This patch splits the el2_sysregs_context_save/restore functions
into multiple functions based on features. This will allow us to
selectively save and restore EL2 context registers based on
features enabled for a particular configuration.
For now feature build flags are used to decide which registers
to save and restore. The long term plan is to dynamically check
for features that are enabled and then save/restore registers
accordingly. Splitting el2_sysregs_context_save/restore functions
into smaller assembly functions makes that task easier. For more
information please take a look at:
https://trustedfirmware-a.readthedocs.io/en/latest/design_documents/context_mgmt_rework.html
Signed-off-by: Zelalem Aweke <zelalem.aweke@arm.com>
Change-Id: I1819a9de8b70fa35c8f45568908025f790c4808c
When SPMC is present at S-EL2, EL1 context registers don't need to be
initialized for Secure state. This patch makes sure that EL1 context
registers are initialized only for Non-secure state, and when SPMC is
not present at S-EL2
Signed-off-by: Zelalem Aweke <zelalem.aweke@arm.com>
Change-Id: I4a60b258c31ce5f6472a243e2687159cc495259b
In cm_prepare_el3_exit_ns, SCR_EL3.NS bit change (to non-secure) is not
committed before the EL2 restoration sequence happens.
At ICC_SRE_EL2 write in cm_el2_sysregs_context_restore, NS is still 0
from CPU perspective (with EEL2=0) which is an invalid condition and
triggers a fault. By adding ISB, SCR_EL3 gets synced with NS=1/EEL2=0
before ICC_SRE_EL2 write.
Signed-off-by: Olivier Deprez <olivier.deprez@arm.com>
Change-Id: Ie72a6152aa7729e66b3344c1b7b0749f54cafb6f
This patch enables access to the branch record buffer control registers
in non-secure EL2 and EL1 using the new build option ENABLE_BRBE_FOR_NS.
It is disabled for all secure world, and cannot be used with ENABLE_RME.
This option is disabled by default, however, the FVP platform makefile
enables it for FVP builds.
Signed-off-by: John Powell <john.powell@arm.com>
Change-Id: I576a49d446a8a73286ea6417c16bd0b8de71fca0
Previously the SCR_EL3.AMVOFFEN bit was set for all contexts, this
behavior is incorrect as it allows secure world to access the virtual
offset registers when it should not be able to. This patch only sets
AMVOFFEN for non-secure world.
Signed-off-by: John Powell <john.powell@arm.com>
Change-Id: I2c61fe0a8a0092df089f1cb2c0d8a45c8c8ad0d3
The current implementation uses plat_arm API under generic code.
"plat_arm" API is a convention used with Arm common platform layer
and is reserved for that purpose. In addition, the function has a
weak definition which is not encouraged in TF-A.
Henceforth, removing the weak API with a configurable macro "TWED_DELAY"
of numeric data type in generic code and simplifying the implementation.
By default "TWED_DELAY" is defined to zero, and the delay value need to
be explicitly set by the platforms during buildtime.
Signed-off-by: Jayanth Dodderi Chidanand <jayanthdodderi.chidanand@arm.com>
Change-Id: I25cd6f628e863dc40415ced3a82d0662fdf2d75a
As part of the RFC:
https://review.trustedfirmware.org/c/TF-A/trusted-firmware-a/+/13651,
this patch adds the 'cm_prepare_el3_exit_ns' function. The function is
a wrapper to 'cm_prepare_el3_exit' function for Non-secure state.
When EL2 sysregs context exists (CTX_INCLUDE_EL2_REGS is
enabled) EL1 and EL2 sysreg values are restored from the context
instead of directly updating the registers.
Signed-off-by: Zelalem Aweke <zelalem.aweke@arm.com>
Change-Id: I9b071030576bb05500d54090e2a03b3f125d1653
This patch splits the function 'cm_setup_context' into four
functions to make it more readable and easier to maintain.
The function is split into the following functions based on
the security state of the context.
