Skip to contentSkip to conversation

sched: Instrument sched domain flags

PATCH Superseded Valentin Schneider 4 participants 27 messages Aug 12, 2020 – Aug 13, 2020

Subsystems (2)

v5 of 6 Superseded Latest is v6 →

lore
Share & tools
lore
#0 lore raw

v5 cover sched: Instrument sched domain flags

Hi,

I've repeatedly stared at an SD flag and asked myself "how should that be set up in the domain hierarchy anyway?". I figured that if we formalize our flags zoology a bit, we could also do some runtime assertions on them - this is what this series is all about.

Patches
=======

The idea is to associate the flags with metaflags that describes how they should be set in a sched domain hierarchy ("if this SD has it, all its {parents, children} have it") or how they behave wrt degeneration - details are in the comments and commit logs.

The good thing is that the debugging bits go away when CONFIG_SCHED_DEBUG isn't set. The bad thing is that this replaces SD_* flags definitions with some unsavoury macros. This is mainly because I wanted to avoid having to duplicate work between declaring the flags and declaring their metaflags. Conceptually they are pretty close to the macros used for SCHED_FEAT.

o Patches 1-2 remove a derelict flag and align the arm scheduler topology with arm64's
o Patches 3-8 instrument SD flags with metadata and add assertions
o Patches 9-10 are additional topology cleanups
o Patches 11-16 each add a new flag to SD_DEGENERATE_GROUPS_MASK
o Patch 17 leverage the previous 6 patches to further factorize domain degeneration

Revisions
=========

v4 -> v5
--------

The final git diff between v4 and v5 isn't too big; there is no diff on the domain degeneration side of things, it has just been split up in more patches.

o Shuffled the series around to facilitate bisection (Ingo)
I kept the arm bits at the start because that was a bit simpler, and in the unlikely case that needs to be reverted it won't be too hard.
o Split the degeneration mask tweak into individual commits per new flag (Ingo)
o Collected Reviewed-by from Dietmar; since I shuffled the whole lot I didn't keep your Tested-by, sorry!

o Turned the SD flags into an enum with automagic power of 2 assignment. I poked the dwarves and they assured me the SD flag values haven't changed.
o [new patch] Made SD flag debug file output flag names

v3 -> v4
--------

o Reordered the series to have fixes / cleanups first

o Added SD_ASYM_CPUCAPACITY propagation (Quentin)
o Made ARM revert back to the default sched topology (Dietmar)
o Removed SD_SERIALIZE degeneration special case (Peter)

o Made SD_NUMA and SD_SERIALIZE have SDF_NEEDS_GROUPS

As discussed on v3, I thought this wasn't required, but thinking some more about it there can be cases where that changes the current behaviour. For instance, in the following wacky triangle:

      0\ 30
      | \
  20  |  2
      | /
      1/ 30

there are two unique distances thus two NUMA topology levels, however the first one for node 2 would have the same span as its child domain and thus should be degenerated. If we don't give SD_NUMA and SD_SERIALIZE SDF_NEEDS_GROUPS, this domain wouldn't be denegerated since its child *doesn't* have either SD_NUMA or SD_SERIALIZE (it's the first NUMA domain), and we'd have this weird NUMA domain lingering with a single group.

v2 -> v3
--------

o Reworded comment for SD_OVERLAP (it's about the groups, not the domains)

o Added more flags to the SD degeneration mask
o Added generation of an SD flag mask for the degeneration functions (Peter)

RFC -> v2
---------

o Rebased on top of tip/sched/core
o Aligned wording of comments between flags
o Rectified some flag descriptions (Morten)
o Added removal of SD_SHARE_POWERDOMAIN (Morten)

Valentin Schneider (17):
ARM, sched/topology: Remove SD_SHARE_POWERDOMAIN
ARM: Revert back to default scheduler topology. sched/topology: Split out SD_* flags declaration to its own file sched/topology: Define and assign sched_domain flag metadata sched/topology: Verify SD_* flags setup when sched_debug is on sched/debug: Output SD flag names rather than their values sched/topology: Introduce SD metaflag for flags needing > 1 groups sched/topology: Use prebuilt SD flag degeneration mask sched/topology: Remove SD_SERIALIZE degeneration special case sched/topology: Propagate SD_ASYM_CPUCAPACITY upwards sched/topology: Mark SD_PREFER_SIBLING as SDF_NEEDS_GROUPS sched/topology: Mark SD_BALANCE_WAKE as SDF_NEEDS_GROUPS sched/topology: Mark SD_SERIALIZE as SDF_NEEDS_GROUPS sched/topology: Mark SD_ASYM_PACKING as SDF_NEEDS_GROUPS sched/topology: Mark SD_OVERLAP as SDF_NEEDS_GROUPS sched/topology: Mark SD_NUMA as SDF_NEEDS_GROUPS sched/topology: Expand use of SD_DEGENERATE_GROUPS_MASK to flags not
needing groups

 arch/arm/kernel/topology.c     |  26 ------
 include/linux/sched/sd_flags.h | 156 +++++++++++++++++++++++++++++++++
 include/linux/sched/topology.h |  45 +++++++---
 kernel/sched/debug.c           |  53 ++++++++++-
 kernel/sched/topology.c        |  54 ++++++------
 5 files changed, 265 insertions(+), 69 deletions(-)
 create mode 100644 include/linux/sched/sd_flags.h

--
2.27.0

#1 lore raw

PATCH v5 1/17 ARM, sched/topology: Remove SD_SHARE_POWERDOMAIN

This flag was introduced in 2014 by commit

d77b3ed5c9f8 ("sched: Add a new SD_SHARE_POWERDOMAIN for sched_domain")

but AFAIA it was never leveraged by the scheduler. The closest thing I can think of is EAS caring about frequency domains, and it does that by leveraging performance domains.

Remove the flag.

---
 arch/arm/kernel/topology.c     |  2 +-
 include/linux/sched/topology.h | 13 ++++++-------
 kernel/sched/topology.c        | 10 +++-------
 3 files changed, 10 insertions(+), 15 deletions(-)
arch/arm/kernel/topology.c +1 -1
index b5adaf744630..353f3ee660e4 100644
@@ -243,7 +243,7 @@ void store_cpu_topology(unsigned int cpuid)
 
 static inline int cpu_corepower_flags(void)
 {
-	return SD_SHARE_PKG_RESOURCES  | SD_SHARE_POWERDOMAIN;
+	return SD_SHARE_PKG_RESOURCES;
 }
 
 static struct sched_domain_topology_level arm_topology[] = {
include/linux/sched/topology.h +6 -7
index 820511289857..6ec7d7c1d1e3 100644
@@ -18,13 +18,12 @@
 #define SD_WAKE_AFFINE		0x0010	/* Wake task to waking CPU */
 #define SD_ASYM_CPUCAPACITY	0x0020  /* Domain members have different CPU capacities */
 #define SD_SHARE_CPUCAPACITY	0x0040	/* Domain members share CPU capacity */
-#define SD_SHARE_POWERDOMAIN	0x0080	/* Domain members share power domain */
-#define SD_SHARE_PKG_RESOURCES	0x0100	/* Domain members share CPU pkg resources */
-#define SD_SERIALIZE		0x0200	/* Only a single load balancing instance */
-#define SD_ASYM_PACKING		0x0400  /* Place busy groups earlier in the domain */
-#define SD_PREFER_SIBLING	0x0800	/* Prefer to place tasks in a sibling domain */
-#define SD_OVERLAP		0x1000	/* sched_domains of this level overlap */
-#define SD_NUMA			0x2000	/* cross-node balancing */
+#define SD_SHARE_PKG_RESOURCES	0x0080	/* Domain members share CPU pkg resources */
+#define SD_SERIALIZE		0x0100	/* Only a single load balancing instance */
+#define SD_ASYM_PACKING		0x0200  /* Place busy groups earlier in the domain */
+#define SD_PREFER_SIBLING	0x0400	/* Prefer to place tasks in a sibling domain */
+#define SD_OVERLAP		0x0800	/* sched_domains of this level overlap */
+#define SD_NUMA			0x1000	/* cross-node balancing */
 
