Skip to contentSkip to conversation

pseries extended cede cpu offline mode removal

PATCH In review Nathan Lynch 2 participants 5 messages Jun 2, 2020 – Jun 4, 2020

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

cover pseries extended cede cpu offline mode removal

I propose removing the "extended cede" offline mode for CPUs as well as the partition suspend code which accommodates it by temporarily onlining all CPUs prior to suspending the LPAR.

Detailed justifications are within the individual commit messages.

I'm hoping to move the pseries partition suspend implementation to the Linux suspend framework, and I expect that having only one offline mode for CPUs will make that task quite a bit less complex.

Nathan Lynch (2):
powerpc/pseries: remove cede offline state for CPUs powerpc/rtas: don't online CPUs for partition suspend

 Documentation/core-api/cpu_hotplug.rst        |   7 -
 arch/powerpc/include/asm/rtas.h               |   2 -
 arch/powerpc/kernel/rtas.c                    | 122 +------------
 arch/powerpc/platforms/pseries/hotplug-cpu.c  | 170 ++----------------
 .../platforms/pseries/offline_states.h        |  38 ----
 arch/powerpc/platforms/pseries/pmem.c         |   1 -
 arch/powerpc/platforms/pseries/smp.c          |  28 +--
 arch/powerpc/platforms/pseries/suspend.c      |  22 +--
 8 files changed, 18 insertions(+), 372 deletions(-)
 delete mode 100644 arch/powerpc/platforms/pseries/offline_states.h
Signature · 1 line
2.25.4
#1 lore raw

PATCH 1/2 powerpc/pseries: remove cede offline state for CPUs

This effectively reverts commit 3aa565f53c39 ("powerpc/pseries: Add hooks to put the CPU into an appropriate offline state"), which added an offline mode for CPUs which uses the H_CEDE hcall instead of the architected stop-self RTAS function in order to facilitate "folding" of dedicated mode processors on PowerVM platforms to achieve energy savings. This has been the default offline mode since its introduction.

There's nothing about stop-self that would prevent the hypervisor from achieving the energy savings available via H_CEDE, so the original premise of this change appears to be flawed.

I also have encountered the claim that the transition to and from ceded state is much faster than stop-self/start-cpu. Certainly we would not want to use stop-self as an *idle* mode. That is what H_CEDE is for. However, this difference is insignificant in the context of Linux CPU hotplug, where the latency of an offline or online operation on current systems is on the order of 100ms, mainly attributable to all the various subsystems' cpuhp callbacks.

The cede offline mode also prevents accurate accounting, as discussed before:
https://lore.kernel.org/linuxppc-dev/[email protected]/

Unconditionally use stop-self to offline processor threads. This is the architected method for offlining CPUs on PAPR systems.

The "cede_offline" boot parameter is rendered obsolete.

Removing this code enables the removal of the partition suspend code which temporarily onlines all present CPUs.

Fixes 3aa565f53c39 ("powerpc/pseries: Add hooks to put the CPU into an appropriate offline state")
---
 Documentation/core-api/cpu_hotplug.rst        |   7 -
 arch/powerpc/platforms/pseries/hotplug-cpu.c  | 170 ++----------------
 .../platforms/pseries/offline_states.h        |  38 ----
 arch/powerpc/platforms/pseries/pmem.c         |   1 -
 arch/powerpc/platforms/pseries/smp.c          |  28 +--
 5 files changed, 15 insertions(+), 229 deletions(-)
 delete mode 100644 arch/powerpc/platforms/pseries/offline_states.h
Documentation/core-api/cpu_hotplug.rst +0 -7
index 4a50ab7817f7..b1ae1ac159cf 100644
@@ -50,13 +50,6 @@ Command Line Switches
 
   This option is limited to the X86 and S390 architecture.
 
