Jump to content

Mavericks kernel testing on AMD (formerly Mountain Lion kernel testing on AMD)

theconnactic

By theconnactic,
in OS X 10.9 (Mavericks)

 Share
6,414 posts in this topic

Recommended Posts

spakk
spakk
Posted
Posted

I'm still in the installation of Snow Leopard. the installation proceeds very annoying, had many setbacks, I'm almost done and still needs to update 10.6.8. Then I can start the tests. As soon as I have a results I will inform you .

Link to comment
Share on other sites
theconnactic
theconnactic
Posted
  • Author
Posted

Hello, folks!

 

I'm glad to present the new beta revision of my 12.0 kernel for AMD machines. It owes much to the hard work of sinetek a.k.a. slackware: i took his ssse3 emulator from his Snow Leopard kernel and add it to the commpage of my own. I can't stop stressing how important was his achievement, and he deserves all credit for what he's done.

 

That's my first try. Besides adding the emulator's header and C source, i altered the commpage.c file to actually enable the emulator. What i did is quite simple: i just replaced the sse3emu calls in the commpage.c for ssse3emu calls. So, if your CPU is old enough to not to have sse3 support, it won't even begin to work. Sorry for it, but remember - it's just another beta and it's not meant to work at all, but to give us answers, so hopefully we get ways to better patches and proper emulator implementation.

 

Add the kernel to the root and boot with the flag sinactic. Report your results, with pics and vids if possible. Later i'll add the diffs for this new beta, so you guys can make experiments f your own. Good luck!

 

sinactic.zip

 

Here's the diff:

 

 

sinactic.diff.zip

  • Like 2
Link to comment
Share on other sites
mac_carol
mac_carol
Posted
Posted

Hello, folks!

 

I'm glad to present the new beta revision of my 12.0 kernel for AMD machines. It owes much to the hard work of sinetek a.k.a. slackware: i took his ssse3 emulator from his Snow Leopard kernel and add it to the commpage of my own. I can't stop stressing how important was his achievement, and he deserves all credit for what he's done.

 

That's my first try. Besides adding the emulator's header and C source, i altered the commpage.c file to actually enable the emulator. What i did is quite simple: i just replaced the sse3emu calls in the commpage.c for ssse3emu calls. So, if your CPU is old enough to not to have sse3 support, it won't even begin to work. Sorry for it, but remember - it's just another beta and it's not meant to work at all, but to give us answers, so hopefully we get ways to better patches and proper emulator implementation.

 

Add the kernel to the root and boot with the flag sinactic. Report your results, with pics and vids if possible. Later i'll add the diffs for this new beta, so you guys can make experiments f your own. Good luck!

 

sinactic.zip

 

Here's the diff:

 

 

sinactic.diff.zip

 

 

 

Tested on with ML 10.8.0 on AMD P-II x4 965

 

 

 

1. flags: sinactic arch=i386 busratio=17 -f -v

 

result: instant reboot

 

 

 

2. flags: sinactic busratio=17 -f -v

 

result: KP

 

ml_01.gif

Link to comment
Share on other sites
zchef2k
zchef2k
Posted
Posted

well i had a backup of my MBP so i gave it a shot by swapping Kernels here the result

 

tried a couple of flags but no sucess .... as asumed

 

 

 

 

 

test.jpg

Tested on with ML 10.8.0 on AMD P-II x4 965

 

 

 

1. flags: sinactic arch=i386 busratio=17 -f -v

 

result: instant reboot

 

 

 

2. flags: sinactic busratio=17 -f -v

 

result: KP

 

ml_01.gif

 

Does anyone actually read the forum?

 

fips_mode=0

 

AMD's SSE3 interferes with the checking and loading of corecrypto. I know this because I have masked the instruction in ESXi and it loads without error.

 

-Z

Link to comment
Share on other sites
mac_carol
mac_carol
Posted
Posted

Does anyone actually read the forum?

 

fips_mode=0

 

AMD's SSE3 interferes with the checking and loading of corecrypto. I know this because I have masked the instruction in ESXi and it loads without error.

 

-Z

 

 

flags: sinactic fips_mode=0 busratio=17 -f -v

 

result: freeze

 

 

ml_02.gif

Link to comment
Share on other sites
spakk
spakk
Posted
Posted

Tested on with ML 10.8.0 on AMD P-II x4 965

 

 

 

1. flags: sinactic arch=i386 busratio=17 -f -v

 

result: instant reboot

 

2. flags: sinactic busratio=17 -f -v

 

result: KP

 

 

Please test this boot flag:

sinactic arch=x86 usekernelcache=no maxmem=4096 busratio=17 -v -f

Link to comment
Share on other sites
mac_carol
mac_carol
Posted
Posted

Please test this boot flag:

sinactic arch=x86 usekernelcache=no maxmem=4096 busratio=17 -v -f

 

arch=x86 ???

 

if what you mean was arch=i386

 

the kernel won't run in 32-bits mode

it'll give me instant reboot

 

 

but with flags: sinactic arch=x86_64 usekernelcache=no maxmem=4096 busratio=17 -v -f

 

it gave me the same KP

 

ml_03.gif

Link to comment
Share on other sites
spakk
spakk
Posted
Posted

Sorry but I can not see any changes from the first and the second conatic-kernel with the current modified conatic / sinetek kernel. It is the same panic message.

 

 

Edit 1:

akimoa what do you think?

 

Edit 2:

What about boot flag: -force64 and -legacy? have you tested it?

Link to comment
Share on other sites
wastez
wastez
Posted
Posted

Sorry but I can not see any changes from the first and the second conatic-kernel with the current modified conatic / sinetek kernel. It is the same panic message.

 

 

Edit 1:

akimoa what do you think?

