From: Zengruan Ye yezengruan@huawei.com

virt inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I8WMFU CVE: NA

--------------------------------

Introduce a paravirtualization interface for KVM/arm64 to PV-sched.

A hypercall interface is provided for the guest to interrogate the hypervisor's support for this interface and the location of the shared memory structures.

Signed-off-by: Zengruan Ye yezengruan@huawei.com Signed-off-by: lishusen lishusen2@huawei.com --- Documentation/virt/kvm/arm/pvsched.rst | 58 ++++++++++++++++++++++++++ 1 file changed, 58 insertions(+) create mode 100644 Documentation/virt/kvm/arm/pvsched.rst

diff --git a/Documentation/virt/kvm/arm/pvsched.rst b/Documentation/virt/kvm/arm/pvsched.rst new file mode 100644 index 000000000000..8f7112a8a9cd --- /dev/null +++ b/Documentation/virt/kvm/arm/pvsched.rst @@ -0,0 +1,58 @@ +.. SPDX-License-Identifier: GPL-2.0 + +Paravirtualized sched support for arm64 +======================================= + +KVM/arm64 provides some hypervisor service calls to support a paravirtualized +sched. + +Some SMCCC compatible hypercalls are defined: + +* PV_SCHED_FEATURES: 0xC5000090 +* PV_SCHED_IPA_INIT: 0xC5000091 +* PV_SCHED_IPA_RELEASE: 0xC5000092 + +The existence of the PV_SCHED hypercall should be probed using the SMCCC 1.1 +ARCH_FEATURES mechanism before calling it. + +PV_SCHED_FEATURES + ============= ======== ========== + Function ID: (uint32) 0xC5000090 + PV_call_id: (uint32) The function to query for support. + Return value: (int64) NOT_SUPPORTED (-1) or SUCCESS (0) if the relevant + PV-sched feature is supported by the hypervisor. + ============= ======== ========== + +PV_SCHED_IPA_INIT + ============= ======== ========== + Function ID: (uint32) 0xC5000091 + Return value: (int64) NOT_SUPPORTED (-1) or SUCCESS (0) if the IPA of + this vCPU's PV data structure is shared to the + hypervisor. + ============= ======== ========== + +PV_SCHED_IPA_RELEASE + ============= ======== ========== + Function ID: (uint32) 0xC5000092 + Return value: (int64) NOT_SUPPORTED (-1) or SUCCESS (0) if the IPA of + this vCPU's PV data structure is released. + ============= ======== ========== + +PV sched state +-------------- + +The structure pointed to by the PV_SCHED_IPA hypercall is as follows: + ++-----------+-------------+-------------+-----------------------------------+ +| Field | Byte Length | Byte Offset | Description | ++===========+=============+=============+===================================+ +| preempted | 4 | 0 | Indicates that the vCPU that owns | +| | | | this struct is running or not. | +| | | | Non-zero values mean the vCPU has | +| | | | been preempted. Zero means the | +| | | | vCPU is not preempted. | ++-----------+-------------+-------------+-----------------------------------+ + +The preempted field will be updated to 0 by the hypervisor prior to scheduling +a vCPU. When the vCPU is scheduled out, the preempted field will be updated +to 1 by the hypervisor.

From: Zengruan Ye yezengruan@huawei.com

virt inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I8WMFU CVE: NA

--------------------------------

Implement the service call for configuring a shared structure between a vCPU and the hypervisor in which the hypervisor can tell the vCPU that is running or not.