- setup_context_common - performs common initializations
- setup_secure_context - performs Secure state specific
initializations
- setup_realm_context - performs Realm state specific
initializations
- setup_ns_context - performs Non-secure state specific
initializations
Signed-off-by: Zelalem Aweke <zelalem.aweke@arm.com>
Change-Id: Ie14a1c2fc6586087e7aa36537cf9064c80802f8f
The following registers are only accessible from secure state,
therefore don't need to be saved/restored during world switch.
- SDER32_EL2
- VSTCR_EL2
- VSTTBR_EL2
This patch removes these registers from EL2 context.
Signed-off-by: Zelalem Aweke <zelalem.aweke@arm.com>
Change-Id: I24d08aacb1b6def261c7b37d3e1265bb76adafdc
Replacing ARM_ARCH_AT_LEAST macro with feature specific build options
to prevent unconditional accesses to the registers during context save
and restore routines.
Registers are tightly coupled with features more than architecture
versions. Henceforth having a feature-specific build flag guarding the
respective registers, will restrict any undefined actions.
Signed-off-by: Jayanth Dodderi Chidanand <jayanthdodderi.chidanand@arm.com>
Change-Id: I809774df580530803c8a6e05a62d8d4de0910e02
During a transition to a higher EL some of the PSTATE bits are not set
by hardware, this means that their state may be leaked from lower ELs.
This patch sets those bits to a default value upon entry to EL3.
This patch was tested using a debugger to check the PSTATE values
are correctly set. As well as adding a test in the next patch to
ensure the PSTATE in lower ELs is still maintained after this change.
Change-Id: Ie546acbca7b9aa3c86bd68185edded91b2a64ae5
Signed-off-by: Daniel Boulby <daniel.boulby@arm.com>
In the next patch we add an extra step of setting the PSTATE
registers to a known state on el3 entry. In this patch we create
the function prepare_el3_entry to wrap the steps needed for before
el3 entry. For now this is only save_gp_pmcr_pauth_regs.
Change-Id: Ie26dc8d89bfaec308769165d2649e84d41be196c
Signed-off-by: Daniel Boulby <daniel.boulby@arm.com>
The HAFGRTR_EL2 register is UNDEFINED unless the CPU supports both
FEAT_FGT and FEAT_AMUv1. FEAT_FGT is mandatory for v8.6-A and upwards,
but FEAT_AMUv1 is optional (from v8.4-A upwards), and as such any
8.6-A cores today without support for FEAT_AMUv1 will trigger an
undefined instruction exception on accessing this register.
Currently ARM_ARCH_AT_LEAST macro has been used to associate with an
architecture extension allowing to access HAFGRTR_EL2 register. This
condition should be replaced with macros specific to individual
features. This patch adds a new set of macros "ENABLE_FEAT_FGT,
ENABLE_FEAT_AMUv1, ENABLE_FEAT_ECV" under build options to provide
controlled access to the HAFGRTR_EL2 register.
Further to ensure that the the build options passed comply
with the given hardware implementation, a feature detection mechanism,
checking whether build options match with the architecture is required
at bootime. This will be implemented and pushed later in a separate
patch.
Signed-off-by: Jayanth Dodderi Chidanand <jayanthdodderi.chidanand@arm.com>
Change-Id: Ie390f4babe233b8b09455290277edbddecd33ead
This patch adds two new compile time options to enable SME in TF-A:
ENABLE_SME_FOR_NS and ENABLE_SME_FOR_SWD for use in non-secure and
secure worlds respectively. Setting ENABLE_SME_FOR_NS=1 will enable
SME for non-secure worlds and trap SME, SVE, and FPU/SIMD instructions
in secure context. Setting ENABLE_SME_FOR_SWD=1 will disable these
traps, but support for SME context management does not yet exist in
SPM so building with SPD=spmd will fail.
