Re: [PATCH] [EFI,PCI] Allow disabling PCI busmastering on bridges during boot

From: Ard Biesheuvel
Date: Tue Dec 03 2019 - 11:33:48 EST


On Tue, 3 Dec 2019 at 15:30, Andy Lutomirski <luto@xxxxxxxxxxxxxx> wrote:
>
> On Mon, Dec 2, 2019 at 4:41 PM Matthew Garrett
> <matthewgarrett@xxxxxxxxxx> wrote:
> >
> > Add an option to disable the busmaster bit in the control register on
> > all PCI bridges before calling ExitBootServices() and passing control to
> > the runtime kernel. System firmware may configure the IOMMU to prevent
> > malicious PCI devices from being able to attack the OS via DMA. However,
> > since firmware can't guarantee that the OS is IOMMU-aware, it will tear
> > down IOMMU configuration when ExitBootServices() is called. This leaves
> > a window between where a hostile device could still cause damage before
> > Linux configures the IOMMU again.
> >
> > If CONFIG_EFI_NO_BUSMASTER is enabled or the "disable_busmaster=1"
> > commandline argument is passed, the EFI stub will clear the busmaster
> > bit on all PCI bridges before ExitBootServices() is called. This will
> > prevent any malicious PCI devices from being able to perform DMA until
> > the kernel reenables busmastering after configuring the IOMMU.
>
> I hate to be an obnoxious bikeshedder, but I really dislike the
> "disable_busmaster" name. I read this and $SUBJECT as "for some
> reason, the admin wants to operate the system with busmastering off".
> What you really want is something more like "disable busmastering
> before IOMMU initialization". Maybe
> "iommu.disable_busmaster_before_init"?
>
> Similarly, EFI_NO_BUSMASTER sounds like a permanent state of affairs.
>
> Would a similar patch apply to non-EFI boot? That is, in a BIOS boot,
> is busmastering on when the kernel is loaded?
>

Yes, bus mastering is on, but since legacy BIOS may implement things
like PS/2 emulation or other compatibility hacks where the PCI masters
(devices or bridges) may need to be left enabled across the transition
from firmware into the OS, I don't think it is wise to try and
implement this feature for it.

So the EFI stub is a reasonable place to put a feature like this,
except for the fact that [on x86], it does not get invoked unless GRUB
boots your kernel with 'linuxefi' rather than 'linux', and so in the
majority of cases, I guess we are essentially doing legacy BIOS boot,
even on UEFI systems.