[PATCH v4 0/2] Enable PV qspinlock for Hyper-V

[PATCH v4 0/2] Enable PV qspinlock for Hyper-V

[Yi Sun]

v3->v4:
    - To avoid possible hang issue (if hv_qlock_wait can be
      interrupted, kick may happen before reading HV_X64_MSR_GUEST_IDLE
      so that hang happens), disable interrupt before READ_ONCE, then
      restore it after reading HV_X64_MSR_GUEST_IDLE.
v2->v3:
    - use "Hyper-V: " as the message prefix
    - remove unnecessary header files
    - remove unnecessary check in 'hv_qlock_wait'
    - fix compilation error on different platforms

v1->v2:
    - compile hv_spinlock.c only when CONFIG_PARAVIRT_SPINLOCKS enabled
    - merge v1 patch 2/3 to single patch
    - remove part of the boilerplate in hv_spinlock.c
    - declare hv_pvspin as __initdata
    - remove spin_wait_info and hv_notify_long_spin_wait because
      SpinWaitInfo is a standalone feature.
    - add comments for reading HV_X64_MSR_GUEST_IDLE
    - replace pr_warn to pr_info
    - use pr_fmt instead of the 'hv:' prefix
    - register callback function for smp_ops.smp_prepare_boot_cpu
      to initialize hyper-v spinlock

This patch adds the necessary Hyper-V specific code to allow
PV qspinlock work on Hyper-V.

In wait callback function, read HV_X64_MSR_GUEST_IDLE MSR
to trigger the guest's transition to the idle power state
which can be exited by an IPI even if IF flag is disabled.

In kick callback function, just send platform IPI to make
waiting vcpu exit idle state.

In vcpu_is_preempted callback function, return false directly
because Hyper-V does not provide such interface so far.


Cc: "K. Y. Srinivasan" <kys@microsoft.com>
Cc: Haiyang Zhang <haiyangz@microsoft.com>
Cc: Stephen Hemminger <sthemmin@microsoft.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Michael Kelley (EOSG) <Michael.H.Kelley@microsoft.com>
Cc: Juergen Gross <jgross@suse.com>

Yi Sun (2):
  X86/Hyper-V: Add Guest IDLE MSR support
  locking/pvqspinlock, hv: Enable PV qspinlock for Hyper-V

 Documentation/admin-guide/kernel-parameters.txt |  5 ++
 arch/x86/hyperv/Makefile                        |  4 ++
 arch/x86/hyperv/hv_spinlock.c                   | 85 +++++++++++++++++++++++++
 arch/x86/include/asm/hyperv-tlfs.h              |  5 ++
 arch/x86/include/asm/mshyperv.h                 |  1 +
 arch/x86/kernel/cpu/mshyperv.c                  | 14 ++++
 6 files changed, 114 insertions(+)
 create mode 100644 arch/x86/hyperv/hv_spinlock.c

-- 
1.9.1

[PATCH v4 1/2] X86/Hyper-V: Add Guest IDLE MSR support

[Yi Sun]

Hyper-V may expose a HV_X64_MSR_GUEST_IDLE MSR. We can read
HYPERV_CPUID_FEATURES eax to check it. Read this MSR can
trigger the guest's transition to the idle power state which
can be exited by an IPI even if IF flag is disabled.

Cc: "K. Y. Srinivasan" <kys@microsoft.com>
Cc: Haiyang Zhang <haiyangz@microsoft.com>
Cc: Stephen Hemminger <sthemmin@microsoft.com>
Signed-off-by: Yi Sun <yi.y.sun@linux.intel.com>
Reviewed-by: Michael Kelley <mikelley@microsoft.com>
---
 arch/x86/include/asm/hyperv-tlfs.h | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/arch/x86/include/asm/hyperv-tlfs.h b/arch/x86/include/asm/hyperv-tlfs.h
index 00e01d2..4139f76 100644
--- a/arch/x86/include/asm/hyperv-tlfs.h
+++ b/arch/x86/include/asm/hyperv-tlfs.h
@@ -38,6 +38,8 @@
 #define HV_MSR_TIME_REF_COUNT_AVAILABLE		(1 << 1)
 /* Partition reference TSC MSR is available */
 #define HV_MSR_REFERENCE_TSC_AVAILABLE		(1 << 9)
+/* Partition Guest IDLE MSR is available */
+#define HV_X64_MSR_GUEST_IDLE_AVAILABLE		(1 << 10)
 
 /* A partition's reference time stamp counter (TSC) page */
 #define HV_X64_MSR_REFERENCE_TSC		0x40000021
@@ -246,6 +248,9 @@
 #define HV_X64_MSR_STIMER3_CONFIG		0x400000B6
 #define HV_X64_MSR_STIMER3_COUNT		0x400000B7
 
+/* Hyper-V guest idle MSR */
+#define HV_X64_MSR_GUEST_IDLE			0x400000F0
+
 /* Hyper-V guest crash notification MSR's */
 #define HV_X64_MSR_CRASH_P0			0x40000100
 #define HV_X64_MSR_CRASH_P1			0x40000101
-- 
1.9.1

[PATCH v4 2/2] locking/pvqspinlock, hv: Enable PV qspinlock for Hyper-V

[Yi Sun]

Follow PV spinlock mechanism to implement the callback functions
to allow the CPU idling and kicking operations on Hyper-V.

