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:   Mon, 23 Nov 2020 14:56:06 +0000
From:   Will Deacon <will@kernel.org>
To:     Brijesh Singh <brijesh.singh@amd.com>
Cc:     Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>,
        linux-kernel@vger.kernel.org, iommu@lists.linux-foundation.org,
        joro@8bytes.org, Jon.Grimm@amd.com, bp@suse.de
Subject: Re: [PATCH v2] iommu/amd: Enforce 4k mapping for certain IOMMU data
 structures
Message-ID: <20201123145605.GA11033@willie-the-truck>
References: <20201105145832.3065-1-suravee.suthikulpanit@amd.com>
 <c189684a-27e5-c0c2-1629-063b9fb16957@amd.com>
 <35c6f7d8-f889-8c3c-2e01-1a9226babf0a@amd.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <35c6f7d8-f889-8c3c-2e01-1a9226babf0a@amd.com>
User-Agent: Mutt/1.10.1 (2018-07-13)
Precedence: bulk

On Thu, Nov 19, 2020 at 10:19:01PM -0600, Brijesh Singh wrote:
> On 11/19/20 8:30 PM, Suravee Suthikulpanit wrote:
> > On 11/18/20 5:57 AM, Will Deacon wrote:
> > > I think I'm missing something here. set_memory_4k() will break the
> > kernel
> > > linear mapping up into page granular mappings, but the IOMMU isn't
> > using
> > > that mapping, right?
> >
> > That's correct. This does not affect the IOMMU, but it affects the PSP
> > FW.
> >
> > > It's just using the physical address returned by
> > iommu_virt_to_phys(), so why does it matter?
> > >
> > > Just be nice to capture some of this rationale in the log,
> > especially as
> > > I'm not familiar with this device.
> >
> > According to the AMD SEV-SNP white paper
> > (https://www.amd.com/system/files/TechDocs/SEV-SNP-strengthening-vm-isolation-with-integrity-protection-and-more.pdf),
> > the Reverse Map Table (RMP) contains one entry for every 4K page of
> > DRAM that may be used by the VM. In this case, the pages allocated by
> > the IOMMU driver are added as 4K entries in the RMP table by the
> > SEV-SNP FW.
> >
> > During the page table walk, the RMP checks if the page is owned by the
> > hypervisor. Without calling set_memory_4k() to break the mapping up
> > into 4K pages, pages could end up being part of large mapping (e.g. 2M
> > page), in which the page access would be denied and result in #PF.
> 
> 
> Since the page is added as a 4K page in the RMP table by the SEV-SNP FW,
> so we need to split the physmap to ensure that this page will be access
> with a 4K mapping from the x86. If the page is part of large page then
> write access will cause a RMP violation (i.e #PF), this is because SNP
> hardware enforce that the CPU page level walk must match with page-level
> programmed in the RMP table.

Got it; thanks.

> > >> Fixes: commit c69d89aff393 ("iommu/amd: Use 4K page for completion
> > wait write-back semaphore")
> > >
> > > I couldn't figure out how that commit could cause this problem.
> > Please can
> > > you explain that to me?
> >
> > Hope this helps clarify. If so, I'll update the commit log and send
> > out V3.

Cheers. No need for a v2, as I've queued this up with a Link: tag.

Will