Re: drm(4), multi-threaded task queues and barriers
On Fri, Jul 05, 2019 at 12:20:20PM +0200, Mark Kettenis wrote:
> > Date: Fri, 5 Jul 2019 14:20:40 +1000
> > From: Jonathan Gray
> >
> > On Wed, Jun 12, 2019 at 09:15:36AM +0200, Mark Kettenis wrote:
> > > > Date: Wed, 12 Jun 2019 17:04:10 +1000
> > > > From: Jonathan Gray
> > > >
> > > > On Tue, Jun 11, 2019 at 09:10:46PM +0200, Mark Kettenis wrote:
> > > > > The drm(4) codebase really needs multi-threaded task queues since the
> > > > > code has taks that wait for the completion of other tasks that are
> > > > > submitted to the same task queue. Thank you Linux...
> > > > >
> > > > > Unfortunately the code also needs to wait for the completion of
> > > > > submitted tasks from other threads. This implemented using
> > > > > task_barrier(9). But unfortunately our current implementation only
> > > > > works for single-threaded task queues.
> > > > >
> > > > > The diff below fixes this by marking the barrier task with a flag and
> > > > > making sure that all threads of the task queue are syncchronized.
> > > > > This achived through a TASK_BARRIER flag that simply blocks the
> > > > > threads until the last unblocked thread sees the flag and executes the
> > > > > task.
> > > > >
> > > > > The diff also starts 4 threads for each workqueue that gets created by
> > > > > the drm(4) layer. The number 4 is a bit arbitrary but it is the
> > > > > number of threads that Linux creates per CPU for a so-called "unbound"
> > > > > workqueue which hopefully is enough to always make progress.
> > > >
> > > > Ignoring taskletq and systq use in burner_task/switchtask across
> > > > drivers and the local only backlight_schedule_update_status(), was it
> > > > intentional to exclude:
> > > >
> > > > workqueue.h: alloc_workqueue()
> > > > struct taskq *tq = taskq_create(name, 1, IPL_TTY, 0);
> > > > workqueue.h: alloc_ordered_workqueue()
> > > > struct taskq *tq = taskq_create(name, 1, IPL_TTY, 0);
> > > > workqueue.h:
> > > > #define system_power_efficient_wq ((struct workqueue_struct
> > > > *)systq)
> > >
> > > Not entirely. But I don't want to spawn too many threads... If this
> > > diff seems to work, we might want to tweak the numbers a bit to see
> > > what works best. Whatever we do, the ordered workqueues need to stay
> > > single-threaded.
> >
> > While the non-drm parts of this went in the drm_linux.c and intel_hotplug.c
> > parts have not. Was there a particular reason these changes didn't go in?
>
> I wanted to check whether we could get away with using less threads
> for the queues but I didn't get around to doing it. But I guess I
> could just commit this and do some additional tuning later.
>
> You're ok with the change going in?
Yes, sorry if that wasn't clear. ok jsg@
>
> > > > > Please test. If you experience a "hang" with this diff, please try to
> > > > > log in to the machine remotely over ssh and send me and jsg@ the
> > > > > output of "ps -AHwlk".
> > > > >
> > > > > Thanks,
> > > > >
> > > > > Mark
> > > > >
> > > > >
> > > > > Index: dev/pci/drm/drm_linux.c
> > > > > ===
> > > > > RCS file: /cvs/src/sys/dev/pci/drm/drm_linux.c,v
> > > > > retrieving revision 1.38
> > > > > diff -u -p -r1.38 drm_linux.c
> > > > > --- dev/pci/drm/drm_linux.c 9 Jun 2019 12:58:30 - 1.38
> > > > > +++ dev/pci/drm/drm_linux.c 11 Jun 2019 18:54:38 -
> > > > > @@ -1399,15 +1399,15 @@ drm_linux_init(void)
> > > > > {
> > > > > if (system_wq == NULL) {
> > > > > system_wq = (struct workqueue_struct *)
> > > > > - taskq_create("drmwq", 1, IPL_HIGH, 0);
> > > > > + taskq_create("drmwq", 4, IPL_HIGH, 0);
> > > > > }
> > > > > if (system_unbound_wq == NULL) {
> > > > > system_unbound_wq = (struct workqueue_struct *)
> > > > > - taskq_create("drmubwq", 1, IPL_HIGH, 0);
> > > > > + taskq_create("drmubwq", 4, IPL_HIGH, 0);
> > > > > }
> > > > > if (system_long_wq == NULL) {
> > > > > system_long_wq = (struct workqueue_struct *)
> > > > > - taskq_create("drmlwq", 1, IPL_HIGH, 0);
> > > > > + taskq_create("drmlwq", 4, IPL_HIGH, 0);
> > > > > }
> > > > >
> > > > > if (taskletq == NULL)
> > > > > Index: dev/pci/drm/i915/intel_hotplug.c
> > > > > ===
> > > > > RCS file: /cvs/src/sys/dev/pci/drm/i915/intel_hotplug.c,v
> > > > > retrieving revision 1.2
> > > > > diff -u -p -r1.2 intel_hotplug.c
> > > > > --- dev/pci/drm/i915/intel_hotplug.c 14 Apr 2019 10:14:52 -
> > > > > 1.2
> > > > > +++ dev/pci/drm/i915/intel_hotplug.c 11 Jun 2019 18:54:38 -
> > > > > @@ -619,7 +619,6 @@ void intel_hpd_init_work(struct drm_i915
> > > > > INIT_WORK(_priv->hotplug.hotplug_work,
> > > > > i915_hotplug_work_func);
> > > > >
Re: drm(4), multi-threaded task queues and barriers
> Date: Fri, 5 Jul 2019 14:20:40 +1000
> From: Jonathan Gray
>
> On Wed, Jun 12, 2019 at 09:15:36AM +0200, Mark Kettenis wrote:
> > > Date: Wed, 12 Jun 2019 17:04:10 +1000
> > > From: Jonathan Gray
> > >
> > > On Tue, Jun 11, 2019 at 09:10:46PM +0200, Mark Kettenis wrote:
> > > > The drm(4) codebase really needs multi-threaded task queues since the
> > > > code has taks that wait for the completion of other tasks that are
> > > > submitted to the same task queue. Thank you Linux...
> > > >
> > > > Unfortunately the code also needs to wait for the completion of
> > > > submitted tasks from other threads. This implemented using
> > > > task_barrier(9). But unfortunately our current implementation only
> > > > works for single-threaded task queues.
> > > >
> > > > The diff below fixes this by marking the barrier task with a flag and
> > > > making sure that all threads of the task queue are syncchronized.
