[RFC PATCH 14/49] perf/x86: Move CPUHP_AP_PERF_X86_STARTING post stop_machine()

[RFC PATCH 14/49] perf/x86: Move CPUHP_AP_PERF_X86_STARTING post stop_machine()

[Frederic Weisbecker]

Between CPUHP_TEARDOWN_CPU and CPUHP_AP_OFFLINE, stop_machine executes
all the teardown callbacks while stopping all CPUs in the system. This
doesn't mix-up well with running RT or isolated tasks.

In order to transition smoothly from global stop_machine to finer
grained synchronization between teardown callbacks and the other CPUs,
a new stop machine post function has been introduced to be executed
after the main one with IRQs still disabled while other CPUs can exit
stop_machine().

Here only intel has a dying callback (intel_pmu_cpu_dying()) which does:

* Deallocate the per CPU DS structure on the MSR_IA32_DS_AREA. This is
  already racy against offlining CPUs because all possible CPUs's MSR
  are written on a hardware event initialization, which still can happen
  before the CPU actually goes off.

* Deallocate PEBS buffer. This is already well synchronized against
  hotplug up callback.

Therefore the related teardwown callback can run concurrently with other
CPUs.

Move it to post stop_machine() accordingly.

Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
---
 kernel/cpu.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/kernel/cpu.c b/kernel/cpu.c
index 5ac2a925f7d6..c37fa1272d57 100644
--- a/kernel/cpu.c
+++ b/kernel/cpu.c
@@ -1275,7 +1275,7 @@ void clear_tasks_mm_cpumask(int cpu)
 static int take_cpu_down(void *_param)
 {
 	struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
-	enum cpuhp_state target = max((int)st->target, CPUHP_AP_PERF_X86_AMD_UNCORE_STARTING);
+	enum cpuhp_state target = max((int)st->target, CPUHP_AP_PERF_X86_STARTING);
 	int err, cpu = smp_processor_id();
 
 	/* Ensure this CPU doesn't handle any more interrupts. */
@@ -1303,7 +1303,7 @@ static int take_cpu_down_post(void *_param)
 	enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE);
 	int cpu = smp_processor_id();
 
-	WARN_ON(st->state != CPUHP_AP_PERF_X86_AMD_UNCORE_STARTING);
+	WARN_ON(st->state != CPUHP_AP_PERF_X86_STARTING);
 	/*
 	 * Invoke the former CPU_DYING callbacks. DYING must not fail!
 	 */
-- 
2.53.0

Re: [RFC PATCH 14/49] perf/x86: Move CPUHP_AP_PERF_X86_STARTING post stop_machine()

[sashiko-bot]

> Between CPUHP_TEARDOWN_CPU and CPUHP_AP_OFFLINE, stop_machine executes
> all the teardown callbacks while stopping all CPUs in the system. This
> doesn't mix-up well with running RT or isolated tasks.
> 
> In order to transition smoothly from global stop_machine to finer
> grained synchronization between teardown callbacks and the other CPUs,
> a new stop machine post function has been introduced to be executed
> after the main one with IRQs still disabled while other CPUs can exit
> stop_machine().
> 
> Here only intel has a dying callback (intel_pmu_cpu_dying()) which does:
> 
> * Deallocate the per CPU DS structure on the MSR_IA32_DS_AREA. This is
>   already racy against offlining CPUs because all possible CPUs's MSR
>   are written on a hardware event initialization, which still can happen
>   before the CPU actually goes off.
> 
> * Deallocate PEBS buffer. This is already well synchronized against
>   hotplug up callback.
> 
> Therefore the related teardwown callback can run concurrently with other
> CPUs.
> 
> Move it to post stop_machine() accordingly.
> 
> Signed-off-by: Frederic Weisbecker <frederic@kernel.org>

Sashiko has reviewed this patch and found no issues.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261019130856.49711-1-frederic@kernel.org?part=14

[RFC PATCH 15/49] perf/amd/ibs: Move CPUHP_AP_PERF_X86_AMD_IBS_STARTING post stop_machine()

[Frederic Weisbecker]

Between CPUHP_TEARDOWN_CPU and CPUHP_AP_OFFLINE, stop_machine executes
all the teardown callbacks while stopping all CPUs in the system. This
doesn't mix-up well with running RT or isolated tasks.