The existing ENABLE_SVE_FOR_NS and ENABLE_SVE_FOR_SWD options cannot
be used with SME as it is a superset of SVE and will enable SVE and
FPU/SIMD along with SME.
Signed-off-by: John Powell <john.powell@arm.com>
Change-Id: Iaaac9d22fe37b4a92315207891da848a8fd0ed73
This patch adds a new context for realm world and realm world
awareness in context management.
Signed-off-by: Zelalem Aweke <zelalem.aweke@arm.com>
Signed-off-by: Subhasish Ghosh <subhasish.ghosh@arm.com>
Change-Id: Ic17469393603e789d7adc025880346bc3d6233d7
FEAT_HCX adds the extended hypervisor configuration register (HCRX_EL2)
and access to this register must be explicitly enabled through the
SCR_EL3.HXEn bit. This patch adds a new build flag ENABLE_FEAT_HCX to
allow the register to be accessed from EL2.
Signed-off-by: John Powell <john.powell@arm.com>
Change-Id: Ibb36ad90622f1dc857adab4b0d4d7a89456a522b
Introduced a build flag 'ENABLE_TRF_FOR_NS' to enable trace filter
control registers access in NS-EL2, or NS-EL1 (when NS-EL2 is
implemented but unused).
Change-Id: If3f53b8173a5573424b9a405a4bd8c206ffdeb8c
Signed-off-by: Manish V Badarkhe <Manish.Badarkhe@arm.com>
Introduced a build flag 'ENABLE_SYS_REG_TRACE_FOR_NS' to enable trace
system registers access in NS-EL2, or NS-EL1 (when NS-EL2 is
implemented but unused).
Change-Id: Idc1acede4186e101758cbf7bed5af7b634d7d18d
Signed-off-by: Manish V Badarkhe <Manish.Badarkhe@arm.com>
Introduced a build flag 'ENABLE_TRBE_FOR_NS' to enable trace buffer
control registers access in NS-EL2, or NS-EL1 (when NS-EL2 is
implemented but unused).
Change-Id: I285a672ccd395eebd377714c992bb21062a729cc
Signed-off-by: Manish V Badarkhe <Manish.Badarkhe@arm.com>
Trap bits of trace buffer control registers access are in
architecturally UNKNOWN state at boot hence
1. Initialized these bits to zero to prohibit trace buffer control
registers accesses in lower ELs (EL2, EL1) in all security states
when FEAT_TRBE is implemented
2. Also, these bits are RES0 when FEAT_TRBE is not implemented, and
hence setting it to zero also aligns with the Arm ARM reference
recommendation, that mentions software must writes RES0 bits with
all 0s
Change-Id: If2752fd314881219f232f21d8e172a9c6d341ea1
Signed-off-by: Manish V Badarkhe <Manish.Badarkhe@arm.com>
If SVE are enabled for both Non-secure and Secure world along with AMU
extension, then it causes the TAM_BIT in CPTR_EL3 to be set upon exit
from bl31. This restricts access to the AMU register set in normal
world. This fix maintains consistency in both TAM_BIT and CPTR_EZ_BIT
by saving and restoring CPTR_EL3 register from EL3 context.
Signed-off-by: Arunachalam Ganapathy <arunachalam.ganapathy@arm.com>
Change-Id: Id76ce1d27ee48bed65eb32392036377716aff087
Enables SVE support for the secure world via ENABLE_SVE_FOR_SWD.
ENABLE_SVE_FOR_SWD defaults to 0 and has to be explicitly set by the
platform. SVE is configured during initial setup and then uses EL3
context save/restore routine to switch between SVE configurations for
different contexts.
Reset value of CPTR_EL3 changed to be most restrictive by default.
Signed-off-by: Max Shvetsov <maksims.svecovs@arm.com>
Change-Id: I889fbbc2e435435d66779b73a2d90d1188bf4116
This patch adds setting MDCR_EL3.MCCD in 'el3_arch_init_common'
macro to disable cycle counting by PMCCNTR_EL0 in EL3 when
FEAT_PMUv3p7 is implemented. This fixes failing test
'Leak PMU CYCLE counter values from EL3 on PSCI suspend SMC'
on FVP models with 'has_v8_7_pmu_extension' parameter set to
1 or 2.
