Cortex-X3 erratum 2372204 is a Cat B erratum that applies to revisions
r0p0 and r1p0. It is fixed in r1p1.
The workaround is to set bit[40] of CPUACTLR2_EL1 to disable folding
of demand requests into older prefetches with L2 miss requests
outstanding.
SDEN can be found here:
https://developer.arm.com/documentation/SDEN2055130/latest
Change-Id: Iad28f8625c84186fbd8049406d139d4f15c6e069
Signed-off-by: Bipin Ravi <biprav01@u203721.austin.arm.com>
* changes:
refactor(tc): correlate secure world addresses with platform_def
feat(tc): add memory node in the device tree
feat(tc): pass the DTB address to BL33 in R0
feat(tc): add arm_ffa node in dts
chore(tc): add dummy entropy to speed up the Linux boot
feat(tc): choose the DPU address and irq based on the target
feat(tc): add SCMI power domain and IOMMU toggles
refactor(tc): move the FVP RoS to a separate file
feat(tc): factor in FVP/FPGA differences
feat(tc): introduce an FPGA subvariant and TC3 CPUs
feat(tc): add TC3 platform definitions
refactor(tc): sanitise the device tree
feat(tc): add PMU entry
feat(tc): allow booting from DRAM
chore(tc): remove unused hdlcd
feat(tc): add firmware update secure partition
feat(tc): add spmc manifest with trusty sp
refactor(tc): unify all the spmc manifests
feat(arm): add trusty_sp_fw_config build option
fix(tc): do not enable MPMM and Aux AMU counters always
fix(tc): correct interrupts
feat(tc): interrupt numbers for `smmu_700`
feat(tc): enable gpu/dpu scmi power domain and also gpu perf domain
Similarly to the memory node in the NS device tree, platform_def already
defines all the necessary values to populate the spmc manifest and NS
related entries automatically. Use the macros directly so any changes
can propagate automatically.
The result of this is that TC3 and above get correct secure world
manifests automatically. They were previously broken.
One "breaking" change is that the FWU region moves. This should have
happened previously but it was missed when the secure portion of DRAM
was increased, leaving it in secure memory. This was caught when going
over the definitions and correlating them should prevent this in the
future.
Signed-off-by: Boyan Karatotev <boyan.karatotev@arm.com>
Change-Id: I1415e402be8c70f5e22f28eabddcb53298c57a11
With new TC revisions, memory banks move around which requires an update
in platform_def. It also requires an update in the device tree which
doesn't come naturally. To avoid this, add the memory node such that it
uses the macros defined in platform_def.
By doing this we can put u-boot out of its misery in trying to come up
with the correct memory node and tf-a's device tree becomes complete.
Signed-off-by: Boyan Karatotev <boyan.karatotev@arm.com>
Change-Id: Ia92cc6931abb12be2856ac3fb1455e4f3005b326
The DTB that tf-a loads is already used in BL33 directly with the
address hardcoded. As this address is prone to changing, pass it forward
so we can avoid breakage in the future.
Signed-off-by: Boyan Karatotev <boyan.karatotev@arm.com>
Change-Id: I7a42f72ecc00814b9f0a4bf5605d70cb53ce2ff4
For u-boot to use the device tree itself it needs to know about the
arm_ffa module. This is not relevant to linux but it doesn't hurt as it
won't use it.
Signed-off-by: Boyan Karatotev <boyan.karatotev@arm.com>
Change-Id: I6e75659e4950c62ce7377dc7941225eb5d7a3d8d
If the kernel is post 5.19 and is configured with
CONFIG_RANDOM_TRUST_BOOTLOADER=y then entropy can be passed to
Linux via the device tree. This avoids delaying the Linux boot
waiting for entropy. This is particularly noticeable when
booting android but also speeds up the generation of the ssl
certificates.
