Add task resctrl information in procfs
Subsystems (4)
v2 cover Add task resctrl information in procfs
Expose the resctrl information in /proc/{pid}/resctrl so that the user space tool is able to locate the resctrl directory path in O(1) rather than O(n) - n is the number of tasks in the system.
Chen Yu (2):
resctrl: Add CPU_RESCTRL
x86/resctrl: Add task resctrl information display
arch/Kconfig | 4 +++ arch/x86/Kconfig | 1 + arch/x86/kernel/cpu/resctrl/rdtgroup.c | 47 ++++++++++++++++++++++++++ fs/proc/base.c | 7 ++++ include/linux/resctrl.h | 16 +++++++++ 5 files changed, 75 insertions(+) create mode 100644 include/linux/resctrl.h
Signature · 1 line
2.17.1
PATCH v2 1/2 resctrl: Add CPU_RESCTRL
Introduce a generic option called CPU_RESCTRL which is selected by the arch-specific ones CONFIG_X86_RESCTRL or CONFIG_ARM64_RESCTRL in the future. The generic one will cover the resctrl filesystem and other generic and shared bits of functionality.
--- arch/Kconfig | 4 ++++ arch/x86/Kconfig | 1 + 2 files changed, 5 insertions(+)
arch/Kconfig +4 -0
@@ -960,6 +960,10 @@ config RELR config ARCH_HAS_MEM_ENCRYPT bool +config CPU_RESCTRL + bool + def_bool n + source "kernel/gcov/Kconfig" source "scripts/gcc-plugins/Kconfig"
arch/x86/Kconfig +1 -0
@@ -455,6 +455,7 @@ config X86_CPU_RESCTRL bool "x86 CPU resource control support" depends on X86 && (CPU_SUP_INTEL || CPU_SUP_AMD) select KERNFS + select CPU_RESCTRL help Enable x86 CPU resource control support.
Signature · 1 line
2.17.1
Borislav Petkov
maintainerRe: [PATCH 1/2][v2] resctrl: Add CPU_RESCTRL
Quoted 7 lines · Chen Yu
Introduce a generic option called
CPU_RESCTRLwhich is selected by the arch-specific onesCONFIG_X86_RESCTRLorCONFIG_ARM64_RESCTRLin the future. The generic one will cover the resctrl filesystem and other generic and shared bits of functionality.Suggested-by Borislav Petkov <[email protected]>
I don't remember suggesting adding a separate CONFIG option...
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquettePATCH v2 2/2 x86/resctrl: Add task resctrl information display
In reply to Chen YuMonitoring tools that want to find out which resctrl CTRL 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}/resctrl to provide this information.
For example:
cat /proc/1193/resctrl
CTRL_MON:/ctrl_grp0
MON:/ctrl_grp0/mon_groups/mon_grp0
If the resctrl filesystem has not been mounted,
reading /proc/{pid}/resctrl returns an error:
cat: /proc/1193/resctrl: No such device
---
v2: Reduce indentation level in proc_resctrl_show()
according to Boris's suggestion.
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.
--- arch/x86/kernel/cpu/resctrl/rdtgroup.c | 47 ++++++++++++++++++++++++++ fs/proc/base.c | 7 ++++ include/linux/resctrl.h | 16 +++++++++ 3 files changed, 70 insertions(+) create mode 100644 include/linux/resctrl.h
arch/x86/kernel/cpu/resctrl/rdtgroup.c +47 -0
@@ -727,6 +727,53 @@ static int rdtgroup_tasks_show(struct kernfs_open_file *of,return ret; } +int proc_resctrl_show (struct seq_file *s,struct pid_namespace *ns, +struct pid *pid,struct task_struct *tsk) +{ +struct rdtgroup *rdtg; +int ret =0 ; + +mutex_lock (&rdtgroup_mutex); + +/* Make sure resctrl has been mounted. */ +if (!static_branch_unlikely (&rdt_enable_key)) { + ret = -ENODEV ; + goto unlock; + } + +list_for_each_entry (rdtg, &rdt_all_groups, rdtgroup_list) { +struct rdtgroup *crg; + +/* +* Task information is only relevant for shareable +* and exclusive groups. +*/ +if (rdtg->mode !=RDT_MODE_SHAREABLE && + rdtg->mode !=RDT_MODE_EXCLUSIVE ) +continue ; + +if (rdtg->closid != tsk->closid) +continue ; + +seq_printf (s,"CTRL_MON:/%s\n" , rdtg->kn->name); +list_for_each_entry (crg, &rdtg->mon.crdtgrp_list, + mon.crdtgrp_list) { +if (tsk->rmid != crg->mon.rmid) +continue ; +seq_printf (s,"MON:%s%s/mon_groups/%s\n" , + rdtg == &rdtgroup_default ?"" :"/" , + rdtg->kn->name, crg->kn->name); + goto unlock; + } + goto unlock; + } + ret = -ENOENT ; +unlock: +mutex_unlock (&rdtgroup_mutex); + +return ret; +} +static int rdt_last_cmd_status_show (struct kernfs_open_file *of,struct seq_file *seq,void *v) {
fs/proc/base.c +7 -0