Cc: "K. Y. Srinivasan" <kys@microsoft.com>
Cc: Haiyang Zhang <haiyangz@microsoft.com>
Cc: Stephen Hemminger <sthemmin@microsoft.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Michael Kelley (EOSG) <Michael.H.Kelley@microsoft.com>
Cc: Juergen Gross <jgross@suse.com>
Signed-off-by: Yi Sun <yi.y.sun@linux.intel.com>
---
v3->v4:
    - To avoid possible hang issue (if hv_qlock_wait can be
      interrupted, kick may happen before reading HV_X64_MSR_GUEST_IDLE
      so that hang happens), disable interrupt before READ_ONCE, then
      restore it after reading HV_X64_MSR_GUEST_IDLE.
      (suggested by Juergen Gross)
---
 Documentation/admin-guide/kernel-parameters.txt |  5 ++
 arch/x86/hyperv/Makefile                        |  4 ++
 arch/x86/hyperv/hv_spinlock.c                   | 85 +++++++++++++++++++++++++
 arch/x86/include/asm/mshyperv.h                 |  1 +
 arch/x86/kernel/cpu/mshyperv.c                  | 14 ++++
 5 files changed, 109 insertions(+)
 create mode 100644 arch/x86/hyperv/hv_spinlock.c

diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
index 92eb1f4..0fc8448 100644
--- a/Documentation/admin-guide/kernel-parameters.txt
+++ b/Documentation/admin-guide/kernel-parameters.txt
@@ -1385,6 +1385,11 @@
 	hvc_iucv_allow=	[S390]	Comma-separated list of z/VM user IDs.
 				If specified, z/VM IUCV HVC accepts connections
 				from listed z/VM user IDs only.
+
+	hv_nopvspin	[X86,HYPER_V]
+			Disables the ticketlock slowpath using HYPER-V PV
+			optimizations.
+
 	keep_bootcon	[KNL]
 			Do not unregister boot console at start. This is only
 			useful for debugging when something happens in the window
diff --git a/arch/x86/hyperv/Makefile b/arch/x86/hyperv/Makefile
index b21ee65..1c11f94 100644
--- a/arch/x86/hyperv/Makefile
+++ b/arch/x86/hyperv/Makefile
@@ -1,2 +1,6 @@
 obj-y			:= hv_init.o mmu.o nested.o
 obj-$(CONFIG_X86_64)	+= hv_apic.o
+
+ifdef CONFIG_X86_64
+obj-$(CONFIG_PARAVIRT_SPINLOCKS)	+= hv_spinlock.o
+endif
diff --git a/arch/x86/hyperv/hv_spinlock.c b/arch/x86/hyperv/hv_spinlock.c
new file mode 100644
index 0000000..b628343
--- /dev/null
+++ b/arch/x86/hyperv/hv_spinlock.c
@@ -0,0 +1,85 @@
+// SPDX-License-Identifier: GPL-2.0
+
+/*
+ * Hyper-V specific spinlock code.
+ *
+ * Copyright (C) 2018, Intel, Inc.
+ *
+ * Author : Yi Sun <yi.y.sun@intel.com>
+ */
+
+#define pr_fmt(fmt) "Hyper-V: " fmt
+
+#include <linux/spinlock.h>
+
+#include <asm/mshyperv.h>
+#include <asm/hyperv-tlfs.h>
+#include <asm/paravirt.h>
+#include <asm/qspinlock.h>
+#include <asm/apic.h>
+
+static bool __initdata hv_pvspin = true;
+
+static void hv_qlock_kick(int cpu)
+{
+	apic->send_IPI(cpu, X86_PLATFORM_IPI_VECTOR);
+}
+
+static void hv_qlock_wait(u8 *byte, u8 val)
+{
+	unsigned long msr_val;
+	unsigned long flags;
+
+	if (in_nmi())
+		return;
+
+	local_irq_save(flags);
+
+	if (READ_ONCE(*byte) != val)
+		goto out;
+
+	/*
+	 * Read HV_X64_MSR_GUEST_IDLE MSR can trigger the guest's
+	 * transition to the idle power state which can be exited
+	 * by an IPI even if IF flag is disabled.
+	 */
+	rdmsrl(HV_X64_MSR_GUEST_IDLE, msr_val);
+
+out:
+	local_irq_restore(flags);
+}
+
+/*
+ * Hyper-V does not support this so far.
+ */
+bool hv_vcpu_is_preempted(int vcpu)
+{
+	return false;
+}
+PV_CALLEE_SAVE_REGS_THUNK(hv_vcpu_is_preempted);
+
+void __init hv_init_spinlocks(void)
+{
+	if (!hv_pvspin ||
+	    !apic ||
+	    !(ms_hyperv.hints & HV_X64_CLUSTER_IPI_RECOMMENDED) ||
+	    !(ms_hyperv.features & HV_X64_MSR_GUEST_IDLE_AVAILABLE)) {
+		pr_info("PV spinlocks disabled\n");
+		return;
+	}
+	pr_info("PV spinlocks enabled\n");
+
+	__pv_init_lock_hash();
+	pv_lock_ops.queued_spin_lock_slowpath = __pv_queued_spin_lock_slowpath;
+	pv_lock_ops.queued_spin_unlock = PV_CALLEE_SAVE(__pv_queued_spin_unlock);
+	pv_lock_ops.wait = hv_qlock_wait;
+	pv_lock_ops.kick = hv_qlock_kick;
+	pv_lock_ops.vcpu_is_preempted = PV_CALLEE_SAVE(hv_vcpu_is_preempted);
+}
+
+static __init int hv_parse_nopvspin(char *arg)
+{
+	hv_pvspin = false;
+	return 0;
+}
+early_param("hv_nopvspin", hv_parse_nopvspin);
diff --git a/arch/x86/include/asm/mshyperv.h b/arch/x86/include/asm/mshyperv.h
index f377044..759cfd2 100644
--- a/arch/x86/include/asm/mshyperv.h
+++ b/arch/x86/include/asm/mshyperv.h
@@ -351,6 +351,7 @@ static inline int cpumask_to_vpset(struct hv_vpset *vpset,
 
