[80% Solved] Iris Xe iGPU on Tiger Lake successfully loaded ICLLP Frambuffer and VRAM also recognizes 1536MB! + However, some issues.

[benmacfreak]

7 minutes ago, Mastachief said:

Use GPU-Z on windows to figure out what your GPU id and model, and edit your post and put it there instead of creating a new post.

im doing school atm ill do that on my end soon

[luke192]

2 hours ago, Mastachief said:

Use GPU-Z on windows to figure out whats your Intel GPU id and model, and edit your post and put it there instead of creating a new post.

 

Verify you are using "XE" Architecture based on the following list. --> https://dgpu-docs.intel.com/devices/hardware-table.html

 

XE-HPG, Xe-HPC & Xe-LPG and others will definitely never have a chance as those are Intel ARC etc

Would this be compatible? I still haven't got macOS to boot BTW

[Mastachief]

We have no way to advise of what will work or won't right now as the driver is in development, if you don't have 11th gen, it's best to test and let us know what the result on your computer is, but you won't have acceleration. No ETA, best to test using ventura.

[ArtikDiamond]

Hey!

Today, I decided to try NootedBlue with a clean install of Ventura. Is it normal to be stuck on "IOKit Daemon," as shown in the picture? It’s been like this for a while (just booting the installer) and it just went from 2 lines of "IOKit Daemon" to 4 while I was writing this.

Also, Mastachief, would you like me to share my config.plist?

Edit - new photo

 

Now it's moving, but I think it's in some kind of error loop:

Edited September 7 by ArtikDiamond

[Visual Ehrmanntraut]

On 9/6/2024 at 12:37 AM, jalavoui said:

their usage in code is strange. see this code (so i opt for use ml_io_read_ptr() in hookcase)

 

[Visual Ehrmanntraut]

[ArtikDiamond]

8 hours ago, ArtikDiamond said:

Now it's moving, but I think it's in some kind of error loop:

Do you know what could it be? Do you need my plist or EFI?

[Visual Ehrmanntraut]

Y'all, that HookCase code Jalavoui sent is a catastrophic fail. Try this instead.

HookCase.kext.zip

[jalavoui]

Visual check BNinja definitions for the functions

 

 

wtf is this ?

c++filt friendly?

 

anyway can you build your kext for AppleIntelBaseController also ?

 

i bet hwSetupDSBMemory() is a mem allocation but with which parameters and offsets?

 

you sure u wanna make hwSetupDSBMemory() return 1? it will skip this code part

 

btw can you also add

extern "C" void gpu_accumulate_time();

extern "C" void gpu_describe();

for AppleIntelTGLGraphics.kext missing symbols?

 

i got a link error so i added this

extern "C" char *__null_terminated strnstr(const char *__null_terminated s, const char *__null_terminated find, size_t slen) __stateful_pure;

gonna recheck if still need

 

anyway tks for writting a few lines of code

i do hope others follow

 

Visual you forgot to fix symbols in your kext. it won't load cause of

 

don't be lazy and create a class !

 

this is the theory i was looking for related to offsets

in this code the offset is 0x9b0 - it is a similiar readregister function of same class we are interested

so for hookcase we need to use same offset for it to work

fine for some functions but hard to find for others

 

but researching older code (AppleIntelCFLGraphicsFramebuffer) what we need might be close to this (my theory)

Edited September 7 by jalavoui

[Visual Ehrmanntraut]

3 minutes ago, jalavoui said:

Visual check BNinja definitions for the functions

 

did. that kext is my attempt to fix your code. don't know what the DSB thing is supposed to do. found an implementation of probePortMode but it's probably too complicated to implement in this environment, have fun with that.

1 hour ago, jalavoui said:

you sure u wanna make hwSetupDSBMemory() return 1? it will skip this code part

Whoops, misread != for ==.

1 hour ago, jalavoui said:

Visual you forgot to fix symbols in your kext. it won't load cause of

 

don't be lazy and create a class !

What do you mean? Oh....

[Visual Ehrmanntraut]

1 hour ago, jalavoui said:

btw can you also add

extern "C" void gpu_accumulate_time();

extern "C" void gpu_describe();

for AppleIntelTGLGraphics.kext missing symbols?

I see this symbol referenced in other kexts as well as standard macOS builds, so it's not actually missing.

1 hour ago, jalavoui said:

anyway tks for writting a few lines of code

Not just that. I was reverse engineering for a few hours to come up with that.

[jalavoui]

i declared this to get the correct export symbol name

extern "C"  unsigned int AppleIntelBaseController::ReadRegister32(unsigned long a)

and ofc this is not what is need. i need the extra void* parameter

can you share your header declaration. i dont get this...

