Re: [PATCH v5 11/11] intel_sgx: driver documentation

From: Sean Christopherson <sean.j.christopherson@intel.com>
To: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>,
	Thomas Gleixner <tglx@linutronix.de>
Cc: Darren Hart <dvhart@infradead.org>,
	platform-driver-x86@vger.kernel.org,
	LKML <linux-kernel@vger.kernel.org>, Ingo Molnar <mingo@elte.hu>,
	"H. Peter Anvin" <hpa@zytor.com>,
	x86@kernel.org, linux-doc@vger.kernel.org,
	Jonathan Corbet <corbet@lwn.net>
Subject: Re: [PATCH v5 11/11] intel_sgx: driver documentation
Date: Mon, 27 Nov 2017 09:03:39 -0800	[thread overview]
Message-ID: <1511802219.9392.44.camel@intel.com> (raw)
In-Reply-To: <20171120230843.ozqhqhj5h3uuds3f@linux.intel.com>

On Tue, 2017-11-21 at 01:08 +0200, Jarkko Sakkinen wrote:
> On Sat, Nov 18, 2017 at 12:34:33AM +0100, Thomas Gleixner wrote:
> > 
> > This is architecural. From the cursory read of that series it seems there
> > are two parts to it:
> > 
> >   1) The actual core handling, which should be in arch/x86 because that
> >      hardly qualifies as a 'platform' device driver.
> > 
> >   2) The user space interface, which can be separated out perhaps.
> > 
> > I don't know how intertwingled they are, but that's hard to tell from the
> > actual patches w/o doing a deep inspection. Jarkko should be able to answer
> > that.
> > 
> > Thanks,
> > 
> > 	tglx
> Darren, tglx,
> 
> You can leave user space device as separate module as sgx_ioctl.c merely
> calls stuff that I have inside sgx_encl.c. VMA creation is bound to file
> operations.
> 
> My questions would be:
> 
> 1. What is your recommendation on the deployment under arch/x86?
> 2. Which parts should be compilable as a LKM? Only the user interface
>    or both parts?
> 
> /Jarkko

To enable KVM and a cgroup for EPC accounting, at a minimum arch/x86 needs to
manage the EPC pages (alloc/free/lrus/reclaim/etc...) and LE hash MSRs.  IMO,
ideally everything else would be left in the device driver, e.g. anything
involving ENCLS.  Keeping the majority of the driver out of arch/x86 minimizes
the footprint in arch/x86 and thereby the size of KVM's dependency required to
virtualize SGX, and allows the various SGX pieces, e.g. arch, driver and KVM, to
evolve more independently.

Preferably the arch/x86 code would not be a loadable module, e.g. to simplify
KVM support.

I have a branch based on Jarkko's patches (I believe it's up-to-date with v5)
that implements what I described.  I'd be happy to send RFC patches if that
would help.

Branches for those interested:

https://github.com/sean-jc/linux.git sgx/arch   - move core EPC to arch/x86
https://github.com/sean-jc/linux.git sgx/kvm    - KVM support for SGX
https://github.com/sean-jc/linux.git sgx/lc     - KVM support for Launch Control
https://github.com/sean-jc/linux.git sgx/cgroup - EPC cgroup

branch relationships:

    Jarkko's patches
            |
            |
         sgx/arch
        /        \
     sgx/kvm    sgx/cgroup
      /
   sgx/lc