This commit fixed windows compilation errors `struct
lp_wqueue_s/hp_wqueue_s has an illegal zero-sized array`.
Signed-off-by: ouyangxiangzhen <ouyangxiangzhen@xiaomi.com>
This patch is a rework of the NuttX file descriptor implementation. The
goal is two-fold:
1. Improve POSIX compliance. The old implementation tied file description
to inode only, not the file struct. POSIX however dictates otherwise.
2. Fix a bug with descriptor duplication (dup2() and dup3()). There is
an existing race condition with this POSIX API that currently results
in a kernel side crash.
The crash occurs when a partially open / closed file descriptor is
duplicated. The reason for the crash is that even if the descriptor is
closed, the file might still be in use by the kernel (due to e.g. ongoing
write to file). The open file data is changed by file_dup3() and this
causes a crash in the device / drivers themselves as they lose access to
the inode and private data.
The fix is done by separating struct file into file and file descriptor
structs. The file struct can live on even if the descriptor is closed,
fixing the crash. This also fixes the POSIX issue, as two descriptors
can now point to the same file.
Signed-off-by: Ville Juven <ville.juven@unikie.com>
Signed-off-by: dongjiuzhu1 <dongjiuzhu1@xiaomi.com>
For simplicity, better performance and lower memory-overhead, this commit replaced the periodical workqueue APIs with the more expressive work_queue_next. The work_queue_next restarts work based on the last expiration time.
Signed-off-by: ouyangxiangzhen <ouyangxiangzhen@xiaomi.com>
This commit replaced periodical timer with the wd_start_next to improve the timing accuracy.
Signed-off-by: ouyangxiangzhen <ouyangxiangzhen@xiaomi.com>
For better simplicity, this commit replaced the periodical wdog APIs with the more expressive wd_start_next. The wd_start_next restarts watchdog timer based on the last expiration time.
Signed-off-by: ouyangxiangzhen <ouyangxiangzhen@xiaomi.com>
This commit changed the type of the delay ticks to the unsigned, which can reduce the useless branch conditions
Besides, this commit added max delay tick limitation to fix incorrect timing behavior if we delay SCLOCK_MAX in the SMP build.
Signed-off-by: ouyangxiangzhen <ouyangxiangzhen@xiaomi.com>
There is no need to check the holder structure "counts". There are cases
where the counts may be greater than 1 when several tasks block
on the mutex, but there is always just one holder, which must be freed.
Signed-off-by: Jukka Laitinen <jukka.laitinen@tii.ae>
This is not a bug, but unnecessary code. If the mutex is no longer blocking,
the released thread will set the holder and clear the blocking bit in the end
of nxsem_wait.
Signed-off-by: Jukka Laitinen <jukka.laitinen@tii.ae>
Checking only adds a race condition. The checking if the wdog is active
or not must be done inside wd_cancel, where the proper spinlock is held
Signed-off-by: Jukka Laitinen <jukka.laitinen@tii.ae>
nxsig_timeout calls nxsched_add_readytorun and up_switch_context, so
must be in critical section.
nxsig_timeout is used as wdentry in nxsig_clockwait
See wd_expiration CALL_FUNC is not in critical section.
Signed-off-by: Serg Podtynnyi <serg@podtynnyi.com>
In case the wdog has already completed when calling cancel, the cancel is
supposed to just return OK.
Signed-off-by: Jukka Laitinen <jukka.laitinen@tii.ae>
This commit added a macro function clock_delay2abstick to calculate the
absolute tick after the delay.
Signed-off-by: ouyangxiangzhen <ouyangxiangzhen@xiaomi.com>
This commit resolves a timing error caused by the round-up behavior in clock_time2ticks. In rare cases, this could lead to a two-tick increment within a single tick interval. To fix this, we introduced clock_time2ticks_floor, which guarantees the correct semantics for obtaining current system ticks.
