d/freebsd-dev
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Make KSE a kernel option, turned on by default in all GENERIC

Browse Source 
kernel configs except sun4v (which doesn't process signals properly
with KSE).

Reviewed by:	davidxu@
This commit is contained in:
John Birrell 2006-10-26 21:42:22 +00:00
parent 50159fa216
commit 8460a577a4
Notes: svn2git 2020-12-20 02:59:44 +00:00 
svn path=/head/; revision=163709
62 changed files with 1351 additions and 245 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
UPDATING
View File

@ -21,6 +21,15 @@ NOTE TO PEOPLE WHO THINK THAT FreeBSD 7.x IS SLOW:
developers choose to disable these features on build machines
to maximize performance.
20061025:
KSE in the kernel has now been made optional and turned on by
default in the GENERIC kernels. Either add 'options KSE' to your
custom kernel or edit /etc/libmap.conf to redirect libpthread.so.2
to libthr.so.2. All kernel modules *must* be recompiled after
this change. There-after, modules from a KSE kernel should be
compatible with modules from a NOKSE kernel due to the temporary
padding fields added to 'struct proc'.
20060929:
mrouted and its utilities have been removed from the base system.

8
lib/libkvm/kvm_proc.c
View File

@ -114,7 +114,7 @@ kvm_proclist(kd, what, arg, p, bp, maxcnt)
struct prison pr;
struct thread mtd;
/*struct kse mke;*/
struct ksegrp mkg;
/*struct ksegrp mkg;*/
struct proc proc;
struct proc pproc;
struct timeval tv;
@ -137,6 +137,7 @@ kvm_proclist(kd, what, arg, p, bp, maxcnt)
TAILQ_FIRST(&proc.p_threads));
return (-1);
}
#if 0
if ((proc.p_flag & P_SA) == 0) {
if (KREAD(kd,
(u_long)TAILQ_FIRST(&proc.p_ksegrps),
@ -146,7 +147,6 @@ kvm_proclist(kd, what, arg, p, bp, maxcnt)
TAILQ_FIRST(&proc.p_ksegrps));
return (-1);
}
#if 0
if (KREAD(kd,
(u_long)TAILQ_FIRST(&mkg.kg_kseq), &mke)) {
_kvm_err(kd, kd->program,
@ -154,8 +154,8 @@ kvm_proclist(kd, what, arg, p, bp, maxcnt)
TAILQ_FIRST(&mkg.kg_kseq));
return (-1);
}
#endif
}
#endif
}
if (KREAD(kd, (u_long)proc.p_ucred, &ucred) == 0) {
kp->ki_ruid = ucred.cr_ruid;
@ -425,13 +425,13 @@ kvm_proclist(kd, what, arg, p, bp, maxcnt)
kp->ki_oncpu = mtd.td_oncpu;
if (!(proc.p_flag & P_SA)) {
#if 0
/* stuff from the ksegrp */
kp->ki_slptime = mkg.kg_slptime;
kp->ki_pri.pri_class = mkg.kg_pri_class;
kp->ki_pri.pri_user = mkg.kg_user_pri;
kp->ki_estcpu = mkg.kg_estcpu;
#if 0
/* Stuff from the kse */
kp->ki_pctcpu = mke.ke_pctcpu;
kp->ki_rqindex = mke.ke_rqindex;

4
sys/amd64/amd64/machdep.c
View File

@ -1120,7 +1120,11 @@ hammer_time(u_int64_t modulep, u_int64_t physfree)
* This may be done better later if it gets more high level
* components in it. If so just link td->td_proc here.
*/
#ifdef KSE
proc_linkup(&proc0, &ksegrp0, &thread0);
#else
proc_linkup(&proc0, &thread0);
#endif
preload_metadata = (caddr_t)(uintptr_t)(modulep + KERNBASE);
preload_bootstrap_relocate(KERNBASE);

4
sys/amd64/amd64/trap.c
View File

@ -299,8 +299,10 @@ trap(frame)
case T_PAGEFLT: /* page fault */
addr = frame.tf_addr;
#ifdef KSE
if (td->td_pflags & TDP_SA)
thread_user_enter(td);
#endif
i = trap_pfault(&frame, TRUE);
if (i == -1)
goto userout;
@ -757,8 +759,10 @@ syscall(frame)
td->td_frame = &frame;
if (td->td_ucred != p->p_ucred)
cred_update_thread(td);
#ifdef KSE
if (p->p_flag & P_SA)
thread_user_enter(td);
#endif
params = (caddr_t)frame.tf_rsp + sizeof(register_t);
code = frame.tf_rax;
orig_tf_rflags = frame.tf_rflags;

1
sys/amd64/conf/GENERIC
View File

@ -58,6 +58,7 @@ options _KPOSIX_PRIORITY_SCHEDULING # POSIX P1003_1B real-time extensions
options KBD_INSTALL_CDEV # install a CDEV entry in /dev
options ADAPTIVE_GIANT # Giant mutex is adaptive.
options STOP_NMI # Stop CPUS using NMI instead of IPI
options KSE # KSE support
# Debugging for use in -current
options KDB # Enable kernel debugger support.

6
sys/arm/arm/trap.c
View File

@ -263,8 +263,10 @@ data_abort_handler(trapframe_t *tf)
td->td_frame = tf;
if (td->td_ucred != td->td_proc->p_ucred)
cred_update_thread(td);
#ifdef KSE
if (td->td_pflags & TDP_SA)
thread_user_enter(td);
#endif
}
/* Grab the current pcb */
@ -730,8 +732,10 @@ prefetch_abort_handler(trapframe_t *tf)
td->td_frame = tf;
if (td->td_ucred != td->td_proc->p_ucred)
cred_update_thread(td);
#ifdef KSE
if (td->td_proc->p_flag & P_SA)
thread_user_enter(td);
#endif
}
fault_pc = tf->tf_pc;
if (td->td_md.md_spinlock_count == 0) {
@ -1005,8 +1009,10 @@ swi_handler(trapframe_t *frame)
td->td_frame = frame;
td->td_pticks = 0;
#ifdef KSE
if (td->td_proc->p_flag & P_SA)
thread_user_enter(td);
#endif
/*
* Make sure the program counter is correctly aligned so we
* don't take an alignment fault trying to read the opcode.

4
sys/arm/at91/kb920x_machdep.c
View File

@ -456,7 +456,11 @@ initarm(void *arg, void *arg2)
undefined_handler_address = (u_int)undefinedinstruction_bounce;
undefined_init();
#ifdef KSE
proc_linkup(&proc0, &ksegrp0, &thread0);
#else
proc_linkup(&proc0, &thread0);
#endif
thread0.td_kstack = kernelstack.pv_va;
thread0.td_pcb = (struct pcb *)
(thread0.td_kstack + KSTACK_PAGES * PAGE_SIZE) - 1;

1
sys/arm/conf/EP80219
View File

@ -59,6 +59,7 @@ options KBD_INSTALL_CDEV # install a CDEV entry in /dev
options GEOM_GPT # GUID Partition Tables.
options GEOM_MBR # DOS/MBR partitioning
options GEOM_LABEL # Providers labelization.
options KSE # KSE support
options BOOTP
options BOOTP_NFSROOT

1
sys/arm/conf/IQ31244
View File

@ -61,6 +61,7 @@ options BOOTP_NFSV3
options BOOTP_WIRED_TO=em0
options BOOTP_COMPAT
#options PREEMPTION
options KSE # KSE support
device genclock
device loop
device ether

1
sys/arm/conf/KB920X
View File

@ -60,6 +60,7 @@ options MUTEX_NOINLINE
options RWLOCK_NOINLINE
options NO_FFS_SNAPSHOT
options NO_SWAPPING
options KSE # KSE support
device genclock
device random
device pty

1
sys/arm/conf/SIMICS
View File

@ -53,6 +53,7 @@ options SYSVSHM #SYSV-style shared memory
options SYSVMSG #SYSV-style message queues
options SYSVSEM #SYSV-style semaphores
options _KPOSIX_PRIORITY_SCHEDULING #Posix P1003_1B real-time extensions
options KSE # KSE support
device genclock
device loop
device ether

1
sys/arm/conf/SKYEYE
View File

@ -58,6 +58,7 @@ options SYSVMSG #SYSV-style message queues
options SYSVSEM #SYSV-style semaphores
options _KPOSIX_PRIORITY_SCHEDULING #Posix P1003_1B real-time extensions
options KBD_INSTALL_CDEV # install a CDEV entry in /dev
options KSE # KSE support
device genclock
device loop
device ether

4
sys/arm/sa11x0/assabet_machdep.c
View File

@ -423,7 +423,11 @@ initarm(void *arg, void *arg2)
/* Set stack for exception handlers */
#ifdef KSE
proc_linkup(&proc0, &ksegrp0, &thread0);
#else
proc_linkup(&proc0, &thread0);
#endif
thread0.td_kstack = kernelstack.pv_va;
thread0.td_pcb = (struct pcb *)
(thread0.td_kstack + KSTACK_PAGES * PAGE_SIZE) - 1;

4
sys/arm/xscale/i80321/ep80219_machdep.c
View File

@ -429,7 +429,11 @@ initarm(void *arg, void *arg2)
undefined_handler_address = (u_int)undefinedinstruction_bounce;
undefined_init();
#ifdef KSE
proc_linkup(&proc0, &ksegrp0, &thread0);
#else
proc_linkup(&proc0, &thread0);
#endif
thread0.td_kstack = kernelstack.pv_va;
thread0.td_pcb = (struct pcb *)
(thread0.td_kstack + KSTACK_PAGES * PAGE_SIZE) - 1;

4
sys/arm/xscale/i80321/iq31244_machdep.c
View File

@ -427,7 +427,11 @@ initarm(void *arg, void *arg2)
undefined_handler_address = (u_int)undefinedinstruction_bounce;
undefined_init();
#ifdef KSE
proc_linkup(&proc0, &ksegrp0, &thread0);
#else
proc_linkup(&proc0, &thread0);
#endif
thread0.td_kstack = kernelstack.pv_va;
thread0.td_pcb = (struct pcb *)
(thread0.td_kstack + KSTACK_PAGES * PAGE_SIZE) - 1;

4
sys/ddb/db_ps.c
View File

@ -292,8 +292,12 @@ DB_SHOW_COMMAND(thread, db_show_thread)
td = kdb_thread;
db_printf("Thread %d at %p:\n", td->td_tid, td);
#ifdef KSE
db_printf(" proc (pid %d): %p ", td->td_proc->p_pid, td->td_proc);
db_printf(" ksegrp: %p\n", td->td_ksegrp);
#else
db_printf(" proc (pid %d): %p\n", td->td_proc->p_pid, td->td_proc);
#endif
if (td->td_name[0] != '\0')
db_printf(" name: %s\n", td->td_name);
db_printf(" flags: %#x ", td->td_flags);

10
sys/fs/procfs/procfs_status.c
View File

@ -113,6 +113,7 @@ procfs_doprocstatus(PFS_FILL_ARGS)
}
mtx_lock_spin(&sched_lock);
#ifdef KSE
if (p->p_flag & P_SA)
wmesg = "-kse- ";
else {
@ -124,6 +125,15 @@ procfs_doprocstatus(PFS_FILL_ARGS)
} else
wmesg = "nochan";
}
#else
tdfirst = FIRST_THREAD_IN_PROC(p);
if (tdfirst->td_wchan != NULL) {
KASSERT(tdfirst->td_wmesg != NULL,
("wchan %p has no wmesg", tdfirst->td_wchan));
wmesg = tdfirst->td_wmesg;
} else
wmesg = "nochan";
#endif
mtx_unlock_spin(&sched_lock);
if (p->p_sflag & PS_INMEM) {

1
sys/i386/conf/GENERIC
View File

@ -58,6 +58,7 @@ options _KPOSIX_PRIORITY_SCHEDULING # POSIX P1003_1B real-time extensions
options KBD_INSTALL_CDEV # install a CDEV entry in /dev
options ADAPTIVE_GIANT # Giant mutex is adaptive.
options STOP_NMI # Stop CPUS using NMI instead of IPI
options KSE # KSE support
# Debugging for use in -current
options KDB # Enable kernel debugger support.

4
sys/i386/i386/machdep.c
View File

@ -2057,7 +2057,11 @@ init386(first)
* This may be done better later if it gets more high level
* components in it. If so just link td->td_proc here.
*/
#ifdef KSE
proc_linkup(&proc0, &ksegrp0, &thread0);
#else
proc_linkup(&proc0, &thread0);
#endif
metadata_missing = 0;
if (bootinfo.bi_modulep) {

4
sys/i386/i386/trap.c
View File

@ -346,8 +346,10 @@ trap(frame)
break;
case T_PAGEFLT: /* page fault */
#ifdef KSE
if (td->td_pflags & TDP_SA)
thread_user_enter(td);
#endif
i = trap_pfault(&frame, TRUE, eva);
#if defined(I586_CPU) && !defined(NO_F00F_HACK)
@ -935,8 +937,10 @@ syscall(frame)
td->td_frame = &frame;
if (td->td_ucred != p->p_ucred)
cred_update_thread(td);
#ifdef KSE
if (p->p_flag & P_SA)
thread_user_enter(td);
#endif
params = (caddr_t)frame.tf_esp + sizeof(int);
code = frame.tf_eax;
orig_tf_eflags = frame.tf_eflags;

1
sys/ia64/conf/GENERIC
View File

@ -53,6 +53,7 @@ options SYSVSHM # SYSV-style shared memory
options UFS_ACL # Support for access control lists
options UFS_DIRHASH # Hash-based directory lookup scheme
options _KPOSIX_PRIORITY_SCHEDULING # Posix P1003_1B RT extensions
options KSE # KSE support
# Various "busses"
device firewire # FireWire bus code

4
sys/ia64/ia64/machdep.c
View File

@ -775,7 +775,11 @@ ia64_init(void)
msgbufp = (struct msgbuf *)pmap_steal_memory(MSGBUF_SIZE);
msgbufinit(msgbufp, MSGBUF_SIZE);
#ifdef KSE
proc_linkup(&proc0, &ksegrp0, &thread0);
#else
proc_linkup(&proc0, &thread0);
#endif
/*
* Init mapping for kernel stack for proc 0
*/

2
sys/ia64/ia64/trap.c
View File

@ -975,8 +975,10 @@ syscall(struct trapframe *tf)
td->td_pticks = 0;
if (td->td_ucred != p->p_ucred)
cred_update_thread(td);
#ifdef KSE
if (p->p_flag & P_SA)
thread_user_enter(td);
#endif
if (p->p_sysent->sv_prepsyscall) {
/* (*p->p_sysent->sv_prepsyscall)(tf, args, &code, &params); */

28
sys/kern/init_main.c
View File

@ -95,7 +95,9 @@ static struct session session0;
static struct pgrp pgrp0;
struct proc proc0;
struct thread thread0 __aligned(8);
#ifdef KSE
struct ksegrp ksegrp0;
#endif
struct vmspace vmspace0;
struct proc *initproc;
@ -364,18 +366,23 @@ proc0_init(void *dummy __unused)
struct proc *p;
unsigned i;
struct thread *td;
#ifdef KSE
struct ksegrp *kg;
#endif
GIANT_REQUIRED;
p = &proc0;
td = &thread0;
#ifdef KSE
kg = &ksegrp0;
#endif
/*
* Initialize magic number.
*/
p->p_magic = P_MAGIC;
#ifdef KSE
/*
* Initialize thread, process and ksegrp structures.
*/
@ -386,6 +393,18 @@ proc0_init(void *dummy __unused)
* Initialise scheduler resources.
* Add scheduler specific parts to proc, ksegrp, thread as needed.
*/
#else
/*
* Initialize thread and process structures.
*/
procinit(); /* set up proc zone */
threadinit(); /* set up UMA zones */
/*
* Initialise scheduler resources.
* Add scheduler specific parts to proc, thread as needed.
*/
#endif
schedinit(); /* scheduler gets its house in order */
/*
* Initialize sleep queue hash table
@ -421,9 +440,14 @@ proc0_init(void *dummy __unused)
STAILQ_INIT(&p->p_ktr);
p->p_nice = NZERO;
td->td_state = TDS_RUNNING;
#ifdef KSE
kg->kg_pri_class = PRI_TIMESHARE;
kg->kg_user_pri = PUSER;
kg->kg_base_user_pri = PUSER;
#else
td->td_pri_class = PRI_TIMESHARE;
td->td_user_pri = PUSER;
#endif
td->td_priority = PVM;
td->td_base_pri = PUSER;
td->td_oncpu = 0;
@ -733,7 +757,11 @@ kick_init(const void *udata __unused)
td = FIRST_THREAD_IN_PROC(initproc);
mtx_lock_spin(&sched_lock);
TD_SET_CAN_RUN(td);
#ifdef KSE
setrunqueue(td, SRQ_BORING); /* XXXKSE */
#else
setrunqueue(td, SRQ_BORING);
#endif
mtx_unlock_spin(&sched_lock);
}
SYSINIT(kickinit, SI_SUB_KTHREAD_INIT, SI_ORDER_FIRST, kick_init, NULL)

19
sys/kern/kern_clock.c
View File

@ -202,6 +202,7 @@ hardclock_cpu(int usermode)
*/
mtx_lock_spin_flags(&sched_lock, MTX_QUIET);
sched_tick();
#ifdef KSE
if (p->p_flag & P_SA) {
/* XXXKSE What to do? */
} else {
@ -218,6 +219,20 @@ hardclock_cpu(int usermode)
td->td_flags |= TDF_ASTPENDING;
}
}
#else
pstats = p->p_stats;
if (usermode &&
timevalisset(&pstats->p_timer[ITIMER_VIRTUAL].it_value) &&
itimerdecr(&pstats->p_timer[ITIMER_VIRTUAL], tick) == 0) {
p->p_sflag |= PS_ALRMPEND;
td->td_flags |= TDF_ASTPENDING;
}
if (timevalisset(&pstats->p_timer[ITIMER_PROF].it_value) &&
itimerdecr(&pstats->p_timer[ITIMER_PROF], tick) == 0) {
p->p_sflag |= PS_PROFPEND;
td->td_flags |= TDF_ASTPENDING;
}
#endif
mtx_unlock_spin_flags(&sched_lock, MTX_QUIET);
#ifdef HWPMC_HOOKS
@ -414,8 +429,10 @@ statclock(int usermode)
/*
* Charge the time as appropriate.
*/
#ifdef KSE
if (p->p_flag & P_SA)
thread_statclock(1);
#endif
td->td_uticks++;
if (p->p_nice > NZERO)
cp_time[CP_NICE]++;
@ -439,8 +456,10 @@ statclock(int usermode)
td->td_iticks++;
cp_time[CP_INTR]++;
} else {
#ifdef KSE
if (p->p_flag & P_SA)
thread_statclock(0);
#endif
td->td_pticks++;
td->td_sticks++;
if (td != PCPU_GET(idlethread))

12
sys/kern/kern_fork.c
View File

@ -204,7 +204,9 @@ fork1(td, flags, pages, procp)
struct filedesc *fd;
struct filedesc_to_leader *fdtol;
struct thread *td2;
#ifdef KSE
struct ksegrp *kg2;
#endif
struct sigacts *newsigacts;
int error;
@ -471,7 +473,9 @@ fork1(td, flags, pages, procp)
* then copy the section that is copied directly from the parent.
*/
td2 = FIRST_THREAD_IN_PROC(p2);
#ifdef KSE
kg2 = FIRST_KSEGRP_IN_PROC(p2);
#endif
/* Allocate and switch to an alternate kstack if specified. */
if (pages != 0)
@ -484,15 +488,19 @@ fork1(td, flags, pages, procp)
__rangeof(struct proc, p_startzero, p_endzero));
bzero(&td2->td_startzero,
__rangeof(struct thread, td_startzero, td_endzero));
#ifdef KSE
bzero(&kg2->kg_startzero,
__rangeof(struct ksegrp, kg_startzero, kg_endzero));
#endif
bcopy(&p1->p_startcopy, &p2->p_startcopy,
__rangeof(struct proc, p_startcopy, p_endcopy));
bcopy(&td->td_startcopy, &td2->td_startcopy,
__rangeof(struct thread, td_startcopy, td_endcopy));
#ifdef KSE
bcopy(&td->td_ksegrp->kg_startcopy, &kg2->kg_startcopy,
__rangeof(struct ksegrp, kg_startcopy, kg_endcopy));
#endif
td2->td_sigstk = td->td_sigstk;
td2->td_sigmask = td->td_sigmask;
@ -514,7 +522,11 @@ fork1(td, flags, pages, procp)
mtx_unlock_spin(&sched_lock);
p2->p_ucred = crhold(td->td_ucred);
#ifdef KSE
td2->td_ucred = crhold(p2->p_ucred); /* XXXKSE */
#else
td2->td_ucred = crhold(p2->p_ucred);
#endif
#ifdef AUDIT
audit_proc_fork(p1, p2);
#endif

