Skip to contentSkip to conversation

x86/resctrl: Add task resctrl information display

PATCH Superseded Chen Yu 3 participants 3 messages Jan 10, 2020 – Jan 14, 2020

Subsystems (4)

v6 of 7 Superseded Latest is v7 →

lore
Share & tools
lore

Chen Yu

author
#0 lore raw

PATCH v6 1/1 x86/resctrl: Add task resctrl information display

Monitoring tools that want to find out which resctrl control and monitor groups a task belongs to must currently read the "tasks" file in every group until they locate the process ID.

Add an additional file /proc/{pid}/cpu_resctrl to provide this information.

The output is as followed, for example:

1) ""
Resctrl is not available.

2) "/"
Task is part of the root group, task is not associated to any monitor group.

3) "/mon_groups/mon0"
Task is part of the root group and monitor group mon0.

4) "/group0"
Task is part of resctrl control group group0, task is not associated to any monitor group.

5) "/group0/mon_groups/mon1"
Task is part of resctrl control group group0 and monitor group mon1.

Cc Thomas Gleixner <[email protected]>
Cc Borislav Petkov <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc "H. Peter Anvin" <[email protected]>
Cc Alexey Dobriyan <[email protected]>
Cc Andrew Morton <[email protected]>
Cc Reinette Chatre <[email protected]>
Cc Fenghua Yu <[email protected]>
Cc Tony Luck <[email protected]>
Cc Chris Down <[email protected]>
Cc Michal Hocko <[email protected]>
Cc Linus Torvalds <[email protected]>

---
v1: Initial version reviewed by Reinette Chatre,
Fenghua Yu and Tony Luck.

v2: According to Boris's suggestion,
reduce indentation level in proc_resctrl_show(). Create the include/linux/resctrl.h header and declare proc_resctrl_show() in this file, so that other architectures would probably use it in the future. Different architectures should implement architectural specific proc_resctrl_show() accordingly.

v3: According to Boris's suggestion,
Return empty string if the resctrl filesystem has not been mounted.
Rename the config from CPU_RESCTRL to PROC_CPU_RESCTRL to better represent its usage. Move PROC_CPU_RESCTRL from arch/Kconfig to fs/proc/Kconfig.
And let PROC_CPU_RESCTRL to be depended on PROC_FS.

v4: According to Thomas's suggestion, changed the output
from multiple lines to one single line.

v5: According to Alexey's feedback, removed the header file
proc_fs.h in resctrl.h, and changed seq_puts() to seq_putc() for simplicity.

v6: According to Chris Down's suggestion,
1. rename:
/proc/{pid}/resctrl to /proc/{pid}/cpu_resctrl to better reflect its meaning.
2. change the description in comments:
"control group" to "resctrl control group" as the former is confusing for cgroup users.

---
 arch/x86/Kconfig                       |  1 +
 arch/x86/kernel/cpu/resctrl/rdtgroup.c | 79 ++++++++++++++++++++++++++
 fs/proc/Kconfig                        |  4 ++
 fs/proc/base.c                         |  7 +++
 include/linux/resctrl.h                | 14 +++++
 5 files changed, 105 insertions(+)
 create mode 100644 include/linux/resctrl.h
arch/x86/Kconfig +1 -0
index 5e8949953660..6e17a68c7d77 100644
@@ -456,6 +456,7 @@ config X86_CPU_RESCTRL
 	bool "x86 CPU resource control support"
 	depends on X86 && (CPU_SUP_INTEL || CPU_SUP_AMD)
 	select KERNFS
+	select PROC_CPU_RESCTRL		if PROC_FS
 	help
 	  Enable x86 CPU resource control support.
 
arch/x86/kernel/cpu/resctrl/rdtgroup.c +79 -0

85 lines not on this page · raw message

fs/proc/Kconfig +4 -0
index 733881a6387b..27ef84d99f59 100644
@@ -103,3 +103,7 @@ config PROC_CHILDREN
 config PROC_PID_ARCH_STATUS
 	def_bool n
 	depends on PROC_FS
+
+config PROC_CPU_RESCTRL
+	def_bool n
+	depends on PROC_FS
fs/proc/base.c +7 -0
index ebea9501afb8..32c9ff154667 100644
@@ -94,6 +94,7 @@
 #include <linux/sched/debug.h>
 #include <linux/sched/stat.h>
 #include <linux/posix-timers.h>
+#include <linux/resctrl.h>
 #include <trace/events/oom.h>
 #include "internal.h"
 #include "fd.h"
@@ -3060,6 +3061,9 @@ static const struct pid_entry tgid_base_stuff[] = {
 #endif
 #ifdef CONFIG_CGROUPS
 	ONE("cgroup",  S_IRUGO, proc_cgroup_show),
+#endif
+#ifdef CONFIG_PROC_CPU_RESCTRL
+	ONE("cpu_resctrl", S_IRUGO, proc_resctrl_show),
 #endif
 	ONE("oom_score",  S_IRUGO, proc_oom_score),
 	REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
@@ -3460,6 +3464,9 @@ static const struct pid_entry tid_base_stuff[] = {
 #endif
 #ifdef CONFIG_CGROUPS
 	ONE("cgroup",  S_IRUGO, proc_cgroup_show),
+#endif
+#ifdef CONFIG_PROC_CPU_RESCTRL
+	ONE("cpu_resctrl", S_IRUGO, proc_resctrl_show),
 #endif
 	ONE("oom_score", S_IRUGO, proc_oom_score),
 	REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
include/linux/resctrl.h new file +14 -0
new file mode 100644
index 000000000000..daf5cf64c6a6
@@ -0,0 +1,14 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _RESCTRL_H
+#define _RESCTRL_H
+
+#ifdef CONFIG_PROC_CPU_RESCTRL
+
+int proc_resctrl_show(struct seq_file *m,
+		      struct pid_namespace *ns,
+		      struct pid *pid,
+		      struct task_struct *tsk);
+
+#endif
+
+#endif /* _RESCTRL_H */
Signature · 1 line
2.17.1
maintainer
#1 lore raw
Quoted 20 lines · Chen Yu