-``cede_offline={"off","on"}``
-  Use this option to disable/enable putting offlined processors to an extended
-  ``H_CEDE`` state on supported pseries platforms. If nothing is specified,
-  ``cede_offline`` is set to "on".
-
-  This option is limited to the PowerPC architecture.
-
 ``cpu0_hotplug``
   Allow to shutdown CPU0.
 
arch/powerpc/platforms/pseries/hotplug-cpu.c +14 -156

229 lines not on this page · raw message

arch/powerpc/platforms/pseries/offline_states.h deleted +0 -38
deleted file mode 100644
index 51414aee2862..000000000000
@@ -1,38 +0,0 @@
-/* SPDX-License-Identifier: GPL-2.0 */
-#ifndef _OFFLINE_STATES_H_
-#define _OFFLINE_STATES_H_
-
-/* Cpu offline states go here */
-enum cpu_state_vals {
-	CPU_STATE_OFFLINE,
-	CPU_STATE_INACTIVE,
-	CPU_STATE_ONLINE,
-	CPU_MAX_OFFLINE_STATES
-};
-
-#ifdef CONFIG_HOTPLUG_CPU
-extern enum cpu_state_vals get_cpu_current_state(int cpu);
-extern void set_cpu_current_state(int cpu, enum cpu_state_vals state);
-extern void set_preferred_offline_state(int cpu, enum cpu_state_vals state);
-extern void set_default_offline_state(int cpu);
-#else
-static inline enum cpu_state_vals get_cpu_current_state(int cpu)
-{
-	return CPU_STATE_ONLINE;
-}
-
-static inline void set_cpu_current_state(int cpu, enum cpu_state_vals state)
-{
-}
-
-static inline void set_preferred_offline_state(int cpu, enum cpu_state_vals state)
-{
-}
-
-static inline void set_default_offline_state(int cpu)
-{
-}
-#endif
-
-extern enum cpu_state_vals get_preferred_offline_state(int cpu);
-#endif
arch/powerpc/platforms/pseries/pmem.c +0 -1
index f860a897a9e0..f827de7087e9 100644
@@ -24,7 +24,6 @@
 #include <asm/topology.h>
 
 #include "pseries.h"
-#include "offline_states.h"
 
 static struct device_node *pmem_node;
 
arch/powerpc/platforms/pseries/smp.c +1 -27
index ad61e90032da..a8a070269151 100644
@@ -44,8 +44,6 @@
 #include <asm/svm.h>
 
 #include "pseries.h"
-#include "offline_states.h"
-
 
 /*
  * The Primary thread of each non-boot processor was started from the OF client
@@ -108,10 +106,7 @@ static inline int smp_startup_cpu(unsigned int lcpu)
 
 	/* Fixup atomic count: it exited inside IRQ handler. */
 	task_thread_info(paca_ptrs[lcpu]->__current)->preempt_count	= 0;
-#ifdef CONFIG_HOTPLUG_CPU
-	if (get_cpu_current_state(lcpu) == CPU_STATE_INACTIVE)
-		goto out;
-#endif
+
 	/* 
 	 * If the RTAS start-cpu token does not exist then presume the
 	 * cpu is already spinning.
@@ -126,9 +121,6 @@ static inline int smp_startup_cpu(unsigned int lcpu)
 		return 0;
 	}
 
-#ifdef CONFIG_HOTPLUG_CPU
-out:
-#endif
 	return 1;
 }
 
@@ -143,10 +135,6 @@ static void smp_setup_cpu(int cpu)
 		vpa_init(cpu);
 
 	cpumask_clear_cpu(cpu, of_spin_mask);
-#ifdef CONFIG_HOTPLUG_CPU
-	set_cpu_current_state(cpu, CPU_STATE_ONLINE);
-	set_default_offline_state(cpu);
-#endif
 }
 
 static int smp_pSeries_kick_cpu(int nr)
@@ -163,20 +151,6 @@ static int smp_pSeries_kick_cpu(int nr)
 	 * the processor will continue on to secondary_start
 	 */
 	paca_ptrs[nr]->cpu_start = 1;
-#ifdef CONFIG_HOTPLUG_CPU
-	set_preferred_offline_state(nr, CPU_STATE_ONLINE);
-
-	if (get_cpu_current_state(nr) == CPU_STATE_INACTIVE) {
-		long rc;
-		unsigned long hcpuid;
-
-		hcpuid = get_hard_smp_processor_id(nr);
-		rc = plpar_hcall_norets(H_PROD, hcpuid);
-		if (rc != H_SUCCESS)
-			printk(KERN_ERR "Error: Prod to wake up processor %d "
-						"Ret= %ld\n", nr, rc);
-	}
-#endif
 
