Patches Proceed For Disabling Radeon AGP GART, Deprecating TTM AGP

[Mat2]

If you have old decrepit hardware, just use older software and operating systems. It's really that simple. You didn't see people in 2000 whining that Debian 2.2 didn't fully support their IBM PC 5150 from '81 - that's the same equivalent time-span here.

Computers have simply become usable for a longer time period. Currently I think that even computers that are like 15 years old are usable for basic tasks (like browsing the web or word processing) - after simple upgrades like RAM extensions. (Note that I am thinking of mid-range computers, that would exclude netbooks). In 2005 most processors by AMD were 64-bit and Intel was switching to amd64. These computers could still be used with lightweight distros like LXDE and Debian did not remove 32-bit support.

In our family there used to be a computer that was built around 2002 (with an Athlon XP). It worked reasonably well on Ubuntu until Firefox removed support for processors that do not have SSE2 extensions. It was ultimately upgraded only last year.

When it comes to AGP GART, most graphic cards from ~2005 seem to be unaffected by the change, at least when it comes to Radeons.

Support for older hardware is one of the advantages of Linux.

Of course using older software is risky from the security standpoint.

[AmericanLocomotive]

Honestly, 2005-era computers are borderline useless for browsing the web these days. Even Core 2 Duo systems from 2008 are extremely marginal - something like YouTube stutters and frame-drops on those systems. Trying to open more than a few tabs at once in Chrome is an exercise in patience. These days, a $200 Chromebook with a Celeron will mop the floor CPU-performance wise on just about anything from 2005 and earlier. Factor in hardware video decode, and it's not even remotely a competition.

...but most 2005 era systems (especially OEM ones) will be using PCIe, so AGP GART is irrelevant. The real AGP systems are Pentium 3s, Pentium 4s and Athlon XPs. Those systems are absolute dogs today. They're so slow, that any kind of even remotely taxing website will bring the whole system to a screeching halt.

I guess what I'm saying is that the systems that this change impacts are so old and so slow, they aren't even usable as internet machines anyways. So there's really no problem with using an older kernel, because those machines shouldn't even be on the internet anyways (thus, no real worries about security).

[oiaohm]

Dell Inspiron 8100 with Intel 82815 PCI/AGP chipset.

NOTE: AFTER ~5 seconds, a hard freeze/lock occurs, sometimes escapable by kernel sys umount/reboot key, most other times hard a hard freeze.

Has this machine been opened and new thermal paste applied to cpu or at least inspected for having thermal paste on the cpu?

Dell made a large batches of Dell Inspiron 8100 and failed to put thermal paste between the CPU and heat sink that went to government orders. The RMA for these was quote the id numbers and get free one send out to replace it. Please note you did not need to return the broken ones and were legally allowed to give those away for free with zero warranty. Yes they were random-ally locking up under windows when CPU load was substantiated.

This change is going to alter load around and a few bits of hardware that use to work with missing thermal paste most of the time now will show 100 percent predictable crash. This might not be a quirk might be simply you have broken hardware and have failed to notice.

rogerx as your numbers showed its 5% worse cpu load GPU framerates don't drop so usability is barely effected as long as the machine does not crash. Remember most of the machines form AGP time-frame the laptop batteries are dead so you are operating from wall power so the extra power usage is not major problem.
Some old machines will crash because:
1) The machine was defective from new as in no thermal paste between the CPU and heatsink and where converted to Linux because windows would not run stability. There are quite a few different models from the 1998 to 2002 that due to production errors this happened with RMA polices where these machines did not have to be returned and could be given away for free. So yes there are a lot of lemon machines out there.
2) Some old machines will crash because the thermal paste is buggered and need to be replaced and the 5 percent extra load will make this problem show itself where it would not have just yet. Please note the just yet the problem would have still been coming. Yes thermal paste does kind of dry out and lose sections of effectiveness over time really old machines it does need to be replaced. Most thermal pastes are meant to be replaced at 15 years old yes a Dell Inspiron 8100 is well past that. So if did have thermal paste when new and never been replaced it buggered now.
3) Finally something that will need a quirk/setting put into PCI GART code.

