Mikael Olsson [Fri, 4 Nov 2022 14:01:02 +0000 (15:01 +0100)]
feat(ethos-n): add NPU sleeping SMC call
The non-secure world delegation of the register needed to determine if
the Arm(R) Ethos(TM)-N NPU is active or sleeping will be removed in the
future. In preparation for the change, a new SMC call has been added to
allow the non-secure world to ask the SiP service for the state instead.
A minor API version bump has been done with this change to indicate
support for the new functionality.
Signed-off-by: Mikael Olsson <mikael.olsson@arm.com>
Change-Id: I1338341be385cf1891f4809efb7083fae6d928bc
Joshua Pimm [Wed, 9 Nov 2022 11:26:11 +0000 (11:26 +0000)]
feat(ethos-n): add multiple asset allocators
Adds additional asset allocators to the device tree include
file as the non-secure world kernel module for the Arm(R)
Ethos(TM)-N NPU now fully supports having and using multiple
asset allocators.
Signed-off-by: Joshua Pimm <joshua.pimm@arm.com> Signed-off-by: Mikael Olsson <mikael.olsson@arm.com>
Change-Id: I82d53667ef64968ee814f611d0a90abd3b3cf3de
Joshua Pimm [Wed, 19 Oct 2022 14:46:27 +0000 (15:46 +0100)]
feat(ethos-n): add reset type to reset SMC calls
Adds a reset type argument for the soft and hard reset SMC calls to
indicate whether to perform a full reset and setup or only halt the
Arm(R) Ethos(TM)-N NPU. For use in cases where the NPU will not be
used but must be put into a known state, such as suspending the NPU
as part of power management.
Signed-off-by: Joshua Pimm <joshua.pimm@arm.com> Signed-off-by: Mikael Olsson <mikael.olsson@arm.com>
Change-Id: I6018af85a28b0e977166ec29d26f04739123140c
TZMP1 protected memory regions have been added in the Juno platform to
store sensitive data for the Arm(R) Ethos(TM)-N NPU
This is enabled when building TF-A with ARM_ETHOSN_NPU_TZMP1.
The NPU uses two protected memory regions:
1) Firmware region to protect the NPU's firmware from being modified
from the non-secure world
2) Data region for sensitive data used by the NPU
Respective memory region can only be accessed with their unique NSAID.
Signed-off-by: Bjorn Engstrom <bjoern.engstroem@arm.com> Signed-off-by: Mikael Olsson <mikael.olsson@arm.com> Signed-off-by: Rob Hughes <robert.hughes@arm.com>
Change-Id: I65200047f10364ca18681ce348a6edb2ffb9b095
Bjorn Engstrom [Fri, 26 Aug 2022 07:45:45 +0000 (09:45 +0200)]
build(ethos-n): add TZMP1 build flag
For the Arm(R) Ethos(TM)-N NPU Driver to support running inference with
protected memory the TZC must be configured with appropriate regions.
This is controlled in build time by the now added build flag.
The new build flag is only supported with the Arm Juno platform and the
TZC is configured with default memory regions as if TZMP1 wasn't
enabled to facilitate adding the new memory regions later.
Signed-off-by: Bjorn Engstrom <bjoern.engstroem@arm.com> Signed-off-by: Rob Hughes <robert.hughes@arm.com> Signed-off-by: Mikael Olsson <mikael.olsson@arm.com>
Change-Id: I9dc49ac5d091cfbc8c20d7c3ab394a2836438b0f
Boyan Karatotev [Thu, 30 Mar 2023 13:56:45 +0000 (14:56 +0100)]
fix(fvp): work around BL31 progbits exceeded
It is useful to have a single build for the FVP that includes as much
stuff as possible. Such a build allows a single TF-A build to be used on
a wide variety of fvp command lines. Unfortunately, the fvp also has a
(somewhat arbitrary) SRAM limit and enabling a bunch of stuff overruns
what is available.
To workaround this limit, don't enable everything for all
configurations. The offending configuration is when tsp is enabled, so
try to slim the binary down only when building with it.
As this doesn't solve the issue of running out of space for BL31, update
the linker error to give some clue as to what has (likely) caused it
while more permanent fixes are found.
Also add FEAT_RNG to the mix as it got missed in the commotion.
