Boyan Karatotev [Mon, 3 Oct 2022 13:21:28 +0000 (14:21 +0100)]
fix(cpus): workaround for Cortex-A710 erratum 2291219
Cortex-A710 erratum 2291219 is a Cat B erratum that applies to revisions
r0p0, r1p0, and r2p0, and is fixed in r2p1. The workaround is to set
CPUACTLR2_EL1[36] to 1 before the power down sequence that sets
CORE_PWRDN_EN. This allows the cpu to retry the power down and prevents
the deadlock. TF-A never clears this bit even if it wakes up from the
wfi in the sequence since it is not expected to do anything but retry to
power down after and the bit is cleared on reset.
SDEN can be found here:
https://developer.arm.com/documentation/SDEN1775101/latest
Boyan Karatotev [Mon, 3 Oct 2022 13:18:28 +0000 (14:18 +0100)]
fix(cpus): workaround for Cortex-X3 erratum 2313909
Cortex-X3 erratum 2313909 is a Cat B erratum that applies to revisions
r0p0 and r1p0, and is fixed in r1p1. The workaround is to set
CPUACTLR2_EL1[36] to 1 before the power down sequence that sets
CORE_PWRDN_EN. This allows the cpu to retry the power down and prevents
the deadlock. TF-A never clears this bit even if it wakes up from the
wfi in the sequence since it is not expected to do anything but retry to
power down after and the bit is cleared on reset.
SDEN can be found here:
https://developer.arm.com/documentation/SDEN2055130/latest
Boyan Karatotev [Mon, 3 Oct 2022 13:07:08 +0000 (14:07 +0100)]
fix(cpus): workaround for Neoverse-N2 erratum 2326639
Neoverse-N2 erratum 2326639 is a Cat B erratum that applies to revision
r0p0 and is fixed in r0p1. The workaround is to set CPUACTLR2_EL1[36] to
1 before the power down sequence that sets CORE_PWRDN_EN. This allows
the cpu to retry the power down and prevents the deadlock. TF-A never
clears this bit even if it wakes up from the wfi in the sequence since
it is not expected to do anything but retry to power down after and the
bit is cleared on reset.
SDEN can be found here:
https://developer.arm.com/documentation/SDEN1982442/latest/
Platforms which implement pwr_domain_pwr_down_wfi differ substantially
in behaviour. However, different cpus require similar sequences to power
down. This patch tightens the behaviour of these platforms to end on a
wfi loop after performing platform power down. This is required so that
platforms behave more consistently on power down, in cases where the wfi
can fall through.
Add documentation how to build EL3 SPMC,
briefly describes all FF-A interfaces,
SP boot flow, SP Manifest, Power Management,
Boot Info Protocol, Runtime model and state
transition and Interrupt Handling.
Boyan Karatotev [Tue, 25 Oct 2022 10:29:04 +0000 (11:29 +0100)]
chore: rename Makalu ELP to Cortex-X3
The Cortex-X3 cpu port was developed before its public release when it
was known as Makalu ELP. Now that it's released we can use the official
product name.
Bipin Ravi [Thu, 13 Oct 2022 22:25:51 +0000 (17:25 -0500)]
fix(security): optimisations for CVE-2022-23960
Optimised the loop workaround for Spectre_BHB mitigation:
1. use of speculation barrier for cores implementing SB instruction.
2. use str/ldr instead of stp/ldp as the loop uses only X2 register.
Signed-off-by: Bipin Ravi <bipin.ravi@arm.com>
Change-Id: I8ac53ea1e42407ad8004c1d59c05f791011f195d
Shawn Guo [Wed, 26 Oct 2022 08:38:53 +0000 (16:38 +0800)]
fix(imx8m): update poweroff related SNVS_LPCR bits only
Function imx_system_off() writes SNVS_LPCR register to power off the SoC
without bit masking. This clears other bits like LPWUI_EN and breaks
the function of SoC wake-up using RTC alarm. Fix it by updating poweroff
related bits only.
