Jump to content

New AGPGart

Slice

By Slice,
in AGPGart

 Share
941 posts in this topic

Recommended Posts

cybland
cybland
Posted
Posted

Hi Slice,

 

I know that AGP is 32bit, hence "AAPL,gart-width"=0x20. What I was asking is if you know if graphical kexts are checking for the value of "AAPL,gart-width" and if they check for it than maybe, in Leo, kexts need a value 0x40 (64bit PCIe). I guess that, if that is true, it would mean that AGP cannot work in Leo. Anyhow I'll test your new AGPGart later today and report my findings.

Link to comment
Share on other sites
Slice
Slice
Posted
  • Author
Posted
Hi Slice,

 

I know that AGP is 32bit, hence "AAPL,gart-width"=0x20. What I was asking is if you know if graphical kexts are checking for the value of "AAPL,gart-width" and if they check for it than maybe, in Leo, kexts need a value 0x40 (64bit PCIe). I guess that, if that is true, it would mean that AGP cannot work in Leo. Anyhow I'll test your new AGPGart later today and report my findings.

If so it is finish. I know that Tiger's ATIRadeon8500.kext check gart-width and works different way in the two cases. To know that I reversed the kext and read 10000 codes. I am not ready to repeat the procedure with Leo kext.

Link to comment
Share on other sites
cybland
cybland
Posted
Posted
If so it is finish. I know that Tiger's ATIRadeon8500.kext check gart-width and works different way in the two cases. To know that I reversed the kext and read 10000 codes. I am not ready to repeat the procedure with Leo kext.

 

 

Well I suspected that since I've tried everything that I thought of and no success in Leo. All that is left then is to perfect the AGPGart for Tiger (for VIA and Intel chipsets) and maybe hope that someone will stumble on the solution for Leo.

Link to comment
Share on other sites
Wppley
Wppley
Posted
Posted

2 Wppley

Corrected for VIA chipset. I check linux driver from VIA and made all the same.

http://www.viaarena.com/Driver/VT3324-3364...nel-src_200.tgz

Install, repair permissions, boot with -s and then sh /etc/rc. If stop, make dmesg for me to know where is the stop.

 

hi Slice.

 

Well, unfortunately, it doesn‘t works here. Stops in boot at the same point that 265 and 266:

AGP: found VIA VT83xx/VT87xx/KTxxx/Px8xx host to pci bridge

AGP: savebridgestate ok

AGP create nub for 022110de bus=1 CAP=68

AGPGart: publish nub display

after that, nothing more happens.

 

Im sending dmesg for you but without AGPGart loaded, off course, i dont if it help (i‘ve saw something about nub there).

 

Thank you for your hard work with my issue.

 

dmesgWppley.txt.zip

Link to comment
Share on other sites
Slice
Slice
Posted
  • Author
Posted
Well I suspected that since I've tried everything that I thought of and no success in Leo. All that is left then is to perfect the AGPGart for Tiger (for VIA and Intel chipsets) and maybe hope that someone will stumble on the solution for Leo.

I am sure that ATIRadeonXXXX.kext works in Leo with gart-width=0x20. The problem is in AMD64.

 

Slice can i get your help with my ati 9600 card? Here's a copy of my ioreg/dmesg.

You have the same problem: AMD64 in Leo. If you have Tiger you can check in it.

And with aperture=256Mb ATIRadeon9700.kext is not working.

 

hi Slice.

 

Well, unfortunately, it doesn‘t works here. Stops in boot at the same point that 265 and 266:

AGP: found VIA VT83xx/VT87xx/KTxxx/Px8xx host to pci bridge

AGP: savebridgestate ok

AGP create nub for 022110de bus=1 CAP=68

AGPGart: publish nub display

after that, nothing more happens.

 

Im sending dmesg for you but without AGPGart loaded, off course, i dont if it help (i‘ve saw something about nub there).

:D Previously I didn't understand your stop point. It is strange. At this place common part of AGPGart is working. Other users have no difficulties here. I need to think.