 #ifdef CONFIG_SCHED_SMT
 static inline int cpu_smt_flags(void)
kernel/sched/topology.c +3 -7
index 9079d865a935..865fff3ef20a 100644
@@ -148,8 +148,7 @@ static int sd_degenerate(struct sched_domain *sd)
 			 SD_BALANCE_EXEC |
 			 SD_SHARE_CPUCAPACITY |
 			 SD_ASYM_CPUCAPACITY |
-			 SD_SHARE_PKG_RESOURCES |
-			 SD_SHARE_POWERDOMAIN)) {
+			 SD_SHARE_PKG_RESOURCES)) {
 		if (sd->groups != sd->groups->next)
 			return 0;
 	}
@@ -180,8 +179,7 @@ sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent)
 			    SD_ASYM_CPUCAPACITY |
 			    SD_SHARE_CPUCAPACITY |
 			    SD_SHARE_PKG_RESOURCES |
-			    SD_PREFER_SIBLING |
-			    SD_SHARE_POWERDOMAIN);
+			    SD_PREFER_SIBLING);
 		if (nr_node_ids == 1)
 			pflags &= ~SD_SERIALIZE;
 	}
@@ -1292,7 +1290,6 @@ int __read_mostly		node_reclaim_distance = RECLAIM_DISTANCE;
  *   SD_SHARE_CPUCAPACITY   - describes SMT topologies
  *   SD_SHARE_PKG_RESOURCES - describes shared caches
  *   SD_NUMA                - describes NUMA topologies
- *   SD_SHARE_POWERDOMAIN   - describes shared power domain
  *
  * Odd one out, which beside describing the topology has a quirk also
  * prescribes the desired behaviour that goes along with it:
@@ -1303,8 +1300,7 @@ int __read_mostly		node_reclaim_distance = RECLAIM_DISTANCE;
 	(SD_SHARE_CPUCAPACITY	|	\
 	 SD_SHARE_PKG_RESOURCES |	\
 	 SD_NUMA		|	\
-	 SD_ASYM_PACKING	|	\
-	 SD_SHARE_POWERDOMAIN)
+	 SD_ASYM_PACKING)
 
 static struct sched_domain *
 sd_init(struct sched_domain_topology_level *tl,
Signature · 1 line
2.27.0
#2 lore raw

PATCH v5 2/17 ARM: Revert back to default scheduler topology.

In reply to Valentin Schneider

The ARM-specific GMC level is meant to be built using the thread sibling mask, but no devicetree in arch/arm/boot/dts uses the 'thread' cpu-map binding. With SD_SHARE_POWERDOMAIN gone, this topology level can be removed, at which point ARM no longer benefits from having a custom defined topology table.

Delete the GMC topology level by making ARM use the default scheduler topology table. This essentially reverts commit

fb2aa85564f4 ("sched, ARM: Create a dedicated scheduler topology table")

---
 arch/arm/kernel/topology.c | 26 --------------------------
 1 file changed, 26 deletions(-)
arch/arm/kernel/topology.c +0 -26
index 353f3ee660e4..ef0058de432b 100644
@@ -177,15 +177,6 @@ static inline void parse_dt_topology(void) {}
 static inline void update_cpu_capacity(unsigned int cpuid) {}
 #endif
 
-/*
- * The current assumption is that we can power gate each core independently.
- * This will be superseded by DT binding once available.
- */
-const struct cpumask *cpu_corepower_mask(int cpu)
-{
-	return &cpu_topology[cpu].thread_sibling;
-}
-
 /*
  * store_cpu_topology is called at boot when only one cpu is running
  * and with the mutex cpu_hotplug.lock locked, when several cpus have booted,
@@ -241,20 +232,6 @@ void store_cpu_topology(unsigned int cpuid)
 	update_siblings_masks(cpuid);
 }
 
-static inline int cpu_corepower_flags(void)
-{
-	return SD_SHARE_PKG_RESOURCES;
-}
-
-static struct sched_domain_topology_level arm_topology[] = {
-#ifdef CONFIG_SCHED_MC
-	{ cpu_corepower_mask, cpu_corepower_flags, SD_INIT_NAME(GMC) },
-	{ cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) },
-#endif
-	{ cpu_cpu_mask, SD_INIT_NAME(DIE) },
-	{ NULL, },
-};
-
 /*
  * init_cpu_topology is called at boot when only one cpu is running
  * which prevent simultaneous write access to cpu_topology array
@@ -265,7 +242,4 @@ void __init init_cpu_topology(void)
 	smp_wmb();
 
 	parse_dt_topology();
-
-	/* Set scheduler topology descriptor */
-	set_sched_topology(arm_topology);
 }
Signature · 1 line
2.27.0
#3 lore raw

Re: [PATCH v5 02/17] ARM: Revert back to default scheduler topology.

Quoted 9 lines · Valentin Schneider

The ARM-specific GMC level is meant to be built using the thread sibling mask, but no devicetree in arch/arm/boot/dts uses the 'thread' cpu-map binding. With SD_SHARE_POWERDOMAIN gone, this topology level can be removed, at which point ARM no longer benefits from having a custom defined topology table.

Delete the GMC topology level by making ARM use the default scheduler topology table. This essentially reverts commit

Quoted 6 lines

fb2aa85564f4 ("sched, ARM: Create a dedicated scheduler topology table")

Minor changelog nit, it's helpful to add this final sentence:

No change in functionality is expected.

( If indeed no change in functionality is expected. ;-)

Thanks,

Ingo

#4 lore raw

Re: [PATCH v5 02/17] ARM: Revert back to default scheduler topology.

Quoted 18 lines · Ingo Molnar
Quoted 9 lines · Valentin Schneider

The ARM-specific GMC level is meant to be built using the thread sibling mask, but no devicetree in arch/arm/boot/dts uses the 'thread' cpu-map binding. With SD_SHARE_POWERDOMAIN gone, this topology level can be removed, at which point ARM no longer benefits from having a custom defined topology table.

Delete the GMC topology level by making ARM use the default scheduler topology table. This essentially reverts commit

Quoted 6 lines

fb2aa85564f4 ("sched, ARM: Create a dedicated scheduler topology table")

Quoted 6 lines

Minor changelog nit, it's helpful to add this final sentence:

No change in functionality is expected.

( If indeed no change in functionality is expected. ;-)

Right, that's indeed the case here given the GMC domain would always be degenerated anyway.

Quoted 3 lines

Thanks,

Ingo

#5 lore raw

PATCH v5 3/17 sched/topology: Split out SD_* flags declaration to its own file

In reply to Valentin Schneider

To associate the SD flags with some metadata, we need some more structure in the way they are declared.

Rather than shove that in a free-standing macro list, move the declaration in a separate file that can be re-imported with different SD_FLAG definitions. This is inspired by what is done with the syscall table (see uapi/asm/unistd.h and sys_call_table).

The value assigned to a given SD flag now depends on the order it appears in sd_flags.h. No change in functionality.

---
 include/linux/sched/sd_flags.h | 35 ++++++++++++++++++++++++++++++++++
 include/linux/sched/topology.h | 26 ++++++++++++-------------
 2 files changed, 48 insertions(+), 13 deletions(-)
 create mode 100644 include/linux/sched/sd_flags.h
include/linux/sched/sd_flags.h new file +35 -0
new file mode 100644
index 000000000000..5a74751a1a83
@@ -0,0 +1,35 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * sched-domains (multiprocessor balancing) flag declarations.
+ */
+
+#ifndef SD_FLAG
+#error "Incorrect import of SD flags definitions"
+#endif
+
+/* Balance when about to become idle */
+SD_FLAG(SD_BALANCE_NEWIDLE)
+/* Balance on exec */
+SD_FLAG(SD_BALANCE_EXEC)
+/* Balance on fork, clone */
+SD_FLAG(SD_BALANCE_FORK)
+/* Balance on wakeup */
+SD_FLAG(SD_BALANCE_WAKE)
+/* Wake task to waking CPU */
+SD_FLAG(SD_WAKE_AFFINE)
+/* Domain members have different CPU capacities */
+SD_FLAG(SD_ASYM_CPUCAPACITY)
+/* Domain members share CPU capacity */
+SD_FLAG(SD_SHARE_CPUCAPACITY)
+/* Domain members share CPU pkg resources */
+SD_FLAG(SD_SHARE_PKG_RESOURCES)
+/* Only a single load balancing instance */
+SD_FLAG(SD_SERIALIZE)
+/* Place busy groups earlier in the domain */
+SD_FLAG(SD_ASYM_PACKING)
+/* Prefer to place tasks in a sibling domain */
+SD_FLAG(SD_PREFER_SIBLING)
+/* sched_domains of this level overlap */
+SD_FLAG(SD_OVERLAP)
+/* cross-node balancing */
+SD_FLAG(SD_NUMA)
include/linux/sched/topology.h +13 -13
index 6ec7d7c1d1e3..3e41c0401b5f 100644
@@ -11,19 +11,19 @@
  */
 #ifdef CONFIG_SMP
 
