Various fixes and cleanups
v2 cover Various fixes and cleanups
This is a rework/rebase of the patch series I sent last December, addressing the concerns and suggestions brought up during our discussion in January.
Find the patches also at git://git.kiszka.org/xenomai.git queues/assorted
Jan Kiszka (4):
POSIX: Fix SCHED_RR thread creation
Mark libs nodlopen on initial-exec TLS Add --enable-dlopen-skins configure switch
POSIX: Do not auto-shadow main with dlopen enabled
configure.in | 22 +++++++++++++++++++- ksrc/skins/posix/syscall.c | 4 +- src/skins/native/Makefile.am | 2 +- src/skins/posix/Makefile.am | 2 +- src/skins/posix/init.c | 43 ++++++++++++++++++++++++++-------------- src/skins/posix/thread.c | 11 +++++---- src/skins/psos+/Makefile.am | 2 +- src/skins/rtai/Makefile.am | 2 +- src/skins/rtdm/Makefile.am | 2 +- src/skins/uitron/Makefile.am | 2 +- src/skins/vrtx/Makefile.am | 2 +- src/skins/vxworks/Makefile.am | 2 +- 12 files changed, 64 insertions(+), 32 deletions(-)
--
Siemens AG, Corporate Technology, CT SE 2
Corporate Competence Center Embedded Linux
PATCH v2 1/4 POSIX: Fix SCHED_RR thread creation
SCHED_RR threads created via pthread_create do not receive the round-robin property at kernel side. Fix this by pushing the policy as an additional argument to __pse51_thread_create down to the kernel. Moreover, enforce SCHED_FIFO for the Linux part of those threads.
---
ksrc/skins/posix/syscall.c | 4 ++-- src/skins/posix/thread.c | 11 ++++++----- 2 files changed, 8 insertions(+), 7 deletions(-)
ksrc/skins/posix/syscall.c +2 -2
@@ -167,7 +167,7 @@ static int __pthread_create(struct pt_regs *regs)calling context. */ pthread_attr_init (&attr); - attr.policy = p->policy; + attr.policy =__xn_reg_arg2 (regs); attr.detachstate =PTHREAD_CREATE_DETACHED ; attr.schedparam_ex.sched_priority = p->rt_priority; attr.fp =1 ;
@@ -179,7 +179,7 @@ static int __pthread_create(struct pt_regs *regs)return -err;/* Conventionally, our error codes are negative. */ err =xnshadow_map (&k_tid->threadbase,NULL , - (unsigned long __user *)__xn_reg_arg2 (regs)); + (unsigned long __user *)__xn_reg_arg3 (regs));if (!err && !__pthread_hash (&hkey, k_tid)) err = -ENOMEM ;
src/skins/posix/thread.c +6 -5
@@ -192,6 +192,9 @@ static void *__pthread_trampoline(void *arg) param.sched_priority = iargs->prio; policy = iargs->policy; +if (policy ==SCHED_RR ) +/* Restrict round-robin scheduling to the Xenomai domain. */ + policy =SCHED_FIFO ; parent_prio = iargs->parent_prio; mode_buf =xeno_init_current_mode ();
@@ -203,8 +206,8 @@ static void *__pthread_trampoline(void *arg)/* Do _not_ inline the call to pthread_self() in the syscall macro: this trashes the syscall regs on some archs. */ - err =XENOMAI_SKINCALL2 (__pse51_muxid, __pse51_thread_create, tid, - mode_buf); + err =XENOMAI_SKINCALL3 (__pse51_muxid, __pse51_thread_create, tid, + iargs->policy, mode_buf); iargs->ret = -err;/* We must save anything we'll need to use from *iargs on our own
@@ -221,9 +224,7 @@ static void *__pthread_trampoline(void *arg)__real_sem_post (&iargs->sync);if (!err) { -/* Broken pthread libs ignore some of the thread attribute specs -passed to pthread_create(3), so we force the scheduling policy -once again here. */ +/* Adjust to final Linux policy and priority. */ __real_pthread_setschedparam (tid, policy, ¶m);/* If the thread running pthread_create runs with the same
PATCH v2 2/4 Mark libs nodlopen on initial-exec TLS
In reply to Jan KiszkaMark libs with nodlopen if initial-exec __thread variables are used because dlopen and this TLS model are in conflict.
