From: "Gautham R. Shenoy" <[email protected]>
Hi,
This is a v2 of the patch series to parse the extended CEDE
information in the pseries-cpuidle driver.
The v1 of this patchset can be found here :
https://lore.kernel.org/linuxppc-dev/[email protected]/
The change from v1 --> v2 :
* Dropped Patches 4 and 5 which would expose extended idle-states,
that wakeup on external interrupts, to cpuidle framework. These
were RFC patches in v1. Dropped them because currently the only
extended CEDE state that wakesup on external interrupts is CEDE(1)
which adds no signifcant value over CEDE(0).
* Rebased the patches onto powerpc/merge.
* No changes in code for Patches 1-3.
Motivation:
===========
On pseries Dedicated Linux LPARs, apart from the polling snooze idle
state, we currently have the CEDE idle state which cedes the CPU to
the hypervisor with latency-hint = 0.
However, the PowerVM hypervisor supports additional extended CEDE
states, which can be queried through the "ibm,get-systems-parameter"
rtas-call with the CEDE_LATENCY_TOKEN. The hypervisor maps these
extended CEDE states to appropriate platform idle-states in order to
provide energy-savings as well as shifting power to the active
units. On existing pseries LPARs today we have extended CEDE with
latency-hints {1,2} supported.
The patches in this patchset, adds code to parse the CEDE latency
records provided by the hypervisor. We use this information to
determine the wakeup latency of the regular CEDE (which we have been
so far hardcoding to 10us while experimentally it is much lesser ~
1us), by looking at the wakeup latency provided by the hypervisor for
Extended CEDE states. Since the platform currently advertises Extended
CEDE 1 to have wakeup latency of 2us, we can be sure that the wakeup
latency of the regular CEDE is no more than this.
With Patches 1-3, we see an improvement in the single-threaded
performance on ebizzy.
2 ebizzy threads bound to the same big-core. 25% improvement in the
avg records/s (higher the better) with patches 1-3.
x without_patches
* with_patches
N Min Max Median Avg Stddev
x 10 2491089 5834307 5398375 4244335 1596244.9
* 10 2893813 5834474 5832448 5327281.3 1055941.4
We do not observe any major regression in either the context_switch2
benchmark or the schbench benchmark
context_switch2 across CPU0 CPU1 (Both belong to same big-core, but different
small cores). We observe a minor 0.14% regression in the number of
context-switches (higher is better).
x without_patch
* with_patch
N Min Max Median Avg Stddev
x 500 348872 362236 354712 354745.69 2711.827
* 500 349422 361452 353942 354215.4 2576.9258
context_switch2 across CPU0 CPU8 (Different big-cores). We observe a 0.37%
improvement in the number of context-switches (higher is better).
x without_patch
* with_patch
N Min Max Median Avg Stddev
x 500 287956 294940 288896 288977.23 646.59295
* 500 288300 294646 289582 290064.76 1161.9992
schbench:
No major difference could be seen until the 99.9th percentile.
Without-patch
Latency percentiles (usec)
50.0th: 29
75.0th: 39
90.0th: 49
95.0th: 59
*99.0th: 13104
99.5th: 14672
99.9th: 15824
min=0, max=17993
With-patch:
Latency percentiles (usec)
50.0th: 29
75.0th: 40
90.0th: 50
95.0th: 61
*99.0th: 13648
99.5th: 14768
99.9th: 15664
min=0, max=29812
Gautham R. Shenoy (3):
cpuidle-pseries: Set the latency-hint before entering CEDE
cpuidle-pseries: Add function to parse extended CEDE records
cpuidle-pseries : Fixup exit latency for CEDE(0)
drivers/cpuidle/cpuidle-pseries.c | 167 +++++++++++++++++++++++++++++++++++++-
1 file changed, 165 insertions(+), 2 deletions(-)
--
1.9.4
From: "Gautham R. Shenoy" <[email protected]>
We are currently assuming that CEDE(0) has exit latency 10us, since
there is no way for us to query from the platform. However, if the
wakeup latency of an Extended CEDE state is smaller than 10us, then we
can be sure that the exit latency of CEDE(0) cannot be more than that.
that.
In this patch, we fix the exit latency of CEDE(0) if we discover an
Extended CEDE state with wakeup latency smaller than 10us.
