freebsd-dev/sys/kern/kern_thr.c

614 lines
14 KiB
C
Raw Normal View History

/*-
* Copyright (c) 2003, Jeffrey Roberson <jeff@freebsd.org>
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice unmodified, this list of conditions, and the following
* disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
* IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
* OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
* IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
* NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
* THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
2003-06-11 00:56:59 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include "opt_compat.h"
#include "opt_posix.h"
#include <sys/param.h>
#include <sys/kernel.h>
#include <sys/lock.h>
#include <sys/mutex.h>
#include <sys/priv.h>
#include <sys/proc.h>
#include <sys/posix4.h>
#include <sys/ptrace.h>
#include <sys/racct.h>
#include <sys/resourcevar.h>
#include <sys/rwlock.h>
#include <sys/sched.h>
#include <sys/sysctl.h>
Refactor a bunch of scheduler code to give basically the same behaviour but with slightly cleaned up interfaces. The KSE structure has become the same as the "per thread scheduler private data" structure. In order to not make the diffs too great one is #defined as the other at this time. The KSE (or td_sched) structure is now allocated per thread and has no allocation code of its own. Concurrency for a KSEGRP is now kept track of via a simple pair of counters rather than using KSE structures as tokens. Since the KSE structure is different in each scheduler, kern_switch.c is now included at the end of each scheduler. Nothing outside the scheduler knows the contents of the KSE (aka td_sched) structure. The fields in the ksegrp structure that are to do with the scheduler's queueing mechanisms are now moved to the kg_sched structure. (per ksegrp scheduler private data structure). In other words how the scheduler queues and keeps track of threads is no-one's business except the scheduler's. This should allow people to write experimental schedulers with completely different internal structuring. A scheduler call sched_set_concurrency(kg, N) has been added that notifies teh scheduler that no more than N threads from that ksegrp should be allowed to be on concurrently scheduled. This is also used to enforce 'fainess' at this time so that a ksegrp with 10000 threads can not swamp a the run queue and force out a process with 1 thread, since the current code will not set the concurrency above NCPU, and both schedulers will not allow more than that many onto the system run queue at a time. Each scheduler should eventualy develop their own methods to do this now that they are effectively separated. Rejig libthr's kernel interface to follow the same code paths as linkse for scope system threads. This has slightly hurt libthr's performance but I will work to recover as much of it as I can. Thread exit code has been cleaned up greatly. exit and exec code now transitions a process back to 'standard non-threaded mode' before taking the next step. Reviewed by: scottl, peter MFC after: 1 week
2004-09-05 02:09:54 +00:00
#include <sys/smp.h>
#include <sys/syscallsubr.h>
#include <sys/sysent.h>
#include <sys/systm.h>
#include <sys/sysproto.h>
#include <sys/signalvar.h>
#include <sys/sysctl.h>
#include <sys/ucontext.h>
#include <sys/thr.h>
#include <sys/rtprio.h>
#include <sys/umtx.h>
#include <sys/limits.h>
Add an initial NUMA affinity/policy configuration for threads and processes. This is based on work done by jeff@ and jhb@, as well as the numa.diff patch that has been circulating when someone asks for first-touch NUMA on -10 or -11. * Introduce a simple set of VM policy and iterator types. * tie the policy types into the vm_phys path for now, mirroring how the initial first-touch allocation work was enabled. * add syscalls to control changing thread and process defaults. * add a global NUMA VM domain policy. * implement a simple cascade policy order - if a thread policy exists, use it; if a process policy exists, use it; use the default policy. * processes inherit policies from their parent processes, threads inherit policies from their parent threads. * add a simple tool (numactl) to query and modify default thread/process policities. * add documentation for the new syscalls, for numa and for numactl. * re-enable first touch NUMA again by default, as now policies can be set in a variety of methods. This is only relevant for very specific workloads. This doesn't pretend to be a final NUMA solution. The previous defaults in -HEAD (with MAXMEMDOM set) can be achieved by 'sysctl vm.default_policy=rr'. This is only relevant if MAXMEMDOM is set to something other than 1. Ie, if