d/freebsd-nq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
389748c1e8
freebsd-nq/sys/kern/sched_ule.c

1489 lines
35 KiB
C
Raw Normal View History

- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
/*-
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
* Copyright (c) 2002-2003, Jeffrey Roberson <jeff@freebsd.org>
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice unmodified, this list of conditions, and the following
* disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
* IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
* OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
* IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
* NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
* THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
Use __FBSDID().
2003-06-11 00:56:59 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/kernel.h>
#include <sys/ktr.h>
#include <sys/lock.h>
#include <sys/mutex.h>
#include <sys/proc.h>
- Add in support for KSEs with 0 slice values on the run queue. If we try to select a KSE with a slice of 0 we will update its slice and insert it onto the next queue. - Pass the KSE instead of the ksegrp into sched_slice(). This more accurately reflects the behavior of the code. Slices are granted to kses. - Add a function kseq_nice_min() which finds the smallest nice value assigned to the kseg of any KSE on the queue. - Rewrite the logic in sched_slice(). Add a large comment describing the new slice selection scheme. To summarize, slices are assigned based on the nice value. Priorities are still calculated based on the nice and interactivity of a process. Slice sizes of 0 may be granted for KSEs whos nice is 20 or futher away from the lowest nice on the run queue. Other nice values are scaled across the range [min, min+20]. This fixes ULEs bad behavior with positively niced processes.
2003-04-02 06:46:43 +00:00
#include <sys/resource.h>
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
#include <sys/sched.h>
#include <sys/smp.h>
#include <sys/sx.h>
#include <sys/sysctl.h>
#include <sys/sysproto.h>
#include <sys/vmmeter.h>
#ifdef DDB
#include <ddb/ddb.h>
#endif
#ifdef KTRACE
#include <sys/uio.h>
#include <sys/ktrace.h>
#endif
#include <machine/cpu.h>
- Add static to local functions and data where it was missing. - Add an IPI based mechanism for migrating kses. This mechanism is broken down into several components. This is intended to reduce cache thrashing by eliminating most cases where one cpu touches another's run queues. - kseq_notify() appends a kse to a lockless singly linked list and conditionally sends an IPI to the target processor. Right now this is protected by sched_lock but at some point I'd like to get rid of the global lock. This is why I used something more complicated than a standard queue. - kseq_assign() processes our list of kses that have been assigned to us by other processors. This simply calls sched_add() for each item on the list after clearing the new KEF_ASSIGNED flag. This flag is used to indicate that we have been appeneded to the assigned queue but not added to the run queue yet. - In sched_add(), instead of adding a KSE to another processor's queue we use kse_notify() so that we don't touch their queue. Also in sched_add(), if KEF_ASSIGNED is already set return immediately. This can happen if a thread is removed and readded so that the priority is recorded properly. - In sched_rem() return immediately if KEF_ASSIGNED is set. All callers immediately readd simply to adjust priorites etc. - In sched_choose(), if we're running an IDLE task or the per cpu idle thread set our cpumask bit in 'kseq_idle' so that other processors may know that we are idle. Before this, make a single pass through the run queues of other processors so that we may find work more immediately if it is available. - In sched_runnable(), don't scan each processor's run queue, they will IPI us if they have work for us to do. - In sched_add(), if we're adding a thread that can be migrated and we have plenty of work to do, try to migrate the thread to an idle kseq. - Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into consideration. - No longer use kseq_choose() to steal threads, it can lose it's last argument. - Create a new function runq_steal() which operates like runq_choose() but skips threads based on some criteria. Currently it will not steal PRI_ITHD threads. In the future this will be used for CPU binding. - Create a kseq_steal() that checks each run queue with runq_steal(), use kseq_steal() in the places where we used kseq_choose() to steal with before.
2003-10-31 11:16:04 +00:00
#include <machine/smp.h>
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
#define KTR_ULE KTR_NFS
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
/* decay 95% of `p_pctcpu' in 60 seconds; see CCPU_SHIFT before changing */
/* XXX This is bogus compatability crap for ps */
static fixpt_t ccpu = 0.95122942450071400909 * FSCALE; /* exp(-1/20) */
SYSCTL_INT(_kern, OID_AUTO, ccpu, CTLFLAG_RD, &ccpu, 0, "");
static void sched_setup(void *dummy);
SYSINIT(sched_setup, SI_SUB_RUN_QUEUE, SI_ORDER_FIRST, sched_setup, NULL)
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
static SYSCTL_NODE(_kern, OID_AUTO, sched, CTLFLAG_RW, 0, "SCHED");
static int sched_strict;
SYSCTL_INT(_kern_sched, OID_AUTO, strict, CTLFLAG_RD, &sched_strict, 0, "");
- Create a function sched_interact_score() which decides on the interactivity of a kseg and assigns it a value of 0 through 100. - Use sched_interact_score() to determine the dynamic priority. - Define SCHED_CURR() in terms of sched_interact_score(). - Adjust the maximum slice back down to 100ms. - Remove redundant clearing of ke_runq in sched_wakeup() - Clean up #defines and comment them.
2003-03-04 02:45:59 +00:00
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
static int slice_min = 1;
SYSCTL_INT(_kern_sched, OID_AUTO, slice_min, CTLFLAG_RW, &slice_min, 0, "");
- Fix the maximum slice value. I accidentally checked in a value of '2' which meant no process would run for longer than 20ms. - Slightly redo the interactivity scorer. It follows the same algorithm but in a slightly more correct way. Previously values above half were incorrect. - Lower the interactivity threshold to 20. It seems that in testing non- interactive tasks are hardly ever near there and expensive interactive tasks can sometimes surpass it. This area needs more testing. - Remove an unnecessary KTR. - Fix a case where an idle thread that had an elevated priority due to priority prop. would be placed back on the idle queue. - Delay setting NEEDRESCHED until userret() for threads that haad their priority elevated while in kernel. This gives us the same context switch optimization as SCHED_4BSD. - Limit the child's slice to 1 in sched_fork_kse() so we detect its behavior more quickly. - Inhert some of the run/slp time from the child in sched_exit_ksegrp(). - Redo some of the priority comparisons so they are more clear. - Throttle the frequency of sched_pctcpu_update() so that rounding errors do not make it invalid.
2003-06-15 02:18:29 +00:00
static int slice_max = 10;
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
SYSCTL_INT(_kern_sched, OID_AUTO, slice_max, CTLFLAG_RW, &slice_max, 0, "");
int realstathz;
int tickincr = 1;
- Enable STRICT_RESCHED until code that dynamically decides on resched strictness based on the current workload is finished.
2003-02-10 14:11:23 +00:00
- Add a simple CPU load balancing algorithm. This works by executing once a second and equalizing the load between the two most imbalanced CPU. This is intended to clear up long term load imbalances that would not be handled by the 'pull' method in sched_choose(). - Pull out some bits of sched_choose() into a kseq_move() function that moves an arbitrary thread from one kseq to another.
2003-06-09 00:39:09 +00:00
#ifdef SMP
/* Callout to handle load balancing SMP systems. */
static struct callout kseq_lb_callout;
#endif
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
/*
* These datastructures are allocated within their parent datastructure but
* are scheduler specific.
*/
struct ke_sched {
int ske_slice;
struct runq *ske_runq;
/* The following variables are only used for pctcpu calculation */
int ske_ltick; /* Last tick that we were running on */
int ske_ftick; /* First tick that we were running on */
int ske_ticks; /* Tick count */
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
/* CPU that we have affinity for. */
- Stop abusing oncpu for our cpu binding. Define a scheduler local element in the kse datastructure called ke_cpu. This is the cpu which we are currently bound to. Some flags may be added later to support hard binding.
2003-02-03 02:26:28 +00:00
u_char ske_cpu;
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
};
#define ke_slice ke_sched->ske_slice
#define ke_runq ke_sched->ske_runq
#define ke_ltick ke_sched->ske_ltick
#define ke_ftick ke_sched->ske_ftick
#define ke_ticks ke_sched->ske_ticks
- Stop abusing oncpu for our cpu binding. Define a scheduler local element in the kse datastructure called ke_cpu. This is the cpu which we are currently bound to. Some flags may be added later to support hard binding.
2003-02-03 02:26:28 +00:00
#define ke_cpu ke_sched->ske_cpu
- Add static to local functions and data where it was missing. - Add an IPI based mechanism for migrating kses. This mechanism is broken down into several components. This is intended to reduce cache thrashing by eliminating most cases where one cpu touches another's run queues. - kseq_notify() appends a kse to a lockless singly linked list and conditionally sends an IPI to the target processor. Right now this is protected by sched_lock but at some point I'd like to get rid of the global lock. This is why I used something more complicated than a standard queue. - kseq_assign() processes our list of kses that have been assigned to us by other processors. This simply calls sched_add() for each item on the list after clearing the new KEF_ASSIGNED flag. This flag is used to indicate that we have been appeneded to the assigned queue but not added to the run queue yet. - In sched_add(), instead of adding a KSE to another processor's queue we use kse_notify() so that we don't touch their queue. Also in sched_add(), if KEF_ASSIGNED is already set return immediately. This can happen if a thread is removed and readded so that the priority is recorded properly. - In sched_rem() return immediately if KEF_ASSIGNED is set. All callers immediately readd simply to adjust priorites etc. - In sched_choose(), if we're running an IDLE task or the per cpu idle thread set our cpumask bit in 'kseq_idle' so that other processors may know that we are idle. Before this, make a single pass through the run queues of other processors so that we may find work more immediately if it is available. - In sched_runnable(), don't scan each processor's run queue, they will IPI us if they have work for us to do. - In sched_add(), if we're adding a thread that can be migrated and we have plenty of work to do, try to migrate the thread to an idle kseq. - Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into consideration. - No longer use kseq_choose() to steal threads, it can lose it's last argument. - Create a new function runq_steal() which operates like runq_choose() but skips threads based on some criteria. Currently it will not steal PRI_ITHD threads. In the future this will be used for CPU binding. - Create a kseq_steal() that checks each run queue with runq_steal(), use kseq_steal() in the places where we used kseq_choose() to steal with before.
2003-10-31 11:16:04 +00:00
#define ke_assign ke_procq.tqe_next
#define KEF_ASSIGNED KEF_SCHED0 /* KSE is being migrated. */
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
struct kg_sched {
- Add a new variable 'kg_runtime' that tracks the amount of time we've run. - Use the ratio of kg_runtime / kg_slptime to determine our dynamic priority. - Scale kg_runtime and kg_slptime back when the sum of the two exceeds SCHED_SLP_RUN_MAX. This allows us to slowly forget old behavior. - Scale back the runtime and slptime in fork so that the new process has the same ratio but much less accumulated time. This causes new behavior to be noticed more quickly.
2003-02-10 14:03:45 +00:00
int skg_slptime; /* Number of ticks we vol. slept */
int skg_runtime; /* Number of ticks we were running */
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
};
#define kg_slptime kg_sched->skg_slptime
- Add a new variable 'kg_runtime' that tracks the amount of time we've run. - Use the ratio of kg_runtime / kg_slptime to determine our dynamic priority. - Scale kg_runtime and kg_slptime back when the sum of the two exceeds SCHED_SLP_RUN_MAX. This allows us to slowly forget old behavior. - Scale back the runtime and slptime in fork so that the new process has the same ratio but much less accumulated time. This causes new behavior to be noticed more quickly.
2003-02-10 14:03:45 +00:00
#define kg_runtime kg_sched->skg_runtime
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
struct td_sched {
int std_slptime;
};
#define td_slptime td_sched->std_slptime
- Make some context switches conditional on SCHED_STRICT_RESCHED. This may have some negative effect on interactivity but it yields great perf. gains. This also brings the conditions under which ULE context switches inline with SCHED_4BSD. - Define some new kseq_* functions for manipulating the run queue. - Add a new kseq member ksq_rslices and ksq_bload. rslices is the sum of the slices of runnable kses. This will be used for push load balance decisions. bload is the number of threads blocked waiting on IO.
2003-02-03 05:30:07 +00:00
struct td_sched td_sched;
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
struct ke_sched ke_sched;
struct kg_sched kg_sched;
struct ke_sched *kse0_sched = &ke_sched;
struct kg_sched *ksegrp0_sched = &kg_sched;
struct p_sched *proc0_sched = NULL;
struct td_sched *thread0_sched = &td_sched;
/*
- Don't allow nice to have such a large effect on priority. This was causing poor interactive performance while unnice processes were running. The new scheme still allows nice to have an effect on priority but it is not as dramatic as the effect of the interactivity score.
2003-06-21 02:22:47 +00:00
* The priority is primarily determined by the interactivity score. Thus, we
* give lower(better) priorities to kse groups that use less CPU. The nice
* value is then directly added to this to allow nice to have some effect
* on latency.
- Create a function sched_interact_score() which decides on the interactivity of a kseg and assigns it a value of 0 through 100. - Use sched_interact_score() to determine the dynamic priority. - Define SCHED_CURR() in terms of sched_interact_score(). - Adjust the maximum slice back down to 100ms. - Remove redundant clearing of ke_runq in sched_wakeup() - Clean up #defines and comment them.
2003-03-04 02:45:59 +00:00
*
* PRI_RANGE: Total priority range for timeshare threads.
- Don't allow nice to have such a large effect on priority. This was causing poor interactive performance while unnice processes were running. The new scheme still allows nice to have an effect on priority but it is not as dramatic as the effect of the interactivity score.
2003-06-21 02:22:47 +00:00
* PRI_NRESV: Number of nice values.
- Create a function sched_interact_score() which decides on the interactivity of a kseg and assigns it a value of 0 through 100. - Use sched_interact_score() to determine the dynamic priority. - Define SCHED_CURR() in terms of sched_interact_score(). - Adjust the maximum slice back down to 100ms. - Remove redundant clearing of ke_runq in sched_wakeup() - Clean up #defines and comment them.
2003-03-04 02:45:59 +00:00
* PRI_BASE: The start of the dynamic range.
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
*/
- Create a function sched_interact_score() which decides on the interactivity of a kseg and assigns it a value of 0 through 100. - Use sched_interact_score() to determine the dynamic priority. - Define SCHED_CURR() in terms of sched_interact_score(). - Adjust the maximum slice back down to 100ms. - Remove redundant clearing of ke_runq in sched_wakeup() - Clean up #defines and comment them.
2003-03-04 02:45:59 +00:00
#define SCHED_PRI_RANGE (PRI_MAX_TIMESHARE - PRI_MIN_TIMESHARE + 1)
- Add in support for KSEs with 0 slice values on the run queue. If we try to select a KSE with a slice of 0 we will update its slice and insert it onto the next queue. - Pass the KSE instead of the ksegrp into sched_slice(). This more accurately reflects the behavior of the code. Slices are granted to kses. - Add a function kseq_nice_min() which finds the smallest nice value assigned to the kseg of any KSE on the queue. - Rewrite the logic in sched_slice(). Add a large comment describing the new slice selection scheme. To summarize, slices are assigned based on the nice value. Priorities are still calculated based on the nice and interactivity of a process. Slice sizes of 0 may be granted for KSEs whos nice is 20 or futher away from the lowest nice on the run queue. Other nice values are scaled across the range [min, min+20]. This fixes ULEs bad behavior with positively niced processes.
2003-04-02 06:46:43 +00:00
#define SCHED_PRI_NRESV PRIO_TOTAL
- I meant divide by two and not shift by two in SCHED_PRI_NHALF.
2003-04-02 08:21:24 +00:00
#define SCHED_PRI_NHALF (PRIO_TOTAL / 2)
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
#define SCHED_PRI_NTHRESH (SCHED_PRI_NHALF - 1)
- Don't allow nice to have such a large effect on priority. This was causing poor interactive performance while unnice processes were running. The new scheme still allows nice to have an effect on priority but it is not as dramatic as the effect of the interactivity score.
2003-06-21 02:22:47 +00:00
#define SCHED_PRI_BASE (PRI_MIN_TIMESHARE)
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
#define SCHED_PRI_INTERACT(score) \
- Don't allow nice to have such a large effect on priority. This was causing poor interactive performance while unnice processes were running. The new scheme still allows nice to have an effect on priority but it is not as dramatic as the effect of the interactivity score.
2003-06-21 02:22:47 +00:00
((score) * SCHED_PRI_RANGE / SCHED_INTERACT_MAX)
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
/*
- Create a function sched_interact_score() which decides on the interactivity of a kseg and assigns it a value of 0 through 100. - Use sched_interact_score() to determine the dynamic priority. - Define SCHED_CURR() in terms of sched_interact_score(). - Adjust the maximum slice back down to 100ms. - Remove redundant clearing of ke_runq in sched_wakeup() - Clean up #defines and comment them.
2003-03-04 02:45:59 +00:00
* These determine the interactivity of a process.
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
*
- Add a new variable 'kg_runtime' that tracks the amount of time we've run. - Use the ratio of kg_runtime / kg_slptime to determine our dynamic priority. - Scale kg_runtime and kg_slptime back when the sum of the two exceeds SCHED_SLP_RUN_MAX. This allows us to slowly forget old behavior. - Scale back the runtime and slptime in fork so that the new process has the same ratio but much less accumulated time. This causes new behavior to be noticed more quickly.
2003-02-10 14:03:45 +00:00
* SLP_RUN_MAX: Maximum amount of sleep time + run time we'll accumulate
* before throttling back.
- Throttle the inherited sleep and run time in sched_fork_kseg(). This allows us to learn the behavior of a thread much more quickly after it starts up.
2003-06-28 06:19:56 +00:00
* SLP_RUN_THROTTLE: Divisor for reducing slp/run time at fork time.
- Fix the maximum slice value. I accidentally checked in a value of '2' which meant no process would run for longer than 20ms. - Slightly redo the interactivity scorer. It follows the same algorithm but in a slightly more correct way. Previously values above half were incorrect. - Lower the interactivity threshold to 20. It seems that in testing non- interactive tasks are hardly ever near there and expensive interactive tasks can sometimes surpass it. This area needs more testing. - Remove an unnecessary KTR. - Fix a case where an idle thread that had an elevated priority due to priority prop. would be placed back on the idle queue. - Delay setting NEEDRESCHED until userret() for threads that haad their priority elevated while in kernel. This gives us the same context switch optimization as SCHED_4BSD. - Limit the child's slice to 1 in sched_fork_kse() so we detect its behavior more quickly. - Inhert some of the run/slp time from the child in sched_exit_ksegrp(). - Redo some of the priority comparisons so they are more clear. - Throttle the frequency of sched_pctcpu_update() so that rounding errors do not make it invalid.
2003-06-15 02:18:29 +00:00
* INTERACT_MAX: Maximum interactivity value. Smaller is better.
- Create a function sched_interact_score() which decides on the interactivity of a kseg and assigns it a value of 0 through 100. - Use sched_interact_score() to determine the dynamic priority. - Define SCHED_CURR() in terms of sched_interact_score(). - Adjust the maximum slice back down to 100ms. - Remove redundant clearing of ke_runq in sched_wakeup() - Clean up #defines and comment them.
2003-03-04 02:45:59 +00:00
* INTERACT_THRESH: Threshhold for placement on the current runq.
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
*/
- The non iterative algorithm for interact_update was broken due to rounding errors. This was the source of the majority of the interactivity problems. Reintroduce the old algorithm and its XXX. - Up the interactivity threshold to 30. It really could stand to be even a tiny bit higher. - Let the sleep and run time accumulate up to 5 seconds of history rather than two. This helps stop XFree86 from becoming non-interactive during bursts of activity.
2003-10-16 08:17:43 +00:00
#define SCHED_SLP_RUN_MAX ((hz * 5) << 10)
- Throttle the inherited sleep and run time in sched_fork_kseg(). This allows us to learn the behavior of a thread much more quickly after it starts up.
2003-06-28 06:19:56 +00:00
#define SCHED_SLP_RUN_THROTTLE (100)
- Fix the maximum slice value. I accidentally checked in a value of '2' which meant no process would run for longer than 20ms. - Slightly redo the interactivity scorer. It follows the same algorithm but in a slightly more correct way. Previously values above half were incorrect. - Lower the interactivity threshold to 20. It seems that in testing non- interactive tasks are hardly ever near there and expensive interactive tasks can sometimes surpass it. This area needs more testing. - Remove an unnecessary KTR. - Fix a case where an idle thread that had an elevated priority due to priority prop. would be placed back on the idle queue. - Delay setting NEEDRESCHED until userret() for threads that haad their priority elevated while in kernel. This gives us the same context switch optimization as SCHED_4BSD. - Limit the child's slice to 1 in sched_fork_kse() so we detect its behavior more quickly. - Inhert some of the run/slp time from the child in sched_exit_ksegrp(). - Redo some of the priority comparisons so they are more clear. - Throttle the frequency of sched_pctcpu_update() so that rounding errors do not make it invalid.
2003-06-15 02:18:29 +00:00
#define SCHED_INTERACT_MAX (100)
#define SCHED_INTERACT_HALF (SCHED_INTERACT_MAX / 2)
- The non iterative algorithm for interact_update was broken due to rounding errors. This was the source of the majority of the interactivity problems. Reintroduce the old algorithm and its XXX. - Up the interactivity threshold to 30. It really could stand to be even a tiny bit higher. - Let the sleep and run time accumulate up to 5 seconds of history rather than two. This helps stop XFree86 from becoming non-interactive during bursts of activity.
2003-10-16 08:17:43 +00:00
#define SCHED_INTERACT_THRESH (30)
- Create a function sched_interact_score() which decides on the interactivity of a kseg and assigns it a value of 0 through 100. - Use sched_interact_score() to determine the dynamic priority. - Define SCHED_CURR() in terms of sched_interact_score(). - Adjust the maximum slice back down to 100ms. - Remove redundant clearing of ke_runq in sched_wakeup() - Clean up #defines and comment them.
2003-03-04 02:45:59 +00:00
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
/*
* These parameters and macros determine the size of the time slice that is
* granted to each thread.
*
* SLICE_MIN: Minimum time slice granted, in units of ticks.
* SLICE_MAX: Maximum time slice granted.
* SLICE_RANGE: Range of available time slices scaled by hz.
- Add in support for KSEs with 0 slice values on the run queue. If we try to select a KSE with a slice of 0 we will update its slice and insert it onto the next queue. - Pass the KSE instead of the ksegrp into sched_slice(). This more accurately reflects the behavior of the code. Slices are granted to kses. - Add a function kseq_nice_min() which finds the smallest nice value assigned to the kseg of any KSE on the queue. - Rewrite the logic in sched_slice(). Add a large comment describing the new slice selection scheme. To summarize, slices are assigned based on the nice value. Priorities are still calculated based on the nice and interactivity of a process. Slice sizes of 0 may be granted for KSEs whos nice is 20 or futher away from the lowest nice on the run queue. Other nice values are scaled across the range [min, min+20]. This fixes ULEs bad behavior with positively niced processes.
2003-04-02 06:46:43 +00:00
* SLICE_SCALE: The number slices granted per val in the range of [0, max].
* SLICE_NICE: Determine the amount of slice granted to a scaled nice.
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
*/
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
#define SCHED_SLICE_MIN (slice_min)
#define SCHED_SLICE_MAX (slice_max)
- Create a function sched_interact_score() which decides on the interactivity of a kseg and assigns it a value of 0 through 100. - Use sched_interact_score() to determine the dynamic priority. - Define SCHED_CURR() in terms of sched_interact_score(). - Adjust the maximum slice back down to 100ms. - Remove redundant clearing of ke_runq in sched_wakeup() - Clean up #defines and comment them.
2003-03-04 02:45:59 +00:00
#define SCHED_SLICE_RANGE (SCHED_SLICE_MAX - SCHED_SLICE_MIN + 1)
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
#define SCHED_SLICE_SCALE(val, max) (((val) * SCHED_SLICE_RANGE) / (max))
- Add in support for KSEs with 0 slice values on the run queue. If we try to select a KSE with a slice of 0 we will update its slice and insert it onto the next queue. - Pass the KSE instead of the ksegrp into sched_slice(). This more accurately reflects the behavior of the code. Slices are granted to kses. - Add a function kseq_nice_min() which finds the smallest nice value assigned to the kseg of any KSE on the queue. - Rewrite the logic in sched_slice(). Add a large comment describing the new slice selection scheme. To summarize, slices are assigned based on the nice value. Priorities are still calculated based on the nice and interactivity of a process. Slice sizes of 0 may be granted for KSEs whos nice is 20 or futher away from the lowest nice on the run queue. Other nice values are scaled across the range [min, min+20]. This fixes ULEs bad behavior with positively niced processes.
2003-04-02 06:46:43 +00:00
#define SCHED_SLICE_NICE(nice) \
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
(SCHED_SLICE_MAX - SCHED_SLICE_SCALE((nice), SCHED_PRI_NTHRESH))
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
/*
* This macro determines whether or not the kse belongs on the current or
* next run queue.
- Add a new variable 'kg_runtime' that tracks the amount of time we've run. - Use the ratio of kg_runtime / kg_slptime to determine our dynamic priority. - Scale kg_runtime and kg_slptime back when the sum of the two exceeds SCHED_SLP_RUN_MAX. This allows us to slowly forget old behavior. - Scale back the runtime and slptime in fork so that the new process has the same ratio but much less accumulated time. This causes new behavior to be noticed more quickly.
2003-02-10 14:03:45 +00:00
*
* XXX nice value should effect how interactive a kg is.
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
*/
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
#define SCHED_INTERACTIVE(kg) \
(sched_interact_score(kg) < SCHED_INTERACT_THRESH)
- Unbreak priority prop. for timeshare threads. Always place something on the current queue if its priority is really elevated. This needs more work as there are cases where a next queue kse could be holding up what would be a curr queue kse, and thus hurting interactivity. Also, when a thread with an elevated priority has its priority lowered it should be placed back on the next queue.
2003-04-12 22:33:24 +00:00
#define SCHED_CURR(kg, ke) \
- If our user_pri doesn't match our actual priority our priority has been elevated either due to priority propagation or because we're in the kernel in either case, put us on the current queue so that we dont stop others from using important resources. At some point the priority elevations from sleeping in the kernel should go away. - Remove an optimization in sched_userret(). Before we would only set NEEDRESCHED if there was something of a higher priority available. This is a trivial optimization and it breaks priority propagation because it doesn't take threads which we may be blocking into account. Notice that the thread which is blocking others gets up to one tick of cpu time before we honor this NEEDRESCHED in sched_clock().
2003-10-15 07:47:06 +00:00
(ke->ke_thread->td_priority != kg->kg_user_pri || \
SCHED_INTERACTIVE(kg))
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
/*
* Cpu percentage computation macros and defines.
*
* SCHED_CPU_TIME: Number of seconds to average the cpu usage across.
* SCHED_CPU_TICKS: Number of hz ticks to average the cpu usage across.
*/
- Make the interactivity calculator decay faster. - Make the pcpu estimator update faster.
2003-04-02 08:22:33 +00:00
#define SCHED_CPU_TIME 10
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
#define SCHED_CPU_TICKS (hz * SCHED_CPU_TIME)
/*
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
* kseq - per processor runqs and statistics.
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
*/
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
#define KSEQ_NCLASS (PRI_IDLE + 1) /* Number of run classes. */
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
struct kseq {
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
struct runq ksq_idle; /* Queue of IDLE threads. */
struct runq ksq_timeshare[2]; /* Run queues for !IDLE. */
struct runq *ksq_next; /* Next timeshare queue. */
struct runq *ksq_curr; /* Current queue. */
int ksq_loads[KSEQ_NCLASS]; /* Load for each class */
int ksq_load; /* Aggregate load. */
short ksq_nice[PRIO_TOTAL + 1]; /* KSEs in each nice bin. */
short ksq_nicemin; /* Least nice. */
- Make some context switches conditional on SCHED_STRICT_RESCHED. This may have some negative effect on interactivity but it yields great perf. gains. This also brings the conditions under which ULE context switches inline with SCHED_4BSD. - Define some new kseq_* functions for manipulating the run queue. - Add a new kseq member ksq_rslices and ksq_bload. rslices is the sum of the slices of runnable kses. This will be used for push load balance decisions. bload is the number of threads blocked waiting on IO.
