PSCI: Add framework to handle composite power states ...

The state-id field in the power-state parameter of a CPU_SUSPEND call can be
used to describe composite power states specific to a platform. The current PSCI
implementation does not interpret the state-id field. It relies on the target
power level and the state type fields in the power-state parameter to perform
state coordination and power management operations. The framework introduced
in this patch allows the PSCI implementation to intepret generic global states
like RUN, RETENTION or OFF from the State-ID to make global state coordination
decisions and reduce the complexity of platform ports. It adds support to
involve the platform in state coordination which facilitates the use of
composite power states and improves the support for entering standby states
at multiple power domains.

The patch also includes support for extended state-id format for the power
state parameter as specified by PSCIv1.0.

The PSCI implementation now defines a generic representation of the power-state
parameter. It depends on the platform port to convert the power-state parameter
(possibly encoding a composite power state) passed in a CPU_SUSPEND call to this
representation via the `validate_power_state()` plat_psci_ops handler. It is an
array where each index corresponds to a power level. Each entry contains the
local power state the power domain at that power level could enter.

The meaning of the local power state values is platform defined, and may vary
between levels in a single platform. The PSCI implementation constrains the
values only so that it can classify the state as RUN, RETENTION or OFF as
required by the specification:
   * zero means RUN
   * all OFF state values at all levels must be higher than all RETENTION
     state values at all levels
   * the platform provides PLAT_MAX_RET_STATE and PLAT_MAX_OFF_STATE values
     to the framework

The platform also must define the macros PLAT_MAX_RET_STATE and
PLAT_MAX_OFF_STATE which lets the PSCI implementation find out which power
domains have been requested to enter a retention or power down state. The PSCI
implementation does not interpret the local power states defined by the
platform. The only constraint is that the PLAT_MAX_RET_STATE <
PLAT_MAX_OFF_STATE.

For a power domain tree, the generic implementation maintains an array of local
power states. These are the states requested for each power domain by all the
cores contained within the domain. During a request to place multiple power
domains in a low power state, the platform is passed an array of requested
power-states for each power domain through the plat_get_target_pwr_state()
API. It coordinates amongst these states to determine a target local power
state for the power domain. A default weak implementation of this API is
provided in the platform layer which returns the minimum of the requested
power-states back to the PSCI state coordination.

Finally, the plat_psci_ops power management handlers are passed the target
local power states for each affected power domain using the generic
representation described above. The platform executes operations specific to
these target states.

The platform power management handler for placing a power domain in a standby
state (plat_pm_ops_t.pwr_domain_standby()) is now only used as a fast path for
placing a core power domain into a standby or retention state should now be
used to only place the core power domain in a standby or retention state.

The extended state-id power state format can be enabled by setting the
build flag PSCI_EXTENDED_STATE_ID=1 and it is disabled by default.

Change-Id: I9d4123d97e179529802c1f589baaa4101759d80c

PSCI: Introduce new platform interface to describe topology ...

This patch removes the assumption in the current PSCI implementation that MPIDR
based affinity levels map directly to levels in a power domain tree. This
enables PSCI generic code to support complex power domain topologies as
envisaged by PSCIv1.0 specification. The platform interface for querying
the power domain topology has been changed such that:

1. The generic PSCI code does not generate MPIDRs and use them to query the
   platform about the number of power domains at a particular power level. The
   platform now provides a description of the power domain tree on the SoC
   through a data structure. The existing platform APIs to provide the same
   information have been removed.

2. The linear indices returned by plat_core_pos_by_mpidr() and
   plat_my_core_pos() are used to retrieve core power domain nodes from the
   power domain tree. Power domains above the core level are accessed using a
   'parent' field in the tree node descriptors.

The platform describes the power domain tree in an array of 'unsigned
char's. The first entry in the array specifies the number of power domains at
the highest power level implemented in the system. Each susbsequent entry
corresponds to a power domain and contains the number of power domains that are
its direct children. This array is exported to the generic PSCI implementation
via the new `plat_get_power_domain_tree_desc()` platform API.

The PSCI generic code uses this array to populate its internal power domain tree
using the Breadth First Search like algorithm. The tree is split into two
arrays:

1. An array that contains all the core power domain nodes

2. An array that contains all the other power domain nodes

A separate array for core nodes allows certain core specific optimisations to
be implemented e.g. remove the bakery lock, re-use per-cpu data framework for
storing some information.