you're using GENERIC or a modified kernel with non-NUMA, then this is a glorified no-op for you. Thank you to Norse Corp for giving me access to rather large (for FreeBSD!) NUMA machines in order to develop and verify this. Thank you to Dell for providing me with dual socket sandybridge and westmere v3 hardware to do NUMA development with. Thank you to Scott Long at Netflix for providing me with access to the two-socket, four-domain haswell v3 hardware. Thank you to Peter Holm for running the stress testing suite against the NUMA branch during various stages of development! Tested: * MIPS (regression testing; non-NUMA) * i386 (regression testing; non-NUMA GENERIC) * amd64 (regression testing; non-NUMA GENERIC) * westmere, 2 socket (thankyou norse!) * sandy bridge, 2 socket (thankyou dell!) * ivy bridge, 2 socket (thankyou norse!) * westmere-EX, 4 socket / 1TB RAM (thankyou norse!) * haswell, 2 socket (thankyou norse!) * haswell v3, 2 socket (thankyou dell) * haswell v3, 2x18 core (thankyou scott long / netflix!) * Peter Holm ran a stress test suite on this work and found one issue, but has not been able to verify it (it doesn't look NUMA related, and he only saw it once over many testing runs.) * I've tested bhyve instances running in fixed NUMA domains and cpusets; all seems to work correctly. Verified: * intel-pcm - pcm-numa.x and pcm-memory.x, whilst selecting different NUMA policies for processes under test. Review: This was reviewed through phabricator (https://reviews.freebsd.org/D2559) as well as privately and via emails to freebsd-arch@. The git history with specific attributes is available at https://github.com/erikarn/freebsd/ in the NUMA branch (https://github.com/erikarn/freebsd/compare/local/adrian_numa_policy). This has been reviewed by a number of people (stas, rpaulo, kib, ngie, wblock) but not achieved a clear consensus. My hope is that with further exposure and testing more functionality can be implemented and evaluated. Notes: * The VM doesn't handle unbalanced domains very well, and if you have an overly unbalanced memory setup whilst under high memory pressure, VM page allocation may fail leading to a kernel panic. This was a problem in the past, but it's much more easily triggered now with these tools. * This work only controls the path through vm_phys; it doesn't yet strongly/predictably affect contigmalloc, KVA placement, UMA, etc. So, driver placement of memory isn't really guaranteed in any way. That's next on my plate. Sponsored by: Norse Corp, Inc.; Dell
2015-07-11 15:21:37 +00:00
#include <vm/vm_domain.h>
#include <machine/frame.h>
#include <security/audit/audit.h>
static SYSCTL_NODE(_kern, OID_AUTO, threads, CTLFLAG_RW, 0,
"thread allocation");
static int max_threads_per_proc = 1500;
SYSCTL_INT(_kern_threads, OID_AUTO, max_threads_per_proc, CTLFLAG_RW,
&max_threads_per_proc, 0, "Limit on threads per proc");
static int max_threads_hits;
SYSCTL_INT(_kern_threads, OID_AUTO, max_threads_hits, CTLFLAG_RD,
&max_threads_hits, 0, "kern.threads.max_threads_per_proc hit count");
#ifdef COMPAT_FREEBSD32
static inline int
suword_lwpid(void *addr, lwpid_t lwpid)
{
int error;
if (SV_CURPROC_FLAG(SV_LP64))
error = suword(addr, lwpid);
else
error = suword32(addr, lwpid);
return (error);
}
#else
#define suword_lwpid suword
#endif
/*
* System call interface.
*/
struct thr_create_initthr_args {
ucontext_t ctx;
long *tid;
};
static int
thr_create_initthr(struct thread *td, void *thunk)
{
struct thr_create_initthr_args *args;
/* Copy out the child tid. */
args = thunk;
if (args->tid != NULL && suword_lwpid(args->tid, td->td_tid))
return (EFAULT);
return (set_mcontext(td, &args->ctx.uc_mcontext));
}
int
sys_thr_create(struct thread *td, struct thr_create_args *uap)
/* ucontext_t *ctx, long *id, int flags */
{
struct thr_create_initthr_args args;
int error;
if ((error = copyin(uap->ctx, &args.ctx, sizeof(args.ctx))))
return (error);
args.tid = uap->id;
return (thread_create(td, NULL, thr_create_initthr, &args));
}
int
sys_thr_new(struct thread *td, struct thr_new_args *uap)
/* struct thr_param * */
{
struct thr_param param;
int error;
if (uap->param_size < 0 || uap->param_size > sizeof(param))
return (EINVAL);
bzero(&param, sizeof(param));
if ((error = copyin(uap->param, &param, uap->param_size)))
return (error);
return (kern_thr_new(td, &param));
}
static int
thr_new_initthr(struct thread *td, void *thunk)
{
stack_t stack;
struct thr_param *param;
/*
* Here we copy out tid to two places, one for child and one
* for parent, because pthread can create a detached thread,
* if parent wants to safely access child tid, it has to provide
* its storage, because child thread may exit quickly and
* memory is freed before parent thread can access it.