P.S. dmesg without AGPGart is not informative.

Link to comment
Share on other sites
jeffc2005
jeffc2005
Posted
Posted
I am sure that ATIRadeonXXXX.kext works in Leo with gart-width=0x20. The problem is in AMD64.

You have the same problem: AMD64 in Leo. If you have Tiger you can check in it.

And with aperture=256Mb ATIRadeon9700.kext is not working.

:angel: Previously I didn't understand your stop point. It is strange. At this place common part of AGPGart is working. Other users have no difficulties here. I need to think.

P.S. dmesg without AGPGart is not informative.

 

 

can you tell me how to change the aperture..

withoutagp.zip

Link to comment
Share on other sites
cybland
cybland
Posted
Posted
I am sure that ATIRadeonXXXX.kext works in Leo with gart-width=0x20. The problem is in AMD64.

 

Hi Slice

 

I've started testing AGPGart 267. I'll go through several settings of AGP_Base, AGP_Mode and flush and try to collect as much data as I can. It'll take some time. In the meantime here are some docs about AMD64. Maybe it will help you.

 

1.) BIOS and Kernel Developer's Guide for the AMDAthlonTM 64 and AMDOpteronTM Processors –> http://www.amd.com/us-en/assets/content_ty..._docs/26094.PDF

 

2.) Software Optimization Guide for AMD64 Processors –> http://www.amd.com/us-en/assets/content_ty..._docs/25112.PDF -> Appendix D AGP Considerations

 

I have made a PCI dump in WinXP. Maybe you'll find some useful info.

PCI_report_128Mb_Apreture.txt

SIV_report_128Mb_Apreture.txt

Link to comment
Share on other sites
zorglups
zorglups
Posted
Posted

Hi Slice,

 

Thanks, it is partely clearer now.

 

I guess I gave too many details on what I try; The short version is:

5) ATIRadeon9700* from 10.4.11 and AGPGart266 - various AGP_Base

Nothing worked at all: black screen on all Bases/fastwrite enabled/disabled (<key>AGP_Mode</key><integer>0xffffffff</integer> / <integer>0x7fffffff</integer> ).

<key>IOProviderClass</key>

<string>IOFramebuffer</string>

with IONDRVSupport1441, IONDRVSupport1442 (yours), IONDRVSupport1.4.8 (10.4.11)

with IOPCIFamily2.2 (10.4.11) and IOPCIFamily2.2.4 (yours)

The difficulty is to get any info on either combination without GUI (command line)

 

7+8) New ATIRadeon9700* from 10.4.11 , Callisto003 and AGPGart266 (AGP_Base?)->

You are walking around my statement that Callisto is not compatible with AGPGart. To know why I need to make new framebuffer.

Sorry about that, I did not read that Callisto is not compatible with AGPGart before.

But I guess taht is all off topic for you...

 

9) ATIRadeon9700* from 10.4.11 , AGPGart266 (AGP_Base?) AND Modifyed ATIRadeon9700GA.plugin

Without correct name you will have software QE/CI and no ATIR300 - it is strict rule.

<string>10024a49-0000-0000-0000-000a27898d3e</string> - I am really doubt that it has influence on somehow. If somebody show me tests with differencies it would be great contribution to the project.

You are most probably right: it is ATIRadeon9700GA.plugin that AGPGart refuse to work with (at least its ATIR300 part). Now I could probably just remove the plugin all together, and AGPGart would work but is that the point?

So the trick for me at least is that modifying

<string>10024a49-0000-0000-0000-000a27898d3e</string>

simply prevents the ATIRadeon9700GA.plugin from loading (evrything or just the ATIR300 part?),

and I get a working GUI to dump info on the kext changes.

