Skip to contentSkip to conversation

sched/uclamp: Introduce SCHED_FLAG_RESET_UCLAMP_ON_FORK flag

RFC PATCH In review David Dai 5 participants 18 messages Apr 16, 2023 – May 3, 2023 Scheduler

lore
Share & tools
lore

David Dai

author
#0 lore raw

RFC PATCH 1/1 sched/uclamp: Introduce SCHED_FLAG_RESET_UCLAMP_ON_FORK flag

A userspace service may manage uclamp dynamically for individual tasks and a child task will unintentionally inherit a pesudo-random uclamp setting. This could result in the child task being stuck with a static uclamp value that results in poor performance or poor power.

Using SCHED_FLAG_RESET_ON_FORK is too coarse for this usecase and will reset other useful scheduler attributes. Adding a SCHED_FLAG_RESET_UCLAMP_ON_FORK will allow userspace to have finer control over scheduler attributes of child processes.

Cc Qais Yousef <[email protected]>
Cc Quentin Perret <[email protected]>
Cc Saravana Kannan <[email protected]>
---
 include/linux/sched.h            | 3 +++
 include/uapi/linux/sched.h       | 4 +++-
 kernel/sched/core.c              | 6 +++++-
 tools/include/uapi/linux/sched.h | 4 +++-
 4 files changed, 14 insertions(+), 3 deletions(-)
include/linux/sched.h +3 -0
index 63d242164b1a..b1676b9381f9 100644
@@ -885,6 +885,9 @@ struct task_struct {
 	unsigned			sched_reset_on_fork:1;
 	unsigned			sched_contributes_to_load:1;
 	unsigned			sched_migrated:1;
+#ifdef CONFIG_UCLAMP_TASK
+	unsigned			sched_reset_uclamp_on_fork:1;
+#endif
 
 	/* Force alignment to the next boundary: */
 	unsigned			:0;
include/uapi/linux/sched.h +3 -1
index 3bac0a8ceab2..7515106e1f1a 100644
@@ -132,12 +132,14 @@ struct clone_args {
 #define SCHED_FLAG_KEEP_PARAMS		0x10
 #define SCHED_FLAG_UTIL_CLAMP_MIN	0x20
 #define SCHED_FLAG_UTIL_CLAMP_MAX	0x40
+#define SCHED_FLAG_RESET_UCLAMP_ON_FORK	0x80
 
 #define SCHED_FLAG_KEEP_ALL	(SCHED_FLAG_KEEP_POLICY | \
 				 SCHED_FLAG_KEEP_PARAMS)
 
 #define SCHED_FLAG_UTIL_CLAMP	(SCHED_FLAG_UTIL_CLAMP_MIN | \
-				 SCHED_FLAG_UTIL_CLAMP_MAX)
+				 SCHED_FLAG_UTIL_CLAMP_MAX | \
+				 SCHED_FLAG_RESET_UCLAMP_ON_FORK)
 
 #define SCHED_FLAG_ALL	(SCHED_FLAG_RESET_ON_FORK	| \
 			 SCHED_FLAG_RECLAIM		| \
kernel/sched/core.c +5 -1
index 0d18c3969f90..f2d5f7911855 100644
@@ -1943,6 +1943,10 @@ static void __setscheduler_uclamp(struct task_struct *p,
 		uclamp_se_set(&p->uclamp_req[UCLAMP_MAX],
 			      attr->sched_util_max, true);
 	}
+
+	p->sched_reset_uclamp_on_fork = !!(attr->sched_flags &
+				     SCHED_FLAG_RESET_UCLAMP_ON_FORK);
+
 }
 
 static void uclamp_fork(struct task_struct *p)
@@ -1956,7 +1960,7 @@ static void uclamp_fork(struct task_struct *p)
 	for_each_clamp_id(clamp_id)
 		p->uclamp[clamp_id].active = false;
 
-	if (likely(!p->sched_reset_on_fork))
+	if (likely(!p->sched_reset_on_fork && !p->sched_reset_uclamp_on_fork))
 		return;
 
 	for_each_clamp_id(clamp_id) {
tools/include/uapi/linux/sched.h +3 -1
index 3bac0a8ceab2..d52c59a2e0d0 100644
@@ -132,12 +132,14 @@ struct clone_args {
 #define SCHED_FLAG_KEEP_PARAMS		0x10
 #define SCHED_FLAG_UTIL_CLAMP_MIN	0x20
 #define SCHED_FLAG_UTIL_CLAMP_MAX	0x40
+#define SCHED_FLAG_RESET_UCLAMP_ON_FORK 0x80
 
 #define SCHED_FLAG_KEEP_ALL	(SCHED_FLAG_KEEP_POLICY | \
 				 SCHED_FLAG_KEEP_PARAMS)
 
 #define SCHED_FLAG_UTIL_CLAMP	(SCHED_FLAG_UTIL_CLAMP_MIN | \
-				 SCHED_FLAG_UTIL_CLAMP_MAX)
+				 SCHED_FLAG_UTIL_CLAMP_MAX | \
+				 SCHED_FLAG_RESET_UCLAMP_ON_FORK)
 
 #define SCHED_FLAG_ALL	(SCHED_FLAG_RESET_ON_FORK	| \
 			 SCHED_FLAG_RECLAIM		| \
Signature · 1 line
2.40.0.634.g4ca3ef3211-goog
#1 lore raw
Quoted 3 lines · David Dai

A userspace service may manage uclamp dynamically for individual tasks and a child task will unintentionally inherit a pesudo-random uclamp setting. This could result in the child task being stuck with a static uclamp value

Could you explain this with a little bit more detail? Why isn't the child task also managed by the userspace service?

The child task will only make a difference if it's on the rq.

Does this issue happen with uclamp mainline or only with Android's slightly different version (max- vs. sum aggregation)?

Quoted 6 lines

that results in poor performance or poor power.

Using SCHED_FLAG_RESET_ON_FORK is too coarse for this usecase and will reset other useful scheduler attributes. Adding a SCHED_FLAG_RESET_UCLAMP_ON_FORK will allow userspace to have finer control over scheduler attributes of child processes.

[...]

author
#2 lore raw
Quoted 1 line · Dietmar Eggemann

Hi Dietmar, thanks for your time,

Quoted 7 lines · David Dai
Quoted 3 lines

A userspace service may manage uclamp dynamically for individual tasks and a child task will unintentionally inherit a pesudo-random uclamp setting. This could result in the child task being stuck with a static uclamp value

Could you explain this with a little bit more detail? Why isn't the child task also managed by the userspace service?

See Qais’ reply that contains more detail on how it’s being used in Android. In general, if a dynamic userspace service will adjust uclamp on the fly for a given task, but has no knowledge or control over if or when a task forks. Depending on the timing of the fork, a child task may inherit a very large or a small uclamp_min or uclamp_max value. The intent of this patch is to provide more flexibility to the uclamp APIs such that child tasks do not get stuck with a poor uclamp value when spawned while retaining other sched attributes. When RESET_ON_FORK is set on the parent task, it will reset uclamp values for the child but also reset other sched attributes as well.