*/
param = thunk;
if ((param->child_tid != NULL &&
suword_lwpid(param->child_tid, td->td_tid)) ||
(param->parent_tid != NULL &&
suword_lwpid(param->parent_tid, td->td_tid)))
return (EFAULT);
/* Set up our machine context. */
stack.ss_sp = param->stack_base;
stack.ss_size = param->stack_size;
/* Set upcall address to user thread entry function. */
cpu_set_upcall(td, param->start_func, param->arg, &stack);
/* Setup user TLS address and TLS pointer register. */
return (cpu_set_user_tls(td, param->tls_base));
}
int
kern_thr_new(struct thread *td, struct thr_param *param)
{
struct rtprio rtp, *rtpp;
int error;
rtpp = NULL;
if (param->rtp != 0) {
error = copyin(param->rtp, &rtp, sizeof(struct rtprio));
if (error)
return (error);
rtpp = &rtp;
}
return (thread_create(td, rtpp, thr_new_initthr, param));
}
int
thread_create(struct thread *td, struct rtprio *rtp,
int (*initialize_thread)(struct thread *, void *), void *thunk)
{
struct thread *newtd;
Refactor a bunch of scheduler code to give basically the same behaviour but with slightly cleaned up interfaces. The KSE structure has become the same as the "per thread scheduler private data" structure. In order to not make the diffs too great one is #defined as the other at this time. The KSE (or td_sched) structure is now allocated per thread and has no allocation code of its own. Concurrency for a KSEGRP is now kept track of via a simple pair of counters rather than using KSE structures as tokens. Since the KSE structure is different in each scheduler, kern_switch.c is now included at the end of each scheduler. Nothing outside the scheduler knows the contents of the KSE (aka td_sched) structure. The fields in the ksegrp structure that are to do with the scheduler's queueing mechanisms are now moved to the kg_sched structure. (per ksegrp scheduler private data structure). In other words how the scheduler queues and keeps track of threads is no-one's business except the scheduler's. This should allow people to write experimental schedulers with completely different internal structuring. A scheduler call sched_set_concurrency(kg, N) has been added that notifies teh scheduler that no more than N threads from that ksegrp should be allowed to be on concurrently scheduled. This is also used to enforce 'fainess' at this time so that a ksegrp with 10000 threads can not swamp a the run queue and force out a process with 1 thread, since the current code will not set the concurrency above NCPU, and both schedulers will not allow more than that many onto the system run queue at a time. Each scheduler should eventualy develop their own methods to do this now that they are effectively separated. Rejig libthr's kernel interface to follow the same code paths as linkse for scope system threads. This has slightly hurt libthr's performance but I will work to recover as much of it as I can. Thread exit code has been cleaned up greatly. exit and exec code now transitions a process back to 'standard non-threaded mode' before taking the next step. Reviewed by: scottl, peter MFC after: 1 week
2004-09-05 02:09:54 +00:00
struct proc *p;
int error;
Refactor a bunch of scheduler code to give basically the same behaviour but with slightly cleaned up interfaces. The KSE structure has become the same as the "per thread scheduler private data" structure. In order to not make the diffs too great one is #defined as the other at this time. The KSE (or td_sched) structure is now allocated per thread and has no allocation code of its own. Concurrency for a KSEGRP is now kept track of via a simple pair of counters rather than using KSE structures as tokens. Since the KSE structure is different in each scheduler, kern_switch.c is now included at the end of each scheduler. Nothing outside the scheduler knows the contents of the KSE (aka td_sched) structure. The fields in the ksegrp structure that are to do with the scheduler's queueing mechanisms are now moved to the kg_sched structure. (per ksegrp scheduler private data structure). In other words how the scheduler queues and keeps track of threads is no-one's business except the scheduler's. This should allow people to write experimental schedulers with completely different internal structuring. A scheduler call sched_set_concurrency(kg, N) has been added that notifies teh scheduler that no more than N threads from that ksegrp should be allowed to be on concurrently scheduled. This is also used to enforce 'fainess' at this time so that a ksegrp with 10000 threads can not swamp a the run queue and force out a process with 1 thread, since the current code will not set the concurrency above NCPU, and both schedulers will not allow more than that many onto the system run queue at a time. Each scheduler should eventualy develop their own methods to do this now that they are effectively separated. Rejig libthr's kernel interface to follow the same code paths as linkse for scope system threads. This has slightly hurt libthr's performance but I will work to recover as much of it as I can. Thread exit code has been cleaned up greatly. exit and exec code now transitions a process back to 'standard non-threaded mode' before taking the next step. Reviewed by: scottl, peter MFC after: 1 week
2004-09-05 02:09:54 +00:00
p = td->td_proc;
if (rtp != NULL) {
switch(rtp->type) {
case RTP_PRIO_REALTIME:
case RTP_PRIO_FIFO:
2006-07-11 08:19:57 +00:00
/* Only root can set scheduler policy */
if (priv_check(td, PRIV_SCHED_SETPOLICY) != 0)
return (EPERM);
if (rtp->prio > RTP_PRIO_MAX)
return (EINVAL);
2006-07-11 08:19:57 +00:00
break;
case RTP_PRIO_NORMAL:
rtp->prio = 0;
2006-07-11 08:19:57 +00:00
break;
default:
return (EINVAL);
}
}
#ifdef RACCT
if (racct_enable) {
PROC_LOCK(p);
error = racct_add(p, RACCT_NTHR, 1);
PROC_UNLOCK(p);
if (error != 0)
return (EPROCLIM);
}
#endif
/* Initialize our td */
error = kern_thr_alloc(p, 0, &newtd);
if (error)
goto fail;
cpu_copy_thread(newtd, td);