 	return 0;
 }
Signature · 1 line
2.25.4
Reviewed-by on v1 1/2
#2 lore raw

Re: [PATCH 1/2] powerpc/pseries: remove cede offline state for CPUs

Hello Nathan,

Quoted 11 lines · Nathan Lynch

This effectively reverts commit 3aa565f53c39 ("powerpc/pseries: Add hooks to put the CPU into an appropriate offline state"), which added an offline mode for CPUs which uses the H_CEDE hcall instead of the architected stop-self RTAS function in order to facilitate "folding" of dedicated mode processors on PowerVM platforms to achieve energy savings. This has been the default offline mode since its introduction.

There's nothing about stop-self that would prevent the hypervisor from achieving the energy savings available via H_CEDE, so the original premise of this change appears to be flawed.

IIRC, back in 2009, when the Extended-CEDE was introduced, it couldn't be exposed via the cpuidle subsystem since this state needs an explicit H_PROD as opposed to the H_IPI which wakes up the regular CEDE call. So, the alterative was to use the CPU-Hotplug way by having a userspace daemon fold the cores which weren't needed currently and bring them back online when they were needed. Back then, Long Term CEDE was definitely faster compared to stop-self call (It is a pity that I didn't post the numbers when I wrote the patch) and the time-taken to unfold a core was definitely one of the concerns.
(https://lkml.org/lkml/2009/9/23/522).

However, on current systems, this is not the case as you rightly mention below. Based on the measurements that I did last year, there is no significant difference in the time taken for a CPU-Hotplug operation while using stop-self vs Extended-CEDE CPU-Hotplug.

Time taken to offline a core (in seconds)
With stop-self
N Min Max Median Avg Stddev
20 0.93 1.31 1.161 1.1308 0.11776721
With Extended cede
N Min Max Median Avg Stddev
20 1.12 1.65 1.39 1.3929 0.13363379

Time taken to online a core (in seconds)
With stop-self
N Min Max Median Avg Stddev
20 1.82 2.22 1.96 1.97605 0.093984587
With Extended cede
N Min Max Median Avg Stddev
20 1.65 2.1 1.9 1.8916 0.13074902

Quoted 4 lines

I also have encountered the claim that the transition to and from ceded state is much faster than stop-self/start-cpu. Certainly we would not want to use stop-self as an *idle* mode. That is what H_CEDE is for.

True. However the regular H_CEDE doesn't get mapped to a deeper platform idle state, while Extended H_CEDE does. And the deeper platform idle states have associated performance benefits, such as the online cores being able to execute Turbo frequencies. However, the right thing to do would be for the hypervisor provide us with a new latency-hint for a extended H_CEDE which can wakeup on H_IPI, instead of needing an explicit H_PROD.

Quoted 12 lines

However, this difference is insignificant in the context of Linux CPU hotplug, where the latency of an offline or online operation on current systems is on the order of 100ms, mainly attributable to all the various subsystems' cpuhp callbacks.

The cede offline mode also prevents accurate accounting, as discussed before:
https://lore.kernel.org/linuxppc-dev/[email protected]/

Unconditionally use stop-self to offline processor threads. This is the architected method for offlining CPUs on PAPR systems.

Quoted 8 lines

The "cede_offline" boot parameter is rendered obsolete.

Removing this code enables the removal of the partition suspend code which temporarily onlines all present CPUs.

Fixes 3aa565f53c39 ("powerpc/pseries: Add hooks to put the CPU into an appropriate offline state")

The patch looks good to me.

Quoted 394 lines
---
 Documentation/core-api/cpu_hotplug.rst        |   7 -
 arch/powerpc/platforms/pseries/hotplug-cpu.c  | 170 ++----------------
 .../platforms/pseries/offline_states.h        |  38 ----
 arch/powerpc/platforms/pseries/pmem.c         |   1 -
 arch/powerpc/platforms/pseries/smp.c          |  28 +--
 5 files changed, 15 insertions(+), 229 deletions(-)
 delete mode 100644 arch/powerpc/platforms/pseries/offline_states.h
Documentation/core-api/cpu_hotplug.rst +0 -7
index 4a50ab7817f7..b1ae1ac159cf 100644
@@ -50,13 +50,6 @@ Command Line Switches
 
   This option is limited to the X86 and S390 architecture.
 