Quoted 5 lines

The child task will only make a difference if it's on the rq.

Does this issue happen with uclamp mainline or only with Android's slightly different version (max- vs. sum aggregation)?

I’m using the version of uclamp that’s in Linus’ tree. How uclamp is aggregated is unrelated to the problem I’m trying to solve with this patch. Which is to extend the uclamp APIs to have finer control for the uclamp inheritance of child tasks.

Thanks,
David

Quoted 10 lines
Quoted 7 lines

that results in poor performance or poor power.

Using SCHED_FLAG_RESET_ON_FORK is too coarse for this usecase and will reset other useful scheduler attributes. Adding a SCHED_FLAG_RESET_UCLAMP_ON_FORK will allow userspace to have finer control over scheduler attributes of child processes.

[...]

#3 lore raw
Quoted 14 lines · David Dai
Quoted 1 line · Dietmar Eggemann

Hi Dietmar, thanks for your time,

Quoted 7 lines · David Dai
Quoted 3 lines

A userspace service may manage uclamp dynamically for individual tasks and a child task will unintentionally inherit a pesudo-random uclamp setting. This could result in the child task being stuck with a static uclamp value

Could you explain this with a little bit more detail? Why isn't the child task also managed by the userspace service?

Quoted 10 lines

See Qais’ reply that contains more detail on how it’s being used in Android. In general, if a dynamic userspace service will adjust uclamp on the fly for a given task, but has no knowledge or control over if or when a task forks. Depending on the timing of the fork, a child task may inherit a very large or a small uclamp_min or uclamp_max value. The intent of this patch is to provide more flexibility to the uclamp APIs such that child tasks do not get stuck with a poor uclamp value when spawned while retaining other sched attributes. When RESET_ON_FORK is set on the parent task, it will reset uclamp values for the child but also reset other sched attributes as well.

OK, in this case, why not just change behavior and always reset the uclamp values at fork?

Do we anticipate a use-case in which uclamp inheritance would be required?

Let's not over-complicate the sched_[sg]etattr() unnecessarily.

[...]

Quoted 7 lines
Quoted 2 lines

Does this issue happen with uclamp mainline or only with Android's slightly different version (max- vs. sum aggregation)?

I’m using the version of uclamp that’s in Linus’ tree. How uclamp is aggregated is unrelated to the problem I’m trying to solve with this patch. Which is to extend the uclamp APIs to have finer control for the uclamp inheritance of child tasks.

OK, I see.

maintainer
#4 lore raw
Quoted 27 lines · Dietmar Eggemann
Quoted 14 lines · David Dai
Quoted 1 line · Dietmar Eggemann

Hi Dietmar, thanks for your time,

Quoted 7 lines · David Dai
Quoted 3 lines

A userspace service may manage uclamp dynamically for individual tasks and a child task will unintentionally inherit a pesudo-random uclamp setting. This could result in the child task being stuck with a static uclamp value

Could you explain this with a little bit more detail? Why isn't the child task also managed by the userspace service?

Quoted 10 lines

See Qais’ reply that contains more detail on how it’s being used in Android. In general, if a dynamic userspace service will adjust uclamp on the fly for a given task, but has no knowledge or control over if or when a task forks. Depending on the timing of the fork, a child task may inherit a very large or a small uclamp_min or uclamp_max value. The intent of this patch is to provide more flexibility to the uclamp APIs such that child tasks do not get stuck with a poor uclamp value when spawned while retaining other sched attributes. When RESET_ON_FORK is set on the parent task, it will reset uclamp values for the child but also reset other sched attributes as well.

Quoted 6 lines

OK, in this case, why not just change behavior and always reset the uclamp values at fork?

Do we anticipate a use-case in which uclamp inheritance would be required?

Let's not over-complicate the sched_[sg]etattr() unnecessarily.

I was about to ask the same question and I'm aligned with Dietmar. Use RESET_ON_FORK and set all attributes

Quoted 12 lines

[...]

Quoted 7 lines
Quoted 2 lines

Does this issue happen with uclamp mainline or only with Android's slightly different version (max- vs. sum aggregation)?

I’m using the version of uclamp that’s in Linus’ tree. How uclamp is aggregated is unrelated to the problem I’m trying to solve with this patch. Which is to extend the uclamp APIs to have finer control for the uclamp inheritance of child tasks.

OK, I see.

#5 lore raw
Quoted 32 lines · Vincent Guittot
Quoted 30 lines · Dietmar Eggemann
Quoted 14 lines · David Dai
Quoted 1 line · Dietmar Eggemann

Hi Dietmar, thanks for your time,

Quoted 7 lines · David Dai
Quoted 3 lines

A userspace service may manage uclamp dynamically for individual tasks and a child task will unintentionally inherit a pesudo-random uclamp setting. This could result in the child task being stuck with a static uclamp value

Could you explain this with a little bit more detail? Why isn't the child task also managed by the userspace service?

Quoted 10 lines

See Qais’ reply that contains more detail on how it’s being used in Android. In general, if a dynamic userspace service will adjust uclamp on the fly for a given task, but has no knowledge or control over if or when a task forks. Depending on the timing of the fork, a child task may inherit a very large or a small uclamp_min or uclamp_max value. The intent of this patch is to provide more flexibility to the uclamp APIs such that child tasks do not get stuck with a poor uclamp value when spawned while retaining other sched attributes. When RESET_ON_FORK is set on the parent task, it will reset uclamp values for the child but also reset other sched attributes as well.

OK, in this case, why not just change behavior and always reset the uclamp values at fork?

Quoted 6 lines
Quoted 3 lines

Do we anticipate a use-case in which uclamp inheritance would be required?

Let's not over-complicate the sched_[sg]etattr() unnecessarily.

I was about to ask the same question and I'm aligned with Dietmar. Use RESET_ON_FORK and set all attributes

That's racy though. If we have an external service (that's only responsible for setting uclamp) setting all the attributes, the forked thread could also be trying to set some of the attributes. Also, how is this external service going to keep track of all the threads being forked and set the right attributes for all of them?

If it's not considered a UAPI breakage, I'd rather we never inherit uclamp.

-Saravana

Quoted 13 lines
Quoted 13 lines

[...]

Quoted 7 lines
Quoted 2 lines

Does this issue happen with uclamp mainline or only with Android's slightly different version (max- vs. sum aggregation)?

I’m using the version of uclamp that’s in Linus’ tree. How uclamp is aggregated is unrelated to the problem I’m trying to solve with this patch. Which is to extend the uclamp APIs to have finer control for the uclamp inheritance of child tasks.

OK, I see.