Signed-off-by: Alexei Fedorov <Alexei.Fedorov@arm.com>
Change-Id: I2ad3ef501b31ee11306f76cb5a61032ecfd0fbda
Since there is a secure and non-secure version of the timer registers
there is no need to preserve their context in EL3.
With that, following registers were removed from EL3 save/restore
routine:
cnthps_ctl_el2
cnthps_tval_el2
cnthps_cval_el2
cnthvs_ctl_el2
cnthvs_tval_el2
cnthvs_cval_el2
cnthp_ctl_el2
cnthp_cval_el2
cnthp_tval_el2
cnthv_ctl_el2
cnthv_cval_el2
cnthv_tval_el2
Signed-off-by: Max Shvetsov <maksims.svecovs@arm.com>
Change-Id: I6e2fc09c74a7375c4fccc11f12af4e39e6dc616b
ARMv8.6 adds virtual offset registers to support virtualization of the
event counters in EL1 and EL0. This patch enables support for this
feature in EL3 firmware.
Signed-off-by: John Powell <john.powell@arm.com>
Change-Id: I7ee1f3d9f554930bf5ef6f3d492e932e6d95b217
For SoCs which do not implement RAS, use DSB as a barrier to
synchronize pending external aborts at the entry and exit of
exception handlers. This is needed to isolate the SErrors to
appropriate context.
However, this introduces an unintended side effect as discussed
in the https://review.trustedfirmware.org/c/TF-A/trusted-firmware-a/+/3440
A summary of the side effect and a quick workaround is provided as
part of this patch and summarized here:
The explicit DSB at the entry of various exception vectors in BL31
for handling exceptions from lower ELs can inadvertently trigger an
SError exception in EL3 due to pending asyncrhonouus aborts in lower
ELs. This will end up being handled by serror_sp_elx in EL3 which will
ultimately panic and die.
The way to workaround is to update a flag to indicate if the exception
truly came from EL3. This flag is allocated in the cpu_context
structure. This is not a bullet proof solution to the problem at hand
because we assume the instructions following "isb" that help to update
the flag (lines 100-102 & 139-141) execute without causing further
exceptions.
Change-Id: I4d345b07d746a727459435ddd6abb37fda24a9bf
Signed-off-by: Madhukar Pappireddy <madhukar.pappireddy@arm.com>
This patch provides the following changes:
- Adds definition for FEAT_MTE3 value in ID_AA64PFR1_EL1 register
- Enables Memory Tagging Extension for FEAT_MTE3.
Change-Id: I735988575466fdc083892ec12c1aee89b5faa472
Signed-off-by: Alexei Fedorov <Alexei.Fedorov@arm.com>
Include EL2 registers related to SPE in EL2 context save/restore
routines if architecture supports it and platform wants to use these
features in Secure world.
Change-Id: Ie01a2c38fa5f6c907276eddec120fdfb222561a6
Signed-off-by: Arunachalam Ganapathy <arunachalam.ganapathy@arm.com>
Include EL2 registers related to Nested Virtualization in EL2 context
save/restore routines if architecture supports it and platform wants to
use these features in Secure world.
Change-Id: If006ab83bbc2576488686f5ffdff88b91adced5c
Signed-off-by: Arunachalam Ganapathy <arunachalam.ganapathy@arm.com>
AArch64-only platforms do not implement AArch32 at EL1 and higher ELs.
In such cases the build option CTX_INCLUDE_AARCH32_REGS is set to 0.
So don't save/restore aarch32 system registers in el2_sysregs_context
save/restore routines if CTX_INCLUDE_AARCH32_REGS is set to 0.
