lkml.org 
[lkml]   [2019]   [Aug]   [2]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH 0/5] Fix FIFO-99 abuse
On 08/02/19 11:32, Peter Zijlstra wrote:
> On Thu, Aug 01, 2019 at 02:17:07PM +0100, Qais Yousef wrote:
> > On 08/01/19 13:13, Peter Zijlstra wrote:
> > > I noticed a bunch of kthreads defaulted to FIFO-99, fix them.
> > >
> > > The generic default is FIFO-50, the admin will have to configure the system
> > > anyway.
> > >
> > > For some the purpose is to be above OTHER and then FIFO-1 really is sufficient.
> >
> > I was looking in this area too and was thinking of a way to consolidate the
> > creation of RT/DL tasks in the kernel and the way we set the priority.
> >
> > Does it make sense to create a new header for RT priorities for kthreads
> > created in the kernel so that we can easily track and rationale about the
> > relative priorities of in-kernel RT tasks?
> >
> > When working in the FW world such a header helped a lot in understanding what
> > runs at each priority level and how to reason about what priority level makes
> > sense for a new item. It could be a nice single point of reference; even for
> > admins.
>
> Well, SCHED_FIFO is a broken scheduler model; that is, it is
> fundamentally incapable of resource management, which is the one thing
> an OS really should be doing.
>
> This is of course the reason it is limited to privileged users only.
>
> Worse still; it is fundamentally impossible to compose static priority
> workloads. You cannot take two correctly working static prio workloads
> and smash them together and still expect them to work.
>
> For this reason 'all' FIFO tasks the kernel creates are basically at:
>
> MAX_RT_PRIO / 2
>
> The administrator _MUST_ configure the system, the kernel simply doesn't
> know enough information to make a sensible choice.
>
> Now, Geert suggested so make make a define for that, but how about we do
> something like:
>
> /*
> * ${the above explanation}
> */
> int kernel_setscheduler_fifo(struct task_struct *p)
> {
> struct sched_param sp = { .sched_priority = MAX_RT_PRIO / 2 };
> return sched_setscheduler_nocheck(p, SCHED_FIFO, &sp);
> }
>
> And then take away sched_setscheduler*().

Yes a somewhat enforced default makes more sense to me. I assume you no longer
want to put the kthreads that just need to be above OTHER in FIFO-1?

While at it, since we will cram all kthreads on the same priority, isn't
a SCHED_RR a better choice now? I think the probability of a clash is pretty
low, but when it happens, shouldn't we try to guarantee some fairness?

--
Qais Yousef

\
 
 \ /
  Last update: 2019-08-02 12:27    [W:0.084 / U:0.060 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site