#6 lore raw
Quoted 38 lines · Saravana Kannan
Quoted 36 lines · Vincent Guittot
Quoted 33 lines · Dietmar Eggemann
Quoted 14 lines · David Dai
Quoted 1 line · Dietmar Eggemann

Hi Dietmar, thanks for your time,

Quoted 7 lines · David Dai
Quoted 3 lines

A userspace service may manage uclamp dynamically for individual tasks and a child task will unintentionally inherit a pesudo-random uclamp setting. This could result in the child task being stuck with a static uclamp value

Could you explain this with a little bit more detail? Why isn't the child task also managed by the userspace service?

Quoted 10 lines

See Qais’ reply that contains more detail on how it’s being used in Android. In general, if a dynamic userspace service will adjust uclamp on the fly for a given task, but has no knowledge or control over if or when a task forks. Depending on the timing of the fork, a child task may inherit a very large or a small uclamp_min or uclamp_max value. The intent of this patch is to provide more flexibility to the uclamp APIs such that child tasks do not get stuck with a poor uclamp value when spawned while retaining other sched attributes. When RESET_ON_FORK is set on the parent task, it will reset uclamp values for the child but also reset other sched attributes as well.

OK, in this case, why not just change behavior and always reset the uclamp values at fork?

Do we anticipate a use-case in which uclamp inheritance would be required?

Let's not over-complicate the sched_[sg]etattr() unnecessarily.

Quoted 8 lines
Quoted 2 lines

I was about to ask the same question and I'm aligned with Dietmar. Use RESET_ON_FORK and set all attributes

That's racy though. If we have an external service (that's only responsible for setting uclamp) setting all the attributes, the forked thread could also be trying to set some of the attributes. Also, how is this external service going to keep track of all the threads being forked and set the right attributes for all of them?

I didn't understand what Vincent is suggesting. But based on your response it seems we are supposed to track all the forks and reset everything. If we can track all the forks and reset everything, then why use RESET_ON_FORK even or rely on the kernel do anything actually..

This also smells like a hack.

Cheers

--
Qais Yousef

maintainer
#7 lore raw
In reply to Saravana Kannan
Quoted 39 lines · Saravana Kannan
Quoted 36 lines · Vincent Guittot
Quoted 33 lines · Dietmar Eggemann
Quoted 14 lines · David Dai
Quoted 1 line · Dietmar Eggemann

Hi Dietmar, thanks for your time,

Quoted 7 lines · David Dai
Quoted 3 lines

A userspace service may manage uclamp dynamically for individual tasks and a child task will unintentionally inherit a pesudo-random uclamp setting. This could result in the child task being stuck with a static uclamp value

Could you explain this with a little bit more detail? Why isn't the child task also managed by the userspace service?

Quoted 10 lines

See Qais’ reply that contains more detail on how it’s being used in Android. In general, if a dynamic userspace service will adjust uclamp on the fly for a given task, but has no knowledge or control over if or when a task forks. Depending on the timing of the fork, a child task may inherit a very large or a small uclamp_min or uclamp_max value. The intent of this patch is to provide more flexibility to the uclamp APIs such that child tasks do not get stuck with a poor uclamp value when spawned while retaining other sched attributes. When RESET_ON_FORK is set on the parent task, it will reset uclamp values for the child but also reset other sched attributes as well.

OK, in this case, why not just change behavior and always reset the uclamp values at fork?

Do we anticipate a use-case in which uclamp inheritance would be required?

Let's not over-complicate the sched_[sg]etattr() unnecessarily.

Quoted 8 lines
Quoted 2 lines

I was about to ask the same question and I'm aligned with Dietmar. Use RESET_ON_FORK and set all attributes

That's racy though. If we have an external service (that's only responsible for setting uclamp) setting all the attributes, the forked thread could also be trying to set some of the attributes. Also, how is this external service going to keep track of all the threads being forked and set the right attributes for all of them?

My assumption was that you didn't use RESET_ON_FORK because there were other attributes that you wanted to keep from parent but it doesn't seem to be the case so use RESET_ON_FORK and if needed the forked thread will set its other attributes

Quoted 18 lines

If it's not considered a UAPI breakage, I'd rather we never inherit uclamp.

-Saravana

Quoted 13 lines
Quoted 13 lines

[...]

Quoted 7 lines
Quoted 2 lines

Does this issue happen with uclamp mainline or only with Android's slightly different version (max- vs. sum aggregation)?

I’m using the version of uclamp that’s in Linus’ tree. How uclamp is aggregated is unrelated to the problem I’m trying to solve with this patch. Which is to extend the uclamp APIs to have finer control for the uclamp inheritance of child tasks.

OK, I see.

#8 lore raw

Hi Vincent

Sorry for the late response.

Quoted 43 lines · Vincent Guittot
Quoted 42 lines · Saravana Kannan
Quoted 38 lines · Vincent Guittot
Quoted 27 lines · Dietmar Eggemann
Quoted 14 lines · David Dai
Quoted 1 line · Dietmar Eggemann

Hi Dietmar, thanks for your time,

Quoted 7 lines · David Dai
Quoted 3 lines

A userspace service may manage uclamp dynamically for individual tasks and a child task will unintentionally inherit a pesudo-random uclamp setting. This could result in the child task being stuck with a static uclamp value

Could you explain this with a little bit more detail? Why isn't the child task also managed by the userspace service?

Quoted 10 lines

See Qais’ reply that contains more detail on how it’s being used in Android. In general, if a dynamic userspace service will adjust uclamp on the fly for a given task, but has no knowledge or control over if or when a task forks. Depending on the timing of the fork, a child task may inherit a very large or a small uclamp_min or uclamp_max value. The intent of this patch is to provide more flexibility to the uclamp APIs such that child tasks do not get stuck with a poor uclamp value when spawned while retaining other sched attributes. When RESET_ON_FORK is set on the parent task, it will reset uclamp values for the child but also reset other sched attributes as well.

Quoted 6 lines

OK, in this case, why not just change behavior and always reset the uclamp values at fork?

Do we anticipate a use-case in which uclamp inheritance would be required?

Let's not over-complicate the sched_[sg]etattr() unnecessarily.

I was about to ask the same question and I'm aligned with Dietmar. Use RESET_ON_FORK and set all attributes

Quoted 8 lines
Quoted 5 lines

That's racy though. If we have an external service (that's only responsible for setting uclamp) setting all the attributes, the forked thread could also be trying to set some of the attributes. Also, how is this external service going to keep track of all the threads being forked and set the right attributes for all of them?

My assumption was that you didn't use RESET_ON_FORK because there were other attributes that you wanted to keep from parent but it doesn't

Correct.

Quoted 2 lines

seem to be the case so use RESET_ON_FORK and if needed the forked thread will set its other attributes

Hmm, it seems you think it's the parent who's trying to control the fork for the child. But the situation we're in is different.

ADPF is a shared library/middleware/system service that provides higher level API to manage the performance requirements for a group of tasks. They provide a set of PIDs and a deadline, then continuously tell it how long they took to finish their work. ADPF uses this info to manipulate uclamp_min dynamically on their behalf so they finish their work in time. Explaining it very simply and briefly here. It should make using uclamp_max easier too (though we are not there yet) since the most efficient point is platform specific and the middleware can help abstract this better in a portable way.

