Date: Tue, 28 Jan 2020 12:11:53 -0800 From: bob prohaska <fbsd@www.zefox.net> To: Mark Millard <marklmi@yahoo.com> Cc: Konstantin Belousov <kib@freebsd.org>, freebsd-arm <freebsd-arm@freebsd.org>, FreeBSD Current <freebsd-current@freebsd.org>, bob prohaska <fbsd@www.zefox.net> Subject: Re: OOMA kill with vm.pfault_oom_attempts="-1" on RPi3 at r357147 (a vm_pfault_oom_attempts < 0 handling bug as of head -r357026) Message-ID: <20200128201152.GA15110@www.zefox.net> In-Reply-To: <94E68249-7751-4B27-AE95-E9C2776D730B@yahoo.com> References: <20200127190709.GA11328@www.zefox.net> <D20DCE73-29D6-4184-80BF-7698EC907B60@yahoo.com> <20200128035317.GA12644@www.zefox.net> <18150258-6210-451E-A5B9-528129A05974@yahoo.com> <9BF68EF1-F83A-473B-9A7B-B3956D6A5EFD@yahoo.com> <20200128170518.GA14654@www.zefox.net> <5A3CE2DA-C5B8-4CC1-BEEA-8B9649A20B8B@yahoo.com> <20200128190210.GA14784@www.zefox.net> <94E68249-7751-4B27-AE95-E9C2776D730B@yahoo.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Tue, Jan 28, 2020 at 11:28:14AM -0800, Mark Millard wrote:
>
>
> On 2020-Jan-28, at 11:02, bob prohaska <fbsd at www.zefox.net> wrote:
>
> > On Tue, Jan 28, 2020 at 09:42:17AM -0800, Mark Millard wrote:
> >>
> >>
> >>
> > The (partly)modified kernel compiled and booted without
> > obvious trouble. It's trying to finish buildworld now.
> >
Stopped already, with
Jan 28 11:41:59 www kernel: pid 29909 (cc), jid 0, uid 0, was killed: fault's page allocation failed
> >> If you are testing with vm.pfault_oom_attempts="-1" then
> >> the vm_fault printf message should never happen anyway.
> >>
> > Would it not be interesting if the message appeared in that
> > case?
>
> Thanks for the question: looking at the new code found a bug
> causing oom where it used to be avoided in head -r357025 and
> before.
Glad to be of service, even if inadvertently 8-)
> After vm_waitpfault(dset, vm_pfault_oom_wait * hz)
> the -r357026 code does a vm_pageout_oom(VM_OOM_MEM_PF) no
> matter what, even when vm_pfault_oom_attempts < 0 ||
> fs->oom < vm_pfault_oom_attempts :
>
> New code in head -r357026
> ( nothing to avoid the vm_pageout_oom(VM_OOM_MEM_PF)
> for vm_pfault_oom_attempts < 0 ||
> fs->oom < vm_pfault_oom_attempts ):
>
> if (fs->m == NULL) {
> unlock_and_deallocate(fs);
> if (vm_pfault_oom_attempts < 0 ||
> fs->oom < vm_pfault_oom_attempts) {
> fs->oom++;
> vm_waitpfault(dset, vm_pfault_oom_wait * hz);
> }
> if (bootverbose)
> printf(
> "proc %d (%s) failed to alloc page on fault, starting OOM\n",
> curproc->p_pid, curproc->p_comm);
> vm_pageout_oom(VM_OOM_MEM_PF);
> return (KERN_RESOURCE_SHORTAGE);
> }
>
> Old code in head -r357025
> ( has the goto RetryFault_oom after vm_waitpfault(. . .),
> thereby avoiding the vm_pageout_oom(VM_OOM_MEM_PF) for
> vm_pfault_oom_attempts < 0 || fs->oom < vm_pfault_oom_attempts ) :
>
> if (fs.m == NULL) {
> unlock_and_deallocate(&fs);
> if (vm_pfault_oom_attempts < 0 ||
> oom < vm_pfault_oom_attempts) {
> oom++;
> vm_waitpfault(dset,
> vm_pfault_oom_wait * hz);
> goto RetryFault_oom;
> }
> if (bootverbose)
> printf(
> "proc %d (%s) failed to alloc page on fault, starting OOM\n",
> curproc->p_pid, curproc->p_comm);
> vm_pageout_oom(VM_OOM_MEM_PF);
> goto RetryFault;
> }
>
> I expect this is the source of the behavioral
> difference folks have been seeing for OOM kills.
>
>
> As for "gather evidence" messages . . .
>
> >> You may be able to just look and manually delete or
> >> comment out the bootverbose line in the more modern
> >> source that currently looks like:
> >>
> >> if (bootverbose)
> >> printf(
> >> "proc %d (%s) failed to alloc page on fault, starting OOM\n",
> >> curproc->p_pid, curproc->p_comm);
> >> vm_pageout_oom(VM_OOM_MEM_PF);
> >> return (KERN_RESOURCE_SHORTAGE);
> >>
> >
> > I can find those lines in /usr/src/sys/vm/vm_fault.c, but
> > unclear on the motivation to comment the lines out. Perhaps
> > to eliminate the return(...) ? Anyway, is it sufficient
> > to insert /* before and */ after?
>
> The only line to delete or comment out in that
> code block is:
>
> if (bootverbose)
>
> Disabling that line makes the following printf
> always happen, even when a verbose boot was not
> done.
Oops, it's commented out now and the kernel is rebuilding.
>
> Based on the above reported code change, having
> a message before vm_pageout_oom(VM_OOM_MEM_PF) is
> important to getting a report of the kill being
> via that code.
>
Thank you!
bob prohaska
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20200128201152.GA15110>