Daniel Boulby [Wed, 5 Oct 2022 10:05:22 +0000 (11:05 +0100)]
feat(zlib): update zlib source files
Upgrade the zlib source files to the ones present in the version 1.2.13
of zlib [1]. Since 1.2.11 the use of Arm crc32 instructions has been
introduced so update the files to make use of this.
[1] https://github.com/madler/zlib/tree/v1.2.13
Change-Id: Ideef78c56f05ae7daec390d00dcaa8f66b18729e Signed-off-by: Daniel Boulby <daniel.boulby@arm.com>
Merge changes from topic "imx8m-hab-support" into integration
* changes:
docs(imx8m): update for high assurance boot
feat(imx8m): add support for high assurance boot
feat(imx8mp): add hab and map required memory blocks
feat(imx8mn): add hab and map required memory blocks
feat(imx8mm): add hab and map required memory blocks
build: deprecate Arm rdn1edge and sgi575 FVP platforms
Arm has decided to deprecate the sgi575 and rdn1edge platforms.
The development of software and fast models for these platforms
has been discontinued. rdn1edge platform has been superseded by the
rdn2 platform, which is already supported in TF-A and CI work is
underway for this platform.
Change-Id: If2228fb73549b244c3a5b0e5746617b3f24fe771 Signed-off-by: Manish V Badarkhe <Manish.Badarkhe@arm.com>
Marco Felsch [Fri, 1 Jul 2022 13:44:09 +0000 (15:44 +0200)]
refactor(imx8mp): cleanup the mmap region settings
Introduce the bl_regions array to gather all regions and make use of the
MAP_REGION_FLAT() macro. The array is than passed to mmap_add() to map
all regions. While on it introduce some defines so the addr, size and
flags can be read more easily. No functional change done.
Signed-off-by: Marco Felsch <m.felsch@pengutronix.de>
Change-Id: I7f637beea61138a86d691cd78fba2dd17e4dc925
Marco Felsch [Tue, 5 Jul 2022 13:00:44 +0000 (15:00 +0200)]
feat(imx8m): make psci common code pie compatible
Swap the BL31_BASE define with the BL31_START symbol. This is required
for later added PIE support because the symbol location can be relocated
whereas the define can't be relocated. In case of disabled PIE support
BL31_START equals BL31_BASE and so we don't need a ifdef.
Signed-off-by: Marco Felsch <m.felsch@pengutronix.de>
Change-Id: Ic1bbf3af5b346898bfcbb207ffc27d9a5bdcaae7
Marco Felsch [Wed, 21 Sep 2022 15:48:35 +0000 (17:48 +0200)]
fix(imx8m): fix dram retention fsp_table access
The fsp_table access by [i-1] can cause invalid memory access in case of
i=0. This can be the case if no fsp_table is available. Fix this by
adding the idx variable which tracks the correct index.
Signed-off-by: Marco Felsch <m.felsch@pengutronix.de>
Change-Id: If2285517eb9fe837f3ad54360307a77a658bf62c
Andre Przywara [Fri, 7 Oct 2022 11:19:05 +0000 (12:19 +0100)]
fix(aarch64): make AArch64 FGT feature detection more robust
The ARMv8 ARM says about the values in the ID register scheme:
==== D17.1.3 Principles of the ID scheme for fields in ID registers ===
The ID fields, which are either signed or unsigned, use increasing
numerical values to indicate increases in functionality. Therefore,
if a value of 0x1 indicates the presence of some instructions, then
the value 0x2 will indicate the presence of those instructions plus
some additional instructions or functionality. This means software
can be written in the form:
if (value >= number) {
// do something that relies on the value of the feature
}
=======================================================================
So to check for the presence of a certain architecture feature, we
should not check against a certain specific value, as it's done right
now in several cases.
Relax the test for Fine Grained Trapping (FGT) to just check against
the field being 0 or not.