 

anyway your kext won't load cause of _start symbol.

just grab them all to /L/E and fix permissions

when you load it you will see the error

 

oh and the write function must not return a value. check your code

 

i'm trying to inplement the validation seen in icl code so

if ((*(byte *)(*(long *)&(this->m_AppleIntelFramebufferController).field_0xcf0 + 0xb2) & 1) != 0) icl code translates to

 

    if ((*(*(arg1 + 0xd40) + 0xfa) & 1) != 0) tgl offsets

 

then there are some range check that maybe easier to ignore and just return zero (nha those validations for range might help avoid crashes)

 

and finally there's a size check that is made against an offset 

if (param_1 < *(int *)&(this->m_AppleIntelFramebufferController).field_0xc30 - 4) { (this is icl code)

so the tgl offset will be...

if (zx.q(*(arg1 + 0x60) - 4) u<= arg3) maybe here for tgl

 

btw apple code as a bug in here (i think the check must be 0x60)-8 as its a 64bits reading acording to old icl code

 

Edited September 7 by jalavoui

[Visual Ehrmanntraut]

AppleIntelICLLPGraphicsFramebuffer.__ZN17AppleIntelPortHAL13probePortModeEv.txtHookCase-master.zipHookCase.zip

[Visual Ehrmanntraut]

39 minutes ago, jalavoui said:

oh and the write function must not return a value. check your code

just misdecompilation

[jalavoui]

gonna check your code meanwhile i wrote this for new tgl

extern "C"  unsigned int AppleIntelBaseController::ReadRegister32(void volatile* a, unsigned long b)
{
    long *plVar1;
    uint32_t uVar2;
    
    if ((*(int *)(*(long *)((long *)a + 0xd40) + 0xfa) & 1) != 0)
    {
      if (0x1fff < b) {
        if (b < 0x40000) {
        }
        else {
          if (b < 0x1c0000 || 0x23ffff < b)
          goto LAB_00047bfb;
        }
        return 0;
      }
    }
  LAB_00047bfb:
    if (b < *(int *)((long *)a + 0x60) - 4)
      uVar2 = *(uint32_t *)((long)a + b);
    else {
      uVar2 = 0;

return uVar2;

 

that was smart you got the rmio address

uint64_t rmmio = getMember<uint64_t>(this, 0x9B0)

 

very nice !

 

ok test results

 

 

gonna try on real hardware now

Edited September 7 by jalavoui

[Visual Ehrmanntraut]

11 minutes ago, jalavoui said:

that was smart you got the rmio address

uint64_t rmmio = getMember<uint64_t>(this, 0x9B0)

 

very nice !

I'm

(lol, hope you get the reference)

Edited September 7 by Visual Ehrmanntraut

[Visual Ehrmanntraut]

Oh yeah. I think I forgot to say that they probably need to go to /S/L/E and not /L/E for proper operation

[jalavoui]

with no patches i get this

 

 

anyway it's a good panic

time to test some patches

 

testing on L/E atm 

 

Edited September 7 by jalavoui

[Mastachief]

So which hookcase should we use?

For S/L/E should I use the in internal tglgraphic kexts? I can always test with the seal broken on Sonoma.

I still don't know where iglogs go when we have igloglevel=8

[Visual Ehrmanntraut]

@jalavoui Are you using the diags kext? That's the only one where the offset in the panic makes any sense.

but probePortMode here should be populated in both HAL and HALDiags, according to the decompilation

[jalavoui]

yes the diags help a lot

 

this bug again with you hook - grrrrr

 

 

todo:

test both internal and production tgl versions

Edited September 8 by jalavoui

[Visual Ehrmanntraut]

Honestly this specifically is likely because you're not putting them in /S/L/E

[jalavoui]

maybe. can you believe i have to rewrite hookcase because i was geting symbol conflicts with icl framebufer kext. well nvm here's a panic from production version

 

the current patches i have are all for internal version atm

 

here's the "compatible" hookcase

 

 

 

but yeah S/L/E will be needed i'm thiking on bundles loading, etc

wonder if other users are testing this. maybe alder lake wih device and ig-platform injection?

this are nblue defaults that match existing tgl osinfo structure

static uint8_t "AAPL,ig-platform-id[] = {0x00, 0x00, 0x49, 0x9A};

static uint8_t device-id[] = {0x49, 0x9A,0x00,0x00};

Edited September 9 by jalavoui

[benmacfreak]

29 minutes ago, jalavoui said:

maybe. can you believe i have to rewrite hookcase because i was geting symbol conflicts with icl framebufer kext. well nvm here's a panic from production version

 

the current patches i have are all for internal version atm

 

here's the "compatible" hookcase

 

HookCase.kext.zip 69.67 kB · 0 downloads HookCase-master.zip 185.48 kB · 0 downloads

 

but yeah S/L/E will be needed i'm thiking on bundles loading, etc

wonder if other users are testing this. maybe alder lake wih device and ig-platform injection?

this are nblue defaults that match existing tgl osinfo structure

static uint8_t "AAPL,ig-platform-id[] = {0x00, 0x00, 0x49, 0x9A};

static uint8_t device-id[] = {0x49, 0x9A,0x00,0x00};

i have a 9A68 device ID jala, just downloaded the HookCase posted 8 mins ago. Will let you know how it goes.

[Visual Ehrmanntraut]

35 minutes ago, jalavoui said:

but yeah S/L/E will be needed i'm thiking on bundles loading, etc

 

I think the bundles can go in /Library/GPUBundles.