Entries in the core power domain array are allocated such that the
array index of the domain is equal to the linear index returned by
plat_core_pos_by_mpidr() and plat_my_core_pos() for the MPIDR
corresponding to that domain. This relationship is key to be able to use
an MPIDR to find the corresponding core power domain node, traverse to higher
power domain nodes and index into arrays that contain core specific
information.

An introductory document has been added to briefly describe the new interface.

Change-Id: I4b444719e8e927ba391cae48a23558308447da13

PSCI: Introduce new platform and CM helper APIs ...

This patch introduces new platform APIs and context management helper APIs
to support the new topology framework based on linear core position. This
framework will be introduced in the follwoing patch and it removes the
assumption that the MPIDR based affinity levels map directly to levels
in a power domain tree. The new platforms APIs and context management
helpers based on core position are as described below:

* plat_my_core_pos() and plat_core_pos_by_mpidr()

These 2 new mandatory platform APIs are meant to replace the existing
'platform_get_core_pos()' API. The 'plat_my_core_pos()' API returns the
linear index of the calling core and 'plat_core_pos_by_mpidr()' returns
the linear index of a core specified by its MPIDR. The latter API will also
validate the MPIDR passed as an argument and will return an error code (-1)
if an invalid MPIDR is passed as the argument. This enables the caller to
safely convert an MPIDR of another core to its linear index without querying
the PSCI topology tree e.g. during a call to PSCI CPU_ON.

Since the 'plat_core_pos_by_mpidr()' API verifies an MPIDR, which is always
platform specific, it is no longer possible to maintain a default implementation
of this API. Also it might not be possible for a platform port to verify an
MPIDR before the C runtime has been setup or the topology has been initialized.
This would prevent 'plat_core_pos_by_mpidr()' from being callable prior to
topology setup. As a result, the generic Trusted Firmware code does not call
this API before the topology setup has been done.

The 'plat_my_core_pos' API should be able to run without a C runtime.
Since this API needs to return a core position which is equal to the one
returned by 'plat_core_pos_by_mpidr()' API for the corresponding MPIDR,
this too cannot have default implementation and is a mandatory API for
platform ports. These APIs will be implemented by the ARM reference platform
ports later in the patch stack.

* plat_get_my_stack() and plat_set_my_stack()

These APIs are the stack management APIs which set/return stack addresses
appropriate for the calling core. These replace the 'platform_get_stack()' and
'platform_set_stack()' APIs. A default weak MP version and a global UP version
of these APIs are provided for the platforms.

* Context management helpers based on linear core position

A set of new context management(CM) helpers viz cm_get_context_by_index(),
cm_set_context_by_index(), cm_init_my_context() and cm_init_context_by_index()
are defined which are meant to replace the old helpers which took MPIDR
as argument. The old CM helpers are implemented based on the new helpers to
allow for code consolidation and will be deprecated once the switch to the new
framework is done.

Change-Id: I89758632b370c2812973a4b2efdd9b81a41f9b69

PSCI: Remove references to affinity based power management ...

As per Section 4.2.2. in the PSCI specification, the term "affinity"
is used in the context of describing the hierarchical arrangement
of cores. This often, but not always, maps directly to the processor
power domain topology of the system. The current PSCI implementation
assumes that this is always the case i.e. MPIDR based levels of
affinity always map to levels in a power domain topology tree.

This patch is the first in a series of patches which remove this
assumption. It removes all occurences of the terms "affinity
instances and levels" when used to describe the power domain
topology. Only the terminology is changed in this patch. Subsequent
patches will implement functional changes to remove the above
mentioned assumption.

Change-Id: Iee162f051b228828310610c5a320ff9d31009b4e

PSCI: Invoke PM hooks only for the highest level ...

This patch optimizes the invocation of the platform power management hooks for
ON, OFF and SUSPEND such that they are called only for the highest affinity
level which will be powered off/on. Earlier, the hooks were being invoked for
all the intermediate levels as well.

This patch requires that the platforms migrate to the new semantics of the PM
hooks.  It also removes the `state` parameter from the pm hooks as the `afflvl`
parameter now indicates the highest affinity level for which power management
operations are required.

Change-Id: I57c87931d8a2723aeade14acc710e5b78ac41732

PSCI: Create new directory to implement new frameworks ...

