A new locking mechanism: read/write locks
When there is a writer it is not possible to put on a read lock or a write lock; when there is a reader it is possible to reenter the read lock but not the write lock.
Writers are exclusive locks, readers are shared locks.
At the same time through the waiter count to determine whether there is currently a blocked task, if there is then in the unlock time to wake up all the waiter, through the priority of the competition to complete the blocked lock execution.
For example:
When we have a reader blocking two waiter writers, when the reader is unlocked it wakes up both writers. The writer with higher priority wakes up and checks for a successful condition and locks the lock, the second writer wakes up and fails to check for a condition and continues to block the lock.
Signed-off-by: chenrun1 <chenrun1@xiaomi.com>
This moves all the public POSIX semaphore functions into libc and with
this most of the user-space logic is also moved; namely cancel point and
errno handling.
This also removes the need for the _SEM_XX macros used to differentiate
which API is used per user-/kernel mode. Such macros are henceforth
unnecessary.
PR #11165 causes an unnecessary regression; task_delete no longer works,
if the deleted task is from another group.
The logic that prevents this comes from:
nxnotify_cancellation() ->
tls_get_info_pid() ->
nxsched_get_stackinfo()
Which checks for permissions, which does not make sense in this case since
it is the kernel asking for the stack information.
Fix this by partially reverting 11165 and implementing a direct path for
the kernel to query for any tasks TLS.
Commit 9244b5a737 added support
for non-standard field si_user that is useful for passing context
pointers to signal handlers.
This commits makes it work for all signals, not just SA_KERNELHAND.
Previously si_user for normal signals was uninitialized garbage.
The task files should consult the "spawn action" and "O_CLOEXEC flags"
to determine further whether the file should be duplicated.
This PR will further optimize file list duplicating to avoid the performance
regression caused by additional file operations.
Signed-off-by: chao an <anchao@xiaomi.com>
This moves task / thread cancel point logic from the NuttX kernel into
libc, while the data needed by the cancel point logic is moved to TLS.
The change is an enabler to move user-space APIs to libc as well, for
a coherent user/kernel separation.
Some assertions in extreme cases will cause syslog to be unable to
output logs normally, so this PR will restore the input registers
into the array of last registers to ensure that we can also obtain
some important informations.
Signed-off-by: chao an <anchao@xiaomi.com>
If the semaphore is shared, the holder has put its own mmapped address
to pholder->sem. This means we must switch to the holder's address
environment when going through the held semaphores list.
A better option would be to get the kernel mapped address for the
semaphore's physical page, but that mechanism is not functional yet.
This fixes a full system crash when CONFIG_PRIORITY_INHERITANCE=y and
CONFIG_BUILD_KERNEL=y and user makes shared semaphore via:
int semfd = shm_open("sem", O_CREAT | O_RDWR, 0666);
sem_t *sem = mmap(0, sizeof(sem_t), PROT_READ | PROT_WRITE, MAP_SHARED, semfd, 0);
Previous adjtime() implementation was limited to adjusting system
timer tick period. This commit reimplements the internals to use
a kernel watchdog timer. Platform-independent part of the code now
works also for adjusting hires RTC and tickless timer rate.
User code facing API is unchanged. Architecture code API has changed:
up_adj_timer_period() is replaced by up_adjtime().
Other improvements:
- Support query of remaining adjustment by passing NULL to first
argument of adjtime(). This matches Linux behavior.
- Improve resolution available for architecture driver, previously
limited to 1 microsecond per tick. Now 1 nanosecond per second.
Set the newly spawned process's signal mask, if the caller has instructed
to do so by setting POSIX_SPAWN_SETSIGMASK.
This is called after the task has been created but has NOT been started
yet.
Like the name implies, it is supposed to set the spawn attributes for
the NuttX specific "spawn proxy task" which was historically used as
a proxy to spawn new tasks. The proxy handled file actions and the signal
mask which are inherited from the parent.
