GitLab

The LDNP/STNP instructions as implemented on Cortex-A53 and
Cortex-A57 do not behave in a way most programmers expect, and will
most probably result in a significant speed degradation to any code
that employs them. The ARMv8-A architecture (see Document ARM DDI
0487A.h, section D3.4.3) allows cores to ignore the non-temporal hint
and treat LDNP/STNP as LDP/STP instead.

This patch introduces 2 new build flags:
A53_DISABLE_NON_TEMPORAL_HINT and A57_DISABLE_NON_TEMPORAL_HINT
to enforce this behaviour on Cortex-A53 and Cortex-A57. They are
enabled by default.

The string printed in debug builds when a specific CPU errata
workaround is compiled in but skipped at runtime has been
generalised, so that it can be reused for the non-temporal hint use
case as well.

Change-Id: I3e354f4797fd5d3959872a678e160322b13867a1

The debug prints used to debug translation table setup in xlat_tables.c
used the `printf()` standard library function instead of the stack
optimized `tf_printf()` API. DEBUG_XLAT_TABLE option was used to enable
debug logs within xlat_tables.c and it configured a much larger stack
size for the platform in case it was enabled. This patch modifies these
debug prints within xlat_tables.c to use tf_printf() and modifies the format
specifiers to be compatible with tf_printf(). The debug prints are now enabled
if the VERBOSE prints are enabled in Trusted Firmware via LOG_LEVEL build
option.

The much larger stack size definition when DEBUG_XLAT_TABLE is defined
is no longer required and the platform ports are modified to remove this
stack size definition.

Change-Id: I2f7d77ea12a04b827fa15e2adc3125b1175e4c23

Migrate all direct usage of __attribute__ to usage of their
corresponding macros from cdefs.h.
e.g.:
 - __attribute__((unused)) -> __unused
Signed-off-by: Soren Brinkmann <soren.brinkmann@xilinx.com>

This patch adds support for ARM Cortex-A35 processor in the CPU
specific framework, as described in the Cortex-A35 TRM (r0p0).

Change-Id: Ief930a0bdf6cd82f6cb1c3b106f591a71c883464

On the ARMv8 architecture, cache maintenance operations by set/way on the last
level of integrated cache do not affect the system cache. This means that such a
flush or clean operation could result in the data being pushed out to the system
cache rather than main memory. Another CPU could access this data before it
enables its data cache or MMU. Such accesses could be serviced from the main
memory instead of the system cache. If the data in the sysem cache has not yet
been flushed or evicted to main memory then there could be a loss of
coherency. The only mechanism to guarantee that the main memory will be updated
is to use cache maintenance operations to the PoC by MVA(See section D3.4.11
(System level caches) of ARMv8-A Reference Manual (Issue A.g/ARM DDI0487A.G).

This patch removes the reliance of Trusted Firmware on the flush by set/way
operation to ensure visibility of data in the main memory. Cache maintenance
operations by MVA are now used instead. The following are the broad category of
changes:

1. The RW areas of BL2/BL31/BL32 are invalidated by MVA before the C runtime is
   initialised. This ensures that any stale cache lines at any level of cache
   are removed.

2. Updates to global data in runtime firmware (BL31) by the primary CPU are made
   visible to secondary CPUs using a cache clean operation by MVA.

3. Cache maintenance by set/way operations are only used prior to power down.

NOTE: NON-UPSTREAM TRUSTED FIRMWARE CODE SHOULD MAKE EQUIVALENT CHANGES IN
ORDER TO FUNCTION CORRECTLY ON PLATFORMS WITH SUPPORT FOR SYSTEM CACHES.

Fixes ARM-software/tf-issues#205

Change-Id: I64f1b398de0432813a0e0881d70f8337681f6e9a

This patch unifies the bakery lock api's across coherent and normal
memory implementation of locks by using same data type `bakery_lock_t`
and similar arguments to functions.