-#define SD_BALANCE_NEWIDLE	0x0001	/* Balance when about to become idle */
-#define SD_BALANCE_EXEC		0x0002	/* Balance on exec */
-#define SD_BALANCE_FORK		0x0004	/* Balance on fork, clone */
-#define SD_BALANCE_WAKE		0x0008  /* Balance on wakeup */
-#define SD_WAKE_AFFINE		0x0010	/* Wake task to waking CPU */
-#define SD_ASYM_CPUCAPACITY	0x0020  /* Domain members have different CPU capacities */
-#define SD_SHARE_CPUCAPACITY	0x0040	/* Domain members share CPU capacity */
-#define SD_SHARE_PKG_RESOURCES	0x0080	/* Domain members share CPU pkg resources */
-#define SD_SERIALIZE		0x0100	/* Only a single load balancing instance */
-#define SD_ASYM_PACKING		0x0200  /* Place busy groups earlier in the domain */
-#define SD_PREFER_SIBLING	0x0400	/* Prefer to place tasks in a sibling domain */
-#define SD_OVERLAP		0x0800	/* sched_domains of this level overlap */
-#define SD_NUMA			0x1000	/* cross-node balancing */
+/* Generate SD flag indexes */
+#define SD_FLAG(name) __##name,
+enum {
+	#include <linux/sched/sd_flags.h>
+	__SD_FLAG_CNT,
+};
+#undef SD_FLAG
+/* Generate SD flag bits */
+#define SD_FLAG(name) name = 1 << __##name,
+enum {
+	#include <linux/sched/sd_flags.h>
+};
+#undef SD_FLAG
 
 #ifdef CONFIG_SCHED_SMT
 static inline int cpu_smt_flags(void)
Signature · 1 line
2.27.0
#6 lore raw

PATCH v5 4/17 sched/topology: Define and assign sched_domain flag metadata

In reply to Valentin Schneider

There are some expectations regarding how sched domain flags should be laid out, but none of them are checked or asserted in sched_domain_debug_one(). After staring at said flags for a while, I've come to realize there's two repeating patterns:

- Shared with children: those flags are set from the base CPU domain upwards. Any domain that has it set will have it set in its children. It hints at "some property holds true / some behaviour is enabled until this level".

- Shared with parents: those flags are set from the topmost domain downwards. Any domain that has it set will have it set in its parents. It hints at "some property isn't visible / some behaviour is disabled until this level".

There are two outliers that (currently) do not map to either of these:

o SD_PREFER_SIBLING, which is cleared below levels with SD_ASYM_CPUCAPACITY. The change was introduced by commit

9c63e84db29b ("sched/core: Disable SD_PREFER_SIBLING on asymmetric CPU capacity domains")

as it could break misfit migration on some systems. In light of this, we might want to change it back to make it fit one of the two categories and fix the issue another way.

o SD_ASYM_CPUCAPACITY, which gets set on a single level and isn't propagated up nor down. From a topology description point of view, it really wants to be SDF_SHARED_PARENT; this will be rectified in a later patch.

Tweak the sched_domain flag declaration to assign each flag an expected layout, and include the rationale for each flag "meta type" assignment as a comment. Consolidate the flag metadata into an array; the index of a flag's metadata can easily be found with log2(flag), IOW __ffs(flag).

---
 include/linux/sched/sd_flags.h | 147 +++++++++++++++++++++++++++------
 include/linux/sched/topology.h |  15 +++-
 2 files changed, 134 insertions(+), 28 deletions(-)
include/linux/sched/sd_flags.h +121 -26
index 5a74751a1a83..ea0ec1a33da4 100644
@@ -7,29 +7,124 @@
 #error "Incorrect import of SD flags definitions"
 #endif
 
-/* Balance when about to become idle */
-SD_FLAG(SD_BALANCE_NEWIDLE)
-/* Balance on exec */
-SD_FLAG(SD_BALANCE_EXEC)
-/* Balance on fork, clone */
-SD_FLAG(SD_BALANCE_FORK)
-/* Balance on wakeup */
-SD_FLAG(SD_BALANCE_WAKE)
-/* Wake task to waking CPU */
-SD_FLAG(SD_WAKE_AFFINE)
-/* Domain members have different CPU capacities */
-SD_FLAG(SD_ASYM_CPUCAPACITY)
-/* Domain members share CPU capacity */
-SD_FLAG(SD_SHARE_CPUCAPACITY)
-/* Domain members share CPU pkg resources */
-SD_FLAG(SD_SHARE_PKG_RESOURCES)
-/* Only a single load balancing instance */
-SD_FLAG(SD_SERIALIZE)
-/* Place busy groups earlier in the domain */
-SD_FLAG(SD_ASYM_PACKING)
-/* Prefer to place tasks in a sibling domain */
-SD_FLAG(SD_PREFER_SIBLING)
-/* sched_domains of this level overlap */
-SD_FLAG(SD_OVERLAP)
-/* cross-node balancing */
-SD_FLAG(SD_NUMA)
+/*
+ * Expected flag uses
+ *
+ * SHARED_CHILD: These flags are meant to be set from the base domain upwards.
+ * If a domain has this flag set, all of its children should have it set. This
+ * is usually because the flag describes some shared resource (all CPUs in that
+ * domain share the same resource), or because they are tied to a scheduling
+ * behaviour that we want to disable at some point in the hierarchy for
+ * scalability reasons.
+ *
+ * In those cases it doesn't make sense to have the flag set for a domain but
+ * not have it in (some of) its children: sched domains ALWAYS span their child
+ * domains, so operations done with parent domains will cover CPUs in the lower
+ * child domains.
+ *
+ *
+ * SHARED_PARENT: These flags are meant to be set from the highest domain
+ * downwards. If a domain has this flag set, all of its parents should have it
+ * set. This is usually for topology properties that start to appear above a
+ * certain level (e.g. domain starts spanning CPUs outside of the base CPU's
+ * socket).
+ */
+#define SDF_SHARED_CHILD 0x1
+#define SDF_SHARED_PARENT 0x2
+
+/*
+ * Balance when about to become idle
+ *
+ * SHARED_CHILD: Set from the base domain up to cpuset.sched_relax_domain_level.
+ */
+SD_FLAG(SD_BALANCE_NEWIDLE, SDF_SHARED_CHILD)
+
+/*
+ * Balance on exec
+ *
+ * SHARED_CHILD: Set from the base domain up to the NUMA reclaim level.
+ */
+SD_FLAG(SD_BALANCE_EXEC, SDF_SHARED_CHILD)
+
+/*
+ * Balance on fork, clone
+ *
+ * SHARED_CHILD: Set from the base domain up to the NUMA reclaim level.
+ */
+SD_FLAG(SD_BALANCE_FORK, SDF_SHARED_CHILD)
+
+/*
+ * Balance on wakeup
+ *
+ * SHARED_CHILD: Set from the base domain up to cpuset.sched_relax_domain_level.
+ */
+SD_FLAG(SD_BALANCE_WAKE, SDF_SHARED_CHILD)
+
+/*
+ * Consider waking task on waking CPU.
+ *
+ * SHARED_CHILD: Set from the base domain up to the NUMA reclaim level.
+ */
+SD_FLAG(SD_WAKE_AFFINE, SDF_SHARED_CHILD)
+
+/*
+ * Domain members have different CPU capacities
+ */
+SD_FLAG(SD_ASYM_CPUCAPACITY, 0)
+
+/*
+ * Domain members share CPU capacity (i.e. SMT)
+ *
+ * SHARED_CHILD: Set from the base domain up until spanned CPUs no longer share
+ * CPU capacity.
+ */
+SD_FLAG(SD_SHARE_CPUCAPACITY, SDF_SHARED_CHILD)
+
+/*
+ * Domain members share CPU package resources (i.e. caches)
+ *
+ * SHARED_CHILD: Set from the base domain up until spanned CPUs no longer share
+ * the same cache(s).
+ */
+SD_FLAG(SD_SHARE_PKG_RESOURCES, SDF_SHARED_CHILD)
+
+/*
+ * Only a single load balancing instance
+ *
+ * SHARED_PARENT: Set for all NUMA levels above NODE. Could be set from a
+ * different level upwards, but it doesn't change that if a domain has this flag
+ * set, then all of its parents need to have it too (otherwise the serialization
+ * doesn't make sense).
+ */
+SD_FLAG(SD_SERIALIZE, SDF_SHARED_PARENT)
+
+/*
+ * Place busy tasks earlier in the domain
+ *
+ * SHARED_CHILD: Usually set on the SMT level. Technically could be set further
+ * up, but currently assumed to be set from the base domain upwards (see
+ * update_top_cache_domain()).
+ */
+SD_FLAG(SD_ASYM_PACKING, SDF_SHARED_CHILD)
+
+/*
+ * Prefer to place tasks in a sibling domain
+ *
+ * Set up until domains start spanning NUMA nodes. Close to being a SHARED_CHILD
+ * flag, but cleared below domains with SD_ASYM_CPUCAPACITY.
+ */
+SD_FLAG(SD_PREFER_SIBLING, 0)
+
+/*
+ * sched_groups of this level overlap
+ *
+ * SHARED_PARENT: Set for all NUMA levels above NODE.
+ */
+SD_FLAG(SD_OVERLAP, SDF_SHARED_PARENT)
+
+/*
+ * Cross-node balancing
+ *
+ * SHARED_PARENT: Set for all NUMA levels above NODE.
+ */
+SD_FLAG(SD_NUMA, SDF_SHARED_PARENT)
include/linux/sched/topology.h +13 -2
index 3e41c0401b5f..32f602ff37a0 100644
@@ -12,19 +12,30 @@
 #ifdef CONFIG_SMP
 
