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]
Message-ID: <ZuHC-G575S4A-S_m@google.com>
Date: Wed, 11 Sep 2024 09:19:04 -0700
From: Sean Christopherson <seanjc@...gle.com>
To: Dave Hansen <dave.hansen@...el.com>
Cc: Alexey Gladkov <legion@...nel.org>, linux-kernel@...r.kernel.org, 
	linux-coco@...ts.linux.dev, Thomas Gleixner <tglx@...utronix.de>, 
	Ingo Molnar <mingo@...hat.com>, Borislav Petkov <bp@...en8.de>, 
	Dave Hansen <dave.hansen@...ux.intel.com>, "H. Peter Anvin" <hpa@...or.com>, 
	"Kirill A. Shutemov" <kirill.shutemov@...ux.intel.com>, Andrew Morton <akpm@...ux-foundation.org>, 
	Yuan Yao <yuan.yao@...el.com>, Geert Uytterhoeven <geert@...ux-m68k.org>, 
	Yuntao Wang <ytcoode@...il.com>, Kai Huang <kai.huang@...el.com>, Baoquan He <bhe@...hat.com>, 
	Oleg Nesterov <oleg@...hat.com>, cho@...rosoft.com, decui@...rosoft.com, 
	John.Starks@...rosoft.com, Paolo Bonzini <pbonzini@...hat.com>
Subject: Re: [PATCH v6 0/6] x86/tdx: Allow MMIO instructions from userspace

On Wed, Sep 11, 2024, Dave Hansen wrote:
> On 9/6/24 14:13, Sean Christopherson wrote:
> > Ditto for what behavior is supported/allowed.  The kernel could choose to disallow
> > userspace MMIO entirely, limit what instructions are supported, etc, in the name
> > of security, simplicity, or whatever.   Doing so would likely cause friction with
> > folks that want to run their workloads in an SNP/TDX VM, but that friction is very
> > much with the guest kernel, not with KVM.
> 
> I think by "guest kernel" you really mean "x86 maintainers".  Thanks for
> throwing us under the bus, Sean. ;)

Heh, I would argue that you tried to push me under the bus, but I'm slippery fast
and danced out of the way, and you got hit instead :-D

> I do agree with you, though.  In the process of taking the VMM out of
> the TCB, confidential computing has to fill the gap with _something_ and
> that something is usually arch-specific code in the guest kernel.
> 
> By dragging the KVM folks in here, I was less asking what KVM does per
> se and more asking for some advice from the experienced VMM folks.
> 
> > FWIW, emulating MMIO that isn't controlled by the kernel gets to be a bit of a
> > slippery slope, e.g. there are KVM patches on the list to support emulating AVX
> > instructions[*].  But, a major use case of any hypervisor is to lift-and-shift
> > workloads, and so KVM users, developers, and maintainers are quite motivated to
> > ensure that anything that works on bare metal also works on KVM.
> 
> Do you have a link for that AVX discussion?  I searched a bit but came
> up empty.

Gah, of course I forgot to paste the link.

https://lore.kernel.org/all/20240820230431.3850991-1-kbusch@meta.com

> The slippery slope is precisely what I'm worried about.  I suspect the
> AVX instructions are a combination of compilers that are increasingly
> happy to spit out AVX and users who just want to use whatever the
> compiler spits out on "pointers" in their apps that just happen to be
> pointed at MMIO.

Yep.  Based on the original report[*], it sounds like the userspace program is
doing a memcpy(), so it's hard to even argue that userspace is being silly.

[*] https://lore.kernel.org/kvm/20240304145932.4e685a38.alex.williamson@redhat.com

> But before we start digging in to avoid the slippery slope, we really do
> need to know more about the friction.  Who are we causing it for and how
> bad is it for them?

This type of issue will most likely show up in the form of an end customer moving
their workload into a TDX/SNP VM, and that workload crashing despite working just
fine when run in a regular VM.

One "answer" could be to tell users that they need to recompile with AVX+
explicitly disabled, but that's an answer that will make everyone unhappy.  E.g.
customers won't like recompiling, CSPs don't like unhappy customers, and CSPs and
hardware vendors don't want their CoCo solutions to be hard(er) to adopt.

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