The proxy task does not exist anymore, thus the proxy task attributes
do not need to be set anymore either.
Also, the function is currently still used, but the signal mask is set
for the spawning process, not the proxy process, and this is most
DEFINITELY an error (as the spawning process's signal mask changes
unexpectedly).
Setting the signal mask for the newly spawned process is simple, just
set it directly, if instructed to do so. This will be done in a later
patch!
VELAPLATFO-18473
refs:
https://man7.org/linux/man-pages/man2/fcntl.2.html
If the FD_CLOEXEC bit is set, the file descriptor will automatically
be closed during a successful execve(2).
(If the execve(2) fails, the file descriptor is left open.)
modify:
1. Ensure that the child task copies all fds of the parent task,
including those with O_CLOEXE.
2. Make sure spawn_file_action is executed under fd with O_CLOEXEC,
otherwise it will fail.
3. When a new task is activated or exec is called, close all fds
with O_CLOEXEC flags.
Signed-off-by: dongjiuzhu1 <dongjiuzhu1@xiaomi.com>
modlib/modlib_symbols.c: In function ‘modlib_symcallback’:
modlib/modlib_symbols.c:215:13: warning: implicit declaration of function ‘modlib_depend’; did you mean ‘modlib_read’? [-Wimplicit-function-declaration]
215 | ret = modlib_depend(exportinfo->modp, modp);
| ^~~~~~~~~~~~~
| modlib_read
Signed-off-by: dongjiuzhu1 <dongjiuzhu1@xiaomi.com>
Handle task spawn attributes as task spawn file actions are handled.
Why? This removes the need for sched_lock() when the task is being
spawned. When loading the new task from a file the scheduler can be
locked for a VERY LONG time, in the order of hundreds of milliseconds!
This is unacceptable for real time operation.
Also fixes a latent bug in exec_module, spawn_file_actions is executed
at a bad location; when CONFIG_ARCH_ADDRENV=y actions will point to the
new process's address environment (as it is temporarily instantiated at
that point). Fix this by moving it to after addrenv_restore.
This commit adds support for custom stream via fopencookie function.
The function allows the programmer the create his own custom stream
for IO operations and hook his custom functions to it.
This is a non POSIX interface defined in Standard C library and implemented
according to it. The only difference is in usage of off_t instead of
off64_t. Programmer can use 64 bits offset if CONFIG_FS_LARGEFILE is
enabled. In that case off_t is defined as int64_t (int32_t otherwise).
Field fs_fd is removed from file_struct and fs_cookie is used instead
as a shared variable for file descriptor or user defined cookie.
The interface will be useful for future fmemopen implementation.
Signed-off-by: Michal Lenc <michallenc@seznam.cz>
Exit immediately when finished processing the current CPU
if there are no other CPUs to be processed.
Signed-off-by: zhangyuan21 <zhangyuan21@xiaomi.com>
Support smp function call, calling smp_call_function allows
a specific core to execute a function. It should be noted
that there should be no waiting operations in the executed
function.
Signed-off-by: zhangyuan21 <zhangyuan21@xiaomi.com>
both functions aren't suitable to be put into libc,
because they call the kernel internal functions directly.
Signed-off-by: Xiang Xiao <xiaoxiang@xiaomi.com>
This path just for modify Mac sim-02 issue.
The compiler require the firt paramter of atomic_compare_exchange_strong
is atomic type and second parameter is int type.
Signed-off-by: TaiJu Wu <tjwu1217@gmail.com>
spinlock.c:
Implement read write spinlock.
Readers can take lock simultaneously but only one writer can take lock.
irq_spinlock.c:
Align g_irq_spin_count.
If the lock is NULL, the caller will get global lock (e.g. g_irq_spin) and spin_lock_irqsave() support nest on the same CPU.
If the CPU can write lock, it can call write_lock_irqsave() again (e.g. support nest).