This patch creates a copy of the existing PSCI files and related psci.h and
platform.h header files in a new `PSCI1.0` directory. The changes for the
new PSCI power domain topology and extended state-ID frameworks will be
added incrementally to these files. This incremental approach will
aid in review and in understanding the changes better. Once all the
changes have been introduced, these files will replace the existing PSCI
files.

Change-Id: Ibb8a52e265daa4204e34829ed050bddd7e3316ff

cortex_a53: Add A53 errata #826319, #836870 ...

- Apply a53 errata #826319 to revision <= r0p2
- Apply a53 errata #836870 to revision <= r0p3
- Update docs/cpu-specific-build-macros.md for newly added errata build flags

Change-Id: I44918e36b47dca1fa29695b68700ff9bf888865e
Signed-off-by: Jimmy Huang <jimmy.huang@mediatek.com>

Add mmio utility functions ...

- Add mmio 16 bits read/write functions.
- Add clear/set/clear-and-set utility functions.

Change-Id: I00fdbdf24af537424f8666b1cadaa5f77a2a46ed
Signed-off-by: Jimmy Huang <jimmy.huang@mediatek.com>

Fix build error with optimizations disabled (-O0) ...

If Trusted Firmware is built with optimizations disabled (-O0), the
linker throws the following error:

    undefined reference to 'xxx'

Where 'xxx' is a raw inline function defined in a header file. The
reason is that, with optimizations disabled, GCC may decide to skip
the inlining. If that is the case, an external definition to the
compilation unit must be provided. Because no external definition
is present, the linker throws the error.

This patch fixes the problem by declaring the following inline
functions static, so the internal definition is used:

    inline void soc_css_security_setup(void)
    inline const arm_config_t *get_arm_config(void)

Change-Id: Id650d6be1b1396bdb48af1ac8a4c7900d212e95f

Add "Project Denver" CPU support ...

Denver is NVIDIA's own custom-designed, 64-bit, dual-core CPU which is
fully ARMv8 architecture compatible.  Each of the two Denver cores
implements a 7-way superscalar microarchitecture (up to 7 concurrent
micro-ops can be executed per clock), and includes a 128KB 4-way L1
instruction cache, a 64KB 4-way L1 data cache, and a 2MB 16-way L2
cache, which services both cores.

Denver implements an innovative process called Dynamic Code Optimization,
which optimizes frequently used software routines at runtime into dense,
highly tuned microcode-equivalent routines. These are stored in a
dedicated, 128MB main-memory-based optimization cache. After being read
into the instruction cache, the optimized micro-ops are executed,
re-fetched and executed from the instruction cache as long as needed and
capacity allows.

Effectively, this reduces the need to re-optimize the software routines.
Instead of using hardware to extract the instruction-level parallelism
(ILP) inherent in the code, Denver extracts the ILP once via software
techniques, and then executes those routines repeatedly, thus amortizing
the cost of ILP extraction over the many execution instances.

Denver also features new low latency power-state transitions, in addition
to extensive power-gating and dynamic voltage and clock scaling based on
workloads.

Signed-off-by: Varun Wadekar <vwadekar@nvidia.com>

Merge pull request #332 from jcastillo-arm/jc/tf-issues/214 ...

Use uintptr_t as base address type in ARM driver APIs

Use uintptr_t as base address type in ARM driver APIs ...

This patch changes the type of the base address parameter in the
ARM device driver APIs to uintptr_t (GIC, CCI, TZC400, PL011). The
uintptr_t type allows coverage of the whole memory space and to
perform arithmetic operations on the addresses. ARM platform code
has also been updated to use uintptr_t as GIC base address in the
configuration.

Fixes ARM-software/tf-issues#214

Change-Id: I1b87daedadcc8b63e8f113477979675e07d788f1

Merge pull request #324 from soby-mathew/sm/sys_suspend ...

PSCI: Add SYSTEM_SUSPEND API support

TBB: delete deprecated plat_match_rotpk() ...

The authentication framework deprecates plat_match_rotpk()
in favour of plat_get_rotpk_info(). This patch removes
plat_match_rotpk() from the platform port.

Change-Id: I2250463923d3ef15496f9c39678b01ee4b33883b

TBB: switch to the new authentication framework ...

This patch modifies the Trusted Board Boot implementation to use
the new authentication framework, making use of the authentication
module, the cryto module and the image parser module to
authenticate the images in the Chain of Trust.

