Author Topic: eCS install problems on a recent mainboard  (Read 4325 times)

Mikkel Simonsen

  • Newbie
  • *
  • Thank You
  • -Given: 0
  • -Receive: 2
  • Posts: 12
    • View Profile
Re: eCS install problems on a recent mainboard
« Reply #15 on: September 03, 2016, 12:20:53 am »
Ouch. That is pretty slow for a modern computer.

Yes, but this is a 10W CPU.  I'm looking for a replacement for an 8-year old system with a 100W+ CPU. Something quiet with low power consumption. I don't think it will be much slower (if it is at all), than the system it's meant to replace.


Quote
Yes, Sysbench runs CPU test multi-threaded. Most OS/2 usage is single threaded, so the higher speed for four cores can be pretty misleading.

The system runs multi-threaded software like Weasel and Goserve, so the extra cores do come in handy. But yes. A 12-core system would be a waste - 2-4 cores are useable.

OS4User

  • Full Member
  • ***
  • Thank You
  • -Given: 5
  • -Receive: 4
  • Posts: 119
    • View Profile
Re: eCS install problems on a recent mainboard
« Reply #16 on: September 03, 2016, 09:08:13 am »

Code: [Select]
The system detected an internal processing error at
location ##1800:1b4e - 0003:1b4e
60000, 2008

10860673
Internal revision 14.105_SMP

It is not "Trap", it is "Panic" -  kernel recognized condition in which it is impossible to continue execution of its own code.

Shown Panic is called "DevHlp Usage Error" and in the original kernel it may happen for quite a few reasons.

At the same time, I saw a positive report about running OS/2 on similar mobo  - Asrock Q1900B-ITX (I do not know how important the letter "B" is) but with OS/4 kernel http://www.os2world.com/wiki/index.php/Phoenix_OS/4#OS.2F4_Tested_Hardware_List

I do not think that it was working because of using OS/4  - so keep trying to change you settings.

Mikkel Simonsen

  • Newbie
  • *
  • Thank You
  • -Given: 0
  • -Receive: 2
  • Posts: 12
    • View Profile
Re: eCS install problems on a recent mainboard
« Reply #17 on: September 27, 2016, 08:50:11 pm »
At the same time, I saw a positive report about running OS/2 on similar mobo  - Asrock Q1900B-ITX (I do not know how important the letter "B" is) but with OS/4 kernel http://www.os2world.com/wiki/index.php/Phoenix_OS/4#OS.2F4_Tested_Hardware_List

I do not think that it was working because of using OS/4  - so keep trying to change you settings.

This report does not mention USB, which is one of the problems I had left.

The J1900 mainboard is now in a FreeBSD system. The NIC works fine, using the same "re" driver that the Arca Noae MMRE driver is based on - which is slightly weird...

I bought two new mainboards to test - I hope to eventually find one that works completely :)

These two are ASRocks again - I have no particular like/dislike for ASRock, but my supplier Reichelt stocks a lot of them. They are the C70M1 and the QC5000M-ITX/PH. Both are based on AMD APUs, but the C70M1 is very cheap, and not very fast.

The good news first: The C70M1 system booted without any problems! Both the MMRE and the R8169 NIC drivers work. USB seems to work also. AHCI and ACPI works with default settings. The only thing not working (I haven't tested stuff like audio), is the PS/2 combo port. Connecting a keyboard works, but if you connect both keyboard and mouse neither work - but this is clearly a hardware problem, as the problem is also present in the setup utility.

The not so good news: The QC5000 system also boots (with both keyboard and mouse working), but networking is flaky. If you ping another local system, you get 15-20% dropped packets, and some take 2-3 seconds to get through. According to acpistat the systems runs in VW-mode, meaning nearly everything is on one shared (low) IRQ. I guess that could be the cause of the network drop-outs?
According to a changelog document on Arca Noae's website, the later ACPI drivers automatically switch to VW-mode, when they encounter a non-supported system configuration.

The C70 is about half as fast as the QC5000 (per core), so I would like to get the faster system running... The SysBench numbers (int/float) are 1911/129 for the C70 (dual core) and 6817/400 for the QC5000 (quad core) - so the QC5000 is about the same speed as the J1900.

Doug Bissett

  • Hero Member
  • *****
  • Thank You
  • -Given: 1
  • -Receive: 38
  • Posts: 928
    • View Profile
Re: eCS install problems on a recent mainboard
« Reply #18 on: September 27, 2016, 10:00:35 pm »
Quote
The not so good news: The QC5000 system also boots (with both keyboard and mouse working), but networking is flaky. If you ping another local system, you get 15-20% dropped packets, and some take 2-3 seconds to get through. According to acpistat the systems runs in VW-mode, meaning nearly everything is on one shared (low) IRQ. I guess that could be the cause of the network drop-outs?