Signed-off-by: Ben Horgan <ben.horgan@arm.com>
Change-Id: I4c6136c54f0e971802a2a9de9f88cd32b610dce9
Currently there are two configurations for DPU
Config 1: Address - 0x2CC0_0000 IRQ - 101
Config 2: Address - 0x40_0000_0000 IRQ - 547
Config 1 is used by all FPGA and TC0, TC1 and TC2 FVPs
Config 2 is used by TC3 FVP currently
Signed-off-by: Davidson K <davidson.kumaresan@arm.com>
Change-Id: If0097441b6ab90f58911df032e45f6bf06fc7909
Compile-time controls have been added for the following:
* SCMI power domain use for DPU and GPU.
* SMMU-700: planned rework required to use IOMMU correctly
for DPU and GPU.
These will allow easier experimentation in the future without
ad-hoc changes needed in the dts file for any sort of analysis
that requires testing different paths.
For TC3 however, the DPU is in an always on power domain so SCMI power
domains are not supported.
Co-developed-by: Tintu Thomas <tintu.thomas@arm.com>
Signed-off-by: Kshitij Sisodia <kshitij.sisodia@arm.com>
Change-Id: If6179a3e4784c1b69f0338a8d52b552452c0eac1
In trying to use the same DTS for the FVP and FPGA subvariants we need
to keep track of what is different. Move the FVP RoS, which is different
to the FPGA's, to reduce the number of ifdefs and make FVP-only changes
easier.
Signed-off-by: Boyan Karatotev <boyan.karatotev@arm.com>
Change-Id: Ib7999d3e39de55ab4a30e68dd81f95120be15a8c
Even though the FVP and FPGA are meant to be identical their RoS's (rest
of system) are different. Factor these in so the device tree works for
both. The differences are:
* addresses of GIC and UART
* displays (FPGA uses 4k)
* ethernet devices and SD card (it's non removable on the FPGA)
Their frequencies are also different. The FVP simulates certain
frequencies but isn't very sensitive when we disregard them. To keep
code similar, update them with the FPGA values. This keeps working on
FVP even if slightly incorrect.
Also add an option for the DPU to either use fixed clocks or SCMI set
clocks, hidden behind a flag. This is useful during bringup and because
SCMI may not necessarily work on FPGA.
Co-developed-by: Kshitij Sisodia <kshitij.sisodia@arm.com>
Co-developed-by: Arunachalam Ganapathy <arunachalam.ganapathy@arm.com>
Co-developed-by: Usama Arif <usama.arif@arm.com>
Co-developed-by: Angel Rodriguez Garcia <angel.rodriguezgarcia@arm.com>
Signed-off-by: Boyan Karatotev <boyan.karatotev@arm.com>
Change-Id: Ic7a4bfc302673a3a6571757e23a9e6184fba2a13
TC is getting an FPGA port alongside the FVP. It is meant to be
identical, but the core configurations on TC2 differ (there are 14 in an
odd arrangement).
Introduce these differences and gate them behind a new TARGET_FLAVOUR
flag which defaults to FVP for compatibility.
While updating CPUs, it's a good time to do TC3 too. It has different
cores in a different configuration again, so it needs different capacity
values. Those have been derived using GeekBench 6.0 ST on the FPGA.
Finally GPU and DPU power domains are 1 above the CPUs so make that
relative.
In the end, the big/mid/little configurations are:
* TC2 FVP: 1/3/4
* TC2 FPGA: 2/3/5/4 (the 3 is a big "min" core)
* TC3 both: 2/4/2 (with new capacities)
Co-developed-by: Tintu Thomas <tintu.thomas@arm.com>
Co-developed-by: Kshitij Sisodia <kshitij.sisodia@arm.com>
Signed-off-by: Boyan Karatotev <boyan.karatotev@arm.com>
Change-Id: I3c3a10d6727f5010fd9026a404df27e9262dff6b
TC3 is a little different from TC2:
* new address for its second DRAM bank
* new CPUs
* a few interrupts have changed
* new SCP MHU base address.