2003-02-03 05:30:07 +00:00
#ifdef SMP
unsigned int ksq_rslices; /* Slices on run queue */
- Add static to local functions and data where it was missing. - Add an IPI based mechanism for migrating kses. This mechanism is broken down into several components. This is intended to reduce cache thrashing by eliminating most cases where one cpu touches another's run queues. - kseq_notify() appends a kse to a lockless singly linked list and conditionally sends an IPI to the target processor. Right now this is protected by sched_lock but at some point I'd like to get rid of the global lock. This is why I used something more complicated than a standard queue. - kseq_assign() processes our list of kses that have been assigned to us by other processors. This simply calls sched_add() for each item on the list after clearing the new KEF_ASSIGNED flag. This flag is used to indicate that we have been appeneded to the assigned queue but not added to the run queue yet. - In sched_add(), instead of adding a KSE to another processor's queue we use kse_notify() so that we don't touch their queue. Also in sched_add(), if KEF_ASSIGNED is already set return immediately. This can happen if a thread is removed and readded so that the priority is recorded properly. - In sched_rem() return immediately if KEF_ASSIGNED is set. All callers immediately readd simply to adjust priorites etc. - In sched_choose(), if we're running an IDLE task or the per cpu idle thread set our cpumask bit in 'kseq_idle' so that other processors may know that we are idle. Before this, make a single pass through the run queues of other processors so that we may find work more immediately if it is available. - In sched_runnable(), don't scan each processor's run queue, they will IPI us if they have work for us to do. - In sched_add(), if we're adding a thread that can be migrated and we have plenty of work to do, try to migrate the thread to an idle kseq. - Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into consideration. - No longer use kseq_choose() to steal threads, it can lose it's last argument. - Create a new function runq_steal() which operates like runq_choose() but skips threads based on some criteria. Currently it will not steal PRI_ITHD threads. In the future this will be used for CPU binding. - Create a kseq_steal() that checks each run queue with runq_steal(), use kseq_steal() in the places where we used kseq_choose() to steal with before.
2003-10-31 11:16:04 +00:00
int ksq_cpus; /* Count of CPUs in this kseq. */
struct kse *ksq_assigned; /* KSEs assigned by another CPU. */
- Make some context switches conditional on SCHED_STRICT_RESCHED. This may have some negative effect on interactivity but it yields great perf. gains. This also brings the conditions under which ULE context switches inline with SCHED_4BSD. - Define some new kseq_* functions for manipulating the run queue. - Add a new kseq member ksq_rslices and ksq_bload. rslices is the sum of the slices of runnable kses. This will be used for push load balance decisions. bload is the number of threads blocked waiting on IO.
2003-02-03 05:30:07 +00:00
#endif
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
};
/*
* One kse queue per processor.
*/
- Use ksq_load as the authoritive count of kses on the pair of kseqs for sched_runnable() et all. - Remove some dead code in sched_clock(). - Define two macros KSEQ_SELF() and KSEQ_CPU() for getting the kseq of the current cpu or some alternate cpu. - Start introducing kseq_() functions, such as kseq_choose() and kseq_setup().
2003-01-29 07:00:51 +00:00
#ifdef SMP
- Add static to local functions and data where it was missing. - Add an IPI based mechanism for migrating kses. This mechanism is broken down into several components. This is intended to reduce cache thrashing by eliminating most cases where one cpu touches another's run queues. - kseq_notify() appends a kse to a lockless singly linked list and conditionally sends an IPI to the target processor. Right now this is protected by sched_lock but at some point I'd like to get rid of the global lock. This is why I used something more complicated than a standard queue. - kseq_assign() processes our list of kses that have been assigned to us by other processors. This simply calls sched_add() for each item on the list after clearing the new KEF_ASSIGNED flag. This flag is used to indicate that we have been appeneded to the assigned queue but not added to the run queue yet. - In sched_add(), instead of adding a KSE to another processor's queue we use kse_notify() so that we don't touch their queue. Also in sched_add(), if KEF_ASSIGNED is already set return immediately. This can happen if a thread is removed and readded so that the priority is recorded properly. - In sched_rem() return immediately if KEF_ASSIGNED is set. All callers immediately readd simply to adjust priorites etc. - In sched_choose(), if we're running an IDLE task or the per cpu idle thread set our cpumask bit in 'kseq_idle' so that other processors may know that we are idle. Before this, make a single pass through the run queues of other processors so that we may find work more immediately if it is available. - In sched_runnable(), don't scan each processor's run queue, they will IPI us if they have work for us to do. - In sched_add(), if we're adding a thread that can be migrated and we have plenty of work to do, try to migrate the thread to an idle kseq. - Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into consideration. - No longer use kseq_choose() to steal threads, it can lose it's last argument. - Create a new function runq_steal() which operates like runq_choose() but skips threads based on some criteria. Currently it will not steal PRI_ITHD threads. In the future this will be used for CPU binding. - Create a kseq_steal() that checks each run queue with runq_steal(), use kseq_steal() in the places where we used kseq_choose() to steal with before.
2003-10-31 11:16:04 +00:00
static int kseq_idle;
static struct kseq kseq_cpu[MAXCPU];
static struct kseq *kseq_idmap[MAXCPU];
- Parse the cpu topology map in sched_setup(). - Associate logical CPUs on the same physical core with the same kseq. - Adjust code that assumed there would only be one running thread in any kseq. - Wrap the HTT code with a ULE_HTT_EXPERIMENTAL ifdef. This is a start towards HyperThreading support but it isn't quite there yet.
2003-07-04 19:59:00 +00:00
#define KSEQ_SELF() (kseq_idmap[PCPU_GET(cpuid)])
#define KSEQ_CPU(x) (kseq_idmap[(x)])
- Use ksq_load as the authoritive count of kses on the pair of kseqs for sched_runnable() et all. - Remove some dead code in sched_clock(). - Define two macros KSEQ_SELF() and KSEQ_CPU() for getting the kseq of the current cpu or some alternate cpu. - Start introducing kseq_() functions, such as kseq_choose() and kseq_setup().
2003-01-29 07:00:51 +00:00
#else
- Add static to local functions and data where it was missing. - Add an IPI based mechanism for migrating kses. This mechanism is broken down into several components. This is intended to reduce cache thrashing by eliminating most cases where one cpu touches another's run queues. - kseq_notify() appends a kse to a lockless singly linked list and conditionally sends an IPI to the target processor. Right now this is protected by sched_lock but at some point I'd like to get rid of the global lock. This is why I used something more complicated than a standard queue. - kseq_assign() processes our list of kses that have been assigned to us by other processors. This simply calls sched_add() for each item on the list after clearing the new KEF_ASSIGNED flag. This flag is used to indicate that we have been appeneded to the assigned queue but not added to the run queue yet. - In sched_add(), instead of adding a KSE to another processor's queue we use kse_notify() so that we don't touch their queue. Also in sched_add(), if KEF_ASSIGNED is already set return immediately. This can happen if a thread is removed and readded so that the priority is recorded properly. - In sched_rem() return immediately if KEF_ASSIGNED is set. All callers immediately readd simply to adjust priorites etc. - In sched_choose(), if we're running an IDLE task or the per cpu idle thread set our cpumask bit in 'kseq_idle' so that other processors may know that we are idle. Before this, make a single pass through the run queues of other processors so that we may find work more immediately if it is available. - In sched_runnable(), don't scan each processor's run queue, they will IPI us if they have work for us to do. - In sched_add(), if we're adding a thread that can be migrated and we have plenty of work to do, try to migrate the thread to an idle kseq. - Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into consideration. - No longer use kseq_choose() to steal threads, it can lose it's last argument. - Create a new function runq_steal() which operates like runq_choose() but skips threads based on some criteria. Currently it will not steal PRI_ITHD threads. In the future this will be used for CPU binding. - Create a kseq_steal() that checks each run queue with runq_steal(), use kseq_steal() in the places where we used kseq_choose() to steal with before.
2003-10-31 11:16:04 +00:00
static struct kseq kseq_cpu;
- Use ksq_load as the authoritive count of kses on the pair of kseqs for sched_runnable() et all. - Remove some dead code in sched_clock(). - Define two macros KSEQ_SELF() and KSEQ_CPU() for getting the kseq of the current cpu or some alternate cpu. - Start introducing kseq_() functions, such as kseq_choose() and kseq_setup().
2003-01-29 07:00:51 +00:00
#define KSEQ_SELF() (&kseq_cpu)
#define KSEQ_CPU(x) (&kseq_cpu)
#endif
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
- Add in support for KSEs with 0 slice values on the run queue. If we try to select a KSE with a slice of 0 we will update its slice and insert it onto the next queue. - Pass the KSE instead of the ksegrp into sched_slice(). This more accurately reflects the behavior of the code. Slices are granted to kses. - Add a function kseq_nice_min() which finds the smallest nice value assigned to the kseg of any KSE on the queue. - Rewrite the logic in sched_slice(). Add a large comment describing the new slice selection scheme. To summarize, slices are assigned based on the nice value. Priorities are still calculated based on the nice and interactivity of a process. Slice sizes of 0 may be granted for KSEs whos nice is 20 or futher away from the lowest nice on the run queue. Other nice values are scaled across the range [min, min+20]. This fixes ULEs bad behavior with positively niced processes.
2003-04-02 06:46:43 +00:00
static void sched_slice(struct kse *ke);
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
static void sched_priority(struct ksegrp *kg);
- Create a function sched_interact_score() which decides on the interactivity of a kseg and assigns it a value of 0 through 100. - Use sched_interact_score() to determine the dynamic priority. - Define SCHED_CURR() in terms of sched_interact_score(). - Adjust the maximum slice back down to 100ms. - Remove redundant clearing of ke_runq in sched_wakeup() - Clean up #defines and comment them.
2003-03-04 02:45:59 +00:00
static int sched_interact_score(struct ksegrp *kg);
- Add a new function "sched_interact_update()" that scales back the sleep and run time. - Scale the sleep and run time back via sched_interact_update() in more places. This is to keep the statistic more accurate. - Charge a parent one tick for forking a child. - Add only the run time and not the sleep time to the parents kg when a thread exits. This allows us to give a penalty for having an expensive thread exit but does not give a bonus for having an interactive thread exit. - Change the SLP_RUN_THROTTLE to limit us to 4/5th and not 1/2. - Change the SLP_RUN_MAX to two seconds. This keeps bursty interactive applications like mozilla and openoffice in the interactive range even through expensive tasks. - Recalculate the slice after every sleep. This ensures that once a task has been marked interactive it only has a slice of 1 at the risk of giving tasks that sleep for a very brief period a longer time slice.
2003-06-17 06:39:51 +00:00
static void sched_interact_update(struct ksegrp *kg);
- Add static to local functions and data where it was missing. - Add an IPI based mechanism for migrating kses. This mechanism is broken down into several components. This is intended to reduce cache thrashing by eliminating most cases where one cpu touches another's run queues. - kseq_notify() appends a kse to a lockless singly linked list and conditionally sends an IPI to the target processor. Right now this is protected by sched_lock but at some point I'd like to get rid of the global lock. This is why I used something more complicated than a standard queue. - kseq_assign() processes our list of kses that have been assigned to us by other processors. This simply calls sched_add() for each item on the list after clearing the new KEF_ASSIGNED flag. This flag is used to indicate that we have been appeneded to the assigned queue but not added to the run queue yet. - In sched_add(), instead of adding a KSE to another processor's queue we use kse_notify() so that we don't touch their queue. Also in sched_add(), if KEF_ASSIGNED is already set return immediately. This can happen if a thread is removed and readded so that the priority is recorded properly. - In sched_rem() return immediately if KEF_ASSIGNED is set. All callers immediately readd simply to adjust priorites etc. - In sched_choose(), if we're running an IDLE task or the per cpu idle thread set our cpumask bit in 'kseq_idle' so that other processors may know that we are idle. Before this, make a single pass through the run queues of other processors so that we may find work more immediately if it is available. - In sched_runnable(), don't scan each processor's run queue, they will IPI us if they have work for us to do. - In sched_add(), if we're adding a thread that can be migrated and we have plenty of work to do, try to migrate the thread to an idle kseq. - Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into consideration. - No longer use kseq_choose() to steal threads, it can lose it's last argument. - Create a new function runq_steal() which operates like runq_choose() but skips threads based on some criteria. Currently it will not steal PRI_ITHD threads. In the future this will be used for CPU binding. - Create a kseq_steal() that checks each run queue with runq_steal(), use kseq_steal() in the places where we used kseq_choose() to steal with before.
2003-10-31 11:16:04 +00:00
static void sched_pctcpu_update(struct kse *ke);
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
- Make some context switches conditional on SCHED_STRICT_RESCHED. This may have some negative effect on interactivity but it yields great perf. gains. This also brings the conditions under which ULE context switches inline with SCHED_4BSD. - Define some new kseq_* functions for manipulating the run queue. - Add a new kseq member ksq_rslices and ksq_bload. rslices is the sum of the slices of runnable kses. This will be used for push load balance decisions. bload is the number of threads blocked waiting on IO.
2003-02-03 05:30:07 +00:00
/* Operations on per processor queues */
- Add static to local functions and data where it was missing. - Add an IPI based mechanism for migrating kses. This mechanism is broken down into several components. This is intended to reduce cache thrashing by eliminating most cases where one cpu touches another's run queues. - kseq_notify() appends a kse to a lockless singly linked list and conditionally sends an IPI to the target processor. Right now this is protected by sched_lock but at some point I'd like to get rid of the global lock. This is why I used something more complicated than a standard queue. - kseq_assign() processes our list of kses that have been assigned to us by other processors. This simply calls sched_add() for each item on the list after clearing the new KEF_ASSIGNED flag. This flag is used to indicate that we have been appeneded to the assigned queue but not added to the run queue yet. - In sched_add(), instead of adding a KSE to another processor's queue we use kse_notify() so that we don't touch their queue. Also in sched_add(), if KEF_ASSIGNED is already set return immediately. This can happen if a thread is removed and readded so that the priority is recorded properly. - In sched_rem() return immediately if KEF_ASSIGNED is set. All callers immediately readd simply to adjust priorites etc. - In sched_choose(), if we're running an IDLE task or the per cpu idle thread set our cpumask bit in 'kseq_idle' so that other processors may know that we are idle. Before this, make a single pass through the run queues of other processors so that we may find work more immediately if it is available. - In sched_runnable(), don't scan each processor's run queue, they will IPI us if they have work for us to do. - In sched_add(), if we're adding a thread that can be migrated and we have plenty of work to do, try to migrate the thread to an idle kseq. - Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into consideration. - No longer use kseq_choose() to steal threads, it can lose it's last argument. - Create a new function runq_steal() which operates like runq_choose() but skips threads based on some criteria. Currently it will not steal PRI_ITHD threads. In the future this will be used for CPU binding. - Create a kseq_steal() that checks each run queue with runq_steal(), use kseq_steal() in the places where we used kseq_choose() to steal with before.
2003-10-31 11:16:04 +00:00
static struct kse * kseq_choose(struct kseq *kseq);
- Use ksq_load as the authoritive count of kses on the pair of kseqs for sched_runnable() et all. - Remove some dead code in sched_clock(). - Define two macros KSEQ_SELF() and KSEQ_CPU() for getting the kseq of the current cpu or some alternate cpu. - Start introducing kseq_() functions, such as kseq_choose() and kseq_setup().
2003-01-29 07:00:51 +00:00
static void kseq_setup(struct kseq *kseq);
- Keep seperate statistics and run queues for different scheduling classes. - Treat each class specially in kseq_{choose,add,rem}. Let the rest of the code be less aware of scheduling classes. - Skip the interactivity calculation for non TIMESHARE ksegrps. - Move slice and runq selection into kseq_add(). Uninline it now that it's big.
2003-04-03 00:29:28 +00:00
static void kseq_add(struct kseq *kseq, struct kse *ke);
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
static void kseq_rem(struct kseq *kseq, struct kse *ke);
static void kseq_nice_add(struct kseq *kseq, int nice);
static void kseq_nice_rem(struct kseq *kseq, int nice);
- Set the ke_cpu field in sched_add() for interrupt and realtime threads since they are going on the current cpu and not their previously assigned cpu. - sched_runnable() should only return true in the SMP case if the other processor has more than one thread that is runnable. We can not steal curthread. - Change kseq_print() to accept the cpuid instead of a kseq pointer. This makes use of this function in ddb much easier.
2003-04-18 05:24:10 +00:00
void kseq_print(int cpu);
- Make some context switches conditional on SCHED_STRICT_RESCHED. This may have some negative effect on interactivity but it yields great perf. gains. This also brings the conditions under which ULE context switches inline with SCHED_4BSD. - Define some new kseq_* functions for manipulating the run queue. - Add a new kseq member ksq_rslices and ksq_bload. rslices is the sum of the slices of runnable kses. This will be used for push load balance decisions. bload is the number of threads blocked waiting on IO.
2003-02-03 05:30:07 +00:00
#ifdef SMP
- Add static to local functions and data where it was missing. - Add an IPI based mechanism for migrating kses. This mechanism is broken down into several components. This is intended to reduce cache thrashing by eliminating most cases where one cpu touches another's run queues. - kseq_notify() appends a kse to a lockless singly linked list and conditionally sends an IPI to the target processor. Right now this is protected by sched_lock but at some point I'd like to get rid of the global lock. This is why I used something more complicated than a standard queue. - kseq_assign() processes our list of kses that have been assigned to us by other processors. This simply calls sched_add() for each item on the list after clearing the new KEF_ASSIGNED flag. This flag is used to indicate that we have been appeneded to the assigned queue but not added to the run queue yet. - In sched_add(), instead of adding a KSE to another processor's queue we use kse_notify() so that we don't touch their queue. Also in sched_add(), if KEF_ASSIGNED is already set return immediately. This can happen if a thread is removed and readded so that the priority is recorded properly. - In sched_rem() return immediately if KEF_ASSIGNED is set. All callers immediately readd simply to adjust priorites etc. - In sched_choose(), if we're running an IDLE task or the per cpu idle thread set our cpumask bit in 'kseq_idle' so that other processors may know that we are idle. Before this, make a single pass through the run queues of other processors so that we may find work more immediately if it is available. - In sched_runnable(), don't scan each processor's run queue, they will IPI us if they have work for us to do. - In sched_add(), if we're adding a thread that can be migrated and we have plenty of work to do, try to migrate the thread to an idle kseq. - Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into consideration. - No longer use kseq_choose() to steal threads, it can lose it's last argument. - Create a new function runq_steal() which operates like runq_choose() but skips threads based on some criteria. Currently it will not steal PRI_ITHD threads. In the future this will be used for CPU binding. - Create a kseq_steal() that checks each run queue with runq_steal(), use kseq_steal() in the places where we used kseq_choose() to steal with before.
2003-10-31 11:16:04 +00:00
#if 0
static int sched_pickcpu(void);
#endif
static struct kse *runq_steal(struct runq *rq);
static struct kseq *kseq_load_highest(void);
static void kseq_balance(void *arg);
static void kseq_move(struct kseq *from, int cpu);
static int kseq_find(void);
static void kseq_notify(struct kse *ke, int cpu);
static void kseq_assign(struct kseq *);
static struct kse *kseq_steal(struct kseq *kseq);
- Make some context switches conditional on SCHED_STRICT_RESCHED. This may have some negative effect on interactivity but it yields great perf. gains. This also brings the conditions under which ULE context switches inline with SCHED_4BSD. - Define some new kseq_* functions for manipulating the run queue. - Add a new kseq member ksq_rslices and ksq_bload. rslices is the sum of the slices of runnable kses. This will be used for push load balance decisions. bload is the number of threads blocked waiting on IO.
2003-02-03 05:30:07 +00:00
#endif
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
void
- Set the ke_cpu field in sched_add() for interrupt and realtime threads since they are going on the current cpu and not their previously assigned cpu. - sched_runnable() should only return true in the SMP case if the other processor has more than one thread that is runnable. We can not steal curthread. - Change kseq_print() to accept the cpuid instead of a kseq pointer. This makes use of this function in ddb much easier.
2003-04-18 05:24:10 +00:00
kseq_print(int cpu)
- Make some context switches conditional on SCHED_STRICT_RESCHED. This may have some negative effect on interactivity but it yields great perf. gains. This also brings the conditions under which ULE context switches inline with SCHED_4BSD. - Define some new kseq_* functions for manipulating the run queue. - Add a new kseq member ksq_rslices and ksq_bload. rslices is the sum of the slices of runnable kses. This will be used for push load balance decisions. bload is the number of threads blocked waiting on IO.
2003-02-03 05:30:07 +00:00
{
- Set the ke_cpu field in sched_add() for interrupt and realtime threads since they are going on the current cpu and not their previously assigned cpu. - sched_runnable() should only return true in the SMP case if the other processor has more than one thread that is runnable. We can not steal curthread. - Change kseq_print() to accept the cpuid instead of a kseq pointer. This makes use of this function in ddb much easier.
2003-04-18 05:24:10 +00:00
struct kseq *kseq;
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
int i;
- Keep seperate statistics and run queues for different scheduling classes. - Treat each class specially in kseq_{choose,add,rem}. Let the rest of the code be less aware of scheduling classes. - Skip the interactivity calculation for non TIMESHARE ksegrps. - Move slice and runq selection into kseq_add(). Uninline it now that it's big.
2003-04-03 00:29:28 +00:00
- Set the ke_cpu field in sched_add() for interrupt and realtime threads since they are going on the current cpu and not their previously assigned cpu. - sched_runnable() should only return true in the SMP case if the other processor has more than one thread that is runnable. We can not steal curthread. - Change kseq_print() to accept the cpuid instead of a kseq pointer. This makes use of this function in ddb much easier.
2003-04-18 05:24:10 +00:00
kseq = KSEQ_CPU(cpu);
- Keep seperate statistics and run queues for different scheduling classes. - Treat each class specially in kseq_{choose,add,rem}. Let the rest of the code be less aware of scheduling classes. - Skip the interactivity calculation for non TIMESHARE ksegrps. - Move slice and runq selection into kseq_add(). Uninline it now that it's big.
2003-04-03 00:29:28 +00:00
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
printf("kseq:\n");
printf("\tload: %d\n", kseq->ksq_load);
printf("\tload ITHD: %d\n", kseq->ksq_loads[PRI_ITHD]);
printf("\tload REALTIME: %d\n", kseq->ksq_loads[PRI_REALTIME]);
printf("\tload TIMESHARE: %d\n", kseq->ksq_loads[PRI_TIMESHARE]);
printf("\tload IDLE: %d\n", kseq->ksq_loads[PRI_IDLE]);
printf("\tnicemin:\t%d\n", kseq->ksq_nicemin);
printf("\tnice counts:\n");
for (i = 0; i < PRIO_TOTAL + 1; i++)
if (kseq->ksq_nice[i])
printf("\t\t%d = %d\n",
i - SCHED_PRI_NHALF, kseq->ksq_nice[i]);
}
- Keep seperate statistics and run queues for different scheduling classes. - Treat each class specially in kseq_{choose,add,rem}. Let the rest of the code be less aware of scheduling classes. - Skip the interactivity calculation for non TIMESHARE ksegrps. - Move slice and runq selection into kseq_add(). Uninline it now that it's big.
2003-04-03 00:29:28 +00:00
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
static void
kseq_add(struct kseq *kseq, struct kse *ke)
{
- When a new thread is added to a kseq the load is incremented prior to adding it to the nice tables. Therefore, in kseq_add_nice, we should keep in mind that the load will be 1 if we are the only thread, and not 0. - Assert that the sched lock is held in all the appropriate places. - Increase the scope of the sched lock in sched_pctcpu_update(). - Hold the sched lock in sched_runnable(). It is not held by the caller.
2003-06-08 00:47:33 +00:00
mtx_assert(&sched_lock, MA_OWNED);
- We only care about the base priority. Ignore the SCHED_FIFO_BIT so that we dont get confused. Reported and debugged by: Steve Kargl <sgk@troutmask.apl.washington.edu>
2003-04-12 07:00:16 +00:00
kseq->ksq_loads[PRI_BASE(ke->ke_ksegrp->kg_pri_class)]++;
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
kseq->ksq_load++;
if (ke->ke_ksegrp->kg_pri_class == PRI_TIMESHARE)
CTR6(KTR_ULE, "Add kse %p to %p (slice: %d, pri: %d, nice: %d(%d))",
ke, ke->ke_runq, ke->ke_slice, ke->ke_thread->td_priority,
ke->ke_ksegrp->kg_nice, kseq->ksq_nicemin);
if (ke->ke_ksegrp->kg_pri_class == PRI_TIMESHARE)
kseq_nice_add(kseq, ke->ke_ksegrp->kg_nice);
- Make some context switches conditional on SCHED_STRICT_RESCHED. This may have some negative effect on interactivity but it yields great perf. gains. This also brings the conditions under which ULE context switches inline with SCHED_4BSD. - Define some new kseq_* functions for manipulating the run queue. - Add a new kseq member ksq_rslices and ksq_bload. rslices is the sum of the slices of runnable kses. This will be used for push load balance decisions. bload is the number of threads blocked waiting on IO.
2003-02-03 05:30:07 +00:00
#ifdef SMP
kseq->ksq_rslices += ke->ke_slice;
#endif
}
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
- Keep seperate statistics and run queues for different scheduling classes. - Treat each class specially in kseq_{choose,add,rem}. Let the rest of the code be less aware of scheduling classes. - Skip the interactivity calculation for non TIMESHARE ksegrps. - Move slice and runq selection into kseq_add(). Uninline it now that it's big.
2003-04-03 00:29:28 +00:00
static void
- Make some context switches conditional on SCHED_STRICT_RESCHED. This may have some negative effect on interactivity but it yields great perf. gains. This also brings the conditions under which ULE context switches inline with SCHED_4BSD. - Define some new kseq_* functions for manipulating the run queue. - Add a new kseq member ksq_rslices and ksq_bload. rslices is the sum of the slices of runnable kses. This will be used for push load balance decisions. bload is the number of threads blocked waiting on IO.
2003-02-03 05:30:07 +00:00
kseq_rem(struct kseq *kseq, struct kse *ke)
{
- When a new thread is added to a kseq the load is incremented prior to adding it to the nice tables. Therefore, in kseq_add_nice, we should keep in mind that the load will be 1 if we are the only thread, and not 0. - Assert that the sched lock is held in all the appropriate places. - Increase the scope of the sched lock in sched_pctcpu_update(). - Hold the sched lock in sched_runnable(). It is not held by the caller.
2003-06-08 00:47:33 +00:00
mtx_assert(&sched_lock, MA_OWNED);
- We only care about the base priority. Ignore the SCHED_FIFO_BIT so that we dont get confused. Reported and debugged by: Steve Kargl <sgk@troutmask.apl.washington.edu>
2003-04-12 07:00:16 +00:00
kseq->ksq_loads[PRI_BASE(ke->ke_ksegrp->kg_pri_class)]--;
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
kseq->ksq_load--;
ke->ke_runq = NULL;
if (ke->ke_ksegrp->kg_pri_class == PRI_TIMESHARE)
kseq_nice_rem(kseq, ke->ke_ksegrp->kg_nice);
- Make some context switches conditional on SCHED_STRICT_RESCHED. This may have some negative effect on interactivity but it yields great perf. gains. This also brings the conditions under which ULE context switches inline with SCHED_4BSD. - Define some new kseq_* functions for manipulating the run queue. - Add a new kseq member ksq_rslices and ksq_bload. rslices is the sum of the slices of runnable kses. This will be used for push load balance decisions. bload is the number of threads blocked waiting on IO.
2003-02-03 05:30:07 +00:00
#ifdef SMP
kseq->ksq_rslices -= ke->ke_slice;
#endif
}
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
static void
kseq_nice_add(struct kseq *kseq, int nice)
- Make some context switches conditional on SCHED_STRICT_RESCHED. This may have some negative effect on interactivity but it yields great perf. gains. This also brings the conditions under which ULE context switches inline with SCHED_4BSD. - Define some new kseq_* functions for manipulating the run queue. - Add a new kseq member ksq_rslices and ksq_bload. rslices is the sum of the slices of runnable kses. This will be used for push load balance decisions. bload is the number of threads blocked waiting on IO.
2003-02-03 05:30:07 +00:00
{
- When a new thread is added to a kseq the load is incremented prior to adding it to the nice tables. Therefore, in kseq_add_nice, we should keep in mind that the load will be 1 if we are the only thread, and not 0. - Assert that the sched lock is held in all the appropriate places. - Increase the scope of the sched lock in sched_pctcpu_update(). - Hold the sched lock in sched_runnable(). It is not held by the caller.
2003-06-08 00:47:33 +00:00
mtx_assert(&sched_lock, MA_OWNED);
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
/* Normalize to zero. */
kseq->ksq_nice[nice + SCHED_PRI_NHALF]++;
- When a new thread is added to a kseq the load is incremented prior to adding it to the nice tables. Therefore, in kseq_add_nice, we should keep in mind that the load will be 1 if we are the only thread, and not 0. - Assert that the sched lock is held in all the appropriate places. - Increase the scope of the sched lock in sched_pctcpu_update(). - Hold the sched lock in sched_runnable(). It is not held by the caller.