> > > > This achived through a TASK_BARRIER flag that simply blocks the
> > > > threads until the last unblocked thread sees the flag and executes the
> > > > task.
> > > >
> > > > The diff also starts 4 threads for each workqueue that gets created by
> > > > the drm(4) layer. The number 4 is a bit arbitrary but it is the
> > > > number of threads that Linux creates per CPU for a so-called "unbound"
> > > > workqueue which hopefully is enough to always make progress.
> > >
> > > Ignoring taskletq and systq use in burner_task/switchtask across
> > > drivers and the local only backlight_schedule_update_status(), was it
> > > intentional to exclude:
> > >
> > > workqueue.h: alloc_workqueue()
> > > struct taskq *tq = taskq_create(name, 1, IPL_TTY, 0);
> > > workqueue.h: alloc_ordered_workqueue()
> > > struct taskq *tq = taskq_create(name, 1, IPL_TTY, 0);
> > > workqueue.h:
> > > #define system_power_efficient_wq ((struct workqueue_struct *)systq)
> >
> > Not entirely. But I don't want to spawn too many threads... If this
> > diff seems to work, we might want to tweak the numbers a bit to see
> > what works best. Whatever we do, the ordered workqueues need to stay
> > single-threaded.
>
> While the non-drm parts of this went in the drm_linux.c and intel_hotplug.c
> parts have not. Was there a particular reason these changes didn't go in?
I wanted to check whether we could get away with using less threads
for the queues but I didn't get around to doing it. But I guess I
could just commit this and do some additional tuning later.
You're ok with the change going in?
> > > > Please test. If you experience a "hang" with this diff, please try to
> > > > log in to the machine remotely over ssh and send me and jsg@ the
> > > > output of "ps -AHwlk".
> > > >
> > > > Thanks,
> > > >
> > > > Mark
> > > >
> > > >
> > > > Index: dev/pci/drm/drm_linux.c
> > > > ===
> > > > RCS file: /cvs/src/sys/dev/pci/drm/drm_linux.c,v
> > > > retrieving revision 1.38
> > > > diff -u -p -r1.38 drm_linux.c
> > > > --- dev/pci/drm/drm_linux.c 9 Jun 2019 12:58:30 - 1.38
> > > > +++ dev/pci/drm/drm_linux.c 11 Jun 2019 18:54:38 -
> > > > @@ -1399,15 +1399,15 @@ drm_linux_init(void)
> > > > {
> > > > if (system_wq == NULL) {
> > > > system_wq = (struct workqueue_struct *)
> > > > - taskq_create("drmwq", 1, IPL_HIGH, 0);
> > > > + taskq_create("drmwq", 4, IPL_HIGH, 0);
> > > > }
> > > > if (system_unbound_wq == NULL) {
> > > > system_unbound_wq = (struct workqueue_struct *)
> > > > - taskq_create("drmubwq", 1, IPL_HIGH, 0);
> > > > + taskq_create("drmubwq", 4, IPL_HIGH, 0);
> > > > }
> > > > if (system_long_wq == NULL) {
> > > > system_long_wq = (struct workqueue_struct *)
> > > > - taskq_create("drmlwq", 1, IPL_HIGH, 0);
> > > > + taskq_create("drmlwq", 4, IPL_HIGH, 0);
> > > > }
> > > >
> > > > if (taskletq == NULL)
> > > > Index: dev/pci/drm/i915/intel_hotplug.c
> > > > ===
> > > > RCS file: /cvs/src/sys/dev/pci/drm/i915/intel_hotplug.c,v
> > > > retrieving revision 1.2
> > > > diff -u -p -r1.2 intel_hotplug.c
> > > > --- dev/pci/drm/i915/intel_hotplug.c14 Apr 2019 10:14:52 -
> > > > 1.2
> > > > +++ dev/pci/drm/i915/intel_hotplug.c11 Jun 2019 18:54:38 -
> > > > @@ -619,7 +619,6 @@ void intel_hpd_init_work(struct drm_i915
> > > > INIT_WORK(_priv->hotplug.hotplug_work,
> > > > i915_hotplug_work_func);
> > > > INIT_WORK(_priv->hotplug.dig_port_work,
> > > > i915_digport_work_func);
> > > > INIT_WORK(_priv->hotplug.poll_init_work,
> > > > i915_hpd_poll_init_work);
> > > > - dev_priv->hotplug.poll_init_work.tq = systq;
> > > > INIT_DELAYED_WORK(_priv->hotplug.reenable_work,
> > > >
Re: drm(4), multi-threaded task queues and barriers
On Wed, Jun 12, 2019 at 09:15:36AM +0200, Mark Kettenis wrote:
> > Date: Wed, 12 Jun 2019 17:04:10 +1000
> > From: Jonathan Gray
> >
> > On Tue, Jun 11, 2019 at 09:10:46PM +0200, Mark Kettenis wrote:
> > > The drm(4) codebase really needs multi-threaded task queues since the
> > > code has taks that wait for the completion of other tasks that are
> > > submitted to the same task queue. Thank you Linux...
> > >
> > > Unfortunately the code also needs to wait for the completion of
> > > submitted tasks from other threads. This implemented using
> > > task_barrier(9). But unfortunately our current implementation only
> > > works for single-threaded task queues.
> > >
> > > The diff below fixes this by marking the barrier task with a flag and
> > > making sure that all threads of the task queue are syncchronized.
> > > This achived through a TASK_BARRIER flag that simply blocks the
> > > threads until the last unblocked thread sees the flag and executes the
> > > task.
> > >
> > > The diff also starts 4 threads for each workqueue that gets created by
> > > the drm(4) layer. The number 4 is a bit arbitrary but it is the
> > > number of threads that Linux creates per CPU for a so-called "unbound"
> > > workqueue which hopefully is enough to always make progress.
> >
> > Ignoring taskletq and systq use in burner_task/switchtask across
> > drivers and the local only backlight_schedule_update_status(), was it
> > intentional to exclude:
> >
> > workqueue.h: alloc_workqueue()
> > struct taskq *tq = taskq_create(name, 1, IPL_TTY, 0);
> > workqueue.h: alloc_ordered_workqueue()
> > struct taskq *tq = taskq_create(name, 1, IPL_TTY, 0);
> > workqueue.h:
> > #define system_power_efficient_wq ((struct workqueue_struct *)systq)
>
> Not entirely. But I don't want to spawn too many threads... If this
> diff seems to work, we might want to tweak the numbers a bit to see
> what works best. Whatever we do, the ordered workqueues need to stay
> single-threaded.