Change-Id: I229cdd46136c4b4bc9623b02eb444d904e09ce5a
Signed-off-by: Arunachalam Ganapathy <arunachalam.ganapathy@arm.com>
Usually, C has no problem up-converting types to larger bit sizes. MISRA
rule 10.7 requires that you not do this, or be very explicit about this.
This resolves the following required rule:
bl1/aarch64/bl1_context_mgmt.c:81:[MISRA C-2012 Rule 10.7 (required)]<None>
The width of the composite expression "0U | ((mode & 3U) << 2U) | 1U |
0x3c0U" (32 bits) is less that the right hand operand
"18446744073709547519ULL" (64 bits).
This also resolves MISRA defects such as:
bl2/aarch64/bl2arch_setup.c:18:[MISRA C-2012 Rule 12.2 (required)]
In the expression "3U << 20", shifting more than 7 bits, the number
of bits in the essential type of the left expression, "3U", is
not allowed.
Further, MISRA requires that all shifts don't overflow. The definition of
PAGE_SIZE was (1U << 12), and 1U is 8 bits. This caused about 50 issues.
This fixes the violation by changing the definition to 1UL << 12. Since
this uses 32bits, it should not create any issues for aarch32.
This patch also contains a fix for a build failure in the sun50i_a64
platform. Specifically, these misra fixes removed a single and
instruction,
92407e73 and x19, x19, #0xffffffff
from the cm_setup_context function caused a relocation in
psci_cpus_on_start to require a linker-generated stub. This increased the
size of the .text section and caused an alignment later on to go over a
page boundary and round up to the end of RAM before placing the .data
section. This sectionn is of non-zero size and therefore causes a link
error.
The fix included in this reorders the functions during link time
without changing their ording with respect to alignment.
Change-Id: I76b4b662c3d262296728a8b9aab7a33b02087f16
Signed-off-by: Jimmy Brisson <jimmy.brisson@arm.com>
Make this more scalable by explicitly checking internal and hardware
states at run_time
Signed-off-by: Ruari Phipps <ruari.phipps@arm.com>
Change-Id: I1c6ed1c1badb3538a93bff3ac5b5189b59cccfa1
As per latest mailing communication [1], we decided to
update AT speculative workaround implementation in order to
disable page table walk for lower ELs(EL1 or EL0) immediately
after context switching to EL3 from lower ELs.
Previous implementation of AT speculative workaround is available
here: 45aecff00
AT speculative workaround is updated as below:
1. Avoid saving and restoring of SCTLR and TCR registers for EL1
in context save and restore routine respectively.
2. On EL3 entry, save SCTLR and TCR registers for EL1.
3. On EL3 entry, update EL1 system registers to disable stage 1
page table walk for lower ELs (EL1 and EL0) and enable EL1
MMU.
4. On EL3 exit, restore SCTLR and TCR registers for EL1 which
are saved in step 2.
[1]:
https://lists.trustedfirmware.org/pipermail/tf-a/2020-July/000586.html
Change-Id: Iee8de16f81dc970a8f492726f2ddd57e7bd9ffb5
Signed-off-by: Manish V Badarkhe <Manish.Badarkhe@arm.com>
SCTLR and TCR registers of EL1 plays role in enabling/disabling of
page table walk for lower ELs (EL0 and EL1).
Hence re-arranged EL1 context offsets to have SCTLR and TCR registers
values one after another in the stack so that these registers values
can be saved and restored using stp and ldp instruction respectively.
Change-Id: Iaa28fd9eba82a60932b6b6d85ec8857a9acd5f8b
Signed-off-by: Manish V Badarkhe <Manish.Badarkhe@arm.com>
As per latest mailing communication [1], we decided
not to update SCTLR and TCR registers in EL1 and EL2 context
restore routine when AT speculative workaround is enabled
hence reverted the changes done as part of this commit: 45aecff00.
