The struct z_page_frame is marked __packed to avoid extra padding as
such padding may represent significant memory waste when lots of page
frames are used. However this is a bad strategy.
The code contained this somewhat dubious comment and code in
free_page_frame_list_put():
/* The structure is packed, which ensures that this is true */
void *node = pf;
sys_slist_append(&free_page_frame_list, node);
This is bad for many reasons:
- type checking is completely bypassed;
- if the sys_snode_t node member is no longer located at the front of
struct z_page_frame then the code will still compile and possibly run
but be broken with memory corruption as a likely outcome;
- the sys_slist_append() code is completely unaware of the packed
attribute which breaks architectures with alignment restrictions.
Let's improve code efficiency as well as memory usage by removing the
packed attribute and manually packing the flags in the unused virtual
address bits. This way the page frame array remains naturally aligned,
data access becomes optimal and the actual array size gets even smaller.
Signed-off-by: Nicolas Pitre <npitre@baylibre.com>
Introduce z_page_frame_set() and z_page_frame_clear() to manipulate
flags. Obtain the virtual address using the existing
z_page_frame_to_virt(). This will make changes to the page frame
structure easier.
Signed-off-by: Nicolas Pitre <npitre@baylibre.com>
Dynamic code execution applications not using LLEXT for "extension"
loading are subject to the same linker optimization symbol resolution
issue described in commit 321e395 (in summary, libkernel.a syscalls
not used directly by the application result in weak symbol resolution
of their z_mrsh_ wrapper).
To support usecases where an application is using alternative methods
to load and execute code calling syscalls (likely from userspace) or
is using a mechanism where the linker may not be aware, the configuration
option has been decoupled from CONFIG_LLEXT (who is now a selector) to
KERNEL_WHOLE_ARCHIVE.
Signed-off-by: Daniel Apperloo <daniel.apperloo@intel.com>
- modified parameter types to receive a const pointer when a
non-const pointer is not needed
- avoided redundant casts
Signed-off-by: Hess Nathan <nhess@baumer.com>
limit is unsigned int and K_SEM_MAX_LIMIT is defined as UINT_MAX this
means that limit will never be greater K_SEM_MAX_LIMIT.
Signed-off-by: Flavio Ceolin <flavio.ceolin@intel.com>
Spell checking tools do not recognize "iff", replace with "if and only if".
See https://en.wikipedia.org/wiki/If_and_only_if
Signed-off-by: Pieter De Gendt <pieter.degendt@basalte.be>
update kernel timeout logic based on retrieve system timer clock
frequency at runtime or static way based on Kconfig
TIMER_READS_ITS_FREQUENCY_AT_RUNTIME
Signed-off-by: Najumon B.A <najumon.ba@intel.com>
The previous abort-lifecycle fix missed a case: other threads can
enter k_thread_join(), see that the thread is already dead, and then
need to call z_thread_switch_spin() to wait for a context switch. But
the new "dummification" code was (by design!) terminating the thread
such that no context would be saved to it. So switch_handle stayed
NULL and if you hit that timing case correctly[1] you'd deadlock
waiting for a switch that would never come.
Fix is just to set switch_handle when dummifying to any non-NULL
value.
Also add an assertion to catch the obvious case that a thread is
actually dead on the exit path of k_thread_abort() to make sure the
variant path continues to set flags correctly
[1] CI was doing it fairly reliably via tests/kernel/smp_abort on
qemu_cortex_a53 only. Only one of my dev systems could see it,
and then only about 15% of the time.
Signed-off-by: Andy Ross <andyross@google.com>
We've had threads spinning on the thread state bits, but weren't being
careful to ensure that those bits were the last things seen to change
in a halting thread. Move it to the end, and add a barrier for
correctness.
Signed-off-by: Andy Ross <andyross@google.com>
Nicolas Pitre points out that since these thread structs are just
dummies for the context swtiching, they can be presumed to be "write
only" and thus there's no point in having one per CPU, everyone can
share the same one.
The only gotcha is that we never really documented (nor really have a
place to document) that rule, so it's not theoretically impossible for
an architecture to read back what it might have written underneath
arch_switch(). Leave this in a separate commit for bisection
purposes, but the risk seems very low.
Signed-off-by: Andy Ross <andyross@google.com>
After a k_thread_abort(), the resulting thread struct is documented as
unused/free memory that may be re-used (for example, to respawn a new
thread).
But in the special case of aborting the current thread from within an
ISR, that wasn't quite happening. The scheduler cleanup would
complete, but the architecture layer would still try to context switch
away from the aborted thread on exit, and that can include writes to
the now-reused thread struct! The specifics will depend on
architecture (some do a full context save on entry, most don't), but
in the case of USE_SWITCH=y it will at the very least write the
switch_handle field.