While the non-drm parts of this went in the drm_linux.c and intel_hotplug.c
parts have not. Was there a particular reason these changes didn't go in?
>
> > > Please test. If you experience a "hang" with this diff, please try to
> > > log in to the machine remotely over ssh and send me and jsg@ the
> > > output of "ps -AHwlk".
> > >
> > > Thanks,
> > >
> > > Mark
> > >
> > >
> > > Index: dev/pci/drm/drm_linux.c
> > > ===
> > > RCS file: /cvs/src/sys/dev/pci/drm/drm_linux.c,v
> > > retrieving revision 1.38
> > > diff -u -p -r1.38 drm_linux.c
> > > --- dev/pci/drm/drm_linux.c 9 Jun 2019 12:58:30 - 1.38
> > > +++ dev/pci/drm/drm_linux.c 11 Jun 2019 18:54:38 -
> > > @@ -1399,15 +1399,15 @@ drm_linux_init(void)
> > > {
> > > if (system_wq == NULL) {
> > > system_wq = (struct workqueue_struct *)
> > > - taskq_create("drmwq", 1, IPL_HIGH, 0);
> > > + taskq_create("drmwq", 4, IPL_HIGH, 0);
> > > }
> > > if (system_unbound_wq == NULL) {
> > > system_unbound_wq = (struct workqueue_struct *)
> > > - taskq_create("drmubwq", 1, IPL_HIGH, 0);
> > > + taskq_create("drmubwq", 4, IPL_HIGH, 0);
> > > }
> > > if (system_long_wq == NULL) {
> > > system_long_wq = (struct workqueue_struct *)
> > > - taskq_create("drmlwq", 1, IPL_HIGH, 0);
> > > + taskq_create("drmlwq", 4, IPL_HIGH, 0);
> > > }
> > >
> > > if (taskletq == NULL)
> > > Index: dev/pci/drm/i915/intel_hotplug.c
> > > ===
> > > RCS file: /cvs/src/sys/dev/pci/drm/i915/intel_hotplug.c,v
> > > retrieving revision 1.2
> > > diff -u -p -r1.2 intel_hotplug.c
> > > --- dev/pci/drm/i915/intel_hotplug.c 14 Apr 2019 10:14:52 -
> > > 1.2
> > > +++ dev/pci/drm/i915/intel_hotplug.c 11 Jun 2019 18:54:38 -
> > > @@ -619,7 +619,6 @@ void intel_hpd_init_work(struct drm_i915
> > > INIT_WORK(_priv->hotplug.hotplug_work, i915_hotplug_work_func);
> > > INIT_WORK(_priv->hotplug.dig_port_work, i915_digport_work_func);
> > > INIT_WORK(_priv->hotplug.poll_init_work, i915_hpd_poll_init_work);
> > > - dev_priv->hotplug.poll_init_work.tq = systq;
> > > INIT_DELAYED_WORK(_priv->hotplug.reenable_work,
> > > intel_hpd_irq_storm_reenable_work);
> > > }
> > > Index: kern/kern_task.c
> > > ===
> > > RCS file: /cvs/src/sys/kern/kern_task.c,v
> > > retrieving revision 1.25
> > > diff -u -p -r1.25 kern_task.c
> > > --- kern/kern_task.c 28 Apr 2019 04:20:40 - 1.25
> > > +++ kern/kern_task.c 11 Jun 2019 18:54:39 -
> > > @@ -43,6 +43,7 @@ struct taskq {
> > > TQ_S_DESTROYED
> > > }tq_state;
> > > unsigned int tq_running;
> > > + unsigned int tq_barrier;
> > > unsigned
Re: drm(4), multi-threaded task queues and barriers
Mark Kettenis on Thu, Jun 13 2019: > So here is a diff that fixes the problem as far as I can tell. > Jonathan, Sven, can you give this a go? Works for me. > + tq->tq_waiting++; > msleep(tq, >tq_mtx, PWAIT, "bored", 0); > + tq->tq_waiting--; Heh, didn't spot this either. State machines are hard. :) Regards.
Re: drm(4), multi-threaded task queues and barriers
On Thu, Jun 13, 2019 at 11:27:57PM +0200, Mark Kettenis wrote:
> > Date: Wed, 12 Jun 2019 23:57:27 +0200 (CEST)
> > From: Mark Kettenis
> >
> > > From: "Sven M. Hallberg"
> > > Date: Wed, 12 Jun 2019 23:18:24 +0200
> > >
> > > Mark Kettenis on Tue, Jun 11 2019:
> > > > The drm(4) codebase really needs multi-threaded task queues [...]
> > > >
> > > > The diff also starts 4 threads for each workqueue that gets created by
> > > > the drm(4) layer. The number 4 is a bit arbitrary but it is the
> > > > number of threads that Linux creates per CPU for a so-called "unbound"
> > > > workqueue which hopefully is enough to always make progress.
> > > >
> > > > Please test.
> > >
> > > Looks good and appears to work fine with two displays (one internal, one
> > > external). Will test with three at work tomorrow.
> > >
> > >
> > > > - dev_priv->hotplug.poll_init_work.tq = systq;
> > >
> > > Intentional?
> >
> > Yes. It removes a local modification that should no longer be necessary.
> >
> > Unfortunately the diff doesn't work with amdgpu. Some more thinking
> > needed...
>
> So here is a diff that fixes the problem as far as I can tell.
> Jonathan, Sven, can you give this a go?
This revised diff does not introduce any new problems with amdgpu as
far as I can tell. Xorg starts/runs, piglit starts and later gets stuck
in dmafence/scheduler related wait channels (as it did before).