In order to transition smoothly from global stop_machine to finer
grained synchronization between teardown callbacks and the other CPUs,
a new stop machine post function has been introduced to be executed
after the main one with IRQs still disabled while other CPUs can exit
stop_machine().

Here the callback only restores the IBS interrupt to the normal priority
on the local APIC and running it concurrently against other CPUs
shouldn't expose any kind of race.

Move this teardown callback to post stop_machine() accordingly.

Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
---
 kernel/cpu.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/kernel/cpu.c b/kernel/cpu.c
index c37fa1272d57..5a46372a8501 100644
--- a/kernel/cpu.c
+++ b/kernel/cpu.c
@@ -1275,7 +1275,7 @@ void clear_tasks_mm_cpumask(int cpu)
 static int take_cpu_down(void *_param)
 {
 	struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
-	enum cpuhp_state target = max((int)st->target, CPUHP_AP_PERF_X86_STARTING);
+	enum cpuhp_state target = max((int)st->target, CPUHP_AP_PERF_X86_AMD_IBS_STARTING);
 	int err, cpu = smp_processor_id();
 
 	/* Ensure this CPU doesn't handle any more interrupts. */
@@ -1303,7 +1303,7 @@ static int take_cpu_down_post(void *_param)
 	enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE);
 	int cpu = smp_processor_id();
 
-	WARN_ON(st->state != CPUHP_AP_PERF_X86_STARTING);
+	WARN_ON(st->state != CPUHP_AP_PERF_X86_AMD_IBS_STARTING);
 	/*
 	 * Invoke the former CPU_DYING callbacks. DYING must not fail!
 	 */
-- 
2.53.0

Re: [RFC PATCH 15/49] perf/amd/ibs: Move CPUHP_AP_PERF_X86_AMD_IBS_STARTING post stop_machine()

[sashiko-bot]

> Between CPUHP_TEARDOWN_CPU and CPUHP_AP_OFFLINE, stop_machine executes
> all the teardown callbacks while stopping all CPUs in the system. This
> doesn't mix-up well with running RT or isolated tasks.
> 
> In order to transition smoothly from global stop_machine to finer
> grained synchronization between teardown callbacks and the other CPUs,
> a new stop machine post function has been introduced to be executed
> after the main one with IRQs still disabled while other CPUs can exit
> stop_machine().
> 
> Here the callback only restores the IBS interrupt to the normal priority
> on the local APIC and running it concurrently against other CPUs
> shouldn't expose any kind of race.
> 
> Move this teardown callback to post stop_machine() accordingly.
> 
> Signed-off-by: Frederic Weisbecker <frederic@kernel.org>

Sashiko has reviewed this patch and found no issues.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261019130856.49711-1-frederic@kernel.org?part=15

[RFC PATCH 16/49] cpu: Move CPUHP_AP_PERF_XTENSA_STARTING post stop_machine()

[Frederic Weisbecker]

Between CPUHP_TEARDOWN_CPU and CPUHP_AP_OFFLINE, stop_machine executes
all the teardown callbacks while stopping all CPUs in the system. This
doesn't mix-up well with running RT or isolated tasks.

In order to transition smoothly from global stop_machine to finer
grained synchronization between teardown callbacks and the other CPUs,
a new stop machine post function has been introduced to be executed
after the main one with IRQs still disabled while other CPUs can exit
stop_machine().

This state doesn't have a teardown callback.

Move it to post stop_machine() accordingly.

Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
---
 kernel/cpu.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/kernel/cpu.c b/kernel/cpu.c
index 5a46372a8501..c20a822df803 100644
--- a/kernel/cpu.c
+++ b/kernel/cpu.c
@@ -1275,7 +1275,7 @@ void clear_tasks_mm_cpumask(int cpu)
 static int take_cpu_down(void *_param)
 {
 	struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
-	enum cpuhp_state target = max((int)st->target, CPUHP_AP_PERF_X86_AMD_IBS_STARTING);
+	enum cpuhp_state target = max((int)st->target, CPUHP_AP_PERF_XTENSA_STARTING);
 	int err, cpu = smp_processor_id();
 
 	/* Ensure this CPU doesn't handle any more interrupts. */
@@ -1303,7 +1303,7 @@ static int take_cpu_down_post(void *_param)
 	enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE);
 	int cpu = smp_processor_id();
 
-	WARN_ON(st->state != CPUHP_AP_PERF_X86_AMD_IBS_STARTING);
+	WARN_ON(st->state != CPUHP_AP_PERF_XTENSA_STARTING);
 	/*
 	 * Invoke the former CPU_DYING callbacks. DYING must not fail!
 	 */
-- 
2.53.0

Re: [RFC PATCH 16/49] cpu: Move CPUHP_AP_PERF_XTENSA_STARTING post stop_machine()

[sashiko-bot]

> Between CPUHP_TEARDOWN_CPU and CPUHP_AP_OFFLINE, stop_machine executes
> all the teardown callbacks while stopping all CPUs in the system. This
> doesn't mix-up well with running RT or isolated tasks.
> 
> In order to transition smoothly from global stop_machine to finer
> grained synchronization between teardown callbacks and the other CPUs,
> a new stop machine post function has been introduced to be executed
> after the main one with IRQs still disabled while other CPUs can exit
> stop_machine().
> 
> This state doesn't have a teardown callback.
> 
> Move it to post stop_machine() accordingly.
> 
> Signed-off-by: Frederic Weisbecker <frederic@kernel.org>

Sashiko has reviewed this patch and found no issues.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261019130856.49711-1-frederic@kernel.org?part=16

[RFC PATCH 19/49] perf/arm_pmu: Move CPUHP_AP_PERF_ARM_STARTING post stop_machine()

[Frederic Weisbecker]

Between CPUHP_TEARDOWN_CPU and CPUHP_AP_OFFLINE, stop_machine executes
all the teardown callbacks while stopping all CPUs in the system. This
doesn't mix-up well with running RT or isolated tasks.

In order to transition smoothly from global stop_machine to finer
grained synchronization between teardown callbacks and the other CPUs,
a new stop machine post function has been introduced to be executed
after the main one with IRQs still disabled while other CPUs can exit
stop_machine().

The teardown callback here just disables the IRQs for each PMUs, which
shouldn't expose any specific race while other CPUs are running.

Move it to post stop_machine() accordingly.

Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
---
 kernel/cpu.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/kernel/cpu.c b/kernel/cpu.c
index 7d7e2d1eda9b..95cc6655cf7b 100644
--- a/kernel/cpu.c
+++ b/kernel/cpu.c
@@ -1275,7 +1275,7 @@ void clear_tasks_mm_cpumask(int cpu)
 static int take_cpu_down(void *_param)
 {
 	struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
-	enum cpuhp_state target = max((int)st->target, CPUHP_AP_PERF_ARM_ACPI_STARTING);
+	enum cpuhp_state target = max((int)st->target, CPUHP_AP_PERF_ARM_STARTING);
 	int err, cpu = smp_processor_id();
 
 	/* Ensure this CPU doesn't handle any more interrupts. */
@@ -1303,7 +1303,7 @@ static int take_cpu_down_post(void *_param)
 	enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE);
 	int cpu = smp_processor_id();
 
-	WARN_ON(st->state != CPUHP_AP_PERF_ARM_ACPI_STARTING);
+	WARN_ON(st->state != CPUHP_AP_PERF_ARM_STARTING);
 	/*
 	 * Invoke the former CPU_DYING callbacks. DYING must not fail!
 	 */
-- 
2.53.0

Re: [RFC PATCH 19/49] perf/arm_pmu: Move CPUHP_AP_PERF_ARM_STARTING post stop_machine()

[sashiko-bot]

