[CDG5] ATI frozen mouse bug, cross-post from MacOS9Lives

Sun Oct 13 10:25:10 AWST 2019

I have just noticed that Mac OS 9 graphics acceleration and Mac OS X CoreImage support are claimed to be mutually exclusive. Is there any hard technical reason for this, or is it just driver support? Could the Radeon 9600 and other R350 cards be made to work using similar hacks to what we have tried?

> On 11 Oct 2019, at 6:28 pm, Elliot Nunn <elliotnunn at fastmail.com> wrote:
> 
> That could be very useful when we want to make novel hardware work with a disk-based driver.
> 
> I settled with bin-patching the individual driver because I knew that all the other ATI stuff got its information from Status calls to the driver. It also seemed more elegant to keep all the hacks in one place. And I was deterred by the complexity of applying a native patch very early in the boot process.
> 
> Care to try? Patching the ROM from Python is super fun.
> 
>> On 11 Oct 2019, at 3:31 am, Daniel B-J <danielbj314 at verizon.net> wrote:
>> 
>> It would be interesting to make a patch to the ExpMgrConfigReadXXXX routines, so that you can make any device ID appear to be any other device ID without patching individual drivers.
>> 
>>> On Oct 10, 2019, at 9:33 AM, Elliot Nunn <elliotnunn at fastmail.com> wrote:
>>> 
>>> Very likely! I have a few newly acquired iMac G4s, so I might even be able to do tests on the exact model. Link me to an EveryMac page?
>>> 
>>>> On 10 Oct 2019, at 6:29 pm, Jd Lyons <lyons_dj at yahoo.com> wrote:
>>>> 
>>>> On that note, do you think we could patch the Mac OS ROM to support the NV18 GPU this way?
>>>> 
>>>> I have a iMac G4 with the NV18, just a Geforce 4MX with 8x AGP support. It should work with OS 9’s nVidia drivers(NV17), but because the device ID is unknown to OS 9 it doesn’t.
>>>> 
>>>>> On Oct 10, 2019, at 6:22 AM, Jd Lyons <lyons_dj at yahoo.com> wrote:
>>>>> 
>>>>> It’s interesting you were able to patch the device ID, normally OS 9 style drivers are not fooled by Device ID patching, such as:
>>>>> 
>>>>> dev agp/@0
>>>>> 00005960 encode-int “ device-id” property
>>>>> 
>>>>> As most OS 9 style drivers don’t read the device ID from name space, they read it directly from the PCI Registers.
>>>>> 
>>>>> OS X, however, seems to read all PCI devices from name space, but not the CPU PVR.
>>>>> 
>>>>> nVidia FCode ROM's after the FX series can append the last half byte of the device ID in the NVStrap, but I’m not sure how OS 9 deals with that, as I’ve never been able to figure out how to hack nVidia driver support for OS 9.
>>>>> 
>>>>>> On Oct 9, 2019, at 11:44 PM, Elliot Nunn <elliotnunn at fastmail.com> wrote:
>>>>>> 
>>>>>> I have fixed the frozen mouse bug!
>>>>>> 
>>>>>> It was, as darthnVader hinted earlier, caused by the ATI extensions. So I took a
>>>>>> deep dive into the ATI driver architecture. This is the typical "life cycle" of
>>>>>> an ATI card in a NewWorld Mac:
>>>>>> 
>>>>>> 1.  If the card firmware lacks an ndrv (e.g. the Radeon 9200 in the mini), the
>>>>>> MacOS uses its generic "COFB" ndrv to get basic video.
>>>>>> 2.  Extensions are loaded from disk. The ATI ROM Xtender or iMic's hacked "ATI
>>>>>> Via Driver" extension can replace the previously loaded ndrv. No graphical
>>>>>> acceleration yet.
>>>>>> 3.  The complex array of ATI extensions steps in, asks the loaded ndrv about the
>>>>>> card using some private Status calls, and loads some acceleration code if the
>>>>>> card is recognised. Accelerated 2D requires "ATI Resource Manager" and "ATI
