- p_sflag was mostly protected by PROC_LOCK rather than the PROC_SLOCK or
previously the sched_lock. These bugs have existed for some time.
- Allow swapout to try each thread in a process individually and then
swapin the whole process if any of these fail. This allows us to move
most scheduler related swap flags into td_flags.
- Keep ki_sflag for backwards compat but change all in source tools to
use the new and more correct location of P_INMEM.
Reported by: pho
Reviewed by: attilio, kib
Approved by: re (kensmith)
on 2cpu machines by reducing it to 1 by default. This improves loaded
operation on 8cpu machines by increasing it to 3 where the extra idle
time is not as critical.
Approved by: re
tdq_group structure. Hyper-threaded cores won't really benefit from
seperate locks anyway.
- Seperate out the migration case from sched_switch to simplify the main
switch code. We only migrate here if called via sched_bind().
- When preempted place the preempted thread back in the same queue at
the head.
- Improve the cpu group and topology infrastructure.
Tested by: many on current@
Approved by: re
machines.
- Leave the long-term load balancer running by default once per second.
- Enable stealing load from the idle thread only when the remote processor
has more than two transferable tasks. Setting this to one further
improves buildworld. Setting it higher improves mysql.
- Remove the bogus pick_zero option. I had not intended to commit this.
- Entirely disallow migration for threads with SRQ_YIELDING set. This
balances out the extra migration allowed for with the load balancers.
It also makes pick_pri perform better as I had anticipated.
Tested by: Dmitry Morozovsky <marck@rinet.ru>
Approved by: re
properly. We have to temporarily unlock the TDQ lock so we can lock
the thread and add it to the run queue. This is used only for KSE.
- When we add a thread from the tdq_move() via sched_balance() we need to
ipi the target if it's sitting in the idle thread or it'll never run.
Reported by: Rene Landan
Approved by: re
been in development for over 6 months as SCHED_SMP.
- Implement one spin lock per thread-queue. Threads assigned to a
run-queue point to this lock via td_lock.
- Improve the facility for assigning threads to CPUs now that sched_lock
contention no longer dominates scheduling decisions on larger SMP
machines.
- Re-write idle time stealing in an attempt to make it less damaging to
general performance. This is still disabled by default. See
kern.sched.steal_idle.
- Call the long-term load balancer from a callout rather than sched_clock()
so there are no locks held. This is disabled by default. See
kern.sched.balance.
- Parameterize many scheduling decisions via sysctls. Try to document
these via sysctl descriptions.
- General structural and naming cleanups.
- Document each function with comments.
Tested by: current@ amd64, x86, UP, SMP.
Approved by: re
- In tdq_choose() only assert that a thread does not have too high a
priority (low value) for the queue we removed it from. This will catch
bugs in priority elevation. It's not a serious error for the thread
to have too low a priority as we don't change queues in this case as
an optimization.
Reported by: kris
- Move all scheduler locking into the schedulers utilizing a technique
similar to solaris's container locking.
- A per-process spinlock is now used to protect the queue of threads,
thread count, suspension count, p_sflags, and other process
related scheduling fields.
- The new thread lock is actually a pointer to a spinlock for the
container that the thread is currently owned by. The container may
be a turnstile, sleepqueue, or run queue.
- thread_lock() is now used to protect access to thread related scheduling
fields. thread_unlock() unlocks the lock and thread_set_lock()
implements the transition from one lock to another.
- A new "blocked_lock" is used in cases where it is not safe to hold the
actual thread's lock yet we must prevent access to the thread.
- sched_throw() and sched_fork_exit() are introduced to allow the
schedulers to fix-up locking at these points.
- Add some minor infrastructure for optionally exporting scheduler
statistics that were invaluable in solving performance problems with
this patch. Generally these statistics allow you to differentiate
between different causes of context switches.
Tested by: kris, current@
Tested on: i386, amd64, ULE, 4BSD, libthr, libkse, PREEMPTION, etc.
Discussed with: kris, attilio, kmacy, jhb, julian, bde (small parts each)
unsigned char. Weirdly, casting the 1 constant to u_char still produces
a signed integer result that is then used in the % computation. This
avoids that mess all together and causes a 0 pri to turn into 255 % 64
as we expect.