9
sys/kern/kern_idle.c
View File

@ -79,7 +79,11 @@ idle_setup(void *dummy)
td = FIRST_THREAD_IN_PROC(p);
TD_SET_CAN_RUN(td);
td->td_flags |= TDF_IDLETD;
#ifdef KSE
sched_class(td->td_ksegrp, PRI_IDLE);
#else
sched_class(td, PRI_IDLE);
#endif
sched_prio(td, PRI_MAX_IDLE);
mtx_unlock_spin(&sched_lock);
PROC_UNLOCK(p);
@ -118,7 +122,12 @@ idle_proc(void *dummy)
#ifdef SMP
idle_cpus_mask &= ~mycpu;
#endif
#ifdef KSE
mi_switch(SW_VOL, NULL);
#else
if ((td = choosethread()) != curthread)
mi_switch(SW_VOL, td);
#endif
#ifdef SMP
idle_cpus_mask |= mycpu;
#endif

4
sys/kern/kern_intr.c
View File

@ -296,7 +296,11 @@ ithread_create(const char *name)
panic("kthread_create() failed with %d", error);
td = FIRST_THREAD_IN_PROC(p); /* XXXKSE */
mtx_lock_spin(&sched_lock);
#ifdef KSE
td->td_ksegrp->kg_pri_class = PRI_ITHD;
#else
td->td_pri_class = PRI_ITHD;
#endif
TD_SET_IWAIT(td);
mtx_unlock_spin(&sched_lock);
td->td_pflags |= TDP_ITHREAD;

28
sys/kern/kern_kse.c
View File

@ -47,6 +47,7 @@ __FBSDID("$FreeBSD$");
#include <sys/ktr.h>
#include <vm/uma.h>
#ifdef KSE
/*
* KSEGRP related storage.
*/
@ -117,6 +118,7 @@ upcall_remove(struct thread *td)
td->td_upcall = NULL;
}
}
#endif
#ifndef _SYS_SYSPROTO_H_
struct kse_switchin_args {
@ -128,6 +130,7 @@ struct kse_switchin_args {
int
kse_switchin(struct thread *td, struct kse_switchin_args *uap)
{
#ifdef KSE
struct kse_thr_mailbox tmbx;
struct kse_upcall *ku;
int error;
@ -167,6 +170,9 @@ kse_switchin(struct thread *td, struct kse_switchin_args *uap)
PROC_UNLOCK(td->td_proc);
}
return ((error == 0) ? EJUSTRETURN : error);
#else /* !KSE */
return (EOPNOTSUPP);
#endif
}
/*
@ -179,6 +185,7 @@ struct kse_thr_interrupt_args {
int
kse_thr_interrupt(struct thread *td, struct kse_thr_interrupt_args *uap)
{
#ifdef KSE
struct kse_execve_args args;
struct image_args iargs;
struct proc *p;
@ -283,6 +290,9 @@ kse_thr_interrupt(struct thread *td, struct kse_thr_interrupt_args *uap)
return (EINVAL);
}
return (0);
#else /* !KSE */
return (EOPNOTSUPP);
#endif
}
/*
@ -293,6 +303,7 @@ struct kse_exit_args {
int
kse_exit(struct thread *td, struct kse_exit_args *uap)
{
#ifdef KSE
struct proc *p;
struct ksegrp *kg;
struct kse_upcall *ku, *ku2;
@ -379,6 +390,9 @@ kse_exit(struct thread *td, struct kse_exit_args *uap)
#else
exit1(td, 0);
#endif
#else /* !KSE */
return (EOPNOTSUPP);
#endif
}
/*
@ -393,6 +407,7 @@ struct kse_release_args {
int
kse_release(struct thread *td, struct kse_release_args *uap)
{
#ifdef KSE
struct proc *p;
struct ksegrp *kg;
struct kse_upcall *ku;
@ -454,6 +469,9 @@ kse_release(struct thread *td, struct kse_release_args *uap)
mtx_unlock_spin(&sched_lock);
}
return (0);
#else /* !KSE */
return (EOPNOTSUPP);
#endif
}
/* struct kse_wakeup_args {
@ -462,6 +480,7 @@ kse_release(struct thread *td, struct kse_release_args *uap)
int
kse_wakeup(struct thread *td, struct kse_wakeup_args *uap)
{
#ifdef KSE
struct proc *p;
struct ksegrp *kg;
struct kse_upcall *ku;
@ -517,6 +536,9 @@ kse_wakeup(struct thread *td, struct kse_wakeup_args *uap)
}
PROC_UNLOCK(p);
return (0);
#else /* !KSE */
return (EOPNOTSUPP);
#endif
}
/*
@ -534,6 +556,7 @@ kse_wakeup(struct thread *td, struct kse_wakeup_args *uap)
int
kse_create(struct thread *td, struct kse_create_args *uap)
{
#ifdef KSE
struct ksegrp *newkg;
struct ksegrp *kg;
struct proc *p;
@ -805,8 +828,12 @@ kse_create(struct thread *td, struct kse_create_args *uap)
mtx_unlock_spin(&sched_lock);
}
return (0);
#else /* !KSE */
return (EOPNOTSUPP);
#endif
}
#ifdef KSE
/*
* Initialize global thread allocation resources.
*/
@ -1479,3 +1506,4 @@ thread_continued(struct proc *p)
}
}
}
#endif

4
sys/kern/kern_poll.c
View File

@ -581,7 +581,11 @@ poll_idle(void)
rtp.prio = RTP_PRIO_MAX; /* lowest priority */
rtp.type = RTP_PRIO_IDLE;
mtx_lock_spin(&sched_lock);
#ifdef KSE
rtp_to_pri(&rtp, td->td_ksegrp);
#else
rtp_to_pri(&rtp, td);
#endif
mtx_unlock_spin(&sched_lock);
for (;;) {

27
sys/kern/kern_proc.c
View File

@ -141,7 +141,7 @@ proc_dtor(void *mem, int size, void *arg)
{
struct proc *p;
struct thread *td;
#ifdef INVARIANTS
#if defined(INVARIANTS) && defined(KSE)
struct ksegrp *kg;
#endif
@ -151,10 +151,14 @@ proc_dtor(void *mem, int size, void *arg)
#ifdef INVARIANTS
KASSERT((p->p_numthreads == 1),
("bad number of threads in exiting process"));
#ifdef KSE
KASSERT((p->p_numksegrps == 1), ("free proc with > 1 ksegrp"));
#endif
KASSERT((td != NULL), ("proc_dtor: bad thread pointer"));
#ifdef KSE
kg = FIRST_KSEGRP_IN_PROC(p);
KASSERT((kg != NULL), ("proc_dtor: bad kg pointer"));
#endif
KASSERT(STAILQ_EMPTY(&p->p_ktr), ("proc_dtor: non-empty p_ktr"));
#endif
@ -177,17 +181,25 @@ proc_init(void *mem, int size, int flags)
{
struct proc *p;
struct thread *td;
#ifdef KSE
struct ksegrp *kg;
#endif
p = (struct proc *)mem;
p->p_sched = (struct p_sched *)&p[1];
td = thread_alloc();
#ifdef KSE
kg = ksegrp_alloc();
#endif
bzero(&p->p_mtx, sizeof(struct mtx));
mtx_init(&p->p_mtx, "process lock", NULL, MTX_DEF | MTX_DUPOK);
p->p_stats = pstats_alloc();
#ifdef KSE
proc_linkup(p, kg, td);
sched_newproc(p, kg, td);
#else
proc_linkup(p, td);
#endif
return (0);
}
@ -203,7 +215,9 @@ proc_fini(void *mem, int size)
p = (struct proc *)mem;
pstats_free(p->p_stats);
#ifdef KSE
ksegrp_free(FIRST_KSEGRP_IN_PROC(p));
#endif
thread_free(FIRST_THREAD_IN_PROC(p));
mtx_destroy(&p->p_mtx);
if (p->p_ksi != NULL)
@ -768,7 +782,9 @@ fill_kinfo_proc_only(struct proc *p, struct kinfo_proc *kp)
static void
fill_kinfo_thread(struct thread *td, struct kinfo_proc *kp)
{
#ifdef KSE
struct ksegrp *kg;
#endif
struct proc *p;
p = td->td_proc;
@ -808,6 +824,7 @@ fill_kinfo_thread(struct thread *td, struct kinfo_proc *kp)
kp->ki_stat = SIDL;
}
#ifdef KSE
kg = td->td_ksegrp;
/* things in the KSE GROUP */
@ -815,7 +832,7 @@ fill_kinfo_thread(struct thread *td, struct kinfo_proc *kp)
kp->ki_slptime = kg->kg_slptime;
kp->ki_pri.pri_user = kg->kg_user_pri;
kp->ki_pri.pri_class = kg->kg_pri_class;
#endif
/* Things in the thread */
kp->ki_wchan = td->td_wchan;
kp->ki_pri.pri_level = td->td_priority;
@ -828,6 +845,12 @@ fill_kinfo_thread(struct thread *td, struct kinfo_proc *kp)
kp->ki_pcb = td->td_pcb;
kp->ki_kstack = (void *)td->td_kstack;
kp->ki_pctcpu = sched_pctcpu(td);
#ifndef KSE
kp->ki_estcpu = td->td_estcpu;
kp->ki_slptime = td->td_slptime;
kp->ki_pri.pri_class = td->td_pri_class;
kp->ki_pri.pri_user = td->td_user_pri;
#endif
/* We can't get this anymore but ps etc never used it anyway. */
kp->ki_rqindex = 0;

86
sys/kern/kern_resource.c
View File

@ -318,7 +318,11 @@ rtprio_thread(struct thread *td, struct rtprio_thread_args *uap)
else
td1 = thread_find(p, uap->lwpid);
if (td1 != NULL)
#ifdef KSE
pri_to_rtp(td1->td_ksegrp, &rtp);
#else
pri_to_rtp(td1, &rtp);
#endif
else
error = ESRCH;
mtx_unlock_spin(&sched_lock);
@ -354,7 +358,11 @@ rtprio_thread(struct thread *td, struct rtprio_thread_args *uap)
else
td1 = thread_find(p, uap->lwpid);
if (td1 != NULL)
#ifdef KSE
error = rtp_to_pri(&rtp, td1->td_ksegrp);
#else
error = rtp_to_pri(&rtp, td1);
#endif
else
error = ESRCH;
mtx_unlock_spin(&sched_lock);
@ -387,7 +395,11 @@ rtprio(td, uap)
{
struct proc *curp;
struct proc *p;
#ifdef KSE
struct ksegrp *kg;
#else
struct thread *tdp;
#endif
struct rtprio rtp;
int cierror, error;
@ -423,14 +435,23 @@ rtprio(td, uap)
* as leaving it zero.
*/
if (uap->pid == 0) {
#ifdef KSE
pri_to_rtp(td->td_ksegrp, &rtp);
#else
pri_to_rtp(td, &rtp);
#endif
} else {
struct rtprio rtp2;
rtp.type = RTP_PRIO_IDLE;
rtp.prio = RTP_PRIO_MAX;
#ifdef KSE
FOREACH_KSEGRP_IN_PROC(p, kg) {
pri_to_rtp(kg, &rtp2);
#else
FOREACH_THREAD_IN_PROC(p, tdp) {
pri_to_rtp(tdp, &rtp2);
#endif
if (rtp2.type < rtp.type ||
(rtp2.type == rtp.type &&
rtp2.prio < rtp.prio)) {
@ -471,20 +492,39 @@ rtprio(td, uap)
}
}
#ifdef KSE
/*
* If we are setting our own priority, set just our
* KSEGRP but if we are doing another process,
* do all the groups on that process. If we
* specify our own pid we do the latter.
*/
#else
/*
* If we are setting our own priority, set just our
* thread but if we are doing another process,
* do all the threads on that process. If we
* specify our own pid we do the latter.
*/
#endif
mtx_lock_spin(&sched_lock);
if (uap->pid == 0) {
#ifdef KSE
error = rtp_to_pri(&rtp, td->td_ksegrp);
#else
error = rtp_to_pri(&rtp, td);
#endif
} else {
#ifdef KSE
FOREACH_KSEGRP_IN_PROC(p, kg) {
if ((error = rtp_to_pri(&rtp, kg)) != 0) {
break;
}
#else
FOREACH_THREAD_IN_PROC(p, td) {
if ((error = rtp_to_pri(&rtp, td)) != 0)
break;
#endif
}
}
mtx_unlock_spin(&sched_lock);
@ -498,7 +538,11 @@ rtprio(td, uap)
}
int
#ifdef KSE
rtp_to_pri(struct rtprio *rtp, struct ksegrp *kg)
#else
rtp_to_pri(struct rtprio *rtp, struct thread *td)
#endif
{
mtx_assert(&sched_lock, MA_OWNED);
@ -506,43 +550,85 @@ rtp_to_pri(struct rtprio *rtp, struct ksegrp *kg)
return (EINVAL);
switch (RTP_PRIO_BASE(rtp->type)) {
case RTP_PRIO_REALTIME:
#ifdef KSE
kg->kg_user_pri = PRI_MIN_REALTIME + rtp->prio;
#else
td->td_user_pri = PRI_MIN_REALTIME + rtp->prio;
#endif
break;
case RTP_PRIO_NORMAL:
#ifdef KSE
kg->kg_user_pri = PRI_MIN_TIMESHARE + rtp->prio;
#else
td->td_user_pri = PRI_MIN_TIMESHARE + rtp->prio;
#endif
break;
case RTP_PRIO_IDLE:
#ifdef KSE
kg->kg_user_pri = PRI_MIN_IDLE + rtp->prio;
#else
td->td_user_pri = PRI_MIN_IDLE + rtp->prio;
#endif
break;
default:
return (EINVAL);
}
#ifdef KSE
sched_class(kg, rtp->type);
if (curthread->td_ksegrp == kg) {
sched_prio(curthread, kg->kg_user_pri); /* XXX dubious */
}
#else
sched_class(td, rtp->type); /* XXX fix */
if (curthread == td)
sched_prio(curthread, td->td_user_pri); /* XXX dubious */
#endif
return (0);
}
void
#ifdef KSE
pri_to_rtp(struct ksegrp *kg, struct rtprio *rtp)
#else
pri_to_rtp(struct thread *td, struct rtprio *rtp)
#endif
{
mtx_assert(&sched_lock, MA_OWNED);
#ifdef KSE
switch (PRI_BASE(kg->kg_pri_class)) {
#else
switch (PRI_BASE(td->td_pri_class)) {
#endif
case PRI_REALTIME:
#ifdef KSE
rtp->prio = kg->kg_user_pri - PRI_MIN_REALTIME;
#else
rtp->prio = td->td_user_pri - PRI_MIN_REALTIME;
#endif
break;
case PRI_TIMESHARE:
#ifdef KSE
rtp->prio = kg->kg_user_pri - PRI_MIN_TIMESHARE;
#else
rtp->prio = td->td_user_pri - PRI_MIN_TIMESHARE;
#endif
break;
case PRI_IDLE:
#ifdef KSE
rtp->prio = kg->kg_user_pri - PRI_MIN_IDLE;
#else
rtp->prio = td->td_user_pri - PRI_MIN_IDLE;
#endif
break;
default:
break;
}
#ifdef KSE
rtp->type = kg->kg_pri_class;
#else
rtp->type = td->td_pri_class;
#endif
}
#if defined(COMPAT_43)

60
sys/kern/kern_sig.c
View File

@ -94,7 +94,9 @@ static int filt_sigattach(struct knote *kn);
static void filt_sigdetach(struct knote *kn);
static int filt_signal(struct knote *kn, long hint);
static struct thread *sigtd(struct proc *p, int sig, int prop);
#ifdef KSE
static int do_tdsignal(struct proc *, struct thread *, int, ksiginfo_t *);
#endif
static void sigqueue_start(void);
static uma_zone_t ksiginfo_zone = NULL;
@ -565,7 +567,11 @@ void
signotify(struct thread *td)
{
struct proc *p;
#ifdef KSE
sigset_t set, saved;
#else
sigset_t set;
#endif
p = td->td_proc;
@ -576,8 +582,10 @@ signotify(struct thread *td)
* previously masked by all threads to our sigqueue.
*/
set = p->p_sigqueue.sq_signals;
#ifdef KSE
if (p->p_flag & P_SA)
saved = p->p_sigqueue.sq_signals;
#endif
SIGSETNAND(set, td->td_sigmask);
if (! SIGISEMPTY(set))
sigqueue_move_set(&p->p_sigqueue, &td->td_sigqueue, &set);
@ -586,6 +594,7 @@ signotify(struct thread *td)
td->td_flags |= TDF_NEEDSIGCHK | TDF_ASTPENDING;
mtx_unlock_spin(&sched_lock);
}
#ifdef KSE
if ((p->p_flag & P_SA) && !(p->p_flag & P_SIGEVENT)) {
if (!SIGSETEQ(saved, p->p_sigqueue.sq_signals)) {
/* pending set changed */
@ -593,6 +602,7 @@ signotify(struct thread *td)
wakeup(&p->p_siglist);
}
}
#endif
}
int
@ -744,11 +754,13 @@ kern_sigaction(td, sig, act, oact, flags)
if (ps->ps_sigact[_SIG_IDX(sig)] == SIG_IGN ||
(sigprop(sig) & SA_IGNORE &&
ps->ps_sigact[_SIG_IDX(sig)] == SIG_DFL)) {
#ifdef KSE
if ((p->p_flag & P_SA) &&
SIGISMEMBER(p->p_sigqueue.sq_signals, sig)) {
p->p_flag |= P_SIGEVENT;
wakeup(&p->p_siglist);
}
#endif
/* never to be seen again */
sigqueue_delete_proc(p, sig);
if (sig != SIGCONT)
@ -1206,10 +1218,12 @@ kern_sigtimedwait(struct thread *td, sigset_t waitset, ksiginfo_t *ksi,
continue;
if (!SIGISMEMBER(td->td_sigqueue.sq_signals, i)) {
if (SIGISMEMBER(p->p_sigqueue.sq_signals, i)) {
#ifdef KSE
if (p->p_flag & P_SA) {
p->p_flag |= P_SIGEVENT;
wakeup(&p->p_siglist);
}
#endif
sigqueue_move(&p->p_sigqueue,
&td->td_sigqueue, i);
} else
@ -1882,7 +1896,9 @@ trapsignal(struct thread *td, ksiginfo_t *ksi)
{
struct sigacts *ps;
struct proc *p;
#ifdef KSE
int error;
#endif
int sig;
int code;
@ -1891,6 +1907,7 @@ trapsignal(struct thread *td, ksiginfo_t *ksi)
code = ksi->ksi_code;
KASSERT(_SIG_VALID(sig), ("invalid signal"));
#ifdef KSE
if (td->td_pflags & TDP_SA) {
if (td->td_mailbox == NULL)
thread_user_enter(td);
@ -1908,6 +1925,9 @@ trapsignal(struct thread *td, ksiginfo_t *ksi)
} else {
PROC_LOCK(p);
}
#else
PROC_LOCK(p);
#endif
ps = p->p_sigacts;
mtx_lock(&ps->ps_mtx);
if ((p->p_flag & P_TRACED) == 0 && SIGISMEMBER(ps->ps_sigcatch, sig) &&
@ -1918,9 +1938,15 @@ trapsignal(struct thread *td, ksiginfo_t *ksi)
ktrpsig(sig, ps->ps_sigact[_SIG_IDX(sig)],
&td->td_sigmask, code);
#endif
#ifdef KSE
if (!(td->td_pflags & TDP_SA))
(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)],
ksi, &td->td_sigmask);
#else
(*p->p_sysent->sv_sendsig)(ps->ps_sigact[_SIG_IDX(sig)],
ksi, &td->td_sigmask);
#endif
#ifdef KSE
else if (td->td_mailbox == NULL) {
mtx_unlock(&ps->ps_mtx);
/* UTS caused a sync signal */
@ -1939,6 +1965,7 @@ trapsignal(struct thread *td, ksiginfo_t *ksi)
sigexit(td, SIGSEGV);
mtx_lock(&ps->ps_mtx);
}
#endif
SIGSETOR(td->td_sigmask, ps->ps_catchmask[_SIG_IDX(sig)]);
if (!SIGISMEMBER(ps->ps_signodefer, sig))
SIGADDSET(td->td_sigmask, sig);
@ -2052,6 +2079,7 @@ psignal_event(struct proc *p, struct sigevent *sigev, ksiginfo_t *ksi)
int
tdsignal(struct proc *p, struct thread *td, int sig, ksiginfo_t *ksi)
{
#ifdef KSE
sigset_t saved;
int ret;
@ -2071,6 +2099,7 @@ tdsignal(struct proc *p, struct thread *td, int sig, ksiginfo_t *ksi)
static int
do_tdsignal(struct proc *p, struct thread *td, int sig, ksiginfo_t *ksi)
{
#endif
sig_t action;
sigqueue_t *sigqueue;
int prop;
@ -2081,9 +2110,17 @@ do_tdsignal(struct proc *p, struct thread *td, int sig, ksiginfo_t *ksi)
PROC_LOCK_ASSERT(p, MA_OWNED);
if (!_SIG_VALID(sig))
#ifdef KSE
panic("do_tdsignal(): invalid signal");
#else
panic("tdsignal(): invalid signal");
#endif
#ifdef KSE
KASSERT(ksi == NULL || !KSI_ONQ(ksi), ("do_tdsignal: ksi on queue"));
#else
KASSERT(ksi == NULL || !KSI_ONQ(ksi), ("tdsignal: ksi on queue"));
#endif
/*
* IEEE Std 1003.1-2001: return success when killing a zombie.
@ -2240,6 +2277,7 @@ do_tdsignal(struct proc *p, struct thread *td, int sig, ksiginfo_t *ksi)
if (action == SIG_DFL) {
sigqueue_delete(sigqueue, sig);
} else if (action == SIG_CATCH) {
#ifdef KSE
/*
* The process wants to catch it so it needs
* to run at least one thread, but which one?
@ -2250,6 +2288,12 @@ do_tdsignal(struct proc *p, struct thread *td, int sig, ksiginfo_t *ksi)
* single thread is runnable asap.
* XXXKSE for now however, make them all run.
*/
#else
/*
* The process wants to catch it so it needs
* to run at least one thread, but which one?
*/
#endif
goto runfast;
}
/*
@ -2541,8 +2585,10 @@ issignal(td)
*/
if (SIGISMEMBER(ps->ps_sigignore, sig) && (traced == 0)) {
sigqueue_delete(&td->td_sigqueue, sig);
#ifdef KSE
if (td->td_pflags & TDP_SA)
SIGADDSET(td->td_sigmask, sig);
#endif
continue;
}
if (p->p_flag & P_TRACED && (p->p_flag & P_PPWAIT) == 0) {
@ -2553,9 +2599,11 @@ issignal(td)
newsig = ptracestop(td, sig);
mtx_lock(&ps->ps_mtx);
#ifdef KSE
if (td->td_pflags & TDP_SA)
SIGADDSET(td->td_sigmask, sig);
#endif
if (sig != newsig) {
ksiginfo_t ksi;
/*
@ -2579,8 +2627,10 @@ issignal(td)
* signal is being masked, look for other signals.
*/
SIGADDSET(td->td_sigqueue.sq_signals, sig);
#ifdef KSE
if (td->td_pflags & TDP_SA)
SIGDELSET(td->td_sigmask, sig);
#endif
if (SIGISMEMBER(td->td_sigmask, sig))
continue;
signotify(td);
@ -2743,7 +2793,11 @@ postsig(sig)
mtx_lock(&ps->ps_mtx);
}
#ifdef KSE
if (!(td->td_pflags & TDP_SA) && action == SIG_DFL) {
#else
if (action == SIG_DFL) {
#endif
/*
* Default action, where the default is to kill
* the process. (Other cases were ignored above.)
@ -2752,6 +2806,7 @@ postsig(sig)
sigexit(td, sig);
/* NOTREACHED */
} else {
#ifdef KSE
if (td->td_pflags & TDP_SA) {
if (sig == SIGKILL) {
mtx_unlock(&ps->ps_mtx);
@ -2759,6 +2814,7 @@ postsig(sig)
}
}
#endif
/*
* If we get here, the signal must be caught.
*/
@ -2801,10 +2857,14 @@ postsig(sig)
p->p_code = 0;
p->p_sig = 0;
}
#ifdef KSE
if (td->td_pflags & TDP_SA)
thread_signal_add(curthread, &ksi);
else
(*p->p_sysent->sv_sendsig)(action, &ksi, &returnmask);
#else
(*p->p_sysent->sv_sendsig)(action, &ksi, &returnmask);
#endif
}
}

