL31 needs
to know about this address. As we can't change the build default to 0
without breaking older kernels, we should try to make a build dealing
with both versions:
This patch introduces a small trampoline code, which gets loaded at
512KB of DRAM, and branches up to 2MB. If users load their newer
kernels at 2MB, this trampoline will cover them. In case an older kernel
is loaded at 512KB, it will overwrite this trampoline code, so it would
still work.
Change-Id: If49ca86f5dca380036caf2555349748722901277
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Up until now we relied on the GICs used in our FPGA images to be GICv3
compliant, without the "direct virtual injection" feature (aka GICv4)
enabled.
To support newer images which have GICv4 compliant GICs, enable the
newly introduced GICv4 detection code, and use that also when we adjust
the redistributor region size in the devicetree.
This allows the same BL31 image to be used with GICv3 or GICv4 FPGA
images.
Change-Id: I9f6435a6d5150983625efe3650a8b7d1ef11b1d1
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
As done recently for plat/tc0 in b5863cab9, enable AMU explicitly.
This is necessary as the recent changes that enable SVE for the secure
world disable AMU by default in the CPTR_EL3 reset value.
Change-Id: Ie3abf1dee8a4e1c8c39f934da8e32d67891f5f09
Signed-off-by: Tom Cosgrove <tom.cosgrove@arm.com>
This patch renames the Matterhorn, Matterhorn ELP, and Klein CPUs to
Cortex A710, Cortex X2, and Cortex A510 respectively.
Signed-off-by: John Powell <john.powell@arm.com>
Change-Id: I056d3114210db71c2840a24562b51caf2546e195
ELP processors can sometimes have different MIDR values or features so
we are adding the "_arm" suffix to differentiate the reference
implementation from other future versions.
Signed-off-by: John Powell <john.powell@arm.com>
Change-Id: Ieea444288587c7c18a397d279ee4b22b7ad79e20
Enable basic support for Neoverse-N2 CPUs.
Signed-off-by: Javier Almansa Sobrino <javier.almansasobrino@arm.com>
Change-Id: I498adc2d9fc61ac6e1af8ece131039410872e8ad
* changes:
arm_fpga: Add platform documentation
arm_fpga: Add post-build linker script
arm_fpga: Add ROM trampoline
arm_fpga: Add devicetree file
arm_fpga: Remove SPE PMU DT node if SPE is not available
arm_fpga: Adjust GICR size in DT to match number of cores
fdt: Add function to adjust GICv3 redistributor size
drivers: arm: gicv3: Allow detecting number of cores
For the Arm Ltd. FPGAs to run, we need to load several payloads into the
FPGA's memory:
- Some trampoline code at address 0x0, to jump to BL31's entry point.
- The actual BL31 binary at the beginning of DRAM.
- The (generic) DTB image to describe the hardware.
- The actual non-secure payloads (kernel, ramdisks, ...)
The latter is application specific, but the first three blobs are rather
generic.
Since the uploader tool supports ELF binaries, it seems helpful to
combine these three images into one .axf file, as this also simplifies
the command line.
Add a post-build linker script, that combines those three bits into one
ELF file, together with their specific load addresses.
Include a call to "ld" with this linker script in the platform Makefile,
so it will be build automatically. The result will be called "bl31.axf".
Change-Id: I4a90da16fa1e0e83b51d19e5b1daf61f5a0bbfca
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
The application cores of the FPGAs used in Arm Ltd. start execution at
address 0x0. This is the location of some (emulated) ROM area (which can
be written to by the uploading tool).
Since the arm_fpga port is configured to run from DRAM, we load BL31 to
the beginning of DRAM (mapped at 2GB). This requires some small
trampoline code in the "ROM" to jump to the BL31 entry point.
To avoid some extra magic binary, add a tiny assembly file with that
trivial jump instruction to the tree, so this binary can be created
alongside BL31.
Change-Id: I9e4439fc0f093fa24dd49a8377c9edb030fbb477
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
The FPGA images used in Arm Ltd. focus on CPU cores, so they share a
common platform, with a minimal set of peripherals (interconnect, GIC,
UART).
This allows to support most platforms with a single devicetree file.
The topology and number of CPU cores differ, but those will added at
runtime, in BL31. Other adjustments (GICR size, SPE node, command line)
are also done at this point.
Add the common devicetree file to TF-A's build system, so it can be
build together with BL31. At runtime, the resulting .dtb file should be
uploaded to the address given with FPGA_PRELOADED_DTB_BASE at build time.
Change-Id: I3206d6131059502ec96896e95329865452c9d83e
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
This patch allows the system to fallback to a default CPU library
in case the MPID does not match with any of the supported ones.
This feature can be enabled by setting SUPPORT_UNKNOWN_MPID build
option to 1 (enabled by default only on arm_fpga platform).
This feature can be very dangerous on a production image and
therefore it MUST be disabled for Release images.
Signed-off-by: Javier Almansa Sobrino <javier.almansasobrino@arm.com>
Change-Id: I0df7ef2b012d7d60a4fd5de44dea1fbbb46881ba
At the moment BL31 dynamically discovers the CPU topology of an FPGA
system at runtime, but does not export it to the non-secure world.
Any BL33 user would typically looks at the devicetree to learn about
existing CPUs.
