New ATI Framebuffer Xcode project to start from

[dong]

As we all known, native ATI drivers are named as ATY_Wormy, ATY_Caretta, etc. Following that, I construct a new Xcode project named ATY_HD, and put in it preliminary stuffs that I learn from dissembled native drivers.

 

ATY_HD now can compile and work, but only with the same function of IONDRVSupport. You may need modify "OSBundleRequired" part in plist file to fit for your system version.

 

One extra is that I combined the natit like thing into it, that means it can inject required entries into IORegistry by itself and then ATY_HD is attached to IONDRVDevice just like native driver do. However, this behavior is not so important at this time as ATY_HD is not a fully functional driver yet.

 

Another thing is that I defined a DriverGlobal structure in ATY_Struct.h, anyone interested in porting linux code can put their shared structures into it, thus provide a way to transport them between all functions.

 

Also I define some Debug Macros people may found useful in ATY_Struct.h. I'm now using it to check what cmd system sends to my driver.

 

In the plist file, "Personality Injection" represents the natit part, I put in a "IOPCIMatch" to limit the side-effect of the driver to other native drivers. So just change its value into your device id to start working with it.

ATY_HD.zip

[samm5506]

I'm sort of new to this stuff so sorry if this sounds dumb, but will this work for any ATI HD gpu's? Such as my ATI Radeon HD 2900 XT? (Which I can't seem to find any working drivers for atm) :/

[dong]

I'm sort of new to this stuff so sorry if this sounds dumb, but will this work for any ATI HD gpu's? Such as my ATI Radeon HD 2900 XT? (Which I can't seem to find any working drivers for atm) :/

 

The goal is to make ATY_HD work for Radeon HD, but not now. If IONDRVSupport(VESA) does not work for you, ATY_HD currently won't either.

Now it is only a start point with basic functions as IONDRVSupport. I need to figure more out from either Linux codes or dissembled native driver code to add to it.

At the mean time, I put the code here in case any other developers are interested in this.

[Slice]

Great dong!

Wish you be successful. If any question I am here.

My laptop with ATI is dead so I am out of the problem now.

[Boombeng]

Isn't it the project we, all radeon mobility HD users, were waiting for ?!

Thank you Dong! I'm sad I don't have knowledge to help...

[rschultz101]

great! for laptop ATI mobility !

 

would need source code for ATI .

would be great, to have native open osx drivers,

not to rely on osx compiled versions.

 

be even great to have it for nvidia .

 

but think, it's to big of a project,

and be to hard to get the sourcecode officially,...

 

at least a working wrapper be good

 

wish I could help, but I'am not a coder

[cybercow]

Hello to all brave devs here, i`m kinda new here, but i think i readed every line of infinite posts about ati cards especially on laptops with mobility series in which topic i`m interested though because i`m the owner of one. When i come here the first time i though things will be more easy and i don`t understood very well the main problematics why the hell my ati card not works, and was very magnetic to posts like "help me my card not works !!!" or "Finally the solution for card XXXX !!!". When i saw things will not go that way with forced applying this and that, jumping from one post to another, rebooting my laptop for the "last time" and praying god that finally boot in native resolution i decided to stop and think a litlle.

 

- i have the HP 6820s with mobility x1350, the "damned" one. Ok, i know is "old" i know nothing is supported bellow x1600, and i readed many times people encourage others to sell their uncompatible hardware as an easy way to get things up. Ok i can do that, i can even think my HP is a piece of obsoleted hardware, but ,,,,, i swear i see many people happily running their original G4`s at 700 Mhz, they know that this hw is "old", but still they re using it because it sodisfy their needs ... well same is here, i can sell my HP and get another one, and i`m sure many other users can do that, but they don`t do that - because they re happy with it, same as i`m happy with my HP. If i wanted to do advanced 3D stuff or gaming on a portable, i will choose another model in the first place either ....

 

- ok, so i`ll stop talking to not spam this post unnecessarly .... what i wanted to tell and ask .... i`m new to OSx and OSx developing in C/C++, but i have some development background in other languages/platforms enough to understand the logic of things.