 /* Generate SD flag indexes */
-#define SD_FLAG(name) __##name,
+#define SD_FLAG(name, mflags) __##name,
 enum {
 	#include <linux/sched/sd_flags.h>
 	__SD_FLAG_CNT,
 };
 #undef SD_FLAG
 /* Generate SD flag bits */
-#define SD_FLAG(name) name = 1 << __##name,
+#define SD_FLAG(name, mflags) name = 1 << __##name,
 enum {
 	#include <linux/sched/sd_flags.h>
 };
 #undef SD_FLAG
 
+#ifdef CONFIG_SCHED_DEBUG
+#define SD_FLAG(_name, mflags) [__##_name] = { .meta_flags = mflags, .name = #_name },
+static const struct {
+	unsigned int meta_flags;
+	char *name;
+} sd_flag_debug[] = {
+#include <linux/sched/sd_flags.h>
+};
+#undef SD_FLAG
+#endif
+
 #ifdef CONFIG_SCHED_SMT
 static inline int cpu_smt_flags(void)
 {
Signature · 1 line
2.27.0
#7 lore raw

PATCH v5 5/17 sched/topology: Verify SD_* flags setup when sched_debug is on

In reply to Valentin Schneider

Now that we have some description of what we expect the flags layout to be, we can use that to assert at runtime that the actual layout is sane.

---
 kernel/sched/topology.c | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)
kernel/sched/topology.c +17 -0
index 865fff3ef20a..f128fcf46a41 100644
@@ -29,6 +29,8 @@ static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
 				  struct cpumask *groupmask)
 {
 	struct sched_group *group = sd->groups;
+	unsigned long flags = sd->flags;
+	unsigned int idx;
 
 	cpumask_clear(groupmask);
 
@@ -43,6 +45,21 @@ static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level,
 		printk(KERN_ERR "ERROR: domain->groups does not contain CPU%d\n", cpu);
 	}
 
+	for_each_set_bit(idx, &flags, __SD_FLAG_CNT) {
+		unsigned int flag = BIT(idx);
+		unsigned int meta_flags = sd_flag_debug[idx].meta_flags;
+
+		if ((meta_flags & SDF_SHARED_CHILD) && sd->child &&
+		    !(sd->child->flags & flag))
+			printk(KERN_ERR "ERROR: flag %s set here but not in child\n",
+			       sd_flag_debug[idx].name);
+
+		if ((meta_flags & SDF_SHARED_PARENT) && sd->parent &&
+		    !(sd->parent->flags & flag))
+			printk(KERN_ERR "ERROR: flag %s set here but not in parent\n",
+			       sd_flag_debug[idx].name);
+	}
+
 	printk(KERN_DEBUG "%*s groups:", level + 1, "");
 	do {
 		if (!group) {
Signature · 1 line
2.27.0
#8 lore raw

PATCH v5 6/17 sched/debug: Output SD flag names rather than their values

In reply to Valentin Schneider

Decoding the output of /proc/sys/kernel/sched_domain/cpu*/domain*/flags has always been somewhat annoying, as one needs to go fetch the bit -> name mapping from the source code itself. This encoding can be saved in a script somewhere, but that isn't safe from flags being added, removed or even shuffled around.

What matters for debugging purposes is to get *which* flags are set in a given domain, their associated value is pretty much meaningless.

Make the sd flags debug file output flag names.

---
 kernel/sched/debug.c | 53 +++++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 52 insertions(+), 1 deletion(-)
kernel/sched/debug.c +52 -1
index 36c54265bb2b..e9e036a41c55 100644
@@ -245,6 +245,57 @@ set_table_entry(struct ctl_table *entry,
 	entry->proc_handler = proc_handler;
 }
 
+static int sd_ctl_doflags(struct ctl_table *table, int write,
+		      void *buffer, size_t *lenp, loff_t *ppos)
+{
+	unsigned long flags = *(unsigned long *)table->data;
+	size_t data_size = 0;
+	size_t len = 0;
+	char *tmp;
+	int idx;
+
+	if (write)
+		return 0;
+
+	for_each_set_bit(idx, &flags, __SD_FLAG_CNT) {
+		char *name = sd_flag_debug[idx].name;
+
+		/* Name plus whitespace */
+		data_size += strlen(name) + 1;
+	}
+
+	if (*ppos > data_size) {
+		*lenp = 0;
+		return 0;
+	}
+
+	tmp = kcalloc(data_size + 1, sizeof(tmp), GFP_KERNEL);
+	for_each_set_bit(idx, &flags, __SD_FLAG_CNT) {
+		char *name = sd_flag_debug[idx].name;
+
+		len += snprintf(tmp + len, strlen(name) + 2, "%s ", name);
+	}
+
+	tmp += *ppos;
+	len -= *ppos;
+
+	if (len > *lenp)
+		len = *lenp;
+	if (len)
+		memcpy(buffer, tmp, len);
+	if (len < *lenp) {
+		((char *)buffer)[len] = '\n';
+		len++;
+	}
+
+	*lenp = len;
+	*ppos += len;
+
+	kfree(tmp);
+
+	return 0;
+}
+
 static struct ctl_table *
 sd_alloc_ctl_domain_table(struct sched_domain *sd)
 {
@@ -258,7 +309,7 @@ sd_alloc_ctl_domain_table(struct sched_domain *sd)
 	set_table_entry(&table[2], "busy_factor",	  &sd->busy_factor,	    sizeof(int),  0644, proc_dointvec_minmax);
 	set_table_entry(&table[3], "imbalance_pct",	  &sd->imbalance_pct,	    sizeof(int),  0644, proc_dointvec_minmax);
 	set_table_entry(&table[4], "cache_nice_tries",	  &sd->cache_nice_tries,    sizeof(int),  0644, proc_dointvec_minmax);
-	set_table_entry(&table[5], "flags",		  &sd->flags,		    sizeof(int),  0444, proc_dointvec_minmax);
+	set_table_entry(&table[5], "flags",		  &sd->flags,		    sizeof(int),  0444, sd_ctl_doflags);
 	set_table_entry(&table[6], "max_newidle_lb_cost", &sd->max_newidle_lb_cost, sizeof(long), 0644, proc_doulongvec_minmax);
 	set_table_entry(&table[7], "name",		  sd->name,	       CORENAME_MAX_SIZE, 0444, proc_dostring);
 	/* &table[8] is terminator */
Signature · 1 line
2.27.0
#9 lore raw

Re: [PATCH v5 06/17] sched/debug: Output SD flag names rather than their values

Thank you for the patch! Perhaps something to improve:

Hi Valentin,

Thank you for the patch! Perhaps something to improve:

