ron minnich rminnich@lanl.gov writes:
The issue is that at the amdk8 northbridge, we need to know bus/dev/fn for the hardware so we can set up the VGA_EN bit in the right PCIIO pair as well as an MMIO entry for it.
Huh? There is a fairly definitive association between resources and non-coherent hypertransport chains. What do you need the device for?
It's not enough to just set a bit in the bridge on the K8 -- you have to set up routing to the right Hypertransport bus.
Right and we know that bus is the bus we set the VGA_EN bit for.
You have to know where the device is.
Yes but you should not need to know which device it is.
By far the easiest way to do this is to add a simple structure member to the bus structure: struct device *vgadev;
Huh?
so we have: struct bus { device_t dev; /* This bridge device */ device_t children; /* devices behind this bridge */ unsigned bridge_ctrl; /* Bridge control register */ /* NEW */ struct device *vgadev; /* if bridge_ctl has * PCI_CB_BRIDGE_CTL_VGA set, * this contains pointer to * the device. */ /* END NEW */ unsigned char link; /* The index of this link */ unsigned char secondary; /* secondary bus number */ unsigned char subordinate; /* max subordinate bus number */ unsigned char cap; /* PCi capability offset */ };
Setting vgadev is then trivial in the allocate_vga_resource since in that function you already have a pointer to the vga device; or just set the pointer. Either way, when you are at a bridge and know that the bridge has vga on the bus somewhere, you can easily get the info you need to set up the bridge if it is a complex bridge like the K8.
Yes the simple enough it just appears to be unnecessary.
while(bus) { bus->bridge_ctrl |= PCI_BRIDGE_CTL_VGA; /* NEW */ bus->vgadev = vga; /* END NEW */ bus = (bus == bus->dev->bus)? 0 : bus->dev->bus; }Unless there is a huge problem with this I will try to get it done tomorrow. It's a new structure member and one line of code and we're on the air.
It is not totally general but ... VGA is "special". As in, really ugly.
Yes, the vga routing at least needs a special case. It is a common case so there is no problem with that. The idea is to have a model that we progressively refine into what we need.
What I don't see is what you need vgadev for. The only case I can think of is short cutting to the emulator. In that case one global would be fine.
Looking at the code though there is a bug. When it finds a device to give the legacy vga resources to it does not allocate any MMIO resources. For bridges this is a normal resource so this looks a real bug in the generic code. If that is why you need the vgadev, let's fix the generic code to handle that part properly.
Eric