Editing custom personalities for ATI Radeon HD[45]xx

bcc9 · May 5, 2013

Interesting. Since your EDID is something that is checked at run-time, then you were looking at a shadow copy of your bios in ram, and you never had a dump of your actual bios. Glad you got the underlying problem figured out.

Slice · May 21, 2013

I did already post the source in this very thread, here: http://www.insanelym...ost__p__1843525

(Zero feedback from anyone on that BTW). The code builds fine under linux last I touched it.

No feedback because I have nothing to report. The idea to add something into bootloader remains idea only.

bcc9 · May 22, 2013

No feedback because I have nothing to report. The idea to add something into bootloader remains idea only.

For this kind of involved patching to OSX, I believe it's best to keep the smarts out of the bootloader. But I know some have the motivation to keep /S/L/E unmodified and wish the bootloader to do all the actual patching on the fly at every boot. For that scenario, perhaps we can standardize on an xml patch file syntax (that is not boot loader specific) that can provide this. Something general purpose that could be used for any hackintosh kext changes. Maybe bsdiff encapsulated in xml for binary patches, and a lighter weight match/replace for text diffs. I know clover has some of my patches built-in but the syntax is clover-specific. If there was a standard general purpose patch syntax we could agree upon then scripts like the one here could just generate the patch config for the bootloader.

Slice · May 28, 2013

I think this is very common syntax for kext patching

 <key>KextsToPatch</key>
 <array>
  <dict>
   <key>Name</key>
   <string>VoodooHDA</string>
   <key>Find</key>
   <data>SGVhZHBob25lcwA=</data>
   <key>Replace</key>
   <data>VGVsZXBob25lcwA=</data>
  </dict>
  <dict>
   <key>Name</key>
   <string>AppleHDAController</string>
   <key>Comment</key>
   <string>Patch_to_not_load_this_driver</string>
   <key>InfoPlistPatch</key>
   <true/>
   <key>Find</key>
   <string>0x04020000</string>
   <key>Replace</key>
   <string>0x44220000</string>
  </dict>
  <dict>
   <key>Name</key>
   <string>IOBlockStorage</string>
   <key>Find</key>
   <data>TA==</data>
   <key>Replace</key>
   <data>TA==</data>
  </dict>
  <dict>
   <key>Name</key>
   <string>AppleAHCIPort</string>
   <key>Find</key>
   <data>RXh0ZXJuYWw=</data>
   <key>Replace</key>
   <data>SW50ZXJuYWw=</data>
  </dict>
 </array>

Do you want more keys?

bcc9 · June 5, 2013

I think this is very common syntax for kext patching

I don't think this clover syntax is sufficient to be a general purpose solution for supporting add/change/delete operations on the system kexts.

For adding to kexts, such as adding a plugin, you'd likely want the kext name to include the path to the kext. Textual additions to kexts (such as Info.plist-only kext additions) would be a lot more legible if the addition wasn't encoded in binary.
Some examples: an Info.plist that contains most of one's codec specific AppleHDA plist changes, support for one of the ethernet drivers that isn't included in the base OS
For changing kexts, such as your patch example, you'd want to reference the script or thread that can be used to maintain the patch across releases. Also a range of OSX release numbers and or kext version numbers for which the patch is known to apply would help. The patch syntax should probably also provide for multiple different versions of the patch depending upon the versions. (At least with enough version information to handle the version of OSX currently running and the version of OSX that the system will boot into next).
Some examples: My AppleHDA patch that varies depending upon OSX version but where the perl script works across a wide range of releases.
Likewise for my AppleACPIPlatform patch for the nvidia mcp79 chipset.
The CMOS reset patch could be handled this way as well (instead of having information about the support thread and versioning details hardcoded in the clover source).
For deleting, perhaps these can all be handled by modifying the kext match clauses, but it would be more failsafe to just effectively delete the kext (then you don't have to worry about changes in the kext's personality lists from release to release).

Slice · June 10, 2013

Bootloaders can't modify kexts because HFS+ drivers are read-only. It is possible to create RW driver but this is not still happen.

A possible way is to write an startup script.... No! Kexts already loaded.

If you want to do patch for a kext then you can do it once or at every boot on the fly.

Other question. Will your method work with UEFI compatible VBIOS? My attempt with 7850 failed.

bcc9 · June 26, 2013

Bootloaders can't modify kexts because HFS+ drivers are read-only.

Bootloaders effectively can; clover does exactly this in kext_patcher.c, which is the model I thought we were discussing as a starting point.

It is possible to create RW driver but this is not still happen.

HFS+ is read/write in linux open source, changes could be ported into UEFI code if one desired. Even hfs+ journaling has opensource support at this point (complete enough for an efi solution at least).

Other question. Will your method work with UEFI compatible VBIOS? My attempt with 7850 failed.

