The user stack is dependent on the user address environment. As the
process exits, this address environment is destroyed anyway, so the
stack does not need to be released separately.
There is also an issue with this when the process exits via exit(). The
problem is that the task group is released prior to this "up_release_stack()"
call along with the address environment, and trying to free the memory
either causes an immediate crash (no valid addrenv), or frees memory into
another process' heap (addrenv from a different process).
Signed-off-by: Ville Juven <ville.juven@unikie.com>
It takes about 10 cycles to obtain the task list according to the task
status. In most cases, we know the task status, so we can directly
add the task from the specified task list to reduce time consuming.
It takes about 10 cycles to obtain the task list according to the task
status. In most cases, we know the task status, so we can directly
delete the task from the specified task list to reduce time consuming.
1.Don't include unwind.h when arch specific backtrace is enable
2.Built arch specific backtrace wrapper only when enable
Signed-off-by: Xiang Xiao <xiaoxiang@xiaomi.com>
Optimize sched_note_begin/end, replace note_printf with note_string
sched_note_begin/end optimized from 50us to 1us per consumption
Signed-off-by: yinshengkai <yinshengkai@xiaomi.com>
Situation:
Assume we have 2 cpus, and busy run task0.
CPU0 CPU1
task0 -> task1 task2 -> task0
1. remove task0 form runninglist
2. take task1 as new tcb
3. add task0 to blocklist
4. clear spinlock
4.1 remove task2 form runninglist
4.2 take task0 as new tcb
4.3 add task2 to blocklist
4.4 use svc ISR swith to task0
4.5 crash
5. use svc ISR swith to task1
Fix:
Move clear spinlock to the end of svc ISR
Signed-off-by: ligd <liguiding1@xiaomi.com>
it inappropriate to apply volatile to the task list:
1.The code access task list is already protected by critical section
2.The queue is complex struct, it isn't enough to protect by volatile
Signed-off-by: Xiang Xiao <xiaoxiang@xiaomi.com>
because not all compiler support the weak attribute, and
many features are either always used or guarded by config.
Signed-off-by: Xiang Xiao <xiaoxiang@xiaomi.com>
1.Forward nxsched_process_cpuload to nxsched_process_cpuload_ticks directly
2.Define the dummy nxsched_process_cpuload_ticks when CPULOAD isn't enabled
3.Remove the weak attribute from nxsched_process_cpuload_ticks
Signed-off-by: Xiang Xiao <xiaoxiang@xiaomi.com>
and remove CONFIG_LIBC_LONG_LONG option to simplify the usage.
note: the size will increase 668
before change:
text data bss dec hex filename
168440 348 4480 173268 2a4d4 nuttx
after change:
text data bss dec hex filename
169108 348 4480 173936 2a770 nuttx
Signed-off-by: Xiang Xiao <xiaoxiang@xiaomi.com>
1. fix typo spilock -> spinlock
2. fix build break
sched/sched_note.c: In function ‘note_spincommon’:
sched/sched_note.c:435:3: error: aggregate value used where an integer was expected
435 | note.nsp_value = (uint8_t)*spinlock;
| ^~~~
Signed-off-by: chao.an <anchao@xiaomi.com>
The implementation of this feature is based on android systrace:
https://source.android.com/devices/tech/debug/ftrace
Application developers are more concerned about the performance of
the specified application section,
added two APIs to implement performance measurement:
void sched_note_begin(uintptr_t ip, FAR const char *buf);
void sched_note_end(uintptr_t ip, FAR const char *buf);
or
SCHED_NOTE_BEGIN(); /* defined to sched_note_begin(_THIS_IP_, __FUNCTION__) */
SCHED_NOTE_END(); /* defined to sched_note_end(_THIS_IP_, __FUNCTION__) */
Signed-off-by: chao.an <anchao@xiaomi.com>
Perf timer interface generally uses the hardware cycle counter
provided by the arch chip directly(such as DWT_CYCCNT(cortex-m)),
CYCCNT is a free running counter and counting upwards.
It wraps around to 0 on overflow.
Signed-off-by: chao.an <anchao@xiaomi.com>
The BUG is as follow:
1. Suppose there are two ready tasks t1 and t2 have the same priority, and t1 is the running.
2. A new, higher priority task t0 (such as hpwork) is ready, then switch to t0.
3. After t0 is suspended, then switch to t1 and reset the t1's timeslice.
4. goto 2
The t2 will have no chance to run.