Refactor a bunch of scheduler code to give basically the same behaviour but with slightly cleaned up interfaces. The KSE structure has become the same as the "per thread scheduler private data" structure. In order to not make the diffs too great one is #defined as the other at this time. The KSE (or td_sched) structure is now allocated per thread and has no allocation code of its own. Concurrency for a KSEGRP is now kept track of via a simple pair of counters rather than using KSE structures as tokens. Since the KSE structure is different in each scheduler, kern_switch.c is now included at the end of each scheduler. Nothing outside the scheduler knows the contents of the KSE (aka td_sched) structure. The fields in the ksegrp structure that are to do with the scheduler's queueing mechanisms are now moved to the kg_sched structure. (per ksegrp scheduler private data structure). In other words how the scheduler queues and keeps track of threads is no-one's business except the scheduler's. This should allow people to write experimental schedulers with completely different internal structuring. A scheduler call sched_set_concurrency(kg, N) has been added that notifies teh scheduler that no more than N threads from that ksegrp should be allowed to be on concurrently scheduled. This is also used to enforce 'fainess' at this time so that a ksegrp with 10000 threads can not swamp a the run queue and force out a process with 1 thread, since the current code will not set the concurrency above NCPU, and both schedulers will not allow more than that many onto the system run queue at a time. Each scheduler should eventualy develop their own methods to do this now that they are effectively separated. Rejig libthr's kernel interface to follow the same code paths as linkse for scope system threads. This has slightly hurt libthr's performance but I will work to recover as much of it as I can. Thread exit code has been cleaned up greatly. exit and exec code now transitions a process back to 'standard non-threaded mode' before taking the next step. Reviewed by: scottl, peter MFC after: 1 week
2004-09-05 02:09:54 +00:00
bzero(&newtd->td_startzero,
__rangeof(struct thread, td_startzero, td_endzero));
Refactor a bunch of scheduler code to give basically the same behaviour but with slightly cleaned up interfaces. The KSE structure has become the same as the "per thread scheduler private data" structure. In order to not make the diffs too great one is #defined as the other at this time. The KSE (or td_sched) structure is now allocated per thread and has no allocation code of its own. Concurrency for a KSEGRP is now kept track of via a simple pair of counters rather than using KSE structures as tokens. Since the KSE structure is different in each scheduler, kern_switch.c is now included at the end of each scheduler. Nothing outside the scheduler knows the contents of the KSE (aka td_sched) structure. The fields in the ksegrp structure that are to do with the scheduler's queueing mechanisms are now moved to the kg_sched structure. (per ksegrp scheduler private data structure). In other words how the scheduler queues and keeps track of threads is no-one's business except the scheduler's. This should allow people to write experimental schedulers with completely different internal structuring. A scheduler call sched_set_concurrency(kg, N) has been added that notifies teh scheduler that no more than N threads from that ksegrp should be allowed to be on concurrently scheduled. This is also used to enforce 'fainess' at this time so that a ksegrp with 10000 threads can not swamp a the run queue and force out a process with 1 thread, since the current code will not set the concurrency above NCPU, and both schedulers will not allow more than that many onto the system run queue at a time. Each scheduler should eventualy develop their own methods to do this now that they are effectively separated. Rejig libthr's kernel interface to follow the same code paths as linkse for scope system threads. This has slightly hurt libthr's performance but I will work to recover as much of it as I can. Thread exit code has been cleaned up greatly. exit and exec code now transitions a process back to 'standard non-threaded mode' before taking the next step. Reviewed by: scottl, peter MFC after: 1 week
2004-09-05 02:09:54 +00:00
bcopy(&td->td_startcopy, &newtd->td_startcopy,
__rangeof(struct thread, td_startcopy, td_endcopy));
Refactor a bunch of scheduler code to give basically the same behaviour but with slightly cleaned up interfaces. The KSE structure has become the same as the "per thread scheduler private data" structure. In order to not make the diffs too great one is #defined as the other at this time. The KSE (or td_sched) structure is now allocated per thread and has no allocation code of its own. Concurrency for a KSEGRP is now kept track of via a simple pair of counters rather than using KSE structures as tokens. Since the KSE structure is different in each scheduler, kern_switch.c is now included at the end of each scheduler. Nothing outside the scheduler knows the contents of the KSE (aka td_sched) structure. The fields in the ksegrp structure that are to do with the scheduler's queueing mechanisms are now moved to the kg_sched structure. (per ksegrp scheduler private data structure). In other words how the scheduler queues and keeps track of threads is no-one's business except the scheduler's. This should allow people to write experimental schedulers with completely different internal structuring. A scheduler call sched_set_concurrency(kg, N) has been added that notifies teh scheduler that no more than N threads from that ksegrp should be allowed to be on concurrently scheduled. This is also used to enforce 'fainess' at this time so that a ksegrp with 10000 threads can not swamp a the run queue and force out a process with 1 thread, since the current code will not set the concurrency above NCPU, and both schedulers will not allow more than that many onto the system run queue at a time. Each scheduler should eventualy develop their own methods to do this now that they are effectively separated. Rejig libthr's kernel interface to follow the same code paths as linkse for scope system threads. This has slightly hurt libthr's performance but I will work to recover as much of it as I can. Thread exit code has been cleaned up greatly. exit and exec code now transitions a process back to 'standard non-threaded mode' before taking the next step. Reviewed by: scottl, peter MFC after: 1 week
2004-09-05 02:09:54 +00:00
newtd->td_proc = td->td_proc;
newtd->td_rb_list = newtd->td_rbp_list = newtd->td_rb_inact = 0;
thread_cow_get(newtd, td);
error = initialize_thread(newtd, thunk);
if (error != 0) {
thread_cow_free(newtd);
thread_free(newtd);
goto fail;
}
PROC_LOCK(p);
p->p_flag |= P_HADTHREADS;
