Re: [RFC PATCH 00/12] hw: Forbid DMA write accesses to MMIO regions

[Stefan Hajnoczi]

On Wed, Sep 9, 2020 at 2:23 PM Peter Maydell <peter.maydell@linaro.org> wrote:
> On Wed, 9 Sep 2020 at 10:12, Stefan Hajnoczi <stefanha@gmail.com> wrote:
> > On Thu, Sep 03, 2020 at 01:08:19PM +0200, Philippe Mathieu-Daudé wrote:
> > > The main patch is:
> > > "exec/memattrs: Introduce MemTxAttrs::direct_access field".
> > > This way we can restrict accesses to ROM/RAM by setting the
> > > 'direct_access' field. Illegal accesses return MEMTX_BUS_ERROR.
> >
> > QEMU needs to simulate the behavior of real hardware. What is the
> > behavior of real hardware?
>
> It varies, depending on the hardware. The most common thing
> is probably "happens to work by luck", which is OK for hardware
> but doesn't help us much since our software implementation is
> naturally more serialized than hardware is and since we don't
> want to allow guests to make QEMU crash or misbehave.

The memory API bounce buffer mechanism is evidence that some board(s)
need or needed it. At a minimum we need to find out the reason for the
bounce buffer mechanism to avoid breaking guests.

Stefan