Monitoring tools that want to find out which resctrl control and monitor groups a task belongs to must currently read the "tasks" file in every group until they locate the process ID.

Add an additional file /proc/{pid}/cpu_resctrl to provide this information.

The output is as followed, for example:

1) ""
Resctrl is not available.

2) "/"
Task is part of the root group, task is not associated to any monitor group.

3) "/mon_groups/mon0"
Task is part of the root group and monitor group mon0.

Quoted 7 lines

4) "/group0"
Task is part of resctrl control group group0, task is not associated to any monitor group.

5) "/group0/mon_groups/mon1"
Task is part of resctrl control group group0 and monitor group mon1.

So this way to present the information is totally non-intuitive, IMNSVHO. What's wrong with:

1)
res_group:
mon_group:

2)
res_group: /
mon_group:

3)
res_group: /
mon_group: mon0

4)
res_group: group0
mon_group:

5)
res_group: group0
mon_group: mon1

?

You can even call the file "cpu_resctrl_groups" so that it is clear that it will dump groups and then do:

res: group0
mon: mon1

which is both human-readable and easily greppable.

Quoted 5 lines
arch/x86/kernel/cpu/resctrl/rdtgroup.c +0 -0
index 2e3b06d6bbc6..dcbf62d6b689 100644

arch/x86/kernel/cpu/resctrl/rdtgroup.c static int rdtgroup_tasks_show(struct kernfs_open_file *of, · line 734

Quoted 10 lines
arch/x86/kernel/cpu/resctrl/rdtgroup.c +7 -0
@@ -725,6 +725,85 @@ static int rdtgroup_tasks_show(struct kernfs_open_file *of,
 	return ret;
 }
 
+#ifdef CONFIG_PROC_CPU_RESCTRL
+
+/*
+ * A task can only be part of one resctrl
+ * control group and of one monitor
+ * group which is associated to that resctrl
+ * control group.

Extend those comments to 80 cols.

arch/x86/kernel/cpu/resctrl/rdtgroup.c static int rdtgroup_tasks_show(struct kernfs_open_file *of, · line 735

Quoted 1 line
arch/x86/kernel/cpu/resctrl/rdtgroup.c +1 -0
+ * So one line is simple and clear enough:

Actually, the one line format you've done is confusing and can be done much more human- and tool-readable.

Thx.

Signature · 4 lines
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette
#2 lore raw

Hi Boris,

Quoted 55 lines · Borislav Petkov
Quoted 20 lines · Chen Yu

Monitoring tools that want to find out which resctrl control and monitor groups a task belongs to must currently read the "tasks" file in every group until they locate the process ID.

Add an additional file /proc/{pid}/cpu_resctrl to provide this information.

The output is as followed, for example:

1) ""
Resctrl is not available.

2) "/"
Task is part of the root group, task is not associated to any monitor group.

3) "/mon_groups/mon0"
Task is part of the root group and monitor group mon0.

Quoted 7 lines

4) "/group0"
Task is part of resctrl control group group0, task is not associated to any monitor group.

5) "/group0/mon_groups/mon1"
Task is part of resctrl control group group0 and monitor group mon1.

So this way to present the information is totally non-intuitive, IMNSVHO. What's wrong with:

1)
res_group:
mon_group:

2)
res_group: /
mon_group:

3)
res_group: /
mon_group: mon0

4)
res_group: group0
mon_group:

5)
res_group: group0
mon_group: mon1

?

Quoted 8 lines

You can even call the file "cpu_resctrl_groups" so that it is clear that it will dump groups and then do:

res: group0
mon: mon1

which is both human-readable and easily greppable.

Yes, to display resctrl control and monitor group separately might be more friendly to the user. Although I was thinking if the user would like to see the full path of
the resource, which might make it easier to be parsed:
A) res: group0
mon: mon1
vs
B) res: /group0
mon: /group0/mon_groups/mon1
as proposal B might introduce duplication I'll send a new version based on proposal A.

Quoted 8 lines
Quoted 5 lines

+/*
+ * A task can only be part of one resctrl
+ * control group and of one monitor
+ * group which is associated to that resctrl
+ * control group.

Extend those comments to 80 cols.

Okay. will do.

Quoted 5 lines
Quoted 1 line

+ * So one line is simple and clear enough:

Actually, the one line format you've done is confusing and can be done much more human- and tool-readable.

Got it.
Thanks,
Chenyu

Quoted 7 lines

Thx.

--
Regards/Gruss,
Boris.

https://people.kernel.org/tglx/notes-about-netiquette