4
sys/kern/kern_subr.c
View File

@ -430,7 +430,11 @@ uio_yield(void)
td = curthread;
mtx_lock_spin(&sched_lock);
DROP_GIANT();
#ifdef KSE
sched_prio(td, td->td_ksegrp->kg_user_pri); /* XXXKSE */
#else
sched_prio(td, td->td_user_pri);
#endif
mi_switch(SW_INVOL, NULL);
mtx_unlock_spin(&sched_lock);
PICKUP_GIANT();

68
sys/kern/kern_switch.c
View File

@ -24,6 +24,7 @@
* SUCH DAMAGE.
*/
#ifdef KSE
/***
Here is the logic..
@ -84,6 +85,7 @@ queued at the priorities they have inherrited from the M highest priority
threads for that KSEGROUP. If this situation changes, the KSEs are
reassigned to keep this true.
***/
#endif
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
@ -142,31 +144,48 @@ SYSCTL_INT(_kern_sched, OID_AUTO, preemption, CTLFLAG_RD,
/************************************************************************
* Functions that manipulate runnability from a thread perspective. *
************************************************************************/
#ifdef KSE
/*
* Select the KSE that will be run next. From that find the thread, and
* remove it from the KSEGRP's run queue. If there is thread clustering,
* this will be what does it.
*/
#else
/*
* Select the thread that will be run next.
*/
#endif
struct thread *
choosethread(void)
{
#ifdef KSE
struct kse *ke;
#endif
struct thread *td;
#ifdef KSE
struct ksegrp *kg;
#endif
#if defined(SMP) && (defined(__i386__) || defined(__amd64__))
if (smp_active == 0 && PCPU_GET(cpuid) != 0) {
/* Shutting down, run idlethread on AP's */
td = PCPU_GET(idlethread);
#ifdef KSE
ke = td->td_kse;
#endif
CTR1(KTR_RUNQ, "choosethread: td=%p (idle)", td);
#ifdef KSE
ke->ke_flags |= KEF_DIDRUN;
#else
td->td_kse->ke_flags |= KEF_DIDRUN;
#endif
TD_SET_RUNNING(td);
return (td);
}
#endif
retry:
#ifdef KSE
ke = sched_choose();
if (ke) {
td = ke->ke_thread;
@ -179,15 +198,25 @@ choosethread(void)
}
TAILQ_REMOVE(&kg->kg_runq, td, td_runq);
}
#else
td = sched_choose();
if (td) {
#endif
CTR2(KTR_RUNQ, "choosethread: td=%p pri=%d",
td, td->td_priority);
} else {
/* Simulate runq_choose() having returned the idle thread */
td = PCPU_GET(idlethread);
#ifdef KSE
ke = td->td_kse;
#endif
CTR1(KTR_RUNQ, "choosethread: td=%p (idle)", td);
}
#ifdef KSE
ke->ke_flags |= KEF_DIDRUN;
#else
td->td_kse->ke_flags |= KEF_DIDRUN;
#endif
/*
* If we are in panic, only allow system threads,
@ -204,6 +233,7 @@ choosethread(void)
return (td);
}
#ifdef KSE
/*
* Given a surplus system slot, try assign a new runnable thread to it.
* Called from:
@ -287,6 +317,7 @@ remrunqueue(struct thread *td)
}
}
#endif
#endif
/*
* Change the priority of a thread that is on the run queue.
@ -294,7 +325,9 @@ remrunqueue(struct thread *td)
void
adjustrunqueue( struct thread *td, int newpri)
{
#ifdef KSE
struct ksegrp *kg;
#endif
struct kse *ke;
mtx_assert(&sched_lock, MA_OWNED);
@ -302,6 +335,7 @@ adjustrunqueue( struct thread *td, int newpri)
ke = td->td_kse;
CTR1(KTR_RUNQ, "adjustrunqueue: td%p", td);
#ifdef KSE
/*
* If it is not a threaded process, take the shortcut.
*/
@ -338,8 +372,22 @@ adjustrunqueue( struct thread *td, int newpri)
TD_SET_CAN_RUN(td);
td->td_priority = newpri;
setrunqueue(td, SRQ_BORING);
#else
/* We only care about the kse in the run queue. */
td->td_priority = newpri;
#ifndef SCHED_CORE
if (ke->ke_rqindex != (newpri / RQ_PPQ))
#else
if (ke->ke_rqindex != newpri)
#endif
{
sched_rem(td);
sched_add(td, SRQ_BORING);
}
#endif
}
#ifdef KSE
/*
* This function is called when a thread is about to be put on a
* ksegrp run queue because it has been made runnable or its
@ -485,15 +533,21 @@ maybe_preempt_in_ksegrp(struct thread *td)
int limitcount;
#endif
void
setrunqueue(struct thread *td, int flags)
{
#ifdef KSE
struct ksegrp *kg;
struct thread *td2;
struct thread *tda;
CTR3(KTR_RUNQ, "setrunqueue: td:%p kg:%p pid:%d",
td, td->td_ksegrp, td->td_proc->p_pid);
#else
CTR2(KTR_RUNQ, "setrunqueue: td:%p pid:%d",
td, td->td_proc->p_pid);
#endif
CTR5(KTR_SCHED, "setrunqueue: %p(%s) prio %d by %p(%s)",
td, td->td_proc->p_comm, td->td_priority, curthread,
curthread->td_proc->p_comm);
@ -503,6 +557,7 @@ setrunqueue(struct thread *td, int flags)
KASSERT((TD_CAN_RUN(td) || TD_IS_RUNNING(td)),
("setrunqueue: bad thread state"));
TD_SET_RUNQ(td);
#ifdef KSE
kg = td->td_ksegrp;
if ((td->td_proc->p_flag & P_HADTHREADS) == 0) {
/*
@ -594,6 +649,9 @@ setrunqueue(struct thread *td, int flags)
if ((flags & SRQ_YIELDING) == 0)
maybe_preempt_in_ksegrp(td);
}
#else
sched_add(td, flags);
#endif
}
/*
@ -705,6 +763,7 @@ maybe_preempt(struct thread *td)
*/
MPASS(TD_ON_RUNQ(td));
MPASS(td->td_sched->ke_state != KES_ONRUNQ);
#ifdef KSE
if (td->td_proc->p_flag & P_HADTHREADS) {
/*
* If this is a threaded process we actually ARE on the
@ -721,6 +780,7 @@ maybe_preempt(struct thread *td)
TAILQ_REMOVE(&kg->kg_runq, td, td_runq);
}
#endif
TD_SET_RUNNING(td);
CTR3(KTR_PROC, "preempting to thread %p (pid %d, %s)\n", td,
td->td_proc->p_pid, td->td_proc->p_comm);
@ -926,7 +986,11 @@ runq_remove(struct runq *rq, struct kse *ke)
struct rqhead *rqh;
int pri;
#ifdef KSE
KASSERT(ke->ke_proc->p_sflag & PS_INMEM,
#else
KASSERT(ke->ke_thread->td_proc->p_sflag & PS_INMEM,
#endif
("runq_remove: process swapped out"));
pri = ke->ke_rqindex;
rqh = &rq->rq_queues[pri];
@ -944,6 +1008,7 @@ runq_remove(struct runq *rq, struct kse *ke)
#include <vm/uma.h>
extern struct mtx kse_zombie_lock;
#ifdef KSE
/*
* Allocate scheduler specific per-process resources.
* The thread and ksegrp have already been linked in.
@ -959,6 +1024,7 @@ sched_newproc(struct proc *p, struct ksegrp *kg, struct thread *td)
/* This can go in sched_fork */
sched_init_concurrency(kg);
}
#endif
/*
* thread is being either created or recycled.
@ -980,6 +1046,7 @@ sched_newthread(struct thread *td)
ke->ke_state = KES_THREAD;
}
#ifdef KSE
/*
* Set up an initial concurrency of 1
* and set the given thread (if given) to be using that
@ -1036,5 +1103,6 @@ sched_thread_exit(struct thread *td)
SLOT_RELEASE(td->td_ksegrp);
slot_fill(td->td_ksegrp);
}
#endif
#endif /* KERN_SWITCH_INCLUDE */

2
sys/kern/kern_synch.c
View File

@ -405,8 +405,10 @@ mi_switch(int flags, struct thread *newtd)
PCPU_SET(switchticks, ticks);
CTR4(KTR_PROC, "mi_switch: old thread %p (kse %p, pid %ld, %s)",
(void *)td, td->td_sched, (long)p->p_pid, p->p_comm);
#ifdef KSE
if ((flags & SW_VOL) && (td->td_proc->p_flag & P_SA))
newtd = thread_switchout(td, flags, newtd);
#endif
#if (KTR_COMPILE & KTR_SCHED) != 0
if (td == PCPU_GET(idlethread))
CTR3(KTR_SCHED, "mi_switch: %p(%s) prio %d idle",

24
sys/kern/kern_thr.c
View File

@ -142,14 +142,18 @@ create_thread(struct thread *td, mcontext_t *ctx,
{
stack_t stack;
struct thread *newtd;
#ifdef KSE
struct ksegrp *kg, *newkg;
#endif
struct proc *p;
long id;
int error;
error = 0;
p = td->td_proc;
#ifdef KSE
kg = td->td_ksegrp;
#endif
/* Have race condition but it is cheap. */
if (p->p_numthreads >= max_threads_per_proc)
@ -225,6 +229,7 @@ create_thread(struct thread *td, mcontext_t *ctx,
}
}
#ifdef KSE
newkg = ksegrp_alloc();
bzero(&newkg->kg_startzero,
__rangeof(struct ksegrp, kg_startzero, kg_endzero));
@ -238,7 +243,16 @@ create_thread(struct thread *td, mcontext_t *ctx,
ksegrp_link(newkg, p);
thread_link(newtd, newkg);
PROC_UNLOCK(p);
#else
PROC_LOCK(td->td_proc);
td->td_proc->p_flag |= P_HADTHREADS;
newtd->td_sigmask = td->td_sigmask;
mtx_lock_spin(&sched_lock);
thread_link(newtd, p);
PROC_UNLOCK(p);
#endif
#ifdef KSE
/* let the scheduler know about these things. */
sched_fork_ksegrp(td, newkg);
sched_fork_thread(td, newtd);
@ -249,6 +263,16 @@ create_thread(struct thread *td, mcontext_t *ctx,
sched_prio(newtd, newkg->kg_user_pri);
} /* ignore timesharing class */
}
#else
sched_fork(td, 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 */
}
#endif
TD_SET_CAN_RUN(newtd);
/* if ((flags & THR_SUSPENDED) == 0) */
setrunqueue(newtd, SRQ_BORING);

