kvm.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* [patch 0/5] cpuidle haltpoll driver and governor (v4)
@ 2019-06-13 22:45 Marcelo Tosatti
  2019-06-13 22:45 ` [patch 1/5] drivers/cpuidle: add cpuidle-haltpoll driver Marcelo Tosatti
                   ` (4 more replies)
  0 siblings, 5 replies; 14+ messages in thread
From: Marcelo Tosatti @ 2019-06-13 22:45 UTC (permalink / raw)
  To: kvm-devel
  Cc: Paolo Bonzini, Radim Krcmar, Andrea Arcangeli, Rafael J. Wysocki,
	Peter Zijlstra, Wanpeng Li, Konrad Rzeszutek Wilk,
	Raslan KarimAllah, Boris Ostrovsky, Ankur Arora,
	Christian Borntraeger, linux-pm

The cpuidle-haltpoll driver with haltpoll governor allows the guest
vcpus to poll for a specified amount of time before halting. 
This provides the following benefits to host side polling:

         1) The POLL flag is set while polling is performed, which allows
            a remote vCPU to avoid sending an IPI (and the associated
            cost of handling the IPI) when performing a wakeup.

         2) The VM-exit cost can be avoided.

The downside of guest side polling is that polling is performed
even with other runnable tasks in the host.

Results comparing halt_poll_ns and server/client application
where a small packet is ping-ponged:

host                                        --> 31.33
halt_poll_ns=300000 / no guest busy spin    --> 33.40   (93.8%)
halt_poll_ns=0 / guest_halt_poll_ns=300000  --> 32.73   (95.7%)

For the SAP HANA benchmarks (where idle_spin is a parameter
of the previous version of the patch, results should be the
same):

hpns == halt_poll_ns

                           idle_spin=0/   idle_spin=800/    idle_spin=0/
                           hpns=200000    hpns=0            hpns=800000
DeleteC06T03 (100 thread) 1.76           1.71 (-3%)        1.78   (+1%)
InsertC16T02 (100 thread) 2.14           2.07 (-3%)        2.18   (+1.8%)
DeleteC00T01 (1 thread)   1.34           1.28 (-4.5%)      1.29   (-3.7%)
UpdateC00T03 (1 thread)   4.72           4.18 (-12%)       4.53   (-5%)

V2:

- Move from x86 to generic code (Paolo/Christian)
- Add auto-tuning logic (Paolo)
- Add MSR to disable host side polling (Paolo)

V3:

- Do not be specific about HLT VM-exit in the documentation (Ankur Arora)
- Mark tuning parameters static and __read_mostly (Andrea Arcangeli)
- Add WARN_ON if host does not support poll control (Joao Martins)
- Use sched_clock and cleanup haltpoll_enter_idle (Peter Zijlstra)
- Mark certain functions in kvm.c as static (kernel test robot)
- Remove tracepoints as they use RCU from extended quiescent state (kernel
test robot)

V4:
- Use a haltpoll governor, use poll_state.c poll code (Rafael J. Wysocki)




^ permalink raw reply	[flat|nested] 14+ messages in thread
* [patch 0/5] cpuidle haltpoll driver and governor (v5)
@ 2019-07-01 18:53 Marcelo Tosatti
  2019-07-01 18:53 ` [patch 5/5] cpuidle-haltpoll: disable host side polling when kvm virtualized Marcelo Tosatti
  0 siblings, 1 reply; 14+ messages in thread
From: Marcelo Tosatti @ 2019-07-01 18:53 UTC (permalink / raw)
  To: kvm, linux-pm
  Cc: Paolo Bonzini, Radim Krcmar, Andrea Arcangeli, Rafael J. Wysocki,
	Peter Zijlstra, Wanpeng Li, Konrad Rzeszutek Wilk,
	Raslan KarimAllah, Boris Ostrovsky, Ankur Arora,
	Christian Borntraeger

The cpuidle-haltpoll driver with haltpoll governor allows the guest
vcpus to poll for a specified amount of time before halting.
This provides the following benefits to host side polling:

         1) The POLL flag is set while polling is performed, which allows
            a remote vCPU to avoid sending an IPI (and the associated
            cost of handling the IPI) when performing a wakeup.