>
> Index: dev/pci/drm/drm_linux.c
> ===
> RCS file: /cvs/src/sys/dev/pci/drm/drm_linux.c,v
> retrieving revision 1.38
> diff -u -p -r1.38 drm_linux.c
> --- dev/pci/drm/drm_linux.c 9 Jun 2019 12:58:30 - 1.38
> +++ dev/pci/drm/drm_linux.c 13 Jun 2019 20:53:23 -
> @@ -1399,15 +1399,15 @@ drm_linux_init(void)
> {
> if (system_wq == NULL) {
> system_wq = (struct workqueue_struct *)
> - taskq_create("drmwq", 1, IPL_HIGH, 0);
> + taskq_create("drmwq", 4, IPL_HIGH, 0);
> }
> if (system_unbound_wq == NULL) {
> system_unbound_wq = (struct workqueue_struct *)
> - taskq_create("drmubwq", 1, IPL_HIGH, 0);
> + taskq_create("drmubwq", 4, IPL_HIGH, 0);
> }
> if (system_long_wq == NULL) {
> system_long_wq = (struct workqueue_struct *)
> - taskq_create("drmlwq", 1, IPL_HIGH, 0);
> + taskq_create("drmlwq", 4, IPL_HIGH, 0);
> }
>
> if (taskletq == NULL)
> Index: dev/pci/drm/i915/intel_hotplug.c
> ===
> RCS file: /cvs/src/sys/dev/pci/drm/i915/intel_hotplug.c,v
> retrieving revision 1.2
> diff -u -p -r1.2 intel_hotplug.c
> --- dev/pci/drm/i915/intel_hotplug.c 14 Apr 2019 10:14:52 - 1.2
> +++ dev/pci/drm/i915/intel_hotplug.c 13 Jun 2019 20:53:23 -
> @@ -619,7 +619,6 @@ void intel_hpd_init_work(struct drm_i915
> INIT_WORK(_priv->hotplug.hotplug_work, i915_hotplug_work_func);
> INIT_WORK(_priv->hotplug.dig_port_work, i915_digport_work_func);
> INIT_WORK(_priv->hotplug.poll_init_work, i915_hpd_poll_init_work);
> - dev_priv->hotplug.poll_init_work.tq = systq;
> INIT_DELAYED_WORK(_priv->hotplug.reenable_work,
> intel_hpd_irq_storm_reenable_work);
> }
> Index: kern/kern_task.c
> ===
> RCS file: /cvs/src/sys/kern/kern_task.c,v
> retrieving revision 1.25
> diff -u -p -r1.25 kern_task.c
> --- kern/kern_task.c 28 Apr 2019 04:20:40 - 1.25
> +++ kern/kern_task.c 13 Jun 2019 20:53:23 -
> @@ -43,6 +43,7 @@ struct taskq {
> TQ_S_DESTROYED
> }tq_state;
> unsigned int tq_running;
> + unsigned int tq_waiting;
> unsigned int tq_nthreads;
> unsigned int tq_flags;
> const char *tq_name;
> @@ -59,6 +60,7 @@ static const char taskq_sys_name[] = "sy
> struct taskq taskq_sys = {
> TQ_S_CREATED,
> 0,
> + 0,
> 1,
> 0,
> taskq_sys_name,
> @@ -77,6 +79,7 @@ static const char taskq_sys_mp_name[] =
> struct taskq taskq_sys_mp = {
> TQ_S_CREATED,
> 0,
> + 0,
> 1,
> TASKQ_MPSAFE,
> taskq_sys_mp_name,
> @@ -122,6 +125,7 @@ taskq_create(const char *name, unsigned
>
> tq->tq_state = TQ_S_CREATED;
> tq->tq_running = 0;
> + tq->tq_waiting = 0;
> tq->tq_nthreads = nthreads;
> tq->tq_name = name;
> tq->tq_flags = flags;
> @@ -223,6 +227,7 @@ taskq_barrier(struct taskq *tq)
>
> WITNESS_CHECKORDER(>tq_lock_object, LOP_NEWORDER, NULL);
>
> + SET(t.t_flags, TASK_BARRIER);
> task_add(tq, );
> cond_wait(, "tqbar");
> }
> @@ -238,6 +243,7 @@ taskq_del_barrier(struct taskq *tq, stru
> if (task_del(tq, del))
> return;
>
> +
Re: drm(4), multi-threaded task queues and barriers
> Date: Wed, 12 Jun 2019 23:57:27 +0200 (CEST)
> From: Mark Kettenis
>
> > From: "Sven M. Hallberg"
> > Date: Wed, 12 Jun 2019 23:18:24 +0200
> >
> > Mark Kettenis on Tue, Jun 11 2019:
> > > The drm(4) codebase really needs multi-threaded task queues [...]
> > >
> > > The diff also starts 4 threads for each workqueue that gets created by
> > > the drm(4) layer. The number 4 is a bit arbitrary but it is the
> > > number of threads that Linux creates per CPU for a so-called "unbound"
> > > workqueue which hopefully is enough to always make progress.
> > >
> > > Please test.
> >
> > Looks good and appears to work fine with two displays (one internal, one
> > external). Will test with three at work tomorrow.
> >
> >
> > > - dev_priv->hotplug.poll_init_work.tq = systq;
> >
> > Intentional?
>
> Yes. It removes a local modification that should no longer be necessary.
>
> Unfortunately the diff doesn't work with amdgpu. Some more thinking
> needed...
So here is a diff that fixes the problem as far as I can tell.
Jonathan, Sven, can you give this a go?