---
configure.in | 3 +++ src/skins/native/Makefile.am | 2 +- src/skins/posix/Makefile.am | 2 +- src/skins/psos+/Makefile.am | 2 +- src/skins/rtai/Makefile.am | 2 +- src/skins/rtdm/Makefile.am | 2 +- src/skins/uitron/Makefile.am | 2 +- src/skins/vrtx/Makefile.am | 2 +- src/skins/vxworks/Makefile.am | 2 +- 9 files changed, 11 insertions(+), 8 deletions(-)
configure.in +3 -0
@@ -758,6 +758,7 @@ LIBS="$LIBS -lrt" AC_CHECK_FUNCS([shm_open shm_unlink]) LIBS="$save_LIBS" +XENO_DLOPEN_CONSTRAINT= AC_ARG_WITH([__thread], AC_HELP_STRING([--without-__thread], [do not use TLS features (allows for dlopen'ing Xenomai libs)]),
@@ -778,6 +779,7 @@ EOF rm -f conftest*]) if test "$libc_cv_gcc___thread" = yes; then AC_DEFINE(HAVE___THREAD,1,[config]) + XENO_DLOPEN_CONSTRAINT="-Wl,-z -Wl,nodlopen" fi fi
@@ -799,6 +801,7 @@ AC_SUBST(XENO_USER_CFLAGS) AC_SUBST(XENO_USER_LDFLAGS) AC_SUBST(XENO_USER_APP_CFLAGS) AC_SUBST(XENO_USER_APP_LDFLAGS) +AC_SUBST(XENO_DLOPEN_CONSTRAINT) AC_SUBST([CONFIG_STATUS_DEPENDENCIES], ['$(top_srcdir)/src/skins/posix/posix.wrappers']) AC_SUBST(XENO_POSIX_WRAPPERS)
src/skins/native/Makefile.am +1 -1
@@ -1,6 +1,6 @@ lib_LTLIBRARIES = libnative.la -libnative_la_LDFLAGS = -version-info 3:0:0 -lpthread +libnative_la_LDFLAGS = @XENO_DLOPEN_CONSTRAINT@ -version-info 3:0:0 -lpthread libnative_la_SOURCES = \ alarm.c \
src/skins/posix/Makefile.am +1 -1
@@ -4,7 +4,7 @@ lib_LTLIBRARIES = libpthread_rt.la CPPFLAGS+=-I$(top_srcdir)/ksrc/skins -libpthread_rt_la_LDFLAGS = -version-info 1:0:0 -lpthread +libpthread_rt_la_LDFLAGS = @XENO_DLOPEN_CONSTRAINT@ -version-info 1:0:0 -lpthread libpthread_rt_la_SOURCES = \ init.c \
src/skins/psos+/Makefile.am +1 -1
@@ -1,6 +1,6 @@ lib_LTLIBRARIES = libpsos.la -libpsos_la_LDFLAGS = -version-info 0:0:0 -lpthread +libpsos_la_LDFLAGS = @XENO_DLOPEN_CONSTRAINT@ -version-info 0:0:0 -lpthread libpsos_la_SOURCES = \ asr.c \
src/skins/rtai/Makefile.am +1 -1
@@ -1,6 +1,6 @@ lib_LTLIBRARIES = librtai.la -librtai_la_LDFLAGS = -version-info 0:0:0 +librtai_la_LDFLAGS = @XENO_DLOPEN_CONSTRAINT@ -version-info 0:0:0 librtai_la_SOURCES = init.c \ shm.c
src/skins/rtdm/Makefile.am +1 -1
@@ -1,6 +1,6 @@ lib_LTLIBRARIES = librtdm.la -librtdm_la_LDFLAGS = -version-info 1:0:0 -lpthread +librtdm_la_LDFLAGS = @XENO_DLOPEN_CONSTRAINT@ -version-info 1:0:0 -lpthread librtdm_la_SOURCES = \ core.c \
src/skins/uitron/Makefile.am +1 -1
@@ -1,6 +1,6 @@ lib_LTLIBRARIES = libuitron.la -libuitron_la_LDFLAGS = -version-info 0:0:0 -lpthread +libuitron_la_LDFLAGS = @XENO_DLOPEN_CONSTRAINT@ -version-info 0:0:0 -lpthread libuitron_la_SOURCES = \ flag.c \
src/skins/vrtx/Makefile.am +1 -1
@@ -1,6 +1,6 @@ lib_LTLIBRARIES = libvrtx.la -libvrtx_la_LDFLAGS = -version-info 0:0:0 -lpthread +libvrtx_la_LDFLAGS = @XENO_DLOPEN_CONSTRAINT@ -version-info 0:0:0 -lpthread libvrtx_la_SOURCES = \ event.c \
src/skins/vxworks/Makefile.am +1 -1
@@ -1,6 +1,6 @@ lib_LTLIBRARIES = libvxworks.la -libvxworks_la_LDFLAGS = -version-info 1:0:0 -lpthread +libvxworks_la_LDFLAGS = @XENO_DLOPEN_CONSTRAINT@ -version-info 1:0:0 -lpthread libvxworks_la_SOURCES = \ errnoLib.c \
PATCH v2 3/4 Add --enable-dlopen-skins configure switch
In reply to Jan KiszkaTo clarify the relation between disabling TLS support and allowing to dlopen Xenomai skin library, introduce an explicit configure switch. This way we will also have a control to disable auto-shadowing in the POSIX skin for dlopen scenarios.
---
configure.in | 19 +++++++++++++++++-- 1 files changed, 17 insertions(+), 2 deletions(-)
configure.in +17 -2
@@ -758,15 +758,30 @@ LIBS="$LIBS -lrt" AC_CHECK_FUNCS([shm_open shm_unlink]) LIBS="$save_LIBS" +AC_MSG_CHECKING(whether to enable dlopen support for skin libraries) +AC_ARG_ENABLE(dlopen-skins, + AC_HELP_STRING([--enable-dlopen-skins], [Disable TLS features and +automatic main thread mapping by the POSIX skin to allows dlopen'ing Xenomai +libs.]), + [case "$enableval" in + y | yes) CONFIG_XENO_LIBS_DLOPEN=y ;; + *) unset CONFIG_XENO_LIBS_DLOPEN ;; + esac]) +AC_MSG_RESULT(${CONFIG_XENO_LIBS_DLOPEN:-no}) +if test x$CONFIG_XENO_LIBS_DLOPEN = xy; then + AC_DEFINE(CONFIG_XENO_LIBS_DLOPEN,1,[config]) +fi + XENO_DLOPEN_CONSTRAINT= + AC_ARG_WITH([__thread], AC_HELP_STRING([--without-__thread], - [do not use TLS features (allows for dlopen'ing Xenomai libs)]), + [do not use TLS features]), [use__thread=$withval], [use__thread=yes]) dnl Check whether the compiler supports the __thread keyword. -if test "x$use__thread" != xno; then +if test "x$use__thread" != xno -a "x$CONFIG_XENO_LIBS_DLOPEN" != xy; then AC_CACHE_CHECK([for __thread], libc_cv_gcc___thread, [cat > conftest.c <<\EOF __thread int a __attribute__ ((tls_model ("initial-exec"))) = 42;
PATCH v2 4/4 POSIX: Do not auto-shadow main with dlopen enabled
In reply to Jan KiszkaDon't perform auto-shadowing in POSIX skin if we might be loaded via dlopen. Otherwise the wrong thread, the undefined dlopen caller, may be (re-)shadowed, assigning wrong scheduling settings.
