Linux Virtualization - Nov 2016 - [GIT PULL v2 1/5] processor.h: introduce cpu_relax_yield

On Tue, Oct 25, 2016 at 11:03:11AM +0200, Christian Borntraeger
wrote:
> For spinning loops people do often use barrier() or cpu_relax().
> For most architectures cpu_relax and barrier are the same, but on
> some architectures cpu_relax can add some latency.
> For example on power,sparc64 and arc, cpu_relax can shift the CPU
> towards other hardware threads in an SMT environment.
> On s390 cpu_relax does even more, it uses an hypercall to the
> hypervisor to give up the timeslice.
> In contrast to the SMT yielding this can result in larger latencies.
> In some places this latency is unwanted, so another variant
> "cpu_relax_lowlatency" was introduced. Before this is used in
more
> and more places, lets revert the logic and provide a cpu_relax_yield
> that can be called in places where yielding is more important than
> latency. By default this is the same as cpu_relax on all architectures.
Rather than having to update all these architectures in this way, can't
we put in some linux/*.h header something like:

#ifndef cpu_relax_yield
#define cpu_relax_yield() cpu_relax()
#endif

so only those architectures that need to do something need to be
modified?

-- 
RMK's Patch system: http://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line: currently at 9.6Mbps down 400kbps up
according to speedtest.net.

On 11/15/2016 01:30 PM, Russell King - ARM Linux wrote:
> On Tue, Oct 25, 2016 at 11:03:11AM +0200, Christian Borntraeger wrote:
>> For spinning loops people do often use barrier() or cpu_relax().
>> For most architectures cpu_relax and barrier are the same, but on
>> some architectures cpu_relax can add some latency.
>> For example on power,sparc64 and arc, cpu_relax can shift the CPU
>> towards other hardware threads in an SMT environment.
>> On s390 cpu_relax does even more, it uses an hypercall to the
>> hypervisor to give up the timeslice.
>> In contrast to the SMT yielding this can result in larger latencies.
>> In some places this latency is unwanted, so another variant
>> "cpu_relax_lowlatency" was introduced. Before this is used in
more
>> and more places, lets revert the logic and provide a cpu_relax_yield
>> that can be called in places where yielding is more important than
>> latency. By default this is the same as cpu_relax on all architectures.
> 
> Rather than having to update all these architectures in this way, can't
> we put in some linux/*.h header something like:
> 
> #ifndef cpu_relax_yield
> #define cpu_relax_yield() cpu_relax()
> #endif
> 
> so only those architectures that need to do something need to be
> modified?
These patches are part of linux-next since a month or so, changing that 
would invalidate all the next testing. If people want that, I can certainly
do that, though.

On Tue, Nov 15, 2016 at 02:19:53PM +0100, Christian Borntraeger
wrote:
> On 11/15/2016 01:30 PM, Russell King - ARM Linux wrote:
> > On Tue, Oct 25, 2016 at 11:03:11AM +0200, Christian Borntraeger wrote:
> >> For spinning loops people do often use barrier() or cpu_relax().
> >> For most architectures cpu_relax and barrier are the same, but on
> >> some architectures cpu_relax can add some latency.
> >> For example on power,sparc64 and arc, cpu_relax can shift the CPU
> >> towards other hardware threads in an SMT environment.
> >> On s390 cpu_relax does even more, it uses an hypercall to the
> >> hypervisor to give up the timeslice.
> >> In contrast to the SMT yielding this can result in larger
latencies.
> >> In some places this latency is unwanted, so another variant
> >> "cpu_relax_lowlatency" was introduced. Before this is
used in more
> >> and more places, lets revert the logic and provide a
cpu_relax_yield
> >> that can be called in places where yielding is more important than
> >> latency. By default this is the same as cpu_relax on all
architectures.
> > 
> > Rather than having to update all these architectures in this way,
can't
> > we put in some linux/*.h header something like:
> > 
> > #ifndef cpu_relax_yield
> > #define cpu_relax_yield() cpu_relax()
> > #endif
> > 
> > so only those architectures that need to do something need to be
> > modified?
> 
> These patches are part of linux-next since a month or so, changing that 
> would invalidate all the next testing. If people want that, I can certainly
> do that, though.
It's three weeks since you posted them.  For one of those weeks (the
week you posted them) I was away, and missed them while catching up.
Sorry, but it sometimes takes a while to spot things amongst the
backlog, and normally takes some subsequent activity on the thread to
bring it back into view.

-- 
RMK's Patch system: http://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line: currently at 9.6Mbps down 400kbps up
according to speedtest.net.