Skip to contentSkip to conversation

affine kernel threads to specified cpumask (v3)

PATCH In review Marcelo Tosatti 4 participants 12 messages Mar 28, 2020 – Apr 4, 2020

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

cover affine kernel threads to specified cpumask (v3)

This is a kernel enhancement to configure the cpu affinity of kernel threads via kernel boot option isolcpus=no_kthreads,<isolcpus_params>,<cpulist>

When this option is specified, the cpumask is immediately applied upon thread launch. This does not affect kernel threads that specify cpu and node.

This allows CPU isolation (that is not allowing certain threads to execute on certain CPUs) without using the isolcpus=domain parameter, making it possible to enable load balancing on such CPUs during runtime.

Note-1: this is based off on Wind River's patch at
https://github.com/starlingx-staging/stx-integ/blob/master/kernel/kernel-std/centos/patches/affine-compute-kernel-threads.patch

Difference being that this patch is limited to modifying kernel thread cpumask: Behaviour of other threads can be controlled via cgroups or sched_setaffinity.

Note-2: Wind River's patch was based off Christoph Lameter's patch at
https://lwn.net/Articles/565932/ with the only difference being the kernel parameter changed from kthread to kthread_cpus.

Changelog:

v2: use isolcpus= subcommand (Thomas Gleixner)

v3: s/MontaVista/Wind River/ on changelog (Chris Friesen)
documentation updates (Chris Friesen)
undeprecate isolcpus (Chris Friesen)
general cleanups (Frederic Weisbecker)
separate cpu_possible_mask kthread
mask change (Frederic Weisbecker)

#1 lore raw

PATCH 1/3 kthread: switch to cpu_possible_mask

Next patch will switch unbound kernel threads mask to housekeeping_cpumask(), a subset of cpu_possible_mask.

Switch from cpu_all_mask to cpu_possible_mask separately, to ease bisection.

---
 kernel/kthread.c |    4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
kernel/kthread.c +2 -2
Index: linux-2.6/kernel/kthread.c
===================================================================
@@ -347,7 +347,7 @@ struct task_struct *__kthread_create_on_
 		 * The kernel thread should not inherit these properties.
 		 */
 		sched_setscheduler_nocheck(task, SCHED_NORMAL, &param);
-		set_cpus_allowed_ptr(task, cpu_all_mask);
+		set_cpus_allowed_ptr(task, cpu_possible_mask);
 	}
 	kfree(create);
 	return task;
@@ -572,7 +572,7 @@ int kthreadd(void *unused)
 	/* Setup a clean context for our children to inherit. */
 	set_task_comm(tsk, "kthreadd");
 	ignore_signals(tsk);
-	set_cpus_allowed_ptr(tsk, cpu_all_mask);
+	set_cpus_allowed_ptr(tsk, cpu_possible_mask);
 	set_mems_allowed(node_states[N_MEMORY]);
 
 	current->flags |= PF_NOFREEZE;
#2 lore raw

PATCH 2/3 isolcpus: affine kernel threads to specified cpumask

In reply to Marcelo Tosatti

This is a kernel enhancement to configure the cpu affinity of kernel threads via kernel boot option isolcpus=no_kthreads,<isolcpus_params>,<cpulist>

When this option is specified, the cpumask is immediately applied upon thread launch. This does not affect kernel threads that specify cpu and node.

This allows CPU isolation (that is not allowing certain threads to execute on certain CPUs) without using the isolcpus=domain parameter, making it possible to enable load balancing on such CPUs during runtime (see kernel-parameters.txt).

Note-1: this is based off on Wind River's patch at
https://github.com/starlingx-staging/stx-integ/blob/master/kernel/kernel-std/centos/patches/affine-compute-kernel-threads.patch

Difference being that this patch is limited to modifying kernel thread cpumask: Behaviour of other threads can be controlled via cgroups or sched_setaffinity.

Note-2: Wind River's patch was based off Christoph Lameter's patch at
https://lwn.net/Articles/565932/ with the only difference being the kernel parameter changed from kthread to kthread_cpus.