>>>>>> Graphics Accelerator" at minimum.
>>>>>> 
>>>>>> This was darthnVader's solution to get graphics acceleration working:
>>>>>> 
>>>>>> 1.  Insert a Mac OS X ndrv into iMic "ATI Via Driver" (better called "ATI
>>>>>> RockHopper2 Driver" after the change).
>>>>>> 2.  Replace all or most occurrences of the 5961 device ID in the other extensions
>>>>>> with 5962. Both IDs refer to the Radeon 9200, but the extensions only knew about
>>>>>> the 5961 ID. (Occurrences of 5960 were also replaced -- see below.)
>>>>>> 
>>>>>> DarthnVader effectively turned our minis into Mac OS 9 gaming rigs!
>>>>>> Unfortunately, one of the binary patches intermittently broke hardware cursor
>>>>>> support. It might be due to the way that the PowerPC compiler optimised
>>>>>> select/case blocks. Sometimes I also experienced a black screen when the ATI
>>>>>> extensions loaded, which I think was related.
>>>>>> 
>>>>>> This is my solution:
>>>>>> 
>>>>>> 1.  Insert the same Mac OS X ndrv (ATY,RockHopper2 v1.0.1f63, from Mac OS X
>>>>>> 10.3.6) into the Mac OS ROM as an ndrv parcel, to mimic a firmware driver.
>>>>>> 
>>>>>> 2.  Patch that ndrv to replace 5962 with 5961 whenever it reads from PCI
>>>>>> Configuration Space.
>>>>>> - To achieve this, I got sidetracked and wrote a single-file PowerPC assembler
>>>>>> script (ppcasm.py) and PEF dump/rebuild script (cfmtool.py):
>>>>>> https://github.com/elliotnunn/tbxi-patches/
>>>>>> 
>>>>>> 3.  Use the unmodified ATI extensions from January 2005.
>>>>>> - The extensions that we host in "9200os9.sit", supposedly from the rare July
>>>>>> 2005 release, do not work.
>>>>>> - I think that they were binary-patched to enable support for the DVI-I Radeon
>>>>>> 9200 card, ID 5960. Specifically, the accl 4 resource of the "ATI Graphics
>>>>>> Accelerator" is an identical build to January 2005, including the build
>>>>>> date, but has 2/2 occurrences of 5961 replaced with 5960. It would be good
>>>>>> to find the originals.
>>>>>> 
>>>>>> So here is my first ROM release of the year. It was derived from the
>>>>>> "OS9General" Mac OS ROM (v9.6.1) using my portable Mac mini patching system:
>>>>>> https://github.com/elliotnunn/tbxi-patches/blob/7a5d179/macmini.py
>>>>>> 
>>>>>> My ATALoad patch has also been applied, to allow Mac OS 9 to boot from volumes
>>>>>> that were formatted for Mac OS X, without first wiping the disk with "Mac OS 9
>>>>>> Drivers". I think that this is especially useful on machines that were formerly
>>>>>> Mac OS X-only.
>>>>>> 
>>>>>> I recommend that this ROM be included in a new release of the Mac mini ISO. It
>>>>>> should be combined with the unmodified ATI extensions from January 2005.
>>>>>> 
>>>>>> _______________________________________________
>>>>>> cdg5 mailing list
>>>>>> cdg5 at ucc.asn.au
>>>>>> https://lists.ucc.gu.uwa.edu.au/mailman/listinfo/cdg5
>>>>> 
>>>> 
>>>> _______________________________________________
>>>> cdg5 mailing list
>>>> cdg5 at ucc.asn.au
>>>> https://lists.ucc.gu.uwa.edu.au/mailman/listinfo/cdg5
>>> 
>>> _______________________________________________
>>> cdg5 mailing list
>>> cdg5 at ucc.asn.au
>>> https://lists.ucc.gu.uwa.edu.au/mailman/listinfo/cdg5
>> 
>> _______________________________________________
>> cdg5 mailing list
>> cdg5 at ucc.asn.au
>> https://lists.ucc.gu.uwa.edu.au/mailman/listinfo/cdg5
> 
> _______________________________________________
> cdg5 mailing list
> cdg5 at ucc.asn.au
> https://lists.ucc.gu.uwa.edu.au/mailman/listinfo/cdg5