Signed-off-by: Zengruan Ye yezengruan@huawei.com Signed-off-by: lishusen lishusen2@huawei.com --- arch/arm64/include/asm/kvm_host.h | 23 +++++++++++++++++++++++ arch/arm64/kvm/arm.c | 8 ++++++++ arch/arm64/kvm/hypercalls.c | 11 +++++++++++ arch/arm64/kvm/pvsched.c | 28 ++++++++++++++++++++++++++++ 4 files changed, 70 insertions(+)

diff --git a/arch/arm64/include/asm/kvm_host.h b/arch/arm64/include/asm/kvm_host.h index 1d046ebae9c3..e256b091f9c6 100644 --- a/arch/arm64/include/asm/kvm_host.h +++ b/arch/arm64/include/asm/kvm_host.h @@ -596,6 +596,11 @@ struct kvm_vcpu_arch { gpa_t base; } steal;

+ /* Guest PV sched state */ + struct { + gpa_t base; + } pvsched; + /* Per-vcpu CCSIDR override or NULL */ u32 *ccsidr;

@@ -1053,11 +1058,29 @@ static inline bool kvm_arm_is_pvtime_enabled(struct kvm_vcpu_arch *vcpu_arch)

#ifdef CONFIG_PARAVIRT_SCHED long kvm_hypercall_pvsched_features(struct kvm_vcpu *vcpu); +void kvm_update_pvsched_preempted(struct kvm_vcpu *vcpu, u32 preempted); + +static inline void kvm_arm_pvsched_vcpu_init(struct kvm_vcpu_arch *vcpu_arch) +{ + vcpu_arch->pvsched.base = INVALID_GPA; +} + +static inline bool kvm_arm_is_pvsched_enabled(struct kvm_vcpu_arch *vcpu_arch) +{ + return (vcpu_arch->pvsched.base != INVALID_GPA); +} #else static inline long kvm_hypercall_pvsched_features(struct kvm_vcpu *vcpu) { return 0; } +static inline void kvm_update_pvsched_preempted(struct kvm_vcpu *vcpu, + u32 preempted) {} +static inline void kvm_arm_pvsched_vcpu_init(struct kvm_vcpu_arch *vcpu_arch) {} +static inline bool kvm_arm_is_pvsched_enabled(struct kvm_vcpu_arch *vcpu_arch) +{ + return false; +} #endif

void kvm_set_sei_esr(struct kvm_vcpu *vcpu, u64 syndrome); diff --git a/arch/arm64/kvm/arm.c b/arch/arm64/kvm/arm.c index e7e35b95821f..c3b47357ade5 100644 --- a/arch/arm64/kvm/arm.c +++ b/arch/arm64/kvm/arm.c @@ -408,6 +408,8 @@ int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)

kvm_arm_pvtime_vcpu_init(&vcpu->arch);

+ kvm_arm_pvsched_vcpu_init(&vcpu->arch); + vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu;

err = kvm_vgic_vcpu_init(vcpu); @@ -495,6 +497,9 @@ void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu) vcpu_set_on_unsupported_cpu(vcpu);

kvm_tlbi_dvmbm_vcpu_load(vcpu); + + if (kvm_arm_is_pvsched_enabled(&vcpu->arch)) + kvm_update_pvsched_preempted(vcpu, 0); }

void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) @@ -512,6 +517,9 @@ void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) vcpu->cpu = -1;

kvm_tlbi_dvmbm_vcpu_put(vcpu); + + if (kvm_arm_is_pvsched_enabled(&vcpu->arch)) + kvm_update_pvsched_preempted(vcpu, 1); }

static void __kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu) diff --git a/arch/arm64/kvm/hypercalls.c b/arch/arm64/kvm/hypercalls.c index a7c4090a5dbc..48bf2e0b7a1d 100644 --- a/arch/arm64/kvm/hypercalls.c +++ b/arch/arm64/kvm/hypercalls.c @@ -348,6 +348,17 @@ int kvm_smccc_call_handler(struct kvm_vcpu *vcpu) case ARM_SMCCC_HV_PV_SCHED_FEATURES: val[0] = kvm_hypercall_pvsched_features(vcpu); break; + case ARM_SMCCC_HV_PV_SCHED_IPA_INIT: + gpa = smccc_get_arg1(vcpu); + if (gpa != INVALID_GPA) { + vcpu->arch.pvsched.base = gpa; + val[0] = SMCCC_RET_SUCCESS; + } + break; + case ARM_SMCCC_HV_PV_SCHED_IPA_RELEASE: + vcpu->arch.pvsched.base = INVALID_GPA; + val[0] = SMCCC_RET_SUCCESS; + break; case ARM_SMCCC_VENDOR_HYP_CALL_UID_FUNC_ID: val[0] = ARM_SMCCC_VENDOR_HYP_UID_KVM_REG_0; val[1] = ARM_SMCCC_VENDOR_HYP_UID_KVM_REG_1; diff --git a/arch/arm64/kvm/pvsched.c b/arch/arm64/kvm/pvsched.c index b54237707ec5..ee7fed4f539e 100644 --- a/arch/arm64/kvm/pvsched.c +++ b/arch/arm64/kvm/pvsched.c @@ -6,9 +6,35 @@

