Commit Graph

309 Commits

Author SHA1 Message Date
Xiang Xiao
11c8f3125d sched/tls: Wrap up_info_size into tls_info_size
Signed-off-by: Xiang Xiao <xiaoxiang@xiaomi.com>
2022-06-08 19:07:33 +09:00
Xiang Xiao
47b707bbf8 sched/tls: Add tls_init_info and tls_dup_info
Signed-off-by: Xiang Xiao <xiaoxiang@xiaomi.com>
2022-06-03 23:56:50 +03:00
Xiang Xiao
3e32b605fe libc/tls: Make tls_get_info as the pulibc function instead up_tls_info
Signed-off-by: Xiang Xiao <xiaoxiang@xiaomi.com>
2022-06-03 23:56:50 +03:00
Xiang Xiao
33a6aa2f48 sched/tls: Don't compensate the tls size to the stack size
it isn't good to change the stack size passed by caller

Signed-off-by: Xiang Xiao <xiaoxiang@xiaomi.com>
2022-06-03 23:56:50 +03:00
Xiang Xiao
3b136771ff sched: Move the call of pthread_mutex_inconsistent to nxtask_recover
to avoid it spread around all possible exit path

Signed-off-by: Xiang Xiao <xiaoxiang@xiaomi.com>
2022-05-27 09:39:09 +03:00
Ville Juven
a54c3d13f9 sched: Remove SCHED_ATEXIT / SCHED_ONEXIT
Remove the kernel side implementations altogether. These will be
replaced by user land implementations.
2022-05-25 15:28:43 +08:00
Ville Juven
622677d4a1 libc: Implement exit, atexit, on_exit and cxa_exit on the user side
For CONFIG_BUILD_KERNEL using the sched/task/task_exithook implementation
will just not work. It calls user code with kernel privileges which is
a bit of a security issue.
2022-05-25 15:28:43 +08:00
Ville Juven
5bcd1dbb64 sched: Remove task_restart in case of CONFIG_BUILD_KERNEL
Same treatment as task_delete, this is not usable in kernel build
either.
2022-05-12 22:08:19 +08:00
Ville Juven
b1d92159fa sched: Remove task_delete in case of CONFIG_BUILD_KERNEL
Deleting a task from another task's context will not do, so shut
this gate down for BUILD_KERNEL. In this case if a task wants another
task to terminate, it must ask the other task to politely kill itself.

Note: kthreads still need this, also, the kernel can delete a task
without asking.
2022-05-12 03:27:25 +08:00
Petro Karashchenko
1b4d8b3734 sched: remove DEBUGASSERT from nx_waitpid
Signed-off-by: Petro Karashchenko <petro.karashchenko@gmail.com>
2022-04-22 15:06:26 -03:00
Xiang Xiao
55b5561fdb sched/task: Implement execle and execve
Signed-off-by: Xiang Xiao <xiaoxiang@xiaomi.com>
2022-04-22 10:40:31 +03:00
Xiang Xiao
76803f4a07 sched/environ: Replace get_environ_ptr with environ
Signed-off-by: Xiang Xiao <xiaoxiang@xiaomi.com>
2022-04-21 22:55:47 +03:00
Xiang Xiao
ce49c80976 sched/task: Correct the comment about environment variable
Signed-off-by: Xiang Xiao <xiaoxiang@xiaomi.com>
2022-04-21 22:55:47 +03:00
Ville Juven
4c1b66246d env_dup: Fix copying of env between address environments
If address environments are in use, it is not possible to simply
memcpy from from one process to another. The current implementation
of env_dup does precisely this and thus, it fails at once when it is
attempted between two user processes.

The solution is to use the kernel's heap as an intermediate buffer.
This is a simple, effective and common way to do a fork().

Obviously this is not needed for kernel processes.
2022-04-21 18:38:37 +08:00
ligd
e9f17947b7 task: don't set default signal in kernal thread
Signed-off-by: ligd <liguiding1@xiaomi.com>
2022-03-30 10:53:14 +09:00
Petro Karashchenko
68902d8732 pid_t: unify usage of special task IDs
Signed-off-by: Petro Karashchenko <petro.karashchenko@gmail.com>
2022-03-22 21:22:32 +08:00
Petro Karashchenko
6472a698b6 style/typo: fix few style and typo issues
Signed-off-by: Petro Karashchenko <petro.karashchenko@gmail.com>
2022-03-18 17:05:10 +01:00
Xiang Xiao
a50d87d5b7 sched/group: Simplify the allocation and deallocation logic
1.Move tg_membe allocation to group_alloc
2.Merge group_deallocate to group_release

Signed-off-by: Xiang Xiao <xiaoxiang@xiaomi.com>
Signed-off-by: chao.an <anchao@xiaomi.com>
2022-03-04 12:11:24 +09:00
Huang Qi
e516c6247e sched: Implement task local storage
Signed-off-by: Huang Qi <huangqi3@xiaomi.com>
2022-02-23 22:05:02 +08:00
Huang Qi
c0a0de97ce Revert "libc: Call pthread_exit in user-space by up_pthread_exit"
This reverts commit f4a0b7aedd.