Be sure you are looking at a quirk the thermal paste issue need to be ruled out.

Shrugs, at this rate, I may consider buying a Mac laptop instead, in the future. At least I'll know I'll have a snapshot of an O/S working 10-20 years down the road.

Really I am totally sick of people saying this because saying this says you have not done any homework..

That Dell Inspiron 8100 is 2001 machine.

Lets pull in a few facts here. Firefox current version requires 10.9 Mac OS to run. Libreoffice requires 10.10 to run.

https://en.wikipedia.org/wiki/MacOS#Hardware_compatibility

So that is a 2007 or new machine required to run current software like current version of Firefox and LibreOffice. Basically pcie based system the old AGP mac systems if you want to run current software you have to convert them to Linux. Please note 10.9 and 10.10 are not getting secure updates to the OS. If you are talking current version of Mac OS so you have all security updates you machine has to be 2012 or newer.

Apple really only provides 8 years of proper hardware support. Company and organisations providing applications for Mac OS only provide 15 years of support. So 8-15 years of usability not 10-20 as you claimed.

With the modern t2 chips in Apple laptops they are not that repairable or dependably life extendable by OS replacement.

https://www.forbes.com/sites/jasonevangelho/2018/11/13/apple-warns-newer-macs-have-a-serious-limitation

Yes the t2 chip also gets in way with modern Apple laptop from being converted to Linux. Basically you don't buy a modern Apple laptop and expect that once Apple stops supporting it that you can put Linux on it to extend it life. Yes when the 8-15 years is up you have to class Modern apple laptop as brick at that point for functional usability.

So the idea of going Apple will make you life better is bogus.

Also remember you laptop is going to be support by Ubuntu 20.04 for modern software for about another 5 years with the old AGP code. So you hardware from the Linux world is getting 25 years of support before you will have to change over to PCI GART or use a old distribution no longer supported distribution. Even that change is happening now it will be a while before major trouble hits you and in that time quirk fixes could have got into the mainline kernel. Yes there 5 years of testing that can be performed before the shoe finally drops.

Linux world when it comes to changing way hardware is supported does not move quickly.

[rogerx]

Has this machine been opened and new thermal paste applied to cpu or at least inspected for having thermal paste on the cpu?

Dell made a large batches of Dell Inspiron 8100 and failed to put thermal paste between the CPU and heat sink that went to government orders. The RMA for these was quote the id numbers and get free one send out to replace it. Please note you did not need to return the broken ones and were legally allowed to give those away for free with zero warranty. Yes they were random-ally locking up under windows when CPU load was substantiated.

Yes. Rebuilt over the years and maintained well, None of what you stated is anything of the sort of I have experienced with this Dell Inspiron 8100. The unit works well from initial purchase, and is well maintained. Thermal paste cleaned/replaced over the years on both CPU/GPU. (The Dell Inspirons/Lattitudes from this era have one of the best typing keyboards, almost similar in performance to the older electronic typewriters. All other keyboards have performed poorly, in my opinion.)

The two to three series of quick repeated glxgears tests were constistent and reproducible 100%. Only problem, I do not have time to run a debugger as nobody maintains my house except I. And, I'm not rich enough to afford a maid, etc.

My Dell laptop experience was awesome, along with easily acquiring replacement parts, likely due to Dell's excellent manufacturing/distribution channel logistics.

rogerx as your numbers showed its 5% worse cpu load GPU framerates don't drop so usability is barely effected as long as the machine does not crash. Remember most of the machines form AGP time-frame the laptop batteries are dead so you are operating from wall power so the extra power usage is not major problem.

Can replace the batteries, just do not have the time. Is using wall power on an older laptop a sin?

Some old machines will crash because:
1) The machine was defective from new as in no thermal paste between the CPU and heatsink and where converted to Linux because windows would not run stability. There are quite a few different models from the 1998 to 2002 that due to production errors this happened with RMA polices where these machines did not have to be returned and could be given away for free. So yes there are a lot of lemon machines out there.