thread_link(newtd, p);
bcopy(p->p_comm, newtd->td_name, sizeof(newtd->td_name));
thread_lock(td);
Refactor a bunch of scheduler code to give basically the same behaviour but with slightly cleaned up interfaces. The KSE structure has become the same as the "per thread scheduler private data" structure. In order to not make the diffs too great one is #defined as the other at this time. The KSE (or td_sched) structure is now allocated per thread and has no allocation code of its own. Concurrency for a KSEGRP is now kept track of via a simple pair of counters rather than using KSE structures as tokens. Since the KSE structure is different in each scheduler, kern_switch.c is now included at the end of each scheduler. Nothing outside the scheduler knows the contents of the KSE (aka td_sched) structure. The fields in the ksegrp structure that are to do with the scheduler's queueing mechanisms are now moved to the kg_sched structure. (per ksegrp scheduler private data structure). In other words how the scheduler queues and keeps track of threads is no-one's business except the scheduler's. This should allow people to write experimental schedulers with completely different internal structuring. A scheduler call sched_set_concurrency(kg, N) has been added that notifies teh scheduler that no more than N threads from that ksegrp should be allowed to be on concurrently scheduled. This is also used to enforce 'fainess' at this time so that a ksegrp with 10000 threads can not swamp a the run queue and force out a process with 1 thread, since the current code will not set the concurrency above NCPU, and both schedulers will not allow more than that many onto the system run queue at a time. Each scheduler should eventualy develop their own methods to do this now that they are effectively separated. Rejig libthr's kernel interface to follow the same code paths as linkse for scope system threads. This has slightly hurt libthr's performance but I will work to recover as much of it as I can. Thread exit code has been cleaned up greatly. exit and exec code now transitions a process back to 'standard non-threaded mode' before taking the next step. Reviewed by: scottl, peter MFC after: 1 week
2004-09-05 02:09:54 +00:00
/* let the scheduler know about these things. */
sched_fork_thread(td, newtd);
thread_unlock(td);
if (P_SHOULDSTOP(p))
newtd->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK;
if (p->p_ptevents & PTRACE_LWP)
newtd->td_dbgflags |= TDB_BORN;
Add an initial NUMA affinity/policy configuration for threads and processes. This is based on work done by jeff@ and jhb@, as well as the numa.diff patch that has been circulating when someone asks for first-touch NUMA on -10 or -11. * Introduce a simple set of VM policy and iterator types. * tie the policy types into the vm_phys path for now, mirroring how the initial first-touch allocation work was enabled. * add syscalls to control changing thread and process defaults. * add a global NUMA VM domain policy. * implement a simple cascade policy order - if a thread policy exists, use it; if a process policy exists, use it; use the default policy. * processes inherit policies from their parent processes, threads inherit policies from their parent threads. * add a simple tool (numactl) to query and modify default thread/process policities. * add documentation for the new syscalls, for numa and for numactl. * re-enable first touch NUMA again by default, as now policies can be set in a variety of methods. This is only relevant for very specific workloads. This doesn't pretend to be a final NUMA solution. The previous defaults in -HEAD (with MAXMEMDOM set) can be achieved by 'sysctl vm.default_policy=rr'. This is only relevant if MAXMEMDOM is set to something other than 1. Ie, if you're using GENERIC or a modified kernel with non-NUMA, then this is a glorified no-op for you. Thank you to Norse Corp for giving me access to rather large (for FreeBSD!) NUMA machines in order to develop and verify this. Thank you to Dell for providing me with dual socket sandybridge and westmere v3 hardware to do NUMA development with. Thank you to Scott Long at Netflix for providing me with access to the two-socket, four-domain haswell v3 hardware. Thank you to Peter Holm for running the stress testing suite against the NUMA branch during various stages of development! Tested: * MIPS (regression testing; non-NUMA) * i386 (regression testing; non-NUMA GENERIC) * amd64 (regression testing; non-NUMA GENERIC) * westmere, 2 socket (thankyou norse!) * sandy bridge, 2 socket (thankyou dell!) * ivy bridge, 2 socket (thankyou norse!) * westmere-EX, 4 socket / 1TB RAM (thankyou norse!) * haswell, 2 socket (thankyou norse!) * haswell v3, 2 socket (thankyou dell) * haswell v3, 2x18 core (thankyou scott long / netflix!) * Peter Holm ran a stress test suite on this work and found one issue, but has not been able to verify it (it doesn't look NUMA related, and he only saw it once over many testing runs.) * I've tested bhyve instances running in fixed NUMA domains and cpusets; all seems to work correctly. Verified: * intel-pcm - pcm-numa.x and pcm-memory.x, whilst selecting different NUMA policies for processes under test. Review: This was reviewed through phabricator (https://reviews.freebsd.org/D2559) as well as privately and via emails to freebsd-arch@. The git history with specific attributes is available at https://github.com/erikarn/freebsd/ in the NUMA branch (https://github.com/erikarn/freebsd/compare/local/adrian_numa_policy). This has been reviewed by a number of people (stas, rpaulo, kib, ngie, wblock) but not achieved a clear consensus. My hope is that with further exposure and testing more functionality can be implemented and evaluated. Notes: * The VM doesn't handle unbalanced domains very well, and if you have an overly unbalanced memory setup whilst under high memory pressure, VM page allocation may fail leading to a kernel panic. This was a problem in the past, but it's much more easily triggered now with these tools. * This work only controls the path through vm_phys; it doesn't yet strongly/predictably affect contigmalloc, KVA placement, UMA, etc. So, driver placement of memory isn't really guaranteed in any way. That's next on my plate. Sponsored by: Norse Corp, Inc.; Dell
2015-07-11 15:21:37 +00:00
/*
* Copy the existing thread VM policy into the new thread.