[1]:
https://lists.trustedfirmware.org/pipermail/tf-a/2020-July/000586.html
Change-Id: I8c5f31d81fcd53770a610e302a5005d98772b71f
Signed-off-by: Manish V Badarkhe <Manish.Badarkhe@arm.com>
This patch adds a build config 'RAS_TRAP_LOWER_EL_ERR_ACCESS' to set
SCR_EL3.TERR during CPU boot. This bit enables trapping RAS register
accesses from EL1 or EL2 to EL3.
RAS_TRAP_LOWER_EL_ERR_ACCESS is disabled by default.
Signed-off-by: Varun Wadekar <vwadekar@nvidia.com>
Change-Id: Ifb0fb0afedea7dd2a29a0b0491a1161ecd241438
Enhanced Counter Virtualization, ECV, is an architecture extension introduced
in ARMv8.6. This extension allows the hypervisor, at EL2, to setup
self-synchronizing views of the timers for it's EL1 Guests. This patch pokes the
control register to enable this extension when booting a hypervisor at EL2.
Change-Id: I4e929ecdf400cea17eff1de5cf8704aa7e40973d
Signed-off-by: Jimmy Brisson <jimmy.brisson@arm.com>
The Fine Grained Traps (FGT) architecture extension was added to aarch64 in
ARMv8.6. This extension primarily allows hypervisors, at EL2, to trap specific
instructions in a more fine grained manner, with an enable bit for each
instruction. This patch adds support for this extension by enabling the
extension when booting an hypervisor at EL2.
Change-Id: Idb9013ed118b6a1b7b76287237096de992ca4da3
Signed-off-by: Jimmy Brisson <jimmy.brisson@arm.com>
This patch enables the v8.6 extension to add a delay before WFE traps
are taken. A weak hook plat_arm_set_twedel_scr_el3 has been added in
plat/common/aarch64/plat_common.c that disables this feature by default
but platform-specific code can override it when needed.
The only hook provided sets the TWED fields in SCR_EL3, there are similar
fields in HCR_EL2, SCTLR_EL2, and SCTLR_EL1 to control WFE trap delays in
lower ELs but these should be configured by code running at EL2 and/or EL1
depending on the platform configuration and is outside the scope of TF-A.
Signed-off-by: John Powell <john.powell@arm.com>
Change-Id: I0a9bb814205efeab693a3d0a0623e62144abba2d
Removing FPEXC32_EL2 from the register save/restore routine for EL2
registers since it is already a part of save/restore routine for
fpregs.
Signed-off-by: Max Shvetsov <maksims.svecovs@arm.com>
Change-Id: I5ed45fdbf7c8efa8dcfcd96586328d4f6b256bc4
During context switching from higher EL (EL2 or higher)
to lower EL can cause incorrect translation in TLB due to
speculative execution of AT instruction using out-of-context
translation regime.
Workaround is implemented as below during EL's (EL1 or EL2)
"context_restore" operation:
1. Disable page table walk using SCTLR.M and TCR.EPD0 & EPD1
bits for EL1 or EL2 (stage1 and stage2 disabled)
2. Save all system registers except TCR and SCTLR (for EL1 and EL2)
3. Do memory barrier operation (isb) to ensure all
system register writes are done.
4. Restore TCR and SCTLR registers (for EL1 and EL2)
Errata details are available for various CPUs as below:
Cortex-A76: 1165522
Cortex-A72: 1319367
Cortex-A57: 1319537
Cortex-A55: 1530923
Cortex-A53: 1530924
More details can be found in mail-chain:
https://lists.trustedfirmware.org/pipermail/tf-a/2020-April/000445.html
Currently, Workaround is implemented as build option which is default
disabled.
Signed-off-by: Manish V Badarkhe <Manish.Badarkhe@arm.com>
Change-Id: If8545e61f782cb0c2dda7ffbaf50681c825bd2f0
TPIDR_EL2 is missing from the EL2 state register save/restore
sequence. This patch adds it to the context save restore routines.