 #ifdef CONFIG_X86_64
 void hv_apic_init(void);
+void __init hv_init_spinlocks(void);
 #else
 static inline void hv_apic_init(void) {}
 #endif
diff --git a/arch/x86/kernel/cpu/mshyperv.c b/arch/x86/kernel/cpu/mshyperv.c
index ad12733..a5cc219 100644
--- a/arch/x86/kernel/cpu/mshyperv.c
+++ b/arch/x86/kernel/cpu/mshyperv.c
@@ -199,6 +199,16 @@ static unsigned long hv_get_tsc_khz(void)
 	return freq / 1000;
 }
 
+#if defined(CONFIG_SMP) && IS_ENABLED(CONFIG_HYPERV)
+static void __init hv_smp_prepare_boot_cpu(void)
+{
+	native_smp_prepare_boot_cpu();
+#if defined(CONFIG_X86_64) && defined(CONFIG_PARAVIRT_SPINLOCKS)
+	hv_init_spinlocks();
+#endif
+}
+#endif
+
 static void __init ms_hyperv_init_platform(void)
 {
 	int hv_host_info_eax;
@@ -303,6 +313,10 @@ static void __init ms_hyperv_init_platform(void)
 	if (ms_hyperv.misc_features & HV_STIMER_DIRECT_MODE_AVAILABLE)
 		alloc_intr_gate(HYPERV_STIMER0_VECTOR,
 				hv_stimer0_callback_vector);
+
+#if defined(CONFIG_SMP)
+	smp_ops.smp_prepare_boot_cpu = hv_smp_prepare_boot_cpu;
+#endif
 #endif
 }
 
-- 
1.9.1

Re: [PATCH v4 2/2] locking/pvqspinlock, hv: Enable PV qspinlock for Hyper-V

[Juergen Gross]

On 08/10/2018 10:29, Yi Sun wrote:
> Follow PV spinlock mechanism to implement the callback functions
> to allow the CPU idling and kicking operations on Hyper-V.
> 
> Cc: "K. Y. Srinivasan" <kys@microsoft.com>
> Cc: Haiyang Zhang <haiyangz@microsoft.com>
> Cc: Stephen Hemminger <sthemmin@microsoft.com>
> Cc: Thomas Gleixner <tglx@linutronix.de>
> Cc: Michael Kelley (EOSG) <Michael.H.Kelley@microsoft.com>
> Cc: Juergen Gross <jgross@suse.com>
> Signed-off-by: Yi Sun <yi.y.sun@linux.intel.com>

