Re: [Qemu-devel] [PATCH 2/7] pci: memory access API and IOMMU support

[Eduard - Gabriel Munteanu]

On Wed, Sep 01, 2010 at 10:10:30PM +0200, Stefan Weil wrote:
> Please see my comments at the end of this mail.
> 
> 
> Am 30.08.2010 00:08, schrieb Eduard - Gabriel Munteanu:
> > PCI devices should access memory through pci_memory_*() instead of
> > cpu_physical_memory_*(). This also provides support for translation and
> > access checking in case an IOMMU is emulated.
> >
> > Memory maps are treated as remote IOTLBs (that is, translation caches
> > belonging to the IOMMU-aware device itself). Clients (devices) must
> > provide callbacks for map invalidation in case these maps are
> > persistent beyond the current I/O context, e.g. AIO DMA transfers.
> >
> > Signed-off-by: Eduard - Gabriel Munteanu <address@hidden>
> > ---

[snip]

> > +static inline void pci_memory_read(PCIDevice *dev,
> > + pcibus_t addr,
> > + uint8_t *buf,
> > + pcibus_t len)
> > +{
> > + pci_memory_rw(dev, addr, buf, len, 0);
> > +}
> > +
> > +static inline void pci_memory_write(PCIDevice *dev,
> > + pcibus_t addr,
> > + const uint8_t *buf,
> > + pcibus_t len)
> > +{
> > + pci_memory_rw(dev, addr, (uint8_t *) buf, len, 1);
> > +}
> > +
> > #endif
> 
> The functions pci_memory_read and pci_memory_write not only read
> or write byte data but many different data types which leads to
> a lot of type casts in your other patches.
> 
> I'd prefer "void *buf" and "const void *buf" in the argument lists.
> Then all those type casts could be removed.
> 
> Regards
> Stefan Weil

I only followed an approach similar to how cpu_physical_memory_{read,write}()
is defined. I think I should change both cpu_physical_memory_* stuff and
pci_memory_* stuff, not only the latter, if I decide to go on that
approach.


        Eduard