A separate section `bakery_lock` has been created and used to allocate
memory for bakery locks using `DEFINE_BAKERY_LOCK`. When locks are
allocated in normal memory, each lock for a core has to spread
across multiple cache lines. By using the total size allocated in a
separate cache line for a single core at compile time, the memory for
other core locks is allocated at link time by multiplying the single
core locks size with (PLATFORM_CORE_COUNT - 1). The normal memory lock
algorithm now uses lock address instead of the `id` in the per_cpu_data.
For locks allocated in coherent memory, it moves locks from
tzfw_coherent_memory to bakery_lock section.

The bakery locks are allocated as part of bss or in coherent memory
depending on usage of coherent memory. Both these regions are
initialised to zero as part of run_time_init before locks are used.
Hence, bakery_lock_init() is made an empty function as the lock memory
is already initialised to zero.

The above design lead to the removal of psci bakery locks from
non_cpu_power_pd_node to psci_locks.

NOTE: THE BAKERY LOCK API WHEN USE_COHERENT_MEM IS NOT SET HAS CHANGED.
THIS IS A BREAKING CHANGE FOR ALL PLATFORM PORTS THAT ALLOCATE BAKERY
LOCKS IN NORMAL MEMORY.

Change-Id: Ic3751c0066b8032dcbf9d88f1d4dc73d15f61d8b

This patch migrates the rest of Trusted Firmware excluding Secure Payload and
the dispatchers to the new platform and context management API. The per-cpu
data framework APIs which took MPIDRs as their arguments are deleted and only
the ones which take core index as parameter are retained.

Change-Id: I839d05ad995df34d2163a1cfed6baa768a5a595d

- 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>

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>

The return value from the SYS_WRITE semihosting operation is 0 if
the call is successful or the number of bytes not written, if there
is an error. The implementation of the write function in the
semihosting driver treats the return value as the number of bytes
written, which is wrong. This patch fixes it.

Change-Id: Id39dac3d17b5eac557408b8995abe90924c85b85

This patch fixes an issue in the cpu specific register reporting
of FVP AEM model whereby crash reporting itself triggers an exception
thus resulting in recursive crash prints. The input to the
'size_controlled_print' in the crash reporting framework should
be a NULL terminated string. As there were no cpu specific register
to be reported on FVP AEM model, the issue was caused by passing 0
instead of NULL terminated string to the above mentioned function.

Change-Id: I664427b22b89977b389175dfde84c815f02c705a

In order for the symbol table in the ELF file to contain the size of
functions written in assembly, it is necessary to report it to the
assembler using the .size directive.

To fulfil the above requirements, this patch introduces an 'endfunc'
macro which contains the .endfunc and .size directives. It also adds
a .func directive to the 'func' assembler macro.

The .func/.endfunc have been used so the assembler can fail if
endfunc is omitted.

Fixes ARM-Software/tf-issues#295

Change-Id: If8cb331b03d7f38fe7e3694d4de26f1075b278fc
Signed-off-by: Kévin Petit <kevin.petit@arm.com>

This patch removes the `owner` field in bakery_lock_t structure which
is the data structure used in the bakery lock implementation that uses
coherent memory. The assertions to protect against recursive lock
acquisition were based on the 'owner' field. They are now done based
on the bakery lock ticket number. These assertions are also added
to the bakery lock implementation that uses normal memory as well.

Change-Id: If4850a00dffd3977e218c0f0a8d145808f36b470

This patch optimizes the data structure used with the bakery lock
implementation for coherent memory to save memory and minimize memory
accesses. These optimizations were already part of the bakery lock
implementation for normal memory and this patch now implements
it for the coherent memory implementation as well. Also
included in the patch is a cleanup to use the do-while loop while
waiting for other contenders to finish choosing their tickets.

Change-Id: Iedb305473133dc8f12126726d8329b67888b70f1

This patch adds support for ARM Cortex-A72 processor in the CPU
specific framework.

Change-Id: I5986855fc1b875aadf3eba8c36e989d8a05e5175

This patch defines the ARRAY_SIZE macro for calculating number of elements
in an array and uses it where appropriate.