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

How much of you! :)

1.) BIOS and Kernel Developer's Guide for the AMDAthlonTM 64 and AMDOpteronTM Processors –> http://www.amd.com/us-en/assets/content_ty..._docs/26094.PDF2.) Software Optimization Guide for AMD64 Processors –> http://www.amd.com/us-en/assets/content_ty..._docs/25112.PDF -> Appendix D AGP ConsiderationsI have made a PCI dump in WinXP. Maybe you'll find some useful info.
It is really help. Thank! No comment while I am reading.Another question. If AMD64 work good in Tiger but not work in Leo then we need to know what is the difference between the systems concerning CPU or memory controller.
Slice can you tell me what "May 29 19:57:15 jcs-7426gx kernel[0]: VGA: match category IOFramebuffer exists" means? I have attached a copy of my sys log.
You probably write IOFramebuffer into not good place.And I notice some bad place in your syslog
May 29 20:12:44 jcs-7426gx kernel[0]: AGPAMD trace host PCI spaceMay 29 20:12:44 jcs-7426gx kernel[0]: (00)=31881106 (04)=22300006 (08)=06000001 (0c)=00000800 May 29 20:12:44 jcs-7426gx kernel[0]: (10)=00000008 (14)=00000000 (18)=00000000 (1c)=00000000
It is AGP_Base. The value must not be 0. There must be e0000008. My mistake?
The difficulty is to get any info on either combination without GUI (command line)So the trick for me at least is that modifying10024a49-0000-0000-0000-000a27898d3esimply prevents the ATIRadeon9700GA.plugin from loading (evrything or just the ATIR300 part?),and I get a working GUI to dump info on the kext changes.
Yes, it is a trick. Without GA.plugin you can get all informations except speed.Try to get new information by means of RadeonPCIhttp://forum.insanelymac.com/index.php?sho...96664&st=27Without GA in two configurations: with and without AGPGart. It will be great!

Thank Cybland!It is a mistake in previous AMD driver. We need no shift Gart_Physical address!!!

Picture_1.png

Link to comment
Share on other sites
cybland
cybland
Posted
Posted
Thank Cybland!It is a mistake in previous AMD driver. We need no shift Gart_Physical address!!!

 

 

You're welcome Slice. I'm glad that info has helped. I'll wait for the corrected driver to do the tests.

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

2 Toadspit

I see you are here :wacko:

Increase flush

AGP_FlushTimeout

10

to 100, 1000, 10000, 100000 and more and check speed

 

Changes for all. Some changes for AMD64. Hope it is better for Leo.

The version is still 2.6.7. Previous one is not good.

AGPGart267.kext.zip

Link to comment
Share on other sites
Wppley
Wppley
Posted
Posted
2 Toadspit

I see you are here -_-

Increase flush

<key>AGP_FlushTimeout</key>

<integer>10</integer>

to 100, 1000, 10000, 100000 and more and check speed

 

Changes for all. Some changes for AMD64. Hope it is better for Leo.

The version is still 2.6.7. Previous one is not good.

 

Hi Slice.

 

Do you think its better for me too?

 

Thanks in advance.

 

Edit.

Nothing changes for me. :hysterical:

Link to comment
Share on other sites
Slice
Slice
Posted
  • Author
Posted
Hi Slice.

 

Do you think its better for me too?

 

Thanks in advance.

 

Edit.

Nothing changes for me. :(

Up to now I don't understand what is the mistake for you. Did you see any new messages after "publish nub"?

Is there any old version that don't freeze your system?

Check once more what information can you give me.

I will be absent up to 6 of june.

Link to comment
Share on other sites
toadspit
toadspit
Posted
Posted
2 Toadspit

I see you are here :)

Increase flush

<key>AGP_FlushTimeout</key>

<integer>10</integer>

to 100, 1000, 10000, 100000 and more and check speed

 

Changes for all. Some changes for AMD64. Hope it is better for Leo.

The version is still 2.6.7. Previous one is not good.

 

I've never gone away! :P

I'll give you some result ASAP

Link to comment
Share on other sites
Wppley
Wppley
Posted
Posted
Up to now I don't understand what is the mistake for you. Did you see any new messages after "publish nub"?

Is there any old version that don't freeze your system?