Reported by: kkenn (about 4 times, thanks)
a thread is an idle thread, just see if it has the IDLETD
flag set. That flag will probably move to the pflags word
as it's permenent and never chenges for the life of the
system so it doesn't need locking.
- only collect timestamps when a lock is contested - this reduces the overhead
of collecting profiles from 20x to 5x
- remove unused function from subr_lock.c
- generalize cnt_hold and cnt_lock statistics to be kept for all locks
- NOTE: rwlock profiling generates invalid statistics (and most likely always has)
someone familiar with that should review
- Fix these types in ULE as well. This fixes bugs in priority index
calculations in certain edge cases. (int)-1 % 64 != (uint)-1 % 64.
Reported by: kkenn using pho's stress2.
minimize IPIs and rescheduling when scheduling like tasks while keeping
latency low for important threads.
1) An idle thread is running.
2) The current thread is worse than realtime and the new thread is
better than realtime. Realtime to realtime doesn't preempt.
3) The new thread's priority is less than the threshold.
support sched_4bsd.
- Rename the KTR level for non schedgraph parsed events. They take event
space from things we'd like to graph.
- Reset our slice value after we sleep. The slice is simply there to
prevent starvation among equal priorities. A thread which had almost
exhausted it's slice and then slept doesn't need to be rescheduled a
tick after it wakes up.
- Set the maximum slice value to a more conservative 100ms now that it is
more accurately enforced.
negative. Use unsigned integers for sleep and run time so this doesn't
disturb sched_interact_score(). This should fix the invalid interactive
priority panics reported by several users.
- Define our own maybe_preempt() as sched_preempt(). We want to be able
to preempt idlethread in all cases.
- Define our idlethread to require preemption to exit.
- Get the cpu estimation tick from sched_tick() so we don't have to worry
about errors from a sampling interval that differs from the time
domain. This was the source of sched_priority prints/panics and
inaccurate pctcpu display in top.
a power saving mode otherwise.
- If the thread is already bound in sched_bind() unbind it before
re-binding it to a new cpu. I don't like these semantics but they are
expected by some code in the tree. Patch by jkoshy.
the ipi settings. If NEEDRESCHED is set and an ipi is later delivered
it will clear it rather than cause extra context switches. However, if
we miss setting it we can have terrible latency.
- In sched_bind() correctly implement bind. Also be slightly more
tolerant of code which calls bind multiple times. However, we don't
change binding if another call is made with a different cpu. This
does not presently work with hwpmc which I believe should be changed.
- Switch back to direct modification of remote CPU run queues. This added
a lot of complexity with questionable gain. It's easy enough to
reimplement if it's shown to help on huge machines.
- Re-implement the old tdq_transfer() call as tdq_pickidle(). Change
sched_add() so we have selectable cpu choosers and simplify the logic
a bit here.
- Implement tdq_pickpri() as the new default cpu chooser. This algorithm
is similar to Solaris in that it tries to always run the threads with
the best priorities. It is actually slightly more complex than
solaris's algorithm because we also tend to favor the local cpu over
other cpus which has a boost in latency but also potentially enables
cache sharing between the waking thread and the woken thread.
- Add a bunch of tunables that can be used to measure effects of different
load balancing strategies. Most of these will go away once the
algorithm is more definite.
- Add a new mechanism to steal threads from busy cpus when we idle. This
is enabled with kern.sched.steal_busy and kern.sched.busy_thresh. The
threshold is the required length of a tdq's run queue before another
cpu will be able to steal runnable threads. This prevents most queue
imbalances that contribute the long latencies.
of max() when computing the divisor in SCHED_TICK_PRI(). This prevents
cases where rounding down would allow the quotient to exceed
SCHED_PRI_RANGE.
- Garbage collect some unused flags and fields.
- Replace TDF_HOLD with sched_pin_td()/sched_unpin_td() since it simply
duplicated this functionality.
- Re-enable the rebalancer by default and fix the sysctl so it can be
modified.
marked idle, thus breaking cpu load balancing.
- Change sched_interact_update() to fix cases where the stored history
has expanded significantly rather than handling them in the callers. This
fixes a case where sched_priority() could compute a bad value.
- Add a sysctl to disable the global load balancer for experimentation.
setting ftick = ltick = ticks in schedinit().
