This solution to the problem noted by EunBong Song results in major memory fragmentation and and out-of-memory conditions on the PX4 platform. On that platform the lower priority work queue is very low priority and essentially never runs when the system is busy. As a result, the systems gets slowly starved of memory until failures and bad behaviors begin to occur.
This is an addition patch coming later to result the original problem in a different way that does not have cause memory starvation.
This reverts commit 91aa26774b.
fs/procfs/fs_procfsproc: Extended the process ID ProcFS output to show per-thread maximum time for pre-emption disabled and maximum time within a critical section.
sched/sched/sched_critmonitor.c: Adds data collection logic in support of monitoring critical sections and pre-emption state.
1. Move exepath_*() related code to libc/misc
1. Rename exepath_ to envpath_
2. Rename BINFMT_EXEPATH to LIB_ENVPATH
libs/libc/modlib: Add pre module library symbol table support
many locations: Change occurences of open() followed by file_detach() to file_open(). Change most non-controversion calls to open() to nx_open().
fs/inode/fs_fileopen.c: Flesh out file_open() with some interim, placeholder logic.
fs/inode/fs_fileopen.c: Add a framework for a file_open() implementation (no real logic in place yet).
fs/vfs/fs_open.c: Add nx_open() which is the same as open() except that it does not create a cancellation point nor does it modify the errno variable.
sched/wqueue/kwork_notifier.c: Redesign some data structures. struct works_s must appear at the beginning of the notifier entry structure. That is because it contains the work queue indices. This solves a harfault issue.
net/tcp/tcp_netpoll.c: tcp_iob_work() needs to free the allocated argument when it is finished.
net/tcp/tcp_send_buffered.c: Extend psock_tcp_cansend() so that it also requires that at least on IOB is also avaialble.
mm/iob: iob_navail() was returning the number of free IOB chain queue entries, not the number of free IOBs. Completely misnamed.
net/tcp/tcp_netpoll.c: Add logic to receive notifications when IOBs are freed (Needs CONFIG_NET_TCP_WRITE_BUFFERS and CONFIG_IOB_NOTIFIER). At present, does nothing because the logic in in psock_tcp_cansend() does not check for the availability of IOBs. That will change.
TASK A TASK B
malloc()
mm_takesemaphore()
heap holder is set to TASK B
<--- preempt
...
task_exit()
Set to current task to
TASK B
Try to release tcb, and
stack memory
free()
mm_takesemaphore()
- Successfully obtain
semaphore because current
task and heap holder is
same.
Free memory....
Heap corrupt.
This change forces all de-allocations via sched_kfree() and sched_ufree()
to be delayed. Eliminating the immediate de-allocation prevents the
above problem with the the re-entrant semaphore because the deallocation
always occurs on the worker thread, never on TASK B.
There could be consequences in the timing of memory availability. We
will see.
Squashed commit of the following:
Fix up some final compile isses.
net/netdev: Convert the network down notification logic to use the new wqueue-based notification factility.
net/udp: Convert the UDP readahead notification logic to use the new wqueue-based notification factility.
net/tcp: Convert the TCP readahead notification logic to use the new wqueue-based notification factility.
mm/iob: Convert the IOB notification logic to use the new wqueue-based notification factility.
sched/wqueue: Signals are not good IPCs to support the target poll functionality for several reasons including the amount of data that can be passed with a signal and in the fact that in protected and kernel modes, user threads executing signal handlers in protected, kernel memory is problematic. Instead, convert the same logic to perform the notifications via function callback on the high priority work queue.
Squashed commit of the following:
mm/iob: The IOB available notifier is now just a wrapper around the common signal notifier.
sched/signal: Add a generic signal notification facility.
sched/signal/sig_evthread.c: More trivial naming changes.
sched/signal: Rename nxsig_notification() to nxsig_evthread() to make forthcoming naming additions more consistent.