2003-06-08 00:47:33 +00:00
if (nice < kseq->ksq_nicemin || kseq->ksq_loads[PRI_TIMESHARE] == 1)
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
kseq->ksq_nicemin = nice;
- Make some context switches conditional on SCHED_STRICT_RESCHED. This may have some negative effect on interactivity but it yields great perf. gains. This also brings the conditions under which ULE context switches inline with SCHED_4BSD. - Define some new kseq_* functions for manipulating the run queue. - Add a new kseq member ksq_rslices and ksq_bload. rslices is the sum of the slices of runnable kses. This will be used for push load balance decisions. bload is the number of threads blocked waiting on IO.
2003-02-03 05:30:07 +00:00
}
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
static void
kseq_nice_rem(struct kseq *kseq, int nice)
- Make some context switches conditional on SCHED_STRICT_RESCHED. This may have some negative effect on interactivity but it yields great perf. gains. This also brings the conditions under which ULE context switches inline with SCHED_4BSD. - Define some new kseq_* functions for manipulating the run queue. - Add a new kseq member ksq_rslices and ksq_bload. rslices is the sum of the slices of runnable kses. This will be used for push load balance decisions. bload is the number of threads blocked waiting on IO.
2003-02-03 05:30:07 +00:00
{
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
int n;
- When a new thread is added to a kseq the load is incremented prior to adding it to the nice tables. Therefore, in kseq_add_nice, we should keep in mind that the load will be 1 if we are the only thread, and not 0. - Assert that the sched lock is held in all the appropriate places. - Increase the scope of the sched lock in sched_pctcpu_update(). - Hold the sched lock in sched_runnable(). It is not held by the caller.
2003-06-08 00:47:33 +00:00
mtx_assert(&sched_lock, MA_OWNED);
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
/* Normalize to zero. */
n = nice + SCHED_PRI_NHALF;
kseq->ksq_nice[n]--;
KASSERT(kseq->ksq_nice[n] >= 0, ("Negative nice count."));
/*
* If this wasn't the smallest nice value or there are more in
* this bucket we can just return. Otherwise we have to recalculate
* the smallest nice.
*/
if (nice != kseq->ksq_nicemin ||
kseq->ksq_nice[n] != 0 ||
kseq->ksq_loads[PRI_TIMESHARE] == 0)
return;
for (; n < SCHED_PRI_NRESV + 1; n++)
if (kseq->ksq_nice[n]) {
kseq->ksq_nicemin = n - SCHED_PRI_NHALF;
return;
}
- Make some context switches conditional on SCHED_STRICT_RESCHED. This may have some negative effect on interactivity but it yields great perf. gains. This also brings the conditions under which ULE context switches inline with SCHED_4BSD. - Define some new kseq_* functions for manipulating the run queue. - Add a new kseq member ksq_rslices and ksq_bload. rslices is the sum of the slices of runnable kses. This will be used for push load balance decisions. bload is the number of threads blocked waiting on IO.
2003-02-03 05:30:07 +00:00
}
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
#ifdef SMP
- Add a simple CPU load balancing algorithm. This works by executing once a second and equalizing the load between the two most imbalanced CPU. This is intended to clear up long term load imbalances that would not be handled by the 'pull' method in sched_choose(). - Pull out some bits of sched_choose() into a kseq_move() function that moves an arbitrary thread from one kseq to another.
2003-06-09 00:39:09 +00:00
/*
* kseq_balance is a simple CPU load balancing algorithm. It operates by
* finding the least loaded and most loaded cpu and equalizing their load
* by migrating some processes.
*
* Dealing only with two CPUs at a time has two advantages. Firstly, most
* installations will only have 2 cpus. Secondly, load balancing too much at
* once can have an unpleasant effect on the system. The scheduler rarely has
* enough information to make perfect decisions. So this algorithm chooses
* algorithm simplicity and more gradual effects on load in larger systems.
*
* It could be improved by considering the priorities and slices assigned to
* each task prior to balancing them. There are many pathological cases with
* any approach and so the semi random algorithm below may work as well as any.
*
*/
- Add static to local functions and data where it was missing. - Add an IPI based mechanism for migrating kses. This mechanism is broken down into several components. This is intended to reduce cache thrashing by eliminating most cases where one cpu touches another's run queues. - kseq_notify() appends a kse to a lockless singly linked list and conditionally sends an IPI to the target processor. Right now this is protected by sched_lock but at some point I'd like to get rid of the global lock. This is why I used something more complicated than a standard queue. - kseq_assign() processes our list of kses that have been assigned to us by other processors. This simply calls sched_add() for each item on the list after clearing the new KEF_ASSIGNED flag. This flag is used to indicate that we have been appeneded to the assigned queue but not added to the run queue yet. - In sched_add(), instead of adding a KSE to another processor's queue we use kse_notify() so that we don't touch their queue. Also in sched_add(), if KEF_ASSIGNED is already set return immediately. This can happen if a thread is removed and readded so that the priority is recorded properly. - In sched_rem() return immediately if KEF_ASSIGNED is set. All callers immediately readd simply to adjust priorites etc. - In sched_choose(), if we're running an IDLE task or the per cpu idle thread set our cpumask bit in 'kseq_idle' so that other processors may know that we are idle. Before this, make a single pass through the run queues of other processors so that we may find work more immediately if it is available. - In sched_runnable(), don't scan each processor's run queue, they will IPI us if they have work for us to do. - In sched_add(), if we're adding a thread that can be migrated and we have plenty of work to do, try to migrate the thread to an idle kseq. - Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into consideration. - No longer use kseq_choose() to steal threads, it can lose it's last argument. - Create a new function runq_steal() which operates like runq_choose() but skips threads based on some criteria. Currently it will not steal PRI_ITHD threads. In the future this will be used for CPU binding. - Create a kseq_steal() that checks each run queue with runq_steal(), use kseq_steal() in the places where we used kseq_choose() to steal with before.
2003-10-31 11:16:04 +00:00
static void
- Add a simple CPU load balancing algorithm. This works by executing once a second and equalizing the load between the two most imbalanced CPU. This is intended to clear up long term load imbalances that would not be handled by the 'pull' method in sched_choose(). - Pull out some bits of sched_choose() into a kseq_move() function that moves an arbitrary thread from one kseq to another.
2003-06-09 00:39:09 +00:00
kseq_balance(void *arg)
{
struct kseq *kseq;
int high_load;
int low_load;
int high_cpu;
int low_cpu;
int move;
int diff;
int i;
high_cpu = 0;
low_cpu = 0;
high_load = 0;
low_load = -1;
mtx_lock_spin(&sched_lock);
- If smp is not started yet don't try to load balance or we'll put threads on cpus that aren't running yet.
2003-06-28 08:24:42 +00:00
if (smp_started == 0)
goto out;
- Add a simple CPU load balancing algorithm. This works by executing once a second and equalizing the load between the two most imbalanced CPU. This is intended to clear up long term load imbalances that would not be handled by the 'pull' method in sched_choose(). - Pull out some bits of sched_choose() into a kseq_move() function that moves an arbitrary thread from one kseq to another.
2003-06-09 00:39:09 +00:00
for (i = 0; i < mp_maxid; i++) {
- Don't migrate to stopped cpus.
2003-06-28 09:09:33 +00:00
if (CPU_ABSENT(i) || (i & stopped_cpus) != 0)
- Add a simple CPU load balancing algorithm. This works by executing once a second and equalizing the load between the two most imbalanced CPU. This is intended to clear up long term load imbalances that would not be handled by the 'pull' method in sched_choose(). - Pull out some bits of sched_choose() into a kseq_move() function that moves an arbitrary thread from one kseq to another.
2003-06-09 00:39:09 +00:00
continue;
kseq = KSEQ_CPU(i);
if (kseq->ksq_load > high_load) {
high_load = kseq->ksq_load;
high_cpu = i;
}
if (low_load == -1 || kseq->ksq_load < low_load) {
low_load = kseq->ksq_load;
low_cpu = i;
}
}
- Parse the cpu topology map in sched_setup(). - Associate logical CPUs on the same physical core with the same kseq. - Adjust code that assumed there would only be one running thread in any kseq. - Wrap the HTT code with a ULE_HTT_EXPERIMENTAL ifdef. This is a start towards HyperThreading support but it isn't quite there yet.
2003-07-04 19:59:00 +00:00
kseq = KSEQ_CPU(high_cpu);
- Add static to local functions and data where it was missing. - Add an IPI based mechanism for migrating kses. This mechanism is broken down into several components. This is intended to reduce cache thrashing by eliminating most cases where one cpu touches another's run queues. - kseq_notify() appends a kse to a lockless singly linked list and conditionally sends an IPI to the target processor. Right now this is protected by sched_lock but at some point I'd like to get rid of the global lock. This is why I used something more complicated than a standard queue. - kseq_assign() processes our list of kses that have been assigned to us by other processors. This simply calls sched_add() for each item on the list after clearing the new KEF_ASSIGNED flag. This flag is used to indicate that we have been appeneded to the assigned queue but not added to the run queue yet. - In sched_add(), instead of adding a KSE to another processor's queue we use kse_notify() so that we don't touch their queue. Also in sched_add(), if KEF_ASSIGNED is already set return immediately. This can happen if a thread is removed and readded so that the priority is recorded properly. - In sched_rem() return immediately if KEF_ASSIGNED is set. All callers immediately readd simply to adjust priorites etc. - In sched_choose(), if we're running an IDLE task or the per cpu idle thread set our cpumask bit in 'kseq_idle' so that other processors may know that we are idle. Before this, make a single pass through the run queues of other processors so that we may find work more immediately if it is available. - In sched_runnable(), don't scan each processor's run queue, they will IPI us if they have work for us to do. - In sched_add(), if we're adding a thread that can be migrated and we have plenty of work to do, try to migrate the thread to an idle kseq. - Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into consideration. - No longer use kseq_choose() to steal threads, it can lose it's last argument. - Create a new function runq_steal() which operates like runq_choose() but skips threads based on some criteria. Currently it will not steal PRI_ITHD threads. In the future this will be used for CPU binding. - Create a kseq_steal() that checks each run queue with runq_steal(), use kseq_steal() in the places where we used kseq_choose() to steal with before.
2003-10-31 11:16:04 +00:00
high_load = kseq->ksq_loads[PRI_IDLE] + kseq->ksq_loads[PRI_TIMESHARE] +
kseq->ksq_loads[PRI_REALTIME];
- Add a simple CPU load balancing algorithm. This works by executing once a second and equalizing the load between the two most imbalanced CPU. This is intended to clear up long term load imbalances that would not be handled by the 'pull' method in sched_choose(). - Pull out some bits of sched_choose() into a kseq_move() function that moves an arbitrary thread from one kseq to another.
2003-06-09 00:39:09 +00:00
/*
* Nothing to do.
*/
- Parse the cpu topology map in sched_setup(). - Associate logical CPUs on the same physical core with the same kseq. - Adjust code that assumed there would only be one running thread in any kseq. - Wrap the HTT code with a ULE_HTT_EXPERIMENTAL ifdef. This is a start towards HyperThreading support but it isn't quite there yet.
2003-07-04 19:59:00 +00:00
if (high_load < kseq->ksq_cpus + 1)
goto out;
high_load -= kseq->ksq_cpus;
if (low_load >= high_load)
- Add a simple CPU load balancing algorithm. This works by executing once a second and equalizing the load between the two most imbalanced CPU. This is intended to clear up long term load imbalances that would not be handled by the 'pull' method in sched_choose(). - Pull out some bits of sched_choose() into a kseq_move() function that moves an arbitrary thread from one kseq to another.
2003-06-09 00:39:09 +00:00
goto out;
diff = high_load - low_load;
move = diff / 2;
if (diff & 0x1)
move++;
for (i = 0; i < move; i++)
- Parse the cpu topology map in sched_setup(). - Associate logical CPUs on the same physical core with the same kseq. - Adjust code that assumed there would only be one running thread in any kseq. - Wrap the HTT code with a ULE_HTT_EXPERIMENTAL ifdef. This is a start towards HyperThreading support but it isn't quite there yet.
2003-07-04 19:59:00 +00:00
kseq_move(kseq, low_cpu);
- Add a simple CPU load balancing algorithm. This works by executing once a second and equalizing the load between the two most imbalanced CPU. This is intended to clear up long term load imbalances that would not be handled by the 'pull' method in sched_choose(). - Pull out some bits of sched_choose() into a kseq_move() function that moves an arbitrary thread from one kseq to another.
2003-06-09 00:39:09 +00:00
out:
mtx_unlock_spin(&sched_lock);
callout_reset(&kseq_lb_callout, hz, kseq_balance, NULL);
return;
}
- Add static to local functions and data where it was missing. - Add an IPI based mechanism for migrating kses. This mechanism is broken down into several components. This is intended to reduce cache thrashing by eliminating most cases where one cpu touches another's run queues. - kseq_notify() appends a kse to a lockless singly linked list and conditionally sends an IPI to the target processor. Right now this is protected by sched_lock but at some point I'd like to get rid of the global lock. This is why I used something more complicated than a standard queue. - kseq_assign() processes our list of kses that have been assigned to us by other processors. This simply calls sched_add() for each item on the list after clearing the new KEF_ASSIGNED flag. This flag is used to indicate that we have been appeneded to the assigned queue but not added to the run queue yet. - In sched_add(), instead of adding a KSE to another processor's queue we use kse_notify() so that we don't touch their queue. Also in sched_add(), if KEF_ASSIGNED is already set return immediately. This can happen if a thread is removed and readded so that the priority is recorded properly. - In sched_rem() return immediately if KEF_ASSIGNED is set. All callers immediately readd simply to adjust priorites etc. - In sched_choose(), if we're running an IDLE task or the per cpu idle thread set our cpumask bit in 'kseq_idle' so that other processors may know that we are idle. Before this, make a single pass through the run queues of other processors so that we may find work more immediately if it is available. - In sched_runnable(), don't scan each processor's run queue, they will IPI us if they have work for us to do. - In sched_add(), if we're adding a thread that can be migrated and we have plenty of work to do, try to migrate the thread to an idle kseq. - Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into consideration. - No longer use kseq_choose() to steal threads, it can lose it's last argument. - Create a new function runq_steal() which operates like runq_choose() but skips threads based on some criteria. Currently it will not steal PRI_ITHD threads. In the future this will be used for CPU binding. - Create a kseq_steal() that checks each run queue with runq_steal(), use kseq_steal() in the places where we used kseq_choose() to steal with before.
2003-10-31 11:16:04 +00:00
static struct kseq *
- Make some context switches conditional on SCHED_STRICT_RESCHED. This may have some negative effect on interactivity but it yields great perf. gains. This also brings the conditions under which ULE context switches inline with SCHED_4BSD. - Define some new kseq_* functions for manipulating the run queue. - Add a new kseq member ksq_rslices and ksq_bload. rslices is the sum of the slices of runnable kses. This will be used for push load balance decisions. bload is the number of threads blocked waiting on IO.
2003-02-03 05:30:07 +00:00
kseq_load_highest(void)
{
struct kseq *kseq;
int load;
int cpu;
int i;
- When a new thread is added to a kseq the load is incremented prior to adding it to the nice tables. Therefore, in kseq_add_nice, we should keep in mind that the load will be 1 if we are the only thread, and not 0. - Assert that the sched lock is held in all the appropriate places. - Increase the scope of the sched lock in sched_pctcpu_update(). - Hold the sched lock in sched_runnable(). It is not held by the caller.
2003-06-08 00:47:33 +00:00
mtx_assert(&sched_lock, MA_OWNED);
- Make some context switches conditional on SCHED_STRICT_RESCHED. This may have some negative effect on interactivity but it yields great perf. gains. This also brings the conditions under which ULE context switches inline with SCHED_4BSD. - Define some new kseq_* functions for manipulating the run queue. - Add a new kseq member ksq_rslices and ksq_bload. rslices is the sum of the slices of runnable kses. This will be used for push load balance decisions. bload is the number of threads blocked waiting on IO.
2003-02-03 05:30:07 +00:00
cpu = 0;
load = 0;
for (i = 0; i < mp_maxid; i++) {
- Don't migrate to stopped cpus.
2003-06-28 09:09:33 +00:00
if (CPU_ABSENT(i) || (i & stopped_cpus) != 0)
- Make some context switches conditional on SCHED_STRICT_RESCHED. This may have some negative effect on interactivity but it yields great perf. gains. This also brings the conditions under which ULE context switches inline with SCHED_4BSD. - Define some new kseq_* functions for manipulating the run queue. - Add a new kseq member ksq_rslices and ksq_bload. rslices is the sum of the slices of runnable kses. This will be used for push load balance decisions. bload is the number of threads blocked waiting on IO.
2003-02-03 05:30:07 +00:00
continue;
kseq = KSEQ_CPU(i);
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
if (kseq->ksq_load > load) {
load = kseq->ksq_load;
- Make some context switches conditional on SCHED_STRICT_RESCHED. This may have some negative effect on interactivity but it yields great perf. gains. This also brings the conditions under which ULE context switches inline with SCHED_4BSD. - Define some new kseq_* functions for manipulating the run queue. - Add a new kseq member ksq_rslices and ksq_bload. rslices is the sum of the slices of runnable kses. This will be used for push load balance decisions. bload is the number of threads blocked waiting on IO.
2003-02-03 05:30:07 +00:00
cpu = i;
}
}
- Parse the cpu topology map in sched_setup(). - Associate logical CPUs on the same physical core with the same kseq. - Adjust code that assumed there would only be one running thread in any kseq. - Wrap the HTT code with a ULE_HTT_EXPERIMENTAL ifdef. This is a start towards HyperThreading support but it isn't quite there yet.
2003-07-04 19:59:00 +00:00
kseq = KSEQ_CPU(cpu);
- Add static to local functions and data where it was missing. - Add an IPI based mechanism for migrating kses. This mechanism is broken down into several components. This is intended to reduce cache thrashing by eliminating most cases where one cpu touches another's run queues. - kseq_notify() appends a kse to a lockless singly linked list and conditionally sends an IPI to the target processor. Right now this is protected by sched_lock but at some point I'd like to get rid of the global lock. This is why I used something more complicated than a standard queue. - kseq_assign() processes our list of kses that have been assigned to us by other processors. This simply calls sched_add() for each item on the list after clearing the new KEF_ASSIGNED flag. This flag is used to indicate that we have been appeneded to the assigned queue but not added to the run queue yet. - In sched_add(), instead of adding a KSE to another processor's queue we use kse_notify() so that we don't touch their queue. Also in sched_add(), if KEF_ASSIGNED is already set return immediately. This can happen if a thread is removed and readded so that the priority is recorded properly. - In sched_rem() return immediately if KEF_ASSIGNED is set. All callers immediately readd simply to adjust priorites etc. - In sched_choose(), if we're running an IDLE task or the per cpu idle thread set our cpumask bit in 'kseq_idle' so that other processors may know that we are idle. Before this, make a single pass through the run queues of other processors so that we may find work more immediately if it is available. - In sched_runnable(), don't scan each processor's run queue, they will IPI us if they have work for us to do. - In sched_add(), if we're adding a thread that can be migrated and we have plenty of work to do, try to migrate the thread to an idle kseq. - Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into consideration. - No longer use kseq_choose() to steal threads, it can lose it's last argument. - Create a new function runq_steal() which operates like runq_choose() but skips threads based on some criteria. Currently it will not steal PRI_ITHD threads. In the future this will be used for CPU binding. - Create a kseq_steal() that checks each run queue with runq_steal(), use kseq_steal() in the places where we used kseq_choose() to steal with before.
2003-10-31 11:16:04 +00:00
if ((kseq->ksq_loads[PRI_IDLE] + kseq->ksq_loads[PRI_TIMESHARE] +
kseq->ksq_loads[PRI_REALTIME]) > kseq->ksq_cpus)
- Parse the cpu topology map in sched_setup(). - Associate logical CPUs on the same physical core with the same kseq. - Adjust code that assumed there would only be one running thread in any kseq. - Wrap the HTT code with a ULE_HTT_EXPERIMENTAL ifdef. This is a start towards HyperThreading support but it isn't quite there yet.
2003-07-04 19:59:00 +00:00
return (kseq);
- Make some context switches conditional on SCHED_STRICT_RESCHED. This may have some negative effect on interactivity but it yields great perf. gains. This also brings the conditions under which ULE context switches inline with SCHED_4BSD. - Define some new kseq_* functions for manipulating the run queue. - Add a new kseq member ksq_rslices and ksq_bload. rslices is the sum of the slices of runnable kses. This will be used for push load balance decisions. bload is the number of threads blocked waiting on IO.
2003-02-03 05:30:07 +00:00
return (NULL);
}
- Add a simple CPU load balancing algorithm. This works by executing once a second and equalizing the load between the two most imbalanced CPU. This is intended to clear up long term load imbalances that would not be handled by the 'pull' method in sched_choose(). - Pull out some bits of sched_choose() into a kseq_move() function that moves an arbitrary thread from one kseq to another.
2003-06-09 00:39:09 +00:00
- Add static to local functions and data where it was missing. - Add an IPI based mechanism for migrating kses. This mechanism is broken down into several components. This is intended to reduce cache thrashing by eliminating most cases where one cpu touches another's run queues. - kseq_notify() appends a kse to a lockless singly linked list and conditionally sends an IPI to the target processor. Right now this is protected by sched_lock but at some point I'd like to get rid of the global lock. This is why I used something more complicated than a standard queue. - kseq_assign() processes our list of kses that have been assigned to us by other processors. This simply calls sched_add() for each item on the list after clearing the new KEF_ASSIGNED flag. This flag is used to indicate that we have been appeneded to the assigned queue but not added to the run queue yet. - In sched_add(), instead of adding a KSE to another processor's queue we use kse_notify() so that we don't touch their queue. Also in sched_add(), if KEF_ASSIGNED is already set return immediately. This can happen if a thread is removed and readded so that the priority is recorded properly. - In sched_rem() return immediately if KEF_ASSIGNED is set. All callers immediately readd simply to adjust priorites etc. - In sched_choose(), if we're running an IDLE task or the per cpu idle thread set our cpumask bit in 'kseq_idle' so that other processors may know that we are idle. Before this, make a single pass through the run queues of other processors so that we may find work more immediately if it is available. - In sched_runnable(), don't scan each processor's run queue, they will IPI us if they have work for us to do. - In sched_add(), if we're adding a thread that can be migrated and we have plenty of work to do, try to migrate the thread to an idle kseq. - Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into consideration. - No longer use kseq_choose() to steal threads, it can lose it's last argument. - Create a new function runq_steal() which operates like runq_choose() but skips threads based on some criteria. Currently it will not steal PRI_ITHD threads. In the future this will be used for CPU binding. - Create a kseq_steal() that checks each run queue with runq_steal(), use kseq_steal() in the places where we used kseq_choose() to steal with before.
2003-10-31 11:16:04 +00:00
static void
- Add a simple CPU load balancing algorithm. This works by executing once a second and equalizing the load between the two most imbalanced CPU. This is intended to clear up long term load imbalances that would not be handled by the 'pull' method in sched_choose(). - Pull out some bits of sched_choose() into a kseq_move() function that moves an arbitrary thread from one kseq to another.
2003-06-09 00:39:09 +00:00
kseq_move(struct kseq *from, int cpu)
{
struct kse *ke;
- Add static to local functions and data where it was missing. - Add an IPI based mechanism for migrating kses. This mechanism is broken down into several components. This is intended to reduce cache thrashing by eliminating most cases where one cpu touches another's run queues. - kseq_notify() appends a kse to a lockless singly linked list and conditionally sends an IPI to the target processor. Right now this is protected by sched_lock but at some point I'd like to get rid of the global lock. This is why I used something more complicated than a standard queue. - kseq_assign() processes our list of kses that have been assigned to us by other processors. This simply calls sched_add() for each item on the list after clearing the new KEF_ASSIGNED flag. This flag is used to indicate that we have been appeneded to the assigned queue but not added to the run queue yet. - In sched_add(), instead of adding a KSE to another processor's queue we use kse_notify() so that we don't touch their queue. Also in sched_add(), if KEF_ASSIGNED is already set return immediately. This can happen if a thread is removed and readded so that the priority is recorded properly. - In sched_rem() return immediately if KEF_ASSIGNED is set. All callers immediately readd simply to adjust priorites etc. - In sched_choose(), if we're running an IDLE task or the per cpu idle thread set our cpumask bit in 'kseq_idle' so that other processors may know that we are idle. Before this, make a single pass through the run queues of other processors so that we may find work more immediately if it is available. - In sched_runnable(), don't scan each processor's run queue, they will IPI us if they have work for us to do. - In sched_add(), if we're adding a thread that can be migrated and we have plenty of work to do, try to migrate the thread to an idle kseq. - Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into consideration. - No longer use kseq_choose() to steal threads, it can lose it's last argument. - Create a new function runq_steal() which operates like runq_choose() but skips threads based on some criteria. Currently it will not steal PRI_ITHD threads. In the future this will be used for CPU binding. - Create a kseq_steal() that checks each run queue with runq_steal(), use kseq_steal() in the places where we used kseq_choose() to steal with before.
2003-10-31 11:16:04 +00:00
ke = kseq_steal(from);
- Add a simple CPU load balancing algorithm. This works by executing once a second and equalizing the load between the two most imbalanced CPU. This is intended to clear up long term load imbalances that would not be handled by the 'pull' method in sched_choose(). - Pull out some bits of sched_choose() into a kseq_move() function that moves an arbitrary thread from one kseq to another.
2003-06-09 00:39:09 +00:00
runq_remove(ke->ke_runq, ke);
ke->ke_state = KES_THREAD;
kseq_rem(from, ke);
ke->ke_cpu = cpu;
- Call sched_add() with the correct argument on SMP. Reported by: Valentin Chopov <valentin@valcho.net>
2003-10-16 20:06:19 +00:00
sched_add(ke->ke_thread);
- Add a simple CPU load balancing algorithm. This works by executing once a second and equalizing the load between the two most imbalanced CPU. This is intended to clear up long term load imbalances that would not be handled by the 'pull' method in sched_choose(). - Pull out some bits of sched_choose() into a kseq_move() function that moves an arbitrary thread from one kseq to another.
2003-06-09 00:39:09 +00:00
}
- Add static to local functions and data where it was missing. - Add an IPI based mechanism for migrating kses. This mechanism is broken down into several components. This is intended to reduce cache thrashing by eliminating most cases where one cpu touches another's run queues. - kseq_notify() appends a kse to a lockless singly linked list and conditionally sends an IPI to the target processor. Right now this is protected by sched_lock but at some point I'd like to get rid of the global lock. This is why I used something more complicated than a standard queue. - kseq_assign() processes our list of kses that have been assigned to us by other processors. This simply calls sched_add() for each item on the list after clearing the new KEF_ASSIGNED flag. This flag is used to indicate that we have been appeneded to the assigned queue but not added to the run queue yet. - In sched_add(), instead of adding a KSE to another processor's queue we use kse_notify() so that we don't touch their queue. Also in sched_add(), if KEF_ASSIGNED is already set return immediately. This can happen if a thread is removed and readded so that the priority is recorded properly. - In sched_rem() return immediately if KEF_ASSIGNED is set. All callers immediately readd simply to adjust priorites etc. - In sched_choose(), if we're running an IDLE task or the per cpu idle thread set our cpumask bit in 'kseq_idle' so that other processors may know that we are idle. Before this, make a single pass through the run queues of other processors so that we may find work more immediately if it is available. - In sched_runnable(), don't scan each processor's run queue, they will IPI us if they have work for us to do. - In sched_add(), if we're adding a thread that can be migrated and we have plenty of work to do, try to migrate the thread to an idle kseq. - Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into consideration. - No longer use kseq_choose() to steal threads, it can lose it's last argument. - Create a new function runq_steal() which operates like runq_choose() but skips threads based on some criteria. Currently it will not steal PRI_ITHD threads. In the future this will be used for CPU binding. - Create a kseq_steal() that checks each run queue with runq_steal(), use kseq_steal() in the places where we used kseq_choose() to steal with before.
2003-10-31 11:16:04 +00:00
static int
kseq_find(void)
{
struct kseq *high;
if (!smp_started)
return (0);
if (kseq_idle & PCPU_GET(cpumask))
return (0);
/*
* Find the cpu with the highest load and steal one proc.
*/
if ((high = kseq_load_highest()) == NULL ||
high == KSEQ_SELF()) {
/*
* If we couldn't find one, set ourselves in the
* idle map.