- Update the priority when we are pulled off of the run queue and when we
are inserted onto the run queue so that it more accurately reflects our
present status. This is important for efficient priority propagation
functioning.
- Move the frequency test into sched_pctcpu_update() so we don't repeat it
each time we'd like to call it.
- Put some temporary work-around code in sched_priority() in case the tick
mechanism produces a bad priority. Eventually this should revert to an
assert again.
the most recently chosen index. This significantly improves nice
behavior. This allows a lower priority thread to run some multiple of
times before the higher priority thread makes it to the front of
the queue. A nice +20 cpu hog now only gets ~5% of the cpu when running
with a nice 0 cpu hog and about 1.5% with a nice -20 hog. A nice
difference of 1 makes a 4% difference in cpu usage between two hogs.
- Track a seperate insert and removal index. When the removal index is
empty it is updated to point at the current insert index.
- Don't remove and re-add a thread to the runq when it is being adjusted
down in priority.
- Pull some conditional code out of sched_tick(). It's looking a bit
large now.
- Remove the double queue mechanism for timeshare threads. It was slow
due to excess cache lines in play, caused suboptimal scheduling behavior
with niced and other non-interactive processes, complicated priority
lending, etc.
- Use a circular queue with a floating starting index for timeshare threads.
Enforces fairness by moving the insertion point closer to threads with
worse priorities over time.
- Give interactive timeshare threads real-time user-space priorities and
place them on the realtime/ithd queue.
- Select non-interactive timeshare thread priorities based on their cpu
utilization over the last 10 seconds combined with the nice value. This
gives us more sane priorities and behavior in a loaded system as
compared to the old method of using the interactivity score. The
interactive score quickly hit a ceiling if threads were non-interactive
and penalized new hog threads.
- Use one slice size for all threads. The slice is not currently
dynamically set to adjust scheduling behavior of different threads.
- Add some new sysctls for scheduling parameters.
Bug fixes/Clean up:
- Fix zeroing of td_sched after initialization in sched_fork_thread() caused
by recent ksegrp removal.
- Fix KSE interactivity issues related to frequent forking and exiting of
kse threads. We simply disable the penalty for thread creation and exit
for kse threads.
- Cleanup the cpu estimator by using tickincr here as well. Keep ticks and
ltick/ftick in the same frequency. Previously ticks were stathz and
others were hz.
- Lots of new and updated comments.
- Many many others.
Tested on: up x86/amd64, 8way amd64.
Make part of John Birrell's KSE patch permanent..
Specifically, remove:
Any reference of the ksegrp structure. This feature was
never fully utilised and made things overly complicated.
All code in the scheduler that tried to make threaded programs
fair to unthreaded programs. Libpthread processes will already
do this to some extent and libthr processes already disable it.
Also:
Since this makes such a big change to the scheduler(s), take the opportunity
to rename some structures and elements that had to be moved anyhow.
This makes the code a lot more readable.
The ULE scheduler compiles again but I have no idea if it works.
The 4bsd scheduler still reqires a little cleaning and some functions that now do
ALMOST nothing will go away, but I thought I'd do that as a separate commit.
Tested by David Xu, and Dan Eischen using libthr and libpthread.
yield() and sched_yield() syscalls. Every scheduler has its own way
to relinquish cpu, the ULE and CORE schedulers have two internal run-
queues, a timesharing thread which calls yield() syscall should be
moved to inactive queue.
I picked it up again. The scheduler is forked from ULE, but the
algorithm to detect an interactive process is almost completely
different with ULE, it comes from Linux paper "Understanding the
Linux 2.6.8.1 CPU Scheduler", although I still use same word
"score" as a priority boost in ULE scheduler.
Briefly, the scheduler has following characteristic:
1. Timesharing process's nice value is seriously respected,
timeslice and interaction detecting algorithm are based
on nice value.
2. per-cpu scheduling queue and load balancing.
3. O(1) scheduling.
4. Some cpu affinity code in wakeup path.
5. Support POSIX SCHED_FIFO and SCHED_RR.
Unlike scheduler 4BSD and ULE which using fuzzy RQ_PPQ, the scheduler
uses 256 priority queues. Unlike ULE which using pull and push, the
scheduelr uses pull method, the main reason is to let relative idle
cpu do the work, but current the whole scheduler is protected by the
big sched_lock, so the benefit is not visible, it really can be worse
than nothing because all other cpu are locked out when we are doing
balancing work, which the 4BSD scheduelr does not have this problem.