Signed-off-by: Zeng Zhaoxiu <zhaoxiu.zeng@gmail.com>
1. If CONFIG_SCHED_INSTRUMENTATION_SWITCH is enabled, then these additional interfaces are expected:
sched_note_suspend
sched_note_resume
at the same time,If CONFIG_SMP is enabled,then expected:
sched_note_cpu_pause
sched_note_cpu_paused
sched_note_cpu_resume
sched_note_cpu_resumed
2. other board if enable SCHED_INSTRUMENTATION,SCHED_INSTRUMENTATION_SWITCH to defconfig.
If a task registers SIGCHLD, uses waitpid(-1)
in the handler, and the task also calls waitpid,
it will fail to delete the child function
Signed-off-by: anjiahao <anjiahao@xiaomi.com>
Fix the regression by commit:
commit 8b67944c75
Author: Xiang Xiao <xiaoxiang@xiaomi.com>
Date: Thu Oct 14 11:03:07 2021 +0800
sched: Remove pidhash_s and move ticks to tcb_s
simplify the code logic and reduce memory a little bit
Signed-off-by: Xiang Xiao <xiaoxiang@xiaomi.com>
Signed-off-by: Xiang Xiao <xiaoxiang@xiaomi.com>
the logical may choose the head of g_readytorun(the greatest priority)
task as the next while ignoring the cpu affinity of it.
Signed-off-by: chenhonglin <chenhonglin@xiaomi.com>
In "nxsched_nexttcb": the task may not running on this_cpu,
and rtrtcb->affinity(the affinity of the task in g_readytorun)
may not include the current cpu which should be the tcb->cpu.
Signed-off-by: chenhonglin <chenhonglin@xiaomi.com>
Fail case:
exit -> nxtask_terminate -> nxtask_exithook -> nxsched_release_tcb
group_leave || nxsched_releasepid & group_leave
/\
/ \
switch out & waitpid()
Thread A group_leave in nxtask_exithook, switch out,
Thread B do waitpid(thread A) then meet traget thread A group is NULL, error.
Change-Id: Ia181d7a13aa645ec1c3141a45839fbf79db35b17
Signed-off-by: ligd <liguiding1@xiaomi.com>
Summary:
- SP_SECTION was introduced to allocate spinlock in non-cachable
region mainly for Cortex-A to stabilize the NuttX SMP kernel
- However, all spinlocks are now allocated in cachable area and
works without any problems
- So SP_SECTION should be removed to simplify the kernel code
Impact:
- None
Testing:
- Build test only
Signed-off-by: Masayuki Ishikawa <Masayuki.Ishikawa@jp.sony.com>
arch: Allocate the space from the beginning in up_stack_frame
and modify the affected portion:
1.Correct the stack dump and check
2.Allocate tls_info_s by up_stack_frame too
3.Move the stack fork allocation from arch to sched
Signed-off-by: Xiang Xiao <xiaoxiang@xiaomi.com>
Fix missing semicolon at the end of a DEBUGASSERT statement:
sched/sched_sporadic.c: In function 'sporadic_budget_expire':
sched/sched_sporadic.c:512:15: error: expected ';' before 'period'
512 | period = (sporadic->repl_period >> 1) - unrealized;
| ^~~~~~
sched/sched_sporadic.c: In function 'nxsched_resume_sporadic':
sched/sched_sporadic.c:1078:19: error: expected ';' before 'period'
1078 | period = (sporadic->repl_period >> 1) - unrealized;
| ^~~~~~
Fix use of uninitialized variable in DEBUGASSERT statement:
sched/sched_sporadic.c:466:27: warning: 'sporadic' may be used uninitialized in this function [-Wmaybe-uninitialized]
466 | sporadic->nrepls > 0);
Also fixes some typos.
There should be no unexpected side-effects of this changed.
Tested with the stm32f4discovery:sporadic configuration (see PR #3097
it is wrong to define a new grpid_t, but not reuse pid_t,
because it make getpid(parent) == getppid(child) impossible.
Signed-off-by: Xiang Xiao <xiaoxiang@xiaomi.com>
Summary:
- Because this_task() and nxsched_get_tcb() are protected inside the funtcions
- Also, enter_critical_section() is used in nxsched_set_affinity()
Impact:
- No impact
Testing:
- Tested with spresense:wifi_smp
Signed-off-by: Masayuki Ishikawa <Masayuki.Ishikawa@jp.sony.com>
Summary:
- Calling sched_lock()/unlock() is unnecessary for SMP
because we've been using the critical section API
Impact:
- SMP only
Testing:
- Tested with smp and ostest with the following configurations
- sabre-6quad:smp (QEMU, dev board)
- spresense:smp, spresense:wifi_smp
- sim:smp, sim:ostest
- maix-bit:smp (QEMU)
- esp32-devkitc:smp (QEMU)
- lc823450-xgevk:rndis
Signed-off-by: Masayuki Ishikawa <Masayuki.Ishikawa@jp.sony.com>
Summary:
- This commit fixes global IRQ control logic
- In previous implementation, g_cpu_irqset for a remote CPU was
set in sched_add_readytorun(), sched_remove_readytorun() and
up_schedule_sigaction()
- In this implementation, they are removed.
