Re: [PATCH v6 1/5] KVM: Implement dirty quota-based throttling of vcpus

[Shivam Kumar <shivam.kumar1@nutanix.com>]

On 10/10/22 11:11 am, Shivam Kumar wrote:
> 
> 
> On 15/09/22 3:40 pm, Shivam Kumar wrote:
>> Define variables to track and throttle memory dirtying for every vcpu.
>>
>> dirty_count:    Number of pages the vcpu has dirtied since its creation,
>>                  while dirty logging is enabled.
>> dirty_quota:    Number of pages the vcpu is allowed to dirty. To dirty
>>                  more, it needs to request more quota by exiting to
>>                  userspace.
>>
>> Implement the flow for throttling based on dirty quota.
>>
>> i) Increment dirty_count for the vcpu whenever it dirties a page.
>> ii) Exit to userspace whenever the dirty quota is exhausted (i.e. dirty
>> count equals/exceeds dirty quota) to request more dirty quota.
>>
>> Suggested-by: Shaju Abraham <shaju.abraham@nutanix.com>
>> Suggested-by: Manish Mishra <manish.mishra@nutanix.com>
>> Co-developed-by: Anurag Madnawat <anurag.madnawat@nutanix.com>
>> Signed-off-by: Anurag Madnawat <anurag.madnawat@nutanix.com>
>> Signed-off-by: Shivam Kumar <shivam.kumar1@nutanix.com>
>> ---
>>   Documentation/virt/kvm/api.rst | 35 ++++++++++++++++++++++++++++++++++
>>   include/linux/kvm_host.h       | 20 ++++++++++++++++++-
>>   include/linux/kvm_types.h      |  1 +
>>   include/uapi/linux/kvm.h       | 12 ++++++++++++
>>   virt/kvm/kvm_main.c            | 26 ++++++++++++++++++++++---
>>   5 files changed, 90 insertions(+), 4 deletions(-)
>>
>> diff --git a/Documentation/virt/kvm/api.rst 
>> b/Documentation/virt/kvm/api.rst
>> index abd7c32126ce..97030a6a35b4 100644
>> --- a/Documentation/virt/kvm/api.rst
>> +++ b/Documentation/virt/kvm/api.rst
>> @@ -6614,6 +6614,26 @@ array field represents return values. The 
>> userspace should update the return
>>   values of SBI call before resuming the VCPU. For more details on 
>> RISC-V SBI
>>   spec refer, https://github.com/riscv/riscv-sbi-doc.
>> +::
>> +
>> +        /* KVM_EXIT_DIRTY_QUOTA_EXHAUSTED */
>> +        struct {
>> +            __u64 count;
>> +            __u64 quota;
>> +        } dirty_quota_exit;
>> +
>> +If exit reason is KVM_EXIT_DIRTY_QUOTA_EXHAUSTED, it indicates that 
>> the VCPU has
>> +exhausted its dirty quota. The 'dirty_quota_exit' member of kvm_run 
>> structure
>> +makes the following information available to the userspace:
>> +    count: the current count of pages dirtied by the VCPU, can be
>> +    skewed based on the size of the pages accessed by each vCPU.
>> +    quota: the observed dirty quota just before the exit to userspace.
>> +
>> +The userspace can design a strategy to allocate the overall scope of 
>> dirtying
>> +for the VM among the vcpus. Based on the strategy and the current 
>> state of dirty
>> +quota throttling, the userspace can make a decision to either update 
>> (increase)
>> +the quota or to put the VCPU to sleep for some time.
>> +
>>   ::
>>       /* KVM_EXIT_NOTIFY */
>> @@ -6668,6 +6688,21 @@ values in kvm_run even if the corresponding bit 
>> in kvm_dirty_regs is not set.
>>   ::
>> +    /*
>> +     * Number of pages the vCPU is allowed to have dirtied over its 
>> entire
>> +     * lifetime.  KVM_RUN exits with KVM_EXIT_DIRTY_QUOTA_EXHAUSTED 
>> if the quota
>> +     * is reached/exceeded.