The problem is these tasks can be RT and/or can have their priorities/nice values set by something else. But what we want is to make sure that we control uclamp values only for the PIDs we were asked to manage, and anything forked must have their uclamp values reset, but not the rest as this is outside of this service available context. It gets delegated to handle performance hints only and make using uclamp easier and generic. But this should not interfere with policy and priority management done outside of its scope.

Apps that manage their own uclamp values wouldn't need this. But if we are to build a smart middleware that provides a consistent and easier to use higher level APIs, which is what we have here, we need this selective reset on fork.

Does this explain the problem better?

Note we noticed because we saw problems here. So we are not trying to be theoretically cautious.

Thanks!

--
Qais Yousef

#9 lore raw
In reply to Dietmar Eggemann
Quoted 27 lines · Dietmar Eggemann
Quoted 14 lines · David Dai
Quoted 1 line · Dietmar Eggemann

Hi Dietmar, thanks for your time,

Quoted 7 lines · David Dai
Quoted 3 lines

A userspace service may manage uclamp dynamically for individual tasks and a child task will unintentionally inherit a pesudo-random uclamp setting. This could result in the child task being stuck with a static uclamp value

Could you explain this with a little bit more detail? Why isn't the child task also managed by the userspace service?

Quoted 10 lines

See Qais’ reply that contains more detail on how it’s being used in Android. In general, if a dynamic userspace service will adjust uclamp on the fly for a given task, but has no knowledge or control over if or when a task forks. Depending on the timing of the fork, a child task may inherit a very large or a small uclamp_min or uclamp_max value. The intent of this patch is to provide more flexibility to the uclamp APIs such that child tasks do not get stuck with a poor uclamp value when spawned while retaining other sched attributes. When RESET_ON_FORK is set on the parent task, it will reset uclamp values for the child but also reset other sched attributes as well.

Quoted 2 lines

OK, in this case, why not just change behavior and always reset the uclamp values at fork?

Personally, I'd have preferred uclamp was never inherited in the first place, but wouldn't that be considered as breaking UAPI if we change it now?

-Saravana

Quoted 15 lines

Do we anticipate a use-case in which uclamp inheritance would be required?

Let's not over-complicate the sched_[sg]etattr() unnecessarily.

[...]

Quoted 7 lines
Quoted 2 lines

Does this issue happen with uclamp mainline or only with Android's slightly different version (max- vs. sum aggregation)?

I’m using the version of uclamp that’s in Linus’ tree. How uclamp is aggregated is unrelated to the problem I’m trying to solve with this patch. Which is to extend the uclamp APIs to have finer control for the uclamp inheritance of child tasks.

OK, I see.

#10 lore raw
Quoted 29 lines · Saravana Kannan
Quoted 27 lines · Dietmar Eggemann
Quoted 24 lines · David Dai
Quoted 1 line · Dietmar Eggemann

Hi Dietmar, thanks for your time,

Quoted 7 lines · David Dai
Quoted 3 lines

A userspace service may manage uclamp dynamically for individual tasks and a child task will unintentionally inherit a pesudo-random uclamp setting. This could result in the child task being stuck with a static uclamp value

Could you explain this with a little bit more detail? Why isn't the child task also managed by the userspace service?

See Qais’ reply that contains more detail on how it’s being used in Android. In general, if a dynamic userspace service will adjust uclamp on the fly for a given task, but has no knowledge or control over if or when a task forks. Depending on the timing of the fork, a child task may inherit a very large or a small uclamp_min or uclamp_max value. The intent of this patch is to provide more flexibility to the uclamp APIs such that child tasks do not get stuck with a poor uclamp value when spawned while retaining other sched attributes. When RESET_ON_FORK is set on the parent task, it will reset uclamp values for the child but also reset other sched attributes as well.

Quoted 6 lines
Quoted 2 lines

OK, in this case, why not just change behavior and always reset the uclamp values at fork?

Personally, I'd have preferred uclamp was never inherited in the first place, but wouldn't that be considered as breaking UAPI if we change it now?

Why is it okay to inherit priority but not uclamp? I see this is hacky and setting a policy in the kernel. Let's not tailor it to specific use cases only please.

--
Qais Yousef

#11 lore raw
In reply to David Dai

Hi David!

Quoted 9 lines · David Dai

A userspace service may manage uclamp dynamically for individual tasks and a child task will unintentionally inherit a pesudo-random uclamp setting. This could result in the child task being stuck with a static uclamp value that results in poor performance or poor power.

Using SCHED_FLAG_RESET_ON_FORK is too coarse for this usecase and will reset other useful scheduler attributes. Adding a SCHED_FLAG_RESET_UCLAMP_ON_FORK will allow userspace to have finer control over scheduler attributes of child processes.

Thanks a lot for the patch. This has a been a known limitation for a while but didn't manage to find the time to push anything yet.

ADPF (Android Dynamic Performance Framework) exposes APIs to manage performance for a set of pids [1]. Only these tasks belong to the session and any forked tasked is expected to have its uclamp values reset. But as you pointed out, the current RESET_ON_FORK resets everything, but we don't want that as these attributes don't belong to ADPF to decide whether they should be reset too or not. And not resetting them means we can end up with tasks inheriting random uclamp values unintentionally. We can't tell these tasks not to fork anything. If the forked tasks are expected to be part of the session, then their pids must be added explicitly.

[1] https://developer.android.com/reference/android/os/PerformanceHintManager#createHintSession(int%5B%5D,%20long)

Quoted 17 lines
Cc Qais Yousef <[email protected]>
Cc Quentin Perret <[email protected]>
Cc Saravana Kannan <[email protected]>
---
 include/linux/sched.h            | 3 +++
 include/uapi/linux/sched.h       | 4 +++-
 kernel/sched/core.c              | 6 +++++-
 tools/include/uapi/linux/sched.h | 4 +++-
 4 files changed, 14 insertions(+), 3 deletions(-)
include/linux/sched.h +0 -0
index 63d242164b1a..b1676b9381f9 100644

include/linux/sched.h struct task_struct { · line 885

Quoted 1 line
include/linux/sched.h +0 -0
@@ -885,6 +885,9 @@ struct task_struct {
 	unsigned			sched_reset_on_fork:1;

nit: can't we convert to a flag and re-use?