---
src/skins/posix/init.c | 43 ++++++++++++++++++++++++++++--------------- 1 files changed, 28 insertions(+), 15 deletions(-)
src/skins/posix/init.c +28 -15
@@ -47,7 +47,6 @@ void pse51_clock_init(int);static __attribute__ ((constructor))void __init_posix_interface (void ) { -struct sched_param parm;int muxid, err; muxid =
@@ -67,28 +66,42 @@ void __init_posix_interface(void) __rtdm_fdcount); } -/* Shadow the main thread. mlock the whole memory for the time of the -syscall, in order to avoid the SIGXCPU signal. */ +#ifdef CONFIG_XENO_LIBS_DLOPEN +/* Don't use auto-shadowing if we are likely invoked from dlopen, but +take care of auto-mlockall. */ +#ifdef CONFIG_XENO_POSIX_AUTO_MLOCKALL if (mlockall (MCL_CURRENT |MCL_FUTURE )) {perror ("Xenomai Posix skin init: mlockall" );exit (EXIT_FAILURE ); } +#endif /* CONFIG_XENO_POSIX_AUTO_MLOCKALL */ +#else /* !CONFIG_XENO_LIBS_DLOPEN */ + { +struct sched_param parm = { .sched_priority =0 }; + +/* Shadow the main thread. mlock the whole memory for the time +of the syscall, in order to avoid the SIGXCPU signal. */ +if (mlockall (MCL_CURRENT |MCL_FUTURE )) { +perror ("Xenomai Posix skin init: mlockall" ); +exit (EXIT_FAILURE ); + } - parm.sched_priority =0 ; - err =__wrap_pthread_setschedparam (pthread_self (),SCHED_OTHER , - &parm); -if (err) { -fprintf (stderr,"Xenomai Posix skin init: " -"pthread_setschedparam: %s\n" ,strerror (err)); -exit (EXIT_FAILURE ); - } + err =__wrap_pthread_setschedparam (pthread_self (),SCHED_OTHER , + &parm); +if (err) { +fprintf (stderr,"Xenomai Posix skin init: " +"pthread_setschedparam: %s\n" ,strerror (err)); +exit (EXIT_FAILURE ); + }#ifndef CONFIG_XENO_POSIX_AUTO_MLOCKALL -if (munlockall ()) { -perror ("Xenomai Posix skin init: munlockall" ); -exit (EXIT_FAILURE ); - } +if (munlockall ()) { +perror ("Xenomai Posix skin init: munlockall" ); +exit (EXIT_FAILURE ); + }#endif /* !CONFIG_XENO_POSIX_AUTO_MLOCKALL */ + } +#endif /* !CONFIG_XENO_LIBS_DLOPEN */ if (!fork_handler_registered) { err =pthread_atfork (NULL ,NULL , &__init_posix_interface);
Re: [Xenomai-core] [PATCH v2 1/4] POSIX: Fix SCHED_RR thread creation
In reply to Jan Kiszkasrc/skins/posix/thread.c static void *__pthread_trampoline(void *arg) · line 197
Quoted 7 lines · Jan Kiszka
src/skins/posix/thread.c +3
-0@@ -192,6 +192,9 @@ static void *__pthread_trampoline(void *arg) param.sched_priority = iargs->prio; policy = iargs->policy; +if (policy ==SCHED_RR ) +/* Restrict round-robin scheduling to the Xenomai domain. */ + policy =SCHED_FIFO ;
Should not there be the same thing in __wrap_pthread_setschedparam ?
Signature · 1 line
Gilles.
Re: [Xenomai-core] [PATCH v2 1/4] POSIX: Fix SCHED_RR thread creation
Quoted 10 lines · Gilles Chanteperdrix
Review static void *__pthread_trampoline(void *arg) · line 197
Quoted 7 lines · Jan Kiszka
(quoted diff) +3
-0@@ -192,6 +192,9 @@ static void *__pthread_trampoline(void *arg) param.sched_priority = iargs->prio; policy = iargs->policy; + if (policy == SCHED_RR) + /* Restrict round-robin scheduling to the Xenomai domain. */ + policy = SCHED_FIFO;Should not there be the same thing in
__wrap_pthread_setschedparam?
Yes, and setschedparam_ex, here we go:
--------->
SCHED_RR threads created via pthread_create do not receive the round-robin property at kernel side. Fix this by pushing the policy as an additional argument to __pse51_thread_create down to the kernel. Moreover, enforce SCHED_FIFO for the Linux part of those threads.