#ifdef CONFIG_PARAVIRT_SCHED #include <linux/arm-smccc.h> +#include <linux/kvm_host.h> + +#include <asm/pvsched-abi.h>

#include <kvm/arm_hypercalls.h>

+void kvm_update_pvsched_preempted(struct kvm_vcpu *vcpu, u32 preempted) +{ + struct kvm *kvm = vcpu->kvm; + u64 base = vcpu->arch.pvsched.base; + u64 offset = offsetof(struct pvsched_vcpu_state, preempted); + int idx; + + if (base == INVALID_GPA) + return; + + /* + * This function is called from atomic context, so we need to + * disable page faults. + */ + pagefault_disable(); + + idx = srcu_read_lock(&kvm->srcu); + kvm_put_guest(kvm, base + offset, cpu_to_le32(preempted)); + srcu_read_unlock(&kvm->srcu, idx); + + pagefault_enable(); +} + long kvm_hypercall_pvsched_features(struct kvm_vcpu *vcpu) { u32 feature = smccc_get_arg1(vcpu); @@ -16,6 +42,8 @@ long kvm_hypercall_pvsched_features(struct kvm_vcpu *vcpu)

switch (feature) { case ARM_SMCCC_HV_PV_SCHED_FEATURES: + case ARM_SMCCC_HV_PV_SCHED_IPA_INIT: + case ARM_SMCCC_HV_PV_SCHED_IPA_RELEASE: val = SMCCC_RET_SUCCESS; break; }

From: Zengruan Ye yezengruan@huawei.com

virt inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I8WMFU CVE: NA

--------------------------------

Support the vcpu_is_preempted() functionality under KVM/arm64. This will enhance lock performance on overcommitted hosts (more runnable vCPUs than physical CPUs in the system) as doing busy waits for preempted vCPUs will hurt system performance far worse than early yielding.

unix benchmark result: host: kernel 4.19.87, HiSilicon Kunpeng920, 8 CPUs guest: kernel 4.19.87, 16 vCPUs

test-case | after-patch | before-patch ----------------------------------------+-------------------+------------------ Dhrystone 2 using register variables | 338955728.5 lps | 339266319.5 lps Double-Precision Whetstone | 30634.9 MWIPS | 30884.4 MWIPS Execl Throughput | 6753.2 lps | 3580.1 lps File Copy 1024 bufsize 2000 maxblocks | 490048.0 KBps | 313282.3 KBps File Copy 256 bufsize 500 maxblocks | 129662.5 KBps | 83550.7 KBps File Copy 4096 bufsize 8000 maxblocks | 1552551.5 KBps | 814327.0 KBps Pipe Throughput | 8976422.5 lps | 9048628.4 lps Pipe-based Context Switching | 258641.7 lps | 252925.9 lps Process Creation | 5312.2 lps | 4507.9 lps Shell Scripts (1 concurrent) | 8704.2 lpm | 6720.9 lpm Shell Scripts (8 concurrent) | 1708.8 lpm | 607.2 lpm System Call Overhead | 3714444.7 lps | 3746386.8 lps ----------------------------------------+-------------------+------------------ System Benchmarks Index Score | 2270.6 | 1679.2

