Currently CACHE domain is not enabled on shared processor mode PowerVM
LPARS. On PowerVM systems, 'ibm,thread-group' device-tree property 2
under cpu-device-node indicates which all CPUs share L2-cache. However
'ibm,thread-group' device-tree property 2 is a relatively new property.
In absence of 'ibm,thread-group' property 2, 'l2-cache' device property
under cpu-device-node could help system to identify CPUs sharing L2-cache.
However this property is not exposed by PhyP in shared processor mode
configurations.

In absence of properties that inform OS about which CPUs share L2-cache,
fallback on core boundary.

Here are some stats from Power9 shared LPAR with the changes.

$ lscpu
Architecture:        ppc64le
Byte Order:          Little Endian
CPU(s):              32
On-line CPU(s) list: 0-31
Thread(s) per core:  8
Core(s) per socket:  1
Socket(s):           3
NUMA node(s):        2
Model:               2.2 (pvr 004e 0202)
Model name:          POWER9 (architected), altivec supported
Hypervisor vendor:   pHyp
Virtualization type: para
L1d cache:           32K
L1i cache:           32K
NUMA node0 CPU(s):   16-23
NUMA node1 CPU(s):   0-15,24-31
Physical sockets:    2
Physical chips:      1
Physical cores/chip: 10

Before patch
$ grep -r . /sys/kernel/debug/sched/domains/cpu0/domain*/name
Before
/sys/kernel/debug/sched/domains/cpu0/domain0/name:SMT
/sys/kernel/debug/sched/domains/cpu0/domain1/name:DIE
/sys/kernel/debug/sched/domains/cpu0/domain2/name:NUMA

After
/sys/kernel/debug/sched/domains/cpu0/domain0/name:SMT
/sys/kernel/debug/sched/domains/cpu0/domain1/name:CACHE
/sys/kernel/debug/sched/domains/cpu0/domain2/name:DIE
/sys/kernel/debug/sched/domains/cpu0/domain3/name:NUMA

$  awk '/domain/{print $1, $2}' /proc/schedstat | sort -u | sed -e 's/00000000,//g'
Before
domain0 00000055
domain0 000000aa
domain0 00005500
domain0 0000aa00
domain0 00550000
domain0 00aa0000
domain0 55000000
domain0 aa000000
domain1 00ff0000
domain1 ff00ffff
domain2 ffffffff

After
domain0 00000055
domain0 000000aa
domain0 00005500
domain0 0000aa00
domain0 00550000
domain0 00aa0000
domain0 55000000
domain0 aa000000
domain1 000000ff
domain1 0000ff00
domain1 00ff0000
domain1 ff000000
domain2 ff00ffff
domain2 ffffffff
domain3 ffffffff

(Lower is better)
perf stat -a -r 5 -n perf bench sched pipe  | tail -n 2
Before
           153.798 +- 0.142 seconds time elapsed  ( +-  0.09% )

After
           111.545 +- 0.652 seconds time elapsed  ( +-  0.58% )

which is an improvement of 27.47%
Signed-off-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Reviewed-by: Gautham R. Shenoy <ego@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210826100401.412519-4-srikar@linux.vnet.ibm.com