Index: dev/pci/drm/drm_linux.c
===
RCS file: /cvs/src/sys/dev/pci/drm/drm_linux.c,v
retrieving revision 1.38
diff -u -p -r1.38 drm_linux.c
--- dev/pci/drm/drm_linux.c 9 Jun 2019 12:58:30 - 1.38
+++ dev/pci/drm/drm_linux.c 13 Jun 2019 20:53:23 -
@@ -1399,15 +1399,15 @@ drm_linux_init(void)
{
if (system_wq == NULL) {
system_wq = (struct workqueue_struct *)
- taskq_create("drmwq", 1, IPL_HIGH, 0);
+ taskq_create("drmwq", 4, IPL_HIGH, 0);
}
if (system_unbound_wq == NULL) {
system_unbound_wq = (struct workqueue_struct *)
- taskq_create("drmubwq", 1, IPL_HIGH, 0);
+ taskq_create("drmubwq", 4, IPL_HIGH, 0);
}
if (system_long_wq == NULL) {
system_long_wq = (struct workqueue_struct *)
- taskq_create("drmlwq", 1, IPL_HIGH, 0);
+ taskq_create("drmlwq", 4, IPL_HIGH, 0);
}
if (taskletq == NULL)
Index: dev/pci/drm/i915/intel_hotplug.c
===
RCS file: /cvs/src/sys/dev/pci/drm/i915/intel_hotplug.c,v
retrieving revision 1.2
diff -u -p -r1.2 intel_hotplug.c
--- dev/pci/drm/i915/intel_hotplug.c14 Apr 2019 10:14:52 - 1.2
+++ dev/pci/drm/i915/intel_hotplug.c13 Jun 2019 20:53:23 -
@@ -619,7 +619,6 @@ void intel_hpd_init_work(struct drm_i915
INIT_WORK(_priv->hotplug.hotplug_work, i915_hotplug_work_func);
INIT_WORK(_priv->hotplug.dig_port_work, i915_digport_work_func);
INIT_WORK(_priv->hotplug.poll_init_work, i915_hpd_poll_init_work);
- dev_priv->hotplug.poll_init_work.tq = systq;
INIT_DELAYED_WORK(_priv->hotplug.reenable_work,
intel_hpd_irq_storm_reenable_work);
}
Index: kern/kern_task.c
===
RCS file: /cvs/src/sys/kern/kern_task.c,v
retrieving revision 1.25
diff -u -p -r1.25 kern_task.c
--- kern/kern_task.c28 Apr 2019 04:20:40 - 1.25
+++ kern/kern_task.c13 Jun 2019 20:53:23 -
@@ -43,6 +43,7 @@ struct taskq {
TQ_S_DESTROYED
}tq_state;
unsigned int tq_running;
+ unsigned int tq_waiting;
unsigned int tq_nthreads;
unsigned int tq_flags;
const char *tq_name;
@@ -59,6 +60,7 @@ static const char taskq_sys_name[] = "sy
struct taskq taskq_sys = {
TQ_S_CREATED,
0,
+ 0,
1,
0,
taskq_sys_name,
@@ -77,6 +79,7 @@ static const char taskq_sys_mp_name[] =
struct taskq taskq_sys_mp = {
TQ_S_CREATED,
0,
+ 0,
1,
TASKQ_MPSAFE,
taskq_sys_mp_name,
@@ -122,6 +125,7 @@ taskq_create(const char *name, unsigned
tq->tq_state = TQ_S_CREATED;
tq->tq_running = 0;
+ tq->tq_waiting = 0;
tq->tq_nthreads = nthreads;
tq->tq_name = name;
tq->tq_flags = flags;
@@ -223,6 +227,7 @@ taskq_barrier(struct taskq *tq)
WITNESS_CHECKORDER(>tq_lock_object, LOP_NEWORDER, NULL);
+ SET(t.t_flags, TASK_BARRIER);
task_add(tq, );
cond_wait(, "tqbar");
}
@@ -238,6 +243,7 @@ taskq_del_barrier(struct taskq *tq, stru
if (task_del(tq, del))
return;
+ SET(t.t_flags, TASK_BARRIER);
task_add(tq, );
cond_wait(, "tqbar");
}
@@ -304,13 +310,26 @@ taskq_next_work(struct taskq *tq, struct
struct task *next;
mtx_enter(>tq_mtx);
+retry:
while ((next = TAILQ_FIRST(>tq_worklist)) == NULL) {
if (tq->tq_state != TQ_S_RUNNING) {
mtx_leave(>tq_mtx);
return (0);
Re: drm(4), multi-threaded task queues and barriers
Mark Kettenis on Wed, Jun 12 2019: >> Looks good and appears to work fine with two displays (one internal, one >> external). Will test with three at work tomorrow. Your diff also works for me with three displays (inteldrm). > Unfortunately the diff doesn't work with amdgpu. Some more thinking > needed... I would be interested to hear what the issue is. Regards, pesco
Re: drm(4), multi-threaded task queues and barriers
> From: "Sven M. Hallberg"
> Date: Wed, 12 Jun 2019 23:18:24 +0200
>
> Mark Kettenis on Tue, Jun 11 2019:
> > The drm(4) codebase really needs multi-threaded task queues [...]
> >
> > The diff also starts 4 threads for each workqueue that gets created by
> > the drm(4) layer. The number 4 is a bit arbitrary but it is the
> > number of threads that Linux creates per CPU for a so-called "unbound"
> > workqueue which hopefully is enough to always make progress.
> >
> > Please test.
>
> Looks good and appears to work fine with two displays (one internal, one
> external). Will test with three at work tomorrow.
>
>
> > - dev_priv->hotplug.poll_init_work.tq = systq;
>
> Intentional?
Yes. It removes a local modification that should no longer be necessary.
Unfortunately the diff doesn't work with amdgpu. Some more thinking
needed...
> > Index: dev/pci/drm/drm_linux.c
> > ===
> > RCS file: /cvs/src/sys/dev/pci/drm/drm_linux.c,v
> > retrieving revision 1.38
> > diff -u -p -r1.38 drm_linux.c
> > --- dev/pci/drm/drm_linux.c 9 Jun 2019 12:58:30 - 1.38
> > +++ dev/pci/drm/drm_linux.c 11 Jun 2019 18:54:38 -
> > @@ -1399,15 +1399,15 @@ drm_linux_init(void)
> > {
> > if (system_wq == NULL) {
> > system_wq = (struct workqueue_struct *)
> > - taskq_create("drmwq", 1, IPL_HIGH, 0);
> > + taskq_create("drmwq", 4, IPL_HIGH, 0);
> > }
> > if (system_unbound_wq == NULL) {
> > system_unbound_wq = (struct workqueue_struct *)
> > - taskq_create("drmubwq", 1, IPL_HIGH, 0);
> > + taskq_create("drmubwq", 4, IPL_HIGH, 0);
> > }
> > if (system_long_wq == NULL) {
> > system_long_wq = (struct workqueue_struct *)
> > - taskq_create("drmlwq", 1, IPL_HIGH, 0);
> > + taskq_create("drmlwq", 4, IPL_HIGH, 0);
> > }
> >
> > if (taskletq == NULL)
> > Index: dev/pci/drm/i915/intel_hotplug.c
> > ===
> > RCS file: /cvs/src/sys/dev/pci/drm/i915/intel_hotplug.c,v
> > retrieving revision 1.2
> > diff -u -p -r1.2 intel_hotplug.c
> > --- dev/pci/drm/i915/intel_hotplug.c14 Apr 2019 10:14:52 -
> > 1.2
> > +++ dev/pci/drm/i915/intel_hotplug.c11 Jun 2019 18:54:38 -
> > @@ -619,7 +619,6 @@ void intel_hpd_init_work(struct drm_i915