Reviewed-by: Juergen Gross <jgross@suse.com>


Juergen

[tip:x86/hyperv] x86/hyperv: Enable PV qspinlock for Hyper-V

[tip-bot for Yi Sun]

Commit-ID:  b36bcb7ce0b353bd4889e06b61e036744b02ca51
Gitweb:     https://git.kernel.org/tip/b36bcb7ce0b353bd4889e06b61e036744b02ca51
Author:     Yi Sun <yi.y.sun@linux.intel.com>
AuthorDate: Mon, 8 Oct 2018 16:29:34 +0800
Committer:  Thomas Gleixner <tglx@linutronix.de>
CommitDate: Tue, 9 Oct 2018 11:26:31 +0200

x86/hyperv: Enable PV qspinlock for Hyper-V

Implement the required wait and kick callbacks to support PV spinlocks in
Hyper-V guests.

[ tglx: Document the requirement for disabling interrupts in the wait()
  	callback. Remove goto and unnecessary includes. Add prototype
	for hv_vcpu_is_preempted(). ]

Signed-off-by: Yi Sun <yi.y.sun@linux.intel.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Juergen Gross <jgross@suse.com>
Cc: "K. Y. Srinivasan" <kys@microsoft.com>
Cc: Haiyang Zhang <haiyangz@microsoft.com>
Cc: Stephen Hemminger <sthemmin@microsoft.com>
Cc: Michael Kelley (EOSG) <Michael.H.Kelley@microsoft.com>
Cc: chao.p.peng@intel.com
Cc: chao.gao@intel.com
Cc: isaku.yamahata@intel.com
Cc: tianyu.lan@microsoft.com
Link: https://lkml.kernel.org/r/1538987374-51217-3-git-send-email-yi.y.sun@linux.intel.com

---
 Documentation/admin-guide/kernel-parameters.txt |  5 ++
 arch/x86/hyperv/Makefile                        |  4 ++
 arch/x86/hyperv/hv_spinlock.c                   | 88 +++++++++++++++++++++++++
 arch/x86/include/asm/mshyperv.h                 |  2 +
 arch/x86/kernel/cpu/mshyperv.c                  | 14 ++++
 5 files changed, 113 insertions(+)

diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
index 92eb1f42240d..ca6ca93e8281 100644
--- a/Documentation/admin-guide/kernel-parameters.txt
+++ b/Documentation/admin-guide/kernel-parameters.txt
@@ -1385,6 +1385,11 @@
 	hvc_iucv_allow=	[S390]	Comma-separated list of z/VM user IDs.
 				If specified, z/VM IUCV HVC accepts connections
 				from listed z/VM user IDs only.
+
+	hv_nopvspin	[X86,HYPER_V] Disables the paravirt spinlock optimizations
+				      which allow the hypervisor to 'idle' the
+				      guest on lock contention.
+
 	keep_bootcon	[KNL]
 			Do not unregister boot console at start. This is only
 			useful for debugging when something happens in the window
diff --git a/arch/x86/hyperv/Makefile b/arch/x86/hyperv/Makefile
index b21ee65c4101..1c11f9420a82 100644
--- a/arch/x86/hyperv/Makefile
+++ b/arch/x86/hyperv/Makefile
@@ -1,2 +1,6 @@
 obj-y			:= hv_init.o mmu.o nested.o
 obj-$(CONFIG_X86_64)	+= hv_apic.o
+
+ifdef CONFIG_X86_64
+obj-$(CONFIG_PARAVIRT_SPINLOCKS)	+= hv_spinlock.o
+endif
diff --git a/arch/x86/hyperv/hv_spinlock.c b/arch/x86/hyperv/hv_spinlock.c
new file mode 100644
index 000000000000..2b6382cbb769
--- /dev/null
+++ b/arch/x86/hyperv/hv_spinlock.c
@@ -0,0 +1,88 @@
+// SPDX-License-Identifier: GPL-2.0
+
+/*
+ * Hyper-V specific spinlock code.
+ *
+ * Copyright (C) 2018, Intel, Inc.
+ *
+ * Author : Yi Sun <yi.y.sun@intel.com>
+ */
+
+#define pr_fmt(fmt) "Hyper-V: " fmt
+
+#include <linux/spinlock.h>
+
+#include <asm/mshyperv.h>
+#include <asm/paravirt.h>
+#include <asm/apic.h>
+
+static bool __initdata hv_pvspin = true;
+
+static void hv_qlock_kick(int cpu)
+{
+	apic->send_IPI(cpu, X86_PLATFORM_IPI_VECTOR);
+}
+
+static void hv_qlock_wait(u8 *byte, u8 val)
+{
+	unsigned long msr_val;
+	unsigned long flags;
+
+	if (in_nmi())
+		return;
+
+	/*
+	 * Reading HV_X64_MSR_GUEST_IDLE MSR tells the hypervisor that the
+	 * vCPU can be put into 'idle' state. This 'idle' state is
+	 * terminated by an IPI, usually from hv_qlock_kick(), even if
+	 * interrupts are disabled on the vCPU.
+	 *
+	 * To prevent a race against the unlock path it is required to
+	 * disable interrupts before accessing the HV_X64_MSR_GUEST_IDLE
+	 * MSR. Otherwise, if the IPI from hv_qlock_kick() arrives between
+	 * the lock value check and the rdmsrl() then the vCPU might be put
+	 * into 'idle' state by the hypervisor and kept in that state for
+	 * an unspecified amount of time.
+	 */
+	local_irq_save(flags);
+	/*
+	 * Only issue the rdmsrl() when the lock state has not changed.
+	 */
+	if (READ_ONCE(*byte) == val)
+		rdmsrl(HV_X64_MSR_GUEST_IDLE, msr_val);
+	local_irq_restore(flags);
+}
+
+/*
+ * Hyper-V does not support this so far.
+ */
+bool hv_vcpu_is_preempted(int vcpu)
+{
+	return false;
+}
+PV_CALLEE_SAVE_REGS_THUNK(hv_vcpu_is_preempted);
+
+void __init hv_init_spinlocks(void)
+{
+	if (!hv_pvspin || !apic ||
+	    !(ms_hyperv.hints & HV_X64_CLUSTER_IPI_RECOMMENDED) ||
+	    !(ms_hyperv.features & HV_X64_MSR_GUEST_IDLE_AVAILABLE)) {
+		pr_info("PV spinlocks disabled\n");
+		return;
+	}
+	pr_info("PV spinlocks enabled\n");
+
+	__pv_init_lock_hash();
+	pv_lock_ops.queued_spin_lock_slowpath = __pv_queued_spin_lock_slowpath;
+	pv_lock_ops.queued_spin_unlock = PV_CALLEE_SAVE(__pv_queued_spin_unlock);
+	pv_lock_ops.wait = hv_qlock_wait;
+	pv_lock_ops.kick = hv_qlock_kick;
+	pv_lock_ops.vcpu_is_preempted = PV_CALLEE_SAVE(hv_vcpu_is_preempted);
+}
+
+static __init int hv_parse_nopvspin(char *arg)
+{
+	hv_pvspin = false;
+	return 0;
+}
+early_param("hv_nopvspin", hv_parse_nopvspin);
diff --git a/arch/x86/include/asm/mshyperv.h b/arch/x86/include/asm/mshyperv.h
index f37704497d8f..0d6271cce198 100644
--- a/arch/x86/include/asm/mshyperv.h
+++ b/arch/x86/include/asm/mshyperv.h
@@ -351,6 +351,8 @@ int hyperv_flush_guest_mapping(u64 as);
 
 #ifdef CONFIG_X86_64
 void hv_apic_init(void);
+void __init hv_init_spinlocks(void);
+bool hv_vcpu_is_preempted(int vcpu);
 #else
 static inline void hv_apic_init(void) {}
 #endif
diff --git a/arch/x86/kernel/cpu/mshyperv.c b/arch/x86/kernel/cpu/mshyperv.c
index ad12733f6058..2179c7efc1f4 100644
--- a/arch/x86/kernel/cpu/mshyperv.c
+++ b/arch/x86/kernel/cpu/mshyperv.c
@@ -199,6 +199,16 @@ static unsigned long hv_get_tsc_khz(void)
 	return freq / 1000;
 }
 
+#if defined(CONFIG_SMP) && IS_ENABLED(CONFIG_HYPERV)
+static void __init hv_smp_prepare_boot_cpu(void)
+{
+	native_smp_prepare_boot_cpu();
+#if defined(CONFIG_X86_64) && defined(CONFIG_PARAVIRT_SPINLOCKS)
+	hv_init_spinlocks();
+#endif
+}
+#endif
+
 static void __init ms_hyperv_init_platform(void)
 {
 	int hv_host_info_eax;
@@ -303,6 +313,10 @@ static void __init ms_hyperv_init_platform(void)
 	if (ms_hyperv.misc_features & HV_STIMER_DIRECT_MODE_AVAILABLE)
 		alloc_intr_gate(HYPERV_STIMER0_VECTOR,
 				hv_stimer0_callback_vector);
+
+# if defined(CONFIG_SMP)
+	smp_ops.smp_prepare_boot_cpu = hv_smp_prepare_boot_cpu;
+# endif
 #endif
 }
 

Re: [PATCH v4 2/2] locking/pvqspinlock, hv: Enable PV qspinlock for Hyper-V

[Ingo Molnar]

* Yi Sun <yi.y.sun@linux.intel.com> wrote:

> Follow PV spinlock mechanism to implement the callback functions
> to allow the CPU idling and kicking operations on Hyper-V.