 

- not related with this new stuff - i downloaded the decompiled version of Caretta someone posted, i opened the code, there is ~300K of code and few header files, someone remarked the known points in the code blocks. I would ask why using this with ATIInject was an dead end ? I understood the LVDS problematics, the EDID problematics though... does anyone try to compile this decompiled version and integrate it with new features added ? where it resides on the system ?

 

- I know i`m asking stupid questions, but i want to learn, and even help to do some nice stuff, i started to use Xcode and explore this all new world to me.... I will try to download and compile the ATY_HD here posted and hope to get some results that will help though. Can i try it on my mobility - or is only for HD series ?

 

Cheers to all and good luck

[dong]

The dissembled Carreta code is posted by me, so I can say something here.

 

When you trying to dissemble the binary code, the most difficult thing is figure out the data structure it has used because all the meaningful names are lost during compiling and it's hard to differentiate struct type data from array or many separate single ones. For a device driver, there are so many structs involved and you don't have a pre-knowledge what they might be since we are not from ATI. To have a rough idea how complex this could be, you can just take a look of the source code of the open linux radeonHD driver on http://www.x.org/wiki/radeonhd.

 

What I mean above is simply that the code I dissembled contains many mistakes in it. When you try to use it directly, you have to recheck with the binary code to make sure it does not interpret wrong, that's what I'm trying to do now.

Also, that code is not complete, lacking definitions of many structs or variables, and some low level hardware handling functions are not dissembled yet.

 

If you are brave enough to pick up all this, you have to learn some reverse engineering knowledge.

 

As for ATY_HD I posted here, like what I have said, it's only a start point without any extra function for any ATI cards including HD series. You can use it for your purpose for any cards if only you add in the required codes.

 

You may want to know, ATI mobility x1350 is actually one of the HD cards, thus you may borrow codes from the linux driver.

[lebidou]

Hello,

 

As a laptop user, I'm interested in an Open ATi framebuffer because atm I'm stuck with Tiger's ATINDRV.kext and I'm not sure it's going to work with possible future updates. I'm not a coder, but I have enough notions to decrypt a c or cpp file.

 

So, I was thinking about the LVDS and EDID thing. I looked in my EDID to see precisely what informations about my internal panel it contains. In fact, it doesn't contain any info about the different resolutions I have access to, exept the native one (1280x800).

 

So I downloaded RadeonDump and compiled AtomDis to look for informations about those modes in the atombios. There is informations about the panel (Horizontal/Vertical Active/Blankig Pixels,Pixel Clock…) but nothing about the other resolutions, I only found infos for the native one.

 

The only "useful" informations in the EDID are color infos, the vendor/product id and the actual names of the manufacturer and the product. So maybe getting the EDID for the internal panel isn't to important after all, A bios dump/parse would be enough to make a fake one.

 

But a question remains, where does my various resolutions and "DetailedTimings" in IOReg come from ? Are they automatically computed by the framebuffer ?

 

Maybe I missed somthing…

 

Here are the various resolutions I have :

 

Note : My ATI Mobility X1900 is a replacement card, when I had a nVidia 7600 Go I hadn't the 800x500 and the 1024x640 modes, they appeared with the ATi.

[dong]

sounds like automatically computed by the framebuffer. And you notice that all other resolutions are universal VESA ones.

[cybercow]

Hello dong, thank you much for your reply, i really appreciate your hard work on ATI field, i think i readed every post of yours in this field with great enthusiasm, u give nice ideas and hope to the rest of us with some kind of bastarduos ATI hardware. I have few more questions for you and your ATY_HD project.

 

1. It would be nice if u can write 1 page of simple dirrections, what is done, what is still to do and how to use the whole thing. I know that some "pro" will catch the whole thing in a minute, and that you already wrote here how to do it, but still having one simple *.txt description page would be nice to resume all of your results.

 

2. btw can this -> VideoHardwareInfo also attached be of some help ?