---
 Documentation/admin-guide/kernel-parameters.txt |    8 ++++++++
 include/linux/sched/isolation.h                 |    1 +
 kernel/kthread.c                                |    6 ++++--
 kernel/sched/isolation.c                        |    6 ++++++
 4 files changed, 19 insertions(+), 2 deletions(-)
Documentation/admin-guide/kernel-parameters.txt +8 -0
Index: linux-2.6/Documentation/admin-guide/kernel-parameters.txt
===================================================================
@@ -1959,6 +1959,14 @@
 			  the CPU affinity syscalls or cpuset.
 			  <cpu number> begins at 0 and the maximum value is
 			  "number of CPUs in system - 1".
+			  When using cpusets, use the isolcpus option "kthread"
+			  to avoid creation of kernel threads on isolated CPUs.
+
+			kthread
+			  Adjust the CPU affinity mask of unbound kernel threads to
+			  not contain CPUs on the isolated list. This complements
+			  the isolation provided by the cpusets mechanism described
+			  above and by managed_irq option.
 
 			managed_irq
 
include/linux/sched/isolation.h +1 -0
Index: linux-2.6/include/linux/sched/isolation.h
===================================================================
@@ -14,6 +14,7 @@ enum hk_flags {
 	HK_FLAG_DOMAIN		= (1 << 5),
 	HK_FLAG_WQ		= (1 << 6),
 	HK_FLAG_MANAGED_IRQ	= (1 << 7),
+	HK_FLAG_KTHREAD		= (1 << 8),
 };
 
 #ifdef CONFIG_CPU_ISOLATION
kernel/kthread.c +4 -2
Index: linux-2.6/kernel/kthread.c
===================================================================
@@ -23,6 +23,7 @@
 #include <linux/ptrace.h>
 #include <linux/uaccess.h>
 #include <linux/numa.h>
+#include <linux/sched/isolation.h>
 #include <trace/events/sched.h>
 
 static DEFINE_SPINLOCK(kthread_create_lock);
@@ -347,7 +348,8 @@ struct task_struct *__kthread_create_on_
 		 * The kernel thread should not inherit these properties.
 		 */
 		sched_setscheduler_nocheck(task, SCHED_NORMAL, &param);
-		set_cpus_allowed_ptr(task, cpu_possible_mask);
+		set_cpus_allowed_ptr(task,
+				     housekeeping_cpumask(HK_FLAG_KTHREAD));
 	}
 	kfree(create);
 	return task;
@@ -572,7 +574,7 @@ int kthreadd(void *unused)
 	/* Setup a clean context for our children to inherit. */
 	set_task_comm(tsk, "kthreadd");
 	ignore_signals(tsk);
-	set_cpus_allowed_ptr(tsk, cpu_possible_mask);
+	set_cpus_allowed_ptr(tsk, housekeeping_cpumask(HK_FLAG_KTHREAD));
 	set_mems_allowed(node_states[N_MEMORY]);
 
 	current->flags |= PF_NOFREEZE;
kernel/sched/isolation.c +6 -0
Index: linux-2.6/kernel/sched/isolation.c
===================================================================
@@ -169,6 +169,12 @@ static int __init housekeeping_isolcpus_
 			continue;
 		}
 
+		if (!strncmp(str, "kthread,", 8)) {
+			str += 8;
+			flags |= HK_FLAG_KTHREAD;
+			continue;
+		}
+
 		pr_warn("isolcpus: Error, unknown flag\n");
 		return 0;
 	}
#3 lore raw

Re: [patch 2/3] isolcpus: affine kernel threads to specified cpumask

Quoted 37 lines · Marcelo Tosatti

This is a kernel enhancement to configure the cpu affinity of kernel threads via kernel boot option isolcpus=no_kthreads,<isolcpus_params>,<cpulist>

When this option is specified, the cpumask is immediately applied upon thread launch. This does not affect kernel threads that specify cpu and node.

This allows CPU isolation (that is not allowing certain threads to execute on certain CPUs) without using the isolcpus=domain parameter, making it possible to enable load balancing on such CPUs during runtime (see kernel-parameters.txt).

Note-1: this is based off on Wind River's patch at
https://github.com/starlingx-staging/stx-integ/blob/master/kernel/kernel-std/centos/patches/affine-compute-kernel-threads.patch

Difference being that this patch is limited to modifying kernel thread cpumask: Behaviour of other threads can be controlled via cgroups or sched_setaffinity.

Note-2: Wind River's patch was based off Christoph Lameter's patch at
https://lwn.net/Articles/565932/ with the only difference being the kernel parameter changed from kthread to kthread_cpus.