Squashed commit of the following:
Add procfs support to show stopped tasks. Add nxsig_action() to solve a chicken and egg problem: We needed to use sigaction to set default actions, but sigaction() would refuse to set actions if the default actions could not be caught or ignored.
sched/signal: Add configuration option to selectively enabled/disable default signal actions for SIGSTOP/SIGSTP/SIGCONT and SIGKILL/SIGINT. Fix some compilation issues.
sched/sched: Okay.. I figured out a way to handle state changes that may occur while they were stopped. If a task/thread was already blocked when SIGSTOP/SIGSTP was received, it will restart in the running state. I will appear that to the task/thread that the blocked condition was interrupt by a signal and returns the EINTR error.
sched/group and sched/sched: Finish framework for continue/resume logic.
sched/signal: Roughing out basic structure to support task suspend/resume
drivers/serial/Kconfig: It is no longer necessary to restrict Ctrl-C handling to the FLAT build
sched/signal: Add a new configuration option to select signal default actions, separate handling of signal default actions from both task startup logic and from the serial TTY Ctrl-C logic. Add a signal set in the group structure to keep track of what signals have been set to the default action. In dispatching signals in PROTECTED or KERNEL mode, use this signal set to determine if the default signal handler is attached and dispatch the signal in kernel mode for the default actions.
drivers/serial: Add Ctrl-C/Ctrl-X support. This is an initial working implementation that still requires that several details be resolved.
sched/task/task_start.c: Add default signal action to SIGKILL.
sched/wqueue: Modify high priority work queue to support multiple threads.
sched/wqueue and libs/libc/wqueue: workqueues don't need set global data to zero since .bss is cleared automatically. Removing this unnecessary initialization also avoids the loss the work items queued before initialization.
include/nuttx/arch.h: Add prototype for an architecture-specific up_trigger_irq function
arch/, include/nuttx, sched/sched: Add the garbage collection hook so each architecture can do custom memory cleanup if necesary.
arch/Kconfig: Add configureation CONFIG_ARCH_GNU_NO_WEAKFUNCTIONS to suppress use of weak functions. Some gnu derived toolchains do not support weak symbols
Author: Xiang Xiao <xiaoxiang@xiaomi.com>
arch/arm/src/armv7-m: Implement SYSTICK timer driver
Author: anchao <anchao@pinecone.net>
configs/sim/loadable: Add loadable demo for arch sim
sched/init: Restore previously reverted option to mount block device for INIT_FILEPATH case
configs/stm3240g-eval/knxwm: Fix build breakage.
Author: Gregory Nutt <gnutt@nuttx.org>
configs: With recent changes to apps/ all configurations that build applications as modules must now also select CONFIG_APPS_LOADABLE.
Author: anchao <anchao@pinecone.net>
configs/sim/loadable: Add loadable demo for arch sim
Kconfig and tools/Makefile.unix and : add dynamic application loadable support
sched/group: Fix a deadlock when loading an ELF
Signed-off-by: Masayuki Ishikawa <Masayuki.Ishikawa@jp.sony.com>
Approved-by: GregoryN <gnutt@nuttx.org>
sched/mqueue/mq_notify.c: Fix - only part of struct copied to internal structure
As source adress the adress of the struct member sigev_value was used,
and it is not located the start of the struct.
This leads to invalid data being copied to internal structure.
Both source and destination should be of type struct sigevent* .
Approved-by: GregoryN <gnutt@nuttx.org>
binfmt/, libs/libc/unistd, and others: Rename CONFIG_EXECFUNCS_SYMTAB to CONFIG_EXECFUNCS_SYMTAB_ARRAY. Rename CONFIG_EXECFUNCS_NSYMBOLS to CONFIG_EXECFUNCS_NSYMBOLS_VAR. Unlike CONFIG_EXECFUNCS_NSYMBOLS, CONFIG_EXECFUNCS_NSYMBOLS_VAR holds the name of an 'int' variable that contains the number of symbols in the symbol table.
configs/sama5d4-ek: Partial update to knsh build instructions.
sched/clock: Move the implementation of clock() from libs/libc/time to sched/clock. This is necessary because it calls the (now) internal OS function clock_systimer. clock() is now accessed only via a system call in certain configuratins.
libs/libc/wqueue: Replace calls to clock_systimer() with calls to the equivalent clock().
Squashed commit of the following:
sched: Rename all use of system_t to clock_t.
syscall: Rename all use of system_t to clock_t.
net: Rename all use of system_t to clock_t.
libs: Rename all use of system_t to clock_t.
fs: Rename all use of system_t to clock_t.
drivers: Rename all use of system_t to clock_t.
arch: Rename all use of system_t to clock_t.
include: Remove definition of systime_t; rename all use of system_t to clock_t.
The time slice check will decrement the currently running task's time slice allotment. If such a context switch occurs, then the newly started task will lost one could of that allotment before it even has a chance to run.
Change all Sony related copyright to conform with our company's internal rules.