---
0 files changed, 0 insertions(+), 0 deletions(-)
ksrc/skins/posix/syscall.c +2 -2
@@ -167,7 +167,7 @@ static int __pthread_create(struct pt_regs *regs)calling context. */ pthread_attr_init (&attr); - attr.policy = p->policy; + attr.policy =__xn_reg_arg2 (regs); attr.detachstate =PTHREAD_CREATE_DETACHED ; attr.schedparam_ex.sched_priority = p->rt_priority; attr.fp =1 ;
@@ -179,7 +179,7 @@ static int __pthread_create(struct pt_regs *regs)return -err;/* Conventionally, our error codes are negative. */ err =xnshadow_map (&k_tid->threadbase,NULL , - (unsigned long __user *)__xn_reg_arg2 (regs)); + (unsigned long __user *)__xn_reg_arg3 (regs));if (!err && !__pthread_hash (&hkey, k_tid)) err = -ENOMEM ;
src/skins/posix/thread.c +16 -6
@@ -61,8 +61,13 @@ int __wrap_pthread_setschedparam(pthread_t thread, __pse51_thread_setschedparam, thread, policy, param, mode_buf, &promoted); -if (err ==EPERM ) +if (err ==EPERM ) { +if (policy ==SCHED_RR ) +/* Restrict round-robin scheduling to the Xenomai +domain. */ + policy =SCHED_FIFO ;return __real_pthread_setschedparam (thread, policy, param); + }if (!err && promoted) {sigshadow_install_once ();
@@ -104,6 +109,10 @@ int pthread_setschedparam_ex(pthread_t thread,if (err ==EPERM ) { short_param.sched_priority = param->sched_priority; +if (policy ==SCHED_RR ) +/* Restrict round-robin scheduling to the Xenomai +domain. */ + policy =SCHED_FIFO ;return __real_pthread_setschedparam (thread, policy, &short_param); }
@@ -192,6 +201,9 @@ static void *__pthread_trampoline(void *arg) param.sched_priority = iargs->prio; policy = iargs->policy; +if (policy ==SCHED_RR ) +/* Restrict round-robin scheduling to the Xenomai domain. */ + policy =SCHED_FIFO ; parent_prio = iargs->parent_prio; mode_buf =xeno_init_current_mode ();
@@ -203,8 +215,8 @@ static void *__pthread_trampoline(void *arg)/* Do _not_ inline the call to pthread_self() in the syscall macro: this trashes the syscall regs on some archs. */ - err =XENOMAI_SKINCALL2 (__pse51_muxid, __pse51_thread_create, tid, - mode_buf); + err =XENOMAI_SKINCALL3 (__pse51_muxid, __pse51_thread_create, tid, + iargs->policy, mode_buf); iargs->ret = -err;/* We must save anything we'll need to use from *iargs on our own
@@ -221,9 +233,7 @@ static void *__pthread_trampoline(void *arg)__real_sem_post (&iargs->sync);if (!err) { -/* Broken pthread libs ignore some of the thread attribute specs -passed to pthread_create(3), so we force the scheduling policy -once again here. */ +/* Adjust to final Linux policy and priority. */ __real_pthread_setschedparam (tid, policy, ¶m);/* If the thread running pthread_create runs with the same
Re: [Xenomai-core] [PATCH v2 1/4] POSIX: Fix SCHED_RR thread creation
Quoted 4 lines · Jan Kiszka
Quoted 3 lines · Gilles Chanteperdrix
Quoted 2 lines · Jan Kiszka
(quoted diff) +0
-0@@ -192,6 +192,9 @@ static void *__pthread_trampoline(void *arg)
Quoted 8 lines
Quoted 6 lines
src/skins/posix/thread.c static void *__pthread_trampoline(void *arg) · line 206
Quoted 5 lines
src/skins/posix/thread.c +3
-0param.sched_priority = iargs->prio; policy = iargs->policy; +if (policy ==SCHED_RR ) +/* Restrict round-robin scheduling to the Xenomai domain. */ + policy =SCHED_FIFO ;Should not there be the same thing in
__wrap_pthread_setschedparam?Yes, and
setschedparam_ex, here we go:
Actually, I am wondering if we can not get rid of these calls to __real_pthread_setschedparam, now that propagating kernel-space priority to user-space is done by a signal ?
Signature · 1 line
Gilles.
Re: [Xenomai-core] [PATCH v2 1/4] POSIX: Fix SCHED_RR thread creation
Quoted 11 lines · Gilles Chanteperdrix
Quoted 10 lines · Jan Kiszka
Quoted 9 lines · Gilles Chanteperdrix
src/skins/posix/thread.c static void *__pthread_trampoline(void *arg) · line 197
Quoted 7 lines · Jan Kiszka
src/skins/posix/thread.c +3
-0@@ -192,6 +192,9 @@ static void *__pthread_trampoline(void *arg) param.sched_priority = iargs->prio; policy = iargs->policy; +if (policy ==SCHED_RR ) +/* Restrict round-robin scheduling to the Xenomai domain. */ + policy =SCHED_FIFO ;Should not there be the same thing in
__wrap_pthread_setschedparam?
Quoted 5 lines
Quoted 1 line
Yes, and
setschedparam_ex, here we go:Actually, I am wondering if we can not get rid of these calls to
__real_pthread_setschedparam, now that propagating kernel-space priority to user-space is done by a signal ?
Not with the existing code, as that only forwards prio changes, but no policy changes.
Jan
Signature · 2 lines
Siemens AG, Corporate Technology, CT SE 2 Corporate Competence Center Embedded Linux
Re: [Xenomai-core] [PATCH v2 1/4] POSIX: Fix SCHED_RR thread creation
Quoted 12 lines · Jan Kiszka
Quoted 11 lines · Gilles Chanteperdrix
Quoted 10 lines · Jan Kiszka
Quoted 9 lines · Gilles Chanteperdrix
src/skins/posix/thread.c static void *__pthread_trampoline(void *arg) · line 197
Quoted 7 lines · Jan Kiszka
src/skins/posix/thread.c +3
-0@@ -192,6 +192,9 @@ static void *__pthread_trampoline(void *arg) param.sched_priority = iargs->prio; policy = iargs->policy; +if (policy ==SCHED_RR ) +/* Restrict round-robin scheduling to the Xenomai domain. */ + policy =SCHED_FIFO ;Should not there be the same thing in
__wrap_pthread_setschedparam?
Quoted 7 lines
Quoted 4 lines
Quoted 1 line
Yes, and
setschedparam_ex, here we go:Actually, I am wondering if we can not get rid of these calls to
__real_pthread_setschedparam, now that propagating kernel-space priority to user-space is done by a signal ?Not with the existing code, as that only forwards prio changes, but no policy changes.
If we map SCHED_RR to SCHED_FIFO, are there any policy changes?
Signature · 1 line
Gilles.