As I already stated, thermal place has always been in place to my knowledge. Tests were reproducible.

2) Some old machines will crash because the thermal paste is buggered and need to be replaced and the 5 percent extra load will make this problem show itself where it would not have just yet. Please note the just yet the problem would have still been coming. Yes thermal paste does kind of dry out and lose sections of effectiveness over time really old machines it does need to be replaced. Most thermal pastes are meant to be replaced at 15 years old yes a Dell Inspiron 8100 is well past that. So if did have thermal paste when new and never been replaced it buggered now.

Repeated yourself from #1 above.

3) Finally something that will need a quirk/setting put into PCI GART code.

You'll need to explain this one. I think I know what you might be touching on, but I'll let you explain.

Really I am totally sick of people saying this because saying this says you have not done any homework..

That Dell Inspiron 8100 is 2001 machine.

Lets pull in a few facts here. Firefox current version requires 10.9 Mac OS to run. Libreoffice requires 10.10 to run.

https://en.wikipedia.org/wiki/MacOS#..._compatibility

So that is a 2007 or new machine required to run current software like current version of Firefox and LibreOffice. Basically pcie based system the old AGP mac systems if you want to run current software you have to convert them to Linux. Please note 10.9 and 10.10 are not getting secure updates to the OS. If you are talking current version of Mac OS so you have all security updates you machine has to be 2012 or newer.

Apple really only provides 8 years of proper hardware support. Company and organisations providing applications for Mac OS only provide 15 years of support. So 8-15 years of usability not 10-20 as you claimed.

With the modern t2 chips in Apple laptops they are not that repairable or dependably life extendable by OS replacement.

https://www.forbes.com/sites/jasonev...ous-limitation

Yes the t2 chip also gets in way with modern Apple laptop from being converted to Linux. Basically you don't buy a modern Apple laptop and expect that once Apple stops supporting it that you can put Linux on it to extend it life. Yes when the 8-15 years is up you have to class Modern apple laptop as brick at that point for functional usability.

So the idea of going Apple will make you life better is bogus.

Also remember you laptop is going to be support by Ubuntu 20.04 for modern software for about another 5 years with the old AGP code. So you hardware from the Linux world is getting 25 years of support before you will have to change over to PCI GART or use a old distribution no longer supported distribution. Even that change is happening now it will be a while before major trouble hits you and in that time quirk fixes could have got into the mainline kernel. Yes there 5 years of testing that can be performed before the shoe finally drops.

Linux world when it comes to changing way hardware is supported does not move quickly.

Lots of facts, however misconstrued. Not everybody uses Firefox.

I use, most times, a terminal. I sometimes rely upon Xorg/DWM.

I rarely use phone support for any purchased software, unless I've tracked a bug and desire to report the facts. Even then, it's futile. The commerical software support for patches, does not mean a company will break their working software platform after ~8 years. The Windows XP O/S works just fine on my laptop, granted, the laptop has limited access, if ever, to the Internet.

A better solution, is to take snapshots the kernel and Xorg, along with any related drivers to prevent such older hardware from breaking. The kernel is sometimes agnostic, as due to the Linux kernel's wonderful modular feature, allowing modules from being compiled and/or installed. Also, the kernel can be compiled to working around such 32/64 bit problems.

Facts are facts. I spent an hour recompiling the kernel several times and redundantly testing. AGP kernel module works, kernel PCI GART failed. The laptop constantly spends hours a day recompiling/installing software for the source based operating system, so quite stable. Also, not a Ubunta fan here, and using another operating system Linux distribution.

A lot of older proven machines are usually relied upon for other tasks. I'm sure somebody out there will probably also start citing reasons why NASA should upgrade it's older machines as well. Spending time trying to prove 2000-2015 machines no longer exist or are usable, sounds futile to me.

Intentionally breaking code that works, seems like a bad idea.

