[patch v3 00/36] cpu/hotplug, x86: Reworked parallel CPU bringup

[patch v3 00/36] cpu/hotplug, x86: Reworked parallel CPU bringup

[Thomas Gleixner]

Hi!

This is version 3 of the reworked parallel bringup series. Version 2 can be
found here:

   https://lore.kernel.org/lkml/20230504185733.126511787@linutronix.de

This is just a quick reiteration to address the following details:

  1) Drop the two extended topology leaf patches as they are not longer
     relevant (Andrew Cooper)

  2) Make the announce_cpu() fixup work for real (Micheal Kelley)

Other than that there are no changes and the other details are all the same
as in V2.

Thanks,

	tglx

[patch v3 01/36] [patch V2 01/38] x86/smpboot: Cleanup topology_phys_to_logical_pkg()/die()

[Thomas Gleixner]

From: Thomas Gleixner <tglx@linutronix.de>

Make topology_phys_to_logical_pkg_die() static as it's only used in
smpboot.c and fixup the kernel-doc warnings for both functions.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Michael Kelley <mikelley@microsoft.com>


---
 arch/x86/include/asm/topology.h |    3 ---
 arch/x86/kernel/smpboot.c       |   10 ++++++----
 2 files changed, 6 insertions(+), 7 deletions(-)
---

--- a/arch/x86/include/asm/topology.h
+++ b/arch/x86/include/asm/topology.h
@@ -139,7 +139,6 @@ static inline int topology_max_smt_threa
 int topology_update_package_map(unsigned int apicid, unsigned int cpu);
 int topology_update_die_map(unsigned int dieid, unsigned int cpu);
 int topology_phys_to_logical_pkg(unsigned int pkg);
-int topology_phys_to_logical_die(unsigned int die, unsigned int cpu);
 bool topology_is_primary_thread(unsigned int cpu);
 bool topology_smt_supported(void);
 #else
@@ -149,8 +148,6 @@ topology_update_package_map(unsigned int
 static inline int
 topology_update_die_map(unsigned int dieid, unsigned int cpu) { return 0; }
 static inline int topology_phys_to_logical_pkg(unsigned int pkg) { return 0; }
-static inline int topology_phys_to_logical_die(unsigned int die,
-		unsigned int cpu) { return 0; }
 static inline int topology_max_die_per_package(void) { return 1; }
 static inline int topology_max_smt_threads(void) { return 1; }
 static inline bool topology_is_primary_thread(unsigned int cpu) { return true; }
--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -288,6 +288,7 @@ bool topology_smt_supported(void)
 
 /**
  * topology_phys_to_logical_pkg - Map a physical package id to a logical
+ * @phys_pkg:	The physical package id to map
  *
  * Returns logical package id or -1 if not found
  */
@@ -304,15 +305,17 @@ int topology_phys_to_logical_pkg(unsigne
 	return -1;
 }
 EXPORT_SYMBOL(topology_phys_to_logical_pkg);
+
 /**
  * topology_phys_to_logical_die - Map a physical die id to logical
+ * @die_id:	The physical die id to map
+ * @cur_cpu:	The CPU for which the mapping is done
  *
  * Returns logical die id or -1 if not found
  */
-int topology_phys_to_logical_die(unsigned int die_id, unsigned int cur_cpu)
+static int topology_phys_to_logical_die(unsigned int die_id, unsigned int cur_cpu)
 {
-	int cpu;
-	int proc_id = cpu_data(cur_cpu).phys_proc_id;
+	int cpu, proc_id = cpu_data(cur_cpu).phys_proc_id;
 
 	for_each_possible_cpu(cpu) {
 		struct cpuinfo_x86 *c = &cpu_data(cpu);
@@ -323,7 +326,6 @@ int topology_phys_to_logical_die(unsigne
 	}
 	return -1;
 }
-EXPORT_SYMBOL(topology_phys_to_logical_die);
 
 /**
  * topology_update_package_map - Update the physical to logical package map

[patch v3 02/36] cpu/hotplug: Mark arch_disable_smp_support() and bringup_nonboot_cpus() __init

[Thomas Gleixner]

From: Thomas Gleixner <tglx@linutronix.de>

No point in keeping them around.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Michael Kelley <mikelley@microsoft.com>


---
 arch/x86/kernel/smpboot.c |    4 ++--
 kernel/cpu.c              |    2 +-
 kernel/smp.c              |    2 +-
 3 files changed, 4 insertions(+), 4 deletions(-)
---

--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -1269,9 +1269,9 @@ int native_cpu_up(unsigned int cpu, stru
 }
 
 /**
- * arch_disable_smp_support() - disables SMP support for x86 at runtime
+ * arch_disable_smp_support() - Disables SMP support for x86 at boottime
  */
-void arch_disable_smp_support(void)
+void __init arch_disable_smp_support(void)
 {
 	disable_ioapic_support();
 }
--- a/kernel/cpu.c
+++ b/kernel/cpu.c
@@ -1502,7 +1502,7 @@ int bringup_hibernate_cpu(unsigned int s
 	return 0;
 }
 
-void bringup_nonboot_cpus(unsigned int setup_max_cpus)
+void __init bringup_nonboot_cpus(unsigned int setup_max_cpus)
 {
 	unsigned int cpu;
 
--- a/kernel/smp.c
+++ b/kernel/smp.c
@@ -892,7 +892,7 @@ EXPORT_SYMBOL(setup_max_cpus);
  * SMP mode to <NUM>.
  */
 
-void __weak arch_disable_smp_support(void) { }
+void __weak __init arch_disable_smp_support(void) { }
 
 static int __init nosmp(char *str)
 {

[patch v3 03/36] x86/smpboot: Avoid pointless delay calibration if TSC is synchronized

[Thomas Gleixner]

From: Thomas Gleixner <tglx@linutronix.de>

When TSC is synchronized across sockets then there is no reason to
calibrate the delay for the first CPU which comes up on a socket.

Just reuse the existing calibration value.

This removes 100ms pointlessly wasted time from CPU hotplug per socket.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Michael Kelley <mikelley@microsoft.com>


---
 arch/x86/kernel/smpboot.c |   38 ++++++++++++++++++++++++--------------
 arch/x86/kernel/tsc.c     |   20 ++++++++++++++++----
 2 files changed, 40 insertions(+), 18 deletions(-)
---

--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -178,10 +178,7 @@ static void smp_callin(void)
 	 */
 	apic_ap_setup();
 
-	/*
-	 * Save our processor parameters. Note: this information
-	 * is needed for clock calibration.
-	 */
+	/* Save our processor parameters. */
 	smp_store_cpu_info(cpuid);
 
 	/*
@@ -192,14 +189,6 @@ static void smp_callin(void)
 
 	ap_init_aperfmperf();
 
-	/*
-	 * Get our bogomips.
-	 * Update loops_per_jiffy in cpu_data. Previous call to
-	 * smp_store_cpu_info() stored a value that is close but not as
-	 * accurate as the value just calculated.
-	 */
-	calibrate_delay();
-	cpu_data(cpuid).loops_per_jiffy = loops_per_jiffy;
 	pr_debug("Stack at about %p\n", &cpuid);
 
 	wmb();
@@ -212,8 +201,24 @@ static void smp_callin(void)
 	cpumask_set_cpu(cpuid, cpu_callin_mask);
 }
 
+static void ap_calibrate_delay(void)
+{
+	/*
+	 * Calibrate the delay loop and update loops_per_jiffy in cpu_data.
+	 * smp_store_cpu_info() stored a value that is close but not as
+	 * accurate as the value just calculated.
+	 *
+	 * As this is invoked after the TSC synchronization check,
+	 * calibrate_delay_is_known() will skip the calibration routine
+	 * when TSC is synchronized across sockets.
+	 */
+	calibrate_delay();
+	cpu_data(smp_processor_id()).loops_per_jiffy = loops_per_jiffy;
+}
+
 static int cpu0_logical_apicid;
 static int enable_start_cpu0;
+
 /*
  * Activate a secondary processor.
  */
@@ -240,10 +245,15 @@ static void notrace start_secondary(void
 
 	/* otherwise gcc will move up smp_processor_id before the cpu_init */
 	barrier();
+	/* Check TSC synchronization with the control CPU: */
+	check_tsc_sync_target();
+
 	/*
-	 * Check TSC synchronization with the boot CPU:
+	 * Calibrate the delay loop after the TSC synchronization check.
+	 * This allows to skip the calibration when TSC is synchronized
+	 * across sockets.
 	 */
-	check_tsc_sync_target();
+	ap_calibrate_delay();
 
 	speculative_store_bypass_ht_init();
 
--- a/arch/x86/kernel/tsc.c
+++ b/arch/x86/kernel/tsc.c
@@ -1598,10 +1598,7 @@ void __init tsc_init(void)
 
 #ifdef CONFIG_SMP
 /*
- * If we have a constant TSC and are using the TSC for the delay loop,
- * we can skip clock calibration if another cpu in the same socket has already
- * been calibrated. This assumes that CONSTANT_TSC applies to all
- * cpus in the socket - this should be a safe assumption.
+ * Check whether existing calibration data can be reused.
  */
 unsigned long calibrate_delay_is_known(void)
 {
@@ -1609,6 +1606,21 @@ unsigned long calibrate_delay_is_known(v
 	int constant_tsc = cpu_has(&cpu_data(cpu), X86_FEATURE_CONSTANT_TSC);
 	const struct cpumask *mask = topology_core_cpumask(cpu);
 
+	/*
+	 * If TSC has constant frequency and TSC is synchronized across
+	 * sockets then reuse CPU0 calibration.
+	 */
+	if (constant_tsc && !tsc_unstable)
+		return cpu_data(0).loops_per_jiffy;
+
+	/*
+	 * If TSC has constant frequency and TSC is not synchronized across
+	 * sockets and this is not the first CPU in the socket, then reuse
+	 * the calibration value of an already online CPU on that socket.
+	 *
+	 * This assumes that CONSTANT_TSC is consistent for all CPUs in a
+	 * socket.
+	 */
 	if (!constant_tsc || !mask)
 		return 0;
 

[patch v3 04/36] x86/smpboot: Rename start_cpu0() to soft_restart_cpu()

[Thomas Gleixner]

From: Thomas Gleixner <tglx@linutronix.de>

This is used in the SEV play_dead() implementation to re-online CPUs. But
that has nothing to do with CPU0.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Michael Kelley <mikelley@microsoft.com>


---
 arch/x86/include/asm/cpu.h   |    2 +-
 arch/x86/kernel/callthunks.c |    2 +-
 arch/x86/kernel/head_32.S    |   10 +++++-----
 arch/x86/kernel/head_64.S    |   10 +++++-----
 arch/x86/kernel/sev.c        |    2 +-
 5 files changed, 13 insertions(+), 13 deletions(-)
---

--- a/arch/x86/include/asm/cpu.h
+++ b/arch/x86/include/asm/cpu.h
@@ -30,7 +30,7 @@ struct x86_cpu {
 #ifdef CONFIG_HOTPLUG_CPU
 extern int arch_register_cpu(int num);
 extern void arch_unregister_cpu(int);
-extern void start_cpu0(void);
+extern void soft_restart_cpu(void);
 #ifdef CONFIG_DEBUG_HOTPLUG_CPU0
 extern int _debug_hotplug_cpu(int cpu, int action);
 #endif
--- a/arch/x86/kernel/callthunks.c
+++ b/arch/x86/kernel/callthunks.c
@@ -134,7 +134,7 @@ static bool skip_addr(void *dest)
 	if (dest == ret_from_fork)
 		return true;
 #ifdef CONFIG_HOTPLUG_CPU
-	if (dest == start_cpu0)
+	if (dest == soft_restart_cpu)
 		return true;
 #endif
 #ifdef CONFIG_FUNCTION_TRACER
--- a/arch/x86/kernel/head_32.S
+++ b/arch/x86/kernel/head_32.S
@@ -140,16 +140,16 @@ SYM_CODE_END(startup_32)
 
 #ifdef CONFIG_HOTPLUG_CPU
 /*
- * Boot CPU0 entry point. It's called from play_dead(). Everything has been set
- * up already except stack. We just set up stack here. Then call
- * start_secondary().
+ * Entry point for soft restart of a CPU. Invoked from xxx_play_dead() for
+ * restarting the boot CPU or for restarting SEV guest CPUs after CPU hot
+ * unplug. Everything is set up already except the stack.
  */
-SYM_FUNC_START(start_cpu0)
+SYM_FUNC_START(soft_restart_cpu)
 	movl initial_stack, %ecx
 	movl %ecx, %esp
 	call *(initial_code)
 1:	jmp 1b
-SYM_FUNC_END(start_cpu0)
+SYM_FUNC_END(soft_restart_cpu)
 #endif
 
 /*
--- a/arch/x86/kernel/head_64.S
+++ b/arch/x86/kernel/head_64.S
@@ -377,11 +377,11 @@ SYM_CODE_END(secondary_startup_64)
 
 #ifdef CONFIG_HOTPLUG_CPU
 /*
- * Boot CPU0 entry point. It's called from play_dead(). Everything has been set
- * up already except stack. We just set up stack here. Then call
- * start_secondary() via .Ljump_to_C_code.
+ * Entry point for soft restart of a CPU. Invoked from xxx_play_dead() for
+ * restarting the boot CPU or for restarting SEV guest CPUs after CPU hot
+ * unplug. Everything is set up already except the stack.
  */
-SYM_CODE_START(start_cpu0)
+SYM_CODE_START(soft_restart_cpu)
 	ANNOTATE_NOENDBR
 	UNWIND_HINT_END_OF_STACK
 
@@ -390,7 +390,7 @@ SYM_CODE_START(start_cpu0)
 	movq	TASK_threadsp(%rcx), %rsp
 
 	jmp	.Ljump_to_C_code
-SYM_CODE_END(start_cpu0)
+SYM_CODE_END(soft_restart_cpu)
 #endif
 
 #ifdef CONFIG_AMD_MEM_ENCRYPT
--- a/arch/x86/kernel/sev.c
+++ b/arch/x86/kernel/sev.c
@@ -1328,7 +1328,7 @@ static void sev_es_play_dead(void)
 	 * If we get here, the VCPU was woken up again. Jump to CPU
 	 * startup code to get it back online.
 	 */
-	start_cpu0();
+	soft_restart_cpu();
 }
 #else  /* CONFIG_HOTPLUG_CPU */
 #define sev_es_play_dead	native_play_dead

[patch v3 05/36] x86/topology: Remove CPU0 hotplug option

[Thomas Gleixner]

From: Thomas Gleixner <tglx@linutronix.de>

This was introduced together with commit e1c467e69040 ("x86, hotplug: Wake
up CPU0 via NMI instead of INIT, SIPI, SIPI") to eventually support
physical hotplug of CPU0:

 "We'll change this code in the future to wake up hard offlined CPU0 if
  real platform and request are available."

11 years later this has not happened and physical hotplug is not officially
supported. Remove the cruft.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Michael Kelley <mikelley@microsoft.com>


---
 Documentation/admin-guide/kernel-parameters.txt |   14 ---
 Documentation/core-api/cpu_hotplug.rst          |   13 ---
 arch/x86/Kconfig                                |   43 ----------
 arch/x86/include/asm/cpu.h                      |    3 
 arch/x86/kernel/topology.c                      |   98 ------------------------
 arch/x86/power/cpu.c                            |   37 ---------
 6 files changed, 6 insertions(+), 202 deletions(-)
---

--- a/Documentation/admin-guide/kernel-parameters.txt
+++ b/Documentation/admin-guide/kernel-parameters.txt
@@ -818,20 +818,6 @@
 			Format:
 			<first_slot>,<last_slot>,<port>,<enum_bit>[,<debug>]
 
-	cpu0_hotplug	[X86] Turn on CPU0 hotplug feature when
-			CONFIG_BOOTPARAM_HOTPLUG_CPU0 is off.
-			Some features depend on CPU0. Known dependencies are:
-			1. Resume from suspend/hibernate depends on CPU0.
-			Suspend/hibernate will fail if CPU0 is offline and you
-			need to online CPU0 before suspend/hibernate.
-			2. PIC interrupts also depend on CPU0. CPU0 can't be
-			removed if a PIC interrupt is detected.
-			It's said poweroff/reboot may depend on CPU0 on some
-			machines although I haven't seen such issues so far
-			after CPU0 is offline on a few tested machines.
-			If the dependencies are under your control, you can
-			turn on cpu0_hotplug.
-
 	cpuidle.off=1	[CPU_IDLE]
 			disable the cpuidle sub-system
 
--- a/Documentation/core-api/cpu_hotplug.rst
+++ b/Documentation/core-api/cpu_hotplug.rst
@@ -127,17 +127,8 @@ Once the CPU is shutdown, it will be rem
  $ echo 1 > /sys/devices/system/cpu/cpu4/online
  smpboot: Booting Node 0 Processor 4 APIC 0x1
 
-The CPU is usable again. This should work on all CPUs. CPU0 is often special
-and excluded from CPU hotplug. On X86 the kernel option
-*CONFIG_BOOTPARAM_HOTPLUG_CPU0* has to be enabled in order to be able to
-shutdown CPU0. Alternatively the kernel command option *cpu0_hotplug* can be
-used. Some known dependencies of CPU0:
-
-* Resume from hibernate/suspend. Hibernate/suspend will fail if CPU0 is offline.
-* PIC interrupts. CPU0 can't be removed if a PIC interrupt is detected.
-
-Please let Fenghua Yu <fenghua.yu@intel.com> know if you find any dependencies
-on CPU0.
+The CPU is usable again. This should work on all CPUs, but CPU0 is often special
+and excluded from CPU hotplug.
 
 The CPU hotplug coordination
 ============================
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -2305,49 +2305,6 @@ config HOTPLUG_CPU
 	def_bool y
 	depends on SMP
 
-config BOOTPARAM_HOTPLUG_CPU0
-	bool "Set default setting of cpu0_hotpluggable"
-	depends on HOTPLUG_CPU
-	help
-	  Set whether default state of cpu0_hotpluggable is on or off.
-
-	  Say Y here to enable CPU0 hotplug by default. If this switch
-	  is turned on, there is no need to give cpu0_hotplug kernel
-	  parameter and the CPU0 hotplug feature is enabled by default.
-
-	  Please note: there are two known CPU0 dependencies if you want
-	  to enable the CPU0 hotplug feature either by this switch or by
-	  cpu0_hotplug kernel parameter.
-
-	  First, resume from hibernate or suspend always starts from CPU0.
-	  So hibernate and suspend are prevented if CPU0 is offline.
-
-	  Second dependency is PIC interrupts always go to CPU0. CPU0 can not
-	  offline if any interrupt can not migrate out of CPU0. There may
-	  be other CPU0 dependencies.
-
-	  Please make sure the dependencies are under your control before
-	  you enable this feature.
-
-	  Say N if you don't want to enable CPU0 hotplug feature by default.
-	  You still can enable the CPU0 hotplug feature at boot by kernel
-	  parameter cpu0_hotplug.
-
-config DEBUG_HOTPLUG_CPU0
-	def_bool n
-	prompt "Debug CPU0 hotplug"
-	depends on HOTPLUG_CPU
-	help
-	  Enabling this option offlines CPU0 (if CPU0 can be offlined) as
-	  soon as possible and boots up userspace with CPU0 offlined. User
-	  can online CPU0 back after boot time.
-
-	  To debug CPU0 hotplug, you need to enable CPU0 offline/online
-	  feature by either turning on CONFIG_BOOTPARAM_HOTPLUG_CPU0 during
-	  compilation or giving cpu0_hotplug kernel parameter at boot.
-
-	  If unsure, say N.
-
 config COMPAT_VDSO
 	def_bool n
 	prompt "Disable the 32-bit vDSO (needed for glibc 2.3.3)"
--- a/arch/x86/include/asm/cpu.h
+++ b/arch/x86/include/asm/cpu.h
@@ -31,9 +31,6 @@ struct x86_cpu {
 extern int arch_register_cpu(int num);
 extern void arch_unregister_cpu(int);
 extern void soft_restart_cpu(void);
-#ifdef CONFIG_DEBUG_HOTPLUG_CPU0
-extern int _debug_hotplug_cpu(int cpu, int action);
-#endif
 #endif
 
 extern void ap_init_aperfmperf(void);
--- a/arch/x86/kernel/topology.c
+++ b/arch/x86/kernel/topology.c
@@ -38,102 +38,12 @@
 static DEFINE_PER_CPU(struct x86_cpu, cpu_devices);
 
 #ifdef CONFIG_HOTPLUG_CPU
-
-#ifdef CONFIG_BOOTPARAM_HOTPLUG_CPU0
-static int cpu0_hotpluggable = 1;
-#else
-static int cpu0_hotpluggable;
-static int __init enable_cpu0_hotplug(char *str)
-{
-	cpu0_hotpluggable = 1;
-	return 1;
-}
-
-__setup("cpu0_hotplug", enable_cpu0_hotplug);
-#endif
-
-#ifdef CONFIG_DEBUG_HOTPLUG_CPU0
-/*
- * This function offlines a CPU as early as possible and allows userspace to
- * boot up without the CPU. The CPU can be onlined back by user after boot.
- *
- * This is only called for debugging CPU offline/online feature.
- */
-int _debug_hotplug_cpu(int cpu, int action)
-{
-	int ret;
-
-	if (!cpu_is_hotpluggable(cpu))
-		return -EINVAL;
-
-	switch (action) {
-	case 0:
-		ret = remove_cpu(cpu);
-		if (!ret)
-			pr_info("DEBUG_HOTPLUG_CPU0: CPU %u is now offline\n", cpu);
-		else
-			pr_debug("Can't offline CPU%d.\n", cpu);
-		break;
-	case 1:
-		ret = add_cpu(cpu);
-		if (ret)
-			pr_debug("Can't online CPU%d.\n", cpu);
-
-		break;
-	default:
-		ret = -EINVAL;
-	}
-
-	return ret;
-}
-
-static int __init debug_hotplug_cpu(void)
+int arch_register_cpu(int cpu)
 {
-	_debug_hotplug_cpu(0, 0);
-	return 0;
-}
-
-late_initcall_sync(debug_hotplug_cpu);
-#endif /* CONFIG_DEBUG_HOTPLUG_CPU0 */
-
-int arch_register_cpu(int num)
-{
-	struct cpuinfo_x86 *c = &cpu_data(num);
-
-	/*
-	 * Currently CPU0 is only hotpluggable on Intel platforms. Other
-	 * vendors can add hotplug support later.
-	 * Xen PV guests don't support CPU0 hotplug at all.
-	 */
-	if (c->x86_vendor != X86_VENDOR_INTEL ||
-	    cpu_feature_enabled(X86_FEATURE_XENPV))
-		cpu0_hotpluggable = 0;
-
-	/*
-	 * Two known BSP/CPU0 dependencies: Resume from suspend/hibernate
-	 * depends on BSP. PIC interrupts depend on BSP.
-	 *
-	 * If the BSP dependencies are under control, one can tell kernel to
-	 * enable BSP hotplug. This basically adds a control file and
-	 * one can attempt to offline BSP.
-	 */
-	if (num == 0 && cpu0_hotpluggable) {
-		unsigned int irq;
-		/*
-		 * We won't take down the boot processor on i386 if some
-		 * interrupts only are able to be serviced by the BSP in PIC.
-		 */
-		for_each_active_irq(irq) {
-			if (!IO_APIC_IRQ(irq) && irq_has_action(irq)) {
-				cpu0_hotpluggable = 0;
-				break;
-			}
-		}
-	}
-	if (num || cpu0_hotpluggable)
-		per_cpu(cpu_devices, num).cpu.hotpluggable = 1;
+	struct x86_cpu *xc = per_cpu_ptr(&cpu_devices, cpu);
 
-	return register_cpu(&per_cpu(cpu_devices, num).cpu, num);
+	xc->cpu.hotpluggable = cpu > 0;
+	return register_cpu(&xc->cpu, cpu);
 }
 EXPORT_SYMBOL(arch_register_cpu);
 
--- a/arch/x86/power/cpu.c
+++ b/arch/x86/power/cpu.c
@@ -351,43 +351,6 @@ static int bsp_pm_callback(struct notifi
 	case PM_HIBERNATION_PREPARE:
 		ret = bsp_check();
 		break;
-#ifdef CONFIG_DEBUG_HOTPLUG_CPU0
-	case PM_RESTORE_PREPARE:
-		/*
-		 * When system resumes from hibernation, online CPU0 because
-		 * 1. it's required for resume and
-		 * 2. the CPU was online before hibernation
-		 */
-		if (!cpu_online(0))
-			_debug_hotplug_cpu(0, 1);
-		break;
-	case PM_POST_RESTORE:
-		/*
-		 * When a resume really happens, this code won't be called.
-		 *
-		 * This code is called only when user space hibernation software
-		 * prepares for snapshot device during boot time. So we just
-		 * call _debug_hotplug_cpu() to restore to CPU0's state prior to
-		 * preparing the snapshot device.
-		 *
-		 * This works for normal boot case in our CPU0 hotplug debug
-		 * mode, i.e. CPU0 is offline and user mode hibernation
-		 * software initializes during boot time.
-		 *
-		 * If CPU0 is online and user application accesses snapshot
-		 * device after boot time, this will offline CPU0 and user may
-		 * see different CPU0 state before and after accessing
-		 * the snapshot device. But hopefully this is not a case when
-		 * user debugging CPU0 hotplug. Even if users hit this case,
-		 * they can easily online CPU0 back.
-		 *
-		 * To simplify this debug code, we only consider normal boot
-		 * case. Otherwise we need to remember CPU0's state and restore
-		 * to that state and resolve racy conditions etc.
-		 */
-		_debug_hotplug_cpu(0, 0);
-		break;
-#endif
 	default:
 		break;
 	}

[patch v3 06/36] x86/smpboot: Remove the CPU0 hotplug kludge

[Thomas Gleixner]

From: Thomas Gleixner <tglx@linutronix.de>

This was introduced with commit e1c467e69040 ("x86, hotplug: Wake up CPU0
via NMI instead of INIT, SIPI, SIPI") to eventually support physical
hotplug of CPU0:

 "We'll change this code in the future to wake up hard offlined CPU0 if
  real platform and request are available."

11 years later this has not happened and physical hotplug is not officially
supported. Remove the cruft.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Michael Kelley <mikelley@microsoft.com>


---
 arch/x86/include/asm/apic.h   |    1 
 arch/x86/include/asm/smp.h    |    1 
 arch/x86/kernel/smpboot.c     |  170 +++---------------------------------------
 drivers/acpi/processor_idle.c |    4 
 4 files changed, 14 insertions(+), 162 deletions(-)
---

--- a/arch/x86/include/asm/apic.h
+++ b/arch/x86/include/asm/apic.h
@@ -377,7 +377,6 @@ extern struct apic *__apicdrivers[], *__
  * APIC functionality to boot other CPUs - only used on SMP:
  */
 #ifdef CONFIG_SMP
-extern int wakeup_secondary_cpu_via_nmi(int apicid, unsigned long start_eip);
 extern int lapic_can_unplug_cpu(void);
 #endif
 
--- a/arch/x86/include/asm/smp.h
+++ b/arch/x86/include/asm/smp.h
@@ -130,7 +130,6 @@ void native_play_dead(void);
 void play_dead_common(void);
 void wbinvd_on_cpu(int cpu);
 int wbinvd_on_all_cpus(void);
-void cond_wakeup_cpu0(void);
 
 void native_smp_send_reschedule(int cpu);
 void native_send_call_func_ipi(const struct cpumask *mask);
--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -216,9 +216,6 @@ static void ap_calibrate_delay(void)
 	cpu_data(smp_processor_id()).loops_per_jiffy = loops_per_jiffy;
 }
 
-static int cpu0_logical_apicid;
-static int enable_start_cpu0;
-
 /*
  * Activate a secondary processor.
  */
@@ -241,8 +238,6 @@ static void notrace start_secondary(void
 	x86_cpuinit.early_percpu_clock_init();
 	smp_callin();
 
-	enable_start_cpu0 = 0;
-
 	/* otherwise gcc will move up smp_processor_id before the cpu_init */
 	barrier();
 	/* Check TSC synchronization with the control CPU: */
@@ -410,7 +405,7 @@ void smp_store_cpu_info(int id)
 	c->cpu_index = id;
 	/*
 	 * During boot time, CPU0 has this setup already. Save the info when
-	 * bringing up AP or offlined CPU0.
+	 * bringing up an AP.
 	 */
 	identify_secondary_cpu(c);
 	c->initialized = true;
@@ -807,51 +802,14 @@ static void __init smp_quirk_init_udelay
 }
 
 /*
- * Poke the other CPU in the eye via NMI to wake it up. Remember that the normal
- * INIT, INIT, STARTUP sequence will reset the chip hard for us, and this
- * won't ... remember to clear down the APIC, etc later.
- */
-int
-wakeup_secondary_cpu_via_nmi(int apicid, unsigned long start_eip)
-{
-	u32 dm = apic->dest_mode_logical ? APIC_DEST_LOGICAL : APIC_DEST_PHYSICAL;
-	unsigned long send_status, accept_status = 0;
-	int maxlvt;
-
-	/* Target chip */
-	/* Boot on the stack */
-	/* Kick the second */
-	apic_icr_write(APIC_DM_NMI | dm, apicid);
-
-	pr_debug("Waiting for send to finish...\n");
-	send_status = safe_apic_wait_icr_idle();
-
-	/*
-	 * Give the other CPU some time to accept the IPI.
-	 */
-	udelay(200);
-	if (APIC_INTEGRATED(boot_cpu_apic_version)) {
-		maxlvt = lapic_get_maxlvt();
-		if (maxlvt > 3)			/* Due to the Pentium erratum 3AP.  */
-			apic_write(APIC_ESR, 0);
-		accept_status = (apic_read(APIC_ESR) & 0xEF);
-	}
-	pr_debug("NMI sent\n");
-
-	if (send_status)
-		pr_err("APIC never delivered???\n");
-	if (accept_status)
-		pr_err("APIC delivery error (%lx)\n", accept_status);
-
-	return (send_status | accept_status);
-}
-
-static int
-wakeup_secondary_cpu_via_init(int phys_apicid, unsigned long start_eip)
+ * Wake up AP by INIT, INIT, STARTUP sequence.
+ */
+static int wakeup_secondary_cpu_via_init(int phys_apicid, unsigned long start_eip)
 {
 	unsigned long send_status = 0, accept_status = 0;
 	int maxlvt, num_starts, j;
 
+	preempt_disable();
 	maxlvt = lapic_get_maxlvt();
 
 	/*
@@ -957,6 +915,7 @@ wakeup_secondary_cpu_via_init(int phys_a
 	if (accept_status)
 		pr_err("APIC delivery error (%lx)\n", accept_status);
 
+	preempt_enable();
 	return (send_status | accept_status);
 }
 
@@ -997,67 +956,6 @@ static void announce_cpu(int cpu, int ap
 			node, cpu, apicid);
 }
 
-static int wakeup_cpu0_nmi(unsigned int cmd, struct pt_regs *regs)
-{
-	int cpu;
-
-	cpu = smp_processor_id();
-	if (cpu == 0 && !cpu_online(cpu) && enable_start_cpu0)
-		return NMI_HANDLED;
-
-	return NMI_DONE;
-}
-
-/*
- * Wake up AP by INIT, INIT, STARTUP sequence.
- *
- * Instead of waiting for STARTUP after INITs, BSP will execute the BIOS
- * boot-strap code which is not a desired behavior for waking up BSP. To
- * void the boot-strap code, wake up CPU0 by NMI instead.
- *
- * This works to wake up soft offlined CPU0 only. If CPU0 is hard offlined
- * (i.e. physically hot removed and then hot added), NMI won't wake it up.
- * We'll change this code in the future to wake up hard offlined CPU0 if
- * real platform and request are available.
- */
-static int
-wakeup_cpu_via_init_nmi(int cpu, unsigned long start_ip, int apicid,
-	       int *cpu0_nmi_registered)
-{
-	int id;
-	int boot_error;
-
-	preempt_disable();
-
-	/*
-	 * Wake up AP by INIT, INIT, STARTUP sequence.
-	 */
-	if (cpu) {
-		boot_error = wakeup_secondary_cpu_via_init(apicid, start_ip);
-		goto out;
-	}
-
-	/*
-	 * Wake up BSP by nmi.
-	 *
-	 * Register a NMI handler to help wake up CPU0.
-	 */
-	boot_error = register_nmi_handler(NMI_LOCAL,
-					  wakeup_cpu0_nmi, 0, "wake_cpu0");
-
-	if (!boot_error) {
-		enable_start_cpu0 = 1;
-		*cpu0_nmi_registered = 1;
-		id = apic->dest_mode_logical ? cpu0_logical_apicid : apicid;
-		boot_error = wakeup_secondary_cpu_via_nmi(id, start_ip);
-	}
-
-out:
-	preempt_enable();
-
-	return boot_error;
-}
-
 int common_cpu_up(unsigned int cpu, struct task_struct *idle)
 {
 	int ret;
@@ -1086,8 +984,7 @@ int common_cpu_up(unsigned int cpu, stru
  * Returns zero if CPU booted OK, else error code from
  * ->wakeup_secondary_cpu.
  */
-static int do_boot_cpu(int apicid, int cpu, struct task_struct *idle,
-		       int *cpu0_nmi_registered)
+static int do_boot_cpu(int apicid, int cpu, struct task_struct *idle)
 {
 	/* start_ip had better be page-aligned! */
 	unsigned long start_ip = real_mode_header->trampoline_start;
@@ -1120,7 +1017,6 @@ static int do_boot_cpu(int apicid, int c
 	 * This grunge runs the startup process for
 	 * the targeted processor.
 	 */
-
 	if (x86_platform.legacy.warm_reset) {
 
 		pr_debug("Setting warm reset code and vector.\n");
@@ -1149,15 +1045,14 @@ static int do_boot_cpu(int apicid, int c
 	 * - Use a method from the APIC driver if one defined, with wakeup
 	 *   straight to 64-bit mode preferred over wakeup to RM.
 	 * Otherwise,
-	 * - Use an INIT boot APIC message for APs or NMI for BSP.
+	 * - Use an INIT boot APIC message
 	 */
 	if (apic->wakeup_secondary_cpu_64)
 		boot_error = apic->wakeup_secondary_cpu_64(apicid, start_ip);
 	else if (apic->wakeup_secondary_cpu)
 		boot_error = apic->wakeup_secondary_cpu(apicid, start_ip);
 	else
-		boot_error = wakeup_cpu_via_init_nmi(cpu, start_ip, apicid,
-						     cpu0_nmi_registered);
+		boot_error = wakeup_secondary_cpu_via_init(apicid, start_ip);
 
 	if (!boot_error) {
 		/*
@@ -1206,9 +1101,8 @@ static int do_boot_cpu(int apicid, int c
 int native_cpu_up(unsigned int cpu, struct task_struct *tidle)
 {
 	int apicid = apic->cpu_present_to_apicid(cpu);
-	int cpu0_nmi_registered = 0;
 	unsigned long flags;
-	int err, ret = 0;
+	int err;
 
 	lockdep_assert_irqs_enabled();
 
@@ -1247,11 +1141,10 @@ int native_cpu_up(unsigned int cpu, stru
 	if (err)
 		return err;
 
-	err = do_boot_cpu(apicid, cpu, tidle, &cpu0_nmi_registered);
+	err = do_boot_cpu(apicid, cpu, tidle);
 	if (err) {
 		pr_err("do_boot_cpu failed(%d) to wakeup CPU#%u\n", err, cpu);
-		ret = -EIO;
-		goto unreg_nmi;
+		return err;
 	}
 
 	/*
@@ -1267,15 +1160,7 @@ int native_cpu_up(unsigned int cpu, stru
 		touch_nmi_watchdog();
 	}
 