Merge changes from topic "jc/sve" into integration
* changes:
fix(qemu): enable dynamic feature detection of FEAT_SVE for NormalWorld
fix(tc): enable dynamic feature detection of FEAT_SVE for NormalWorld
FEAT_SVE build macro, "ENABLE_SVE_FOR_NS" default value has been updated
to 2, to support its existing behavior of dynamic detection as well as
keep it aligned with the changes concerning STATE=FEAT_STATE_CHECKED(2),
part of Feature Detection procedure.
fix(qemu): enable dynamic feature detection of FEAT_SVE for NormalWorld
Currently, TF-A supports three states for feature flags:
0: FEAT_DISABLED
1: FEAT_STATE_ALWAYS ( for fixed/real platforms)
2: FEAT_STATE_CHECK ( for configurable platforms)
to meet the feature detection requirements dynamically, mainly
targetting configurable/Fixed Virtual platforms.
With this mechanism in place, we are refactoring all the existing
feature flags to the FEAT_STATE_CHECK option(=2), including
FEAT_SVE explicitly for FVPs.
SVE Patch Reference:
https://review.trustedfirmware.org/c/TF-A/trusted-firmware-a/+/19888/25
This newly introduced change, breaks the existing behaviour especially
for virtual platforms, who have set the ENABLE_SVE_FOR_NS flag to 1.
Moving ahead, we advise the platforms to take the following steps while
enabling the features:
1. If the platform is configurable (virtual), and want to ensure feature
detection happens dynamically at runtime, set the build flags to
FEAT_STATE_CHECK(=2).
2. For real(fixed) platforms, depending on the features supported by the
hardware and platform wants to enable it, platforms could set build
flags to FEAT_STATE_ALWAYS(=1).
(Note: Only the non-secure world enablement related build flags have
been refactored to take the values within 0 to 2. As earlier Secure
world enablement flags will still remain boolean.)
Henceforth, in order to keep it aligned with this tri-state mechanism,
changing the qemu platform default to the now supported dynamic
option(=2), so the right decision can be made by the code at runtime.
fix(tc): enable dynamic feature detection of FEAT_SVE for NormalWorld
Currently, TF-A supports three states for feature flags:
0: FEAT_DISABLED
1: FEAT_STATE_ALWAYS (for fixed/real platforms)
2: FEAT_STATE_CHECK (for configurable platforms)
to meet the feature detection requirements dynamically, mainly
targetting configurable/Fixed Virtual platforms.
With this mechanism in place, we are refactoring all the existing
feature flags to the FEAT_STATE_CHECK option(=2), including
FEAT_SVE explicitly for FVPs.
SVE Patch Reference:
https://review.trustedfirmware.org/c/TF-A/trusted-firmware-a/+/19888/25
This newly introduced change, breaks the existing behaviour especially
for virtual platforms, who have set the ENABLE_SVE_FOR_NS flag to 1.
Moving ahead, we advise the platforms to take the following steps while
enabling the features:
1. If the platform is configurable (virtual), and want to ensure feature
detection happens dynamically at runtime, set the build flags to
FEAT_STATE_CHECK(=2).
2. For real(fixed) platforms, depending on the features supported by the
hardware and platform wants to enable it, platforms could set build
flags to FEAT_STATE_ALWAYS(=1).
(Note: Only the non-secure world enablement related build flags have
been refactored to take the values within 0 to 2. As earlier Secure
world enablement flags will still remain boolean.)
Henceforth, in order to keep it aligned with this tri-state mechanism,
changing the TC platform default to the now supported dynamic
option(=2), so the right decision can be made by the code at runtime.
Andre Przywara [Tue, 28 Mar 2023 15:55:06 +0000 (16:55 +0100)]
fix(aarch64): allow build with ARM_ARCH_MINOR=4
When building the FVP platform with SPMD (which activates the context
switch code), but keeping ARM_ARCH_MINOR to 4 or lower, the assembler
will complain about the SCXTNUM_EL2 system register not being supported
by the "selected processor".
Allow building this combination of options by defining the SCXTNUM_EL2
register via the generic S3_ encoding, so any assembler, with any -march
settings, will generate the access without any warnings.
We do protect accesses to this register by runtime checks, if not
explicitly requested otherwise, so can override the toolchain in this
case.
