Re: [PATCH 4/9] KVM: arm64: Support stolen time reporting via shared structure

From: Steven Price <steven.price@arm.com>
To: Marc Zyngier <maz@kernel.org>
Cc: kvm@vger.kernel.org, linux-doc@vger.kernel.org,
	Catalin Marinas <catalin.marinas@arm.com>,
	linux-kernel@vger.kernel.org,
	Russell King <linux@armlinux.org.uk>,
	Paolo Bonzini <pbonzini@redhat.com>,
	Will Deacon <will@kernel.org>,
	kvmarm@lists.cs.columbia.edu,
	linux-arm-kernel@lists.infradead.org
Subject: Re: [PATCH 4/9] KVM: arm64: Support stolen time reporting via shared structure
Date: Mon, 5 Aug 2019 15:18:04 +0100	[thread overview]
Message-ID: <4a373dc5-4a6b-efc6-4309-e7180eb0c696@arm.com> (raw)
In-Reply-To: <20190803191303.02e9bcc9@why>

On 03/08/2019 19:13, Marc Zyngier wrote:
> On Sat, 3 Aug 2019 18:58:17 +0100
> Marc Zyngier <maz@kernel.org> wrote:
> 
>> On Fri,  2 Aug 2019 15:50:12 +0100
>> Steven Price <steven.price@arm.com> wrote:
>>
>>> Implement the service call for configuring a shared structre between a
>>> VCPU and the hypervisor in which the hypervisor can write the time
>>> stolen from the VCPU's execution time by other tasks on the host.
>>>
>>> The hypervisor allocates memory which is placed at an IPA chosen by user
>>> space. The hypervisor then uses WRITE_ONCE() to update the shared
>>> structre ensuring single copy atomicity of the 64-bit unsigned value
>>> that reports stolen time in nanoseconds.
>>>
>>> Whenever stolen time is enabled by the guest, the stolen time counter is
>>> reset.
>>>
>>> The stolen time itself is retrieved from the sched_info structure
>>> maintained by the Linux scheduler code. We enable SCHEDSTATS when
>>> selecting KVM Kconfig to ensure this value is meaningful.
>>>
>>> Signed-off-by: Steven Price <steven.price@arm.com>
>>> ---
>>>  arch/arm64/include/asm/kvm_host.h | 13 +++++-
>>>  arch/arm64/kvm/Kconfig            |  1 +
>>>  include/kvm/arm_hypercalls.h      |  1 +
>>>  include/linux/kvm_types.h         |  2 +
>>>  virt/kvm/arm/arm.c                | 18 ++++++++
>>>  virt/kvm/arm/hypercalls.c         | 70 +++++++++++++++++++++++++++++++
>>>  6 files changed, 104 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/arch/arm64/include/asm/kvm_host.h b/arch/arm64/include/asm/kvm_host.h
>>> index f656169db8c3..78f270190d43 100644
>>> --- a/arch/arm64/include/asm/kvm_host.h
>>> +++ b/arch/arm64/include/asm/kvm_host.h
>>> @@ -44,6 +44,7 @@
>>>  	KVM_ARCH_REQ_FLAGS(0, KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
>>>  #define KVM_REQ_IRQ_PENDING	KVM_ARCH_REQ(1)
>>>  #define KVM_REQ_VCPU_RESET	KVM_ARCH_REQ(2)
>>> +#define KVM_REQ_RECORD_STEAL	KVM_ARCH_REQ(3)
>>>  
>>>  DECLARE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
>>>  
>>> @@ -83,6 +84,11 @@ struct kvm_arch {
>>>  
>>>  	/* Mandated version of PSCI */
>>>  	u32 psci_version;
>>> +
>>> +	struct kvm_arch_pvtime {
>>> +		void *st;
>>> +		gpa_t st_base;
>>> +	} pvtime;
>>>  };
>>>  
>>>  #define KVM_NR_MEM_OBJS     40
>>> @@ -338,8 +344,13 @@ struct kvm_vcpu_arch {
>>>  	/* True when deferrable sysregs are loaded on the physical CPU,
>>>  	 * see kvm_vcpu_load_sysregs and kvm_vcpu_put_sysregs. */
>>>  	bool sysregs_loaded_on_cpu;
>>> -};
>>>  
>>> +	/* Guest PV state */
>>> +	struct {
>>> +		u64 steal;
>>> +		u64 last_steal;
>>> +	} steal;
>>> +};
>>>  /* Pointer to the vcpu's SVE FFR for sve_{save,load}_state() */
>>>  #define vcpu_sve_pffr(vcpu) ((void *)((char *)((vcpu)->arch.sve_state) + \
>>>  				      sve_ffr_offset((vcpu)->arch.sve_max_vl)))
>>> diff --git a/arch/arm64/kvm/Kconfig b/arch/arm64/kvm/Kconfig
>>> index a67121d419a2..d8b88e40d223 100644
>>> --- a/arch/arm64/kvm/Kconfig
>>> +++ b/arch/arm64/kvm/Kconfig
>>> @@ -39,6 +39,7 @@ config KVM
>>>  	select IRQ_BYPASS_MANAGER
