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;
MIME-Version: 1.0
References: <161663543465.1867664.5674061943008380442.stgit@dwillia2-desk3.amr.corp.intel.com>
 <20210326161247.GA819704@bjorn-Precision-5520>
In-Reply-To: <20210326161247.GA819704@bjorn-Precision-5520>
From:   Dan Williams <dan.j.williams@intel.com>
Date:   Mon, 29 Mar 2021 09:46:15 -0700
Message-ID: <CAPcyv4h_c46tBpemKksciHL4DWu356h7T8A-0eHKUW9H3CZkKw@mail.gmail.com>
Subject: Re: [PATCH] PCI: Allow drivers to claim exclusive access to config regions
To:     Bjorn Helgaas <helgaas@kernel.org>
Cc:     Bjorn Helgaas <bhelgaas@google.com>,
        Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        Jonathan Cameron <Jonathan.Cameron@huawei.com>,
        Linux PCI <linux-pci@vger.kernel.org>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        Christoph Hellwig <hch@infradead.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Fri, Mar 26, 2021 at 9:12 AM Bjorn Helgaas <helgaas@kernel.org> wrote:
>
> [+cc Christoph]
>
> On Wed, Mar 24, 2021 at 06:23:54PM -0700, Dan Williams wrote:
> > The PCIE Data Object Exchange (DOE) mailbox is a protocol run over
> > configuration cycles. It assumes one initiator at a time is
> > reading/writing the data registers. If userspace reads from the response
> > data payload it may steal data that a kernel driver was expecting to
> > read. If userspace writes to the request payload it may corrupt the
> > request a driver was trying to send.
>
> IIUC the problem we're talking about is that userspace config access,
> e.g., via "lspci" or "setpci" may interfere with kernel usage of DOE.
> I attached what I think are the relevant bits from the spec.
>
> It looks to me like config *reads* should not be a problem: A read of
> Write Data Mailbox always returns 0 and looks innocuous.  A userspace
> read of Read Data Mailbox may return a DW of the data object, but it
> doesn't advance the cursor, so it shouldn't interfere with a kernel
> read.
>
> A write to Write Data Mailbox could obviously corrupt an object being
> written to the device.  A config write to Read Data Mailbox *does*
> advance the cursor, so that would definitely interfere with a kernel
> user.
>
> So I think we're really talking about an issue with "setpci" and I
> don't expect "lspci" to be a problem.  "setpci" is a valuable tool,
> and the fact that it can hose your system is not really news.  I don't
> know how hard we should work to protect against that.

True, the threat is smaller than I was reading, I apologize for that
noise. Temporary blocking over kernel DOE cycles seems sufficient for
now.