Signed-off-by: ouyangxiangzhen <ouyangxiangzhen@xiaomi.com>
This commit added CONFIG_TIMER_ADJUST_USEC to support the time compensation for wdog timer. Normally, timer event cannot triggered on exact time due to the existence of interrupt latency. Assuming that the interrupt latency is distributed within [Best-Case Execution Time, Worst-Case Excution Time], we can set the timer adjustment value to the BCET to reduce the latency.
Signed-off-by: ouyangxiangzhen <ouyangxiangzhen@xiaomi.com>
To improve the absolute timer accuracy, this commit move the tick++ to relative wdog timer.
Signed-off-by: ouyangxiangzhen <ouyangxiangzhen@xiaomi.com>
Or will be catch by codespell, when do checkpatch.sh
Also fix the relative comment file changed.
include/nuttx/scsi.h
drivers/syslog/ramlog.c
excluded as we have to modify field name in struct
Signed-off-by: buxiasen <buxiasen@xiaomi.com>
This commit improve the performance of the work_queue by reducing
unnecessary wdog timer setting.
Signed-off-by: ouyangxiangzhen <ouyangxiangzhen@xiaomi.com>
This commit fixed work_cancel_sync at a very rare boundary case. When a worker thread re-enqueues the work data structure during the execution of work, the user thread cannot directly dequeue the work in work_cancel_sync. Instead, it should wait until all workers' references to the work data structure have been eliminated after dequeuing.
Signed-off-by: ouyangxiangzhen <ouyangxiangzhen@xiaomi.com>
When using pthread_kill, the signal should be delivered to the
specified thread. Current implementation, however, may add the
signal to the groups pending list, if the signal is masked at the
time of dispatch. From the group's pending list it can be delivered
to any thread of the group, which is wrong.
Fix this by adding a new field "FAR struct tcb_s *tcb" to
"struct sigpendq", marking if the signal needs to be delivered
to a specific thread. Use NULL for the value if delivery to any
thread in the group is ok.
Signed-off-by: Jukka Laitinen <jukka.laitinen@tii.ae>
The kmm_alloc can break the critical section, if it sleeps on the
heap mutex. If we run out of pending signal structures, allocate more
right after entering the critical section but before checking if the signal
needs to be added to the pending queue.
Signed-off-by: Jukka Laitinen <jukka.laitinen@tii.ae>
The TCB used in find_action is locked by spinlock, so it doesn't belong
inside critical section. Just find the possible action already in the
beginning of nxsig_tcbdispatch.
Signed-off-by: Jukka Laitinen <jukka.laitinen@tii.ae>
- Remove the redundant holder, as nxsem now manages hoder TID
- Remove DEBUGASSERTIONS which are managed in nxsem
- Remove the "reset" handling logic, as it is now managed in nxsem
- Inline the simplest functions
Signed-off-by: Jukka Laitinen <jukka.laitinen@tii.ae>
This puts the mutex support fully inside nxsem, allowing
locking the mutex and setting the holder with single atomic
operation.
This enables fast mutex locking from userspace, avoiding taking
critical_sections, which may be heavy in SMP and cleanup
of nxmutex library in the future.
Signed-off-by: Jukka Laitinen <jukka.laitinen@tii.ae>
The task which is deleted should be removed from the semaphores waitlist,
if the task happens to be blocked.
Signed-off-by: Jukka Laitinen <jukka.laitinen@tii.ae>
For the watchdog list and the workqueue list, we wonder whether the list head has changed after the insertion. In the original implementation, we have to access the list->next field and compare the result to the currently inserted node. In this commit, we mark the list head before the insertion and comparing the current traversed node with the list head we marked, which can avoid accessing the list->next and is more cache-friendly.
Signed-off-by: ouyangxiangzhen <ouyangxiangzhen@xiaomi.com>
This commit refactors the logic of workqueue processing delayed and periodic work, and changes the timer to be set in `work_thread`. The improvements of this change are as follows:
- Fixed the memory reuse problem of the original periodic workqueue implementation.
- By removing the `wdog_s` structure in the `work_s` structure, the memory overhead of each `work_s` structure is reduced by about 30 bytes.
- Set the timer for each workqueue instead of each work, which improves system performance.