>> +     */
>> +    __u64 dirty_quota;
>> +
>> +Please note that enforcing the quota is best effort, as the guest may 
>> dirty
>> +multiple pages before KVM can recheck the quota.  However, unless KVM 
>> is using
>> +a hardware-based dirty ring buffer, e.g. Intel's Page Modification 
>> Logging,
>> +KVM will detect quota exhaustion within a handful of dirtied pages.  
>> If a
>> +hardware ring buffer is used, the overrun is bounded by the size of 
>> the buffer
>> +(512 entries for PML).
>> +
>> +::
>>     };
>> diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h
>> index f4519d3689e1..9acb28635d94 100644
>> --- a/include/linux/kvm_host.h
>> +++ b/include/linux/kvm_host.h
>> @@ -151,12 +151,13 @@ static inline bool is_error_page(struct page *page)
>>   #define KVM_REQUEST_NO_ACTION      BIT(10)
>>   /*
>>    * Architecture-independent vcpu->requests bit members
>> - * Bits 4-7 are reserved for more arch-independent bits.
>> + * Bits 5-7 are reserved for more arch-independent bits.
>>    */
>>   #define KVM_REQ_TLB_FLUSH         (0 | KVM_REQUEST_WAIT | 
>> KVM_REQUEST_NO_WAKEUP)
>>   #define KVM_REQ_VM_DEAD           (1 | KVM_REQUEST_WAIT | 
>> KVM_REQUEST_NO_WAKEUP)
>>   #define KVM_REQ_UNBLOCK           2
>>   #define KVM_REQ_UNHALT            3
>> +#define KVM_REQ_DIRTY_QUOTA_EXIT  4
>>   #define KVM_REQUEST_ARCH_BASE     8
>>   /*
>> @@ -380,6 +381,8 @@ struct kvm_vcpu {
>>        */
>>       struct kvm_memory_slot *last_used_slot;
>>       u64 last_used_slot_gen;
>> +
>> +    u64 dirty_quota;
>>   };
>>   /*
>> @@ -542,6 +545,21 @@ static inline int 
>> kvm_vcpu_exiting_guest_mode(struct kvm_vcpu *vcpu)
>>       return cmpxchg(&vcpu->mode, IN_GUEST_MODE, EXITING_GUEST_MODE);
>>   }
>> +static inline int kvm_vcpu_check_dirty_quota(struct kvm_vcpu *vcpu)
>> +{
>> +    struct kvm_run *run = vcpu->run;
>> +    u64 dirty_quota = READ_ONCE(run->dirty_quota);
>> +    u64 pages_dirtied = vcpu->stat.generic.pages_dirtied;
>> +
>> +    if (!dirty_quota || (pages_dirtied < dirty_quota))
>> +        return 1;
>> +
>> +    run->exit_reason = KVM_EXIT_DIRTY_QUOTA_EXHAUSTED;
>> +    run->dirty_quota_exit.count = pages_dirtied;
>> +    run->dirty_quota_exit.quota = dirty_quota;
>> +    return 0;
>> +}
>> +
>>   /*
>>    * Some of the bitops functions do not support too long bitmaps.
>>    * This number must be determined not to exceed such limits.
>> diff --git a/include/linux/kvm_types.h b/include/linux/kvm_types.h
>> index 3ca3db020e0e..263a588f3cd3 100644
>> --- a/include/linux/kvm_types.h
>> +++ b/include/linux/kvm_types.h
>> @@ -118,6 +118,7 @@ struct kvm_vcpu_stat_generic {
>>       u64 halt_poll_fail_hist[HALT_POLL_HIST_COUNT];
>>       u64 halt_wait_hist[HALT_POLL_HIST_COUNT];
>>       u64 blocking;
>> +    u64 pages_dirtied;
> I am reworking the QEMU patches and I am not sure how I can access the
> pages_dirtied info from the userspace side when the migration starts, i.e.
> without a dirty quota exit.
> 
> I need this info to initialise the dirty quota. This is what I am looking
> to do on the userspace side at the start of dirty quota migration:
>      dirty_quota = pages_dirtied + some initial quota
> 
> Hoping if you could help, Sean. Thanks in advance.
I think I can set dirty_quota initially to 1 and let the vpcu exit with 
exit reason KVM_EXIT_DIRTY_QUOTA_EXHAUSTED. Then, I can set the quota.