Signed-off-by: Huang Qi <huangqi3@xiaomi.com>
2022-02-09 21:21:43 +08:00
Huang Qi
6814b1416f Revert "pthread: Avoid recursive pthread_exit call"
This reverts commit a2941532bd.

Signed-off-by: Huang Qi <huangqi3@xiaomi.com>
2022-02-09 21:21:43 +08:00
Xiang Xiao
0000e61da6 task: Always implement posix_spawnp as macro
Since there is no benefit to implement it as
function when CONFIG_LIBC_ENVPATH equals y

Signed-off-by: Xiang Xiao <xiaoxiang@xiaomi.com>
2022-02-09 20:09:11 +08:00
Xiang Xiao
d48946063a Revert "sched: Don't duplicate caller file handler when creating kernel thread"
This reverts commit 512676cb06.
2022-02-01 08:22:45 +09:00
Xiang Xiao
512676cb06 sched: Don't duplicate caller file handler when creating kernel thread
kernel thread should have only the starndard file i/o just like idle thread

Signed-off-by: Xiang Xiao <xiaoxiang@xiaomi.com>
2022-01-31 14:44:29 +01:00
Petro Karashchenko
c0881feca9 boards/arm/samv7/same70-qmtech: fix description in README.txt
Signed-off-by: Petro Karashchenko <petro.karashchenko@gmail.com>
2022-01-23 20:08:10 -03:00
Xiang Xiao
1af8cd4de8 sched: Move argv from tcb_s to task_info_s
argv is allocated from stack and then belong to userspace,
so task_info_s is a best location to hold this information.

Signed-off-by: Xiang Xiao <xiaoxiang@xiaomi.com>
2021-12-11 10:08:53 -06:00
Huang Qi
58e0781e2e arch/arm: Implement TLS support
Signed-off-by: Huang Qi <no1wudi@qq.com>
2021-12-07 23:31:41 -06:00
Xiang Xiao
b1879d5d24 sched/vfork: allocate tls_info_s instead task_info_s from stack
Signed-off-by: Xiang Xiao <xiaoxiang@xiaomi.com>
2021-11-01 20:53:07 -03:00
Xiang Xiao
8b67944c75 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>
2021-10-16 06:19:17 -03:00
Jiuzhu Dong
35b3898ce1 task_spawn: release g_spawn_parmsem and sched_unlock at the ending
Change-Id: Ifcb5b9921e82fc495c4457fdb5f0607f40b07fc0
Signed-off-by: Jiuzhu Dong <dongjiuzhu1@xiaomi.com>
2021-08-19 20:43:54 -07:00
Xiang Xiao
5025fbef8d Rename LIB_ to LIBC_ for all libc Kconfig
follow other libc component naming convention

Signed-off-by: Xiang Xiao <xiaoxiang@xiaomi.com>
2021-08-05 19:45:24 +02:00
SPRESENSE
5db4d2a83b sched/task: Fix wrong return value from nxspawn_open()
nxspawn_open() is expected to return "OK" when it success, but
it doesn't return it in case of executing dup2.
Because of this, the "Command as parameter" couldn't work with
Builtin Apps.
2021-07-20 09:26:01 -03:00
Jiuzhu Dong
9b1f554429 sched: Dynamically extend the pid hash table
Change-Id: I3a719bc575cd408cd530a39efa203f507b44fa9c
Signed-off-by: Jiuzhu Dong <dongjiuzhu1@xiaomi.com>
2021-07-11 19:42:30 -07:00
Huang Qi
65fa11634b sched: Refine tls_info_s and task_info_s
Signed-off-by: Huang Qi <huangqi3@xiaomi.com>
2021-07-02 10:33:34 -05:00
Xiang Xiao
13780dada3 sched: Add kthread_create_with_stack
let caller customize the stack buffer

Signed-off-by: Xiang Xiao <xiaoxiang@xiaomi.com>
Change-Id: I878818ec1ce5650165dbbe2d72a4618cb2e6a094
2021-06-27 12:35:13 -03:00
Juha Niskanen
ea1144630d sched/task/task_spawn.c: initialize variables explicitly
'pid' cannot really be used uninitialized, but Clang analyzer does not
see it.  Add initializer to silence it and also make debugging slightly
easier.

Explicitly set pid's address to NULL to fix this complaint:

  "Address of stack memory associated with local variable 'pid' is still
   referred to by the global variable 'g_spawn_parms' upon returning to
   the caller."

No functional change.

Signed-off-by: Juha Niskanen <juha.niskanen@haltian.com>
2021-06-22 03:44:02 -05:00
Xiang Xiao
bec68ad8ea sched/posix_spawn: Don't insert name at the begin of argv
since the standard require the caller pass the name explicitly
https://pubs.opengroup.org/onlinepubs/009695399/functions/posix_spawn.html:
The argument argv is an array of character pointers to null-terminated strings.
The last member of this array shall be a null pointer and is not counted in argc.
These strings constitute the argument list available to the new process image.
The value in argv[0] should point to a filename that is associated with the
process image being started by the posix_spawn() or posix_spawnp() function.

