similar to: [PATCH/RFC 0/5] cpu_relax: introduce yield, remove lowlatency

Peter, here is v2 with some improved patch descriptions and some fixes. The previous version has survived one day of linux-next and I only changed small parts. So unless there is some other issue, feel free to pull (or to apply the patches) to tip/locking. The following changes since commit 07d9a380680d1c0eb51ef87ff2eab5c994949e69: Linux 4.9-rc2 (2016-10-23 17:10:14 -0700) are available in

Peter, here is v2 with some improved patch descriptions and some fixes. The previous version has survived one day of linux-next and I only changed small parts. So unless there is some other issue, feel free to pull (or to apply the patches) to tip/locking. The following changes since commit 07d9a380680d1c0eb51ef87ff2eab5c994949e69: Linux 4.9-rc2 (2016-10-23 17:10:14 -0700) are available in

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 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 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

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

On Fri, Oct 21, 2016 at 01:58:55PM +0200, Christian Borntraeger wrote: > stop_machine can take a very long time if the hypervisor does > overcommitment for guest CPUs. When waiting for "the one", lets > give up our CPU by using the new cpu_relax_yield. This seems something that would apply to most other virt stuff. Lets Cc a few more lists for that. > Signed-off-by:

On Fri, Oct 21, 2016 at 01:58:55PM +0200, Christian Borntraeger wrote: > stop_machine can take a very long time if the hypervisor does > overcommitment for guest CPUs. When waiting for "the one", lets > give up our CPU by using the new cpu_relax_yield. This seems something that would apply to most other virt stuff. Lets Cc a few more lists for that. > Signed-off-by:

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.

On 11/15/2016 02:37 PM, Russell King - ARM Linux wrote: > 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

No need to duplicate the same define everywhere. Since the only user is stop-machine and the only provider is s390, we can use a default implementation of cpu_relax_yield in sched.h. Suggested-by: Russell King <linux at armlinux.org.uk> Signed-off-by: Christian Borntraeger <borntraeger at de.ibm.com> --- arch/alpha/include/asm/processor.h | 1 - arch/arc/include/asm/processor.h

No need to duplicate the same define everywhere. Since the only user is stop-machine and the only provider is s390, we can use a default implementation of cpu_relax_yield in sched.h. Suggested-by: Russell King <linux at armlinux.org.uk> Signed-off-by: Christian Borntraeger <borntraeger at de.ibm.com> --- arch/alpha/include/asm/processor.h | 1 - arch/arc/include/asm/processor.h

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

stop_machine seemed to be the only important place for yielding during cpu_relax. This was fixed by using cpu_relax_yield. Therefore, we can now redefine cpu_relax to be a barrier instead on s390, making s390 identical to all other architectures. Signed-off-by: Christian Borntraeger <borntraeger at de.ibm.com> --- arch/s390/include/asm/processor.h | 2 +- 1 file changed, 1 insertion(+), 1

As there are no users left, we can remove cpu_relax_lowlatency. Signed-off-by: Christian Borntraeger <borntraeger at de.ibm.com> --- arch/alpha/include/asm/processor.h | 1 - arch/arc/include/asm/processor.h | 2 -- arch/arm/include/asm/processor.h | 1 - arch/arm64/include/asm/processor.h | 1 - arch/avr32/include/asm/processor.h | 1 -

On 10/22/2016 02:06 AM, Nicholas Piggin wrote: > On Fri, 21 Oct 2016 14:05:36 +0200 > Peter Zijlstra <peterz at infradead.org> wrote: > >> On Fri, Oct 21, 2016 at 01:58:55PM +0200, Christian Borntraeger wrote: >>> stop_machine can take a very long time if the hypervisor does >>> overcommitment for guest CPUs. When waiting for "the one", lets

On Fri, 21 Oct 2016 14:05:36 +0200 Peter Zijlstra <peterz at infradead.org> wrote: > On Fri, Oct 21, 2016 at 01:58:55PM +0200, Christian Borntraeger wrote: > > stop_machine can take a very long time if the hypervisor does > > overcommitment for guest CPUs. When waiting for "the one", lets > > give up our CPU by using the new cpu_relax_yield. > > This

Some time ago commit 57f2ffe14fd125c2 ("s390: remove diag 44 calls from cpu_relax()") did stop cpu_relax on s390 yielding to the hypervisor. As it turns out this made stop_machine run really slow on virtualized overcommited systems. For example the kprobes test during bootup took several seconds instead of just running unnoticed with large guests. Therefore, the yielding was

On 02/26/2016 09:42 AM, Jason Wang wrote: > This patch tries to poll for new added tx buffer or socket receive > queue for a while at the end of tx/rx processing. The maximum time > spent on polling were specified through a new kind of vring ioctl. > > Signed-off-by: Jason Wang <jasowang at redhat.com> > --- > drivers/vhost/net.c | 79

On 02/26/2016 09:42 AM, Jason Wang wrote: > This patch tries to poll for new added tx buffer or socket receive > queue for a while at the end of tx/rx processing. The maximum time > spent on polling were specified through a new kind of vring ioctl. > > Signed-off-by: Jason Wang <jasowang at redhat.com> > --- > drivers/vhost/net.c | 79