Check once more what information can you give me.

I will be absent up to 6 of june.

 

Hi Slice.

 

Nothing happens after publish nub, it freezes.

v2.1 is loaded, system profiler says AGP but Open GL suffer decrease: without AGPGart=85 / with AGPGart=74 in Xbench.

v2.2 is loaded, system profiler says AGP but no CI/QE.

 

Thanks in advance and good rest.

Link to comment
Share on other sites
jeffc2005
jeffc2005
Posted
Posted

Slice, what does this mean?

 

1 13:00:08 localhost kernel[0]: IONDRV: VGA baseAddr=d8000000 rowBytes=00001000 Width=00000400 height=00000300 depth=00000020

Jun 1 13:00:08 localhost com.apple.launchd.peruser.501[80]: BootCacheControl: could not insert prefetch tag: No such file or directory

Jun 1 13:00:09 localhost kernel[0]: CODE SIGNING: cs_invalid_page: p=78[ATSServer] clearing CS_VALID

Jun 1 13:00:09 localhost kernel[0]: CODE SIGNING: cs_invalid_page: p=76[mdworker] clearing CS_VALID

Jun 1 13:00:11 localhost loginwindow[36]: Login Window Started Security Agent

 

I have included a copy of my ioregistry and my xbench score jumped from 30's/40's -> 80's after i updated my AppleVIAATA.kext (1.02). Can you explain?

 

Oh, I am using ati9700(1.4.58)/agpgart(2.6.7)/iopcifamily(2.4.4)/iondrvsupport(1.4.42).

Archive.zip

Link to comment
Share on other sites
jeffc2005
jeffc2005
Posted
Posted

Slice or anyone, how do i get agpgart to reconize another pciclass(0x06000000)?

 

Jun 1 16:38:03 localhost kernel[0]: AGP: Found VIA 8385 host to PCI bridge

Jun 1 16:38:03 localhost kernel[0]: AGPBridge has secondaryBus 0

Jun 1 16:38:03 localhost kernel[0]: AppleVIAATADriver: VIA 8235 (CMD 0x1f0, CTR 0x3f4, IRQ 14, BM 0x1ce0)AppleVIAATADriver: VIA 8235 (CMD 0x170, CTR 0x374, IRQ 15, BM 0x1ce8)

Link to comment
Share on other sites
zorglups
zorglups
Posted
Posted

Hi Slice,

 

While you are away (not at Computex by any chance?)

Update:

5) ATIRadeon9700* from 10.4.11 and AGPGart266 - various AGP_Base or AGPGart267 AND Modifyed ATIRadeon9700GA.plugin

Nothing worked at all: black screen on all Bases/fastwrite enabled/disabled (<key>AGP_Mode</key><integer>0xffffffff</integer> / <integer>0x7fffffff</integer> ).

<key>IOProviderClass</key>

<string>IOFramebuffer</string>

with IONDRVSupport1442 (yours), and IOPCIFamily2.2.4 (yours)

with RadeonPCI (RadeonDump Ok), and IORadeonSupport101 (both recompiled with Xcode).

Both AGPGART ok, and very tolerant to AGP_Base values.

 

Black screen with

<key>IOProviderClass</key>

<string>IONDRVFramebuffer</string>

 

Original ATIRadeon9700GA.plugin (i.e. with acceleration):

black screen on both Framebuffers.

Link to comment
Share on other sites
cybland
cybland
Posted
Posted

Hi Slice,

 

Slice: I tried to modify flushtime with following vals: 10 100000 100000000

No performance improvements at all.

Test results in attachment.

 

t.

 

Same here. Tested in Tiger using Xbench, OpenGl Viewer, Cinebench 9.5 and Cinebench 10. All results consistent from AGPGart 264 to AGPGart 267_new (corrected). On the subject of the corrected AGPGart I have seen no difference between the old one and corrected one. Both work same on my system.

 

Hi Slice,

 

While you are away (not at Computex by any chance?)