Change-Id: I0941f4c4dcf541bd968c153b9c3fac61ca23f7ef Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Joanna Farley [Tue, 28 Mar 2023 16:38:49 +0000 (18:38 +0200)]
Merge changes from topic "set-wake-source-for-versal-net" into integration
* changes:
refactor(xilinx): move enum to common place
fix(xilinx): fix misra defects
fix(xilinx): remove unnecessary condition
feat(versal): replace irq array with switch case
feat(versal-net): add support for set wakeup source
refactor(versal): move set wake src fn to common place
feat(cpufeat): enable FEAT_SVE for FEAT_STATE_CHECKED
Add support for runtime detection (ENABLE_SVE_FOR_NS=2), by splitting
sve_supported() into an ID register reading function and a
second function to report the support status. That function considers
both build time settings and runtime information (if needed), and is
used before we do SVE specific setup.
Change the FVP platform default to the now supported dynamic
option (=2), so the right decision can be made by the code at runtime.
feat(cpufeat): enable FEAT_SME for FEAT_STATE_CHECKED
Add support for runtime detection (ENABLE_SME_FOR_NS=2), by splitting
feat_sme_supported() into an ID register reading function and a
second function to report the support status. That function considers
both build time settings and runtime information (if needed), and is
used before we do SME specific setup.
Change the FVP platform default to the now supported dynamic option
(=2),so the right decision can be made by the code at runtime.
Manish Pandey [Tue, 28 Mar 2023 10:27:37 +0000 (12:27 +0200)]
Merge changes from topic "psci-osi" into integration
* changes:
feat(sc7280): add support for PSCI_OS_INIT_MODE
feat(fvp): enable support for PSCI OS-initiated mode
feat(psci): update PSCI_FEATURES
feat(psci): add support for OS-initiated mode
feat(psci): add support for PSCI_SET_SUSPEND_MODE
build(psci): add build option for OS-initiated mode
docs(psci): add design proposal for OS-initiated mode
Jay Buddhabhatti [Fri, 23 Dec 2022 06:27:01 +0000 (22:27 -0800)]
feat(versal): replace irq array with switch case
Replaced array of interrupt to PM node index map with switch-case for
Versal. As a result, the size of code got reduced by 527 bytes. In case
of error return invalid node index i.e. XPM_NODEIDX_DEV_MIN.
Signed-off-by: Jay Buddhabhatti <jay.buddhabhatti@amd.com>
Change-Id: Ifb17366362e2d1757d8933e1ce29083f7ad86b8f
Jay Buddhabhatti [Fri, 23 Dec 2022 04:56:02 +0000 (20:56 -0800)]
feat(versal-net): add support for set wakeup source
Currently wakeup source is not getting setup during suspend resume.
Add support to set wakeup source as per IRQ enabled using switch-case
instead of static array as it is more efficient.
Signed-off-by: Jay Buddhabhatti <jay.buddhabhatti@amd.com>
Change-Id: I30d7ceb3a1d56ba5174fc7334f3a29081c918c92
Jay Buddhabhatti [Tue, 28 Feb 2023 09:23:04 +0000 (01:23 -0800)]
refactor(xilinx): rename gic macros to make common
Rename macros PLAT_VERSAL_GICD_BASE, PLAT_VERSAL_GICR_BASE,
PLAT_VERSAL_NET_GICD_BASE and PLAT_VERSAL_NET_GICR_BASE to
PLAT_GICD_BASE_VALUE and PLAT_GICR_BASE_VALUE to make common
for both Versal and Versal NET platforms.
Signed-off-by: Jay Buddhabhatti <jay.buddhabhatti@amd.com>
Change-Id: Ibcebfb8e741e828ef272b32cbedfb4dcbf8629b6
Andre Przywara [Fri, 3 Mar 2023 10:30:06 +0000 (10:30 +0000)]
refactor(amu): use new AMU feature check routines
The AMU extension code was using its own feature detection routines.
Replace them with the generic CPU feature handlers (defined in
arch_features.h), which get updated to cover the v1p1 variant as well.
Change-Id: I8540f1e745d7b02a25a6c6cdf2a39d6f5e21f2aa Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Andre Przywara [Tue, 21 Mar 2023 13:53:19 +0000 (13:53 +0000)]
refactor(amu): unify ENABLE_AMU and ENABLE_FEAT_AMUv1
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>
Amit Nagal [Thu, 23 Mar 2023 08:46:01 +0000 (14:16 +0530)]
feat(zynqmp): build pm code as library
Build Platform Management(PM) code as an Library.
Building PM code as library provides an option to switch to different
firmware interfaces like custom packages.