@@ -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_CPU_RESCTRL +ONE ("resctrl" ,S_IRUGO , proc_resctrl_show), #endifONE ("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_CPU_RESCTRL +ONE ("resctrl" ,S_IRUGO , proc_resctrl_show), #endifONE ("oom_score" ,S_IRUGO , proc_oom_score), REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations),
include/linux/resctrl.h new file +16 -0
@@ -0,0 +1,16 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +#ifndef _CPU_RESCTRL_H +#define _CPU_RESCTRL_H + +#ifdef CONFIG_CPU_RESCTRL + +#include <linux/proc_fs.h> + +int proc_resctrl_show (struct seq_file *m, +struct pid_namespace *ns, +struct pid *pid, +struct task_struct *tsk); + +#endif + +#endif /* _CPU_RESCTRL_H */
Signature · 1 line
2.17.1
Borislav Petkov
maintainerRe: [PATCH 2/2][v2] x86/resctrl: Add task resctrl information display
Quoted 8 lines · Chen Yu
Monitoring tools that want to find out which resctrl CTRL 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}/resctrlto provide this information.
Quoted 8 lines
For example:
cat/proc/1193/resctrl
CTRL_MON:/ctrl_grp0
MON:/ctrl_grp0/mon_groups/mon_grp0If the resctrl filesystem has not been mounted, reading
/proc/{pid}/resctrlreturns an error:
cat:/proc/1193/resctrl: No such device
Eww, this doesn't sound very user-friendly. How is the user supposed to know that the resctrl fs needs to be mounted for this to work?
Why does the resctrl fs need to be mounted at all to show this?
I'm guessing if it is not mounted, you have no groups so you don't have to return an error - you simply return "". Right?
Quoted 4 lines
Tested-by Jinshi Chen <[email protected]>Reviewed-by Reinette Chatre <[email protected]>Reviewed-by Fenghua Yu <[email protected]>Reviewed-by Tony Luck <[email protected]>
When you send a new version which has non-trivial changes, you should drop those tags because they don't apply anymore. Unless those people have managed to review and test the new version ...
Looking at CONFIG_PROC_PID_ARCH_STATUS for an example of proc/ calling arch-specific functions, I guess you need to do:
select CPU_RESCTRL if PROC_FS
Thx.
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquetteRe: [PATCH 2/2][v2] x86/resctrl: Add task resctrl information display
Hi Boris,
Quoted 19 lines · Borislav Petkov
Quoted 8 lines · Chen Yu
Monitoring tools that want to find out which resctrl CTRL 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}/resctrlto provide this information.Quoted 8 lines
For example:
cat/proc/1193/resctrl
CTRL_MON:/ctrl_grp0
MON:/ctrl_grp0/mon_groups/mon_grp0If the resctrl filesystem has not been mounted, reading
/proc/{pid}/resctrlreturns an error:
cat:/proc/1193/resctrl: No such device
Quoted 8 lines
Eww, this doesn't sound very user-friendly. How is the user supposed to know that the resctrl fs needs to be mounted for this to work?
Why does the resctrl fs need to be mounted at all to show this?
I'm guessing if it is not mounted, you have no groups so you don't have to return an error - you simply return "". Right?
Right, we can return 'blank' to user and let the user to parse the information. And there is a similar behavior in cgroup that, for kernel thread that does not belong
to any cgroup, /proc/{pid}/cgroup just show 'blank' without returning an error.
Quoted 9 lines
Quoted 4 lines
Tested-by Jinshi Chen <[email protected]>Reviewed-by Reinette Chatre <[email protected]>Reviewed-by Fenghua Yu <[email protected]>Reviewed-by Tony Luck <[email protected]>When you send a new version which has non-trivial changes, you should drop those tags because they don't apply anymore. Unless those people have managed to review and test the new version ...
Got it, thanks for telling me this.
Quoted 5 lines
Looking at
CONFIG_PROC_PID_ARCH_STATUSfor an example of proc/ calling arch-specific functions, I guess you need to do:select
CPU_RESCTRLifPROC_FS
Yes, only when PROC_FS is set, /proc/{pid}/resctrl
can be displayed. However, CPU_RESCTRL might not
depend on proc fs, it is possible that the CPU_RESCTRL
is enabled but without PROC_FS set. If I understand correctly,
CPU_RESCTRL is the 'root' config for X86_CPU_RESCTRL,
after reading this thread:
https://lists.gt.net/linux/kernel/3211659
If this is the case, shall we add the new file at kernel/resctrl/resctrl.c?
