cpu_smt_mask tracks topology_sibling_cpumask. This would be good for
most architectures. One of the users of cpu_smt_mask(), would be to
identify idle-cores. On Power9, a pair of SMT4 cores can be presented by
the firmware as a SMT8 core for backward compatibility reasons.
Powerpc allows LPARs to be live migrated from Power8 to Power9. Do note
Power8 had only SMT8 cores. Existing software which has been
developed/configured for Power8 would expect to see SMT8 core.
Maintaining the illusion of SMT8 core is a requirement to make that
work.
In order to maintain above userspace backward compatibility with
previous versions of processor, Power9 onwards there is option to the
firmware to advertise a pair of SMT4 cores as a fused cores aka SMT8
core. On Power9 this pair shares the L2 cache as well. However, from the
scheduler's point of view, a core should be determined by SMT4, since
its a completely independent unit of compute. Hence allow PowerPc
architecture to override the default cpu_smt_mask() to point to the SMT4
cores in a SMT8 mode.
This will ensure the scheduler is always given the right information.
Cc: linuxppc-dev <[email protected]>
Cc: LKML <[email protected]>
Cc: Michael Ellerman <[email protected]>
Cc: Michael Neuling <[email protected]>
Cc: Gautham R Shenoy <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Valentin Schneider <[email protected]>
Cc: Dietmar Eggemann <[email protected]>
Cc: Mel Gorman <[email protected]>
Cc: Vincent Guittot <[email protected]>
Cc: Vaidyanathan Srinivasan <[email protected]>
Acked-by; Peter Zijlstra (Intel) <[email protected]>
Signed-off-by: Srikar Dronamraju <[email protected]>
---
Changelog v1->v2:
Update the commit msg based on the discussion in community esp
with Peter Zijlstra and Michael Ellerman.
include/linux/topology.h | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/include/linux/topology.h b/include/linux/topology.h
index 608fa4aadf0e..ad03df1cc266 100644
--- a/include/linux/topology.h
+++ b/include/linux/topology.h
@@ -198,7 +198,7 @@ static inline int cpu_to_mem(int cpu)
#define topology_die_cpumask(cpu) cpumask_of(cpu)
#endif
-#ifdef CONFIG_SCHED_SMT
+#if defined(CONFIG_SCHED_SMT) && !defined(cpu_smt_mask)
static inline const struct cpumask *cpu_smt_mask(int cpu)
{
return topology_sibling_cpumask(cpu);
--
2.18.2
On Power9, a pair of SMT4 cores can be presented by the firmware as a SMT8
core for backward compatibility reasons, with the fusion of two SMT4 cores.
Powerpc allows LPARs to be live migrated from Power8 to Power9. Existing
software developed/configured for Power8, expects to see a SMT8 core.
In order to maintain userspace backward compatibility (with Power8 chips in
case of Power9) in enterprise Linux systems, the topology_sibling_cpumask
has to be set to SMT8 core.
cpu_smt_mask() should generally point to the cpu mask of the SMT4 core.
Hence override the default cpu_smt_mask() to be powerpc specific
allowing for better scheduling behaviour on Power.
