[PATCH 0/2] measure latency of cpu hotplug path

[PATCH 0/2] measure latency of cpu hotplug path

[Prasad Sodagudi]

There are all changes related to cpu hotplug path and would like to seek
upstream review. These are all patches in Qualcomm downstream kernel
for a quite long time. First patch sets the rt prioity to hotplug
task and second patch adds cpuhp trace events.

1) cpu-hotplug: Always use real time scheduling when hotplugging a CPU
2) cpu/hotplug: Add cpuhp_latency trace event

Example logs:-  
cpu online -
         cpuhp/4-200   [004] ....   223.891886: cpuhp_enter: cpu: 0004 target: 213 step: 212 (sched_cpu_activate)
         cpuhp/4-200   [004] ....   223.891894: cpuhp_exit:  cpu: 0004  state: 212 step: 212 ret: 0
              sh-176   [000] ....   223.891912: cpuhp_exit:  cpu: 0004  state: 213 step:  86 ret: 0
              sh-176   [000] ....   223.891915: cpuhp_latency:  cpu:4 state:online latency:3874 USEC ret: 0

cpu offline - 
              sh-176   [000] ....   265.193490: cpuhp_exit:  cpu: 0004  state:   2 step:   2 ret: 0
              sh-176   [000] ....   265.193494: cpuhp_latency:  cpu:4 state:offline latency:57431 USEC ret: 0


Prasad Sodagudi (1):
  cpu/hotplug: Add cpuhp_latency trace event

Syed Rameez Mustafa (1):
  cpu-hotplug: Always use real time scheduling when hotplugging a CPU

 include/trace/events/cpuhp.h | 29 +++++++++++++++++++++++++
 kernel/cpu.c                 | 50 ++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 79 insertions(+)

-- 
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project

[PATCH 1/2] cpu/hotplug: Add cpuhp_latency trace event

[Prasad Sodagudi]

Add ftrace event trace_cpuhp_latency to track cpu
hotplug latency. It helps to track the hotplug latency
impact by firmware changes and kernel cpu hotplug callbacks.

Signed-off-by: Prasad Sodagudi <psodagud@codeaurora.org>
---
 include/trace/events/cpuhp.h | 29 +++++++++++++++++++++++++++++
 kernel/cpu.c                 |  9 +++++++++
 2 files changed, 38 insertions(+)

diff --git a/include/trace/events/cpuhp.h b/include/trace/events/cpuhp.h
index ad16f77..c871850 100644
--- a/include/trace/events/cpuhp.h
+++ b/include/trace/events/cpuhp.h
@@ -6,6 +6,7 @@
 #define _TRACE_CPUHP_H
 
 #include <linux/tracepoint.h>
+#include <linux/sched/clock.h>
 
 TRACE_EVENT(cpuhp_enter,
 
@@ -89,6 +90,34 @@ TRACE_EVENT(cpuhp_exit,
 		  __entry->cpu, __entry->state, __entry->idx,  __entry->ret)
 );
 
+TRACE_EVENT(cpuhp_latency,
+
+	TP_PROTO(unsigned int cpu, unsigned int state,
+		 u64 start_time,  int ret),
+
+	TP_ARGS(cpu, state, start_time, ret),
+
+	TP_STRUCT__entry(
+		__field(unsigned int,	cpu)
+		__field(unsigned int,	state)
+		__field(u64,		time)
+		__field(int,		ret)
+	),
+
+	TP_fast_assign(
+		__entry->cpu	= cpu;
+		__entry->state	= state;
+		__entry->time	= div64_u64(sched_clock() - start_time, 1000);
+		__entry->ret	= ret;
+	),
+
+	TP_printk(" cpu:%d state:%s latency:%llu USEC ret: %d",
+		__entry->cpu, __entry->state ? "online" : "offline",
+		__entry->time, __entry->ret)
+);
+
+
+
 #endif
 
 /* This part must be outside protection */
diff --git a/kernel/cpu.c b/kernel/cpu.c
index 6ff2578..68b3740 100644
--- a/kernel/cpu.c
+++ b/kernel/cpu.c
@@ -36,6 +36,7 @@
 #include <trace/events/power.h>
 #define CREATE_TRACE_POINTS
 #include <trace/events/cpuhp.h>
+#include <linux/sched/clock.h>
 
 #include "smpboot.h"
 