This patch exports a minimum /cpus node in a devicetree to satisfy
the binding. This means that no cpumaps or caches are described.
This could be added later if needed.
An existing /cpus node in the DT will make the code bail out with a
message.
Signed-off-by: Javier Almansa Sobrino <javier.almansasobrino@arm.com>
Change-Id: I589a2b3412411a3660134bdcef3a65e8200e1d7e
The command line for BL33 payloads is typically taken from the DTB. On
"normal" systems the bootloader will put the right version in there, but
we typically don't use one on the FPGAs.
To avoid editing (and possibly re-packaging) the DTB for every change in
the command line, try to read it from some "magic" memory location
instead. It can be easily placed there by the tool that uploads the
other payloads to the FPGA's memory. BL31 will then replace the existing
command line in the DTB with that new string.
To avoid reading garbage, check the memory location for containing a
magic value. This is conveniently chosen to be a simple ASCII string, so
it can just preceed the actual command line in a text file:
--------------------------------
CMD:console=ttyAMA0,38400n8 debug loglevel=8
--------------------------------
Change-Id: I5923a80332c9fac3b4afd1a6aaa321233d0f60da
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
The memory layout for the FPGA is fairly uniform for most of the FPGA
images, and we already assume that DRAM starts at 2GB by default.
Prepopulate PRELOADED_BL33_BASE and FPGA_PRELOADED_DTB_BASE to some
sane default values, to simplify building some stock image.
If people want to deviate from that, they can always override those
addresses on the make command line.
Change-Id: I2238fafb3f8253a01ad2d88d45827c141d9b29dd
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
To support FPGAs with those cores as well, as the respective cpulib
files to the Makefile.
Change-Id: I1a60867d5937be88b32b210c7817be4274554a76
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
The only difference between GIC-500 and GIC-600 relevant to TF-A is the
differing power management sequence.
A certain GIC implementation is detectable at runtime, for instance by
checking the IIDR register. Let's add that test before initiating the
GIC-600 specific sequence, so the code can be used on both GIC-600 and
GIC-500 chips alike, without deciding on a GIC chip at compile time.
This means that the GIC-500 "driver" is now redundant. To allow minimal
platform support, add a switch to disable GIC-600 support.
Change-Id: I17ea97d9fb05874772ebaa13e6678b4ba3415557
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
This should allow git to easily track file moves
Signed-off-by: Jimmy Brisson <jimmy.brisson@arm.com>
Change-Id: I1592cf39a4f94209c560dc6d1a8bc1bfb21d8327
The ARM Generic Timer DT binding describes an (optional) property to
declare the counter frequency. Its usage is normally discouraged, as the
value should be read from the CNTFRQ_EL0 system register.
However in our case we can use it to program this register in the first
place, which avoids us to hard code a counter frequency into the code.
We keep some default value in, if the DT lacks that property for
whatever reason.
Change-Id: I5b71176db413f904f21eb16f3302fbb799cb0305
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
The SCP firmware on the ARM FPGA initialises the UART already. This allows
us to treat the PL011 as an SBSA Generic UART, which does not require
any further setup.
This in particular removes the need for any baudrate and base clock related
settings to be hard coded into the BL31 image.
Change-Id: I16fc943526267356b97166a7068459e06ff77f0f
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
with commit a6ea06f5, the way platform includes gicv3 files has been
modified, this patch adapts to new method of including gicv3 files
for arm_fpga platform.
Signed-off-by: Manish Pandey <manish.pandey2@arm.com>
Change-Id: Ic5ccae842b39b7db06d4f23c5738b174c42edf63
This change is part of the goal of enabling the port to be compatible
with multiple FPGA images.
BL31 behaves differently depending on whether or not the CPUs in the
system use cache coherency, and as a result any CPU libraries that are
compiled together must serve processors that are consistent in this
regard.
This compiles a different set of CPU libraries depending on whether or
not the HW_ASSISTED_COHERENCY is enabled at build-time to indicate the
CPUs support hardware-level support for cache coherency. This build
flag is used in the makefile in the same way as the Arm FVP port.
Signed-off-by: Oliver Swede <oli.swede@arm.com>
Change-Id: I18300b4443176b89767015e3688c0f315a91c27e
This change is part of the goal of enabling the port to be compatible
with multiple FPGA images.
The BL31 port that is uploaded as a payload to the FPGA with an image
should cater for a wide variety of system configurations. This patch
makes the necessary changes to enable it to function with images whose
cluster configurations may be larger (either by utilizing more
clusters, more CPUs per cluster, more threads in each CPU, or a
combination) than the initial image being used for testing.
As part of this, the hard-coded values that configure the size of the
array describing the topology of the power domain tree are increased
to max. 8 clusters, max. 8 cores per cluster & max 4 threads per core.
This ensures the port works with cluster configurations up to these
sizes. When there are too many entries for the number of available PEs,
e.g. if there is a variable number of CPUs between clusters, then there
will be empty entries in the array. This is permitted and the PSCI
library will still function as expected. While this increases its size,
this shouldn't be an issue in the context of the size of BL31, and is
worth the trade-off for the extra compatibility.
Signed-off-by: Oliver Swede <oli.swede@arm.com>
Change-Id: I7d4ae1e20b2e99fdbac428d122a2cf9445394363
