On Wed, Jun 21, 2023 at 04:40:59PM -0500, Scott Cheloha wrote:
> On Wed, Jun 21, 2023 at 01:03:03PM -0600, Chris Waddey wrote:
> > Sorry for breaking the thread, I wasn't subscribed to misc, but found
> > this in the archives.
> >
> > After some testing, it looks like the recent uvm_meter() commit is what
> > did this (to my machine at least).
> >
> > The git commit for that is 71d823ace2523fb9fee2d1ab9b4d92a18d3f5714.
> >
> > I compiled the commit right before it in the logs and booted no problems
> > with a GENERIC.MP kernel config, but that one broke it.
> >
> > I'm not as familiar with CVS, so apologies for not having the commit
> > from there.
> >
> > Here is the commit message if that helps, though I those on tech will
> > know it regardless:
> >
> > schedcpu, uvm_meter(9): make uvm_meter() an independent timeout
> >
> > uvm_meter(9) should not base its periodic uvm_loadav() call on the UTC
> > clock. It also no longer needs to periodically wake up proc0 because
> > proc0 doesn't do any work. schedcpu() itself may change or go away,
> > but as kettenis@ notes we probably can't completely remove the concept
> > of a "load average" from OpenBSD, given its long Unix heritage.
> >
> > So, (1) remove the uvm_meter() call from schedcpu(), (2) make
> > uvm_meter() an independent timeout started alongside schedcpu() during
> > scheduler_start(), and (3) delete the vestigial periodic proc0 wakeup.
> >
> > With input from deraadt@, kettenis@, and claudio@. deraadt@ cautions
> > that this change may confuse administrators who hold the load average
> > in high regard.
> >
> > Thread: https://marc.info/?l=openbsd-tech&m=168710929409153&w=2
> >
> > general agreement with this direction from kettenis@
> > ok claudio@
> >
> > If I should repost on tech, let me know.
>
> Thank you for bisecting the issue.
>
> Could you try this tweaked version of that patch? I am trying to
> narrow down the chunk responsible for the hang. I cannot trigger the
> hang on any of my machines. Clearly you have something in your setup
> that I don't.
>
> There aren't many possibilities.
>
> 1. Separating uvm_meter() from schedcpu().
> 2. Not using the UTC clock to decide when to update load averages.
> 3. Not periodically awakening proc0.
>
> Let's try 3.
>
> Index: kern/sched_bsd.c
> ===================================================================
> RCS file: /cvs/src/sys/kern/sched_bsd.c,v
> retrieving revision 1.76
> diff -u -p -r1.76 sched_bsd.c
> --- kern/sched_bsd.c 21 Jun 2023 21:16:21 -0000 1.76
> +++ kern/sched_bsd.c 21 Jun 2023 21:40:17 -0000
> @@ -234,7 +234,6 @@ schedcpu(void *arg)
> }
> SCHED_UNLOCK(s);
> }
> - uvm_meter();
> wakeup(&lbolt);
> timeout_add_sec(to, 1);
> }
> @@ -669,6 +668,7 @@ scheduler_start(void)
>
> rrticks_init = hz / 10;
> schedcpu(&schedcpu_to);
> + uvm_meter(NULL);
>
> #ifndef SMALL_KERNEL
> if (perfpolicy == PERFPOL_AUTO)
> Index: uvm/uvm_meter.c
> ===================================================================
> RCS file: /cvs/src/sys/uvm/uvm_meter.c,v
> retrieving revision 1.44
> diff -u -p -r1.44 uvm_meter.c
> --- uvm/uvm_meter.c 21 Jun 2023 21:16:21 -0000 1.44
> +++ uvm/uvm_meter.c 21 Jun 2023 21:40:17 -0000
> @@ -65,6 +65,9 @@
> int maxslp = MAXSLP; /* patchable ... */
> struct loadavg averunnable;
>
> +#define UVM_METER_INTVL 5
> +struct timeout uvm_meter_to = TIMEOUT_INITIALIZER(uvm_meter, NULL);
> +
> /*
> * constants for averages over 1, 5, and 15 minutes when sampling at
> * 5 second intervals.
> @@ -85,10 +88,10 @@ void uvmexp_read(struct uvmexp *);
> * uvm_meter: calculate load average and wake up the swapper (if needed)
> */
> void
> -uvm_meter(void)
> +uvm_meter(void *unused)
> {
> - if ((gettime() % 5) == 0)
> - uvm_loadav(&averunnable);
> + timeout_add_sec(&uvm_meter_to, UVM_METER_INTVL);
> + uvm_loadav(&averunnable);
> if (proc0.p_slptime > (maxslp / 2))
> wakeup(&proc0);
> }
> Index: uvm/uvm_extern.h
> ===================================================================
> RCS file: /cvs/src/sys/uvm/uvm_extern.h,v
> retrieving revision 1.170
> diff -u -p -r1.170 uvm_extern.h
> --- uvm/uvm_extern.h 21 Jun 2023 21:16:21 -0000 1.170
> +++ uvm/uvm_extern.h 21 Jun 2023 21:40:17 -0000
> @@ -414,7 +414,7 @@ void uvmspace_free(struct vmspace *);
> struct vmspace *uvmspace_share(struct process *);
> int uvm_share(vm_map_t, vaddr_t, vm_prot_t,
> vm_map_t, vaddr_t, vsize_t);
> -void uvm_meter(void);
> +void uvm_meter(void *);
> int uvm_sysctl(int *, u_int, void *, size_t *,
> void *, size_t, struct proc *);
> struct vm_page *uvm_pagealloc(struct uvm_object *,
That worked. Applied it to the last commit that worked for me, built GENERIC.MP,
rebooted and everything seems normal.
No comments:
Post a Comment