-unreg_nmi:
-	/*
-	 * Clean up the nmi handler. Do this after the callin and callout sync
-	 * to avoid impact of possible long unregister time.
-	 */
-	if (cpu0_nmi_registered)
-		unregister_nmi_handler(NMI_LOCAL, "wake_cpu0");
-
-	return ret;
+	return 0;
 }
 
 /**
@@ -1373,14 +1258,6 @@ static void __init smp_cpu_index_default
 	}
 }
 
-static void __init smp_get_logical_apicid(void)
-{
-	if (x2apic_mode)
-		cpu0_logical_apicid = apic_read(APIC_LDR);
-	else
-		cpu0_logical_apicid = GET_APIC_LOGICAL_ID(apic_read(APIC_LDR));
-}
-
 void __init smp_prepare_cpus_common(void)
 {
 	unsigned int i;
@@ -1443,8 +1320,6 @@ void __init native_smp_prepare_cpus(unsi
 	/* Setup local timer */
 	x86_init.timers.setup_percpu_clockev();
 
-	smp_get_logical_apicid();
-
 	pr_info("CPU0: ");
 	print_cpu_info(&cpu_data(0));
 
@@ -1752,18 +1627,6 @@ void play_dead_common(void)
 	local_irq_disable();
 }
 
-/**
- * cond_wakeup_cpu0 - Wake up CPU0 if needed.
- *
- * If NMI wants to wake up CPU0, start CPU0.
- */
-void cond_wakeup_cpu0(void)
-{
-	if (smp_processor_id() == 0 && enable_start_cpu0)
-		start_cpu0();
-}
-EXPORT_SYMBOL_GPL(cond_wakeup_cpu0);
-
 /*
  * We need to flush the caches before going to sleep, lest we have
  * dirty data in our caches when we come back up.
@@ -1831,8 +1694,6 @@ static inline void mwait_play_dead(void)
 		__monitor(mwait_ptr, 0, 0);
 		mb();
 		__mwait(eax, 0);
-
-		cond_wakeup_cpu0();
 	}
 }
 
@@ -1841,11 +1702,8 @@ void __noreturn hlt_play_dead(void)
 	if (__this_cpu_read(cpu_info.x86) >= 4)
 		wbinvd();
 
-	while (1) {
+	while (1)
 		native_halt();
-
-		cond_wakeup_cpu0();
-	}
 }
 
 void native_play_dead(void)
--- a/drivers/acpi/processor_idle.c
+++ b/drivers/acpi/processor_idle.c
@@ -597,10 +597,6 @@ static int acpi_idle_play_dead(struct cp
 			io_idle(cx->address);
 		} else
 			return -ENODEV;
-
-#if defined(CONFIG_X86) && defined(CONFIG_HOTPLUG_CPU)
-		cond_wakeup_cpu0();
-#endif
 	}
 
 	/* Never reached */

[patch v3 07/36] x86/smpboot: Restrict soft_restart_cpu() to SEV

[Thomas Gleixner]

From: Thomas Gleixner <tglx@linutronix.de>

Now that the CPU0 hotplug cruft is gone, the only user is AMD SEV.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Michael Kelley <mikelley@microsoft.com>


---
 arch/x86/kernel/callthunks.c |    2 +-
 arch/x86/kernel/head_32.S    |   14 --------------
 arch/x86/kernel/head_64.S    |    2 +-
 3 files changed, 2 insertions(+), 16 deletions(-)
---

--- a/arch/x86/kernel/callthunks.c
+++ b/arch/x86/kernel/callthunks.c
@@ -133,7 +133,7 @@ static bool skip_addr(void *dest)
 	/* Accounts directly */
 	if (dest == ret_from_fork)
 		return true;
-#ifdef CONFIG_HOTPLUG_CPU
+#if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_AMD_MEM_ENCRYPT)
 	if (dest == soft_restart_cpu)
 		return true;
 #endif
--- a/arch/x86/kernel/head_32.S
+++ b/arch/x86/kernel/head_32.S
@@ -138,20 +138,6 @@ SYM_CODE_START(startup_32)
 	jmp .Ldefault_entry
 SYM_CODE_END(startup_32)
 
-#ifdef CONFIG_HOTPLUG_CPU
-/*
- * Entry point for soft restart of a CPU. Invoked from xxx_play_dead() for
- * restarting the boot CPU or for restarting SEV guest CPUs after CPU hot
- * unplug. Everything is set up already except the stack.
- */
-SYM_FUNC_START(soft_restart_cpu)
-	movl initial_stack, %ecx
-	movl %ecx, %esp
-	call *(initial_code)
-1:	jmp 1b
-SYM_FUNC_END(soft_restart_cpu)
-#endif
-
 /*
  * Non-boot CPU entry point; entered from trampoline.S
  * We can't lgdt here, because lgdt itself uses a data segment, but
--- a/arch/x86/kernel/head_64.S
+++ b/arch/x86/kernel/head_64.S
@@ -375,7 +375,7 @@ SYM_CODE_END(secondary_startup_64)
 #include "verify_cpu.S"
 #include "sev_verify_cbit.S"
 
-#ifdef CONFIG_HOTPLUG_CPU
+#if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_AMD_MEM_ENCRYPT)
 /*
  * Entry point for soft restart of a CPU. Invoked from xxx_play_dead() for
  * restarting the boot CPU or for restarting SEV guest CPUs after CPU hot

[patch v3 08/36] x86/smpboot: Split up native_cpu_up() into separate phases and document them

[Thomas Gleixner]

From: David Woodhouse <dwmw@amazon.co.uk>

There are four logical parts to what native_cpu_up() does on the BSP (or
on the controlling CPU for a later hotplug):

 1) Wake the AP by sending the INIT/SIPI/SIPI sequence.

 2) Wait for the AP to make it as far as wait_for_master_cpu() which
    sets that CPU's bit in cpu_initialized_mask, then sets the bit in
    cpu_callout_mask to let the AP proceed through cpu_init().

 3) Wait for the AP to finish cpu_init() and get as far as the
    smp_callin() call, which sets that CPU's bit in cpu_callin_mask.

 4) Perform the TSC synchronization and wait for the AP to actually
    mark itself online in cpu_online_mask.

In preparation to allow these phases to operate in parallel on multiple
APs, split them out into separate functions and document the interactions
a little more clearly in both the BP and AP code paths.

No functional change intended.

Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Michael Kelley <mikelley@microsoft.com>


---
 arch/x86/kernel/smpboot.c |  187 +++++++++++++++++++++++++++++-----------------
 1 file changed, 121 insertions(+), 66 deletions(-)
---

--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -193,6 +193,10 @@ static void smp_callin(void)
 
 	wmb();
 
+	/*
+	 * This runs the AP through all the cpuhp states to its target
+	 * state (CPUHP_ONLINE in the case of serial bringup).
+	 */
 	notify_cpu_starting(cpuid);
 
 	/*
@@ -233,14 +237,31 @@ static void notrace start_secondary(void
 	load_cr3(swapper_pg_dir);
 	__flush_tlb_all();
 #endif
+	/*
+	 * Sync point with wait_cpu_initialized(). Before proceeding through
+	 * cpu_init(), the AP will call wait_for_master_cpu() which sets its
+	 * own bit in cpu_initialized_mask and then waits for the BSP to set
+	 * its bit in cpu_callout_mask to release it.
+	 */
 	cpu_init_secondary();
 	rcu_cpu_starting(raw_smp_processor_id());
 	x86_cpuinit.early_percpu_clock_init();
+
+	/*
+	 * Sync point with wait_cpu_callin(). The AP doesn't wait here
+	 * but just sets the bit to let the controlling CPU (BSP) know that
+	 * it's got this far.
+	 */
 	smp_callin();
 
-	/* otherwise gcc will move up smp_processor_id before the cpu_init */
+	/* Otherwise gcc will move up smp_processor_id() before cpu_init() */
 	barrier();
-	/* Check TSC synchronization with the control CPU: */
+
+	/*
+	 * Check TSC synchronization with the control CPU, which will do
+	 * its part of this from wait_cpu_online(), making it an implicit
+	 * synchronization point.
+	 */
 	check_tsc_sync_target();
 
 	/*
@@ -259,6 +280,7 @@ static void notrace start_secondary(void
 	 * half valid vector space.
 	 */
 	lock_vector_lock();
+	/* Sync point with do_wait_cpu_online() */
 	set_cpu_online(smp_processor_id(), true);
 	lapic_online();
 	unlock_vector_lock();
@@ -981,17 +1003,13 @@ int common_cpu_up(unsigned int cpu, stru
 /*
  * NOTE - on most systems this is a PHYSICAL apic ID, but on multiquad
  * (ie clustered apic addressing mode), this is a LOGICAL apic ID.
- * Returns zero if CPU booted OK, else error code from
+ * Returns zero if startup was successfully sent, else error code from
  * ->wakeup_secondary_cpu.
  */
 static int do_boot_cpu(int apicid, int cpu, struct task_struct *idle)
 {
-	/* start_ip had better be page-aligned! */
 	unsigned long start_ip = real_mode_header->trampoline_start;
 
-	unsigned long boot_error = 0;
-	unsigned long timeout;
-
 #ifdef CONFIG_X86_64
 	/* If 64-bit wakeup method exists, use the 64-bit mode trampoline IP */
 	if (apic->wakeup_secondary_cpu_64)
@@ -1048,60 +1066,89 @@ static int do_boot_cpu(int apicid, int c
 	 * - Use an INIT boot APIC message
 	 */
 	if (apic->wakeup_secondary_cpu_64)
-		boot_error = apic->wakeup_secondary_cpu_64(apicid, start_ip);
+		return apic->wakeup_secondary_cpu_64(apicid, start_ip);
 	else if (apic->wakeup_secondary_cpu)
-		boot_error = apic->wakeup_secondary_cpu(apicid, start_ip);
-	else
-		boot_error = wakeup_secondary_cpu_via_init(apicid, start_ip);
+		return apic->wakeup_secondary_cpu(apicid, start_ip);
 
-	if (!boot_error) {
-		/*
-		 * Wait 10s total for first sign of life from AP
-		 */
-		boot_error = -1;
-		timeout = jiffies + 10*HZ;
-		while (time_before(jiffies, timeout)) {
-			if (cpumask_test_cpu(cpu, cpu_initialized_mask)) {
-				/*
-				 * Tell AP to proceed with initialization
-				 */
-				cpumask_set_cpu(cpu, cpu_callout_mask);
-				boot_error = 0;
-				break;
-			}
-			schedule();
-		}
-	}
+	return wakeup_secondary_cpu_via_init(apicid, start_ip);
+}
 
-	if (!boot_error) {
-		/*
-		 * Wait till AP completes initial initialization
-		 */
-		while (!cpumask_test_cpu(cpu, cpu_callin_mask)) {
-			/*
-			 * Allow other tasks to run while we wait for the
-			 * AP to come online. This also gives a chance
-			 * for the MTRR work(triggered by the AP coming online)
-			 * to be completed in the stop machine context.
-			 */
-			schedule();
-		}
-	}
+static int wait_cpu_cpumask(unsigned int cpu, const struct cpumask *mask)
+{
+	unsigned long timeout;
 
-	if (x86_platform.legacy.warm_reset) {
-		/*
-		 * Cleanup possible dangling ends...
-		 */
-		smpboot_restore_warm_reset_vector();
+	/*
+	 * Wait up to 10s for the CPU to report in.
+	 */
+	timeout = jiffies + 10*HZ;
+	while (time_before(jiffies, timeout)) {
+		if (cpumask_test_cpu(cpu, mask))
+			return 0;
+
+		schedule();
 	}
+	return -1;
+}
 
-	return boot_error;
+/*
+ * Bringup step two: Wait for the target AP to reach cpu_init_secondary()
+ * and thus wait_for_master_cpu(), then set cpu_callout_mask to allow it
+ * to proceed.  The AP will then proceed past setting its 'callin' bit
+ * and end up waiting in check_tsc_sync_target() until we reach
+ * do_wait_cpu_online() to tend to it.
+ */
+static int wait_cpu_initialized(unsigned int cpu)
+{
+	/*
+	 * Wait for first sign of life from AP.
+	 */
+	if (wait_cpu_cpumask(cpu, cpu_initialized_mask))
+		return -1;
+
+	cpumask_set_cpu(cpu, cpu_callout_mask);
+	return 0;
 }
 
-int native_cpu_up(unsigned int cpu, struct task_struct *tidle)
+/*
+ * Bringup step three: Wait for the target AP to reach smp_callin().
+ * The AP is not waiting for us here so we don't need to parallelise
+ * this step. Not entirely clear why we care about this, since we just
+ * proceed directly to TSC synchronization which is the next sync
+ * point with the AP anyway.
+ */
+static void wait_cpu_callin(unsigned int cpu)
+{
+	while (!cpumask_test_cpu(cpu, cpu_callin_mask))
+		schedule();
+}
+
+/*
+ * Bringup step four: Synchronize the TSC and wait for the target AP
+ * to reach set_cpu_online() in start_secondary().
+ */
+static void wait_cpu_online(unsigned int cpu)
 {
-	int apicid = apic->cpu_present_to_apicid(cpu);
 	unsigned long flags;
+
+	/*
+	 * Check TSC synchronization with the AP (keep irqs disabled
+	 * while doing so):
+	 */
+	local_irq_save(flags);
+	check_tsc_sync_source(cpu);
+	local_irq_restore(flags);
+
+	/*
+	 * Wait for the AP to mark itself online, so the core caller
+	 * can drop sparse_irq_lock.
+	 */
+	while (!cpu_online(cpu))
+		schedule();
+}
+
+static int native_kick_ap(unsigned int cpu, struct task_struct *tidle)
+{
+	int apicid = apic->cpu_present_to_apicid(cpu);
 	int err;
 
 	lockdep_assert_irqs_enabled();
@@ -1142,25 +1189,33 @@ int native_cpu_up(unsigned int cpu, stru
 		return err;
 
 	err = do_boot_cpu(apicid, cpu, tidle);
-	if (err) {
+	if (err)
 		pr_err("do_boot_cpu failed(%d) to wakeup CPU#%u\n", err, cpu);
-		return err;
-	}
 
-	/*
-	 * Check TSC synchronization with the AP (keep irqs disabled
-	 * while doing so):
-	 */
-	local_irq_save(flags);
-	check_tsc_sync_source(cpu);
-	local_irq_restore(flags);
+	return err;
+}
 
-	while (!cpu_online(cpu)) {
-		cpu_relax();
-		touch_nmi_watchdog();
-	}
+int native_cpu_up(unsigned int cpu, struct task_struct *tidle)
+{
+	int ret;
 
-	return 0;
+	ret = native_kick_ap(cpu, tidle);
+	if (ret)
+		goto out;
+
+	ret = wait_cpu_initialized(cpu);
+	if (ret)
+		goto out;
+
+	wait_cpu_callin(cpu);
+	wait_cpu_online(cpu);
+
+out:
+	/* Cleanup possible dangling ends... */
+	if (x86_platform.legacy.warm_reset)
+		smpboot_restore_warm_reset_vector();
+
+	return ret;
 }
 
 /**

Re: [patch v3 08/36] x86/smpboot: Split up native_cpu_up() into separate phases and document them

[Peter Zijlstra]

On Mon, May 08, 2023 at 09:43:39PM +0200, Thomas Gleixner wrote:

> @@ -233,14 +237,31 @@ static void notrace start_secondary(void
>  	load_cr3(swapper_pg_dir);
>  	__flush_tlb_all();
>  #endif
> +	/*
> +	 * Sync point with wait_cpu_initialized(). Before proceeding through
> +	 * cpu_init(), the AP will call wait_for_master_cpu() which sets its
> +	 * own bit in cpu_initialized_mask and then waits for the BSP to set
> +	 * its bit in cpu_callout_mask to release it.
> +	 */
>  	cpu_init_secondary();
>  	rcu_cpu_starting(raw_smp_processor_id());
>  	x86_cpuinit.early_percpu_clock_init();
> +
> +	/*
> +	 * Sync point with wait_cpu_callin(). The AP doesn't wait here
> +	 * but just sets the bit to let the controlling CPU (BSP) know that
> +	 * it's got this far.
> +	 */
>  	smp_callin();
>  
> -	/* otherwise gcc will move up smp_processor_id before the cpu_init */
> +	/* Otherwise gcc will move up smp_processor_id() before cpu_init() */
>  	barrier();

Not to the detriment of this patch, but this barrier() and it's comment
seem weird vs smp_callin(). That function ends with an atomic bitop (it
has to, at the very least it must not be weaker than store-release) but
also has an explicit wmb() to order setup vs CPU_STARTING.

(arguably that should be a full fence *AND* get a comment)

There is no way the smp_processor_id() referred to in this comment can
land before cpu_init() even without the barrier().

> -	/* Check TSC synchronization with the control CPU: */
> +
> +	/*
> +	 * Check TSC synchronization with the control CPU, which will do
> +	 * its part of this from wait_cpu_online(), making it an implicit
> +	 * synchronization point.
> +	 */
>  	check_tsc_sync_target();
>  
>  	/*

Re: [patch v3 08/36] x86/smpboot: Split up native_cpu_up() into separate phases and document them

[Thomas Gleixner]

On Tue, May 09 2023 at 12:04, Peter Zijlstra wrote:
> On Mon, May 08, 2023 at 09:43:39PM +0200, Thomas Gleixner wrote:
>> +	/*
>> +	 * Sync point with wait_cpu_callin(). The AP doesn't wait here
>> +	 * but just sets the bit to let the controlling CPU (BSP) know that
>> +	 * it's got this far.
>> +	 */
>>  	smp_callin();
>>  
>> -	/* otherwise gcc will move up smp_processor_id before the cpu_init */
>> +	/* Otherwise gcc will move up smp_processor_id() before cpu_init() */
>>  	barrier();
>
> Not to the detriment of this patch, but this barrier() and it's comment
> seem weird vs smp_callin(). That function ends with an atomic bitop (it
> has to, at the very least it must not be weaker than store-release) but
> also has an explicit wmb() to order setup vs CPU_STARTING.
>
> (arguably that should be a full fence *AND* get a comment)
>
> There is no way the smp_processor_id() referred to in this comment can
> land before cpu_init() even without the barrier().

Right. Let me clean that up.

Thanks,

        tglx

Re: [patch v3 08/36] x86/smpboot: Split up native_cpu_up() into separate phases and document them

[Thomas Gleixner]

On Tue, May 09 2023 at 14:07, Thomas Gleixner wrote:
> On Tue, May 09 2023 at 12:04, Peter Zijlstra wrote:
>> On Mon, May 08, 2023 at 09:43:39PM +0200, Thomas Gleixner wrote:
>>> +	/*
>>> +	 * Sync point with wait_cpu_callin(). The AP doesn't wait here
>>> +	 * but just sets the bit to let the controlling CPU (BSP) know that
>>> +	 * it's got this far.
>>> +	 */
>>>  	smp_callin();
>>>  
>>> -	/* otherwise gcc will move up smp_processor_id before the cpu_init */
>>> +	/* Otherwise gcc will move up smp_processor_id() before cpu_init() */
>>>  	barrier();
>>
>> Not to the detriment of this patch, but this barrier() and it's comment
>> seem weird vs smp_callin(). That function ends with an atomic bitop (it
>> has to, at the very least it must not be weaker than store-release) but
>> also has an explicit wmb() to order setup vs CPU_STARTING.
>>
>> (arguably that should be a full fence *AND* get a comment)
>>
>> There is no way the smp_processor_id() referred to in this comment can
>> land before cpu_init() even without the barrier().
>
> Right. Let me clean that up.

So I went and tried to figure out where this comes from. It's from
d8f19f2cac70 ("[PATCH] x86-64 merge") in the history tree. One of those
well documented combo patches which change world and some more. The
context back then was:

	/*
	 * Dont put anything before smp_callin(), SMP
	 * booting is too fragile that we want to limit the
	 * things done here to the most necessary things.
	 */
	cpu_init();
	smp_callin();

	Dprintk("cpu %d: waiting for commence\n", smp_processor_id()); 

That still does not explain what the barrier is doing. I tried to
harvest mailing list archives, but did not find anything. The back then
list discuss@x86-64.org was never publicly archived... Boris gave me an
tarball, but this 'barrier()' add on was nowhere discussed in public.

As the barrier has no obvious value, I'm adding a patch upfront which
removes it.

Thanks,

        tglx

Re: [patch v3 08/36] x86/smpboot: Split up native_cpu_up() into separate phases and document them

[Thomas Gleixner]

On Tue, May 09 2023 at 12:04, Peter Zijlstra wrote:
> On Mon, May 08, 2023 at 09:43:39PM +0200, Thomas Gleixner wrote:
> Not to the detriment of this patch, but this barrier() and it's comment
> seem weird vs smp_callin(). That function ends with an atomic bitop (it
> has to, at the very least it must not be weaker than store-release) but
> also has an explicit wmb() to order setup vs CPU_STARTING.
>
> (arguably that should be a full fence *AND* get a comment)

TBH: I'm grasping for something 'arguable': What's the point of this
wmb() or even a mb()?

Most of the [w]mb()'s in smpboot.c except those in mwait_play_dead()
have a very distinct voodoo programming smell.

Thanks,

        tglx

Re: [patch v3 08/36] x86/smpboot: Split up native_cpu_up() into separate phases and document them

[Peter Zijlstra]

On Tue, May 09, 2023 at 10:11:05PM +0200, Thomas Gleixner wrote:
> On Tue, May 09 2023 at 12:04, Peter Zijlstra wrote:
> > On Mon, May 08, 2023 at 09:43:39PM +0200, Thomas Gleixner wrote:
> > Not to the detriment of this patch, but this barrier() and it's comment
> > seem weird vs smp_callin(). That function ends with an atomic bitop (it
> > has to, at the very least it must not be weaker than store-release) but
> > also has an explicit wmb() to order setup vs CPU_STARTING.
> >
> > (arguably that should be a full fence *AND* get a comment)
> 
> TBH: I'm grasping for something 'arguable': What's the point of this
> wmb() or even a mb()?
> 
> Most of the [w]mb()'s in smpboot.c except those in mwait_play_dead()
> have a very distinct voodoo programming smell.

Oh fully agreed, esp. without a comment these things are hugely suspect.
I could not immediately see purpose either.

My arguably argument was about IF it was needed at all, then it would
make more sense to me to also constrain loads. But I'd be more than
happy to see the whole thing go. But perhaps not in this series?

Re: [patch v3 08/36] x86/smpboot: Split up native_cpu_up() into separate phases and document them

[Peter Zijlstra]

Again, not really this patch, but since I had to look at this code ....

On Mon, May 08, 2023 at 09:43:39PM +0200, Thomas Gleixner wrote:
> @@ -1048,60 +1066,89 @@ static int do_boot_cpu(int apicid, int c

	/*
	 * AP might wait on cpu_callout_mask in cpu_init() with
	 * cpu_initialized_mask set if previous attempt to online
	 * it timed-out. Clear cpu_initialized_mask so that after
	 * INIT/SIPI it could start with a clean state.
	 */
	cpumask_clear_cpu(cpu, cpu_initialized_mask);
	smp_mb();

^^^ that barrier is weird too, cpumask_clear_cpu() is an atomic op and
implies much the same (this is x86 after all). If you want to be super
explicit about it write:

	smp_mb__after_atomic();

(which is a no-op) but then it still very much requires a comment as to
what exactly it orders against what.


	/*
	 * Wake up a CPU in difference cases:
	 * - Use a method from the APIC driver if one defined, with wakeup
	 *   straight to 64-bit mode preferred over wakeup to RM.
	 * Otherwise,
>  	 * - Use an INIT boot APIC message
>  	 */
>  	if (apic->wakeup_secondary_cpu_64)
> +		return apic->wakeup_secondary_cpu_64(apicid, start_ip);
>  	else if (apic->wakeup_secondary_cpu)
> +		return apic->wakeup_secondary_cpu(apicid, start_ip);
>  
> +	return wakeup_secondary_cpu_via_init(apicid, start_ip);
> +}

Re: [patch v3 08/36] x86/smpboot: Split up native_cpu_up() into separate phases and document them

[Thomas Gleixner]

On Tue, May 09 2023 at 12:19, Peter Zijlstra wrote:
> On Mon, May 08, 2023 at 09:43:39PM +0200, Thomas Gleixner wrote:
>> @@ -1048,60 +1066,89 @@ static int do_boot_cpu(int apicid, int c
>
> 	/*
> 	 * AP might wait on cpu_callout_mask in cpu_init() with
> 	 * cpu_initialized_mask set if previous attempt to online
> 	 * it timed-out. Clear cpu_initialized_mask so that after
> 	 * INIT/SIPI it could start with a clean state.
> 	 */
> 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
> 	smp_mb();
>
> ^^^ that barrier is weird too, cpumask_clear_cpu() is an atomic op and
> implies much the same (this is x86 after all). If you want to be super
> explicit about it write:
>
> 	smp_mb__after_atomic();
>
> (which is a no-op) but then it still very much requires a comment as to
> what exactly it orders against what.

As this is gone a few patches later, I just be lazy and leave it alone.

Re: [patch v3 08/36] x86/smpboot: Split up native_cpu_up() into separate phases and document them

[Thomas Gleixner]

On Tue, May 09 2023 at 12:19, Peter Zijlstra wrote:
> Again, not really this patch, but since I had to look at this code ....
>
> On Mon, May 08, 2023 at 09:43:39PM +0200, Thomas Gleixner wrote:
>> @@ -1048,60 +1066,89 @@ static int do_boot_cpu(int apicid, int c
>
> 	/*
> 	 * AP might wait on cpu_callout_mask in cpu_init() with
> 	 * cpu_initialized_mask set if previous attempt to online
> 	 * it timed-out. Clear cpu_initialized_mask so that after
> 	 * INIT/SIPI it could start with a clean state.
> 	 */
> 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
> 	smp_mb();
>
> ^^^ that barrier is weird too, cpumask_clear_cpu() is an atomic op and
> implies much the same (this is x86 after all). If you want to be super
> explicit about it write:
>
> 	smp_mb__after_atomic();
>
> (which is a no-op) but then it still very much requires a comment as to
> what exactly it orders against what.

Won't bother either as that mask is gone a few patches later.

Re: [patch v3 08/36] x86/smpboot: Split up native_cpu_up() into separate phases and document them

[Peter Zijlstra]

And since I'm commenting on existing things anyway, let me continue...

On Mon, May 08, 2023 at 09:43:39PM +0200, Thomas Gleixner wrote:

> +static int wait_cpu_cpumask(unsigned int cpu, const struct cpumask *mask)
> +{
> +	unsigned long timeout;
>  
> +	/*
> +	 * Wait up to 10s for the CPU to report in.
> +	 */
> +	timeout = jiffies + 10*HZ;
> +	while (time_before(jiffies, timeout)) {
> +		if (cpumask_test_cpu(cpu, mask))
> +			return 0;
> +
> +		schedule();
>  	}
> +	return -1;
> +}

> +/*
> + * Bringup step three: Wait for the target AP to reach smp_callin().
> + * The AP is not waiting for us here so we don't need to parallelise
> + * this step. Not entirely clear why we care about this, since we just
> + * proceed directly to TSC synchronization which is the next sync
> + * point with the AP anyway.
> + */
> +static void wait_cpu_callin(unsigned int cpu)
> +{
> +	while (!cpumask_test_cpu(cpu, cpu_callin_mask))
> +		schedule();
> +}
> +
> +/*
> + * Bringup step four: Synchronize the TSC and wait for the target AP
> + * to reach set_cpu_online() in start_secondary().
> + */
> +static void wait_cpu_online(unsigned int cpu)
>  {
>  	unsigned long flags;
> +
> +	/*
> +	 * Check TSC synchronization with the AP (keep irqs disabled
> +	 * while doing so):
> +	 */
> +	local_irq_save(flags);
> +	check_tsc_sync_source(cpu);
> +	local_irq_restore(flags);
> +
> +	/*
> +	 * Wait for the AP to mark itself online, so the core caller
> +	 * can drop sparse_irq_lock.
> +	 */
> +	while (!cpu_online(cpu))
> +		schedule();
> +}

These schedule() loops make me itch... this is basically Ye Olde yield()
loop with all it's known 'benefits'.

Now, I don't think it's horribly broken, we're explicitly waiting on
another CPU and can't have priority inversions, but yuck!

It could all be somewhat cleaned up with wait_var_event{_timeout}() and
wake_up_var(), but I'm really not sure that's worth it. But at least it
requires a comment to justify.

Re: [patch v3 08/36] x86/smpboot: Split up native_cpu_up() into separate phases and document them

[Thomas Gleixner]

On Tue, May 09 2023 at 12:31, Peter Zijlstra wrote:
> On Mon, May 08, 2023 at 09:43:39PM +0200, Thomas Gleixner wrote:
>> +	/*
>> +	 * Wait for the AP to mark itself online, so the core caller
>> +	 * can drop sparse_irq_lock.
>> +	 */
>> +	while (!cpu_online(cpu))
>> +		schedule();
>> +}
>
> These schedule() loops make me itch... this is basically Ye Olde yield()
> loop with all it's known 'benefits'.
>
> Now, I don't think it's horribly broken, we're explicitly waiting on
> another CPU and can't have priority inversions, but yuck!
>
> It could all be somewhat cleaned up with wait_var_event{_timeout}() and
> wake_up_var(), but I'm really not sure that's worth it. But at least it
> requires a comment to justify.

All of them are gone with the later patches and the control CPU will
just go straight to wait for the completion in the core code.

[patch v3 09/36] x86/smpboot: Get rid of cpu_init_secondary()

[Thomas Gleixner]

From: Thomas Gleixner <tglx@linutronix.de>

The synchronization of the AP with the control CPU is a SMP boot problem
and has nothing to do with cpu_init().

Open code cpu_init_secondary() in start_secondary() and move
wait_for_master_cpu() into the SMP boot code.

No functional change.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Michael Kelley <mikelley@microsoft.com>


---
 arch/x86/include/asm/processor.h |    1 -
 arch/x86/kernel/cpu/common.c     |   27 ---------------------------
 arch/x86/kernel/smpboot.c        |   24 +++++++++++++++++++-----
 3 files changed, 19 insertions(+), 33 deletions(-)
---

--- a/arch/x86/include/asm/processor.h
+++ b/arch/x86/include/asm/processor.h
@@ -551,7 +551,6 @@ extern void switch_gdt_and_percpu_base(i
 extern void load_direct_gdt(int);
 extern void load_fixmap_gdt(int);
 extern void cpu_init(void);
-extern void cpu_init_secondary(void);
 extern void cpu_init_exception_handling(void);
 extern void cr4_init(void);
 
--- a/arch/x86/kernel/cpu/common.c
+++ b/arch/x86/kernel/cpu/common.c
@@ -2123,19 +2123,6 @@ static void dbg_restore_debug_regs(void)
 #define dbg_restore_debug_regs()
 #endif /* ! CONFIG_KGDB */
 
-static void wait_for_master_cpu(int cpu)
-{
-#ifdef CONFIG_SMP
-	/*
-	 * wait for ACK from master CPU before continuing
-	 * with AP initialization
-	 */
-	WARN_ON(cpumask_test_and_set_cpu(cpu, cpu_initialized_mask));
-	while (!cpumask_test_cpu(cpu, cpu_callout_mask))
-		cpu_relax();
-#endif
-}
-
 static inline void setup_getcpu(int cpu)
 {
 	unsigned long cpudata = vdso_encode_cpunode(cpu, early_cpu_to_node(cpu));
@@ -2239,8 +2226,6 @@ void cpu_init(void)
 	struct task_struct *cur = current;
 	int cpu = raw_smp_processor_id();
 
-	wait_for_master_cpu(cpu);
-
 	ucode_cpu_init(cpu);
 
 #ifdef CONFIG_NUMA
@@ -2293,18 +2278,6 @@ void cpu_init(void)
 	load_fixmap_gdt(cpu);
 }
 
-#ifdef CONFIG_SMP
-void cpu_init_secondary(void)
-{
-	/*
-	 * Relies on the BP having set-up the IDT tables, which are loaded
-	 * on this CPU in cpu_init_exception_handling().
-	 */
-	cpu_init_exception_handling();
-	cpu_init();
-}
-#endif
-
 #ifdef CONFIG_MICROCODE_LATE_LOADING
 /**
  * store_cpu_caps() - Store a snapshot of CPU capabilities
--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -220,6 +220,17 @@ static void ap_calibrate_delay(void)
 	cpu_data(smp_processor_id()).loops_per_jiffy = loops_per_jiffy;
 }
 
+static void wait_for_master_cpu(int cpu)
+{
+	/*
+	 * Wait for release by control CPU before continuing with AP
+	 * initialization.
+	 */
+	WARN_ON(cpumask_test_and_set_cpu(cpu, cpu_initialized_mask));
+	while (!cpumask_test_cpu(cpu, cpu_callout_mask))
+		cpu_relax();
+}
+
 /*
  * Activate a secondary processor.
  */
@@ -237,13 +248,16 @@ static void notrace start_secondary(void
 	load_cr3(swapper_pg_dir);
 	__flush_tlb_all();
 #endif
+	cpu_init_exception_handling();
+
 	/*
-	 * Sync point with wait_cpu_initialized(). Before proceeding through
-	 * cpu_init(), the AP will call wait_for_master_cpu() which sets its
-	 * own bit in cpu_initialized_mask and then waits for the BSP to set
-	 * its bit in cpu_callout_mask to release it.
+	 * Sync point with wait_cpu_initialized(). Sets AP in
+	 * cpu_initialized_mask and then waits for the control CPU
+	 * to release it.
 	 */
-	cpu_init_secondary();
+	wait_for_master_cpu(raw_smp_processor_id());
+
+	cpu_init();
 	rcu_cpu_starting(raw_smp_processor_id());
 	x86_cpuinit.early_percpu_clock_init();
 

[patch v3 10/36] [patch V2 10/38] x86/cpu/cacheinfo: Remove cpu_callout_mask dependency

[Thomas Gleixner]

From: Thomas Gleixner <tglx@linutronix.de>

cpu_callout_mask is used for the stop machine based MTRR/PAT init.

In preparation of moving the BP/AP synchronization to the core hotplug
code, use a private CPU mask for cacheinfo and manage it in the
starting/dying hotplug state.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Michael Kelley <mikelley@microsoft.com>


---
 arch/x86/kernel/cpu/cacheinfo.c |   21 +++++++++++++++++----
 1 file changed, 17 insertions(+), 4 deletions(-)
---

--- a/arch/x86/kernel/cpu/cacheinfo.c
+++ b/arch/x86/kernel/cpu/cacheinfo.c
@@ -39,6 +39,8 @@ DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t
 /* Shared L2 cache maps */
 DEFINE_PER_CPU_READ_MOSTLY(cpumask_var_t, cpu_l2c_shared_map);
 
+static cpumask_var_t cpu_cacheinfo_mask;
+
 /* Kernel controls MTRR and/or PAT MSRs. */
 unsigned int memory_caching_control __ro_after_init;
 
@@ -1172,8 +1174,10 @@ void cache_bp_restore(void)
 		cache_cpu_init();
 }
 
-static int cache_ap_init(unsigned int cpu)
+static int cache_ap_online(unsigned int cpu)
 {
+	cpumask_set_cpu(cpu, cpu_cacheinfo_mask);
+
 	if (!memory_caching_control || get_cache_aps_delayed_init())
 		return 0;
 
@@ -1191,11 +1195,17 @@ static int cache_ap_init(unsigned int cp
 	 *      lock to prevent MTRR entry changes
 	 */
 	stop_machine_from_inactive_cpu(cache_rendezvous_handler, NULL,
-				       cpu_callout_mask);
+				       cpu_cacheinfo_mask);
 
 	return 0;
 }
 
+static int cache_ap_offline(unsigned int cpu)
+{
+	cpumask_clear_cpu(cpu, cpu_cacheinfo_mask);
+	return 0;
+}
+
 /*
  * Delayed cache initialization for all AP's
  */
@@ -1210,9 +1220,12 @@ void cache_aps_init(void)
 
 static int __init cache_ap_register(void)
 {
+	zalloc_cpumask_var(&cpu_cacheinfo_mask, GFP_KERNEL);
+	cpumask_set_cpu(smp_processor_id(), cpu_cacheinfo_mask);
+
 	cpuhp_setup_state_nocalls(CPUHP_AP_CACHECTRL_STARTING,
 				  "x86/cachectrl:starting",
-				  cache_ap_init, NULL);
+				  cache_ap_online, cache_ap_offline);
 	return 0;
 }
-core_initcall(cache_ap_register);
+early_initcall(cache_ap_register);

[patch v3 11/36] [patch V2 11/38] x86/smpboot: Move synchronization masks to SMP boot code

[Thomas Gleixner]

From: Thomas Gleixner <tglx@linutronix.de>

The usage is in smpboot.c and not in the CPU initialization code.