*/
atomic_set_int(&kseq_idle, PCPU_GET(cpumask));
return (0);
}
/*
* Remove this kse from this kseq and runq and then requeue
* on the current processor. We now have a load of one!
*/
kseq_move(high, PCPU_GET(cpuid));
return (1);
}
static void
kseq_assign(struct kseq *kseq)
{
struct kse *nke;
struct kse *ke;
do {
ke = kseq->ksq_assigned;
} while(!atomic_cmpset_ptr(&kseq->ksq_assigned, ke, NULL));
for (; ke != NULL; ke = nke) {
nke = ke->ke_assign;
ke->ke_flags &= ~KEF_ASSIGNED;
sched_add(ke->ke_thread);
}
}
static void
kseq_notify(struct kse *ke, int cpu)
{
struct kseq *kseq;
struct thread *td;
struct pcpu *pcpu;
ke->ke_flags |= KEF_ASSIGNED;
kseq = KSEQ_CPU(cpu);
/*
* Place a KSE on another cpu's queue and force a resched.
*/
do {
ke->ke_assign = kseq->ksq_assigned;
} while(!atomic_cmpset_ptr(&kseq->ksq_assigned, ke->ke_assign, ke));
pcpu = pcpu_find(cpu);
td = pcpu->pc_curthread;
if (ke->ke_thread->td_priority < td->td_priority ||
td == pcpu->pc_idlethread) {
td->td_flags |= TDF_NEEDRESCHED;
ipi_selected(1 << cpu, IPI_AST);
}
}
static struct kse *
runq_steal(struct runq *rq)
{
struct rqhead *rqh;
struct rqbits *rqb;
struct kse *ke;
int word;
int bit;
mtx_assert(&sched_lock, MA_OWNED);
rqb = &rq->rq_status;
for (word = 0; word < RQB_LEN; word++) {
if (rqb->rqb_bits[word] == 0)
continue;
for (bit = 0; bit < RQB_BPW; bit++) {
if ((rqb->rqb_bits[word] & (1 << bit)) == 0)
continue;
rqh = &rq->rq_queues[bit + (word << RQB_L2BPW)];
TAILQ_FOREACH(ke, rqh, ke_procq) {
if (PRI_BASE(ke->ke_ksegrp->kg_pri_class) !=
PRI_ITHD)
return (ke);
}
}
}
return (NULL);
}
static struct kse *
kseq_steal(struct kseq *kseq)
{
struct kse *ke;
if ((ke = runq_steal(kseq->ksq_curr)) != NULL)
return (ke);
if ((ke = runq_steal(kseq->ksq_next)) != NULL)
return (ke);
return (runq_steal(&kseq->ksq_idle));
}
#endif /* SMP */
- Make some context switches conditional on SCHED_STRICT_RESCHED. This may have some negative effect on interactivity but it yields great perf. gains. This also brings the conditions under which ULE context switches inline with SCHED_4BSD. - Define some new kseq_* functions for manipulating the run queue. - Add a new kseq member ksq_rslices and ksq_bload. rslices is the sum of the slices of runnable kses. This will be used for push load balance decisions. bload is the number of threads blocked waiting on IO.
2003-02-03 05:30:07 +00:00
- When stealing a kse in kseq_move() ignore the current kseq's min nice value. We want to steal any thread, even one that is not given a slice on its current queue.
2003-07-08 06:19:40 +00:00
/*
- Add static to local functions and data where it was missing. - Add an IPI based mechanism for migrating kses. This mechanism is broken down into several components. This is intended to reduce cache thrashing by eliminating most cases where one cpu touches another's run queues. - kseq_notify() appends a kse to a lockless singly linked list and conditionally sends an IPI to the target processor. Right now this is protected by sched_lock but at some point I'd like to get rid of the global lock. This is why I used something more complicated than a standard queue. - kseq_assign() processes our list of kses that have been assigned to us by other processors. This simply calls sched_add() for each item on the list after clearing the new KEF_ASSIGNED flag. This flag is used to indicate that we have been appeneded to the assigned queue but not added to the run queue yet. - In sched_add(), instead of adding a KSE to another processor's queue we use kse_notify() so that we don't touch their queue. Also in sched_add(), if KEF_ASSIGNED is already set return immediately. This can happen if a thread is removed and readded so that the priority is recorded properly. - In sched_rem() return immediately if KEF_ASSIGNED is set. All callers immediately readd simply to adjust priorites etc. - In sched_choose(), if we're running an IDLE task or the per cpu idle thread set our cpumask bit in 'kseq_idle' so that other processors may know that we are idle. Before this, make a single pass through the run queues of other processors so that we may find work more immediately if it is available. - In sched_runnable(), don't scan each processor's run queue, they will IPI us if they have work for us to do. - In sched_add(), if we're adding a thread that can be migrated and we have plenty of work to do, try to migrate the thread to an idle kseq. - Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into consideration. - No longer use kseq_choose() to steal threads, it can lose it's last argument. - Create a new function runq_steal() which operates like runq_choose() but skips threads based on some criteria. Currently it will not steal PRI_ITHD threads. In the future this will be used for CPU binding. - Create a kseq_steal() that checks each run queue with runq_steal(), use kseq_steal() in the places where we used kseq_choose() to steal with before.
2003-10-31 11:16:04 +00:00
* Pick the highest priority task we have and return it.
- When stealing a kse in kseq_move() ignore the current kseq's min nice value. We want to steal any thread, even one that is not given a slice on its current queue.
2003-07-08 06:19:40 +00:00
*/
- Add static to local functions and data where it was missing. - Add an IPI based mechanism for migrating kses. This mechanism is broken down into several components. This is intended to reduce cache thrashing by eliminating most cases where one cpu touches another's run queues. - kseq_notify() appends a kse to a lockless singly linked list and conditionally sends an IPI to the target processor. Right now this is protected by sched_lock but at some point I'd like to get rid of the global lock. This is why I used something more complicated than a standard queue. - kseq_assign() processes our list of kses that have been assigned to us by other processors. This simply calls sched_add() for each item on the list after clearing the new KEF_ASSIGNED flag. This flag is used to indicate that we have been appeneded to the assigned queue but not added to the run queue yet. - In sched_add(), instead of adding a KSE to another processor's queue we use kse_notify() so that we don't touch their queue. Also in sched_add(), if KEF_ASSIGNED is already set return immediately. This can happen if a thread is removed and readded so that the priority is recorded properly. - In sched_rem() return immediately if KEF_ASSIGNED is set. All callers immediately readd simply to adjust priorites etc. - In sched_choose(), if we're running an IDLE task or the per cpu idle thread set our cpumask bit in 'kseq_idle' so that other processors may know that we are idle. Before this, make a single pass through the run queues of other processors so that we may find work more immediately if it is available. - In sched_runnable(), don't scan each processor's run queue, they will IPI us if they have work for us to do. - In sched_add(), if we're adding a thread that can be migrated and we have plenty of work to do, try to migrate the thread to an idle kseq. - Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into consideration. - No longer use kseq_choose() to steal threads, it can lose it's last argument. - Create a new function runq_steal() which operates like runq_choose() but skips threads based on some criteria. Currently it will not steal PRI_ITHD threads. In the future this will be used for CPU binding. - Create a kseq_steal() that checks each run queue with runq_steal(), use kseq_steal() in the places where we used kseq_choose() to steal with before.
2003-10-31 11:16:04 +00:00
static struct kse *
kseq_choose(struct kseq *kseq)
- Make some context switches conditional on SCHED_STRICT_RESCHED. This may have some negative effect on interactivity but it yields great perf. gains. This also brings the conditions under which ULE context switches inline with SCHED_4BSD. - Define some new kseq_* functions for manipulating the run queue. - Add a new kseq member ksq_rslices and ksq_bload. rslices is the sum of the slices of runnable kses. This will be used for push load balance decisions. bload is the number of threads blocked waiting on IO.
2003-02-03 05:30:07 +00:00
{
struct kse *ke;
struct runq *swap;
- When a new thread is added to a kseq the load is incremented prior to adding it to the nice tables. Therefore, in kseq_add_nice, we should keep in mind that the load will be 1 if we are the only thread, and not 0. - Assert that the sched lock is held in all the appropriate places. - Increase the scope of the sched lock in sched_pctcpu_update(). - Hold the sched lock in sched_runnable(). It is not held by the caller.
2003-06-08 00:47:33 +00:00
mtx_assert(&sched_lock, MA_OWNED);
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
swap = NULL;
for (;;) {
ke = runq_choose(kseq->ksq_curr);
if (ke == NULL) {
/*
* We already swaped once and didn't get anywhere.
*/
if (swap)
break;
swap = kseq->ksq_curr;
kseq->ksq_curr = kseq->ksq_next;
kseq->ksq_next = swap;
continue;
}
/*
* If we encounter a slice of 0 the kse is in a
* TIMESHARE kse group and its nice was too far out
* of the range that receives slices.
*/
- Add static to local functions and data where it was missing. - Add an IPI based mechanism for migrating kses. This mechanism is broken down into several components. This is intended to reduce cache thrashing by eliminating most cases where one cpu touches another's run queues. - kseq_notify() appends a kse to a lockless singly linked list and conditionally sends an IPI to the target processor. Right now this is protected by sched_lock but at some point I'd like to get rid of the global lock. This is why I used something more complicated than a standard queue. - kseq_assign() processes our list of kses that have been assigned to us by other processors. This simply calls sched_add() for each item on the list after clearing the new KEF_ASSIGNED flag. This flag is used to indicate that we have been appeneded to the assigned queue but not added to the run queue yet. - In sched_add(), instead of adding a KSE to another processor's queue we use kse_notify() so that we don't touch their queue. Also in sched_add(), if KEF_ASSIGNED is already set return immediately. This can happen if a thread is removed and readded so that the priority is recorded properly. - In sched_rem() return immediately if KEF_ASSIGNED is set. All callers immediately readd simply to adjust priorites etc. - In sched_choose(), if we're running an IDLE task or the per cpu idle thread set our cpumask bit in 'kseq_idle' so that other processors may know that we are idle. Before this, make a single pass through the run queues of other processors so that we may find work more immediately if it is available. - In sched_runnable(), don't scan each processor's run queue, they will IPI us if they have work for us to do. - In sched_add(), if we're adding a thread that can be migrated and we have plenty of work to do, try to migrate the thread to an idle kseq. - Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into consideration. - No longer use kseq_choose() to steal threads, it can lose it's last argument. - Create a new function runq_steal() which operates like runq_choose() but skips threads based on some criteria. Currently it will not steal PRI_ITHD threads. In the future this will be used for CPU binding. - Create a kseq_steal() that checks each run queue with runq_steal(), use kseq_steal() in the places where we used kseq_choose() to steal with before.
2003-10-31 11:16:04 +00:00
if (ke->ke_slice == 0) {
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
runq_remove(ke->ke_runq, ke);
sched_slice(ke);
ke->ke_runq = kseq->ksq_next;
runq_add(ke->ke_runq, ke);
continue;
}
return (ke);
- Make some context switches conditional on SCHED_STRICT_RESCHED. This may have some negative effect on interactivity but it yields great perf. gains. This also brings the conditions under which ULE context switches inline with SCHED_4BSD. - Define some new kseq_* functions for manipulating the run queue. - Add a new kseq member ksq_rslices and ksq_bload. rslices is the sum of the slices of runnable kses. This will be used for push load balance decisions. bload is the number of threads blocked waiting on IO.
2003-02-03 05:30:07 +00:00
}
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
return (runq_choose(&kseq->ksq_idle));
- Add in support for KSEs with 0 slice values on the run queue. If we try to select a KSE with a slice of 0 we will update its slice and insert it onto the next queue. - Pass the KSE instead of the ksegrp into sched_slice(). This more accurately reflects the behavior of the code. Slices are granted to kses. - Add a function kseq_nice_min() which finds the smallest nice value assigned to the kseg of any KSE on the queue. - Rewrite the logic in sched_slice(). Add a large comment describing the new slice selection scheme. To summarize, slices are assigned based on the nice value. Priorities are still calculated based on the nice and interactivity of a process. Slice sizes of 0 may be granted for KSEs whos nice is 20 or futher away from the lowest nice on the run queue. Other nice values are scaled across the range [min, min+20]. This fixes ULEs bad behavior with positively niced processes.
2003-04-02 06:46:43 +00:00
}
- Use ksq_load as the authoritive count of kses on the pair of kseqs for sched_runnable() et all. - Remove some dead code in sched_clock(). - Define two macros KSEQ_SELF() and KSEQ_CPU() for getting the kseq of the current cpu or some alternate cpu. - Start introducing kseq_() functions, such as kseq_choose() and kseq_setup().
2003-01-29 07:00:51 +00:00
static void
kseq_setup(struct kseq *kseq)
{
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
runq_init(&kseq->ksq_timeshare[0]);
runq_init(&kseq->ksq_timeshare[1]);
- Keep seperate statistics and run queues for different scheduling classes. - Treat each class specially in kseq_{choose,add,rem}. Let the rest of the code be less aware of scheduling classes. - Skip the interactivity calculation for non TIMESHARE ksegrps. - Move slice and runq selection into kseq_add(). Uninline it now that it's big.
2003-04-03 00:29:28 +00:00
runq_init(&kseq->ksq_idle);
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
kseq->ksq_curr = &kseq->ksq_timeshare[0];
kseq->ksq_next = &kseq->ksq_timeshare[1];
kseq->ksq_loads[PRI_ITHD] = 0;
kseq->ksq_loads[PRI_REALTIME] = 0;
kseq->ksq_loads[PRI_TIMESHARE] = 0;
kseq->ksq_loads[PRI_IDLE] = 0;
- Set the ke_cpu field in sched_add() for interrupt and realtime threads since they are going on the current cpu and not their previously assigned cpu. - sched_runnable() should only return true in the SMP case if the other processor has more than one thread that is runnable. We can not steal curthread. - Change kseq_print() to accept the cpuid instead of a kseq pointer. This makes use of this function in ddb much easier.
2003-04-18 05:24:10 +00:00
kseq->ksq_load = 0;
- Make some context switches conditional on SCHED_STRICT_RESCHED. This may have some negative effect on interactivity but it yields great perf. gains. This also brings the conditions under which ULE context switches inline with SCHED_4BSD. - Define some new kseq_* functions for manipulating the run queue. - Add a new kseq member ksq_rslices and ksq_bload. rslices is the sum of the slices of runnable kses. This will be used for push load balance decisions. bload is the number of threads blocked waiting on IO.
2003-02-03 05:30:07 +00:00
#ifdef SMP
kseq->ksq_rslices = 0;
- Add static to local functions and data where it was missing. - Add an IPI based mechanism for migrating kses. This mechanism is broken down into several components. This is intended to reduce cache thrashing by eliminating most cases where one cpu touches another's run queues. - kseq_notify() appends a kse to a lockless singly linked list and conditionally sends an IPI to the target processor. Right now this is protected by sched_lock but at some point I'd like to get rid of the global lock. This is why I used something more complicated than a standard queue. - kseq_assign() processes our list of kses that have been assigned to us by other processors. This simply calls sched_add() for each item on the list after clearing the new KEF_ASSIGNED flag. This flag is used to indicate that we have been appeneded to the assigned queue but not added to the run queue yet. - In sched_add(), instead of adding a KSE to another processor's queue we use kse_notify() so that we don't touch their queue. Also in sched_add(), if KEF_ASSIGNED is already set return immediately. This can happen if a thread is removed and readded so that the priority is recorded properly. - In sched_rem() return immediately if KEF_ASSIGNED is set. All callers immediately readd simply to adjust priorites etc. - In sched_choose(), if we're running an IDLE task or the per cpu idle thread set our cpumask bit in 'kseq_idle' so that other processors may know that we are idle. Before this, make a single pass through the run queues of other processors so that we may find work more immediately if it is available. - In sched_runnable(), don't scan each processor's run queue, they will IPI us if they have work for us to do. - In sched_add(), if we're adding a thread that can be migrated and we have plenty of work to do, try to migrate the thread to an idle kseq. - Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into consideration. - No longer use kseq_choose() to steal threads, it can lose it's last argument. - Create a new function runq_steal() which operates like runq_choose() but skips threads based on some criteria. Currently it will not steal PRI_ITHD threads. In the future this will be used for CPU binding. - Create a kseq_steal() that checks each run queue with runq_steal(), use kseq_steal() in the places where we used kseq_choose() to steal with before.
2003-10-31 11:16:04 +00:00
kseq->ksq_assigned = NULL;
- Make some context switches conditional on SCHED_STRICT_RESCHED. This may have some negative effect on interactivity but it yields great perf. gains. This also brings the conditions under which ULE context switches inline with SCHED_4BSD. - Define some new kseq_* functions for manipulating the run queue. - Add a new kseq member ksq_rslices and ksq_bload. rslices is the sum of the slices of runnable kses. This will be used for push load balance decisions. bload is the number of threads blocked waiting on IO.
2003-02-03 05:30:07 +00:00
#endif
- Use ksq_load as the authoritive count of kses on the pair of kseqs for sched_runnable() et all. - Remove some dead code in sched_clock(). - Define two macros KSEQ_SELF() and KSEQ_CPU() for getting the kseq of the current cpu or some alternate cpu. - Start introducing kseq_() functions, such as kseq_choose() and kseq_setup().
2003-01-29 07:00:51 +00:00
}
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
static void
sched_setup(void *dummy)
{
- Clean up an unused variable. Submitted by: Steve Kargl <skg@routmask.apl.washington.edu>
2003-07-07 21:08:28 +00:00
#ifdef SMP
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
int i;
- Clean up an unused variable. Submitted by: Steve Kargl <skg@routmask.apl.washington.edu>
2003-07-07 21:08:28 +00:00
#endif
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
- Adjust the default maximum slice value to ~140ms. This has improved the nice distribution without significantly impacting interactive response. As a side effect it should also allow batch processes to run for a slightly longer period which will positively impact their performance.
2003-06-28 06:04:47 +00:00
slice_min = (hz/100); /* 10ms */
slice_max = (hz/7); /* ~140ms */
- Create a function sched_interact_score() which decides on the interactivity of a kseg and assigns it a value of 0 through 100. - Use sched_interact_score() to determine the dynamic priority. - Define SCHED_CURR() in terms of sched_interact_score(). - Adjust the maximum slice back down to 100ms. - Remove redundant clearing of ke_runq in sched_wakeup() - Clean up #defines and comment them.
2003-03-04 02:45:59 +00:00
- Parse the cpu topology map in sched_setup(). - Associate logical CPUs on the same physical core with the same kseq. - Adjust code that assumed there would only be one running thread in any kseq. - Wrap the HTT code with a ULE_HTT_EXPERIMENTAL ifdef. This is a start towards HyperThreading support but it isn't quite there yet.
2003-07-04 19:59:00 +00:00
#ifdef SMP
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
/* init kseqs */
- Parse the cpu topology map in sched_setup(). - Associate logical CPUs on the same physical core with the same kseq. - Adjust code that assumed there would only be one running thread in any kseq. - Wrap the HTT code with a ULE_HTT_EXPERIMENTAL ifdef. This is a start towards HyperThreading support but it isn't quite there yet.
2003-07-04 19:59:00 +00:00
/* Create the idmap. */
#ifdef ULE_HTT_EXPERIMENTAL
if (smp_topology == NULL) {
#else
if (1) {
#endif
for (i = 0; i < MAXCPU; i++) {
kseq_setup(&kseq_cpu[i]);
kseq_idmap[i] = &kseq_cpu[i];
kseq_cpu[i].ksq_cpus = 1;
}
} else {
int j;
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
- Parse the cpu topology map in sched_setup(). - Associate logical CPUs on the same physical core with the same kseq. - Adjust code that assumed there would only be one running thread in any kseq. - Wrap the HTT code with a ULE_HTT_EXPERIMENTAL ifdef. This is a start towards HyperThreading support but it isn't quite there yet.
2003-07-04 19:59:00 +00:00
for (i = 0; i < smp_topology->ct_count; i++) {
struct cpu_group *cg;
cg = &smp_topology->ct_group[i];
kseq_setup(&kseq_cpu[i]);
for (j = 0; j < MAXCPU; j++)
if ((cg->cg_mask & (1 << j)) != 0)
kseq_idmap[j] = &kseq_cpu[i];
kseq_cpu[i].ksq_cpus = cg->cg_count;
}
}
Change instances of callout_init that specify MPSAFE behaviour to use CALLOUT_MPSAFE instead of "1" for the second parameter. This does not change the behaviour; it just makes the intent more clear.
2003-08-19 17:51:11 +00:00
callout_init(&kseq_lb_callout, CALLOUT_MPSAFE);
- Add a simple CPU load balancing algorithm. This works by executing once a second and equalizing the load between the two most imbalanced CPU. This is intended to clear up long term load imbalances that would not be handled by the 'pull' method in sched_choose(). - Pull out some bits of sched_choose() into a kseq_move() function that moves an arbitrary thread from one kseq to another.
2003-06-09 00:39:09 +00:00
kseq_balance(NULL);
- Parse the cpu topology map in sched_setup(). - Associate logical CPUs on the same physical core with the same kseq. - Adjust code that assumed there would only be one running thread in any kseq. - Wrap the HTT code with a ULE_HTT_EXPERIMENTAL ifdef. This is a start towards HyperThreading support but it isn't quite there yet.
2003-07-04 19:59:00 +00:00
#else
kseq_setup(KSEQ_SELF());
- Add a simple CPU load balancing algorithm. This works by executing once a second and equalizing the load between the two most imbalanced CPU. This is intended to clear up long term load imbalances that would not be handled by the 'pull' method in sched_choose(). - Pull out some bits of sched_choose() into a kseq_move() function that moves an arbitrary thread from one kseq to another.
2003-06-09 00:39:09 +00:00
#endif
- Parse the cpu topology map in sched_setup(). - Associate logical CPUs on the same physical core with the same kseq. - Adjust code that assumed there would only be one running thread in any kseq. - Wrap the HTT code with a ULE_HTT_EXPERIMENTAL ifdef. This is a start towards HyperThreading support but it isn't quite there yet.
2003-07-04 19:59:00 +00:00
mtx_lock_spin(&sched_lock);
kseq_add(KSEQ_SELF(), &kse0);
mtx_unlock_spin(&sched_lock);
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
}
/*
* Scale the scheduling priority according to the "interactivity" of this
* process.
*/
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
static void
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
sched_priority(struct ksegrp *kg)
{
int pri;
if (kg->kg_pri_class != PRI_TIMESHARE)
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
return;
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
pri = SCHED_PRI_INTERACT(sched_interact_score(kg));
- Create a function sched_interact_score() which decides on the interactivity of a kseg and assigns it a value of 0 through 100. - Use sched_interact_score() to determine the dynamic priority. - Define SCHED_CURR() in terms of sched_interact_score(). - Adjust the maximum slice back down to 100ms. - Remove redundant clearing of ke_runq in sched_wakeup() - Clean up #defines and comment them.
2003-03-04 02:45:59 +00:00
pri += SCHED_PRI_BASE;
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
pri += kg->kg_nice;
if (pri > PRI_MAX_TIMESHARE)
pri = PRI_MAX_TIMESHARE;
else if (pri < PRI_MIN_TIMESHARE)
pri = PRI_MIN_TIMESHARE;
kg->kg_user_pri = pri;
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
return;
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
}
/*
- Add in support for KSEs with 0 slice values on the run queue. If we try to select a KSE with a slice of 0 we will update its slice and insert it onto the next queue. - Pass the KSE instead of the ksegrp into sched_slice(). This more accurately reflects the behavior of the code. Slices are granted to kses. - Add a function kseq_nice_min() which finds the smallest nice value assigned to the kseg of any KSE on the queue. - Rewrite the logic in sched_slice(). Add a large comment describing the new slice selection scheme. To summarize, slices are assigned based on the nice value. Priorities are still calculated based on the nice and interactivity of a process. Slice sizes of 0 may be granted for KSEs whos nice is 20 or futher away from the lowest nice on the run queue. Other nice values are scaled across the range [min, min+20]. This fixes ULEs bad behavior with positively niced processes.
2003-04-02 06:46:43 +00:00
* Calculate a time slice based on the properties of the kseg and the runq
- Keep seperate statistics and run queues for different scheduling classes. - Treat each class specially in kseq_{choose,add,rem}. Let the rest of the code be less aware of scheduling classes. - Skip the interactivity calculation for non TIMESHARE ksegrps. - Move slice and runq selection into kseq_add(). Uninline it now that it's big.
2003-04-03 00:29:28 +00:00
* that we're on. This is only for PRI_TIMESHARE ksegrps.
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
*/
- Add in support for KSEs with 0 slice values on the run queue. If we try to select a KSE with a slice of 0 we will update its slice and insert it onto the next queue. - Pass the KSE instead of the ksegrp into sched_slice(). This more accurately reflects the behavior of the code. Slices are granted to kses. - Add a function kseq_nice_min() which finds the smallest nice value assigned to the kseg of any KSE on the queue. - Rewrite the logic in sched_slice(). Add a large comment describing the new slice selection scheme. To summarize, slices are assigned based on the nice value. Priorities are still calculated based on the nice and interactivity of a process. Slice sizes of 0 may be granted for KSEs whos nice is 20 or futher away from the lowest nice on the run queue. Other nice values are scaled across the range [min, min+20]. This fixes ULEs bad behavior with positively niced processes.
2003-04-02 06:46:43 +00:00
static void
sched_slice(struct kse *ke)
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
{
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
struct kseq *kseq;
- Add in support for KSEs with 0 slice values on the run queue. If we try to select a KSE with a slice of 0 we will update its slice and insert it onto the next queue. - Pass the KSE instead of the ksegrp into sched_slice(). This more accurately reflects the behavior of the code. Slices are granted to kses. - Add a function kseq_nice_min() which finds the smallest nice value assigned to the kseg of any KSE on the queue. - Rewrite the logic in sched_slice(). Add a large comment describing the new slice selection scheme. To summarize, slices are assigned based on the nice value. Priorities are still calculated based on the nice and interactivity of a process. Slice sizes of 0 may be granted for KSEs whos nice is 20 or futher away from the lowest nice on the run queue. Other nice values are scaled across the range [min, min+20]. This fixes ULEs bad behavior with positively niced processes.
2003-04-02 06:46:43 +00:00
struct ksegrp *kg;
kg = ke->ke_ksegrp;
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
kseq = KSEQ_CPU(ke->ke_cpu);
- Add in support for KSEs with 0 slice values on the run queue. If we try to select a KSE with a slice of 0 we will update its slice and insert it onto the next queue. - Pass the KSE instead of the ksegrp into sched_slice(). This more accurately reflects the behavior of the code. Slices are granted to kses. - Add a function kseq_nice_min() which finds the smallest nice value assigned to the kseg of any KSE on the queue. - Rewrite the logic in sched_slice(). Add a large comment describing the new slice selection scheme. To summarize, slices are assigned based on the nice value. Priorities are still calculated based on the nice and interactivity of a process. Slice sizes of 0 may be granted for KSEs whos nice is 20 or futher away from the lowest nice on the run queue. Other nice values are scaled across the range [min, min+20]. This fixes ULEs bad behavior with positively niced processes.
2003-04-02 06:46:43 +00:00
/*
* Rationale:
* KSEs in interactive ksegs get the minimum slice so that we
* quickly notice if it abuses its advantage.
*
* KSEs in non-interactive ksegs are assigned a slice that is
* based on the ksegs nice value relative to the least nice kseg
* on the run queue for this cpu.
*
* If the KSE is less nice than all others it gets the maximum
* slice and other KSEs will adjust their slice relative to
* this when they first expire.
*
* 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.
*
* If you are outside of the window you will get no slice and
* you will be reevaluated each time you are selected on the
* run queue.
*
*/
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
if (!SCHED_INTERACTIVE(kg)) {
- Add in support for KSEs with 0 slice values on the run queue. If we try to select a KSE with a slice of 0 we will update its slice and insert it onto the next queue. - Pass the KSE instead of the ksegrp into sched_slice(). This more accurately reflects the behavior of the code. Slices are granted to kses. - Add a function kseq_nice_min() which finds the smallest nice value assigned to the kseg of any KSE on the queue. - Rewrite the logic in sched_slice(). Add a large comment describing the new slice selection scheme. To summarize, slices are assigned based on the nice value. Priorities are still calculated based on the nice and interactivity of a process. Slice sizes of 0 may be granted for KSEs whos nice is 20 or futher away from the lowest nice on the run queue. Other nice values are scaled across the range [min, min+20]. This fixes ULEs bad behavior with positively niced processes.