 

[edit -> i compiled ATY_HD today, and it worked for me (HP 6820s) -> i attached screen pic and syslog part if it helps dunno]

 

besides this i also attached the hp6820s edid dump if someone needs it [not related for now]

 

 

thank you in advance

VideoHardwareInfo.zip

hp6820s_display_edid_dump.txt

syslog_atyhd.txt

[0xdeadbeef]

Hi dong,

 

Very interested in the porting work you did to xf86-video-radeonhd. I have done a lot of work looking through Iago for my HD2400M and after tweaking your code for RadeonHD, I actually got my internal display to switch on and provide a garbled display. ( "first time ever" yay!) However using the accelerator failed on set_display_mode_and_vram() Which tells me something about what the accelerator is doing.

 

I have a feeling the reason the linux driver wasn't more successful on OSX was the lack of DDC communication - this is what breaks most of the apple drivers too. Many Laptops have no DDC on their internal display and even on desktop cards the DDC channels can be switched around compared to apples. If we get the linux AtomBIOS parser working, we can at least make an effort at constructing a viable display structure + connector info for displays that have no DDC. I have tried quite hard to get the xf86 AtomBIOS parser to compile, but there are very many problems yet.

 

OSX doesn't actually require a framebuffer to be very complex, if the correct display structures are returned and set in the corresponding doDriverIO() for a couple of functions, that is all the OS needs to know. However it is unlikely such a framebuffer would work in conjunction with the ATI accelerator - there are function codes in the ATI framebuffers for things like "DoATICommunication()" etc which might need to be implemented as well.

 

Are you going to be deriving this driver from xf86 too, or building it up manually?

[dong]

Frankly say, I kind of stuck into the dissembling of ATY_Carreta and ATY_Wormy when try to put the native initialize codes into ATY_HD and make it working. Maybe I should first look at DoATICommunication().

[0xdeadbeef]

Frankly say, I kind of stuck into the dissembling of ATY_Carreta and ATY_Wormy when try to put the native initialize codes into ATY_HD and make it working. Maybe I should first look at DoATICommunication().

 

I don't think so. A framebuffer should just provide information to the OS about the supported resolutions, and connected displays called through the subclass to IONDRV, it shouldn't need the accelerator to work to provide that information. What I mean is the interface to the native framebuffer only needs to implement the IONDrvFramebuffer calls to be useable by the OS for resolution change, gamma, detect display EDID, etc. the CoreImage and IOAccelerator stuff shouldn't be needed for that, unless I am very much mistaken. So there might be very much hardware/card specific code to get the card setup and started, but no hardware specific communications with the OS Graphics subsystem.

[dong]

Well, all stuffs in ATY_Carreta and ATY_Wormy are related to resolution change, gamma, detect display EDID, etc. The CoreImage and IOAccelerator stuffs are in other kexts. The problem here is how to get the hardware/card specific code to get the card setup and started.

[Slice]

However using the accelerator failed on set_display_mode_and_vram() Which tells me something about what the accelerator is doing.

The accelerator driver (ATIRadeonX1000.kext) use the method set_display_mode_and_vram() to initialize internal variables and GPU registers with information calculated from register 0x00f8 that filled by BIOS. Correction of the register is my key setting in ATILead.kext (previously ATISlice).

[lebidou]

I've done more searching on my side, and as I thought, building a fake EDID with AtomBios informations gives enough information to the "official" framebuffer.

I've modified ATILead to dump these infos (using AtomDis code), build an EDID string and inject it as "LVDS,EDID".

However as there is no color information it brokes ColorSync.

 

Can't be a workaround for non DDC internal panels ? I've seen something in the xf86 radeon hd driver about fake edid and the sLVDSInfos structure, so maybe it is what it uses to.

 

I'll make the code nicer and post it if you want to.

[0xdeadbeef]

Well, all stuffs in ATY_Carreta and ATY_Wormy are related to resolution change, gamma, detect display EDID, etc. The CoreImage and IOAccelerator stuffs are in other kexts. The problem here is how to get the hardware/card specific code to get the card setup and started.