IRQ usage, or VW mode, is not likely to cause network drop-out problems, although I have never seen a case where ACPI would drop back to VW mode.

I had a problem using a plug in PCI card, with a Realtek 8169 on it, when using the R8169 driver. It would work fine, for about a minute, then it got really slow. PING dropped packets, and if you looked at the data that was returned, the packets were returned out of order. The latest MMRE driver fixed the problem, for me.

Dave Yeo

  • Hero Member
  • *****
  • Thank You
  • -Given: 7
  • -Receive: 87
  • Posts: 1098
    • View Profile
Re: eCS install problems on a recent mainboard
« Reply #19 on: September 28, 2016, 04:30:21 am »
I had a Realtek PCI card that would work for a few minutes, getting slower and slower and then stop working. I moved it to a PCI slot with a dedicated IRQ and that fixed the problem.
My understanding of Realtek network cards/chips is that they use the CPU more then other network chips, which is one reason for their crappy performance.

Mikkel Simonsen

  • Newbie
  • *
  • Thank You
  • -Given: 0
  • -Receive: 2
  • Posts: 12
    • View Profile
Re: eCS install problems on a recent mainboard
« Reply #20 on: September 28, 2016, 03:52:14 pm »
My understanding of Realtek network cards/chips is that they use the CPU more then other network chips, which is one reason for their crappy performance.

This is a problem with some of them at least. This is also why I suspect sharing the IRQ with every other device in the system, could cause problems. If the interrupts are not serviced fast enough, some packets could get lost in the (probably too small) buffers.

Pete

  • Hero Member
  • *****
  • Thank You
  • -Given: 0
  • -Receive: 26
  • Posts: 529
    • View Profile
Re: eCS install problems on a recent mainboard
« Reply #21 on: September 30, 2016, 05:28:04 pm »
Hi All

@Mikkel

I'm running a Gigabyte 990FXA-UD3 mainboard which has an onboard Realtek RTL8168 using the current Arca Noae r8169.os2 driver which works fine. I have read somewhere, possibly within these forums, that success with the Realtek nics can depend on the Revision number; this nic is a 10ECh:8168h and is Revision 006 - only discovered by running the Windows Device Manager. I suspect the Revision number is related to some sort of change to the nic hardware.

It may be that I am lucky in that the nic shares IRQ10 with only 3 other devices.


@Doug

Prior to very recent releases of ACPI this mainboard actually required the /VW parameter on the acpi config.sys line to avoid hardware problems including the nic suddenly stopping working. The later versions of ACPI do not need the /VW parameter as the driver works out if the system needs to work in VW mode and switches to it if necessary - as with this mainboard. If I remember correctly VW mode limits the number of IRQs available to either equal to or less than the maximum number of IRQs that OS/2 can handle - I think I read somewhere that is 32. So, Yes, current ACPI drops back to VW mode if necessary.



Regards

Pete



Andreas Schnellbacher

  • Full Member
  • ***
  • Thank You
  • -Given: 1
  • -Receive: 14
  • Posts: 235
    • View Profile
Re: eCS install problems on a recent mainboard
« Reply #22 on: September 30, 2016, 06:45:41 pm »
So, Yes, current ACPI drops back to VW mode if necessary.
Unfortunately not for every hardware: My board requires /VW to work. I have a MSI A88X-G41 PC Mate.

Moreover, David's Realtek driver doesn't work here. Newer version won't get an IRQ assigned (see \ibmcom\lantran.log). One older version has good transfer rates, but connected extremely slow -- useless for web pages.

Doug Bissett

  • Hero Member
  • *****
  • Thank You
  • -Given: 1
  • -Receive: 38
  • Posts: 928
    • View Profile
Re: eCS install problems on a recent mainboard
« Reply #23 on: September 30, 2016, 11:10:13 pm »
Quote
I'm running a Gigabyte 990FXA-UD3 mainboard which has an onboard Realtek RTL8168 using the current Arca Noae r8169.os2 driver which works fine.

You should try the MMRE driver. It works with all of my Realtek gigabit NICS.

Quote
If I remember correctly VW mode limits the number of IRQs available to either equal to or less than the maximum number of IRQs that OS/2 can handle - I think I read somewhere that is 32.

I believe that VW just uses the lower 16 IRQs. whether that is what your main board needs, or not, I don't know.

Quote
Moreover, David's Realtek driver doesn't work here. Newer version won't get an IRQ assigned (see \ibmcom\lantran.log).

MMRE is from the BSD driver. It works for every Realtek gigabit NIC that I have  (6 of them). R8169 is from Linux (Red Hat, I think), and it won't work with two of them. I hope you submitted a problem report, so the problem can be fixed. It seems that MMRE will be used going forward.

FWIW, Realtek NICs are not very good, but motherboard builders like them because they are cheap. It has been my experience, that Intel NICs are the way to go for reliability, and performance.