105
sys/kern/kern_thread.c
View File

@ -50,10 +50,16 @@ __FBSDID("$FreeBSD$");
#include <vm/vm_extern.h>
#include <vm/uma.h>
#ifdef KSE
/*
* KSEGRP related storage.
*/
static uma_zone_t ksegrp_zone;
#else
/*
* thread related storage.
*/
#endif
static uma_zone_t thread_zone;
/* DEBUG ONLY */
@ -74,13 +80,18 @@ int max_threads_hits;
SYSCTL_INT(_kern_threads, OID_AUTO, max_threads_hits, CTLFLAG_RD,
&max_threads_hits, 0, "");
#ifdef KSE
int virtual_cpu;
#endif
TAILQ_HEAD(, thread) zombie_threads = TAILQ_HEAD_INITIALIZER(zombie_threads);
#ifdef KSE
TAILQ_HEAD(, ksegrp) zombie_ksegrps = TAILQ_HEAD_INITIALIZER(zombie_ksegrps);
#endif
struct mtx kse_zombie_lock;
MTX_SYSINIT(kse_zombie_lock, &kse_zombie_lock, "kse zombie lock", MTX_SPIN);
#ifdef KSE
static int
sysctl_kse_virtual_cpu(SYSCTL_HANDLER_ARGS)
{
@ -105,6 +116,7 @@ sysctl_kse_virtual_cpu(SYSCTL_HANDLER_ARGS)
SYSCTL_PROC(_kern_threads, OID_AUTO, virtual_cpu, CTLTYPE_INT|CTLFLAG_RW,
0, sizeof(virtual_cpu), sysctl_kse_virtual_cpu, "I",
"debug virtual cpus");
#endif
struct mtx tid_lock;
static struct unrhdr *tid_unrhdr;
@ -216,6 +228,7 @@ thread_fini(void *mem, int size)
vm_thread_dispose(td);
}
#ifdef KSE
/*
* Initialize type-stable parts of a ksegrp (when newly created).
*/
@ -271,6 +284,7 @@ ksegrp_unlink(struct ksegrp *kg)
if (p->p_procscopegrp == kg)
p->p_procscopegrp = NULL;
}
#endif
/*
* For a newly created process,
@ -281,10 +295,16 @@ ksegrp_unlink(struct ksegrp *kg)
* proc_init()
*/
void
#ifdef KSE
proc_linkup(struct proc *p, struct ksegrp *kg, struct thread *td)
#else
proc_linkup(struct proc *p, struct thread *td)
#endif
{
#ifdef KSE
TAILQ_INIT(&p->p_ksegrps); /* all ksegrps in proc */
#endif
TAILQ_INIT(&p->p_threads); /* all threads in proc */
TAILQ_INIT(&p->p_suspended); /* Threads suspended */
sigqueue_init(&p->p_sigqueue, p);
@ -294,11 +314,17 @@ proc_linkup(struct proc *p, struct ksegrp *kg, struct thread *td)
p->p_ksi->ksi_flags = KSI_EXT | KSI_INS;
}
LIST_INIT(&p->p_mqnotifier);
#ifdef KSE
p->p_numksegrps = 0;
#endif
p->p_numthreads = 0;
#ifdef KSE
ksegrp_link(kg, p);
thread_link(td, kg);
#else
thread_link(td, p);
#endif
}
/*
@ -314,10 +340,12 @@ threadinit(void)
thread_zone = uma_zcreate("THREAD", sched_sizeof_thread(),
thread_ctor, thread_dtor, thread_init, thread_fini,
UMA_ALIGN_CACHE, 0);
#ifdef KSE
ksegrp_zone = uma_zcreate("KSEGRP", sched_sizeof_ksegrp(),
ksegrp_ctor, NULL, NULL, NULL,
UMA_ALIGN_CACHE, 0);
kseinit(); /* set up kse specific stuff e.g. upcall zone*/
#endif
}
/*
@ -331,6 +359,7 @@ thread_stash(struct thread *td)
mtx_unlock_spin(&kse_zombie_lock);
}
#ifdef KSE
/*
* Stash an embarasingly extra ksegrp into the zombie ksegrp queue.
*/
@ -341,6 +370,7 @@ ksegrp_stash(struct ksegrp *kg)
TAILQ_INSERT_HEAD(&zombie_ksegrps, kg, kg_ksegrp);
mtx_unlock_spin(&kse_zombie_lock);
}
#endif
/*
* Reap zombie kse resource.
@ -349,21 +379,31 @@ void
thread_reap(void)
{
struct thread *td_first, *td_next;
#ifdef KSE
struct ksegrp *kg_first, * kg_next;
#endif
/*
* Don't even bother to lock if none at this instant,
* we really don't care about the next instant..
*/
#ifdef KSE
if ((!TAILQ_EMPTY(&zombie_threads))
|| (!TAILQ_EMPTY(&zombie_ksegrps))) {
#else
if (!TAILQ_EMPTY(&zombie_threads)) {
#endif
mtx_lock_spin(&kse_zombie_lock);
td_first = TAILQ_FIRST(&zombie_threads);
#ifdef KSE
kg_first = TAILQ_FIRST(&zombie_ksegrps);
#endif
if (td_first)
TAILQ_INIT(&zombie_threads);
#ifdef KSE
if (kg_first)
TAILQ_INIT(&zombie_ksegrps);
#endif
mtx_unlock_spin(&kse_zombie_lock);
while (td_first) {
td_next = TAILQ_NEXT(td_first, td_runq);
@ -372,6 +412,7 @@ thread_reap(void)
thread_free(td_first);
td_first = td_next;
}
#ifdef KSE
while (kg_first) {
kg_next = TAILQ_NEXT(kg_first, kg_ksegrp);
ksegrp_free(kg_first);
@ -382,9 +423,11 @@ thread_reap(void)
* is there.
*/
kse_GC();
#endif
}
}
#ifdef KSE
/*
* Allocate a ksegrp.
*/
@ -393,6 +436,7 @@ ksegrp_alloc(void)
{
return (uma_zalloc(ksegrp_zone, M_WAITOK));
}
#endif
/*
* Allocate a thread.
@ -400,10 +444,12 @@ ksegrp_alloc(void)
struct thread *
thread_alloc(void)
{
thread_reap(); /* check if any zombies to get */
return (uma_zalloc(thread_zone, M_WAITOK));
}
#ifdef KSE
/*
* Deallocate a ksegrp.
*/
@ -412,6 +458,7 @@ ksegrp_free(struct ksegrp *td)
{
uma_zfree(ksegrp_zone, td);
}
#endif
/*
* Deallocate a thread.
@ -449,8 +496,10 @@ thread_free(struct thread *td)
* exit1()
* kse_exit()
* thr_exit()
* ifdef KSE
* thread_user_enter()
* thread_userret()
* endif
* thread_suspend_check()
*/
void
@ -459,17 +508,23 @@ thread_exit(void)
uint64_t new_switchtime;
struct thread *td;
struct proc *p;
#ifdef KSE
struct ksegrp *kg;
#endif
td = curthread;
#ifdef KSE
kg = td->td_ksegrp;
#endif
p = td->td_proc;
mtx_assert(&sched_lock, MA_OWNED);
mtx_assert(&Giant, MA_NOTOWNED);
PROC_LOCK_ASSERT(p, MA_OWNED);
KASSERT(p != NULL, ("thread exiting without a process"));
#ifdef KSE
KASSERT(kg != NULL, ("thread exiting without a kse group"));
#endif
CTR3(KTR_PROC, "thread_exit: thread %p (pid %ld, %s)", td,
(long)p->p_pid, p->p_comm);
KASSERT(TAILQ_EMPTY(&td->td_sigqueue.sq_list), ("signal pending"));
@ -478,6 +533,7 @@ thread_exit(void)
AUDIT_SYSCALL_EXIT(0, td);
#endif
#ifdef KSE
if (td->td_standin != NULL) {
/*
* Note that we don't need to free the cred here as it
@ -486,6 +542,7 @@ thread_exit(void)
thread_stash(td->td_standin);
td->td_standin = NULL;
}
#endif
umtx_thread_exit(td);
@ -496,6 +553,7 @@ thread_exit(void)
*/
cpu_thread_exit(td); /* XXXSMP */
#ifdef KSE
/*
* The thread is exiting. scheduler can release its stuff
* and collect stats etc.
@ -503,6 +561,7 @@ thread_exit(void)
* need scheduler stuff.
*/
sched_thread_exit(td);
#endif
/* Do the same timestamp bookkeeping that mi_switch() would do. */
new_switchtime = cpu_ticks();
@ -529,9 +588,13 @@ thread_exit(void)
if (p->p_flag & P_HADTHREADS) {
if (p->p_numthreads > 1) {
thread_unlink(td);
#ifdef KSE
/* XXX first arg not used in 4BSD or ULE */
sched_exit_thread(FIRST_THREAD_IN_PROC(p), td);
#else
sched_exit(p, td);
#endif
/*
* The test below is NOT true if we are the
@ -544,6 +607,7 @@ thread_exit(void)
}
}
#ifdef KSE
/*
* Because each upcall structure has an owner thread,
* owner thread exits only when process is in exiting
@ -582,8 +646,11 @@ thread_exit(void)
ksegrp_unlink(kg);
ksegrp_stash(kg);
}
#endif
PROC_UNLOCK(p);
#ifdef KSE
td->td_ksegrp = NULL;
#endif
PCPU_SET(deadthread, td);
} else {
/*
@ -593,8 +660,10 @@ thread_exit(void)
* exit1() - clears threading flags before coming here
* kse_exit() - treats last thread specially
* thr_exit() - treats last thread specially
* ifdef KSE
* thread_user_enter() - only if more exist
* thread_userret() - only if more exist
* endif
* thread_suspend_check() - only if more exist
*/
panic ("thread_exit: Last thread exiting on its own");
@ -625,8 +694,11 @@ thread_wait(struct proc *p)
mtx_assert(&Giant, MA_NOTOWNED);
KASSERT((p->p_numthreads == 1), ("Multiple threads in wait1()"));
#ifdef KSE
KASSERT((p->p_numksegrps == 1), ("Multiple ksegrps in wait1()"));
#endif
FOREACH_THREAD_IN_PROC(p, td) {
#ifdef KSE
if (td->td_standin != NULL) {
if (td->td_standin->td_ucred != NULL) {
crfree(td->td_standin->td_ucred);
@ -635,6 +707,7 @@ thread_wait(struct proc *p)
thread_free(td->td_standin);
td->td_standin = NULL;
}
#endif
cpu_thread_clean(td);
crfree(td->td_ucred);
}
@ -650,28 +723,46 @@ thread_wait(struct proc *p)
* The thread is linked as if running but no KSE assigned.
* Called from:
* proc_linkup()
* ifdef KSE
* thread_schedule_upcall()
* endif
* thr_create()
*/
void
#ifdef KSE
thread_link(struct thread *td, struct ksegrp *kg)
#else
thread_link(struct thread *td, struct proc *p)
#endif
{
#ifdef KSE
struct proc *p;
#endif
#ifdef KSE
p = kg->kg_proc;
#endif
td->td_state = TDS_INACTIVE;
td->td_proc = p;
#ifdef KSE
td->td_ksegrp = kg;
#endif
td->td_flags = 0;
#ifdef KSE
td->td_kflags = 0;
#endif
LIST_INIT(&td->td_contested);
sigqueue_init(&td->td_sigqueue, p);
callout_init(&td->td_slpcallout, CALLOUT_MPSAFE);
TAILQ_INSERT_HEAD(&p->p_threads, td, td_plist);
#ifdef KSE
TAILQ_INSERT_HEAD(&kg->kg_threads, td, td_kglist);
#endif
p->p_numthreads++;
#ifdef KSE
kg->kg_numthreads++;
#endif
}
/*
@ -686,6 +777,7 @@ thread_unthread(struct thread *td)
struct proc *p = td->td_proc;
KASSERT((p->p_numthreads == 1), ("Unthreading with >1 threads"));
#ifdef KSE
upcall_remove(td);
p->p_flag &= ~(P_SA|P_HADTHREADS);
td->td_mailbox = NULL;
@ -695,6 +787,9 @@ thread_unthread(struct thread *td)
td->td_standin = NULL;
}
sched_set_concurrency(td->td_ksegrp, 1);
#else
p->p_flag &= ~P_HADTHREADS;
#endif
}
/*
@ -705,15 +800,23 @@ void
thread_unlink(struct thread *td)
{
struct proc *p = td->td_proc;
#ifdef KSE
struct ksegrp *kg = td->td_ksegrp;
#endif
mtx_assert(&sched_lock, MA_OWNED);
TAILQ_REMOVE(&p->p_threads, td, td_plist);
p->p_numthreads--;
#ifdef KSE
TAILQ_REMOVE(&kg->kg_threads, td, td_kglist);
kg->kg_numthreads--;
#endif
/* could clear a few other things here */
#ifdef KSE
/* Must NOT clear links to proc and ksegrp! */
#else
/* Must NOT clear links to proc! */
#endif
}
/*
@ -1032,7 +1135,9 @@ thread_single_end(void)
p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT | P_SINGLE_BOUNDARY);
mtx_lock_spin(&sched_lock);
p->p_singlethread = NULL;
#ifdef KSE
p->p_procscopegrp = NULL;
#endif
/*
* If there are other threads they mey now run,
* unless of course there is a blanket 'stop order'

6
sys/kern/kern_umtx.c
View File

@ -166,9 +166,15 @@ struct umtxq_chain {
* if it is using 100%CPU, this is unfair to other processes.
*/
#ifdef KSE
#define UPRI(td) (((td)->td_ksegrp->kg_user_pri >= PRI_MIN_TIMESHARE &&\
(td)->td_ksegrp->kg_user_pri <= PRI_MAX_TIMESHARE) ?\
PRI_MAX_TIMESHARE : (td)->td_ksegrp->kg_user_pri)
#else
#define UPRI(td) (((td)->td_user_pri >= PRI_MIN_TIMESHARE &&\
(td)->td_user_pri <= PRI_MAX_TIMESHARE) ?\
PRI_MAX_TIMESHARE : (td)->td_user_pri)
#endif
#define GOLDEN_RATIO_PRIME 2654404609U
#define UMTX_CHAINS 128

18
sys/kern/ksched.c
View File

@ -106,7 +106,11 @@ getscheduler(struct ksched *ksched, struct thread *td, int *policy)
int e = 0;
mtx_lock_spin(&sched_lock);
#ifdef KSE
pri_to_rtp(td->td_ksegrp, &rtp);
#else
pri_to_rtp(td, &rtp);
#endif
mtx_unlock_spin(&sched_lock);
switch (rtp.type)
{
@ -153,7 +157,11 @@ ksched_getparam(struct ksched *ksched,
struct rtprio rtp;
mtx_lock_spin(&sched_lock);
#ifdef KSE
pri_to_rtp(td->td_ksegrp, &rtp);
#else
pri_to_rtp(td, &rtp);
#endif
mtx_unlock_spin(&sched_lock);
if (RTP_PRIO_IS_REALTIME(rtp.type))
param->sched_priority = rtpprio_to_p4prio(rtp.prio);
@ -174,7 +182,9 @@ ksched_setscheduler(struct ksched *ksched,
{
int e = 0;
struct rtprio rtp;
#ifdef KSE
struct ksegrp *kg = td->td_ksegrp;
#endif
switch(policy)
{
@ -189,6 +199,7 @@ ksched_setscheduler(struct ksched *ksched,
? RTP_PRIO_FIFO : RTP_PRIO_REALTIME;
mtx_lock_spin(&sched_lock);
#ifdef KSE
rtp_to_pri(&rtp, kg);
FOREACH_THREAD_IN_GROUP(kg, td) { /* XXXKSE */
if (TD_IS_RUNNING(td)) {
@ -199,6 +210,9 @@ ksched_setscheduler(struct ksched *ksched,
}
}
}
#else
rtp_to_pri(&rtp, td);
#endif
mtx_unlock_spin(&sched_lock);
}
else
@ -212,6 +226,7 @@ ksched_setscheduler(struct ksched *ksched,
rtp.type = RTP_PRIO_NORMAL;
rtp.prio = p4prio_to_rtpprio(param->sched_priority);
mtx_lock_spin(&sched_lock);
#ifdef KSE
rtp_to_pri(&rtp, kg);
/* XXX Simply revert to whatever we had for last
@ -230,6 +245,9 @@ ksched_setscheduler(struct ksched *ksched,
}
}
#else
rtp_to_pri(&rtp, td);
#endif
mtx_unlock_spin(&sched_lock);
}
break;