Quoted 15 lines
include/linux/sched.h +3 -0
 	unsigned			sched_contributes_to_load:1;
 	unsigned			sched_migrated:1;
+#ifdef CONFIG_UCLAMP_TASK
+	unsigned			sched_reset_uclamp_on_fork:1;
+#endif
 
 	/* Force alignment to the next boundary: */
 	unsigned			:0;
include/uapi/linux/sched.h +0 -0
index 3bac0a8ceab2..7515106e1f1a 100644
@@ -132,12 +132,14 @@ struct clone_args {
 #define SCHED_FLAG_KEEP_PARAMS		0x10
 #define SCHED_FLAG_UTIL_CLAMP_MIN	0x20

include/uapi/linux/sched.h struct clone_args { · line 142

Quoted 10 lines
include/uapi/linux/sched.h +3 -1
 #define SCHED_FLAG_UTIL_CLAMP_MAX	0x40
+#define SCHED_FLAG_RESET_UCLAMP_ON_FORK	0x80
 
 #define SCHED_FLAG_KEEP_ALL	(SCHED_FLAG_KEEP_POLICY | \
 				 SCHED_FLAG_KEEP_PARAMS)
 
 #define SCHED_FLAG_UTIL_CLAMP	(SCHED_FLAG_UTIL_CLAMP_MIN | \
-				 SCHED_FLAG_UTIL_CLAMP_MAX)
+				 SCHED_FLAG_UTIL_CLAMP_MAX | \
+				 SCHED_FLAG_RESET_UCLAMP_ON_FORK)

I was considering to have something a bit more generic that allows selecting which attributes to reset.

For example a syscall with SCHED_FLAG_RESET_ON_FORK_SEL combined with SCHED_FLAG_UCLAMP_MIN/MAX will only reset those. This should make it extensible if we have other similar use cases in the future. The downside it *might* require to be done in a separate syscall to the one that sets these parameter. But it should be done once.

Maybe there's a better interface, but I think it makes sense to do it in a way that we won't have to do this again. Would be good to hear from maintainers first before you take my word for it ;-)

Cheers

--
Qais Yousef

Quoted 50 lines
include/uapi/linux/sched.h +0 -0
 #define SCHED_FLAG_ALL	(SCHED_FLAG_RESET_ON_FORK	| \
 			 SCHED_FLAG_RECLAIM		| \
kernel/sched/core.c +5 -1
index 0d18c3969f90..f2d5f7911855 100644
@@ -1943,6 +1943,10 @@ static void __setscheduler_uclamp(struct task_struct *p,
 		uclamp_se_set(&p->uclamp_req[UCLAMP_MAX],
 			      attr->sched_util_max, true);
 	}
+
+	p->sched_reset_uclamp_on_fork = !!(attr->sched_flags &
+				     SCHED_FLAG_RESET_UCLAMP_ON_FORK);
+
 }
 
 static void uclamp_fork(struct task_struct *p)
@@ -1956,7 +1960,7 @@ static void uclamp_fork(struct task_struct *p)
 	for_each_clamp_id(clamp_id)
 		p->uclamp[clamp_id].active = false;
 
-	if (likely(!p->sched_reset_on_fork))
+	if (likely(!p->sched_reset_on_fork && !p->sched_reset_uclamp_on_fork))
 		return;
 
 	for_each_clamp_id(clamp_id) {
tools/include/uapi/linux/sched.h +3 -1
index 3bac0a8ceab2..d52c59a2e0d0 100644
@@ -132,12 +132,14 @@ struct clone_args {
 #define SCHED_FLAG_KEEP_PARAMS		0x10
 #define SCHED_FLAG_UTIL_CLAMP_MIN	0x20
 #define SCHED_FLAG_UTIL_CLAMP_MAX	0x40
+#define SCHED_FLAG_RESET_UCLAMP_ON_FORK 0x80
 
 #define SCHED_FLAG_KEEP_ALL	(SCHED_FLAG_KEEP_POLICY | \
 				 SCHED_FLAG_KEEP_PARAMS)
 
 #define SCHED_FLAG_UTIL_CLAMP	(SCHED_FLAG_UTIL_CLAMP_MIN | \
-				 SCHED_FLAG_UTIL_CLAMP_MAX)
+				 SCHED_FLAG_UTIL_CLAMP_MAX | \
+				 SCHED_FLAG_RESET_UCLAMP_ON_FORK)
 
 #define SCHED_FLAG_ALL	(SCHED_FLAG_RESET_ON_FORK	| \
 			 SCHED_FLAG_RECLAIM		| \
Signature · 1 line
2.40.0.634.g4ca3ef3211-goog
#12 lore raw
Quoted 79 lines · Qais Yousef

Hi David!

Quoted 9 lines · David Dai

A userspace service may manage uclamp dynamically for individual tasks and a child task will unintentionally inherit a pesudo-random uclamp setting. This could result in the child task being stuck with a static uclamp value that results in poor performance or poor power.

Using SCHED_FLAG_RESET_ON_FORK is too coarse for this usecase and will reset other useful scheduler attributes. Adding a SCHED_FLAG_RESET_UCLAMP_ON_FORK will allow userspace to have finer control over scheduler attributes of child processes.

Thanks a lot for the patch. This has a been a known limitation for a while but didn't manage to find the time to push anything yet.

ADPF (Android Dynamic Performance Framework) exposes APIs to manage performance for a set of pids [1]. Only these tasks belong to the session and any forked tasked is expected to have its uclamp values reset. But as you pointed out, the current RESET_ON_FORK resets everything, but we don't want that as these attributes don't belong to ADPF to decide whether they should be reset too or not. And not resetting them means we can end up with tasks inheriting random uclamp values unintentionally. We can't tell these tasks not to fork anything. If the forked tasks are expected to be part of the session, then their pids must be added explicitly.

[1] https://developer.android.com/reference/android/os/PerformanceHintManager#createHintSession(int%5B%5D,%20long)

Quoted 17 lines
Cc Qais Yousef <[email protected]>
Cc Quentin Perret <[email protected]>
Cc Saravana Kannan <[email protected]>
---
 include/linux/sched.h            | 3 +++
 include/uapi/linux/sched.h       | 4 +++-
 kernel/sched/core.c              | 6 +++++-
 tools/include/uapi/linux/sched.h | 4 +++-
 4 files changed, 14 insertions(+), 3 deletions(-)
include/linux/sched.h +0 -0
index 63d242164b1a..b1676b9381f9 100644

include/linux/sched.h struct task_struct { · line 885

Quoted 1 line
include/linux/sched.h +0 -0
@@ -885,6 +885,9 @@ struct task_struct {
      unsigned                        sched_reset_on_fork:1;

nit: can't we convert to a flag and re-use?

Quoted 15 lines
include/linux/sched.h +3 -0
      unsigned                        sched_contributes_to_load:1;
      unsigned                        sched_migrated:1;
+#ifdef CONFIG_UCLAMP_TASK
+     unsigned                        sched_reset_uclamp_on_fork:1;
+#endif
 
      /* Force alignment to the next boundary: */
      unsigned                        :0;
include/uapi/linux/sched.h +0 -0
index 3bac0a8ceab2..7515106e1f1a 100644
@@ -132,12 +132,14 @@ struct clone_args {
 #define SCHED_FLAG_KEEP_PARAMS               0x10
 #define SCHED_FLAG_UTIL_CLAMP_MIN    0x20

include/uapi/linux/sched.h struct clone_args { · line 142

Quoted 10 lines
include/uapi/linux/sched.h +3 -1
 #define SCHED_FLAG_UTIL_CLAMP_MAX    0x40
+#define SCHED_FLAG_RESET_UCLAMP_ON_FORK      0x80
 
 #define SCHED_FLAG_KEEP_ALL  (SCHED_FLAG_KEEP_POLICY | \
                               SCHED_FLAG_KEEP_PARAMS)
 
 #define SCHED_FLAG_UTIL_CLAMP        (SCHED_FLAG_UTIL_CLAMP_MIN | \
-                              SCHED_FLAG_UTIL_CLAMP_MAX)
+                              SCHED_FLAG_UTIL_CLAMP_MAX | \
+                              SCHED_FLAG_RESET_UCLAMP_ON_FORK)

I was considering to have something a bit more generic that allows selecting which attributes to reset.