The XEN_PV usage of cpu_callout_mask is obsolete as cpu_init() not longer
waits and cacheinfo has its own CPU mask now, so cpu_callout_mask can be
made static too.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Michael Kelley <mikelley@microsoft.com>


---
 arch/x86/include/asm/cpumask.h |    5 -----
 arch/x86/kernel/cpu/common.c   |   17 -----------------
 arch/x86/kernel/smpboot.c      |   16 ++++++++++++++++
 arch/x86/xen/smp_pv.c          |    3 ---
 4 files changed, 16 insertions(+), 25 deletions(-)
---

--- a/arch/x86/include/asm/cpumask.h
+++ b/arch/x86/include/asm/cpumask.h
@@ -4,11 +4,6 @@
 #ifndef __ASSEMBLY__
 #include <linux/cpumask.h>
 
-extern cpumask_var_t cpu_callin_mask;
-extern cpumask_var_t cpu_callout_mask;
-extern cpumask_var_t cpu_initialized_mask;
-extern cpumask_var_t cpu_sibling_setup_mask;
-
 extern void setup_cpu_local_masks(void);
 
 /*
--- a/arch/x86/kernel/cpu/common.c
+++ b/arch/x86/kernel/cpu/common.c
@@ -67,14 +67,6 @@
 
 u32 elf_hwcap2 __read_mostly;
 
-/* all of these masks are initialized in setup_cpu_local_masks() */
-cpumask_var_t cpu_initialized_mask;
-cpumask_var_t cpu_callout_mask;
-cpumask_var_t cpu_callin_mask;
-
-/* representing cpus for which sibling maps can be computed */
-cpumask_var_t cpu_sibling_setup_mask;
-
 /* Number of siblings per CPU package */
 int smp_num_siblings = 1;
 EXPORT_SYMBOL(smp_num_siblings);
@@ -169,15 +161,6 @@ static void ppin_init(struct cpuinfo_x86
 	clear_cpu_cap(c, info->feature);
 }
 
-/* correctly size the local cpu masks */
-void __init setup_cpu_local_masks(void)
-{
-	alloc_bootmem_cpumask_var(&cpu_initialized_mask);
-	alloc_bootmem_cpumask_var(&cpu_callin_mask);
-	alloc_bootmem_cpumask_var(&cpu_callout_mask);
-	alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask);
-}
-
 static void default_init(struct cpuinfo_x86 *c)
 {
 #ifdef CONFIG_X86_64
--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -101,6 +101,13 @@ EXPORT_PER_CPU_SYMBOL(cpu_die_map);
 DEFINE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info);
 EXPORT_PER_CPU_SYMBOL(cpu_info);
 
+/* All of these masks are initialized in setup_cpu_local_masks() */
+static cpumask_var_t cpu_initialized_mask;
+static cpumask_var_t cpu_callout_mask;
+static cpumask_var_t cpu_callin_mask;
+/* Representing CPUs for which sibling maps can be computed */
+static cpumask_var_t cpu_sibling_setup_mask;
+
 /* Logical package management. We might want to allocate that dynamically */
 unsigned int __max_logical_packages __read_mostly;
 EXPORT_SYMBOL(__max_logical_packages);
@@ -1548,6 +1555,15 @@ early_param("possible_cpus", _setup_poss
 		set_cpu_possible(i, true);
 }
 
+/* correctly size the local cpu masks */
+void __init setup_cpu_local_masks(void)
+{
+	alloc_bootmem_cpumask_var(&cpu_initialized_mask);
+	alloc_bootmem_cpumask_var(&cpu_callin_mask);
+	alloc_bootmem_cpumask_var(&cpu_callout_mask);
+	alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask);
+}
+
 #ifdef CONFIG_HOTPLUG_CPU
 
 /* Recompute SMT state for all CPUs on offline */
--- a/arch/x86/xen/smp_pv.c
+++ b/arch/x86/xen/smp_pv.c
@@ -254,15 +254,12 @@ cpu_initialize_context(unsigned int cpu,
 	struct desc_struct *gdt;
 	unsigned long gdt_mfn;
 
-	/* used to tell cpu_init() that it can proceed with initialization */
-	cpumask_set_cpu(cpu, cpu_callout_mask);
 	if (cpumask_test_and_set_cpu(cpu, xen_cpu_initialized_map))
 		return 0;
 
 	ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL);
 	if (ctxt == NULL) {
 		cpumask_clear_cpu(cpu, xen_cpu_initialized_map);
-		cpumask_clear_cpu(cpu, cpu_callout_mask);
 		return -ENOMEM;
 	}
 

[patch v3 12/36] [patch V2 12/38] x86/smpboot: Make TSC synchronization function call based

[Thomas Gleixner]

From: Thomas Gleixner <tglx@linutronix.de>

Spin-waiting on the control CPU until the AP reaches the TSC
synchronization is just a waste especially in the case that there is no
synchronization required.

As the synchronization has to run with interrupts disabled the control CPU
part can just be done from a SMP function call. The upcoming AP issues that
call async only in the case that synchronization is required.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Michael Kelley <mikelley@microsoft.com>


---
 arch/x86/include/asm/tsc.h |    2 --
 arch/x86/kernel/smpboot.c  |   20 +++-----------------
 arch/x86/kernel/tsc_sync.c |   36 +++++++++++-------------------------
 3 files changed, 14 insertions(+), 44 deletions(-)
---

--- a/arch/x86/include/asm/tsc.h
+++ b/arch/x86/include/asm/tsc.h
@@ -55,12 +55,10 @@ extern bool tsc_async_resets;
 #ifdef CONFIG_X86_TSC
 extern bool tsc_store_and_check_tsc_adjust(bool bootcpu);
 extern void tsc_verify_tsc_adjust(bool resume);
-extern void check_tsc_sync_source(int cpu);
 extern void check_tsc_sync_target(void);
 #else
 static inline bool tsc_store_and_check_tsc_adjust(bool bootcpu) { return false; }
 static inline void tsc_verify_tsc_adjust(bool resume) { }
-static inline void check_tsc_sync_source(int cpu) { }
 static inline void check_tsc_sync_target(void) { }
 #endif
 
--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -278,11 +278,7 @@ static void notrace start_secondary(void
 	/* Otherwise gcc will move up smp_processor_id() before cpu_init() */
 	barrier();
 
-	/*
-	 * Check TSC synchronization with the control CPU, which will do
-	 * its part of this from wait_cpu_online(), making it an implicit
-	 * synchronization point.
-	 */
+	/* Check TSC synchronization with the control CPU. */
 	check_tsc_sync_target();
 
 	/*
@@ -1144,21 +1140,11 @@ static void wait_cpu_callin(unsigned int
 }
 
 /*
- * Bringup step four: Synchronize the TSC and wait for the target AP
- * to reach set_cpu_online() in start_secondary().
+ * Bringup step four: Wait for the target AP to reach set_cpu_online() in
+ * start_secondary().
  */
 static void wait_cpu_online(unsigned int cpu)
 {
-	unsigned long flags;
-
-	/*
-	 * Check TSC synchronization with the AP (keep irqs disabled
-	 * while doing so):
-	 */
-	local_irq_save(flags);
-	check_tsc_sync_source(cpu);
-	local_irq_restore(flags);
-
 	/*
 	 * Wait for the AP to mark itself online, so the core caller
 	 * can drop sparse_irq_lock.
--- a/arch/x86/kernel/tsc_sync.c
+++ b/arch/x86/kernel/tsc_sync.c
@@ -245,7 +245,6 @@ bool tsc_store_and_check_tsc_adjust(bool
  */
 static atomic_t start_count;
 static atomic_t stop_count;
-static atomic_t skip_test;
 static atomic_t test_runs;
 
 /*
@@ -344,21 +343,14 @@ static inline unsigned int loop_timeout(
 }
 
 /*
- * Source CPU calls into this - it waits for the freshly booted
- * target CPU to arrive and then starts the measurement:
+ * The freshly booted CPU initiates this via an async SMP function call.
  */
-void check_tsc_sync_source(int cpu)
+static void check_tsc_sync_source(void *__cpu)
 {
+	unsigned int cpu = (unsigned long)__cpu;
 	int cpus = 2;
 
 	/*
-	 * No need to check if we already know that the TSC is not
-	 * synchronized or if we have no TSC.
-	 */
-	if (unsynchronized_tsc())
-		return;
-
-	/*
 	 * Set the maximum number of test runs to
 	 *  1 if the CPU does not provide the TSC_ADJUST MSR
 	 *  3 if the MSR is available, so the target can try to adjust
@@ -368,16 +360,9 @@ void check_tsc_sync_source(int cpu)
 	else
 		atomic_set(&test_runs, 3);
 retry:
-	/*
-	 * Wait for the target to start or to skip the test:
-	 */
-	while (atomic_read(&start_count) != cpus - 1) {
-		if (atomic_read(&skip_test) > 0) {
-			atomic_set(&skip_test, 0);
-			return;
-		}
+	/* Wait for the target to start. */
+	while (atomic_read(&start_count) != cpus - 1)
 		cpu_relax();
-	}
 
 	/*
 	 * Trigger the target to continue into the measurement too:
@@ -397,14 +382,14 @@ void check_tsc_sync_source(int cpu)
 	if (!nr_warps) {
 		atomic_set(&test_runs, 0);
 
-		pr_debug("TSC synchronization [CPU#%d -> CPU#%d]: passed\n",
+		pr_debug("TSC synchronization [CPU#%d -> CPU#%u]: passed\n",
 			smp_processor_id(), cpu);
 
 	} else if (atomic_dec_and_test(&test_runs) || random_warps) {
 		/* Force it to 0 if random warps brought us here */
 		atomic_set(&test_runs, 0);
 
-		pr_warn("TSC synchronization [CPU#%d -> CPU#%d]:\n",
+		pr_warn("TSC synchronization [CPU#%d -> CPU#%u]:\n",
 			smp_processor_id(), cpu);
 		pr_warn("Measured %Ld cycles TSC warp between CPUs, "
 			"turning off TSC clock.\n", max_warp);
@@ -457,11 +442,12 @@ void check_tsc_sync_target(void)
 	 * SoCs the TSC is frequency synchronized, but still the TSC ADJUST
 	 * register might have been wreckaged by the BIOS..
 	 */
-	if (tsc_store_and_check_tsc_adjust(false) || tsc_clocksource_reliable) {
-		atomic_inc(&skip_test);
+	if (tsc_store_and_check_tsc_adjust(false) || tsc_clocksource_reliable)
 		return;
-	}
 
+	/* Kick the control CPU into the TSC synchronization function */
+	smp_call_function_single(cpumask_first(cpu_online_mask), check_tsc_sync_source,
+				 (unsigned long *)(unsigned long)cpu, 0);
 retry:
 	/*
 	 * Register this CPU's participation and wait for the

[patch v3 13/36] x86/smpboot: Remove cpu_callin_mask

[Thomas Gleixner]

From: Thomas Gleixner <tglx@linutronix.de>

Now that TSC synchronization is SMP function call based there is no reason
to wait for the AP to be set in smp_callin_mask. The control CPU waits for
the AP to set itself in the online mask anyway.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Michael Kelley <mikelley@microsoft.com>


---
 arch/x86/kernel/smpboot.c |   61 +++++++---------------------------------------
 1 file changed, 10 insertions(+), 51 deletions(-)
---

--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -104,7 +104,6 @@ EXPORT_PER_CPU_SYMBOL(cpu_info);
 /* All of these masks are initialized in setup_cpu_local_masks() */
 static cpumask_var_t cpu_initialized_mask;
 static cpumask_var_t cpu_callout_mask;
-static cpumask_var_t cpu_callin_mask;
 /* Representing CPUs for which sibling maps can be computed */
 static cpumask_var_t cpu_sibling_setup_mask;
 
@@ -167,21 +166,16 @@ static inline void smpboot_restore_warm_
  */
 static void smp_callin(void)
 {
-	int cpuid;
+	int cpuid = smp_processor_id();
 
 	/*
 	 * If waken up by an INIT in an 82489DX configuration
-	 * cpu_callout_mask guarantees we don't get here before
-	 * an INIT_deassert IPI reaches our local APIC, so it is
-	 * now safe to touch our local APIC.
-	 */
-	cpuid = smp_processor_id();
-
-	/*
-	 * the boot CPU has finished the init stage and is spinning
-	 * on callin_map until we finish. We are free to set up this
-	 * CPU, first the APIC. (this is probably redundant on most
-	 * boards)
+	 * cpu_callout_mask guarantees we don't get here before an
+	 * INIT_deassert IPI reaches our local APIC, so it is now safe to
+	 * touch our local APIC.
+	 *
+	 * Set up this CPU, first the APIC, which is probably redundant on
+	 * most boards.
 	 */
 	apic_ap_setup();
 
@@ -192,7 +186,7 @@ static void smp_callin(void)
 	 * The topology information must be up to date before
 	 * calibrate_delay() and notify_cpu_starting().
 	 */
-	set_cpu_sibling_map(raw_smp_processor_id());
+	set_cpu_sibling_map(cpuid);
 
 	ap_init_aperfmperf();
 
@@ -205,11 +199,6 @@ static void smp_callin(void)
 	 * state (CPUHP_ONLINE in the case of serial bringup).
 	 */
 	notify_cpu_starting(cpuid);
-
-	/*
-	 * Allow the master to continue.
-	 */
-	cpumask_set_cpu(cpuid, cpu_callin_mask);
 }
 
 static void ap_calibrate_delay(void)
@@ -268,11 +257,6 @@ static void notrace start_secondary(void
 	rcu_cpu_starting(raw_smp_processor_id());
 	x86_cpuinit.early_percpu_clock_init();
 
-	/*
-	 * Sync point with wait_cpu_callin(). The AP doesn't wait here
-	 * but just sets the bit to let the controlling CPU (BSP) know that
-	 * it's got this far.
-	 */
 	smp_callin();
 
 	/* Otherwise gcc will move up smp_processor_id() before cpu_init() */
@@ -1112,7 +1096,7 @@ static int wait_cpu_cpumask(unsigned int
  * and thus wait_for_master_cpu(), then set cpu_callout_mask to allow it
  * to proceed.  The AP will then proceed past setting its 'callin' bit
  * and end up waiting in check_tsc_sync_target() until we reach
- * do_wait_cpu_online() to tend to it.
+ * wait_cpu_online() to tend to it.
  */
 static int wait_cpu_initialized(unsigned int cpu)
 {
@@ -1127,20 +1111,7 @@ static int wait_cpu_initialized(unsigned
 }
 
 /*
- * Bringup step three: Wait for the target AP to reach smp_callin().
- * The AP is not waiting for us here so we don't need to parallelise
- * this step. Not entirely clear why we care about this, since we just
- * proceed directly to TSC synchronization which is the next sync
- * point with the AP anyway.
- */
-static void wait_cpu_callin(unsigned int cpu)
-{
-	while (!cpumask_test_cpu(cpu, cpu_callin_mask))
-		schedule();
-}
-
-/*
- * Bringup step four: Wait for the target AP to reach set_cpu_online() in
+ * Bringup step three: Wait for the target AP to reach set_cpu_online() in
  * start_secondary().
  */
 static void wait_cpu_online(unsigned int cpu)
@@ -1170,14 +1141,6 @@ static int native_kick_ap(unsigned int c
 	}
 
 	/*
-	 * Already booted CPU?
-	 */
-	if (cpumask_test_cpu(cpu, cpu_callin_mask)) {
-		pr_debug("do_boot_cpu %d Already started\n", cpu);
-		return -ENOSYS;
-	}
-
-	/*
 	 * Save current MTRR state in case it was changed since early boot
 	 * (e.g. by the ACPI SMI) to initialize new CPUs with MTRRs in sync:
 	 */
@@ -1214,7 +1177,6 @@ int native_cpu_up(unsigned int cpu, stru
 	if (ret)
 		goto out;
 
-	wait_cpu_callin(cpu);
 	wait_cpu_online(cpu);
 
 out:
@@ -1330,7 +1292,6 @@ void __init smp_prepare_cpus_common(void
 	 * Setup boot CPU information
 	 */
 	smp_store_boot_cpu_info(); /* Final full version of the data */
-	cpumask_copy(cpu_callin_mask, cpumask_of(0));
 	mb();
 
 	for_each_possible_cpu(i) {
@@ -1545,7 +1506,6 @@ early_param("possible_cpus", _setup_poss
 void __init setup_cpu_local_masks(void)
 {
 	alloc_bootmem_cpumask_var(&cpu_initialized_mask);
-	alloc_bootmem_cpumask_var(&cpu_callin_mask);
 	alloc_bootmem_cpumask_var(&cpu_callout_mask);
 	alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask);
 }
@@ -1609,7 +1569,6 @@ static void remove_cpu_from_maps(int cpu
 {
 	set_cpu_online(cpu, false);
 	cpumask_clear_cpu(cpu, cpu_callout_mask);
-	cpumask_clear_cpu(cpu, cpu_callin_mask);
 	/* was set by cpu_init() */
 	cpumask_clear_cpu(cpu, cpu_initialized_mask);
 	numa_remove_cpu(cpu);

Re: [patch v3 13/36] x86/smpboot: Remove cpu_callin_mask

[Peter Zijlstra]

On Mon, May 08, 2023 at 09:43:47PM +0200, Thomas Gleixner wrote:

> --- a/arch/x86/kernel/smpboot.c
> +++ b/arch/x86/kernel/smpboot.c

> @@ -167,21 +166,16 @@ static inline void smpboot_restore_warm_
>   */
>  static void smp_callin(void)
>  {
> -	int cpuid;
> +	int cpuid = smp_processor_id();
>  
>  	/*
>  	 * If waken up by an INIT in an 82489DX configuration
> -	 * cpu_callout_mask guarantees we don't get here before
> -	 * an INIT_deassert IPI reaches our local APIC, so it is
> -	 * now safe to touch our local APIC.
> -	 */
> -	cpuid = smp_processor_id();
> -
> -	/*
> -	 * the boot CPU has finished the init stage and is spinning
> -	 * on callin_map until we finish. We are free to set up this
> -	 * CPU, first the APIC. (this is probably redundant on most
> -	 * boards)
> +	 * cpu_callout_mask guarantees we don't get here before an
> +	 * INIT_deassert IPI reaches our local APIC, so it is now safe to
> +	 * touch our local APIC.
> +	 *
> +	 * Set up this CPU, first the APIC, which is probably redundant on
> +	 * most boards.
>  	 */
>  	apic_ap_setup();
>  
> @@ -192,7 +186,7 @@ static void smp_callin(void)
>  	 * The topology information must be up to date before
>  	 * calibrate_delay() and notify_cpu_starting().
>  	 */
> -	set_cpu_sibling_map(raw_smp_processor_id());
> +	set_cpu_sibling_map(cpuid);
>  
>  	ap_init_aperfmperf();
>  
> @@ -205,11 +199,6 @@ static void smp_callin(void)
>  	 * state (CPUHP_ONLINE in the case of serial bringup).
>  	 */
>  	notify_cpu_starting(cpuid);
> -
> -	/*
> -	 * Allow the master to continue.
> -	 */
> -	cpumask_set_cpu(cpuid, cpu_callin_mask);
>  }
>  
>  static void ap_calibrate_delay(void)
> @@ -268,11 +257,6 @@ static void notrace start_secondary(void
>  	rcu_cpu_starting(raw_smp_processor_id());
>  	x86_cpuinit.early_percpu_clock_init();
>  
> -	/*
> -	 * Sync point with wait_cpu_callin(). The AP doesn't wait here
> -	 * but just sets the bit to let the controlling CPU (BSP) know that
> -	 * it's got this far.
> -	 */
>  	smp_callin();
>  
>  	/* Otherwise gcc will move up smp_processor_id() before cpu_init() */

Good riddance to that mask; however is smp_callin() still an appropriate
name for that function?

Would smp_starting() -- seeing how this kicks of CPU_STARTING not be a
better name?

Re: [patch v3 13/36] x86/smpboot: Remove cpu_callin_mask

[Thomas Gleixner]

On Tue, May 09 2023 at 12:49, Peter Zijlstra wrote:
> On Mon, May 08, 2023 at 09:43:47PM +0200, Thomas Gleixner wrote:
>> -	/*
>> -	 * Sync point with wait_cpu_callin(). The AP doesn't wait here
>> -	 * but just sets the bit to let the controlling CPU (BSP) know that
>> -	 * it's got this far.
>> -	 */
>>  	smp_callin();
>>  
>>  	/* Otherwise gcc will move up smp_processor_id() before cpu_init() */
>
> Good riddance to that mask; however is smp_callin() still an appropriate
> name for that function?
>
> Would smp_starting() -- seeing how this kicks of CPU_STARTING not be a
> better name?

Something like that, yes.

[patch v3 14/36] [patch V2 14/38] cpu/hotplug: Rework sparse_irq locking in bringup_cpu()

[Thomas Gleixner]

From: Thomas Gleixner <tglx@linutronix.de>

There is no harm to hold sparse_irq lock until the upcoming CPU completes
in cpuhp_online_idle(). This allows to remove cpu_online() synchronization
from architecture code.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Michael Kelley <mikelley@microsoft.com>


---
 kernel/cpu.c |   28 +++++++++++++++++++---------
 1 file changed, 19 insertions(+), 9 deletions(-)
---

--- a/kernel/cpu.c
+++ b/kernel/cpu.c
@@ -558,7 +558,7 @@ static int cpuhp_kick_ap(int cpu, struct
 	return ret;
 }
 
-static int bringup_wait_for_ap(unsigned int cpu)
+static int bringup_wait_for_ap_online(unsigned int cpu)
 {
 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
 
@@ -579,15 +579,12 @@ static int bringup_wait_for_ap(unsigned
 	 */
 	if (!cpu_smt_allowed(cpu))
 		return -ECANCELED;
-
-	if (st->target <= CPUHP_AP_ONLINE_IDLE)
-		return 0;
-
-	return cpuhp_kick_ap(cpu, st, st->target);
+	return 0;
 }
 
 static int bringup_cpu(unsigned int cpu)
 {
+	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
 	struct task_struct *idle = idle_thread_get(cpu);
 	int ret;
 
@@ -606,10 +603,23 @@ static int bringup_cpu(unsigned int cpu)
 
 	/* Arch-specific enabling code. */
 	ret = __cpu_up(cpu, idle);
-	irq_unlock_sparse();
 	if (ret)
-		return ret;
-	return bringup_wait_for_ap(cpu);
+		goto out_unlock;
+
+	ret = bringup_wait_for_ap_online(cpu);
+	if (ret)
+		goto out_unlock;
+
+	irq_unlock_sparse();
+
+	if (st->target <= CPUHP_AP_ONLINE_IDLE)
+		return 0;
+
+	return cpuhp_kick_ap(cpu, st, st->target);
+
+out_unlock:
+	irq_unlock_sparse();
+	return ret;
 }
 
 static int finish_cpu(unsigned int cpu)

Re: [patch v3 14/36] [patch V2 14/38] cpu/hotplug: Rework sparse_irq locking in bringup_cpu()

[Peter Zijlstra]

On Mon, May 08, 2023 at 09:43:49PM +0200, Thomas Gleixner wrote:
> From: Thomas Gleixner <tglx@linutronix.de>
> 
> There is no harm to hold sparse_irq lock until the upcoming CPU completes
> in cpuhp_online_idle(). This allows to remove cpu_online() synchronization
> from architecture code.

Uuuuuhh.. damn. Can you at the very least ammend the comment near
irq_lock_sparse() to mention these extra duties?

Re: [patch v3 14/36] [patch V2 14/38] cpu/hotplug: Rework sparse_irq locking in bringup_cpu()

[Thomas Gleixner]

On Tue, May 09 2023 at 13:02, Peter Zijlstra wrote:
> On Mon, May 08, 2023 at 09:43:49PM +0200, Thomas Gleixner wrote:
>> From: Thomas Gleixner <tglx@linutronix.de>
>> 
>> There is no harm to hold sparse_irq lock until the upcoming CPU completes
>> in cpuhp_online_idle(). This allows to remove cpu_online() synchronization
>> from architecture code.
>
> Uuuuuhh.. damn. Can you at the very least ammend the comment near
> irq_lock_sparse() to mention these extra duties?

Will do.

[patch v3 15/36] x86/smpboot: Remove wait for cpu_online()

[Thomas Gleixner]

From: Thomas Gleixner <tglx@linutronix.de>

Now that the core code drops sparse_irq_lock after the idle thread
synchronized, it's pointless to wait for the AP to mark itself online.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Michael Kelley <mikelley@microsoft.com>


---
 arch/x86/kernel/smpboot.c |   26 ++------------------------
 1 file changed, 2 insertions(+), 24 deletions(-)
---

--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -281,7 +281,6 @@ static void notrace start_secondary(void
 	 * half valid vector space.
 	 */
 	lock_vector_lock();
-	/* Sync point with do_wait_cpu_online() */
 	set_cpu_online(smp_processor_id(), true);
 	lapic_online();
 	unlock_vector_lock();
@@ -1110,20 +1109,6 @@ static int wait_cpu_initialized(unsigned
 	return 0;
 }
 
-/*
- * Bringup step three: Wait for the target AP to reach set_cpu_online() in
- * start_secondary().
- */
-static void wait_cpu_online(unsigned int cpu)
-{
-	/*
-	 * Wait for the AP to mark itself online, so the core caller
-	 * can drop sparse_irq_lock.
-	 */
-	while (!cpu_online(cpu))
-		schedule();
-}
-
 static int native_kick_ap(unsigned int cpu, struct task_struct *tidle)
 {
 	int apicid = apic->cpu_present_to_apicid(cpu);
@@ -1170,16 +1155,9 @@ int native_cpu_up(unsigned int cpu, stru
 	int ret;
 
 	ret = native_kick_ap(cpu, tidle);
-	if (ret)
-		goto out;
-
-	ret = wait_cpu_initialized(cpu);
-	if (ret)
-		goto out;
-
-	wait_cpu_online(cpu);
+	if (!ret)
+		ret = wait_cpu_initialized(cpu);
 
-out:
 	/* Cleanup possible dangling ends... */
 	if (x86_platform.legacy.warm_reset)
 		smpboot_restore_warm_reset_vector();

[patch v3 16/36] x86/xen/smp_pv: Remove wait for CPU online

[Thomas Gleixner]

From: Thomas Gleixner <tglx@linutronix.de>

Now that the core code drops sparse_irq_lock after the idle thread
synchronized, it's pointless to wait for the AP to mark itself online.

Whether the control CPU runs in a wait loop or sleeps in the core code
waiting for the online operation to complete makes no difference.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Michael Kelley <mikelley@microsoft.com>


---
 arch/x86/xen/smp_pv.c |   10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)
---

--- a/arch/x86/xen/smp_pv.c
+++ b/arch/x86/xen/smp_pv.c
@@ -340,11 +340,11 @@ static int xen_pv_cpu_up(unsigned int cp
 
 	xen_pmu_init(cpu);
 
-	rc = HYPERVISOR_vcpu_op(VCPUOP_up, xen_vcpu_nr(cpu), NULL);
-	BUG_ON(rc);
-
-	while (cpu_report_state(cpu) != CPU_ONLINE)
-		HYPERVISOR_sched_op(SCHEDOP_yield, NULL);
+	/*
+	 * Why is this a BUG? If the hypercall fails then everything can be
+	 * rolled back, no?
+	 */
+	BUG_ON(HYPERVISOR_vcpu_op(VCPUOP_up, xen_vcpu_nr(cpu), NULL));
 
 	return 0;
 }

[patch v3 17/36] x86/xen/hvm: Get rid of DEAD_FROZEN handling

[Thomas Gleixner]

From: Thomas Gleixner <tglx@linutronix.de>

No point in this conditional voodoo. Un-initializing the lock mechanism is
safe to be called unconditionally even if it was already invoked when the
CPU died.

Remove the invocation of xen_smp_intr_free() as that has been already
cleaned up in xen_cpu_dead_hvm().

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Michael Kelley <mikelley@microsoft.com>


---
 arch/x86/xen/enlighten_hvm.c |   11 +++++------
 1 file changed, 5 insertions(+), 6 deletions(-)
---

--- a/arch/x86/xen/enlighten_hvm.c
+++ b/arch/x86/xen/enlighten_hvm.c
@@ -161,13 +161,12 @@ static int xen_cpu_up_prepare_hvm(unsign
 	int rc = 0;
 
 	/*
-	 * This can happen if CPU was offlined earlier and
-	 * offlining timed out in common_cpu_die().
+	 * If a CPU was offlined earlier and offlining timed out then the
+	 * lock mechanism is still initialized. Uninit it unconditionally
+	 * as it's safe to call even if already uninited. Interrupts and
+	 * timer have already been handled in xen_cpu_dead_hvm().
 	 */
-	if (cpu_report_state(cpu) == CPU_DEAD_FROZEN) {
-		xen_smp_intr_free(cpu);
-		xen_uninit_lock_cpu(cpu);
-	}
+	xen_uninit_lock_cpu(cpu);
 
 	if (cpu_acpi_id(cpu) != U32_MAX)
 		per_cpu(xen_vcpu_id, cpu) = cpu_acpi_id(cpu);

[patch v3 18/36] [patch V2 18/38] cpu/hotplug: Add CPU state tracking and synchronization

[Thomas Gleixner]

From: Thomas Gleixner <tglx@linutronix.de>

The CPU state tracking and synchronization mechanism in smpboot.c is
completely independent of the hotplug code and all logic around it is
implemented in architecture specific code.

Except for the state reporting of the AP there is absolutely nothing
architecture specific and the sychronization and decision functions can be
moved into the generic hotplug core code.

Provide an integrated variant and add the core synchronization and decision
points. This comes in two flavours:

  1) DEAD state synchronization

     Updated by the architecture code once the AP reaches the point where
     it is ready to be torn down by the control CPU, e.g. by removing power
     or clocks or tear down via the hypervisor.

     The control CPU waits for this state to be reached with a timeout. If
     the state is reached an architecture specific cleanup function is
     invoked.

  2) Full state synchronization

     This extends #1 with AP alive synchronization. This is new
     functionality, which allows to replace architecture specific wait
     mechanims, e.g. cpumasks, completely.

     It also prevents that an AP which is in a limbo state can be brought
     up again. This can happen when an AP failed to report dead state
     during a previous off-line operation.