* utility space address (needed for MPAM) is different
* no CMN (and therefore cmn-pmu)
* the uart clock is different
This requires the dts to be different between revisions for the first
time. Introduce a tc_vers.dtsi that includes only definitions for things
that are different.
Signed-off-by: Tintu Thomas <tintu.thomas@arm.com>
Signed-off-by: Boyan Karatotev <boyan.karatotev@arm.com>
Change-Id: I2940d87a69ea93502b7f5a22a539e4b70a63e827
We have lots of errors in our device tree when running dt-validate.
Remove the majority so that dt-validate-ing is useful now.
Do this by renaming nodes to conform to spec, making addresses lowercase
with no 0x at the front, and removing nodes that shouldn't be there.
Signed-off-by: Boyan Karatotev <boyan.karatotev@arm.com>
Change-Id: I1840f0f5de34a56ee240c07eff08d73c856b338e
TC has PMUs with interrupts in all cores and Linux needs to be told
about them.
Signed-off-by: Boyan Karatotev <boyan.karatotev@arm.com>
Change-Id: Ice0e6dab396b90c05f4b9668623ba7b3556a53ac
In some occasions it is useful to boot with the rest of system (RoS)
disabled. With no RoS there's no flash so we need to put images
somewhere and that's in the DRAM1 bank. If we want to access it it needs
to be mapped to memory.
Signed-off-by: Boyan Karatotev <boyan.karatotev@arm.com>
Change-Id: I45e0fbb016e8f615d41b6ad9da0d1e7b466ece72
The hdlcd device tree node is not in use for any TC incarnation. The DPU
replaces it. So drop it.
Signed-off-by: Boyan Karatotev <boyan.karatotev@arm.com>
Change-Id: I5393435e36d8307bef909a6519cb40305b77f0cf
Firmware update is a trusted service secure partition that implements
the PSA firmware update specification. It executes in the secure world
in total compute platform. To make it fit with Op-tee we need to reduce
its available memory.
Also, reserve 4 MB for stmm communication used for firmware update.
The firmware update secure partition and u-boot communicates using the
stmm communication layer and it needs a dedicated memory region.
Co-developed-by: Sergio Alves <sergio.dasilvalves@arm.com>
Co-developed-by: Davidson K <davidson.kumaresan@arm.com>
Signed-off-by: Tudor Cretu <tudor.cretu@arm.com>
Signed-off-by: Boyan Karatotev <boyan.karatotev@arm.com>
Change-Id: I0427549845f6c7650b8ef4e450d387fe9702a847
The manifests describe the same hardware layout with only the secure
partitions being different. Factor it out so it can be shared and only
add the VM information separately.
This has some deliberate side effects: the test configuration gets the
full secure memory address space and drops the 0x7000000 region as that
was accidentally copied over from the FVP platform and doesn't apply to
TC.
Also optee unconditionally gets the smaller mem_size as it's been
working fine and simplifies the manifest.
Small touch up is that mem_size-s are now in hex but otherwise the same
number.
Signed-off-by: Boyan Karatotev <boyan.karatotev@arm.com>
Change-Id: Iea23f9769235eea32afa374952b9a0e4f6d3e9a1
Also increase add PLAT_ARM_SP_MAX_SIZE to override the default
ARM_SP_MAX_SIZE to support Trusty image and move OPTEE_SP_FW_CONFIG
documentation to build-internals.rst as it's not externally set-able.
Signed-off-by: Arunachalam Ganapathy <arunachalam.ganapathy@arm.com>
Signed-off-by: Boyan Karatotev <boyan.karatotev@arm.com>
Change-Id: Ief90ae9113d32265ee2200f35f3e517b7b9a4bea
There are requirements in which the MPMM and Auxiliary AMU counters have
to be disabled. Hence removing the "override" here which helps in
disabling them during the build.
