[OpenBIOS] [PATCHv2 2/6] ppc: fix ESCC reg properties

[Mark Cave-Ayland]

On 23/02/16 03:58, Programmingkid wrote:

> On Feb 20, 2016, at 12:18 PM, Mark Cave-Ayland wrote:
> 
>> I/O offsets were wrong in legacy mode and some were missing in non-legacy mode.
>> This fixes serial port detection in NetBSD, which was ignoring it.
>> This also partly fixes MacOS 9.2, which relies on them to initialize OpenTransport.
>>
>> Based upon an original patch from Hervé Poussineau <hpoussin at reactos.org>.
>>
>> Signed-off-by: Mark Cave-Ayland <mark.cave-ayland at ilande.co.uk>
>> ---
>> openbios-devel/drivers/escc.c |   43 ++++++++++++++++++++++++++++-------------
>> 1 file changed, 30 insertions(+), 13 deletions(-)
>>
>> diff --git a/openbios-devel/drivers/escc.c b/openbios-devel/drivers/escc.c
>> index 9b87083..44b73f0 100644
>> --- a/openbios-devel/drivers/escc.c
>> +++ b/openbios-devel/drivers/escc.c
>> @@ -388,7 +388,20 @@ escc_add_channel(const char *path, const char *node, phys_addr_t addr,
>>     cell props[10];
>>     int index;
>>     int legacy;
>> -
>> +    int reg_offsets[2][2][3] = {
>> +        {
>> +            /* ch-b */
>> +            { 0x00, 0x10, 0x40 },
>> +            /* ch-a */
>> +            { 0x20, 0x30, 0x50 }
>> +        },{
>> +            /* legacy ch-b */
>> +            { 0x0, 0x2, 0x8 },
>> +            /* legacy ch-a */
>> +            { 0x4, 0x6, 0xa }
>> +        }
>> +    };
>> +    
>>     switch (esnum) {
>>         case 2: index = 1; legacy = 0; break;
>>         case 3: index = 0; legacy = 0; break;
>> @@ -425,17 +438,21 @@ escc_add_channel(const char *path, const char *node, phys_addr_t addr,
>>     set_property(dnode, "compatible", buf, 9);
>>
>>     if (legacy) {
>> -        props[0] = IO_ESCC_LEGACY_OFFSET + index * 0x4;
>> -        props[1] = 0x00000001;
>> -        props[2] = IO_ESCC_LEGACY_OFFSET + index * 0x4 + 2;
>> -        props[3] = 0x00000001;
>> -        props[4] = IO_ESCC_LEGACY_OFFSET + index * 0x4 + 6;
>> -        props[5] = 0x00000001;
>> +        props[0] = IO_ESCC_LEGACY_OFFSET + reg_offsets[legacy][index][0];
>> +        props[1] = 0x1;
>> +        props[2] = IO_ESCC_LEGACY_OFFSET + reg_offsets[legacy][index][1];
>> +        props[3] = 0x1;
>> +        props[4] = IO_ESCC_LEGACY_OFFSET + reg_offsets[legacy][index][2];
>> +        props[5] = 0x1;
>>         set_property(dnode, "reg", (char *)&props, 6 * sizeof(cell));
>>     } else {
>> -        props[0] = IO_ESCC_OFFSET + index * 0x20;
>> -        props[1] = 0x00000020;
>> -        set_property(dnode, "reg", (char *)&props, 2 * sizeof(cell));
>> +        props[0] = IO_ESCC_OFFSET + reg_offsets[legacy][index][0];
>> +        props[1] = 0x1;
>> +        props[2] = IO_ESCC_OFFSET + reg_offsets[legacy][index][1];
>> +        props[3] = 0x1;
>> +        props[4] = IO_ESCC_OFFSET + reg_offsets[legacy][index][2];
>> +        props[5] = 0x1;
>> +        set_property(dnode, "reg", (char *)&props, 6 * sizeof(cell));
>>     }
>>
>>     if (legacy) {
>> @@ -446,13 +463,13 @@ escc_add_channel(const char *path, const char *node, phys_addr_t addr,
>>     OLDWORLD(set_property(dnode, "AAPL,address",
>>             (char *)&props, 1 * sizeof(cell)));
>>
>> -    props[0] = 0x00000010 - index;
>> +    props[0] = 0x10 - index;
>>     OLDWORLD(set_property(dnode, "AAPL,interrupts",
>>             (char *)&props, 1 * sizeof(cell)));
>>
>>     props[0] = (0x24) + index;
>> -    props[1] = 0;
>> -    props[2] = 0;
>> +    props[1] = 0x0;
>> +    props[2] = 0x0;
>>     NEWWORLD(set_property(dnode, "interrupts",
>>              (char *)&props, 3 * sizeof(cell)));
>>
>> -- 
>> 1.7.10.4
> 
> 
> I had a lot of problems applying this patch. Eventually I just applied it by hand. Testing your v2 patch set with Mac OS 9.2 ends with an Illegal Instruction error for me. When I tested Mac OS 9.0, the OS just stops booting and does nothing. It doesn't crash. I just sits there doing nothing. The mouse cursor still can move. I tried this OS again and it crashed on the second time. The third time it just sat there doing nothing like the first time. 

Just this one patch, or the entire patchset? Pretty much everything just
works with "git am" these days?

I think there are two issues here: firstly something does cause memory
corruption which can cause random crashes under OS 9, and secondly the
changes to the serial ports. We know that OT probe causes OS 9 to
crash/hang when booting from a vanilla image, but the probe doesn't take
place on pre-installed images. In these cases the hang is almost better
than the crash in that at least the OS gets to the point where it can
probe the serial ports, it is just that the QEMU emulation is lacking.

If the change in behaviour occurs when booting from a fresh ISO then I'm
not too worried because everyone testing at the moment will have removed
these extensions from their images. However if existing installed images
start to fail then perhaps the patch needs more work.

Howard, is this something that you can test? I believe this is a good
candidate to be applied to trunk this week in time for QEMU freeze since
it definitely fixes the properties in terms of the reference device tree
and indeed it works to the point where NetBSD/Linux can now see the
serial ports and use them.


ATB,

Mark.