GitLab

Consider a RAS scenario:
Enter EL3 by sync exception, then call spm_mm_sp_call() enter
EL0s to handle this error, then call sdei_dispatch_event() to
inform OS. Finally, return back to OS from sync exception flow.
Similar flow is sgi_ras_intr_handler() in sgi_ras.c.

The icc_pmr_el1 register will be change in above flow:
1 cm_el1_sysregs_context_save(NON_SECURE);
  -> ehf_exited_normal_world();
    ##icc_pmr_el1: 0xf8 => 0x80
2 spm_mm_sp_call();
3 sdei_dispatch_event();
4 ehf_activate_priority(sdei_event_priority(map));
    ##icc_pmr_el1: 0x80 => 0x60
5 restore_and_resume_ns_context();
  -> ehf_exited_normal_world();
     ##return due to has_valid_pri_activations(pe_data) == 1
6 ehf_deactivate_priority(sdei_event_priority(map));
    ##icc_pmr_el1: 0x60 => 0x80
The icc_pmr_el1 was rewrote from 0xf8 to 0x80. This issue will
result in OS hang when eret to OS from RAS flow.

Move ehf_activate_priority(sdei_event_priority(map)) after
restore_and_resum...

The SDEI specification now says that during an SDEI
event handler dispatch the SPSR should be set according
to the TakeException() pseudocode function defined in
the Arm Architecture Reference Manual. This patch sets
the SPSR according to the function given in
ARM DDI 0487F.c page J1-7635

Change-Id: Id2f8f2464fd69c701d81626162827e5c4449b658
Signed-off-by: Daniel Boulby <daniel.boulby@arm.com>

Instead of having a custom implementation of setjmp() and longjmp() it
is better to follow the C standard.

The comments in setjmp.h are no longer needed as there are no deviations
from the expected one, so they have been removed.

All SDEI code that relied on them has been fixed to use the new function
prototypes and structs.

Change-Id: I6cd2e21cb5a5bcf81ba12283f2e4c067bd5172ca
Signed-off-by: Antonio Nino Diaz <antonio.ninodiaz@arm.com>

Enforce full include path for includes. Deprecate old paths.

The following folders inside include/lib have been left unchanged:

- include/lib/cpus/${ARCH}
- include/lib/el3_runtime/${ARCH}

The reason for this change is that having a global namespace for
includes isn't a good idea. It defeats one of the advantages of having
folders and it introduces problems that are sometimes subtle (because
you may not know the header you are actually including if there are two
of them).

For example, this patch had to be created because two headers were
called the same way: e0ea0928 ("Fix gpio includes of mt8173 platform
to avoid collision."). More recently, this patch has had similar
problems: 46f9b2c3 ("drivers: add tzc380 support").

This problem was introduced in commit 4ecca339

 ("Move include and
source files to logical locations"). At that time, there weren't too
many headers so it wasn't a real issue. However, time has shown that
this creates problems.

Platforms that want to preserve the way they include headers may add the
removed paths to PLAT_INCLUDES, but this is discouraged.

Change-Id: I39dc53ed98f9e297a5966e723d1936d6ccf2fc8f
Signed-off-by: Antonio Nino Diaz <antonio.ninodiaz@arm.com>

Secure world execution nearly always expect a controlled exit to
Non-secure world. SDEI interrupts, although targets EL3, occur on behalf
of Non-secure world, and may have higher priority than Secure world
interrupts. Therefore they might preempt Secure execution, and yield
execution to Non-secure SDEI handler. Upon completion of SDEI event
handling (regardless of whether it's COPLETE or COMPLETE_AND_RESUME), we
must resume Secure execution if it was preempted.

Change-Id: I6edd991032588588427ba2fe6c3d7668f7080e3d
Signed-off-by: Jeenu Viswambharan <jeenu.viswambharan@arm.com>

The event lock for a shared event was being unlocked twice, and the
locking sequence for event complete was misplaced. This patch fixes both
issues.

Change-Id: Ie2fb15c6ec240af132d7d438946ca160bd5c63dc
Signed-off-by: Jeenu Viswambharan <jeenu.viswambharan@arm.com>

These changes address most of the required MISRA rules. In the process,
some from generic code is also fixed.

No functional changes.

Change-Id: I6235a355e006f0b1c7c1c4d811b3964a64d0434f
Signed-off-by: Jeenu Viswambharan <jeenu.viswambharan@arm.com>

The function end_sdei_explicit_dispatch() was intended to be
end_sdei_synchronous_dispatch() which does the opposite of
begin_sdei_synchronous_dispatch(). This patch fixes that.

No functional changes.

Change-Id: I141bd91eb342ecf4ddfd05b49513eee4549e7a56
Signed-off-by: Jeenu Viswambharan <jeenu.viswambharan@arm.com>

The Commit cdb6ac94

 introduced a bug
because of which the SDEI dispatcher wrongly panic when an SDEI event
dispatched earlier as a result of interrupt. This patch fixes the check
for a bound interrupt.