[auto build test WARNING on tip/sched/core]
[also build test WARNING on tip/auto-latest linux/master linus/master v5.8 next-20200812] [If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch]

url:    https://github.com/0day-ci/linux/commits/Valentin-Schneider/sched-Instrument-sched-domain-flags/20200812-205638
base:   https://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git 949bcb8135a96a6923e676646bd29cbe69e8350f
config: i386-randconfig-s001-20200811 (attached as .config)
compiler: gcc-9 (Debian 9.3.0-15) 9.3.0
reproduce:
        # apt-get install sparse
        # sparse version: v0.6.2-168-g9554805c-dirty
        # save the attached .config to linux build tree
        make W=1 C=1 CF='-fdiagnostic-prefix -D__CHECK_ENDIAN__' ARCH=i386 

If you fix the issue, kindly add following tag as appropriate

sparse warnings: (new ones prefixed by >>)

kernel/sched/debug.c:327:9: sparse: sparse: incorrect type in assignment (different address spaces) @@ expected struct sched_domain *[assigned] sd @@ got struct sched_domain [noderef] __rcu *parent @@
kernel/sched/debug.c:327:9: sparse: expected struct sched_domain *[assigned] sd
kernel/sched/debug.c:327:9: sparse: got struct sched_domain [noderef] __rcu *parent
kernel/sched/debug.c:334:9: sparse: sparse: incorrect type in assignment (different address spaces) @@ expected struct sched_domain *[assigned] sd @@ got struct sched_domain [noderef] __rcu *parent @@
kernel/sched/debug.c:334:9: sparse: expected struct sched_domain *[assigned] sd
kernel/sched/debug.c:334:9: sparse: got struct sched_domain [noderef] __rcu *parent
kernel/sched/debug.c:486:22: sparse: sparse: incompatible types in comparison expression (different address spaces):
kernel/sched/debug.c:486:22: sparse: struct task_struct [noderef] __rcu *
kernel/sched/debug.c:486:22: sparse: struct task_struct *
kernel/sched/debug.c:694:9: sparse: sparse: incorrect type in argument 1 (different address spaces) @@ expected struct task_struct *tsk @@ got struct task_struct [noderef] __rcu *curr @@
kernel/sched/debug.c:694:9: sparse: expected struct task_struct *tsk
kernel/sched/debug.c:694:9: sparse: got struct task_struct [noderef] __rcu *curr
kernel/sched/debug.c:694:9: sparse: sparse: incorrect type in argument 1 (different address spaces) @@ expected struct task_struct *tsk @@ got struct task_struct [noderef] __rcu *curr @@
kernel/sched/debug.c:694:9: sparse: expected struct task_struct *tsk
kernel/sched/debug.c:694:9: sparse: got struct task_struct [noderef] __rcu *curr

>> kernel/sched/debug.c:279:17: sparse: sparse: non size-preserving pointer to integer cast
>> kernel/sched/debug.c:279:17: sparse: sparse: non size-preserving integer to pointer cast

vim +279 kernel/sched/debug.c

247
248 static int sd_ctl_doflags(struct ctl_table *table, int write,
249 void *buffer, size_t *lenp, loff_t *ppos)
250 {
251 unsigned long flags = *(unsigned long *)table->data;
252 size_t data_size = 0;
253 size_t len = 0;
254 char *tmp;
255 int idx;
256
257 if (write)
258 return 0;
259
260 for_each_set_bit(idx, &flags, __SD_FLAG_CNT) {
261 char *name = sd_flag_debug[idx].name;
262
263 /* Name plus whitespace */
264 data_size += strlen(name) + 1;
265 }
266
267 if (*ppos > data_size) {
268 *lenp = 0;
269 return 0;
270 }
271
272 tmp = kcalloc(data_size + 1, sizeof(tmp), GFP_KERNEL);
273 for_each_set_bit(idx, &flags, __SD_FLAG_CNT) {
274 char *name = sd_flag_debug[idx].name;
275
276 len += snprintf(tmp + len, strlen(name) + 2, "%s ", name);
277 }
278

 > 279		tmp += *ppos;

280 len -= *ppos;
281
282 if (len > *lenp)
283 len = *lenp;
284 if (len)
285 memcpy(buffer, tmp, len);
286 if (len < *lenp) {
287 ((char *)buffer)[len] = '\n';
288 len++;
289 }
290
291 *lenp = len;
292 *ppos += len;
293
294 kfree(tmp);
295
296 return 0;
297 }
298

---
0-DAY CI Kernel Test Service, Intel Corporation
https://lists.01.org/hyperkitty/list/[email protected]

#10 lore raw

Re: [PATCH v5 06/17] sched/debug: Output SD flag names rather than their values

Quoted 4 lines · kernel test robot

Hi Valentin,

Thank you for the patch! Perhaps something to improve:

[...]

Quoted 14 lines
url:    https://github.com/0day-ci/linux/commits/Valentin-Schneider/sched-Instrument-sched-domain-flags/20200812-205638
base:   https://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git 949bcb8135a96a6923e676646bd29cbe69e8350f
config: i386-randconfig-s001-20200811 (attached as .config)
compiler: gcc-9 (Debian 9.3.0-15) 9.3.0
reproduce:
        # apt-get install sparse
        # sparse version: v0.6.2-168-g9554805c-dirty
        # save the attached .config to linux build tree
        make W=1 C=1 CF='-fdiagnostic-prefix -D__CHECK_ENDIAN__' ARCH=i386
Quoted 2 lines
Quoted 2 lines

kernel/sched/debug.c:279:17: sparse: sparse: non size-preserving pointer to integer cast kernel/sched/debug.c:279:17: sparse: sparse: non size-preserving integer to pointer cast

Quoted 9 lines
   271
   272		tmp = kcalloc(data_size + 1, sizeof(tmp), GFP_KERNEL);
   273		for_each_set_bit(idx, &flags, __SD_FLAG_CNT) {
   274			char *name = sd_flag_debug[idx].name;
   275
   276			len += snprintf(tmp + len, strlen(name) + 2, "%s ", name);
   277		}
   278
Quoted 1 line

279 tmp += *ppos;

I pretty much copied kernel/sysctl.c::_proc_do_string() and I think that's exactly the same types here: char* buffer incremented by loff_t offset. It does look fine to me, but I can't really parse that warning.

Quoted 23 lines
   280		len -= *ppos;
   281
   282		if (len > *lenp)
   283			len = *lenp;
   284		if (len)
   285			memcpy(buffer, tmp, len);
   286		if (len < *lenp) {
   287			((char *)buffer)[len] = '\n';
   288			len++;
   289		}
   290
   291		*lenp = len;
   292		*ppos += len;
   293
   294		kfree(tmp);
   295
   296		return 0;
   297	}
   298

---
0-DAY CI Kernel Test Service, Intel Corporation
https://lists.01.org/hyperkitty/list/kbuild-all(a)lists.01.org

#11 lore raw

Re: [PATCH v5 06/17] sched/debug: Output SD flag names rather than their values

Quoted 7 lines · Valentin Schneider
Quoted 5 lines · kernel test robot

config: i386-randconfig-s001-20200811 (attached as .config) reproduce:

        # sparse version: v0.6.2-168-g9554805c-dirty
        make W=1 C=1 CF='-fdiagnostic-prefix -D__CHECK_ENDIAN__' ARCH=i386
Quoted 8 lines
Quoted 4 lines
Quoted 4 lines
Quoted 2 lines

kernel/sched/debug.c:279:17: sparse: sparse: non size-preserving pointer to integer cast kernel/sched/debug.c:279:17: sparse: sparse: non size-preserving integer to pointer cast

279 tmp += *ppos;

I pretty much copied kernel/sysctl.c::_proc_do_string() and I think that's exactly the same types here: char* buffer incremented by loff_t offset. It does look fine to me, but I can't really parse that warning.

The warnings mean that there is a cast from a pointer to an integer with a size other than the size of a pointer and the other way around.

I's indeed the case here, on i386, where pointers are 32-bit and loff_t is 64-bit. But yes, I agree:
1) these messages are far from clear
2) these casts are internal and are probably not appropriate here.

I'll look later what can be done at sparse level.