Change-Id: I2c0a808d5d9968082a508a9206e34f7a57f2e33a
Signed-off-by: Davidson K <davidson.kumaresan@arm.com>
The gic and trbe0 are listed as active high, but the spec says they are
triggered on active low. Correct according to the spec.
While we're at it, convert all interrupts to use the macros so hopefully
no such confusion happens again.
Signed-off-by: Boyan Karatotev <boyan.karatotev@arm.com>
Change-Id: I2fc01cf0a34b031b95219b9656b613a19a2e9b2a
The scmi-perf-domain property is a custom property while we
wait for proper support in the kernel for using the scmi_pm_domain
driver and scmi_perf_domain driver at the same time.
GPU operating points are now in the SCP firmware.
Change-Id: Ib6d8f52c8bf69194b1d2da4e065a34c4a341c221
Signed-off-by: Ben Horgan <ben.horgan@arm.com>
This is an effort to optimise the unused members in the cpu_context_t
structure. TF-A statically allocates memory for context entry for
each wolrd per PE. Some of the members in this struct are not used
for all the build combinations.
RMM in particular, is not using this context member and henceforth
removing them.
Change-Id: Ia5bf9c8dfef6e856ba6d88fa678876c704d42858
Signed-off-by: Jayanth Dodderi Chidanand <jayanthdodderi.chidanand@arm.com>
Similar to the refactoring process followed for EL2 system registers,
moving the save and restore routines of EL1 system registers into C
file, thereby reducing assembly code.
Change-Id: Ib59fbbe2eef2aa815effe854cf962fc4ac62a2ae
Signed-off-by: Jayanth Dodderi Chidanand <jayanthdodderi.chidanand@arm.com>
This change includes adding new CORSTONE1000_WITH_BL32 preprocessor
flag on the basis of NEED_BL32 flag. This flag allows us to run the
TF-A with or without loading BL32 image. This feature is required to
add the support of Corstone-1000 FVP in TF-A open CI.
After this, we can run the TF-A tftf tests with or without
executing BL32 image, which is optee in case of Corstone-1000.
Signed-off-by: Harsimran Singh Tungal <harsimransingh.tungal@arm.com>
Change-Id: Idacbd3883473473841481a2032314db8c9715b1f
Default values for toolchain tools are instead provided by the main
toolchain makefile, rather than a parent makefile. This was an oversight
from a previous version of the original toolchain refactor patch.
Change-Id: I75752ed7874b36e1c679d94292a2664e234c484b
Signed-off-by: Chris Kay <chris.kay@arm.com>
Address the coding style issue that arose from patch [1], which
was inadvertently overlooked during the CI check.
[1]: https://review.trustedfirmware.org/c/TF-A/trusted-firmware-a/+/26263
Signed-off-by: Manish V Badarkhe <Manish.Badarkhe@arm.com>
Change-Id: I280766fddf0e9e366bb2376c52a6907093b0d958
* changes:
docs: update FVP TC2 model version and build (11.23/17)
fix(tc): increase BL2 maximum size limit
refactor(tc): update platform tests
feat(rss): add defines for 'type' range and use them in psa_call()
feat(rss): adjust parameter packing to match TF-M changes
refactor(tc): remap console logs
This patch does following two changes
- Create a separate routine for saving/restoring GIC el2 system registers
- To access ICC_SRE_EL2 register there was a workaround to set
SCR_EL3.NS before accessing it. This was required because SCR_EL3.EEL2
was zero. But with commit f105dd5fa this bit has been set to one early
on in booting process for a system with FEAT_SEL2 present and S-EL2
enabled.
However, we still need the workaround for a system which needs
save/restore of EL2 registers without secure EL2 being enabled e.g.
system with Non-secure and Realm world present.
Signed-off-by: Manish Pandey <manish.pandey2@arm.com>
Change-Id: I8d55c3dc6a17c4749748822d4a738912c1e13298