And the generic proc_resctrl_show() could be put into this file. In the future the generic code for resctrl could be added/moved to kernel/resctrl/resctrl.c
Thanks,
Chenyu
Quoted 7 lines
Thx.
--
Regards/Gruss,
Boris.
Borislav Petkov
maintainerRe: [PATCH 2/2][v2] x86/resctrl: Add task resctrl information display
Quoted 1 line · Ryan Chen
Right, we can return 'blank' to user and let the user to parse the information.
There is nothing to parse - the task doesn't belong to any groups. That's it.
Quoted 3 lines
And there is a similar behavior in cgroup that, for kernel thread that does not belong
to any cgroup,/proc/{pid}/cgroupjust show 'blank' without returning an error.
By 'blank' I assume you mean the empty string '' ?
Quoted 7 lines
Yes, only when
PROC_FSis set,/proc/{pid}/resctrlcan be displayed. However,CPU_RESCTRLmight not depend on proc fs, it is possible that theCPU_RESCTRLis enabled but withoutPROC_FSset. If I understand correctly,
CPU_RESCTRLis the 'root' config forX86_CPU_RESCTRL,
after reading this thread:
https://lists.gt.net/linux/kernel/3211659
I'm not sure I know what you mean here. There's no CPU_RESCTRL option - you've added it in the previous patch:
[ ~/kernel/linux> git grep -E CONFIG_CPU_RESCTRL [ ~/kernel/linux> git grep -E "\WCPU_RESCTRL"
[ ~/kernel/linux>
And if you want to use that option in proc/, then it needs to depend on PROC_FS, like the the example I gave you with CONFIG_PROC_PID_ARCH_STATUS.
Or do you mean something else?
Quoted 3 lines
If this is the case, shall we add the new file at
kernel/resctrl/resctrl.c?
And the genericproc_resctrl_show()could be put into this file. In the future the generic code for resctrl could be added/moved tokernel/resctrl/resctrl.c
Not worth it for a single function. Leave it in arch/x86/kernel/cpu/resctrl/rdtgroup.c where you had it.
Thx.
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquetteRe: [PATCH 2/2][v2] x86/resctrl: Add task resctrl information display
Hi Boris,
Quoted 4 lines · Borislav Petkov
Quoted 1 line · Ryan Chen
Right, we can return 'blank' to user and let the user to parse the information.
Quoted 8 lines
There is nothing to parse - the task doesn't belong to any groups. That's it.
Quoted 3 lines
And there is a similar behavior in cgroup that, for kernel thread that does not belong
to any cgroup,/proc/{pid}/cgroupjust show 'blank' without returning an error.By 'blank' I assume you mean the empty string '' ?
Yes, it is 'empty string'.
Quoted 15 lines
Quoted 7 lines
Yes, only when
PROC_FSis set,/proc/{pid}/resctrlcan be displayed. However,CPU_RESCTRLmight not depend on proc fs, it is possible that theCPU_RESCTRLis enabled but withoutPROC_FSset. If I understand correctly,
CPU_RESCTRLis the 'root' config forX86_CPU_RESCTRL,
after reading this thread:
https://lists.gt.net/linux/kernel/3211659I'm not sure I know what you mean here. There's no
CPU_RESCTRLoption - you've added it in the previous patch:[ ~/kernel/linux> git grep -E
CONFIG_CPU_RESCTRL[ ~/kernel/linux> git grep -E "\WCPU_RESCTRL"
[ ~/kernel/linux>
Quoted 6 lines
And if you want to use that option in proc/, then it needs to depend on
PROC_FS, like the the example I gave you withCONFIG_PROC_PID_ARCH_STATUS.Or do you mean something else?
Previously I was thinking of introducing CONFIG_CPU_RESCTRL
to the kernel, so that platform-independent resctrl code could be moved under this config. The idea was seen in the following commit log:
commit e6d429313ea5c776d (x86/resctrl: Avoid confusion over the new X86_RESCTRL config)
But since we only touch proc fs in this patch, I think a config named CONFIG_PROC_CPU_RESCTRL declared in fs/proc/Kconfig
might be more appropriate(similar to CONFIG_PROC_PID_ARCH_STATUS).
Quoted 7 lines
Quoted 3 lines
If this is the case, shall we add the new file at
kernel/resctrl/resctrl.c?
And the genericproc_resctrl_show()could be put into this file. In the future the generic code for resctrl could be added/moved tokernel/resctrl/resctrl.cNot worth it for a single function. Leave it in
arch/x86/kernel/cpu/resctrl/rdtgroup.cwhere you had it.
Okay, got it. Let me send the v3 patch out and to see if it is suitable. Thanks,
Chenyu
Quoted 7 lines
Thx.
--
Regards/Gruss,
Boris.