Change-Id: I872d45edf55ac83a6efb86591d12a0fef7b598cb Signed-off-by: Amit Nagal <amit.nagal@amd.com> Signed-off-by: Akshay Belsare <akshay.belsare@amd.com>
Akshay Belsare [Mon, 27 Mar 2023 05:11:54 +0000 (10:41 +0530)]
chore(zynqmp): print entry address to Secure and NS world
The base address for BL32 and BL33 is read from the FSBL to TF-A
handoff params.
Print the base address for BL32 and BL33 as entry to the secure and
non-secure world respectively in the release build.
Manish Pandey [Mon, 27 Mar 2023 11:08:26 +0000 (13:08 +0200)]
Merge changes from topic "feat_state_part4" into integration
* changes:
refactor(cpufeat): enable FEAT_RNG for FEAT_STATE_CHECKED
refactor(cpufeat): align FEAT_SEL2 to new feature handling
refactor(cpufeat): enable FEAT_NV2 for FEAT_STATE_CHECKED
refactor(cpufeat): enable FEAT_TWED for FEAT_STATE_CHECKED
refactor(cpufeat): enable FEAT_CSV2_2 for FEAT_STATE_CHECKED
refactor(cpufeat): enable FEAT_ECV for FEAT_STATE_CHECKED
refactor(cpufeat): enable FEAT_PAN for FEAT_STATE_CHECKED
refactor(cpufeat): align FEAT_SB to new feature handling
refactor(cpufeat): use alternative encoding for "SB" barrier
refactor(cpufeat): enable SYS_REG_TRACE for FEAT_STATE_CHECKED
fix(cpufeat): make stub enable functions "static inline"
fix(mpam): feat_detect: support major/minor
Varun Wadekar [Wed, 8 Mar 2023 16:47:38 +0000 (16:47 +0000)]
fix(gicv3): workaround for NVIDIA erratum T241-FABRIC-4
The purpose of this patch is to address the T241 erratum T241-FABRIC-4,
which causes unexpected behavior in the GIC when multiple transactions
are received simultaneously from different sources. This hardware issue
impacts NVIDIA server platforms that use more than two T241 chips
interconnected. Each chip has support for 320 {E}SPIs.
This issue occurs when multiple packets from different GICs are
incorrectly interleaved at the target chip. The erratum text below
specifies exactly what can cause multiple transfer packets susceptible
to interleaving and GIC state corruption. GIC state corruption can
lead to a range of problems, including kernel panics, and unexpected
behavior.
The workaround is to ensure that MMIO accesses target the GIC on the
socket that holds the data, for example SPI ranges owned by the socket’s
GIC. This ensures that the GIC will not utilize the inter-socket AXI
Stream interface for servicing these GIC MMIO accesses.
This patch updates the functions that use the GICD_In{E} registers to
ensure that the accesses are directed to the chip that owns the SPI,
instead of using the global alias.
Andre Przywara [Thu, 23 Mar 2023 11:43:22 +0000 (11:43 +0000)]
fix(fpga): include missing header file
Since transitioning over FEAT_SPE to the new feature checking scheme, we
make use of the new is_feat_spe_supported() function in the Arm FPGA
platform code. However this missed to include the header file, so the
build broke.
Add the arch_features.h header to make arm_fpga compile again.
Change-Id: I5c8feecfcc6fb5845a6671842850df1943086a58 Signed-off-by: Andre Przywara <andre.przywara@arm.com>
fix(versal-net): use spin_lock instead of bakery_lock
In ARM v8.2 the cache will turn off automatically when cpu power down.
Therefore use the spin_lock instead of bakery_lock for the platform in
which HW_ASSISTED_COHERENCY is enabled.
In Versal NET platform HW_ASSISTED_COHERENCY is enabled so it will use
spin lock. In ZynqMP and Versal HW_ASSISTED_COHERENCY is not enabled so
it will use bakery_lock.
Also remove bakery_lock_init() because it is empty.
Signed-off-by: Jay Buddhabhatti <jay.buddhabhatti@amd.com>
Change-Id: I18ff939b51f16d7d3484d8564d6ee6c586f363d8
Jay Buddhabhatti [Thu, 23 Mar 2023 05:44:16 +0000 (22:44 -0700)]
fix(versal-net): correct aff level for cpu off
CPU suspend is calling validate_power_state PSCI opps which returns
power domain state for CPU suspend according to PSTATE type. In case of
power down it assigns PLAT_MAX_OFF_STATE to all affinity level which is
incorrect since for CPU suspend we need to set only MPIDR_AFFLVL0 which
is CPU state. So correct affinity level for CPU suspend.