-``cede_offline={"off","on"}``
-  Use this option to disable/enable putting offlined processors to an extended
-  ``H_CEDE`` state on supported pseries platforms. If nothing is specified,
-  ``cede_offline`` is set to "on".
-
-  This option is limited to the PowerPC architecture.
-
 ``cpu0_hotplug``
   Allow to shutdown CPU0.
 
arch/powerpc/platforms/pseries/hotplug-cpu.c +14 -156
index 3e8cbfe7a80f..d4b346355bb9 100644
@@ -35,54 +35,10 @@
 #include <asm/topology.h>
 
 #include "pseries.h"
-#include "offline_states.h"
 
 /* This version can't take the spinlock, because it never returns */
 static int rtas_stop_self_token = RTAS_UNKNOWN_SERVICE;
 
-static DEFINE_PER_CPU(enum cpu_state_vals, preferred_offline_state) =
-							CPU_STATE_OFFLINE;
-static DEFINE_PER_CPU(enum cpu_state_vals, current_state) = CPU_STATE_OFFLINE;
-
-static enum cpu_state_vals default_offline_state = CPU_STATE_OFFLINE;
-
-static bool cede_offline_enabled __read_mostly = true;
-
-/*
- * Enable/disable cede_offline when available.
- */
-static int __init setup_cede_offline(char *str)
-{
-	return (kstrtobool(str, &cede_offline_enabled) == 0);
-}
-
-__setup("cede_offline=", setup_cede_offline);
-
-enum cpu_state_vals get_cpu_current_state(int cpu)
-{
-	return per_cpu(current_state, cpu);
-}
-
-void set_cpu_current_state(int cpu, enum cpu_state_vals state)
-{
-	per_cpu(current_state, cpu) = state;
-}
-
-enum cpu_state_vals get_preferred_offline_state(int cpu)
-{
-	return per_cpu(preferred_offline_state, cpu);
-}
-
-void set_preferred_offline_state(int cpu, enum cpu_state_vals state)
-{
-	per_cpu(preferred_offline_state, cpu) = state;
-}
-
-void set_default_offline_state(int cpu)
-{
-	per_cpu(preferred_offline_state, cpu) = default_offline_state;
-}
-
 static void rtas_stop_self(void)
 {
 	static struct rtas_args args;
@@ -101,9 +57,7 @@ static void rtas_stop_self(void)
 
 static void pseries_mach_cpu_die(void)
 {
-	unsigned int cpu = smp_processor_id();
 	unsigned int hwcpu = hard_smp_processor_id();
-	u8 cede_latency_hint = 0;
 
 	local_irq_disable();
 	idle_task_exit();
@@ -112,49 +66,6 @@ static void pseries_mach_cpu_die(void)
 	else
 		xics_teardown_cpu();
 
-	if (get_preferred_offline_state(cpu) == CPU_STATE_INACTIVE) {
-		set_cpu_current_state(cpu, CPU_STATE_INACTIVE);
-		if (ppc_md.suspend_disable_cpu)
-			ppc_md.suspend_disable_cpu();
-
-		cede_latency_hint = 2;
-
-		get_lppaca()->idle = 1;
-		if (!lppaca_shared_proc(get_lppaca()))
-			get_lppaca()->donate_dedicated_cpu = 1;
-
-		while (get_preferred_offline_state(cpu) == CPU_STATE_INACTIVE) {
-			while (!prep_irq_for_idle()) {
-				local_irq_enable();
-				local_irq_disable();
-			}
-
-			extended_cede_processor(cede_latency_hint);
-		}
-
-		local_irq_disable();
-
-		if (!lppaca_shared_proc(get_lppaca()))
-			get_lppaca()->donate_dedicated_cpu = 0;
-		get_lppaca()->idle = 0;
-
-		if (get_preferred_offline_state(cpu) == CPU_STATE_ONLINE) {
-			unregister_slb_shadow(hwcpu);
-
-			hard_irq_disable();
-			/*
-			 * Call to start_secondary_resume() will not return.
-			 * Kernel stack will be reset and start_secondary()
-			 * will be called to continue the online operation.
-			 */
-			start_secondary_resume();
-		}
-	}
-
-	/* Requested state is CPU_STATE_OFFLINE at this point */
-	WARN_ON(get_preferred_offline_state(cpu) != CPU_STATE_OFFLINE);
-
-	set_cpu_current_state(cpu, CPU_STATE_OFFLINE);
 	unregister_slb_shadow(hwcpu);
 	rtas_stop_self();
 