*/
vm_domain_policy_localcopy(&newtd->td_vm_dom_policy,
&td->td_vm_dom_policy);
PROC_UNLOCK(p);
tidhash_add(newtd);
thread_lock(newtd);
if (rtp != NULL) {
if (!(td->td_pri_class == PRI_TIMESHARE &&
rtp->type == RTP_PRIO_NORMAL)) {
rtp_to_pri(rtp, newtd);
sched_prio(newtd, newtd->td_user_pri);
} /* ignore timesharing class */
}
Refactor a bunch of scheduler code to give basically the same behaviour but with slightly cleaned up interfaces. The KSE structure has become the same as the "per thread scheduler private data" structure. In order to not make the diffs too great one is #defined as the other at this time. The KSE (or td_sched) structure is now allocated per thread and has no allocation code of its own. Concurrency for a KSEGRP is now kept track of via a simple pair of counters rather than using KSE structures as tokens. Since the KSE structure is different in each scheduler, kern_switch.c is now included at the end of each scheduler. Nothing outside the scheduler knows the contents of the KSE (aka td_sched) structure. The fields in the ksegrp structure that are to do with the scheduler's queueing mechanisms are now moved to the kg_sched structure. (per ksegrp scheduler private data structure). In other words how the scheduler queues and keeps track of threads is no-one's business except the scheduler's. This should allow people to write experimental schedulers with completely different internal structuring. A scheduler call sched_set_concurrency(kg, N) has been added that notifies teh scheduler that no more than N threads from that ksegrp should be allowed to be on concurrently scheduled. This is also used to enforce 'fainess' at this time so that a ksegrp with 10000 threads can not swamp a the run queue and force out a process with 1 thread, since the current code will not set the concurrency above NCPU, and both schedulers will not allow more than that many onto the system run queue at a time. Each scheduler should eventualy develop their own methods to do this now that they are effectively separated. Rejig libthr's kernel interface to follow the same code paths as linkse for scope system threads. This has slightly hurt libthr's performance but I will work to recover as much of it as I can. Thread exit code has been cleaned up greatly. exit and exec code now transitions a process back to 'standard non-threaded mode' before taking the next step. Reviewed by: scottl, peter MFC after: 1 week
2004-09-05 02:09:54 +00:00
TD_SET_CAN_RUN(newtd);
sched_add(newtd, SRQ_BORING);
thread_unlock(newtd);
return (0);
fail:
#ifdef RACCT
if (racct_enable) {
PROC_LOCK(p);
racct_sub(p, RACCT_NTHR, 1);
PROC_UNLOCK(p);
}
#endif
return (error);
}
int
sys_thr_self(struct thread *td, struct thr_self_args *uap)
/* long *id */
{
int error;
error = suword_lwpid(uap->id, (unsigned)td->td_tid);
if (error == -1)
return (EFAULT);
return (0);
}
int
sys_thr_exit(struct thread *td, struct thr_exit_args *uap)
/* long *state */
{
Add implementation of robust mutexes, hopefully close enough to the intention of the POSIX IEEE Std 1003.1TM-2008/Cor 1-2013. A robust mutex is guaranteed to be cleared by the system upon either thread or process owner termination while the mutex is held. The next mutex locker is then notified about inconsistent mutex state and can execute (or abandon) corrective actions. The patch mostly consists of small changes here and there, adding neccessary checks for the inconsistent and abandoned conditions into existing paths. Additionally, the thread exit handler was extended to iterate over the userspace-maintained list of owned robust mutexes, unlocking and marking as terminated each of them. The list of owned robust mutexes cannot be maintained atomically synchronous with the mutex lock state (it is possible in kernel, but is too expensive). Instead, for the duration of lock or unlock operation, the current mutex is remembered in a special slot that is also checked by the kernel at thread termination. Kernel must be aware about the per-thread location of the heads of robust mutex lists and the current active mutex slot. When a thread touches a robust mutex for the first time, a new umtx op syscall is issued which informs about location of lists heads. The umtx sleep queues for PP and PI mutexes are split between non-robust and robust. Somewhat unrelated changes in the patch: 1. Style. 2. The fix for proper tdfind() call use in umtxq_sleep_pi() for shared pi mutexes. 3. Removal of the userspace struct pthread_mutex m_owner field. 4. The sysctl kern.ipc.umtx_vnode_persistent is added, which controls the lifetime of the shared mutex associated with a vnode' page. Reviewed by: jilles (previous version, supposedly the objection was fixed) Discussed with: brooks, Martin Simmons <martin@lispworks.com> (some aspects) Tested by: pho Sponsored by: The FreeBSD Foundation
2016-05-17 09:56:22 +00:00
umtx_thread_exit(td);
/* Signal userland that it can free the stack. */
if ((void *)uap->state != NULL) {
suword_lwpid(uap->state, 1);
2008-04-29 05:48:05 +00:00
kern_umtx_wake(td, uap->state, INT_MAX, 0);
}
return (kern_thr_exit(td));
}
int
kern_thr_exit(struct thread *td)
{
struct proc *p;
p = td->td_proc;
/*
* If all of the threads in a process call this routine to
* exit (e.g. all threads call pthread_exit()), exactly one
* thread should return to the caller to terminate the process
* instead of the thread.