Signed-off-by: Masayuki Ishikawa <Masayuki.Ishikawa@jp.sony.com>
Approved-by: Gregory Nutt <gnutt@nuttx.org>
Fix SMP related bugs
* sched/sched: Fix a deadlock in SMP mode
Two months ago, I introduced sched_tasklist_lock() and
sched_tasklist_unlock() to protect tasklists in SMP mode.
Actually, this change works pretty well for HTTP audio
streaming aging test with lc823450-xgevk.
However, I found a deadlock in the scheduler when I tried
similar aging tests with DVFS autonomous mode where CPU
clock speed changed based on cpu load. In this case, call
sequences were as follows;
cpu1: sched_unlock()->sched_mergepending()->sched_addreadytorun()->up_cpu_pause()
cpu0: sched_lock()->sched_mergepending()
To avoid this deadlock, I added sched_tasklist_unlock() when calling
up_cpu_pause() and sched_addreadytorun(). Also, added
sched_tasklist_lock() after the call.
Signed-off-by: Masayuki Ishikawa <Masayuki.Ishikawa@jp.sony.com>
* libc: Add critical section in lib_filesem.c for SMP
To set my_pid into fs_folder atomically in SMP mode,
critical section API must be used.
Signed-off-by: Masayuki Ishikawa <Masayuki.Ishikawa@jp.sony.com>
* mm: Add critical section in mm_sem.c for SMP
To set my_pid into mm_folder atomically in SMP mode,
critical section API must be used.
Signed-off-by: Masayuki Ishikawa <Masayuki.Ishikawa@jp.sony.com>
* net: Add critical section in net_lock.c for SMP
To set my pid (me) into fs_folder atomically in SMP mode,
critical section API must be used.
Signed-off-by: Masayuki Ishikawa <Masayuki.Ishikawa@jp.sony.com>
Approved-by: Gregory Nutt <gnutt@nuttx.org>
lc823450 smp test
* sched/clock: Replace critical section APIs with spin lock APIs in clock_gettime.c
This change will improve performance for SMP systems but nothing
changes for non-SMP systems. (Pls see include/nuttx/irq.h)
Signed-off-by: Masayuki Ishikawa <Masayuki.Ishikawa@jp.sony.com>
* sched/sched: Remove unnecessary DEBUGASSERT in sched_removereadytorun.c
In SMP mode, rtrtcb is not always at the g_readytorun.head.
This change removes DEBUGASSERT() to avoid this condition.
Signed-off-by: Masayuki Ishikawa <Masayuki.Ishikawa@jp.sony.com>
Approved-by: Gregory Nutt <gnutt@nuttx.org>
Fix signal handing for smp
* sched/signal: Remove SMP related logic in sig_dispatch.c
This change prevents from a deadlock in up_schedulesigaction.c
where inter-CPU signal handling is actually implemented.
Signed-off-by: Masayuki Ishikawa <Masayuki.Ishikawa@jp.sony.com>
* arch/arm/src/armv7-m: Fix signal handling for SMP
In previous implementation, signal handling for SMP was incorrect.
Thus, for example, if an inter-CPU signal happened an incorret tcb
was signaled and caused ASSERT().
This change fixes the issues and works for both inter-CPU signal
handling and signal handling on the same CPU.
Signed-off-by: Masayuki Ishikawa <Masayuki.Ishikawa@jp.sony.com>
Approved-by: Gregory Nutt <gnutt@nuttx.org>
sched/: Convert legitimate uses of task_create() to nxtask_create(). Review handling of returned values from all uses of kthread_create() (as well as nxtask_create()).
graphics/: Review return values for all calls to kthread_start() because it no longer returns an errno.
drivers/: threads started by drivers should be kernel threads, not user tasks. Review return values for all calls to kthread_start() because it no longer returns an errno.
configs/: threads started by board bringup logic should be kernel threads, not user tasksi (part 2 of 2).
sched/task: Add nxtask_create(). Kthread_create() and nxtask_create() are internal OS functions and should not modify the errno variable. configs/: threads started by board bringup logic should be kernel threads, not user tasks.
The previous implementation of clearing global IRQ in sched_addreadytorun()
and sched_removereadytorun() was done too early. As a result, nxsem_post()
would have a chance to enter the critical section even nxsem_wait() is
still not in blocked state. This patch moves clearing global IRQ controls
from sched_addreadytorun() and sched_removereadytorun() to sched_resumescheduler()
to ensure that nxsem_post() can enter the critical section correctly.