Re: [Xenomai-core] [PATCH v2 1/4] POSIX: Fix SCHED_RR thread creation
Quoted 13 lines · Gilles Chanteperdrix
Quoted 12 lines · Jan Kiszka
Quoted 11 lines · Gilles Chanteperdrix
Quoted 10 lines · Jan Kiszka
Quoted 9 lines · Gilles Chanteperdrix
src/skins/posix/thread.c static void *__pthread_trampoline(void *arg) · line 197
Quoted 7 lines · Jan Kiszka
src/skins/posix/thread.c +3
-0@@ -192,6 +192,9 @@ static void *__pthread_trampoline(void *arg) param.sched_priority = iargs->prio; policy = iargs->policy; +if (policy ==SCHED_RR ) +/* Restrict round-robin scheduling to the Xenomai domain. */ + policy =SCHED_FIFO ;Should not there be the same thing in
__wrap_pthread_setschedparam?
Quoted 8 lines
Quoted 6 lines
Quoted 4 lines
Quoted 1 line
Yes, and
setschedparam_ex, here we go:Actually, I am wondering if we can not get rid of these calls to
__real_pthread_setschedparam, now that propagating kernel-space priority to user-space is done by a signal ?Not with the existing code, as that only forwards prio changes, but no policy changes.
If we map
SCHED_RRtoSCHED_FIFO, are there any policy changes?
Yes, the initial one. Keep in mind that not all pthread implementations may respect the pthread_attr or have problems assigning SCHED_FIFO to threads of non-root users.
Jan
Signature · 2 lines
Siemens AG, Corporate Technology, CT SE 2 Corporate Competence Center Embedded Linux
Re: [Xenomai-core] [PATCH v2 1/4] POSIX: Fix SCHED_RR thread creation
Quoted 18 lines · Jan Kiszka
Quoted 17 lines · Gilles Chanteperdrix
Quoted 16 lines · Jan Kiszka
Quoted 15 lines · Gilles Chanteperdrix
Quoted 11 lines · Jan Kiszka
Quoted 9 lines · Gilles Chanteperdrix
src/skins/posix/thread.c static void *__pthread_trampoline(void *arg) · line 197
Quoted 7 lines · Jan Kiszka
src/skins/posix/thread.c +3
-0@@ -192,6 +192,9 @@ static void *__pthread_trampoline(void *arg) param.sched_priority = iargs->prio; policy = iargs->policy; +if (policy ==SCHED_RR ) +/* Restrict round-robin scheduling to the Xenomai domain. */ + policy =SCHED_FIFO ;Should not there be the same thing in
__wrap_pthread_setschedparam?Yes, and
setschedparam_ex, here we go:Actually, I am wondering if we can not get rid of these calls to
__real_pthread_setschedparam, now that propagating kernel-space priority to user-space is done by a signal ?
Quoted 7 lines
Quoted 3 lines
Quoted 2 lines
Not with the existing code, as that only forwards prio changes, but no policy changes.
If we map
SCHED_RRtoSCHED_FIFO, are there any policy changes?Yes, the initial one. Keep in mind that not all pthread implementations may respect the
pthread_attror have problems assigningSCHED_FIFOto threads of non-root users.
Ok. What about calling xnshadow_renice in xnshadow_map ?
Signature · 1 line
Gilles.
Re: [Xenomai-core] [PATCH v2 1/4] POSIX: Fix SCHED_RR thread creation
Quoted 19 lines · Gilles Chanteperdrix
Quoted 18 lines · Jan Kiszka
Quoted 17 lines · Gilles Chanteperdrix
Quoted 16 lines · Jan Kiszka
Quoted 15 lines · Gilles Chanteperdrix
Quoted 11 lines · Jan Kiszka
Quoted 9 lines · Gilles Chanteperdrix
src/skins/posix/thread.c static void *__pthread_trampoline(void *arg) · line 197
Quoted 7 lines · Jan Kiszka
src/skins/posix/thread.c +3
-0@@ -192,6 +192,9 @@ static void *__pthread_trampoline(void *arg) param.sched_priority = iargs->prio; policy = iargs->policy; +if (policy ==SCHED_RR ) +/* Restrict round-robin scheduling to the Xenomai domain. */ + policy =SCHED_FIFO ;Should not there be the same thing in
__wrap_pthread_setschedparam?Yes, and
setschedparam_ex, here we go:Actually, I am wondering if we can not get rid of these calls to
__real_pthread_setschedparam, now that propagating kernel-space priority to user-space is done by a signal ?
Quoted 8 lines
Quoted 6 lines
Quoted 3 lines
Quoted 2 lines
Not with the existing code, as that only forwards prio changes, but no policy changes.
If we map
SCHED_RRtoSCHED_FIFO, are there any policy changes?Yes, the initial one. Keep in mind that not all pthread implementations may respect the
pthread_attror have problems assigningSCHED_FIFOto threads of non-root users.Ok. What about calling
xnshadow_reniceinxnshadow_map?
Sorry, I can't follow. xnshadow_renice does not help user space to find the right policy, or what do you mean?
Jan
Signature · 2 lines
Siemens AG, Corporate Technology, CT SE 2 Corporate Competence Center Embedded Linux
Re: [Xenomai-core] [PATCH v2 1/4] POSIX: Fix SCHED_RR thread creation
Quoted 22 lines · Jan Kiszka
Quoted 21 lines · Gilles Chanteperdrix
Quoted 20 lines · Jan Kiszka
Quoted 19 lines · Gilles Chanteperdrix
Quoted 18 lines · Jan Kiszka
Quoted 11 lines · Gilles Chanteperdrix
Quoted 10 lines · Jan Kiszka
Quoted 9 lines · Gilles Chanteperdrix
Review static void *__pthread_trampoline(void *arg) · line 197
Quoted 7 lines · Jan Kiszka
(quoted diff) +3
-0@@ -192,6 +192,9 @@ static void *__pthread_trampoline(void *arg) param.sched_priority = iargs->prio; policy = iargs->policy; + if (policy == SCHED_RR) + /* Restrict round-robin scheduling to the Xenomai domain. */ + policy = SCHED_FIFO;Should not there be the same thing in
__wrap_pthread_setschedparam?Quoted 4 lines
Quoted 1 line
Yes, and
setschedparam_ex, here we go:Actually, I am wondering if we can not get rid of these calls to
__real_pthread_setschedparam, now that propagating kernel-space priority to user-space is done by a signal ?Not with the existing code, as that only forwards prio changes, but no policy changes.