Signed-off-by: Jay Buddhabhatti <jay.buddhabhatti@amd.com>
Change-Id: I39f92790ea74e4cab8e87342e73e1ac211a46fcd
refactor(fvp): use RSS API to retrieve attestation token and key
Retrieved the platform attestation token and delegated realm attestation
key through the PSA delegated attestation layer.
Even though FVP doesn't support RSS hardware today, it can still
leverage the RSS implementation of these PSA interfaces in their mocking
form (see PLAT_RSS_NOT_SUPPORTED).
Therefore, platform APIs now call these PSA interfaces instead of
directly providing these hardcoded values.
Change-Id: I31d0ca58f6f1a444f513d954da4e3e67757321ad Signed-off-by: Manish V Badarkhe <Manish.Badarkhe@arm.com>
Andre Przywara [Wed, 22 Feb 2023 17:55:59 +0000 (17:55 +0000)]
refactor(cpufeat): enable FEAT_RNG for FEAT_STATE_CHECKED
At the moment we only support for FEAT_RNG to be either unconditionally
compiled in, or to be not supported at all.
Add support for runtime detection (FEAT_RNG=2), by splitting
is_armv8_5_rng_present() into an ID register reading function and a second
function to report the support status. That function considers both build
time settings and runtime information (if needed), and is used before we
access the RNDRRS system register.
Change the QEMU platform default to the now supported dynamic option (=2),
so the right decision can be made by the code at runtime.
Change-Id: I1a4a538d5ad395fead7324f297d0056bda4f84cb Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Andre Przywara [Wed, 22 Feb 2023 16:53:50 +0000 (16:53 +0000)]
refactor(cpufeat): align FEAT_SEL2 to new feature handling
In ARMv8.4, the EL2 exception level got added to the secure world.
Adapt and rename the existing is_armv8_4_sel2_present() function, to
align its handling with the other CPU features.
Change-Id: If11e1942fdeb63c63f36ab9e89be810347d1a952 Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Andre Przywara [Fri, 27 Jan 2023 14:09:20 +0000 (14:09 +0000)]
refactor(cpufeat): enable FEAT_NV2 for FEAT_STATE_CHECKED
At the moment we only support for FEAT_NV2 to be either unconditionally
compiled in, or to be not supported at all.
Add support for runtime detection (CTX_INCLUDE_NEVE_REGS=2), by
splitting get_armv8_4_feat_nv_support() into an ID register reading
function and a second function to report the support status. That
function considers both build time settings and runtime information
(if needed), and is used before we access the VNCR_EL2 system register.
Also move the context saving code from assembly to C, and use the new
is_feat_nv2_supported() function to guard its execution.
Change the FVP platform default to the now supported dynamic option (=2),
so the right decision can be made by the code at runtime.
Change-Id: I85b080641995fb72cfd4ac933f7a3f75770c2cb9 Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Andre Przywara [Fri, 27 Jan 2023 12:25:49 +0000 (12:25 +0000)]
refactor(cpufeat): enable FEAT_TWED for FEAT_STATE_CHECKED
At the moment we only support FEAT_TWED to be either unconditionally
compiled in, or to be not supported at all.
Add support for runtime detection (ENABLE_FEAT_TWED=2), by splitting
is_armv8_6_twed_present() into an ID register reading function and a
second function to report the support status. That function considers
both build time settings and runtime information (if needed), and is
used before we set the trap delay time.
Change the FVP platform default to the now supported dynamic option (=2),
so the right decision can be made by the code at runtime.
Change-Id: I58626230ef0af49886c0a197abace01e81f661d2 Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Andre Przywara [Thu, 17 Nov 2022 17:30:43 +0000 (17:30 +0000)]
refactor(cpufeat): enable FEAT_CSV2_2 for FEAT_STATE_CHECKED
At the moment we only support FEAT_CSV2_2 to be either unconditionally
compiled in, or to be not supported at all.
Add support for runtime detection (ENABLE_FEAT_CSV2_2=2), by splitting
is_armv8_0_feat_csv2_2_present() into an ID register reading function
and a second function to report the support status. That function
considers both build time settings and runtime information (if needed),
and is used before we access the SCXTNUM_EL2 system register.
Also move the context saving code from assembly to C, and use the new
is_feat_csv2_2_supported() function to guard its execution.
Change the FVP platform default to the now supported dynamic option (=2),
so the right decision can be made by the code at runtime.