---
 Documentation/admin-guide/kernel-parameters.txt |    8 ++++++++
 include/linux/sched/isolation.h                 |    1 +
 kernel/kthread.c                                |    6 ++++--
 kernel/sched/isolation.c                        |    6 ++++++
 4 files changed, 19 insertions(+), 2 deletions(-)

Index: linux-2.6/Documentation/admin-guide/kernel-parameters.txt
===================================================================

Documentation/admin-guide/kernel-parameters.txt +0 -0
@@ -1959,6 +1959,14 @@
 			  the CPU affinity syscalls or cpuset.

Documentation/admin-guide/kernel-parameters.txt line 1969

Quoted 10 lines
Documentation/admin-guide/kernel-parameters.txt +8 -0
 			  <cpu number> begins at 0 and the maximum value is
 			  "number of CPUs in system - 1".
+			  When using cpusets, use the isolcpus option "kthread"
+			  to avoid creation of kernel threads on isolated CPUs.
+
+			kthread
+			  Adjust the CPU affinity mask of unbound kernel threads to
+			  not contain CPUs on the isolated list. This complements
+			  the isolation provided by the cpusets mechanism described
+			  above and by managed_irq option.

So, what about what I suggested with having "unbound" instead, which includes all the CPU-unbound work?

HK_FLAG_WQ | HK_FLAG_TIMER | HK_FLAG_RCU | HK_FLAG_MISC | HK_FLAG_KTHREAD | HK_FLAG_SCHED

(and yes your suggestion of including HK_FLAG_SCHED is good).

Because I don't see the point of exposing kthread isolation alone as an ABI so far.

Later I suspect I'll turn all these flags into a single HK_FLAG_UNBOUND.

Thanks.

#4 lore raw

Re: [patch 2/3] isolcpus: affine kernel threads to specified cpumask

Quoted 32 lines · Frederic Weisbecker
Quoted 31 lines · Marcelo Tosatti

This is a kernel enhancement to configure the cpu affinity of kernel threads via kernel boot option isolcpus=no_kthreads,<isolcpus_params>,<cpulist>

When this option is specified, the cpumask is immediately applied upon thread launch. This does not affect kernel threads that specify cpu and node.

This allows CPU isolation (that is not allowing certain threads to execute on certain CPUs) without using the isolcpus=domain parameter, making it possible to enable load balancing on such CPUs during runtime (see kernel-parameters.txt).

Note-1: this is based off on Wind River's patch at
https://github.com/starlingx-staging/stx-integ/blob/master/kernel/kernel-std/centos/patches/affine-compute-kernel-threads.patch

Difference being that this patch is limited to modifying kernel thread cpumask: Behaviour of other threads can be controlled via cgroups or sched_setaffinity.

Note-2: Wind River's patch was based off Christoph Lameter's patch at
https://lwn.net/Articles/565932/ with the only difference being the kernel parameter changed from kthread to kthread_cpus.

---
 Documentation/admin-guide/kernel-parameters.txt |    8 ++++++++
 include/linux/sched/isolation.h                 |    1 +
 kernel/kthread.c                                |    6 ++++--
 kernel/sched/isolation.c                        |    6 ++++++
 4 files changed, 19 insertions(+), 2 deletions(-)
Quoted 16 lines
Quoted 6 lines

Index: linux-2.6/Documentation/admin-guide/kernel-parameters.txt
===================================================================

Documentation/admin-guide/kernel-parameters.txt +0 -0
@@ -1959,6 +1959,14 @@
 			  the CPU affinity syscalls or cpuset.

Documentation/admin-guide/kernel-parameters.txt line 1969

Quoted 10 lines
Documentation/admin-guide/kernel-parameters.txt +8 -0
 			  <cpu number> begins at 0 and the maximum value is
 			  "number of CPUs in system - 1".
+			  When using cpusets, use the isolcpus option "kthread"
+			  to avoid creation of kernel threads on isolated CPUs.
+
+			kthread
+			  Adjust the CPU affinity mask of unbound kernel threads to
+			  not contain CPUs on the isolated list. This complements
+			  the isolation provided by the cpusets mechanism described
+			  above and by managed_irq option.

Hi Frederic,

Quoted 4 lines

So, what about what I suggested with having "unbound" instead, which includes all the CPU-unbound work?

HK_FLAG_WQ | HK_FLAG_TIMER | HK_FLAG_RCU | HK_FLAG_MISC | HK_FLAG_KTHREAD | HK_FLAG_SCHED

After more thought, it would share certain flags with nohz_full=, which seemed confusing.