Change-Id: I72746a9229f0b259323972b498b9a3999731bc9b

The cpu-ops pointer was initialized before enabling the data cache in the cold
and warm boot paths. This required a DCIVAC cache maintenance operation to
invalidate any stale cache lines resident in other cpus.

This patch moves this initialization to the bl31_arch_setup() function
which is always called after the data cache and MMU has been enabled.

This change removes the need:
 1. for the DCIVAC cache maintenance operation.
 2. to initialise the CPU ops upon resumption from a PSCI CPU_SUSPEND
    call since memory contents are always preserved in this case.

Change-Id: Ibb2fa2f7460d1a1f1e721242025e382734c204c6

The CPU specific reset handlers no longer have the freedom
of using any general purpose register because it is being invoked
by the BL3-1 entry point in addition to BL1. The Cortex-A57 CPU
specific reset handler was overwriting x20 register which was being
used by the BL3-1 entry point to save the entry point information.
This patch fixes this bug by reworking the register allocation in the
Cortex-A57 reset handler to avoid using x20. The patch also
explicitly mentions the register clobber list for each of the
callee functions invoked by the reset handler

Change-Id: I28fcff8e742aeed883eaec8f6c4ee2bd3fce30df

This patch adds the missing features to the C library included
in the Trusted Firmware to build PolarSSL:

  - strcasecmp() function
  - exit() function
  - sscanf()* function
  - time.h header file (and its dependencies)

* NOTE: the sscanf() function is not a real implementation. It just
returns the number of expected arguments by counting the number of
'%' characters present in the formar string. This return value is
good enough for PolarSSL because during the certificate parsing
only the return value is checked. The certificate validity period
is ignored.

Change-Id: I43bb3742f26f0bd458272fccc3d72a7f2176ab3d

This patch adds support to call the reset_handler() function in BL3-1 in the
cold and warm boot paths when another Boot ROM reset_handler() has already run.

This means the BL1 and BL3-1 versions of the CPU and platform specific reset
handlers may execute different code to each other. This enables a developer to
perform additional actions or undo actions already performed during the first
call of the reset handlers e.g. apply additional errata workarounds.

Typically, the reset handler will be first called from the BL1 Boot ROM. Any
additional functionality can be added to the reset handler when it is called
from BL3-1 resident in RW memory. The constant FIRST_RESET_HANDLER_CALL is used
to identify whether this is the first version of the reset handler code to be
executed or an overridden version of the code.

The Cortex-A57 errata workarounds are applied only if they have not already been
applied.

Fixes ARM-software/tf-issue#275

Change-Id: Id295f106e4fda23d6736debdade2ac7f2a9a9053

This patch moves the bakery locks out of coherent memory to normal memory.
This implies that the lock information needs to be placed on a separate cache
line for each cpu. Hence the bakery_lock_info_t structure is allocated in the
per-cpu data so as to minimize memory wastage. A similar platform per-cpu
data is introduced for the platform locks.

As a result of the above changes, the bakery lock api is completely changed.
Earlier, a reference to the lock structure was passed to the lock implementation.
Now a unique-id (essentially an index into the per-cpu data array) and an offset
into the per-cpu data for bakery_info_t needs to be passed to the lock
implementation.

Change-Id: I1e76216277448713c6c98b4c2de4fb54198b39e0

This patch is an optimization in the bakery_lock_get() function
which removes the wfe() when waiting for other contenders to choose
their ticket i.e when their `entering` flag is set. Since the time
taken to execute bakery_get_ticket() by other contenders is bounded,
this wait is a bounded time wait. Hence the removal of wfe() and the
corresponding sev() and dsb() in bakery_get_ticket() may result
in better time performance during lock acquisition.

Change-Id: I141bb21294226b54cb6e89e7cac0175c553afd8d

This patch fixes a crash due to corruption of cpu_ops
data structure. During the secondary CPU boot, after the
cpu_ops has been initialized in the per cpu-data, the
dcache lines need to invalidated so that the update in
memory can be seen later on when the dcaches are turned ON.
Also, after initializing the psci per cpu data, the dcache
lines are flushed so that they are written back to memory
and dirty dcache lines are avoided.