Benchmark results:
ebizzy:
2 ebizzy threads bound to the same big-core. 25% improvement in the
avg records/s with patch.
x without_patch
+ with_patch
N Min Max Median Avg Stddev
x 10 2491089 5834307 5398375 4244335 1596244.9
+ 10 2893813 5834474 5832448 5327281.3 1055941.4
context_switch2 :
There is no major regression observed with this patch as seen from the
context_switch2 benchmark.
context_switch2 across CPU0 CPU1 (Both belong to same big-core, but different
small cores). We observe a minor 0.14% regression in the number of
context-switches (higher is better).
x without_patch
+ with_patch
N Min Max Median Avg Stddev
x 500 348872 362236 354712 354745.69 2711.827
+ 500 349422 361452 353942 354215.4 2576.9258
Difference at 99.0% confidence
-530.288 +/- 430.963
-0.149484% +/- 0.121485%
(Student's t, pooled s = 2645.24)
context_switch2 across CPU0 CPU8 (Different big-cores). We observe a 0.37%
improvement in the number of context-switches (higher is better).
x without_patch
+ with_patch
N Min Max Median Avg Stddev
x 500 287956 294940 288896 288977.23 646.59295
+ 500 288300 294646 289582 290064.76 1161.9992
Difference at 99.0% confidence
1087.53 +/- 153.194
0.376337% +/- 0.0530125%
(Student's t, pooled s = 940.299)
schbench:
No major difference could be seen until the 99.9th percentile.
Without-patch
Latency percentiles (usec)
50.0th: 29
75.0th: 39
90.0th: 49
95.0th: 59
*99.0th: 13104
99.5th: 14672
99.9th: 15824
min=0, max=17993
With-patch:
Latency percentiles (usec)
50.0th: 29
75.0th: 40
90.0th: 50
95.0th: 61
*99.0th: 13648
99.5th: 14768
99.9th: 15664
min=0, max=29812
Reviewed-by: Vaidyanathan Srinivasan <[email protected]>
Signed-off-by: Gautham R. Shenoy <[email protected]>
---
drivers/cpuidle/cpuidle-pseries.c | 34 ++++++++++++++++++++++++++++++++--
1 file changed, 32 insertions(+), 2 deletions(-)
diff --git a/drivers/cpuidle/cpuidle-pseries.c b/drivers/cpuidle/cpuidle-pseries.c
index b1dc24d..0b2f115 100644
--- a/drivers/cpuidle/cpuidle-pseries.c
+++ b/drivers/cpuidle/cpuidle-pseries.c
@@ -334,12 +334,42 @@ static int pseries_cpuidle_driver_init(void)
static int add_pseries_idle_states(void)
{
int nr_states = 2; /* By default we have snooze, CEDE */
+ int i;
+ u64 min_latency_us = dedicated_states[1].exit_latency; /* CEDE latency */
if (parse_cede_parameters())
return nr_states;
- pr_info("cpuidle : Skipping the %d Extended CEDE idle states\n",
- nr_xcede_records);
+ for (i = 0; i < nr_xcede_records; i++) {
+ u64 latency_tb = xcede_records[i].wakeup_latency_tb_ticks;
+ u64 latency_us = tb_to_ns(latency_tb) / NSEC_PER_USEC;
+
+ if (latency_us < min_latency_us)
+ min_latency_us = latency_us;
+ }
+
+ /*
+ * We are currently assuming that CEDE(0) has exit latency
+ * 10us, since there is no way for us to query from the
+ * platform.
+ *
+ * However, if the wakeup latency of an Extended CEDE state is
+ * smaller than 10us, then we can be sure that CEDE(0)
+ * requires no more than that.
+ *
+ * Perform the fix-up.
+ */
+ if (min_latency_us < dedicated_states[1].exit_latency) {
+ u64 cede0_latency = min_latency_us - 1;
+
+ if (cede0_latency <= 0)
+ cede0_latency = min_latency_us;
+
+ dedicated_states[1].exit_latency = cede0_latency;
+ dedicated_states[1].target_residency = 10 * (cede0_latency);
+ pr_info("cpuidle : Fixed up CEDE exit latency to %llu us\n",
+ cede0_latency);
+ }
return nr_states;
}
--
1.9.4