Fix this simply, with a per-cpu "switch dummy" thread struct for use
as a target for context switches like this. There is some non-trivial
memory cost to that; thread structs on many architectures are large.
Pleasingly, this also addresses a known deadlock on SMP: because the
"spin in ISR" step now happens as the very last stage of
k_thread_abort() handling, the existing scheduler lock works to
serialize calls such that it's impossible for a cycle of threads to
independently decide to spin on each other: at least one will see
itself as "already aborting" and break the cycle.
Fixes#64646
Signed-off-by: Andy Ross <andyross@google.com>
Spinlocks taken in ISRs were storing the _current thread pointer of
the interrupted thread as the owner, which was never strictly correct
but was benign as the thread would never run until the lock was
released.
But now k_thread_abort(_current) in an ISR has been fixed to eliminate
all references to the (now aborted) thread struct, and _current points
to a dummy thread. Handle that edge case in the validation framework.
Signed-off-by: Andy Ross <andyross@google.com>
This will reset time slice in k_thread_time_slice_set()
when slice per thread api is used.
Currently it will reset it only in standard slice_set
Signed-off-by: Adrian Bonislawski <adrian.bonislawski@intel.com>
Updates z_get_next_switch_handle() to set the new thread's base.cpu
value as it is done in do_swap(). This helps to ensure that the
last CPU on which the thread executed remains current.
Signed-off-by: Peter Mitsis <peter.mitsis@intel.com>
There is a duplicate write in `z_sched_thread_usage()` that can be
removed. Also modified the docstrings to `k_thread_runtime_stats` to
help better describe the differences between execution_cycles and
total_cycles when getting stats for the CPU or a thread
Signed-off-by: Eric Johnson <eric@memfault.com>
This reverts commit 93dc7e7438.
This PR introduced 2 regressions in main CI:
71977 & 71978
Let's revert it by now to get main's CI passing again.
Signed-off-by: Alberto Escolar Piedras <alberto.escolar.piedras@nordicsemi.no>
This reverts commit 61c70626a5.
This PR introduced 2 regressions in main CI:
71977 & 71978
Let's revert it by now to get main's CI passing again.
Signed-off-by: Alberto Escolar Piedras <alberto.escolar.piedras@nordicsemi.no>
This reverts commit 20611f13ca.
This PR introduced 2 regressions in main CI:
71977 & 71978
Let's revert it by now to get main's CI passing again.
Signed-off-by: Alberto Escolar Piedras <alberto.escolar.piedras@nordicsemi.no>
This reverts commit 02b24911f7.
This PR introduced 2 regressions in main CI:
71977 & 71978
Let's revert it by now to get main's CI passing again.
Signed-off-by: Alberto Escolar Piedras <alberto.escolar.piedras@nordicsemi.no>
Marking sysworkq as essential, so when it fails, the system will halt
instead of continuously working, and dependent components stay
in a broken state.
Signed-off-by: Mohamed ElShahawi <ExtremeGTX@hotmail.com>
Allow the creator of a work_queue instance to choose whether
the work_queue thread should be marked as ESSENTIAL or not.
Signed-off-by: Mohamed ElShahawi <ExtremeGTX@hotmail.com>
We've had threads spinning on the thread state bits, but weren't being
careful to ensure that those bits were the last things seen to change
in a halting thread. Move it to the end, and add a barrier for
correctness.
Signed-off-by: Andy Ross <andyross@google.com>
Nicolas Pitre points out that since these thread structs are just
dummies for the context swtiching, they can be presumed to be "write
only" and thus there's no point in having one per CPU, everyone can
share the same one.
The only gotcha is that we never really documented (nor really have a
place to document) that rule, so it's not theoretically impossible for
an architecture to read back what it might have written underneath
arch_switch(). Leave this in a separate commit for bisection
purposes, but the risk seems very low.
Signed-off-by: Andy Ross <andyross@google.com>
After a k_thread_abort(), the resulting thread struct is documented as
unused/free memory that may be re-used (for example, to respawn a new
thread).
But in the special case of aborting the current thread from within an
ISR, that wasn't quite happening. The scheduler cleanup would
complete, but the architecture layer would still try to context switch
away from the aborted thread on exit, and that can include writes to
the now-reused thread struct! The specifics will depend on
architecture (some do a full context save on entry, most don't), but
in the case of USE_SWITCH=y it will at the very least write the
switch_handle field.
Fix this simply, with a per-cpu "switch dummy" thread struct for use
as a target for context switches like this. There is some non-trivial
memory cost to that; thread structs on many architectures are large.
