Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Thu, 15 Dec 2022 16:31:00 -0800
From:   Isaku Yamahata <isaku.yamahata@gmail.com>
To:     Vishal Annapurve <vannapurve@google.com>
Cc:     binbin.wu@linux.intel.com, dmatlack@google.com,
        erdemaktas@google.com, isaku.yamahata@gmail.com,
        isaku.yamahata@intel.com, kvm@vger.kernel.org,
        linux-kernel@vger.kernel.org, pbonzini@redhat.com,
        sagis@google.com, seanjc@google.com
Subject: Re: [PATCH v10 062/108] KVM: x86/tdp_mmu: implement MapGPA hypercall
 for TDX
Message-ID: <20221216003100.GM3632095@ls.amr.corp.intel.com>
References: <b49d953a-61f0-e466-90d2-713e25dff770@linux.intel.com>
 <20221209000130.1380904-1-vannapurve@google.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
In-Reply-To: <20221209000130.1380904-1-vannapurve@google.com>
Precedence: bulk

On Fri, Dec 09, 2022 at 12:01:30AM +0000,
Vishal Annapurve <vannapurve@google.com> wrote:

> > +int __kvm_mmu_map_gpa(struct kvm *kvm, gfn_t *startp, gfn_t end,
> > +		      bool map_private)
> > +{
> > +	gfn_t start = *startp;
> > +	int attr;
> > +	int ret;
> > +
> > +	if (!kvm_gfn_shared_mask(kvm))
> > +		return -EOPNOTSUPP;
> > +
> > +	attr = map_private ? KVM_MEM_ATTR_PRIVATE : KVM_MEM_ATTR_SHARED;
> > +	start = start & ~kvm_gfn_shared_mask(kvm);
> > +	end = end & ~kvm_gfn_shared_mask(kvm);
> > +
> > +	/*
> > +	 * To make the following kvm_vm_set_mem_attr() success within spinlock
> > +	 * without memory allocation.
> > +	 */
> > +	ret = kvm_vm_reserve_mem_attr(kvm, start, end);
> > +	if (ret)
> > +		return ret;
> > +
> > +	write_lock(&kvm-> mmu_lock);
> > +	if (is_tdp_mmu_enabled(kvm)) {
> > +		gfn_t s = start;
> > +
> > +		ret = kvm_tdp_mmu_map_gpa(kvm, &s, end, map_private);
> > +		if (!ret) {
> > +			KVM_BUG_ON(kvm_vm_set_mem_attr(kvm, attr, start, end), kvm);
> 
> This will result in no exits to userspace during memory conversion requests from
> guests. And as a result, userspace will not be able to explicitly back/unback
> shared/private memory during conversions leading to double allocation of memory.
> 
> Is this an intended behavior for memory conversion with TDX VMs as per earlier
> discussion?

It was because UPM ioctl interface was not stable yet. With v10, it seems mostly
stable, I'll change the patch of "[PATCH v10 101/108] KVM: TDX: Handle TDX PV
map_gpa hypercall" to exit to user space.
-- 
Isaku Yamahata <isaku.yamahata@gmail.com>