[oiaohm]

Yes. Rebuilt over the years and maintained well, None of what you stated is anything of the sort of I have experienced with this Dell Inspiron 8100. The unit works well from initial purchase, and is well maintained. Thermal paste cleaned/replaced over the years on both CPU/GPU.

So this machine has had it correct maintenance. To be correct there are still a lot machines throwing issues out there that people have not done this maintenance. Just had to check.

My Dell laptop experience was awesome, along with easily acquiring replacement parts, likely due to Dell's excellent manufacturing/distribution channel logistics.

That is true for Dell except for they have done return merchandise authorization (RMA) process using the section where the product does not have to be returned to be replaced. 1998-2002 Please note its not just Dells different parts of HP and others distribution channels in the same time frame did the same thing, Its just something to be aware of when dealing with machines from that time frame. That Lemons computer to governments and major companies in that time frame were not always returned to make for refurbishment. Companies worked out this was causing reputation damaged and stopped doing it in 2002.

Can replace the batteries, just do not have the time. Is using wall power on an older laptop a sin?

Its not a sin to have not replaced those batteries but when you are looking at how these old machines are used its more likely that people will not be using them on battery any more and will be using them from wall.

You'll need to explain this one. I think I know what you might be touching on, but I'll let you explain.

The fact that the basic hardware causes are dealt this leaves bug. AGP code in AGP gart had a few quirks on max speed it is possible the PCI GART tried to transfer stuff faster than works. There are other quirks that the PCI GART code does not have again its not been testing on the AGP platform that much on x86. So some teething problems by change would be expected. Even so this will still end up with simpler to maintain code.

Lots of facts, however misconstrued. Not everybody uses Firefox.

The reality here firefox is about the last maintained bit of software to leave platform. Chrome only goes back to 10.10 mac os same as libreoffice. MS Office needs 10.13 mac os or better so 2009-2010 yes Microsoft does move this number forwards.

The commerical software support for patches, does not mean a company will break their working software platform after ~8 years.

Except that really does not apply on a Mac OS machine provided by Apple with at T2 chip.

https://eclecticlight.co/2019/09/10/dont-try-reverting-a-t2-to-older-firmware/

T2 chips are one directional upgrades. Yes you have upgraded T2 chip for newer version of Mac OS now older version of Mac OS no longer works correctly or not working at all. Its nicely possible with T2 to upgrade it to only be compatible with a version of Mac OS that is not compatible with your hardware at all.

After 8 years Apple is not going to fix any T2 issue that happens to you. This is not like PC firmware that you can get old version of firmware by some means put it back in the system so old operating system works. Also T2 can break before the 8 year point and people are already taking machines back to Apple to have the T2 fixed.

Modern Apple machines don't have the same means to run older OS as your normal PC class hardware.

A better solution, is to take snapshots the kernel and Xorg, along with any related drivers to prevent such older hardware from breaking.

From a security point of view that not a valid plan. That almost as bad as just install a old Linux distribution that is no longer supported.

A lot of older proven machines are usually relied upon for other tasks. I'm sure somebody out there will probably also start citing reasons why NASA should upgrade it's older machines as well. Spending time trying to prove 2000-2015 machines no longer exist or are usable, sounds futile to me.

Except NASA and others fund work like

https://www.cip-project.org/

To keep old kernel alive with security updates for the hardware platforms they have.

The last AGP motherboards from 2005 had a pcie x16 slot on them. Laptops made after start of 2004 don't have anything AGP in chipsets they are all pcie.

In laptops you have 6 years of production from 1997 to end 2003 that can contain a AGP chipset. Please note can not all did even in .

Desktop PC motherboards you have 1997 to 2005 that 8 years of production with production in 2004 and latter having pciex16 slots. So again only a 6 year window of hardware that had to use AGP.

Interesting enough before 1992-1997 graphics cards were PCI. Joining both AGP and PCI under 1 driver means now maintaining a 10 year time frame of hardware with 1 driver.

Intentionally breaking code that works, seems like a bad idea.