@@ -200,24 +111,13 @@ static void pseries_cpu_die(unsigned int cpu)
 	int cpu_status = 1;
 	unsigned int pcpu = get_hard_smp_processor_id(cpu);
 
-	if (get_preferred_offline_state(cpu) == CPU_STATE_INACTIVE) {
-		cpu_status = 1;
-		for (tries = 0; tries < 5000; tries++) {
-			if (get_cpu_current_state(cpu) == CPU_STATE_INACTIVE) {
-				cpu_status = 0;
-				break;
-			}
-			msleep(1);
-		}
-	} else if (get_preferred_offline_state(cpu) == CPU_STATE_OFFLINE) {
+	for (tries = 0; tries < 25; tries++) {
+		cpu_status = smp_query_cpu_stopped(pcpu);
+		if (cpu_status == QCSS_STOPPED ||
+		    cpu_status == QCSS_HARDWARE_ERROR)
+			break;
+		cpu_relax();
 
-		for (tries = 0; tries < 25; tries++) {
-			cpu_status = smp_query_cpu_stopped(pcpu);
-			if (cpu_status == QCSS_STOPPED ||
-			    cpu_status == QCSS_HARDWARE_ERROR)
-				break;
-			cpu_relax();
-		}
 	}
 
 	if (cpu_status != 0) {
@@ -359,28 +259,15 @@ static int dlpar_offline_cpu(struct device_node *dn)
 			if (get_hard_smp_processor_id(cpu) != thread)
 				continue;
 
-			if (get_cpu_current_state(cpu) == CPU_STATE_OFFLINE)
+			if (!cpu_online(cpu))
 				break;
 
-			if (get_cpu_current_state(cpu) == CPU_STATE_ONLINE) {
-				set_preferred_offline_state(cpu,
-							    CPU_STATE_OFFLINE);
-				cpu_maps_update_done();
-				timed_topology_update(1);
-				rc = device_offline(get_cpu_device(cpu));
-				if (rc)
-					goto out;
-				cpu_maps_update_begin();
-				break;
-			}
-
-			/*
-			 * The cpu is in CPU_STATE_INACTIVE.
-			 * Upgrade it's state to CPU_STATE_OFFLINE.
-			 */
-			set_preferred_offline_state(cpu, CPU_STATE_OFFLINE);
-			WARN_ON(plpar_hcall_norets(H_PROD, thread) != H_SUCCESS);
-			__cpu_die(cpu);
+			cpu_maps_update_done();
+			timed_topology_update(1);
+			rc = device_offline(get_cpu_device(cpu));
+			if (rc)
+				goto out;
+			cpu_maps_update_begin();
 			break;
 		}
 		if (cpu == num_possible_cpus()) {
@@ -414,8 +301,6 @@ static int dlpar_online_cpu(struct device_node *dn)
 		for_each_present_cpu(cpu) {
 			if (get_hard_smp_processor_id(cpu) != thread)
 				continue;
-			BUG_ON(get_cpu_current_state(cpu)
-					!= CPU_STATE_OFFLINE);
 			cpu_maps_update_done();
 			timed_topology_update(1);
 			find_and_online_cpu_nid(cpu);
@@ -1013,27 +898,8 @@ static struct notifier_block pseries_smp_nb = {
 	.notifier_call = pseries_smp_notifier,
 };
 