Signed-off-by: Zengruan Ye yezengruan@huawei.com Signed-off-by: lishusen lishusen2@huawei.com --- arch/arm64/include/asm/kvm_host.h | 2 + arch/arm64/include/asm/paravirt.h | 5 ++ arch/arm64/include/asm/pvsched-abi.h | 2 + arch/arm64/kernel/paravirt-spinlocks.c | 2 + arch/arm64/kernel/paravirt.c | 108 +++++++++++++++++++++++++ arch/arm64/kvm/arm.c | 2 + arch/arm64/kvm/hypercalls.c | 4 + include/linux/arm-smccc.h | 2 + include/linux/cpuhotplug.h | 1 + 9 files changed, 128 insertions(+)

diff --git a/arch/arm64/include/asm/kvm_host.h b/arch/arm64/include/asm/kvm_host.h index e256b091f9c6..f4d25426aef3 100644 --- a/arch/arm64/include/asm/kvm_host.h +++ b/arch/arm64/include/asm/kvm_host.h @@ -596,10 +596,12 @@ struct kvm_vcpu_arch { gpa_t base; } steal;

+#ifdef CONFIG_PARAVIRT_SCHED /* Guest PV sched state */ struct { gpa_t base; } pvsched; +#endif

/* Per-vcpu CCSIDR override or NULL */ u32 *ccsidr; diff --git a/arch/arm64/include/asm/paravirt.h b/arch/arm64/include/asm/paravirt.h index 53282dc429f8..5ccead71bf87 100644 --- a/arch/arm64/include/asm/paravirt.h +++ b/arch/arm64/include/asm/paravirt.h @@ -20,6 +20,9 @@ static inline u64 paravirt_steal_clock(int cpu)

int __init pv_time_init(void);

+#ifdef CONFIG_PARAVIRT_SCHED +int __init pv_sched_init(void); + __visible bool __native_vcpu_is_preempted(int cpu); DECLARE_STATIC_CALL(pv_vcpu_preempted, __native_vcpu_is_preempted);

@@ -27,10 +30,12 @@ static inline bool pv_vcpu_is_preempted(int cpu) { return static_call(pv_vcpu_preempted)(cpu); } +#endif /* CONFIG_PARAVIRT_SCHED */

#else

#define pv_time_init() do {} while (0) +#define pv_sched_init() do {} while (0)

#endif // CONFIG_PARAVIRT

diff --git a/arch/arm64/include/asm/pvsched-abi.h b/arch/arm64/include/asm/pvsched-abi.h index 80e50e7a1a31..cc8e27dd266a 100644 --- a/arch/arm64/include/asm/pvsched-abi.h +++ b/arch/arm64/include/asm/pvsched-abi.h @@ -7,10 +7,12 @@ #ifndef __ASM_PVSCHED_ABI_H #define __ASM_PVSCHED_ABI_H

+#ifdef CONFIG_PARAVIRT_SCHED struct pvsched_vcpu_state { __le32 preempted; /* Structure must be 64 byte aligned, pad to that size */ u8 padding[60]; } __packed; +#endif /* CONFIG_PARAVIRT_SCHED */

#endif diff --git a/arch/arm64/kernel/paravirt-spinlocks.c b/arch/arm64/kernel/paravirt-spinlocks.c index 3566897070d9..f402e7e6c301 100644 --- a/arch/arm64/kernel/paravirt-spinlocks.c +++ b/arch/arm64/kernel/paravirt-spinlocks.c @@ -4,6 +4,7 @@ * Author: Zengruan Ye yezengruan@huawei.com */

+#ifdef CONFIG_PARAVIRT_SCHED #include <linux/static_call.h> #include <linux/spinlock.h> #include <asm/paravirt.h> @@ -14,3 +15,4 @@ __visible bool __native_vcpu_is_preempted(int cpu) }

