Re: [Qemu-devel] [PATCH v2] mac99: Change memory layout to better match PowerMac3, 1

[BALATON Zoltan]

On Mon, 23 Jun 2014, Alexander Graf wrote:
> On 17.06.14 12:24, BALATON Zoltan wrote:
> > On Tue, 17 Jun 2014, Alexander Graf wrote:
> > > On 17.06.14 11:36, BALATON Zoltan wrote:
> > > > On Tue, 17 Jun 2014, Alexander Graf wrote:
> > > > > On 12.04.14 11:20, BALATON Zoltan wrote:
> > > > > > Bring the memory map closer to a PowerMac3,1 model by removing unused
> > > > > > areas and adding the VGA and network cards after the macio to let the
> > > > > > latter be mapped from 0x80000000 like on real hardware. (On real
> > > > > > hardware the graphics and network cards are on separate buses but we
> > > > > > don't model that yet.)
> > > > > >
> > > > > > Signed-off-by: BALATON Zoltan <address@hidden>
> > > > > > ---
> > > > > >
> > > > > > This patch is intended to bring memory layout closer to what's seen in
> > > > > > these dumps:
> > > > > >
> > > > > > http://nandra.segv.jp/NetBSD/G4.dump-device-tree.txt
> > > > > > http://raveland.org/ports/eeprom.txt
> > > > > > http://mail-index.netbsd.org/port-macppc/2007/10/24/0000.html
> > > > > > https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=604134
> > > > > >
> > > > > > v2: Added back unin2_memory region that Darwin seems to like better
> > > > > >
> > > > > > ---
> > > > > >   hw/pci-host/uninorth.c |  2 +-
> > > > > >   hw/ppc/mac_newworld.c  | 14 +++++++-------
> > > > > >   2 files changed, 8 insertions(+), 8 deletions(-)
> > > > > >
> > > > > > diff --git a/hw/pci-host/uninorth.c b/hw/pci-host/uninorth.c
> > > > > > index e72fe2a..21f805f 100644
> > > > > > --- a/hw/pci-host/uninorth.c
> > > > > > +++ b/hw/pci-host/uninorth.c
> > > > > > @@ -230,7 +230,7 @@ PCIBus *pci_pmac_init(qemu_irq *pic,
> > > > > >       d = UNI_NORTH_PCI_HOST_BRIDGE(dev);
> > > > > >       memory_region_init(&d->pci_mmio, OBJECT(d), "pci-mmio", 
> > > > > > 0x100000000ULL);
> > > > > >       memory_region_init_alias(&d->pci_hole, OBJECT(d), "pci-hole", 
> > > > > > &d->pci_mmio,
> > > > > > -                             0x80000000ULL, 0x70000000ULL);
> > > > > > +                             0x80000000ULL, 0x10000000ULL);
> > > > >
> > > > > Doesn't OpenBIOS need to know about this change so it can update its 
> > > > > device tree?
> > > >
> > > > No.
> > >
> > > We don't expose the pci-hole size in device tree?
> >
> > We do but already as 0x10000000. See:
> > http://tracker.coreboot.org/trac/openbios/browser/trunk/openbios-devel/arch/ppc/qemu/init.c 
>
> Ugh. Please send a single patch that fixes this with a proper patch 
> description indicating that OpenBIOS already exposes the pci hole with 
> 0x10000000 size.

Sent to the list.

