[Qemu-devel] Re: [PATCH 14/23] pci: factor out the logic to get pci device from address.

[Michael S. Tsirkin]

On Tue, Oct 06, 2009 at 06:50:05PM +0900, Isaku Yamahata wrote:
> On Mon, Oct 05, 2009 at 02:45:33PM +0200, Michael S. Tsirkin wrote:
> > On Mon, Oct 05, 2009 at 07:06:54PM +0900, Isaku Yamahata wrote:
> > > factor out conversion logic from io port address into bus+dev+func
> > > with bit shift operation and conversion bus+dev+func into pci device.
> > > They will be used later by pcie support.
> > > This patch also eliminates the logic duplication.
> > 
> > Please split removing logic duplication into separate patch.
> > It might be applicable already, while express stuff needs more work IMO.
> > 
> > > Signed-off-by: Isaku Yamahata <address@hidden>
> > > ---
> > >  hw/pci.c |   83 
> > > ++++++++++++++++++++++++++++++++++---------------------------
> > >  1 files changed, 46 insertions(+), 37 deletions(-)
> > > 
> > > diff --git a/hw/pci.c b/hw/pci.c
> > > index 6ddd256..5f808ff 100644
> > > --- a/hw/pci.c
> > > +++ b/hw/pci.c
> > > @@ -629,44 +629,25 @@ void pci_default_write_config(PCIDevice *d, 
> > > uint32_t addr, uint32_t val, int l)
> > >          pci_update_mappings(d);
> > >  }
> > >  
> > > -void pci_data_write(void *opaque, uint32_t addr, uint32_t val, int len)
> > > +static void pci_data_write_common(PCIDevice *pci_dev,
> > > +                                  uint32_t config_addr, uint32_t val, 
> > > int len)
> > 
> > So as I said in the previous comment, pci express should just be able to
> > use pci code directly, instead of reusing this low-level stuff.
> > All you need to do is range-check address and ignore extended registers.
> 
> Maybe my patch description isn't good enough,
> I'm afraid that you misunderstood the patch purpose.
> 
> PCI configuration and PCI express MMCONFIG uses different address scheme
> as follows.
> This patch purpose is to separate out its address parsing logic
> from pci config space update logic.
> 
> PCI:
>  0- 7 bit: offset in the configuration space (256bytes)
>  7-15 bit: devfn
> 16-14 bit: bus
> 
> PCI express:
>  0-11 bit: offset in the configuration space (4KBytes)
> 12-19 bit: devfn
> 20-28 bit: bus

The comment is basically that there's no common code:
you don't need len checks that you added for mmconfig (they are there
just for documentation purposes), and so all pci_data_write does is
address parsing.  So just call pci_dev->config_write for both mmconfig
and config.