Quoted 5 lines

For example a syscall with SCHED_FLAG_RESET_ON_FORK_SEL combined with SCHED_FLAG_UCLAMP_MIN/MAX will only reset those. This should make it extensible if we have other similar use cases in the future. The downside it *might* require to be done in a separate syscall to the one that sets these parameter. But it should be done once.

In addition to the downside you mentioned, I'm not a huge fan of this suggestion since the meaning of the SCHED_FLAG_RESET_ON_FORK_SEL flag changes based on what other flags or attrs are set. I'd rather we have explicit flags.

SCHED_FLAG_RESET_ON_FORK_SEL makes it harder to maintain the userspace code/makes it easy to accidentally introduce bugs. For example, a syscall could be setting UCLAMP_MIN and RESET_ON_FORK_SEL. Someone else might come and change the call to also set a nice value but not remember to split it up into two calls. Whereas with an explicit flag like David's proposal, we won't hit such an issue.

Also, we'll need to have separate flags internally to track what needs to be reset on fork vs not. So we really aren't saving anything by adding RESET_ON_FORK_SEL.

At least, that's my 2 cents.

-Saravana

Quoted 60 lines

Maybe there's a better interface, but I think it makes sense to do it in a way that we won't have to do this again. Would be good to hear from maintainers first before you take my word for it ;-)

Cheers

--
Qais Yousef

Quoted 50 lines
include/uapi/linux/sched.h +0 -0
 #define SCHED_FLAG_ALL       (SCHED_FLAG_RESET_ON_FORK       | \
                       SCHED_FLAG_RECLAIM             | \
kernel/sched/core.c +5 -1
index 0d18c3969f90..f2d5f7911855 100644
@@ -1943,6 +1943,10 @@ static void __setscheduler_uclamp(struct task_struct *p,
              uclamp_se_set(&p->uclamp_req[UCLAMP_MAX],
                            attr->sched_util_max, true);
      }
+
+     p->sched_reset_uclamp_on_fork = !!(attr->sched_flags &
+                                  SCHED_FLAG_RESET_UCLAMP_ON_FORK);
+
 }
 
 static void uclamp_fork(struct task_struct *p)
@@ -1956,7 +1960,7 @@ static void uclamp_fork(struct task_struct *p)
      for_each_clamp_id(clamp_id)
              p->uclamp[clamp_id].active = false;
 
-     if (likely(!p->sched_reset_on_fork))
+     if (likely(!p->sched_reset_on_fork && !p->sched_reset_uclamp_on_fork))
              return;
 
      for_each_clamp_id(clamp_id) {
tools/include/uapi/linux/sched.h +3 -2
index 3bac0a8ceab2..d52c59a2e0d0 100644
@@ -132,12 +132,14 @@ struct clone_args {
 #define SCHED_FLAG_KEEP_PARAMS               0x10
 #define SCHED_FLAG_UTIL_CLAMP_MIN    0x20
 #define SCHED_FLAG_UTIL_CLAMP_MAX    0x40
+#define SCHED_FLAG_RESET_UCLAMP_ON_FORK 0x80
 
 #define SCHED_FLAG_KEEP_ALL  (SCHED_FLAG_KEEP_POLICY | \
                               SCHED_FLAG_KEEP_PARAMS)
 
 #define SCHED_FLAG_UTIL_CLAMP        (SCHED_FLAG_UTIL_CLAMP_MIN | \
-                              SCHED_FLAG_UTIL_CLAMP_MAX)
+                              SCHED_FLAG_UTIL_CLAMP_MAX | \
+                              SCHED_FLAG_RESET_UCLAMP_ON_FORK)
 
 #define SCHED_FLAG_ALL       (SCHED_FLAG_RESET_ON_FORK       | \
                       SCHED_FLAG_RECLAIM             | \
--

2.40.0.634.g4ca3ef3211-goog

#13 lore raw
Quoted 86 lines · Saravana Kannan
Quoted 79 lines · Qais Yousef

Hi David!

Quoted 9 lines · David Dai

A userspace service may manage uclamp dynamically for individual tasks and a child task will unintentionally inherit a pesudo-random uclamp setting. This could result in the child task being stuck with a static uclamp value that results in poor performance or poor power.

Using SCHED_FLAG_RESET_ON_FORK is too coarse for this usecase and will reset other useful scheduler attributes. Adding a SCHED_FLAG_RESET_UCLAMP_ON_FORK will allow userspace to have finer control over scheduler attributes of child processes.

Thanks a lot for the patch. This has a been a known limitation for a while but didn't manage to find the time to push anything yet.

ADPF (Android Dynamic Performance Framework) exposes APIs to manage performance for a set of pids [1]. Only these tasks belong to the session and any forked tasked is expected to have its uclamp values reset. But as you pointed out, the current RESET_ON_FORK resets everything, but we don't want that as these attributes don't belong to ADPF to decide whether they should be reset too or not. And not resetting them means we can end up with tasks inheriting random uclamp values unintentionally. We can't tell these tasks not to fork anything. If the forked tasks are expected to be part of the session, then their pids must be added explicitly.

[1] https://developer.android.com/reference/android/os/PerformanceHintManager#createHintSession(int%5B%5D,%20long)

Quoted 17 lines
Cc Qais Yousef <[email protected]>
Cc Quentin Perret <[email protected]>
Cc Saravana Kannan <[email protected]>
---
 include/linux/sched.h            | 3 +++
 include/uapi/linux/sched.h       | 4 +++-
 kernel/sched/core.c              | 6 +++++-
 tools/include/uapi/linux/sched.h | 4 +++-
 4 files changed, 14 insertions(+), 3 deletions(-)
include/linux/sched.h +0 -0
index 63d242164b1a..b1676b9381f9 100644

include/linux/sched.h struct task_struct { · line 885

Quoted 1 line
include/linux/sched.h +0 -0
@@ -885,6 +885,9 @@ struct task_struct {
      unsigned                        sched_reset_on_fork:1;

nit: can't we convert to a flag and re-use?