358
sys/kern/sched_4bsd.c
View File

@ -75,6 +75,7 @@ __FBSDID("$FreeBSD$");
#endif
#define NICE_WEIGHT 1 /* Priorities per nice level. */
#ifdef KSE
/*
* The schedulable entity that can be given a context to run.
* A process may have several of these. Probably one per processor
@ -82,6 +83,13 @@ __FBSDID("$FreeBSD$");
* with a KSEG that contains the priority and niceness
* for the group.
*/
#else
/*
* The schedulable entity that runs a context.
* A process may have several of these. Probably one per processor
* but posibly a few more.
*/
#endif
struct kse {
TAILQ_ENTRY(kse) ke_procq; /* (j/z) Run queue. */
struct thread *ke_thread; /* (*) Active associated thread. */
@ -95,8 +103,10 @@ struct kse {
struct runq *ke_runq; /* runq the kse is currently on */
};
#ifdef KSE
#define ke_proc ke_thread->td_proc
#define ke_ksegrp ke_thread->td_ksegrp
#endif
#define td_kse td_sched
@ -113,6 +123,7 @@ struct kse {
#define SKE_RUNQ_PCPU(ke) \
((ke)->ke_runq != 0 && (ke)->ke_runq != &runq)
#ifdef KSE
struct kg_sched {
struct thread *skg_last_assigned; /* (j) Last thread assigned to */
/* the system scheduler. */
@ -144,6 +155,7 @@ do { \
/* KASSERT((kg->kg_avail_opennings >= 0), \
("slots out of whack"));*/ \
} while (0)
#endif
/*
* KSE_CAN_MIGRATE macro returns true if the kse can migrate between
@ -153,7 +165,9 @@ do { \
((ke)->ke_thread->td_pinned == 0 && ((ke)->ke_flags & KEF_BOUND) == 0)
static struct kse kse0;
#ifdef KSE
static struct kg_sched kg_sched0;
#endif
static int sched_tdcnt; /* Total runnable threads in the system. */
static int sched_quantum; /* Roundrobin scheduling quantum in ticks. */
@ -161,8 +175,12 @@ static int sched_quantum; /* Roundrobin scheduling quantum in ticks. */
static struct callout roundrobin_callout;
#ifdef KSE
static void slot_fill(struct ksegrp *kg);
static struct kse *sched_choose(void); /* XXX Should be thread * */
#else
static struct thread *sched_choose(void);
#endif
static void setup_runqs(void);
static void roundrobin(void *arg);
@ -171,9 +189,15 @@ static void schedcpu_thread(void);
static void sched_priority(struct thread *td, u_char prio);
static void sched_setup(void *dummy);
static void maybe_resched(struct thread *td);
#ifdef KSE
static void updatepri(struct ksegrp *kg);
static void resetpriority(struct ksegrp *kg);
static void resetpriority_thread(struct thread *td, struct ksegrp *kg);
#else
static void updatepri(struct thread *td);
static void resetpriority(struct thread *td);
static void resetpriority_thread(struct thread *td);
#endif
#ifdef SMP
static int forward_wakeup(int cpunum);
#endif
@ -276,6 +300,7 @@ SYSCTL_INT(_kern_sched_ipiwakeup, OID_AUTO, htt2, CTLFLAG_RW,
"account for htt");
#endif
#ifdef KSE
static int sched_followon = 0;
SYSCTL_INT(_kern_sched, OID_AUTO, followon, CTLFLAG_RW,
&sched_followon, 0,
@ -290,6 +315,7 @@ static int sched_kgfollowons = 0;
SYSCTL_INT(_kern_sched, OID_AUTO, kgfollowons, CTLFLAG_RD,
&sched_kgfollowons, 0,
"number of followons done in a ksegrp");
#endif
static __inline void
sched_load_add(void)
@ -340,20 +366,40 @@ roundrobin(void *arg)
/*
* Constants for digital decay and forget:
* ifdef KSE
* 90% of (kg_estcpu) usage in 5 * loadav time
* else
* 90% of (td_estcpu) usage in 5 * loadav time
* endif
* 95% of (ke_pctcpu) usage in 60 seconds (load insensitive)
* Note that, as ps(1) mentions, this can let percentages
* total over 100% (I've seen 137.9% for 3 processes).
*
* ifdef KSE
* Note that schedclock() updates kg_estcpu and p_cpticks asynchronously.
* else
* Note that schedclock() updates td_estcpu and p_cpticks asynchronously.
* endif
*
* ifdef KSE
* We wish to decay away 90% of kg_estcpu in (5 * loadavg) seconds.
* else
* We wish to decay away 90% of td_estcpu in (5 * loadavg) seconds.
* endif
* That is, the system wants to compute a value of decay such
* that the following for loop:
* for (i = 0; i < (5 * loadavg); i++)
* ifdef KSE
* kg_estcpu *= decay;
* else
* td_estcpu *= decay;
* endif
* will compute
* ifdef KSE
* kg_estcpu *= 0.1;
* else
* td_estcpu *= 0.1;
* endif
* for all values of loadavg:
*
* Mathematically this loop can be expressed by saying:
@ -436,7 +482,9 @@ schedcpu(void)
struct thread *td;
struct proc *p;
struct kse *ke;
#ifdef KSE
struct ksegrp *kg;
#endif
int awake, realstathz;
realstathz = stathz ? stathz : hz;
@ -451,8 +499,13 @@ schedcpu(void)
* 16-bit int's (remember them?) overflow takes 45 days.
*/
p->p_swtime++;
#ifdef KSE
FOREACH_KSEGRP_IN_PROC(p, kg) {
#else
FOREACH_THREAD_IN_PROC(p, td) {
#endif
awake = 0;
#ifdef KSE
FOREACH_THREAD_IN_GROUP(kg, td) {
ke = td->td_kse;
/*
@ -502,12 +555,70 @@ schedcpu(void)
#endif
ke->ke_cpticks = 0;
} /* end of kse loop */
#else
ke = td->td_kse;
/*
* Increment sleep time (if sleeping). We
* ignore overflow, as above.
*/
/*
* The kse slptimes are not touched in wakeup
* because the thread may not HAVE a KSE.
*/
if (ke->ke_state == KES_ONRUNQ) {
awake = 1;
ke->ke_flags &= ~KEF_DIDRUN;
} else if ((ke->ke_state == KES_THREAD) &&
(TD_IS_RUNNING(td))) {
awake = 1;
/* Do not clear KEF_DIDRUN */
} else if (ke->ke_flags & KEF_DIDRUN) {
awake = 1;
ke->ke_flags &= ~KEF_DIDRUN;
}
/*
* ke_pctcpu is only for ps and ttyinfo().
* Do it per kse, and add them up at the end?
* XXXKSE
*/
ke->ke_pctcpu = (ke->ke_pctcpu * ccpu) >>
FSHIFT;
/*
* If the kse has been idle the entire second,
* stop recalculating its priority until
* it wakes up.
*/
if (ke->ke_cpticks == 0)
continue;
#if (FSHIFT >= CCPU_SHIFT)
ke->ke_pctcpu += (realstathz == 100)
? ((fixpt_t) ke->ke_cpticks) <<
(FSHIFT - CCPU_SHIFT) :
100 * (((fixpt_t) ke->ke_cpticks)
<< (FSHIFT - CCPU_SHIFT)) / realstathz;
#else
ke->ke_pctcpu += ((FSCALE - ccpu) *
(ke->ke_cpticks *
FSCALE / realstathz)) >> FSHIFT;
#endif
ke->ke_cpticks = 0;
#endif
/*
* ifdef KSE
* If there are ANY running threads in this KSEGRP,
* else
* If there are ANY running threads in this process,
* endif
* then don't count it as sleeping.
*/
if (awake) {
#ifdef KSE
if (kg->kg_slptime > 1) {
#else
if (td->td_slptime > 1) {
#endif
/*
* In an ideal world, this should not
* happen, because whoever woke us
@ -517,6 +628,7 @@ schedcpu(void)
* priority. Should KASSERT at some
* point when all the cases are fixed.
*/
#ifdef KSE
updatepri(kg);
}
kg->kg_slptime = 0;
@ -530,6 +642,19 @@ schedcpu(void)
resetpriority_thread(td, kg);
}
} /* end of ksegrp loop */
#else
updatepri(td);
}
td->td_slptime = 0;
} else
td->td_slptime++;
if (td->td_slptime > 1)
continue;
td->td_estcpu = decay_cpu(loadfac, td->td_estcpu);
resetpriority(td);
resetpriority_thread(td);
} /* end of thread loop */
#endif
mtx_unlock_spin(&sched_lock);
} /* end of process loop */
sx_sunlock(&allproc_lock);
@ -551,24 +676,48 @@ schedcpu_thread(void)
/*
* Recalculate the priority of a process after it has slept for a while.
* ifdef KSE
* For all load averages >= 1 and max kg_estcpu of 255, sleeping for at
* least six times the loadfactor will decay kg_estcpu to zero.
* else
* For all load averages >= 1 and max td_estcpu of 255, sleeping for at
* least six times the loadfactor will decay td_estcpu to zero.
* endif
*/
static void
#ifdef KSE
updatepri(struct ksegrp *kg)
#else
updatepri(struct thread *td)
#endif
{
register fixpt_t loadfac;
register unsigned int newcpu;
loadfac = loadfactor(averunnable.ldavg[0]);
#ifdef KSE
if (kg->kg_slptime > 5 * loadfac)
kg->kg_estcpu = 0;
#else
if (td->td_slptime > 5 * loadfac)
td->td_estcpu = 0;
#endif
else {
#ifdef KSE
newcpu = kg->kg_estcpu;
kg->kg_slptime--; /* was incremented in schedcpu() */
while (newcpu && --kg->kg_slptime)
#else
newcpu = td->td_estcpu;
td->td_slptime--; /* was incremented in schedcpu() */
while (newcpu && --td->td_slptime)
#endif
newcpu = decay_cpu(loadfac, newcpu);
#ifdef KSE
kg->kg_estcpu = newcpu;
#else
td->td_estcpu = newcpu;
#endif
}
}
@ -578,16 +727,30 @@ updatepri(struct ksegrp *kg)
* than that of the current process.
*/
static void
#ifdef KSE
resetpriority(struct ksegrp *kg)
#else
resetpriority(struct thread *td)
#endif
{
register unsigned int newpriority;
#ifdef KSE
if (kg->kg_pri_class == PRI_TIMESHARE) {
newpriority = PUSER + kg->kg_estcpu / INVERSE_ESTCPU_WEIGHT +
NICE_WEIGHT * (kg->kg_proc->p_nice - PRIO_MIN);
#else
if (td->td_pri_class == PRI_TIMESHARE) {
newpriority = PUSER + td->td_estcpu / INVERSE_ESTCPU_WEIGHT +
NICE_WEIGHT * (td->td_proc->p_nice - PRIO_MIN);
#endif
newpriority = min(max(newpriority, PRI_MIN_TIMESHARE),
PRI_MAX_TIMESHARE);
#ifdef KSE
sched_user_prio(kg, newpriority);
#else
sched_user_prio(td, newpriority);
#endif
}
}
@ -596,7 +759,11 @@ resetpriority(struct ksegrp *kg)
* priority changes.
*/
static void
#ifdef KSE
resetpriority_thread(struct thread *td, struct ksegrp *kg)
#else
resetpriority_thread(struct thread *td)
#endif
{
/* Only change threads with a time sharing user priority. */
@ -607,7 +774,11 @@ resetpriority_thread(struct thread *td, struct ksegrp *kg)
/* XXX the whole needresched thing is broken, but not silly. */
maybe_resched(td);
#ifdef KSE
sched_prio(td, kg->kg_user_pri);
#else
sched_prio(td, td->td_user_pri);
#endif
}
/* ARGSUSED */
@ -643,12 +814,16 @@ schedinit(void)
* Set up the scheduler specific parts of proc0.
*/
proc0.p_sched = NULL; /* XXX */
#ifdef KSE
ksegrp0.kg_sched = &kg_sched0;
#endif
thread0.td_sched = &kse0;
kse0.ke_thread = &thread0;
kse0.ke_state = KES_THREAD;
#ifdef KSE
kg_sched0.skg_concurrency = 1;
kg_sched0.skg_avail_opennings = 0; /* we are already running */
#endif
}
int
@ -672,8 +847,13 @@ sched_rr_interval(void)
/*
* We adjust the priority of the current process. The priority of
* a process gets worse as it accumulates CPU time. The cpu usage
* ifdef KSE
* estimator (kg_estcpu) is increased here. resetpriority() will
* compute a different priority each time kg_estcpu increases by
* else
* estimator (td_estcpu) is increased here. resetpriority() will
* compute a different priority each time td_estcpu increases by
* endif
* INVERSE_ESTCPU_WEIGHT
* (until MAXPRI is reached). The cpu usage estimator ramps up
* quite quickly when the process is running (linearly), and decays
@ -686,21 +866,33 @@ sched_rr_interval(void)
void
sched_clock(struct thread *td)
{
#ifdef KSE
struct ksegrp *kg;
#endif
struct kse *ke;
mtx_assert(&sched_lock, MA_OWNED);
#ifdef KSE
kg = td->td_ksegrp;
#endif
ke = td->td_kse;
ke->ke_cpticks++;
#ifdef KSE
kg->kg_estcpu = ESTCPULIM(kg->kg_estcpu + 1);
if ((kg->kg_estcpu % INVERSE_ESTCPU_WEIGHT) == 0) {
resetpriority(kg);
resetpriority_thread(td, kg);
#else
td->td_estcpu = ESTCPULIM(td->td_estcpu + 1);
if ((td->td_estcpu % INVERSE_ESTCPU_WEIGHT) == 0) {
resetpriority(td);
resetpriority_thread(td);
#endif
}
}
#ifdef KSE
/*
* charge childs scheduling cpu usage to parent.
*
@ -709,13 +901,30 @@ sched_clock(struct thread *td)
* all ksegrps, this is strictly as expected. Assume that the child process
* aggregated all the estcpu into the 'built-in' ksegrp.
*/
#else
/*
* charge childs scheduling cpu usage to parent.
*/
#endif
void
sched_exit(struct proc *p, struct thread *td)
{
#ifdef KSE
sched_exit_ksegrp(FIRST_KSEGRP_IN_PROC(p), td);
sched_exit_thread(FIRST_THREAD_IN_PROC(p), td);
#else
struct thread *parent = FIRST_THREAD_IN_PROC(p);
CTR3(KTR_SCHED, "sched_exit: %p(%s) prio %d",
td, td->td_proc->p_comm, td->td_priority);
parent->td_estcpu = ESTCPULIM(parent->td_estcpu + td->td_estcpu);
if ((td->td_proc->p_flag & P_NOLOAD) == 0)
sched_load_rem();
#endif
}
#ifdef KSE
void
sched_exit_ksegrp(struct ksegrp *kg, struct thread *childtd)
{
@ -732,14 +941,21 @@ sched_exit_thread(struct thread *td, struct thread *child)
if ((child->td_proc->p_flag & P_NOLOAD) == 0)
sched_load_rem();
}
#endif
void
sched_fork(struct thread *td, struct thread *childtd)
{
#ifdef KSE
sched_fork_ksegrp(td, childtd->td_ksegrp);
sched_fork_thread(td, childtd);
#else
childtd->td_estcpu = td->td_estcpu;
sched_newthread(childtd);
#endif
}
#ifdef KSE
void
sched_fork_ksegrp(struct thread *td, struct ksegrp *child)
{
@ -752,37 +968,61 @@ sched_fork_thread(struct thread *td, struct thread *childtd)
{
sched_newthread(childtd);
}
#endif
void
sched_nice(struct proc *p, int nice)
{
#ifdef KSE
struct ksegrp *kg;
#endif
struct thread *td;
PROC_LOCK_ASSERT(p, MA_OWNED);
mtx_assert(&sched_lock, MA_OWNED);
p->p_nice = nice;
#ifdef KSE
FOREACH_KSEGRP_IN_PROC(p, kg) {
resetpriority(kg);
FOREACH_THREAD_IN_GROUP(kg, td) {
resetpriority_thread(td, kg);
}
}
#else
FOREACH_THREAD_IN_PROC(p, td) {
resetpriority(td);
resetpriority_thread(td);
}
#endif
}
void
#ifdef KSE
sched_class(struct ksegrp *kg, int class)
#else
sched_class(struct thread *td, int class)
#endif
{
mtx_assert(&sched_lock, MA_OWNED);
#ifdef KSE
kg->kg_pri_class = class;
#else
td->td_pri_class = class;
#endif
}
#ifdef KSE
/*
* Adjust the priority of a thread.
* This may include moving the thread within the KSEGRP,
* changing the assignment of a kse to the thread,
* and moving a KSE in the system run queue.
*/
#else
/*
* Adjust the priority of a thread.
*/
#endif
static void
sched_priority(struct thread *td, u_char prio)
{
@ -827,7 +1067,11 @@ sched_unlend_prio(struct thread *td, u_char prio)
if (td->td_base_pri >= PRI_MIN_TIMESHARE &&
td->td_base_pri <= PRI_MAX_TIMESHARE)
#ifdef KSE
base_pri = td->td_ksegrp->kg_user_pri;
#else
base_pri = td->td_user_pri;
#endif
else
base_pri = td->td_base_pri;
if (prio >= base_pri) {
@ -865,11 +1109,18 @@ sched_prio(struct thread *td, u_char prio)
}
void
#ifdef KSE
sched_user_prio(struct ksegrp *kg, u_char prio)
#else
sched_user_prio(struct thread *td, u_char prio)
#endif
{
#ifdef KSE
struct thread *td;
#endif
u_char oldprio;
#ifdef KSE
kg->kg_base_user_pri = prio;
/* XXXKSE only for 1:1 */
@ -885,6 +1136,12 @@ sched_user_prio(struct ksegrp *kg, u_char prio)
oldprio = kg->kg_user_pri;
kg->kg_user_pri = prio;
#else
td->td_base_user_pri = prio;
oldprio = td->td_user_pri;
td->td_user_pri = prio;
#endif
if (TD_ON_UPILOCK(td) && oldprio != prio)
umtx_pi_adjust(td, oldprio);
@ -897,8 +1154,13 @@ sched_lend_user_prio(struct thread *td, u_char prio)
td->td_flags |= TDF_UBORROWING;
#ifdef KSE
oldprio = td->td_ksegrp->kg_user_pri;
td->td_ksegrp->kg_user_pri = prio;
#else
oldprio = td->td_user_pri;
td->td_user_pri = prio;
#endif
if (TD_ON_UPILOCK(td) && oldprio != prio)
umtx_pi_adjust(td, oldprio);
@ -907,13 +1169,23 @@ sched_lend_user_prio(struct thread *td, u_char prio)
void
sched_unlend_user_prio(struct thread *td, u_char prio)
{
#ifdef KSE
struct ksegrp *kg = td->td_ksegrp;
#endif
u_char base_pri;
#ifdef KSE
base_pri = kg->kg_base_user_pri;
#else
base_pri = td->td_base_user_pri;
#endif
if (prio >= base_pri) {
td->td_flags &= ~TDF_UBORROWING;
#ifdef KSE
sched_user_prio(kg, base_pri);
#else
sched_user_prio(td, base_pri);
#endif
} else
sched_lend_user_prio(td, prio);
}
@ -923,16 +1195,24 @@ sched_sleep(struct thread *td)
{
mtx_assert(&sched_lock, MA_OWNED);
#ifdef KSE
td->td_ksegrp->kg_slptime = 0;
#else
td->td_slptime = 0;
#endif
}
#ifdef KSE
static void remrunqueue(struct thread *td);
#endif
void
sched_switch(struct thread *td, struct thread *newtd, int flags)
{
struct kse *ke;
#ifdef KSE
struct ksegrp *kg;
#endif
struct proc *p;
ke = td->td_kse;
@ -942,6 +1222,7 @@ sched_switch(struct thread *td, struct thread *newtd, int flags)
if ((p->p_flag & P_NOLOAD) == 0)
sched_load_rem();
#ifdef KSE
/*
* We are volunteering to switch out so we get to nominate
* a successor for the rest of our quantum
@ -967,6 +1248,7 @@ sched_switch(struct thread *td, struct thread *newtd, int flags)
}
}
}
#endif
if (newtd)
newtd->td_flags |= (td->td_flags & TDF_NEEDRESCHED);
@ -984,12 +1266,15 @@ sched_switch(struct thread *td, struct thread *newtd, int flags)
if (td == PCPU_GET(idlethread))
TD_SET_CAN_RUN(td);
else {
#ifdef KSE
SLOT_RELEASE(td->td_ksegrp);
#endif
if (TD_IS_RUNNING(td)) {
/* Put us back on the run queue (kse and all). */
setrunqueue(td, (flags & SW_PREEMPT) ?
SRQ_OURSELF|SRQ_YIELDING|SRQ_PREEMPTED :
SRQ_OURSELF|SRQ_YIELDING);
#ifdef KSE
} else if (p->p_flag & P_HADTHREADS) {
/*
* We will not be on the run queue. So we must be
@ -999,6 +1284,7 @@ sched_switch(struct thread *td, struct thread *newtd, int flags)
*/
if (newtd == NULL || newtd->td_ksegrp != td->td_ksegrp)
slot_fill(td->td_ksegrp);
#endif
}
}
if (newtd) {
@ -1007,12 +1293,16 @@ sched_switch(struct thread *td, struct thread *newtd, int flags)
* as if it had been added to the run queue and selected.
* It came from:
* * A preemption
* ifdef KSE
* * An upcall
* endif
* * A followon
*/
KASSERT((newtd->td_inhibitors == 0),
("trying to run inhibitted thread"));
#ifdef KSE
SLOT_USE(newtd->td_ksegrp);
#endif
newtd->td_kse->ke_flags |= KEF_DIDRUN;
TD_SET_RUNNING(newtd);
if ((newtd->td_proc->p_flag & P_NOLOAD) == 0)
@ -1026,6 +1316,7 @@ sched_switch(struct thread *td, struct thread *newtd, int flags)
if (PMC_PROC_IS_USING_PMCS(td->td_proc))
PMC_SWITCH_CONTEXT(td, PMC_FN_CSW_OUT);
#endif
cpu_switch(td, newtd);
#ifdef HWPMC_HOOKS
if (PMC_PROC_IS_USING_PMCS(td->td_proc))
@ -1040,15 +1331,25 @@ sched_switch(struct thread *td, struct thread *newtd, int flags)
void
sched_wakeup(struct thread *td)
{
#ifdef KSE
struct ksegrp *kg;
#endif
mtx_assert(&sched_lock, MA_OWNED);
#ifdef KSE
kg = td->td_ksegrp;
if (kg->kg_slptime > 1) {
updatepri(kg);
resetpriority(kg);
}
kg->kg_slptime = 0;
#else
if (td->td_slptime > 1) {
updatepri(td);
resetpriority(td);
}
td->td_slptime = 0;
#endif
setrunqueue(td, SRQ_BORING);
}
@ -1188,8 +1489,13 @@ sched_add(struct thread *td, int flags)
mtx_assert(&sched_lock, MA_OWNED);
KASSERT(ke->ke_state != KES_ONRUNQ,
("sched_add: kse %p (%s) already in run queue", ke,
#ifdef KSE
ke->ke_proc->p_comm));
KASSERT(ke->ke_proc->p_sflag & PS_INMEM,
#else
td->td_proc->p_comm));
KASSERT(td->td_proc->p_sflag & PS_INMEM,
#endif
("sched_add: process swapped out"));
CTR5(KTR_SCHED, "sched_add: %p(%s) prio %d by %p(%s)",
td, td->td_proc->p_comm, td->td_priority, curthread,
@ -1239,7 +1545,9 @@ sched_add(struct thread *td, int flags)
if ((td->td_proc->p_flag & P_NOLOAD) == 0)
sched_load_add();
#ifdef KSE
SLOT_USE(td->td_ksegrp);
#endif
runq_add(ke->ke_runq, ke, flags);
ke->ke_state = KES_ONRUNQ;
}
@ -1250,8 +1558,13 @@ sched_add(struct thread *td, int flags)
mtx_assert(&sched_lock, MA_OWNED);
KASSERT(ke->ke_state != KES_ONRUNQ,
("sched_add: kse %p (%s) already in run queue", ke,
#ifdef KSE
ke->ke_proc->p_comm));
KASSERT(ke->ke_proc->p_sflag & PS_INMEM,
#else
td->td_proc->p_comm));
KASSERT(td->td_proc->p_sflag & PS_INMEM,
#endif
("sched_add: process swapped out"));
CTR5(KTR_SCHED, "sched_add: %p(%s) prio %d by %p(%s)",
td, td->td_proc->p_comm, td->td_priority, curthread,
@ -1276,7 +1589,9 @@ sched_add(struct thread *td, int flags)
}
if ((td->td_proc->p_flag & P_NOLOAD) == 0)
sched_load_add();
#ifdef KSE
SLOT_USE(td->td_ksegrp);
#endif
runq_add(ke->ke_runq, ke, flags);
ke->ke_state = KES_ONRUNQ;
maybe_resched(td);
@ -1289,7 +1604,11 @@ sched_rem(struct thread *td)
struct kse *ke;
ke = td->td_kse;
#ifdef KSE
KASSERT(ke->ke_proc->p_sflag & PS_INMEM,
#else
KASSERT(td->td_proc->p_sflag & PS_INMEM,
#endif
("sched_rem: process swapped out"));
KASSERT((ke->ke_state == KES_ONRUNQ),
("sched_rem: KSE not on run queue"));
@ -1300,7 +1619,9 @@ sched_rem(struct thread *td)
if ((td->td_proc->p_flag & P_NOLOAD) == 0)
sched_load_rem();
#ifdef KSE
SLOT_RELEASE(td->td_ksegrp);
#endif
runq_remove(ke->ke_runq, ke);
ke->ke_state = KES_THREAD;
@ -1310,7 +1631,11 @@ sched_rem(struct thread *td)
* Select threads to run.
* Notice that the running threads still consume a slot.
*/
#ifdef KSE
struct kse *
#else
struct thread *
#endif
sched_choose(void)
{
struct kse *ke;
@ -1339,20 +1664,36 @@ sched_choose(void)
ke = runq_choose(&runq);
#endif
#ifdef KSE
if (ke != NULL) {
#else
if (ke) {
#endif
runq_remove(rq, ke);
ke->ke_state = KES_THREAD;
#ifdef KSE
KASSERT(ke->ke_proc->p_sflag & PS_INMEM,
("sched_choose: process swapped out"));
#else
KASSERT(ke->ke_thread->td_proc->p_sflag & PS_INMEM,
("sched_choose: process swapped out"));
return (ke->ke_thread);
#endif
}
#ifdef KSE
return (ke);
#else
return (NULL);
#endif
}
void
sched_userret(struct thread *td)
{
#ifdef KSE
struct ksegrp *kg;
#endif
/*
* XXX we cheat slightly on the locking here to avoid locking in
* the usual case. Setting td_priority here is essentially an
@ -1364,6 +1705,7 @@ sched_userret(struct thread *td)
*/
KASSERT((td->td_flags & TDF_BORROWING) == 0,
("thread with borrowed priority returning to userland"));
#ifdef KSE
kg = td->td_ksegrp;
if (td->td_priority != kg->kg_user_pri) {
mtx_lock_spin(&sched_lock);
@ -1371,6 +1713,14 @@ sched_userret(struct thread *td)
td->td_base_pri = kg->kg_user_pri;
mtx_unlock_spin(&sched_lock);
}
#else
if (td->td_priority != td->td_user_pri) {
mtx_lock_spin(&sched_lock);
td->td_priority = td->td_user_pri;
td->td_base_pri = td->td_user_pri;
mtx_unlock_spin(&sched_lock);
}
#endif
}
void
@ -1413,11 +1763,17 @@ sched_is_bound(struct thread *td)
void
sched_relinquish(struct thread *td)
{
#ifdef KSE
struct ksegrp *kg;
kg = td->td_ksegrp;
#endif
mtx_lock_spin(&sched_lock);
#ifdef KSE
if (kg->kg_pri_class == PRI_TIMESHARE)
#else
if (td->td_pri_class == PRI_TIMESHARE)
#endif
sched_prio(td, PRI_MAX_TIMESHARE);
mi_switch(SW_VOL, NULL);
mtx_unlock_spin(&sched_lock);
@ -1429,11 +1785,13 @@ sched_load(void)
return (sched_tdcnt);
}
#ifdef KSE
int
sched_sizeof_ksegrp(void)
{
return (sizeof(struct ksegrp) + sizeof(struct kg_sched));
}
#endif
int
sched_sizeof_proc(void)

