Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Wed, 26 May 2021 21:10:47 +0000
From:   Sean Christopherson <seanjc@google.com>
To:     Isaku Yamahata <isaku.yamahata@intel.com>
Cc:     kvm@vger.kernel.org, linux-kernel@vger.kernel.org,
        Paolo Bonzini <pbonzini@redhat.com>,
        Vitaly Kuznetsov <vkuznets@redhat.com>,
        Wanpeng Li <wanpengli@tencent.com>,
        Jim Mattson <jmattson@google.com>, isaku.yamahata@gmail.com
Subject: Re: [RFC PATCH 00/10] KVM: x86/mmu: simplify argument to kvm page
 fault handler
Message-ID: <YK65V++S2Kt1OLTu@google.com>
References: <cover.1618914692.git.isaku.yamahata@intel.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <cover.1618914692.git.isaku.yamahata@intel.com>
Precedence: bulk

On Tue, Apr 20, 2021, Isaku Yamahata wrote:
> This is a preliminary clean up for TDX which complicates KVM page fault
> execution path.

Ooh, a series to complicate the page fault path!  ;-)

Grammatical snarkiness aside, I'm all in favor of adding a struct to collect the
page fault collateral.  Overarching feedback:

  - Have kvm_mmu_do_page_fault() handle initialization of the struct.  That
    will allow making most of the fields const, and will avoid the rather painful
    kvm_page_fault_init().

  - Pass @vcpu separately.  Yes, it's associated with the fault, but literally
    the first line in every consumer is "struct kvm_vcpu *vcpu = kpf->vcpu;".

  - Use "fault" instead of "kpf", mostly because it reads better for people that
    aren't intimately familiar with the code, but also to avoid having to refactor
    a huge amount of code if we decide to rename kvm_page_fault, e.g. if we decide
    to use that name to return fault information to userspace.

  - Snapshot anything that is computed in multiple places, even if it is
    derivative of existing info.  E.g. it probably makes sense to grab
    write/fetch (or exec).


E.g. I'm thinking something like

struct kvm_page_fault {
	const gpa_t cr2_or_gpa;
	const u32 error_code;
	const bool write;
	const bool read;
	const bool fetch;
	const bool prefault;
	const bool is_tdp;

	gfn_t gfn;
	hva_t hva;
	int max_level;

	kvm_pfn_t pfn;
	bool map_writable;
};

int kvm_tdp_page_fault(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault);

static inline int kvm_mmu_do_page_fault(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
					u32 err, bool prefault)
{
	struct kvm_page_fault fault = {
		.cr2_or_gpa = cr2_or_gpa,
		.error_code = err,
		.write	    = err & PFERR_WRITE_MASK,
		.fetch	    = err & PFERR_FETCH_MASK,
		.perm	    = ...
		.rsvd	    = err & PFERR_RSVD_MASK,

		.is_tdp	    = vcpu->arch.mmu->page_fault == kvm_tdp_page_fault,

		...
	};

#ifdef CONFIG_RETPOLINE
	if (likely(fault.is_tdp))
		return kvm_tdp_page_fault(vcpu, &fault);
#endif
	return vcpu->arch.mmu->page_fault(vcpu, &fault);
}