Received-SPF: pass (google.com: domain of linux-kernel+bounces-40353-linux.lists.archive=gmail.com@vger.kernel.org designates 139.178.88.99 as permitted sender) client-ip=139.178.88.99;
Date: Fri, 26 Jan 2024 08:39:20 -0800
In-Reply-To: <87sf2k83qb.fsf@redhat.com>
Precedence: bulk
Mime-Version: 1.0
References: <20240110011533.503302-1-seanjc@google.com> <20240110011533.503302-4-seanjc@google.com>
 <87sf2k83qb.fsf@redhat.com>
Message-ID: <ZbPgOLVk4armOGDi@google.com>
Subject: Re: [PATCH 3/4] KVM: Get reference to VM's address space in the async
 #PF worker
From: Sean Christopherson <seanjc@google.com>
To: Vitaly Kuznetsov <vkuznets@redhat.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>, kvm@vger.kernel.org, linux-kernel@vger.kernel.org, 
	David Matlack <dmatlack@google.com>, Xu Yilun <yilun.xu@linux.intel.com>
Content-Type: text/plain; charset="us-ascii"

On Fri, Jan 26, 2024, Vitaly Kuznetsov wrote:
> Sean Christopherson <seanjc@google.com> writes:
> 
> > Get a reference to the target VM's address space in async_pf_execute()
> > instead of gifting a reference from kvm_setup_async_pf().  Keeping the
> > address space alive just to service an async #PF is counter-productive,
> > i.e. if the process is exiting and all vCPUs are dead, then NOT doing
> > get_user_pages_remote() and freeing the address space asap is
> > desirable.
> 
> It took me a while to realize why all vCPU fds are managed by the same
> mm which did KVM_CREATE_VM as (AFAIU) fds can be passed around. Turns
> out, we explicitly forbid this in kvm_vcpu_ioctl():
> 
>         if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
>                 return -EIO;
> 
> so this indeed means that grabbing current->mm in kvm_setup_async_pf()
> can be avoided. I'm not sure whether it's just me or a "all vCPUs are
> quired to be managed by the same mm" comment somewhere would be helpful.

It's definitely not just you.  Documentation/virt/kvm/* could use thorough
documentation of what all in KVM relies on vCPUs, and all meaningful ioctls(),
to be executed in the same mm_struct (address space).  Because that requirement
is pervasive throughout KVM.  E.g. sharing KVM page tables across vCPUs is safe
iff all vCPUs are in the same address space, otherwise the hva=>pfn translations
through the memslot would diverge, mmu_notifiers would be all kinds of broken, etc.