 

 

Indeed, but I think the linux driver does all this quite well. For me at least, the biggest problem with the ATY_* framebuffers has been figuring out the mappings from display->transducer->CRTC->connector. The ATY_* framebuffers do most of this automatically, going through CheckLVDSConnection(), CheckDACConnection(), CheckTMDSConnection() etc etc on startup, but if the DDC line fails, it has problems and must rely totally on the _DriversConnectors[] tables - i don't think it parses AtomBIOS at all. However the latest ATY_* framebuffers (ATY_Quail and ATY_Motmot) do parse AtomBIOS on startup (I don't know yet if they read connector+detailed timing info from the bios).

 

The biggest change framebuffer-wise between the R500->R600 drivers appears to have been the 1-register shift for initializing the likes of TMDS2 (LVDS) I had thought of trying to patch all the register offsets in Wormy to try to get it to work with my R600, but I realised I would probably just end up making an Iago

 

Would it be very difficult, you think using the xf86 code to start the card and mapping the structures on-the-fly in DriverIO so that we can extract resolution, edid and gamma info from the running linux driver through the pScrn and returning it to IOGraphics? And the corresponding setResolution & gamma functions too?

[0xdeadbeef]

I've done more searching on my side, and as I thought, building a fake EDID with AtomBios informations gives enough information to the "official" framebuffer.

 

 

This doesn't usually work on the R600 framebuffers for laptop internal displays ( it is possible to inject your real display EDID with Natit ). the problem is that the R600 framebuffers like Iago will not bring up the LVDS connection even if you do that.

 

There are other settings you can use to inject EDID - ( "override-no-connect" is one of them) if you put your EDID in there, the framebuffer will read your display EDID and try to construct a display according to the type you specify in "connector-type". Unfortunately Iago will still refuse to switch LVDS on

(maybe because _HW_InitLVDS() looks like this )

 

void __cdecl HW_InitLVDS()
{
;
}

 

I have tried to patch in code from Triakis' HW_InitTDMS2() and set LVDS mode without any success so far.

 

 

the override-no-connect property is read like this;

 

signed int __cdecl DoDDCForceRead(int a1, int a2, char a3, unsigned __int8 a4, char a5, char a6)
{
 __int16 v7; // ax@6
 char v8; // [sp+3Ch] [bp-1Ch]@1
 unsigned __int8 v9; // [sp+38h] [bp-20h]@1
 char v10; // [sp+34h] [bp-24h]@1
 char v11; // [sp+30h] [bp-28h]@1
 signed int v12; // [sp+2Ch] [bp-2Ch]@2
 char *v13; // [sp+4Ch] [bp-Ch]@6
 __int16 v14; // [sp+4Ah] [bp-Eh]@6

 v8 = a3;
 v9 = a4;
 v10 = a5;
 v11 = a6;
 if ( *(_DWORD *)(a1 + 8) & 1 )
 {
if ( *(_DWORD *)(a1 + 4) )
{
  v12 = -50;
}
else
{
  if ( v10 & v9 )
  {
	v13 = "override-no-connect";
	v7 = ReadOverrideEDIDProperty(a2, v9, (unsigned __int8)*(_DWORD *)a1, "override-no-connect", a1 + 16);
	v14 = v7;
	if ( v7 )
	  return -19;
  }
  else
  {
	DDCInit(a2);
	if ( (unsigned __int8)DDCGetEDID(a2, (unsigned __int8)*(_DWORD *)a1, a1 + 16, v9) == 1 )
	{
	  if ( (unsigned __int8)EdidDigital(a2, a1 + 16) )
		v13 = "override-has-edid-digital";
	  else
		v13 = "override-has-edid";
	  ReadOverrideEDIDProperty(a2, v9, (unsigned __int8)*(_DWORD *)a1, v13, a1 + 16);
	}
	else
	{
	  v14 = -19;
	  if ( v11 & v9 )
	  {
		v13 = "override-no-edid";
		v14 = ReadOverrideEDIDProperty(a2, v9, (unsigned __int8)*(_DWORD *)a1, "override-no-edid", a1 + 16);
	  }
	  if ( v14 )
		return -19;
	}
  }
  if ( (unsigned __int8)DDCCheckEDID(a2, a1 + 16, (unsigned __int8)((unsigned __int8)*(_DWORD *)a1 - 1), 128) == 1 )
	v12 = 0;
  else
	v12 = -19;
}
 }
 else
 {
v12 = -50;
 }
 return v12;
}