Quoted 8 lines
Quoted 5 lines
Quoted 4 lines
Quoted 1 line
If we map
SCHED_RRtoSCHED_FIFO, are there any policy changes?Yes, the initial one. Keep in mind that not all pthread implementations may respect the
pthread_attror have problems assigningSCHED_FIFOto threads of non-root users.Ok. What about calling
xnshadow_reniceinxnshadow_map?Sorry, I can't follow.
xnshadow_renicedoes not help user space to find the right policy, or what do you mean?
xnshadow_renice sends a signal to the user-space thread which in turn calls __real_pthread_setschedparam.
The advantage of this solution is that it avoids calling __real_pthread_setschedparam in many places in the user-space skins.
Signature · 1 line
Gilles.
Re: [Xenomai-core] [PATCH v2 1/4] POSIX: Fix SCHED_RR thread creation
Quoted 27 lines · Gilles Chanteperdrix
Quoted 26 lines · Jan Kiszka
Quoted 25 lines · Gilles Chanteperdrix
Quoted 24 lines · Jan Kiszka
Quoted 13 lines · Gilles Chanteperdrix
Quoted 12 lines · Jan Kiszka
Quoted 11 lines · Gilles Chanteperdrix
Quoted 10 lines · Jan Kiszka
Quoted 9 lines · Gilles Chanteperdrix
Review static void *__pthread_trampoline(void *arg) · line 197
Quoted 7 lines · Jan Kiszka
(quoted diff) +3
-0@@ -192,6 +192,9 @@ static void *__pthread_trampoline(void *arg) param.sched_priority = iargs->prio; policy = iargs->policy; + if (policy == SCHED_RR) + /* Restrict round-robin scheduling to the Xenomai domain. */ + policy = SCHED_FIFO;Should not there be the same thing in
__wrap_pthread_setschedparam?Quoted 7 lines
Quoted 6 lines
Quoted 4 lines
Quoted 1 line
Yes, and
setschedparam_ex, here we go:Actually, I am wondering if we can not get rid of these calls to
__real_pthread_setschedparam, now that propagating kernel-space priority to user-space is done by a signal ?Not with the existing code, as that only forwards prio changes, but no policy changes.
If we map
SCHED_RRtoSCHED_FIFO, are there any policy changes?Yes, the initial one. Keep in mind that not all pthread implementations may respect the
pthread_attror have problems assigningSCHED_FIFOto threads of non-root users.
Quoted 6 lines
Quoted 3 lines
Quoted 1 line
Ok. What about calling
xnshadow_reniceinxnshadow_map?Sorry, I can't follow.
xnshadow_renicedoes not help user space to find the right policy, or what do you mean?
xnshadow_renicesends a signal to the user-space thread which in turn calls__real_pthread_setschedparam.
Ah, and user space derives the policy from the prio. That's what I missed.
Quoted 3 lines
The advantage of this solution is that it avoids calling
__real_pthread_setschedparamin many places in the user-space skins.
We would save the call in the trampolines at least (my other changes in the second revision were nonsense BTW - those paths belong to the non-Xenomai case). We would just have to pass the priority down on pthread_create as well.
Will look into this and come up with a new version for this patch. But that need not delay merging the others.
Jan
Signature · 2 lines
Siemens AG, Corporate Technology, CT SE 2 Corporate Competence Center Embedded Linux
Re: [Xenomai-core] [PATCH v2 1/4] POSIX: Fix SCHED_RR thread creation
Quoted 39 lines · Jan Kiszka
Quoted 24 lines · Gilles Chanteperdrix
Quoted 23 lines · Jan Kiszka
Quoted 22 lines · Gilles Chanteperdrix
Quoted 21 lines · Jan Kiszka
Quoted 20 lines · Gilles Chanteperdrix
Quoted 12 lines · Jan Kiszka
Quoted 11 lines · Gilles Chanteperdrix
Quoted 10 lines · Jan Kiszka
Quoted 9 lines · Gilles Chanteperdrix
Review static void *__pthread_trampoline(void *arg) · line 197
Quoted 7 lines · Jan Kiszka
(quoted diff) +3
-0@@ -192,6 +192,9 @@ static void *__pthread_trampoline(void *arg) param.sched_priority = iargs->prio; policy = iargs->policy; + if (policy == SCHED_RR) + /* Restrict round-robin scheduling to the Xenomai domain. */ + policy = SCHED_FIFO;Should not there be the same thing in
__wrap_pthread_setschedparam?Quoted 6 lines
Quoted 4 lines
Quoted 1 line
Yes, and
setschedparam_ex, here we go:Actually, I am wondering if we can not get rid of these calls to
__real_pthread_setschedparam, now that propagating kernel-space priority to user-space is done by a signal ?Not with the existing code, as that only forwards prio changes, but no policy changes.
If we map
SCHED_RRtoSCHED_FIFO, are there any policy changes?Quoted 8 lines
Quoted 6 lines
Quoted 4 lines
Quoted 3 lines
Yes, the initial one. Keep in mind that not all pthread implementations may respect the
pthread_attror have problems assigningSCHED_FIFOto threads of non-root users.Ok. What about calling
xnshadow_reniceinxnshadow_map?Sorry, I can't follow.
xnshadow_renicedoes not help user space to find the right policy, or what do you mean?
xnshadow_renicesends a signal to the user-space thread which in turn calls__real_pthread_setschedparam.Ah, and user space derives the policy from the prio. That's what I missed.
Quoted 2 lines
The advantage of this solution is that it avoids calling
__real_pthread_setschedparamin many places in the user-space skins.
Quoted 7 lines
We would save the call in the trampolines at least (my other changes in the second revision were nonsense BTW - those paths belong to the non-Xenomai case). We would just have to pass the priority down on
pthread_createas well.Will look into this and come up with a new version for this patch. But that need not delay merging the others.