Andreas Schnellbacher

  • Full Member
  • ***
  • Thank You
  • -Given: 1
  • -Receive: 14
  • Posts: 235
    • View Profile
Re: eCS install problems on a recent mainboard
« Reply #24 on: October 02, 2016, 08:48:40 pm »
MMRE is from the BSD driver. It works for every Realtek gigabit NIC that I have  (6 of them). R8169 is from Linux (Red Hat, I think), and it won't work with two of them. I hope you submitted a problem report, so the problem can be fixed. It seems that MMRE will be used going forward.
In the past David said that he doesn't own a Realtek device and that therefore it's not well supported. IIRC, he just made the driver compile with his toolkit and the newest Free BSD sources. See the bugtracker for details.

Doug Bissett

  • Hero Member
  • *****
  • Thank You
  • -Given: 1
  • -Receive: 38
  • Posts: 928
    • View Profile
Re: eCS install problems on a recent mainboard
« Reply #25 on: October 02, 2016, 10:45:49 pm »
Quote
In the past David said that he doesn't own a Realtek device and that therefore it's not well supported. IIRC, he just made the driver compile with his toolkit and the newest Free BSD sources. See the bugtracker for details.

I believe that David has a few Realtek devices now. The statement that he just compiled from the BSD driver source seems to be true for all of the drivers, not just the Realtek versions. I also use the E1000B driver, for Intel, and the MMLEM driver, for use with virtual machines (VBOX), and they work well too.

From what i understand, the sources for Linux have been incorporated into the Linux Kernel, so it has become very difficult to simply lift that code, and make it work for OS/2. The BSD sources are much more user friendly, so that will be where we get future NIC drivers.

Paul Smedley

  • Sr. Member
  • ****
  • Thank You
  • -Given: 11
  • -Receive: 46
  • Posts: 354
    • View Profile
Re: eCS install problems on a recent mainboard
« Reply #26 on: October 02, 2016, 10:49:10 pm »
From what i understand, the sources for Linux have been incorporated into the Linux Kernel, so it has become very difficult to simply lift that code, and make it work for OS/2. The BSD sources are much more user friendly, so that will be where we get future NIC drivers.

The same as the above happened with ALSA - makes it very difficult to support new sound devices on OS/2.

Cheers,

Paul

Pete

  • Hero Member
  • *****
  • Thank You
  • -Given: 0
  • -Receive: 26
  • Posts: 529
    • View Profile
Re: eCS install problems on a recent mainboard
« Reply #27 on: October 11, 2016, 03:23:47 pm »
Hi All

Following Dougs recommendation I decided to give the MMRE driver a try so Changed nic driver from R8169 to MMRE and rebooted.  After a short period of use the system locked up - no idea why - and required a press of the Reset button to restart. After the restart I had no nic connectivity. I tried another reboot - still no connectivity; driver shows as loaded but ping fails to contact any ip address.

To check the problem was not hardware failure I booted my Windows7 installation. No problem with the nic, all works fine.

I rebooted into eCS2.1+ and the nic worked fine and has since, 3 days, with no unexplained lockups.

I am wondering if booting Windows7 reset something in the hardware making it more "OS/2 friendly"? - Might be worth a try for those people with Realtek nics that load and do not work.


I see the latest ACPI supports 64 interrupts removing the need for switching to /VW mode which may help with problems mentioned in this discussion.


Regards

Pete

Doug Bissett

  • Hero Member
  • *****
  • Thank You
  • -Given: 1
  • -Receive: 38
  • Posts: 928
    • View Profile
Re: eCS install problems on a recent mainboard
« Reply #28 on: October 11, 2016, 07:47:50 pm »
Quote
Following Dougs recommendation I decided to give the MMRE driver a try so Changed nic driver from R8169 to MMRE and rebooted.  After a short period of use the system locked up - no idea why - and required a press of the Reset button to restart. After the restart I had no nic connectivity. I tried another reboot - still no connectivity; driver shows as loaded but ping fails to contact any ip address.

To check the problem was not hardware failure I booted my Windows7 installation. No problem with the nic, all works fine.

I rebooted into eCS2.1+ and the nic worked fine and has since, 3 days, with no unexplained lockups.

I am wondering if booting Windows7 reset something in the hardware making it more "OS/2 friendly"? - Might be worth a try for those people with Realtek nics that load and do not work.

Did you use the 1005 version., or the 1008 version. The 1005 version has a problem that caused a TRAP 0008 when it loaded, for me. Apparently all of the 1005 drivers have the same problem, so be sure to update if necessary.

Pete

  • Hero Member
  • *****
  • Thank You
  • -Given: 0
  • -Receive: 26
  • Posts: 529
    • View Profile
Re: eCS install problems on a recent mainboard
« Reply #29 on: October 12, 2016, 12:55:17 am »