Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::3:7 as permitted sender) client-ip=2620:137:e000::3:7;
From:   Bibo Mao <maobibo@loongson.cn>
To:     Tianrui Zhao <zhaotianrui@loongson.cn>,
        Huacai Chen <chenhuacai@kernel.org>
Cc:     WANG Xuerui <kernel@xen0n.name>, kvm@vger.kernel.org,
        loongarch@lists.linux.dev, linux-kernel@vger.kernel.org
Subject: [PATCH v4 1/3] LoongArch: KVM: Remove SW timer switch when vcpu is halt polling
Date:   Thu, 16 Nov 2023 10:30:34 +0800
Message-Id: <20231116023036.2324371-2-maobibo@loongson.cn>
In-Reply-To: <20231116023036.2324371-1-maobibo@loongson.cn>
References: <20231116023036.2324371-1-maobibo@loongson.cn>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Precedence: bulk

With halt-polling supported, there is checking for pending events
or interrupts when vcpu executes idle instruction. Pending interrupts
include injected SW interrupts and passthrough HW interrupts, such as
HW timer interrupts, since HW timer works still even if vcpu exists
from VM mode.

Since HW timer pending interrupt can be set directly with CSR status
register, and pending HW timer interrupt checking is used in vcpu block
checking function, it is not necessary to switch to sw timer during
halt-polling. This patch adds preemption disabling in function
kvm_cpu_has_pending_timer, and removes SW timer switching in idle
instruction emulation function.

Signed-off-by: Bibo Mao <maobibo@loongson.cn>
---
 arch/loongarch/kvm/exit.c  | 13 ++-----------
 arch/loongarch/kvm/timer.c | 13 ++++++++++---
 arch/loongarch/kvm/vcpu.c  |  9 ++++++++-
 3 files changed, 20 insertions(+), 15 deletions(-)

diff --git a/arch/loongarch/kvm/exit.c b/arch/loongarch/kvm/exit.c
index ce8de3fa472c..e708a1786d6b 100644
--- a/arch/loongarch/kvm/exit.c
+++ b/arch/loongarch/kvm/exit.c
@@ -200,17 +200,8 @@ int kvm_emu_idle(struct kvm_vcpu *vcpu)
 	++vcpu->stat.idle_exits;
 	trace_kvm_exit_idle(vcpu, KVM_TRACE_EXIT_IDLE);
 
-	if (!kvm_arch_vcpu_runnable(vcpu)) {
-		/*
-		 * Switch to the software timer before halt-polling/blocking as
-		 * the guest's timer may be a break event for the vCPU, and the
-		 * hypervisor timer runs only when the CPU is in guest mode.
-		 * Switch before halt-polling so that KVM recognizes an expired
-		 * timer before blocking.
-		 */
-		kvm_save_timer(vcpu);
-		kvm_vcpu_block(vcpu);
-	}
+	if (!kvm_arch_vcpu_runnable(vcpu))
+		kvm_vcpu_halt(vcpu);
 
 	return EMULATE_DONE;
 }
diff --git a/arch/loongarch/kvm/timer.c b/arch/loongarch/kvm/timer.c
index 284bf553fefe..437e960d8fdb 100644
--- a/arch/loongarch/kvm/timer.c
+++ b/arch/loongarch/kvm/timer.c
@@ -155,11 +155,18 @@ static void _kvm_save_timer(struct kvm_vcpu *vcpu)
 		 */
 		hrtimer_cancel(&vcpu->arch.swtimer);
 		hrtimer_start(&vcpu->arch.swtimer, expire, HRTIMER_MODE_ABS_PINNED);
-	} else
+	} else if (vcpu->stat.generic.blocking) {
 		/*
-		 * Inject timer interrupt so that hall polling can dectect and exit
+		 * Inject timer interrupt so that hall polling can dectect and
+		 * exit.
+		 * VCPU is scheduled out already and sleeps in rcuwait queue and
+		 * will not poll pending events again. kvm_queue_irq is not
+		 * enough, hrtimer swtimer should be used here.
 		 */
-		kvm_queue_irq(vcpu, INT_TI);
+		expire = ktime_add_ns(ktime_get(), 10);  // 10ns is enough here?
+		vcpu->arch.expire = expire;
+		hrtimer_start(&vcpu->arch.swtimer, expire, HRTIMER_MODE_ABS_PINNED);
+	}
 }
 
 /*
diff --git a/arch/loongarch/kvm/vcpu.c b/arch/loongarch/kvm/vcpu.c
index 73d0c2b9c1a5..42663a345bd1 100644
--- a/arch/loongarch/kvm/vcpu.c
+++ b/arch/loongarch/kvm/vcpu.c
@@ -187,8 +187,15 @@ int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
 
 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 {
-	return kvm_pending_timer(vcpu) ||
+	int ret;
+
+	/* protect from TOD sync and vcpu_load/put */
+	preempt_disable();
+	ret = kvm_pending_timer(vcpu) ||
 		kvm_read_hw_gcsr(LOONGARCH_CSR_ESTAT) & (1 << INT_TI);
+	preempt_enable();
+
+	return ret;
 }
 
 int kvm_arch_vcpu_dump_regs(struct kvm_vcpu *vcpu)
-- 
2.39.3