I played with it, but raising a signal from xnshadow_map is no good
idea: If we inject the signal before the migration, xnshadow_map fails due to the pending signal, we exit to user space, run the signal, restart the call, raise the signal, and so on. If we raise after migration, we immediately force the new thread into secondary mode again, definitely causing problems when the mode switch warning is active, but also risking other regressions for xnshadow_map users that expect to find themselves in primary mode on return.
In short: I think it is better to stick with my original version of this patch.
Jan
Re: [Xenomai-core] [PATCH v2 1/4] POSIX: Fix SCHED_RR thread creation
Quoted 9 lines · Jan Kiszka
I played with it, but raising a signal from
xnshadow_mapis no good
idea: If we inject the signal before the migration,xnshadow_mapfails due to the pending signal, we exit to user space, run the signal, restart the call, raise the signal, and so on. If we raise after migration, we immediately force the new thread into secondary mode again, definitely causing problems when the mode switch warning is active, but also risking other regressions forxnshadow_mapusers that expect to find themselves in primary mode on return.
Meditating over it again, I realized that there is actually no need to raise the renice signal if we migrate in xnshadow_map. This case will be handled automatically on next relax. That pattern seems to work and allows indeed a few cleanups (just uitron still needs pthread_setschedparam). This is just a first draft, needs more testing:
ksrc/nucleus/shadow.c +6 -0
@@ -1250,6 +1250,12 @@ int xnshadow_map(xnthread_t *thread, xncompletion_t __user *u_completion, thread->u_mode = u_mode;if (u_completion) { +/* +* Send the renice signal if we are not migrating so that user +* space will immediately align Linux sched policy and prio. +*/ +xnshadow_renice (thread); +/* * We still have the XNDORMANT bit set, so we can't * link to the RPI queue which only links _runnable_
ksrc/skins/posix/syscall.c +3 -3
@@ -167,9 +167,9 @@ static int __pthread_create(struct pt_regs *regs)calling context. */ pthread_attr_init (&attr); - attr.policy = p->policy; + attr.policy =__xn_reg_arg2 (regs); attr.detachstate =PTHREAD_CREATE_DETACHED ; - attr.schedparam_ex.sched_priority = p->rt_priority; + attr.schedparam_ex.sched_priority =__xn_reg_arg3 (regs); attr.fp =1 ; attr.name = p->comm;
@@ -179,7 +179,7 @@ static int __pthread_create(struct pt_regs *regs)return -err;/* Conventionally, our error codes are negative. */ err =xnshadow_map (&k_tid->threadbase,NULL , - (unsigned long __user *)__xn_reg_arg2 (regs)); + (unsigned long __user *)__xn_reg_arg4 (regs));if (!err && !__pthread_hash (&hkey, k_tid)) err = -ENOMEM ;
src/skins/native/task.c +0 -19
@@ -55,21 +55,10 @@ static void *rt_task_trampoline(void *cookie) {struct rt_task_iargs *iargs = (struct rt_task_iargs *)cookie;void (*entry) (void *cookie); -struct sched_param param;struct rt_arg_bulk bulk;RT_TASK *task;long err; -if (iargs->prio >0 ) { -/* -* Re-apply sched params here as some libpthread -* implementations fail doing this via pthread_create. -*/ -memset (¶m,0 ,sizeof (param)); - param.sched_priority = iargs->prio; -__real_pthread_setschedparam (pthread_self (),SCHED_FIFO , ¶m); - } -/* rt_task_delete requires asynchronous cancellation */ pthread_setcanceltype (PTHREAD_CANCEL_ASYNCHRONOUS ,NULL );
@@ -178,7 +167,6 @@ int rt_task_start(RT_TASK *task, void (*entry) (void *cookie), void *cookie)int rt_task_shadow (RT_TASK *task,const char *name,int prio,int mode) { -struct sched_param param;struct rt_arg_bulk bulk;RT_TASK task_desc;int err;
@@ -190,13 +178,6 @@ int rt_task_shadow(RT_TASK *task, const char *name, int prio, int mode)pthread_setcanceltype (PTHREAD_CANCEL_ASYNCHRONOUS ,NULL );sigshadow_install_once (); -if (prio >0 ) { -/* Make sure the POSIX library caches the right priority. */ -memset (¶m,0 ,sizeof (param)); - param.sched_priority = prio; -__real_pthread_setschedparam (pthread_self (),SCHED_FIFO , ¶m); - } - bulk.a1 = (u_long )task; bulk.a2 = (u_long )name; bulk.a3 = (u_long )prio;
src/skins/native/wrappers.c +0 -7
@@ -33,13 +33,6 @@*/ __attribute__ ((weak)) -int __real_pthread_setschedparam (pthread_t thread, -int policy,const struct sched_param *param) -{ -return pthread_setschedparam (thread, policy, param); -} - -__attribute__ ((weak))int __real_pthread_create (pthread_t *tid,const pthread_attr_t * attr,void *(*start) (void *),void *arg)
src/skins/native/wrappers.h +0 -3
@@ -8,9 +8,6 @@ int __real_pthread_create(pthread_t *tid,const pthread_attr_t * attr, void *(*start) (void *), void *arg); -int __real_pthread_setschedparam (pthread_t thread, -int policy,const struct sched_param *param); -int __real_pthread_kill (pthread_t tid,int sig);int __real_open (const char *path,int oflag, ...);
src/skins/posix/thread.c +2 -7
@@ -203,8 +203,8 @@ static void *__pthread_trampoline(void *arg)/* Do _not_ inline the call to pthread_self() in the syscall macro: this trashes the syscall regs on some archs. */ - err =XENOMAI_SKINCALL2 (__pse51_muxid, __pse51_thread_create, tid, - mode_buf); + err =XENOMAI_SKINCALL4 (__pse51_muxid, __pse51_thread_create, tid, + iargs->policy, iargs->prio, mode_buf); iargs->ret = -err;/* We must save anything we'll need to use from *iargs on our own