> > INIT_WORK(_priv->hotplug.hotplug_work, i915_hotplug_work_func);
> > INIT_WORK(_priv->hotplug.dig_port_work, i915_digport_work_func);
> > INIT_WORK(_priv->hotplug.poll_init_work, i915_hpd_poll_init_work);
> > - dev_priv->hotplug.poll_init_work.tq = systq;
> > INIT_DELAYED_WORK(_priv->hotplug.reenable_work,
> > intel_hpd_irq_storm_reenable_work);
> > }
> > Index: kern/kern_task.c
> > ===
> > RCS file: /cvs/src/sys/kern/kern_task.c,v
> > retrieving revision 1.25
> > diff -u -p -r1.25 kern_task.c
> > --- kern/kern_task.c28 Apr 2019 04:20:40 - 1.25
> > +++ kern/kern_task.c11 Jun 2019 18:54:39 -
> > @@ -43,6 +43,7 @@ struct taskq {
> > TQ_S_DESTROYED
> > }tq_state;
> > unsigned int tq_running;
> > + unsigned int tq_barrier;
> > unsigned int tq_nthreads;
> > unsigned int tq_flags;
> > const char *tq_name;
> > @@ -59,6 +60,7 @@ static const char taskq_sys_name[] = "sy
> > struct taskq taskq_sys = {
> > TQ_S_CREATED,
> > 0,
> > + 0,
> > 1,
> > 0,
> > taskq_sys_name,
> > @@ -77,6 +79,7 @@ static const char taskq_sys_mp_name[] =
> > struct taskq taskq_sys_mp = {
> > TQ_S_CREATED,
> > 0,
> > + 0,
> > 1,
> > TASKQ_MPSAFE,
> > taskq_sys_mp_name,
> > @@ -122,6 +125,7 @@ taskq_create(const char *name, unsigned
> >
> > tq->tq_state = TQ_S_CREATED;
> > tq->tq_running = 0;
> > + tq->tq_barrier = 0;
> > tq->tq_nthreads = nthreads;
> > tq->tq_name = name;
> > tq->tq_flags = flags;
> > @@ -161,6 +165,7 @@ taskq_destroy(struct taskq *tq)
> > panic("unexpected %s tq state %u", tq->tq_name, tq->tq_state);
> > }
> >
> > + tq->tq_barrier = 0;
> > while (tq->tq_running > 0) {
> > wakeup(tq);
> > msleep(>tq_running, >tq_mtx, PWAIT, "tqdestroy", 0);
> > @@ -223,6 +228,7 @@ taskq_barrier(struct taskq *tq)
> >
> > WITNESS_CHECKORDER(>tq_lock_object, LOP_NEWORDER, NULL);
> >
> > + SET(t.t_flags, TASK_BARRIER);
> > task_add(tq, );
> > cond_wait(, "tqbar");
> > }
> > @@ -238,6 +244,7 @@ taskq_del_barrier(struct taskq *tq, stru
> > if (task_del(tq, del))
> > return;
> >
> > + SET(t.t_flags, TASK_BARRIER);
> > task_add(tq, );
> > cond_wait(, "tqbar");
> > }
> > @@ -304,6 +311,7 @@ taskq_next_work(struct taskq *tq, struct
>
Re: drm(4), multi-threaded task queues and barriers
Mark Kettenis on Tue, Jun 11 2019:
> The drm(4) codebase really needs multi-threaded task queues [...]
>
> The diff also starts 4 threads for each workqueue that gets created by
> the drm(4) layer. The number 4 is a bit arbitrary but it is the
> number of threads that Linux creates per CPU for a so-called "unbound"
> workqueue which hopefully is enough to always make progress.
>
> Please test.
Looks good and appears to work fine with two displays (one internal, one
external). Will test with three at work tomorrow.
> - dev_priv->hotplug.poll_init_work.tq = systq;
Intentional?
-p
> Index: dev/pci/drm/drm_linux.c
> ===
> RCS file: /cvs/src/sys/dev/pci/drm/drm_linux.c,v
> retrieving revision 1.38
> diff -u -p -r1.38 drm_linux.c
> --- dev/pci/drm/drm_linux.c 9 Jun 2019 12:58:30 - 1.38
> +++ dev/pci/drm/drm_linux.c 11 Jun 2019 18:54:38 -
> @@ -1399,15 +1399,15 @@ drm_linux_init(void)
> {
> if (system_wq == NULL) {
> system_wq = (struct workqueue_struct *)
> - taskq_create("drmwq", 1, IPL_HIGH, 0);
> + taskq_create("drmwq", 4, IPL_HIGH, 0);
> }
> if (system_unbound_wq == NULL) {
> system_unbound_wq = (struct workqueue_struct *)
> - taskq_create("drmubwq", 1, IPL_HIGH, 0);
> + taskq_create("drmubwq", 4, IPL_HIGH, 0);
> }
> if (system_long_wq == NULL) {
> system_long_wq = (struct workqueue_struct *)
> - taskq_create("drmlwq", 1, IPL_HIGH, 0);
> + taskq_create("drmlwq", 4, IPL_HIGH, 0);
> }
>
> if (taskletq == NULL)
> Index: dev/pci/drm/i915/intel_hotplug.c
> ===
> RCS file: /cvs/src/sys/dev/pci/drm/i915/intel_hotplug.c,v
> retrieving revision 1.2
> diff -u -p -r1.2 intel_hotplug.c
> --- dev/pci/drm/i915/intel_hotplug.c 14 Apr 2019 10:14:52 - 1.2
> +++ dev/pci/drm/i915/intel_hotplug.c 11 Jun 2019 18:54:38 -
> @@ -619,7 +619,6 @@ void intel_hpd_init_work(struct drm_i915
> INIT_WORK(_priv->hotplug.hotplug_work, i915_hotplug_work_func);
> INIT_WORK(_priv->hotplug.dig_port_work, i915_digport_work_func);
> INIT_WORK(_priv->hotplug.poll_init_work, i915_hpd_poll_init_work);
> - dev_priv->hotplug.poll_init_work.tq = systq;
> INIT_DELAYED_WORK(_priv->hotplug.reenable_work,
> intel_hpd_irq_storm_reenable_work);
> }
> Index: kern/kern_task.c
> ===
> RCS file: /cvs/src/sys/kern/kern_task.c,v
> retrieving revision 1.25
> diff -u -p -r1.25 kern_task.c
> --- kern/kern_task.c 28 Apr 2019 04:20:40 - 1.25
> +++ kern/kern_task.c 11 Jun 2019 18:54:39 -
> @@ -43,6 +43,7 @@ struct taskq {
> TQ_S_DESTROYED
> }tq_state;
> unsigned int tq_running;
> + unsigned int tq_barrier;
> unsigned int tq_nthreads;
> unsigned int tq_flags;
> const char *tq_name;
> @@ -59,6 +60,7 @@ static const char taskq_sys_name[] = "sy
> struct taskq taskq_sys = {
> TQ_S_CREATED,
> 0,
> + 0,
> 1,
> 0,
> taskq_sys_name,
> @@ -77,6 +79,7 @@ static const char taskq_sys_mp_name[] =
> struct taskq taskq_sys_mp = {
> TQ_S_CREATED,
> 0,
> + 0,
> 1,
> TASKQ_MPSAFE,
> taskq_sys_mp_name,
> @@ -122,6 +125,7 @@ taskq_create(const char *name, unsigned
>
> tq->tq_state = TQ_S_CREATED;
> tq->tq_running = 0;
> + tq->tq_barrier = 0;
> tq->tq_nthreads = nthreads;
> tq->tq_name = name;
> tq->tq_flags = flags;
> @@ -161,6 +165,7 @@ taskq_destroy(struct taskq *tq)
> panic("unexpected %s tq state %u", tq->tq_name, tq->tq_state);
> }
>
> + tq->tq_barrier = 0;
> while (tq->tq_running > 0) {
> wakeup(tq);