The dead synchronization is what most architectures use. Only x86 makes a
bringup decision based on that state at the moment.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Michael Kelley <mikelley@microsoft.com>


---
 arch/Kconfig               |   15 +++
 include/linux/cpuhotplug.h |   12 ++
 kernel/cpu.c               |  193 ++++++++++++++++++++++++++++++++++++++++++++-
 kernel/smpboot.c           |    2 
 4 files changed, 221 insertions(+), 1 deletion(-)
---

--- a/arch/Kconfig
+++ b/arch/Kconfig
@@ -34,6 +34,21 @@ config ARCH_HAS_SUBPAGE_FAULTS
 config HOTPLUG_SMT
 	bool
 
+# Selected by HOTPLUG_CORE_SYNC_DEAD or HOTPLUG_CORE_SYNC_FULL
+config HOTPLUG_CORE_SYNC
+	bool
+
+# Basic CPU dead synchronization selected by architecture
+config HOTPLUG_CORE_SYNC_DEAD
+	bool
+	select HOTPLUG_CORE_SYNC
+
+# Full CPU synchronization with alive state selected by architecture
+config HOTPLUG_CORE_SYNC_FULL
+	bool
+	select HOTPLUG_CORE_SYNC_DEAD if HOTPLUG_CPU
+	select HOTPLUG_CORE_SYNC
+
 config GENERIC_ENTRY
 	bool
 
--- a/include/linux/cpuhotplug.h
+++ b/include/linux/cpuhotplug.h
@@ -517,4 +517,16 @@ void cpuhp_online_idle(enum cpuhp_state
 static inline void cpuhp_online_idle(enum cpuhp_state state) { }
 #endif
 
+void cpuhp_ap_sync_alive(void);
+void arch_cpuhp_sync_state_poll(void);
+void arch_cpuhp_cleanup_kick_cpu(unsigned int cpu);
+
+#ifdef CONFIG_HOTPLUG_CORE_SYNC_DEAD
+void cpuhp_ap_report_dead(void);
+void arch_cpuhp_cleanup_dead_cpu(unsigned int cpu);
+#else
+static inline void cpuhp_ap_report_dead(void) { }
+static inline void arch_cpuhp_cleanup_dead_cpu(unsigned int cpu) { }
+#endif
+
 #endif
--- a/kernel/cpu.c
+++ b/kernel/cpu.c
@@ -17,6 +17,7 @@
 #include <linux/cpu.h>
 #include <linux/oom.h>
 #include <linux/rcupdate.h>
+#include <linux/delay.h>
 #include <linux/export.h>
 #include <linux/bug.h>
 #include <linux/kthread.h>
@@ -59,6 +60,7 @@
  * @last:	For multi-instance rollback, remember how far we got
  * @cb_state:	The state for a single callback (install/uninstall)
  * @result:	Result of the operation
+ * @ap_sync_state:	State for AP synchronization
  * @done_up:	Signal completion to the issuer of the task for cpu-up
  * @done_down:	Signal completion to the issuer of the task for cpu-down
  */
@@ -76,6 +78,7 @@ struct cpuhp_cpu_state {
 	struct hlist_node	*last;
 	enum cpuhp_state	cb_state;
 	int			result;
+	atomic_t		ap_sync_state;
 	struct completion	done_up;
 	struct completion	done_down;
 #endif
@@ -276,6 +279,182 @@ static bool cpuhp_is_atomic_state(enum c
 	return CPUHP_AP_IDLE_DEAD <= state && state < CPUHP_AP_ONLINE;
 }
 
+/* Synchronization state management */
+enum cpuhp_sync_state {
+	SYNC_STATE_DEAD,
+	SYNC_STATE_KICKED,
+	SYNC_STATE_SHOULD_DIE,
+	SYNC_STATE_ALIVE,
+	SYNC_STATE_SHOULD_ONLINE,
+	SYNC_STATE_ONLINE,
+};
+
+#ifdef CONFIG_HOTPLUG_CORE_SYNC
+/**
+ * cpuhp_ap_update_sync_state - Update synchronization state during bringup/teardown
+ * @state:	The synchronization state to set
+ *
+ * No synchronization point. Just update of the synchronization state.
+ */
+static inline void cpuhp_ap_update_sync_state(enum cpuhp_sync_state state)
+{
+	atomic_t *st = this_cpu_ptr(&cpuhp_state.ap_sync_state);
+	int sync = atomic_read(st);
+
+	while (!atomic_try_cmpxchg(st, &sync, state));
+}
+
+void __weak arch_cpuhp_sync_state_poll(void) { cpu_relax(); }
+
+static bool cpuhp_wait_for_sync_state(unsigned int cpu, enum cpuhp_sync_state state,
+				      enum cpuhp_sync_state next_state)
+{
+	atomic_t *st = per_cpu_ptr(&cpuhp_state.ap_sync_state, cpu);
+	ktime_t now, end, start = ktime_get();
+	int sync;
+
+	end = start + 10ULL * NSEC_PER_SEC;
+
+	sync = atomic_read(st);
+	while (1) {
+		if (sync == state) {
+			if (!atomic_try_cmpxchg(st, &sync, next_state))
+				continue;
+			return true;
+		}
+
+		now = ktime_get();
+		if (now > end) {
+			/* Timeout. Leave the state unchanged */
+			return false;
+		} else if (now - start < NSEC_PER_MSEC) {
+			/* Poll for one millisecond */
+			arch_cpuhp_sync_state_poll();
+		} else {
+			usleep_range_state(USEC_PER_MSEC, 2 * USEC_PER_MSEC, TASK_UNINTERRUPTIBLE);
+		}
+		sync = atomic_read(st);
+	}
+	return true;
+}
+#else  /* CONFIG_HOTPLUG_CORE_SYNC */
+static inline void cpuhp_ap_update_sync_state(enum cpuhp_sync_state state) { }
+#endif /* !CONFIG_HOTPLUG_CORE_SYNC */
+
+#ifdef CONFIG_HOTPLUG_CORE_SYNC_DEAD
+/**
+ * cpuhp_ap_report_dead - Update synchronization state to DEAD
+ *
+ * No synchronization point. Just update of the synchronization state.
+ */
+void cpuhp_ap_report_dead(void)
+{
+	cpuhp_ap_update_sync_state(SYNC_STATE_DEAD);
+}
+
+void __weak arch_cpuhp_cleanup_dead_cpu(unsigned int cpu) { }
+
+/*
+ * Late CPU shutdown synchronization point. Cannot use cpuhp_state::done_down
+ * because the AP cannot issue complete() at this stage.
+ */
+static void cpuhp_bp_sync_dead(unsigned int cpu)
+{
+	atomic_t *st = per_cpu_ptr(&cpuhp_state.ap_sync_state, cpu);
+	int sync = atomic_read(st);
+
+	do {
+		/* CPU can have reported dead already. Don't overwrite that! */
+		if (sync == SYNC_STATE_DEAD)
+			break;
+	} while (!atomic_try_cmpxchg(st, &sync, SYNC_STATE_SHOULD_DIE));
+
+	if (cpuhp_wait_for_sync_state(cpu, SYNC_STATE_DEAD, SYNC_STATE_DEAD)) {
+		/* CPU reached dead state. Invoke the cleanup function */
+		arch_cpuhp_cleanup_dead_cpu(cpu);
+		return;
+	}
+
+	/* No further action possible. Emit message and give up. */
+	pr_err("CPU%u failed to report dead state\n", cpu);
+}
+#else /* CONFIG_HOTPLUG_CORE_SYNC_DEAD */
+static inline void cpuhp_bp_sync_dead(unsigned int cpu) { }
+#endif /* !CONFIG_HOTPLUG_CORE_SYNC_DEAD */
+
+#ifdef CONFIG_HOTPLUG_CORE_SYNC_FULL
+/**
+ * cpuhp_ap_sync_alive - Synchronize AP with the control CPU once it is alive
+ *
+ * Updates the AP synchronization state to SYNC_STATE_ALIVE and waits
+ * for the BP to release it.
+ */
+void cpuhp_ap_sync_alive(void)
+{
+	atomic_t *st = this_cpu_ptr(&cpuhp_state.ap_sync_state);
+
+	cpuhp_ap_update_sync_state(SYNC_STATE_ALIVE);
+
+	/* Wait for the control CPU to release it. */
+	while (atomic_read(st) != SYNC_STATE_SHOULD_ONLINE)
+		cpu_relax();
+}
+
+static bool cpuhp_can_boot_ap(unsigned int cpu)
+{
+	atomic_t *st = per_cpu_ptr(&cpuhp_state.ap_sync_state, cpu);
+	int sync = atomic_read(st);
+
+again:
+	switch (sync) {
+	case SYNC_STATE_DEAD:
+		/* CPU is properly dead */
+		break;
+	case SYNC_STATE_KICKED:
+		/* CPU did not come up in previous attempt */
+		break;
+	case SYNC_STATE_ALIVE:
+		/* CPU is stuck cpuhp_ap_sync_alive(). */
+		break;
+	default:
+		/* CPU failed to report online or dead and is in limbo state. */
+		return false;
+	}
+
+	/* Prepare for booting */
+	if (!atomic_try_cmpxchg(st, &sync, SYNC_STATE_KICKED))
+		goto again;
+
+	return true;
+}
+
+void __weak arch_cpuhp_cleanup_kick_cpu(unsigned int cpu) { }
+
+/*
+ * Early CPU bringup synchronization point. Cannot use cpuhp_state::done_up
+ * because the AP cannot issue complete() so early in the bringup.
+ */
+static int cpuhp_bp_sync_alive(unsigned int cpu)
+{
+	int ret = 0;
+
+	if (!IS_ENABLED(CONFIG_HOTPLUG_CORE_SYNC_FULL))
+		return 0;
+
+	if (!cpuhp_wait_for_sync_state(cpu, SYNC_STATE_ALIVE, SYNC_STATE_SHOULD_ONLINE)) {
+		pr_err("CPU%u failed to report alive state\n", cpu);
+		ret = -EIO;
+	}
+
+	/* Let the architecture cleanup the kick alive mechanics. */
+	arch_cpuhp_cleanup_kick_cpu(cpu);
+	return ret;
+}
+#else /* CONFIG_HOTPLUG_CORE_SYNC_FULL */
+static inline int cpuhp_bp_sync_alive(unsigned int cpu) { return 0; }
+static inline bool cpuhp_can_boot_ap(unsigned int cpu) { return true; }
+#endif /* !CONFIG_HOTPLUG_CORE_SYNC_FULL */
+
 /* Serializes the updates to cpu_online_mask, cpu_present_mask */
 static DEFINE_MUTEX(cpu_add_remove_lock);
 bool cpuhp_tasks_frozen;
@@ -588,6 +767,9 @@ static int bringup_cpu(unsigned int cpu)
 	struct task_struct *idle = idle_thread_get(cpu);
 	int ret;
 
+	if (!cpuhp_can_boot_ap(cpu))
+		return -EAGAIN;
+
 	/*
 	 * Reset stale stack state from the last time this CPU was online.
 	 */
@@ -606,6 +788,10 @@ static int bringup_cpu(unsigned int cpu)
 	if (ret)
 		goto out_unlock;
 
+	ret = cpuhp_bp_sync_alive(cpu);
+	if (ret)
+		goto out_unlock;
+
 	ret = bringup_wait_for_ap_online(cpu);
 	if (ret)
 		goto out_unlock;
@@ -1109,6 +1295,8 @@ static int takedown_cpu(unsigned int cpu
 	/* This actually kills the CPU. */
 	__cpu_die(cpu);
 
+	cpuhp_bp_sync_dead(cpu);
+
 	tick_cleanup_dead_cpu(cpu);
 	rcutree_migrate_callbacks(cpu);
 	return 0;
@@ -1355,8 +1543,10 @@ void cpuhp_online_idle(enum cpuhp_state
 	if (state != CPUHP_AP_ONLINE_IDLE)
 		return;
 
+	cpuhp_ap_update_sync_state(SYNC_STATE_ONLINE);
+
 	/*
-	 * Unpart the stopper thread before we start the idle loop (and start
+	 * Unpark the stopper thread before we start the idle loop (and start
 	 * scheduling); this ensures the stopper task is always available.
 	 */
 	stop_machine_unpark(smp_processor_id());
@@ -2733,6 +2923,7 @@ void __init boot_cpu_hotplug_init(void)
 {
 #ifdef CONFIG_SMP
 	cpumask_set_cpu(smp_processor_id(), &cpus_booted_once_mask);
+	atomic_set(this_cpu_ptr(&cpuhp_state.ap_sync_state), SYNC_STATE_ONLINE);
 #endif
 	this_cpu_write(cpuhp_state.state, CPUHP_ONLINE);
 	this_cpu_write(cpuhp_state.target, CPUHP_ONLINE);
--- a/kernel/smpboot.c
+++ b/kernel/smpboot.c
@@ -326,6 +326,7 @@ void smpboot_unregister_percpu_thread(st
 }
 EXPORT_SYMBOL_GPL(smpboot_unregister_percpu_thread);
 
+#ifndef CONFIG_HOTPLUG_CORE_SYNC
 static DEFINE_PER_CPU(atomic_t, cpu_hotplug_state) = ATOMIC_INIT(CPU_POST_DEAD);
 
 /*
@@ -488,3 +489,4 @@ bool cpu_report_death(void)
 }
 
 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
+#endif /* !CONFIG_HOTPLUG_CORE_SYNC */

Re: [patch v3 18/36] [patch V2 18/38] cpu/hotplug: Add CPU state tracking and synchronization

[Peter Zijlstra]

On Mon, May 08, 2023 at 09:43:55PM +0200, Thomas Gleixner wrote:

> +static inline void cpuhp_ap_update_sync_state(enum cpuhp_sync_state state)
> +{
> +	atomic_t *st = this_cpu_ptr(&cpuhp_state.ap_sync_state);
> +	int sync = atomic_read(st);
> +
> +	while (!atomic_try_cmpxchg(st, &sync, state));
> +}

Why isn't:

	atomic_set(st, state);

any good?

Re: [patch v3 18/36] [patch V2 18/38] cpu/hotplug: Add CPU state tracking and synchronization

[Peter Zijlstra]

On Tue, May 09, 2023 at 01:07:23PM +0200, Peter Zijlstra wrote:
> On Mon, May 08, 2023 at 09:43:55PM +0200, Thomas Gleixner wrote:
> 
> > +static inline void cpuhp_ap_update_sync_state(enum cpuhp_sync_state state)
> > +{
> > +	atomic_t *st = this_cpu_ptr(&cpuhp_state.ap_sync_state);
> > +	int sync = atomic_read(st);
> > +
> > +	while (!atomic_try_cmpxchg(st, &sync, state));
> > +}
> 
> Why isn't:
> 
> 	atomic_set(st, state);
> 
> any good?

Hmm, should at the very least be atomic_set_release(), but if you want
the full barrier then:

	(void)atomic_xchg(st, state);

is the much saner way to write the above.

Re: [patch v3 18/36] [patch V2 18/38] cpu/hotplug: Add CPU state tracking and synchronization

[Thomas Gleixner]

On Tue, May 09 2023 at 13:07, Peter Zijlstra wrote:
> On Mon, May 08, 2023 at 09:43:55PM +0200, Thomas Gleixner wrote:
>
>> +static inline void cpuhp_ap_update_sync_state(enum cpuhp_sync_state state)
>> +{
>> +	atomic_t *st = this_cpu_ptr(&cpuhp_state.ap_sync_state);
>> +	int sync = atomic_read(st);
>> +
>> +	while (!atomic_try_cmpxchg(st, &sync, state));
>> +}
>
> Why isn't:
>
> 	atomic_set(st, state);
>
> any good?

Good question. It's only the other side (control CPU) which needs to be
careful.

Thanks,

        tglx

[patch v3 19/36] x86/smpboot: Switch to hotplug core state synchronization

[Thomas Gleixner]

From: Thomas Gleixner <tglx@linutronix.de>

The new AP state tracking and synchronization mechanism in the CPU hotplug
core code allows to remove quite some x86 specific code:

  1) The AP alive synchronization based on cpumasks

  2) The decision whether an AP can be brought up again

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Michael Kelley <mikelley@microsoft.com>

---
V2: Use for_each_online_cpu() - Brian
---
 arch/x86/Kconfig           |    1 
 arch/x86/include/asm/smp.h |    7 +
 arch/x86/kernel/smp.c      |    1 
 arch/x86/kernel/smpboot.c  |  161 ++++++++++-----------------------------------
 arch/x86/xen/smp_hvm.c     |   16 +---
 arch/x86/xen/smp_pv.c      |   39 ++++++----
 6 files changed, 73 insertions(+), 152 deletions(-)
---

--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -274,6 +274,7 @@ config X86
 	select HAVE_UNSTABLE_SCHED_CLOCK
 	select HAVE_USER_RETURN_NOTIFIER
 	select HAVE_GENERIC_VDSO
+	select HOTPLUG_CORE_SYNC_FULL		if SMP
 	select HOTPLUG_SMT			if SMP
 	select IRQ_FORCED_THREADING
 	select NEED_PER_CPU_EMBED_FIRST_CHUNK
--- a/arch/x86/include/asm/smp.h
+++ b/arch/x86/include/asm/smp.h
@@ -38,6 +38,8 @@ struct smp_ops {
 	void (*crash_stop_other_cpus)(void);
 	void (*smp_send_reschedule)(int cpu);
 
+	void (*cleanup_dead_cpu)(unsigned cpu);
+	void (*poll_sync_state)(void);
 	int (*cpu_up)(unsigned cpu, struct task_struct *tidle);
 	int (*cpu_disable)(void);
 	void (*cpu_die)(unsigned int cpu);
@@ -90,7 +92,8 @@ static inline int __cpu_disable(void)
 
 static inline void __cpu_die(unsigned int cpu)
 {
-	smp_ops.cpu_die(cpu);
+	if (smp_ops.cpu_die)
+		smp_ops.cpu_die(cpu);
 }
 
 static inline void __noreturn play_dead(void)
@@ -123,8 +126,6 @@ void native_smp_cpus_done(unsigned int m
 int common_cpu_up(unsigned int cpunum, struct task_struct *tidle);
 int native_cpu_up(unsigned int cpunum, struct task_struct *tidle);
 int native_cpu_disable(void);
-int common_cpu_die(unsigned int cpu);
-void native_cpu_die(unsigned int cpu);
 void __noreturn hlt_play_dead(void);
 void native_play_dead(void);
 void play_dead_common(void);
--- a/arch/x86/kernel/smp.c
+++ b/arch/x86/kernel/smp.c
@@ -269,7 +269,6 @@ struct smp_ops smp_ops = {
 	.smp_send_reschedule	= native_smp_send_reschedule,
 
 	.cpu_up			= native_cpu_up,
-	.cpu_die		= native_cpu_die,
 	.cpu_disable		= native_cpu_disable,
 	.play_dead		= native_play_dead,
 
--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -57,6 +57,7 @@
 #include <linux/pgtable.h>
 #include <linux/overflow.h>
 #include <linux/stackprotector.h>
+#include <linux/cpuhotplug.h>
 
 #include <asm/acpi.h>
 #include <asm/cacheinfo.h>
@@ -101,9 +102,6 @@ EXPORT_PER_CPU_SYMBOL(cpu_die_map);
 DEFINE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info);
 EXPORT_PER_CPU_SYMBOL(cpu_info);
 
-/* All of these masks are initialized in setup_cpu_local_masks() */
-static cpumask_var_t cpu_initialized_mask;
-static cpumask_var_t cpu_callout_mask;
 /* Representing CPUs for which sibling maps can be computed */
 static cpumask_var_t cpu_sibling_setup_mask;
 
@@ -169,8 +167,8 @@ static void smp_callin(void)
 	int cpuid = smp_processor_id();
 
 	/*
-	 * If waken up by an INIT in an 82489DX configuration
-	 * cpu_callout_mask guarantees we don't get here before an
+	 * If waken up by an INIT in an 82489DX configuration the alive
+	 * synchronization guarantees we don't get here before an
 	 * INIT_deassert IPI reaches our local APIC, so it is now safe to
 	 * touch our local APIC.
 	 *
@@ -216,17 +214,6 @@ static void ap_calibrate_delay(void)
 	cpu_data(smp_processor_id()).loops_per_jiffy = loops_per_jiffy;
 }
 
-static void wait_for_master_cpu(int cpu)
-{
-	/*
-	 * Wait for release by control CPU before continuing with AP
-	 * initialization.
-	 */
-	WARN_ON(cpumask_test_and_set_cpu(cpu, cpu_initialized_mask));
-	while (!cpumask_test_cpu(cpu, cpu_callout_mask))
-		cpu_relax();
-}
-
 /*
  * Activate a secondary processor.
  */
@@ -247,11 +234,11 @@ static void notrace start_secondary(void
 	cpu_init_exception_handling();
 
 	/*
-	 * Sync point with wait_cpu_initialized(). Sets AP in
-	 * cpu_initialized_mask and then waits for the control CPU
-	 * to release it.
+	 * Synchronization point with the hotplug core. Sets the
+	 * synchronization state to ALIVE and waits for the control CPU to
+	 * release this CPU for further bringup.
 	 */
-	wait_for_master_cpu(raw_smp_processor_id());
+	cpuhp_ap_sync_alive();
 
 	cpu_init();
 	rcu_cpu_starting(raw_smp_processor_id());
@@ -284,7 +271,6 @@ static void notrace start_secondary(void
 	set_cpu_online(smp_processor_id(), true);
 	lapic_online();
 	unlock_vector_lock();
-	cpu_set_state_online(smp_processor_id());
 	x86_platform.nmi_init();
 
 	/* enable local interrupts */
@@ -735,9 +721,9 @@ static void impress_friends(void)
 	 * Allow the user to impress friends.
 	 */
 	pr_debug("Before bogomips\n");
-	for_each_possible_cpu(cpu)
-		if (cpumask_test_cpu(cpu, cpu_callout_mask))
-			bogosum += cpu_data(cpu).loops_per_jiffy;
+	for_each_online_cpu(cpu)
+		bogosum += cpu_data(cpu).loops_per_jiffy;
+
 	pr_info("Total of %d processors activated (%lu.%02lu BogoMIPS)\n",
 		num_online_cpus(),
 		bogosum/(500000/HZ),
@@ -1009,6 +995,7 @@ int common_cpu_up(unsigned int cpu, stru
 static int do_boot_cpu(int apicid, int cpu, struct task_struct *idle)
 {
 	unsigned long start_ip = real_mode_header->trampoline_start;
+	int ret;
 
 #ifdef CONFIG_X86_64
 	/* If 64-bit wakeup method exists, use the 64-bit mode trampoline IP */
@@ -1049,13 +1036,6 @@ static int do_boot_cpu(int apicid, int c
 		}
 	}
 
-	/*
-	 * AP might wait on cpu_callout_mask in cpu_init() with
-	 * cpu_initialized_mask set if previous attempt to online
-	 * it timed-out. Clear cpu_initialized_mask so that after
-	 * INIT/SIPI it could start with a clean state.
-	 */
-	cpumask_clear_cpu(cpu, cpu_initialized_mask);
 	smp_mb();
 
 	/*
@@ -1066,47 +1046,16 @@ static int do_boot_cpu(int apicid, int c
 	 * - Use an INIT boot APIC message
 	 */
 	if (apic->wakeup_secondary_cpu_64)
-		return apic->wakeup_secondary_cpu_64(apicid, start_ip);
+		ret = apic->wakeup_secondary_cpu_64(apicid, start_ip);
 	else if (apic->wakeup_secondary_cpu)
-		return apic->wakeup_secondary_cpu(apicid, start_ip);
-
-	return wakeup_secondary_cpu_via_init(apicid, start_ip);
-}
-
-static int wait_cpu_cpumask(unsigned int cpu, const struct cpumask *mask)
-{
-	unsigned long timeout;
-
-	/*
-	 * Wait up to 10s for the CPU to report in.
-	 */
-	timeout = jiffies + 10*HZ;
-	while (time_before(jiffies, timeout)) {
-		if (cpumask_test_cpu(cpu, mask))
-			return 0;
-
-		schedule();
-	}
-	return -1;
-}
-
-/*
- * Bringup step two: Wait for the target AP to reach cpu_init_secondary()
- * and thus wait_for_master_cpu(), then set cpu_callout_mask to allow it
- * to proceed.  The AP will then proceed past setting its 'callin' bit
- * and end up waiting in check_tsc_sync_target() until we reach
- * wait_cpu_online() to tend to it.
- */
-static int wait_cpu_initialized(unsigned int cpu)
-{
-	/*
-	 * Wait for first sign of life from AP.
-	 */
-	if (wait_cpu_cpumask(cpu, cpu_initialized_mask))
-		return -1;
+		ret = apic->wakeup_secondary_cpu(apicid, start_ip);
+	else
+		ret = wakeup_secondary_cpu_via_init(apicid, start_ip);
 
-	cpumask_set_cpu(cpu, cpu_callout_mask);
-	return 0;
+	/* If the wakeup mechanism failed, cleanup the warm reset vector */
+	if (ret)
+		arch_cpuhp_cleanup_kick_cpu(cpu);
+	return ret;
 }
 
 static int native_kick_ap(unsigned int cpu, struct task_struct *tidle)
@@ -1131,11 +1080,6 @@ static int native_kick_ap(unsigned int c
 	 */
 	mtrr_save_state();
 
-	/* x86 CPUs take themselves offline, so delayed offline is OK. */
-	err = cpu_check_up_prepare(cpu);
-	if (err && err != -EBUSY)
-		return err;
-
 	/* the FPU context is blank, nobody can own it */
 	per_cpu(fpu_fpregs_owner_ctx, cpu) = NULL;
 
@@ -1152,17 +1096,29 @@ static int native_kick_ap(unsigned int c
 
 int native_cpu_up(unsigned int cpu, struct task_struct *tidle)
 {
-	int ret;
-
-	ret = native_kick_ap(cpu, tidle);
-	if (!ret)
-		ret = wait_cpu_initialized(cpu);
+	return native_kick_ap(cpu, tidle);
+}
 
+void arch_cpuhp_cleanup_kick_cpu(unsigned int cpu)
+{
 	/* Cleanup possible dangling ends... */
-	if (x86_platform.legacy.warm_reset)
+	if (smp_ops.cpu_up == native_cpu_up && x86_platform.legacy.warm_reset)
 		smpboot_restore_warm_reset_vector();
+}
 
-	return ret;
+void arch_cpuhp_cleanup_dead_cpu(unsigned int cpu)
+{
+	if (smp_ops.cleanup_dead_cpu)
+		smp_ops.cleanup_dead_cpu(cpu);
+
+	if (system_state == SYSTEM_RUNNING)
+		pr_info("CPU %u is now offline\n", cpu);
+}
+
+void arch_cpuhp_sync_state_poll(void)
+{
+	if (smp_ops.poll_sync_state)
+		smp_ops.poll_sync_state();
 }
 
 /**
@@ -1354,9 +1310,6 @@ void __init native_smp_prepare_boot_cpu(
 	if (!IS_ENABLED(CONFIG_SMP))
 		switch_gdt_and_percpu_base(me);
 
-	/* already set me in cpu_online_mask in boot_cpu_init() */
-	cpumask_set_cpu(me, cpu_callout_mask);
-	cpu_set_state_online(me);
 	native_pv_lock_init();
 }
 
@@ -1483,8 +1436,6 @@ early_param("possible_cpus", _setup_poss
 /* correctly size the local cpu masks */
 void __init setup_cpu_local_masks(void)
 {
-	alloc_bootmem_cpumask_var(&cpu_initialized_mask);
-	alloc_bootmem_cpumask_var(&cpu_callout_mask);
 	alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask);
 }
 
@@ -1546,9 +1497,6 @@ static void remove_siblinginfo(int cpu)
 static void remove_cpu_from_maps(int cpu)
 {
 	set_cpu_online(cpu, false);
-	cpumask_clear_cpu(cpu, cpu_callout_mask);
-	/* was set by cpu_init() */
-	cpumask_clear_cpu(cpu, cpu_initialized_mask);
 	numa_remove_cpu(cpu);
 }
 
@@ -1599,36 +1547,11 @@ int native_cpu_disable(void)
 	return 0;
 }
 
-int common_cpu_die(unsigned int cpu)
-{
-	int ret = 0;
-
-	/* We don't do anything here: idle task is faking death itself. */
-
-	/* They ack this in play_dead() by setting CPU_DEAD */
-	if (cpu_wait_death(cpu, 5)) {
-		if (system_state == SYSTEM_RUNNING)
-			pr_info("CPU %u is now offline\n", cpu);
-	} else {
-		pr_err("CPU %u didn't die...\n", cpu);
-		ret = -1;
-	}
-
-	return ret;
-}
-
-void native_cpu_die(unsigned int cpu)
-{
-	common_cpu_die(cpu);
-}
-
 void play_dead_common(void)
 {
 	idle_task_exit();
 
-	/* Ack it */
-	(void)cpu_report_death();
-
+	cpuhp_ap_report_dead();
 	/*
 	 * With physical CPU hotplug, we should halt the cpu
 	 */
@@ -1730,12 +1653,6 @@ int native_cpu_disable(void)
 	return -ENOSYS;
 }
 
-void native_cpu_die(unsigned int cpu)
-{
-	/* We said "no" in __cpu_disable */
-	BUG();
-}
-
 void native_play_dead(void)
 {
 	BUG();
--- a/arch/x86/xen/smp_hvm.c
+++ b/arch/x86/xen/smp_hvm.c
@@ -55,18 +55,16 @@ static void __init xen_hvm_smp_prepare_c
 }
 
 #ifdef CONFIG_HOTPLUG_CPU
-static void xen_hvm_cpu_die(unsigned int cpu)
+static void xen_hvm_cleanup_dead_cpu(unsigned int cpu)
 {
-	if (common_cpu_die(cpu) == 0) {
-		if (xen_have_vector_callback) {
-			xen_smp_intr_free(cpu);
-			xen_uninit_lock_cpu(cpu);
-			xen_teardown_timer(cpu);
-		}
+	if (xen_have_vector_callback) {
+		xen_smp_intr_free(cpu);
+		xen_uninit_lock_cpu(cpu);
+		xen_teardown_timer(cpu);
 	}
 }
 #else
-static void xen_hvm_cpu_die(unsigned int cpu)
+static void xen_hvm_cleanup_dead_cpu(unsigned int cpu)
 {
 	BUG();
 }
@@ -77,7 +75,7 @@ void __init xen_hvm_smp_init(void)
 	smp_ops.smp_prepare_boot_cpu = xen_hvm_smp_prepare_boot_cpu;
 	smp_ops.smp_prepare_cpus = xen_hvm_smp_prepare_cpus;
 	smp_ops.smp_cpus_done = xen_smp_cpus_done;
-	smp_ops.cpu_die = xen_hvm_cpu_die;
+	smp_ops.cleanup_dead_cpu = xen_hvm_cleanup_dead_cpu;
 
 	if (!xen_have_vector_callback) {
 #ifdef CONFIG_PARAVIRT_SPINLOCKS
--- a/arch/x86/xen/smp_pv.c
+++ b/arch/x86/xen/smp_pv.c
@@ -62,6 +62,7 @@ static void cpu_bringup(void)
 	int cpu;
 
 	cr4_init();
+	cpuhp_ap_sync_alive();
 	cpu_init();
 	touch_softlockup_watchdog();
 
@@ -83,7 +84,7 @@ static void cpu_bringup(void)
 
 	set_cpu_online(cpu, true);
 
-	cpu_set_state_online(cpu);  /* Implies full memory barrier. */
+	smp_mb();
 
 	/* We can take interrupts now: we're officially "up". */
 	local_irq_enable();
@@ -323,14 +324,6 @@ static int xen_pv_cpu_up(unsigned int cp
 
 	xen_setup_runstate_info(cpu);
 
-	/*
-	 * PV VCPUs are always successfully taken down (see 'while' loop
-	 * in xen_cpu_die()), so -EBUSY is an error.
-	 */
-	rc = cpu_check_up_prepare(cpu);
-	if (rc)
-		return rc;
-
 	/* make sure interrupts start blocked */
 	per_cpu(xen_vcpu, cpu)->evtchn_upcall_mask = 1;
 
@@ -349,6 +342,11 @@ static int xen_pv_cpu_up(unsigned int cp
 	return 0;
 }
 
+static void xen_pv_poll_sync_state(void)
+{
+	HYPERVISOR_sched_op(SCHEDOP_yield, NULL);
+}
+
 #ifdef CONFIG_HOTPLUG_CPU
 static int xen_pv_cpu_disable(void)
 {
@@ -364,18 +362,18 @@ static int xen_pv_cpu_disable(void)
 
 static void xen_pv_cpu_die(unsigned int cpu)
 {
-	while (HYPERVISOR_vcpu_op(VCPUOP_is_up,
-				  xen_vcpu_nr(cpu), NULL)) {
+	while (HYPERVISOR_vcpu_op(VCPUOP_is_up, xen_vcpu_nr(cpu), NULL)) {
 		__set_current_state(TASK_UNINTERRUPTIBLE);
 		schedule_timeout(HZ/10);
 	}
+}
 
-	if (common_cpu_die(cpu) == 0) {
-		xen_smp_intr_free(cpu);
-		xen_uninit_lock_cpu(cpu);
-		xen_teardown_timer(cpu);
-		xen_pmu_finish(cpu);
-	}
+static void xen_pv_cleanup_dead_cpu(unsigned int cpu)
+{
+	xen_smp_intr_free(cpu);
+	xen_uninit_lock_cpu(cpu);
+	xen_teardown_timer(cpu);
+	xen_pmu_finish(cpu);
 }
 
 static void __noreturn xen_pv_play_dead(void) /* used only with HOTPLUG_CPU */
@@ -397,6 +395,11 @@ static void xen_pv_cpu_die(unsigned int
 	BUG();
 }
 
+static void xen_pv_cleanup_dead_cpu(unsigned int cpu)
+{
+	BUG();
+}
+
 static void __noreturn xen_pv_play_dead(void)
 {
 	BUG();
@@ -437,6 +440,8 @@ static const struct smp_ops xen_smp_ops
 
 	.cpu_up = xen_pv_cpu_up,
 	.cpu_die = xen_pv_cpu_die,
+	.cleanup_dead_cpu = xen_pv_cleanup_dead_cpu,
+	.poll_sync_state = xen_pv_poll_sync_state,
 	.cpu_disable = xen_pv_cpu_disable,
 	.play_dead = xen_pv_play_dead,
 

[patch v3 20/36] cpu/hotplug: Remove cpu_report_state() and related unused cruft

[Thomas Gleixner]

From: Thomas Gleixner <tglx@linutronix.de>

No more users.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Michael Kelley <mikelley@microsoft.com>


---
 include/linux/cpu.h |    2 -
 kernel/smpboot.c    |   90 ----------------------------------------------------
 2 files changed, 92 deletions(-)
---

--- a/include/linux/cpu.h
+++ b/include/linux/cpu.h
@@ -184,8 +184,6 @@ void arch_cpu_idle_enter(void);
 void arch_cpu_idle_exit(void);
 void __noreturn arch_cpu_idle_dead(void);
 
-int cpu_report_state(int cpu);
-int cpu_check_up_prepare(int cpu);
 void cpu_set_state_online(int cpu);
 void play_idle_precise(u64 duration_ns, u64 latency_ns);
 
--- a/kernel/smpboot.c
+++ b/kernel/smpboot.c
@@ -329,97 +329,7 @@ EXPORT_SYMBOL_GPL(smpboot_unregister_per
 #ifndef CONFIG_HOTPLUG_CORE_SYNC
 static DEFINE_PER_CPU(atomic_t, cpu_hotplug_state) = ATOMIC_INIT(CPU_POST_DEAD);
 
-/*
- * Called to poll specified CPU's state, for example, when waiting for
- * a CPU to come online.
- */
-int cpu_report_state(int cpu)
-{
-	return atomic_read(&per_cpu(cpu_hotplug_state, cpu));
-}
-
-/*
- * If CPU has died properly, set its state to CPU_UP_PREPARE and
- * return success.  Otherwise, return -EBUSY if the CPU died after
- * cpu_wait_death() timed out.  And yet otherwise again, return -EAGAIN
- * if cpu_wait_death() timed out and the CPU still hasn't gotten around
- * to dying.  In the latter two cases, the CPU might not be set up
- * properly, but it is up to the arch-specific code to decide.
- * Finally, -EIO indicates an unanticipated problem.
- *
- * Note that it is permissible to omit this call entirely, as is
- * done in architectures that do no CPU-hotplug error checking.
- */
-int cpu_check_up_prepare(int cpu)
-{
-	if (!IS_ENABLED(CONFIG_HOTPLUG_CPU)) {
-		atomic_set(&per_cpu(cpu_hotplug_state, cpu), CPU_UP_PREPARE);
-		return 0;
-	}
-
-	switch (atomic_read(&per_cpu(cpu_hotplug_state, cpu))) {
-
-	case CPU_POST_DEAD:
-
-		/* The CPU died properly, so just start it up again. */
-		atomic_set(&per_cpu(cpu_hotplug_state, cpu), CPU_UP_PREPARE);
-		return 0;
-
-	case CPU_DEAD_FROZEN:
-
-		/*
-		 * Timeout during CPU death, so let caller know.
-		 * The outgoing CPU completed its processing, but after
-		 * cpu_wait_death() timed out and reported the error. The
-		 * caller is free to proceed, in which case the state
-		 * will be reset properly by cpu_set_state_online().
-		 * Proceeding despite this -EBUSY return makes sense
-		 * for systems where the outgoing CPUs take themselves
-		 * offline, with no post-death manipulation required from
-		 * a surviving CPU.
-		 */
-		return -EBUSY;
-
-	case CPU_BROKEN:
-
-		/*
-		 * The most likely reason we got here is that there was
-		 * a timeout during CPU death, and the outgoing CPU never
-		 * did complete its processing.  This could happen on
-		 * a virtualized system if the outgoing VCPU gets preempted
-		 * for more than five seconds, and the user attempts to
-		 * immediately online that same CPU.  Trying again later
-		 * might return -EBUSY above, hence -EAGAIN.
-		 */
-		return -EAGAIN;
-
-	case CPU_UP_PREPARE:
-		/*
-		 * Timeout while waiting for the CPU to show up. Allow to try
-		 * again later.
-		 */
-		return 0;
-
-	default:
-
-		/* Should not happen.  Famous last words. */
-		return -EIO;
-	}
-}
-
-/*
- * Mark the specified CPU online.
- *
- * Note that it is permissible to omit this call entirely, as is
- * done in architectures that do no CPU-hotplug error checking.
- */
-void cpu_set_state_online(int cpu)
-{
-	(void)atomic_xchg(&per_cpu(cpu_hotplug_state, cpu), CPU_ONLINE);
-}
-
 #ifdef CONFIG_HOTPLUG_CPU
-
 /*
  * Wait for the specified CPU to exit the idle loop and die.
  */

[patch v3 21/36] [patch V2 21/38] ARM: smp: Switch to hotplug core state synchronization

[Thomas Gleixner]

From: Thomas Gleixner <tglx@linutronix.de>

Switch to the CPU hotplug core state tracking and synchronization
mechanim. No functional change intended.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Michael Kelley <mikelley@microsoft.com>


---
 arch/arm/Kconfig           |    1 +
 arch/arm/include/asm/smp.h |    2 +-
 arch/arm/kernel/smp.c      |   18 +++++++-----------
 3 files changed, 9 insertions(+), 12 deletions(-)
---

--- a/arch/arm/Kconfig
+++ b/arch/arm/Kconfig
@@ -124,6 +124,7 @@ config ARM
 	select HAVE_SYSCALL_TRACEPOINTS
 	select HAVE_UID16
 	select HAVE_VIRT_CPU_ACCOUNTING_GEN
+	select HOTPLUG_CORE_SYNC_DEAD if HOTPLUG_CPU
 	select IRQ_FORCED_THREADING
 	select MODULES_USE_ELF_REL
 	select NEED_DMA_MAP_STATE
--- a/arch/arm/include/asm/smp.h
+++ b/arch/arm/include/asm/smp.h
@@ -64,7 +64,7 @@ extern void secondary_startup_arm(void);
 
 extern int __cpu_disable(void);
 
-extern void __cpu_die(unsigned int cpu);
+static inline void __cpu_die(unsigned int cpu) { }
 
 extern void arch_send_call_function_single_ipi(int cpu);
 extern void arch_send_call_function_ipi_mask(const struct cpumask *mask);
--- a/arch/arm/kernel/smp.c
+++ b/arch/arm/kernel/smp.c
@@ -288,15 +288,11 @@ int __cpu_disable(void)
 }
 