For this change, sched_resumescheduler.c is always necessary for SMP configuration.
In addition, by this change, task_exit() had to be modified so that it calls
sched_resumescheduler() because it calls sched_removescheduler() inside the
function, otherwise it will cause a deadlock.
However, I encountered another DEBUGASSERT() in sched_cpu_select() during
HTTP streaming aging test on lc823450-xgevk. Actually sched_cpu_select()
accesses the g_assignedtasks which might be changed by another CPU. Similarly,
other tasklists might be modified simultaneously if both CPUs are executing
scheduling logic. To avoid this, I introduced tasklist protetion APIs.
With these changes, SMP kernel stability has been much improved.
Signed-off-by: Masayuki Ishikawa <Masayuki.Ishikawa@jp.sony.com>
binfmt/, configs/, grahics/, libc/, mm/, net/, sched/: OS references to the errno variable should always use the set_errno(), get_errno() macros
arch/arm/src/stm32 and stm32f7: Architecture-specific code is not permitted to modify the errno variable. drivers/ and libc/: OS references to the errno variable should always use the set_errno(), get_errno() macros
sched/sched: Correct some build issues introduced by last set of changes.
sched/sched: Add new internal OS function nxsched_setaffinity() that is identical to sched_isetaffinity() except that it does not modify the errno value. All usage of sched_setaffinity() within the OS is replaced with nxsched_setaffinity().
sched/sched: Internal functions sched_reprioritize() and sched_setpriority() no longer movidify the errno value. Also renamed to nxsched_reprioritize() and sched_setpriority().
sched/sched: Add new internal OS function nxsched_getscheduler() that is identical to sched_getscheduler() except that it does not modify the errno value. All usage of sched_getscheduler() within the OS is replaced with nxsched_getscheduler().
sched/sched: Add new internal OS function nxsched_setparam() that is identical to sched_setparam() except that it does not modify the errno value. All usage of sched_setparam() within the OS is replaced with nxsched_setparam().
sched/sched: Add new internal OS function nxsched_getparam() that is identical to sched_getparam() except that it does not modify the errno value (actually, the previous value erroneously neglected to set the errno value to begin with, but this fixes both issues). All usage of sched_getparam() within the OS is replaced with nxsched_getparam().
These APIs are used in sched_note.c to protect instumentation data.
The deffrence between these APIs to exsiting spin_lock() and spin_unlock()
is that they do not perform insturumentation to avoid recursive call
when SCHED_INSTRUMENTATION_SPINLOCKS=y.
Signed-off-by: Masayuki Ishikawa <Masayuki.Ishikawa@jp.sony.com>
If SMP=n or SMP=y && SPINLOCK_IRQ=n, this works in the same way as before.
If SMP=y && SPINLOCK_IRQ=y, performance will be improved.
Signed-off-by: Masayuki Ishikawa <Masayuki.Ishikawa@jp.sony.com>
In ARM document regarding memory barrires, SP_DMB() must be issued
before changing a spinlock state to SP_UNLOCKED. However, we found
that SP_DSB() is also needed to ensure that spin_unlock() works
correctly for network streaming aging test.
Signed-off-by: Masayuki Ishikawa <Masayuki.Ishikawa@jp.sony.com>
sched/sched: Remove DEBUGASSERT() in sched_mergepending()
Because this DEBUGASSERT() assumes that while loop executes only once.
Signed-off-by: Masayuki Ishikawa <Masayuki.Ishikawa@jp.sony.com>
Approved-by: Gregory Nutt <gnutt@nuttx.org>
SMP: Introduce spin_lock_irqsave() and spin_unlock_irqrestore()
These APIs are simplified version of enter_critical_section() and
leave_critical_section() to protect data (e.g. registers) in SMP mode.
By using these APIs inside drivers, performace will be improved.
Signed-off-by: Masayuki Ishikawa <Masayuki.Ishikawa@jp.sony.com>
Approved-by: Gregory Nutt <gnutt@nuttx.org>
Change all calls to mq_receive() and mq_timedreceive() in the OS to calls to nxmq_receive() and nxmq_timedreceive(), making appropriate changes for differences in return values.
sched/mqueue: Add nxmq_receive() and mxmq_timedreceive() which are functionally equivalent to the standard mq_receive and mq_timedreceive() except that (1) they do not create cancellation points, and (2) the do not modify the application's errno variable.