@@ -994,6 +995,7 @@ static int __ref _cpu_down(unsigned int cpu, int tasks_frozen,
 {
 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
 	int prev_state, ret = 0;
+	u64 start_time = 0;
 
 	if (num_online_cpus() == 1)
 		return -EBUSY;
@@ -1002,6 +1004,8 @@ static int __ref _cpu_down(unsigned int cpu, int tasks_frozen,
 		return -EINVAL;
 
 	cpus_write_lock();
+	if (trace_cpuhp_latency_enabled())
+		start_time = sched_clock();
 
 	cpuhp_tasks_frozen = tasks_frozen;
 
@@ -1040,6 +1044,7 @@ static int __ref _cpu_down(unsigned int cpu, int tasks_frozen,
 	}
 
 out:
+	trace_cpuhp_latency(cpu, 0, start_time, ret);
 	cpus_write_unlock();
 	/*
 	 * Do post unplug cleanup. This is still protected against
@@ -1192,8 +1197,11 @@ static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target)
 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
 	struct task_struct *idle;
 	int ret = 0;
+	u64 start_time = 0;
 
 	cpus_write_lock();
+	if (trace_cpuhp_latency_enabled())
+		start_time = sched_clock();
 
 	if (!cpu_present(cpu)) {
 		ret = -EINVAL;
@@ -1241,6 +1249,7 @@ static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target)
 	target = min((int)target, CPUHP_BRINGUP_CPU);
 	ret = cpuhp_up_callbacks(cpu, st, target);
 out:
+	trace_cpuhp_latency(cpu, 1, start_time, ret);
 	cpus_write_unlock();
 	arch_smt_update();
 	return ret;
-- 
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project

Re: [PATCH 1/2] cpu/hotplug: Add cpuhp_latency trace event

[Thomas Gleixner]

On Wed, Sep 23 2020 at 16:37, Prasad Sodagudi wrote:
> Add ftrace event trace_cpuhp_latency to track cpu
> hotplug latency. It helps to track the hotplug latency
> impact by firmware changes and kernel cpu hotplug callbacks.

Why?

And even if that makes sense, the implementation makes absolutely no
sense at all.

      trace_cpuhp_callback_enter();
      callback();
      trace_cpuhp_callback_exit();

No point in all this start time and conditional sched_clock() muck.

But then again, hotplug is slow by definition and nobody cares.

Thanks,

        tglx

[PATCH 2/2] cpu-hotplug: Always use real time scheduling when hotplugging a CPU

[Prasad Sodagudi]

From: Syed Rameez Mustafa <rameezmustafa@codeaurora.org>

CPU hotplug operations take place in preemptible context. This leaves
the hotplugging thread at the mercy of overall system load and CPU
availability. If the hotplugging thread does not get an opportunity
to execute after it has already begun a hotplug operation, CPUs can
end up being stuck in a quasi online state. In the worst case a CPU
can be stuck in a state where the migration thread is parked while
another task is executing and changing affinity in a loop. This
combination can result in unbounded execution time for the running
task until the hotplugging thread gets the chance to run to complete
the hotplug operation.

Fix the said problem by ensuring that hotplug can only occur from
threads belonging to the RT sched class. This allows the hotplugging
thread priority on the CPU no matter what the system load or the
number of available CPUs are. If a SCHED_NORMAL task attempts to
hotplug a CPU, we temporarily elevate it's scheduling policy to RT.
Furthermore, we disallow hotplugging operations to begin if the
calling task belongs to the idle and deadline classes or those that
use the SCHED_BATCH policy.

Signed-off-by: Syed Rameez Mustafa <rameezmustafa@codeaurora.org>
Signed-off-by: Prasad Sodagudi <psodagud@codeaurora.org>
---
 kernel/cpu.c | 41 +++++++++++++++++++++++++++++++++++++++++
 1 file changed, 41 insertions(+)

diff --git a/kernel/cpu.c b/kernel/cpu.c
index 68b3740..aea4ce2 100644
--- a/kernel/cpu.c
+++ b/kernel/cpu.c
@@ -32,6 +32,7 @@
 #include <linux/relay.h>
 #include <linux/slab.h>
 #include <linux/percpu-rwsem.h>
+#include <uapi/linux/sched/types.h>
 
 #include <trace/events/power.h>
 #define CREATE_TRACE_POINTS
@@ -1191,6 +1192,33 @@ void cpuhp_online_idle(enum cpuhp_state state)
 	complete_ap_thread(st, true);
 }
 