A new function 'load_auth_image()' has been implemented. When TBB
is enabled, this function will call the authentication module to
authenticate parent images following the CoT up to the root of
trust to finally load and authenticate the requested image.

The platform is responsible for picking up the right makefiles to
build the corresponding cryptographic and image parser libraries.
ARM platforms use the mbedTLS based libraries.

The platform may also specify what key algorithm should be used
to sign the certificates. This is done by declaring the 'KEY_ALG'
variable in the platform makefile. FVP and Juno use ECDSA keys.

On ARM platforms, BL2 and BL1-RW regions have been increased 4KB
each to accommodate the ECDSA code.

REMOVED BUILD OPTIONS:

  * 'AUTH_MOD'

Change-Id: I47d436589fc213a39edf5f5297bbd955f15ae867

TBB: add TBBR Chain of Trust ...

This patch adds a CoT based on the Trusted Board Boot Requirements
document*. The CoT consists of an array of authentication image
descriptors indexed by the image identifiers.

A new header file with TBBR image identifiers has been added.
Platforms that use the TBBR (i.e. ARM platforms) may reuse these
definitions as part of their platform porting.

PLATFORM PORT - IMPORTANT:

Default image IDs have been removed from the platform common
definitions file (common_def.h). As a consequence, platforms that
used those common definitons must now either include the IDs
provided by the TBBR header file or define their own IDs.

*The NVCounter authentication method has not been implemented yet.

Change-Id: I7c4d591863ef53bb0cd4ce6c52a60b06fa0102d5

TBB: add mbedTLS authentication related libraries ...

This patch adds the following mbedTLS based libraries:

* Cryptographic library

It is used by the crypto module to verify a digital signature
and a hash. This library relies on mbedTLS to perform the
cryptographic operations. mbedTLS sources must be obtained
separately.

Two key algorithms are currently supported:

    * RSA-2048
    * ECDSA-SECP256R1

The platform is responsible for picking up the required
algorithm by defining the 'MBEDTLS_KEY_ALG' variable in the
platform makefile. Available options are:

    * 'rsa' (for RSA-2048) (default option)
    * 'ecdsa' (for ECDSA-SECP256R1)

Hash algorithm currently supported is SHA-256.

* Image parser library

Used by the image parser module to extract the authentication
parameters stored in X509v3 certificates.

Change-Id: I597c4be3d29287f2f18b82846973afc142ee0bf0

TBB: add authentication framework ...

This patch adds the authentication framework that will be used as
the base to implement Trusted Board Boot in the Trusted Firmware.
The framework comprises the following modules:

- Image Parser Module (IPM)

    This module is responsible for interpreting images, check
    their integrity and extract authentication information from
    them during Trusted Board Boot.

    The module currently supports three types of images i.e.
    raw binaries, X509v3 certificates and any type specific to
    a platform. An image parser library must be registered for
    each image type (the only exception is the raw image parser,
    which is included in the main module by default).

    Each parser library (if used) must export a structure in a
    specific linker section which contains function pointers to:

        1. Initialize the library
        2. Check the integrity of the image type supported by
           the library
        3. Extract authentication information from the image

- Cryptographic Module (CM)

    This module is responsible for verifying digital signatures
    and hashes. It relies on an external cryptographic library
    to perform the cryptographic operations.

    To register a cryptographic library, the library must use the
    REGISTER_CRYPTO_LIB macro, passing function pointers to:

        1. Initialize the library
        2. Verify a digital signature
        3. Verify a hash

    Failing to register a cryptographic library will generate
    a build time error.

- Authentication Module (AM)

    This module provides methods to authenticate an image, like
    hash comparison or digital signatures. It uses the image parser
    module to extract authentication parameters, the crypto module
    to perform cryptographic operations and the Chain of Trust to
    authenticate the images.

    The Chain of Trust (CoT) is a data structure that defines the
    dependencies between images and the authentication methods
    that must be followed to authenticate an image.

The Chain of Trust, when added, must provide a header file named
cot_def.h with the following definitions:

- COT_MAX_VERIFIED_PARAMS

    Integer value indicating the maximum number of authentication
    parameters an image can present. This value will be used by the
    authentication module to allocate the memory required to load
    the parameters in the image descriptor.

Change-Id: Ied11bd5cd410e1df8767a1df23bb720ce7e58178

TBB: add platform API to read the ROTPK information ...