Signed-off-by: Olivier Deprez <olivier.deprez@arm.com>
Change-Id: I35fc5ee82f97b72bcedac57c791312e7b3a45251
This patch adds EL2 registers that are supported up to ARMv8.6.
ARM_ARCH_MINOR has to specified to enable save/restore routine.
Note: Following registers are still not covered in save/restore.
* AMEVCNTVOFF0<n>_EL2
* AMEVCNTVOFF1<n>_EL2
* ICH_AP0R<n>_EL2
* ICH_AP1R<n>_EL2
* ICH_LR<n>_EL2
Change-Id: I4813f3243e56e21cb297b31ef549a4b38d4876e1
Signed-off-by: Max Shvetsov <maksims.svecovs@arm.com>
NOTE: Not all EL-2 system registers are saved/restored.
This subset includes registers recognized by ARMv8.0
Change-Id: I9993c7d78d8f5f8e72d1c6c8d6fd871283aa3ce0
Signed-off-by: Jose Marinho <jose.marinho@arm.com>
Signed-off-by: Olivier Deprez <olivier.deprez@arm.com>
Signed-off-by: Artsem Artsemenka <artsem.artsemenka@arm.com>
Signed-off-by: Max Shvetsov <maksims.svecovs@arm.com>
The Secure Configuration Register is 64-bits in AArch64 and 32-bits in
AArch32. Use u_register_t instead of unsigned int to reflect this.
Change-Id: I51b69467baba36bf0cfaec2595dc8837b1566934
Signed-off-by: Louis Mayencourt <louis.mayencourt@arm.com>
Since BL31 PROGBITS and BL31 NOBITS sections are going to be
in non-adjacent memory regions, potentially far from each other,
some fixes are needed to support it completely.
1. adr instruction only allows computing the effective address
of a location only within 1MB range of the PC. However, adrp
instruction together with an add permits position independent
address of any location with 4GB range of PC.
2. Since BL31 _RW_END_ marks the end of BL31 image, care must be
taken that it is aligned to page size since we map this memory
region in BL31 using xlat_v2 lib utils which mandate alignment of
image size to page granularity.
Change-Id: I3451cc030d03cb2032db3cc088f0c0e2c84bffda
Signed-off-by: Madhukar Pappireddy <madhukar.pappireddy@arm.com>
Even though ERET always causes a jump to another address, aarch64 CPUs
speculatively execute following instructions as if the ERET
instruction was not a jump instruction.
The speculative execution does not cross privilege-levels (to the jump
target as one would expect), but it continues on the kernel privilege
level as if the ERET instruction did not change the control flow -
thus execution anything that is accidentally linked after the ERET
instruction. Later, the results of this speculative execution are
always architecturally discarded, however they can leak data using
microarchitectural side channels. This speculative execution is very
reliable (seems to be unconditional) and it manages to complete even
relatively performance-heavy operations (e.g. multiple dependent
fetches from uncached memory).
This was fixed in Linux, FreeBSD, OpenBSD and Optee OS:
679db7080129fb48ace43a08873eceabfd092aa1
It is demonstrated in a SafeSide example:
https://github.com/google/safeside/blob/master/demos/eret_hvc_smc_wrapper.cchttps://github.com/google/safeside/blob/master/kernel_modules/kmod_eret_hvc_smc/eret_hvc_smc_module.c
Signed-off-by: Anthony Steinhauser <asteinhauser@google.com>
Change-Id: Iead39b0b9fb4b8d8b5609daaa8be81497ba63a0f
Zelalem Aweke
Olivier Deprez
johpow01
Jayanth Dodderi Chidanand
Daniel Boulby
Manish V Badarkhe
Arunachalam Ganapathy
Max Shvetsov
Alexei Fedorov
Madhukar Pappireddy
Jimmy Brisson
Ruari Phipps
Varun Wadekar
Louis Mayencourt
Mark Dykes
Anthony Steinhauser