Regards,
-- Luc

#12 lore raw

Re: [PATCH v5 06/17] sched/debug: Output SD flag names rather than their values

Quoted 20 lines · Luc Van Oostenryck
Quoted 7 lines · Valentin Schneider
Quoted 5 lines · kernel test robot

config: i386-randconfig-s001-20200811 (attached as .config) reproduce:

        # sparse version: v0.6.2-168-g9554805c-dirty
        make W=1 C=1 CF='-fdiagnostic-prefix -D__CHECK_ENDIAN__' ARCH=i386
Quoted 8 lines
Quoted 4 lines
Quoted 4 lines
Quoted 2 lines

kernel/sched/debug.c:279:17: sparse: sparse: non size-preserving pointer to integer cast kernel/sched/debug.c:279:17: sparse: sparse: non size-preserving integer to pointer cast

279 tmp += *ppos;

I pretty much copied kernel/sysctl.c::_proc_do_string() and I think that's exactly the same types here: char* buffer incremented by loff_t offset. It does look fine to me, but I can't really parse that warning.

The warnings mean that there is a cast from a pointer to an integer with a size other than the size of a pointer and the other way around.

Quoted 7 lines

I's indeed the case here, on i386, where pointers are 32-bit and loff_t is 64-bit. But yes, I agree:
1) these messages are far from clear
2) these casts are internal and are probably not appropriate here.

I'll look later what can be done at sparse level.

Thanks!

Quoted 3 lines

Regards,
-- Luc

#13 lore raw

Re: [PATCH v5 06/17] sched/debug: Output SD flag names rather than their values

In reply to Valentin Schneider
Thank you for the patch! Perhaps something to improve:

Hi Valentin,

Thank you for the patch! Perhaps something to improve:

[auto build test WARNING on tip/sched/core]
[also build test WARNING on tip/auto-latest linux/master linus/master v5.8 next-20200812] [If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch]

url:    https://github.com/0day-ci/linux/commits/Valentin-Schneider/sched-Instrument-sched-domain-flags/20200812-205638
base:   https://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git 949bcb8135a96a6923e676646bd29cbe69e8350f
config: riscv-randconfig-c003-20200811 (attached as .config)
compiler: riscv64-linux-gcc (GCC) 9.3.0

If you fix the issue, kindly add following tag as appropriate

coccinelle warnings: (new ones prefixed by >>)

>> kernel/sched/debug.c:272:30-36: ERROR: application of sizeof to pointer

Please review and possibly fold the followup patch.

---
0-DAY CI Kernel Test Service, Intel Corporation
https://lists.01.org/hyperkitty/list/kbuild-all(a)lists.01.org

#14 lore raw

[PATCH] sched/debug: fix noderef.cocci warnings

In reply to Valentin Schneider
kernel/sched/debug.c:272:30-36: ERROR: application of sizeof to pointer

From: kernel test robot <[email protected]>

kernel/sched/debug.c:272:30-36: ERROR: application of sizeof to pointer

sizeof when applied to a pointer typed expression gives the size of the pointer

Generated by: scripts/coccinelle/misc/noderef.cocci

CC Valentin Schneider <[email protected]>

---

url:    https://github.com/0day-ci/linux/commits/Valentin-Schneider/sched-Instrument-sched-domain-flags/20200812-205638
base:   https://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git 949bcb8135a96a6923e676646bd29cbe69e8350f
 debug.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
kernel/sched/debug.c +1 -1
@@ -269,7 +269,7 @@ static int sd_ctl_doflags(struct ctl_tab
 		return 0;
 	}
 
-	tmp = kcalloc(data_size + 1, sizeof(tmp), GFP_KERNEL);
+	tmp = kcalloc(data_size + 1, sizeof(*tmp), GFP_KERNEL);
 	for_each_set_bit(idx, &flags, __SD_FLAG_CNT) {
 		char *name = sd_flag_debug[idx].name;
 
#15 lore raw

Re: [PATCH] sched/debug: fix noderef.cocci warnings

Quoted 22 lines · kernel test robot

From: kernel test robot <[email protected]>

kernel/sched/debug.c:272:30-36: ERROR: application of sizeof to pointer

sizeof when applied to a pointer typed expression gives the size of the pointer

Generated by: scripts/coccinelle/misc/noderef.cocci

CC Valentin Schneider <[email protected]>

---

url:    https://github.com/0day-ci/linux/commits/Valentin-Schneider/sched-Instrument-sched-domain-flags/20200812-205638
base:   https://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git 949bcb8135a96a6923e676646bd29cbe69e8350f
 debug.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

kernel/sched/debug.c static int sd_ctl_doflags(struct ctl_tab · line 272

Quoted 5 lines
kernel/sched/debug.c +1 -1
@@ -269,7 +269,7 @@ static int sd_ctl_doflags(struct ctl_tab
              return 0;
      }
 
-	tmp = kcalloc(data_size + 1, sizeof(tmp), GFP_KERNEL);
+	tmp = kcalloc(data_size + 1, sizeof(*tmp), GFP_KERNEL);

Praised be coccinelle for rubbing in my face that I can't write code; also I'm not even checking if the allocation succeeded which is clearly daft, even if this is debug stuff. I'll blame the heat and try to move on...

Quoted 3 lines
kernel/sched/debug.c +0 -0
      for_each_set_bit(idx, &flags, __SD_FLAG_CNT) {
              char *name = sd_flag_debug[idx].name;
 
#16 lore raw

PATCH v5 7/17 sched/topology: Introduce SD metaflag for flags needing > 1 groups

In reply to Valentin Schneider

In preparation of cleaning up the sd_degenerate*() functions, mark flags used in sd_degenerate() with the new SDF_NEEDS_GROUPS flag. With this, build a compile-time mask of those SD flags.

Note that sd_parent_degenerate() uses an extra flag in its mask, SD_PREFER_SIBLING, which remains singled out for now.

---
 include/linux/sched/sd_flags.h | 39 ++++++++++++++++++++++++----------
 include/linux/sched/topology.h |  7 ++++++
 2 files changed, 35 insertions(+), 11 deletions(-)
include/linux/sched/sd_flags.h +28 -11
index ea0ec1a33da4..21a43ad6f26a 100644
@@ -8,7 +8,7 @@
 #endif
 
 /*
- * Expected flag uses
+ * Hierarchical metaflags
  *
  * SHARED_CHILD: These flags are meant to be set from the base domain upwards.
  * If a domain has this flag set, all of its children should have it set. This
@@ -29,29 +29,42 @@
  * certain level (e.g. domain starts spanning CPUs outside of the base CPU's
  * socket).
  */
-#define SDF_SHARED_CHILD 0x1
-#define SDF_SHARED_PARENT 0x2
+#define SDF_SHARED_CHILD       0x1
+#define SDF_SHARED_PARENT      0x2
+
+/*
+ * Behavioural metaflags
+ *
+ * NEEDS_GROUPS: These flags are only relevant if the domain they are set on has
+ * more than one group. This is usually for balancing flags (load balancing
+ * involves equalizing a metric between groups), or for flags describing some
+ * shared resource (which would be shared between groups).
+ */
+#define SDF_NEEDS_GROUPS       0x4
 
 /*
  * Balance when about to become idle
  *
  * SHARED_CHILD: Set from the base domain up to cpuset.sched_relax_domain_level.
+ * NEEDS_GROUPS: Load balancing flag.
  */
-SD_FLAG(SD_BALANCE_NEWIDLE, SDF_SHARED_CHILD)
+SD_FLAG(SD_BALANCE_NEWIDLE, SDF_SHARED_CHILD | SDF_NEEDS_GROUPS)
 
 /*
  * Balance on exec
  *
  * SHARED_CHILD: Set from the base domain up to the NUMA reclaim level.
+ * NEEDS_GROUPS: Load balancing flag.
  */
-SD_FLAG(SD_BALANCE_EXEC, SDF_SHARED_CHILD)
+SD_FLAG(SD_BALANCE_EXEC, SDF_SHARED_CHILD | SDF_NEEDS_GROUPS)
 
 /*
  * Balance on fork, clone
  *
  * SHARED_CHILD: Set from the base domain up to the NUMA reclaim level.
+ * NEEDS_GROUPS: Load balancing flag.
  */
-SD_FLAG(SD_BALANCE_FORK, SDF_SHARED_CHILD)
+SD_FLAG(SD_BALANCE_FORK, SDF_SHARED_CHILD | SDF_NEEDS_GROUPS)
 
 /*
  * Balance on wakeup
@@ -69,24 +82,28 @@ SD_FLAG(SD_WAKE_AFFINE, SDF_SHARED_CHILD)
 
 /*
  * Domain members have different CPU capacities
+ *
+ * NEEDS_GROUPS: Per-CPU capacity is asymmetric between groups.
  */
-SD_FLAG(SD_ASYM_CPUCAPACITY, 0)
+SD_FLAG(SD_ASYM_CPUCAPACITY, SDF_NEEDS_GROUPS)
 
