v3 patches and question

[Myles Watson]

serengeti.diff and dtc-links.diff make serengeti find all of its devices
correctly.

serengeti.diff
- takes out the multiple links for the Opteron (they weren't working
correctly)
- adds the 8132 devices and makes them pci_a so they don't conflict with the
8111
- corrects the disabled flags (I think we should have a build error
somewhere when there are incorrect flags)
- moves the lpc device to the lpc bus
- adds back in the second Opteron

Signed-off-by: Myles Watson

dtc-links.diff
- removes multiple link code.  I know someone could probably fix it, but
it's broken and not needed for now.
It made siblings that were bridges be multiple links on the same device.
That messed a lot of things up for Serengeti.  I think it could be changed
so that siblings with the same PCI identifier that are bridges become
multiple links.  It's ugly, though.

Signed-off-by: Myles Watson

pci_match_v2.diff:
Even though Serengeti seems to work the same as v2, it still won't show
anything on the display.  The VGA device is not claiming the 0xa0000-0xc0000
address space.  Since SimNOW is a simulator, I trace the memory space to see
that the VGA writes are going to RAM.

I put this in to make the PCI config space bit-for-bit identical at VGA ROM
run.  It still doesn't work. :(  Any help would be appreciated.

Thanks,
Myles

[Marc Jones]

On Wed, Dec 17, 2008 at 1:42 PM, Myles Watson  wrote:
> serengeti.diff and dtc-links.diff make serengeti find all of its devices
> correctly.
>
> serengeti.diff
> takes out the multiple links for the Opteron (they weren't working
>
> correctly)
> adds the 8132 devices and makes them pci_a so they don't conflict with the
>
> 8111
> corrects the disabled flags (I think we should have a build error
>
> somewhere when there are incorrect flags)
> moves the lpc device to the lpc bus
> adds back in the second Opteron
>
> Signed-off-by: Myles Watson

Acked-by: Marc Jones 
But what about amd8151?
> dtc-links.diff
> removes multiple link code.  I know someone could probably fix it, but
>
> it's broken and not needed for now.
> It made siblings that were bridges be multiple links on the same device.
> That messed a lot of things up for Serengeti.  I think it could be changed
> so that siblings with the same PCI identifier that are bridges become
> multiple links.  It's ugly, though.
>
> Signed-off-by: Myles Watson

So how does this allow amd8151 to be on a separate link?
> pci_match_v2.diff:
> Even though Serengeti seems to work the same as v2, it still won't show
> anything on the display.  The VGA device is not claiming the 0xa0000-0xc0000
> address space.  Since SimNOW is a simulator, I trace the memory space to see
> that the VGA writes are going to RAM.
>
> I put this in to make the PCI config space bit-for-bit identical at VGA ROM
> run.  It still doesn't work. :(  Any help would be appreciated.

You have probably look but amdk8_create_vga_resource is working correctly?

Marc

[Myles Watson]

On Thu, Dec 18, 2008 at 11:41 AM, Marc Jones  wrote:
> On Wed, Dec 17, 2008 at 1:42 PM, Myles Watson  wrote:
> > serengeti.diff and dtc-links.diff make serengeti find all of its devices
> > correctly.
> >
> > serengeti.diff
> > takes out the multiple links for the Opteron (they weren't working
> >
> > correctly)
> > adds the 8132 devices and makes them pci_a so they don't conflict with
>
> the
> > 8111
> > corrects the disabled flags (I think we should have a build error
> >
> > somewhere when there are incorrect flags)
> > moves the lpc device to the lpc bus
> > adds back in the second Opteron
> >
> > Signed-off-by: Myles Watson
>
> Acked-by: Marc Jones 
> But what about amd8151?

I don't know how to put it in correctly right now.  There isn't a working
way to represent multiple busses from a single device in the dts.  Since I
can't test the 8151 (free version of SimNow), this is as complete as I can
make it.