This fixes TF-A crashing due to an unhandled exception, when running a
Linux kernel on an FVP enabling ARMv8.9 features. The value of
ID_AA64MMFR0_EL1.FGT went from 0b0001 to 0b0010 there.
Change-Id: Ic3f1625a7650306ed388a0660429ca8823c673c2 Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Add a section into documentation listing the support for High Assurance
Boot (HABv4), note on the DRAM mapping, and reference to the external
documentation.
Introduce support for High Assurance Boot (HABv4), which is used to
establish and extend the Root-of-Trust during FW loading at any given
boot stage.
This commit introduces support for HAB ROM Vector Table (RVT) API, which
is normally used by post-ROM code to authenticate additional boot images
(Kernel, FDT, FIT, etc.) that are taking part in the Root-of-Trust.
feat(imx8mp): add hab and map required memory blocks
In order for HAB to perform operations, memory regions has to be mapped
in TF-A, which HAB ROM code would use internally.
Include those memory blocks for i.MX8M+ SoC. Of a special note, the DRAM
block is mapped with complete size available on the platform and uses
MT_RW attributes, this is required to minimize the size of translation
tables and provide a possibility to exchange the execution results
between EL3 and EL1&2, see details in [1].
feat(imx8mn): add hab and map required memory blocks
In order for HAB to perform operations, memory regions has to be mapped
in TF-A, which HAB ROM code would use internally.
Include those memory blocks for i.MX8MN SoC. Of a special note, the DRAM
block is mapped with complete size available on the platform and uses
MT_RW attributes, this is required to minimize the size of translation
tables and provide a possibility to exchange the execution results
between EL3 and EL1&2, see details in [1].
feat(imx8mm): add hab and map required memory blocks
In order for HAB to perform operations, memory regions has to be mapped
in TF-A, which HAB ROM code would use internally.
Include those memory blocks for i.MX8MM SoC. Of a special note, the DRAM
block is mapped with complete size available on the platform and uses
MT_RW attributes, this is required to minimize the size of translation
tables and provide a possibility to exchange the execution results
between EL3 and EL1&2, see details in [1].
Daniel Boulby [Fri, 23 Sep 2022 08:37:20 +0000 (09:37 +0100)]
docs(prerequisites): upgrade to Mbed TLS 2.28.1
In anticpation of the next Trusted Firmware release update the to newest
2.x Mbed TLS library [1].
Note that the Mbed TLS project published version 3.x some time ago.
However, as this is a major release with API breakages, upgrading to
this one might require some more involved changes in TF-A, which we are
not ready to do. We shall upgrade to Mbed TLS 3.x after the v2.8 release
of TF-A.
Akshay Belsare [Tue, 11 Oct 2022 09:42:02 +0000 (15:12 +0530)]
fix(versal_net): Enable a78 errata workarounds
TF-A is reporting that erratum are missing to be enabled.
Enable the Following errata workaround to Cortex-A78 AE CPU for versal_net
ERRATA_A78_AE_1941500
ERRATA_A78_AE_1951502
ERRATA_A78_AE_2376748
ERRATA_A78_AE_2395408
For further information refer to
https://developer.arm.com/documentation/SDEN1707912/1300/
Right now, the delegated attestation module is not used in TF-A. This
means it's not even getting built and so the CI system cannot detect
build regressions.
Eventually, delegated attestation will be involved in a new runtime
service exposed by BL31 to lower exception levels. We are not there
yet but let's already include it into BL31 image, so we get build
coverage and static analysis on the code. Note that we make sure to
cover both PLAT_RSS_NOT_SUPPORTED=0 and PLAT_RSS_NOT_SUPPORTED=1
configurations.
Delegated attestation is currently made dependent on measured boot
support. This dependency is not at the source code level (attestation
code does not invoke any measured boot interfaces) but it is rather a
logical dependency: attestation without boot measurements is not very
useful...
For now, this is good enough for our purpose but the conditions under
which the attestation code is included might change in the future.