Change all calls to mq_send() and mq_timedsend() in the OS to calls to nxmq_send() and nxmq_timedsend(), making appropriate changes for differences in return values.
sched/mqueue: Add internal function nxmq_send() and nxmq_timedsend() that are equivalent to mq_send() and mq_timedsend() except that they do not create cancellation points and do to not modify the errno variable.
sched/mqueue: Rename all private static functions for use the nxmq_ vs. mq_ naming.
sched/mqueue: Rename all OS internal functions declared in sched/mqueue/mqueue.h to begin with nxmq_ vs. mq_. The mq_ prefix is reserved for standard application interfaces.
Replace all calls to sigprocmask() in the OS proper with calls to nxsig_procmask().
sched/signal: Add internal OS interface nxsig_procmask(). This internal interface is equivalent to the standard sigprocmask() used by applications except that it does not modify the errno value. Also fixes a problem in that the original sigprocmask() was not setting the errno.
Replace all calls to sigqueue() in the OS proper with calls to nxsig_queue() to avoid accessing the errno variable.
sched/signal: Add nxsig_queue() which is functionally equivalent to sigqueue() except that it does not modify the errno variable.
Replace all usage kill() in the OS proper with nxsig_kill().
sched/signal: Add nxsig_kill() which is functionally equivalent to kill() except that it does not modify the errno variable.
Squashed commit of the following:
Change all calls to usleep() in the OS proper to calls to nxsig_usleep()
sched/signal: Add a new OS internal function nxsig_usleep() that is functionally equivalent to usleep() but does not cause a cancellaption point and does not modify the errno variable.
sched/signal: Add a new OS internal function nxsig_sleep() that is functionally equivalent to sleep() but does not cause a cancellaption point.
sigtimedwait() -> nxsig_timedwait()
sigwaitinfo() -> nxsig_waitinfo()
nanosleep() -> nxsig_nanosleep()
The internal OS versions differ from the standard application interfaces in that:
- They do not create cancellation points, and
- they do not modify the application's errno variable
Squashed commit of the following:
sched/signal: Replace all usage of sigwaitinfo(), sigtimedwait(), and nanosleep() with the OS internal counterparts nxsig_waitinfo(), nxsig_timedwait(), and nxsig_nanosleep().
sched/signal: Add nxsig_nanosleep(). This is an internal OS version of nanosleep(). It differs in that it does not set the errno varaiable and does not create a cancellation point.
sched/signal: Add nxsig_timedwait() and nxsig_waitinfo(). These are internal OS versions of sigtimedwait() and sigwaitinfo(). They differ in that they do not set the errno varaiable and they do not create cancellation points.
What was I thinking? I missed that litle minus sign and the possibility that the errno might be some positive non-zero value.
This reverts commit 43880878e4.
This is analogous to similar renaming that was done previously for semaphores.
Squashed commit of the following:
sched/signal: Fix a few compile warnings introduced by naming changes.
sched/signal: Rename all private, internal signl functions to use the nxsig_ prefix.
sched/signal: Rename sig_removependingsignal, sig_unmaskpendingsignal, and sig_mqnotempty to nxsig_remove_pendingsignal, nxsig_unmask_pendingsignal, and nxsig_mqnotempty to make it clear that these are OS internal interfaces.
sched/signal: Rename sig_findaction and sig_lowest to nxsig_find_action and nxsig_lowest to make it clear that these are OS internal interfaces.
sched/signal: Rename sig_allocatepingsigaction and sig_deliver to nxsig_alloc_pendingsigaction and nxsig_deliver to make it clear that these are OS internal interfaces.
sched/signal: Rename sig_cleanup, sig_release, sig_releasependingaction, and sig_releasependingsignal to nxsig_cleanup, nxsig_release, nxsig_release_pendingaction, and nxsig_release_pendingsignal to make it clear that these are OS internal interfaces.
sched/signal: Rename sig_tcbdispatch and sig_dispatch to nxsig_tcbdispatch and nxsig_dispatch to make it clear that these are OS internal interfaces.
sched/signal: Rename sig_releaseaction and sig_pendingset to nxsig_release_action and nxsig_pendingset to make it clear that these are OS internal interfaces.
sched/signal: Rename sig_initialize and sig_allocateactionblock to nxsig_initialize and nxsig_alloc_actionblock to make it clear that these are OS internal interfaces.