Quoted 6 lines

(and yes your suggestion of including HK_FLAG_SCHED is good).

Because I don't see the point of exposing kthread isolation alone as an ABI so far.

Later I suspect I'll turn all these flags into a single HK_FLAG_UNBOUND.

How about keeping the flags separate, and then on top of that do an "unbound" flag (less typing needed).

This would allow users to combine the individual flags (again, useful for debugging) while at the same time having an option which groups all the others?

#5 lore raw

Re: [patch 2/3] isolcpus: affine kernel threads to specified cpumask

Quoted 11 lines · Marcelo Tosatti
Quoted 1 line · Frederic Weisbecker

On Sat, Mar 28, 2020 at 12:21:19PM -0300, Marcelo Tosatti wrote:

Hi Frederic,

Quoted 4 lines

So, what about what I suggested with having "unbound" instead, which includes all the CPU-unbound work?

HK_FLAG_WQ | HK_FLAG_TIMER | HK_FLAG_RCU | HK_FLAG_MISC | HK_FLAG_KTHREAD | HK_FLAG_SCHED

After more thought, it would share certain flags with nohz_full=, which seemed confusing.

In fact I think we should also add HK_FLAG_KTHREAD | HK_FLAG_SCHED to nohz_full=

nohz_full is merely just a shortcut to isolate the tick and the unbound works anyway.

Quoted 6 lines
Quoted 6 lines

(and yes your suggestion of including HK_FLAG_SCHED is good).

Because I don't see the point of exposing kthread isolation alone as an ABI so far.

Quoted 8 lines
Quoted 1 line

Later I suspect I'll turn all these flags into a single HK_FLAG_UNBOUND.

How about keeping the flags separate, and then on top of that do an "unbound" flag (less typing needed).

This would allow users to combine the individual flags (again, useful for debugging) while at the same time having an option which groups all the others?

Well, I'd rather not expose all the individual bits to the user. They rely on kernel implementation details that shouldn't be much useful to the user.

As for internal kernel use, not sure this will ease debugging...

Thanks.

#6 lore raw

PATCH 3/3 isolcpus: undeprecate on documentation

In reply to Marcelo Tosatti

isolcpus is used to control steering of interrupts to managed_irqs and kernel threads, therefore its incorrect to state that its deprecated.

Remove deprecation warning.

---
 Documentation/admin-guide/kernel-parameters.txt |    1 -
 1 file changed, 1 deletion(-)
Documentation/admin-guide/kernel-parameters.txt +0 -1
Index: linux-2.6/Documentation/admin-guide/kernel-parameters.txt
===================================================================
@@ -1926,7 +1926,6 @@
 			Format: <RDP>,<reset>,<pci_scan>,<verbosity>
 
 	isolcpus=	[KNL,SMP,ISOL] Isolate a given set of CPUs from disturbance.
-			[Deprecated - use cpusets instead]
 			Format: [flag-list,]<cpu-list>
 
 			Specify one or more CPUs to isolate from disturbances
maintainer
#7 lore raw

Re: [patch 3/3] isolcpus: undeprecate on documentation

Quoted 17 lines · Marcelo Tosatti

isolcpus is used to control steering of interrupts to managed_irqs and kernel threads, therefore its incorrect to state that its deprecated.

Remove deprecation warning.

---
 Documentation/admin-guide/kernel-parameters.txt |    1 -
 1 file changed, 1 deletion(-)

Index: linux-2.6/Documentation/admin-guide/kernel-parameters.txt
===================================================================

Documentation/admin-guide/kernel-parameters.txt line 1930

Quoted 6 lines
Documentation/admin-guide/kernel-parameters.txt +0 -1
@@ -1926,7 +1926,6 @@
 			Format: <RDP>,<reset>,<pci_scan>,<verbosity>
 
 	isolcpus=	[KNL,SMP,ISOL] Isolate a given set of CPUs from disturbance.
-			[Deprecated - use cpusets instead]
 			Format: [flag-list,]<cpu-list>
 

It's still an absolute horrible piece of crap though. nozh_full piling more and more shit on it doesn't make it more shiny.

#8 lore raw

Re: [patch 3/3] isolcpus: undeprecate on documentation

Quoted 20 lines · Peter Zijlstra
Quoted 19 lines · Marcelo Tosatti

isolcpus is used to control steering of interrupts to managed_irqs and kernel threads, therefore its incorrect to state that its deprecated.

Remove deprecation warning.

