[RFC PATCH V3 4/4] RISC-V: paravirt: Support nopvspin to disable PARAVIRT_SPINLOCKS

[guoren@kernel.org]

From: "Guo Ren (Alibaba DAMO Academy)" <guoren@kernel.org>

The VM guests should fall back to a Test-and-Set spinlock when
PARAVIRT_SPINLOCKS disabled, because fair locks have horrible lock
'holder' preemption issues. The virt_spin_lock_key would shortcut for
the queued_spin_lock_- slowpath() function that allow virt_spin_lock
to hijack it. ref: 43b3f02899f7 ("locking/qspinlock/x86: Fix
performance regression under unaccelerated VMs").

Add a static key controlling whether virt_spin_lock() should be
called or not. Add nopvspin support as x86.

Signed-off-by: Guo Ren (Alibaba DAMO Academy) <guoren@kernel.org>
---
 .../admin-guide/kernel-parameters.txt         |  2 +-
 arch/riscv/include/asm/qspinlock.h            | 24 +++++++++++++++++++
 arch/riscv/kernel/qspinlock_paravirt.c        |  8 +++++++
 3 files changed, 33 insertions(+), 1 deletion(-)

diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
index 6c42061ca20e..9e895e9ca655 100644
--- a/Documentation/admin-guide/kernel-parameters.txt
+++ b/Documentation/admin-guide/kernel-parameters.txt
@@ -4393,7 +4393,7 @@
 			as generic guest with no PV drivers. Currently support
 			XEN HVM, KVM, HYPER_V and VMWARE guest.
 
-	nopvspin	[X86,XEN,KVM,EARLY]
+	nopvspin	[X86,RISCV,XEN,KVM,EARLY]
 			Disables the qspinlock slow path using PV optimizations
 			which allow the hypervisor to 'idle' the guest on lock
 			contention.
diff --git a/arch/riscv/include/asm/qspinlock.h b/arch/riscv/include/asm/qspinlock.h
index b39f23415ec1..70ad7679fce0 100644
--- a/arch/riscv/include/asm/qspinlock.h
+++ b/arch/riscv/include/asm/qspinlock.h
@@ -14,6 +14,8 @@
 /* How long a lock should spin before we consider blocking */
 #define SPIN_THRESHOLD		(1 << 15)
 
+extern bool nopvspin;
+
 void native_queued_spin_lock_slowpath(struct qspinlock *lock, u32 val);
 void __pv_init_lock_hash(void);
 void __pv_queued_spin_lock_slowpath(struct qspinlock *lock, u32 val);
@@ -31,5 +33,27 @@ static inline void queued_spin_unlock(struct qspinlock *lock)
 #endif /* CONFIG_PARAVIRT_SPINLOCKS */
 
 #include <asm-generic/qspinlock.h>
+#include <asm/jump_label.h>
+
+/*
+ * The KVM guests fall back to a Test-and-Set spinlock, because fair locks
+ * have horrible lock 'holder' preemption issues. The test_and_set_spinlock_key
+ * would shortcut for the queued_spin_lock_slowpath() function that allow
+ * virt_spin_lock to hijack it.
+ */
+DECLARE_STATIC_KEY_FALSE(virt_spin_lock_key);
+
+#define virt_spin_lock rv_virt_spin_lock
+static inline bool rv_virt_spin_lock(struct qspinlock *lock)
+{
+	if (!static_branch_likely(&virt_spin_lock_key))
+		return false;
+
+	do {
+		smp_cond_load_relaxed((s32 *)&lock->val, VAL == 0);
+	} while (atomic_cmpxchg(&lock->val, 0, _Q_LOCKED_VAL) != 0);
+
+	return true;
+}
 
 #endif /* _ASM_RISCV_QSPINLOCK_H */
diff --git a/arch/riscv/kernel/qspinlock_paravirt.c b/arch/riscv/kernel/qspinlock_paravirt.c
index cae991139abe..b89f13d0d6e8 100644
--- a/arch/riscv/kernel/qspinlock_paravirt.c
+++ b/arch/riscv/kernel/qspinlock_paravirt.c
@@ -50,6 +50,8 @@ EXPORT_STATIC_CALL(pv_queued_spin_lock_slowpath);
 DEFINE_STATIC_CALL(pv_queued_spin_unlock, native_queued_spin_unlock);
 EXPORT_STATIC_CALL(pv_queued_spin_unlock);
 
+DEFINE_STATIC_KEY_FALSE(virt_spin_lock_key);
+
 bool __init pv_qspinlock_init(void)
 {
 	if (num_possible_cpus() == 1)
@@ -58,6 +60,12 @@ bool __init pv_qspinlock_init(void)
 	if (!sbi_probe_extension(SBI_EXT_PVLOCK))
 		return false;
 
+	if (nopvspin) {
+		static_branch_enable(&virt_spin_lock_key);
+		pr_info("virt_spin_lock enabled by nopvspin\n");
+		return true;
+	}
+
 	pr_info("PV qspinlocks enabled\n");
 	__pv_init_lock_hash();
 
-- 
2.40.1