@@ -221,11 +221,6 @@ static void *__pthread_trampoline(void *arg)__real_sem_post (&iargs->sync);if (!err) { -/* Broken pthread libs ignore some of the thread attribute specs -passed to pthread_create(3), so we force the scheduling policy -once again here. */ -__real_pthread_setschedparam (tid, policy, ¶m); -/* If the thread running pthread_create runs with the same priority as us, we should leave it running, as if there never was a synchronization with a semaphore. */
src/skins/psos+/task.c +0 -5
@@ -63,13 +63,8 @@ static void *psos_task_trampoline(void *cookie)void (*entry)(u_long ,u_long ,u_long ,u_long );u_long dummy_args[4 ] = {0 ,0 ,0 ,0 }, *targs;struct psos_arg_bulk bulk; -struct sched_param param; -int policy;long err; - policy =psos_task_set_posix_priority (iargs->prio, ¶m); -pthread_setschedparam (pthread_self (), policy, ¶m); -pthread_setcanceltype (PTHREAD_CANCEL_ASYNCHRONOUS ,NULL );sigshadow_install_once ();
src/skins/vrtx/task.c +0 -9
@@ -73,8 +73,6 @@ static void *vrtx_task_trampoline(void *cookie)struct vrtx_task_iargs *iargs = (struct vrtx_task_iargs *)cookie, _iargs;struct vrtx_arg_bulk bulk; -struct sched_param param; -int policy;long err;#ifndef HAVE___THREAD TCB *tcb;
@@ -83,13 +81,6 @@ static void *vrtx_task_trampoline(void *cookie)/* Backup the arg struct, it might vanish after completion. */ memcpy (&_iargs, iargs,sizeof (_iargs)); -/* -* Apply sched params here as some libpthread implementations -* fail doing this properly via pthread_create. -*/ - policy =vrtx_task_set_posix_priority (iargs->prio, ¶m); -pthread_setschedparam (pthread_self (), policy, ¶m); -/* vrtx_task_delete requires asynchronous cancellation */ pthread_setcanceltype (PTHREAD_CANCEL_ASYNCHRONOUS ,NULL );
src/skins/vxworks/taskLib.c +0 -9
@@ -86,21 +86,12 @@ static void *wind_task_trampoline(void *cookie)struct wind_task_iargs *iargs = (struct wind_task_iargs *)cookie, _iargs;struct wind_arg_bulk bulk; -struct sched_param param;WIND_TCB *pTcb; -int policy;long err;/* Backup the arg struct, it might vanish after completion. */ memcpy (&_iargs, iargs,sizeof (_iargs)); -/* -* Apply sched params here as some libpthread implementations -* fail doing this properly via pthread_create. -*/ - policy =wind_task_set_posix_priority (iargs->prio, ¶m); -__real_pthread_setschedparam (pthread_self (), policy, ¶m); -/* wind_task_delete requires asynchronous cancellation */ pthread_setcanceltype (PTHREAD_CANCEL_ASYNCHRONOUS ,NULL );sigshadow_install_once ();
src/skins/vxworks/wrappers.c +0 -7
@@ -33,13 +33,6 @@*/ __attribute__ ((weak)) -int __real_pthread_setschedparam (pthread_t thread, -int policy,const struct sched_param *param) -{ -return pthread_setschedparam (thread, policy, param); -} - -__attribute__ ((weak))int __real_pthread_create (pthread_t *tid,const pthread_attr_t * attr,void *(*start) (void *),void *arg)
src/skins/vxworks/wrappers.h +0 -3
@@ -8,9 +8,6 @@ int __real_pthread_create(pthread_t *tid,const pthread_attr_t * attr, void *(*start) (void *), void *arg); -int __real_pthread_setschedparam (pthread_t thread, -int policy,const struct sched_param *param); -int __real_pthread_kill (pthread_t tid,int sig);int __real_open (const char *path,int oflag, ...);
Quoted 5 lines · Jan Kiszka
This is a rework/rebase of the patch series I sent last December, addressing the concerns and suggestions brought up during our discussion in January.
Quoted 7 lines
Find the patches also at
git://git.kiszka.org/xenomai.gitqueues/assortedJan Kiszka (4):
POSIX: FixSCHED_RRthread creation
Mark libs nodlopen on initial-exec TLS Add --enable-dlopen-skins configure switch
POSIX: Do not auto-shadow main with dlopen enabled
What is the schedule for these patches? 2..4 shouldn't be controversial anymore (find them rebased as patch 1..3 under queues/assorted). And for the first one I already sent a rework (now patch 4 in that queue) that still looks fine to me. But I can also repost some/all of them if preferred.
Jan
Signature · 2 lines
Siemens AG, Corporate Technology, CT SE 2 Corporate Competence Center Embedded Linux
Quoted 10 lines · Jan Kiszka
Quoted 9 lines · Jan Kiszka
This is a rework/rebase of the patch series I sent last December, addressing the concerns and suggestions brought up during our discussion in January.
Find the patches also at
git://git.kiszka.org/xenomai.gitqueues/assortedJan Kiszka (4):
POSIX: FixSCHED_RRthread creation
This one needs more review and testing. It will be merged in 2.5-rc1 to get more exposure first, before any backport to 2.4 in order to reduce the impact of any regression we might find.
Quoted 4 lines
Quoted 3 lines
Mark libs nodlopen on initial-exec TLS Add --enable-dlopen-skins configure switch
POSIX: Do not auto-shadow main with dlopen enabled
Will make their way to 2.4.7 and 2.5-rc1 asap.
Quoted 7 lines
What is the schedule for these patches? 2..4 shouldn't be controversial anymore (find them rebased as patch 1..3 under queues/assorted). And for the first one I already sent a rework (now patch 4 in that queue) that still looks fine to me. But I can also repost some/all of them if preferred.
Jan
Signature · 1 line
Philippe.