2003-04-02 06:46:43 +00:00
int nice;
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
nice = kg->kg_nice + (0 - kseq->ksq_nicemin);
if (kseq->ksq_loads[PRI_TIMESHARE] == 0 ||
kg->kg_nice < kseq->ksq_nicemin)
- Add in support for KSEs with 0 slice values on the run queue. If we try to select a KSE with a slice of 0 we will update its slice and insert it onto the next queue. - Pass the KSE instead of the ksegrp into sched_slice(). This more accurately reflects the behavior of the code. Slices are granted to kses. - Add a function kseq_nice_min() which finds the smallest nice value assigned to the kseg of any KSE on the queue. - Rewrite the logic in sched_slice(). Add a large comment describing the new slice selection scheme. To summarize, slices are assigned based on the nice value. Priorities are still calculated based on the nice and interactivity of a process. Slice sizes of 0 may be granted for KSEs whos nice is 20 or futher away from the lowest nice on the run queue. Other nice values are scaled across the range [min, min+20]. This fixes ULEs bad behavior with positively niced processes.
2003-04-02 06:46:43 +00:00
ke->ke_slice = SCHED_SLICE_MAX;
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
else if (nice <= SCHED_PRI_NTHRESH)
- Add in support for KSEs with 0 slice values on the run queue. If we try to select a KSE with a slice of 0 we will update its slice and insert it onto the next queue. - Pass the KSE instead of the ksegrp into sched_slice(). This more accurately reflects the behavior of the code. Slices are granted to kses. - Add a function kseq_nice_min() which finds the smallest nice value assigned to the kseg of any KSE on the queue. - Rewrite the logic in sched_slice(). Add a large comment describing the new slice selection scheme. To summarize, slices are assigned based on the nice value. Priorities are still calculated based on the nice and interactivity of a process. Slice sizes of 0 may be granted for KSEs whos nice is 20 or futher away from the lowest nice on the run queue. Other nice values are scaled across the range [min, min+20]. This fixes ULEs bad behavior with positively niced processes.
2003-04-02 06:46:43 +00:00
ke->ke_slice = SCHED_SLICE_NICE(nice);
else
ke->ke_slice = 0;
} else
ke->ke_slice = SCHED_SLICE_MIN;
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
CTR6(KTR_ULE,
"Sliced %p(%d) (nice: %d, nicemin: %d, load: %d, interactive: %d)",
ke, ke->ke_slice, kg->kg_nice, kseq->ksq_nicemin,
kseq->ksq_loads[PRI_TIMESHARE], SCHED_INTERACTIVE(kg));
- Add a new variable 'kg_runtime' that tracks the amount of time we've run. - Use the ratio of kg_runtime / kg_slptime to determine our dynamic priority. - Scale kg_runtime and kg_slptime back when the sum of the two exceeds SCHED_SLP_RUN_MAX. This allows us to slowly forget old behavior. - Scale back the runtime and slptime in fork so that the new process has the same ratio but much less accumulated time. This causes new behavior to be noticed more quickly.
2003-02-10 14:03:45 +00:00
/*
- Keep seperate statistics and run queues for different scheduling classes. - Treat each class specially in kseq_{choose,add,rem}. Let the rest of the code be less aware of scheduling classes. - Skip the interactivity calculation for non TIMESHARE ksegrps. - Move slice and runq selection into kseq_add(). Uninline it now that it's big.
2003-04-03 00:29:28 +00:00
* Check to see if we need to scale back the slp and run time
* in the kg. This will cause us to forget old interactivity
* while maintaining the current ratio.
- Add a new variable 'kg_runtime' that tracks the amount of time we've run. - Use the ratio of kg_runtime / kg_slptime to determine our dynamic priority. - Scale kg_runtime and kg_slptime back when the sum of the two exceeds SCHED_SLP_RUN_MAX. This allows us to slowly forget old behavior. - Scale back the runtime and slptime in fork so that the new process has the same ratio but much less accumulated time. This causes new behavior to be noticed more quickly.
2003-02-10 14:03:45 +00:00
*/
- Add a new function "sched_interact_update()" that scales back the sleep and run time. - Scale the sleep and run time back via sched_interact_update() in more places. This is to keep the statistic more accurate. - Charge a parent one tick for forking a child. - Add only the run time and not the sleep time to the parents kg when a thread exits. This allows us to give a penalty for having an expensive thread exit but does not give a bonus for having an interactive thread exit. - Change the SLP_RUN_THROTTLE to limit us to 4/5th and not 1/2. - Change the SLP_RUN_MAX to two seconds. This keeps bursty interactive applications like mozilla and openoffice in the interactive range even through expensive tasks. - Recalculate the slice after every sleep. This ensures that once a task has been marked interactive it only has a slice of 1 at the risk of giving tasks that sleep for a very brief period a longer time slice.
2003-06-17 06:39:51 +00:00
sched_interact_update(kg);
- Add a new variable 'kg_runtime' that tracks the amount of time we've run. - Use the ratio of kg_runtime / kg_slptime to determine our dynamic priority. - Scale kg_runtime and kg_slptime back when the sum of the two exceeds SCHED_SLP_RUN_MAX. This allows us to slowly forget old behavior. - Scale back the runtime and slptime in fork so that the new process has the same ratio but much less accumulated time. This causes new behavior to be noticed more quickly.
2003-02-10 14:03:45 +00:00
- Add in support for KSEs with 0 slice values on the run queue. If we try to select a KSE with a slice of 0 we will update its slice and insert it onto the next queue. - Pass the KSE instead of the ksegrp into sched_slice(). This more accurately reflects the behavior of the code. Slices are granted to kses. - Add a function kseq_nice_min() which finds the smallest nice value assigned to the kseg of any KSE on the queue. - Rewrite the logic in sched_slice(). Add a large comment describing the new slice selection scheme. To summarize, slices are assigned based on the nice value. Priorities are still calculated based on the nice and interactivity of a process. Slice sizes of 0 may be granted for KSEs whos nice is 20 or futher away from the lowest nice on the run queue. Other nice values are scaled across the range [min, min+20]. This fixes ULEs bad behavior with positively niced processes.
2003-04-02 06:46:43 +00:00
return;
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
}
- Add a new function "sched_interact_update()" that scales back the sleep and run time. - Scale the sleep and run time back via sched_interact_update() in more places. This is to keep the statistic more accurate. - Charge a parent one tick for forking a child. - Add only the run time and not the sleep time to the parents kg when a thread exits. This allows us to give a penalty for having an expensive thread exit but does not give a bonus for having an interactive thread exit. - Change the SLP_RUN_THROTTLE to limit us to 4/5th and not 1/2. - Change the SLP_RUN_MAX to two seconds. This keeps bursty interactive applications like mozilla and openoffice in the interactive range even through expensive tasks. - Recalculate the slice after every sleep. This ensures that once a task has been marked interactive it only has a slice of 1 at the risk of giving tasks that sleep for a very brief period a longer time slice.
2003-06-17 06:39:51 +00:00
static void
sched_interact_update(struct ksegrp *kg)
{
- Use a better algorithm in sched_pctcpu_update() Contributed by: Thomaswuerfl@gmx.de - In sched_prio(), adjust the run queue for threads which may need to move to the current queue due to priority propagation . - In sched_switch(), fix style bug introduced when the KSE support went in. Columns are 80 chars wide, not 90. - In sched_switch(), Fix the comparison in the idle case and explicitly re-initialize the runq in the not propagated case. - Remove dead code in sched_clock(). - In sched_clock(), If we're an IDLE class td set NEEDRESCHED so that threads that have become runnable will get a chance to. - In sched_runnable(), if we're not the IDLETD, we should not consider curthread when examining the load. This mimics the 4BSD behavior of returning 0 when the only runnable thread is running. - In sched_userret(), remove the code for setting NEEDRESCHED entirely. This is not necessary and is not implemented in 4BSD. - Use the correct comparison in sched_add() when checking to see if an idle prio task has had it's priority temporarily elevated.
2003-10-27 06:47:05 +00:00
int ratio;
if ((kg->kg_runtime + kg->kg_slptime) > SCHED_SLP_RUN_MAX) {
ratio = ((SCHED_SLP_RUN_MAX * 15) / (kg->kg_runtime +
kg->kg_slptime ));
kg->kg_runtime = (kg->kg_runtime * ratio) / 16;
kg->kg_slptime = (kg->kg_slptime * ratio) / 16;
- Add a new function "sched_interact_update()" that scales back the sleep and run time. - Scale the sleep and run time back via sched_interact_update() in more places. This is to keep the statistic more accurate. - Charge a parent one tick for forking a child. - Add only the run time and not the sleep time to the parents kg when a thread exits. This allows us to give a penalty for having an expensive thread exit but does not give a bonus for having an interactive thread exit. - Change the SLP_RUN_THROTTLE to limit us to 4/5th and not 1/2. - Change the SLP_RUN_MAX to two seconds. This keeps bursty interactive applications like mozilla and openoffice in the interactive range even through expensive tasks. - Recalculate the slice after every sleep. This ensures that once a task has been marked interactive it only has a slice of 1 at the risk of giving tasks that sleep for a very brief period a longer time slice.
2003-06-17 06:39:51 +00:00
}
}
- Create a function sched_interact_score() which decides on the interactivity of a kseg and assigns it a value of 0 through 100. - Use sched_interact_score() to determine the dynamic priority. - Define SCHED_CURR() in terms of sched_interact_score(). - Adjust the maximum slice back down to 100ms. - Remove redundant clearing of ke_runq in sched_wakeup() - Clean up #defines and comment them.
2003-03-04 02:45:59 +00:00
static int
sched_interact_score(struct ksegrp *kg)
{
- Fix the maximum slice value. I accidentally checked in a value of '2' which meant no process would run for longer than 20ms. - Slightly redo the interactivity scorer. It follows the same algorithm but in a slightly more correct way. Previously values above half were incorrect. - Lower the interactivity threshold to 20. It seems that in testing non- interactive tasks are hardly ever near there and expensive interactive tasks can sometimes surpass it. This area needs more testing. - Remove an unnecessary KTR. - Fix a case where an idle thread that had an elevated priority due to priority prop. would be placed back on the idle queue. - Delay setting NEEDRESCHED until userret() for threads that haad their priority elevated while in kernel. This gives us the same context switch optimization as SCHED_4BSD. - Limit the child's slice to 1 in sched_fork_kse() so we detect its behavior more quickly. - Inhert some of the run/slp time from the child in sched_exit_ksegrp(). - Redo some of the priority comparisons so they are more clear. - Throttle the frequency of sched_pctcpu_update() so that rounding errors do not make it invalid.
2003-06-15 02:18:29 +00:00
int div;
- Create a function sched_interact_score() which decides on the interactivity of a kseg and assigns it a value of 0 through 100. - Use sched_interact_score() to determine the dynamic priority. - Define SCHED_CURR() in terms of sched_interact_score(). - Adjust the maximum slice back down to 100ms. - Remove redundant clearing of ke_runq in sched_wakeup() - Clean up #defines and comment them.
2003-03-04 02:45:59 +00:00
if (kg->kg_runtime > kg->kg_slptime) {
- Fix the maximum slice value. I accidentally checked in a value of '2' which meant no process would run for longer than 20ms. - Slightly redo the interactivity scorer. It follows the same algorithm but in a slightly more correct way. Previously values above half were incorrect. - Lower the interactivity threshold to 20. It seems that in testing non- interactive tasks are hardly ever near there and expensive interactive tasks can sometimes surpass it. This area needs more testing. - Remove an unnecessary KTR. - Fix a case where an idle thread that had an elevated priority due to priority prop. would be placed back on the idle queue. - Delay setting NEEDRESCHED until userret() for threads that haad their priority elevated while in kernel. This gives us the same context switch optimization as SCHED_4BSD. - Limit the child's slice to 1 in sched_fork_kse() so we detect its behavior more quickly. - Inhert some of the run/slp time from the child in sched_exit_ksegrp(). - Redo some of the priority comparisons so they are more clear. - Throttle the frequency of sched_pctcpu_update() so that rounding errors do not make it invalid.
2003-06-15 02:18:29 +00:00
div = max(1, kg->kg_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);
- Create a function sched_interact_score() which decides on the interactivity of a kseg and assigns it a value of 0 through 100. - Use sched_interact_score() to determine the dynamic priority. - Define SCHED_CURR() in terms of sched_interact_score(). - Adjust the maximum slice back down to 100ms. - Remove redundant clearing of ke_runq in sched_wakeup() - Clean up #defines and comment them.
2003-03-04 02:45:59 +00:00
}
- Fix the maximum slice value. I accidentally checked in a value of '2' which meant no process would run for longer than 20ms. - Slightly redo the interactivity scorer. It follows the same algorithm but in a slightly more correct way. Previously values above half were incorrect. - Lower the interactivity threshold to 20. It seems that in testing non- interactive tasks are hardly ever near there and expensive interactive tasks can sometimes surpass it. This area needs more testing. - Remove an unnecessary KTR. - Fix a case where an idle thread that had an elevated priority due to priority prop. would be placed back on the idle queue. - Delay setting NEEDRESCHED until userret() for threads that haad their priority elevated while in kernel. This gives us the same context switch optimization as SCHED_4BSD. - Limit the child's slice to 1 in sched_fork_kse() so we detect its behavior more quickly. - Inhert some of the run/slp time from the child in sched_exit_ksegrp(). - Redo some of the priority comparisons so they are more clear. - Throttle the frequency of sched_pctcpu_update() so that rounding errors do not make it invalid.
2003-06-15 02:18:29 +00:00
/*
* This can happen if slptime and runtime are 0.
*/
return (0);
- Create a function sched_interact_score() which decides on the interactivity of a kseg and assigns it a value of 0 through 100. - Use sched_interact_score() to determine the dynamic priority. - Define SCHED_CURR() in terms of sched_interact_score(). - Adjust the maximum slice back down to 100ms. - Remove redundant clearing of ke_runq in sched_wakeup() - Clean up #defines and comment them.
2003-03-04 02:45:59 +00:00
}
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
/*
* This is only somewhat accurate since given many processes of the same
* priority they will switch when their slices run out, which will be
* at most SCHED_SLICE_MAX.
*/
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
int
sched_rr_interval(void)
{
return (SCHED_SLICE_MAX);
}
- Add static to local functions and data where it was missing. - Add an IPI based mechanism for migrating kses. This mechanism is broken down into several components. This is intended to reduce cache thrashing by eliminating most cases where one cpu touches another's run queues. - kseq_notify() appends a kse to a lockless singly linked list and conditionally sends an IPI to the target processor. Right now this is protected by sched_lock but at some point I'd like to get rid of the global lock. This is why I used something more complicated than a standard queue. - kseq_assign() processes our list of kses that have been assigned to us by other processors. This simply calls sched_add() for each item on the list after clearing the new KEF_ASSIGNED flag. This flag is used to indicate that we have been appeneded to the assigned queue but not added to the run queue yet. - In sched_add(), instead of adding a KSE to another processor's queue we use kse_notify() so that we don't touch their queue. Also in sched_add(), if KEF_ASSIGNED is already set return immediately. This can happen if a thread is removed and readded so that the priority is recorded properly. - In sched_rem() return immediately if KEF_ASSIGNED is set. All callers immediately readd simply to adjust priorites etc. - In sched_choose(), if we're running an IDLE task or the per cpu idle thread set our cpumask bit in 'kseq_idle' so that other processors may know that we are idle. Before this, make a single pass through the run queues of other processors so that we may find work more immediately if it is available. - In sched_runnable(), don't scan each processor's run queue, they will IPI us if they have work for us to do. - In sched_add(), if we're adding a thread that can be migrated and we have plenty of work to do, try to migrate the thread to an idle kseq. - Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into consideration. - No longer use kseq_choose() to steal threads, it can lose it's last argument. - Create a new function runq_steal() which operates like runq_choose() but skips threads based on some criteria. Currently it will not steal PRI_ITHD threads. In the future this will be used for CPU binding. - Create a kseq_steal() that checks each run queue with runq_steal(), use kseq_steal() in the places where we used kseq_choose() to steal with before.
2003-10-31 11:16:04 +00:00
static void
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
sched_pctcpu_update(struct kse *ke)
{
/*
* Adjust counters and watermark for pctcpu calc.
- Fix the maximum slice value. I accidentally checked in a value of '2' which meant no process would run for longer than 20ms. - Slightly redo the interactivity scorer. It follows the same algorithm but in a slightly more correct way. Previously values above half were incorrect. - Lower the interactivity threshold to 20. It seems that in testing non- interactive tasks are hardly ever near there and expensive interactive tasks can sometimes surpass it. This area needs more testing. - Remove an unnecessary KTR. - Fix a case where an idle thread that had an elevated priority due to priority prop. would be placed back on the idle queue. - Delay setting NEEDRESCHED until userret() for threads that haad their priority elevated while in kernel. This gives us the same context switch optimization as SCHED_4BSD. - Limit the child's slice to 1 in sched_fork_kse() so we detect its behavior more quickly. - Inhert some of the run/slp time from the child in sched_exit_ksegrp(). - Redo some of the priority comparisons so they are more clear. - Throttle the frequency of sched_pctcpu_update() so that rounding errors do not make it invalid.
2003-06-15 02:18:29 +00:00
*/
- Somewhere along the line I stupidly removed critical logic from sched_ptcpu_update(). This caused erroneous cpu times in TOP for processes that were asleep. Replace the code that was removed.
2003-09-20 02:05:58 +00:00
if (ke->ke_ltick > ticks - SCHED_CPU_TICKS) {
/*
* Shift the tick count out so that the divide doesn't
* round away our results.
*/
ke->ke_ticks <<= 10;
ke->ke_ticks = (ke->ke_ticks / (ticks - ke->ke_ftick)) *
SCHED_CPU_TICKS;
ke->ke_ticks >>= 10;
} else
ke->ke_ticks = 0;
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
ke->ke_ltick = ticks;
ke->ke_ftick = ke->ke_ltick - SCHED_CPU_TICKS;
}
- Add static to local functions and data where it was missing. - Add an IPI based mechanism for migrating kses. This mechanism is broken down into several components. This is intended to reduce cache thrashing by eliminating most cases where one cpu touches another's run queues. - kseq_notify() appends a kse to a lockless singly linked list and conditionally sends an IPI to the target processor. Right now this is protected by sched_lock but at some point I'd like to get rid of the global lock. This is why I used something more complicated than a standard queue. - kseq_assign() processes our list of kses that have been assigned to us by other processors. This simply calls sched_add() for each item on the list after clearing the new KEF_ASSIGNED flag. This flag is used to indicate that we have been appeneded to the assigned queue but not added to the run queue yet. - In sched_add(), instead of adding a KSE to another processor's queue we use kse_notify() so that we don't touch their queue. Also in sched_add(), if KEF_ASSIGNED is already set return immediately. This can happen if a thread is removed and readded so that the priority is recorded properly. - In sched_rem() return immediately if KEF_ASSIGNED is set. All callers immediately readd simply to adjust priorites etc. - In sched_choose(), if we're running an IDLE task or the per cpu idle thread set our cpumask bit in 'kseq_idle' so that other processors may know that we are idle. Before this, make a single pass through the run queues of other processors so that we may find work more immediately if it is available. - In sched_runnable(), don't scan each processor's run queue, they will IPI us if they have work for us to do. - In sched_add(), if we're adding a thread that can be migrated and we have plenty of work to do, try to migrate the thread to an idle kseq. - Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into consideration. - No longer use kseq_choose() to steal threads, it can lose it's last argument. - Create a new function runq_steal() which operates like runq_choose() but skips threads based on some criteria. Currently it will not steal PRI_ITHD threads. In the future this will be used for CPU binding. - Create a kseq_steal() that checks each run queue with runq_steal(), use kseq_steal() in the places where we used kseq_choose() to steal with before.
2003-10-31 11:16:04 +00:00
#if 0
- Make some context switches conditional on SCHED_STRICT_RESCHED. This may have some negative effect on interactivity but it yields great perf. gains. This also brings the conditions under which ULE context switches inline with SCHED_4BSD. - Define some new kseq_* functions for manipulating the run queue. - Add a new kseq member ksq_rslices and ksq_bload. rslices is the sum of the slices of runnable kses. This will be used for push load balance decisions. bload is the number of threads blocked waiting on IO.
2003-02-03 05:30:07 +00:00
/* XXX Should be changed to kseq_load_lowest() */
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
int
sched_pickcpu(void)
{
- Use ksq_load as the authoritive count of kses on the pair of kseqs for sched_runnable() et all. - Remove some dead code in sched_clock(). - Define two macros KSEQ_SELF() and KSEQ_CPU() for getting the kseq of the current cpu or some alternate cpu. - Start introducing kseq_() functions, such as kseq_choose() and kseq_setup().
2003-01-29 07:00:51 +00:00
struct kseq *kseq;
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
int load;
- Use ksq_load as the authoritive count of kses on the pair of kseqs for sched_runnable() et all. - Remove some dead code in sched_clock(). - Define two macros KSEQ_SELF() and KSEQ_CPU() for getting the kseq of the current cpu or some alternate cpu. - Start introducing kseq_() functions, such as kseq_choose() and kseq_setup().
2003-01-29 07:00:51 +00:00
int cpu;
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
int i;
- When a new thread is added to a kseq the load is incremented prior to adding it to the nice tables. Therefore, in kseq_add_nice, we should keep in mind that the load will be 1 if we are the only thread, and not 0. - Assert that the sched lock is held in all the appropriate places. - Increase the scope of the sched lock in sched_pctcpu_update(). - Hold the sched lock in sched_runnable(). It is not held by the caller.
2003-06-08 00:47:33 +00:00
mtx_assert(&sched_lock, MA_OWNED);
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
if (!smp_started)
return (0);
- Use ksq_load as the authoritive count of kses on the pair of kseqs for sched_runnable() et all. - Remove some dead code in sched_clock(). - Define two macros KSEQ_SELF() and KSEQ_CPU() for getting the kseq of the current cpu or some alternate cpu. - Start introducing kseq_() functions, such as kseq_choose() and kseq_setup().
2003-01-29 07:00:51 +00:00
load = 0;
cpu = 0;
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
for (i = 0; i < mp_maxid; i++) {
- Don't migrate to stopped cpus.
2003-06-28 09:09:33 +00:00
if (CPU_ABSENT(i) || (i & stopped_cpus) != 0)
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
continue;
- Use ksq_load as the authoritive count of kses on the pair of kseqs for sched_runnable() et all. - Remove some dead code in sched_clock(). - Define two macros KSEQ_SELF() and KSEQ_CPU() for getting the kseq of the current cpu or some alternate cpu. - Start introducing kseq_() functions, such as kseq_choose() and kseq_setup().
2003-01-29 07:00:51 +00:00
kseq = KSEQ_CPU(i);
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
if (kseq->ksq_load < load) {
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
cpu = i;
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
load = kseq->ksq_load;
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
}
}
CTR1(KTR_RUNQ, "sched_pickcpu: %d", cpu);
return (cpu);
}
#endif
void
sched_prio(struct thread *td, u_char prio)
{
- Use a better algorithm in sched_pctcpu_update() Contributed by: Thomaswuerfl@gmx.de - In sched_prio(), adjust the run queue for threads which may need to move to the current queue due to priority propagation . - In sched_switch(), fix style bug introduced when the KSE support went in. Columns are 80 chars wide, not 90. - In sched_switch(), Fix the comparison in the idle case and explicitly re-initialize the runq in the not propagated case. - Remove dead code in sched_clock(). - In sched_clock(), If we're an IDLE class td set NEEDRESCHED so that threads that have become runnable will get a chance to. - In sched_runnable(), if we're not the IDLETD, we should not consider curthread when examining the load. This mimics the 4BSD behavior of returning 0 when the only runnable thread is running. - In sched_userret(), remove the code for setting NEEDRESCHED entirely. This is not necessary and is not implemented in 4BSD. - Use the correct comparison in sched_add() when checking to see if an idle prio task has had it's priority temporarily elevated.
2003-10-27 06:47:05 +00:00
struct kse *ke;
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
- Use a better algorithm in sched_pctcpu_update() Contributed by: Thomaswuerfl@gmx.de - In sched_prio(), adjust the run queue for threads which may need to move to the current queue due to priority propagation . - In sched_switch(), fix style bug introduced when the KSE support went in. Columns are 80 chars wide, not 90. - In sched_switch(), Fix the comparison in the idle case and explicitly re-initialize the runq in the not propagated case. - Remove dead code in sched_clock(). - In sched_clock(), If we're an IDLE class td set NEEDRESCHED so that threads that have become runnable will get a chance to. - In sched_runnable(), if we're not the IDLETD, we should not consider curthread when examining the load. This mimics the 4BSD behavior of returning 0 when the only runnable thread is running. - In sched_userret(), remove the code for setting NEEDRESCHED entirely. This is not necessary and is not implemented in 4BSD. - Use the correct comparison in sched_add() when checking to see if an idle prio task has had it's priority temporarily elevated.
2003-10-27 06:47:05 +00:00
ke = td->td_kse;
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
mtx_assert(&sched_lock, MA_OWNED);
if (TD_ON_RUNQ(td)) {
- Use a better algorithm in sched_pctcpu_update() Contributed by: Thomaswuerfl@gmx.de - In sched_prio(), adjust the run queue for threads which may need to move to the current queue due to priority propagation . - In sched_switch(), fix style bug introduced when the KSE support went in. Columns are 80 chars wide, not 90. - In sched_switch(), Fix the comparison in the idle case and explicitly re-initialize the runq in the not propagated case. - Remove dead code in sched_clock(). - In sched_clock(), If we're an IDLE class td set NEEDRESCHED so that threads that have become runnable will get a chance to. - In sched_runnable(), if we're not the IDLETD, we should not consider curthread when examining the load. This mimics the 4BSD behavior of returning 0 when the only runnable thread is running. - In sched_userret(), remove the code for setting NEEDRESCHED entirely. This is not necessary and is not implemented in 4BSD. - Use the correct comparison in sched_add() when checking to see if an idle prio task has had it's priority temporarily elevated.
2003-10-27 06:47:05 +00:00
/*
* If the priority has been elevated due to priority
* propagation, we may have to move ourselves to a new
* queue. We still call adjustrunqueue below in case kse
* needs to fix things up.
*/
- Add static to local functions and data where it was missing. - Add an IPI based mechanism for migrating kses. This mechanism is broken down into several components. This is intended to reduce cache thrashing by eliminating most cases where one cpu touches another's run queues. - kseq_notify() appends a kse to a lockless singly linked list and conditionally sends an IPI to the target processor. Right now this is protected by sched_lock but at some point I'd like to get rid of the global lock. This is why I used something more complicated than a standard queue. - kseq_assign() processes our list of kses that have been assigned to us by other processors. This simply calls sched_add() for each item on the list after clearing the new KEF_ASSIGNED flag. This flag is used to indicate that we have been appeneded to the assigned queue but not added to the run queue yet. - In sched_add(), instead of adding a KSE to another processor's queue we use kse_notify() so that we don't touch their queue. Also in sched_add(), if KEF_ASSIGNED is already set return immediately. This can happen if a thread is removed and readded so that the priority is recorded properly. - In sched_rem() return immediately if KEF_ASSIGNED is set. All callers immediately readd simply to adjust priorites etc. - In sched_choose(), if we're running an IDLE task or the per cpu idle thread set our cpumask bit in 'kseq_idle' so that other processors may know that we are idle. Before this, make a single pass through the run queues of other processors so that we may find work more immediately if it is available. - In sched_runnable(), don't scan each processor's run queue, they will IPI us if they have work for us to do. - In sched_add(), if we're adding a thread that can be migrated and we have plenty of work to do, try to migrate the thread to an idle kseq. - Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into consideration. - No longer use kseq_choose() to steal threads, it can lose it's last argument. - Create a new function runq_steal() which operates like runq_choose() but skips threads based on some criteria. Currently it will not steal PRI_ITHD threads. In the future this will be used for CPU binding. - Create a kseq_steal() that checks each run queue with runq_steal(), use kseq_steal() in the places where we used kseq_choose() to steal with before.
