linux-kernel - Re: [PATCH v3 03/28] drm/i915/gvt: Verify hugepages are contiguous in physical address space

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

Date:   Thu, 18 May 2023 11:04:46 -0700
From:   Sean Christopherson <seanjc@...gle.com>
To:     Yan Zhao <yan.y.zhao@...el.com>
Cc:     Paolo Bonzini <pbonzini@...hat.com>,
        Zhenyu Wang <zhenyuw@...ux.intel.com>,
        Zhi Wang <zhi.a.wang@...el.com>, kvm@...r.kernel.org,
        intel-gvt-dev@...ts.freedesktop.org,
        intel-gfx@...ts.freedesktop.org, linux-kernel@...r.kernel.org,
        Ben Gardon <bgardon@...gle.com>
Subject: Re: [PATCH v3 03/28] drm/i915/gvt: Verify hugepages are contiguous in
 physical address space

On Thu, May 18, 2023, Yan Zhao wrote:
> On Wed, May 17, 2023 at 07:50:26AM -0700, Sean Christopherson wrote:
> > On Tue, May 16, 2023, Yan Zhao wrote:
> > > hi Sean
> > > 
> > > Do you think it's necessary to double check that struct page pointers
> > > are also contiguous?
> > 
> > No, the virtual address space should be irrelevant.  The only way it would be
> > problematic is if something in dma_map_page() expected to be able to access the
> > entire chunk of memory by getting the virtual address of only the first page,
> > but I can't imagine that code is reading or writing memory, let alone doing so
> > across a huge range of memory.
> Yes, I do find arm_iommu version of dma_map_page() access the memory by getting
> virtual address of pages passed in, but it's implemented as page by page, not only
> from the first page.
> 
> dma_map_page
>   dma_map_page_attrs
>     ops->map_page
>       arm_iommu_map_page

Heh, thankfully this is ARM specific, which presumably doesn't collide with KVMGT.

>          __dma_page_cpu_to_dev
>            dma_cache_maint_page
> 
> Just a little worried about the condition of PFNs are contiguous
> while they belong to different backends, e.g. one from system memory and
> one from MMIO.
> But I don't know how to avoid this without complicated checks.
> And this condition might not happen in practice.

IMO, assuming that contiguous pfns are vritually contiguous is wrong, i.e. would
be a bug in the other code.  The above dma_cache_maint_page() get's this right,
and even has a well written comment to boot.