This patch extends the platform port by adding an API that returns
either the Root of Trust public key (ROTPK) or its hash. This is
usually stored in ROM or eFUSE memory. The ROTPK returned must be
encoded in DER format according to the following ASN.1 structure:

    SubjectPublicKeyInfo  ::=  SEQUENCE  {
        algorithm           AlgorithmIdentifier,
        subjectPublicKey    BIT STRING
    }

In case the platform returns a hash of the key:

    DigestInfo  ::= SEQUENCE {
        digestAlgorithm     AlgorithmIdentifier,
        keyDigest           OCTET STRING
    }

An implementation for ARM development platforms is provided in this
patch. When TBB is enabled, the ROTPK hash location must be specified
using the build option 'ARM_ROTPK_LOCATION'. Available options are:

    - 'regs' : return the ROTPK hash stored in the Trusted
      root-key storage registers.

    - 'devel_rsa' : return a ROTPK hash embedded in the BL1 and
      BL2 binaries. This hash has been obtained from the development
      RSA public key located in 'plat/arm/board/common/rotpk'.

On FVP, the number of MMU tables has been increased to map and
access the ROTPK registers.

A new file 'board_common.mk' has been added to improve code sharing
in the ARM develelopment platforms.

Change-Id: Ib25862e5507d1438da10773e62bd338da8f360bf

Use numbers to identify images instead of names ...

The Trusted firmware code identifies BL images by name. The platform
port defines a name for each image e.g. the IO framework uses this
mechanism in the platform function plat_get_image_source(). For
a given image name, it returns the handle to the image file which
involves comparing images names. In addition, if the image is
packaged in a FIP, a name comparison is required to find the UUID
for the image. This method is not optimal.

This patch changes the interface between the generic and platform
code with regard to identifying images. The platform port must now
allocate a unique number (ID) for every image. The generic code will
use the image ID instead of the name to access its attributes.

As a result, the plat_get_image_source() function now takes an image
ID as an input parameter. The organisation of data structures within
the IO framework has been rationalised to use an image ID as an index
into an array which contains attributes of the image such as UUID and
name. This prevents the name comparisons.

A new type 'io_uuid_spec_t' has been introduced in the IO framework
to specify images identified by UUID (i.e. when the image is contained
in a FIP file). There is no longer need to maintain a look-up table
[iname_name --> uuid] in the io_fip driver code.

Because image names are no longer mandatory in the platform port, the
debug messages in the generic code will show the image identifier
instead of the file name. The platforms that support semihosting to
load images (i.e. FVP) must provide the file names as definitions
private to the platform.

The ARM platform ports and documentation have been updated accordingly.
All ARM platforms reuse the image IDs defined in the platform common
code. These IDs will be used to access other attributes of an image in
subsequent patches.

IMPORTANT: applying this patch breaks compatibility for platforms that
use TF BL1 or BL2 images or the image loading code. The platform port
must be updated to match the new interface.

Change-Id: I9c1b04cb1a0684c6ee65dee66146dd6731751ea5

Bug fix: Build time condition to relocate RW data ...

This patch fixes the build time condition deciding whether the
read-write data should be relocated from ROM to RAM. It was incorrectly
using __DATA_ROM_START__, which is a linker symbol and not a compiler
build flag. As a result, the relocation code was always compiled out.

This bug has been introduced by the following patch:
"Rationalize reset handling code"

Change-Id: I1c8d49de32f791551ab4ac832bd45101d6934045

Merge pull request #310 from sandrine-bailleux/sb/tf-issue-304-phase1 ...

Enhance BL3-1 entrypoint handling to support non-TF boot firmware - Phase 1

PSCI: Add SYSTEM_SUSPEND API support ...

This patch adds support for SYSTEM_SUSPEND API as mentioned in the PSCI 1.0
specification. This API, on being invoked on the last running core on a
supported platform, will put the system into a low power mode with memory
retention.

The psci_afflvl_suspend() internal API has been reused as most of the actions
to suspend a system are the same as invoking the PSCI CPU_SUSPEND API with the
target affinity level as 'system'. This API needs the 'power state' parameter
for the target low power state. This parameter is not passed by the caller of
the SYSTEM_SUSPEND API. Hence, the platform needs to implement the
get_sys_suspend_power_state() platform function to provide this information.
Also, the platform also needs to add support for suspending the system to the
existing 'plat_pm_ops' functions: affinst_suspend() and
affinst_suspend_finish().