Update:

5) ATIRadeon9700* from 10.4.11 and AGPGart266 - various AGP_Base or AGPGart267 AND Modifyed ATIRadeon9700GA.plugin

Nothing worked at all: black screen on all Bases/fastwrite enabled/disabled (<key>AGP_Mode</key><integer>0xffffffff</integer> / <integer>0x7fffffff</integer> ).

<key>IOProviderClass</key>

<string>IOFramebuffer</string>

with IONDRVSupport1442 (yours), and IOPCIFamily2.2.4 (yours)

with RadeonPCI (RadeonDump Ok), and IORadeonSupport101 (both recompiled with Xcode).

Both AGPGART ok, and very tolerant to AGP_Base values.

 

Black screen with

<key>IOProviderClass</key>

<string>IONDRVFramebuffer</string>

 

Original ATIRadeon9700GA.plugin (i.e. with acceleration):

black screen on both Framebuffers.

 

Also same here when using AGPGart in Leo. I think the problem lies with GA.plugin (GeForceGA.plugin in my case). If you disable it from loading you can boot Leo to GUI but without CI/QE. Also installing one from Tiger gets you to GUI but without CI/QE. Slice I still think that GA.plugin in Leo checks for gart-width=0x40 and will not work correctly if gart-width=0x20. The only way to make AGPGart work in Leo would be to make Tiger's GA.plugin compatible with Leo, but I don't think that is possible.

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

I was in Germany at Drupa-2008. I was not resting and not working about the our problem.

Some comments.

Nothing happens after publish nub, it freezes.

v2.1 is loaded, system profiler says AGP but Open GL suffer decrease: without AGPGart=85 / with AGPGart=74 in Xbench.

v2.2 is loaded, system profiler says AGP but no CI/QE.

If 2.1 and 2.2 is loaded and you can boot into GUI then two possibility

1. They are both not working at all. Only said "AGP".

2. When I switch on VIA hardware the output goes to bad address. Did you see IDE LED flashing?

 

Slice: I tried to modify flushtime with following vals: 10 100000 100000000

No performance improvements at all.

:blink: I see your have aperture=0x02000000=32Mb. With this aperture you can't have QE! See my previous post with my answer to JaS. So no changes with flushtime.

 

Slice or anyone, how do i get agpgart to reconize another pciclass(0x06000000)?

AGPGart choose the only device with pciclass=0600 that have AGP capabilities. It is already done. Forget it!

Jun 1 16:38:03 localhost kernel[0]: AGPBridge has secondaryBus 0

It is a mistake of bus numbering. Did you use my IOPCIFamily 2.4.4?

Did you see a message that AGPGart has corrected secondaryBus?

 

Hi Slice,

 

While you are away (not at Computex by any chance?)

At Drupa.

Nothing worked at all: black screen on all Bases/fastwrite enabled/disabled

Something works?

with RadeonPCI (RadeonDump Ok), and IORadeonSupport101 (both recompiled with Xcode).

The IORadeonSupport101 is not working yet!

Both AGPGART ok, and very tolerant to AGP_Base values.

Without GA.plugin all is good with any values. But without acceleration.

 

Hi Slice,

Same here. Tested in Tiger using Xbench, OpenGl Viewer, Cinebench 9.5 and Cinebench 10. All results consistent from AGPGart 264 to AGPGart 267_new (corrected). On the subject of the corrected AGPGart I have seen no difference between the old one and corrected one. Both work same on my system.

Also same here when using AGPGart in Leo. I think the problem lies with GA.plugin (GeForceGA.plugin in my case). If you disable it from loading you can boot Leo to GUI but without CI/QE. Also installing one from Tiger gets you to GUI but without CI/QE. Slice I still think that GA.plugin in Leo checks for gart-width=0x40 and will not work correctly if gart-width=0x20. The only way to make AGPGart work in Leo would be to make Tiger's GA.plugin compatible with Leo, but I don't think that is possible.

I don't see if GA.plugin is reading gart-width. But I can say nothing about nVidia drivers.

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