 /*
- * called on the thread which is asking for a CPU to be shutdown -
- * waits until shutdown has completed, or it is timed out.
+ * called on the thread which is asking for a CPU to be shutdown after the
+ * shutdown completed.
  */
-void __cpu_die(unsigned int cpu)
+void arch_cpuhp_cleanup_dead_cpu(unsigned int cpu)
 {
-	if (!cpu_wait_death(cpu, 5)) {
-		pr_err("CPU%u: cpu didn't die\n", cpu);
-		return;
-	}
 	pr_debug("CPU%u: shutdown\n", cpu);
 
 	clear_tasks_mm_cpumask(cpu);
@@ -336,11 +332,11 @@ void __noreturn arch_cpu_idle_dead(void)
 	flush_cache_louis();
 
 	/*
-	 * Tell __cpu_die() that this CPU is now safe to dispose of.  Once
-	 * this returns, power and/or clocks can be removed at any point
-	 * from this CPU and its cache by platform_cpu_kill().
+	 * Tell cpuhp_bp_sync_dead() that this CPU is now safe to dispose
+	 * of. Once this returns, power and/or clocks can be removed at
+	 * any point from this CPU and its cache by platform_cpu_kill().
 	 */
-	(void)cpu_report_death();
+	cpuhp_ap_report_dead();
 
 	/*
 	 * Ensure that the cache lines associated with that completion are

[patch v3 22/36] arm64: smp: Switch to hotplug core state synchronization

[Thomas Gleixner]

From: Thomas Gleixner <tglx@linutronix.de>

Switch to the CPU hotplug core state tracking and synchronization
mechanim. No functional change intended.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Mark Rutland <mark.rutland@arm.com>
Tested-by: Michael Kelley <mikelley@microsoft.com>

---
 arch/arm64/Kconfig           |    1 +
 arch/arm64/include/asm/smp.h |    2 +-
 arch/arm64/kernel/smp.c      |   14 +++++---------
 3 files changed, 7 insertions(+), 10 deletions(-)
---

--- a/arch/arm64/Kconfig
+++ b/arch/arm64/Kconfig
@@ -222,6 +222,7 @@ config ARM64
 	select HAVE_KPROBES
 	select HAVE_KRETPROBES
 	select HAVE_GENERIC_VDSO
+	select HOTPLUG_CORE_SYNC_DEAD if HOTPLUG_CPU
 	select IRQ_DOMAIN
 	select IRQ_FORCED_THREADING
 	select KASAN_VMALLOC if KASAN
--- a/arch/arm64/include/asm/smp.h
+++ b/arch/arm64/include/asm/smp.h
@@ -99,7 +99,7 @@ static inline void arch_send_wakeup_ipi_
 
 extern int __cpu_disable(void);
 
-extern void __cpu_die(unsigned int cpu);
+static inline void __cpu_die(unsigned int cpu) { }
 extern void __noreturn cpu_die(void);
 extern void __noreturn cpu_die_early(void);
 
--- a/arch/arm64/kernel/smp.c
+++ b/arch/arm64/kernel/smp.c
@@ -332,17 +332,13 @@ static int op_cpu_kill(unsigned int cpu)
 }
 
 /*
- * called on the thread which is asking for a CPU to be shutdown -
- * waits until shutdown has completed, or it is timed out.
+ * Called on the thread which is asking for a CPU to be shutdown after the
+ * shutdown completed.
  */
-void __cpu_die(unsigned int cpu)
+void arch_cpuhp_cleanup_dead_cpu(unsigned int cpu)
 {
 	int err;
 
-	if (!cpu_wait_death(cpu, 5)) {
-		pr_crit("CPU%u: cpu didn't die\n", cpu);
-		return;
-	}
 	pr_debug("CPU%u: shutdown\n", cpu);
 
 	/*
@@ -369,8 +365,8 @@ void __noreturn cpu_die(void)
 
 	local_daif_mask();
 
-	/* Tell __cpu_die() that this CPU is now safe to dispose of */
-	(void)cpu_report_death();
+	/* Tell cpuhp_bp_sync_dead() that this CPU is now safe to dispose of */
+	cpuhp_ap_report_dead();
 
 	/*
 	 * Actually shutdown the CPU. This must never fail. The specific hotplug

[patch v3 23/36] [patch V2 23/38] csky/smp: Switch to hotplug core state synchronization

[Thomas Gleixner]

From: Thomas Gleixner <tglx@linutronix.de>

Switch to the CPU hotplug core state tracking and synchronization
mechanim. No functional change intended.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Michael Kelley <mikelley@microsoft.com>


---
 arch/csky/Kconfig           |    1 +
 arch/csky/include/asm/smp.h |    2 +-
 arch/csky/kernel/smp.c      |    8 ++------
 3 files changed, 4 insertions(+), 7 deletions(-)
---

--- a/arch/csky/Kconfig
+++ b/arch/csky/Kconfig
@@ -96,6 +96,7 @@ config CSKY
 	select HAVE_REGS_AND_STACK_ACCESS_API
 	select HAVE_STACKPROTECTOR
 	select HAVE_SYSCALL_TRACEPOINTS
+	select HOTPLUG_CORE_SYNC_DEAD if HOTPLUG_CPU
 	select MAY_HAVE_SPARSE_IRQ
 	select MODULES_USE_ELF_RELA if MODULES
 	select OF
--- a/arch/csky/include/asm/smp.h
+++ b/arch/csky/include/asm/smp.h
@@ -23,7 +23,7 @@ void __init set_send_ipi(void (*func)(co
 
 int __cpu_disable(void);
 
-void __cpu_die(unsigned int cpu);
+static inline void __cpu_die(unsigned int cpu) { }
 
 #endif /* CONFIG_SMP */
 
--- a/arch/csky/kernel/smp.c
+++ b/arch/csky/kernel/smp.c
@@ -291,12 +291,8 @@ int __cpu_disable(void)
 	return 0;
 }
 
-void __cpu_die(unsigned int cpu)
+void arch_cpuhp_cleanup_dead_cpu(unsigned int cpu)
 {
-	if (!cpu_wait_death(cpu, 5)) {
-		pr_crit("CPU%u: shutdown failed\n", cpu);
-		return;
-	}
 	pr_notice("CPU%u: shutdown\n", cpu);
 }
 
@@ -304,7 +300,7 @@ void __noreturn arch_cpu_idle_dead(void)
 {
 	idle_task_exit();
 
-	cpu_report_death();
+	cpuhp_ap_report_dead();
 
 	while (!secondary_stack)
 		arch_cpu_idle();

[patch v3 24/36] [patch V2 24/38] MIPS: SMP_CPS: Switch to hotplug core state synchronization

[Thomas Gleixner]

From: Thomas Gleixner <tglx@linutronix.de>

Switch to the CPU hotplug core state tracking and synchronization
mechanim. This unfortunately requires to add dead reporting to the non CPS
platforms as CPS is the only user, but it allows an overall consolidation
of this functionality.

No functional change intended.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Michael Kelley <mikelley@microsoft.com>


---
 arch/mips/Kconfig               |    1 +
 arch/mips/cavium-octeon/smp.c   |    1 +
 arch/mips/include/asm/smp-ops.h |    1 +
 arch/mips/kernel/smp-bmips.c    |    1 +
 arch/mips/kernel/smp-cps.c      |   14 +++++---------
 arch/mips/kernel/smp.c          |    8 ++++++++
 arch/mips/loongson64/smp.c      |    1 +
 7 files changed, 18 insertions(+), 9 deletions(-)
---

--- a/arch/mips/Kconfig
+++ b/arch/mips/Kconfig
@@ -2285,6 +2285,7 @@ config MIPS_CPS
 	select MIPS_CM
 	select MIPS_CPS_PM if HOTPLUG_CPU
 	select SMP
+	select HOTPLUG_CORE_SYNC_DEAD if HOTPLUG_CPU
 	select SYNC_R4K if (CEVT_R4K || CSRC_R4K)
 	select SYS_SUPPORTS_HOTPLUG_CPU
 	select SYS_SUPPORTS_SCHED_SMT if CPU_MIPSR6
--- a/arch/mips/cavium-octeon/smp.c
+++ b/arch/mips/cavium-octeon/smp.c
@@ -345,6 +345,7 @@ void play_dead(void)
 	int cpu = cpu_number_map(cvmx_get_core_num());
 
 	idle_task_exit();
+	cpuhp_ap_report_dead();
 	octeon_processor_boot = 0xff;
 	per_cpu(cpu_state, cpu) = CPU_DEAD;
 
--- a/arch/mips/include/asm/smp-ops.h
+++ b/arch/mips/include/asm/smp-ops.h
@@ -33,6 +33,7 @@ struct plat_smp_ops {
 #ifdef CONFIG_HOTPLUG_CPU
 	int (*cpu_disable)(void);
 	void (*cpu_die)(unsigned int cpu);
+	void (*cleanup_dead_cpu)(unsigned cpu);
 #endif
 #ifdef CONFIG_KEXEC
 	void (*kexec_nonboot_cpu)(void);
--- a/arch/mips/kernel/smp-bmips.c
+++ b/arch/mips/kernel/smp-bmips.c
@@ -392,6 +392,7 @@ static void bmips_cpu_die(unsigned int c
 void __ref play_dead(void)
 {
 	idle_task_exit();
+	cpuhp_ap_report_dead();
 
 	/* flush data cache */
 	_dma_cache_wback_inv(0, ~0);
--- a/arch/mips/kernel/smp-cps.c
+++ b/arch/mips/kernel/smp-cps.c
@@ -503,8 +503,7 @@ void play_dead(void)
 		}
 	}
 
-	/* This CPU has chosen its way out */
-	(void)cpu_report_death();
+	cpuhp_ap_report_dead();
 
 	cps_shutdown_this_cpu(cpu_death);
 
@@ -527,7 +526,9 @@ static void wait_for_sibling_halt(void *
 	} while (!(halted & TCHALT_H));
 }
 
-static void cps_cpu_die(unsigned int cpu)
+static void cps_cpu_die(unsigned int cpu) { }
+
+static void cps_cleanup_dead_cpu(unsigned cpu)
 {
 	unsigned core = cpu_core(&cpu_data[cpu]);
 	unsigned int vpe_id = cpu_vpe_id(&cpu_data[cpu]);
@@ -535,12 +536,6 @@ static void cps_cpu_die(unsigned int cpu
 	unsigned stat;
 	int err;
 
-	/* Wait for the cpu to choose its way out */
-	if (!cpu_wait_death(cpu, 5)) {
-		pr_err("CPU%u: didn't offline\n", cpu);
-		return;
-	}
-
 	/*
 	 * Now wait for the CPU to actually offline. Without doing this that
 	 * offlining may race with one or more of:
@@ -624,6 +619,7 @@ static const struct plat_smp_ops cps_smp
 #ifdef CONFIG_HOTPLUG_CPU
 	.cpu_disable		= cps_cpu_disable,
 	.cpu_die		= cps_cpu_die,
+	.cleanup_dead_cpu	= cps_cleanup_dead_cpu,
 #endif
 #ifdef CONFIG_KEXEC
 	.kexec_nonboot_cpu	= cps_kexec_nonboot_cpu,
--- a/arch/mips/kernel/smp.c
+++ b/arch/mips/kernel/smp.c
@@ -690,6 +690,14 @@ void flush_tlb_one(unsigned long vaddr)
 EXPORT_SYMBOL(flush_tlb_page);
 EXPORT_SYMBOL(flush_tlb_one);
 
+#ifdef CONFIG_HOTPLUG_CPU
+void arch_cpuhp_cleanup_dead_cpu(unsigned int cpu)
+{
+	if (mp_ops->cleanup_dead_cpu)
+		mp_ops->cleanup_dead_cpu(cpu);
+}
+#endif
+
 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
 
 static void tick_broadcast_callee(void *info)
--- a/arch/mips/loongson64/smp.c
+++ b/arch/mips/loongson64/smp.c
@@ -775,6 +775,7 @@ void play_dead(void)
 	void (*play_dead_at_ckseg1)(int *);
 
 	idle_task_exit();
+	cpuhp_ap_report_dead();
 
 	prid_imp = read_c0_prid() & PRID_IMP_MASK;
 	prid_rev = read_c0_prid() & PRID_REV_MASK;

[patch v3 25/36] parisc: Switch to hotplug core state synchronization

[Thomas Gleixner]

From: Thomas Gleixner <tglx@linutronix.de>

Switch to the CPU hotplug core state tracking and synchronization
mechanim. No functional change intended.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Michael Kelley <mikelley@microsoft.com>


---
 arch/parisc/Kconfig          |    1 +
 arch/parisc/kernel/process.c |    4 ++--
 arch/parisc/kernel/smp.c     |    7 +++----
 3 files changed, 6 insertions(+), 6 deletions(-)
---

--- a/arch/parisc/Kconfig
+++ b/arch/parisc/Kconfig
@@ -57,6 +57,7 @@ config PARISC
 	select HAVE_ARCH_SECCOMP_FILTER
 	select HAVE_ARCH_TRACEHOOK
 	select HAVE_REGS_AND_STACK_ACCESS_API
+	select HOTPLUG_CORE_SYNC_DEAD if HOTPLUG_CPU
 	select GENERIC_SCHED_CLOCK
 	select GENERIC_IRQ_MIGRATION if SMP
 	select HAVE_UNSTABLE_SCHED_CLOCK if SMP
--- a/arch/parisc/kernel/process.c
+++ b/arch/parisc/kernel/process.c
@@ -166,8 +166,8 @@ void __noreturn arch_cpu_idle_dead(void)
 
 	local_irq_disable();
 
-	/* Tell __cpu_die() that this CPU is now safe to dispose of. */
-	(void)cpu_report_death();
+	/* Tell the core that this CPU is now safe to dispose of. */
+	cpuhp_ap_report_dead();
 
 	/* Ensure that the cache lines are written out. */
 	flush_cache_all_local();
--- a/arch/parisc/kernel/smp.c
+++ b/arch/parisc/kernel/smp.c
@@ -500,11 +500,10 @@ int __cpu_disable(void)
 void __cpu_die(unsigned int cpu)
 {
 	pdc_cpu_rendezvous_lock();
+}
 
-	if (!cpu_wait_death(cpu, 5)) {
-		pr_crit("CPU%u: cpu didn't die\n", cpu);
-		return;
-	}
+void arch_cpuhp_cleanup_dead_cpu(unsigned int cpu)
+{
 	pr_info("CPU%u: is shutting down\n", cpu);
 
 	/* set task's state to interruptible sleep */

[patch v3 26/36] riscv: Switch to hotplug core state synchronization

[Thomas Gleixner]

From: Thomas Gleixner <tglx@linutronix.de>

Switch to the CPU hotplug core state tracking and synchronization
mechanim. No functional change intended.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Palmer Dabbelt <palmer@rivosinc.com>
Tested-by: Michael Kelley <mikelley@microsoft.com>


---
 arch/riscv/Kconfig              |    1 +
 arch/riscv/include/asm/smp.h    |    2 +-
 arch/riscv/kernel/cpu-hotplug.c |   14 +++++++-------
 3 files changed, 9 insertions(+), 8 deletions(-)
---

--- a/arch/riscv/Kconfig
+++ b/arch/riscv/Kconfig
@@ -122,6 +122,7 @@ config RISCV
 	select HAVE_RSEQ
 	select HAVE_STACKPROTECTOR
 	select HAVE_SYSCALL_TRACEPOINTS
+	select HOTPLUG_CORE_SYNC_DEAD if HOTPLUG_CPU
 	select IRQ_DOMAIN
 	select IRQ_FORCED_THREADING
 	select KASAN_VMALLOC if KASAN
--- a/arch/riscv/include/asm/smp.h
+++ b/arch/riscv/include/asm/smp.h
@@ -70,7 +70,7 @@ asmlinkage void smp_callin(void);
 
 #if defined CONFIG_HOTPLUG_CPU
 int __cpu_disable(void);
-void __cpu_die(unsigned int cpu);
+static inline void __cpu_die(unsigned int cpu) { }
 #endif /* CONFIG_HOTPLUG_CPU */
 
 #else
--- a/arch/riscv/kernel/cpu-hotplug.c
+++ b/arch/riscv/kernel/cpu-hotplug.c
@@ -8,6 +8,7 @@
 #include <linux/sched.h>
 #include <linux/err.h>
 #include <linux/irq.h>
+#include <linux/cpuhotplug.h>
 #include <linux/cpu.h>
 #include <linux/sched/hotplug.h>
 #include <asm/irq.h>
@@ -49,17 +50,15 @@ int __cpu_disable(void)
 	return ret;
 }
 
+#ifdef CONFIG_HOTPLUG_CPU
 /*
- * Called on the thread which is asking for a CPU to be shutdown.
+ * Called on the thread which is asking for a CPU to be shutdown, if the
+ * CPU reported dead to the hotplug core.
  */
-void __cpu_die(unsigned int cpu)
+void arch_cpuhp_cleanup_dead_cpu(unsigned int cpu)
 {
 	int ret = 0;
 
-	if (!cpu_wait_death(cpu, 5)) {
-		pr_err("CPU %u: didn't die\n", cpu);
-		return;
-	}
 	pr_notice("CPU%u: off\n", cpu);
 
 	/* Verify from the firmware if the cpu is really stopped*/
@@ -76,9 +75,10 @@ void __noreturn arch_cpu_idle_dead(void)
 {
 	idle_task_exit();
 
-	(void)cpu_report_death();
+	cpuhp_ap_report_dead();
 
 	cpu_ops[smp_processor_id()]->cpu_stop();
 	/* It should never reach here */
 	BUG();
 }
+#endif

[patch v3 27/36] cpu/hotplug: Remove unused state functions

[Thomas Gleixner]

From: Thomas Gleixner <tglx@linutronix.de>

All users converted to the hotplug core mechanism.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Michael Kelley <mikelley@microsoft.com>


---
 include/linux/cpu.h |    2 -
 kernel/smpboot.c    |   75 ----------------------------------------------------
 2 files changed, 77 deletions(-)
---

--- a/include/linux/cpu.h
+++ b/include/linux/cpu.h
@@ -193,8 +193,6 @@ static inline void play_idle(unsigned lo
 }
 
 #ifdef CONFIG_HOTPLUG_CPU
-bool cpu_wait_death(unsigned int cpu, int seconds);
-bool cpu_report_death(void);
 void cpuhp_report_idle_dead(void);
 #else
 static inline void cpuhp_report_idle_dead(void) { }
--- a/kernel/smpboot.c
+++ b/kernel/smpboot.c
@@ -325,78 +325,3 @@ void smpboot_unregister_percpu_thread(st
 	cpus_read_unlock();
 }
 EXPORT_SYMBOL_GPL(smpboot_unregister_percpu_thread);
-
-#ifndef CONFIG_HOTPLUG_CORE_SYNC
-static DEFINE_PER_CPU(atomic_t, cpu_hotplug_state) = ATOMIC_INIT(CPU_POST_DEAD);
-
-#ifdef CONFIG_HOTPLUG_CPU
-/*
- * Wait for the specified CPU to exit the idle loop and die.
- */
-bool cpu_wait_death(unsigned int cpu, int seconds)
-{
-	int jf_left = seconds * HZ;
-	int oldstate;
-	bool ret = true;
-	int sleep_jf = 1;
-
-	might_sleep();
-
-	/* The outgoing CPU will normally get done quite quickly. */
-	if (atomic_read(&per_cpu(cpu_hotplug_state, cpu)) == CPU_DEAD)
-		goto update_state_early;
-	udelay(5);
-
-	/* But if the outgoing CPU dawdles, wait increasingly long times. */
-	while (atomic_read(&per_cpu(cpu_hotplug_state, cpu)) != CPU_DEAD) {
-		schedule_timeout_uninterruptible(sleep_jf);
-		jf_left -= sleep_jf;
-		if (jf_left <= 0)
-			break;
-		sleep_jf = DIV_ROUND_UP(sleep_jf * 11, 10);
-	}
-update_state_early:
-	oldstate = atomic_read(&per_cpu(cpu_hotplug_state, cpu));
-update_state:
-	if (oldstate == CPU_DEAD) {
-		/* Outgoing CPU died normally, update state. */
-		smp_mb(); /* atomic_read() before update. */
-		atomic_set(&per_cpu(cpu_hotplug_state, cpu), CPU_POST_DEAD);
-	} else {
-		/* Outgoing CPU still hasn't died, set state accordingly. */
-		if (!atomic_try_cmpxchg(&per_cpu(cpu_hotplug_state, cpu),
-					&oldstate, CPU_BROKEN))
-			goto update_state;
-		ret = false;
-	}
-	return ret;
-}
-
-/*
- * Called by the outgoing CPU to report its successful death.  Return
- * false if this report follows the surviving CPU's timing out.
- *
- * A separate "CPU_DEAD_FROZEN" is used when the surviving CPU
- * timed out.  This approach allows architectures to omit calls to
- * cpu_check_up_prepare() and cpu_set_state_online() without defeating
- * the next cpu_wait_death()'s polling loop.
- */
-bool cpu_report_death(void)
-{
-	int oldstate;
-	int newstate;
-	int cpu = smp_processor_id();
-
-	oldstate = atomic_read(&per_cpu(cpu_hotplug_state, cpu));
-	do {
-		if (oldstate != CPU_BROKEN)
-			newstate = CPU_DEAD;
-		else
-			newstate = CPU_DEAD_FROZEN;
-	} while (!atomic_try_cmpxchg(&per_cpu(cpu_hotplug_state, cpu),
-				     &oldstate, newstate));
-	return newstate == CPU_DEAD;
-}
-
-#endif /* #ifdef CONFIG_HOTPLUG_CPU */
-#endif /* !CONFIG_HOTPLUG_CORE_SYNC */

[patch v3 28/36] cpu/hotplug: Reset task stack state in _cpu_up()

[Thomas Gleixner]

From: David Woodhouse <dwmw@amazon.co.uk>

Commit dce1ca0525bf ("sched/scs: Reset task stack state in bringup_cpu()")
ensured that the shadow call stack and KASAN poisoning were removed from
a CPU's stack each time that CPU is brought up, not just once.

This is not incorrect. However, with parallel bringup the idle thread setup
will happen at a different step. As a consequence the cleanup in
bringup_cpu() would be too late.

Move the SCS/KASAN cleanup to the generic _cpu_up() function instead,
which already ensures that the new CPU's stack is available, purely to
allow for early failure. This occurs when the CPU to be brought up is
in the CPUHP_OFFLINE state, which should correctly do the cleanup any
time the CPU has been taken down to the point where such is needed.

Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Mark Rutland <mark.rutland@arm.com>
Reviewed-by: Mark Rutland <mark.rutland@arm.com>
Tested-by: Michael Kelley <mikelley@microsoft.com>


---
 kernel/cpu.c |   12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)
---

--- a/kernel/cpu.c
+++ b/kernel/cpu.c
@@ -771,12 +771,6 @@ static int bringup_cpu(unsigned int cpu)
 		return -EAGAIN;
 
 	/*
-	 * Reset stale stack state from the last time this CPU was online.
-	 */
-	scs_task_reset(idle);
-	kasan_unpoison_task_stack(idle);
-
-	/*
 	 * Some architectures have to walk the irq descriptors to
 	 * setup the vector space for the cpu which comes online.
 	 * Prevent irq alloc/free across the bringup.
@@ -1583,6 +1577,12 @@ static int _cpu_up(unsigned int cpu, int
 			ret = PTR_ERR(idle);
 			goto out;
 		}
+
+		/*
+		 * Reset stale stack state from the last time this CPU was online.
+		 */
+		scs_task_reset(idle);
+		kasan_unpoison_task_stack(idle);
 	}
 
 	cpuhp_tasks_frozen = tasks_frozen;

[patch v3 29/36] [patch V2 29/38] cpu/hotplug: Provide a split up CPUHP_BRINGUP mechanism

[Thomas Gleixner]

From: Thomas Gleixner <tglx@linutronix.de>

The bring up logic of a to be onlined CPU consists of several parts, which
are considered to be a single hotplug state:

  1) Control CPU issues the wake-up

  2) To be onlined CPU starts up, does the minimal initialization,
     reports to be alive and waits for release into the complete bring-up.

  3) Control CPU waits for the alive report and releases the upcoming CPU
     for the complete bring-up.

Allow to split this into two states:

  1) Control CPU issues the wake-up

     After that the to be onlined CPU starts up, does the minimal
     initialization, reports to be alive and waits for release into the
     full bring-up. As this can run after the control CPU dropped the
     hotplug locks the code which is executed on the AP before it reports
     alive has to be carefully audited to not violate any of the hotplug
     constraints, especially not modifying any of the various cpumasks.

     This is really only meant to avoid waiting for the AP to react on the
     wake-up. Of course an architecture can move strict CPU related setup
     functionality, e.g. microcode loading, with care before the
     synchronization point to save further pointless waiting time.

  2) Control CPU waits for the alive report and releases the upcoming CPU
     for the complete bring-up.

This allows that the two states can be split up to run all to be onlined
CPUs up to state #1 on the control CPU and then at a later point run state
#2. This spares some of the latencies of the full serialized per CPU
bringup by avoiding the per CPU wakeup/wait serialization. The assumption
is that the first AP already waits when the last AP has been woken up. This
obvioulsy depends on the hardware latencies and depending on the timings
this might still not completely eliminate all wait scenarios.

This split is just a preparatory step for enabling the parallel bringup
later. The boot time bringup is still fully serialized. It has a separate
config switch so that architectures which want to support parallel bringup
can test the split of the CPUHP_BRINGUG step separately.

To enable this the architecture must support the CPU hotplug core sync
mechanism and has to be audited that there are no implicit hotplug state
dependencies which require a fully serialized bringup.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Michael Kelley <mikelley@microsoft.com>


---
 arch/Kconfig               |    4 ++
 include/linux/cpuhotplug.h |    4 ++
 kernel/cpu.c               |   70 +++++++++++++++++++++++++++++++++++++++++++--
 3 files changed, 76 insertions(+), 2 deletions(-)
---

--- a/arch/Kconfig
+++ b/arch/Kconfig
@@ -49,6 +49,10 @@ config HOTPLUG_CORE_SYNC_FULL
 	select HOTPLUG_CORE_SYNC_DEAD if HOTPLUG_CPU
 	select HOTPLUG_CORE_SYNC
 
+config HOTPLUG_SPLIT_STARTUP
+	bool
+	select HOTPLUG_CORE_SYNC_FULL
+
 config GENERIC_ENTRY
 	bool
 
--- a/include/linux/cpuhotplug.h
+++ b/include/linux/cpuhotplug.h
@@ -133,6 +133,7 @@ enum cpuhp_state {
 	CPUHP_MIPS_SOC_PREPARE,
 	CPUHP_BP_PREPARE_DYN,
 	CPUHP_BP_PREPARE_DYN_END		= CPUHP_BP_PREPARE_DYN + 20,
+	CPUHP_BP_KICK_AP,
 	CPUHP_BRINGUP_CPU,
 
 	/*
@@ -517,9 +518,12 @@ void cpuhp_online_idle(enum cpuhp_state
 static inline void cpuhp_online_idle(enum cpuhp_state state) { }
 #endif
 
+struct task_struct;
+
 void cpuhp_ap_sync_alive(void);
 void arch_cpuhp_sync_state_poll(void);
 void arch_cpuhp_cleanup_kick_cpu(unsigned int cpu);
+int arch_cpuhp_kick_ap_alive(unsigned int cpu, struct task_struct *tidle);
 
 #ifdef CONFIG_HOTPLUG_CORE_SYNC_DEAD
 void cpuhp_ap_report_dead(void);
--- a/kernel/cpu.c
+++ b/kernel/cpu.c
@@ -761,6 +761,47 @@ static int bringup_wait_for_ap_online(un
 	return 0;
 }
 
+#ifdef CONFIG_HOTPLUG_SPLIT_STARTUP
+static int cpuhp_kick_ap_alive(unsigned int cpu)
+{
+	if (!cpuhp_can_boot_ap(cpu))
+		return -EAGAIN;
+
+	return arch_cpuhp_kick_ap_alive(cpu, idle_thread_get(cpu));
+}
+
+static int cpuhp_bringup_ap(unsigned int cpu)
+{
+	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
+	int ret;
+
+	/*
+	 * Some architectures have to walk the irq descriptors to
+	 * setup the vector space for the cpu which comes online.
+	 * Prevent irq alloc/free across the bringup.
+	 */
+	irq_lock_sparse();
+
+	ret = cpuhp_bp_sync_alive(cpu);
+	if (ret)
+		goto out_unlock;
+
+	ret = bringup_wait_for_ap_online(cpu);
+	if (ret)
+		goto out_unlock;
+
+	irq_unlock_sparse();
+
+	if (st->target <= CPUHP_AP_ONLINE_IDLE)
+		return 0;
+
+	return cpuhp_kick_ap(cpu, st, st->target);
+
+out_unlock:
+	irq_unlock_sparse();
+	return ret;
+}
+#else
 static int bringup_cpu(unsigned int cpu)
 {
 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
@@ -777,7 +818,6 @@ static int bringup_cpu(unsigned int cpu)
 	 */
 	irq_lock_sparse();
 
-	/* Arch-specific enabling code. */
 	ret = __cpu_up(cpu, idle);
 	if (ret)
 		goto out_unlock;
@@ -801,6 +841,7 @@ static int bringup_cpu(unsigned int cpu)
 	irq_unlock_sparse();
 	return ret;
 }
+#endif
 
 static int finish_cpu(unsigned int cpu)
 {
@@ -1940,13 +1981,38 @@ static struct cpuhp_step cpuhp_hp_states
 		.startup.single		= timers_prepare_cpu,
 		.teardown.single	= timers_dead_cpu,
 	},
-	/* Kicks the plugged cpu into life */
+
+#ifdef CONFIG_HOTPLUG_SPLIT_STARTUP
+	/*
+	 * Kicks the AP alive. AP will wait in cpuhp_ap_sync_alive() until
+	 * the next step will release it.
+	 */
+	[CPUHP_BP_KICK_AP] = {
+		.name			= "cpu:kick_ap",
+		.startup.single		= cpuhp_kick_ap_alive,
+	},
+
+	/*
+	 * Waits for the AP to reach cpuhp_ap_sync_alive() and then
+	 * releases it for the complete bringup.
+	 */
+	[CPUHP_BRINGUP_CPU] = {
+		.name			= "cpu:bringup",
+		.startup.single		= cpuhp_bringup_ap,
+		.teardown.single	= finish_cpu,
+		.cant_stop		= true,
+	},
+#else
+	/*
+	 * All-in-one CPU bringup state which includes the kick alive.
+	 */
 	[CPUHP_BRINGUP_CPU] = {
 		.name			= "cpu:bringup",
 		.startup.single		= bringup_cpu,
 		.teardown.single	= finish_cpu,
 		.cant_stop		= true,
 	},
+#endif
 	/* Final state before CPU kills itself */
 	[CPUHP_AP_IDLE_DEAD] = {
 		.name			= "idle:dead",

[patch v3 30/36] x86/smpboot: Enable split CPU startup

[Thomas Gleixner]

From: Thomas Gleixner <tglx@linutronix.de>

The x86 CPU bringup state currently does AP wake-up, wait for AP to
respond and then release it for full bringup.

It is safe to be split into a wake-up and and a separate wait+release
state.