413
sys/kern/sched_ule.c
View File

@ -95,7 +95,7 @@ int tickincr = 1 << 10;
* The schedulable entity that can be given a context to run. A process may
* have several of these.
*/
struct kse {
struct td_sched { /* really kse */
TAILQ_ENTRY(kse) ke_procq; /* (j/z) Run queue. */
int ke_flags; /* (j) KEF_* flags. */
struct thread *ke_thread; /* (*) Active associated thread. */
@ -114,11 +114,11 @@ struct kse {
int ke_ftick; /* First tick that we were running on */
int ke_ticks; /* Tick count */
/* originally from kg_sched */
int skg_slptime; /* Number of ticks we vol. slept */
int skg_runtime; /* Number of ticks we were running */
};
#define td_kse td_sched
#define td_slptime td_kse->ke_slptime
#define ke_proc ke_thread->td_proc
#define ke_ksegrp ke_thread->td_ksegrp
#define ke_assign ke_procq.tqe_next
/* flags kept in ke_flags */
#define KEF_ASSIGNED 0x0001 /* Thread is being migrated. */
@ -131,25 +131,7 @@ struct kse {
#define KEF_DIDRUN 0x02000 /* Thread actually ran. */
#define KEF_EXIT 0x04000 /* Thread is being killed. */
struct kg_sched {
struct thread *skg_last_assigned; /* (j) Last thread assigned to */
/* the system scheduler */
int skg_slptime; /* Number of ticks we vol. slept */
int skg_runtime; /* Number of ticks we were running */
int skg_avail_opennings; /* (j) Num unfilled slots in group.*/
int skg_concurrency; /* (j) Num threads requested in group.*/
};
#define kg_last_assigned kg_sched->skg_last_assigned
#define kg_avail_opennings kg_sched->skg_avail_opennings
#define kg_concurrency kg_sched->skg_concurrency
#define kg_runtime kg_sched->skg_runtime
#define kg_slptime kg_sched->skg_slptime
#define SLOT_RELEASE(kg) (kg)->kg_avail_opennings++
#define SLOT_USE(kg) (kg)->kg_avail_opennings--
static struct kse kse0;
static struct kg_sched kg_sched0;
/*
* The priority is primarily determined by the interactivity score. Thus, we
@ -207,11 +189,11 @@ static struct kg_sched kg_sched0;
* This macro determines whether or not the thread belongs on the current or
* next run queue.
*/
#define SCHED_INTERACTIVE(kg) \
(sched_interact_score(kg) < SCHED_INTERACT_THRESH)
#define SCHED_CURR(kg, ke) \
#define SCHED_INTERACTIVE(td) \
(sched_interact_score(td) < SCHED_INTERACT_THRESH)
#define SCHED_CURR(td, ke) \
((ke->ke_thread->td_flags & TDF_BORROWING) || \
(ke->ke_flags & KEF_PREEMPTED) || SCHED_INTERACTIVE(kg))
(ke->ke_flags & KEF_PREEMPTED) || SCHED_INTERACTIVE(td))
/*
* Cpu percentage computation macros and defines.
@ -288,14 +270,13 @@ static struct kseq kseq_cpu;
#define KSEQ_CPU(x) (&kseq_cpu)
#endif
static void slot_fill(struct ksegrp *);
static struct kse *sched_choose(void); /* XXX Should be thread * */
static void sched_slice(struct kse *);
static void sched_priority(struct ksegrp *);
static void sched_priority(struct thread *);
static void sched_thread_priority(struct thread *, u_char);
static int sched_interact_score(struct ksegrp *);
static void sched_interact_update(struct ksegrp *);
static void sched_interact_fork(struct ksegrp *);
static int sched_interact_score(struct thread *);
static void sched_interact_update(struct thread *);
static void sched_interact_fork(struct thread *);
static void sched_pctcpu_update(struct kse *);
/* Operations on per processor queues */
@ -379,19 +360,19 @@ kseq_load_add(struct kseq *kseq, struct kse *ke)
{
int class;
mtx_assert(&sched_lock, MA_OWNED);
class = PRI_BASE(ke->ke_ksegrp->kg_pri_class);
class = PRI_BASE(ke->ke_thread->td_pri_class);
if (class == PRI_TIMESHARE)
kseq->ksq_load_timeshare++;
kseq->ksq_load++;
CTR1(KTR_SCHED, "load: %d", kseq->ksq_load);
if (class != PRI_ITHD && (ke->ke_proc->p_flag & P_NOLOAD) == 0)
if (class != PRI_ITHD && (ke->ke_thread->td_proc->p_flag & P_NOLOAD) == 0)
#ifdef SMP
kseq->ksq_group->ksg_load++;
#else
kseq->ksq_sysload++;
#endif
if (ke->ke_ksegrp->kg_pri_class == PRI_TIMESHARE)
kseq_nice_add(kseq, ke->ke_proc->p_nice);
if (ke->ke_thread->td_pri_class == PRI_TIMESHARE)
kseq_nice_add(kseq, ke->ke_thread->td_proc->p_nice);
}
static void
@ -399,10 +380,10 @@ kseq_load_rem(struct kseq *kseq, struct kse *ke)
{
int class;
mtx_assert(&sched_lock, MA_OWNED);
class = PRI_BASE(ke->ke_ksegrp->kg_pri_class);
class = PRI_BASE(ke->ke_thread->td_pri_class);
if (class == PRI_TIMESHARE)
kseq->ksq_load_timeshare--;
if (class != PRI_ITHD && (ke->ke_proc->p_flag & P_NOLOAD) == 0)
if (class != PRI_ITHD && (ke->ke_thread->td_proc->p_flag & P_NOLOAD) == 0)
#ifdef SMP
kseq->ksq_group->ksg_load--;
#else
@ -411,8 +392,8 @@ kseq_load_rem(struct kseq *kseq, struct kse *ke)
kseq->ksq_load--;
CTR1(KTR_SCHED, "load: %d", kseq->ksq_load);
ke->ke_runq = NULL;
if (ke->ke_ksegrp->kg_pri_class == PRI_TIMESHARE)
kseq_nice_rem(kseq, ke->ke_proc->p_nice);
if (ke->ke_thread->td_pri_class == PRI_TIMESHARE)
kseq_nice_rem(kseq, ke->ke_thread->td_proc->p_nice);
}
static void
@ -686,7 +667,7 @@ kseq_notify(struct kse *ke, int cpu)
kseq = KSEQ_CPU(cpu);
/* XXX */
class = PRI_BASE(ke->ke_ksegrp->kg_pri_class);
class = PRI_BASE(ke->ke_thread->td_pri_class);
if ((class == PRI_TIMESHARE || class == PRI_REALTIME) &&
(kseq_idle & kseq->ksq_group->ksg_mask))
atomic_clear_int(&kseq_idle, kseq->ksq_group->ksg_mask);
@ -889,10 +870,10 @@ kseq_choose(struct kseq *kseq)
* TIMESHARE kse group and its nice was too far out
* of the range that receives slices.
*/
nice = ke->ke_proc->p_nice + (0 - kseq->ksq_nicemin);
nice = ke->ke_thread->td_proc->p_nice + (0 - kseq->ksq_nicemin);
#if 0
if (ke->ke_slice == 0 || (nice > SCHED_SLICE_NTHRESH &&
ke->ke_proc->p_nice != 0)) {
ke->ke_thread->td_proc->p_nice != 0)) {
runq_remove(ke->ke_runq, ke);
sched_slice(ke);
ke->ke_runq = kseq->ksq_next;
@ -1045,41 +1026,45 @@ sched_initticks(void *dummy)
* process.
*/
static void
sched_priority(struct ksegrp *kg)
sched_priority(struct thread *td)
{
int pri;
if (kg->kg_pri_class != PRI_TIMESHARE)
if (td->td_pri_class != PRI_TIMESHARE)
return;
pri = SCHED_PRI_INTERACT(sched_interact_score(kg));
pri = SCHED_PRI_INTERACT(sched_interact_score(td));
pri += SCHED_PRI_BASE;
pri += kg->kg_proc->p_nice;
pri += td->td_proc->p_nice;
if (pri > PRI_MAX_TIMESHARE)
pri = PRI_MAX_TIMESHARE;
else if (pri < PRI_MIN_TIMESHARE)
pri = PRI_MIN_TIMESHARE;
#ifdef KSE
sched_user_prio(kg, pri);
#else
sched_user_prio(td, pri);
#endif
return;
}
/*
* Calculate a time slice based on the properties of the kseg and the runq
* that we're on. This is only for PRI_TIMESHARE ksegrps.
* that we're on. This is only for PRI_TIMESHARE threads.
*/
static void
sched_slice(struct kse *ke)
{
struct kseq *kseq;
struct ksegrp *kg;
struct thread *td;
kg = ke->ke_ksegrp;
td = ke->ke_thread;
kseq = KSEQ_CPU(ke->ke_cpu);
if (ke->ke_thread->td_flags & TDF_BORROWING) {
if (td->td_flags & TDF_BORROWING) {
ke->ke_slice = SCHED_SLICE_MIN;
return;
}
@ -1099,7 +1084,7 @@ sched_slice(struct kse *ke)
*
* There is 20 point window that starts relative to the least
* nice kse on the run queue. Slice size is determined by
* the kse distance from the last nice ksegrp.
* the kse distance from the last nice thread.
*
* If the kse is outside of the window it will get no slice
* and will be reevaluated each time it is selected on the
@ -1107,16 +1092,16 @@ sched_slice(struct kse *ke)
* a nice -20 is running. They are always granted a minimum
* slice.
*/
if (!SCHED_INTERACTIVE(kg)) {
if (!SCHED_INTERACTIVE(td)) {
int nice;
nice = kg->kg_proc->p_nice + (0 - kseq->ksq_nicemin);
nice = td->td_proc->p_nice + (0 - kseq->ksq_nicemin);
if (kseq->ksq_load_timeshare == 0 ||
kg->kg_proc->p_nice < kseq->ksq_nicemin)
td->td_proc->p_nice < kseq->ksq_nicemin)
ke->ke_slice = SCHED_SLICE_MAX;
else if (nice <= SCHED_SLICE_NTHRESH)
ke->ke_slice = SCHED_SLICE_NICE(nice);
else if (kg->kg_proc->p_nice == 0)
else if (td->td_proc->p_nice == 0)
ke->ke_slice = SCHED_SLICE_MIN;
else
ke->ke_slice = SCHED_SLICE_MIN; /* 0 */
@ -1133,11 +1118,11 @@ sched_slice(struct kse *ke)
* adjusted to more than double their maximum.
*/
static void
sched_interact_update(struct ksegrp *kg)
sched_interact_update(struct thread *td)
{
int sum;
sum = kg->kg_runtime + kg->kg_slptime;
sum = td->td_sched->skg_runtime + td->td_sched->skg_slptime;
if (sum < SCHED_SLP_RUN_MAX)
return;
/*
@ -1146,40 +1131,40 @@ sched_interact_update(struct ksegrp *kg)
* us into the range of [4/5 * SCHED_INTERACT_MAX, SCHED_INTERACT_MAX]
*/
if (sum > (SCHED_SLP_RUN_MAX / 5) * 6) {
kg->kg_runtime /= 2;
kg->kg_slptime /= 2;
td->td_sched->skg_runtime /= 2;
td->td_sched->skg_slptime /= 2;
return;
}
kg->kg_runtime = (kg->kg_runtime / 5) * 4;
kg->kg_slptime = (kg->kg_slptime / 5) * 4;
td->td_sched->skg_runtime = (td->td_sched->skg_runtime / 5) * 4;
td->td_sched->skg_slptime = (td->td_sched->skg_slptime / 5) * 4;
}
static void
sched_interact_fork(struct ksegrp *kg)
sched_interact_fork(struct thread *td)
{
int ratio;
int sum;
sum = kg->kg_runtime + kg->kg_slptime;
sum = td->td_sched->skg_runtime + td->td_sched->skg_slptime;
if (sum > SCHED_SLP_RUN_FORK) {
ratio = sum / SCHED_SLP_RUN_FORK;
kg->kg_runtime /= ratio;
kg->kg_slptime /= ratio;
td->td_sched->skg_runtime /= ratio;
td->td_sched->skg_slptime /= ratio;
}
}
static int
sched_interact_score(struct ksegrp *kg)
sched_interact_score(struct thread *td)
{
int div;
if (kg->kg_runtime > kg->kg_slptime) {
div = max(1, kg->kg_runtime / SCHED_INTERACT_HALF);
if (td->td_sched->skg_runtime > td->td_sched->skg_slptime) {
div = max(1, td->td_sched->skg_runtime / SCHED_INTERACT_HALF);
return (SCHED_INTERACT_HALF +
(SCHED_INTERACT_HALF - (kg->kg_slptime / div)));
} if (kg->kg_slptime > kg->kg_runtime) {
div = max(1, kg->kg_slptime / SCHED_INTERACT_HALF);
return (kg->kg_runtime / div);
(SCHED_INTERACT_HALF - (td->td_sched->skg_slptime / div)));
} if (td->td_sched->skg_slptime > td->td_sched->skg_runtime) {
div = max(1, td->td_sched->skg_slptime / SCHED_INTERACT_HALF);
return (td->td_sched->skg_runtime / div);
}
/*
@ -1202,12 +1187,9 @@ schedinit(void)
* Set up the scheduler specific parts of proc0.
*/
proc0.p_sched = NULL; /* XXX */
ksegrp0.kg_sched = &kg_sched0;
thread0.td_sched = &kse0;
kse0.ke_thread = &thread0;
kse0.ke_state = KES_THREAD;
kg_sched0.skg_concurrency = 1;
kg_sched0.skg_avail_opennings = 0; /* we are already running */
}
/*
@ -1307,7 +1289,7 @@ sched_unlend_prio(struct thread *td, u_char prio)
if (td->td_base_pri >= PRI_MIN_TIMESHARE &&
td->td_base_pri <= PRI_MAX_TIMESHARE)
base_pri = td->td_ksegrp->kg_user_pri;
base_pri = td->td_user_pri;
else
base_pri = td->td_base_pri;
if (prio >= base_pri) {
@ -1345,11 +1327,18 @@ sched_prio(struct thread *td, u_char prio)
}
void
#ifdef KSE
sched_user_prio(struct ksegrp *kg, u_char prio)
#else
sched_user_prio(struct thread *td, u_char prio)
#endif
{
#ifdef KSE
struct thread *td;
#endif
u_char oldprio;
#ifdef KSE
kg->kg_base_user_pri = prio;
/* XXXKSE only for 1:1 */
@ -1365,6 +1354,12 @@ sched_user_prio(struct ksegrp *kg, u_char prio)
oldprio = kg->kg_user_pri;
kg->kg_user_pri = prio;
#else
td->td_base_user_pri = prio;
oldprio = td->td_user_pri;
td->td_user_pri = prio;
#endif
if (TD_ON_UPILOCK(td) && oldprio != prio)
umtx_pi_adjust(td, oldprio);
@ -1377,8 +1372,13 @@ sched_lend_user_prio(struct thread *td, u_char prio)
td->td_flags |= TDF_UBORROWING;
#ifdef KSE
oldprio = td->td_ksegrp->kg_user_pri;
td->td_ksegrp->kg_user_pri = prio;
#else
oldprio = td->td__user_pri;
td->td_user_pri = prio;
#endif
if (TD_ON_UPILOCK(td) && oldprio != prio)
umtx_pi_adjust(td, oldprio);
@ -1387,13 +1387,23 @@ sched_lend_user_prio(struct thread *td, u_char prio)
void
sched_unlend_user_prio(struct thread *td, u_char prio)
{
#ifdef KSE
struct ksegrp *kg = td->td_ksegrp;
#endif
u_char base_pri;
#ifdef KSE
base_pri = kg->kg_base_user_pri;
#else
base_pri = td->td_base_user_pri;
#endif
if (prio >= base_pri) {
td->td_flags &= ~TDF_UBORROWING;
#ifdef KSE
sched_user_prio(kg, base_pri);
#else
sched_user_prio(td, base_pri);
#endif
} else
sched_lend_user_prio(td, prio);
}
@ -1422,7 +1432,6 @@ sched_switch(struct thread *td, struct thread *newtd, int flags)
TD_SET_CAN_RUN(td);
} else if ((ke->ke_flags & KEF_ASSIGNED) == 0) {
/* We are ending our run so make our slot available again */
SLOT_RELEASE(td->td_ksegrp);
kseq_load_rem(ksq, ke);
if (TD_IS_RUNNING(td)) {
/*
@ -1434,15 +1443,7 @@ sched_switch(struct thread *td, struct thread *newtd, int flags)
SRQ_OURSELF|SRQ_YIELDING|SRQ_PREEMPTED :
SRQ_OURSELF|SRQ_YIELDING);
ke->ke_flags &= ~KEF_HOLD;
} else if ((td->td_proc->p_flag & P_HADTHREADS) &&
(newtd == NULL || newtd->td_ksegrp != td->td_ksegrp))
/*
* We will not be on the run queue.
* So we must be sleeping or similar.
* Don't use the slot if we will need it
* for newtd.
*/
slot_fill(td->td_ksegrp);
}
}
if (newtd != NULL) {
/*
@ -1453,15 +1454,6 @@ sched_switch(struct thread *td, struct thread *newtd, int flags)
newtd->td_kse->ke_runq = ksq->ksq_curr;
TD_SET_RUNNING(newtd);
kseq_load_add(KSEQ_SELF(), newtd->td_kse);
/*
* XXX When we preempt, we've already consumed a slot because
* we got here through sched_add(). However, newtd can come
* from thread_switchout() which can't SLOT_USE() because
* the SLOT code is scheduler dependent. We must use the
* slot here otherwise.
*/
if ((flags & SW_PREEMPT) == 0)
SLOT_USE(newtd->td_ksegrp);
} else
newtd = choosethread();
if (td != newtd) {
@ -1469,6 +1461,7 @@ sched_switch(struct thread *td, struct thread *newtd, int flags)
if (PMC_PROC_IS_USING_PMCS(td->td_proc))
PMC_SWITCH_CONTEXT(td, PMC_FN_CSW_OUT);
#endif
cpu_switch(td, newtd);
#ifdef HWPMC_HOOKS
if (PMC_PROC_IS_USING_PMCS(td->td_proc))
@ -1484,7 +1477,6 @@ sched_switch(struct thread *td, struct thread *newtd, int flags)
void
sched_nice(struct proc *p, int nice)
{
struct ksegrp *kg;
struct kse *ke;
struct thread *td;
struct kseq *kseq;
@ -1494,23 +1486,20 @@ sched_nice(struct proc *p, int nice)
/*
* We need to adjust the nice counts for running KSEs.
*/
FOREACH_KSEGRP_IN_PROC(p, kg) {
if (kg->kg_pri_class == PRI_TIMESHARE) {
FOREACH_THREAD_IN_GROUP(kg, td) {
ke = td->td_kse;
if (ke->ke_runq == NULL)
continue;
kseq = KSEQ_CPU(ke->ke_cpu);
kseq_nice_rem(kseq, p->p_nice);
kseq_nice_add(kseq, nice);
}
FOREACH_THREAD_IN_PROC(p, td) {
if (td->td_pri_class == PRI_TIMESHARE) {
ke = td->td_kse;
if (ke->ke_runq == NULL)
continue;
kseq = KSEQ_CPU(ke->ke_cpu);
kseq_nice_rem(kseq, p->p_nice);
kseq_nice_add(kseq, nice);
}
}
p->p_nice = nice;
FOREACH_KSEGRP_IN_PROC(p, kg) {
sched_priority(kg);
FOREACH_THREAD_IN_GROUP(kg, td)
td->td_flags |= TDF_NEEDRESCHED;
FOREACH_THREAD_IN_PROC(p, td) {
sched_priority(td);
td->td_flags |= TDF_NEEDRESCHED;
}
}
@ -1519,7 +1508,7 @@ sched_sleep(struct thread *td)
{
mtx_assert(&sched_lock, MA_OWNED);
td->td_slptime = ticks;
td->td_kse->ke_slptime = ticks;
}
void
@ -1531,22 +1520,20 @@ sched_wakeup(struct thread *td)
* Let the kseg know how long we slept for. This is because process
* interactivity behavior is modeled in the kseg.
*/
if (td->td_slptime) {
struct ksegrp *kg;
if (td->td_kse->ke_slptime) {
int hzticks;
kg = td->td_ksegrp;
hzticks = (ticks - td->td_slptime) << 10;
hzticks = (ticks - td->td_kse->ke_slptime) << 10;
if (hzticks >= SCHED_SLP_RUN_MAX) {
kg->kg_slptime = SCHED_SLP_RUN_MAX;
kg->kg_runtime = 1;
td->td_sched->skg_slptime = SCHED_SLP_RUN_MAX;
td->td_sched->skg_runtime = 1;
} else {
kg->kg_slptime += hzticks;
sched_interact_update(kg);
td->td_sched->skg_slptime += hzticks;
sched_interact_update(td);
}
sched_priority(kg);
sched_priority(td);
sched_slice(td->td_kse);
td->td_slptime = 0;
td->td_kse->ke_slptime = 0;
}
setrunqueue(td, SRQ_BORING);
}
@ -1556,37 +1543,23 @@ sched_wakeup(struct thread *td)
* priority.
*/
void
sched_fork(struct thread *td, struct thread *childtd)
{
mtx_assert(&sched_lock, MA_OWNED);
sched_fork_ksegrp(td, childtd->td_ksegrp);
sched_fork_thread(td, childtd);
}
void
sched_fork_ksegrp(struct thread *td, struct ksegrp *child)
{
struct ksegrp *kg = td->td_ksegrp;
mtx_assert(&sched_lock, MA_OWNED);
child->kg_slptime = kg->kg_slptime;
child->kg_runtime = kg->kg_runtime;
child->kg_user_pri = kg->kg_user_pri;
child->kg_base_user_pri = kg->kg_base_user_pri;
sched_interact_fork(child);
kg->kg_runtime += tickincr;
sched_interact_update(kg);
}
void
sched_fork_thread(struct thread *td, struct thread *child)
sched_fork(struct thread *td, struct thread *child)
{
struct kse *ke;
struct kse *ke2;
mtx_assert(&sched_lock, MA_OWNED);
child->td_sched->skg_slptime = td->td_sched->skg_slptime;
child->td_sched->skg_runtime = td->td_sched->skg_runtime;
child->td_user_pri = td->td_user_pri;
child->kg_base_user_pri = kg->kg_base_user_pri;
sched_interact_fork(child);
td->td_sched->skg_runtime += tickincr;
sched_interact_update(td);
sched_newthread(child);
ke = td->td_kse;
ke2 = child->td_kse;
ke2->ke_slice = 1; /* Attempt to quickly learn interactivity. */
@ -1600,55 +1573,52 @@ sched_fork_thread(struct thread *td, struct thread *child)
}
void
sched_class(struct ksegrp *kg, int class)
sched_class(struct thread *td, int class)
{
struct kseq *kseq;
struct kse *ke;
struct thread *td;
int nclass;
int oclass;
mtx_assert(&sched_lock, MA_OWNED);
if (kg->kg_pri_class == class)
if (td->td_pri_class == class)
return;
nclass = PRI_BASE(class);
oclass = PRI_BASE(kg->kg_pri_class);
FOREACH_THREAD_IN_GROUP(kg, td) {
ke = td->td_kse;
if ((ke->ke_state != KES_ONRUNQ &&
ke->ke_state != KES_THREAD) || ke->ke_runq == NULL)
continue;
kseq = KSEQ_CPU(ke->ke_cpu);
oclass = PRI_BASE(td->td_pri_class);
ke = td->td_kse;
if ((ke->ke_state != KES_ONRUNQ &&
ke->ke_state != KES_THREAD) || ke->ke_runq == NULL)
continue;
kseq = KSEQ_CPU(ke->ke_cpu);
#ifdef SMP
/*
* On SMP if we're on the RUNQ we must adjust the transferable
* count because could be changing to or from an interrupt
* class.
*/
if (ke->ke_state == KES_ONRUNQ) {
if (KSE_CAN_MIGRATE(ke)) {
kseq->ksq_transferable--;
kseq->ksq_group->ksg_transferable--;
}
if (KSE_CAN_MIGRATE(ke)) {
kseq->ksq_transferable++;
kseq->ksq_group->ksg_transferable++;
}
/*
* On SMP if we're on the RUNQ we must adjust the transferable
* count because could be changing to or from an interrupt
* class.
*/
if (ke->ke_state == KES_ONRUNQ) {
if (KSE_CAN_MIGRATE(ke)) {
kseq->ksq_transferable--;
kseq->ksq_group->ksg_transferable--;
}
#endif
if (oclass == PRI_TIMESHARE) {
kseq->ksq_load_timeshare--;
kseq_nice_rem(kseq, kg->kg_proc->p_nice);
}
if (nclass == PRI_TIMESHARE) {
kseq->ksq_load_timeshare++;
kseq_nice_add(kseq, kg->kg_proc->p_nice);
if (KSE_CAN_MIGRATE(ke)) {
kseq->ksq_transferable++;
kseq->ksq_group->ksg_transferable++;
}
}
#endif
if (oclass == PRI_TIMESHARE) {
kseq->ksq_load_timeshare--;
kseq_nice_rem(kseq, td->td_proc->p_nice);
}
if (nclass == PRI_TIMESHARE) {
kseq->ksq_load_timeshare++;
kseq_nice_add(kseq, td->td_proc->p_nice);
}
kg->kg_pri_class = class;
td->td_pri_class = class;
}
/*
@ -1657,24 +1627,16 @@ sched_class(struct ksegrp *kg, int class)
void
sched_exit(struct proc *p, struct thread *childtd)
{
struct thread *parent = FIRST_THREAD_IN_PROC(p);
mtx_assert(&sched_lock, MA_OWNED);
sched_exit_ksegrp(FIRST_KSEGRP_IN_PROC(p), childtd);
sched_exit_thread(NULL, childtd);
}
void
sched_exit_ksegrp(struct ksegrp *kg, struct thread *td)
{
/* kg->kg_slptime += td->td_ksegrp->kg_slptime; */
kg->kg_runtime += td->td_ksegrp->kg_runtime;
sched_interact_update(kg);
}
void
sched_exit_thread(struct thread *td, struct thread *childtd)
{
CTR3(KTR_SCHED, "sched_exit_thread: %p(%s) prio %d",
CTR3(KTR_SCHED, "sched_exit: %p(%s) prio %d",
childtd, childtd->td_proc->p_comm, childtd->td_priority);
/* parent->td_sched->skg_slptime += childtd->td_sched->skg_slptime; */
parent->td_sched->skg_runtime += childtd->td_sched->skg_runtime;
sched_interact_update(parent);
kseq_load_rem(KSEQ_CPU(childtd->td_kse->ke_cpu), childtd->td_kse);
}
@ -1682,7 +1644,6 @@ void
sched_clock(struct thread *td)
{
struct kseq *kseq;
struct ksegrp *kg;
struct kse *ke;
mtx_assert(&sched_lock, MA_OWNED);
@ -1700,7 +1661,6 @@ sched_clock(struct thread *td)
kseq_assign(kseq); /* Potentially sets NEEDRESCHED */
#endif
ke = td->td_kse;
kg = ke->ke_ksegrp;
/* Adjust ticks for pctcpu */
ke->ke_ticks++;
@ -1713,16 +1673,16 @@ sched_clock(struct thread *td)
if (td->td_flags & TDF_IDLETD)
return;
/*
* We only do slicing code for TIMESHARE ksegrps.
* We only do slicing code for TIMESHARE threads.
*/
if (kg->kg_pri_class != PRI_TIMESHARE)
if (td->td_pri_class != PRI_TIMESHARE)
return;
/*
* We used a tick charge it to the ksegrp so that we can compute our
* We used a tick charge it to the thread so that we can compute our
* interactivity.
*/
kg->kg_runtime += tickincr;
sched_interact_update(kg);
td->td_sched->skg_runtime += tickincr;
sched_interact_update(td);
/*
* We used up one time slice.
@ -1733,9 +1693,9 @@ sched_clock(struct thread *td)
* We're out of time, recompute priorities and requeue.
*/
kseq_load_rem(kseq, ke);
sched_priority(kg);
sched_priority(td);
sched_slice(ke);
if (SCHED_CURR(kg, ke))
if (SCHED_CURR(td, ke))
ke->ke_runq = kseq->ksq_curr;
else
ke->ke_runq = kseq->ksq_next;
@ -1770,22 +1730,6 @@ sched_runnable(void)
return (load);
}
void
sched_userret(struct thread *td)
{
struct ksegrp *kg;
KASSERT((td->td_flags & TDF_BORROWING) == 0,
("thread with borrowed priority returning to userland"));
kg = td->td_ksegrp;
if (td->td_priority != kg->kg_user_pri) {
mtx_lock_spin(&sched_lock);
td->td_priority = kg->kg_user_pri;
td->td_base_pri = kg->kg_user_pri;
mtx_unlock_spin(&sched_lock);
}
}
struct kse *
sched_choose(void)
{
@ -1802,7 +1746,7 @@ sched_choose(void)
ke = kseq_choose(kseq);
if (ke) {
#ifdef SMP
if (ke->ke_ksegrp->kg_pri_class == PRI_IDLE)
if (ke->ke_thread->td_pri_class == PRI_IDLE)
if (kseq_idled(kseq) == 0)
goto restart;
#endif
@ -1822,7 +1766,6 @@ void
sched_add(struct thread *td, int flags)
{
struct kseq *kseq;
struct ksegrp *kg;
struct kse *ke;
int preemptive;
int canmigrate;
@ -1833,13 +1776,10 @@ sched_add(struct thread *td, int flags)
curthread->td_proc->p_comm);
mtx_assert(&sched_lock, MA_OWNED);
ke = td->td_kse;
kg = td->td_ksegrp;
canmigrate = 1;
preemptive = !(flags & SRQ_YIELDING);
class = PRI_BASE(kg->kg_pri_class);
class = PRI_BASE(td->td_pri_class);
kseq = KSEQ_SELF();
if ((ke->ke_flags & KEF_INTERNAL) == 0)
SLOT_USE(td->td_ksegrp);
ke->ke_flags &= ~KEF_INTERNAL;
#ifdef SMP
if (ke->ke_flags & KEF_ASSIGNED) {
@ -1859,8 +1799,8 @@ sched_add(struct thread *td, int flags)
#endif
KASSERT(ke->ke_state != KES_ONRUNQ,
("sched_add: kse %p (%s) already in run queue", ke,
ke->ke_proc->p_comm));
KASSERT(ke->ke_proc->p_sflag & PS_INMEM,
td->td_proc->p_comm));
KASSERT(td->td_proc->p_sflag & PS_INMEM,
("sched_add: process swapped out"));
KASSERT(ke->ke_runq == NULL,
("sched_add: KSE %p is still assigned to a run queue", ke));
@ -1875,7 +1815,7 @@ sched_add(struct thread *td, int flags)
ke->ke_cpu = PCPU_GET(cpuid);
break;
case PRI_TIMESHARE:
if (SCHED_CURR(kg, ke))
if (SCHED_CURR(td, ke))
ke->ke_runq = kseq->ksq_curr;
else
ke->ke_runq = kseq->ksq_next;
@ -1947,7 +1887,6 @@ sched_rem(struct thread *td)
curthread->td_proc->p_comm);
mtx_assert(&sched_lock, MA_OWNED);
ke = td->td_kse;
SLOT_RELEASE(td->td_ksegrp);
ke->ke_flags &= ~KEF_PREEMPTED;
if (ke->ke_flags & KEF_ASSIGNED) {
ke->ke_flags |= KEF_REMOVED;
@ -1990,7 +1929,7 @@ sched_pctcpu(struct thread *td)
pctcpu = (FSCALE * ((FSCALE * rtick)/realstathz)) >> FSHIFT;
}
ke->ke_proc->p_swtime = ke->ke_ltick - ke->ke_ftick;
td->td_proc->p_swtime = ke->ke_ltick - ke->ke_ftick;
mtx_unlock_spin(&sched_lock);
return (pctcpu);
@ -2033,11 +1972,17 @@ sched_is_bound(struct thread *td)
void
sched_relinquish(struct thread *td)
{
#ifdef KSE
struct ksegrp *kg;
kg = td->td_ksegrp;
#endif
mtx_lock_spin(&sched_lock);
#ifdef KSE
if (kg->kg_pri_class == PRI_TIMESHARE)
#else
if (td->td_pri_class == PRI_TIMESHARE)
#endif
sched_prio(td, PRI_MAX_TIMESHARE);
mi_switch(SW_VOL, NULL);
mtx_unlock_spin(&sched_lock);
@ -2059,12 +2004,6 @@ sched_load(void)
#endif
}
int
sched_sizeof_ksegrp(void)
{
return (sizeof(struct ksegrp) + sizeof(struct kg_sched));
}
int
sched_sizeof_proc(void)
{