Pleasingly, this also addresses a known deadlock on SMP: because the
"spin in ISR" step now happens as the very last stage of
k_thread_abort() handling, the existing scheduler lock works to
serialize calls such that it's impossible for a cycle of threads to
independently decide to spin on each other: at least one will see
itself as "already aborting" and break the cycle.
Fixes#64646
Signed-off-by: Andy Ross <andyross@google.com>
Spinlocks taken in ISRs were storing the _current thread pointer of
the interrupted thread as the owner, which was never strictly correct
but was benign as the thread would never run until the lock was
released.
But now k_thread_abort(_current) in an ISR has been fixed to eliminate
all references to the (now aborted) thread struct, and _current points
to a dummy thread. Handle that edge case in the validation framework.
Signed-off-by: Andy Ross <andyross@google.com>
Big change is to factor out a thread_halt_spin() utility to manage the
core complexity of this code: the situation where an ISR is asked to
abort a thread already running on another SMP CPU.
With that gone, things can be cleaned up quite a bit. Remove early
returns, most of the "#if CONFIG_SMP" usage was superfluous and will
optimize out, unify and clean up the comments, etc...
No behavioral changes (hopefully), just refactoring.
Signed-off-by: Andy Ross <andyross@google.com>
After the move to C files we got some drop in the performance when
running latency_measure. This patch declares the priority queue
functions as static inlines with minor optimizations.
The result for one metric (on qemu):
3.6 and before the anything was changed:
Get data from LIFO (w/ ctx switch): 13087 ns
after original change (46484da502):
Get data from LIFO (w/ ctx switch): 13663 ns
with this change:
Get data from LIFO (w/ ctx switch): 12543 ns
So overall, a net gain of ~ 500ns that can be seen across the board on many
of the metrics.
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
Check possible overflow in k_stack data struct. An overflow
can happens resulting in a much smaller amount of memory allocation.
Signed-off-by: Flavio Ceolin <flavio.ceolin@intel.com>
Currently, all devices are initialized at boot time (following their
level and priority order). This patch introduces deferred
initialization: by setting the property `zephyr,deferred-init` on a
device on the devicetree, Zephyr will not initialized the device.
To initialize such devices, one has to call `device_init()`.
Deferred initialization is done by grouping all deferred devices on a
different ELF section. In this way, there's no need to consume more
memory to keep track of deferred devices. When `device_init()` is
called, Zephyr will scan the deferred devices section and call the
initialization function for the matching device. As this scanning is
done only during deferred device initialization, its cost should be
bearable.
Signed-off-by: Ederson de Souza <ederson.desouza@intel.com>
This introduces support for memory mapped thread stacks,
where each thread stack is mapped into virtual memory
address space with two guard pages to catch
under-/over-flowing the stack. This is just on the kernel
side. Additional architecture code is required to fully
support this feature.
Signed-off-by: Daniel Leung <daniel.leung@intel.com>
This is similar to k_mem_map()/_unmap(). But instead of using
anonymous memory, the provided physical region is mapped
into virtual address instead. In addition to simple mapping
physical ro virtual addresses, the mapping also adds two
guard pages before and after the virtual region to catch
buffer under-/over-flow.
Signed-off-by: Daniel Leung <daniel.leung@intel.com>
This adds the mechanism to do cleanup after k_thread_abort()
is called with the current thread. This is mainly used for
cleaning up things when the thread cannot be running, e.g.,
cleanup the thread stack.
Signed-off-by: Daniel Leung <daniel.leung@intel.com>
Next timeout was set unconditionally at the end of sys_clock_announce.
However, if one of the current expired timeouts was setting a new
timeout which is the first to execute then system clock was configured
twice. Lets configure system clock only once in the isr at the and of
sys_clock_announce.
If timeouts are frequent this optimization can reduce CPU load. In
many cases setting the new sys_clock timeout is the most time
consuming operation in the sys_clock isr handler. As an example,
on the target I used setting new sys_clock timeout is taking 6 uS of
9 uS spent in the isr and it takes 16 uS with the redundant call.
Signed-off-by: Krzysztof Chruściński <krzysztof.chruscinski@nordicsemi.no>
Do not build threading support when CONFIG_MULTITHREADING=n is set and
move needed calls to a new file with the changes needed instead of the
ifdef party in sched.c
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
The `BUILD_VERSION` can be defined but empty when built
without git, causing version to be missing from the banner:
```
*** Booting Zephyr OS build ***
Hello World! qemu_riscv64
```
Let's check if it is empty before using it, so that
`KERNEL_VERSION_STRING`, which is generated independently
with cmake can be used as a fallback:
```
*** Booting Zephyr OS build 3.5.0 ***
Hello World! qemu_riscv64
```
Signed-off-by: Yong Cong Sin <ycsin@meta.com>