Boyan Karatotev [Thu, 13 Oct 2022 12:51:05 +0000 (13:51 +0100)]
fix(sme): add missing ISBs
EL3 is configured to trap accesses to SME registers (via
CPTR_EL3.ESM=0). To allow SME instructions, this needs to be temporarily
disabled before changing system registers. If the PE delays the effects
of writes to system registers then accessing the SME registers will trap
without an isb. This patch adds the isb to restore functionality.
Michal Simek [Fri, 7 Oct 2022 06:15:19 +0000 (08:15 +0200)]
fix(versal): enable a72 erratum 859971 and 1319367
TF-A is reporting that above two erratum are missing to be enabled that's
why enable them by default.
For futher information please refer to
https://developer.arm.com/documentation/epm012079/11/
where
859971 is "Speculative instruction prefetch to Execute-never (XN) memory
could cause deadlock or data integrity issue" and 1319367 is "Speculative AT instruction using out-of-context translation
regime could cause subsequent request to generate an incorrect
translation".
Change-Id: I408706713a169e53db63ac5657751b0b003e646d Signed-off-by: Michal Simek <michal.simek@amd.com>
Boyan Karatotev [Wed, 5 Oct 2022 13:43:54 +0000 (14:43 +0100)]
chore(rpi3): remove redundant code
The pwr_domain_pwr_down_wfi entry is overridden by a newer
implementation. This removes the last reference to
rpi3_pwr_domain_pwr_down_wfi. Remove both as they are not needed
Scott Parlane [Mon, 5 Sep 2022 22:59:57 +0000 (10:59 +1200)]
fix(rk3399): explicitly define the sys_sleep_flag_sram type
Recent GCC versions now do array-bounds checking which fails for
sys_sleep_flag_sram because the struct is larger than the 8-bytes
size that (void *) is
This variable is only used in one place as the struct,
so it can be defined with the struct type.
Resolves:
plat/rockchip/px30/drivers/pmu/pmu.c: In function 'rockchip_soc_sys_pwr_dm_suspend':
plat/rockchip/px30/drivers/pmu/pmu.c:977:23: error: array subscript 'struct psram_data_t[0]' is partly outside array bounds of 'void[8]' [-Werror=array-bounds]
977 | psram_boot_cfg->pm_flag &= ~PM_WARM_BOOT_BIT;
Change-Id: Ifbe42d11d0c7875f6cb23dc0b7ffb3f3f90c55a8 Signed-off-by: Scott Parlane <scott@parlanenz.com>
fix(gicv3/multichip): fix overflow caused by left shift
When spi_id_max is 5119, the expression `(spi_id_max - 4096U + 1U >> 5)`
evaluates to 32 leading to undefined behavior when using it to left
shift 1. Fix this undefined behavior.
```
large_shift: In expression 1 << (spi_id_max - 4096U + 1U >> 5), left
shifting by more than 31 bits has undefined behavior. The shift
amount, spi_id_max - 4096U + 1U >> 5, is as much as 32.
```
The arm,vexpress,config-bus DT binding restricts the possible (sub)node
names.
Adjust the current node names, to drop the unneeded address specifier,
and make the node names binding compliant.
Change-Id: Ic48c6969268c960ce92c8ec3a756ed1d89e61b08 Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Andre Przywara [Mon, 22 Aug 2022 14:50:22 +0000 (15:50 +0100)]
fix(fvp): fdts: fix memtimer subframe addressing
The arm,armv7-timer-mem DT binding documentation demands that the
#size-cells property should be <1> only.
Adjust the value to be <1> and drop the now needless leading 0 in the
frame's reg property. Convert to #address-cell = <1> on the way.
Also adjust the interrupts property to use the proper GIC macros.
Change-Id: Ia2224663b1e6aaa7cf94af777473641de6a840d2 Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Andre Przywara [Fri, 19 Aug 2022 15:21:29 +0000 (16:21 +0100)]
feat(fvp): fdts: update rtsm_ve DT files from the Linux kernel
The existing DT files for the base FVP model are having some issues,
that lead to warnings reported by the device tree compiler.
