lkml.org 
[lkml]   [2015]   [May]   [29]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH] cpuidle: powernv/pseries: Decrease the snooze residency
Hi Shilpa,

The subject does not convey the purpose of this patch clearly IMO.
I would definitely suggest changing the subject to something like
"Auto promotion of snooze to deeper idle state" or similar.

On 05/29/2015 06:02 PM, Shilpasri G Bhat wrote:
> The idle cpus which stay in snooze for a long period can degrade the
> perfomance of the sibling cpus. If the cpu stays in snooze for more
> than target residency of the next available idle state, then exit from
> snooze. This gives a chance to the cpuidle governor to re-evaluate the
> last idle state of the cpu to promote it to deeper idle states.
>
> Signed-off-by: Shilpasri G Bhat <shilpa.bhat@linux.vnet.ibm.com>
> ---
> drivers/cpuidle/cpuidle-powernv.c | 12 ++++++++++++
> drivers/cpuidle/cpuidle-pseries.c | 11 +++++++++++
> 2 files changed, 23 insertions(+)
>
> diff --git a/drivers/cpuidle/cpuidle-pseries.c b/drivers/cpuidle/cpuidle-pseries.c
> index bb9e2b6..07135e0 100644
> --- a/drivers/cpuidle/cpuidle-pseries.c
> +++ b/drivers/cpuidle/cpuidle-pseries.c
> @@ -27,6 +27,8 @@ struct cpuidle_driver pseries_idle_driver = {
>
> static int max_idle_state;
> static struct cpuidle_state *cpuidle_state_table;
> +static u64 snooze_timeout;
> +static bool snooze_timeout_en;
>
> static inline void idle_loop_prolog(unsigned long *in_purr)
> {
> @@ -58,14 +60,18 @@ static int snooze_loop(struct cpuidle_device *dev,
> int index)
> {
> unsigned long in_purr;
> + u64 snooze_exit_time;
>
> idle_loop_prolog(&in_purr);
> local_irq_enable();
> set_thread_flag(TIF_POLLING_NRFLAG);
> + snooze_exit_time = get_tb() + snooze_timeout;
>
> while (!need_resched()) {
> HMT_low();
> HMT_very_low();
> + if (snooze_timeout_en && get_tb() > snooze_exit_time)
> + break;
> }
>
> HMT_medium();
> @@ -244,6 +250,11 @@ static int pseries_idle_probe(void)
> } else
> return -ENODEV;
>
> + if (max_idle_state > 1) {
> + snooze_timeout_en = true;
> + snooze_timeout = cpuidle_state_table[1].target_residency *
> + tb_ticks_per_usec;
> + }

Any idea why we don't have snooze defined on the shared lpar configuration ?

Regards
Preeti U Murthy
> return 0;
> }
>
>



\
 
 \ /
  Last update: 2015-05-29 16:01    [W:0.065 / U:0.040 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site