-#define MAX_CEDE_LATENCY_LEVELS		4
-#define	CEDE_LATENCY_PARAM_LENGTH	10
-#define CEDE_LATENCY_PARAM_MAX_LENGTH	\
-	(MAX_CEDE_LATENCY_LEVELS * CEDE_LATENCY_PARAM_LENGTH * sizeof(char))
-#define CEDE_LATENCY_TOKEN		45
-
-static char cede_parameters[CEDE_LATENCY_PARAM_MAX_LENGTH];
-
-static int parse_cede_parameters(void)
-{
-	memset(cede_parameters, 0, CEDE_LATENCY_PARAM_MAX_LENGTH);
-	return rtas_call(rtas_token("ibm,get-system-parameter"), 3, 1,
-			 NULL,
-			 CEDE_LATENCY_TOKEN,
-			 __pa(cede_parameters),
-			 CEDE_LATENCY_PARAM_MAX_LENGTH);
-}
-
 static int __init pseries_cpu_hotplug_init(void)
 {
-	int cpu;
 	int qcss_tok;
 
 #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE
@@ -1056,16 +922,8 @@ static int __init pseries_cpu_hotplug_init(void)
 	smp_ops->cpu_die = pseries_cpu_die;
 
 	/* Processors can be added/removed only on LPAR */
-	if (firmware_has_feature(FW_FEATURE_LPAR)) {
+	if (firmware_has_feature(FW_FEATURE_LPAR))
 		of_reconfig_notifier_register(&pseries_smp_nb);
-		cpu_maps_update_begin();
-		if (cede_offline_enabled && parse_cede_parameters() == 0) {
-			default_offline_state = CPU_STATE_INACTIVE;
-			for_each_online_cpu(cpu)
-				set_default_offline_state(cpu);
-		}
-		cpu_maps_update_done();
-	}
 
 	return 0;
 }
arch/powerpc/platforms/pseries/offline_states.h deleted +0 -38
deleted file mode 100644
index 51414aee2862..000000000000
@@ -1,38 +0,0 @@
-/* SPDX-License-Identifier: GPL-2.0 */
-#ifndef _OFFLINE_STATES_H_
-#define _OFFLINE_STATES_H_
-
-/* Cpu offline states go here */
-enum cpu_state_vals {
-	CPU_STATE_OFFLINE,
-	CPU_STATE_INACTIVE,
-	CPU_STATE_ONLINE,
-	CPU_MAX_OFFLINE_STATES
-};
-
-#ifdef CONFIG_HOTPLUG_CPU
-extern enum cpu_state_vals get_cpu_current_state(int cpu);
-extern void set_cpu_current_state(int cpu, enum cpu_state_vals state);
-extern void set_preferred_offline_state(int cpu, enum cpu_state_vals state);
-extern void set_default_offline_state(int cpu);
-#else
-static inline enum cpu_state_vals get_cpu_current_state(int cpu)
-{
-	return CPU_STATE_ONLINE;
-}
-
-static inline void set_cpu_current_state(int cpu, enum cpu_state_vals state)
-{
-}
-
-static inline void set_preferred_offline_state(int cpu, enum cpu_state_vals state)
-{
-}
-
-static inline void set_default_offline_state(int cpu)
-{
-}
-#endif
-
-extern enum cpu_state_vals get_preferred_offline_state(int cpu);
-#endif
arch/powerpc/platforms/pseries/pmem.c +0 -1
index f860a897a9e0..f827de7087e9 100644
@@ -24,7 +24,6 @@
 #include <asm/topology.h>
 
 #include "pseries.h"
-#include "offline_states.h"
 
 static struct device_node *pmem_node;
 
arch/powerpc/platforms/pseries/smp.c +1 -27
index ad61e90032da..a8a070269151 100644
@@ -44,8 +44,6 @@
 #include <asm/svm.h>
 
 #include "pseries.h"
-#include "offline_states.h"
-
 
 /*
  * The Primary thread of each non-boot processor was started from the OF client
@@ -108,10 +106,7 @@ static inline int smp_startup_cpu(unsigned int lcpu)
 
 	/* Fixup atomic count: it exited inside IRQ handler. */
 	task_thread_info(paca_ptrs[lcpu]->__current)->preempt_count	= 0;
-#ifdef CONFIG_HOTPLUG_CPU
-	if (get_cpu_current_state(lcpu) == CPU_STATE_INACTIVE)
-		goto out;
-#endif
+
 	/* 
 	 * If the RTAS start-cpu token does not exist then presume the
 	 * cpu is already spinning.
@@ -126,9 +121,6 @@ static inline int smp_startup_cpu(unsigned int lcpu)
 		return 0;
 	}
 