Signed-off-by: TaiJu Wu <tjwu1217@gmail.com>
Co-authored-by: David Sidrane <David.Sidrane@Nscdg.com>
In addition to printing out the thread name (task name in flat mode),
print the parent process's name as well.
It is quite useful to know which process is the parent of a faulting
thread, although this information can be read from the assert dump, in
some cases the dump might be incomplete (due to e.g. stack corruption,
which causes another exception and PANIC().)
test config: ./tools/configure.sh -l qemu-armv8a:nsh_smp
Pass ostest
No matter big-endian or little-endian, ticket spinlock only check the
next and the owner is equal or not.
If they are equal, it means there is a task hold the lock or lock is
free.
Signed-off-by: TaiJu Wu <tjwu1217@gmail.com>
Co-authored-by: Xiang Xiao <xiaoxiang781216@gmail.com>
clock_getcycle always returns an incremented cycle value
If the hardware does not support perf event it will use arch_alarm's up_perf_gettime
Signed-off-by: yinshengkai <yinshengkai@xiaomi.com>
1) Previously adjustments less than 1 microsecond per tick would be
completely ignored. Now they are applied over a shorter period at
a rate of 1 us per tick.
2) Previously CLOCK_ADJTIME_PERIOD was in units of 1/100th of second.
Change to milliseconds to be more generally useful unit.
Change setting name to CLOCK_ADJTIME_PERIOD_MS to make the unit change
easier to notice.
3) Previously CLOCK_ADJTIME_SLEWLIMIT was in percentage.
Most clock crystals have better accuracy than 1%, so the minimum slew
rate was excessive. Change to CLOCK_ADJTIME_SLEWLIMIT_PPM with setting
value in parts per million.
4) No need to use floating point math in clock_adjtime.c.
CONFIG_SCHED_CPULOAD_EXTCLK doesn't actually require tickless mode.
As long as the platform provides external call to nxsched_process_cpuload(),
it will work in either tickless or ticking mode.
Removed Kconfig dependency.
Instead, CONFIG_SCHED_CPULOAD_SYSCLK does require ticking mode to work,
as documented in CONFIG_SCHED_CPULOAD help text.
Added the dependency to Kconfig also.
* build-globals.sh
- Only look in the nuttx for external symbols used when loading
dynamic shared objects
* include/elf64.h
- Correct the type of fields in the Elf64_Phdr structure
* libs/libc/dlfcn/lib_dlclose.c
- Distinguish between ET_DYN and other objects as the former
has both text and data in a single allocation to reserve
GOT offsets
* libs/libc/dlfcn/lib_dlopen.c
- Code formatting
* libs/libc/modlib/modlib_bind.c
- Distinguish between relocation entry sizes by section type
- Handle RELA style relocations
* libs/libc/modlib/modlib_globals.S
- Formatting fixes
- Symbols should not be weak - they exist or they don't
* include/nuttx/lib/modlib.h
- Add an inidcator to module_s to distinguish between ET_DYN and other
* libs/libc/modlib/modlib_load.c
- ET_DYN objects need to keep the relative displacement between the text
and data sections due to GOT references from the former to the latter.
This also implies that linking may require modification from the default
for the shared objects being produced. For example, default alignment may
mean nearly 64K of wasted space.
* libs/libc/modlib/modlib_unload.c
sched/module/mod_rmmod.c
- Distingusih between freeing of ET_DYN storage and other as the former
is a single allocation.
* libs/libc/modlib/mod_insmod.c
- Cater for ET_DYN objects having init and preinit sections
bug:
user thread: hpwork:
timer_create() with SIGEV_THREAD
timer_settime()
irq -> work_queue() add nxsig_notification_worker to Q
timer_delete()
nxsig_cancel_notification()
call nxsig_notification_worker()
work_cancel()
timer_free()
nxsig_notification_worker() used after free
root cause:
work_cancel() can't cancel work completely, the worker may alreay be running.
resolve:
use work_cancel_sync() API to cancel the work completely
Signed-off-by: ligd <liguiding1@xiaomi.com>
