A simple WAITI isn't sufficient in all cases. The cAVS 2.5 hardware
uses WAITI as the entry state for per-core power gating, which is very
difficult to debug. Provide a fallback that simply spins in the idle
loop waiting for interrupts to provide a stable system while this
feature stabilizes.
Also, the SOF code for those platforms references a known bug with the
Xtensa LX6 core IP (or at least some versions), and will prefix the
WAIT instruction with 128 NOP.N's followed by an ISYNC and EXTW. This
bug hasn't been seen under Zephyr yet, and details are sketchy. But
the code is simply enough to import and works correctly.
Place both workaround under new kconfig variables and select them both
(even though they're actually mutually exclusive -- if you select both
CPU_IDLE_SPIN overrides) for cavs_v25.
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
On CPU startup, When we reach the cache flush code in arch_switch(),
the outgoing thread is a dummy. The behavior of the existing code was
to leave the existing value in the SR unchanged (probably NULL at
startup). Then the context switch would walk from that address up to
the top of the outgoing stack, flushing everything in between. That's
wrong, because the outgoing stack is a real pointer (generally the
interrupt stack of the current CPU), and we're flushing everything in
memory underneath it.
This also reverts commit 29abc8adc0 ("xtensa: fix booting secondary
cores on the dummy thread"), which appears to have been an early
attempt to address this issue. It worked (modulo all the extra and
potentially incorrect flushing) on cavs v1.5/1.8 because of the way
the entry code worked there. But on 2.5 we now hit the first context
switch in a case where those extra lines are in address space already
marked unwritable by the CPU, so the flush explodes.
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
For IMX, for timer interrupt, the interrupt handler
was not the correct one executed and that’s because
the handlers were not at the expected address.
For IMX the size constraint of the interrupt vector
table entry is 0x1C bytes of code, less than usual.
I've added a small indirection to bypass this size
constraint and moved the default handlers to the end
of vector table, renaming them to
_Level\LVL\()VectorHelper.
Signed-off-by: Iuliana Prodan <iuliana.prodan@nxp.com>
When secondary cores are booted, they use the dummy thread and
the IRQ stack until they switch over to a real thread. Therefore
dummy threads shouldn't be skipped when cohering outgoing thread
stack, only threads with zero stack size should be skipped.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
Both operands of an operator in which the usual arithmetic
conversions are performed shall have the same essential
type category.
Signed-off-by: Flavio Ceolin <flavio.ceolin@intel.com>
When we reach this code in interrupt context, our upper GPRs contain a
cross-stack call that may still include some registers from the
interrupted thread. Those need to go out to memory before we can do
our cache coherence dance here.
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
Both new thread creation and context switch had the same mistake in
cache management: the bottom of the stack (the "unused" region between
the lower memory bound and the live stack pointer) needs to be
invalidated before we switch, because otherwise any dirty lines we
might have left over can get flushed out on top of the same thread on
another CPU that is putting live data there.
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
The Xtensa L1 cache layer has straightforward semantics accessible via
single-instructions that operate on cache lines via physical
addresses. These are very amenable to inlining.
Unfortunately the Xtensa HAL layer requires function calls to do this,
leading to significant code waste at the calling site, an extra frame
on the stack and needless runtime instructions for situations where
the call is over a constant region that could elide the loop. This is
made even worse because the HAL library is not built with
-ffunction-sections, so pulling in even one of these tiny cache
functions has the effect of importing a 1500-byte object file into the
link!
Add our own tiny cache layer to include/arch/xtensa/cache.h and use
that instead.
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
Back when I started work on this stuff, I had a set of notes on
register windows that slowly evolved into something that looks like
formal documentation. There really isn't any overview-style
documentation of this stuff on the public internet, so it couldn't
hurt to commit it here for posterity.