Problem is agp gart has been works for you. There have been on going issues with it saying breaking code that works is not exactly true here. agp gart has not been working properly for ages. Of course there comes a point where you need to try different option instead of beating head against same wall. PCI GART on x86 is new for driving agp based cards. Note not new on powerpc. Any time you make a change like this there is going to be a few teething problems.

[rogerx]

So this machine has had it correct maintenance. To be correct there are still a lot machines throwing issues out there that people have not done this maintenance. Just had to check. ...

Sorry. I don't have any time to read all that. As much time spent reading and typing in a correct response, my time would probably be best spent opening a debugger and reading-up on the AGP specification and Linux kernel code.

One of the additional reasons I keep legacy hardware and operating systems around aside from preventing waste, is for legacy hardware such as portable GPS units requiring costly proprietary GPS maps to install. One of my GPS units with costly proprietary maps requires Windows XP for installing using the bare-metal USB bus. The same task redundantly fails within virtual installs, due to the proprietary software bugs.

When one person gets rid of one computer with a specific operating system, they also likely have to trash/loose all the software connected portable devices to the computer/operating system combo, unless the hardware has universal drivers/software. This can get quite costly after acquiring numerous portable pieces of hardware.

[oiaohm]

Sorry. I don't have any time to read all that. As much time spent reading and typing in a correct response, my time would probably be best spent opening a debugger and reading-up on the AGP specification and Linux kernel code.

Reading the AGP specification it does clearly state that PCI Gart should work stable there is a interesting point in the AGP specification that says the AGP GART stuff is optional this is one of the reasons why breakage due to AGP GART removed should not happen by specification. The fact you have a crash either the Linux PCI Gart code has a bug or the hardware you have has a quirk that is not documented yes this means that hardware will not be to AGP specification. Basically PCI GART mode should work with slightly higher CPU usage and that should be it. Anything else will be a bug somewhere in the PCI GART code or hardware quirk.

Due to the specification once using PCI GART mode is done more and the bugs are reported and fixed with hardware quirks and PCI GART code bugs the difference between AGP GART and PCI GART driving a AGP card should basically drop less than 5% higher cpu usage at worst without the AGP GART race condition stability issues.

So you have hit a teething problem.

One of the additional reasons I keep legacy hardware and operating systems around aside from preventing waste, is for legacy hardware such as portable GPS units requiring costly proprietary GPS maps to install. One of my GPS units with costly proprietary maps requires Windows XP for installing using the bare-metal USB bus. The same task redundantly fails within virtual installs, due to the proprietary software bugs.

There is work at the moment to make WINUSB stack work in wine by libusb for existing applications modern applications like this. But no one is investing the developer time to support the XP and before usb stack in wine.

So this problem is lack of investment and lack of making sure device can be updated without OS unique software. Its really sad to be paying for costly proprietary maps and they have not updated the software to update your device they must not care about making money from you long term or are wanting to force you to buy new devices to make more money from you.

When one person gets rid of one computer with a specific operating system, they also likely have to trash/loose all the software connected portable devices to the computer/operating system combo, unless the hardware has universal drivers/software. This can get quite costly after acquiring numerous portable pieces of hardware.

This is not 100 percent true. There is a line in the sand where you can get rid of one very old computer and replace it with a new system that will run exactly the same stuff.

https://github.com/MiSTer-devel/ao486_MiSTer

This is basically i386dx without floating point done in fpga. This is a modern platform that will run up to Windows 95 operating systems. Basically your pre agp hardware can be made in modern day fpga units so you don't need the old motherboards of pre AGP to test drivers for them.

It would be possible with investment as demo by this https://github.com/lmEshoo/sp-i586 to move into the i586 time frame with it running on modern FPGA units yes that would get you as modern platform that would run XP.

Remember computer hardware does not run forever thinking i586/i686 hardware lot of it is now coming up on its physical end of life that there very little investment in FPGA being used to replace it. Once something is over 20 years old patents on the technology has expired. In theory there should be a nice line in sand at 20 years old for hardware where you can throw old hardware away and replace it with new and lose nothing. The reality is the line in sand closer to 25-40 years old where you can bin old hardware.