> > >  {
> > > -    PCIBus *s = opaque;
> > > -    PCIDevice *pci_dev;
> > > -    int config_addr, bus_num;
> > > -
> > > -#if 0
> > > -    PCI_DPRINTF("pci_data_write: addr=%08"PRIx32" val=%08"PRIx32" 
> > > len=%d\n",
> > > -                addr, val, len);
> > > -#endif
> > > -    bus_num = (addr >> 16) & 0xff;
> > > -    s = pci_find_bus(s, bus_num);
> > > -    if (!s)
> > > -        return;
> > > -    pci_dev = s->devices[(addr >> 8) & 0xff];
> > > +    assert(len == 1 || len == 2 || len == 4);
> > >      if (!pci_dev)
> > >          return;
> > > -    config_addr = addr & 0xff;
> > > -    PCI_DPRINTF("pci_config_write: %s: "
> > > -                "addr=%02x val=%08"PRI32x" len=%d\n",
> > > -                pci_dev->name, config_addr, val, len);
> > > +
> > > +    PCI_DPRINTF("%s: %s: addr=%02"PRIx32" val=%08"PRI32x" len=%d\n",
> > > +                __func__, pci_dev->name, config_addr, val, len);
> > >      pci_dev->config_write(pci_dev, config_addr, val, len);
> > >  }
> > >  
> > > -uint32_t pci_data_read(void *opaque, uint32_t addr, int len)
> > > +static uint32_t pci_data_read_common(PCIDevice *pci_dev,
> > > +                                     uint32_t config_addr, int len)
> > >  {
> > > -    PCIBus *s = opaque;
> > > -    PCIDevice *pci_dev;
> > > -    int config_addr, bus_num;
> > >      uint32_t val;
> > >  
> > > -    bus_num = (addr >> 16) & 0xff;
> > > -    s = pci_find_bus(s, bus_num);
> > > -    if (!s)
> > > -        goto fail;
> > > -    pci_dev = s->devices[(addr >> 8) & 0xff];
> > > +    assert(len == 1 || len == 2 || len == 4);
> > >      if (!pci_dev) {
> > > -    fail:
> > >          switch(len) {
> > >          case 1:
> > >              val = 0xff;
> > > @@ -679,21 +660,49 @@ uint32_t pci_data_read(void *opaque, uint32_t addr, 
> > > int len)
> > >              val = 0xffffffff;
> > >              break;
> > >          }
> > > -        goto the_end;
> > > +    } else {
> > > +        val = pci_dev->config_read(pci_dev, config_addr, len);
> > > +        PCI_DPRINTF("%s: %s: addr=%02"PRIx32" val=%08"PRIx32" len=%d\n",
> > > +                    __func__, pci_dev->name, config_addr, val, len);
> > >      }
> > > -    config_addr = addr & 0xff;
> > > -    val = pci_dev->config_read(pci_dev, config_addr, len);
> > > -    PCI_DPRINTF("pci_config_read: %s: "
> > > -                "addr=%02x val=%08"PRIx32" len=%d\n",
> > > -                pci_dev->name, config_addr, val, len);
> > > - the_end:
> > > +
> > >  #if 0
> > > -    PCI_DPRINTF("pci_data_read: addr=%08"PRIx32" val=%08"PRIx32" 
> > > len=%d\n",
> > > -                addr, val, len);
> > > +    PCI_DPRINTF("%s: addr=%08"PRIx32" val=%08"PRIx32" len=%d\n",
> > > +                __func__, addr, val, len);
> > >  #endif
> > 
> > move to real pci_data_read?
> > 
> > >      return val;
> > >  }
> > >  
> > > +static inline PCIDevice *pci_addr_to_dev(PCIBus *bus, uint32_t addr)
> > > +{
> > > +    uint8_t bus_num = (addr >> 16) & 0xff;
> > > +    uint8_t devfn = (addr >> 8) & 0xff;
> > > +    return pci_find_device(bus, bus_num, PCI_SLOT(devfn), 
> > > PCI_FUNC(devfn));
> > > +}
> > > +
> > > +static inline uint32_t pci_addr_to_config(uint32_t addr)
> > > +{
> > > +    return addr & 0xff;
> > > +}
> > > +
> > > +void pci_data_write(void *opaque, uint32_t addr, uint32_t val, int len)
> > > +{
> > > +    PCIBus *s = opaque;
> > > +#if 0
> > > +    PCI_DPRINTF("pci_data_write: addr=%08"PRIx32" val=%08"PRIx32" 
> > > len=%d\n",
> > > +                addr, val, len);
> > > +#endif
> > > +
> > > +    pci_data_write_common(pci_addr_to_dev(s, addr), 
> > > pci_addr_to_config(addr),
> > > +                          val, len);
> > 
> > For pci express, there's extended register number in bits 11:8
> > so what we should do here IMO is check these bits and
> > ignore if they are not 0.
> 
> I don't understand what you suggest.
> pci_data_{write, read}() are glue for ioio or mmio to
> access pci configuration space. And 'addr' parameter is used
> to specify configuration space access.
> It includes not only the offset in configuration space, but also
> devfn and bus.
> 

Okay. I was misled by naming.  What I am saying, here and elsewhere, is
that a large part of pain is because of 256 byte/4K byte difference.
Since in express you can hardware addresses > 256 to 0, I propose that
we simply make mmconfig code do this:

if (offset > 256)
        return 0;

And then basically nothing needs to be changed in pci.

> 
> > > +}
> > > +
> > > +uint32_t pci_data_read(void *opaque, uint32_t addr, int len)
> > > +{
> > > +    PCIBus *s = opaque;
> > 
> > For pci express, there's extended register number in bits 11:8
> > so what we should do here IMO is check these bits and
> > return 0 if not 0.
> > 
> > > +    return pci_data_read_common(pci_addr_to_dev(s, addr),
> > > +                                pci_addr_to_config(addr), len);
> > > +}
> > >  /***********************************************************/
> > >  /* generic PCI irq support */
> > >  
> > 
> 
> -- 
> yamahata