-#ifdef CONFIG_HOTPLUG_CPU
-out:
-#endif
 	return 1;
 }
 
@@ -143,10 +135,6 @@ static void smp_setup_cpu(int cpu)
 		vpa_init(cpu);
 
 	cpumask_clear_cpu(cpu, of_spin_mask);
-#ifdef CONFIG_HOTPLUG_CPU
-	set_cpu_current_state(cpu, CPU_STATE_ONLINE);
-	set_default_offline_state(cpu);
-#endif
 }
 
 static int smp_pSeries_kick_cpu(int nr)
@@ -163,20 +151,6 @@ static int smp_pSeries_kick_cpu(int nr)
 	 * the processor will continue on to secondary_start
 	 */
 	paca_ptrs[nr]->cpu_start = 1;
-#ifdef CONFIG_HOTPLUG_CPU
-	set_preferred_offline_state(nr, CPU_STATE_ONLINE);
-
-	if (get_cpu_current_state(nr) == CPU_STATE_INACTIVE) {
-		long rc;
-		unsigned long hcpuid;
-
-		hcpuid = get_hard_smp_processor_id(nr);
-		rc = plpar_hcall_norets(H_PROD, hcpuid);
-		if (rc != H_SUCCESS)
-			printk(KERN_ERR "Error: Prod to wake up processor %d "
-						"Ret= %ld\n", nr, rc);
-	}
-#endif
 
 	return 0;
 }
Signature · 1 line
2.25.4
#3 lore raw

Re: [PATCH 1/2] powerpc/pseries: remove cede offline state for CPUs

Hi Gautham,

Quoted 14 lines · Gautham R Shenoy
Quoted 11 lines · Nathan Lynch

This effectively reverts commit 3aa565f53c39 ("powerpc/pseries: Add hooks to put the CPU into an appropriate offline state"), which added an offline mode for CPUs which uses the H_CEDE hcall instead of the architected stop-self RTAS function in order to facilitate "folding" of dedicated mode processors on PowerVM platforms to achieve energy savings. This has been the default offline mode since its introduction.

There's nothing about stop-self that would prevent the hypervisor from achieving the energy savings available via H_CEDE, so the original premise of this change appears to be flawed.

Quoted 10 lines

IIRC, back in 2009, when the Extended-CEDE was introduced, it couldn't be exposed via the cpuidle subsystem since this state needs an explicit H_PROD as opposed to the H_IPI which wakes up the regular CEDE call. So, the alterative was to use the CPU-Hotplug way by having a userspace daemon fold the cores which weren't needed currently and bring them back online when they were needed. Back then, Long Term CEDE was definitely faster compared to stop-self call (It is a pity that I didn't post the numbers when I wrote the patch) and the time-taken to unfold a core was definitely one of the concerns.
(https://lkml.org/lkml/2009/9/23/522).

Thanks for the background. Ideally, we would understand what has changed since then... certainly the offline path would have been slower when pseries_cpu_die() slept for 200ms between issuing query-cpu-stopped-state calls, but that was changed in 2008 in b906cfa397fd ("powerpc/pseries: Fix cpu hotplug") so perhaps it wasn't a factor in your measurements. Even less sure about what could have made the online path more expensive.

(For the benefit of anybody else referring to the 2009 discussion that you linked: that exchange is a bit strange to me because there seems to be a conception that stop-self releases the processor to the platform in a way that could prevent the OS from restarting it on demand. I don't believe this has ever been the case. Processor deallocation is a related but separate workflow involving different RTAS functions.)

Quoted 2 lines

The patch looks good to me.

I appreciate the review, thanks!

#4 lore raw

PATCH 2/2 powerpc/rtas: don't online CPUs for partition suspend

In reply to Nathan Lynch

Partition suspension, used for hibernation and migration, requires that the OS place all but one of the LPAR's processor threads into one of two states prior to calling the ibm,suspend-me RTAS function:

* the architected offline state (via RTAS stop-self); or
* the H_JOIN hcall, which does not return until the partition resumes execution

Using H_CEDE as the offline mode, introduced by
commit 3aa565f53c39 ("powerpc/pseries: Add hooks to put the CPU into an appropriate offline state"), means that any threads which are offline from Linux's point of view must be moved to one of those two states before a partition suspension can proceed.