> +#if defined(CONFIG_SMP)
> +	smp_ops.smp_prepare_boot_cpu = hv_smp_prepare_boot_cpu;
> +#endif

What's wrong with using the common pattern of:

	#ifdef CONFIG_SMP

?

Thanks,

	Ingo

[tip:x86/paravirt] x86/hyperv: Enable PV qspinlock for Hyper-V

[tip-bot for Yi Sun]

Commit-ID:  3a025de64bf89c84a79909069e3c24ad9e710d27
Gitweb:     https://git.kernel.org/tip/3a025de64bf89c84a79909069e3c24ad9e710d27
Author:     Yi Sun <yi.y.sun@linux.intel.com>
AuthorDate: Mon, 8 Oct 2018 16:29:34 +0800
Committer:  Thomas Gleixner <tglx@linutronix.de>
CommitDate: Tue, 9 Oct 2018 14:21:39 +0200

x86/hyperv: Enable PV qspinlock for Hyper-V

Implement the required wait and kick callbacks to support PV spinlocks in
Hyper-V guests.

[ tglx: Document the requirement for disabling interrupts in the wait()
  	callback. Remove goto and unnecessary includes. Add prototype
	for hv_vcpu_is_preempted(). Adapted to pending paravirt changes. ]

Signed-off-by: Yi Sun <yi.y.sun@linux.intel.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Juergen Gross <jgross@suse.com>
Cc: "K. Y. Srinivasan" <kys@microsoft.com>
Cc: Haiyang Zhang <haiyangz@microsoft.com>
Cc: Stephen Hemminger <sthemmin@microsoft.com>
Cc: Michael Kelley (EOSG) <Michael.H.Kelley@microsoft.com>
Cc: chao.p.peng@intel.com
Cc: chao.gao@intel.com
Cc: isaku.yamahata@intel.com
Cc: tianyu.lan@microsoft.com
Link: https://lkml.kernel.org/r/1538987374-51217-3-git-send-email-yi.y.sun@linux.intel.com
---
 Documentation/admin-guide/kernel-parameters.txt |  5 ++
 arch/x86/hyperv/Makefile                        |  4 ++
 arch/x86/hyperv/hv_spinlock.c                   | 88 +++++++++++++++++++++++++
 arch/x86/include/asm/mshyperv.h                 |  2 +
 arch/x86/kernel/cpu/mshyperv.c                  | 14 ++++
 5 files changed, 113 insertions(+)

diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
index 9871e649ffef..4cdf6a673592 100644
--- a/Documentation/admin-guide/kernel-parameters.txt
+++ b/Documentation/admin-guide/kernel-parameters.txt
@@ -1385,6 +1385,11 @@
 	hvc_iucv_allow=	[S390]	Comma-separated list of z/VM user IDs.
 				If specified, z/VM IUCV HVC accepts connections
 				from listed z/VM user IDs only.
+
+	hv_nopvspin	[X86,HYPER_V] Disables the paravirt spinlock optimizations
+				      which allow the hypervisor to 'idle' the
+				      guest on lock contention.
+
 	keep_bootcon	[KNL]
 			Do not unregister boot console at start. This is only
 			useful for debugging when something happens in the window
diff --git a/arch/x86/hyperv/Makefile b/arch/x86/hyperv/Makefile
index b21ee65c4101..1c11f9420a82 100644
--- a/arch/x86/hyperv/Makefile
+++ b/arch/x86/hyperv/Makefile
@@ -1,2 +1,6 @@
 obj-y			:= hv_init.o mmu.o nested.o
 obj-$(CONFIG_X86_64)	+= hv_apic.o
+
+ifdef CONFIG_X86_64
+obj-$(CONFIG_PARAVIRT_SPINLOCKS)	+= hv_spinlock.o
+endif
diff --git a/arch/x86/hyperv/hv_spinlock.c b/arch/x86/hyperv/hv_spinlock.c
new file mode 100644
index 000000000000..a861b0456b1a
--- /dev/null
+++ b/arch/x86/hyperv/hv_spinlock.c
@@ -0,0 +1,88 @@
+// SPDX-License-Identifier: GPL-2.0
+
+/*
+ * Hyper-V specific spinlock code.
+ *
+ * Copyright (C) 2018, Intel, Inc.
+ *
+ * Author : Yi Sun <yi.y.sun@intel.com>
+ */
+
+#define pr_fmt(fmt) "Hyper-V: " fmt
+
+#include <linux/spinlock.h>
+
+#include <asm/mshyperv.h>
+#include <asm/paravirt.h>
+#include <asm/apic.h>
+
+static bool __initdata hv_pvspin = true;
+
+static void hv_qlock_kick(int cpu)
+{
+	apic->send_IPI(cpu, X86_PLATFORM_IPI_VECTOR);
+}
+
+static void hv_qlock_wait(u8 *byte, u8 val)
+{
+	unsigned long msr_val;
+	unsigned long flags;
+
+	if (in_nmi())
+		return;
+
+	/*
+	 * Reading HV_X64_MSR_GUEST_IDLE MSR tells the hypervisor that the
+	 * vCPU can be put into 'idle' state. This 'idle' state is
+	 * terminated by an IPI, usually from hv_qlock_kick(), even if
+	 * interrupts are disabled on the vCPU.
+	 *
+	 * To prevent a race against the unlock path it is required to
+	 * disable interrupts before accessing the HV_X64_MSR_GUEST_IDLE
+	 * MSR. Otherwise, if the IPI from hv_qlock_kick() arrives between
+	 * the lock value check and the rdmsrl() then the vCPU might be put
+	 * into 'idle' state by the hypervisor and kept in that state for
+	 * an unspecified amount of time.
+	 */
+	local_irq_save(flags);
+	/*
+	 * Only issue the rdmsrl() when the lock state has not changed.
+	 */
+	if (READ_ONCE(*byte) == val)
+		rdmsrl(HV_X64_MSR_GUEST_IDLE, msr_val);
+	local_irq_restore(flags);
+}
+
+/*
+ * Hyper-V does not support this so far.
+ */
+bool hv_vcpu_is_preempted(int vcpu)
+{
+	return false;
+}
+PV_CALLEE_SAVE_REGS_THUNK(hv_vcpu_is_preempted);
+
+void __init hv_init_spinlocks(void)
+{
+	if (!hv_pvspin || !apic ||
+	    !(ms_hyperv.hints & HV_X64_CLUSTER_IPI_RECOMMENDED) ||
+	    !(ms_hyperv.features & HV_X64_MSR_GUEST_IDLE_AVAILABLE)) {
+		pr_info("PV spinlocks disabled\n");
+		return;
+	}
+	pr_info("PV spinlocks enabled\n");
+
+	__pv_init_lock_hash();
+	pv_ops.lock.queued_spin_lock_slowpath = __pv_queued_spin_lock_slowpath;
+	pv_ops.lock.queued_spin_unlock = PV_CALLEE_SAVE(__pv_queued_spin_unlock);
+	pv_ops.lock.wait = hv_qlock_wait;
+	pv_ops.lock.kick = hv_qlock_kick;
+	pv_ops.lock.vcpu_is_preempted = PV_CALLEE_SAVE(hv_vcpu_is_preempted);
+}
+
+static __init int hv_parse_nopvspin(char *arg)
+{
+	hv_pvspin = false;
+	return 0;
+}
+early_param("hv_nopvspin", hv_parse_nopvspin);
diff --git a/arch/x86/include/asm/mshyperv.h b/arch/x86/include/asm/mshyperv.h
index f37704497d8f..0d6271cce198 100644
--- a/arch/x86/include/asm/mshyperv.h
+++ b/arch/x86/include/asm/mshyperv.h
@@ -351,6 +351,8 @@ int hyperv_flush_guest_mapping(u64 as);
 
 #ifdef CONFIG_X86_64
 void hv_apic_init(void);
+void __init hv_init_spinlocks(void);
+bool hv_vcpu_is_preempted(int vcpu);
 #else
 static inline void hv_apic_init(void) {}
 #endif
diff --git a/arch/x86/kernel/cpu/mshyperv.c b/arch/x86/kernel/cpu/mshyperv.c
index ad12733f6058..1c72f3819eb1 100644
--- a/arch/x86/kernel/cpu/mshyperv.c
+++ b/arch/x86/kernel/cpu/mshyperv.c
@@ -199,6 +199,16 @@ static unsigned long hv_get_tsc_khz(void)
 	return freq / 1000;
 }
 