What problem are you having, decoding the connector info from the bios or patching the osx kext? I would think the later would work independent of whether the card was started up in uefi mode.

Personally I haven't used my radeon hd in about 2 years (gave it away actually). I could check with the person who has it now, as it's in a system with UEFI system bios at this point. But I'm not sure if that's what you mean.

MJKhaani · July 20, 2013

Sir, Many thanks for your guide. I'm trying to patch my personality to get LDVS working, Unsuccessful until now . I have some questions, if you answer them I'll be glad.

I use "Pithecia" frame buffer and it work with VGA and with 2 port matched (ATY_ActiveFlags = 10), Port 0 and 1. VGA works perfect and LDVS shows black screen with brightness control working. When I switch from VGA to LDVS, Screen become distorted, But cursor works fine and shown normally.

1. Are the port number of personalities In IOReg ( like Port@0 ), shows index of personalities (from top to down)? If it's true ConnectorType of Port 0 & 1 are not indicate to VGA!? So how VGA works?

Personality: Pithecia

0000000 00 04 00 00 04 03 00 00 00 01 00 00 21 03 02 04 >Port@0

0000010 04 00 00 00 14 02 00 00 00 01 00 00 00 00 04 03 >Port@1

0000020 10 00 00 00 10 00 00 00 00 01 00 00 00 00 00 01 >Port@2

2.When I change one of personalities it cause to load default frame buffer (AMD,FrameBuffer ...) instead of "Pithecia". Why?

Asus K54HR SEYMOUR M2 XTX 
DDR3                                             
Subsystem Vendor ID: 1043       Subsystem ID: 2002
Object Header Structure Size: 199
Connector Object Table Offset: 34
Router Object Table Offset: 0
Encoder Object Table Offset: 99
Display Path Table Offset: 12

Connector Object Id [14] which is [LVDS] 
        encoder obj id [0x1e] which is [INTERNAL_UNIPHY (osx txmit 0x10 [duallink 0x0] enc 0x0)] linkb: false
Connector Object Id [12] which is [HDMI_TYPE_A]
        encoder obj id [0x20] which is [INTERNAL_UNIPHY1 (osx txmit 0x21 [duallink 0x1] enc 0x3)] linkb: true
Connector Object Id [5] which is [VGA]
        encoder obj id [0x15] which is [INTERNAL_KLDSCP_DAC1 (osx txmit 0x00 enc 0x10?)] linkb: false

Sorry for my English.

bcc9 · July 26, 2013

Again, I don't use ATI mobility hardware, so I have no first hand experience getting LVDS ports working with this driver. There is a separate thread for ATI mobility support.

The order that ports show up in the ioregistry doesn't necessarily match the order of ConnectorInfo found in the personality.

MJKhaani · July 28, 2013

Again, I don't use ATI mobility hardware, so I have no first hand experience getting LVDS ports working with this driver. There is a separate thread for ATI mobility support.

The order that ports show up in the ioregistry doesn't necessarily match the order of ConnectorInfo found in the personality.

Thank you very much :thumbsup_anim:

smx · September 1, 2013

Hi bcc9. I'm very happy with you tool, and i use it all the time to patch the personalities after a system upgrade (to remove the ghost display of the stock Kext). The problem is that, after the last upgrade to osx 10.8.4, your tool just says ConnectorInfo count is 0 for all the personalities and the offsets seems wrong (i don't get the usual connector structures in hexdump).
I'm using ati-personality version 0.10 from first post
I attach the normal output, the verbose one and a copy of the AMD Controller (my GPU is a MSI R5770)

ati-personality-logsbin.zip

k3nny · September 1, 2013

@smx: Look here for the fixed version.

bcc9 · September 2, 2013

@smx: Look here for the fixed version.

Thanks, this is the first time I've been made aware that users were bumping into the otool compatibility issue with this script. Xcode as of 4.6 broke my scripts by changing the (default) output style from hex to decimal in otool. I did notice this before with my ahci patch script but I forgot about this one. I'll update.

bcc9 · September 3, 2013

I've updated post #1 with version 0.11. otool is simply invoked as otool -Q to get its former (more useful IMO) behavior. The script also warns the user when otool is missing instead of simply failing to run correctly.

smx · September 3, 2013

It worked like a charm. Thanks kenny, bcc9 and slice

delifruit · October 9, 2013

Hi..

This script doesn't work mavericks' AMD kext.

I ran it at mavericks GM and 10.8.5.

Can someone help me please?

$ perl ati-personality.pl /Volumes/mavericks/System/Library/Extensions/AMD5000Controller.kext

Kext /Volumes/mavericks/System/Library/Extensions/AMD5000Controller.kext/Contents/MacOS/*Controller

Personality: Douc