*
* Checking p_numthreads alone is not sufficient since threads
* might be committed to terminating while the PROC_LOCK is
* dropped in either ptracestop() or while removing this thread
* from the tidhash. Instead, the p_pendingexits field holds
* the count of threads in either of those states and a thread
* is considered the "last" thread if all of the other threads
* in a process are already terminating.
*/
PROC_LOCK(p);
if (p->p_numthreads == p->p_pendingexits + 1) {
/*
* Ignore attempts to shut down last thread in the
* proc. This will actually call _exit(2) in the
* usermode trampoline when it returns.
*/
PROC_UNLOCK(p);
return (0);
Refactor a bunch of scheduler code to give basically the same behaviour but with slightly cleaned up interfaces. The KSE structure has become the same as the "per thread scheduler private data" structure. In order to not make the diffs too great one is #defined as the other at this time. The KSE (or td_sched) structure is now allocated per thread and has no allocation code of its own. Concurrency for a KSEGRP is now kept track of via a simple pair of counters rather than using KSE structures as tokens. Since the KSE structure is different in each scheduler, kern_switch.c is now included at the end of each scheduler. Nothing outside the scheduler knows the contents of the KSE (aka td_sched) structure. The fields in the ksegrp structure that are to do with the scheduler's queueing mechanisms are now moved to the kg_sched structure. (per ksegrp scheduler private data structure). In other words how the scheduler queues and keeps track of threads is no-one's business except the scheduler's. This should allow people to write experimental schedulers with completely different internal structuring. A scheduler call sched_set_concurrency(kg, N) has been added that notifies teh scheduler that no more than N threads from that ksegrp should be allowed to be on concurrently scheduled. This is also used to enforce 'fainess' at this time so that a ksegrp with 10000 threads can not swamp a the run queue and force out a process with 1 thread, since the current code will not set the concurrency above NCPU, and both schedulers will not allow more than that many onto the system run queue at a time. Each scheduler should eventualy develop their own methods to do this now that they are effectively separated. Rejig libthr's kernel interface to follow the same code paths as linkse for scope system threads. This has slightly hurt libthr's performance but I will work to recover as much of it as I can. Thread exit code has been cleaned up greatly. exit and exec code now transitions a process back to 'standard non-threaded mode' before taking the next step. Reviewed by: scottl, peter MFC after: 1 week
2004-09-05 02:09:54 +00:00
}
p->p_pendingexits++;
td->td_dbgflags |= TDB_EXIT;
if (p->p_ptevents & PTRACE_LWP)
ptracestop(td, SIGTRAP);
PROC_UNLOCK(p);
tidhash_remove(td);
PROC_LOCK(p);
p->p_pendingexits--;
/*
* The check above should prevent all other threads from this
* process from exiting while the PROC_LOCK is dropped, so
* there must be at least one other thread other than the
* current thread.