The scheduler does not support hyperthreading very well, in fact,
the scheduler does not make the difference between physical CPU and
logical CPU, this should be improved in feature. The scheduler has
priority inversion problem on MP machine, it is not good for
realtime scheduling, it can cause realtime process starving.
As a result, it seems the MySQL super-smack runs better on my
Pentium-D machine when using libthr, despite on UP or SMP kernel.
a thread holding critical resource, e.g mutex or other implicit
synchronous flags. Give thread which exceeds nice threshold a minimum
time slice.
PR: kern/86087
as this happens via thread_switchout(). I don't particularly like the
structure of the code here. We twice call out to thread code when
a thread is voluntarily switching. Once to thread_switchout() and once
to slot_fill(), while sched_4BSD does even more work which is redundant
to select another thread to use our remaining slice. This should be
simplified in the future, but for now I'm only going to fix the bug not
the bad design.
slot for us. Previously, we would take two slots on every preempt, and
setrunqueue() would fix it up for us in the non threaded case. The
threaded case was simply broken.
- Clean up flags, prototypes, comments.
schedulers a bit to ensure more correct handling of priorities and fewer
priority inversions:
- Add two functions to the sched(9) API to handle priority lending:
sched_lend_prio() and sched_unlend_prio(). The turnstile code uses these
functions to ask the scheduler to lend a thread a set priority and to
tell the scheduler when it thinks it is ok for a thread to stop borrowing
priority. The unlend case is slightly complex in that the turnstile code
tells the scheduler what the minimum priority of the thread needs to be
to satisfy the requirements of any other threads blocked on locks owned
by the thread in question. The scheduler then decides where the thread
can go back to normal mode (if it's normal priority is high enough to
satisfy the pending lock requests) or it it should continue to use the
priority specified to the sched_unlend_prio() call. This involves adding
a new per-thread flag TDF_BORROWING that replaces the ULE-only kse flag
for priority elevation.
- Schedulers now refuse to lower the priority of a thread that is currently
borrowing another therad's priority.
- If a scheduler changes the priority of a thread that is currently sitting
on a turnstile, it will call a new function turnstile_adjust() to inform
the turnstile code of the change. This function resorts the thread on
the priority list of the turnstile if needed, and if the thread ends up
at the head of the list (due to having the highest priority) and its
priority was raised, then it will propagate that new priority to the
owner of the lock it is blocked on.
Some additional fixes specific to the 4BSD scheduler include:
- Common code for updating the priority of a thread when the user priority
of its associated kse group has been consolidated in a new static
function resetpriority_thread(). One change to this function is that
it will now only adjust the priority of a thread if it already has a
time sharing priority, thus preserving any boosts from a tsleep() until
the thread returns to userland. Also, resetpriority() no longer calls
maybe_resched() on each thread in the group. Instead, the code calling
resetpriority() is responsible for calling resetpriority_thread() on
any threads that need to be updated.
- schedcpu() now uses resetpriority_thread() instead of just calling
sched_prio() directly after it updates a kse group's user priority.
- sched_clock() now uses resetpriority_thread() rather than writing
directly to td_priority.
- sched_nice() now updates all the priorities of the threads after the
group priority has been adjusted.
Discussed with: bde
Reviewed by: ups, jeffr
Tested on: 4bsd, ule
Tested on: i386, alpha, sparc64
- Remove the sched_add wrapper that used sched_add_internal() as a backend.
Its only purpose was to interpret one flag and turn it into an int. Do
the right thing and interpret the flag in sched_add() instead.
- Pass the flag argument to sched_add() to kseq_runq_add() so that we can
get the SRQ_PREEMPT optimization too.
- Add a KEF_INTERNAL flag. If KEF_INTERNAL is set we don't adjust the SLOT
counts, otherwise the slot counts are adjusted as soon as we enter
sched_add() or sched_rem() rather than when the thread is actually placed
on the run queue. This greatly simplifies the handling of slots.
- Remove the explicit prevention of migration for ithreads on non-x86
platforms. This was never shown to have any real benefit.
- Remove the unused class argument to KSE_CAN_MIGRATE().