Quoted 15 lines
include/linux/sched.h +3 -0
      unsigned                        sched_contributes_to_load:1;
      unsigned                        sched_migrated:1;
+#ifdef CONFIG_UCLAMP_TASK
+     unsigned                        sched_reset_uclamp_on_fork:1;
+#endif
 
      /* Force alignment to the next boundary: */
      unsigned                        :0;
include/uapi/linux/sched.h +0 -0
index 3bac0a8ceab2..7515106e1f1a 100644
@@ -132,12 +132,14 @@ struct clone_args {
 #define SCHED_FLAG_KEEP_PARAMS               0x10
 #define SCHED_FLAG_UTIL_CLAMP_MIN    0x20

include/uapi/linux/sched.h struct clone_args { · line 142

Quoted 10 lines
include/uapi/linux/sched.h +3 -1
 #define SCHED_FLAG_UTIL_CLAMP_MAX    0x40
+#define SCHED_FLAG_RESET_UCLAMP_ON_FORK      0x80
 
 #define SCHED_FLAG_KEEP_ALL  (SCHED_FLAG_KEEP_POLICY | \
                               SCHED_FLAG_KEEP_PARAMS)
 
 #define SCHED_FLAG_UTIL_CLAMP        (SCHED_FLAG_UTIL_CLAMP_MIN | \
-                              SCHED_FLAG_UTIL_CLAMP_MAX)
+                              SCHED_FLAG_UTIL_CLAMP_MAX | \
+                              SCHED_FLAG_RESET_UCLAMP_ON_FORK)

I was considering to have something a bit more generic that allows selecting which attributes to reset.

Quoted 5 lines

For example a syscall with SCHED_FLAG_RESET_ON_FORK_SEL combined with SCHED_FLAG_UCLAMP_MIN/MAX will only reset those. This should make it extensible if we have other similar use cases in the future. The downside it *might* require to be done in a separate syscall to the one that sets these parameter. But it should be done once.

Quoted 4 lines

In addition to the downside you mentioned, I'm not a huge fan of this suggestion since the meaning of the SCHED_FLAG_RESET_ON_FORK_SEL flag changes based on what other flags or attrs are set. I'd rather we have explicit flags.

The concern is that these flags are limited resources. latency_nice hopefully is coming and I don't see uclamp is an exception to warrant its own unique reset flag. Do you think we should never ever face similar exception again?

Quoted 7 lines

SCHED_FLAG_RESET_ON_FORK_SEL makes it harder to maintain the userspace code/makes it easy to accidentally introduce bugs. For example, a syscall could be setting UCLAMP_MIN and RESET_ON_FORK_SEL. Someone else might come and change the call to also set a nice value but not remember to split it up into two calls. Whereas with an explicit flag like David's proposal, we won't hit such an issue.

I think this mode of failure exists today and not new. You'll have to remember to set the right flag to keep policy etc otherwise you can end up with accidental effect.

That was the first suggestion comes to mind, it could be done another ways I suppose.

Quoted 3 lines

Also, we'll need to have separate flags internally to track what needs to be reset on fork vs not. So we really aren't saving anything by adding RESET_ON_FORK_SEL.

I don't get you here. Do you mean in kernel or userspace we'll have to track? I persume the former. It's just setting a flag in reset_on_fork variable. I don't see the problem.

If preserving the flag space is not a concern, then yeah potentially this is okay. Though in principle I think it doesn't make sense to continue to add new flag for every potential similar exception. History tends to repeat itself. I'm okay with keeping it simple if the maintainers don't share the concern about the flag space.

user_check_sched_setscheduler() prevents none privileged users from clearing reset_on_fork. Shouldn't we do the same?

Also we should make sure to clear it after the fork. Like is done for reset_on_fork.

Cheers

--
Qais Yousef

#14 lore raw
In reply to Qais Yousef

On 04/19/23 18:54, Qais Yousef wrote:

[...]

Quoted 9 lines

I was considering to have something a bit more generic that allows selecting which attributes to reset.

For example a syscall with SCHED_FLAG_RESET_ON_FORK_SEL combined with SCHED_FLAG_UCLAMP_MIN/MAX will only reset those. This should make it extensible if we have other similar use cases in the future. The downside it *might* require to be done in a separate syscall to the one that sets these parameter. But it should be done once.

Quoted 3 lines

Maybe there's a better interface, but I think it makes sense to do it in a way that we won't have to do this again. Would be good to hear from maintainers first before you take my word for it ;-)

Actually I think we can do a better and simpler generic interface. We don't need a new flag. We can just add a new parameter for what to reset on fork. When this value is 0 (which it should be by default), it means reset everything.

// pseudo code

#define RESET_ON_FORK_ALL 0
#define RESET_ON_FORK_POLICY BIT(1) // implies resetting priority
#define RESET_ON_FORK_PRIORITY BIT(2)
#define RESET_ON_FORK_UCLAMP BIT(3)

struct sched_attr {
	...
	__u64 sched_reset_on_fork_flags;
};

Cheers

--
Qais Yousef

#15 lore raw
Quoted 14 lines · Qais Yousef

On 04/19/23 18:54, Qais Yousef wrote:

[...]

Quoted 9 lines

I was considering to have something a bit more generic that allows selecting which attributes to reset.

For example a syscall with SCHED_FLAG_RESET_ON_FORK_SEL combined with SCHED_FLAG_UCLAMP_MIN/MAX will only reset those. This should make it extensible if we have other similar use cases in the future. The downside it *might* require to be done in a separate syscall to the one that sets these parameter. But it should be done once.

Quoted 8 lines
Quoted 3 lines

Maybe there's a better interface, but I think it makes sense to do it in a way that we won't have to do this again. Would be good to hear from maintainers first before you take my word for it ;-)

Actually I think we can do a better and simpler generic interface. We don't need a new flag. We can just add a new parameter for what to reset on fork. When this value is 0 (which it should be by default), it means reset everything.

Isn't he default NOT to reset everything?

Quoted 7 lines

// pseudo code

#define RESET_ON_FORK_ALL       0
#define RESET_ON_FORK_POLICY    BIT(1) // implies resetting priority
#define RESET_ON_FORK_PRIORITY  BIT(2)
#define RESET_ON_FORK_UCLAMP    BIT(3)
Quoted 5 lines
struct sched_attr {
        ...
        __u64 sched_reset_on_fork_flags;
};

Also, honestly I think this is over designing for a hypothetical. We have approximately 53 unused bits. By the time we run out of those, we'd have added at least 20-50 more fields. At that point, we can always add a flags2 field if we need it. I like David's patch as is -- it's clear and simple. Add a flag for explicitly what we are trying to do and extend as needed.

-Saravana

#16 lore raw
Quoted 19 lines · Saravana Kannan
Quoted 18 lines · Qais Yousef

On 04/19/23 18:54, Qais Yousef wrote:

[...]

Quoted 12 lines

I was considering to have something a bit more generic that allows selecting which attributes to reset.

For example a syscall with SCHED_FLAG_RESET_ON_FORK_SEL combined with SCHED_FLAG_UCLAMP_MIN/MAX will only reset those. This should make it extensible if we have other similar use cases in the future. The downside it *might* require to be done in a separate syscall to the one that sets these parameter. But it should be done once.