Change-Id: If55c8f0422ff781731248bbbc9c1b59fa0d3c4b0
Signed-off-by: Jeenu Viswambharan <jeenu.viswambharan@arm.com>

SDEI event dispatches currently only sets up the Non-secure context
before returning to the caller. The actual dispatch only happens upon
exiting EL3 next time.

However, for various error handling scenarios, it's beneficial to have
the dispatch happen synchronously. I.e. when receiving SDEI interrupt,
or for a successful sdei_dispatch_event() call, the event handler is
executed; and upon the event completion, dispatcher execution resumes
after the point of dispatch. The jump primitives introduced in the
earlier patch facilitates this feature.

With this patch:

  - SDEI interrupts and calls to sdei_dispatch_event prepares the NS
    context for event dispatch, then sets a jump point, and immediately
    exits EL3. This results in the client handler executing in
    Non-secure.

  - When the SDEI client completes the dispatched event, the SDEI
    dispatcher does a longjmp to the jump pointer created earlier. For
    the caller of the sdei_dispatch_event() in particular, this would
    appear as if call returned successfully.

The dynamic workaround for CVE_2018_3639 is slightly shifted around as
part of related minor refactoring. It doesn't affect the workaround
functionality.

Documentation updated.

NOTE: This breaks the semantics of the explicit dispatch API, and any
exiting usages should be carefully reviewed.

Change-Id: Ib9c876d27ea2af7fb22de49832e55a0da83da3f9
Signed-off-by: Jeenu Viswambharan <jeenu.viswambharan@arm.com>

Currently, the dispatcher reads from SCR_EL3 register directly to
determine the EL of SDEI client. This is with the assumption that
SCR_EL3 is not modified throughout. However, with RAS work flows, it's
possible that SCR_EL3 register contains values corresponding to Secure
world, and therefore EL determination can go wrong. To mitigate this,
always read the register from the saved Non-secure context.

Change-Id: Ic85e4021deb18eb58757f676f9a001174998543a
Signed-off-by: Jeenu Viswambharan <jeenu.viswambharan@arm.com>

The current macros only allow to define dynamic and statically-bound
SDEI events. However, there ought be a mechanism to define SDEI events
that are explicitly dispatched; i.e., events that are dispatched as a
result of a previous secure interrupt or other exception

This patch introduces SDEI_EXPLICIT_EVENT() macro to define an explicit
event. They must be placed under private mappings. Only the priority
flags are allowed to be additionally specified.

Documentation updated.

Change-Id: I2e12f5571381195d6234c9dfbd5904608ad41db3
Signed-off-by: Jeenu Viswambharan <jeenu.viswambharan@arm.com>

When dynamic mitigation is used, the SDEI handler is required to
execute with the mitigation enabled by default, regardless of the
mitigation state for lower ELs.  This means that if the kernel or
hypervisor explicitly disables the mitigation and then later when the
event is dispatched, the dispatcher will remember the mitigation state
for the lower ELs but force the mitigation to be on during the SDEI
handler execution.  When the SDEI handler returns, it will restore the
mitigation state.

This behaviour is described in "Firmware interfaces for mitigating
cache speculation vulnerabilities System Software on Arm Systems"[0].

[0] https://developer.arm.com/cache-speculation-vulnerability-firmware-specification



Change-Id: I8dd60b736be0aa9e832b0f92d67a401fdeb417f4
Signed-off-by: Dimitris Papastamos <dimitris.papastamos@arm.com>

Currently, when the client attempts to do SDEI_EVENT_COMPLETE or
SDEI_EVENT_COMPLETE_AND_RESUME, the dispatcher pops off the outstanding
dispatch context for sanity check. There are however other checks
following this, which could potentially return failure. If that happens,
by popping the context, the dispatcher has inadvertently discarded a
valid context.

This patch fixes this bug by inspecting (not actually popping) the
outstanding context. The context is popped only after all error checks
are completed.

Change-Id: Ie199f6442f871a8177a8247a0c646543bad76d21
Signed-off-by: Jeenu Viswambharan <jeenu.viswambharan@arm.com>

Change-Id: Ic381ab5d03ec68c7f6e8d357ac2e2cbf0cc6b2e8
Signed-off-by: Jeenu Viswambharan <jeenu.viswambharan@arm.com>

This allows for other EL3 components to schedule an SDEI event dispatch
to Normal world upon the next ERET. The API usage constrains are set out
in the SDEI dispatcher documentation.

Documentation to follow.

Change-Id: Id534bae0fd85afc94523490098c81f85c4e8f019
Signed-off-by: Jeenu Viswambharan <jeenu.viswambharan@arm.com>

The implementation currently supports only interrupt-based SDEI events,
and supports all interfaces as defined by SDEI specification version
1.0 [1].

Introduce the build option SDEI_SUPPORT to include SDEI dispatcher in
BL31.

Update user guide and porting guide. SDEI documentation to follow.

[1] http://infocenter.arm.com/help/topic/com.arm.doc.den0054a/ARM_DEN0054A_Software_Delegated_Exception_Interface.pdf



Change-Id: I758b733084e4ea3b27ac77d0259705565842241a
Co-authored-by: Yousuf A <yousuf.sait@arm.com>
Signed-off-by: Jeenu Viswambharan <jeenu.viswambharan@arm.com>