---
 Documentation/admin-guide/kernel-parameters.txt |    1 -
 1 file changed, 1 deletion(-)

Index: linux-2.6/Documentation/admin-guide/kernel-parameters.txt
===================================================================

Documentation/admin-guide/kernel-parameters.txt +0 -0
@@ -1926,7 +1926,6 @@
 			Format: <RDP>,<reset>,<pci_scan>,<verbosity>
 
Quoted 7 lines

Documentation/admin-guide/kernel-parameters.txt line 1930

Quoted 4 lines
Documentation/admin-guide/kernel-parameters.txt +0 -1
 	isolcpus=	[KNL,SMP,ISOL] Isolate a given set of CPUs from disturbance.
-			[Deprecated - use cpusets instead]
 			Format: [flag-list,]<cpu-list>
 

It's still an absolute horrible piece of crap though. nozh_full piling more and more shit on it doesn't make it more shiny.

Hi Peter,

Why do you dislike it? What you think would be a decent approach?

maintainer
#9 lore raw

Re: [patch 3/3] isolcpus: undeprecate on documentation

Quoted 29 lines · Marcelo Tosatti
Quoted 20 lines · Peter Zijlstra
Quoted 19 lines · Marcelo Tosatti

isolcpus is used to control steering of interrupts to managed_irqs and kernel threads, therefore its incorrect to state that its deprecated.

Remove deprecation warning.

---
 Documentation/admin-guide/kernel-parameters.txt |    1 -
 1 file changed, 1 deletion(-)

Index: linux-2.6/Documentation/admin-guide/kernel-parameters.txt
===================================================================

Documentation/admin-guide/kernel-parameters.txt +0 -0
@@ -1926,7 +1926,6 @@
 			Format: <RDP>,<reset>,<pci_scan>,<verbosity>
 
Quoted 7 lines

Documentation/admin-guide/kernel-parameters.txt line 1930

Quoted 4 lines
Documentation/admin-guide/kernel-parameters.txt +0 -1
 	isolcpus=	[KNL,SMP,ISOL] Isolate a given set of CPUs from disturbance.
-			[Deprecated - use cpusets instead]
 			Format: [flag-list,]<cpu-list>
 

It's still an absolute horrible piece of crap though. nozh_full piling more and more shit on it doesn't make it more shiny.

Quoted 3 lines

Hi Peter,

Why do you dislike it? What you think would be a decent approach?

Try and reconfigure it after boot.

#10 lore raw

Re: [patch 3/3] isolcpus: undeprecate on documentation

Quoted 3 lines · Peter Zijlstra
Quoted 1 line

Why do you dislike it? What you think would be a decent approach?

Try and reconfigure it after boot.

Not really a problem since this is not the only issue that requires a reboot.

You have to reboot most of the time if you change applications on a HPC machine. After all you cannot reliably change the available set of huge pages after boot. So you have to reboot anyways if you want some cores to do different things.

And then the machines performance goes down over time due to memory fragmentation. So you better reboot frequently.

Thus reboot, reboot, reboot. We have become like Windows 95 ...

#11 lore raw

Re: [patch 3/3] isolcpus: undeprecate on documentation

In reply to Peter Zijlstra
Quoted 20 lines · Peter Zijlstra
Quoted 19 lines · Marcelo Tosatti

isolcpus is used to control steering of interrupts to managed_irqs and kernel threads, therefore its incorrect to state that its deprecated.

Remove deprecation warning.

---
 Documentation/admin-guide/kernel-parameters.txt |    1 -
 1 file changed, 1 deletion(-)

Index: linux-2.6/Documentation/admin-guide/kernel-parameters.txt
===================================================================

Documentation/admin-guide/kernel-parameters.txt +0 -0
@@ -1926,7 +1926,6 @@
 			Format: <RDP>,<reset>,<pci_scan>,<verbosity>
 
Quoted 7 lines

Documentation/admin-guide/kernel-parameters.txt line 1930

Quoted 4 lines
Documentation/admin-guide/kernel-parameters.txt +0 -1
 	isolcpus=	[KNL,SMP,ISOL] Isolate a given set of CPUs from disturbance.
-			[Deprecated - use cpusets instead]
 			Format: [flag-list,]<cpu-list>
 

It's still an absolute horrible piece of crap though. nozh_full piling more and more shit on it doesn't make it more shiny.

Right. I still plan to propagate it to the cpuset mountpoints at some future so that this becomes mutable.