schbench
(latency measured in usecs, so lesser is better)
Without patch With patch
Latency percentiles (usec) Latency percentiles (usec)
50.0000th: 34 50.0000th: 38
75.0000th: 47 75.0000th: 52
90.0000th: 54 90.0000th: 60
95.0000th: 57 95.0000th: 64
*99.0000th: 62 *99.0000th: 72
99.5000th: 65 99.5000th: 75
99.9000th: 76 99.9000th: 3452
min=0, max=9205 min=0, max=9344
schbench (With Cede disabled)
Without patch With patch
Latency percentiles (usec) Latency percentiles (usec)
50.0000th: 20 50.0000th: 21
75.0000th: 28 75.0000th: 29
90.0000th: 33 90.0000th: 34
95.0000th: 35 95.0000th: 37
*99.0000th: 40 *99.0000th: 40
99.5000th: 48 99.5000th: 42
99.9000th: 94 99.9000th: 79
min=0, max=791 min=0, max=791
perf bench sched pipe
usec/ops : lesser is better
Without patch
N Min Max Median Avg Stddev
101 5.095113 5.595269 5.204842 5.2298776 0.10762713
5.10 - 5.15 : ################################################## 23% (24)
5.15 - 5.20 : ############################################# 21% (22)
5.20 - 5.25 : ################################################## 23% (24)
5.25 - 5.30 : ######################### 11% (12)
5.30 - 5.35 : ########## 4% (5)
5.35 - 5.40 : ######## 3% (4)
5.40 - 5.45 : ######## 3% (4)
5.45 - 5.50 : #### 1% (2)
5.50 - 5.55 : ## 0% (1)
5.55 - 5.60 : #### 1% (2)
With patch
N Min Max Median Avg Stddev
101 5.134675 8.524719 5.207658 5.2780985 0.34911969
5.1 - 5.5 : ################################################## 94% (95)
5.5 - 5.8 : ## 3% (4)
5.8 - 6.2 : 0% (1)
6.2 - 6.5 :
6.5 - 6.8 :
6.8 - 7.2 :
7.2 - 7.5 :
7.5 - 7.8 :
7.8 - 8.2 :
8.2 - 8.5 :
perf bench sched pipe (cede disabled)
usec/ops : lesser is better
Without patch
N Min Max Median Avg Stddev
101 7.884227 12.576538 7.956474 8.0170722 0.46159054
7.9 - 8.4 : ################################################## 99% (100)
8.4 - 8.8 :
8.8 - 9.3 :
9.3 - 9.8 :
9.8 - 10.2 :
10.2 - 10.7 :
10.7 - 11.2 :
11.2 - 11.6 :
11.6 - 12.1 :
12.1 - 12.6 :
With patch
N Min Max Median Avg Stddev
101 7.956021 8.217284 8.015615 8.0283866 0.049844967
7.96 - 7.98 : ###################### 12% (13)
7.98 - 8.01 : ################################################## 28% (29)
8.01 - 8.03 : #################################### 20% (21)
8.03 - 8.06 : ######################### 14% (15)
8.06 - 8.09 : ###################### 12% (13)
8.09 - 8.11 : ###### 3% (4)
8.11 - 8.14 : ### 1% (2)
8.14 - 8.17 : ### 1% (2)
8.17 - 8.19 :
8.19 - 8.22 : # 0% (1)
Observations: With the patch, the initial run/iteration takes a slight
longer time. This can be attributed to the fact that now we pick a CPU
from a idle core which could be sleep mode. Once we remove the cede,
state the numbers improve in favour of the patch.
ebizzy:
transactions per second (higher is better)
without patch
N Min Max Median Avg Stddev
100 1018433 1304470 1193208 1182315.7 60018.733
1018433 - 1047037 : ###### 3% (3)
1047037 - 1075640 : ######## 4% (4)
1075640 - 1104244 : ######## 4% (4)
1104244 - 1132848 : ############### 7% (7)
1132848 - 1161452 : #################################### 17% (17)
1161452 - 1190055 : ########################## 12% (12)
1190055 - 1218659 : ############################################# 21% (21)
1218659 - 1247263 : ################################################## 23% (23)
1247263 - 1275866 : ######## 4% (4)
1275866 - 1304470 : ######## 4% (4)
with patch
N Min Max Median Avg Stddev
100 967014 1292938 1208819 1185281.8 69815.851
967014 - 999606 : ## 1% (1)
999606 - 1032199 : ## 1% (1)
1032199 - 1064791 : ############ 6% (6)
1064791 - 1097384 : ########## 5% (5)
1097384 - 1129976 : ################## 9% (9)
1129976 - 1162568 : #################### 10% (10)
1162568 - 1195161 : ########################## 13% (13)
1195161 - 1227753 : ############################################ 22% (22)
1227753 - 1260346 : ################################################## 25% (25)
1260346 - 1292938 : ############## 7% (7)
Observations: Not much changes, ebizzy is not much impacted.
