linux-kernel - Re: [PATCH v1] KVM: x86/mmu: refactor kvm_tdp_mmu

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <ZMpy8qvKTtAqaDWM@google.com>
Date:   Wed, 2 Aug 2023 08:14:58 -0700
From:   Sean Christopherson <seanjc@...gle.com>
To:     Wei Wang <wei.w.wang@...el.com>
Cc:     pbonzini@...hat.com, bgardon@...gle.com, dmatlack@...gle.com,
        kvm@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH v1] KVM: x86/mmu: refactor kvm_tdp_mmu_map

On Wed, Aug 02, 2023, Wei Wang wrote:
> The implementation of kvm_tdp_mmu_map is a bit long. It essentially does
> three things:
> 1) adjust the leaf entry level (e.g. 4KB, 2MB or 1GB) to map according to
>    the hugepage configurations;
> 2) map the nonleaf entries of the tdp page table; and
> 3) map the target leaf entry.
> 
> Improve the readabiliy by moving the implementation of 2) above into a
> subfunction, kvm_tdp_mmu_map_nonleaf, and removing the unnecessary
> "goto"s. No functional changes intended.

Eh, I prefer the current code from a readability perspective.  I like being able
to see the entire flow, and I especially like that this

		if (iter.level == fault->goal_level)
			goto map_target_level;

very clearly and explicitly captures that reaching the goal leavel means that it's
time to map the target level, whereas IMO this does not, in no small part because
seeing "continue" in a loop makes me think "continue the loop", not "continue on
to the next part of the page fault"

		if (iter->level == fault->goal_level)
			return RET_PF_CONTINUE;

And the existing code follows the patter of the other page fault paths, direct_map()
and FNAME(fetch).  That doesn't necessarily mean that the existing pattern is
"better", but I personally place a lot of value on consistency.

> +/*
> + * Handle a TDP page fault (NPT/EPT violation/misconfiguration) by installing
> + * page tables and SPTEs to translate the faulting guest physical address.
> + */
> +int kvm_tdp_mmu_map(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault)
> +{
> +	struct tdp_iter iter;
> +	int ret;
> +
> +	kvm_mmu_hugepage_adjust(vcpu, fault);
> +
> +	trace_kvm_mmu_spte_requested(fault);
> +
> +	rcu_read_lock();
> +
> +	ret = kvm_tdp_mmu_map_nonleafs(vcpu, fault, &iter);
> +	if (ret == RET_PF_CONTINUE)
> +		ret = tdp_mmu_map_handle_target_level(vcpu, fault, &iter);

And I also don't like passing in an uninitialized tdp_iter, and then consuming
it too.

>  
> -retry:
>  	rcu_read_unlock();
>  	return ret;
>  }
> -- 
> 2.27.0
>