*/
KASSERT(p->p_numthreads > 1, ("too few threads"));
racct_sub(p, RACCT_NTHR, 1);
tdsigcleanup(td);
PROC_SLOCK(p);
thread_stopped(p);
thread_exit();
/* NOTREACHED */
}
int
sys_thr_kill(struct thread *td, struct thr_kill_args *uap)
/* long id, int sig */
{
ksiginfo_t ksi;
struct thread *ttd;
struct proc *p;
int error;
p = td->td_proc;
ksiginfo_init(&ksi);
ksi.ksi_signo = uap->sig;
ksi.ksi_code = SI_LWP;
ksi.ksi_pid = p->p_pid;
ksi.ksi_uid = td->td_ucred->cr_ruid;
if (uap->id == -1) {
if (uap->sig != 0 && !_SIG_VALID(uap->sig)) {
error = EINVAL;
} else {
error = ESRCH;
PROC_LOCK(p);
FOREACH_THREAD_IN_PROC(p, ttd) {
if (ttd != td) {
error = 0;
if (uap->sig == 0)
break;
tdksignal(ttd, uap->sig, &ksi);
}
}
PROC_UNLOCK(p);
}
} else {
error = 0;
ttd = tdfind((lwpid_t)uap->id, p->p_pid);
if (ttd == NULL)
return (ESRCH);
if (uap->sig == 0)
;
else if (!_SIG_VALID(uap->sig))
error = EINVAL;
else
tdksignal(ttd, uap->sig, &ksi);
PROC_UNLOCK(ttd->td_proc);
}
return (error);
}
int
sys_thr_kill2(struct thread *td, struct thr_kill2_args *uap)
/* pid_t pid, long id, int sig */
{
ksiginfo_t ksi;
struct thread *ttd;
struct proc *p;
int error;
AUDIT_ARG_SIGNUM(uap->sig);
ksiginfo_init(&ksi);
ksi.ksi_signo = uap->sig;
ksi.ksi_code = SI_LWP;
ksi.ksi_pid = td->td_proc->p_pid;
ksi.ksi_uid = td->td_ucred->cr_ruid;
if (uap->id == -1) {
if ((p = pfind(uap->pid)) == NULL)
return (ESRCH);
AUDIT_ARG_PROCESS(p);
error = p_cansignal(td, p, uap->sig);
if (error) {
PROC_UNLOCK(p);
return (error);
}
if (uap->sig != 0 && !_SIG_VALID(uap->sig)) {
error = EINVAL;
} else {
error = ESRCH;
FOREACH_THREAD_IN_PROC(p, ttd) {
if (ttd != td) {
error = 0;
if (uap->sig == 0)
break;
tdksignal(ttd, uap->sig, &ksi);
}
}
}
PROC_UNLOCK(p);
} else {
ttd = tdfind((lwpid_t)uap->id, uap->pid);
if (ttd == NULL)
return (ESRCH);
p = ttd->td_proc;
AUDIT_ARG_PROCESS(p);
error = p_cansignal(td, p, uap->sig);
if (uap->sig == 0)
;
else if (!_SIG_VALID(uap->sig))
error = EINVAL;
else
tdksignal(ttd, uap->sig, &ksi);
PROC_UNLOCK(p);
}
return (error);
}
int
sys_thr_suspend(struct thread *td, struct thr_suspend_args *uap)
/* const struct timespec *timeout */
{
struct timespec ts, *tsp;
int error;
tsp = NULL;
if (uap->timeout != NULL) {
error = umtx_copyin_timeout(uap->timeout, &ts);
if (error != 0)
return (error);
tsp = &ts;
}
return (kern_thr_suspend(td, tsp));
}
int
kern_thr_suspend(struct thread *td, struct timespec *tsp)
{
struct proc *p = td->td_proc;
struct timeval tv;
int error = 0;
int timo = 0;
if (td->td_pflags & TDP_WAKEUP) {
td->td_pflags &= ~TDP_WAKEUP;
return (0);
}
if (tsp != NULL) {
if (tsp->tv_sec == 0 && tsp->tv_nsec == 0)
error = EWOULDBLOCK;
else {
TIMESPEC_TO_TIMEVAL(&tv, tsp);
timo = tvtohz(&tv);
}
}
PROC_LOCK(p);
if (error == 0 && (td->td_flags & TDF_THRWAKEUP) == 0)
error = msleep((void *)td, &p->p_mtx,
PCATCH, "lthr", timo);
if (td->td_flags & TDF_THRWAKEUP) {
thread_lock(td);
td->td_flags &= ~TDF_THRWAKEUP;
thread_unlock(td);
PROC_UNLOCK(p);
return (0);
}
PROC_UNLOCK(p);
if (error == EWOULDBLOCK)
error = ETIMEDOUT;
else if (error == ERESTART) {
if (timo != 0)
error = EINTR;
}
return (error);
}
int
sys_thr_wake(struct thread *td, struct thr_wake_args *uap)
/* long id */
{
struct proc *p;
struct thread *ttd;
if (uap->id == td->td_tid) {
td->td_pflags |= TDP_WAKEUP;
return (0);
}
p = td->td_proc;
ttd = tdfind((lwpid_t)uap->id, p->p_pid);
if (ttd == NULL)
return (ESRCH);
thread_lock(ttd);
ttd->td_flags |= TDF_THRWAKEUP;
thread_unlock(ttd);
wakeup((void *)ttd);
PROC_UNLOCK(p);
return (0);
}
int
sys_thr_set_name(struct thread *td, struct thr_set_name_args *uap)
{
struct proc *p;
char name[MAXCOMLEN + 1];
struct thread *ttd;
int error;
error = 0;
name[0] = '\0';
if (uap->name != NULL) {
error = copyinstr(uap->name, name, sizeof(name),
NULL);
if (error)
return (error);
}
p = td->td_proc;
ttd = tdfind((lwpid_t)uap->id, p->p_pid);
if (ttd == NULL)
return (ESRCH);
strcpy(ttd->td_name, name);
#ifdef KTR
sched_clear_tdname(ttd);
#endif
PROC_UNLOCK(p);
return (error);
}
int
kern_thr_alloc(struct proc *p, int pages, struct thread **ntd)
{
/* Have race condition but it is cheap. */
if (p->p_numthreads >= max_threads_per_proc) {
++max_threads_hits;
return (EPROCLIM);
}
*ntd = thread_alloc(pages);
if (*ntd == NULL)
return (ENOMEM);
return (0);
}