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
Instead of passing the crt1 _start function as the entry code for
auxiliary CPUs, use a tiny assembly stub instead which can avoid the
runtime testing needed to skip the work in _start. All the crt1 code
was doing was clearing BSS (which must not happen on a second CPU) and
setting the stack pointer (which is wrong on the second CPU).
This allows us to clean out the SMP code in crt1.
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
The kernel passes the CPU's interrupt stack expected that it will
start on that, so do it. Pass the initial stack pointer from the SOC
layer in the variable "z_mp_stack_top" and set it in the assembly
startup before calling z_mp_entry().
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
The xtensa atomics layer was written with hand-coded assembly that had
to be called as functions. That's needlessly slow, given that the low
level primitives are a two-instruction sequence. Ideally the compiler
should see this as an inline to permit it to better optimize around
the needed barriers.
There was also a bug with the atomic_cas function, which had a loop
internally instead of returning the old value synchronously on a
failed swap. That's benign right now because our existing spin lock
does nothing but retry it in a tight loop anyway, but it's incorrect
per spec and would have caused a contention hang with more elaborate
algorithms (for example a spinlock with backoff semantics).
Remove the old implementation and replace with a much smaller inline C
one based on just two assembly primitives.
This patch also contains a little bit of refactoring to address the
scheme has been split out into a separate header for each, and the
ATOMIC_OPERATIONS_CUSTOM kconfig has been renamed to
ATOMIC_OPERATIONS_ARCH to better capture what it means.
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
There was a bunch of dead historical cruft floating around in the
arch/xtensa tree, left over from older code versions. It's time to do
a cleanup pass. This is entirely refactoring and size optimization,
no behavior changes on any in-tree devices should be present.
Among the more notable changes:
+ xtensa_context.h offered an elaborate API to deal with a stack frame
and context layout that we no longer use.
+ xtensa_rtos.h was entirely dead code
+ xtensa_timer.h was a parallel abstraction layer implementing in the
architecture layer what we're already doing in our timer driver.
+ The architecture thread structs (_callee_saved and _thread_arch)
aren't used by current code, and had dead fields that were removed.
Unfortunately for standards compliance and C++ compatibility it's
not possible to leave an empty struct here, so they have a single
byte field.
+ xtensa_api.h was really just some interrupt management inlines used
by irq.h, so fold that code into the outer header.
+ Remove the stale assembly offsets. This architecture doesn't use
that facility.
All told, more than a thousand lines have been removed. Not bad.
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
a0 is used as scratch register. Restore value of a0 (return address)
from stack frame before spilling registers on stack
Signed-off-by: Shubham Kulkarni <shubham.kulkarni@espressif.com>
Only the CAVS 1.5 linker script has full support for the coherence
features, don't advertise it on the other SoC's yet.
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
While fixing license headers, identified this script as orphan and not
being used anywhere, so remove.
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
XCC doesn't like the "rsr.<reg name>" style assembly
so fix that to the other style.
Also, XCC doesn't like _CONCAT() with the EPC/EPS
registers so need to spell out all of them.
Signed-off-by: Daniel Leung <daniel.leung@intel.com>
There is a hard-coded value of PS_INTLEVEL(15) to set the PS
register. The correct way is actually to use XCHAL_EXCM_LEVEL
with PS_INTLEVEL() to setup the register. So fix it.
Fixes#31858
Signed-off-by: Daniel Leung <daniel.leung@intel.com>
This change uses stack frame to print backtrace once exception occurs
Printing backtrace helps to identify the cause of exception
Signed-off-by: Shubham Kulkarni <shubham.kulkarni@espressif.com>
Currently Zephyr links reset-vector.S twice in xtensa builds:
into the bootloader and the main image. It is run at the end
of the boot loader execution and immediately after that again
in the beginning of the main code. This patch adds a
configuration option to select whether to link the file to the
bootloader or to the application. The default is to the
application, as needed e.g. for QEMU, SOF links it to the
bootloader like in native builds.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
There may be Xtensa SoCs which don't have high enough interrupt
levels for EPC6/EPS6 to exist in _restore_context. So changes
these to those which should be available according to the ISA
config file.