- Instead, in the pause handler, call enter_critical_setion()
which will call up_cpu_paused() then acquire g_cpu_irqlock
- So if a new task with irqcount > 1 restarts on the remote CPU,
the CPU will only hold a critical section. Thus, the issue such as
'POSSIBLE FOR TWO CPUs TO HOLD A CRITICAL SECTION' could be resolved.
- Fix nxsched_resume_scheduler() so that it does not call spin_clrbit()
if a CPU does not hold a g_cpu_irqset
- Fix nxtask_exit() so that it acquires g_cpu_irqlock
- Update TODO
Impact:
- All SMP implementations
Testing:
- Tested with smp, ostest with the following configurations
- Tested with spresense:wifi_smp (NCPUS=2,4)
- Tested with sabre-6quad:smp (QEMU, dev board)
- Tested with maix-bit:smp (QEMU)
- Tested with esp32-core:smp (QEMU)
- Tested with lc823450-xgevk:rndis
Signed-off-by: Masayuki Ishikawa <Masayuki.Ishikawa@jp.sony.com>
Summary:
- I noticed waitpid_test stops with lc823450-xgevk:rndis
- The condition was CONFIG_DEBUG_ASSERTION=y
- Actually, the child task sent SIGCHILD but the parent couldn't catch the signal
- Then, I found that nx_waitid(), nx_waitpid() use sched_lock()
- However, a parent task and a child task are running on different CPUs
- So, sched_lock() is not enough and need to use a critical section
- Also, signal handling in nxtask_exithook() must be done in a critical section
Impact:
- SMP only
Testing:
- Tested with ostest with the following configurations
- lc823450-xgevk:rndis (CONFIG_DEBUG_ASSERTION=y and n)
- spresense:smp
- spresense:wifi_smp (NCPUS=2 and 4)
- sabre-6quad:smp (QEMU)
- esp32-core:smp (QEMU)
- maix-bit:smp (QEMU)
- sim:smp
Signed-off-by: Masayuki Ishikawa <Masayuki.Ishikawa@jp.sony.com>
Summary:
- sched_tasklistlock.c was introduced to stabilize NuttX SMP
- However, the current SMP implementation is stable by recent fixes
- So I decided to remove the file finally
Impact:
- SMP only
Testing:
- Tested with ostest with the following configurations
- spresense:smp
- spresense:wifi_smp (NCPUS=2 and 4)
- sabre-6quad:smp (QEMU)
- esp32-core:smp (QEMU)
- maix-bit:smp (QEMU)
- sim:smp
- lc823450-xgevk:rndis
Signed-off-by: Masayuki Ishikawa <Masayuki.Ishikawa@jp.sony.com>
Summary:
- I noticed sched_lock() logic is different from sched_unlock()
- I think sched_lock() should use critical section
- Also, the code should be simple like sched_unlock()
- This commit fixes these issues
Impact:
- Affects SMP only
Testing:
- Tested with spresense:wifi_smp (both NCPUS=2 and 3)
- Tested with lc823450-xgevk:rndis
- Tested with maix-bit:smp (QEMU)
- Tested with esp32-core:smp (QEMU)
- Tested with sabre-6quad:smp (QEMU)
Signed-off-by: Masayuki Ishikawa <Masayuki.Ishikawa@jp.sony.com>
Summary:
- I noticed DEBUGASSERT() happens in sched_unlock()
- The test was Wi-Fi audio streaming stress test with spresense 3cores
- Actually, g_cpu_schedlock was locked but g_cpu_lockset was incorrect
- Finally, I found that cpu was obtained before enter_critical_section()
- And the task was moved from one cpu to another cpu
- However, that call should be done within the critical section
- This commit fixes this issue
Impact:
- Affects SMP only
Testing:
- Tested with spresense:wifi_smp (both NCPUS=2 and 3)
- Tested with lc823450-xgevk:rndis
- Tested with maix-bit:smp (QEMU)
- Tested with esp32-core:smp (QEMU)
- Tested with sabre-6quad:smp (QEMU)
Signed-off-by: Masayuki Ishikawa <Masayuki.Ishikawa@jp.sony.com>
1.Reduce the default size of task_group_s(~512B each task)
2.Scale better between simple and complex application
Signed-off-by: Xiang Xiao <xiaoxiang@xiaomi.com>
Change-Id: Ia872137504fddcf64d89c48d6f0593d76d582710
Summary:
- ARCH_GLOBAL_IRQDISABLE was initially introduced for LC823450 SMP
- At that time, i.MX6 (quad Cortex-A9) did not use this config
- However, this option is now used for all CPUs which support SMP
- So it's good timing for refactoring the code
Impact:
- Should have no impact because the logic is the same for SMP
Testing:
- Tested with board: spresense:smp, spresense:wifi_smp
- Tested with qemu: esp32-core:smp, maix-bit:smp, sabre-6quad:smp
- Build only: lc823450-xgevk:rndis, sam4cmp-db:nsh
Signed-off-by: Masayuki Ishikawa <Masayuki.Ishikawa@jp.sony.com>
Found by clang-check:
sched/sched_waitpid.c:380:33: warning: Although the value stored to 'ret' is used in the enclosing expression, the value is never actually read from 'ret'
(pid != (pid_t)-1 && (ret = nxsig_kill(pid, 0)) < 0))
^ ~~~~~~~~~~~~~~~~~~
1 warning generated.