DEFINE_STATIC_CALL(pv_vcpu_preempted, __native_vcpu_is_preempted); +#endif /* CONFIG_PARAVIRT_SCHED */ diff --git a/arch/arm64/kernel/paravirt.c b/arch/arm64/kernel/paravirt.c index aa718d6a9274..53bb6fa76366 100644 --- a/arch/arm64/kernel/paravirt.c +++ b/arch/arm64/kernel/paravirt.c @@ -22,6 +22,7 @@

#include <asm/paravirt.h> #include <asm/pvclock-abi.h> +#include <asm/pvsched-abi.h> #include <asm/smp_plat.h>

struct static_key paravirt_steal_enabled; @@ -174,3 +175,110 @@ int __init pv_time_init(void)

return 0; } + +#ifdef CONFIG_PARAVIRT_SCHED +DEFINE_PER_CPU(struct pvsched_vcpu_state, pvsched_vcpu_region) __aligned(64); +EXPORT_PER_CPU_SYMBOL(pvsched_vcpu_region); + +static bool kvm_vcpu_is_preempted(int cpu) +{ + struct pvsched_vcpu_state *reg; + u32 preempted; + + reg = &per_cpu(pvsched_vcpu_region, cpu); + if (!reg) { + pr_warn_once("PV sched enabled but not configured for cpu %d\n", + cpu); + return false; + } + + preempted = le32_to_cpu(READ_ONCE(reg->preempted)); + + return !!preempted; +} + +static int pvsched_vcpu_state_dying_cpu(unsigned int cpu) +{ + struct pvsched_vcpu_state *reg; + struct arm_smccc_res res; + + reg = this_cpu_ptr(&pvsched_vcpu_region); + if (!reg) + return -EFAULT; + + arm_smccc_1_1_invoke(ARM_SMCCC_HV_PV_SCHED_IPA_RELEASE, &res); + memset(reg, 0, sizeof(*reg)); + + return 0; +} + +static int init_pvsched_vcpu_state(unsigned int cpu) +{ + struct pvsched_vcpu_state *reg; + struct arm_smccc_res res; + + reg = this_cpu_ptr(&pvsched_vcpu_region); + if (!reg) + return -EFAULT; + + /* Pass the memory address to host via hypercall */ + arm_smccc_1_1_invoke(ARM_SMCCC_HV_PV_SCHED_IPA_INIT, + virt_to_phys(reg), &res); + + return 0; +} + +static int kvm_arm_init_pvsched(void) +{ + int ret; + + ret = cpuhp_setup_state(CPUHP_AP_ARM_KVM_PVSCHED_STARTING, + "hypervisor/arm/pvsched:starting", + init_pvsched_vcpu_state, + pvsched_vcpu_state_dying_cpu); + + if (ret < 0) { + pr_warn("PV sched init failed\n"); + return ret; + } + + return 0; +} + +static bool has_kvm_pvsched(void) +{ + struct arm_smccc_res res; + + /* To detect the presence of PV sched support we require SMCCC 1.1+ */ + if (arm_smccc_1_1_get_conduit() == SMCCC_CONDUIT_NONE) + return false; + + arm_smccc_1_1_invoke(ARM_SMCCC_ARCH_FEATURES_FUNC_ID, + ARM_SMCCC_HV_PV_SCHED_FEATURES, &res); + + return (res.a0 == SMCCC_RET_SUCCESS); +} + +int __init pv_sched_init(void) +{ + int ret; + + if (is_hyp_mode_available()) + return 0; + + if (!has_kvm_pvsched()) { + pr_warn("PV sched is not available\n"); + return 0; + } + + ret = kvm_arm_init_pvsched(); + if (ret) + return ret; + + static_call_update(pv_vcpu_preempted, kvm_vcpu_is_preempted); + pr_info("using PV sched preempted\n"); + + return 0; +} +early_initcall(pv_sched_init); +#endif /* CONFIG_PARAVIRT_SCHED */ diff --git a/arch/arm64/kvm/arm.c b/arch/arm64/kvm/arm.c index c3b47357ade5..a6435fa24611 100644 --- a/arch/arm64/kvm/arm.c +++ b/arch/arm64/kvm/arm.c @@ -1377,6 +1377,8 @@ static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,

spin_unlock(&vcpu->arch.mp_state_lock);