2003-10-31 11:16:04 +00:00
if (ke && (ke->ke_flags & KEF_ASSIGNED) == 0 &&
ke->ke_runq != KSEQ_CPU(ke->ke_cpu)->ksq_curr) {
- Use a better algorithm in sched_pctcpu_update() Contributed by: Thomaswuerfl@gmx.de - In sched_prio(), adjust the run queue for threads which may need to move to the current queue due to priority propagation . - In sched_switch(), fix style bug introduced when the KSE support went in. Columns are 80 chars wide, not 90. - In sched_switch(), Fix the comparison in the idle case and explicitly re-initialize the runq in the not propagated case. - Remove dead code in sched_clock(). - In sched_clock(), If we're an IDLE class td set NEEDRESCHED so that threads that have become runnable will get a chance to. - In sched_runnable(), if we're not the IDLETD, we should not consider curthread when examining the load. This mimics the 4BSD behavior of returning 0 when the only runnable thread is running. - In sched_userret(), remove the code for setting NEEDRESCHED entirely. This is not necessary and is not implemented in 4BSD. - Use the correct comparison in sched_add() when checking to see if an idle prio task has had it's priority temporarily elevated.
2003-10-27 06:47:05 +00:00
runq_remove(ke->ke_runq, ke);
ke->ke_runq = KSEQ_CPU(ke->ke_cpu)->ksq_curr;
runq_add(ke->ke_runq, ke);
}
Let SA process work under ULE scheduler, originally it would panic kernel. Reviewed by: jeff
2003-08-26 11:33:15 +00:00
adjustrunqueue(td, prio);
- Use a better algorithm in sched_pctcpu_update() Contributed by: Thomaswuerfl@gmx.de - In sched_prio(), adjust the run queue for threads which may need to move to the current queue due to priority propagation . - In sched_switch(), fix style bug introduced when the KSE support went in. Columns are 80 chars wide, not 90. - In sched_switch(), Fix the comparison in the idle case and explicitly re-initialize the runq in the not propagated case. - Remove dead code in sched_clock(). - In sched_clock(), If we're an IDLE class td set NEEDRESCHED so that threads that have become runnable will get a chance to. - In sched_runnable(), if we're not the IDLETD, we should not consider curthread when examining the load. This mimics the 4BSD behavior of returning 0 when the only runnable thread is running. - In sched_userret(), remove the code for setting NEEDRESCHED entirely. This is not necessary and is not implemented in 4BSD. - Use the correct comparison in sched_add() when checking to see if an idle prio task has had it's priority temporarily elevated.
2003-10-27 06:47:05 +00:00
} else
Let SA process work under ULE scheduler, originally it would panic kernel. Reviewed by: jeff
2003-08-26 11:33:15 +00:00
td->td_priority = prio;
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
}
void
- Collapse sched_switchin() and sched_switchout() into sched_switch(). Now mi_switch() calls sched_switch() which calls cpu_switch(). This is actually one less function call than it had been.
2003-10-16 08:53:46 +00:00
sched_switch(struct thread *td)
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
{
- Collapse sched_switchin() and sched_switchout() into sched_switch(). Now mi_switch() calls sched_switch() which calls cpu_switch(). This is actually one less function call than it had been.
2003-10-16 08:53:46 +00:00
struct thread *newtd;
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
struct kse *ke;
mtx_assert(&sched_lock, MA_OWNED);
ke = td->td_kse;
td->td_last_kse = ke;
Move the _oncpu entry from the KSE to the thread. The entry in the KSE still exists but it's purpose will change a bit when we add the ability to lock a KSE to a cpu.
2003-04-10 17:35:44 +00:00
td->td_lastcpu = td->td_oncpu;
td->td_oncpu = NOCPU;
Move a bunch of flags from the KSE to the thread. I was in two minds as to where to put them in the first case.. I should have listenned to the other mind. Submitted by: parts by davidxu@ Reviewed by: jeff@ mini@
2003-02-17 09:55:10 +00:00
td->td_flags &= ~TDF_NEEDRESCHED;
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
if (TD_IS_RUNNING(td)) {
Let SA process work under ULE scheduler, originally it would panic kernel. Reviewed by: jeff
2003-08-26 11:33:15 +00:00
if (td->td_proc->p_flag & P_SA) {
kseq_rem(KSEQ_CPU(ke->ke_cpu), ke);
setrunqueue(td);
} else {
/*
- Use a better algorithm in sched_pctcpu_update() Contributed by: Thomaswuerfl@gmx.de - In sched_prio(), adjust the run queue for threads which may need to move to the current queue due to priority propagation . - In sched_switch(), fix style bug introduced when the KSE support went in. Columns are 80 chars wide, not 90. - In sched_switch(), Fix the comparison in the idle case and explicitly re-initialize the runq in the not propagated case. - Remove dead code in sched_clock(). - In sched_clock(), If we're an IDLE class td set NEEDRESCHED so that threads that have become runnable will get a chance to. - In sched_runnable(), if we're not the IDLETD, we should not consider curthread when examining the load. This mimics the 4BSD behavior of returning 0 when the only runnable thread is running. - In sched_userret(), remove the code for setting NEEDRESCHED entirely. This is not necessary and is not implemented in 4BSD. - Use the correct comparison in sched_add() when checking to see if an idle prio task has had it's priority temporarily elevated.
2003-10-27 06:47:05 +00:00
* This queue is always correct except for idle threads
* which have a higher priority due to priority
* propagation.
Let SA process work under ULE scheduler, originally it would panic kernel. Reviewed by: jeff
2003-08-26 11:33:15 +00:00
*/
- Use a better algorithm in sched_pctcpu_update() Contributed by: Thomaswuerfl@gmx.de - In sched_prio(), adjust the run queue for threads which may need to move to the current queue due to priority propagation . - In sched_switch(), fix style bug introduced when the KSE support went in. Columns are 80 chars wide, not 90. - In sched_switch(), Fix the comparison in the idle case and explicitly re-initialize the runq in the not propagated case. - Remove dead code in sched_clock(). - In sched_clock(), If we're an IDLE class td set NEEDRESCHED so that threads that have become runnable will get a chance to. - In sched_runnable(), if we're not the IDLETD, we should not consider curthread when examining the load. This mimics the 4BSD behavior of returning 0 when the only runnable thread is running. - In sched_userret(), remove the code for setting NEEDRESCHED entirely. This is not necessary and is not implemented in 4BSD. - Use the correct comparison in sched_add() when checking to see if an idle prio task has had it's priority temporarily elevated.
2003-10-27 06:47:05 +00:00
if (ke->ke_ksegrp->kg_pri_class == PRI_IDLE) {
if (td->td_priority < PRI_MIN_IDLE)
ke->ke_runq = KSEQ_SELF()->ksq_curr;
else
ke->ke_runq = &KSEQ_SELF()->ksq_idle;
}
Let SA process work under ULE scheduler, originally it would panic kernel. Reviewed by: jeff
2003-08-26 11:33:15 +00:00
runq_add(ke->ke_runq, ke);
/* setrunqueue(td); */
}
- Only kse_reassign() in the !running case. Reported by: kris
2003-10-16 20:32:57 +00:00
} else {
if (ke->ke_runq)
kseq_rem(KSEQ_CPU(ke->ke_cpu), ke);
/*
* We will not be on the run queue. So we must be
* sleeping or similar.
*/
if (td->td_proc->p_flag & P_SA)
kse_reassign(ke);
}
- Collapse sched_switchin() and sched_switchout() into sched_switch(). Now mi_switch() calls sched_switch() which calls cpu_switch(). This is actually one less function call than it had been.
2003-10-16 08:53:46 +00:00
newtd = choosethread();
if (td != newtd)
cpu_switch(td, newtd);
sched_lock.mtx_lock = (uintptr_t)td;
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
Move the _oncpu entry from the KSE to the thread. The entry in the KSE still exists but it's purpose will change a bit when we add the ability to lock a KSE to a cpu.
2003-04-10 17:35:44 +00:00
td->td_oncpu = PCPU_GET(cpuid);
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
}
void
sched_nice(struct ksegrp *kg, int nice)
{
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
struct kse *ke;
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
struct thread *td;
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
struct kseq *kseq;
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
- Assert that the proc lock and sched_lock are held in sched_nice(). - For the 4BSD scheduler, this means that all callers of the static function resetpriority() now always hold sched_lock, so don't lock sched_lock explicitly in that function.
2003-04-22 20:50:38 +00:00
PROC_LOCK_ASSERT(kg->kg_proc, MA_OWNED);
mtx_assert(&sched_lock, MA_OWNED);
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
/*
* We need to adjust the nice counts for running KSEs.
*/
if (kg->kg_pri_class == PRI_TIMESHARE)
FOREACH_KSE_IN_GROUP(kg, ke) {
- Use a more robust mechanism for determining whether or not a kse is on a kseq.
2003-06-17 19:49:18 +00:00
if (ke->ke_runq == NULL)
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
continue;
kseq = KSEQ_CPU(ke->ke_cpu);
kseq_nice_rem(kseq, kg->kg_nice);
kseq_nice_add(kseq, nice);
}
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
kg->kg_nice = nice;
sched_priority(kg);
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
FOREACH_THREAD_IN_GROUP(kg, td)
Move a bunch of flags from the KSE to the thread. I was in two minds as to where to put them in the first case.. I should have listenned to the other mind. Submitted by: parts by davidxu@ Reviewed by: jeff@ mini@
2003-02-17 09:55:10 +00:00
td->td_flags |= TDF_NEEDRESCHED;
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
}
void
sched_sleep(struct thread *td, u_char prio)
{
mtx_assert(&sched_lock, MA_OWNED);
td->td_slptime = ticks;
td->td_priority = prio;
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
CTR2(KTR_ULE, "sleep kse %p (tick: %d)",
td->td_kse, td->td_slptime);
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
}
void
sched_wakeup(struct thread *td)
{
mtx_assert(&sched_lock, MA_OWNED);
/*
* Let the kseg know how long we slept for. This is because process
* interactivity behavior is modeled in the kseg.
*/
if (td->td_slptime) {
- Refrain from setting the td_priority in sched_wakeup(). It will be reset before we return to user space.
2003-03-03 04:11:40 +00:00
struct ksegrp *kg;
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
int hzticks;
- Refrain from setting the td_priority in sched_wakeup(). It will be reset before we return to user space.
2003-03-03 04:11:40 +00:00
kg = td->td_ksegrp;
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
hzticks = ticks - td->td_slptime;
kg->kg_slptime += hzticks << 10;
- Add a new function "sched_interact_update()" that scales back the sleep and run time. - Scale the sleep and run time back via sched_interact_update() in more places. This is to keep the statistic more accurate. - Charge a parent one tick for forking a child. - Add only the run time and not the sleep time to the parents kg when a thread exits. This allows us to give a penalty for having an expensive thread exit but does not give a bonus for having an interactive thread exit. - Change the SLP_RUN_THROTTLE to limit us to 4/5th and not 1/2. - Change the SLP_RUN_MAX to two seconds. This keeps bursty interactive applications like mozilla and openoffice in the interactive range even through expensive tasks. - Recalculate the slice after every sleep. This ensures that once a task has been marked interactive it only has a slice of 1 at the risk of giving tasks that sleep for a very brief period a longer time slice.
2003-06-17 06:39:51 +00:00
sched_interact_update(kg);
- Refrain from setting the td_priority in sched_wakeup(). It will be reset before we return to user space.
2003-03-03 04:11:40 +00:00
sched_priority(kg);
- Add a new function "sched_interact_update()" that scales back the sleep and run time. - Scale the sleep and run time back via sched_interact_update() in more places. This is to keep the statistic more accurate. - Charge a parent one tick for forking a child. - Add only the run time and not the sleep time to the parents kg when a thread exits. This allows us to give a penalty for having an expensive thread exit but does not give a bonus for having an interactive thread exit. - Change the SLP_RUN_THROTTLE to limit us to 4/5th and not 1/2. - Change the SLP_RUN_MAX to two seconds. This keeps bursty interactive applications like mozilla and openoffice in the interactive range even through expensive tasks. - Recalculate the slice after every sleep. This ensures that once a task has been marked interactive it only has a slice of 1 at the risk of giving tasks that sleep for a very brief period a longer time slice.
2003-06-17 06:39:51 +00:00
if (td->td_kse)
sched_slice(td->td_kse);
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
CTR2(KTR_ULE, "wakeup kse %p (%d ticks)",
td->td_kse, hzticks);
- Refrain from setting the td_priority in sched_wakeup(). It will be reset before we return to user space.
2003-03-03 04:11:40 +00:00
td->td_slptime = 0;
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
}
setrunqueue(td);
}
/*
* Penalize the parent for creating a new child and initialize the child's
* priority.
*/
void
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
sched_fork(struct proc *p, struct proc *p1)
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
{
mtx_assert(&sched_lock, MA_OWNED);
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
sched_fork_ksegrp(FIRST_KSEGRP_IN_PROC(p), FIRST_KSEGRP_IN_PROC(p1));
sched_fork_kse(FIRST_KSE_IN_PROC(p), FIRST_KSE_IN_PROC(p1));
sched_fork_thread(FIRST_THREAD_IN_PROC(p), FIRST_THREAD_IN_PROC(p1));
}
void
sched_fork_kse(struct kse *ke, struct kse *child)
{
Add lock assertions for various proc/thread/kse/ksegroup fields to the scheduler functions.
2003-04-23 18:51:05 +00:00
- Fix the maximum slice value. I accidentally checked in a value of '2' which meant no process would run for longer than 20ms. - Slightly redo the interactivity scorer. It follows the same algorithm but in a slightly more correct way. Previously values above half were incorrect. - Lower the interactivity threshold to 20. It seems that in testing non- interactive tasks are hardly ever near there and expensive interactive tasks can sometimes surpass it. This area needs more testing. - Remove an unnecessary KTR. - Fix a case where an idle thread that had an elevated priority due to priority prop. would be placed back on the idle queue. - Delay setting NEEDRESCHED until userret() for threads that haad their priority elevated while in kernel. This gives us the same context switch optimization as SCHED_4BSD. - Limit the child's slice to 1 in sched_fork_kse() so we detect its behavior more quickly. - Inhert some of the run/slp time from the child in sched_exit_ksegrp(). - Redo some of the priority comparisons so they are more clear. - Throttle the frequency of sched_pctcpu_update() so that rounding errors do not make it invalid.
2003-06-15 02:18:29 +00:00
child->ke_slice = 1; /* Attempt to quickly learn interactivity. */
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
child->ke_cpu = ke->ke_cpu; /* sched_pickcpu(); */
child->ke_runq = NULL;
- In SCHED_CURR() add holding Giant to the list of criteria that will keep you on the current queue. In the future, it would be nice if priority propagation could deterministicly pluck a thread off of the next queue and put it on the current queue. Until then this hack stops us from holding up our entire current queue, including interrupt handlers, while a thread on the next queue is blocked while holding Giant. - Inherit our pctcpu information from our parent.
2003-10-12 21:07:31 +00:00
/* Grab our parents cpu estimation information. */
child->ke_ticks = ke->ke_ticks;
child->ke_ltick = ke->ke_ltick;
child->ke_ftick = ke->ke_ftick;
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
}
void
sched_fork_ksegrp(struct ksegrp *kg, struct ksegrp *child)
{
Add lock assertions for various proc/thread/kse/ksegroup fields to the scheduler functions.
2003-04-23 18:51:05 +00:00
PROC_LOCK_ASSERT(child->kg_proc, MA_OWNED);
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
/* XXX Need something better here */
- Fix the maximum slice value. I accidentally checked in a value of '2' which meant no process would run for longer than 20ms. - Slightly redo the interactivity scorer. It follows the same algorithm but in a slightly more correct way. Previously values above half were incorrect. - Lower the interactivity threshold to 20. It seems that in testing non- interactive tasks are hardly ever near there and expensive interactive tasks can sometimes surpass it. This area needs more testing. - Remove an unnecessary KTR. - Fix a case where an idle thread that had an elevated priority due to priority prop. would be placed back on the idle queue. - Delay setting NEEDRESCHED until userret() for threads that haad their priority elevated while in kernel. This gives us the same context switch optimization as SCHED_4BSD. - Limit the child's slice to 1 in sched_fork_kse() so we detect its behavior more quickly. - Inhert some of the run/slp time from the child in sched_exit_ksegrp(). - Redo some of the priority comparisons so they are more clear. - Throttle the frequency of sched_pctcpu_update() so that rounding errors do not make it invalid.
2003-06-15 02:18:29 +00:00
- Throttle the inherited sleep and run time in sched_fork_kseg(). This allows us to learn the behavior of a thread much more quickly after it starts up.
2003-06-28 06:19:56 +00:00
child->kg_slptime = kg->kg_slptime / SCHED_SLP_RUN_THROTTLE;
child->kg_runtime = kg->kg_runtime / SCHED_SLP_RUN_THROTTLE;
- Add a new function "sched_interact_update()" that scales back the sleep and run time. - Scale the sleep and run time back via sched_interact_update() in more places. This is to keep the statistic more accurate. - Charge a parent one tick for forking a child. - Add only the run time and not the sleep time to the parents kg when a thread exits. This allows us to give a penalty for having an expensive thread exit but does not give a bonus for having an interactive thread exit. - Change the SLP_RUN_THROTTLE to limit us to 4/5th and not 1/2. - Change the SLP_RUN_MAX to two seconds. This keeps bursty interactive applications like mozilla and openoffice in the interactive range even through expensive tasks. - Recalculate the slice after every sleep. This ensures that once a task has been marked interactive it only has a slice of 1 at the risk of giving tasks that sleep for a very brief period a longer time slice.
2003-06-17 06:39:51 +00:00
kg->kg_runtime += tickincr << 10;
sched_interact_update(kg);
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
child->kg_user_pri = kg->kg_user_pri;
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
child->kg_nice = kg->kg_nice;
}
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
void
sched_fork_thread(struct thread *td, struct thread *child)
{
}
void
sched_class(struct ksegrp *kg, int class)
{
struct kseq *kseq;
struct kse *ke;
Add lock assertions for various proc/thread/kse/ksegroup fields to the scheduler functions.
2003-04-23 18:51:05 +00:00
mtx_assert(&sched_lock, MA_OWNED);
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
if (kg->kg_pri_class == class)
return;
FOREACH_KSE_IN_GROUP(kg, ke) {
if (ke->ke_state != KES_ONRUNQ &&
ke->ke_state != KES_THREAD)
continue;
kseq = KSEQ_CPU(ke->ke_cpu);
- We only care about the base priority. Ignore the SCHED_FIFO_BIT so that we dont get confused. Reported and debugged by: Steve Kargl <sgk@troutmask.apl.washington.edu>
2003-04-12 07:00:16 +00:00
kseq->ksq_loads[PRI_BASE(kg->kg_pri_class)]--;
kseq->ksq_loads[PRI_BASE(class)]++;
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
if (kg->kg_pri_class == PRI_TIMESHARE)
kseq_nice_rem(kseq, kg->kg_nice);
else if (class == PRI_TIMESHARE)
kseq_nice_add(kseq, kg->kg_nice);
- Fix the ksq_load calculation. It now reflects the number of entries on the run queue for each cpu. - Introduce kse stealing into the sched_choose() code. This helps balance cpus better in cases where process turnover is high. This implementation is fairly trivial and will likely be only a temporary measure until something more sophisticated has been written.
2003-01-28 09:28:20 +00:00
}
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
kg->kg_pri_class = class;
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
}
/*
* Return some of the child's priority and interactivity to the parent.
*/
void
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
sched_exit(struct proc *p, struct proc *child)
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
{
/* XXX Need something better here */
mtx_assert(&sched_lock, MA_OWNED);
- Add sched_exit_* - Call sched_exit_kse() from sched_exit() instead of implementing it here.
2003-04-11 19:24:00 +00:00
sched_exit_kse(FIRST_KSE_IN_PROC(p), FIRST_KSE_IN_PROC(child));
- Fix the maximum slice value. I accidentally checked in a value of '2' which meant no process would run for longer than 20ms. - Slightly redo the interactivity scorer. It follows the same algorithm but in a slightly more correct way. Previously values above half were incorrect. - Lower the interactivity threshold to 20. It seems that in testing non- interactive tasks are hardly ever near there and expensive interactive tasks can sometimes surpass it. This area needs more testing. - Remove an unnecessary KTR. - Fix a case where an idle thread that had an elevated priority due to priority prop. would be placed back on the idle queue. - Delay setting NEEDRESCHED until userret() for threads that haad their priority elevated while in kernel. This gives us the same context switch optimization as SCHED_4BSD. - Limit the child's slice to 1 in sched_fork_kse() so we detect its behavior more quickly. - Inhert some of the run/slp time from the child in sched_exit_ksegrp(). - Redo some of the priority comparisons so they are more clear. - Throttle the frequency of sched_pctcpu_update() so that rounding errors do not make it invalid.
2003-06-15 02:18:29 +00:00
sched_exit_ksegrp(FIRST_KSEGRP_IN_PROC(p), FIRST_KSEGRP_IN_PROC(child));
- Add sched_exit_* - Call sched_exit_kse() from sched_exit() instead of implementing it here.
2003-04-11 19:24:00 +00:00
}
void
sched_exit_kse(struct kse *ke, struct kse *child)
{
kseq_rem(KSEQ_CPU(child->ke_cpu), child);
}
void
sched_exit_ksegrp(struct ksegrp *kg, struct ksegrp *child)
{
- Add a new function "sched_interact_update()" that scales back the sleep and run time. - Scale the sleep and run time back via sched_interact_update() in more places. This is to keep the statistic more accurate. - Charge a parent one tick for forking a child. - Add only the run time and not the sleep time to the parents kg when a thread exits. This allows us to give a penalty for having an expensive thread exit but does not give a bonus for having an interactive thread exit. - Change the SLP_RUN_THROTTLE to limit us to 4/5th and not 1/2. - Change the SLP_RUN_MAX to two seconds. This keeps bursty interactive applications like mozilla and openoffice in the interactive range even through expensive tasks. - Recalculate the slice after every sleep. This ensures that once a task has been marked interactive it only has a slice of 1 at the risk of giving tasks that sleep for a very brief period a longer time slice.
2003-06-17 06:39:51 +00:00
/* kg->kg_slptime += child->kg_slptime; */
- Fix the maximum slice value. I accidentally checked in a value of '2' which meant no process would run for longer than 20ms. - Slightly redo the interactivity scorer. It follows the same algorithm but in a slightly more correct way. Previously values above half were incorrect. - Lower the interactivity threshold to 20. It seems that in testing non- interactive tasks are hardly ever near there and expensive interactive tasks can sometimes surpass it. This area needs more testing. - Remove an unnecessary KTR. - Fix a case where an idle thread that had an elevated priority due to priority prop. would be placed back on the idle queue. - Delay setting NEEDRESCHED until userret() for threads that haad their priority elevated while in kernel. This gives us the same context switch optimization as SCHED_4BSD. - Limit the child's slice to 1 in sched_fork_kse() so we detect its behavior more quickly. - Inhert some of the run/slp time from the child in sched_exit_ksegrp(). - Redo some of the priority comparisons so they are more clear. - Throttle the frequency of sched_pctcpu_update() so that rounding errors do not make it invalid.
2003-06-15 02:18:29 +00:00
kg->kg_runtime += child->kg_runtime;
- Add a new function "sched_interact_update()" that scales back the sleep and run time. - Scale the sleep and run time back via sched_interact_update() in more places. This is to keep the statistic more accurate. - Charge a parent one tick for forking a child. - Add only the run time and not the sleep time to the parents kg when a thread exits. This allows us to give a penalty for having an expensive thread exit but does not give a bonus for having an interactive thread exit. - Change the SLP_RUN_THROTTLE to limit us to 4/5th and not 1/2. - Change the SLP_RUN_MAX to two seconds. This keeps bursty interactive applications like mozilla and openoffice in the interactive range even through expensive tasks. - Recalculate the slice after every sleep. This ensures that once a task has been marked interactive it only has a slice of 1 at the risk of giving tasks that sleep for a very brief period a longer time slice.
2003-06-17 06:39:51 +00:00
sched_interact_update(kg);
- Add sched_exit_* - Call sched_exit_kse() from sched_exit() instead of implementing it here.
2003-04-11 19:24:00 +00:00
}
void
sched_exit_thread(struct thread *td, struct thread *child)
{
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
}
void
- Update the sched api. sched_{add,rem,clock,pctcpu} now all accept a td argument rather than a kse.
2003-10-16 08:39:15 +00:00
sched_clock(struct thread *td)
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
{
struct kseq *kseq;
- Use ksq_load as the authoritive count of kses on the pair of kseqs for sched_runnable() et all. - Remove some dead code in sched_clock(). - Define two macros KSEQ_SELF() and KSEQ_CPU() for getting the kseq of the current cpu or some alternate cpu. - Start introducing kseq_() functions, such as kseq_choose() and kseq_setup().
2003-01-29 07:00:51 +00:00
struct ksegrp *kg;
- Update the sched api. sched_{add,rem,clock,pctcpu} now all accept a td argument rather than a kse.
2003-10-16 08:39:15 +00:00
struct kse *ke;
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
/*
* sched_setup() apparently happens prior to stathz being set. We
* need to resolve the timers earlier in the boot so we can avoid
* calculating this here.
*/
if (realstathz == 0) {
realstathz = stathz ? stathz : hz;
tickincr = hz / realstathz;
/*
* XXX This does not work for values of stathz that are much
* larger than hz.
*/
if (tickincr == 0)
tickincr = 1;
}
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
- Update the sched api. sched_{add,rem,clock,pctcpu} now all accept a td argument rather than a kse.
2003-10-16 08:39:15 +00:00
ke = td->td_kse;
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
kg = ke->ke_ksegrp;
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
- Use ksq_load as the authoritive count of kses on the pair of kseqs for sched_runnable() et all. - Remove some dead code in sched_clock(). - Define two macros KSEQ_SELF() and KSEQ_CPU() for getting the kseq of the current cpu or some alternate cpu. - Start introducing kseq_() functions, such as kseq_choose() and kseq_setup().
2003-01-29 07:00:51 +00:00
mtx_assert(&sched_lock, MA_OWNED);
KASSERT((td != NULL), ("schedclock: null thread pointer"));
/* Adjust ticks for pctcpu */
- Shift the tick count by 10 and back around sched_pctcpu_update() calculations. Keep this changes local to the function so the tick count is in its natural form otherwise. Previously 1000 was added each time a tick fired and we divided by 1000 when it was reported. This is done to reduce rounding errors.
2003-03-03 05:29:09 +00:00
ke->ke_ticks++;
- Allow idle's pctcpu time to be calculated.
2003-01-28 09:30:17 +00:00
ke->ke_ltick = ticks;
- Keep seperate statistics and run queues for different scheduling classes. - Treat each class specially in kseq_{choose,add,rem}. Let the rest of the code be less aware of scheduling classes. - Skip the interactivity calculation for non TIMESHARE ksegrps. - Move slice and runq selection into kseq_add(). Uninline it now that it's big.
2003-04-03 00:29:28 +00:00
- Allow idle's pctcpu time to be calculated.
2003-01-28 09:30:17 +00:00
/* Go up to one second beyond our max and then trim back down */
if (ke->ke_ftick + SCHED_CPU_TICKS + hz < ke->ke_ltick)
sched_pctcpu_update(ke);
Fix typo in last commit
2003-05-02 06:18:55 +00:00
if (td->td_flags & TDF_IDLETD)
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
return;
- Use ksq_load as the authoritive count of kses on the pair of kseqs for sched_runnable() et all. - Remove some dead code in sched_clock(). - Define two macros KSEQ_SELF() and KSEQ_CPU() for getting the kseq of the current cpu or some alternate cpu. - Start introducing kseq_() functions, such as kseq_choose() and kseq_setup().
2003-01-29 07:00:51 +00:00
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
CTR4(KTR_ULE, "Tick kse %p (slice: %d, slptime: %d, runtime: %d)",
ke, ke->ke_slice, kg->kg_slptime >> 10, kg->kg_runtime >> 10);
/*
- Use a better algorithm in sched_pctcpu_update() Contributed by: Thomaswuerfl@gmx.de - In sched_prio(), adjust the run queue for threads which may need to move to the current queue due to priority propagation . - In sched_switch(), fix style bug introduced when the KSE support went in. Columns are 80 chars wide, not 90. - In sched_switch(), Fix the comparison in the idle case and explicitly re-initialize the runq in the not propagated case. - Remove dead code in sched_clock(). - In sched_clock(), If we're an IDLE class td set NEEDRESCHED so that threads that have become runnable will get a chance to. - In sched_runnable(), if we're not the IDLETD, we should not consider curthread when examining the load. This mimics the 4BSD behavior of returning 0 when the only runnable thread is running. - In sched_userret(), remove the code for setting NEEDRESCHED entirely. This is not necessary and is not implemented in 4BSD. - Use the correct comparison in sched_add() when checking to see if an idle prio task has had it's priority temporarily elevated.
2003-10-27 06:47:05 +00:00
* We only do slicing code for TIMESHARE ksegrps.