I won't commit since I have to change the dtc code to make this work.
> dtc-links.diff
> > removes multiple link code.  I know someone could probably fix it, but
> >
> > it's broken and not needed for now.
> > It made siblings that were bridges be multiple links on the same device.
> > That messed a lot of things up for Serengeti.  I think it could be
>
> changed
> > so that siblings with the same PCI identifier that are bridges become
> > multiple links.  It's ugly, though.
> >
> > Signed-off-by: Myles Watson
>
> So how does this allow amd8151 to be on a separate link?

It doesn't.  I'm not sure how to do that. (See above.)
> > pci_match_v2.diff:
> > Even though Serengeti seems to work the same as v2, it still won't show
> > anything on the display.  The VGA device is not claiming the
>
> 0xa0000-0xc0000
> > address space.  Since SimNOW is a simulator, I trace the memory space to
>
> see
> > that the VGA writes are going to RAM.
> >
> > I put this in to make the PCI config space bit-for-bit identical at VGA
>
> ROM
> > run.  It still doesn't work. :(  Any help would be appreciated.
>
> You have probably look but amdk8_create_vga_resource is working correctly?

It seems like it.  The register in the routing table gets set up correctly.
I didn't have to mess with that to get it exactly the same as v2.

Thanks,
Myles

[Myles Watson]

On Thu, Dec 18, 2008 at 11:51 AM, Myles Watson  wrote:
> On Thu, Dec 18, 2008 at 11:41 AM, Marc Jones  wrote:
> > On Wed, Dec 17, 2008 at 1:42 PM, Myles Watson  wrote:
> > > serengeti.diff and dtc-links.diff make serengeti find all of its devices
> > > correctly.
> > >
> > > serengeti.diff
> > > takes out the multiple links for the Opteron (they weren't working
> > >
> > > correctly)
> > > adds the 8132 devices and makes them pci_a so they don't conflict with
> >
> > the
> > > 8111
> > > corrects the disabled flags (I think we should have a build error
> > >
> > > somewhere when there are incorrect flags)
> > > moves the lpc device to the lpc bus
> > > adds back in the second Opteron
> > >
> > > Signed-off-by: Myles Watson
> >
> > Acked-by: Marc Jones 
> > But what about amd8151?
>
> I don't know how to put it in correctly right now.  There isn't a working
> way to represent multiple busses from a single device in the dts.  Since I
> can't test the 8151 (free version of SimNow), this is as complete as I can
> make it.

I forgot to add that it should fine if we just add the 8151 anywhere in the
dts.  That will make the drivers get compiled in, and then it will be found
and initialized correctly.  It's not pretty though.

Thanks,
Myles

[Myles Watson]

On Thu, Dec 18, 2008 at 12:21 PM, Myles Watson  wrote:
> On Thu, Dec 18, 2008 at 11:51 AM, Myles Watson  wrote:
> > On Thu, Dec 18, 2008 at 11:41 AM, Marc Jones  wrote:
> > > On Wed, Dec 17, 2008 at 1:42 PM, Myles Watson  wrote:
> > > > serengeti.diff and dtc-links.diff make serengeti find all of its
> > >
> > > devices
> > > > correctly.
> > > >
> > > > serengeti.diff
> > > > takes out the multiple links for the Opteron (they weren't working
> > > >
> > > > correctly)
> > > > adds the 8132 devices and makes them pci_a so they don't conflict
> > >
> > > with the
> > > > 8111
> > > > corrects the disabled flags (I think we should have a build error
> > > >
> > > > somewhere when there are incorrect flags)
> > > > moves the lpc device to the lpc bus
> > > > adds back in the second Opteron
> > > >
> > > > Signed-off-by: Myles Watson
> > >
> > > Acked-by: Marc Jones 
> > > But what about amd8151?

I was thinking the 8151 was on a different link of the same Opteron.  It's
not.  Since it's on a different Opteron, it's easy.  Sorry for the
confusion.

This patch adds the 8151 and makes it compile.  Tested on the free version,
so the 8151 just gets disabled.

Signed-off-by: Myles Watson

BTW: it still needs the dtc patch.  If you don't have it, this adds yet
another case where two links are created instead of one.

