[PATCH] mac: Don't set APPL, address properties on NewWorld

[Benjamin Herrenschmidt]

They are created by the MacOS ROM and in some cases MacOS X to contain
the *virtual* addresses of the BARs as mapped by the kernel.

Having these already there with the wrong values causes the hangs
with MacOS X 10.5 and the video driver.

The reason is that the IONDRVFramebuffer code in OS X 10.5 won't replace
those properties if they already exist, so the driver ends up accessing
the addresses put in there by OpenBIOS which aren't valid mapped virtual
addresses under OS X.

Signed-off-by: Benjamin Herrenschmidt 
---
 drivers/pci.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/drivers/pci.c b/drivers/pci.c
index 0cdf0bf..9813086 100644
--- a/drivers/pci.c
+++ b/drivers/pci.c
@@ -1029,10 +1029,9 @@ static void ob_pci_add_properties(phandle_t phandle,
    }

pci_set_assigned_addresses(phandle, config, num_bars);
-	
-	if (is_apple()) {
+
+	if (is_apple() && is_oldworld())
    	pci_set_AAPL_address(config);
-	}

PCI_DPRINTF("\n");
 }

[Howard Spoelstra]

Great, that fixes the "regression". One thing I noticed while booting with
only the Leopard iso is that Qemu/OpenBios make the Leopard system profiler
show two ATA devices, both containing a QemuDVD drive. Perhaps that is the
reason a hard disk is not recognised?

On Thu, Aug 4, 2016 at 3:16 AM, Benjamin Herrenschmidt <
[e-mail]> wrote:
> They are created by the MacOS ROM and in some cases MacOS X to contain
> the *virtual* addresses of the BARs as mapped by the kernel.
>
> Having these already there with the wrong values causes the hangs
> with MacOS X 10.5 and the video driver.
>
> The reason is that the IONDRVFramebuffer code in OS X 10.5 won't replace
> those properties if they already exist, so the driver ends up accessing
> the addresses put in there by OpenBIOS which aren't valid mapped virtual
> addresses under OS X.
> Signed-off-by: Benjamin Herrenschmidt
>
> drivers/pci.c | 5 ++---
>  1 file changed, 2 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/pci.c b/drivers/pci.c
> index 0cdf0bf..9813086 100644
> --- a/drivers/pci.c
> +++ b/drivers/pci.c
> @@ -1029,10 +1029,9 @@ static void ob_pci_add_properties(phandle_t phandle,
>         }
>     pci_set_assigned_addresses(phandle, config, num_bars);
>   if (is_apple()) {
>   if (is_apple() && is_oldworld())
>           pci_set_AAPL_address(config);
>   }
>
>   PCI_DPRINTF("\n");
>
> }

[Mark Cave-Ayland]

On 04/08/16 02:16, Benjamin Herrenschmidt wrote:
> They are created by the MacOS ROM and in some cases MacOS X to contain
> the *virtual* addresses of the BARs as mapped by the kernel.
>
> Having these already there with the wrong values causes the hangs
> with MacOS X 10.5 and the video driver.
>
> The reason is that the IONDRVFramebuffer code in OS X 10.5 won't replace
> those properties if they already exist, so the driver ends up accessing
> the addresses put in there by OpenBIOS which aren't valid mapped virtual
> addresses under OS X.
> Signed-off-by: Benjamin Herrenschmidt
>
> drivers/pci.c | 5 ++---
>  1 file changed, 2 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/pci.c b/drivers/pci.c
> index 0cdf0bf..9813086 100644
> --- a/drivers/pci.c
> +++ b/drivers/pci.c
> @@ -1029,10 +1029,9 @@ static void ob_pci_add_properties(phandle_t phandle,
>    }
>
> pci_set_assigned_addresses(phandle, config, num_bars);
> if (is_apple()) {
> if (is_apple() && is_oldworld())
> pci_set_AAPL_address(config);
>
> }
>
> PCI_DPRINTF("\n");
>
> }

Yeah this looks entirely sensible.

I've applied this to my next branch (with a typo fix-up in the subject)
- thanks a lot!

ATB,

Mark.