12
sys/kern/subr_trap.c
View File

@ -115,11 +115,13 @@ userret(struct thread *td, struct trapframe *frame)
PROC_UNLOCK(p);
}
#ifdef KSE
/*
* Do special thread processing, e.g. upcall tweaking and such.
*/
if (p->p_flag & P_SA)
thread_userret(td, frame);
#endif
/*
* Charge system time if profiling.
@ -147,7 +149,9 @@ ast(struct trapframe *framep)
{
struct thread *td;
struct proc *p;
#ifdef KSE
struct ksegrp *kg;
#endif
struct rlimit rlim;
int sflag;
int flags;
@ -159,7 +163,9 @@ ast(struct trapframe *framep)
td = curthread;
p = td->td_proc;
#ifdef KSE
kg = td->td_ksegrp;
#endif
CTR3(KTR_SYSC, "ast: thread %p (pid %d, %s)", td, p->p_pid,
p->p_comm);
@ -170,8 +176,10 @@ ast(struct trapframe *framep)
td->td_frame = framep;
td->td_pticks = 0;
#ifdef KSE
if ((p->p_flag & P_SA) && (td->td_mailbox == NULL))
thread_user_enter(td);
#endif
/*
* This updates the p_sflag's for the checks below in one
@ -256,7 +264,11 @@ ast(struct trapframe *framep)
ktrcsw(1, 1);
#endif
mtx_lock_spin(&sched_lock);
#ifdef KSE
sched_prio(td, kg->kg_user_pri);
#else
sched_prio(td, td->td_user_pri);
#endif
mi_switch(SW_INVOL, NULL);
mtx_unlock_spin(&sched_lock);
#ifdef KTRACE

6
sys/kern/sys_process.c
View File

@ -802,7 +802,9 @@ kern_ptrace(struct thread *td, int req, pid_t pid, void *addr, int data)
* continuing process.
*/
mtx_unlock_spin(&sched_lock);
#ifdef KSE
thread_continued(p);
#endif
p->p_flag &= ~(P_STOPPED_TRACE|P_STOPPED_SIG|P_WAITED);
mtx_lock_spin(&sched_lock);
thread_unsuspend(p);
@ -940,6 +942,7 @@ kern_ptrace(struct thread *td, int req, pid_t pid, void *addr, int data)
pl->pl_event = PL_EVENT_SIGNAL;
else
pl->pl_event = 0;
#ifdef KSE
if (td2->td_pflags & TDP_SA) {
pl->pl_flags = PL_FLAG_SA;
if (td2->td_upcall && !TD_CAN_UNBIND(td2))
@ -947,6 +950,9 @@ kern_ptrace(struct thread *td, int req, pid_t pid, void *addr, int data)
} else {
pl->pl_flags = 0;
}
#else
pl->pl_flags = 0;
#endif
pl->pl_sigmask = td2->td_sigmask;
pl->pl_siglist = td2->td_siglist;
break;

11
sys/kern/tty.c
View File

@ -2669,7 +2669,11 @@ proc_compare(struct proc *p1, struct proc *p2)
{
int esta, estb;
#ifdef KSE
struct ksegrp *kg;
#else
struct thread *td;
#endif
mtx_assert(&sched_lock, MA_OWNED);
if (p1 == NULL)
return (1);
@ -2690,12 +2694,19 @@ proc_compare(struct proc *p1, struct proc *p2)
* tie - favor one with highest recent cpu utilization
*/
esta = estb = 0;
#ifdef KSE
FOREACH_KSEGRP_IN_PROC(p1,kg) {
esta += kg->kg_estcpu;
}
FOREACH_KSEGRP_IN_PROC(p2,kg) {
estb += kg->kg_estcpu;
}
#else
FOREACH_THREAD_IN_PROC(p1, td)
esta += td->td_estcpu;
FOREACH_THREAD_IN_PROC(p2, td)
estb += td->td_estcpu;
#endif
if (estb > esta)
return (1);
if (esta > estb)

1
sys/pc98/conf/GENERIC
View File

@ -59,6 +59,7 @@ options SYSVMSG # SYSV-style message queues
options SYSVSEM # SYSV-style semaphores
options _KPOSIX_PRIORITY_SCHEDULING # POSIX P1003_1B real-time extensions
options KBD_INSTALL_CDEV # install a CDEV entry in /dev
options KSE # KSE support
# Debugging for use in -current
options KDB # Enable kernel debugger support.

4
sys/pc98/pc98/machdep.c
View File

@ -1905,7 +1905,11 @@ init386(first)
* This may be done better later if it gets more high level
* components in it. If so just link td->td_proc here.
*/
#ifdef KSE
proc_linkup(&proc0, &ksegrp0, &thread0);
#else
proc_linkup(&proc0, &thread0);
#endif
/*
* Initialize DMAC

18
sys/posix4/ksched.c
View File

@ -106,7 +106,11 @@ getscheduler(struct ksched *ksched, struct thread *td, int *policy)
int e = 0;
mtx_lock_spin(&sched_lock);
#ifdef KSE
pri_to_rtp(td->td_ksegrp, &rtp);
#else
pri_to_rtp(td, &rtp);
#endif
mtx_unlock_spin(&sched_lock);
switch (rtp.type)
{
@ -153,7 +157,11 @@ ksched_getparam(struct ksched *ksched,
struct rtprio rtp;
mtx_lock_spin(&sched_lock);
#ifdef KSE
pri_to_rtp(td->td_ksegrp, &rtp);
#else
pri_to_rtp(td, &rtp);
#endif
mtx_unlock_spin(&sched_lock);
if (RTP_PRIO_IS_REALTIME(rtp.type))
param->sched_priority = rtpprio_to_p4prio(rtp.prio);
@ -174,7 +182,9 @@ ksched_setscheduler(struct ksched *ksched,
{
int e = 0;
struct rtprio rtp;
#ifdef KSE
struct ksegrp *kg = td->td_ksegrp;
#endif
switch(policy)
{
@ -189,6 +199,7 @@ ksched_setscheduler(struct ksched *ksched,
? RTP_PRIO_FIFO : RTP_PRIO_REALTIME;
mtx_lock_spin(&sched_lock);
#ifdef KSE
rtp_to_pri(&rtp, kg);
FOREACH_THREAD_IN_GROUP(kg, td) { /* XXXKSE */
if (TD_IS_RUNNING(td)) {
@ -199,6 +210,9 @@ ksched_setscheduler(struct ksched *ksched,
}
}
}
#else
rtp_to_pri(&rtp, td);
#endif
mtx_unlock_spin(&sched_lock);
}
else
@ -212,6 +226,7 @@ ksched_setscheduler(struct ksched *ksched,
rtp.type = RTP_PRIO_NORMAL;
rtp.prio = p4prio_to_rtpprio(param->sched_priority);
mtx_lock_spin(&sched_lock);
#ifdef KSE
rtp_to_pri(&rtp, kg);
/* XXX Simply revert to whatever we had for last
@ -230,6 +245,9 @@ ksched_setscheduler(struct ksched *ksched,
}
}
#else
rtp_to_pri(&rtp, td);
#endif
mtx_unlock_spin(&sched_lock);
}
break;

4
sys/powerpc/aim/machdep.c
View File

@ -295,7 +295,11 @@ powerpc_init(u_int startkernel, u_int endkernel, u_int basekernel, void *mdp)
/*
* Start initializing proc0 and thread0.
*/
#ifdef KSE
proc_linkup(&proc0, &ksegrp0, &thread0);
#else
proc_linkup(&proc0, &thread0);
#endif
thread0.td_frame = &frame0;
/*

2
sys/powerpc/aim/trap.c
View File

@ -351,8 +351,10 @@ syscall(struct trapframe *frame)
PCPU_LAZY_INC(cnt.v_syscall);
#ifdef KSE
if (p->p_flag & P_SA)
thread_user_enter(td);
#endif
code = frame->fixreg[0];
params = (caddr_t)(frame->fixreg + FIRSTARG);

1
sys/powerpc/conf/GENERIC
View File

@ -56,6 +56,7 @@ options SYSVSHM #SYSV-style shared memory
options SYSVMSG #SYSV-style message queues
options SYSVSEM #SYSV-style semaphores
options _KPOSIX_PRIORITY_SCHEDULING #Posix P1003_1B real-time extensions
options KSE # KSE support
# Debugging for use in -current
options KDB #Enable the kernel debugger

4
sys/powerpc/powerpc/machdep.c
View File

@ -295,7 +295,11 @@ powerpc_init(u_int startkernel, u_int endkernel, u_int basekernel, void *mdp)
/*
* Start initializing proc0 and thread0.
*/
#ifdef KSE
proc_linkup(&proc0, &ksegrp0, &thread0);
#else
proc_linkup(&proc0, &thread0);
#endif
thread0.td_frame = &frame0;
/*

2
sys/powerpc/powerpc/trap.c
View File

@ -351,8 +351,10 @@ syscall(struct trapframe *frame)
PCPU_LAZY_INC(cnt.v_syscall);
#ifdef KSE
if (p->p_flag & P_SA)
thread_user_enter(td);
#endif
code = frame->fixreg[0];
params = (caddr_t)(frame->fixreg + FIRSTARG);

1
sys/sparc64/conf/GENERIC
View File

@ -58,6 +58,7 @@ options SYSVMSG # SYSV-style message queues
options SYSVSEM # SYSV-style semaphores
options _KPOSIX_PRIORITY_SCHEDULING # POSIX P1003_1B real-time extensions
options ADAPTIVE_GIANT # Giant mutex is adaptive.
options KSE # KSE support
# Debugging for use in -current
options KDB # Enable kernel debugger support.