Fixes#30126
Signed-off-by: Daniel Leung <daniel.leung@intel.com>
Most of kernel files where declaring os module without providing
log level. Because of that default log level was used instead of
CONFIG_KERNEL_LOG_LEVEL.
Signed-off-by: Krzysztof Chruscinski <krzysztof.chruscinski@nordicsemi.no>
Since the tracing of thread being switched in/out has the same
instrumentation points, we can roll the tracing function calls
into the one for thread stats gathering functions.
This avoids duplicating code to call another function.
Signed-off-by: Daniel Leung <daniel.leung@intel.com>
Adds the necessary bits to initialize TLS in the stack
area and sets up CPU registers during context switch.
Note that this does not enable TLS for all Xtensa SoC.
This is because Xtensa SoCs are highly configurable
so that each SoC can be considered a whole architecture.
So TLS needs to be enabled on the SoC level, instead of
at the arch level.
Signed-off-by: Daniel Leung <daniel.leung@intel.com>
Implement the kernel "coherence" API on top of the linker
cached/uncached mapping work.
Add Xtensa handling for the stack coherence API.
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
It's legal to have CONFIG_MP_NUM_CPUS > 1 and !CONFIG_SMP. The
tests/kernel/mp test does this as a unit test of the multiprocessor
facilities. Test the right tunable when deciding whether to blow away
static data or not.
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
This code had one purpose only, feed timing information into a test and
was not used by anything else. The custom trace points unfortunatly were
not accurate and this test was delivering informatin that conflicted
with other tests we have due to placement of such trace points in the
architecture and kernel code.
For such measurements we are planning to use the tracing functionality
in a special mode that would be used for metrics without polluting the
architecture and kernel code with additional tracing and timing code.
Furthermore, much of the assembly code used had issues.
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
Signed-off-by: Daniel Leung <daniel.leung@intel.com>
Move tracing switched_in and switched_out to the architecture code and
remove duplications. This changes swap tracing for x86, xtensa.
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
The core kernel computes the initial stack pointer
for a thread, properly aligning it and subtracting out
any random offsets or thread-local storage areas.
arch_new_thread() no longer needs to make any calculations,
an initial stack frame may be placed at the bounds of
the new 'stack_ptr' parameter passed in. This parameter
replaces 'stack_size'.
thread->stack_info is now set before arch_new_thread()
is invoked, z_new_thread_init() has been removed.
The values populated may need to be adjusted on arches
which carve-out MPU guard space from the actual stack
buffer.
thread->stack_info now has a new member 'delta' which
indicates any offset applied for TLS or random offset.
It's used so the calculations don't need to be repeated
if the thread later drops to user mode.
CONFIG_INIT_STACKS logic is now performed inside
z_setup_new_thread(), before arch_new_thread() is called.
thread->stack_info is now defined as the canonical
user-accessible area within the stack object, including
random offsets and TLS. It will never include any
carved-out memory for MPU guards and must be updated at
runtime if guards are removed.
Available stack space is now optimized. Some arches may
need to significantly round up the buffer size to account
for page-level granularity or MPU power-of-two requirements.
This space is now accounted for and used by virtue of
the Z_THREAD_STACK_SIZE_ADJUST() call in z_setup_new_thread.
Signed-off-by: Andrew Boie <andrew.p.boie@intel.com>
MISRA-C wants the parameter names in a function implementaion
to match the names used by the header prototype.
Signed-off-by: Andrew Boie <andrew.p.boie@intel.com>
arch_new_thread() passes along the thread priority and option
flags, but these are already initialized in thread->base and
can be accessed there if needed.
Signed-off-by: Andrew Boie <andrew.p.boie@intel.com>
This operation is formally defined as rounding down a potential
stack pointer value to meet CPU and ABI requirments.