Those (and many other issues around (updated) DT binding compliance)
were fixed in the Linux kernel tree, so let's sync those files back into
TF-A.
We cannot copy the files "as is" for now, since we rely on certain custom
properties to be added (max-pwr-lvl in the PSCI node, SDEI nodes, etc).
Merge in the changed parts of the Linux kernel DT (from Linux v6.0-rc1),
and rework the base file to allow including the motherboard.dtsi
unchanged. This should make any future update less painful.
As this also affects the FVP VE boards (Cortex-A7 and Cortex-A5), since
they share the motherboard include file, fix them up as well.
Change-Id: I4f74d05e5583747f8849e32f246f74aeec7a9c60 Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Andre Przywara [Fri, 19 Aug 2022 10:01:16 +0000 (11:01 +0100)]
refactor(fvp): fdts: consolidate GICv2 base FVP DT files
The GICv2 and GICv3 version of the FVP DT files are unnecessarily split,
as the common part of the peripherals is the same: it's literally just
the interrupt controller node that is different.
Since the GICv3 versions now use a generic DT include file (without any
GIC node), let's reuse that for the GICv2 versions of the FVP as well.
We just add a separate fvp-base-gicv2.dtsi file which describes the
GICv2 interrupt controller. Also shorten the compatible string, since
the GICv2 binding documentation does not allow the current combination.
This allows to remove the mostly redundant nodes from the GICv2 .dts
file.
Change-Id: I9018031bb611fb00ca7dbefc1bff7d40c3f05819 Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Andre Przywara [Fri, 19 Aug 2022 10:00:37 +0000 (11:00 +0100)]
refactor(fvp): fdts: consolidate GICv3 base FVP DT files
The GICv2 and GICv3 version of the FVP DT files are unnecessarily split,
as the common part of the peripherals is the same: it's literally just
the interrupt controller node that is different.
To facilitate a unification, refactor the DT include files to explicitly
include a snippet with just the GICv3 description, and a generic base DT
file for the rest. This generic file can then be reused by the GICv2
versions later.
Since we can only have a /memreserve/ entry *before* any DT nodes, move
that line to each file, to allow including the GIC DT file separately.
Change-Id: I9ff357d3fe0ce46e280c30131aeae97a99631512 Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Andre Przywara [Fri, 19 Aug 2022 09:26:00 +0000 (10:26 +0100)]
feat(fvp): dts: drop 32-bit .dts files
Conceptually the DT is a hardware description, as such it's independent
from the instruction set that a DT client uses. So having separate DTs
for aarch32 and aarch64 does not make sense and is not needed.
Probably due to historic reasons (a Linux bug fixed in 2016 with Linux
commit ba6dea4f7ced, in Linux v4.8) the CPU reg property was using a
different size between aarch64 and aarch32, even though the size of it
is solely governed by the parent's #address-cells property.
Consolidate this to be always 2, and always use two cells to describe
the CPU's MPIDR register.
This removes the last difference of the -aarch32 versions of the FVP
DT files, so just remove all of them. The respective versions without
that suffix can now be used with AArch32 DT clients as well.
Also remove the respective part in the documentation.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Change-Id: I45d3a2cbba8e04595a741e1cf41900377952673e
For no real reason we were shipping two separate DT include files for the
base FVP motherboard peripherals, one for aarch32, one for aarch64.
There is no difference in the hardware description when using a
different instruction set, and the diff between the two files was about
a missing interrupt map for the 64-bit DT files.
Consolidate the situation by just using a single motherboard .dtsi file,
which relies on an interrupt map by the including files.
Provide that map in the two files where it was missing before, and
change the filenames to let all users include the same file now.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Change-Id: I19b77ecc8da9b4bfbd61d02f910b9ab05dbf92e9
Andre Przywara [Thu, 25 Aug 2022 11:59:10 +0000 (12:59 +0100)]
refactor(fvp_ve): fdts: prepare Cortex-A5 and A7 model DTs
The DT files for the Cortex-A5 and Cortex-A7 FVP models include the
shared rtsm_ve-motherboard.dtsi file, which we need to sync with the
upstream Linux version soon.