4
sys/sparc64/sparc64/machdep.c
View File

@ -391,7 +391,11 @@ sparc64_init(caddr_t mdp, u_long o1, u_long o2, u_long o3, ofw_vec_t *vec)
/*
* Initialize proc0 stuff (p_contested needs to be done early).
*/
#ifdef KSE
proc_linkup(&proc0, &ksegrp0, &thread0);
#else
proc_linkup(&proc0, &thread0);
#endif
proc0.p_md.md_sigtramp = NULL;
proc0.p_md.md_utrap = NULL;
thread0.td_kstack = kstack0;

2
sys/sparc64/sparc64/trap.c
View File

@ -529,8 +529,10 @@ syscall(struct trapframe *tf)
td->td_frame = tf;
if (td->td_ucred != p->p_ucred)
cred_update_thread(td);
#ifdef KSE
if (p->p_flag & P_SA)
thread_user_enter(td);
#endif
code = tf->tf_global[1];
/*

4
sys/sun4v/sun4v/machdep.c
View File

@ -376,7 +376,11 @@ sparc64_init(caddr_t mdp, u_long o1, u_long o2, u_long o3, ofw_vec_t *vec)
* Initialize proc0 stuff (p_contested needs to be done early).
*/
#ifdef KSE
proc_linkup(&proc0, &ksegrp0, &thread0);
#else
proc_linkup(&proc0, &thread0);
#endif
proc0.p_md.md_sigtramp = NULL;
proc0.p_md.md_utrap = NULL;
frame0.tf_tstate = TSTATE_IE | TSTATE_PEF | TSTATE_PRIV;

103
sys/sys/proc.h
View File

@ -144,7 +144,7 @@ struct pargs {
* q - td_contested lock
* r - p_peers lock
* x - created at fork, only changes during single threading in exec
* z - zombie threads/ksegroup lock
* z - zombie threads lock
*
* If the locking key specifies two identifiers (for example, p_pptr) then
* either lock is sufficient for read access, but both locks must be held
@ -152,16 +152,26 @@ struct pargs {
*/
struct auditinfo;
struct kaudit_record;
#ifdef KSE
struct kg_sched;
#else
struct td_sched;
#endif
struct nlminfo;
struct kaioinfo;
struct p_sched;
struct proc;
struct sleepqueue;
#ifdef KSE
struct td_sched;
#else
struct thread;
#endif
struct trapframe;
struct turnstile;
struct mqueue_notifier;
#ifdef KSE
/*
* Here we define the three structures used for process information.
*
@ -230,7 +240,9 @@ They would be given priorities calculated from the KSEG.
*
*****************/
#endif
#ifdef KSE
/*
* Kernel runnable context (thread).
* This is what is put to sleep and reactivated.
@ -240,16 +252,33 @@ They would be given priorities calculated from the KSEG.
* With N runnable and queued KSEs in the KSEGRP, the first N threads
* are linked to them. Other threads are not yet assigned.
*/
#else
/*
* Thread context. Processes may have multiple threads.
*/
#endif
struct thread {
struct proc *td_proc; /* (*) Associated process. */
#ifdef KSE
struct ksegrp *td_ksegrp; /* (*) Associated KSEG. */
#else
void *was_td_ksegrp; /* Temporary padding. */
#endif
TAILQ_ENTRY(thread) td_plist; /* (*) All threads in this proc. */
#ifdef KSE
TAILQ_ENTRY(thread) td_kglist; /* (*) All threads in this ksegrp. */
#else
TAILQ_ENTRY(thread) was_td_kglist; /* Temporary padding. */
#endif
/* The two queues below should someday be merged. */
TAILQ_ENTRY(thread) td_slpq; /* (j) Sleep queue. */
TAILQ_ENTRY(thread) td_lockq; /* (j) Lock queue. */
#ifdef KSE
TAILQ_ENTRY(thread) td_runq; /* (j/z) Run queue(s). XXXKSE */
#else
TAILQ_ENTRY(thread) td_runq; /* (j/z) Run queue(s). */
#endif
TAILQ_HEAD(, selinfo) td_selq; /* (p) List of selinfos. */
struct sleepqueue *td_sleepqueue; /* (k) Associated sleep queue. */
@ -278,10 +307,23 @@ struct thread {
struct lock_list_entry *td_sleeplocks; /* (k) Held sleep locks. */
int td_intr_nesting_level; /* (k) Interrupt recursion. */
int td_pinned; /* (k) Temporary cpu pin count. */
#ifdef KSE
struct kse_thr_mailbox *td_mailbox; /* (*) Userland mailbox address. */
#else
void *was_td_mailbox; /* Temporary padding. */
#endif
struct ucred *td_ucred; /* (k) Reference to credentials. */
#ifdef KSE
struct thread *td_standin; /* (k + a) Use this for an upcall. */
struct kse_upcall *td_upcall; /* (k + j) Upcall structure. */
u_int new_td_estcpu; /* Temporary padding. */
u_int new_td_slptime; /* Temporary padding. */
#else
void *was_td_standin; /* Temporary padding. */
void *was_td_upcall; /* Temporary padding. */
u_int td_estcpu; /* (j) Sum of the same field in KSEs. */
u_int td_slptime; /* (j) How long completely blocked. */
#endif
u_int td_pticks; /* (k) Statclock hits for profiling */
u_int td_sticks; /* (k) Statclock hits in system mode. */
u_int td_iticks; /* (k) Statclock hits in intr mode. */
@ -293,7 +335,11 @@ struct thread {
sigset_t td_sigmask; /* (c) Current signal mask. */
volatile u_int td_generation; /* (k) For detection of preemption */
stack_t td_sigstk; /* (k) Stack ptr and on-stack flag. */
#ifdef KSE
int td_kflags; /* (c) Flags for KSE threading. */
#else
int was_td_kflags; /* Temporary padding. */
#endif
int td_xsig; /* (c) Signal for ptrace */
u_long td_profil_addr; /* (k) Temporary addr until AST. */
u_int td_profil_ticks; /* (k) Temporary ticks until AST. */
@ -304,6 +350,15 @@ struct thread {
#define td_startcopy td_endzero
u_char td_base_pri; /* (j) Thread base kernel priority. */
u_char td_priority; /* (j) Thread active priority. */
#ifdef KSE
u_char new_td_pri_class; /* Temporary padding. */
u_char new_td_user_pri; /* Temporary padding. */
u_char new_td_base_user_pri; /* Temporary padding. */
#else
u_char td_pri_class; /* (j) Scheduling class. */
u_char td_user_pri; /* (j) User pri from estcpu and nice. */
u_char td_base_user_pri; /* (j) Base user pri */
#endif
#define td_endcopy td_pcb
/*
@ -372,15 +427,27 @@ struct thread {
#define TDP_OLDMASK 0x00000001 /* Need to restore mask after suspend. */
#define TDP_INKTR 0x00000002 /* Thread is currently in KTR code. */
#define TDP_INKTRACE 0x00000004 /* Thread is currently in KTRACE code. */
#ifdef KSE
#define TDP_UPCALLING 0x00000008 /* This thread is doing an upcall. */
#else
/* 0x00000008 */
#endif
#define TDP_COWINPROGRESS 0x00000010 /* Snapshot copy-on-write in progress. */
#define TDP_ALTSTACK 0x00000020 /* Have alternate signal stack. */
#define TDP_DEADLKTREAT 0x00000040 /* Lock aquisition - deadlock treatment. */
#ifdef KSE
#define TDP_SA 0x00000080 /* A scheduler activation based thread. */
#else
/* 0x00000080 */
#endif
#define TDP_NOSLEEPING 0x00000100 /* Thread is not allowed to sleep on a sq. */
#define TDP_OWEUPC 0x00000200 /* Call addupc() at next AST. */
#define TDP_ITHREAD 0x00000400 /* Thread is an interrupt thread. */
#ifdef KSE
#define TDP_CAN_UNBIND 0x00000800 /* Only temporarily bound. */
#else
/* 0x00000800 */
#endif
#define TDP_SCHED1 0x00001000 /* Reserved for scheduler private use */
#define TDP_SCHED2 0x00002000 /* Reserved for scheduler private use */
#define TDP_SCHED3 0x00004000 /* Reserved for scheduler private use */
@ -399,6 +466,7 @@ struct thread {
#define TDI_LOCK 0x0008 /* Stopped on a lock. */
#define TDI_IWAIT 0x0010 /* Awaiting interrupt. */
#ifdef KSE
/*
* flags (in kflags) related to M:N threading.
*/
@ -409,6 +477,7 @@ struct thread {
#define TD_CAN_UNBIND(td) \
(((td)->td_pflags & TDP_CAN_UNBIND) && \
((td)->td_upcall != NULL))
#endif
#define TD_IS_SLEEPING(td) ((td)->td_inhibitors & TDI_SLEEPING)
#define TD_ON_SLEEPQ(td) ((td)->td_wchan != NULL)
@ -450,6 +519,7 @@ struct thread {
#define TD_SET_RUNQ(td) (td)->td_state = TDS_RUNQ
#define TD_SET_CAN_RUN(td) (td)->td_state = TDS_CAN_RUN
#ifdef KSE
/*
* An upcall is used when returning to userland. If a thread does not have
* an upcall on return to userland the thread exports its context and exits.
@ -498,6 +568,7 @@ struct ksegrp {
int kg_numthreads; /* (j) Num threads in total. */
struct kg_sched *kg_sched; /* (*) Scheduler-specific data. */
};
#endif
/*
* XXX: Does this belong in resource.h or resourcevar.h instead?
@ -525,7 +596,9 @@ struct rusage_ext {
*/
struct proc {
LIST_ENTRY(proc) p_list; /* (d) List of all processes. */
#ifdef KSE
TAILQ_HEAD(, ksegrp) p_ksegrps; /* (c)(kg_ksegrp) All KSEGs. */
#endif
TAILQ_HEAD(, thread) p_threads; /* (j)(td_plist) Threads. (shortcut) */
TAILQ_HEAD(, thread) p_suspended; /* (td_runq) Suspended threads. */
struct ucred *p_ucred; /* (c) Process owner's identity. */
@ -588,7 +661,9 @@ struct proc {
int p_suspcount; /* (c) Num threads in suspended mode. */
struct thread *p_xthread; /* (c) Trap thread */
int p_boundary_count;/* (c) Num threads at user boundary */
#ifdef KSE
struct ksegrp *p_procscopegrp;
#endif
int p_pendingcnt; /* how many signals are pending */
struct itimers *p_itimers; /* (c) POSIX interval timers. */
/* End area that is zeroed on creation. */
@ -609,7 +684,9 @@ struct proc {
u_short p_xstat; /* (c) Exit status; also stop sig. */
struct knlist p_klist; /* (c) Knotes attached to this proc. */
int p_numthreads; /* (j) Number of threads. */
#ifdef KSE
int p_numksegrps; /* (c) Number of ksegrps. */
#endif
struct mdproc p_md; /* Any machine-dependent fields. */
struct callout p_itcallout; /* (h + c) Interval timer callout. */
u_short p_acflag; /* (c) Accounting flags. */
@ -718,18 +795,22 @@ MALLOC_DECLARE(M_ZOMBIE);
#define FOREACH_PROC_IN_SYSTEM(p) \
LIST_FOREACH((p), &allproc, p_list)
#ifdef KSE
#define FOREACH_KSEGRP_IN_PROC(p, kg) \
TAILQ_FOREACH((kg), &(p)->p_ksegrps, kg_ksegrp)
#define FOREACH_THREAD_IN_GROUP(kg, td) \
TAILQ_FOREACH((td), &(kg)->kg_threads, td_kglist)
#define FOREACH_UPCALL_IN_GROUP(kg, ku) \
TAILQ_FOREACH((ku), &(kg)->kg_upcalls, ku_link)
#endif
#define FOREACH_THREAD_IN_PROC(p, td) \
TAILQ_FOREACH((td), &(p)->p_threads, td_plist)
/* XXXKSE the following lines should probably only be used in 1:1 code: */
#define FIRST_THREAD_IN_PROC(p) TAILQ_FIRST(&(p)->p_threads)
#ifdef KSE
#define FIRST_KSEGRP_IN_PROC(p) TAILQ_FIRST(&(p)->p_ksegrps)
#endif
/*
* We use process IDs <= PID_MAX; PID_MAX + 1 must also fit in a pid_t,
@ -840,7 +921,9 @@ extern u_long pgrphash;
extern struct sx allproc_lock;
extern struct sx proctree_lock;
extern struct mtx ppeers_lock;
#ifdef KSE
extern struct ksegrp ksegrp0; /* Primary ksegrp in proc0. */
#endif
extern struct proc proc0; /* Process slot for swapper. */
extern struct thread thread0; /* Primary thread in proc0. */
extern struct vmspace vmspace0; /* VM space for proc0. */
@ -891,7 +974,11 @@ void pargs_drop(struct pargs *pa);
void pargs_free(struct pargs *pa);
void pargs_hold(struct pargs *pa);
void procinit(void);
#ifdef KSE
void proc_linkup(struct proc *p, struct ksegrp *kg, struct thread *td);
#else
void proc_linkup(struct proc *p, struct thread *td);
#endif
void proc_reparent(struct proc *child, struct proc *newparent);
struct pstats *pstats_alloc(void);
void pstats_fork(struct pstats *src, struct pstats *dst);
@ -919,9 +1006,11 @@ void cpu_fork(struct thread *, struct proc *, struct thread *, int);
void cpu_set_fork_handler(struct thread *, void (*)(void *), void *);
/* New in KSE. */
#ifdef KSE
struct ksegrp *ksegrp_alloc(void);
void ksegrp_free(struct ksegrp *kg);
void ksegrp_stash(struct ksegrp *kg);
#endif
void kse_GC(void);
void kseinit(void);
void cpu_set_upcall(struct thread *td, struct thread *td0);
@ -932,16 +1021,24 @@ void cpu_thread_exit(struct thread *);
void cpu_thread_setup(struct thread *td);
void cpu_thread_swapin(struct thread *);
void cpu_thread_swapout(struct thread *);
#ifdef KSE
void ksegrp_link(struct ksegrp *kg, struct proc *p);
void ksegrp_unlink(struct ksegrp *kg);
#endif
struct thread *thread_alloc(void);
void thread_continued(struct proc *p);
void thread_exit(void) __dead2;
int thread_export_context(struct thread *td, int willexit);
void thread_free(struct thread *td);
#ifdef KSE
void thread_link(struct thread *td, struct ksegrp *kg);
#else
void thread_link(struct thread *td, struct proc *p);
#endif
void thread_reap(void);
#ifdef KSE
struct thread *thread_schedule_upcall(struct thread *td, struct kse_upcall *ku);
#endif
void thread_signal_add(struct thread *td, ksiginfo_t *);
int thread_single(int how);
void thread_single_end(void);
@ -959,17 +1056,21 @@ void thread_unlink(struct thread *td);
void thread_unsuspend(struct proc *p);
void thread_unsuspend_one(struct thread *td);
void thread_unthread(struct thread *td);
#ifdef KSE
int thread_userret(struct thread *td, struct trapframe *frame);
void thread_user_enter(struct thread *td);
#endif
void thread_wait(struct proc *p);
struct thread *thread_find(struct proc *p, lwpid_t tid);
void thr_exit1(void);
#ifdef KSE
struct kse_upcall *upcall_alloc(void);
void upcall_free(struct kse_upcall *ku);
void upcall_link(struct kse_upcall *ku, struct ksegrp *kg);
void upcall_unlink(struct kse_upcall *ku);
void upcall_remove(struct thread *td);
void upcall_stash(struct kse_upcall *ke);
#endif
#endif /* _KERNEL */

6
sys/sys/rtprio.h
View File

@ -75,9 +75,15 @@ struct rtprio {
};
#ifdef _KERNEL
#ifdef KSE
struct ksegrp;
int rtp_to_pri(struct rtprio *, struct ksegrp *);
void pri_to_rtp(struct ksegrp *, struct rtprio *);
#else
struct thread;
int rtp_to_pri(struct rtprio *, struct thread *);
void pri_to_rtp(struct thread *, struct rtprio *);
#endif
#endif
#endif

16
sys/sys/sched.h
View File

@ -52,17 +52,23 @@ void sched_fork(struct thread *td, struct thread *childtd);
* KSE Groups contain scheduling priority information. They record the
* behavior of groups of KSEs and threads.
*/
#ifdef KSE
void sched_class(struct ksegrp *kg, int class);
void sched_exit_ksegrp(struct ksegrp *kg, struct thread *childtd);
void sched_fork_ksegrp(struct thread *td, struct ksegrp *child);
#else
void sched_class(struct thread *td, int class);
#endif
void sched_nice(struct proc *p, int nice);
/*
* Threads are switched in and out, block on resources, have temporary
* priorities inherited from their ksegs, and use up cpu time.
*/
#ifdef KSE
void sched_exit_thread(struct thread *td, struct thread *child);
void sched_fork_thread(struct thread *td, struct thread *child);
#endif
void sched_lend_prio(struct thread *td, u_char prio);
void sched_lend_user_prio(struct thread *td, u_char pri);
fixpt_t sched_pctcpu(struct thread *td);
@ -71,7 +77,11 @@ void sched_sleep(struct thread *td);
void sched_switch(struct thread *td, struct thread *newtd, int flags);
void sched_unlend_prio(struct thread *td, u_char prio);
void sched_unlend_user_prio(struct thread *td, u_char pri);
#ifdef KSE
void sched_user_prio(struct ksegrp *kg, u_char prio);
#else
void sched_user_prio(struct thread *td, u_char prio);
#endif
void sched_userret(struct thread *td);
void sched_wakeup(struct thread *td);
@ -98,7 +108,9 @@ int sched_is_bound(struct thread *td);
* These procedures tell the process data structure allocation code how
* many bytes to actually allocate.
*/
#ifdef KSE
int sched_sizeof_ksegrp(void);
#endif
int sched_sizeof_proc(void);
int sched_sizeof_thread(void);
@ -116,11 +128,15 @@ sched_unpin(void)
/* temporarily here */
void schedinit(void);
#ifdef KSE
void sched_init_concurrency(struct ksegrp *kg);
void sched_set_concurrency(struct ksegrp *kg, int cuncurrency);
#endif
void sched_schedinit(void);
#ifdef KSE
void sched_newproc(struct proc *p, struct ksegrp *kg, struct thread *td);
void sched_thread_exit(struct thread *td);
#endif
void sched_newthread(struct thread *td);
#endif /* !_SYS_SCHED_H_ */

33
sys/vm/vm_glue.c
View File

@ -682,7 +682,9 @@ scheduler(dummy)
ppri = INT_MIN;
sx_slock(&allproc_lock);
FOREACH_PROC_IN_SYSTEM(p) {
#ifdef KSE
struct ksegrp *kg;
#endif
if (p->p_sflag & (PS_INMEM | PS_SWAPPINGOUT | PS_SWAPPINGIN)) {
continue;
}
@ -694,14 +696,18 @@ scheduler(dummy)
*
*/
if (td->td_inhibitors == TDI_SWAPPED) {
#ifdef KSE
kg = td->td_ksegrp;
pri = p->p_swtime + kg->kg_slptime;
#else
pri = p->p_swtime + td->td_slptime;
#endif
if ((p->p_sflag & PS_SWAPINREQ) == 0) {
pri -= p->p_nice * 8;
}
/*
* if this ksegrp is higher priority
* if this ksegrp/thread is higher priority
* and there is enough space, then select
* this process instead of the previous
* selection.
@ -810,7 +816,9 @@ int action;
{
struct proc *p;
struct thread *td;
#ifdef KSE
struct ksegrp *kg;
#endif
int didswap = 0;
retry:
@ -884,15 +892,24 @@ int action;
* do not swapout a realtime process
* Check all the thread groups..
*/
#ifdef KSE
FOREACH_KSEGRP_IN_PROC(p, kg) {
if (PRI_IS_REALTIME(kg->kg_pri_class))
#else
FOREACH_THREAD_IN_PROC(p, td) {
if (PRI_IS_REALTIME(td->td_pri_class))
#endif
goto nextproc;
/*
* Guarantee swap_idle_threshold1
* time in memory.
*/
#ifdef KSE
if (kg->kg_slptime < swap_idle_threshold1)
#else
if (td->td_slptime < swap_idle_threshold1)
#endif
goto nextproc;
/*
@ -904,11 +921,16 @@ int action;
* This could be refined to support
* swapping out a thread.
*/
#ifdef KSE
FOREACH_THREAD_IN_GROUP(kg, td) {
if ((td->td_priority) < PSOCK ||
!thread_safetoswapout(td))
goto nextproc;
}
#else
if ((td->td_priority) < PSOCK || !thread_safetoswapout(td))
goto nextproc;
#endif
/*
* If the system is under memory stress,
* or if we are swapping
@ -917,11 +939,20 @@ int action;
*/
if (((action & VM_SWAP_NORMAL) == 0) &&
(((action & VM_SWAP_IDLE) == 0) ||
#ifdef KSE
(kg->kg_slptime < swap_idle_threshold2)))
#else
(td->td_slptime < swap_idle_threshold2)))
#endif
goto nextproc;
#ifdef KSE
if (minslptime > kg->kg_slptime)
minslptime = kg->kg_slptime;
#else
if (minslptime > td->td_slptime)
minslptime = td->td_slptime;
#endif
}
/*

4
sys/vm/vm_zeroidle.c
View File

@ -179,7 +179,11 @@ pagezero_start(void __unused *arg)
PROC_UNLOCK(pagezero_proc);
mtx_lock_spin(&sched_lock);
td = FIRST_THREAD_IN_PROC(pagezero_proc);
#ifdef KSE
sched_class(td->td_ksegrp, PRI_IDLE);
#else
sched_class(td, PRI_IDLE);
#endif
sched_prio(td, PRI_MAX_IDLE);
setrunqueue(td, SRQ_BORING);
mtx_unlock_spin(&sched_lock);