 /*
  * Domain members share CPU capacity (i.e. SMT)
  *
  * SHARED_CHILD: Set from the base domain up until spanned CPUs no longer share
- * CPU capacity.
+ *               CPU capacity.
+ * NEEDS_GROUPS: Capacity is shared between groups.
  */
-SD_FLAG(SD_SHARE_CPUCAPACITY, SDF_SHARED_CHILD)
+SD_FLAG(SD_SHARE_CPUCAPACITY, SDF_SHARED_CHILD | SDF_NEEDS_GROUPS)
 
 /*
  * Domain members share CPU package resources (i.e. caches)
  *
  * SHARED_CHILD: Set from the base domain up until spanned CPUs no longer share
- * the same cache(s).
+ *               the same cache(s).
+ * NEEDS_GROUPS: Caches are shared between groups.
  */
-SD_FLAG(SD_SHARE_PKG_RESOURCES, SDF_SHARED_CHILD)
+SD_FLAG(SD_SHARE_PKG_RESOURCES, SDF_SHARED_CHILD | SDF_NEEDS_GROUPS)
 
 /*
  * Only a single load balancing instance
include/linux/sched/topology.h +7 -0
index 32f602ff37a0..2d59ca77103e 100644
@@ -25,6 +25,13 @@ enum {
 };
 #undef SD_FLAG
 
+/* Generate a mask of SD flags with the SDF_NEEDS_GROUPS metaflag */
+#define SD_FLAG(name, mflags) (name * !!((mflags) & SDF_NEEDS_GROUPS)) |
+static const unsigned int SD_DEGENERATE_GROUPS_MASK =
+#include <linux/sched/sd_flags.h>
+0;
+#undef SD_FLAG
+
 #ifdef CONFIG_SCHED_DEBUG
 #define SD_FLAG(_name, mflags) [__##_name] = { .meta_flags = mflags, .name = #_name },
 static const struct {
Signature · 1 line
2.27.0
#17 lore raw

PATCH v5 8/17 sched/topology: Use prebuilt SD flag degeneration mask

In reply to Valentin Schneider

Leverage SD_DEGENERATE_GROUPS_MASK in sd_degenerate() and sd_parent_degenerate().

---
 kernel/sched/topology.c | 20 ++++----------------
 1 file changed, 4 insertions(+), 16 deletions(-)
kernel/sched/topology.c +4 -16
index f128fcf46a41..5f2bc99ff659 100644
@@ -160,15 +160,9 @@ static int sd_degenerate(struct sched_domain *sd)
 		return 1;
 
 	/* Following flags need at least 2 groups */
-	if (sd->flags & (SD_BALANCE_NEWIDLE |
-			 SD_BALANCE_FORK |
-			 SD_BALANCE_EXEC |
-			 SD_SHARE_CPUCAPACITY |
-			 SD_ASYM_CPUCAPACITY |
-			 SD_SHARE_PKG_RESOURCES)) {
-		if (sd->groups != sd->groups->next)
-			return 0;
-	}
+	if ((sd->flags & SD_DEGENERATE_GROUPS_MASK) &&
+	    (sd->groups != sd->groups->next))
+		return 0;
 
 	/* Following flags don't use groups */
 	if (sd->flags & (SD_WAKE_AFFINE))
@@ -190,13 +184,7 @@ sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent)
 
 	/* Flags needing groups don't count if only 1 group in parent */
 	if (parent->groups == parent->groups->next) {
-		pflags &= ~(SD_BALANCE_NEWIDLE |
-			    SD_BALANCE_FORK |
-			    SD_BALANCE_EXEC |
-			    SD_ASYM_CPUCAPACITY |
-			    SD_SHARE_CPUCAPACITY |
-			    SD_SHARE_PKG_RESOURCES |
-			    SD_PREFER_SIBLING);
+		pflags &= ~(SD_DEGENERATE_GROUPS_MASK | SD_PREFER_SIBLING);
 		if (nr_node_ids == 1)
 			pflags &= ~SD_SERIALIZE;
 	}
Signature · 1 line
2.27.0
#18 lore raw

PATCH v5 9/17 sched/topology: Remove SD_SERIALIZE degeneration special case

In reply to Valentin Schneider

If there is only a single NUMA node in the system, the only NUMA topology level that will be generated will be NODE (identity distance), which doesn't have SD_SERIALIZE.

This means we don't need this special case in sd_parent_degenerate(), as having the NODE level "naturally" covers it. Thus, remove it.

---
 kernel/sched/topology.c | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)
kernel/sched/topology.c +2 -4
index 5f2bc99ff659..00ad7cef2ec1 100644
@@ -183,11 +183,9 @@ sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent)
 		return 0;
 
 	/* Flags needing groups don't count if only 1 group in parent */
-	if (parent->groups == parent->groups->next) {
+	if (parent->groups == parent->groups->next)
 		pflags &= ~(SD_DEGENERATE_GROUPS_MASK | SD_PREFER_SIBLING);
-		if (nr_node_ids == 1)
-			pflags &= ~SD_SERIALIZE;
-	}
+
 	if (~cflags & pflags)
 		return 0;
 
Signature · 1 line
2.27.0
#19 lore raw

PATCH v5 10/17 sched/topology: Propagate SD_ASYM_CPUCAPACITY upwards

In reply to Valentin Schneider

We currently set this flag *only* on domains whose topology level exactly match the level where we detect asymmetry (as returned by asym_cpu_capacity_level()). This is rather problematic.

Say there are two clusters in the system, one with a lone big CPU and the other with a mix of big and LITTLE CPUs (as is allowed by DynamIQ):

DIE [ ]
MC [ ][ ]

     0   1   2   3  4
     L   L   B   B  B

asym_cpu_capacity_level() will figure out that the MC level is the one where all CPUs can see a CPU of max capacity, and we will thus set SD_ASYM_CPUCAPACITY at MC level for all CPUs.

That lone big CPU will degenerate its MC domain, since it would be alone in there, and will end up with just a DIE domain. Since the flag was only set at MC, this CPU ends up not seeing any SD with the flag set, which is broken.

Rather than clearing dflags at every topology level, clear it before entering the topology level loop. This will properly propagate upwards flags that are set starting from a certain level.

---
 include/linux/sched/sd_flags.h | 4 +++-
 kernel/sched/topology.c        | 3 +--
 2 files changed, 4 insertions(+), 3 deletions(-)
include/linux/sched/sd_flags.h +3 -1
index 21a43ad6f26a..4f07b405564e 100644
@@ -83,9 +83,11 @@ SD_FLAG(SD_WAKE_AFFINE, SDF_SHARED_CHILD)
 /*
  * Domain members have different CPU capacities
  *
+ * SHARED_PARENT: Set from the topmost domain down to the first domain where
+ *                asymmetry is detected.
  * NEEDS_GROUPS: Per-CPU capacity is asymmetric between groups.
  */
-SD_FLAG(SD_ASYM_CPUCAPACITY, SDF_NEEDS_GROUPS)
+SD_FLAG(SD_ASYM_CPUCAPACITY, SDF_SHARED_PARENT | SDF_NEEDS_GROUPS)
 
 /*
  * Domain members share CPU capacity (i.e. SMT)
kernel/sched/topology.c +1 -2
index 00ad7cef2ec1..02fd8db747b2 100644
@@ -1988,11 +1988,10 @@ build_sched_domains(const struct cpumask *cpu_map, struct sched_domain_attr *att
 	/* Set up domains for CPUs specified by the cpu_map: */
 	for_each_cpu(i, cpu_map) {
 		struct sched_domain_topology_level *tl;
+		int dflags = 0;
 
 		sd = NULL;
 		for_each_sd_topology(tl) {
-			int dflags = 0;
-
 			if (tl == tl_asym) {
 				dflags |= SD_ASYM_CPUCAPACITY;
 				has_asym = true;
Signature · 1 line
2.27.0
#20 lore raw

PATCH v5 11/17 sched/topology: Mark SD_PREFER_SIBLING as SDF_NEEDS_GROUPS

In reply to Valentin Schneider

SD_PREFER_SIBLING is currently considered in sd_parent_degenerate() but not in sd_degenerate(). It too hinges on load balancing, and thus won't have any effect when set on a domain with a single group. Add it to SD_DEGENERATE_GROUPS_MASK.

