Re: [PATCH v4 02/16] sched/core: uclamp: map TASK's clamp values into CPU's clamp groups

[Patrick Bellasi]

On 12-Sep 18:12, Peter Zijlstra wrote:
> On Wed, Sep 12, 2018 at 04:56:19PM +0100, Patrick Bellasi wrote:
> > On 12-Sep 15:49, Peter Zijlstra wrote:
> > > On Tue, Aug 28, 2018 at 02:53:10PM +0100, Patrick Bellasi wrote:
> 
> > > > +/**
> > > > + * uclamp_map: reference counts a utilization "clamp value"
> > > > + * @value:    the utilization "clamp value" required
> > > > + * @se_count: the number of scheduling entities requiring the "clamp value"
> > > > + * @se_lock:  serialize reference count updates by protecting se_count
> > > 
> > > Why do you have a spinlock to serialize a single value? Don't we have
> > > atomics for that?
> > 
> > There are some code paths where it's used to protect clamp groups
> > mapping and initialization, e.g.
> > 
> >       uclamp_group_get()
> >           spin_lock()
> >               // initialize clamp group (if required) and then...
                     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
               This is actually a couple of function calls

> >               se_count += 1
> >           spin_unlock()
> > 
> > Almost all these paths are triggered from user-space and protected
> > by a global uclamp_mutex, but fork/exit paths.
> > 
> > To serialize these paths I'm using the spinlock above, does it make
> > sense ? Can we use the global uclamp_mutex on forks/exit too ?
> 
> OK, then your comment is misleading; it serializes both fields.

Yes... that definitively needs an update.

> > One additional observations is that, if in the future we want to add a
> > kernel space API, (e.g. driver asking for a new clamp value), maybe we
> > will need to have a serialized non-sleeping uclamp_group_get() API ?
> 
> No idea; but if you want to go all fancy you can replace he whole
> uclamp_map thing with something like:
> 
> struct uclamp_map {
> 	union {
> 		struct {
> 			unsigned long v : 10;
> 			unsigned long c : BITS_PER_LONG - 10;
> 		};
> 		atomic_long_t s;
> 	};
> };

That sounds really cool and scary at the same time :)

The v:10 requires that we never set SCHED_CAPACITY_SCALE>1024
or that we use it to track a percentage value (i.e. [0..100]).

One of the last patches introduces percentage values to userspace.
But, I was considering that in kernel space we should always track
full scale utilization values.

The c:(BITS_PER_LONG-10) restricts the range of concurrently active
SE refcounting the same clamp value. Which, for some 32bit systems is
only 4 milions among tasks and cgroups... maybe still reasonable...


> And use uclamp_map::c == 0 as unused (as per normal refcount
> semantics) and atomic_long_cmpxchg() the whole thing using
> uclamp_map::s.

Yes... that could work for the uclamp_map updates, but as I noted
above, I think I have other calls serialized by that lock... will look
better into what you suggest, thanks!


[...]

> > > What's the purpose of that cacheline align statement?
> > 
> > In uclamp_maps, we still need to scan the array when a clamp value is
> > changed from user-space, i.e. the cases reported above. Thus, that
> > alignment is just to ensure that we minimize the number of cache lines
> > used. Does that make sense ?
> > 
> > Maybe that alignment implicitly generated by the compiler ?
> 
> It is not, but if it really is a slow path, we shouldn't care about
> alignment.

Ok, will remove it.

> > > Note that without that apparently superfluous lock, it would be 8*12 =
> > > 96 bytes, which is 1.5 lines and would indeed suggest you default to
> > > GROUP_COUNT=7 by default to fill 2 lines.
> > 
> > Yes, will check better if we can count on just the uclamp_mutex
> 
> Well, if we don't care about performance (slow path) then keeping he
> lock is fine, just the comment and alignment are misleading.

Ok

[...]

Cheers,
Patrick

-- 
#include <best/regards.h>

Patrick Bellasi