To prepare for its changed structure there, adjust the top-level
#address-cells and #size-cells properties to be compatible with the
expectations of the Linux version.
Also extend the interrupt map to cover all peripherals listed in the
motherboard file, and use the proper GIC macros to make them more
readable on the way.
Change-Id: I7d1493f1a200e8350530f912833f9ffcc5f94b21 Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Andre Przywara [Fri, 10 Dec 2021 18:22:09 +0000 (18:22 +0000)]
fix(fvp): fdts: unify and fix PSCI nodes
The PSCI DT nodes used for the various fvp-base model variants provide
explicit function IDs, as required for the pre-v0.2 PSCI specification.
This prevents them from being used from both AArch32 and AArch64 DT
clients, and using this version of the PSCI spec is long deprecated
anyway.
Remove the old compatible string and the function properties, to
force clients to use the standard function IDs as described in the PSCI
spec. sys_poweroff and sys_reset were never standardised or used anyway.
There should be no client software around that cannot deal with PSCI
v0.2.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Change-Id: Ie87deb9898eae79b7307c15bcefcd4b311d4dc22
Fix a syntax error in the delegated attestation service code.
Unfortunately, this build failure was not caught by the CI system
because right now lib/psa/delegated_attestation.c file is not getting
pulled in by any upstream platform. This will be addressed in a
separate patch.
Andre Przywara [Tue, 4 Oct 2022 12:56:49 +0000 (13:56 +0100)]
fix(rme): relax RME compiler requirements
Currently building TF-A for the FVP with RME enabled requires a
toolchain that understands the -march=armv8.6-a command line option,
even though we actually don't need any ARMv8.6 features from the
compiler.
Relax the requirement to use ARMv8.5, since this is what's the GCC
shipped with Ubuntu 20.04 understands. This is in line what the current
RMM implementation uses as well.
Change-Id: I3806dcff90319a87f003fe2c86b7cdcdebd625e4 Signed-off-by: Andre Przywara <andre.przywara@arm.com>
revert(cpus): "Revert workaround for A77 erratum 1800714"
Reinstate the workaround introduced in commit 9bbc03a6e0608a949d66d9da6db12a455b452bfb. The cited change to the SDEN
could not be found and there are no known problems with the workaround.
Merge changes I072fe5fe,I4066d476,Ie4af38b8,I730e7b04,Iac3356f8, ... into integration
* changes:
fix(psa): extend measured boot logging
fix(rss): determine the size of sw_type in RSS mboot metadata
fix(psa): align with original API in tf-m-extras
fix(rss): clear the message buffer
feat(tc): enable RSS backend based measured boot
feat(tc): increase maximum BL1/BL2/BL31 sizes
Harrison Mutai [Mon, 3 Oct 2022 11:48:35 +0000 (12:48 +0100)]
feat(cpu): add library support for Hunter ELP
Add basic CPU library code to support the Hunter ELP CPU in TF-A.
Hunter-ELP adds v9.2 architecture support and is derived from
Makalu-ELP. As such, the library code is adapted from the
Makalu-ELP support library.
Change-Id: I7e93b9af6b1f0bc4d08c3cf5caf071d2cbdbc89f Signed-off-by: Harrison Mutai <harrison.mutai@arm.com>
David Vincze [Mon, 11 Apr 2022 15:08:20 +0000 (17:08 +0200)]
feat(tc): add RSS-AP message size macro
Define the RSS_COMMS_PAYLOAD_MAX_SIZE macro. Its value is platform
specific and gives the largest message size which are exchanged
on the TC2 platform between RSS and AP.
Change-Id: Id831c282dc9a39755b82befead1a81767e217215 Signed-off-by: David Vincze <david.vincze@arm.com> Signed-off-by: Tamas Ban <tamas.ban@arm.com>
projects / arm-tf.git / log