Change-Id: I89c7bc883e6a65727fdbdd36eb3bfbffb2196da7 Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Andre Przywara [Thu, 17 Nov 2022 17:30:43 +0000 (17:30 +0000)]
refactor(cpufeat): enable FEAT_ECV for FEAT_STATE_CHECKED
At the moment we only support FEAT_ECV to be either unconditionally
compiled in, or to be not supported at all.
Add support for runtime detection (ENABLE_FEAT_ECV=2), by splitting
is_feat_ecv_present() into an ID register reading function and a second
function to report the support status. That function considers both
build time settings and runtime information (if needed), and is used
before we access the CNTPOFF_EL2 system register.
Also move the context saving code from assembly to C, and use the new
is_feat_ecv_supported() function to guard its execution.
Change the FVP platform default to the now supported dynamic option (=2),
so the right decision can be made by the code at runtime.
Change-Id: I4acd5384929f1902b62a87ae073aafa1472cd66b Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Andre Przywara [Thu, 26 Jan 2023 15:27:38 +0000 (15:27 +0000)]
refactor(cpufeat): enable FEAT_PAN for FEAT_STATE_CHECKED
At the moment we only support FEAT_PAN to be either unconditionally
compiled in, or to be not supported at all.
Add support for runtime detection (ENABLE_FEAT_PAN=2), by splitting
is_armv8_1_pan_present() into an ID register reading function and a
second function to report the support status. That function considers
both build time settings and runtime information (if needed), and is
used before we PAN specific setup.
Change the FVP platform default to the now supported dynamic option (=2),
so the right decision can be made by the code at runtime.
Change-Id: I58e5fe8d3c9332820391c7d93a8fb9dba4cf754a Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Andre Przywara [Thu, 17 Nov 2022 16:42:09 +0000 (16:42 +0000)]
refactor(cpufeat): align FEAT_SB to new feature handling
FEAT_SB introduces a new speculation barrier instruction, that is more
lightweight than a "dsb; isb" combination. We use that in a hot path,
so cannot afford and don't want a runtime detection mechanism.
Nevertheless align the implementation of the feature detection part
with the other features, but renaming the detection function, and
updating the FEAT_DETECTION code. Also update the documentation.
Change-Id: I2b86dfd1ad259c3bb99ab5186e2911ace454b54c Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Andre Przywara [Fri, 25 Nov 2022 14:10:13 +0000 (14:10 +0000)]
refactor(cpufeat): use alternative encoding for "SB" barrier
The "sb" barrier instruction is a rather new addition to the AArch64
instruction set, so it is not recognised by all toolchains. On top of
that, the GNU assembler denies this instruction, unless a compatible
processor is selected:
asm_macros.S:223: Error: selected processor does not support `sb'
Provide an alternative encoding of the "sb" instruction, by using a
system register write, as this is the group where the barrier
instructions borrow their encoding space from.
This results in the exact same opcode to be generated, and any
disassembler will decode this instruction as "sb".
Change-Id: I5f44c8321e0cc04c784e02bd838e964602a96a8e Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Andre Przywara [Thu, 17 Nov 2022 16:42:09 +0000 (16:42 +0000)]
refactor(cpufeat): enable SYS_REG_TRACE for FEAT_STATE_CHECKED
At the moment we only support access to the trace unit by system
registers (SYS_REG_TRACE) to be either unconditionally compiled in, or
to be not supported at all.
Add support for runtime detection (ENABLE_SYS_REG_TRACE_FOR_NS=2), by
adding is_feat_sys_reg_trace_supported(). That function considers both
build time settings and runtime information (if needed), and is used
before we access SYS_REG_TRACE related registers.
The FVP platform decided to compile in support unconditionally (=1),
even though this is an optional feature, so it is not available with the
FVP model's default command line.
Change that to the now supported dynamic option (=2), so the right
decision can be made by the code at runtime.
Change-Id: I450a574a4f6bd9fc269887037049c94c906f54b2 Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Andre Przywara [Wed, 22 Mar 2023 13:25:00 +0000 (13:25 +0000)]
fix(cpufeat): make stub enable functions "static inline"
For builds where we don't have a certain feature enabled, we provide
empty stub xxx_enable() functions in a header file. This way we
avoid #ifdef's in the code, and can call the enable function
unconditionally. When compiling with -O1 or higher, the compiler will
even optimise out the whole call, so the symbol will never make it into
any object file.