- Add ktr points for thread migration events.
- Fix a long standing bug on platforms which don't initialize the cpu
topology. The ksg_maxid variable was never correctly set on these
platforms which caused the long term load balancer to never inspect
more than the first group or processor.
- Fix another bug which prevented the long term load balancer from working
properly. If stathz != hz we can't expect sched_clock() to be called
on the exact tick count that we're anticipating.
- Rearrange sched_switch() a bit to reduce indentation levels.
nice of 0. Doing so can cause an infinite loop because they should be
running, but a nice -20 process could prevent them from doing so.
- Add a new flag KEF_PRIOELEV to flag a thread that has had its priority
elevated due to priority propagation. If a thread has had its priority
elevated, we assume that it must go on the current queue and it must
get a slice.
- In sched_userret() if our priority was elevated and we shouldn't have
a timeslice, yield here until we should.
Found/Tested by: glebius
outside of the nice threshold due to a recently awoken thread with a
lower nice value. This further reduces the amount of time a positively
niced thread gets while running in conjunction with a workload that has
many short sleeps (ie buildworld).
check for TD_ON_RUNQ() no longer means the thread is really on a run-
queue. I suspect this state should be re-evaluated as it must mean
something else now. This fixes ULE+KSE+PREEMPTION on UP x86.
fully initialed when the pmap layer tries to call sched_pini() early in the
boot and results in an quick panic. Use ke_pinned instead as was originally
done with Tor's patch.
Approved by: julian
scheduler specific extension to it. Put it in the extension as
the implimentation details of how the pinning is done needn't be visible
outside the scheduler.
Submitted by: tegge (of course!) (with changes)
MFC after: 3 days
but with slightly cleaned up interfaces.
The KSE structure has become the same as the "per thread scheduler
private data" structure. In order to not make the diffs too great
one is #defined as the other at this time.
The KSE (or td_sched) structure is now allocated per thread and has no
allocation code of its own.
Concurrency for a KSEGRP is now kept track of via a simple pair of counters
rather than using KSE structures as tokens.
Since the KSE structure is different in each scheduler, kern_switch.c
is now included at the end of each scheduler. Nothing outside the
scheduler knows the contents of the KSE (aka td_sched) structure.
The fields in the ksegrp structure that are to do with the scheduler's
queueing mechanisms are now moved to the kg_sched structure.
(per ksegrp scheduler private data structure). In other words how the
scheduler queues and keeps track of threads is no-one's business except
the scheduler's. This should allow people to write experimental
schedulers with completely different internal structuring.
A scheduler call sched_set_concurrency(kg, N) has been added that
notifies teh scheduler that no more than N threads from that ksegrp
should be allowed to be on concurrently scheduled. This is also
used to enforce 'fainess' at this time so that a ksegrp with
10000 threads can not swamp a the run queue and force out a process
with 1 thread, since the current code will not set the concurrency above
NCPU, and both schedulers will not allow more than that many
onto the system run queue at a time. Each scheduler should eventualy develop
their own methods to do this now that they are effectively separated.
Rejig libthr's kernel interface to follow the same code paths as
linkse for scope system threads. This has slightly hurt libthr's performance
but I will work to recover as much of it as I can.
Thread exit code has been cleaned up greatly.
exit and exec code now transitions a process back to
'standard non-threaded mode' before taking the next step.
Reviewed by: scottl, peter
MFC after: 1 week
FULL_PREEMPTION is defined. Add a runtime warning to ULE if PREEMPTION is
enabled (code inspired by the PREEMPTION warning in kern_switch.c). This
is a possible MT5 candidate.
preemption and/or the rev 1.79 kern_switch.c change that was backed out.
The thread was being assigned to a runq without adding in the load, which
would cause the counter to hit -1.
migration. Use this in sched_prio() and sched_switch() to stop us from
migrating threads that are in short term sleeps or are runnable. These
extra migrations were added in the patches to support KSE.
- Only set NEEDRESCHED if the thread we're adding in sched_add() is a
lower priority and is being placed on the current queue.
- Fix some minor whitespace problems.
contributed to the transferable load count. This prevents any potential
problems with sched_pin() being used around calls to setrunqueue().
- Change the sched_add() load balancing algorithm to try to migrate on
wakeup. This attempts to place threads that communicate with each other
on the same CPU.