> > > > > > memory_region_add_subregion(address_space_mem, 0x80000000ULL,
> > > > > >                                   &d->pci_hole);
> > > > > >   diff --git a/hw/ppc/mac_newworld.c b/hw/ppc/mac_newworld.c
> > > > > > index 4bdaa8d..a4e5044 100644
> > > > > > --- a/hw/ppc/mac_newworld.c
> > > > > > +++ b/hw/ppc/mac_newworld.c
> > > > > > @@ -371,18 +371,11 @@ static void ppc_core99_init(MachineState 
> > > > > > *machine)
> > > > > >           machine_arch = ARCH_MAC99;
> > > > > >       }
> > > > > >       /* init basic PC hardware */
> > > > > > -    pci_vga_init(pci_bus);
> > > > > > -
> > > > > >       escc_mem = escc_init(0, pic[0x25], pic[0x24],
> > > > > >                            serial_hds[0], serial_hds[1], ESCC_CLOCK, 
> > > > > > 4);
> > > > > >       memory_region_init_alias(escc_bar, NULL, "escc-bar",
> > > > > >                                escc_mem, 0, 
> > > > > > memory_region_size(escc_mem));
> > > > > >   -    for(i = 0; i < nb_nics; i++)
> > > > > > -        pci_nic_init_nofail(&nd_table[i], pci_bus, "ne2k_pci", NULL);
> > > > > > -
> > > > > > -    ide_drive_get(hd, MAX_IDE_BUS);
> > > > > > -
> > > > > >       macio = pci_create(pci_bus, -1, TYPE_NEWWORLD_MACIO);
> > > > > >       dev = DEVICE(macio);
> > > > > >       qdev_connect_gpio_out(dev, 0, pic[0x19]); /* CUDA */
> > > > > > @@ -393,6 +386,8 @@ static void ppc_core99_init(MachineState *machine)
> > > > > >       macio_init(macio, pic_mem, escc_bar);
> > > > > >         /* We only emulate 2 out of 3 IDE controllers for now */
> > > > > > +    ide_drive_get(hd, MAX_IDE_BUS);
> > > > > > +
> > > > > >       macio_ide = 
> > > > > > MACIO_IDE(object_resolve_path_component(OBJECT(macio),
> > > > > > "ide[0]"));
> > > > > >       macio_ide_init_drives(macio_ide, hd);
> > > > > > @@ -418,9 +413,14 @@ static void ppc_core99_init(MachineState *machine)
> > > > > >           }
> > > > > >       }
> > > > > >   +    pci_vga_init(pci_bus);
> > > > > > +
> > > > > >       if (graphic_depth != 15 && graphic_depth != 32 && graphic_depth 
> > > > > > != 8)
> > > > > >           graphic_depth = 15;
> > > > > >   +    for(i = 0; i < nb_nics; i++)
> > > > > > +        pci_nic_init_nofail(&nd_table[i], pci_bus, "ne2k_pci", NULL);
> > > > > > +
> > > > > >       /* The NewWorld NVRAM is not located in the MacIO device */
> > > > > >       dev = qdev_create(NULL, TYPE_MACIO_NVRAM);
> > > > > >       qdev_prop_set_uint32(dev, "size", 0x2000);
> > > > >
> > > > > I presume all the changes above only change the devfn ordering?
> > > >
> > > > It changes the addresses assigned to devices which is needed for MorphOS 
> > > > to work as it hardcodes the mmio locations of some (actually most) 
> > > > devices.
> > >
> > > I don't see how the ordering change here could possibly change MMIO 
> > > locations for anything?
> >
> > It's how OpenBIOS assigns MMIO addresses to pci devices. It does it by 
> > going through them in order and map them starting from the base address 
> > (with some allingment). I guess you could look at drivers/pci.c I think 
> > it's in there somewhere.
>
> I think it'd make more sense to just bolt the PCI devices to their respective 
> devfns that they also have on real hardware. Depending on ordering magic that 
> happens to give us different BAR maps by firmware doesn't really give me a 
> lot of confidence.

I don't understand what you mean. I don't want to rewrite PCI handling 
code in OpenBIOS as that would have a higher chance of breaking something 
else (OpenBIOS is used by other archs as well). Also it would require more 
knowledge about the emulated hardware in OpenBIOS while it aims to be a 
generic implementation and wants to reduce special cases already in it. So 
I don't see a cleaner and easy way to do this. If I'm missing something 
please tell me. On the other hand you've said before that the mac99 
machine is mostly a hack to be enough that some OS-es can run with it. Why 
reordering some devices to get the right BAR maps not fit in this hack?

Regards,
BALATON Zoltan