When compiling with optimisations turned off, the function stub will
survive, and could make it into multiple object files, which would lead
to a multiple definitons error.
Avoid this by defining those stub functions as "static inline". The
"static" will avoid the multiple definitions problems, the "inline" will
avoid a potential compiler warning about unused functions.
This patterns is used extensively in the Linux kernel.
Change-Id: Iad07bb946aab372587c83f2423b4983bf3817990 Reported-by: Chris Kay <chris.kay@arm.com> Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Andre Przywara [Tue, 21 Mar 2023 14:44:59 +0000 (14:44 +0000)]
fix(mpam): feat_detect: support major/minor
The MPAM CPU ID version number is split between two CPU ID register
fields, with the second being a fractional field, allowing for instance
for a "MPAM v1.1" number. The read_feat_mpam_version() function merges
those two fields to form a "4.4" fixed point fractional number, but the
limit check in the check_feature() function was not taking this into
account.
To support MPAM major version 1, extend the limit from "1" to "17", to
cover the current maximum version of "MPAM v1.1".
This fixes FVP runs with "has_mpam=1" and FEATURE_DETECTION enabled.
Change-Id: Icb557741d597e4e43eaf658b78f18af6e9fb439e Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Wing Li [Wed, 14 Sep 2022 20:18:19 +0000 (13:18 -0700)]
feat(psci): update PSCI_FEATURES
This patch updates the PSCI_FEATURES handler to indicate support for
OS-initiated mode per section 5.15.2 of the PSCI spec (DEN0022D.b) based
on the value of `FF_SUPPORTS_OS_INIT_MODE`, which is conditionally
enabled by the `PSCI_OS_INIT_MODE` build option.
Change-Id: I5da8a989b53419ad2ab55b73ddeee6e882c25554 Signed-off-by: Wing Li <wingers@google.com>
Wing Li [Wed, 14 Sep 2022 20:18:17 +0000 (13:18 -0700)]
feat(psci): add support for OS-initiated mode
This patch adds a `psci_validate_state_coordination` function that is
called by `psci_cpu_suspend_start` in OS-initiated mode.
This function validates the request per sections 4.2.3.2, 5.4.5, and 6.3
of the PSCI spec (DEN0022D.b):
- The requested power states are consistent with the system's state
- The calling core is the last running core at the requested power level
This function differs from `psci_do_state_coordination` in that:
- The `psci_req_local_pwr_states` map is not modified if the request
were to be denied
- The `state_info` argument is never modified since it contains the
power states requested by the calling OS
This is conditionally compiled into the build depending on the value of
the `PSCI_OS_INIT_MODE` build option.
Change-Id: I667041c842d2856e9d128c98db4d5ae4e4552df3 Signed-off-by: Wing Li <wingers@google.com>
Wing Li [Wed, 14 Sep 2022 20:18:15 +0000 (13:18 -0700)]
feat(psci): add support for PSCI_SET_SUSPEND_MODE
This patch adds a PSCI_SET_SUSPEND_MODE handler that validates the
request per section 5.20.2 of the PSCI spec (DEN0022D.b), and updates
the suspend mode to the requested mode.
This is conditionally compiled into the build depending on the value of
the `PSCI_OS_INIT_MODE` build option.
Change-Id: Iebf65f5f7846aef6b8643ad6082db99b4dcc4bef Signed-off-by: Wing Li <wingers@google.com>
Bipin Ravi [Mon, 20 Mar 2023 21:21:24 +0000 (22:21 +0100)]
Merge changes from topics "qemu", "qemu_sbsa" into integration
* changes:
feat(qemu): add A76/N1 cpu support for virt
feat(qemu): add "neoverse-n1" cpu support
feat(qemu): make coherent memory section optional
refactor(qemu): make use of setup_page_tables()
Manish Pandey [Mon, 20 Mar 2023 17:25:00 +0000 (18:25 +0100)]
Merge changes from topic "feat_state_part3" into integration
* changes:
refactor(cpufeat): enable FEAT_VHE for FEAT_STATE_CHECKED
refactor(mpam): enable FEAT_MPAM for FEAT_STATE_CHECKED
feat(libc): add support for fallthrough statement
refactor(spe): enable FEAT_SPE for FEAT_STATE_CHECKED
refactor(cpufeat): rename ENABLE_SPE_FOR_LOWER_ELS to ENABLE_SPE_FOR_NS
fix(spe): drop SPE EL2 context switch code
Andre Przywara [Thu, 17 Nov 2022 16:42:09 +0000 (16:42 +0000)]
refactor(cpufeat): enable FEAT_VHE for FEAT_STATE_CHECKED
At the moment we only support FEAT_VHE to be either unconditionally
compiled in, or to be not supported at all.