Since up_release_stack auto detect whether the memory come from builtin heap
if (ttype == TCB_FLAG_TTYPE_KERNEL)
{
if (kmm_heapmember(dtcb->stack_alloc_ptr))
{
kmm_free(dtcb->stack_alloc_ptr);
}
}
else
{
/* Use the user-space allocator if this is a task or pthread */
if (umm_heapmember(dtcb->stack_alloc_ptr))
{
kumm_free(dtcb->stack_alloc_ptr);
}
}
This reverts commit 124e6ee53d.
If there's no child status available immediately,
return 0 without blocking as specified by the standards.
I checked the following version of the standard.
I believe it has always been this way though.
The Open Group Base Specifications Issue 7, 2018 edition
IEEE Std 1003.1-2017 (Revision of IEEE Std 1003.1-2008)
If there's no child status available immediately,
return 0 without blocking as specified by the standards.
The implementation for non CONFIG_SCHED_HAVE_PARENT case
seems ok in this regard.
I checked the following version of the standard.
I believe it has always been this way though.
The Open Group Base Specifications Issue 7, 2018 edition
IEEE Std 1003.1-2017 (Revision of IEEE Std 1003.1-2008)
sched_releasetcb() will normally free the stack allocated for a task. However, a task with a custom, user-managed stack may be created using nxtask_init() followed by nxtask_activer. If such a custom stack is used then it must not be free in this many or a crash will most likely result.
This chagne addes a flag call TCB_FLAG_CUSTOM_STACK that may be passed in the the pre-allocted TCB to nxtask_init(). This flag is not used internally anywhere in the OS except that if set, it will prevent sched_releasetcb() from freeing that custom stack.
1. Internal scheduler functions should begin with nxsched_, not sched_
2. Follow the consistent naming patter of https://cwiki.apache.org/confluence/display/NUTTX/Naming+of+OS+Internal+Functions
# clock_systimer -> clock_systime_tick
# clock_systimespec -> clock_systime_timespec
sched_oneshot_extclk -> nxsched_oneshot_extclk
sched_period_extclk -> nxsched_period_extclk
# nxsem_setprotocol -> nxsem_set_protocol
# nxsem_getprotocol -> nxsem_get_protocol
# nxsem_getvalue -> nxsem_get_value
nxsem_initholders -> nxsem_initialize_holders
nxsem_addholder -> nxsem_add_holder
nxsem_addholder_tcb -> nxsem_add_holder_tcb
nxsem_boostpriority -> nxsem_boost_priority
nxsem_releaseholder -> nxsem_release_holder
nxsem_restorebaseprio -> nxsem_restore_baseprio
Some planned name changed were skipped for now because they effect too many files (and would require many hours of coding style fixups).
All complaints fixed except for those that were not possible to fix:
- Used of Mixed case identifier in ESP32 files. These are references to Expressif ROM functions which are outside of the scope of NuttX.
- Remove per-thread errno from the TCB structure (pterrno)
- Remove get_errno() and set_errno() as functions. The macros are still available as stubs and will be needed in the future if we need to access the errno from a different address environment (KERNEL mode).
- Add errno value to the tls_info_s structure definitions
- Move sched/errno to libs/libc/errno. Replace old TCB access to the errno with TLS access to the errno.