- Simplified the workqueue cancel logic.
Signed-off-by: ouyangxiangzhen <ouyangxiangzhen@xiaomi.com>
In NuttX, the dq and the list are two different implementations of the double-linked list. Comparing to the dq, the list implementation has less branch conditions such as checking whether the head or tail is NULL. In theory and practice, the list is more friendly to the CPU pipeline. This commit changed the dq to the list in the wqueue implementation.
Signed-off-by: ouyangxiangzhen <ouyangxiangzhen@xiaomi.com>
For example a race in managing tcb->sigprocmask may cause signal not being delivered at all.
The mechanism is as follows:
1. cpu 1: nxsig_deliver adds the signal to the sigprocmask for the duration of signal delivery
2. cpu 2: new signal (same signo) is dispatched in nxsig_tcbdispatch
3. cpu 2: nxsig_tcbdispatch detects that signal is masked
4. cpu 2: nxsig_tcbdispatch leaves critical section before calling nxsig_add_pendigsignal
5. cpu 1: nxsig_deliver finishes. The "nxsig_unmask_pendingsignal" is called in the end but does nothing
6. cpu 2: nxsig_tcbdispatch continues and adds the pending signal
In the end, the logic in the end of nxsig_deliver, which tries to handle signals added
to the pending queue during the signal action delivery (step 5) failed, and the pending signal
was not delivered.
Fix this by just keeping the critical section for during the whole duration of nxsig_tcbdispatch,
and move things which can't be executed from within critical section outside.
Signed-off-by: Jukka Laitinen <jukka.laitinen@tii.ae>
For some SMP calls it is necessary to lock the current CPU for the process
receiving the SMP call. This is done by setting the CPU affinity to the
current CPU and preventing the CPU selection algorithm from switching
CPUs.
dtcb->flags |= TCB_FLAG_CPU_LOCKED;
CPU_SET(dtcb->cpu, &dtcb->affinity);
However, this logic is currently broken, as CPU_SET is defined as:
#define CPU_SET(c,s) do { *(s) |= (1u << (c)); } while (0)
In order to assign tcb->cpu (the current CPU) to the affinity mask, the
mask must be cleared first by calling CPU_ZERO.
Renaming "modlib" to "libelf" is more in line with the implementation content,
which makes it easier for individual developers to understand the capabilities of this module.
CONFIG_LIBC_MODLIB -> CONFIG_LIBC_ELF
Signed-off-by: chao an <anchao.archer@bytedance.com>
test:
1.use mps3-an547 build helloxx as module and run it
2.use qemu-armv7a:knsh test kernel build helloxx and run it
Signed-off-by: anjiahao <anjiahao@xiaomi.com>
There is no need for a gettid() syscall, as the thread ID is stable through
the life of the process. It is safe to put a copy of TID to TLS. This way
a user processes can access TID quickly via its own stack, instead of
having to use an expensive syscall.
Signed-off-by: Ville Juven <ville.juven@unikie.com>
This avoids unnecessary syscalls in memory protected builds, when mutex
lock/unlock can be done with only atomic counter access
Signed-off-by: Jukka Laitinen <jukka.laitinen@tii.ae>
When the semaphore priority flags is set to NONE, and the semaphore
is a mutex, the fast locking path can be used, even when
priority inheritance or priority protect are enabled globally.
Signed-off-by: Jukka Laitinen <jukka.laitinen@tii.ae>
After pthread mutexes changed to nxmutex, the priority inheritance
was set on by default; even if one tried to control it with
CONFIG_PTHREAD_MUTEX_DEFAULT_PRIO_INHERIT.
Also the CONFIG_PTHREAD_MUTEX_DEFAULT_PRIO_PROTECT is not effective.
Fix this by setting the default mutex priority adjustment flags according
to CONFIG_PTHREAD_MUTEX_DEFAULT_PRIO_INHERIT and CONFIG_PTHREAD_MUTEX_DEFAULT_PRIO_PROTECT.
Signed-off-by: Jukka Laitinen <jukka.laitinen@tii.ae>