Add support for runtime detection (ENABLE_FEAT_VHE=2), by splitting
is_armv8_1_vhe_present() into an ID register reading function and a
second function to report the support status. That function considers
both build time settings and runtime information (if needed), and is
used before we access VHE related registers.
Also move the context saving code from assembly to C, and use the new
is_feat_vhe_supported() function to guard its execution.
Enable VHE in its runtime detection version for all FVP builds.
Change-Id: Ib397cd0c83e8c709bd6fed603560e39901fa672b Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Andre Przywara [Thu, 17 Nov 2022 16:42:09 +0000 (16:42 +0000)]
refactor(mpam): enable FEAT_MPAM for FEAT_STATE_CHECKED
At the moment we only support FEAT_MPAM to be either unconditionally
compiled in, or to be not supported at all.
Add support for runtime detection (ENABLE_MPAM_FOR_LOWER_ELS=2), by
splitting get_mpam_version() into an ID register reading
function and a second function to report the support status. That
function considers both build time settings and runtime information (if
needed), and is used before we access MPAM related registers.
Also move the context saving code from assembly to C, and use the new
is_feat_mpam_supported() function to guard its execution.
ENABLE_MPAM_FOR_LOWER_ELS defaults to 0, so add a stub enable function
to cover builds with compiler optimisations turned off. The unused
mpam_enable() function call will normally be optimised away (because it
would never be called), but with -O0 the compiler will leave the symbol
in the object file.
Change-Id: I531d87cb855a7c43471f861f625b5a6d4bc61313 Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Andre Przywara [Tue, 21 Feb 2023 12:05:29 +0000 (12:05 +0000)]
feat(libc): add support for fallthrough statement
Modern C compilers warn about unannotated switch/case fallthrough code,
and require either a comment with some magic words, or an explicit
compiler attribute.
Since some TF-A static analysis CI check suggests having a "fallthrough;"
statement instead of a comment, introduce a macro that implements that
statement, and emits the proper compiler attribute.
Change-Id: Ib34e615fb48d0f4a340aabfad4472e08d5c70248 Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Andre Przywara [Thu, 17 Nov 2022 16:42:09 +0000 (16:42 +0000)]
refactor(spe): enable FEAT_SPE for FEAT_STATE_CHECKED
At the moment we only support FEAT_SPE to be either unconditionally
compiled in, or to be not supported at all.
Add support for runtime detection (ENABLE_SPE_FOR_NS=2), by splitting
is_armv8_2_feat_spe_present() into an ID register reading function and
a second function to report the support status. That function considers
both build time settings and runtime information (if needed), and is
used before we access SPE related registers.
Previously SPE was enabled unconditionally for all platforms, change
this now to the runtime detection version.
Change-Id: I830c094107ce6a398bf1f4aef7ffcb79d4f36552 Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Andre Przywara [Fri, 3 Feb 2023 15:30:14 +0000 (15:30 +0000)]
refactor(cpufeat): rename ENABLE_SPE_FOR_LOWER_ELS to ENABLE_SPE_FOR_NS
At the moment we hardcode the SPE functionality to be available on the
non-secure side only, by setting MDCR_EL2.E2PB accordingly.
This should be reflected in the feature selection symbol, so rename that
to ENABLE_SPE_FOR_NS, to make it clearer that SPE is not supported in
the secure world.
Change-Id: I3f9b48eab1a45d6ccfcbb9c90a11eeb66867ad9a Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Andre Przywara [Fri, 3 Feb 2023 15:23:59 +0000 (15:23 +0000)]
fix(spe): drop SPE EL2 context switch code
At the moment we hardcode the SPE functionality to be available on the
non-secure side only, by setting MDCR_EL3.NSPB accordingly.
This also means that the secure world cannot use SPE, so there is no
need to context switch the PMSCR_EL2 register.
Drop the SPE bits from the EL2 context switch code. If any of the other
EL2 worlds wish to start using SPE, this can be brought back.
Change-Id: Ie0fedb2aeb722a2c9db316051fbbe57ca0e3c0c9 Signed-off-by: Andre Przywara <andre.przywara@arm.com>
projects / arm-tf.git / log