Maybe there's a better interface, but I think it makes sense to do it in a way that we won't have to do this again. Would be good to hear from maintainers first before you take my word for it ;-)

Quoted 6 lines
Quoted 4 lines

Actually I think we can do a better and simpler generic interface. We don't need a new flag. We can just add a new parameter for what to reset on fork. When this value is 0 (which it should be by default), it means reset everything.

Isn't he default NOT to reset everything?

The default when the RESET_ON_FORK flag is set. This field will not be used otherwise. Like what happens for the other params.

Quoted 8 lines
Quoted 8 lines

// pseudo code

#define RESET_ON_FORK_ALL       0
#define RESET_ON_FORK_POLICY    BIT(1) // implies resetting priority
#define RESET_ON_FORK_PRIORITY  BIT(2)
#define RESET_ON_FORK_UCLAMP    BIT(3)
Quoted 7 lines
Quoted 5 lines
struct sched_attr {
        ...
        __u64 sched_reset_on_fork_flags;
};

Also, honestly I think this is over designing for a hypothetical. We

latency_nice is coming next and most likely to require something similar. It's not hypothetical nor over designing. I think it's worthwhile spending time to plan for the future. More interfaces are confusing to the end users. And glibc already complained about evolution of sched_setattr, that's why we don't have a wrapper there yet (beside none of us pushed that hard to resolve the concerns due to lack of bandwidth).

https://public-inbox.org/libc-alpha/[email protected]/

(this thread reminded me linux-api must be CCed)

And there has been various discussions of the need of higher level wrappers/libraries that exposes simpler interface to app developers. So I'm actually expecting this to repeat. I think that was at LPC by Len Brown. I can find this thread on libc mailing list.

https://public-inbox.org/libc-alpha/CAMe9rOpUh1pjfEUqf_hNxce8ZX=4mg6W=n+BbdZSNFHLi7wtkw@mail.gmail.com/

These QoS hints might imply manipulating nice values and I can see ending up with a similar situation where we need to reset nice on fork without resetting other params.

Generally I don't think we should restrict users to self-managed model. A delegated model does make sense, and the latter implies the need for finer control on what to reset.

There's rtkit by the way which already an example of a delegating model to enable creating RT tasks by non privileged users.

Should rtkit force resetting uclamp when on fork? I think it's a grey area and I learn towards it shouldn't.

Quoted 5 lines

have approximately 53 unused bits. By the time we run out of those, we'd have added at least 20-50 more fields. At that point, we can always add a flags2 field if we need it. I like David's patch as is -- it's clear and simple. Add a flag for explicitly what we are trying to do and extend as needed.

Fair enough. As I said if the maintainers are okay with current proposal, no objection from my side. Based on my experience I didn't expect them to be. And I do think a generic solution is not really complicated and is the better option. You can consider this as a backup plan ;-)

Cheers

--
Qais Yousef

#17 lore raw
Quoted 45 lines · Qais Yousef
Quoted 19 lines · Saravana Kannan
Quoted 18 lines · Qais Yousef

On 04/19/23 18:54, Qais Yousef wrote:

[...]

Quoted 12 lines

I was considering to have something a bit more generic that allows selecting which attributes to reset.

For example a syscall with SCHED_FLAG_RESET_ON_FORK_SEL combined with SCHED_FLAG_UCLAMP_MIN/MAX will only reset those. This should make it extensible if we have other similar use cases in the future. The downside it *might* require to be done in a separate syscall to the one that sets these parameter. But it should be done once.

Maybe there's a better interface, but I think it makes sense to do it in a way that we won't have to do this again. Would be good to hear from maintainers first before you take my word for it ;-)

Quoted 6 lines
Quoted 4 lines

Actually I think we can do a better and simpler generic interface. We don't need a new flag. We can just add a new parameter for what to reset on fork. When this value is 0 (which it should be by default), it means reset everything.

Isn't he default NOT to reset everything?

The default when the RESET_ON_FORK flag is set. This field will not be used otherwise. Like what happens for the other params.

Quoted 14 lines
Quoted 13 lines

// pseudo code

#define RESET_ON_FORK_ALL       0
#define RESET_ON_FORK_POLICY    BIT(1) // implies resetting priority
#define RESET_ON_FORK_PRIORITY  BIT(2)
#define RESET_ON_FORK_UCLAMP    BIT(3)
struct sched_attr {
        ...
        __u64 sched_reset_on_fork_flags;
};
Quoted 8 lines
Quoted 1 line

Also, honestly I think this is over designing for a hypothetical. We

latency_nice is coming next and most likely to require something similar. It's not hypothetical nor over designing. I think it's worthwhile spending time to plan for the future. More interfaces are confusing to the end users. And glibc already complained about evolution of sched_setattr, that's why we don't have a wrapper there yet (beside none of us pushed that hard to resolve the concerns due to lack of bandwidth).

My comment about over designing/hypothetical is about the "running out of bits" concern you are trying to solve. Totally agree that we'll likely need finer reset control and never disagreed with that. But we can add those as we go to the existing flag field instead of introducing a new one.

Quoted 19 lines

https://public-inbox.org/libc-alpha/[email protected]/

(this thread reminded me linux-api must be CCed)

And there has been various discussions of the need of higher level wrappers/libraries that exposes simpler interface to app developers. So I'm actually expecting this to repeat. I think that was at LPC by Len Brown. I can find this thread on libc mailing list.

https://public-inbox.org/libc-alpha/CAMe9rOpUh1pjfEUqf_hNxce8ZX=4mg6W=n+BbdZSNFHLi7wtkw@mail.gmail.com/

These QoS hints might imply manipulating nice values and I can see ending up with a similar situation where we need to reset nice on fork without resetting other params.

Generally I don't think we should restrict users to self-managed model. A delegated model does make sense, and the latter implies the need for finer control on what to reset.

Quoted 5 lines

There's rtkit by the way which already an example of a delegating model to enable creating RT tasks by non privileged users.

Should rtkit force resetting uclamp when on fork? I think it's a grey area and I learn towards it shouldn't.

Nothing to say to any of the stuff above because I didn't disagree on this in the first place.

Quoted 10 lines
Quoted 5 lines

have approximately 53 unused bits. By the time we run out of those, we'd have added at least 20-50 more fields. At that point, we can always add a flags2 field if we need it. I like David's patch as is -- it's clear and simple. Add a flag for explicitly what we are trying to do and extend as needed.

Fair enough. As I said if the maintainers are okay with current proposal, no objection from my side. Based on my experience I didn't expect them to be. And I do think a generic solution is not really complicated and is the better option. You can consider this as a backup plan ;-)

I think the fact I could misunderstand the API and you had to explain it makes it a bit more complicated (agree, not a lot) over just using the existing bits. And I think that's what the maintainers are pushing back on (in addition to even having finer reset control).

-Saravana