Change-Id: Ib6bf10809cb4e9b92f463755608889aedd83cef5

Merge pull request #320 from danh-arm/rh/timer-api-v10 ...

Add delay timer API v10

FVP: Add SP804 delay timer ...

Add SP804 delay timer support to the FVP BSP.

This commit simply provides the 3 constants needed by the SP804
delay timer driver and calls sp804_timer_init() in
bl2_platform_setup(). The BSP does not currently use the delay
timer functions.

Note that the FVP SP804 is a normal world accessible peripheral
and should not be used by the secure world after transition
to the normal world.

Change-Id: I5f91d2ac9eb336fd81943b3bb388860dfb5f2b39
Co-authored-by: Dan Handley <dan.handley@arm.com>

Add SP804 delay timer driver ...

Add a delay timer driver for the ARM SP804 dual timer.

This driver only uses the first timer, called timer 1 in the
SP804 Technical Reference Manual (ARM DDI 0271D).

To use this driver, the BSP must provide three constants:

*   The base address of the SP804 dual timer
*   The clock multiplier
*   The clock divider

The BSP is responsible for calling sp804_timer_init(). The SP804
driver instantiates a constant timer_ops_t and calls the generic
timer_init().

Change-Id: I49ba0a52bdf6072f403d1d0a20e305151d4bc086
Co-authored-by: Dan Handley <dan.handley@arm.com>

Add a simple delay timer driver API ...

The API is simple. The BSP or specific timer driver creates an
instance of timer_ops_t, fills in the timer specific data, then calls
timer_init(). The timer specific data includes a function pointer
to return the timer value and a clock multiplier/divider. The ratio
of the multiplier and the divider is the clock frequency in MHz.

After that, mdelay() or udelay() can be called to delay execution for
the specified time (milliseconds or microseconds, respectively).

Change-Id: Icf8a295e1d25874f789bf28b7412156329dc975c
Co-authored-by: Dan Handley <dan.handley@arm.com>

CSS: Remove the constants MHU_SECURE_BASE/SIZE ...

For CSS based platforms, the constants MHU_SECURE_BASE and
MHU_SECURE_SIZE used to define the extents of the Trusted Mailboxes.
As such, they were misnamed because the mailboxes are completely
unrelated to the MHU hardware.

This patch removes the MHU_SECURE_BASE and MHU_SECURE_SIZE #defines.
The address of the Trusted Mailboxes is now relative to the base of
the Trusted SRAM.

This patch also introduces a new constant, SCP_COM_SHARED_MEM_BASE,
which is the address of the first memory region used for communication
between AP and SCP. This is used by the BOM and SCPI protocols.

Change-Id: Ib200f057b19816bf05e834d111271c3ea777291f

CSS: Clarify what the SCP boot config is ...

Add a comment explaining what the SCP boot configuration information
is on CSS based platforms like Juno. Also express its address
relatively to the base of the Trusted SRAM rather than hard-coding it.

Change-Id: I82cf708a284c8b8212933074ea8c37bdf48b403b

Rationalize reset handling code ...

The attempt to run the CPU reset code as soon as possible after reset
results in highly complex conditional code relating to the
RESET_TO_BL31 option.

This patch relaxes this requirement a little. In the BL1, BL3-1 and
PSCI entrypoints code, the sequence of operations is now as follows:
 1) Detect whether it is a cold or warm boot;
 2) For cold boot, detect whether it is the primary or a secondary
    CPU. This is needed to handle multiple CPUs entering cold reset
    simultaneously;
 3) Run the CPU init code.

This patch also abstracts the EL3 registers initialisation done by
the BL1, BL3-1 and PSCI entrypoints into common code.

This improves code re-use and consolidates the code flows for
different types of systems.

NOTE: THE FUNCTION plat_secondary_cold_boot() IS NOW EXPECTED TO
NEVER RETURN. THIS PATCH FORCES PLATFORM PORTS THAT RELIED ON THE
FORMER RETRY LOOP AT THE CALL SITE TO MODIFY THEIR IMPLEMENTATION.
OTHERWISE, SECONDARY CPUS WILL PANIC.

Change-Id: If5ecd74d75bee700b1bd718d23d7556b8f863546