Fixes ARM-Software/tf-issues#271

Change-Id: Ia90f55e9882690ead61226eea5a5a9146d35f313

This patch fixes a bug in the bakery lock implementation where a data
synchronisation barrier instruction is not issued before sending an event as
mandated by the ARMv8 ARM. This can cause a event to be signalled before the
related memory accesses have completed resulting in erroneous execution.

Fixes ARM-software/tf-issues#272

Change-Id: I5ce02bf70afb001d967b9fa4c3f77442931d5349

This patch optimizes the Cortex-A57 cluster power down sequence by not
flushing the Level1 data cache. The L1 data cache and the L2 unified
cache are inclusive. A flush of the L2 by set/way flushes any dirty
lines from the L1 as well. This is a known safe deviation from the
Cortex-A57 TRM defined power down sequence. This optimization can be
enabled by the platform through the 'SKIP_A57_L1_FLUSH_PWR_DWN' build
flag. Each Cortex-A57 based platform must make its own decision on
whether to use the optimization.

This patch also renames the cpu-errata-workarounds.md to
cpu-specific-build-macros.md as this facilitates documentation
of both CPU Specific errata and CPU Specific Optimization
build macros.

Change-Id: I299b9fe79e9a7e08e8a0dffb7d345f9a00a71480

This the patch replaces the DSB SY with DSB ISH
after disabling L2 prefetches during the Cortex-A57
power down sequence.

Change-Id: I048d12d830c1b974b161224eff079fb9f8ecf52d

Prior to this patch, the errata workarounds were applied for any version
of the CPU in the release build and in the debug build an assert
failure resulted when the revision did not match. This patch applies
errata workarounds in the Cortex-A57 reset handler only if the 'variant'
and 'revision' fields read from the MIDR_EL1 match. In the debug build,
a warning message is printed for each errata workaround which is not
applied.

The patch modifies the register usage in 'reset_handler` so
as to adhere to ARM procedure calling standards.

Fixes ARM-software/tf-issues#242

Change-Id: I51b1f876474599db885afa03346e38a476f84c29

This patch adds level specific cache maintenance functions
to cache_helpers.S. The new functions 'dcsw_op_levelx',
where '1 <= x <= 3', allow to perform cache maintenance by
set/way for that particular level of cache.  With this patch,
functions to support cache maintenance upto level 3 have
been implemented since it is the highest cache level for
most ARM SoCs.

These functions are now utilized in CPU specific power down
sequences to implement them as mandated by processor specific
technical reference manual.

Change-Id: Icd90ce6b51cff5a12863bcda01b93601417fd45c

This patch adds workarounds for selected errata which affect the Cortex-A57 r0p0
part. Each workaround has a build time flag which should be used by the platform
port to enable or disable the corresponding workaround. The workarounds are
disabled by default. An assertion is raised if the platform enables a workaround
which does not match the CPU revision at runtime.

Change-Id: I9ae96b01c6ff733d04dc733bd4e67dbf77b29fb0

This patch adds handlers for dumping Cortex-A57 and Cortex-A53 specific register
state to the CPU specific operations framework. The contents of CPUECTLR_EL1 are
dumped currently.

Change-Id: I63d3dbfc4ac52fef5e25a8cf6b937c6f0975c8ab

This patch adds CPU core and cluster power down sequences to the CPU specific
operations framework introduced in a earlier patch. Cortex-A53, Cortex-A57 and
generic AEM sequences have been added. The latter is suitable for the
Foundation and Base AEM FVPs. A pointer to each CPU's operations structure is
saved in the per-cpu data so that it can be easily accessed during power down
seqeunces.

An optional platform API has been introduced to allow a platform to disable the
Accelerator Coherency Port (ACP) during a cluster power down sequence. The weak
definition of this function (plat_disable_acp()) does not take any action. It
should be overriden with a strong definition if the ACP is present on a
platform.