---
 include/linux/sched/sd_flags.h | 4 +++-
 kernel/sched/topology.c        | 2 +-
 2 files changed, 4 insertions(+), 2 deletions(-)
include/linux/sched/sd_flags.h +3 -1
index 4f07b405564e..75c9749a6e2d 100644
@@ -131,8 +131,10 @@ SD_FLAG(SD_ASYM_PACKING, SDF_SHARED_CHILD)
  *
  * Set up until domains start spanning NUMA nodes. Close to being a SHARED_CHILD
  * flag, but cleared below domains with SD_ASYM_CPUCAPACITY.
+ *
+ * NEEDS_GROUPS: Load balancing flag.
  */
-SD_FLAG(SD_PREFER_SIBLING, 0)
+SD_FLAG(SD_PREFER_SIBLING, SDF_NEEDS_GROUPS)
 
 /*
  * sched_groups of this level overlap
kernel/sched/topology.c +1 -1
index 02fd8db747b2..8064f495641b 100644
@@ -184,7 +184,7 @@ sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent)
 
 	/* Flags needing groups don't count if only 1 group in parent */
 	if (parent->groups == parent->groups->next)
-		pflags &= ~(SD_DEGENERATE_GROUPS_MASK | SD_PREFER_SIBLING);
+		pflags &= ~SD_DEGENERATE_GROUPS_MASK;
 
 	if (~cflags & pflags)
 		return 0;
Signature · 1 line
2.27.0
#21 lore raw

PATCH v5 12/17 sched/topology: Mark SD_BALANCE_WAKE as SDF_NEEDS_GROUPS

In reply to Valentin Schneider

Even if no mainline topology uses this flag, it is a load balancing flag just like SD_BALANCE_FORK and requires 2+ groups to have any effect.

---
 include/linux/sched/sd_flags.h | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
include/linux/sched/sd_flags.h +2 -1
index 75c9749a6e2d..9043d18ce418 100644
@@ -70,8 +70,9 @@ SD_FLAG(SD_BALANCE_FORK, SDF_SHARED_CHILD | SDF_NEEDS_GROUPS)
  * Balance on wakeup
  *
  * SHARED_CHILD: Set from the base domain up to cpuset.sched_relax_domain_level.
+ * NEEDS_GROUPS: Load balancing flag.
  */
-SD_FLAG(SD_BALANCE_WAKE, SDF_SHARED_CHILD)
+SD_FLAG(SD_BALANCE_WAKE, SDF_SHARED_CHILD | SDF_NEEDS_GROUPS)
 
 /*
  * Consider waking task on waking CPU.
Signature · 1 line
2.27.0
#22 lore raw

PATCH v5 13/17 sched/topology: Mark SD_SERIALIZE as SDF_NEEDS_GROUPS

In reply to Valentin Schneider

There would be no point in preserving a sched_domain with a single group just because it has this flag set. Add it to SD_DEGENERATE_GROUPS_MASK.

---
 include/linux/sched/sd_flags.h | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)
include/linux/sched/sd_flags.h +5 -4
index 9043d18ce418..6eb302528659 100644
@@ -112,11 +112,12 @@ SD_FLAG(SD_SHARE_PKG_RESOURCES, SDF_SHARED_CHILD | SDF_NEEDS_GROUPS)
  * Only a single load balancing instance
  *
  * SHARED_PARENT: Set for all NUMA levels above NODE. Could be set from a
- * different level upwards, but it doesn't change that if a domain has this flag
- * set, then all of its parents need to have it too (otherwise the serialization
- * doesn't make sense).
+ *                different level upwards, but it doesn't change that if a
+ *                domain has this flag set, then all of its parents need to have
+ *                it too (otherwise the serialization doesn't make sense).
+ * NEEDS_GROUPS: No point in preserving domain if it has a single group.
  */
-SD_FLAG(SD_SERIALIZE, SDF_SHARED_PARENT)
+SD_FLAG(SD_SERIALIZE, SDF_SHARED_PARENT | SDF_NEEDS_GROUPS)
 
 /*
  * Place busy tasks earlier in the domain
Signature · 1 line
2.27.0
#23 lore raw

PATCH v5 14/17 sched/topology: Mark SD_ASYM_PACKING as SDF_NEEDS_GROUPS

In reply to Valentin Schneider

Being a load-balancing flag, it requires 2+ groups to have any effect.

---
 include/linux/sched/sd_flags.h | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)
include/linux/sched/sd_flags.h +4 -3
index 6eb302528659..f855f80f0052 100644
@@ -123,10 +123,11 @@ SD_FLAG(SD_SERIALIZE, SDF_SHARED_PARENT | SDF_NEEDS_GROUPS)
  * Place busy tasks earlier in the domain
  *
  * SHARED_CHILD: Usually set on the SMT level. Technically could be set further
- * up, but currently assumed to be set from the base domain upwards (see
- * update_top_cache_domain()).
+ *               up, but currently assumed to be set from the base domain
+ *               upwards (see update_top_cache_domain()).
+ * NEEDS_GROUPS: Load balancing flag.
  */
-SD_FLAG(SD_ASYM_PACKING, SDF_SHARED_CHILD)
+SD_FLAG(SD_ASYM_PACKING, SDF_SHARED_CHILD | SDF_NEEDS_GROUPS)
 
 /*
  * Prefer to place tasks in a sibling domain
Signature · 1 line
2.27.0
#24 lore raw

PATCH v5 15/17 sched/topology: Mark SD_OVERLAP as SDF_NEEDS_GROUPS

In reply to Valentin Schneider

A sched_domain can only have overlapping sched_groups if it has more than one group.

---
 include/linux/sched/sd_flags.h | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
include/linux/sched/sd_flags.h +2 -1
index f855f80f0052..021b909d3941 100644
@@ -143,8 +143,9 @@ SD_FLAG(SD_PREFER_SIBLING, SDF_NEEDS_GROUPS)
  * sched_groups of this level overlap
  *
  * SHARED_PARENT: Set for all NUMA levels above NODE.
+ * NEEDS_GROUPS: Overlaps can only exist with more than one group.
  */
-SD_FLAG(SD_OVERLAP, SDF_SHARED_PARENT)
+SD_FLAG(SD_OVERLAP, SDF_SHARED_PARENT | SDF_NEEDS_GROUPS)
 
 /*
  * Cross-node balancing
Signature · 1 line
2.27.0
#25 lore raw

PATCH v5 16/17 sched/topology: Mark SD_NUMA as SDF_NEEDS_GROUPS

In reply to Valentin Schneider

There would be no point in preserving a sched_domain with a single group just because it has this flag set. Add it to SD_DEGENERATE_GROUPS_MASK.

---
 include/linux/sched/sd_flags.h | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
include/linux/sched/sd_flags.h +2 -1
index 021b909d3941..71675afa55f0 100644
@@ -151,5 +151,6 @@ SD_FLAG(SD_OVERLAP, SDF_SHARED_PARENT | SDF_NEEDS_GROUPS)
  * Cross-node balancing
  *
  * SHARED_PARENT: Set for all NUMA levels above NODE.
+ * NEEDS_GROUPS: No point in preserving domain if it has a single group.
  */
-SD_FLAG(SD_NUMA, SDF_SHARED_PARENT)
+SD_FLAG(SD_NUMA, SDF_SHARED_PARENT | SDF_NEEDS_GROUPS)
Signature · 1 line
2.27.0
#26 lore raw

PATCH v5 17/17 sched/topology: Expand use of SD_DEGENERATE_GROUPS_MASK to flags not needing groups

In reply to Valentin Schneider

All SD flags requiring 2+ sched_group to have any effect are now decorated with the SDF_NEEDS_GROUPS metaflag. This means we can now use the bitwise negation of SD_DEGENERATE_MASK in sd_degenerate() instead of explicitly using SD_WAKE_AFFINE (IOW the only flag without SDF_NEEDS_GROUPS).

From now on, any flag without SDF_NEEDS_GROUPS will be correctly accounted as a flag not requiring 2+ sched_groups.

---
 kernel/sched/topology.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
kernel/sched/topology.c +1 -1
index 8064f495641b..3bb145ef5abd 100644
@@ -165,7 +165,7 @@ static int sd_degenerate(struct sched_domain *sd)
 		return 0;
 
 	/* Following flags don't use groups */
-	if (sd->flags & (SD_WAKE_AFFINE))
+	if (sd->flags & ~SD_DEGENERATE_GROUPS_MASK)
 		return 0;
 
 	return 1;
Signature · 1 line
2.27.0