+ kvm_arm_pvsched_vcpu_init(&vcpu->arch); + return 0; }

diff --git a/arch/arm64/kvm/hypercalls.c b/arch/arm64/kvm/hypercalls.c index 48bf2e0b7a1d..4c9fc5df5142 100644 --- a/arch/arm64/kvm/hypercalls.c +++ b/arch/arm64/kvm/hypercalls.c @@ -332,9 +332,11 @@ int kvm_smccc_call_handler(struct kvm_vcpu *vcpu) &smccc_feat->std_hyp_bmap)) val[0] = SMCCC_RET_SUCCESS; break; +#ifdef CONFIG_PARAVIRT_SCHED case ARM_SMCCC_HV_PV_SCHED_FEATURES: val[0] = SMCCC_RET_SUCCESS; break; +#endif /* CONFIG_PARAVIRT_SCHED */ } break; case ARM_SMCCC_HV_PV_TIME_FEATURES: @@ -345,6 +347,7 @@ int kvm_smccc_call_handler(struct kvm_vcpu *vcpu) if (gpa != INVALID_GPA) val[0] = gpa; break; +#ifdef CONFIG_PARAVIRT_SCHED case ARM_SMCCC_HV_PV_SCHED_FEATURES: val[0] = kvm_hypercall_pvsched_features(vcpu); break; @@ -359,6 +362,7 @@ int kvm_smccc_call_handler(struct kvm_vcpu *vcpu) vcpu->arch.pvsched.base = INVALID_GPA; val[0] = SMCCC_RET_SUCCESS; break; +#endif /* CONFIG_PARAVIRT_SCHED */ case ARM_SMCCC_VENDOR_HYP_CALL_UID_FUNC_ID: val[0] = ARM_SMCCC_VENDOR_HYP_UID_KVM_REG_0; val[1] = ARM_SMCCC_VENDOR_HYP_UID_KVM_REG_1; diff --git a/include/linux/arm-smccc.h b/include/linux/arm-smccc.h index daab363d31ae..7acb9898fbe1 100644 --- a/include/linux/arm-smccc.h +++ b/include/linux/arm-smccc.h @@ -577,6 +577,7 @@ asmlinkage void __arm_smccc_hvc(unsigned long a0, unsigned long a1, method; \ })

+#ifdef CONFIG_PARAVIRT_SCHED /* Paravirtualised sched calls */ #define ARM_SMCCC_HV_PV_SCHED_FEATURES \ ARM_SMCCC_CALL_VAL(ARM_SMCCC_FAST_CALL, \ @@ -595,6 +596,7 @@ asmlinkage void __arm_smccc_hvc(unsigned long a0, unsigned long a1, ARM_SMCCC_SMC_64, \ ARM_SMCCC_OWNER_STANDARD_HYP, \ 0x92) +#endif /* CONFIG_PARAVIRT_SCHED */

#endif /*__ASSEMBLY__*/ #endif /*__LINUX_ARM_SMCCC_H*/ diff --git a/include/linux/cpuhotplug.h b/include/linux/cpuhotplug.h index 624d4a38c358..f94a1b8e34e0 100644 --- a/include/linux/cpuhotplug.h +++ b/include/linux/cpuhotplug.h @@ -190,6 +190,7 @@ enum cpuhp_state { CPUHP_AP_DUMMY_TIMER_STARTING, CPUHP_AP_ARM_XEN_STARTING, CPUHP_AP_ARM_XEN_RUNSTATE_STARTING, + CPUHP_AP_ARM_KVM_PVSCHED_STARTING, CPUHP_AP_ARM_CORESIGHT_STARTING, CPUHP_AP_ARM_CORESIGHT_CTI_STARTING, CPUHP_AP_ARM64_ISNDEP_STARTING,