static config for CPUs with K8 example

[ron minnich]

OK, I am hoping this is the kind of thing that will work for the K8.

First, in cpu/k8, define a file, chip.h:

struct cpu_k8_config {
        struct chip *north, *south, *east, *west;
};

Thenin cpu/k8/Config.lb, add this line:
config chip.h

Note this file could be called anything, but chip.h is a habit for me.

In the mainboard config (mainboard/arima/hdama/Config.lb), you set things 
up as follows:

southbridge amd/amd8111 "amd8111"
end
southbridge amd/amd8131 "amd8131"
end

# pull in all includes, etc. for  the k8. 
# should we have a 'noise keyword' for this, e.g. 'cputype k8'?

dir /cpu/k8

# define the CPUs, their names, and their connections.
cpu k8 "cpu0"
  register "north" = "amd8111"
  register "east"  = "cpu1"
end

cpu k8 "cpu1"
  register "north" = "amd8131"
  register "west"  = "cpu0"
end

This will result in initialized structures, per-cpu, that "do the right 
thing".

I've hit a bug I have to ask Greg about, but this is a rough idea. 
Comments?

ron

[Stefan Reinauer]

* ron minnich  [030731 00:44]:
> OK, I am hoping this is the kind of thing that will work for the K8.
>
> First, in cpu/k8, define a file, chip.h:
>
> struct cpu_k8_config {
>         struct chip *north, *south, *east, *west;
> };

I don't think we need both of east and west really. The Bios Developers
Guide calls these links "UP", "DOWN" and "ACROSS", for LDT0-2. 
It might still be interesting to have a topological view on the
hardware, but at least for hypertransport setup, east and west is equal.
> In the mainboard config (mainboard/arima/hdama/Config.lb), you set things 
> up as follows:
>
> southbridge amd/amd8111 "amd8111"

register "north"     = "cpu0" 
    register "ht_width"  = "8"      # 8bit, even if device reports more
    register "ht_speed"  = "200"    # don't drive faster than 200MHz,
                                    # in case device reports false
                                    # maximum speed
> end
> southbridge amd/amd8131 "amd8131"

register "north" = "cpu1" (for example)
    register "ht_width"  = "8"
    register "ht_speed"  = "200"
> end
> # pull in all includes, etc. for  the k8. 
> # should we have a 'noise keyword' for this, e.g. 'cputype k8'?

noise keyword?
> dir /cpu/k8

^^^^^^^^^^^

can this not implicitly be generated from the below description?
> # define the CPUs, their names, and their connections.
> cpu k8 "cpu0"
>   register "north" = "amd8111"

^^^^^ south (iirc)
> register "east"  = "cpu1"
> end
>
> cpu k8 "cpu1"
>   register "north" = "amd8131"

^^^^^ south
> register "west"  = "cpu0"
> end

Stefan

[ron minnich]

On Thu, 31 Jul 2003, Stefan Reinauer wrote:
> > First, in cpu/k8, define a file, chip.h:
> >
> > struct cpu_k8_config {
> >         struct chip *north, *south, *east, *west;
> > };
>
> I don't think we need both of east and west really. The Bios Developers
> Guide calls these links "UP", "DOWN" and "ACROSS", for LDT0-2. 
> It might still be interesting to have a topological view on the
> hardware, but at least for hypertransport setup, east and west is equal.

OK. I also occured to me last night that we need this:

struct ht_link {
    struct chip *chip;
    unsigned int ht_width, ht_speed;
};

then:

southbridge amd/amd8111 "amd8111"
    register "north" = "{.chip = &cpu0, .ht_width = 8, .ht_speed=200"
end

and so on.  Make sense? Or not? I don't have the data book so don't know 
if the 8111 has more than one link.

For the CPUs we then have

struct cpu_k8_config {
    struct ht_link up, down, across;
};
> > dir /cpu/k8
>
> ^^^^^^^^^^^
>
> can this not implicitly be generated from the below description?

yes, and it should be, I had the same thought.
> > # define the CPUs, their names, and their connections.

cpu k8 "cpu0"

register "down" = "{.chip = &amd8111, .ht_width=8, .ht_speed=200}"
    .
    .
end

better?

ron

p.s. the chip-specific structure for the PCI bridge will, of course, have
a pointer to the top of the PCI tree, and that becomes our link from the
static to the dynamic devices.