Provide the required functions and enable the split CPU bringup, which
prepares for parallel bringup, where the bringup of the non-boot CPUs takes
two iterations: One to prepare and wake all APs and the second to wait and
release them. Depending on timing this can eliminate the wait time
completely.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Michael Kelley <mikelley@microsoft.com>


---
 arch/x86/Kconfig           |    2 +-
 arch/x86/include/asm/smp.h |    9 ++-------
 arch/x86/kernel/smp.c      |    2 +-
 arch/x86/kernel/smpboot.c  |    8 ++++----
 arch/x86/xen/smp_pv.c      |    4 ++--
 5 files changed, 10 insertions(+), 15 deletions(-)
---

--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -274,8 +274,8 @@ config X86
 	select HAVE_UNSTABLE_SCHED_CLOCK
 	select HAVE_USER_RETURN_NOTIFIER
 	select HAVE_GENERIC_VDSO
-	select HOTPLUG_CORE_SYNC_FULL		if SMP
 	select HOTPLUG_SMT			if SMP
+	select HOTPLUG_SPLIT_STARTUP		if SMP
 	select IRQ_FORCED_THREADING
 	select NEED_PER_CPU_EMBED_FIRST_CHUNK
 	select NEED_PER_CPU_PAGE_FIRST_CHUNK
--- a/arch/x86/include/asm/smp.h
+++ b/arch/x86/include/asm/smp.h
@@ -40,7 +40,7 @@ struct smp_ops {
 
 	void (*cleanup_dead_cpu)(unsigned cpu);
 	void (*poll_sync_state)(void);
-	int (*cpu_up)(unsigned cpu, struct task_struct *tidle);
+	int (*kick_ap_alive)(unsigned cpu, struct task_struct *tidle);
 	int (*cpu_disable)(void);
 	void (*cpu_die)(unsigned int cpu);
 	void (*play_dead)(void);
@@ -80,11 +80,6 @@ static inline void smp_cpus_done(unsigne
 	smp_ops.smp_cpus_done(max_cpus);
 }
 
-static inline int __cpu_up(unsigned int cpu, struct task_struct *tidle)
-{
-	return smp_ops.cpu_up(cpu, tidle);
-}
-
 static inline int __cpu_disable(void)
 {
 	return smp_ops.cpu_disable();
@@ -124,7 +119,7 @@ void native_smp_prepare_cpus(unsigned in
 void calculate_max_logical_packages(void);
 void native_smp_cpus_done(unsigned int max_cpus);
 int common_cpu_up(unsigned int cpunum, struct task_struct *tidle);
-int native_cpu_up(unsigned int cpunum, struct task_struct *tidle);
+int native_kick_ap(unsigned int cpu, struct task_struct *tidle);
 int native_cpu_disable(void);
 void __noreturn hlt_play_dead(void);
 void native_play_dead(void);
--- a/arch/x86/kernel/smp.c
+++ b/arch/x86/kernel/smp.c
@@ -268,7 +268,7 @@ struct smp_ops smp_ops = {
 #endif
 	.smp_send_reschedule	= native_smp_send_reschedule,
 
-	.cpu_up			= native_cpu_up,
+	.kick_ap_alive		= native_kick_ap,
 	.cpu_disable		= native_cpu_disable,
 	.play_dead		= native_play_dead,
 
--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -1058,7 +1058,7 @@ static int do_boot_cpu(int apicid, int c
 	return ret;
 }
 
-static int native_kick_ap(unsigned int cpu, struct task_struct *tidle)
+int native_kick_ap(unsigned int cpu, struct task_struct *tidle)
 {
 	int apicid = apic->cpu_present_to_apicid(cpu);
 	int err;
@@ -1094,15 +1094,15 @@ static int native_kick_ap(unsigned int c
 	return err;
 }
 
-int native_cpu_up(unsigned int cpu, struct task_struct *tidle)
+int arch_cpuhp_kick_ap_alive(unsigned int cpu, struct task_struct *tidle)
 {
-	return native_kick_ap(cpu, tidle);
+	return smp_ops.kick_ap_alive(cpu, tidle);
 }
 
 void arch_cpuhp_cleanup_kick_cpu(unsigned int cpu)
 {
 	/* Cleanup possible dangling ends... */
-	if (smp_ops.cpu_up == native_cpu_up && x86_platform.legacy.warm_reset)
+	if (smp_ops.kick_ap_alive == native_kick_ap && x86_platform.legacy.warm_reset)
 		smpboot_restore_warm_reset_vector();
 }
 
--- a/arch/x86/xen/smp_pv.c
+++ b/arch/x86/xen/smp_pv.c
@@ -314,7 +314,7 @@ cpu_initialize_context(unsigned int cpu,
 	return 0;
 }
 
-static int xen_pv_cpu_up(unsigned int cpu, struct task_struct *idle)
+static int xen_pv_kick_ap(unsigned int cpu, struct task_struct *idle)
 {
 	int rc;
 
@@ -438,7 +438,7 @@ static const struct smp_ops xen_smp_ops
 	.smp_prepare_cpus = xen_pv_smp_prepare_cpus,
 	.smp_cpus_done = xen_smp_cpus_done,
 
-	.cpu_up = xen_pv_cpu_up,
+	.kick_ap_alive = xen_pv_kick_ap,
 	.cpu_die = xen_pv_cpu_die,
 	.cleanup_dead_cpu = xen_pv_cleanup_dead_cpu,
 	.poll_sync_state = xen_pv_poll_sync_state,

[patch v3 31/36] x86/apic: Provide cpu_primary_thread mask

[Thomas Gleixner]

From: Thomas Gleixner <tglx@linutronix.de>

Make the primary thread tracking CPU mask based in preparation for simpler
handling of parallel bootup.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Michael Kelley <mikelley@microsoft.com>


---
 arch/x86/include/asm/apic.h     |    2 --
 arch/x86/include/asm/topology.h |   19 +++++++++++++++----
 arch/x86/kernel/apic/apic.c     |   20 +++++++++-----------
 arch/x86/kernel/smpboot.c       |   12 +++---------
 4 files changed, 27 insertions(+), 26 deletions(-)
---

--- a/arch/x86/include/asm/apic.h
+++ b/arch/x86/include/asm/apic.h
@@ -506,10 +506,8 @@ extern int default_check_phys_apicid_pre
 #endif /* CONFIG_X86_LOCAL_APIC */
 
 #ifdef CONFIG_SMP
-bool apic_id_is_primary_thread(unsigned int id);
 void apic_smt_update(void);
 #else
-static inline bool apic_id_is_primary_thread(unsigned int id) { return false; }
 static inline void apic_smt_update(void) { }
 #endif
 
--- a/arch/x86/include/asm/topology.h
+++ b/arch/x86/include/asm/topology.h
@@ -31,9 +31,9 @@
  * CONFIG_NUMA.
  */
 #include <linux/numa.h>
+#include <linux/cpumask.h>
 
 #ifdef CONFIG_NUMA
-#include <linux/cpumask.h>
 
 #include <asm/mpspec.h>
 #include <asm/percpu.h>
@@ -139,9 +139,20 @@ static inline int topology_max_smt_threa
 int topology_update_package_map(unsigned int apicid, unsigned int cpu);
 int topology_update_die_map(unsigned int dieid, unsigned int cpu);
 int topology_phys_to_logical_pkg(unsigned int pkg);
-bool topology_is_primary_thread(unsigned int cpu);
 bool topology_smt_supported(void);
-#else
+
+extern struct cpumask __cpu_primary_thread_mask;
+#define cpu_primary_thread_mask ((const struct cpumask *)&__cpu_primary_thread_mask)
+
+/**
+ * topology_is_primary_thread - Check whether CPU is the primary SMT thread
+ * @cpu:	CPU to check
+ */
+static inline bool topology_is_primary_thread(unsigned int cpu)
+{
+	return cpumask_test_cpu(cpu, cpu_primary_thread_mask);
+}
+#else /* CONFIG_SMP */
 #define topology_max_packages()			(1)
 static inline int
 topology_update_package_map(unsigned int apicid, unsigned int cpu) { return 0; }
@@ -152,7 +163,7 @@ static inline int topology_max_die_per_p
 static inline int topology_max_smt_threads(void) { return 1; }
 static inline bool topology_is_primary_thread(unsigned int cpu) { return true; }
 static inline bool topology_smt_supported(void) { return false; }
-#endif
+#endif /* !CONFIG_SMP */
 
 static inline void arch_fix_phys_package_id(int num, u32 slot)
 {
--- a/arch/x86/kernel/apic/apic.c
+++ b/arch/x86/kernel/apic/apic.c
@@ -2386,20 +2386,16 @@ bool arch_match_cpu_phys_id(int cpu, u64
 }
 
 #ifdef CONFIG_SMP
-/**
- * apic_id_is_primary_thread - Check whether APIC ID belongs to a primary thread
- * @apicid: APIC ID to check
- */
-bool apic_id_is_primary_thread(unsigned int apicid)
+static void cpu_mark_primary_thread(unsigned int cpu, unsigned int apicid)
 {
-	u32 mask;
-
-	if (smp_num_siblings == 1)
-		return true;
 	/* Isolate the SMT bit(s) in the APICID and check for 0 */
-	mask = (1U << (fls(smp_num_siblings) - 1)) - 1;
-	return !(apicid & mask);
+	u32 mask = (1U << (fls(smp_num_siblings) - 1)) - 1;
+
+	if (smp_num_siblings == 1 || !(apicid & mask))
+		cpumask_set_cpu(cpu, &__cpu_primary_thread_mask);
 }
+#else
+static inline void cpu_mark_primary_thread(unsigned int cpu, unsigned int apicid) { }
 #endif
 
 /*
@@ -2544,6 +2540,8 @@ int generic_processor_info(int apicid, i
 	set_cpu_present(cpu, true);
 	num_processors++;
 
+	cpu_mark_primary_thread(cpu, apicid);
+
 	return cpu;
 }
 
--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -102,6 +102,9 @@ EXPORT_PER_CPU_SYMBOL(cpu_die_map);
 DEFINE_PER_CPU_READ_MOSTLY(struct cpuinfo_x86, cpu_info);
 EXPORT_PER_CPU_SYMBOL(cpu_info);
 
+/* CPUs which are the primary SMT threads */
+struct cpumask __cpu_primary_thread_mask __read_mostly;
+
 /* Representing CPUs for which sibling maps can be computed */
 static cpumask_var_t cpu_sibling_setup_mask;
 
@@ -283,15 +286,6 @@ static void notrace start_secondary(void
 }
 
 /**
- * topology_is_primary_thread - Check whether CPU is the primary SMT thread
- * @cpu:	CPU to check
- */
-bool topology_is_primary_thread(unsigned int cpu)
-{
-	return apic_id_is_primary_thread(per_cpu(x86_cpu_to_apicid, cpu));
-}
-
-/**
  * topology_smt_supported - Check whether SMT is supported by the CPUs
  */
 bool topology_smt_supported(void)

Re: [patch v3 31/36] x86/apic: Provide cpu_primary_thread mask

[Kirill A. Shutemov]

On Mon, May 08, 2023 at 09:44:17PM +0200, Thomas Gleixner wrote:
> From: Thomas Gleixner <tglx@linutronix.de>
> 
> Make the primary thread tracking CPU mask based in preparation for simpler
> handling of parallel bootup.
> 
> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
> Tested-by: Michael Kelley <mikelley@microsoft.com>
> 
> 
> ---
>  arch/x86/include/asm/apic.h     |    2 --
>  arch/x86/include/asm/topology.h |   19 +++++++++++++++----
>  arch/x86/kernel/apic/apic.c     |   20 +++++++++-----------
>  arch/x86/kernel/smpboot.c       |   12 +++---------
>  4 files changed, 27 insertions(+), 26 deletions(-)
> ---
> 

...

> @@ -2386,20 +2386,16 @@ bool arch_match_cpu_phys_id(int cpu, u64
>  }
>  
>  #ifdef CONFIG_SMP
> -/**
> - * apic_id_is_primary_thread - Check whether APIC ID belongs to a primary thread
> - * @apicid: APIC ID to check
> - */
> -bool apic_id_is_primary_thread(unsigned int apicid)
> +static void cpu_mark_primary_thread(unsigned int cpu, unsigned int apicid)
>  {
> -	u32 mask;
> -
> -	if (smp_num_siblings == 1)
> -		return true;
>  	/* Isolate the SMT bit(s) in the APICID and check for 0 */
> -	mask = (1U << (fls(smp_num_siblings) - 1)) - 1;
> -	return !(apicid & mask);
> +	u32 mask = (1U << (fls(smp_num_siblings) - 1)) - 1;
> +
> +	if (smp_num_siblings == 1 || !(apicid & mask))
> +		cpumask_set_cpu(cpu, &__cpu_primary_thread_mask);
>  }
> +#else
> +static inline void cpu_mark_primary_thread(unsigned int cpu, unsigned int apicid) { }
>  #endif
>  
>  /*

This patch causes boot regression on TDX guest. The guest crashes on SMP
bring up.

The change makes use of smp_num_siblings earlier than before: the mask get
constructed in acpi_boot_init() codepath. Later on smp_num_siblings gets
updated in detect_ht().

In my setup with 16 vCPUs, smp_num_siblings is 16 before detect_ht() and
set to 1 in detect_ht().

-- 
  Kiryl Shutsemau / Kirill A. Shutemov

Re: [patch v3 31/36] x86/apic: Provide cpu_primary_thread mask

[Thomas Gleixner]

On Wed, May 24 2023 at 23:48, Kirill A. Shutemov wrote:
> On Mon, May 08, 2023 at 09:44:17PM +0200, Thomas Gleixner wrote:
>>  #ifdef CONFIG_SMP
>> -/**
>> - * apic_id_is_primary_thread - Check whether APIC ID belongs to a primary thread
>> - * @apicid: APIC ID to check
>> - */
>> -bool apic_id_is_primary_thread(unsigned int apicid)
>> +static void cpu_mark_primary_thread(unsigned int cpu, unsigned int apicid)
>>  {
>> -	u32 mask;
>> -
>> -	if (smp_num_siblings == 1)
>> -		return true;
>>  	/* Isolate the SMT bit(s) in the APICID and check for 0 */
>> -	mask = (1U << (fls(smp_num_siblings) - 1)) - 1;
>> -	return !(apicid & mask);
>> +	u32 mask = (1U << (fls(smp_num_siblings) - 1)) - 1;
>> +
>> +	if (smp_num_siblings == 1 || !(apicid & mask))
>> +		cpumask_set_cpu(cpu, &__cpu_primary_thread_mask);
>>  }
>> +#else
>> +static inline void cpu_mark_primary_thread(unsigned int cpu, unsigned int apicid) { }
>>  #endif
>>  
>>  /*
>
> This patch causes boot regression on TDX guest. The guest crashes on SMP
> bring up.

I rather call it a security feature: It makes TDX unbreakably secure.

> The change makes use of smp_num_siblings earlier than before: the mask get
> constructed in acpi_boot_init() codepath. Later on smp_num_siblings gets
> updated in detect_ht().
>
> In my setup with 16 vCPUs, smp_num_siblings is 16 before detect_ht() and
> set to 1 in detect_ht().

  early_init_intel(c)
    if (detect_extended_topology_early(c) < 0)
       detect_ht_early(c);

  acpi_boot_init()
    ....

  identify_boot_cpu(c)
    detect_ht(c);

Aaargh. That whole CPU identification code is a complete horrorshow.

I'll have a look....

Re: [patch v3 31/36] x86/apic: Provide cpu_primary_thread mask

[Thomas Gleixner]

On Fri, May 26 2023 at 12:14, Thomas Gleixner wrote:
> On Wed, May 24 2023 at 23:48, Kirill A. Shutemov wrote:
>> This patch causes boot regression on TDX guest. The guest crashes on SMP
>> bring up.

The below should fix that. Sigh...

Thanks,

        tglx
----
Subject: x86/smp: Initialize cpu_primary_thread_mask late
From: Thomas Gleixner <tglx@linutronix.de>
Date: Fri, 26 May 2023 21:38:47 +0200

Marking primary threads in the cpumask during early boot is only correct in
certain configurations, but broken e.g. for the legacy hyperthreading
detection.

This is due to the complete mess in the CPUID evaluation code which
initializes smp_num_siblings only half during early init and fixes it up
later when identify_boot_cpu() is invoked.

So using smp_num_siblings before identify_boot_cpu() leads to incorrect
results.

Fixing the early CPU init code to provide the proper data is a larger scale
surgery as the code has dependencies on data structures which are not
initialized during early boot.

Move the initialization of cpu_primary_thread_mask wich depends on
smp_num_siblings being correct to an early initcall so that it is set up
correctly before SMP bringup.

Fixes: f54d4434c281 ("x86/apic: Provide cpu_primary_thread mask")
Reported-by: "Kirill A. Shutemov" <kirill@shutemov.name>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 arch/x86/kernel/apic/apic.c |   18 +++++++++++++++++-
 1 file changed, 17 insertions(+), 1 deletion(-)

--- a/arch/x86/kernel/apic/apic.c
+++ b/arch/x86/kernel/apic/apic.c
@@ -2398,6 +2398,21 @@ static void cpu_mark_primary_thread(unsi
 	if (smp_num_siblings == 1 || !(apicid & mask))
 		cpumask_set_cpu(cpu, &__cpu_primary_thread_mask);
 }
+
+/*
+ * Due to the utter mess of CPUID evaluation smp_num_siblings is not valid
+ * during early boot. Initialize the primary thread mask before SMP
+ * bringup.
+ */
+static int __init smp_init_primary_thread_mask(void)
+{
+	unsigned int cpu;
+
+	for (cpu = 0; cpu < nr_logical_cpuids; cpu++)
+		cpu_mark_primary_thread(cpu, cpuid_to_apicid[cpu]);
+	return 0;
+}
+early_initcall(smp_init_primary_thread_mask);
 #else
 static inline void cpu_mark_primary_thread(unsigned int cpu, unsigned int apicid) { }
 #endif
@@ -2544,7 +2559,8 @@ int generic_processor_info(int apicid, i
 	set_cpu_present(cpu, true);
 	num_processors++;
 
-	cpu_mark_primary_thread(cpu, apicid);
+	if (system_state >= SYSTEM_BOOTING)
+		cpu_mark_primary_thread(cpu, apicid);
 
 	return cpu;
 }

Re: [patch v3 31/36] x86/apic: Provide cpu_primary_thread mask

[Kirill A. Shutemov]

On Sat, May 27, 2023 at 03:40:02PM +0200, Thomas Gleixner wrote:
> On Fri, May 26 2023 at 12:14, Thomas Gleixner wrote:
> > On Wed, May 24 2023 at 23:48, Kirill A. Shutemov wrote:
> >> This patch causes boot regression on TDX guest. The guest crashes on SMP
> >> bring up.
> 
> The below should fix that. Sigh...

Okay, this gets me fixes the boot for TDX guest:

Tested-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>

But it gets broken again on "x86/smpboot: Implement a bit spinlock to
protect the realmode stack" with

[    0.554079] .... node  #0, CPUs:        #1  #2
[    0.738071] Callback from call_rcu_tasks() invoked.
[   10.562065] CPU2 failed to report alive state
[   10.566337]   #3
[   20.570066] CPU3 failed to report alive state
[   20.574268]   #4
...

Notably CPU1 is missing from "failed to report" list. So CPU1 takes the
lock fine, but seems never unlocks it.

Maybe trampoline_lock(%rip) in head_64.S somehow is not the same as
&tr_lock in trampoline_64.S. I donno.

I haven't find the root cause yet. But bypassing locking in
LOAD_REALMODE_ESP makes the issue go away.

I will look more into it.

-- 
  Kiryl Shutsemau / Kirill A. Shutemov

Re: [patch v3 31/36] x86/apic: Provide cpu_primary_thread mask

[Thomas Gleixner]

On Mon, May 29 2023 at 05:39, Kirill A. Shutemov wrote:
> On Sat, May 27, 2023 at 03:40:02PM +0200, Thomas Gleixner wrote:
> But it gets broken again on "x86/smpboot: Implement a bit spinlock to
> protect the realmode stack" with
>
> [    0.554079] .... node  #0, CPUs:        #1  #2
> [    0.738071] Callback from call_rcu_tasks() invoked.
> [   10.562065] CPU2 failed to report alive state
> [   10.566337]   #3
> [   20.570066] CPU3 failed to report alive state
> [   20.574268]   #4
> ...
>
> Notably CPU1 is missing from "failed to report" list. So CPU1 takes the
> lock fine, but seems never unlocks it.
>
> Maybe trampoline_lock(%rip) in head_64.S somehow is not the same as
> &tr_lock in trampoline_64.S. I donno.

It's definitely the same in the regular startup (16bit mode), but TDX
starts up via:

trampoline_start64
  trampoline_compat
    LOAD_REALMODE_ESP <- lock

That place cannot work with that LOAD_REALMODE_ESP macro. The untested
below should cure it.

Thanks,

        tglx
---
--- a/arch/x86/realmode/rm/trampoline_64.S
+++ b/arch/x86/realmode/rm/trampoline_64.S
@@ -37,12 +37,12 @@
 	.text
 	.code16
 
-.macro LOAD_REALMODE_ESP
+.macro LOAD_REALMODE_ESP lock:req
 	/*
 	 * Make sure only one CPU fiddles with the realmode stack
 	 */
 .Llock_rm\@:
-        lock btsl       $0, tr_lock
+        lock btsl       $0, \lock
         jnc             2f
         pause
         jmp             .Llock_rm\@
@@ -63,7 +63,7 @@ SYM_CODE_START(trampoline_start)
 	mov	%ax, %es
 	mov	%ax, %ss
 
-	LOAD_REALMODE_ESP
+	LOAD_REALMODE_ESP tr_lock
 
 	call	verify_cpu		# Verify the cpu supports long mode
 	testl   %eax, %eax		# Check for return code
@@ -106,7 +106,7 @@ SYM_CODE_START(sev_es_trampoline_start)
 	mov	%ax, %es
 	mov	%ax, %ss
 
-	LOAD_REALMODE_ESP
+	LOAD_REALMODE_ESP tr_lock
 
 	jmp	.Lswitch_to_protected
 SYM_CODE_END(sev_es_trampoline_start)
@@ -189,7 +189,7 @@ SYM_CODE_START(pa_trampoline_compat)
 	 * In compatibility mode.  Prep ESP and DX for startup_32, then disable
 	 * paging and complete the switch to legacy 32-bit mode.
 	 */
-	LOAD_REALMODE_ESP
+	LOAD_REALMODE_ESP pa_tr_lock
 	movw	$__KERNEL_DS, %dx
 
 	movl	$(CR0_STATE & ~X86_CR0_PG), %eax

Re: [patch v3 31/36] x86/apic: Provide cpu_primary_thread mask

[Kirill A. Shutemov]

On Mon, May 29, 2023 at 09:27:13PM +0200, Thomas Gleixner wrote:
> On Mon, May 29 2023 at 05:39, Kirill A. Shutemov wrote:
> > On Sat, May 27, 2023 at 03:40:02PM +0200, Thomas Gleixner wrote:
> > But it gets broken again on "x86/smpboot: Implement a bit spinlock to
> > protect the realmode stack" with
> >
> > [    0.554079] .... node  #0, CPUs:        #1  #2
> > [    0.738071] Callback from call_rcu_tasks() invoked.
> > [   10.562065] CPU2 failed to report alive state
> > [   10.566337]   #3
> > [   20.570066] CPU3 failed to report alive state
> > [   20.574268]   #4
> > ...
> >
> > Notably CPU1 is missing from "failed to report" list. So CPU1 takes the
> > lock fine, but seems never unlocks it.
> >
> > Maybe trampoline_lock(%rip) in head_64.S somehow is not the same as
> > &tr_lock in trampoline_64.S. I donno.
> 
> It's definitely the same in the regular startup (16bit mode), but TDX
> starts up via:
> 
> trampoline_start64
>   trampoline_compat
>     LOAD_REALMODE_ESP <- lock
> 
> That place cannot work with that LOAD_REALMODE_ESP macro. The untested
> below should cure it.

Yep, works for me.

Aaand the next patch that breaks TDX boot is... <drum roll>

	x86/smpboot/64: Implement arch_cpuhp_init_parallel_bringup() and enable it

Disabling parallel bringup helps. I didn't look closer yet. If you have
an idea let me know.

-- 
  Kiryl Shutsemau / Kirill A. Shutemov

Re: [patch v3 31/36] x86/apic: Provide cpu_primary_thread mask

[Kirill A. Shutemov]

On Mon, May 29, 2023 at 11:31:29PM +0300, Kirill A. Shutemov wrote:
> On Mon, May 29, 2023 at 09:27:13PM +0200, Thomas Gleixner wrote:
> > On Mon, May 29 2023 at 05:39, Kirill A. Shutemov wrote:
> > > On Sat, May 27, 2023 at 03:40:02PM +0200, Thomas Gleixner wrote:
> > > But it gets broken again on "x86/smpboot: Implement a bit spinlock to
> > > protect the realmode stack" with
> > >
> > > [    0.554079] .... node  #0, CPUs:        #1  #2
> > > [    0.738071] Callback from call_rcu_tasks() invoked.
> > > [   10.562065] CPU2 failed to report alive state
> > > [   10.566337]   #3
> > > [   20.570066] CPU3 failed to report alive state
> > > [   20.574268]   #4
> > > ...
> > >
> > > Notably CPU1 is missing from "failed to report" list. So CPU1 takes the
> > > lock fine, but seems never unlocks it.
> > >
> > > Maybe trampoline_lock(%rip) in head_64.S somehow is not the same as
> > > &tr_lock in trampoline_64.S. I donno.
> > 
> > It's definitely the same in the regular startup (16bit mode), but TDX
> > starts up via:
> > 
> > trampoline_start64
> >   trampoline_compat
> >     LOAD_REALMODE_ESP <- lock
> > 
> > That place cannot work with that LOAD_REALMODE_ESP macro. The untested
> > below should cure it.
> 
> Yep, works for me.
> 
> Aaand the next patch that breaks TDX boot is... <drum roll>
> 
> 	x86/smpboot/64: Implement arch_cpuhp_init_parallel_bringup() and enable it
> 
> Disabling parallel bringup helps. I didn't look closer yet. If you have
> an idea let me know.

Okay, it crashes around .Lread_apicid due to touching MSRs that trigger #VE.

Looks like the patch had no intention to enable parallel bringup on TDX.

+        * Intel-TDX has a secure RDMSR hypercall, but that needs to be
+        * implemented seperately in the low level startup ASM code.

But CC_ATTR_GUEST_STATE_ENCRYPT that used to filter it out is
SEV-ES-specific thingy and doesn't cover TDX. I don't think we have an
attribute that fits nicely here.

-- 
  Kiryl Shutsemau / Kirill A. Shutemov

Re: [patch v3 31/36] x86/apic: Provide cpu_primary_thread mask

[Thomas Gleixner]

On Tue, May 30 2023 at 03:54, Kirill A. Shutemov wrote:
> On Mon, May 29, 2023 at 11:31:29PM +0300, Kirill A. Shutemov wrote:
>> Disabling parallel bringup helps. I didn't look closer yet. If you have
>> an idea let me know.
>
> Okay, it crashes around .Lread_apicid due to touching MSRs that trigger #VE.
>
> Looks like the patch had no intention to enable parallel bringup on TDX.
>
> +        * Intel-TDX has a secure RDMSR hypercall, but that needs to be
> +        * implemented seperately in the low level startup ASM code.
>
> But CC_ATTR_GUEST_STATE_ENCRYPT that used to filter it out is
> SEV-ES-specific thingy and doesn't cover TDX. I don't think we have an
> attribute that fits nicely here.

Bah. That sucks.

Re: [patch v3 31/36] x86/apic: Provide cpu_primary_thread mask

[Thomas Gleixner]

On Tue, May 30 2023 at 11:26, Thomas Gleixner wrote:
> On Tue, May 30 2023 at 03:54, Kirill A. Shutemov wrote:
>> On Mon, May 29, 2023 at 11:31:29PM +0300, Kirill A. Shutemov wrote:
>>> Disabling parallel bringup helps. I didn't look closer yet. If you have
>>> an idea let me know.
>>
>> Okay, it crashes around .Lread_apicid due to touching MSRs that trigger #VE.
>>
>> Looks like the patch had no intention to enable parallel bringup on TDX.
>>
>> +        * Intel-TDX has a secure RDMSR hypercall, but that needs to be
>> +        * implemented seperately in the low level startup ASM code.
>>
>> But CC_ATTR_GUEST_STATE_ENCRYPT that used to filter it out is
>> SEV-ES-specific thingy and doesn't cover TDX. I don't think we have an
>> attribute that fits nicely here.
>
> Bah. That sucks.

Can we have something consistent in this CC space or needs everything to
be extra magic per CC variant?

Re: [patch v3 31/36] x86/apic: Provide cpu_primary_thread mask

[Kirill A. Shutemov]

On Tue, May 30, 2023 at 12:34:45PM +0200, Thomas Gleixner wrote:
> On Tue, May 30 2023 at 11:26, Thomas Gleixner wrote:
> > On Tue, May 30 2023 at 03:54, Kirill A. Shutemov wrote:
> >> On Mon, May 29, 2023 at 11:31:29PM +0300, Kirill A. Shutemov wrote:
> >>> Disabling parallel bringup helps. I didn't look closer yet. If you have
> >>> an idea let me know.
> >>
> >> Okay, it crashes around .Lread_apicid due to touching MSRs that trigger #VE.
> >>
> >> Looks like the patch had no intention to enable parallel bringup on TDX.
> >>
> >> +        * Intel-TDX has a secure RDMSR hypercall, but that needs to be
> >> +        * implemented seperately in the low level startup ASM code.
> >>
> >> But CC_ATTR_GUEST_STATE_ENCRYPT that used to filter it out is
> >> SEV-ES-specific thingy and doesn't cover TDX. I don't think we have an
> >> attribute that fits nicely here.
> >
> > Bah. That sucks.
> 
> Can we have something consistent in this CC space or needs everything to
> be extra magic per CC variant?

IIUC, CC_ATTR_GUEST_MEM_ENCRYPT should cover all AMD SEV flavours and
Intel TDX. But the name is confusing in this context: memory encryption
has nothing to do with the APIC.

-- 
  Kiryl Shutsemau / Kirill A. Shutemov

[patch] x86/smpboot: Disable parallel bootup if cc_vendor != NONE

[Thomas Gleixner]

The decision to allow parallel bringup of secondary CPUs checks
CC_ATTR_GUEST_STATE_ENCRYPT to detect encrypted guests. Those cannot use
parallel bootup because accessing the local APIC is intercepted and raises
a #VC or #VE, which cannot be handled at that point.

The check works correctly, but only for AMD encrypted guests. TDX does not
set that flag.

Check for cc_vendor != CC_VENDOR_NONE instead. That might be overbroad, but
definitely works for both AMD and Intel.

Fixes: 0c7ffa32dbd6 ("x86/smpboot/64: Implement arch_cpuhp_init_parallel_bringup() and enable it")
Reported-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 arch/x86/kernel/smpboot.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -1282,7 +1282,7 @@ bool __init arch_cpuhp_init_parallel_bri
 	 * Intel-TDX has a secure RDMSR hypercall, but that needs to be
 	 * implemented seperately in the low level startup ASM code.
 	 */
-	if (cc_platform_has(CC_ATTR_GUEST_STATE_ENCRYPT)) {
+	if (cc_get_vendor() != CC_VENDOR_NONE) {
 		pr_info("Parallel CPU startup disabled due to guest state encryption\n");
 		return false;
 	}

Re: [patch] x86/smpboot: Disable parallel bootup if cc_vendor != NONE

[Kirill A. Shutemov]

On Tue, May 30, 2023 at 02:09:17PM +0200, Thomas Gleixner wrote:
> The decision to allow parallel bringup of secondary CPUs checks
> CC_ATTR_GUEST_STATE_ENCRYPT to detect encrypted guests. Those cannot use
> parallel bootup because accessing the local APIC is intercepted and raises
> a #VC or #VE, which cannot be handled at that point.
> 
> The check works correctly, but only for AMD encrypted guests. TDX does not
> set that flag.
> 
> Check for cc_vendor != CC_VENDOR_NONE instead. That might be overbroad, but
> definitely works for both AMD and Intel.

It boots fine with TDX, but I think it is wrong. cc_get_vendor() will
report CC_VENDOR_AMD even on bare metal if SME is enabled. I don't think
we want it.

-- 
  Kiryl Shutsemau / Kirill A. Shutemov

Re: [patch] x86/smpboot: Disable parallel bootup if cc_vendor != NONE

[Thomas Gleixner]

On Tue, May 30 2023 at 15:29, Kirill A. Shutemov wrote:
> On Tue, May 30, 2023 at 02:09:17PM +0200, Thomas Gleixner wrote:
>> The decision to allow parallel bringup of secondary CPUs checks
>> CC_ATTR_GUEST_STATE_ENCRYPT to detect encrypted guests. Those cannot use
>> parallel bootup because accessing the local APIC is intercepted and raises
>> a #VC or #VE, which cannot be handled at that point.
>> 
>> The check works correctly, but only for AMD encrypted guests. TDX does not
>> set that flag.
>> 
>> Check for cc_vendor != CC_VENDOR_NONE instead. That might be overbroad, but
>> definitely works for both AMD and Intel.
>
> It boots fine with TDX, but I think it is wrong. cc_get_vendor() will
> report CC_VENDOR_AMD even on bare metal if SME is enabled. I don't think
> we want it.

Right. Did not think about that.

But the same way is CC_ATTR_GUEST_MEM_ENCRYPT overbroad for AMD. Only
SEV-ES traps RDMSR if I'm understandig that maze correctly.

Thanks,

        tglx

Re: [patch] x86/smpboot: Disable parallel bootup if cc_vendor != NONE

[Sean Christopherson]

On Tue, May 30, 2023, Thomas Gleixner wrote:
> On Tue, May 30 2023 at 15:29, Kirill A. Shutemov wrote:
> > On Tue, May 30, 2023 at 02:09:17PM +0200, Thomas Gleixner wrote:
> >> The decision to allow parallel bringup of secondary CPUs checks
> >> CC_ATTR_GUEST_STATE_ENCRYPT to detect encrypted guests. Those cannot use
> >> parallel bootup because accessing the local APIC is intercepted and raises
> >> a #VC or #VE, which cannot be handled at that point.
> >> 
> >> The check works correctly, but only for AMD encrypted guests. TDX does not
> >> set that flag.
> >> 
> >> Check for cc_vendor != CC_VENDOR_NONE instead. That might be overbroad, but
> >> definitely works for both AMD and Intel.
> >
> > It boots fine with TDX, but I think it is wrong. cc_get_vendor() will
> > report CC_VENDOR_AMD even on bare metal if SME is enabled. I don't think
> > we want it.
> 
> Right. Did not think about that.
> 
> But the same way is CC_ATTR_GUEST_MEM_ENCRYPT overbroad for AMD. Only
> SEV-ES traps RDMSR if I'm understandig that maze correctly.

Ya, regular SEV doesn't encrypt register state.

Re: [patch] x86/smpboot: Disable parallel bootup if cc_vendor != NONE

[Thomas Gleixner]

On Tue, May 30 2023 at 09:56, Sean Christopherson wrote:
> On Tue, May 30, 2023, Thomas Gleixner wrote:
>> On Tue, May 30 2023 at 15:29, Kirill A. Shutemov wrote:
>> > On Tue, May 30, 2023 at 02:09:17PM +0200, Thomas Gleixner wrote:
>> >> The decision to allow parallel bringup of secondary CPUs checks
>> >> CC_ATTR_GUEST_STATE_ENCRYPT to detect encrypted guests. Those cannot use
>> >> parallel bootup because accessing the local APIC is intercepted and raises
>> >> a #VC or #VE, which cannot be handled at that point.
>> >> 
>> >> The check works correctly, but only for AMD encrypted guests. TDX does not
>> >> set that flag.
>> >> 
>> >> Check for cc_vendor != CC_VENDOR_NONE instead. That might be overbroad, but
>> >> definitely works for both AMD and Intel.
>> >
>> > It boots fine with TDX, but I think it is wrong. cc_get_vendor() will
>> > report CC_VENDOR_AMD even on bare metal if SME is enabled. I don't think
>> > we want it.
>> 
>> Right. Did not think about that.
>> 
>> But the same way is CC_ATTR_GUEST_MEM_ENCRYPT overbroad for AMD. Only
>> SEV-ES traps RDMSR if I'm understandig that maze correctly.
>
> Ya, regular SEV doesn't encrypt register state.

That aside. From a semantical POV making this decision about parallel
bootup based on some magic CC encryption attribute is questionable.

I'm tending to just do the below and make this CC agnostic (except that
I couldn't find the right spot for SEV-ES to clear that flag.)

Thanks,

        tglx
---
--- a/arch/x86/coco/tdx/tdx.c
+++ b/arch/x86/coco/tdx/tdx.c
@@ -871,5 +871,7 @@ void __init tdx_early_init(void)
 	x86_platform.guest.enc_tlb_flush_required   = tdx_tlb_flush_required;
 	x86_platform.guest.enc_status_change_finish = tdx_enc_status_changed;
 
+	x86_cpuinit.parallel_bringup = false;
+
 	pr_info("Guest detected\n");
 }
--- a/arch/x86/include/asm/x86_init.h
+++ b/arch/x86/include/asm/x86_init.h
@@ -2,6 +2,7 @@
 #ifndef _ASM_X86_PLATFORM_H
 #define _ASM_X86_PLATFORM_H
 
+#include <linux/bits.h>
 #include <asm/bootparam.h>
 
 struct ghcb;
@@ -177,11 +178,14 @@ struct x86_init_ops {
  * struct x86_cpuinit_ops - platform specific cpu hotplug setups
  * @setup_percpu_clockev:	set up the per cpu clock event device
  * @early_percpu_clock_init:	early init of the per cpu clock event device
+ * @fixup_cpu_id:		fixup function for cpuinfo_x86::phys_proc_id
+ * @parallel_bringup:		Parallel bringup control
  */
 struct x86_cpuinit_ops {
 	void (*setup_percpu_clockev)(void);
 	void (*early_percpu_clock_init)(void);
 	void (*fixup_cpu_id)(struct cpuinfo_x86 *c, int node);
+	bool parallel_bringup;
 };
 
 struct timespec64;
--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -1287,6 +1287,11 @@ bool __init arch_cpuhp_init_parallel_bri
 		return false;
 	}
 
+	if (!x86_cpuinit.parallel_bringup) {
+		pr_info("Parallel CPU startup disabled by the platform\n");
+		return false;
+	}
+
 	smpboot_control = STARTUP_READ_APICID;
 	pr_debug("Parallel CPU startup enabled: 0x%08x\n", smpboot_control);
 	return true;
--- a/arch/x86/kernel/x86_init.c
+++ b/arch/x86/kernel/x86_init.c
@@ -126,6 +126,7 @@ struct x86_init_ops x86_init __initdata
 struct x86_cpuinit_ops x86_cpuinit = {
 	.early_percpu_clock_init	= x86_init_noop,
 	.setup_percpu_clockev		= setup_secondary_APIC_clock,
+	.parallel_bringup		= true,
 };
 
 static void default_nmi_init(void) { };

Re: [patch] x86/smpboot: Disable parallel bootup if cc_vendor != NONE

[Tom Lendacky]

On 5/30/23 14:51, Thomas Gleixner wrote:
> On Tue, May 30 2023 at 09:56, Sean Christopherson wrote:
>> On Tue, May 30, 2023, Thomas Gleixner wrote:
>>> On Tue, May 30 2023 at 15:29, Kirill A. Shutemov wrote:
>>>> On Tue, May 30, 2023 at 02:09:17PM +0200, Thomas Gleixner wrote:
>>>>> The decision to allow parallel bringup of secondary CPUs checks
>>>>> CC_ATTR_GUEST_STATE_ENCRYPT to detect encrypted guests. Those cannot use
>>>>> parallel bootup because accessing the local APIC is intercepted and raises
>>>>> a #VC or #VE, which cannot be handled at that point.
>>>>>
>>>>> The check works correctly, but only for AMD encrypted guests. TDX does not
>>>>> set that flag.
>>>>>
>>>>> Check for cc_vendor != CC_VENDOR_NONE instead. That might be overbroad, but
>>>>> definitely works for both AMD and Intel.
>>>>
>>>> It boots fine with TDX, but I think it is wrong. cc_get_vendor() will
>>>> report CC_VENDOR_AMD even on bare metal if SME is enabled. I don't think
>>>> we want it.
>>>
>>> Right. Did not think about that.
>>>
>>> But the same way is CC_ATTR_GUEST_MEM_ENCRYPT overbroad for AMD. Only
>>> SEV-ES traps RDMSR if I'm understandig that maze correctly.
>>
>> Ya, regular SEV doesn't encrypt register state.
> 
> That aside. From a semantical POV making this decision about parallel
> bootup based on some magic CC encryption attribute is questionable.
> 
> I'm tending to just do the below and make this CC agnostic (except that
> I couldn't find the right spot for SEV-ES to clear that flag.)