Change-Id: I8d09bd40d2f528a28d2d3f19b77101178778685d

This patch adds an optional platform API (plat_reset_handler) which allows the
platform to perform any actions immediately after a cold or warm reset
e.g. implement errata workarounds. The function is called with MMU and caches
turned off. This API is weakly defined and does nothing by default but can be
overriden by a platform with a strong definition.

Change-Id: Ib0acdccbd24bc756528a8bd647df21e8d59707ff

This patch introduces a framework which will allow CPUs to perform
implementation defined actions after a CPU reset, during a CPU or cluster power
down, and when a crash occurs. CPU specific reset handlers have been implemented
in this patch. Other handlers will be implemented in subsequent patches.

Also moved cpu_helpers.S to the new directory lib/cpus/aarch64/.

Change-Id: I1ca1bade4d101d11a898fb30fea2669f9b37b956

Move the remaining IO storage source file (io_storage.c) from the
lib to the drivers directory. This requires that platform ports
explicitly add this file to the list of source files.

Also move the IO header files to a new sub-directory, include/io.

Change-Id: I862b1252a796b3bcac0d93e50b11e7fb2ded93d6

The intent of io_init() was to allow platform ports to provide
a data object (io_plat_data_t) to the IO storage framework to
allocate into. The abstraction was incomplete because io_plat_data_t
uses a platform defined constant and the IO storage framework
internally allocates other arrays using platform defined constants.

This change simplifies the implementation by instantiating the
supporting objects in the IO storage framework itself. There is now
no need for the platform to call io_init().

The FVP port has been updated accordingly.

THIS CHANGE REQUIRES ALL PLATFORM PORTS THAT USE THE IO STORAGE
FRAMEWORK TO BE UDPATED.

Change-Id: Ib48ac334de9e538064734334c773f8b43df3a7dc

Fix the following issues with the console log output:

* Make sure the welcome string is the first thing in the log output
(during normal boot).
* Prefix each message with the BL image name so it's clear which
BL the output is coming from.
* Ensure all output is wrapped in one of the log output macros so it can
be easily compiled out if necessary. Change some of the INFO() messages
to VERBOSE(), especially in the TSP.
* Create some extra NOTICE() and INFO() messages during cold boot.
* Remove all usage of \r in log output.

Fixes ARM-software/tf-issues#231

Change-Id: Ib24f7acb36ce64bbba549f204b9cde2dbb46c8a3

The patch implements a macro ASM_ASSERT() which can
be invoked from assembly code. When assertion happens,
file name and line number of the check is written
to the crash console.

Fixes ARM-software/tf-issues#95

Change-Id: I6f905a068e1c0fa4f746d723f18df60daaa00a86

This patch reworks the manner in which the M,A, C, SA, I, WXN & EE bits of
SCTLR_EL3 & SCTLR_EL1 are managed. The EE bit is cleared immediately after reset
in EL3. The I, A and SA bits are set next in EL3 and immediately upon entry in
S-EL1. These bits are no longer managed in the blX_arch_setup() functions. They
do not have to be saved and restored either. The M, WXN and optionally the C
bit are set in the enable_mmu_elX() function. This is done during both the warm
and cold boot paths.

Fixes ARM-software/tf-issues#226

Change-Id: Ie894d1a07b8697c116960d858cd138c50bc7a069

This patch implements a "tf_printf" which supports only the commonly
used format specifiers in Trusted Firmware, which uses a lot less
stack space than the stdlib printf function.

Fixes ARM-software/tf-issues#116

Change-Id: I7dfa1944f4c1e634b3e2d571f49afe02d109a351

This patch adds a 'flags' parameter to each exception level specific function
responsible for enabling the MMU. At present only a single flag which indicates
whether the data cache should also be enabled is implemented. Subsequent patches
will use this flag when enabling the MMU in the warm boot paths.

Change-Id: I0eafae1e678c9ecc604e680851093f1680e9cefa