Thanks,
Myles

[Marc Jones]

On Thu, Dec 18, 2008 at 12:58 PM, Myles Watson  wrote:
> On Thu, Dec 18, 2008 at 12:21 PM, Myles Watson  wrote:
> > On Thu, Dec 18, 2008 at 11:51 AM, Myles Watson  wrote:
> > > On Thu, Dec 18, 2008 at 11:41 AM, Marc Jones  wrote:
> > > > On Wed, Dec 17, 2008 at 1:42 PM, Myles Watson  wrote:
> > > > > serengeti.diff and dtc-links.diff make serengeti find all of its
> > > > > devices
> > > > > correctly.
> > > > >
> > > > > serengeti.diff
> > > > > takes out the multiple links for the Opteron (they weren't working
> > > > >
> > > > > correctly)
> > > > > adds the 8132 devices and makes them pci_a so they don't conflict
> > > > >
> > > > > with the
> > > > > 8111
> > > > > corrects the disabled flags (I think we should have a build error
> > > > >
> > > > > somewhere when there are incorrect flags)
> > > > > moves the lpc device to the lpc bus
> > > > > adds back in the second Opteron
> > > > >
> > > > > Signed-off-by: Myles Watson
> > > >
> > > > Acked-by: Marc Jones 
> > > > But what about amd8151?
>
> I was thinking the 8151 was on a different link of the same Opteron.  It's
> not.  Since it's on a different Opteron, it's easy.  Sorry for the
> confusion.
>
> This patch adds the 8151 and makes it compile.  Tested on the free version,
> so the 8151 just gets disabled.
>
> Signed-off-by: Myles Watson
>
> BTW: it still needs the dtc patch.  If you don't have it, this adds yet
> another case where two links are created instead of one.

It would be unusual for two IO bridges  to be on the same CPU. I can't
think of every seeing that.
I have no way to test this now but it looks good so, let's get your
work in. This ack is for the dts and the dtc changes.

Acked-by: Marc Jones

[Myles Watson]

On Thu, Dec 18, 2008 at 3:21 PM, Marc Jones  wrote:
> On Thu, Dec 18, 2008 at 12:58 PM, Myles Watson  wrote:
> > On Thu, Dec 18, 2008 at 12:21 PM, Myles Watson
>
> wrote:
> > > On Thu, Dec 18, 2008 at 11:51 AM, Myles Watson
>
> wrote:
> > > > On Thu, Dec 18, 2008 at 11:41 AM, Marc Jones
>
> wrote:
> > > > > On Wed, Dec 17, 2008 at 1:42 PM, Myles Watson
>
> wrote:
> > > > > > serengeti.diff and dtc-links.diff make serengeti find all of its
> > > > > > devices
> > > > > > correctly.
> > > > > >
> > > > > > serengeti.diff
> > > > > > takes out the multiple links for the Opteron (they weren't working
> > > > > >
> > > > > > correctly)
> > > > > > adds the 8132 devices and makes them pci_a so they don't conflict
> > > > > >
> > > > > > with the
> > > > > > 8111
> > > > > > corrects the disabled flags (I think we should have a build error
> > > > > >
> > > > > > somewhere when there are incorrect flags)
> > > > > > moves the lpc device to the lpc bus
> > > > > > adds back in the second Opteron
> > > > > >
> > > > > > Signed-off-by: Myles Watson
> > > > >
> > > > > Acked-by: Marc Jones 
> > > > > But what about amd8151?
> >
> > I was thinking the 8151 was on a different link of the same Opteron.
>
> It's
> > not.  Since it's on a different Opteron, it's easy.  Sorry for the
> > confusion.
> >
> > This patch adds the 8151 and makes it compile.  Tested on the free
>
> version,
> > so the 8151 just gets disabled.
> >
> > Signed-off-by: Myles Watson
> >
> > BTW: it still needs the dtc patch.  If you don't have it, this adds yet
> > another case where two links are created instead of one.
>
> It would be unusual for two IO bridges  to be on the same CPU. I can't
> think of every seeing that.
> I have no way to test this now but it looks good so, let's get your
> work in. This ack is for the dts

Rev 1080.

and the dtc changes.
>

Rev 1079.
> Acked-by: Marc Jones

Thanks,
Myles