> msleep(>tq_running, >tq_mtx, PWAIT, "tqdestroy", 0);
> @@ -223,6 +228,7 @@ taskq_barrier(struct taskq *tq)
>
> WITNESS_CHECKORDER(>tq_lock_object, LOP_NEWORDER, NULL);
>
> + SET(t.t_flags, TASK_BARRIER);
> task_add(tq, );
> cond_wait(, "tqbar");
> }
> @@ -238,6 +244,7 @@ taskq_del_barrier(struct taskq *tq, stru
> if (task_del(tq, del))
> return;
>
> + SET(t.t_flags, TASK_BARRIER);
> task_add(tq, );
> cond_wait(, "tqbar");
> }
> @@ -304,6 +311,7 @@ taskq_next_work(struct taskq *tq, struct
> struct task *next;
>
> mtx_enter(>tq_mtx);
> +retry:
> while ((next = TAILQ_FIRST(>tq_worklist)) == NULL) {
> if (tq->tq_state != TQ_S_RUNNING) {
> mtx_leave(>tq_mtx);
> @@ -311,6 +319,17 @@ taskq_next_work(struct taskq *tq, struct
> }
>
> msleep(tq, >tq_mtx, PWAIT, "bored",
Re: drm(4), multi-threaded task queues and barriers
> Date: Wed, 12 Jun 2019 17:04:10 +1000
> From: Jonathan Gray
>
> On Tue, Jun 11, 2019 at 09:10:46PM +0200, Mark Kettenis wrote:
> > The drm(4) codebase really needs multi-threaded task queues since the
> > code has taks that wait for the completion of other tasks that are
> > submitted to the same task queue. Thank you Linux...
> >
> > Unfortunately the code also needs to wait for the completion of
> > submitted tasks from other threads. This implemented using
> > task_barrier(9). But unfortunately our current implementation only
> > works for single-threaded task queues.
> >
> > The diff below fixes this by marking the barrier task with a flag and
> > making sure that all threads of the task queue are syncchronized.
> > This achived through a TASK_BARRIER flag that simply blocks the
> > threads until the last unblocked thread sees the flag and executes the
> > task.
> >
> > The diff also starts 4 threads for each workqueue that gets created by
> > the drm(4) layer. The number 4 is a bit arbitrary but it is the
> > number of threads that Linux creates per CPU for a so-called "unbound"
> > workqueue which hopefully is enough to always make progress.
>
> Ignoring taskletq and systq use in burner_task/switchtask across
> drivers and the local only backlight_schedule_update_status(), was it
> intentional to exclude:
>
> workqueue.h: alloc_workqueue()
> struct taskq *tq = taskq_create(name, 1, IPL_TTY, 0);
> workqueue.h: alloc_ordered_workqueue()
> struct taskq *tq = taskq_create(name, 1, IPL_TTY, 0);
> workqueue.h:
> #define system_power_efficient_wq ((struct workqueue_struct *)systq)
Not entirely. But I don't want to spawn too many threads... If this
diff seems to work, we might want to tweak the numbers a bit to see
what works best. Whatever we do, the ordered workqueues need to stay
single-threaded.
> > Please test. If you experience a "hang" with this diff, please try to
> > log in to the machine remotely over ssh and send me and jsg@ the
> > output of "ps -AHwlk".
> >
> > Thanks,
> >
> > Mark
> >
> >
> > Index: dev/pci/drm/drm_linux.c
> > ===
> > RCS file: /cvs/src/sys/dev/pci/drm/drm_linux.c,v
> > retrieving revision 1.38
> > diff -u -p -r1.38 drm_linux.c
> > --- dev/pci/drm/drm_linux.c 9 Jun 2019 12:58:30 - 1.38
> > +++ dev/pci/drm/drm_linux.c 11 Jun 2019 18:54:38 -
> > @@ -1399,15 +1399,15 @@ drm_linux_init(void)
> > {
> > if (system_wq == NULL) {
> > system_wq = (struct workqueue_struct *)
> > - taskq_create("drmwq", 1, IPL_HIGH, 0);
> > + taskq_create("drmwq", 4, IPL_HIGH, 0);
> > }
> > if (system_unbound_wq == NULL) {
> > system_unbound_wq = (struct workqueue_struct *)
> > - taskq_create("drmubwq", 1, IPL_HIGH, 0);
> > + taskq_create("drmubwq", 4, IPL_HIGH, 0);
> > }
> > if (system_long_wq == NULL) {
> > system_long_wq = (struct workqueue_struct *)
> > - taskq_create("drmlwq", 1, IPL_HIGH, 0);
> > + taskq_create("drmlwq", 4, IPL_HIGH, 0);
> > }
> >
> > if (taskletq == NULL)
> > Index: dev/pci/drm/i915/intel_hotplug.c
> > ===
> > RCS file: /cvs/src/sys/dev/pci/drm/i915/intel_hotplug.c,v
> > retrieving revision 1.2
> > diff -u -p -r1.2 intel_hotplug.c
> > --- dev/pci/drm/i915/intel_hotplug.c14 Apr 2019 10:14:52 -
> > 1.2
> > +++ dev/pci/drm/i915/intel_hotplug.c11 Jun 2019 18:54:38 -
> > @@ -619,7 +619,6 @@ void intel_hpd_init_work(struct drm_i915
> > INIT_WORK(_priv->hotplug.hotplug_work, i915_hotplug_work_func);
> > INIT_WORK(_priv->hotplug.dig_port_work, i915_digport_work_func);
> > INIT_WORK(_priv->hotplug.poll_init_work, i915_hpd_poll_init_work);
> > - dev_priv->hotplug.poll_init_work.tq = systq;
> > INIT_DELAYED_WORK(_priv->hotplug.reenable_work,
> > intel_hpd_irq_storm_reenable_work);
> > }
> > Index: kern/kern_task.c
> > ===
> > RCS file: /cvs/src/sys/kern/kern_task.c,v
> > retrieving revision 1.25
> > diff -u -p -r1.25 kern_task.c
> > --- kern/kern_task.c28 Apr 2019 04:20:40 - 1.25
> > +++ kern/kern_task.c11 Jun 2019 18:54:39 -
> > @@ -43,6 +43,7 @@ struct taskq {
> > TQ_S_DESTROYED
> > }tq_state;
> > unsigned int tq_running;
> > + unsigned int tq_barrier;
> > unsigned int tq_nthreads;
> > unsigned int tq_flags;
> > const char *tq_name;
> > @@ -59,6 +60,7 @@ static const char taskq_sys_name[] = "sy
> > struct taskq taskq_sys = {
> > TQ_S_CREATED,
> > 0,
> > + 0,
> > 1,
> > 0,
> > taskq_sys_name,
> > @@ -77,6 +79,7 @@ static const char taskq_sys_mp_name[] =
> >
Re: drm(4), multi-threaded task queues and barriers
On Tue, Jun 11, 2019 at 09:10:46PM +0200, Mark Kettenis wrote:
> The drm(4) codebase really needs multi-threaded task queues since the
> code has taks that wait for the completion of other tasks that are
> submitted to the same task queue. Thank you Linux...