+static int switch_to_rt_policy(void)
+{
+	struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 };
+	unsigned int policy = current->policy;
+	int err;
+
+	/* Nobody should be attempting hotplug from these policy contexts. */
+	if (policy == SCHED_BATCH || policy == SCHED_IDLE ||
+					policy == SCHED_DEADLINE)
+		return -EPERM;
+
+	if (policy == SCHED_FIFO || policy == SCHED_RR)
+		return 1;
+
+	/* Only SCHED_NORMAL left. */
+	err = sched_setscheduler_nocheck(current, SCHED_FIFO, &param);
+	return err;
+
+}
+
+static int switch_to_fair_policy(void)
+{
+	struct sched_param param = { .sched_priority = 0 };
+
+	return sched_setscheduler_nocheck(current, SCHED_NORMAL, &param);
+}
+
 /* Requires cpu_add_remove_lock to be held */
 static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target)
 {
@@ -1258,6 +1286,7 @@ static int _cpu_up(unsigned int cpu, int tasks_frozen, enum cpuhp_state target)
 static int cpu_up(unsigned int cpu, enum cpuhp_state target)
 {
 	int err = 0;
+	int switch_err = 0;
 
 	if (!cpu_possible(cpu)) {
 		pr_err("can't online cpu %d because it is not configured as may-hotadd at boot time\n",
@@ -1268,6 +1297,10 @@ static int cpu_up(unsigned int cpu, enum cpuhp_state target)
 		return -EINVAL;
 	}
 
+	switch_err = switch_to_rt_policy();
+	if (switch_err < 0)
+		return switch_err;
+
 	err = try_online_node(cpu_to_node(cpu));
 	if (err)
 		return err;
@@ -1286,6 +1319,14 @@ static int cpu_up(unsigned int cpu, enum cpuhp_state target)
 	err = _cpu_up(cpu, 0, target);
 out:
 	cpu_maps_update_done();
+
+	if (!switch_err) {
+		switch_err = switch_to_fair_policy();
+		if (switch_err)
+			pr_err("Hotplug policy switch err=%d Task %s pid=%d\n",
+				switch_err, current->comm, current->pid);
+	}
+
 	return err;
 }
 
-- 
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project

Re: [PATCH 0/2] measure latency of cpu hotplug path

[peterz]

On Wed, Sep 23, 2020 at 04:37:44PM -0700, Prasad Sodagudi wrote:
> There are all changes related to cpu hotplug path and would like to seek
> upstream review. These are all patches in Qualcomm downstream kernel
> for a quite long time. First patch sets the rt prioity to hotplug
> task and second patch adds cpuhp trace events.
> 
> 1) cpu-hotplug: Always use real time scheduling when hotplugging a CPU
> 2) cpu/hotplug: Add cpuhp_latency trace event

Why? Hotplug is a known super slow path. If you care about hotplug
latency you're doing it wrong.

Re: [PATCH 0/2] measure latency of cpu hotplug path

[Steven Rostedt]

On Thu, 24 Sep 2020 10:34:14 +0200
peterz@infradead.org wrote:

> On Wed, Sep 23, 2020 at 04:37:44PM -0700, Prasad Sodagudi wrote:
> > There are all changes related to cpu hotplug path and would like to seek
> > upstream review. These are all patches in Qualcomm downstream kernel
> > for a quite long time. First patch sets the rt prioity to hotplug
> > task and second patch adds cpuhp trace events.
> > 
> > 1) cpu-hotplug: Always use real time scheduling when hotplugging a CPU
> > 2) cpu/hotplug: Add cpuhp_latency trace event  
> 
> Why? Hotplug is a known super slow path. If you care about hotplug
> latency you're doing it wrong.

I'd like to know the answer to Peter's question too. Why?

-- Steve

Re: [PATCH 0/2] measure latency of cpu hotplug path

[psodagud]

On 2020-09-24 07:58, Steven Rostedt wrote:
> On Thu, 24 Sep 2020 10:34:14 +0200
> peterz@infradead.org wrote:
> 
>> On Wed, Sep 23, 2020 at 04:37:44PM -0700, Prasad Sodagudi wrote:
>> > There are all changes related to cpu hotplug path and would like to seek
>> > upstream review. These are all patches in Qualcomm downstream kernel
>> > for a quite long time. First patch sets the rt prioity to hotplug
>> > task and second patch adds cpuhp trace events.
>> >
>> > 1) cpu-hotplug: Always use real time scheduling when hotplugging a CPU
>> > 2) cpu/hotplug: Add cpuhp_latency trace event
>> 
>> Why? Hotplug is a known super slow path. If you care about hotplug
>> latency you're doing it wrong.
Hi Peter,