Signed-off-by: Xiang Xiao <xiaoxiang@xiaomi.com>
2021-06-15 07:32:33 -06:00
Xiang Xiao
90f71bd017 Revert "sched/posix_spawn: Don't insert name at the begin of argv"
This reverts commit 032086870d.
2021-06-15 07:32:33 -06:00
Xiang Xiao
032086870d sched/posix_spawn: Don't insert name at the begin of argv
since the standard require the caller pass the name explicitly
https://pubs.opengroup.org/onlinepubs/009695399/functions/posix_spawn.html:
The argument argv is an array of character pointers to null-terminated strings.
The last member of this array shall be a null pointer and is not counted in argc.
These strings constitute the argument list available to the new process image.
The value in argv[0] should point to a filename that is associated with the
process image being started by the posix_spawn() or posix_spawnp() function.

Signed-off-by: Xiang Xiao <xiaoxiang@xiaomi.com>
Change-Id: Id79ffcc501ae9552dc4e908418ff555f498be7f1
2021-06-14 07:11:53 -06:00
YAMAMOTO Takashi
b3e8535ad6 Revert "tls: Move pthread key destructor to libc"
This reverts commit cc514d7791.

* It introduced a regression.
  https://github.com/apache/incubator-nuttx/issues/3868

* It seems conceptually wrong to have per-process data in
  the main thread's stack.
2021-06-09 07:59:44 -05:00
Huang Qi
cc514d7791 tls: Move pthread key destructor to libc
Signed-off-by: Huang Qi <huangqi3@xiaomi.com>
2021-06-06 15:25:56 -06:00
Xiang Xiao
2e54df0f35 Don't include assert.h from public header file
Signed-off-by: Xiang Xiao <xiaoxiang@xiaomi.com>
2021-06-03 08:36:03 -07:00
Xiang Xiao
ae216cf9e9 sched/task: Simplify the syscall handling of task_spawn
It's better to save one argument by returning pid directly.
This change also follow the convention of task_create.
BTW, it is reasonable to adjust the function prototype a
little bit from both implementation and consistency since
task_spawn is NuttX specific API.

Signed-off-by: Xiang Xiao <xiaoxiang@xiaomi.com>
2021-06-01 12:45:10 +09:00
Huang Qi
a2941532bd pthread: Avoid recursive pthread_exit call
pthread_exit will be called recursive when pthread_cancel
or other cleanup operation with syscalls that support
cancellation, to avoid this by mark current tcb flag as
TCB_FLAG_CANCEL_DOING instead of TCB_FLAG_CANCEL_PENDING.

Signed-off-by: Huang Qi <huangqi3@xiaomi.com>
2021-05-21 22:46:52 -06:00
Huang Qi
f4a0b7aedd libc: Call pthread_exit in user-space by up_pthread_exit
Drop to user-space in kernel/protected build with up_pthread_exit,
now all pthread_cleanup functions executed in user mode.

* A new syscall SYS_pthread_exit added
* A new tcb flag TCB_FLAG_CANCEL_DOING added
* up_pthread_exit implemented for riscv/arm arch

Signed-off-by: Huang Qi <huangqi3@xiaomi.com>
2021-05-21 22:46:52 -06:00
Huang Qi
f1a92e9cbb libc: Move pthread_cleanup to user space
Signed-off-by: Huang Qi <huangqi3@xiaomi.com>
2021-05-21 22:46:52 -06:00
Xiang Xiao
8afb16ca55 Don't include errno.h from spawn.h
Signed-off-by: Xiang Xiao <xiaoxiang@xiaomi.com>
2021-05-19 10:11:15 +02:00
Xiang Xiao
2335b69120 arch: Allocate the space from the beginning in up_stack_frame
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>
2021-04-16 12:41:41 +09:00
Masayuki Ishikawa
a51a60d485 sched: task: Fix to initialize the task-specific data
Summary:
- I noticed that getopt() test in ostest wailed with
  esp32-devkitc:smp and spresense:smp
- Finally, I found that the task-specific data is not
  initialized.
- This commit fixes this issue

Impact:
- None

Testing:
- Tested with ostest esp32-devkitc:smp and spresense:smp

Signed-off-by: Masayuki Ishikawa <Masayuki.Ishikawa@jp.sony.com>
2021-04-03 08:32:46 -06:00
Gregory Nutt
2ccc0da0c7 Move getopt() variables into TLS
getopt() in the FLAT build environment is not thread safe.  This is because global variables that are process-specific in Unix are truly global in the FLAT build.  Moving the getopt() variables into TLS resolves this issue.

No side-effects are expected other than to getopt()

Tested with sim:nsh
2021-03-31 21:13:01 -05:00