 

Edit 2:

What about boot flag: -force64 and -legacy? have you tested it?

 

-legacy will not work because it is just a 64 bit kernel.

-force64 makes also no sense, i think i don´t have to tell you why ;)

Link to comment
Share on other sites
spakk
spakk
Posted
Posted

as I told you before,

the kernel won't let me running the osx in 32-bits (arch=i386 -legacy / arch=i386 -force64)

It'll give me instant reboot

-legacy will not work because it is just a 64 bit kernel.

-force64 makes also no sense, i think i don´t have to tell you why ;)

 

how was that? with can't see the wood for the trees.... :whistle: I apologize and thanks for the information

:compress:

Link to comment
Share on other sites
Sinetek
Sinetek
Posted
Posted

howdy,

 

thanks connactic for doing this. huge timesaver. i've begun dorking around with a 10.8 kernel actually, so an x64 SSSE3 emu is in the works. The pictures you guys posted are very helpful and they clearly show corecrypto crashing because of a missing instruction, presumably SSSE3.

 

I'm in deep {censored} with calculus finals, so don't expect much this week however XD

TODO: add my own credits to the files i've written, lol

 

-- Sinetek

 

*dons the thinking hat*

 

connactic, there's a few technical details i'd want to precise out,

the commpage has to be filled in, maybe it's best to do so unconditionally? the actual call to the commpage occurs in idt.s/idt64.s, so if you don't load the commpage but still leave the idt modified, then you'll get a hard reboot on any invalid instruction.

If run on a cpu with SSSE3, then the routines will never be called and no performance hit incurs.

 

another thing, there is super magic voodoo that must be done at compile time in order for the commpage binary to work.. this is because of apple's linker.. do a search for "ssse3emu.o" in the makefiles to see how it's done, or if you don't get it, ask me then

 

*doffs the thinking hat*

  • Like 1
Link to comment
Share on other sites
zchef2k
zchef2k
Posted
Posted

howdy,

 

The pictures you guys posted are very helpful and they clearly show corecrypto crashing because of a missing instruction, presumably SSSE3.

 

 

My CPUs do not have SSSE3 and load corecrypto fine. For the 3rd time, it's the SSE3 instruction causing the issue, not the lack of SSSE3.

 

Specific CPUID mask to allow corecrypto to load successfully is:

 

Level 1 ECX

---- ---- ---- ---- ---- ---- ---- ---0

 

Toggling CPUID masks in VMWare got me at least somewhere. This came from masking AMD's SSE3 fature. No more KP on corecrypto. Still hangs though as it does with fips_mode=0.

 

Untitled.jpg

Link to comment
Share on other sites
theconnactic
theconnactic
Posted
  • Author
Posted

Well, i'll try later to tweak the kernel to see if the situation improves a little. Some of the possible causes for the errors we've been getting are:

 

- the ssse3 emulator is incorrectly integrated to the commpage.c file - the proper integration would save us time, even the ssse3 emulator being not ready yet;

- the ssse3 emulator, as sinetek said before, isn't ready, ergo won't work in 64-bit - we need to wait for a 64-bit capable emulator;

- there are bad AMD-specific instructions that are preventing corecrypto to load - zchef2k found a workaround for this;

- bad patches elsewhere - remember: these patches of mine are just an adaptation from RAWx86's Lion patches and maybe we'll need to write ML-specific patches from scratch.

 

Tonight i'll possibly work on a better integration of the ssse3 emulator. Now, just a little gift for Bulldozer users: i compiled a i386 patched kernel, also based on RAWx86's work, for 10.7.5 - we've been using the patched 10.7.4 for it, not an optimal solution at all. This one should work better. Boot with connactic arch=i386 plus the usual boot flags you use for your system. Non-Bulldozer/Piledriver users should add the bootflag -legacy:

connactic.zip

Link to comment
Share on other sites
Shaneee
Shaneee
Posted
Posted

Well, i'll try later to tweak the kernel to see if the situation improves a little. Some of the possible causes for the errors we've been getting are:

 

- the ssse3 emulator is incorrectly integrated to the commpage.c file - the proper integration would save us time, even the ssse3 emulator being not ready yet;

- the ssse3 emulator, as sinetek said before, isn't ready, ergo won't work in 64-bit - we need to wait for a 64-bit capable emulator;

- there are bad AMD-specific instructions that are preventing corecrypto to load - zchef2k found a workaround for this;

- bad patches elsewhere - remember: these patches of mine are just an adaptation from RAWx86's Lion patches and maybe we'll need to write ML-specific patches from scratch.

 

Tonight i'll possibly work on a better integration of the ssse3 emulator. Now, just a little gift for Bulldozer users: i compiled a i386 patched kernel, also based on RAWx86's work, for 10.7.5 - we've been using the patched 10.7.4 for it, not an optimal solution at all. This one should work better. Boot with connactic arch=i386 plus the usual boot flags you use for your system. Non-Bulldozer/Piledriver users should add the bootflag -legacy:

connactic.zip

 

Tried booting ML DP1 with this kernel and got errno 85 on launched witch is EBADEXEC 85 /* Bad executable */

Link to comment
Share on other sites
Shaneee
Shaneee
Posted
Posted

hello ,

 

this is an i386 kernel 10.7, does not work on ML10.8

 

Darwin Kernel Version 11.4.2

: Tue Dec 4 17:27:03 UTC 2012; bronzovkAnVoodoo v0.9.30:/Users/snowyjoe/Desktop/XNU1075PATCHED/BUILD/obj/RELEASE_I386

 

 

Yes but I am trying it with DP1 of ML which is i386 and effectively Lion.

Link to comment
Share on other sites
 Share
Go to topic listing
×
×
  • Create New...