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
*/
- Use a better algorithm in sched_pctcpu_update() Contributed by: Thomaswuerfl@gmx.de - In sched_prio(), adjust the run queue for threads which may need to move to the current queue due to priority propagation . - In sched_switch(), fix style bug introduced when the KSE support went in. Columns are 80 chars wide, not 90. - In sched_switch(), Fix the comparison in the idle case and explicitly re-initialize the runq in the not propagated case. - Remove dead code in sched_clock(). - In sched_clock(), If we're an IDLE class td set NEEDRESCHED so that threads that have become runnable will get a chance to. - In sched_runnable(), if we're not the IDLETD, we should not consider curthread when examining the load. This mimics the 4BSD behavior of returning 0 when the only runnable thread is running. - In sched_userret(), remove the code for setting NEEDRESCHED entirely. This is not necessary and is not implemented in 4BSD. - Use the correct comparison in sched_add() when checking to see if an idle prio task has had it's priority temporarily elevated.
2003-10-27 06:47:05 +00:00
if (kg->kg_pri_class != PRI_TIMESHARE)
return;
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
/*
- Add a new variable 'kg_runtime' that tracks the amount of time we've run. - Use the ratio of kg_runtime / kg_slptime to determine our dynamic priority. - Scale kg_runtime and kg_slptime back when the sum of the two exceeds SCHED_SLP_RUN_MAX. This allows us to slowly forget old behavior. - Scale back the runtime and slptime in fork so that the new process has the same ratio but much less accumulated time. This causes new behavior to be noticed more quickly.
2003-02-10 14:03:45 +00:00
* We used a tick charge it to the ksegrp so that we can compute our
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
* interactivity.
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
*/
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
kg->kg_runtime += tickincr << 10;
- Add a new function "sched_interact_update()" that scales back the sleep and run time. - Scale the sleep and run time back via sched_interact_update() in more places. This is to keep the statistic more accurate. - Charge a parent one tick for forking a child. - Add only the run time and not the sleep time to the parents kg when a thread exits. This allows us to give a penalty for having an expensive thread exit but does not give a bonus for having an interactive thread exit. - Change the SLP_RUN_THROTTLE to limit us to 4/5th and not 1/2. - Change the SLP_RUN_MAX to two seconds. This keeps bursty interactive applications like mozilla and openoffice in the interactive range even through expensive tasks. - Recalculate the slice after every sleep. This ensures that once a task has been marked interactive it only has a slice of 1 at the risk of giving tasks that sleep for a very brief period a longer time slice.
2003-06-17 06:39:51 +00:00
sched_interact_update(kg);
- Add a new variable 'kg_runtime' that tracks the amount of time we've run. - Use the ratio of kg_runtime / kg_slptime to determine our dynamic priority. - Scale kg_runtime and kg_slptime back when the sum of the two exceeds SCHED_SLP_RUN_MAX. This allows us to slowly forget old behavior. - Scale back the runtime and slptime in fork so that the new process has the same ratio but much less accumulated time. This causes new behavior to be noticed more quickly.
2003-02-10 14:03:45 +00:00
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
/*
* We used up one time slice.
*/
ke->ke_slice--;
- Use a better algorithm in sched_pctcpu_update() Contributed by: Thomaswuerfl@gmx.de - In sched_prio(), adjust the run queue for threads which may need to move to the current queue due to priority propagation . - In sched_switch(), fix style bug introduced when the KSE support went in. Columns are 80 chars wide, not 90. - In sched_switch(), Fix the comparison in the idle case and explicitly re-initialize the runq in the not propagated case. - Remove dead code in sched_clock(). - In sched_clock(), If we're an IDLE class td set NEEDRESCHED so that threads that have become runnable will get a chance to. - In sched_runnable(), if we're not the IDLETD, we should not consider curthread when examining the load. This mimics the 4BSD behavior of returning 0 when the only runnable thread is running. - In sched_userret(), remove the code for setting NEEDRESCHED entirely. This is not necessary and is not implemented in 4BSD. - Use the correct comparison in sched_add() when checking to see if an idle prio task has had it's priority temporarily elevated.
2003-10-27 06:47:05 +00:00
kseq = KSEQ_SELF();
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
#ifdef SMP
- When migrating a kse from one kseq to the next actually insert it onto the second kseq's run queue so that it is referenced by the kse when it is switched out. - Spell ksq_rslices properly. Reported by: Ian Freislich <ianf@za.uu.net>
2003-04-11 18:37:34 +00:00
kseq->ksq_rslices--;
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
#endif
if (ke->ke_slice > 0)
return;
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
/*
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
* We're out of time, recompute priorities and requeue.
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
*/
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
kseq_rem(kseq, ke);
sched_priority(kg);
sched_slice(ke);
if (SCHED_CURR(kg, ke))
ke->ke_runq = kseq->ksq_curr;
else
ke->ke_runq = kseq->ksq_next;
kseq_add(kseq, ke);
td->td_flags |= TDF_NEEDRESCHED;
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
}
int
sched_runnable(void)
{
struct kseq *kseq;
- When a new thread is added to a kseq the load is incremented prior to adding it to the nice tables. Therefore, in kseq_add_nice, we should keep in mind that the load will be 1 if we are the only thread, and not 0. - Assert that the sched lock is held in all the appropriate places. - Increase the scope of the sched lock in sched_pctcpu_update(). - Hold the sched lock in sched_runnable(). It is not held by the caller.
2003-06-08 00:47:33 +00:00
int load;
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
- When a new thread is added to a kseq the load is incremented prior to adding it to the nice tables. Therefore, in kseq_add_nice, we should keep in mind that the load will be 1 if we are the only thread, and not 0. - Assert that the sched lock is held in all the appropriate places. - Increase the scope of the sched lock in sched_pctcpu_update(). - Hold the sched lock in sched_runnable(). It is not held by the caller.
2003-06-08 00:47:33 +00:00
load = 1;
mtx_lock_spin(&sched_lock);
- Use ksq_load as the authoritive count of kses on the pair of kseqs for sched_runnable() et all. - Remove some dead code in sched_clock(). - Define two macros KSEQ_SELF() and KSEQ_CPU() for getting the kseq of the current cpu or some alternate cpu. - Start introducing kseq_() functions, such as kseq_choose() and kseq_setup().
2003-01-29 07:00:51 +00:00
kseq = KSEQ_SELF();
- Add static to local functions and data where it was missing. - Add an IPI based mechanism for migrating kses. This mechanism is broken down into several components. This is intended to reduce cache thrashing by eliminating most cases where one cpu touches another's run queues. - kseq_notify() appends a kse to a lockless singly linked list and conditionally sends an IPI to the target processor. Right now this is protected by sched_lock but at some point I'd like to get rid of the global lock. This is why I used something more complicated than a standard queue. - kseq_assign() processes our list of kses that have been assigned to us by other processors. This simply calls sched_add() for each item on the list after clearing the new KEF_ASSIGNED flag. This flag is used to indicate that we have been appeneded to the assigned queue but not added to the run queue yet. - In sched_add(), instead of adding a KSE to another processor's queue we use kse_notify() so that we don't touch their queue. Also in sched_add(), if KEF_ASSIGNED is already set return immediately. This can happen if a thread is removed and readded so that the priority is recorded properly. - In sched_rem() return immediately if KEF_ASSIGNED is set. All callers immediately readd simply to adjust priorites etc. - In sched_choose(), if we're running an IDLE task or the per cpu idle thread set our cpumask bit in 'kseq_idle' so that other processors may know that we are idle. Before this, make a single pass through the run queues of other processors so that we may find work more immediately if it is available. - In sched_runnable(), don't scan each processor's run queue, they will IPI us if they have work for us to do. - In sched_add(), if we're adding a thread that can be migrated and we have plenty of work to do, try to migrate the thread to an idle kseq. - Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into consideration. - No longer use kseq_choose() to steal threads, it can lose it's last argument. - Create a new function runq_steal() which operates like runq_choose() but skips threads based on some criteria. Currently it will not steal PRI_ITHD threads. In the future this will be used for CPU binding. - Create a kseq_steal() that checks each run queue with runq_steal(), use kseq_steal() in the places where we used kseq_choose() to steal with before.
2003-10-31 11:16:04 +00:00
#ifdef SMP
if (kseq->ksq_assigned)
kseq_assign(kseq);
#endif
- Use a better algorithm in sched_pctcpu_update() Contributed by: Thomaswuerfl@gmx.de - In sched_prio(), adjust the run queue for threads which may need to move to the current queue due to priority propagation . - In sched_switch(), fix style bug introduced when the KSE support went in. Columns are 80 chars wide, not 90. - In sched_switch(), Fix the comparison in the idle case and explicitly re-initialize the runq in the not propagated case. - Remove dead code in sched_clock(). - In sched_clock(), If we're an IDLE class td set NEEDRESCHED so that threads that have become runnable will get a chance to. - In sched_runnable(), if we're not the IDLETD, we should not consider curthread when examining the load. This mimics the 4BSD behavior of returning 0 when the only runnable thread is running. - In sched_userret(), remove the code for setting NEEDRESCHED entirely. This is not necessary and is not implemented in 4BSD. - Use the correct comparison in sched_add() when checking to see if an idle prio task has had it's priority temporarily elevated.
2003-10-27 06:47:05 +00:00
if ((curthread->td_flags & TDF_IDLETD) != 0) {
if (kseq->ksq_load > 0)
goto out;
} else
if (kseq->ksq_load - 1 > 0)
goto out;
- When a new thread is added to a kseq the load is incremented prior to adding it to the nice tables. Therefore, in kseq_add_nice, we should keep in mind that the load will be 1 if we are the only thread, and not 0. - Assert that the sched lock is held in all the appropriate places. - Increase the scope of the sched lock in sched_pctcpu_update(). - Hold the sched lock in sched_runnable(). It is not held by the caller.
2003-06-08 00:47:33 +00:00
load = 0;
out:
mtx_unlock_spin(&sched_lock);
return (load);
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
}
void
sched_userret(struct thread *td)
{
struct ksegrp *kg;
- Use a better algorithm in sched_pctcpu_update() Contributed by: Thomaswuerfl@gmx.de - In sched_prio(), adjust the run queue for threads which may need to move to the current queue due to priority propagation . - In sched_switch(), fix style bug introduced when the KSE support went in. Columns are 80 chars wide, not 90. - In sched_switch(), Fix the comparison in the idle case and explicitly re-initialize the runq in the not propagated case. - Remove dead code in sched_clock(). - In sched_clock(), If we're an IDLE class td set NEEDRESCHED so that threads that have become runnable will get a chance to. - In sched_runnable(), if we're not the IDLETD, we should not consider curthread when examining the load. This mimics the 4BSD behavior of returning 0 when the only runnable thread is running. - In sched_userret(), remove the code for setting NEEDRESCHED entirely. This is not necessary and is not implemented in 4BSD. - Use the correct comparison in sched_add() when checking to see if an idle prio task has had it's priority temporarily elevated.
2003-10-27 06:47:05 +00:00
kg = td->td_ksegrp;
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
if (td->td_priority != kg->kg_user_pri) {
mtx_lock_spin(&sched_lock);
td->td_priority = kg->kg_user_pri;
mtx_unlock_spin(&sched_lock);
}
}
- Fix the ksq_load calculation. It now reflects the number of entries on the run queue for each cpu. - Introduce kse stealing into the sched_choose() code. This helps balance cpus better in cases where process turnover is high. This implementation is fairly trivial and will likely be only a temporary measure until something more sophisticated has been written.
2003-01-28 09:28:20 +00:00
struct kse *
sched_choose(void)
{
- Use ksq_load as the authoritive count of kses on the pair of kseqs for sched_runnable() et all. - Remove some dead code in sched_clock(). - Define two macros KSEQ_SELF() and KSEQ_CPU() for getting the kseq of the current cpu or some alternate cpu. - Start introducing kseq_() functions, such as kseq_choose() and kseq_setup().
2003-01-29 07:00:51 +00:00
struct kseq *kseq;
- Fix the ksq_load calculation. It now reflects the number of entries on the run queue for each cpu. - Introduce kse stealing into the sched_choose() code. This helps balance cpus better in cases where process turnover is high. This implementation is fairly trivial and will likely be only a temporary measure until something more sophisticated has been written.
2003-01-28 09:28:20 +00:00
struct kse *ke;
- When a new thread is added to a kseq the load is incremented prior to adding it to the nice tables. Therefore, in kseq_add_nice, we should keep in mind that the load will be 1 if we are the only thread, and not 0. - Assert that the sched lock is held in all the appropriate places. - Increase the scope of the sched lock in sched_pctcpu_update(). - Hold the sched lock in sched_runnable(). It is not held by the caller.
2003-06-08 00:47:33 +00:00
mtx_assert(&sched_lock, MA_OWNED);
- Add static to local functions and data where it was missing. - Add an IPI based mechanism for migrating kses. This mechanism is broken down into several components. This is intended to reduce cache thrashing by eliminating most cases where one cpu touches another's run queues. - kseq_notify() appends a kse to a lockless singly linked list and conditionally sends an IPI to the target processor. Right now this is protected by sched_lock but at some point I'd like to get rid of the global lock. This is why I used something more complicated than a standard queue. - kseq_assign() processes our list of kses that have been assigned to us by other processors. This simply calls sched_add() for each item on the list after clearing the new KEF_ASSIGNED flag. This flag is used to indicate that we have been appeneded to the assigned queue but not added to the run queue yet. - In sched_add(), instead of adding a KSE to another processor's queue we use kse_notify() so that we don't touch their queue. Also in sched_add(), if KEF_ASSIGNED is already set return immediately. This can happen if a thread is removed and readded so that the priority is recorded properly. - In sched_rem() return immediately if KEF_ASSIGNED is set. All callers immediately readd simply to adjust priorites etc. - In sched_choose(), if we're running an IDLE task or the per cpu idle thread set our cpumask bit in 'kseq_idle' so that other processors may know that we are idle. Before this, make a single pass through the run queues of other processors so that we may find work more immediately if it is available. - In sched_runnable(), don't scan each processor's run queue, they will IPI us if they have work for us to do. - In sched_add(), if we're adding a thread that can be migrated and we have plenty of work to do, try to migrate the thread to an idle kseq. - Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into consideration. - No longer use kseq_choose() to steal threads, it can lose it's last argument. - Create a new function runq_steal() which operates like runq_choose() but skips threads based on some criteria. Currently it will not steal PRI_ITHD threads. In the future this will be used for CPU binding. - Create a kseq_steal() that checks each run queue with runq_steal(), use kseq_steal() in the places where we used kseq_choose() to steal with before.
2003-10-31 11:16:04 +00:00
kseq = KSEQ_SELF();
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
#ifdef SMP
- Add in support for KSEs with 0 slice values on the run queue. If we try to select a KSE with a slice of 0 we will update its slice and insert it onto the next queue. - Pass the KSE instead of the ksegrp into sched_slice(). This more accurately reflects the behavior of the code. Slices are granted to kses. - Add a function kseq_nice_min() which finds the smallest nice value assigned to the kseg of any KSE on the queue. - Rewrite the logic in sched_slice(). Add a large comment describing the new slice selection scheme. To summarize, slices are assigned based on the nice value. Priorities are still calculated based on the nice and interactivity of a process. Slice sizes of 0 may be granted for KSEs whos nice is 20 or futher away from the lowest nice on the run queue. Other nice values are scaled across the range [min, min+20]. This fixes ULEs bad behavior with positively niced processes.
2003-04-02 06:46:43 +00:00
retry:
- Add static to local functions and data where it was missing. - Add an IPI based mechanism for migrating kses. This mechanism is broken down into several components. This is intended to reduce cache thrashing by eliminating most cases where one cpu touches another's run queues. - kseq_notify() appends a kse to a lockless singly linked list and conditionally sends an IPI to the target processor. Right now this is protected by sched_lock but at some point I'd like to get rid of the global lock. This is why I used something more complicated than a standard queue. - kseq_assign() processes our list of kses that have been assigned to us by other processors. This simply calls sched_add() for each item on the list after clearing the new KEF_ASSIGNED flag. This flag is used to indicate that we have been appeneded to the assigned queue but not added to the run queue yet. - In sched_add(), instead of adding a KSE to another processor's queue we use kse_notify() so that we don't touch their queue. Also in sched_add(), if KEF_ASSIGNED is already set return immediately. This can happen if a thread is removed and readded so that the priority is recorded properly. - In sched_rem() return immediately if KEF_ASSIGNED is set. All callers immediately readd simply to adjust priorites etc. - In sched_choose(), if we're running an IDLE task or the per cpu idle thread set our cpumask bit in 'kseq_idle' so that other processors may know that we are idle. Before this, make a single pass through the run queues of other processors so that we may find work more immediately if it is available. - In sched_runnable(), don't scan each processor's run queue, they will IPI us if they have work for us to do. - In sched_add(), if we're adding a thread that can be migrated and we have plenty of work to do, try to migrate the thread to an idle kseq. - Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into consideration. - No longer use kseq_choose() to steal threads, it can lose it's last argument. - Create a new function runq_steal() which operates like runq_choose() but skips threads based on some criteria. Currently it will not steal PRI_ITHD threads. In the future this will be used for CPU binding. - Create a kseq_steal() that checks each run queue with runq_steal(), use kseq_steal() in the places where we used kseq_choose() to steal with before.
2003-10-31 11:16:04 +00:00
if (kseq->ksq_assigned)
kseq_assign(kseq);
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
#endif
- Add static to local functions and data where it was missing. - Add an IPI based mechanism for migrating kses. This mechanism is broken down into several components. This is intended to reduce cache thrashing by eliminating most cases where one cpu touches another's run queues. - kseq_notify() appends a kse to a lockless singly linked list and conditionally sends an IPI to the target processor. Right now this is protected by sched_lock but at some point I'd like to get rid of the global lock. This is why I used something more complicated than a standard queue. - kseq_assign() processes our list of kses that have been assigned to us by other processors. This simply calls sched_add() for each item on the list after clearing the new KEF_ASSIGNED flag. This flag is used to indicate that we have been appeneded to the assigned queue but not added to the run queue yet. - In sched_add(), instead of adding a KSE to another processor's queue we use kse_notify() so that we don't touch their queue. Also in sched_add(), if KEF_ASSIGNED is already set return immediately. This can happen if a thread is removed and readded so that the priority is recorded properly. - In sched_rem() return immediately if KEF_ASSIGNED is set. All callers immediately readd simply to adjust priorites etc. - In sched_choose(), if we're running an IDLE task or the per cpu idle thread set our cpumask bit in 'kseq_idle' so that other processors may know that we are idle. Before this, make a single pass through the run queues of other processors so that we may find work more immediately if it is available. - In sched_runnable(), don't scan each processor's run queue, they will IPI us if they have work for us to do. - In sched_add(), if we're adding a thread that can be migrated and we have plenty of work to do, try to migrate the thread to an idle kseq. - Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into consideration. - No longer use kseq_choose() to steal threads, it can lose it's last argument. - Create a new function runq_steal() which operates like runq_choose() but skips threads based on some criteria. Currently it will not steal PRI_ITHD threads. In the future this will be used for CPU binding. - Create a kseq_steal() that checks each run queue with runq_steal(), use kseq_steal() in the places where we used kseq_choose() to steal with before.
2003-10-31 11:16:04 +00:00
ke = kseq_choose(kseq);
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
if (ke) {
- Add static to local functions and data where it was missing. - Add an IPI based mechanism for migrating kses. This mechanism is broken down into several components. This is intended to reduce cache thrashing by eliminating most cases where one cpu touches another's run queues. - kseq_notify() appends a kse to a lockless singly linked list and conditionally sends an IPI to the target processor. Right now this is protected by sched_lock but at some point I'd like to get rid of the global lock. This is why I used something more complicated than a standard queue. - kseq_assign() processes our list of kses that have been assigned to us by other processors. This simply calls sched_add() for each item on the list after clearing the new KEF_ASSIGNED flag. This flag is used to indicate that we have been appeneded to the assigned queue but not added to the run queue yet. - In sched_add(), instead of adding a KSE to another processor's queue we use kse_notify() so that we don't touch their queue. Also in sched_add(), if KEF_ASSIGNED is already set return immediately. This can happen if a thread is removed and readded so that the priority is recorded properly. - In sched_rem() return immediately if KEF_ASSIGNED is set. All callers immediately readd simply to adjust priorites etc. - In sched_choose(), if we're running an IDLE task or the per cpu idle thread set our cpumask bit in 'kseq_idle' so that other processors may know that we are idle. Before this, make a single pass through the run queues of other processors so that we may find work more immediately if it is available. - In sched_runnable(), don't scan each processor's run queue, they will IPI us if they have work for us to do. - In sched_add(), if we're adding a thread that can be migrated and we have plenty of work to do, try to migrate the thread to an idle kseq. - Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into consideration. - No longer use kseq_choose() to steal threads, it can lose it's last argument. - Create a new function runq_steal() which operates like runq_choose() but skips threads based on some criteria. Currently it will not steal PRI_ITHD threads. In the future this will be used for CPU binding. - Create a kseq_steal() that checks each run queue with runq_steal(), use kseq_steal() in the places where we used kseq_choose() to steal with before.
2003-10-31 11:16:04 +00:00
#ifdef SMP
if (ke->ke_ksegrp->kg_pri_class == PRI_IDLE)
if (kseq_find())
goto retry;
#endif
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
runq_remove(ke->ke_runq, ke);
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
ke->ke_state = KES_THREAD;
- Add in support for KSEs with 0 slice values on the run queue. If we try to select a KSE with a slice of 0 we will update its slice and insert it onto the next queue. - Pass the KSE instead of the ksegrp into sched_slice(). This more accurately reflects the behavior of the code. Slices are granted to kses. - Add a function kseq_nice_min() which finds the smallest nice value assigned to the kseg of any KSE on the queue. - Rewrite the logic in sched_slice(). Add a large comment describing the new slice selection scheme. To summarize, slices are assigned based on the nice value. Priorities are still calculated based on the nice and interactivity of a process. Slice sizes of 0 may be granted for KSEs whos nice is 20 or futher away from the lowest nice on the run queue. Other nice values are scaled across the range [min, min+20]. This fixes ULEs bad behavior with positively niced processes.
2003-04-02 06:46:43 +00:00
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
if (ke->ke_ksegrp->kg_pri_class == PRI_TIMESHARE) {
CTR4(KTR_ULE, "Run kse %p from %p (slice: %d, pri: %d)",
ke, ke->ke_runq, ke->ke_slice,
ke->ke_thread->td_priority);
}
return (ke);
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
}
- Fix the ksq_load calculation. It now reflects the number of entries on the run queue for each cpu. - Introduce kse stealing into the sched_choose() code. This helps balance cpus better in cases where process turnover is high. This implementation is fairly trivial and will likely be only a temporary measure until something more sophisticated has been written.
2003-01-28 09:28:20 +00:00
#ifdef SMP
- Add static to local functions and data where it was missing. - Add an IPI based mechanism for migrating kses. This mechanism is broken down into several components. This is intended to reduce cache thrashing by eliminating most cases where one cpu touches another's run queues. - kseq_notify() appends a kse to a lockless singly linked list and conditionally sends an IPI to the target processor. Right now this is protected by sched_lock but at some point I'd like to get rid of the global lock. This is why I used something more complicated than a standard queue. - kseq_assign() processes our list of kses that have been assigned to us by other processors. This simply calls sched_add() for each item on the list after clearing the new KEF_ASSIGNED flag. This flag is used to indicate that we have been appeneded to the assigned queue but not added to the run queue yet. - In sched_add(), instead of adding a KSE to another processor's queue we use kse_notify() so that we don't touch their queue. Also in sched_add(), if KEF_ASSIGNED is already set return immediately. This can happen if a thread is removed and readded so that the priority is recorded properly. - In sched_rem() return immediately if KEF_ASSIGNED is set. All callers immediately readd simply to adjust priorites etc. - In sched_choose(), if we're running an IDLE task or the per cpu idle thread set our cpumask bit in 'kseq_idle' so that other processors may know that we are idle. Before this, make a single pass through the run queues of other processors so that we may find work more immediately if it is available. - In sched_runnable(), don't scan each processor's run queue, they will IPI us if they have work for us to do. - In sched_add(), if we're adding a thread that can be migrated and we have plenty of work to do, try to migrate the thread to an idle kseq. - Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into consideration. - No longer use kseq_choose() to steal threads, it can lose it's last argument. - Create a new function runq_steal() which operates like runq_choose() but skips threads based on some criteria. Currently it will not steal PRI_ITHD threads. In the future this will be used for CPU binding. - Create a kseq_steal() that checks each run queue with runq_steal(), use kseq_steal() in the places where we used kseq_choose() to steal with before.
2003-10-31 11:16:04 +00:00
if (kseq_find())
- When migrating a kse from one kseq to the next actually insert it onto the second kseq's run queue so that it is referenced by the kse when it is switched out. - Spell ksq_rslices properly. Reported by: Ian Freislich <ianf@za.uu.net>
2003-04-11 18:37:34 +00:00
goto retry;
- Fix the ksq_load calculation. It now reflects the number of entries on the run queue for each cpu. - Introduce kse stealing into the sched_choose() code. This helps balance cpus better in cases where process turnover is high. This implementation is fairly trivial and will likely be only a temporary measure until something more sophisticated has been written.
2003-01-28 09:28:20 +00:00
#endif
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
return (NULL);
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
}
void
- Update the sched api. sched_{add,rem,clock,pctcpu} now all accept a td argument rather than a kse.
2003-10-16 08:39:15 +00:00
sched_add(struct thread *td)
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
{
- Make some context switches conditional on SCHED_STRICT_RESCHED. This may have some negative effect on interactivity but it yields great perf. gains. This also brings the conditions under which ULE context switches inline with SCHED_4BSD. - Define some new kseq_* functions for manipulating the run queue. - Add a new kseq member ksq_rslices and ksq_bload. rslices is the sum of the slices of runnable kses. This will be used for push load balance decisions. bload is the number of threads blocked waiting on IO.
2003-02-03 05:30:07 +00:00
struct kseq *kseq;
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
struct ksegrp *kg;
- Update the sched api. sched_{add,rem,clock,pctcpu} now all accept a td argument rather than a kse.
2003-10-16 08:39:15 +00:00
struct kse *ke;
- Add static to local functions and data where it was missing. - Add an IPI based mechanism for migrating kses. This mechanism is broken down into several components. This is intended to reduce cache thrashing by eliminating most cases where one cpu touches another's run queues. - kseq_notify() appends a kse to a lockless singly linked list and conditionally sends an IPI to the target processor. Right now this is protected by sched_lock but at some point I'd like to get rid of the global lock. This is why I used something more complicated than a standard queue. - kseq_assign() processes our list of kses that have been assigned to us by other processors. This simply calls sched_add() for each item on the list after clearing the new KEF_ASSIGNED flag. This flag is used to indicate that we have been appeneded to the assigned queue but not added to the run queue yet. - In sched_add(), instead of adding a KSE to another processor's queue we use kse_notify() so that we don't touch their queue. Also in sched_add(), if KEF_ASSIGNED is already set return immediately. This can happen if a thread is removed and readded so that the priority is recorded properly. - In sched_rem() return immediately if KEF_ASSIGNED is set. All callers immediately readd simply to adjust priorites etc. - In sched_choose(), if we're running an IDLE task or the per cpu idle thread set our cpumask bit in 'kseq_idle' so that other processors may know that we are idle. Before this, make a single pass through the run queues of other processors so that we may find work more immediately if it is available. - In sched_runnable(), don't scan each processor's run queue, they will IPI us if they have work for us to do. - In sched_add(), if we're adding a thread that can be migrated and we have plenty of work to do, try to migrate the thread to an idle kseq. - Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into consideration. - No longer use kseq_choose() to steal threads, it can lose it's last argument. - Create a new function runq_steal() which operates like runq_choose() but skips threads based on some criteria. Currently it will not steal PRI_ITHD threads. In the future this will be used for CPU binding. - Create a kseq_steal() that checks each run queue with runq_steal(), use kseq_steal() in the places where we used kseq_choose() to steal with before.