         2) The VM-exit cost can be avoided.

The downside of guest side polling is that polling is performed
even with other runnable tasks in the host.

Results comparing halt_poll_ns and server/client application
where a small packet is ping-ponged:

host                                        --> 31.33
halt_poll_ns=300000 / no guest busy spin    --> 33.40   (93.8%)
halt_poll_ns=0 / guest_halt_poll_ns=300000  --> 32.73   (95.7%)

For the SAP HANA benchmarks (where idle_spin is a parameter
of the previous version of the patch, results should be the
same):

hpns == halt_poll_ns

                           idle_spin=0/   idle_spin=800/    idle_spin=0/
                           hpns=200000    hpns=0            hpns=800000
DeleteC06T03 (100 thread) 1.76           1.71 (-3%)        1.78   (+1%)
InsertC16T02 (100 thread) 2.14           2.07 (-3%)        2.18   (+1.8%)
DeleteC00T01 (1 thread)   1.34           1.28 (-4.5%)	   1.29   (-3.7%)
UpdateC00T03 (1 thread)   4.72           4.18 (-12%)	   4.53   (-5%)

V2:

- Move from x86 to generic code (Paolo/Christian)
- Add auto-tuning logic (Paolo)
- Add MSR to disable host side polling (Paolo)

V3:

- Do not be specific about HLT VM-exit in the documentation (Ankur Arora)
- Mark tuning parameters static and __read_mostly (Andrea Arcangeli)
- Add WARN_ON if host does not support poll control (Joao Martins)
- Use sched_clock and cleanup haltpoll_enter_idle (Peter Zijlstra)
- Mark certain functions in kvm.c as static (kernel test robot)
- Remove tracepoints as they use RCU from extended quiescent state (kernel
test robot)

V4:
- Use a haltpoll governor, use poll_state.c poll code (Rafael J. Wysocki)

V5:
- Take latency requirement into consideration (Rafael J. Wysocki)
- Set target_residency/exit_latency to 1 (Rafael J. Wysocki)
- Do not load cpuidle driver if not virtualized (Rafael J. Wysocki)



^ permalink raw reply	[flat|nested] 14+ messages in thread
* [patch 0/5] cpuidle haltpoll driver and governor (v6)
@ 2019-07-03 23:51 Marcelo Tosatti
  2019-07-03 23:51 ` [patch 5/5] cpuidle-haltpoll: disable host side polling when kvm virtualized Marcelo Tosatti
  0 siblings, 1 reply; 14+ messages in thread
From: Marcelo Tosatti @ 2019-07-03 23:51 UTC (permalink / raw)
  To: kvm-devel
  Cc: Paolo Bonzini, Radim Krcmar, Andrea Arcangeli, Rafael J. Wysocki,
	Peter Zijlstra, Wanpeng Li, Konrad Rzeszutek Wilk,
	Raslan KarimAllah, Boris Ostrovsky, Ankur Arora,
	Christian Borntraeger, linux-pm

(rebased against queue branch of kvm.git tree)

The cpuidle-haltpoll driver with haltpoll governor allows the guest
vcpus to poll for a specified amount of time before halting.
This provides the following benefits to host side polling:

         1) The POLL flag is set while polling is performed, which allows
            a remote vCPU to avoid sending an IPI (and the associated
            cost of handling the IPI) when performing a wakeup.

         2) The VM-exit cost can be avoided.

The downside of guest side polling is that polling is performed
even with other runnable tasks in the host.