+#if defined(CONFIG_SMP) && IS_ENABLED(CONFIG_HYPERV)
+static void __init hv_smp_prepare_boot_cpu(void)
+{
+	native_smp_prepare_boot_cpu();
+#if defined(CONFIG_X86_64) && defined(CONFIG_PARAVIRT_SPINLOCKS)
+	hv_init_spinlocks();
+#endif
+}
+#endif
+
 static void __init ms_hyperv_init_platform(void)
 {
 	int hv_host_info_eax;
@@ -303,6 +313,10 @@ static void __init ms_hyperv_init_platform(void)
 	if (ms_hyperv.misc_features & HV_STIMER_DIRECT_MODE_AVAILABLE)
 		alloc_intr_gate(HYPERV_STIMER0_VECTOR,
 				hv_stimer0_callback_vector);
+
+# ifdef CONFIG_SMP
+	smp_ops.smp_prepare_boot_cpu = hv_smp_prepare_boot_cpu;
+# endif
 #endif
 }
 

Re: [PATCH v4 2/2] locking/pvqspinlock, hv: Enable PV qspinlock for Hyper-V

[Yi Sun]

On 18-10-09 12:54:27, Ingo Molnar wrote:
> 
> * Yi Sun <yi.y.sun@linux.intel.com> wrote:
> 
> > Follow PV spinlock mechanism to implement the callback functions
> > to allow the CPU idling and kicking operations on Hyper-V.
> 
> > +#if defined(CONFIG_SMP)
> > +	smp_ops.smp_prepare_boot_cpu = hv_smp_prepare_boot_cpu;
> > +#endif
> 
> What's wrong with using the common pattern of:
> 
> 	#ifdef CONFIG_SMP
> 
> ?

There is no difference between "#ifdef" and "#if defined". I just copied
it from "hv_smp_prepare_boot_cpu()". Do you need me submit a fix patch
after this set is merged?

> 
> Thanks,
> 
> 	Ingo

Re: [PATCH v4 2/2] locking/pvqspinlock, hv: Enable PV qspinlock for Hyper-V

[Ingo Molnar]

* Yi Sun <yi.y.sun@linux.intel.com> wrote:

> On 18-10-09 12:54:27, Ingo Molnar wrote:
> > 
> > * Yi Sun <yi.y.sun@linux.intel.com> wrote:
> > 
> > > Follow PV spinlock mechanism to implement the callback functions
> > > to allow the CPU idling and kicking operations on Hyper-V.
> > 
> > > +#if defined(CONFIG_SMP)
> > > +	smp_ops.smp_prepare_boot_cpu = hv_smp_prepare_boot_cpu;
> > > +#endif
> > 
> > What's wrong with using the common pattern of:
> > 
> > 	#ifdef CONFIG_SMP
> > 
> > ?
> 
> There is no difference between "#ifdef" and "#if defined". I just copied
> it from "hv_smp_prepare_boot_cpu()". Do you need me submit a fix patch
> after this set is merged?

It's equivalent code, my point was that '#ifdef CONFIG_XYZ' is the shorter, more canonical 
pattern we use in the kernel most of the time, it's shorter, easier to read. We only use 
defined() for longer preprocessor directive conditions, and it's a pattern for "there's 
something more complex here than a simple CONFIG_XYZ dependency".

Anyway, Thomas fixed it up in the latest iteration.

Thanks,

	Ingo

Re: [PATCH v4 2/2] locking/pvqspinlock, hv: Enable PV qspinlock for Hyper-V

[Yi Sun]

On 18-10-10 07:50:26, Ingo Molnar wrote:
> 
> * Yi Sun <yi.y.sun@linux.intel.com> wrote:
> 
> > On 18-10-09 12:54:27, Ingo Molnar wrote:
> > > 
> > > * Yi Sun <yi.y.sun@linux.intel.com> wrote:
> > > 
> > > > Follow PV spinlock mechanism to implement the callback functions
> > > > to allow the CPU idling and kicking operations on Hyper-V.
> > > 
> > > > +#if defined(CONFIG_SMP)
> > > > +	smp_ops.smp_prepare_boot_cpu = hv_smp_prepare_boot_cpu;
> > > > +#endif
> > > 
> > > What's wrong with using the common pattern of:
> > > 
> > > 	#ifdef CONFIG_SMP
> > > 
> > > ?
> > 
> > There is no difference between "#ifdef" and "#if defined". I just copied
> > it from "hv_smp_prepare_boot_cpu()". Do you need me submit a fix patch
> > after this set is merged?
> 
> It's equivalent code, my point was that '#ifdef CONFIG_XYZ' is the shorter, more canonical 
> pattern we use in the kernel most of the time, it's shorter, easier to read. We only use 
> defined() for longer preprocessor directive conditions, and it's a pattern for "there's 
> something more complex here than a simple CONFIG_XYZ dependency".
> 
> Anyway, Thomas fixed it up in the latest iteration.
> 
> Thanks,
> 
> 	Ingo

Thank you! I will notice it in the future.

BRs,
Yi Sun