[PATCH 1/2] cpu/hotplug: Add cpuhp_latency trace event -
1)	Tracing of the cpuhp operation is important to find whether upstream 
changes or out of tree modules(or firmware changes) caused latency 
regression or not.
2)	Secondary cpus are hotplug out during the device suspend and hotplug 
in during the resume.
3)	firmware(psci calls handling from firmware) changes impact need to be 
tested right?
4)	cpu hotplug framework(CPUHP_AP_ONLINE_DYN) dynamic callbacks may 
impact the hotplug latency.


[PATCH 2/2] cpu-hotplug: Always use real time scheduling when  
hotplugging a CPU –

CPU hotplug operation is stressed and while stress testing with full 
load on the system following problem is observed.
CPU hotplug operations take place in preemptible context. This leaves 
the hotplugging thread at the mercy of overall system load and CPU
availability. If the hotplugging thread does not get an opportunity to 
execute after it has already begun a hotplug operation, CPUs can
end up being stuck in a quasi online state. In the worst case a CPU can 
be stuck in a state where the migration thread is parked while
another task is executing and changing affinity in a loop. This 
combination can result in unbounded execution time for the running
task until the hot plugging thread gets the chance to run to complete 
the hotplug operation.

-Thanks, Prasad

> 
> I'd like to know the answer to Peter's question too. Why?
> 
> -- Steve

Re: [PATCH 0/2] measure latency of cpu hotplug path

[Peter Zijlstra]

On Sun, Sep 27, 2020 at 07:41:45PM -0700, psodagud@codeaurora.org wrote:
> On 2020-09-24 07:58, Steven Rostedt wrote:
> > On Thu, 24 Sep 2020 10:34:14 +0200
> > peterz@infradead.org wrote:
> > 
> > > On Wed, Sep 23, 2020 at 04:37:44PM -0700, Prasad Sodagudi wrote:
> > > > There are all changes related to cpu hotplug path and would like to seek
> > > > upstream review. These are all patches in Qualcomm downstream kernel
> > > > for a quite long time. First patch sets the rt prioity to hotplug
> > > > task and second patch adds cpuhp trace events.
> > > >
> > > > 1) cpu-hotplug: Always use real time scheduling when hotplugging a CPU
> > > > 2) cpu/hotplug: Add cpuhp_latency trace event
> > > 
> > > Why? Hotplug is a known super slow path. If you care about hotplug
> > > latency you're doing it wrong.
> Hi Peter,
> 
> [PATCH 1/2] cpu/hotplug: Add cpuhp_latency trace event -
> 1)	Tracing of the cpuhp operation is important to find whether upstream
> changes or out of tree modules(or firmware changes) caused latency
> regression or not.

This is a contradiction in terms, it is impossible to have a latency
regression is you don't care about the latency in this super slow path
to begin with.

> 2)	Secondary cpus are hotplug out during the device suspend and hotplug in
> during the resume.

Indeed they are.

> 3)	firmware(psci calls handling from firmware) changes impact need to be
> tested right?

Firmware is firmware, it's broken by design and we can't fix it if it's
broken. The only sane solution is not having firmware :-)

> 4)	cpu hotplug framework(CPUHP_AP_ONLINE_DYN) dynamic callbacks may impact
> the hotplug latency.

Again, nobody cares.

> [PATCH 2/2] cpu-hotplug: Always use real time scheduling when  hotplugging a
> CPU –
> 
> CPU hotplug operation is stressed and while stress testing with full load on
> the system following problem is observed.
> CPU hotplug operations take place in preemptible context. This leaves the
> hotplugging thread at the mercy of overall system load and CPU
> availability. If the hotplugging thread does not get an opportunity to
> execute after it has already begun a hotplug operation, CPUs can
> end up being stuck in a quasi online state. In the worst case a CPU can be
> stuck in a state where the migration thread is parked while
> another task is executing and changing affinity in a loop. This combination
> can result in unbounded execution time for the running
> task until the hot plugging thread gets the chance to run to complete the
> hotplug operation.

How is that not an administration problem?

Also, you shouldn't be able to change your affinity _to_ a CPU that's
going down. One of the very first steps in hotplug ensures that.