Results comparing halt_poll_ns and server/client application
where a small packet is ping-ponged:

host                                        --> 31.33
halt_poll_ns=300000 / no guest busy spin    --> 33.40   (93.8%)
halt_poll_ns=0 / guest_halt_poll_ns=300000  --> 32.73   (95.7%)

For the SAP HANA benchmarks (where idle_spin is a parameter
of the previous version of the patch, results should be the
same):

hpns == halt_poll_ns

                           idle_spin=0/   idle_spin=800/    idle_spin=0/
                           hpns=200000    hpns=0            hpns=800000
DeleteC06T03 (100 thread) 1.76           1.71 (-3%)        1.78   (+1%)
InsertC16T02 (100 thread) 2.14           2.07 (-3%)        2.18   (+1.8%)
DeleteC00T01 (1 thread)   1.34           1.28 (-4.5%)	   1.29   (-3.7%)
UpdateC00T03 (1 thread)   4.72           4.18 (-12%)	   4.53   (-5%)

V2:

- Move from x86 to generic code (Paolo/Christian)
- Add auto-tuning logic (Paolo)
- Add MSR to disable host side polling (Paolo)

V3:

- Do not be specific about HLT VM-exit in the documentation (Ankur Arora)
- Mark tuning parameters static and __read_mostly (Andrea Arcangeli)
- Add WARN_ON if host does not support poll control (Joao Martins)
- Use sched_clock and cleanup haltpoll_enter_idle (Peter Zijlstra)
- Mark certain functions in kvm.c as static (kernel test robot)
- Remove tracepoints as they use RCU from extended quiescent state (kernel
test robot)

V4:
- Use a haltpoll governor, use poll_state.c poll code (Rafael J. Wysocki)

V5:
- Take latency requirement into consideration (Rafael J. Wysocki)
- Set target_residency/exit_latency to 1 (Rafael J. Wysocki)
- Do not load cpuidle driver if not virtualized (Rafael J. Wysocki)

V6:
- Switch from callback to poll_limit_ns variable in cpuidle device structure
(Rafael J. Wysocki)
- Move last_used_idx to cpuidle device structure (Rafael J. Wysocki)
- Drop per-cpu device structure in haltpoll governor (Rafael J. Wysocki)



^ permalink raw reply	[flat|nested] 14+ messages in thread

end of thread, other threads:[~2019-07-03 23:59 UTC | newest]

Thread overview: 14+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2019-06-13 22:45 [patch 0/5] cpuidle haltpoll driver and governor (v4) Marcelo Tosatti
2019-06-13 22:45 ` [patch 1/5] drivers/cpuidle: add cpuidle-haltpoll driver Marcelo Tosatti
2019-06-25 21:50   ` Rafael J. Wysocki
2019-06-26  8:40   ` Rafael J. Wysocki
2019-06-13 22:45 ` [patch 2/5] cpuidle: add get_poll_time callback Marcelo Tosatti
2019-06-25 21:52   ` Rafael J. Wysocki
2019-06-13 22:45 ` [patch 3/5] cpuidle: add haltpoll governor Marcelo Tosatti
2019-06-21  0:34   ` Doug Smythies
2019-06-27 18:08     ` Marcelo Tosatti
2019-06-25 21:57   ` Rafael J. Wysocki
2019-06-13 22:45 ` [patch 4/5] kvm: x86: add host poll control msrs Marcelo Tosatti
2019-06-13 22:45 ` [patch 5/5] cpuidle-haltpoll: disable host side polling when kvm virtualized Marcelo Tosatti
2019-07-01 18:53 [patch 0/5] cpuidle haltpoll driver and governor (v5) Marcelo Tosatti
2019-07-01 18:53 ` [patch 5/5] cpuidle-haltpoll: disable host side polling when kvm virtualized Marcelo Tosatti
2019-07-03 23:51 [patch 0/5] cpuidle haltpoll driver and governor (v6) Marcelo Tosatti
2019-07-03 23:51 ` [patch 5/5] cpuidle-haltpoll: disable host side polling when kvm virtualized Marcelo Tosatti

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).