> Between CPUHP_TEARDOWN_CPU and CPUHP_AP_OFFLINE, stop_machine executes
> all the teardown callbacks while stopping all CPUs in the system. This
> doesn't mix-up well with running RT or isolated tasks.
> 
> In order to transition smoothly from global stop_machine to finer
> grained synchronization between teardown callbacks and the other CPUs,
> a new stop machine post function has been introduced to be executed
> after the main one with IRQs still disabled while other CPUs can exit
> stop_machine().
> 
> The teardown callback here just disables the IRQs for each PMUs, which
> shouldn't expose any specific race while other CPUs are running.
> 
> Move it to post stop_machine() accordingly.
> 
> Signed-off-by: Frederic Weisbecker <frederic@kernel.org>

Sashiko has reviewed this patch and found no issues.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261019130856.49711-1-frederic@kernel.org?part=19

[RFC PATCH 20/49] perf: RISC-V: Move CPUHP_AP_PERF_RISCV_STARTING post stop_machine()

[Frederic Weisbecker]

Between CPUHP_TEARDOWN_CPU and CPUHP_AP_OFFLINE, stop_machine executes
all the teardown callbacks while stopping all CPUs in the system. This
doesn't mix-up well with running RT or isolated tasks.

In order to transition smoothly from global stop_machine to finer
grained synchronization between teardown callbacks and the other CPUs,
a new stop machine post function has been introduced to be executed
after the main one with IRQs still disabled while other CPUs can exit
stop_machine().

The teardown callback here:

* Disables the per-cpu IRQ
* Disables user-mode access to the CPU counter
* Disables the shared memory snapshot of the CPU PMU.

All these operations are local to the CPU and should be able to execute
concurrently with other CPUs.

Move this teardown callback to post stop_machine() accordingly.

Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
---
 kernel/cpu.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/kernel/cpu.c b/kernel/cpu.c
index 95cc6655cf7b..c69a265e4383 100644
--- a/kernel/cpu.c
+++ b/kernel/cpu.c
@@ -1275,7 +1275,7 @@ void clear_tasks_mm_cpumask(int cpu)
 static int take_cpu_down(void *_param)
 {
 	struct cpuhp_cpu_state *st = this_cpu_ptr(&cpuhp_state);
-	enum cpuhp_state target = max((int)st->target, CPUHP_AP_PERF_ARM_STARTING);
+	enum cpuhp_state target = max((int)st->target, CPUHP_AP_PERF_RISCV_STARTING);
 	int err, cpu = smp_processor_id();
 
 	/* Ensure this CPU doesn't handle any more interrupts. */
@@ -1303,7 +1303,7 @@ static int take_cpu_down_post(void *_param)
 	enum cpuhp_state target = max((int)st->target, CPUHP_AP_OFFLINE);
 	int cpu = smp_processor_id();
 
-	WARN_ON(st->state != CPUHP_AP_PERF_ARM_STARTING);
+	WARN_ON(st->state != CPUHP_AP_PERF_RISCV_STARTING);
 	/*
 	 * Invoke the former CPU_DYING callbacks. DYING must not fail!
 	 */
-- 
2.53.0

Re: [RFC PATCH 20/49] perf: RISC-V: Move CPUHP_AP_PERF_RISCV_STARTING post stop_machine()

[sashiko-bot]

> Between CPUHP_TEARDOWN_CPU and CPUHP_AP_OFFLINE, stop_machine executes
> all the teardown callbacks while stopping all CPUs in the system. This
> doesn't mix-up well with running RT or isolated tasks.
> 
> In order to transition smoothly from global stop_machine to finer
> grained synchronization between teardown callbacks and the other CPUs,
> a new stop machine post function has been introduced to be executed
> after the main one with IRQs still disabled while other CPUs can exit
> stop_machine().
> 
> The teardown callback here:
> 
> * Disables the per-cpu IRQ
> * Disables user-mode access to the CPU counter
> * Disables the shared memory snapshot of the CPU PMU.
> 
> All these operations are local to the CPU and should be able to execute
> concurrently with other CPUs.
> 
> Move this teardown callback to post stop_machine() accordingly.
> 
> Signed-off-by: Frederic Weisbecker <frederic@kernel.org>

Sashiko has reviewed this patch and found no issues.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261019130856.49711-1-frederic@kernel.org?part=20