Maybe in sme_sev_setup_real_mode() in arch/x86/realmode/init.c? You could 
clear the flag within the CC_ATTR_GUEST_STATE_ENCRYPT check.

Thanks,
Tom

> 
> Thanks,
> 
>          tglx
> ---
> --- a/arch/x86/coco/tdx/tdx.c
> +++ b/arch/x86/coco/tdx/tdx.c
> @@ -871,5 +871,7 @@ void __init tdx_early_init(void)
>   	x86_platform.guest.enc_tlb_flush_required   = tdx_tlb_flush_required;
>   	x86_platform.guest.enc_status_change_finish = tdx_enc_status_changed;
>   
> +	x86_cpuinit.parallel_bringup = false;
> +
>   	pr_info("Guest detected\n");
>   }
> --- a/arch/x86/include/asm/x86_init.h
> +++ b/arch/x86/include/asm/x86_init.h
> @@ -2,6 +2,7 @@
>   #ifndef _ASM_X86_PLATFORM_H
>   #define _ASM_X86_PLATFORM_H
>   
> +#include <linux/bits.h>
>   #include <asm/bootparam.h>
>   
>   struct ghcb;
> @@ -177,11 +178,14 @@ struct x86_init_ops {
>    * struct x86_cpuinit_ops - platform specific cpu hotplug setups
>    * @setup_percpu_clockev:	set up the per cpu clock event device
>    * @early_percpu_clock_init:	early init of the per cpu clock event device
> + * @fixup_cpu_id:		fixup function for cpuinfo_x86::phys_proc_id
> + * @parallel_bringup:		Parallel bringup control
>    */
>   struct x86_cpuinit_ops {
>   	void (*setup_percpu_clockev)(void);
>   	void (*early_percpu_clock_init)(void);
>   	void (*fixup_cpu_id)(struct cpuinfo_x86 *c, int node);
> +	bool parallel_bringup;
>   };
>   
>   struct timespec64;
> --- a/arch/x86/kernel/smpboot.c
> +++ b/arch/x86/kernel/smpboot.c
> @@ -1287,6 +1287,11 @@ bool __init arch_cpuhp_init_parallel_bri
>   		return false;
>   	}
>   
> +	if (!x86_cpuinit.parallel_bringup) {
> +		pr_info("Parallel CPU startup disabled by the platform\n");
> +		return false;
> +	}
> +
>   	smpboot_control = STARTUP_READ_APICID;
>   	pr_debug("Parallel CPU startup enabled: 0x%08x\n", smpboot_control);
>   	return true;
> --- a/arch/x86/kernel/x86_init.c
> +++ b/arch/x86/kernel/x86_init.c
> @@ -126,6 +126,7 @@ struct x86_init_ops x86_init __initdata
>   struct x86_cpuinit_ops x86_cpuinit = {
>   	.early_percpu_clock_init	= x86_init_noop,
>   	.setup_percpu_clockev		= setup_secondary_APIC_clock,
> +	.parallel_bringup		= true,
>   };
>   
>   static void default_nmi_init(void) { };

Re: [patch] x86/smpboot: Disable parallel bootup if cc_vendor != NONE

[Thomas Gleixner]

On Tue, May 30 2023 at 15:03, Tom Lendacky wrote:
> On 5/30/23 14:51, Thomas Gleixner wrote:
>> That aside. From a semantical POV making this decision about parallel
>> bootup based on some magic CC encryption attribute is questionable.
>> 
>> I'm tending to just do the below and make this CC agnostic (except that
>> I couldn't find the right spot for SEV-ES to clear that flag.)
>
> Maybe in sme_sev_setup_real_mode() in arch/x86/realmode/init.c? You could 
> clear the flag within the CC_ATTR_GUEST_STATE_ENCRYPT check.

Eeew.

Can we please have a AMD SEV-ES init specific place and not hijack some
random code which has to check CC_ATTR_GUEST_STATE_ENCRYPT?

Thanks,

        tglx

Re: [patch] x86/smpboot: Disable parallel bootup if cc_vendor != NONE

[Tom Lendacky]

On 5/30/23 15:39, Thomas Gleixner wrote:
> On Tue, May 30 2023 at 15:03, Tom Lendacky wrote:
>> On 5/30/23 14:51, Thomas Gleixner wrote:
>>> That aside. From a semantical POV making this decision about parallel
>>> bootup based on some magic CC encryption attribute is questionable.
>>>
>>> I'm tending to just do the below and make this CC agnostic (except that
>>> I couldn't find the right spot for SEV-ES to clear that flag.)
>>
>> Maybe in sme_sev_setup_real_mode() in arch/x86/realmode/init.c? You could
>> clear the flag within the CC_ATTR_GUEST_STATE_ENCRYPT check.
> 
> Eeew.
> 
> Can we please have a AMD SEV-ES init specific place and not hijack some
> random code which has to check CC_ATTR_GUEST_STATE_ENCRYPT?

As long as it's not too early, you could try sme_early_init() in 
arch/x86/mm/mem_encrypt_amd.c. Add a check for sev_status & 
MSR_AMD64_SEV_ES_ENABLED and clear the flag.

Thanks,
Tom

> 
> Thanks,
> 
>          tglx

[patch] x86/smpboot: Fix the parallel bringup decision

[Thomas Gleixner]

The decision to allow parallel bringup of secondary CPUs checks
CC_ATTR_GUEST_STATE_ENCRYPT to detect encrypted guests. Those cannot use
parallel bootup because accessing the local APIC is intercepted and raises
a #VC or #VE, which cannot be handled at that point.

The check works correctly, but only for AMD encrypted guests. TDX does not
set that flag.

As there is no real connection between CC attributes and the inability to
support parallel bringup, replace this with a generic control flag in
x86_cpuinit and let SEV-ES and TDX init code disable it.

Fixes: 0c7ffa32dbd6 ("x86/smpboot/64: Implement arch_cpuhp_init_parallel_bringup() and enable it")
Reported-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 arch/x86/coco/tdx/tdx.c         |   11 +++++++++++
 arch/x86/include/asm/x86_init.h |    3 +++
 arch/x86/kernel/smpboot.c       |   19 ++-----------------
 arch/x86/kernel/x86_init.c      |    1 +
 arch/x86/mm/mem_encrypt_amd.c   |   15 +++++++++++++++
 5 files changed, 32 insertions(+), 17 deletions(-)

--- a/arch/x86/coco/tdx/tdx.c
+++ b/arch/x86/coco/tdx/tdx.c
@@ -871,5 +871,16 @@ void __init tdx_early_init(void)
 	x86_platform.guest.enc_tlb_flush_required   = tdx_tlb_flush_required;
 	x86_platform.guest.enc_status_change_finish = tdx_enc_status_changed;
 
+	/*
+	 * TDX intercepts the RDMSR to read the X2APIC ID in the parallel
+	 * bringup low level code. That raises #VE which cannot be handled
+	 * there.
+	 *
+	 * Intel-TDX has a secure RDMSR hypercall, but that needs to be
+	 * implemented seperately in the low level startup ASM code.
+	 * Until that is in place, disable parallel bringup for TDX.
+	 */
+	x86_cpuinit.parallel_bringup = false;
+
 	pr_info("Guest detected\n");
 }
--- a/arch/x86/include/asm/x86_init.h
+++ b/arch/x86/include/asm/x86_init.h
@@ -177,11 +177,14 @@ struct x86_init_ops {
  * struct x86_cpuinit_ops - platform specific cpu hotplug setups
  * @setup_percpu_clockev:	set up the per cpu clock event device
  * @early_percpu_clock_init:	early init of the per cpu clock event device
+ * @fixup_cpu_id:		fixup function for cpuinfo_x86::phys_proc_id
+ * @parallel_bringup:		Parallel bringup control
  */
 struct x86_cpuinit_ops {
 	void (*setup_percpu_clockev)(void);
 	void (*early_percpu_clock_init)(void);
 	void (*fixup_cpu_id)(struct cpuinfo_x86 *c, int node);
+	bool parallel_bringup;
 };
 
 struct timespec64;
--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -1267,23 +1267,8 @@ void __init smp_prepare_cpus_common(void
 /* Establish whether parallel bringup can be supported. */
 bool __init arch_cpuhp_init_parallel_bringup(void)
 {
-	/*
-	 * Encrypted guests require special handling. They enforce X2APIC
-	 * mode but the RDMSR to read the APIC ID is intercepted and raises
-	 * #VC or #VE which cannot be handled in the early startup code.
-	 *
-	 * AMD-SEV does not provide a RDMSR GHCB protocol so the early
-	 * startup code cannot directly communicate with the secure
-	 * firmware. The alternative solution to retrieve the APIC ID via
-	 * CPUID(0xb), which is covered by the GHCB protocol, is not viable
-	 * either because there is no enforcement of the CPUID(0xb)
-	 * provided "initial" APIC ID to be the same as the real APIC ID.
-	 *
-	 * Intel-TDX has a secure RDMSR hypercall, but that needs to be
-	 * implemented seperately in the low level startup ASM code.
-	 */
-	if (cc_platform_has(CC_ATTR_GUEST_STATE_ENCRYPT)) {
-		pr_info("Parallel CPU startup disabled due to guest state encryption\n");
+	if (!x86_cpuinit.parallel_bringup) {
+		pr_info("Parallel CPU startup disabled by the platform\n");
 		return false;
 	}
 
--- a/arch/x86/kernel/x86_init.c
+++ b/arch/x86/kernel/x86_init.c
@@ -126,6 +126,7 @@ struct x86_init_ops x86_init __initdata
 struct x86_cpuinit_ops x86_cpuinit = {
 	.early_percpu_clock_init	= x86_init_noop,
 	.setup_percpu_clockev		= setup_secondary_APIC_clock,
+	.parallel_bringup		= true,
 };
 
 static void default_nmi_init(void) { };
--- a/arch/x86/mm/mem_encrypt_amd.c
+++ b/arch/x86/mm/mem_encrypt_amd.c
@@ -501,6 +501,21 @@ void __init sme_early_init(void)
 	x86_platform.guest.enc_status_change_finish  = amd_enc_status_change_finish;
 	x86_platform.guest.enc_tlb_flush_required    = amd_enc_tlb_flush_required;
 	x86_platform.guest.enc_cache_flush_required  = amd_enc_cache_flush_required;
+
+	/*
+	 * AMD-SEV-ES intercepts the RDMSR to read the X2APIC ID in the
+	 * parallel bringup low level code. That raises #VC which cannot be
+	 * handled there.
+	 * It does not provide a RDMSR GHCB protocol so the early startup
+	 * code cannot directly communicate with the secure firmware. The
+	 * alternative solution to retrieve the APIC ID via CPUID(0xb),
+	 * which is covered by the GHCB protocol, is not viable either
+	 * because there is no enforcement of the CPUID(0xb) provided
+	 * "initial" APIC ID to be the same as the real APIC ID.
+	 * Disable parallel bootup.
+	 */
+	if (sev_status & MSR_AMD64_SEV_ES_ENABLED)
+		x86_cpuinit.parallel_bringup = false;
 }
 
 void __init mem_encrypt_free_decrypted_mem(void)

Re: [patch] x86/smpboot: Fix the parallel bringup decision

[Kirill A. Shutemov]

On Wed, May 31, 2023 at 09:44:26AM +0200, Thomas Gleixner wrote:
> The decision to allow parallel bringup of secondary CPUs checks
> CC_ATTR_GUEST_STATE_ENCRYPT to detect encrypted guests. Those cannot use
> parallel bootup because accessing the local APIC is intercepted and raises
> a #VC or #VE, which cannot be handled at that point.
> 
> The check works correctly, but only for AMD encrypted guests. TDX does not
> set that flag.
> 
> As there is no real connection between CC attributes and the inability to
> support parallel bringup, replace this with a generic control flag in
> x86_cpuinit and let SEV-ES and TDX init code disable it.
> 
> Fixes: 0c7ffa32dbd6 ("x86/smpboot/64: Implement arch_cpuhp_init_parallel_bringup() and enable it")
> Reported-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

Tested-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>

-- 
  Kiryl Shutsemau / Kirill A. Shutemov

Re: [patch] x86/smpboot: Fix the parallel bringup decision

[Tom Lendacky]

On 5/31/23 02:44, Thomas Gleixner wrote:
> The decision to allow parallel bringup of secondary CPUs checks
> CC_ATTR_GUEST_STATE_ENCRYPT to detect encrypted guests. Those cannot use
> parallel bootup because accessing the local APIC is intercepted and raises
> a #VC or #VE, which cannot be handled at that point.
> 
> The check works correctly, but only for AMD encrypted guests. TDX does not
> set that flag.
> 
> As there is no real connection between CC attributes and the inability to
> support parallel bringup, replace this with a generic control flag in
> x86_cpuinit and let SEV-ES and TDX init code disable it.
> 
> Fixes: 0c7ffa32dbd6 ("x86/smpboot/64: Implement arch_cpuhp_init_parallel_bringup() and enable it")
> Reported-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

Still works for SEV-ES/SEV-SNP with parallel boot properly disabled.

Tested-by: Tom Lendacky <thomas.lendacky@amd.com>

> ---
>   arch/x86/coco/tdx/tdx.c         |   11 +++++++++++
>   arch/x86/include/asm/x86_init.h |    3 +++
>   arch/x86/kernel/smpboot.c       |   19 ++-----------------
>   arch/x86/kernel/x86_init.c      |    1 +
>   arch/x86/mm/mem_encrypt_amd.c   |   15 +++++++++++++++
>   5 files changed, 32 insertions(+), 17 deletions(-)
> 
> --- a/arch/x86/coco/tdx/tdx.c
> +++ b/arch/x86/coco/tdx/tdx.c
> @@ -871,5 +871,16 @@ void __init tdx_early_init(void)
>   	x86_platform.guest.enc_tlb_flush_required   = tdx_tlb_flush_required;
>   	x86_platform.guest.enc_status_change_finish = tdx_enc_status_changed;
>   
> +	/*
> +	 * TDX intercepts the RDMSR to read the X2APIC ID in the parallel
> +	 * bringup low level code. That raises #VE which cannot be handled
> +	 * there.
> +	 *
> +	 * Intel-TDX has a secure RDMSR hypercall, but that needs to be
> +	 * implemented seperately in the low level startup ASM code.
> +	 * Until that is in place, disable parallel bringup for TDX.
> +	 */
> +	x86_cpuinit.parallel_bringup = false;
> +
>   	pr_info("Guest detected\n");
>   }
> --- a/arch/x86/include/asm/x86_init.h
> +++ b/arch/x86/include/asm/x86_init.h
> @@ -177,11 +177,14 @@ struct x86_init_ops {
>    * struct x86_cpuinit_ops - platform specific cpu hotplug setups
>    * @setup_percpu_clockev:	set up the per cpu clock event device
>    * @early_percpu_clock_init:	early init of the per cpu clock event device
> + * @fixup_cpu_id:		fixup function for cpuinfo_x86::phys_proc_id
> + * @parallel_bringup:		Parallel bringup control
>    */
>   struct x86_cpuinit_ops {
>   	void (*setup_percpu_clockev)(void);
>   	void (*early_percpu_clock_init)(void);
>   	void (*fixup_cpu_id)(struct cpuinfo_x86 *c, int node);
> +	bool parallel_bringup;
>   };
>   
>   struct timespec64;
> --- a/arch/x86/kernel/smpboot.c
> +++ b/arch/x86/kernel/smpboot.c
> @@ -1267,23 +1267,8 @@ void __init smp_prepare_cpus_common(void
>   /* Establish whether parallel bringup can be supported. */
>   bool __init arch_cpuhp_init_parallel_bringup(void)
>   {
> -	/*
> -	 * Encrypted guests require special handling. They enforce X2APIC
> -	 * mode but the RDMSR to read the APIC ID is intercepted and raises
> -	 * #VC or #VE which cannot be handled in the early startup code.
> -	 *
> -	 * AMD-SEV does not provide a RDMSR GHCB protocol so the early
> -	 * startup code cannot directly communicate with the secure
> -	 * firmware. The alternative solution to retrieve the APIC ID via
> -	 * CPUID(0xb), which is covered by the GHCB protocol, is not viable
> -	 * either because there is no enforcement of the CPUID(0xb)
> -	 * provided "initial" APIC ID to be the same as the real APIC ID.
> -	 *
> -	 * Intel-TDX has a secure RDMSR hypercall, but that needs to be
> -	 * implemented seperately in the low level startup ASM code.
> -	 */
> -	if (cc_platform_has(CC_ATTR_GUEST_STATE_ENCRYPT)) {
> -		pr_info("Parallel CPU startup disabled due to guest state encryption\n");
> +	if (!x86_cpuinit.parallel_bringup) {
> +		pr_info("Parallel CPU startup disabled by the platform\n");
>   		return false;
>   	}
>   
> --- a/arch/x86/kernel/x86_init.c
> +++ b/arch/x86/kernel/x86_init.c
> @@ -126,6 +126,7 @@ struct x86_init_ops x86_init __initdata
>   struct x86_cpuinit_ops x86_cpuinit = {
>   	.early_percpu_clock_init	= x86_init_noop,
>   	.setup_percpu_clockev		= setup_secondary_APIC_clock,
> +	.parallel_bringup		= true,
>   };
>   
>   static void default_nmi_init(void) { };
> --- a/arch/x86/mm/mem_encrypt_amd.c
> +++ b/arch/x86/mm/mem_encrypt_amd.c
> @@ -501,6 +501,21 @@ void __init sme_early_init(void)
>   	x86_platform.guest.enc_status_change_finish  = amd_enc_status_change_finish;
>   	x86_platform.guest.enc_tlb_flush_required    = amd_enc_tlb_flush_required;
>   	x86_platform.guest.enc_cache_flush_required  = amd_enc_cache_flush_required;
> +
> +	/*
> +	 * AMD-SEV-ES intercepts the RDMSR to read the X2APIC ID in the
> +	 * parallel bringup low level code. That raises #VC which cannot be
> +	 * handled there.
> +	 * It does not provide a RDMSR GHCB protocol so the early startup
> +	 * code cannot directly communicate with the secure firmware. The
> +	 * alternative solution to retrieve the APIC ID via CPUID(0xb),
> +	 * which is covered by the GHCB protocol, is not viable either
> +	 * because there is no enforcement of the CPUID(0xb) provided
> +	 * "initial" APIC ID to be the same as the real APIC ID.
> +	 * Disable parallel bootup.
> +	 */
> +	if (sev_status & MSR_AMD64_SEV_ES_ENABLED)
> +		x86_cpuinit.parallel_bringup = false;
>   }
>   
>   void __init mem_encrypt_free_decrypted_mem(void)

Re: [patch] x86/smpboot: Disable parallel bootup if cc_vendor != NONE

[Kirill A. Shutemov]

On Tue, May 30, 2023 at 06:00:46PM +0200, Thomas Gleixner wrote:
> On Tue, May 30 2023 at 15:29, Kirill A. Shutemov wrote:
> > On Tue, May 30, 2023 at 02:09:17PM +0200, Thomas Gleixner wrote:
> >> The decision to allow parallel bringup of secondary CPUs checks
> >> CC_ATTR_GUEST_STATE_ENCRYPT to detect encrypted guests. Those cannot use
> >> parallel bootup because accessing the local APIC is intercepted and raises
> >> a #VC or #VE, which cannot be handled at that point.
> >> 
> >> The check works correctly, but only for AMD encrypted guests. TDX does not
> >> set that flag.
> >> 
> >> Check for cc_vendor != CC_VENDOR_NONE instead. That might be overbroad, but
> >> definitely works for both AMD and Intel.
> >
> > It boots fine with TDX, but I think it is wrong. cc_get_vendor() will
> > report CC_VENDOR_AMD even on bare metal if SME is enabled. I don't think
> > we want it.
> 
> Right. Did not think about that.
> 
> But the same way is CC_ATTR_GUEST_MEM_ENCRYPT overbroad for AMD. Only
> SEV-ES traps RDMSR if I'm understandig that maze correctly.

I don't know difference between SEV flavours that well.

I see there's that on SEV-SNP access to x2APIC MSR range (MSR 0x800-0x8FF)
is intercepted regardless if MSR_AMD64_SNP_ALT_INJ feature is present. But
I'm not sure what the state on SEV or SEV-ES.

Tom?

-- 
  Kiryl Shutsemau / Kirill A. Shutemov

Re: [patch] x86/smpboot: Disable parallel bootup if cc_vendor != NONE

[Sean Christopherson]

On Tue, May 30, 2023, Kirill A. Shutemov wrote:
> On Tue, May 30, 2023 at 06:00:46PM +0200, Thomas Gleixner wrote:
> > On Tue, May 30 2023 at 15:29, Kirill A. Shutemov wrote:
> > > On Tue, May 30, 2023 at 02:09:17PM +0200, Thomas Gleixner wrote:
> > >> The decision to allow parallel bringup of secondary CPUs checks
> > >> CC_ATTR_GUEST_STATE_ENCRYPT to detect encrypted guests. Those cannot use
> > >> parallel bootup because accessing the local APIC is intercepted and raises
> > >> a #VC or #VE, which cannot be handled at that point.
> > >> 
> > >> The check works correctly, but only for AMD encrypted guests. TDX does not
> > >> set that flag.
> > >> 
> > >> Check for cc_vendor != CC_VENDOR_NONE instead. That might be overbroad, but
> > >> definitely works for both AMD and Intel.
> > >
> > > It boots fine with TDX, but I think it is wrong. cc_get_vendor() will
> > > report CC_VENDOR_AMD even on bare metal if SME is enabled. I don't think
> > > we want it.
> > 
> > Right. Did not think about that.
> > 
> > But the same way is CC_ATTR_GUEST_MEM_ENCRYPT overbroad for AMD. Only
> > SEV-ES traps RDMSR if I'm understandig that maze correctly.
> 
> I don't know difference between SEV flavours that well.
> 
> I see there's that on SEV-SNP access to x2APIC MSR range (MSR 0x800-0x8FF)
> is intercepted regardless if MSR_AMD64_SNP_ALT_INJ feature is present. But
> I'm not sure what the state on SEV or SEV-ES.

With SEV-ES, if the hypervisor intercepts an MSR access, the VM-Exit is instead
morphed to a #VC (except for EFER).  The guest needs to do an explicit VMGEXIT
(i.e. a hypercall) to explicitly request MSR emulation (this *can* be done in the
#VC handler, but the guest can also do VMGEXIT directly, e.g. in lieu of a RDMSR).

With regular SEV, VM-Exits aren't reflected into the guest.  Register state isn't
encrypted so the hypervisor can emulate MSR accesses (and other instructions)
without needing an explicit hypercall from the guest.

Re: [patch v3 31/36] x86/apic: Provide cpu_primary_thread mask

[Thomas Gleixner]

On Mon, May 29 2023 at 23:31, Kirill A. Shutemov wrote:
> Aaand the next patch that breaks TDX boot is... <drum roll>
>
> 	x86/smpboot/64: Implement arch_cpuhp_init_parallel_bringup() and enable it
>
> Disabling parallel bringup helps. I didn't look closer yet. If you have
> an idea let me know.

So how does TDX end up with actual parallel bringup?

	if (cc_platform_has(CC_ATTR_GUEST_STATE_ENCRYPT)) {
		pr_info("Parallel CPU startup disabled due to guest state encryption\n");
		return false;
	}

It should take that path, no?

Thanks,

        tglx

[patch] x86/realmode: Make stack lock work in trampoline_compat()

[Thomas Gleixner]

The stack locking and stack assignment macro LOAD_REALMODE_ESP fails to
work when invoked from the 64bit trampoline entry point:

trampoline_start64
  trampoline_compat
    LOAD_REALMODE_ESP <- lock

Accessing tr_lock is only possible from 16bit mode. For the compat entry
point this needs to be pa_tr_lock so that the required relocation entry is
generated. Otherwise it locks the non-relocated address which is
aside of being wrong never cleared in secondary_startup_64() causing all
but the first CPU to get stuck on the lock.

Make the macro take an argument lock_pa which defaults to 0 and rename it
to LOCK_AND_LOAD_REALMODE_ESP to make it clear what this is about.

Fixes: f6f1ae9128d2 ("x86/smpboot: Implement a bit spinlock to protect the realmode stack")
Reported-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 arch/x86/realmode/rm/trampoline_64.S |   12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)

--- a/arch/x86/realmode/rm/trampoline_64.S
+++ b/arch/x86/realmode/rm/trampoline_64.S
@@ -37,12 +37,16 @@
 	.text
 	.code16
 
-.macro LOAD_REALMODE_ESP
+.macro LOCK_AND_LOAD_REALMODE_ESP lock_pa=0
 	/*
 	 * Make sure only one CPU fiddles with the realmode stack
 	 */
 .Llock_rm\@:
+	.if \lock_pa
+        lock btsl       $0, pa_tr_lock
+	.else
         lock btsl       $0, tr_lock
+	.endif
         jnc             2f
         pause
         jmp             .Llock_rm\@
@@ -63,7 +67,7 @@ SYM_CODE_START(trampoline_start)
 	mov	%ax, %es
 	mov	%ax, %ss
 
-	LOAD_REALMODE_ESP
+	LOCK_AND_LOAD_REALMODE_ESP
 
 	call	verify_cpu		# Verify the cpu supports long mode
 	testl   %eax, %eax		# Check for return code
@@ -106,7 +110,7 @@ SYM_CODE_START(sev_es_trampoline_start)
 	mov	%ax, %es
 	mov	%ax, %ss
 
-	LOAD_REALMODE_ESP
+	LOCK_AND_LOAD_REALMODE_ESP
 
 	jmp	.Lswitch_to_protected
 SYM_CODE_END(sev_es_trampoline_start)
@@ -189,7 +193,7 @@ SYM_CODE_START(pa_trampoline_compat)
 	 * In compatibility mode.  Prep ESP and DX for startup_32, then disable
 	 * paging and complete the switch to legacy 32-bit mode.
 	 */
-	LOAD_REALMODE_ESP
+	LOCK_AND_LOAD_REALMODE_ESP lock_pa=1
 	movw	$__KERNEL_DS, %dx
 
 	movl	$(CR0_STATE & ~X86_CR0_PG), %eax

Re: [patch] x86/realmode: Make stack lock work in trampoline_compat()

[Kirill A. Shutemov]

On Tue, May 30, 2023 at 12:46:22PM +0200, Thomas Gleixner wrote:
> The stack locking and stack assignment macro LOAD_REALMODE_ESP fails to
> work when invoked from the 64bit trampoline entry point:
> 
> trampoline_start64
>   trampoline_compat
>     LOAD_REALMODE_ESP <- lock
> 
> Accessing tr_lock is only possible from 16bit mode. For the compat entry
> point this needs to be pa_tr_lock so that the required relocation entry is
> generated. Otherwise it locks the non-relocated address which is
> aside of being wrong never cleared in secondary_startup_64() causing all
> but the first CPU to get stuck on the lock.
> 
> Make the macro take an argument lock_pa which defaults to 0 and rename it
> to LOCK_AND_LOAD_REALMODE_ESP to make it clear what this is about.
> 
> Fixes: f6f1ae9128d2 ("x86/smpboot: Implement a bit spinlock to protect the realmode stack")
> Reported-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

Tested-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>

-- 
  Kiryl Shutsemau / Kirill A. Shutemov

Re: [patch] x86/realmode: Make stack lock work in trampoline_compat()

[Yunhong Jiang]

On Tue, May 30, 2023 at 12:46:22PM +0200, Thomas Gleixner wrote:
> The stack locking and stack assignment macro LOAD_REALMODE_ESP fails to
> work when invoked from the 64bit trampoline entry point:
> 
> trampoline_start64
>   trampoline_compat
>     LOAD_REALMODE_ESP <- lock
One possibly dumb question and hope get some hints. The LOAD_REALMODE_ESP is
defined under .code16 directive and will be used by 32-bit mode caller also. Is
it ok because the instructions there will be same for both 16-bit and 32-bit? I
checked
https://ftp.gnu.org/old-gnu/Manuals/gas-2.9.1/html_chapter/as_16.html#SEC205 and
don't find much information there.

Re: [patch] x86/realmode: Make stack lock work in trampoline_compat()

[Andrew Cooper]

On 09/06/2023 12:34 am, Yunhong Jiang wrote:
> On Tue, May 30, 2023 at 12:46:22PM +0200, Thomas Gleixner wrote:
>> The stack locking and stack assignment macro LOAD_REALMODE_ESP fails to
>> work when invoked from the 64bit trampoline entry point:
>>
>> trampoline_start64
>>   trampoline_compat
>>     LOAD_REALMODE_ESP <- lock
> One possibly dumb question and hope get some hints.

There's a phrase.  "The only dumb question is the one not asked".

If you have this question, there's an excellent chance that someone else
reading this thread has the same question.

>  The LOAD_REALMODE_ESP is
> defined under .code16 directive and will be used by 32-bit mode caller also. Is
> it ok because the instructions there will be same for both 16-bit and 32-bit? I
> checked
> https://ftp.gnu.org/old-gnu/Manuals/gas-2.9.1/html_chapter/as_16.html#SEC205 and
> don't find much information there.

The position of the LOAD_REALMODE_ESP .macro itself doesn't matter. 
It's just some text which gets pasted elsewhere.  Imagine it just the
same as running the C preprocessor on a file before compiling it.

As you note, some expansions of the macro are in .code16, and some are
not.  This does result in different bytes being emitted.  The default
operands size flips between .code16 and .code32, so there will be some
0x66 prefixes in one mode, and not in others.

The important point is the l suffix on btsl, which forces it to be long
(32bit) irrespective of the default operand size.

So yes, it will work, but that's because gas is handling the differing
encodings automatically based on the default operand size where the
LOAD_REALMODE_ESP gets expanded.

Hope this helps,

~Andrew

Re: [patch] x86/realmode: Make stack lock work in trampoline_compat()

[Yunhong Jiang]

On Fri, Jun 09, 2023 at 12:57:46AM +0100, Andrew Cooper wrote:
> On 09/06/2023 12:34 am, Yunhong Jiang wrote:
> > On Tue, May 30, 2023 at 12:46:22PM +0200, Thomas Gleixner wrote:
> >> The stack locking and stack assignment macro LOAD_REALMODE_ESP fails to
> >> work when invoked from the 64bit trampoline entry point:
> >>
> >> trampoline_start64
> >>   trampoline_compat
> >>     LOAD_REALMODE_ESP <- lock
> > One possibly dumb question and hope get some hints.
> 
> There's a phrase.  "The only dumb question is the one not asked".
> 
> If you have this question, there's an excellent chance that someone else
> reading this thread has the same question.
> 
> >  The LOAD_REALMODE_ESP is
> > defined under .code16 directive and will be used by 32-bit mode caller also. Is
> > it ok because the instructions there will be same for both 16-bit and 32-bit? I
> > checked
> > https://ftp.gnu.org/old-gnu/Manuals/gas-2.9.1/html_chapter/as_16.html#SEC205 and
> > don't find much information there.
> 
> The position of the LOAD_REALMODE_ESP .macro itself doesn't matter. 
> It's just some text which gets pasted elsewhere.  Imagine it just the
> same as running the C preprocessor on a file before compiling it.
> 
> As you note, some expansions of the macro are in .code16, and some are
> not.  This does result in different bytes being emitted.  The default
> operands size flips between .code16 and .code32, so there will be some
> 0x66 prefixes in one mode, and not in others.
> 
> The important point is the l suffix on btsl, which forces it to be long
> (32bit) irrespective of the default operand size.
> 
> So yes, it will work, but that's because gas is handling the differing
> encodings automatically based on the default operand size where the
> LOAD_REALMODE_ESP gets expanded.
> 
> Hope this helps,
Thank you for the explaination, it's quite clear now.
> 
> ~Andrew

RE: [patch] x86/realmode: Make stack lock work in trampoline_compat()

[David Laight]

From: Andrew Cooper
> Sent: 09 June 2023 00:58
> 
...
> The important point is the l suffix on btsl, which forces it to be long
> (32bit) irrespective of the default operand size.

Does that matter at all?
The 'bit' opcodes (I'm sure 'bts' is 'bit test and set') take
a bit offset from the base address.
This accesses the same bit regardless of the operand size.

The one real issue is that a byte operand will only lock the one byte.
This might be problematic if non-bit locked accesses are also used.
Although it would need to be rather obscure use.
(This may be one of them...)

The only other problem is that btsl always does a locked 32bit
access. If the base address is misaligned this is a misaligned
locked access - problematic if it crosses a cache line boundary.

	David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)

Re: [patch] x86/realmode: Make stack lock work in trampoline_compat()

['Andrew Cooper']

On 10/06/2023 8:50 pm, David Laight wrote:
> From: Andrew Cooper
>> Sent: 09 June 2023 00:58
>>
> ...
>> The important point is the l suffix on btsl, which forces it to be long
>> (32bit) irrespective of the default operand size.
> Does that matter at all?

Yes, or I wouldn't have wasted everyone's time saying otherwise.

I recommend educating yourself with the instruction manual, rather than
presenting a set of statements which are all wrong.

~Andrew

[patch v3 32/36] cpu/hotplug: Allow "parallel" bringup up to CPUHP_BP_KICK_AP_STATE

[Thomas Gleixner]

From: Thomas Gleixner <tglx@linutronix.de>

There is often significant latency in the early stages of CPU bringup, and
time is wasted by waking each CPU (e.g. with SIPI/INIT/INIT on x86) and
then waiting for it to respond before moving on to the next.

Allow a platform to enable parallel setup which brings all to be onlined
CPUs up to the CPUHP_BP_KICK_AP state. While this state advancement on the
control CPU (BP) is single-threaded the important part is the last state
CPUHP_BP_KICK_AP which wakes the to be onlined CPUs up.

This allows the CPUs to run up to the first sychronization point
cpuhp_ap_sync_alive() where they wait for the control CPU to release them
one by one for the full onlining procedure.

This parallelism depends on the CPU hotplug core sync mechanism which
ensures that the parallel brought up CPUs wait for release before touching
any state which would make the CPU visible to anything outside the hotplug
control mechanism.

To handle the SMT constraints of X86 correctly the bringup happens in two
iterations when CONFIG_HOTPLUG_SMT is enabled. The control CPU brings up
the primary SMT threads of each core first, which can load the microcode
without the need to rendevouz with the thread siblings. Once that's
completed it brings up the secondary SMT threads.