>>>  	select HAVE_KVM_IRQ_BYPASS
>>>  	select HAVE_KVM_VCPU_RUN_PID_CHANGE
>>> +	select SCHEDSTATS
>>>  	---help---
>>>  	  Support hosting virtualized guest machines.
>>>  	  We don't support KVM with 16K page tables yet, due to the multiple
>>> diff --git a/include/kvm/arm_hypercalls.h b/include/kvm/arm_hypercalls.h
>>> index 35a5abcc4ca3..9f0710ab4292 100644
>>> --- a/include/kvm/arm_hypercalls.h
>>> +++ b/include/kvm/arm_hypercalls.h
>>> @@ -7,6 +7,7 @@
>>>  #include <asm/kvm_emulate.h>
>>>  
>>>  int kvm_hvc_call_handler(struct kvm_vcpu *vcpu);
>>> +int kvm_update_stolen_time(struct kvm_vcpu *vcpu);
>>>  
>>>  static inline u32 smccc_get_function(struct kvm_vcpu *vcpu)
>>>  {
>>> diff --git a/include/linux/kvm_types.h b/include/linux/kvm_types.h
>>> index bde5374ae021..1c88e69db3d9 100644
>>> --- a/include/linux/kvm_types.h
>>> +++ b/include/linux/kvm_types.h
>>> @@ -35,6 +35,8 @@ typedef unsigned long  gva_t;
>>>  typedef u64            gpa_t;
>>>  typedef u64            gfn_t;
>>>  
>>> +#define GPA_INVALID	(~(gpa_t)0)
>>> +
>>>  typedef unsigned long  hva_t;
>>>  typedef u64            hpa_t;
>>>  typedef u64            hfn_t;
>>> diff --git a/virt/kvm/arm/arm.c b/virt/kvm/arm/arm.c
>>> index f645c0fbf7ec..ebd963d2580b 100644
>>> --- a/virt/kvm/arm/arm.c
>>> +++ b/virt/kvm/arm/arm.c
>>> @@ -40,6 +40,10 @@
>>>  #include <asm/kvm_coproc.h>
>>>  #include <asm/sections.h>
>>>  
>>> +#include <kvm/arm_hypercalls.h>
>>> +#include <kvm/arm_pmu.h>
>>> +#include <kvm/arm_psci.h>
>>> +
>>>  #ifdef REQUIRES_VIRT
>>>  __asm__(".arch_extension	virt");
>>>  #endif
>>> @@ -135,6 +139,7 @@ int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
>>>  	kvm->arch.max_vcpus = vgic_present ?
>>>  				kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
>>>  
>>> +	kvm->arch.pvtime.st_base = GPA_INVALID;
>>>  	return ret;
>>>  out_free_stage2_pgd:
>>>  	kvm_free_stage2_pgd(kvm);
>>> @@ -371,6 +376,7 @@ void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
>>>  	kvm_vcpu_load_sysregs(vcpu);
>>>  	kvm_arch_vcpu_load_fp(vcpu);
>>>  	kvm_vcpu_pmu_restore_guest(vcpu);
>>> +	kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
>>>  
>>>  	if (single_task_running())
>>>  		vcpu_clear_wfe_traps(vcpu);
>>> @@ -617,6 +623,15 @@ static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
>>>  	smp_rmb();
>>>  }
>>>  
>>> +static void vcpu_req_record_steal(struct kvm_vcpu *vcpu)
>>> +{
>>> +	int idx;
>>> +
>>> +	idx = srcu_read_lock(&vcpu->kvm->srcu);
>>> +	kvm_update_stolen_time(vcpu);
>>> +	srcu_read_unlock(&vcpu->kvm->srcu, idx);
>>> +}
>>> +
>>>  static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
>>>  {
>>>  	return vcpu->arch.target >= 0;
>>> @@ -636,6 +651,9 @@ static void check_vcpu_requests(struct kvm_vcpu *vcpu)
>>>  		 * that a VCPU sees new virtual interrupts.
>>>  		 */
>>>  		kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
>>> +
>>> +		if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
>>> +			vcpu_req_record_steal(vcpu);  
>>
>> Something troubles me. Here, you've set the request on load. But you
>> can be preempted at any time (preemption gets disabled just after).
>>
>> I have the feeling that should you get preempted right here, you'll
>> end-up having accumulated the wrong amount of steal time, as the
>> request put via load when you'll get scheduled back in will only get
>> processed after a full round of entry/exit/entry, which doesn't look
>> great.
> 
> Ah, no. We're saved by the check for pending requests right before we
> jump in the guest, causing an early exit and the whole shebang to be
> restarted.

Yes, that's my understanding. Obviously not ideal if it happens in that
small window, but everything is redone to get the right values in the end.

Steve