>
> Unfortunately the code also needs to wait for the completion of
> submitted tasks from other threads. This implemented using
> task_barrier(9). But unfortunately our current implementation only
> works for single-threaded task queues.
>
> The diff below fixes this by marking the barrier task with a flag and
> making sure that all threads of the task queue are syncchronized.
> This achived through a TASK_BARRIER flag that simply blocks the
> threads until the last unblocked thread sees the flag and executes the
> task.
>
> The diff also starts 4 threads for each workqueue that gets created by
> the drm(4) layer. The number 4 is a bit arbitrary but it is the
> number of threads that Linux creates per CPU for a so-called "unbound"
> workqueue which hopefully is enough to always make progress.
Ignoring taskletq and systq use in burner_task/switchtask across
drivers and the local only backlight_schedule_update_status(), was it
intentional to exclude:
workqueue.h: alloc_workqueue()
struct taskq *tq = taskq_create(name, 1, IPL_TTY, 0);
workqueue.h: alloc_ordered_workqueue()
struct taskq *tq = taskq_create(name, 1, IPL_TTY, 0);
workqueue.h:
#define system_power_efficient_wq ((struct workqueue_struct *)systq)
>
> Please test. If you experience a "hang" with this diff, please try to
> log in to the machine remotely over ssh and send me and jsg@ the
> output of "ps -AHwlk".
>
> Thanks,
>
> Mark
>
>
> Index: dev/pci/drm/drm_linux.c
> ===
> RCS file: /cvs/src/sys/dev/pci/drm/drm_linux.c,v
> retrieving revision 1.38
> diff -u -p -r1.38 drm_linux.c
> --- dev/pci/drm/drm_linux.c 9 Jun 2019 12:58:30 - 1.38
> +++ dev/pci/drm/drm_linux.c 11 Jun 2019 18:54:38 -
> @@ -1399,15 +1399,15 @@ drm_linux_init(void)
> {
> if (system_wq == NULL) {
> system_wq = (struct workqueue_struct *)
> - taskq_create("drmwq", 1, IPL_HIGH, 0);
> + taskq_create("drmwq", 4, IPL_HIGH, 0);
> }
> if (system_unbound_wq == NULL) {
> system_unbound_wq = (struct workqueue_struct *)
> - taskq_create("drmubwq", 1, IPL_HIGH, 0);
> + taskq_create("drmubwq", 4, IPL_HIGH, 0);
> }
> if (system_long_wq == NULL) {
> system_long_wq = (struct workqueue_struct *)
> - taskq_create("drmlwq", 1, IPL_HIGH, 0);
> + taskq_create("drmlwq", 4, IPL_HIGH, 0);
> }
>
> if (taskletq == NULL)
> Index: dev/pci/drm/i915/intel_hotplug.c
> ===
> RCS file: /cvs/src/sys/dev/pci/drm/i915/intel_hotplug.c,v
> retrieving revision 1.2
> diff -u -p -r1.2 intel_hotplug.c
> --- dev/pci/drm/i915/intel_hotplug.c 14 Apr 2019 10:14:52 - 1.2
> +++ dev/pci/drm/i915/intel_hotplug.c 11 Jun 2019 18:54:38 -
> @@ -619,7 +619,6 @@ void intel_hpd_init_work(struct drm_i915
> INIT_WORK(_priv->hotplug.hotplug_work, i915_hotplug_work_func);
> INIT_WORK(_priv->hotplug.dig_port_work, i915_digport_work_func);
> INIT_WORK(_priv->hotplug.poll_init_work, i915_hpd_poll_init_work);
> - dev_priv->hotplug.poll_init_work.tq = systq;
> INIT_DELAYED_WORK(_priv->hotplug.reenable_work,
> intel_hpd_irq_storm_reenable_work);
> }
> Index: kern/kern_task.c
> ===
> RCS file: /cvs/src/sys/kern/kern_task.c,v
> retrieving revision 1.25
> diff -u -p -r1.25 kern_task.c
> --- kern/kern_task.c 28 Apr 2019 04:20:40 - 1.25
> +++ kern/kern_task.c 11 Jun 2019 18:54:39 -
> @@ -43,6 +43,7 @@ struct taskq {
> TQ_S_DESTROYED
> }tq_state;
> unsigned int tq_running;
> + unsigned int tq_barrier;
> unsigned int tq_nthreads;
> unsigned int tq_flags;
> const char *tq_name;
> @@ -59,6 +60,7 @@ static const char taskq_sys_name[] = "sy
> struct taskq taskq_sys = {
> TQ_S_CREATED,
> 0,
> + 0,
> 1,
> 0,
> taskq_sys_name,
> @@ -77,6 +79,7 @@ static const char taskq_sys_mp_name[] =
> struct taskq taskq_sys_mp = {
> TQ_S_CREATED,
> 0,
> + 0,
> 1,
> TASKQ_MPSAFE,
> taskq_sys_mp_name,
> @@ -122,6 +125,7 @@ taskq_create(const char *name, unsigned
>
> tq->tq_state = TQ_S_CREATED;
> tq->tq_running = 0;
> + tq->tq_barrier = 0;
> tq->tq_nthreads = nthreads;
> tq->tq_name = name;
> tq->tq_flags = flags;
> @@ -161,6 +165,7 @@ taskq_destroy(struct taskq *tq)
>