Co-developed-by: David Woodhouse <dwmw@amazon.co.uk>
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Michael Kelley <mikelley@microsoft.com>


---
 Documentation/admin-guide/kernel-parameters.txt |    6 +
 arch/Kconfig                                    |    4 
 include/linux/cpuhotplug.h                      |    1 
 kernel/cpu.c                                    |  103 ++++++++++++++++++++++--
 4 files changed, 109 insertions(+), 5 deletions(-)
---

--- a/Documentation/admin-guide/kernel-parameters.txt
+++ b/Documentation/admin-guide/kernel-parameters.txt
@@ -838,6 +838,12 @@
 			on every CPU online, such as boot, and resume from suspend.
 			Default: 10000
 
+	cpuhp.parallel=
+			[SMP] Enable/disable parallel bringup of secondary CPUs
+			Format: <bool>
+			Default is enabled if CONFIG_HOTPLUG_PARALLEL=y. Otherwise
+			the parameter has no effect.
+
 	crash_kexec_post_notifiers
 			Run kdump after running panic-notifiers and dumping
 			kmsg. This only for the users who doubt kdump always
--- a/arch/Kconfig
+++ b/arch/Kconfig
@@ -53,6 +53,10 @@ config HOTPLUG_SPLIT_STARTUP
 	bool
 	select HOTPLUG_CORE_SYNC_FULL
 
+config HOTPLUG_PARALLEL
+	bool
+	select HOTPLUG_SPLIT_STARTUP
+
 config GENERIC_ENTRY
 	bool
 
--- a/include/linux/cpuhotplug.h
+++ b/include/linux/cpuhotplug.h
@@ -524,6 +524,7 @@ void cpuhp_ap_sync_alive(void);
 void arch_cpuhp_sync_state_poll(void);
 void arch_cpuhp_cleanup_kick_cpu(unsigned int cpu);
 int arch_cpuhp_kick_ap_alive(unsigned int cpu, struct task_struct *tidle);
+bool arch_cpuhp_init_parallel_bringup(void);
 
 #ifdef CONFIG_HOTPLUG_CORE_SYNC_DEAD
 void cpuhp_ap_report_dead(void);
--- a/kernel/cpu.c
+++ b/kernel/cpu.c
@@ -649,8 +649,23 @@ bool cpu_smt_possible(void)
 		cpu_smt_control != CPU_SMT_NOT_SUPPORTED;
 }
 EXPORT_SYMBOL_GPL(cpu_smt_possible);
+
+static inline bool cpuhp_smt_aware(void)
+{
+	return topology_smt_supported();
+}
+
+static inline const struct cpumask *cpuhp_get_primary_thread_mask(void)
+{
+	return cpu_primary_thread_mask;
+}
 #else
 static inline bool cpu_smt_allowed(unsigned int cpu) { return true; }
+static inline bool cpuhp_smt_aware(void) { return false; }
+static inline const struct cpumask *cpuhp_get_primary_thread_mask(void)
+{
+	return cpu_present_mask;
+}
 #endif
 
 static inline enum cpuhp_state
@@ -1743,16 +1758,94 @@ int bringup_hibernate_cpu(unsigned int s
 	return 0;
 }
 
-void __init bringup_nonboot_cpus(unsigned int setup_max_cpus)
+static void __init cpuhp_bringup_mask(const struct cpumask *mask, unsigned int ncpus,
+				      enum cpuhp_state target)
 {
 	unsigned int cpu;
 
-	for_each_present_cpu(cpu) {
-		if (num_online_cpus() >= setup_max_cpus)
+	for_each_cpu(cpu, mask) {
+		struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
+
+		if (!--ncpus)
 			break;
-		if (!cpu_online(cpu))
-			cpu_up(cpu, CPUHP_ONLINE);
+
+		if (cpu_up(cpu, target) && can_rollback_cpu(st)) {
+			/*
+			 * If this failed then cpu_up() might have only
+			 * rolled back to CPUHP_BP_KICK_AP for the final
+			 * online. Clean it up. NOOP if already rolled back.
+			 */
+			WARN_ON(cpuhp_invoke_callback_range(false, cpu, st, CPUHP_OFFLINE));
+		}
+	}
+}
+
+#ifdef CONFIG_HOTPLUG_PARALLEL
+static bool __cpuhp_parallel_bringup __ro_after_init = true;
+
+static int __init parallel_bringup_parse_param(char *arg)
+{
+	return kstrtobool(arg, &__cpuhp_parallel_bringup);
+}
+early_param("cpuhp.parallel", parallel_bringup_parse_param);
+
+/*
+ * On architectures which have enabled parallel bringup this invokes all BP
+ * prepare states for each of the to be onlined APs first. The last state
+ * sends the startup IPI to the APs. The APs proceed through the low level
+ * bringup code in parallel and then wait for the control CPU to release
+ * them one by one for the final onlining procedure.
+ *
+ * This avoids waiting for each AP to respond to the startup IPI in
+ * CPUHP_BRINGUP_CPU.
+ */
+static bool __init cpuhp_bringup_cpus_parallel(unsigned int ncpus)
+{
+	const struct cpumask *mask = cpu_present_mask;
+
+	if (__cpuhp_parallel_bringup)
+		__cpuhp_parallel_bringup = arch_cpuhp_init_parallel_bringup();
+	if (!__cpuhp_parallel_bringup)
+		return false;
+
+	if (cpuhp_smt_aware()) {
+		const struct cpumask *pmask = cpuhp_get_primary_thread_mask();
+		static struct cpumask tmp_mask __initdata;
+
+		/*
+		 * X86 requires to prevent that SMT siblings stopped while
+		 * the primary thread does a microcode update for various
+		 * reasons. Bring the primary threads up first.
+		 */
+		cpumask_and(&tmp_mask, mask, pmask);
+		cpuhp_bringup_mask(&tmp_mask, ncpus, CPUHP_BP_KICK_AP);
+		cpuhp_bringup_mask(&tmp_mask, ncpus, CPUHP_ONLINE);
+		/* Account for the online CPUs */
+		ncpus -= num_online_cpus();
+		if (!ncpus)
+			return true;
+		/* Create the mask for secondary CPUs */
+		cpumask_andnot(&tmp_mask, mask, pmask);
+		mask = &tmp_mask;
 	}
+
+	/* Bring the not-yet started CPUs up */
+	cpuhp_bringup_mask(mask, ncpus, CPUHP_BP_KICK_AP);
+	cpuhp_bringup_mask(mask, ncpus, CPUHP_ONLINE);
+	return true;
+}
+#else
+static inline bool cpuhp_bringup_cpus_parallel(unsigned int ncpus) { return false; }
+#endif /* CONFIG_HOTPLUG_PARALLEL */
+
+void __init bringup_nonboot_cpus(unsigned int setup_max_cpus)
+{
+	/* Try parallel bringup optimization if enabled */
+	if (cpuhp_bringup_cpus_parallel(setup_max_cpus))
+		return;
+
+	/* Full per CPU serialized bringup */
+	cpuhp_bringup_mask(cpu_present_mask, setup_max_cpus, CPUHP_ONLINE);
 }
 
 #ifdef CONFIG_PM_SLEEP_SMP

[patch v3 33/36] x86/apic: Save the APIC virtual base address

[Thomas Gleixner]

From: Thomas Gleixner <tglx@linutronix.de>

For parallel CPU brinugp it's required to read the APIC ID in the low level
startup code. The virtual APIC base address is a constant because its a
fix-mapped address. Exposing that constant which is composed via macros to
assembly code is non-trivial dues to header inclusion hell.

Aside of that it's constant only because of the vsyscall ABI
requirement. Once vsyscall is out of the picture the fixmap can be placed
at runtime.

Avoid header hell, stay flexible and store the address in a variable which
can be exposed to the low level startup code.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Michael Kelley <mikelley@microsoft.com>


---
 arch/x86/include/asm/smp.h  |    1 +
 arch/x86/kernel/apic/apic.c |    4 ++++
 2 files changed, 5 insertions(+)
---

--- a/arch/x86/include/asm/smp.h
+++ b/arch/x86/include/asm/smp.h
@@ -196,6 +196,7 @@ extern void nmi_selftest(void);
 #endif
 
 extern unsigned int smpboot_control;
+extern unsigned long apic_mmio_base;
 
 #endif /* !__ASSEMBLY__ */
 
--- a/arch/x86/kernel/apic/apic.c
+++ b/arch/x86/kernel/apic/apic.c
@@ -101,6 +101,9 @@ static int apic_extnmi __ro_after_init =
  */
 static bool virt_ext_dest_id __ro_after_init;
 
+/* For parallel bootup. */
+unsigned long apic_mmio_base __ro_after_init;
+
 /*
  * Map cpu index to physical APIC ID
  */
@@ -2163,6 +2166,7 @@ void __init register_lapic_address(unsig
 
 	if (!x2apic_mode) {
 		set_fixmap_nocache(FIX_APIC_BASE, address);
+		apic_mmio_base = APIC_BASE;
 		apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n",
 			    APIC_BASE, address);
 	}

Re: [patch v3 33/36] x86/apic: Save the APIC virtual base address

[Sergey Shtylyov]

Hello!

On 5/8/23 10:44 PM, Thomas Gleixner wrote:

> From: Thomas Gleixner <tglx@linutronix.de>
> 
> For parallel CPU brinugp it's required to read the APIC ID in the low level
> startup code. The virtual APIC base address is a constant because its a
> fix-mapped address. Exposing that constant which is composed via macros to
> assembly code is non-trivial dues to header inclusion hell.

   s/dues/due/?

> Aside of that it's constant only because of the vsyscall ABI
> requirement. Once vsyscall is out of the picture the fixmap can be placed
> at runtime.
> 
> Avoid header hell, stay flexible and store the address in a variable which
> can be exposed to the low level startup code.
> 
> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
> Tested-by: Michael Kelley <mikelley@microsoft.com>

[...]

MBR, Sergey

[patch v3 34/36] x86/smpboot: Implement a bit spinlock to protect the realmode stack

[Thomas Gleixner]

From: Thomas Gleixner <tglx@linutronix.de>

Parallel AP bringup requires that the APs can run fully parallel through
the early startup code including the real mode trampoline.

To prepare for this implement a bit-spinlock to serialize access to the
real mode stack so that parallel upcoming APs are not going to corrupt each
others stack while going through the real mode startup code.

Co-developed-by: David Woodhouse <dwmw@amazon.co.uk>
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Michael Kelley <mikelley@microsoft.com>


---
 arch/x86/include/asm/realmode.h      |    3 +++
 arch/x86/kernel/head_64.S            |   13 +++++++++++++
 arch/x86/realmode/init.c             |    3 +++
 arch/x86/realmode/rm/trampoline_64.S |   27 ++++++++++++++++++++++-----
 4 files changed, 41 insertions(+), 5 deletions(-)
---

--- a/arch/x86/include/asm/realmode.h
+++ b/arch/x86/include/asm/realmode.h
@@ -52,6 +52,7 @@ struct trampoline_header {
 	u64 efer;
 	u32 cr4;
 	u32 flags;
+	u32 lock;
 #endif
 };
 
@@ -64,6 +65,8 @@ extern unsigned long initial_stack;
 extern unsigned long initial_vc_handler;
 #endif
 
+extern u32 *trampoline_lock;
+
 extern unsigned char real_mode_blob[];
 extern unsigned char real_mode_relocs[];
 
--- a/arch/x86/kernel/head_64.S
+++ b/arch/x86/kernel/head_64.S
@@ -252,6 +252,17 @@ SYM_INNER_LABEL(secondary_startup_64_no_
 	movq	TASK_threadsp(%rax), %rsp
 
 	/*
+	 * Now that this CPU is running on its own stack, drop the realmode
+	 * protection. For the boot CPU the pointer is NULL!
+	 */
+	movq	trampoline_lock(%rip), %rax
+	testq	%rax, %rax
+	jz	.Lsetup_gdt
+	lock
+	btrl	$0, (%rax)
+
+.Lsetup_gdt:
+	/*
 	 * We must switch to a new descriptor in kernel space for the GDT
 	 * because soon the kernel won't have access anymore to the userspace
 	 * addresses where we're currently running on. We have to do that here
@@ -433,6 +444,8 @@ SYM_DATA(initial_code,	.quad x86_64_star
 #ifdef CONFIG_AMD_MEM_ENCRYPT
 SYM_DATA(initial_vc_handler,	.quad handle_vc_boot_ghcb)
 #endif
+
+SYM_DATA(trampoline_lock, .quad 0);
 	__FINITDATA
 
 	__INIT
--- a/arch/x86/realmode/init.c
+++ b/arch/x86/realmode/init.c
@@ -154,6 +154,9 @@ static void __init setup_real_mode(void)
 
 	trampoline_header->flags = 0;
 
+	trampoline_lock = &trampoline_header->lock;
+	*trampoline_lock = 0;
+
 	trampoline_pgd = (u64 *) __va(real_mode_header->trampoline_pgd);
 
 	/* Map the real mode stub as virtual == physical */
--- a/arch/x86/realmode/rm/trampoline_64.S
+++ b/arch/x86/realmode/rm/trampoline_64.S
@@ -37,6 +37,24 @@
 	.text
 	.code16
 
+.macro LOAD_REALMODE_ESP
+	/*
+	 * Make sure only one CPU fiddles with the realmode stack
+	 */
+.Llock_rm\@:
+	btl	$0, tr_lock
+	jnc	2f
+	pause
+	jmp	.Llock_rm\@
+2:
+	lock
+	btsl	$0, tr_lock
+	jc	.Llock_rm\@
+
+	# Setup stack
+	movl	$rm_stack_end, %esp
+.endm
+
 	.balign	PAGE_SIZE
 SYM_CODE_START(trampoline_start)
 	cli			# We should be safe anyway
@@ -49,8 +67,7 @@ SYM_CODE_START(trampoline_start)
 	mov	%ax, %es
 	mov	%ax, %ss
 
-	# Setup stack
-	movl	$rm_stack_end, %esp
+	LOAD_REALMODE_ESP
 
 	call	verify_cpu		# Verify the cpu supports long mode
 	testl   %eax, %eax		# Check for return code
@@ -93,8 +110,7 @@ SYM_CODE_START(sev_es_trampoline_start)
 	mov	%ax, %es
 	mov	%ax, %ss
 
-	# Setup stack
-	movl	$rm_stack_end, %esp
+	LOAD_REALMODE_ESP
 
 	jmp	.Lswitch_to_protected
 SYM_CODE_END(sev_es_trampoline_start)
@@ -177,7 +193,7 @@ SYM_CODE_START(pa_trampoline_compat)
 	 * In compatibility mode.  Prep ESP and DX for startup_32, then disable
 	 * paging and complete the switch to legacy 32-bit mode.
 	 */
-	movl	$rm_stack_end, %esp
+	LOAD_REALMODE_ESP
 	movw	$__KERNEL_DS, %dx
 
 	movl	$(CR0_STATE & ~X86_CR0_PG), %eax
@@ -241,6 +257,7 @@ SYM_DATA_START(trampoline_header)
 	SYM_DATA(tr_efer,		.space 8)
 	SYM_DATA(tr_cr4,		.space 4)
 	SYM_DATA(tr_flags,		.space 4)
+	SYM_DATA(tr_lock,		.space 4)
 SYM_DATA_END(trampoline_header)
 
 #include "trampoline_common.S"

Re: [patch v3 34/36] x86/smpboot: Implement a bit spinlock to protect the realmode stack

[Peter Zijlstra]

On Mon, May 08, 2023 at 09:44:22PM +0200, Thomas Gleixner wrote:

> @@ -252,6 +252,17 @@ SYM_INNER_LABEL(secondary_startup_64_no_
>  	movq	TASK_threadsp(%rax), %rsp
>  
>  	/*
> +	 * Now that this CPU is running on its own stack, drop the realmode
> +	 * protection. For the boot CPU the pointer is NULL!
> +	 */
> +	movq	trampoline_lock(%rip), %rax
	movl	$0, (%rax)

> +.Lsetup_gdt:
> +	/*
>  	 * We must switch to a new descriptor in kernel space for the GDT
>  	 * because soon the kernel won't have access anymore to the userspace
>  	 * addresses where we're currently running on. We have to do that here

> --- a/arch/x86/realmode/rm/trampoline_64.S
> +++ b/arch/x86/realmode/rm/trampoline_64.S
> @@ -37,6 +37,24 @@
>  	.text
>  	.code16
>  
> +.macro LOAD_REALMODE_ESP
> +	/*
> +	 * Make sure only one CPU fiddles with the realmode stack
> +	 */
> +.Llock_rm\@:
> +	btl	$0, tr_lock
> +	jnc	2f
> +	pause
> +	jmp	.Llock_rm\@
> +2:
> +	lock
> +	btsl	$0, tr_lock
> +	jc	.Llock_rm\@

Do we really care about performance here; or should we pick the simpler
form? Also, 'lock' is a prefix, not an instruction.

.Llock_rm\@:
	lock btsl	$0, tr_lock;
	jnc		2f
	pause
	jmp		.Llock_rm\@
2:

> +
> +	# Setup stack
> +	movl	$rm_stack_end, %esp
> +.endm
> +
>  	.balign	PAGE_SIZE

Re: [patch v3 34/36] x86/smpboot: Implement a bit spinlock to protect the realmode stack

[Thomas Gleixner]

On Tue, May 09 2023 at 15:13, Peter Zijlstra wrote:
> On Mon, May 08, 2023 at 09:44:22PM +0200, Thomas Gleixner wrote:
>
> Do we really care about performance here; or should we pick the simpler
> form? Also, 'lock' is a prefix, not an instruction.

Right. KISS is the way to go.

[patch v3 35/36] x86/smpboot: Support parallel startup of secondary CPUs

[Thomas Gleixner]

From: David Woodhouse <dwmw@amazon.co.uk>

In parallel startup mode the APs are kicked alive by the control CPU
quickly after each other and run through the early startup code in
parallel. The real-mode startup code is already serialized with a
bit-spinlock to protect the real-mode stack.

In parallel startup mode the smpboot_control variable obviously cannot
contain the Linux CPU number so the APs have to determine their Linux CPU
number on their own. This is required to find the CPUs per CPU offset in
order to find the idle task stack and other per CPU data.

To achieve this, export the cpuid_to_apicid[] array so that each AP can
find its own CPU number by searching therein based on its APIC ID.

Introduce a flag in the top bits of smpboot_control which indicates that
the AP should find its CPU number by reading the APIC ID from the APIC.

This is required because CPUID based APIC ID retrieval can only provide the
initial APIC ID, which might have been overruled by the firmware. Some AMD
APUs come up with APIC ID = initial APIC ID + 0x10, so the APIC ID to CPU
number lookup would fail miserably if based on CPUID. Also virtualization
can make its own APIC ID assignements. The only requirement is that the
APIC IDs are consistent with the APCI/MADT table.

For the boot CPU or in case parallel bringup is disabled the control bits
are empty and the CPU number is directly available in bit 0-23 of
smpboot_control.

[ tglx: Initial proof of concept patch with bitlock and APIC ID lookup ]
[ dwmw2: Rework and testing, commit message, CPUID 0x1 and CPU0 support ]
[ seanc: Fix stray override of initial_gs in common_cpu_up() ]
[ Oleksandr Natalenko: reported suspend/resume issue fixed in
  x86_acpi_suspend_lowlevel ]
[ tglx: Make it read the APIC ID from the APIC instead of using CPUID,
  	split the bitlock part out ]

Co-developed-by: Thomas Gleixner <tglx@linutronix.de>
Co-developed-by: Brian Gerst <brgerst@gmail.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Brian Gerst <brgerst@gmail.com>
Signed-off-by: David Woodhouse <dwmw@amazon.co.uk>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Michael Kelley <mikelley@microsoft.com>


---
 arch/x86/include/asm/apic.h    |    2 +
 arch/x86/include/asm/apicdef.h |    5 ++-
 arch/x86/include/asm/smp.h     |    6 +++
 arch/x86/kernel/acpi/sleep.c   |    9 +++++
 arch/x86/kernel/apic/apic.c    |    2 -
 arch/x86/kernel/head_64.S      |   62 +++++++++++++++++++++++++++++++++++++++++
 arch/x86/kernel/smpboot.c      |    2 -
 7 files changed, 84 insertions(+), 4 deletions(-)
---

--- a/arch/x86/include/asm/apic.h
+++ b/arch/x86/include/asm/apic.h
@@ -55,6 +55,8 @@ extern int local_apic_timer_c2_ok;
 extern int disable_apic;
 extern unsigned int lapic_timer_period;
 
+extern int cpuid_to_apicid[];
+
 extern enum apic_intr_mode_id apic_intr_mode;
 enum apic_intr_mode_id {
 	APIC_PIC,
--- a/arch/x86/include/asm/apicdef.h
+++ b/arch/x86/include/asm/apicdef.h
@@ -138,7 +138,8 @@
 #define		APIC_EILVT_MASKED	(1 << 16)
 
 #define APIC_BASE (fix_to_virt(FIX_APIC_BASE))
-#define APIC_BASE_MSR	0x800
+#define APIC_BASE_MSR		0x800
+#define APIC_X2APIC_ID_MSR	0x802
 #define XAPIC_ENABLE	(1UL << 11)
 #define X2APIC_ENABLE	(1UL << 10)
 
@@ -162,6 +163,7 @@
 #define APIC_CPUID(apicid)	((apicid) & XAPIC_DEST_CPUS_MASK)
 #define NUM_APIC_CLUSTERS	((BAD_APICID + 1) >> XAPIC_DEST_CPUS_SHIFT)
 
+#ifndef __ASSEMBLY__
 /*
  * the local APIC register structure, memory mapped. Not terribly well
  * tested, but we might eventually use this one in the future - the
@@ -435,4 +437,5 @@ enum apic_delivery_modes {
 	APIC_DELIVERY_MODE_EXTINT	= 7,
 };
 
+#endif /* !__ASSEMBLY__ */
 #endif /* _ASM_X86_APICDEF_H */
--- a/arch/x86/include/asm/smp.h
+++ b/arch/x86/include/asm/smp.h
@@ -200,4 +200,10 @@ extern unsigned long apic_mmio_base;
 
 #endif /* !__ASSEMBLY__ */
 
+/* Control bits for startup_64 */
+#define STARTUP_READ_APICID	0x80000000
+
+/* Top 8 bits are reserved for control */
+#define STARTUP_PARALLEL_MASK	0xFF000000
+
 #endif /* _ASM_X86_SMP_H */
--- a/arch/x86/kernel/acpi/sleep.c
+++ b/arch/x86/kernel/acpi/sleep.c
@@ -16,6 +16,7 @@
 #include <asm/cacheflush.h>
 #include <asm/realmode.h>
 #include <asm/hypervisor.h>
+#include <asm/smp.h>
 
 #include <linux/ftrace.h>
 #include "../../realmode/rm/wakeup.h"
@@ -127,7 +128,13 @@ int x86_acpi_suspend_lowlevel(void)
 	 * value is in the actual %rsp register.
 	 */
 	current->thread.sp = (unsigned long)temp_stack + sizeof(temp_stack);
-	smpboot_control = smp_processor_id();
+	/*
+	 * Ensure the CPU knows which one it is when it comes back, if
+	 * it isn't in parallel mode and expected to work that out for
+	 * itself.
+	 */
+	if (!(smpboot_control & STARTUP_PARALLEL_MASK))
+		smpboot_control = smp_processor_id();
 #endif
 	initial_code = (unsigned long)wakeup_long64;
 	saved_magic = 0x123456789abcdef0L;
--- a/arch/x86/kernel/apic/apic.c
+++ b/arch/x86/kernel/apic/apic.c
@@ -2380,7 +2380,7 @@ static int nr_logical_cpuids = 1;
 /*
  * Used to store mapping between logical CPU IDs and APIC IDs.
  */
-static int cpuid_to_apicid[] = {
+int cpuid_to_apicid[] = {
 	[0 ... NR_CPUS - 1] = -1,
 };
 
--- a/arch/x86/kernel/head_64.S
+++ b/arch/x86/kernel/head_64.S
@@ -24,7 +24,9 @@
 #include "../entry/calling.h"
 #include <asm/export.h>
 #include <asm/nospec-branch.h>
+#include <asm/apicdef.h>
 #include <asm/fixmap.h>
+#include <asm/smp.h>
 
 /*
  * We are not able to switch in one step to the final KERNEL ADDRESS SPACE
@@ -234,8 +236,68 @@ SYM_INNER_LABEL(secondary_startup_64_no_
 	ANNOTATE_NOENDBR // above
 
 #ifdef CONFIG_SMP
+	/*
+	 * For parallel boot, the APIC ID is read from the APIC, and then
+	 * used to look up the CPU number.  For booting a single CPU, the
+	 * CPU number is encoded in smpboot_control.
+	 *
+	 * Bit 31	STARTUP_READ_APICID (Read APICID from APIC)
+	 * Bit 0-23	CPU# if STARTUP_xx flags are not set
+	 */
 	movl	smpboot_control(%rip), %ecx
+	testl	$STARTUP_READ_APICID, %ecx
+	jnz	.Lread_apicid
+	/*
+	 * No control bit set, single CPU bringup. CPU number is provided
+	 * in bit 0-23. This is also the boot CPU case (CPU number 0).
+	 */
+	andl	$(~STARTUP_PARALLEL_MASK), %ecx
+	jmp	.Lsetup_cpu
+
+.Lread_apicid:
+	/* Check whether X2APIC mode is already enabled */
+	mov	$MSR_IA32_APICBASE, %ecx
+	rdmsr
+	testl	$X2APIC_ENABLE, %eax
+	jnz	.Lread_apicid_msr
+
+	/* Read the APIC ID from the fix-mapped MMIO space. */
+	movq	apic_mmio_base(%rip), %rcx
+	addq	$APIC_ID, %rcx
+	movl	(%rcx), %eax
+	shr	$24, %eax
+	jmp	.Llookup_AP
+
+.Lread_apicid_msr:
+	mov	$APIC_X2APIC_ID_MSR, %ecx
+	rdmsr
+
+.Llookup_AP:
+	/* EAX contains the APIC ID of the current CPU */
+	xorq	%rcx, %rcx
+	leaq	cpuid_to_apicid(%rip), %rbx
+
+.Lfind_cpunr:
+	cmpl	(%rbx,%rcx,4), %eax
+	jz	.Lsetup_cpu
+	inc	%ecx
+#ifdef CONFIG_FORCE_NR_CPUS
+	cmpl	$NR_CPUS, %ecx
+#else
+	cmpl	nr_cpu_ids(%rip), %ecx
+#endif
+	jb	.Lfind_cpunr
+
+	/*  APIC ID not found in the table. Drop the trampoline lock and bail. */
+	movq	trampoline_lock(%rip), %rax
+	lock
+	btrl	$0, (%rax)
+
+1:	cli
+	hlt
+	jmp	1b
 
+.Lsetup_cpu:
 	/* Get the per cpu offset for the given CPU# which is in ECX */
 	movq	__per_cpu_offset(,%rcx,8), %rdx
 #else
--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -1002,7 +1002,7 @@ static int do_boot_cpu(int apicid, int c
 	if (IS_ENABLED(CONFIG_X86_32)) {
 		early_gdt_descr.address = (unsigned long)get_cpu_gdt_rw(cpu);
 		initial_stack  = idle->thread.sp;
-	} else {
+	} else if (!(smpboot_control & STARTUP_PARALLEL_MASK)) {
 		smpboot_control = cpu;
 	}
 

Re: [patch v3 35/36] x86/smpboot: Support parallel startup of secondary CPUs

[Peter Zijlstra]

On Mon, May 08, 2023 at 09:44:23PM +0200, Thomas Gleixner wrote:
> +	/*  APIC ID not found in the table. Drop the trampoline lock and bail. */
> +	movq	trampoline_lock(%rip), %rax

Again:

	movl	$0, (%rax)

is sufficient for unlock.

> +	lock
> +	btrl	$0, (%rax)

[patch v3 36/36] x86/smpboot/64: Implement arch_cpuhp_init_parallel_bringup() and enable it

[Thomas Gleixner]

From: Thomas Gleixner <tglx@linutronix.de>

Implement the validation function which tells the core code whether
parallel bringup is possible.

The only condition for now is that the kernel does not run in an encrypted
guest as these will trap the RDMSR via #VC, which cannot be handled at that
point in early startup.

There was an earlier variant for AMD-SEV which used the GHBC protocol for
retrieving the APIC ID via CPUID, but there is no guarantee that the
initial APIC ID in CPUID is the same as the real APIC ID. There is no
enforcement from the secure firmware and the hypervisor can assign APIC IDs
as it sees fit as long as the ACPI/MADT table is consistent with that
assignment.

Unfortunately there is no RDMSR GHCB protocol at the moment, so enabling
AMD-SEV guests for parallel startup needs some more thought.

Intel-TDX provides a secure RDMSR hypercall, but supporting that is outside
the scope of this change.

Fixup announce_cpu() as e.g. on Hyper-V CPU1 is the secondary sibling of
CPU0, which makes the @cpu == 1 logic in announce_cpu() fall apart.

[ mikelley: Reported the announce_cpu() fallout

Originally-by: David Woodhouse <dwmw@amazon.co.uk>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Michael Kelley <mikelley@microsoft.com>

---
V2: Fixup announce_cpu() - Michael Kelley
V3: Fixup announce_cpu() for real - Michael Kelley
---
 arch/x86/Kconfig             |    3 -
 arch/x86/kernel/cpu/common.c |    6 --
 arch/x86/kernel/smpboot.c    |   87 +++++++++++++++++++++++++++++++++++--------
 3 files changed, 75 insertions(+), 21 deletions(-)
---

--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -274,8 +274,9 @@ config X86
 	select HAVE_UNSTABLE_SCHED_CLOCK
 	select HAVE_USER_RETURN_NOTIFIER
 	select HAVE_GENERIC_VDSO
+	select HOTPLUG_PARALLEL			if SMP && X86_64
 	select HOTPLUG_SMT			if SMP
-	select HOTPLUG_SPLIT_STARTUP		if SMP
+	select HOTPLUG_SPLIT_STARTUP		if SMP && X86_32
 	select IRQ_FORCED_THREADING
 	select NEED_PER_CPU_EMBED_FIRST_CHUNK
 	select NEED_PER_CPU_PAGE_FIRST_CHUNK
--- a/arch/x86/kernel/cpu/common.c
+++ b/arch/x86/kernel/cpu/common.c
@@ -2128,11 +2128,7 @@ static inline void setup_getcpu(int cpu)
 }
 
 #ifdef CONFIG_X86_64
-static inline void ucode_cpu_init(int cpu)
-{
-	if (cpu)
-		load_ucode_ap();
-}
+static inline void ucode_cpu_init(int cpu) { }
 
 static inline void tss_setup_ist(struct tss_struct *tss)
 {
--- a/arch/x86/kernel/smpboot.c
+++ b/arch/x86/kernel/smpboot.c
@@ -58,6 +58,7 @@
 #include <linux/overflow.h>
 #include <linux/stackprotector.h>
 #include <linux/cpuhotplug.h>
+#include <linux/mc146818rtc.h>
 
 #include <asm/acpi.h>
 #include <asm/cacheinfo.h>
@@ -75,7 +76,7 @@
 #include <asm/fpu/api.h>
 #include <asm/setup.h>
 #include <asm/uv/uv.h>
-#include <linux/mc146818rtc.h>
+#include <asm/microcode.h>
 #include <asm/i8259.h>
 #include <asm/misc.h>
 #include <asm/qspinlock.h>
@@ -128,7 +129,6 @@ int arch_update_cpu_topology(void)
 	return retval;
 }
 
-
 static unsigned int smpboot_warm_reset_vector_count;
 
 static inline void smpboot_setup_warm_reset_vector(unsigned long start_eip)
@@ -229,16 +229,43 @@ static void notrace start_secondary(void
 	 */
 	cr4_init();
 
-#ifdef CONFIG_X86_32
-	/* switch away from the initial page table */
-	load_cr3(swapper_pg_dir);
-	__flush_tlb_all();
-#endif
+	/*
+	 * 32-bit specific. 64-bit reaches this code with the correct page
+	 * table established. Yet another historical divergence.
+	 */
+	if (IS_ENABLED(CONFIG_X86_32)) {
+		/* switch away from the initial page table */
+		load_cr3(swapper_pg_dir);
+		__flush_tlb_all();
+	}
+
 	cpu_init_exception_handling();
 
 	/*
-	 * Synchronization point with the hotplug core. Sets the
-	 * synchronization state to ALIVE and waits for the control CPU to
+	 * 32-bit systems load the microcode from the ASM startup code for
+	 * historical reasons.
+	 *
+	 * On 64-bit systems load it before reaching the AP alive
+	 * synchronization point below so it is not part of the full per
+	 * CPU serialized bringup part when "parallel" bringup is enabled.
+	 *
+	 * That's even safe when hyperthreading is enabled in the CPU as
+	 * the core code starts the primary threads first and leaves the
+	 * secondary threads waiting for SIPI. Loading microcode on
+	 * physical cores concurrently is a safe operation.
+	 *
+	 * This covers both the Intel specific issue that concurrent
+	 * microcode loading on SMT siblings must be prohibited and the
+	 * vendor independent issue`that microcode loading which changes
+	 * CPUID, MSRs etc. must be strictly serialized to maintain
+	 * software state correctness.
+	 */
+	if (IS_ENABLED(CONFIG_X86_64))
+		load_ucode_ap();
+
+	/*
+	 * Synchronization point with the hotplug core. Sets this CPUs
+	 * synchronization state to ALIVE and spin-waits for the control CPU to
 	 * release this CPU for further bringup.
 	 */
 	cpuhp_ap_sync_alive();
@@ -924,9 +951,9 @@ static int wakeup_secondary_cpu_via_init
 /* reduce the number of lines printed when booting a large cpu count system */
 static void announce_cpu(int cpu, int apicid)
 {
+	static int width, node_width, first = 1;
 	static int current_node = NUMA_NO_NODE;
 	int node = early_cpu_to_node(cpu);
-	static int width, node_width;
 
 	if (!width)
 		width = num_digits(num_possible_cpus()) + 1; /* + '#' sign */
@@ -934,10 +961,10 @@ static void announce_cpu(int cpu, int ap
 	if (!node_width)
 		node_width = num_digits(num_possible_nodes()) + 1; /* + '#' */
 
-	if (cpu == 1)
-		printk(KERN_INFO "x86: Booting SMP configuration:\n");
-
 	if (system_state < SYSTEM_RUNNING) {
+		if (first)
+			pr_info("x86: Booting SMP configuration:\n");
+
 		if (node != current_node) {
 			if (current_node > (-1))
 				pr_cont("\n");
@@ -948,11 +975,11 @@ static void announce_cpu(int cpu, int ap
 		}
 
 		/* Add padding for the BSP */
-		if (cpu == 1)
+		if (first)
 			pr_cont("%*s", width + 1, " ");
+		first = 0;
 
 		pr_cont("%*s#%d", width - num_digits(cpu), " ", cpu);
-
 	} else
 		pr_info("Booting Node %d Processor %d APIC 0x%x\n",
 			node, cpu, apicid);
@@ -1242,6 +1269,36 @@ void __init smp_prepare_cpus_common(void
 	set_cpu_sibling_map(0);
 }
 
+#ifdef CONFIG_X86_64
+/* Establish whether parallel bringup can be supported. */
+bool __init arch_cpuhp_init_parallel_bringup(void)
+{
+	/*
+	 * Encrypted guests require special handling. They enforce X2APIC
+	 * mode but the RDMSR to read the APIC ID is intercepted and raises
+	 * #VC or #VE which cannot be handled in the early startup code.
+	 *
+	 * AMD-SEV does not provide a RDMSR GHCB protocol so the early
+	 * startup code cannot directly communicate with the secure
+	 * firmware. The alternative solution to retrieve the APIC ID via
+	 * CPUID(0xb), which is covered by the GHCB protocol, is not viable
+	 * either because there is no enforcement of the CPUID(0xb)
+	 * provided "initial" APIC ID to be the same as the real APIC ID.
+	 *
+	 * Intel-TDX has a secure RDMSR hypercall, but that needs to be
+	 * implemented seperately in the low level startup ASM code.
+	 */
+	if (cc_platform_has(CC_ATTR_GUEST_STATE_ENCRYPT)) {
+		pr_info("Parallel CPU startup disabled due to guest state encryption\n");
+		return false;
+	}
+
+	smpboot_control = STARTUP_READ_APICID;
+	pr_debug("Parallel CPU startup enabled: 0x%08x\n", smpboot_control);
+	return true;
+}
+#endif
+
 /*
  * Prepare for SMP bootup.
  * @max_cpus: configured maximum number of CPUs, It is a legacy parameter