[dong]

Would it be very difficult, you think using the xf86 code to start the card and mapping the structures on-the-fly in DriverIO so that we can extract resolution, edid and gamma info from the running linux driver through the pScrn and returning it to IOGraphics? And the corresponding setResolution & gamma functions too?

I think in my radeonHD project, I already managed to extract EDID and resolution (but not gamma) from the linux driver and provide the information to OS, which results the display of multiple available resolution choices in the Display Preference Panel. However, directly utilize setResolution function from linux driver failed for me. And I have no clue yet what could be the exact solution. That happened almost one year ago, and since that I'm digging and digging in the native drivers. Sadlly I'm almost lost in the thousands of lines of dissembled code, and do not have a clear mind yet to find out what I need from them.

But nice discussion with you guys, that will someday make it clear for me.

 

As for the DDC thing, here is my dissembled code for them that just was put in one file these days.

ATY_DDC.cpp.txt

[netkas]

Hey, can u make a tiny driver which will just initialize a card and quit, without any modes settings and etc, just like what bios/efi does (without settings display mode) ?

 

it will help us to have multi card ati setups with apple drivers

[dong]

Hi, netkas,

 

If what you mean is a natit like thing with only the function of injecting required entries (already present in plist file) into IORegistry, it should be easy that anyone can write one.

 

But it won't be so easy if these entries need to be determined by checking the card at boot time as I'm not very familiar with the meanings of those entries yet. If anyone has a clue of this, we may give it a try.

[netkas]

Dong, no, injecting works with natit just fine, vga bios does some initialization of card (using atombios prolly) , so the card can be used by osx drivers, if no initializayion was done - osx drivers will not detect any output on card. so, atm, we cant use dual card solutions using ati.

(dual card - for 4 monitors)

[Slice]

Yes, I know that some initialization by BIOS did occur. But how can we know what exactly? I know a set of registers "BIOS_..._scratch", "Video_Memory_...". What else?

[dong]

Dong, no, injecting works with natit just fine, vga bios does some initialization of card (using atombios prolly) , so the card can be used by osx drivers, if no initializayion was done - osx drivers will not detect any output on card. so, atm, we cant use dual card solutions using ati.

(dual card - for 4 monitors)

 

Ok, I now got what you mean.

 

After reading some posts on the web, I got the information like that there is no such initialization problem for dual NVidia cards, but ATI cards just won't POST if plugged in as secondary card.

 

It looks to me that the issue here is how to get the secondary ATI card POSTed. I rechecked codes in radeonHD linux driver, they are using int10 in a simulated real mode to boot the BIOS on the secondary card, which is not so easy to port to OSX in my opinion.

 

And I'm a little confused here about whether just running the card BIOS got it POSTed or you have to additionaly running ATOM init functions to get job done?

 

Without int10, there is also another problem of accessing the video BIOS as it's not located at 0xC0000 (already used by primary card) and you got no idea where it will be (maybe not in system memory at all). Without access to the BIOS image, there is no way to call ATOM init functions.

 

Edit: well, I checked ATY_Megalodon and found it did provide a way to read BIOS by interacting with IO registers of the card (0x1600 - 0x1620, 0x1798 - 0x17A0 for setting card to enter/exit BIOS reading state, and 0xA8, 0xAC for actually reading, 4 bytes each time). I will test it to see if this is general for any ATI cards.

 

Update: I tried these registers on my mobility x1400, but not working. But I saw similar code in radeonHD (RHDReadPCIBios), it does confirm that it depends on the type of card. Some time will be needed to test with it.

 

Please point out if I got something wrong.