- Don't clear the idle counts in kseq_transfer(), let the cpus do that when
they call sched_add() from kseq_assign().
- Correct a few out of date comments.
- Make sure the ke_cpu field is correct when we preempt.
- Call kseq_assign() from sched_clock() to catch any assignments that were
done without IPI. Presently all assignments are done with an IPI, but I'm
trying a patch that limits that.
- Don't migrate a thread if it is still runnable in sched_add(). Previously,
this could only happen for KSE threads, but due to changes to
sched_switch() all threads went through this path.
- Remove some code that was added with preemption but is not necessary.
specify "us" as the thread not the process/ksegrp/kse.
You can always find the others from the thread but the converse is not true.
Theorotically this would lead to runtime being allocated to the wrong
entity in some cases though it is not clear how often this actually happenned.
(would only affect threaded processes and would probably be pretty benign,
but it WAS a bug..)
Reviewed by: peter
since they are only accessed by curthread and thus do not need any
locking.
- Move pr_addr and pr_ticks out of struct uprof (which is per-process)
and directly into struct thread as td_profil_addr and td_profil_ticks
as these variables are really per-thread. (They are used to defer an
addupc_intr() that was too "hard" until ast()).
takes an argument to specify if it should preempt or not. Don't preempt
when sched_add_internal() is called from kseq_idled() or kseq_assign()
as in those cases we are about to call mi_switch() anyways. Also, doing
so during the first context switch on an AP leads to a NULL pointer deref
because curthread is NULL.
- Reenable preemption for ULE.
Submitted by: Taku YAMAMOTO taku at tackymt.homeip.net
hangs due to recent preemption changes. This change appears to remove
the panic that I was running into, but at the cost of increasing
ithread scheduling latency, and as such is a temporary band-aid until
jhb has a chance to resolve the ule<->preemption interaction that is
the source of the problem. If it doesn't fix the problem for others--
sorry!
introduced a KSE_CAN_MIGRATE() invocation with one argument
missing (class). Either this is a genuine forget or it crept
in from JHB's repo where he may have modified it. If it's
the latter then it may require more attention. For now fix
the make depend.
than as one-off hacks in various other parts of the kernel:
- Add a function maybe_preempt() that is called from sched_add() to
determine if a thread about to be added to a run queue should be
preempted to directly. If it is not safe to preempt or if the new
thread does not have a high enough priority, then the function returns
false and sched_add() adds the thread to the run queue. If the thread
should be preempted to but the current thread is in a nested critical
section, then the flag TDF_OWEPREEMPT is set and the thread is added
to the run queue. Otherwise, mi_switch() is called immediately and the
thread is never added to the run queue since it is switch to directly.
When exiting an outermost critical section, if TDF_OWEPREEMPT is set,
then clear it and call mi_switch() to perform the deferred preemption.
- Remove explicit preemption from ithread_schedule() as calling
setrunqueue() now does all the correct work. This also removes the
do_switch argument from ithread_schedule().
- Do not use the manual preemption code in mtx_unlock if the architecture
supports native preemption.
- Don't call mi_switch() in a loop during shutdown to give ithreads a
chance to run if the architecture supports native preemption since
the ithreads will just preempt DELAY().
- Don't call mi_switch() from the page zeroing idle thread for
architectures that support native preemption as it is unnecessary.
- Native preemption is enabled on the same archs that supported ithread
preemption, namely alpha, i386, and amd64.
This change should largely be a NOP for the default case as committed
except that we will do fewer context switches in a few cases and will
avoid the run queues completely when preempting.
Approved by: scottl (with his re@ hat)
switch to. If a non-NULL thread pointer is passed in, then the CPU will
switch to that thread directly rather than calling choosethread() to pick
a thread to choose to.
- Make sched_switch() aware of idle threads and know to do
TD_SET_CAN_RUN() instead of sticking them on the run queue rather than
requiring all callers of mi_switch() to know to do this if they can be
called from an idlethread.
- Move constants for arguments to mi_switch() and thread_single() out of
the middle of the function prototypes and up above into their own
section.
sched_clock() rather than using callouts. This means we no longer have to
take the load of the callout thread into consideration while balancing and
should make the balancing decisions simpler and more accurate.
Tested on: x86/UP, amd64/SMP