Cc: linuxppc-dev <[email protected]>
Cc: LKML <[email protected]>
Cc: Michael Ellerman <[email protected]>
Cc: Michael Neuling <[email protected]>
Cc: Gautham R Shenoy <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Valentin Schneider <[email protected]>
Cc: Dietmar Eggemann <[email protected]>
Cc: Mel Gorman <[email protected]>
Cc: Vincent Guittot <[email protected]>
Cc: Vaidyanathan Srinivasan <[email protected]>
Signed-off-by: Srikar Dronamraju <[email protected]>
---
Changelog v1->v2:
Modified commit msg as per mailing list discussion.
Added performance numbers
arch/powerpc/include/asm/cputhreads.h | 1 -
arch/powerpc/include/asm/smp.h | 13 +++++++++++++
2 files changed, 13 insertions(+), 1 deletion(-)
diff --git a/arch/powerpc/include/asm/cputhreads.h b/arch/powerpc/include/asm/cputhreads.h
index deb99fd6e060..98c8bd155bf9 100644
--- a/arch/powerpc/include/asm/cputhreads.h
+++ b/arch/powerpc/include/asm/cputhreads.h
@@ -23,7 +23,6 @@
extern int threads_per_core;
extern int threads_per_subcore;
extern int threads_shift;
-extern bool has_big_cores;
extern cpumask_t threads_core_mask;
#else
#define threads_per_core 1
diff --git a/arch/powerpc/include/asm/smp.h b/arch/powerpc/include/asm/smp.h
index 9cd0765633c5..bb06aa875131 100644
--- a/arch/powerpc/include/asm/smp.h
+++ b/arch/powerpc/include/asm/smp.h
@@ -131,6 +131,19 @@ static inline struct cpumask *cpu_smallcore_mask(int cpu)
extern int cpu_to_core_id(int cpu);
+extern bool has_big_cores;
+
+#define cpu_smt_mask cpu_smt_mask
+#ifdef CONFIG_SCHED_SMT
+static inline const struct cpumask *cpu_smt_mask(int cpu)
+{
+ if (has_big_cores)
+ return per_cpu(cpu_smallcore_map, cpu);
+
+ return per_cpu(cpu_sibling_map, cpu);
+}
+#endif /* CONFIG_SCHED_SMT */
+
/* Since OpenPIC has only 4 IPIs, we use slightly different message numbers.
*
* Make sure this matches openpic_request_IPIs in open_pic.c, or what shows up
--
2.18.2
On Fri, 7 Aug 2020 13:15:16 +0530, Srikar Dronamraju wrote:
> cpu_smt_mask tracks topology_sibling_cpumask. This would be good for
> most architectures. One of the users of cpu_smt_mask(), would be to
> identify idle-cores. On Power9, a pair of SMT4 cores can be presented by
> the firmware as a SMT8 core for backward compatibility reasons.
>
> Powerpc allows LPARs to be live migrated from Power8 to Power9. Do note
> Power8 had only SMT8 cores. Existing software which has been
> developed/configured for Power8 would expect to see SMT8 core.
> Maintaining the illusion of SMT8 core is a requirement to make that
> work.
>
> [...]
Applied to powerpc/next.
[1/2] sched/topology: Allow archs to override cpu_smt_mask
https://git.kernel.org/powerpc/c/3babbe447d76ac2919ec4d0eb3b0adfb22f5b03c
[2/2] powerpc/topology: Override cpu_smt_mask
https://git.kernel.org/powerpc/c/f3232321db58480804f80d59aeb651a5c859a200
cheers