This was previously defined ad-hoc as STACK_ROUND_DOWN().
A new architecture constant ARCH_STACK_PTR_ALIGN is added.
Z_STACK_PTR_ALIGN() is defined in terms of it. This used to
be inconsistently specified as STACK_ALIGN or STACK_PTR_ALIGN;
in the latter case, STACK_ALIGN meant something else, typically
a required alignment for the base of a stack buffer.
STACK_ROUND_UP() only used in practice by Risc-V, delete
elsewhere.
Signed-off-by: Andrew Boie <andrew.p.boie@intel.com>
The core kernel z_setup_new_thread() calls into arch_new_thread(),
which calls back into the core kernel via z_new_thread_init().
Move everything that doesn't have to be in z_new_thread_init() to
z_setup_new_thread() and convert to an inline function.
Signed-off-by: Andrew Boie <andrew.p.boie@intel.com>
Under multi-processing, only the first CPU#0 needs to go through
setting up the kernel structs and clearing out BSS (among others).
There is no need for other CPUs to do those tasks. Since each
Xtensa core starts using the same boot vector, CPUs other than #0
need to skip all the startup tasks by not calling to z_cstart().
So provide another entry point for those CPUs. Note that Xtensa
arch is highly configurable. So the implementation of the entry
point is up to each individual SoC config.
Signed-off-by: Daniel Leung <daniel.leung@intel.com>
Under SMP, the main BSS section only needs to be zero-ed on CPU #0.
Other CPUs should not zero out BSS, or else it may cause CPU #0 to
crash on invalid data.
Signed-off-by: Daniel Leung <daniel.leung@intel.com>
The set of interrupt stacks is now expressed as an array. We
also define the idle threads and their associated stacks this
way. This allows for iteration in cases where we have multiple
CPUs.
There is now a centralized declaration in kernel_internal.h.
On uniprocessor systems, z_interrupt_stacks has one element
and can be used in the same way as _interrupt_stack.
The IRQ stack for CPU 0 is now set in init.c instead of in
arch code.
The extern definition of the main thread stack is now removed,
this doesn't need to be in a header.
Signed-off-by: Andrew Boie <andrew.p.boie@intel.com>
Xtensa uses two instructions to perform atomic compare-and-set
instruction: first the comparison register, then the actual
instruction to do compare-and-set. There is a potential that
context switching is performed before these two instructions.
A restored context may have the wrong value in the comparison
register. So we need to save and restore the comparison
register during context switching.
Fixes#21800
Signed-off-by: Daniel Leung <daniel.leung@intel.com>
This reverts commit 9987c2e2f9
which spills SoC configs into architecture files and is not
exactly desirable. So revert it.
Signed-off-by: Daniel Leung <daniel.leung@intel.com>
Use BOOTLOADER definition to separate bootloader code. This allows to
use the same file reset-vector.S when building bootloader and when
CONFIG_XTENSA_RESET_VECTOR is enabled.
Signed-off-by: Andrei Emeltchenko <andrei.emeltchenko@intel.com>
The atomic_cas function was using incorrect register when determining
whether value was swapped. The swapping instruction s32c1i in
atomic_cas stores the value at memory location in register a4
regardless of whether swapping is done. In this case, the register a4
should be used to determine whether a swap is done. However, register
a3 (containing the oldValue as function argument) is used instead.
Since register a5 contains the old value at address loaded before
the swapping instruction, a3 and a5 contain the same value.
Since a3 == a5 is always true in this case, the function will always
return 1 even though values are not swapped. So fix it by using
the correct register.
Also, in case the value is not swapped, it jumps to where it returns
zero instead of loading from memory and comparing again.
The function was simply looping until swapping was done, which did not
align with the API where it would return 0 when swapping is not done
(regardless whether the memory location contains the old value or not).
Signed-off-by: Daniel Leung <daniel.leung@intel.com>
