[Qemu-devel] Re: [PATCH V5 15/29] pci: typedef pcibus_t as uint64_t instead of uint32_t.

["Michael S. Tsirkin" <mst@redhat.com>]

On Tue, Oct 13, 2009 at 10:31:33PM +0900, Isaku Yamahata wrote:
> On Sun, Oct 11, 2009 at 12:43:12PM +0200, Michael S. Tsirkin wrote:
> > On Fri, Oct 09, 2009 at 03:28:48PM +0900, Isaku Yamahata wrote:
> > > This patch is preliminary for 64bit bar.
> > > For 64bit bar support, change pcibus_t which represents
> > > pci bus addr/size from uint32_t to uint64_t.
> > > And also change FMT_pcibus for printf.
> > > 
> > > In pci_update_mapping() checks 32bit overflow.
> > > So the check must be updated too.
> > > 
> > > Signed-off-by: Isaku Yamahata <yamahata@valinux.co.jp>
> > 
> > That's all fine, but if you look at users implementing
> > map io, they do: cpu_register_physical_memory()
> > on the address they are given.  And if target_phys_addr_t is 32 bit,
> > this will silently truncate the address.
> > 
> > So I would like to understand how this will all
> > work on 32 bit systems.
> 
> The case is
>   . BAR is memory 64bit and
>   . target_phys_addr_t is 32bit and
>   . bar is set to >4G.
> Hmm, the case isn't checked.
> 
> It would be checked by
> -	last_addr <= new_addr
> +	(target_phys_addr_t)last_addr <= new_addr

That's pretty tricky. Can we just convert everything into
64 bit unconditionally and just do simple range checks?

> 
> I'll fix it with comments added. Nice catch.

Is this the right thing to do though?
I think 32 bit CPU might address something like 64G
memory of highmem support, so a 64 bit value might
actually be valid.

Let's step back and understand what the motivation is?
Maybe declaring all bars as 32 bit for 32 bit targets is enough?

> 
> > 
> > > ---
> > >  hw/pci.c |    9 ++++++++-
> > >  hw/pci.h |    4 ++--
> > >  2 files changed, 10 insertions(+), 3 deletions(-)
> > > 
> > > diff --git a/hw/pci.c b/hw/pci.c
> > > index c0ae66a..e7f8fb4 100644
> > > --- a/hw/pci.c
> > > +++ b/hw/pci.c
> > > @@ -524,7 +524,14 @@ static void pci_update_mappings(PCIDevice *d)
> > >                         mappings, we handle specific values as invalid
> > >                         mappings. */
> > >                      if (last_addr <= new_addr || new_addr == 0 ||
> > > -                        last_addr == PCI_BAR_UNMAPPED) {
> > > +                        last_addr == PCI_BAR_UNMAPPED ||
> > > +
> > > +                        /* Now pcibus_t is 64bit.
> > > +                         * Check if 32 bit BAR wrap around explicitly.
> > > +                         * Without this, PC ide doesn't work well.
> > > +                         * TODO: remove this work around.
> > > +                         */
> > > +                        last_addr >= UINT32_MAX) {
> > >                          new_addr = PCI_BAR_UNMAPPED;
> > >                      }
> > >                  } else {
> > > diff --git a/hw/pci.h b/hw/pci.h
> > > index 8a187c2..8fbd45e 100644
> > > --- a/hw/pci.h
> > > +++ b/hw/pci.h
> > > @@ -71,8 +71,8 @@ extern target_phys_addr_t pci_mem_base;
> > >  #define PCI_DEVICE_ID_VIRTIO_BALLOON     0x1002
> > >  #define PCI_DEVICE_ID_VIRTIO_CONSOLE     0x1003
> > >  
> > > -typedef uint32_t pcibus_t;
> > > -#define FMT_PCIBUS                      PRIx32
> > > +typedef uint64_t pcibus_t;
> > > +#define FMT_PCIBUS                      PRIx64
> > >  
> > >  typedef void PCIConfigWriteFunc(PCIDevice *pci_dev,
> > >                                  uint32_t address, uint32_t data, int len);
> > 
> 
> -- 
> yamahata