[RFC v2 PATCH 00/11] change scheduler domain hierarchy set-up
From: dietmar . eggemann
Date: Mon Jan 20 2014 - 07:40:51 EST
From: Dietmar Eggemann <dietmar.eggemann@xxxxxxx>
This patch-set cleans up the scheduler domain (sd) level initialisation
code. It is based on the idea of Peter Zijlstra to use a single sd
init function sketched here: https://lkml.org/lkml/2013/11/5/239
The sd level data for conventional (SMT, MC, BOOK, CPU) sd levels which
can be set by the arch comprises the cpu mask and the sd topology flags
for this level.
An arch can either use the existing sd level hierarchy defined in the
scheduler core or specify its own complete sd level hierarchy in its
topology shim layer.
This clean-up is necessary because some of the arch specific settings
are just bitrot or accidents. So far, additional arch specific sd
topology flags are set via flag specific weak functions, like
SD_ASYM_PACKING. If in the future, we want to further distinguish the
functionality of the different sd levels based on more topology
information (e.g. Vincent's SD_SHARE_POWERDOMAIN flag) we do need a
better interface than this.
Changes in v2:
1) Export struct sched_domain_topology_info and set_sd_topology_info()
towards the arch to enable it to customize sched domain topology
2) Keep a default topology setup inside the scheduler for the
conventional sched domain levels.
3) Sched domain level topology flags can be specified dependent on
runtime information (e.g. cpu feature) and per cpu.
4) Move struct sched_domain from interface header file into private
scheduler header file for data encapsulation purposes. With the
deletion of the arch specific sd init macros and the sd ptr arg in
arch_scale_freq_power(), struct sched_domain is no longer needed
outside the scheduler core code.
What does the patch-set try to achieve:
1) Let the arch define the conventional (SMT, MC, BOOK, CPU) sched
domain hierarchy. The arch specifies per sched domain level the
pointer to the getter function of the corresponding cpu mask and
the topology flags as well as the name of the sched domain level,
the latter for debug purposes.
2) Unify the set-up code for conventional and NUMA scheduler domains.
All scheduler domain topology levels are now allocated in the same
function sched_init_topology(). All scheduler domains now use a
common init function sd_init() which makes the existing SD_FOO_INIT
3) The arch is no longer limited to the default sched domain levels
(SMT, MC, BOOK, CPU) but can easily define its own sched domain level
4) Prepare the mechanics to make it easier to integrate the provision of
additional topology related data (e.g. energy information) to the
1) The patch-set has only been tested on ARM TC2 (2 clusters, one with 2
Cortex A15 and the other with 3 Cortex A7) and on an Intel i5-520M (2
cores with 2 threads each) platform.
2) For other archs it has only been compile tested for certain
configurations (powerpc: chroma_defconfig, mips: ip27_defconfig,
s390: defconfig, tile: tilegx_defconfig).
The patch-set is against v3.13-rc8.
Dietmar Eggemann (11):
sched: define sched_domain_topology_info
sched: export cpu_smt_mask() and cpu_cpu_mask()
sched: define TOPOLOGY_SD_FLAGS
sched: replace SD_INIT_FUNC with sd_init()
sched: add a name to sched_domain_topology_info
sched: delete redundant sd init macros
sched: consolidate sched_init_numa() and sched_init_conv()
sched: introduce a func ptr for sd topology flags
sched: provide SD_ASYM_PACKING via topology info table
sched: delete sd ptr arg in arch_scale_freq_power()
sched: un-export struct sched_domain
arch/arm/kernel/topology.c | 4 +-
arch/ia64/include/asm/topology.h | 24 ---
arch/metag/include/asm/topology.h | 27 ---
arch/powerpc/kernel/smp.c | 34 +++-
arch/s390/include/asm/topology.h | 2 -
arch/tile/include/asm/topology.h | 33 ----
include/linux/sched.h | 119 ++++---------
include/linux/topology.h | 127 ++------------
kernel/sched/core.c | 351 ++++++++++++++++++-------------------
kernel/sched/fair.c | 10 +-
kernel/sched/sched.h | 169 ++++++++++++++----
11 files changed, 384 insertions(+), 516 deletions(-)
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/