This was eventually addressed in commit 120496ac2d2d ("powerpc: Bring all threads online prior to migration/hibernation"), which added code to temporarily bring up any offline processor threads so they can call H_JOIN. Conceptually this is fine, but the implementation has had multiple races with cpu hotplug operations initiated from user space[1][2][3], the error handling is fragile, and it generates user-visible cpu hotplug events which is a lot of noise for a platform feature that's supposed to minimize disruption to workloads.

With commit 3aa565f53c39 ("powerpc/pseries: Add hooks to put the CPU into an appropriate offline state") reverted, this code becomes unnecessary, so remove it. Since any offline CPUs now are truly offline from the platform's point of view, it is no longer necessary to bring up CPUs only to have them call H_JOIN and then go offline again upon resuming. Only active threads are required to call H_JOIN; stopped threads can be left alone.

[1] commit a6717c01ddc2 ("powerpc/rtas: use device model APIs and serialization during LPM")
[2] commit 9fb603050ffd ("powerpc/rtas: retry when cpu offline races with suspend/migration")
[3] commit dfd718a2ed1f ("powerpc/rtas: Fix a potential race between CPU-Offline & Migration")

Fixes 120496ac2d2d ("powerpc: Bring all threads online prior to migration/hibernation")
---
 arch/powerpc/include/asm/rtas.h          |   2 -
 arch/powerpc/kernel/rtas.c               | 122 +----------------------
 arch/powerpc/platforms/pseries/suspend.c |  22 +---
 3 files changed, 3 insertions(+), 143 deletions(-)
arch/powerpc/include/asm/rtas.h +0 -2
index 3c1887351c71..bd227e0eab07 100644
@@ -368,8 +368,6 @@ extern int rtas_set_indicator_fast(int indicator, int index, int new_value);
 extern void rtas_progress(char *s, unsigned short hex);
 extern int rtas_suspend_cpu(struct rtas_suspend_me_data *data);
 extern int rtas_suspend_last_cpu(struct rtas_suspend_me_data *data);
-extern int rtas_online_cpus_mask(cpumask_var_t cpus);
-extern int rtas_offline_cpus_mask(cpumask_var_t cpus);
 extern int rtas_ibm_suspend_me(u64 handle);
 
 struct rtc_time;
arch/powerpc/kernel/rtas.c +2 -120

157 lines not on this page · raw message

arch/powerpc/platforms/pseries/suspend.c +1 -21
index 0a24a5a185f0..f789693f61f4 100644
@@ -132,15 +132,11 @@ static ssize_t store_hibernate(struct device *dev,
 			       struct device_attribute *attr,
 			       const char *buf, size_t count)
 {
-	cpumask_var_t offline_mask;
 	int rc;
 
 	if (!capable(CAP_SYS_ADMIN))
 		return -EPERM;
 
-	if (!alloc_cpumask_var(&offline_mask, GFP_KERNEL))
-		return -ENOMEM;
-
 	stream_id = simple_strtoul(buf, NULL, 16);
 
 	do {
@@ -150,32 +146,16 @@ static ssize_t store_hibernate(struct device *dev,
 	} while (rc == -EAGAIN);
 
 	if (!rc) {
-		/* All present CPUs must be online */
-		cpumask_andnot(offline_mask, cpu_present_mask,
-				cpu_online_mask);
-		rc = rtas_online_cpus_mask(offline_mask);
-		if (rc) {
-			pr_err("%s: Could not bring present CPUs online.\n",
-					__func__);
-			goto out;
-		}
-
 		stop_topology_update();
 		rc = pm_suspend(PM_SUSPEND_MEM);
 		start_topology_update();
-
-		/* Take down CPUs not online prior to suspend */
-		if (!rtas_offline_cpus_mask(offline_mask))
-			pr_warn("%s: Could not restore CPUs to offline "
-					"state.\n", __func__);
 	}
 
 	stream_id = 0;
 
 	if (!rc)
 		rc = count;
-out:
-	free_cpumask_var(offline_mask);
+
 	return rc;
 }
 
Signature · 1 line
2.25.4