2003-10-31 11:16:04 +00:00
int class;
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
- Add static to local functions and data where it was missing. - Add an IPI based mechanism for migrating kses. This mechanism is broken down into several components. This is intended to reduce cache thrashing by eliminating most cases where one cpu touches another's run queues. - kseq_notify() appends a kse to a lockless singly linked list and conditionally sends an IPI to the target processor. Right now this is protected by sched_lock but at some point I'd like to get rid of the global lock. This is why I used something more complicated than a standard queue. - kseq_assign() processes our list of kses that have been assigned to us by other processors. This simply calls sched_add() for each item on the list after clearing the new KEF_ASSIGNED flag. This flag is used to indicate that we have been appeneded to the assigned queue but not added to the run queue yet. - In sched_add(), instead of adding a KSE to another processor's queue we use kse_notify() so that we don't touch their queue. Also in sched_add(), if KEF_ASSIGNED is already set return immediately. This can happen if a thread is removed and readded so that the priority is recorded properly. - In sched_rem() return immediately if KEF_ASSIGNED is set. All callers immediately readd simply to adjust priorites etc. - In sched_choose(), if we're running an IDLE task or the per cpu idle thread set our cpumask bit in 'kseq_idle' so that other processors may know that we are idle. Before this, make a single pass through the run queues of other processors so that we may find work more immediately if it is available. - In sched_runnable(), don't scan each processor's run queue, they will IPI us if they have work for us to do. - In sched_add(), if we're adding a thread that can be migrated and we have plenty of work to do, try to migrate the thread to an idle kseq. - Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into consideration. - No longer use kseq_choose() to steal threads, it can lose it's last argument. - Create a new function runq_steal() which operates like runq_choose() but skips threads based on some criteria. Currently it will not steal PRI_ITHD threads. In the future this will be used for CPU binding. - Create a kseq_steal() that checks each run queue with runq_steal(), use kseq_steal() in the places where we used kseq_choose() to steal with before.
2003-10-31 11:16:04 +00:00
mtx_assert(&sched_lock, MA_OWNED);
- Update the sched api. sched_{add,rem,clock,pctcpu} now all accept a td argument rather than a kse.
2003-10-16 08:39:15 +00:00
ke = td->td_kse;
kg = td->td_ksegrp;
- Add static to local functions and data where it was missing. - Add an IPI based mechanism for migrating kses. This mechanism is broken down into several components. This is intended to reduce cache thrashing by eliminating most cases where one cpu touches another's run queues. - kseq_notify() appends a kse to a lockless singly linked list and conditionally sends an IPI to the target processor. Right now this is protected by sched_lock but at some point I'd like to get rid of the global lock. This is why I used something more complicated than a standard queue. - kseq_assign() processes our list of kses that have been assigned to us by other processors. This simply calls sched_add() for each item on the list after clearing the new KEF_ASSIGNED flag. This flag is used to indicate that we have been appeneded to the assigned queue but not added to the run queue yet. - In sched_add(), instead of adding a KSE to another processor's queue we use kse_notify() so that we don't touch their queue. Also in sched_add(), if KEF_ASSIGNED is already set return immediately. This can happen if a thread is removed and readded so that the priority is recorded properly. - In sched_rem() return immediately if KEF_ASSIGNED is set. All callers immediately readd simply to adjust priorites etc. - In sched_choose(), if we're running an IDLE task or the per cpu idle thread set our cpumask bit in 'kseq_idle' so that other processors may know that we are idle. Before this, make a single pass through the run queues of other processors so that we may find work more immediately if it is available. - In sched_runnable(), don't scan each processor's run queue, they will IPI us if they have work for us to do. - In sched_add(), if we're adding a thread that can be migrated and we have plenty of work to do, try to migrate the thread to an idle kseq. - Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into consideration. - No longer use kseq_choose() to steal threads, it can lose it's last argument. - Create a new function runq_steal() which operates like runq_choose() but skips threads based on some criteria. Currently it will not steal PRI_ITHD threads. In the future this will be used for CPU binding. - Create a kseq_steal() that checks each run queue with runq_steal(), use kseq_steal() in the places where we used kseq_choose() to steal with before.
2003-10-31 11:16:04 +00:00
if (ke->ke_flags & KEF_ASSIGNED)
return;
kseq = KSEQ_SELF();
- Make some context switches conditional on SCHED_STRICT_RESCHED. This may have some negative effect on interactivity but it yields great perf. gains. This also brings the conditions under which ULE context switches inline with SCHED_4BSD. - Define some new kseq_* functions for manipulating the run queue. - Add a new kseq member ksq_rslices and ksq_bload. rslices is the sum of the slices of runnable kses. This will be used for push load balance decisions. bload is the number of threads blocked waiting on IO.
2003-02-03 05:30:07 +00:00
KASSERT((ke->ke_thread != NULL), ("sched_add: No thread on KSE"));
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
KASSERT((ke->ke_thread->td_kse != NULL),
- Make some context switches conditional on SCHED_STRICT_RESCHED. This may have some negative effect on interactivity but it yields great perf. gains. This also brings the conditions under which ULE context switches inline with SCHED_4BSD. - Define some new kseq_* functions for manipulating the run queue. - Add a new kseq member ksq_rslices and ksq_bload. rslices is the sum of the slices of runnable kses. This will be used for push load balance decisions. bload is the number of threads blocked waiting on IO.
2003-02-03 05:30:07 +00:00
("sched_add: No KSE on thread"));
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
KASSERT(ke->ke_state != KES_ONRUNQ,
- Make some context switches conditional on SCHED_STRICT_RESCHED. This may have some negative effect on interactivity but it yields great perf. gains. This also brings the conditions under which ULE context switches inline with SCHED_4BSD. - Define some new kseq_* functions for manipulating the run queue. - Add a new kseq member ksq_rslices and ksq_bload. rslices is the sum of the slices of runnable kses. This will be used for push load balance decisions. bload is the number of threads blocked waiting on IO.
2003-02-03 05:30:07 +00:00
("sched_add: kse %p (%s) already in run queue", ke,
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
ke->ke_proc->p_comm));
KASSERT(ke->ke_proc->p_sflag & PS_INMEM,
- Make some context switches conditional on SCHED_STRICT_RESCHED. This may have some negative effect on interactivity but it yields great perf. gains. This also brings the conditions under which ULE context switches inline with SCHED_4BSD. - Define some new kseq_* functions for manipulating the run queue. - Add a new kseq member ksq_rslices and ksq_bload. rslices is the sum of the slices of runnable kses. This will be used for push load balance decisions. bload is the number of threads blocked waiting on IO.
2003-02-03 05:30:07 +00:00
("sched_add: process swapped out"));
- Clean up some debug code left over from my earlier megacommit.
2003-04-12 07:28:36 +00:00
KASSERT(ke->ke_runq == NULL,
("sched_add: KSE %p is still assigned to a run queue", ke));
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
- Add static to local functions and data where it was missing. - Add an IPI based mechanism for migrating kses. This mechanism is broken down into several components. This is intended to reduce cache thrashing by eliminating most cases where one cpu touches another's run queues. - kseq_notify() appends a kse to a lockless singly linked list and conditionally sends an IPI to the target processor. Right now this is protected by sched_lock but at some point I'd like to get rid of the global lock. This is why I used something more complicated than a standard queue. - kseq_assign() processes our list of kses that have been assigned to us by other processors. This simply calls sched_add() for each item on the list after clearing the new KEF_ASSIGNED flag. This flag is used to indicate that we have been appeneded to the assigned queue but not added to the run queue yet. - In sched_add(), instead of adding a KSE to another processor's queue we use kse_notify() so that we don't touch their queue. Also in sched_add(), if KEF_ASSIGNED is already set return immediately. This can happen if a thread is removed and readded so that the priority is recorded properly. - In sched_rem() return immediately if KEF_ASSIGNED is set. All callers immediately readd simply to adjust priorites etc. - In sched_choose(), if we're running an IDLE task or the per cpu idle thread set our cpumask bit in 'kseq_idle' so that other processors may know that we are idle. Before this, make a single pass through the run queues of other processors so that we may find work more immediately if it is available. - In sched_runnable(), don't scan each processor's run queue, they will IPI us if they have work for us to do. - In sched_add(), if we're adding a thread that can be migrated and we have plenty of work to do, try to migrate the thread to an idle kseq. - Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into consideration. - No longer use kseq_choose() to steal threads, it can lose it's last argument. - Create a new function runq_steal() which operates like runq_choose() but skips threads based on some criteria. Currently it will not steal PRI_ITHD threads. In the future this will be used for CPU binding. - Create a kseq_steal() that checks each run queue with runq_steal(), use kseq_steal() in the places where we used kseq_choose() to steal with before.
2003-10-31 11:16:04 +00:00
class = PRI_BASE(kg->kg_pri_class);
switch (class) {
- Keep seperate statistics and run queues for different scheduling classes. - Treat each class specially in kseq_{choose,add,rem}. Let the rest of the code be less aware of scheduling classes. - Skip the interactivity calculation for non TIMESHARE ksegrps. - Move slice and runq selection into kseq_add(). Uninline it now that it's big.
2003-04-03 00:29:28 +00:00
case PRI_ITHD:
case PRI_REALTIME:
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
ke->ke_runq = kseq->ksq_curr;
ke->ke_slice = SCHED_SLICE_MAX;
- Set the ke_cpu field in sched_add() for interrupt and realtime threads since they are going on the current cpu and not their previously assigned cpu. - sched_runnable() should only return true in the SMP case if the other processor has more than one thread that is runnable. We can not steal curthread. - Change kseq_print() to accept the cpuid instead of a kseq pointer. This makes use of this function in ddb much easier.
2003-04-18 05:24:10 +00:00
ke->ke_cpu = PCPU_GET(cpuid);
- Keep seperate statistics and run queues for different scheduling classes. - Treat each class specially in kseq_{choose,add,rem}. Let the rest of the code be less aware of scheduling classes. - Skip the interactivity calculation for non TIMESHARE ksegrps. - Move slice and runq selection into kseq_add(). Uninline it now that it's big.
2003-04-03 00:29:28 +00:00
break;
case PRI_TIMESHARE:
- Add static to local functions and data where it was missing. - Add an IPI based mechanism for migrating kses. This mechanism is broken down into several components. This is intended to reduce cache thrashing by eliminating most cases where one cpu touches another's run queues. - kseq_notify() appends a kse to a lockless singly linked list and conditionally sends an IPI to the target processor. Right now this is protected by sched_lock but at some point I'd like to get rid of the global lock. This is why I used something more complicated than a standard queue. - kseq_assign() processes our list of kses that have been assigned to us by other processors. This simply calls sched_add() for each item on the list after clearing the new KEF_ASSIGNED flag. This flag is used to indicate that we have been appeneded to the assigned queue but not added to the run queue yet. - In sched_add(), instead of adding a KSE to another processor's queue we use kse_notify() so that we don't touch their queue. Also in sched_add(), if KEF_ASSIGNED is already set return immediately. This can happen if a thread is removed and readded so that the priority is recorded properly. - In sched_rem() return immediately if KEF_ASSIGNED is set. All callers immediately readd simply to adjust priorites etc. - In sched_choose(), if we're running an IDLE task or the per cpu idle thread set our cpumask bit in 'kseq_idle' so that other processors may know that we are idle. Before this, make a single pass through the run queues of other processors so that we may find work more immediately if it is available. - In sched_runnable(), don't scan each processor's run queue, they will IPI us if they have work for us to do. - In sched_add(), if we're adding a thread that can be migrated and we have plenty of work to do, try to migrate the thread to an idle kseq. - Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into consideration. - No longer use kseq_choose() to steal threads, it can lose it's last argument. - Create a new function runq_steal() which operates like runq_choose() but skips threads based on some criteria. Currently it will not steal PRI_ITHD threads. In the future this will be used for CPU binding. - Create a kseq_steal() that checks each run queue with runq_steal(), use kseq_steal() in the places where we used kseq_choose() to steal with before.
2003-10-31 11:16:04 +00:00
#ifdef SMP
if (ke->ke_cpu != PCPU_GET(cpuid)) {
kseq_notify(ke, ke->ke_cpu);
return;
}
#endif
- Clean up some debug code left over from my earlier megacommit.
2003-04-12 07:28:36 +00:00
if (SCHED_CURR(kg, ke))
ke->ke_runq = kseq->ksq_curr;
else
ke->ke_runq = kseq->ksq_next;
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
break;
- Keep seperate statistics and run queues for different scheduling classes. - Treat each class specially in kseq_{choose,add,rem}. Let the rest of the code be less aware of scheduling classes. - Skip the interactivity calculation for non TIMESHARE ksegrps. - Move slice and runq selection into kseq_add(). Uninline it now that it's big.
2003-04-03 00:29:28 +00:00
case PRI_IDLE:
- Add static to local functions and data where it was missing. - Add an IPI based mechanism for migrating kses. This mechanism is broken down into several components. This is intended to reduce cache thrashing by eliminating most cases where one cpu touches another's run queues. - kseq_notify() appends a kse to a lockless singly linked list and conditionally sends an IPI to the target processor. Right now this is protected by sched_lock but at some point I'd like to get rid of the global lock. This is why I used something more complicated than a standard queue. - kseq_assign() processes our list of kses that have been assigned to us by other processors. This simply calls sched_add() for each item on the list after clearing the new KEF_ASSIGNED flag. This flag is used to indicate that we have been appeneded to the assigned queue but not added to the run queue yet. - In sched_add(), instead of adding a KSE to another processor's queue we use kse_notify() so that we don't touch their queue. Also in sched_add(), if KEF_ASSIGNED is already set return immediately. This can happen if a thread is removed and readded so that the priority is recorded properly. - In sched_rem() return immediately if KEF_ASSIGNED is set. All callers immediately readd simply to adjust priorites etc. - In sched_choose(), if we're running an IDLE task or the per cpu idle thread set our cpumask bit in 'kseq_idle' so that other processors may know that we are idle. Before this, make a single pass through the run queues of other processors so that we may find work more immediately if it is available. - In sched_runnable(), don't scan each processor's run queue, they will IPI us if they have work for us to do. - In sched_add(), if we're adding a thread that can be migrated and we have plenty of work to do, try to migrate the thread to an idle kseq. - Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into consideration. - No longer use kseq_choose() to steal threads, it can lose it's last argument. - Create a new function runq_steal() which operates like runq_choose() but skips threads based on some criteria. Currently it will not steal PRI_ITHD threads. In the future this will be used for CPU binding. - Create a kseq_steal() that checks each run queue with runq_steal(), use kseq_steal() in the places where we used kseq_choose() to steal with before.
2003-10-31 11:16:04 +00:00
#ifdef SMP
if (ke->ke_cpu != PCPU_GET(cpuid)) {
kseq_notify(ke, ke->ke_cpu);
return;
}
#endif
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
/*
* This is for priority prop.
*/
- Use a better algorithm in sched_pctcpu_update() Contributed by: Thomaswuerfl@gmx.de - In sched_prio(), adjust the run queue for threads which may need to move to the current queue due to priority propagation . - In sched_switch(), fix style bug introduced when the KSE support went in. Columns are 80 chars wide, not 90. - In sched_switch(), Fix the comparison in the idle case and explicitly re-initialize the runq in the not propagated case. - Remove dead code in sched_clock(). - In sched_clock(), If we're an IDLE class td set NEEDRESCHED so that threads that have become runnable will get a chance to. - In sched_runnable(), if we're not the IDLETD, we should not consider curthread when examining the load. This mimics the 4BSD behavior of returning 0 when the only runnable thread is running. - In sched_userret(), remove the code for setting NEEDRESCHED entirely. This is not necessary and is not implemented in 4BSD. - Use the correct comparison in sched_add() when checking to see if an idle prio task has had it's priority temporarily elevated.
2003-10-27 06:47:05 +00:00
if (ke->ke_thread->td_priority < PRI_MIN_IDLE)
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
ke->ke_runq = kseq->ksq_curr;
else
ke->ke_runq = &kseq->ksq_idle;
ke->ke_slice = SCHED_SLICE_MIN;
break;
default:
panic("Unknown pri class.\n");
- Keep seperate statistics and run queues for different scheduling classes. - Treat each class specially in kseq_{choose,add,rem}. Let the rest of the code be less aware of scheduling classes. - Skip the interactivity calculation for non TIMESHARE ksegrps. - Move slice and runq selection into kseq_add(). Uninline it now that it's big.
2003-04-03 00:29:28 +00:00
break;
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
}
- Add static to local functions and data where it was missing. - Add an IPI based mechanism for migrating kses. This mechanism is broken down into several components. This is intended to reduce cache thrashing by eliminating most cases where one cpu touches another's run queues. - kseq_notify() appends a kse to a lockless singly linked list and conditionally sends an IPI to the target processor. Right now this is protected by sched_lock but at some point I'd like to get rid of the global lock. This is why I used something more complicated than a standard queue. - kseq_assign() processes our list of kses that have been assigned to us by other processors. This simply calls sched_add() for each item on the list after clearing the new KEF_ASSIGNED flag. This flag is used to indicate that we have been appeneded to the assigned queue but not added to the run queue yet. - In sched_add(), instead of adding a KSE to another processor's queue we use kse_notify() so that we don't touch their queue. Also in sched_add(), if KEF_ASSIGNED is already set return immediately. This can happen if a thread is removed and readded so that the priority is recorded properly. - In sched_rem() return immediately if KEF_ASSIGNED is set. All callers immediately readd simply to adjust priorites etc. - In sched_choose(), if we're running an IDLE task or the per cpu idle thread set our cpumask bit in 'kseq_idle' so that other processors may know that we are idle. Before this, make a single pass through the run queues of other processors so that we may find work more immediately if it is available. - In sched_runnable(), don't scan each processor's run queue, they will IPI us if they have work for us to do. - In sched_add(), if we're adding a thread that can be migrated and we have plenty of work to do, try to migrate the thread to an idle kseq. - Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into consideration. - No longer use kseq_choose() to steal threads, it can lose it's last argument. - Create a new function runq_steal() which operates like runq_choose() but skips threads based on some criteria. Currently it will not steal PRI_ITHD threads. In the future this will be used for CPU binding. - Create a kseq_steal() that checks each run queue with runq_steal(), use kseq_steal() in the places where we used kseq_choose() to steal with before.
2003-10-31 11:16:04 +00:00
#ifdef SMP
/*
* If there are any idle processors, give them our extra load.
*/
if (kseq_idle && class != PRI_ITHD &&
(kseq->ksq_loads[PRI_IDLE] + kseq->ksq_loads[PRI_TIMESHARE] +
kseq->ksq_loads[PRI_REALTIME]) >= kseq->ksq_cpus) {
int cpu;
/*
* Multiple cpus could find this bit simultaneously but the
* race shouldn't be terrible.
*/
cpu = ffs(kseq_idle);
if (cpu) {
cpu--;
atomic_clear_int(&kseq_idle, 1 << cpu);
ke->ke_cpu = cpu;
ke->ke_runq = NULL;
kseq_notify(ke, cpu);
return;
}
}
if (class == PRI_TIMESHARE || class == PRI_REALTIME)
atomic_clear_int(&kseq_idle, PCPU_GET(cpumask));
#endif
if (td->td_priority < curthread->td_priority)
curthread->td_flags |= TDF_NEEDRESCHED;
- Keep seperate statistics and run queues for different scheduling classes. - Treat each class specially in kseq_{choose,add,rem}. Let the rest of the code be less aware of scheduling classes. - Skip the interactivity calculation for non TIMESHARE ksegrps. - Move slice and runq selection into kseq_add(). Uninline it now that it's big.
2003-04-03 00:29:28 +00:00
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
ke->ke_ksegrp->kg_runq_kses++;
ke->ke_state = KES_ONRUNQ;
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
runq_add(ke->ke_runq, ke);
- Clean up some debug code left over from my earlier megacommit.
2003-04-12 07:28:36 +00:00
kseq_add(kseq, ke);
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
}
void
- Update the sched api. sched_{add,rem,clock,pctcpu} now all accept a td argument rather than a kse.
2003-10-16 08:39:15 +00:00
sched_rem(struct thread *td)
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
{
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
struct kseq *kseq;
- Update the sched api. sched_{add,rem,clock,pctcpu} now all accept a td argument rather than a kse.
2003-10-16 08:39:15 +00:00
struct kse *ke;
ke = td->td_kse;
- Add static to local functions and data where it was missing. - Add an IPI based mechanism for migrating kses. This mechanism is broken down into several components. This is intended to reduce cache thrashing by eliminating most cases where one cpu touches another's run queues. - kseq_notify() appends a kse to a lockless singly linked list and conditionally sends an IPI to the target processor. Right now this is protected by sched_lock but at some point I'd like to get rid of the global lock. This is why I used something more complicated than a standard queue. - kseq_assign() processes our list of kses that have been assigned to us by other processors. This simply calls sched_add() for each item on the list after clearing the new KEF_ASSIGNED flag. This flag is used to indicate that we have been appeneded to the assigned queue but not added to the run queue yet. - In sched_add(), instead of adding a KSE to another processor's queue we use kse_notify() so that we don't touch their queue. Also in sched_add(), if KEF_ASSIGNED is already set return immediately. This can happen if a thread is removed and readded so that the priority is recorded properly. - In sched_rem() return immediately if KEF_ASSIGNED is set. All callers immediately readd simply to adjust priorites etc. - In sched_choose(), if we're running an IDLE task or the per cpu idle thread set our cpumask bit in 'kseq_idle' so that other processors may know that we are idle. Before this, make a single pass through the run queues of other processors so that we may find work more immediately if it is available. - In sched_runnable(), don't scan each processor's run queue, they will IPI us if they have work for us to do. - In sched_add(), if we're adding a thread that can be migrated and we have plenty of work to do, try to migrate the thread to an idle kseq. - Simplify the logic in sched_prio() and take the KEF_ASSIGNED flag into consideration. - No longer use kseq_choose() to steal threads, it can lose it's last argument. - Create a new function runq_steal() which operates like runq_choose() but skips threads based on some criteria. Currently it will not steal PRI_ITHD threads. In the future this will be used for CPU binding. - Create a kseq_steal() that checks each run queue with runq_steal(), use kseq_steal() in the places where we used kseq_choose() to steal with before.
2003-10-31 11:16:04 +00:00
/*
* It is safe to just return here because sched_rem() is only ever
* used in places where we're immediately going to add the
* kse back on again. In that case it'll be added with the correct
* thread and priority when the caller drops the sched_lock.
*/
if (ke->ke_flags & KEF_ASSIGNED)
return;
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
mtx_assert(&sched_lock, MA_OWNED);
- Clean up some debug code left over from my earlier megacommit.
2003-04-12 07:28:36 +00:00
KASSERT((ke->ke_state == KES_ONRUNQ), ("KSE not on run queue"));
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
ke->ke_state = KES_THREAD;
ke->ke_ksegrp->kg_runq_kses--;
- Add a SYSCTL node for the ule scheduler. - Allow user adjustable min and max time slices (suggested by hiten). - Change the SLP_RUN_MAX to 100ms from 2 seconds so that we learn whether a process is interactive or not much more quickly. - Place a process on the current run queue if it is interactive or if it is running at an interrupt thread priority due to priority prop. - Use the 'current' timeshare queue for interrupt threads, realtime threads, and idle threads that are running at higher priority due to priority prop. This fixes problems where priorities would have been elevated but we would not check the timeshare run queue until other lower priority tasks were no longer runnable. - Keep an array of loads indexed by the priority class as well as a global load. - Keep an bucket of nice values with a count of the number of kses currently runnable with that nice value. - Keep track of the minimum nice value of any running thread. - Remove the unused short term sleep accounting. I was attempting to use this for load balancing but it didn't work out. - Define a kseq_print() for use with debugging. - Add KTR debugging at useful places so we can easily debug slice and priority assignment. - Decouple the runq assignment from the kseq assignment. kseq_add now keeps track of statistics. This is done so that the nice and load is still tracked for the currently running process. Previously if a niced process was added while a non nice process was running the niced process would still get a slice since it was not aware of the unnice process. - Make adjustments for the sched api changes.
2003-04-11 03:47:14 +00:00
kseq = KSEQ_CPU(ke->ke_cpu);
runq_remove(ke->ke_runq, ke);
kseq_rem(kseq, ke);
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
}
fixpt_t
- Update the sched api. sched_{add,rem,clock,pctcpu} now all accept a td argument rather than a kse.
2003-10-16 08:39:15 +00:00
sched_pctcpu(struct thread *td)
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
{
fixpt_t pctcpu;
- Update the sched api. sched_{add,rem,clock,pctcpu} now all accept a td argument rather than a kse.
2003-10-16 08:39:15 +00:00
struct kse *ke;
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
pctcpu = 0;
- Update the sched api. sched_{add,rem,clock,pctcpu} now all accept a td argument rather than a kse.
2003-10-16 08:39:15 +00:00
ke = td->td_kse;
- If a thread is not bound to a kse return 0 from sched_pctcpu(). Reported by: pawel.worach@nordea.com
2003-10-20 19:55:21 +00:00
if (ke == NULL)
return (0);
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
- When a new thread is added to a kseq the load is incremented prior to adding it to the nice tables. Therefore, in kseq_add_nice, we should keep in mind that the load will be 1 if we are the only thread, and not 0. - Assert that the sched lock is held in all the appropriate places. - Increase the scope of the sched lock in sched_pctcpu_update(). - Hold the sched lock in sched_runnable(). It is not held by the caller.
2003-06-08 00:47:33 +00:00
mtx_lock_spin(&sched_lock);
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
if (ke->ke_ticks) {
int rtick;
- Fix the maximum slice value. I accidentally checked in a value of '2' which meant no process would run for longer than 20ms. - Slightly redo the interactivity scorer. It follows the same algorithm but in a slightly more correct way. Previously values above half were incorrect. - Lower the interactivity threshold to 20. It seems that in testing non- interactive tasks are hardly ever near there and expensive interactive tasks can sometimes surpass it. This area needs more testing. - Remove an unnecessary KTR. - Fix a case where an idle thread that had an elevated priority due to priority prop. would be placed back on the idle queue. - Delay setting NEEDRESCHED until userret() for threads that haad their priority elevated while in kernel. This gives us the same context switch optimization as SCHED_4BSD. - Limit the child's slice to 1 in sched_fork_kse() so we detect its behavior more quickly. - Inhert some of the run/slp time from the child in sched_exit_ksegrp(). - Redo some of the priority comparisons so they are more clear. - Throttle the frequency of sched_pctcpu_update() so that rounding errors do not make it invalid.
2003-06-15 02:18:29 +00:00
/*
* Don't update more frequently than twice a second. Allowing
* this causes the cpu usage to decay away too quickly due to
* rounding errors.
*/
if (ke->ke_ltick < (ticks - (hz / 2)))
sched_pctcpu_update(ke);
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
/* How many rtick per second ? */
- Fix the maximum slice value. I accidentally checked in a value of '2' which meant no process would run for longer than 20ms. - Slightly redo the interactivity scorer. It follows the same algorithm but in a slightly more correct way. Previously values above half were incorrect. - Lower the interactivity threshold to 20. It seems that in testing non- interactive tasks are hardly ever near there and expensive interactive tasks can sometimes surpass it. This area needs more testing. - Remove an unnecessary KTR. - Fix a case where an idle thread that had an elevated priority due to priority prop. would be placed back on the idle queue. - Delay setting NEEDRESCHED until userret() for threads that haad their priority elevated while in kernel. This gives us the same context switch optimization as SCHED_4BSD. - Limit the child's slice to 1 in sched_fork_kse() so we detect its behavior more quickly. - Inhert some of the run/slp time from the child in sched_exit_ksegrp(). - Redo some of the priority comparisons so they are more clear. - Throttle the frequency of sched_pctcpu_update() so that rounding errors do not make it invalid.
2003-06-15 02:18:29 +00:00
rtick = min(ke->ke_ticks / SCHED_CPU_TIME, SCHED_CPU_TICKS);
Use hz if stathz is zero. Adopted from sched_4bsd.
2003-02-02 08:24:32 +00:00
pctcpu = (FSCALE * ((FSCALE * rtick)/realstathz)) >> FSHIFT;
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
}
ke->ke_proc->p_swtime = ke->ke_ltick - ke->ke_ftick;
Protect p_swtime with the sched_lock.
2003-04-22 19:48:25 +00:00
mtx_unlock_spin(&sched_lock);
- Add the ule scheduler. This is intended to be a general purpose process scheduler with many SMP benefits. It is still very experimental and should be used only in test environments.
2003-01-26 05:23:15 +00:00
return (pctcpu);
}
int
sched_sizeof_kse(void)
{
return (sizeof(struct kse) + sizeof(struct ke_sched));
}
int
sched_sizeof_ksegrp(void)
{
return (sizeof(struct ksegrp) + sizeof(struct kg_sched));
}
int
sched_sizeof_proc(void)
{
return (sizeof(struct proc));
}
int
sched_sizeof_thread(void)
{
return (sizeof(struct thread) + sizeof(struct td_sched));
}
Reference in New Issue Copy Permalink