@scabit, Hypex, damned

Argh!!! I had almost finished a large reply when this damn old Firefox version decided to commit suicide.

Well, I'll reply tomorrow...thanks for your patience.

_________________
A1SE: G3@600MHz, 2GB, 1GBit network card
A1XE: G4@933MHz, 2GB, refitted AC'97 codec
microA1: G3@800MHz, 1GB

- A1 Linux support -

@Geri

I hate that. Things are never the same when I retype them five or even ten minutes later. I'm just a slightly different person by then and my writing style isn't the same.

@olegil

apt-get upgrade libcairo for example shows 0 upgraded, 0newly installed, 0 to remove and 0 not upgraded.
Nothing happens.
I changed my sources.list file as shown by Geri above and ran an apt-get update.

Also, when I try to use aptitude to get some things like libcairo2, I get an error that I do not have a public key for deb-multimedia.org squeeze Release

How can I get the required public key?

Scott

_________________
AmigaOne uA1-c 512M RAM - Only Amiga Makes It Possible!
Check my blog AmigaOne Computing

@Hypex

Quote:

So with this card you haven't been able to boot Linux at all? In the event you can test any kernel with any distro you could connect devices through a USB1.1 hub to test the 6212 controller in USB1.1 mode. That should work at least.

From what I can recall, this card works just fine as USB 1.1 in amigaOS4 if I disable the EHCI kickstart module and reboot without it.

In Lenny, this card also worked as USB 1.1, but when I plug in any device that utilizes EHCI, it locks Lenny up solid. So things like USB mice work, but USB memory sticks lock it up.

In Squeeze, I have yet to be able to boot into squeeze from the onboard hard drive controller, so I cannot unplug my SIL0680 and plug this card in and boot into squeeze. (I only have a single PCI slot on my micro A1). Until I can get Squeeze to see my onboard squeeze partition and boot from it all by itself, I cannot even test this card with squeeze.

Oh, great info on the 6212 EHCI driver work arounds by the way.

Seems like byte 0x4b is the sleep timer byte that is wrongly set in earlier revisions of the 6212 chip to 1 us instead of 10 us, so causes task starvation. But I am not sure what registers 0x41 and 0x49 does exactly...but they too need to be changed to get EHCI to work on the 6212.
0x41 is changed in usb-uhci.c while 0x49 and 0x4b are changed in ehci-hcd.c. Interesting....

Scott

_________________
AmigaOne uA1-c 512M RAM - Only Amiga Makes It Possible!
Check my blog AmigaOne Computing

@hypex

Oh, found the spec sheet for the 6212L here: http://www.equinoxefr.org/wp-content/uploads/2008/05/datasheet.pdf

It shows that device specific ehci registers and the definitions of their bits are as follows:

0x49 miscellaneous control 5 default value 0x20

Offset 49 - Miscellaneous Control 5 (20h) ...................... RW
7 Enable MAC (provides more delay between
transactions)
0 Disable................................................... default
1 Enable
6 Enable MAC (provides timeout to device when
error is detected)
0 Disable................................................... default
1 Enable
->5 Clock Auto Stop
->0 Disable, no stop
->1 Enable, auto stop ................................... default
4-0 Reserved ........................................always reads 0


0x4B MAC turn around time parameter default value 09

Offset 4B - MAC turn around time parameter (09h).... RW
7-6 Reserved ........................................always reads 0
->5 EHCI sleep time select
->0 1 μs .................................................... default
->1 10 μs
4 Reserved ........................................always reads 0
3-0 USB 2.0 MAC TX time parameter .........default = 9

and in uhci:

0x41 miscellaneous control 2 default value 0x10

Offset 41 - Miscellaneous Control 2 (10h).......................RW
7-5 Reserved (Do Not Program) ................... default = 0
->4 Hold PCI REQ_ for successive access
->0 Disable
->1 Enable
3 Reserved (Do Not Program) ................... default = 0
2 Trap Option
0 Set trap 60/64 status bits without checking
enable bits ..............................................default
1 Set trap 60/64 status bits only when trap 60/64
enable bits are set
1 A20Gate Pass Through Option
0 Pass through A20Gate command sequence .def
1 Do not pass through write I/O port 64
0 Reserved (Do Not Program) ................... default = 0


the -> arrows indicate which bits are being changed by the VIA 6212L linux enhancements.....

Last edited by scabit on 14-Feb-2013 at 09:45 PM.

_________________
AmigaOne uA1-c 512M RAM - Only Amiga Makes It Possible!
Check my blog AmigaOne Computing

@Hypex

Quote:

Hypex wrote: @Geri Quote: These two packages are also installed on my A1SE, where keyboard and mouse work out of the box. I got it! I installed all three; xserver-xorg-input-evdev was in, xserver-xorg-input-kbd was missing and xserver-xorg-input-mouse couldn't install as it was referecned by another package.

Then I wonder why it works without the xserver-xorg-input-kbd package here!?

Quote:

I also started getting the locale errors. And tried a fsck. Which IIRC is directly missing from the rescue installler. I rebooted and after that it worked.

Maybe it can be installed within the installer by running "anna-install e2fsprogs-udeb"!? I didn't try it out however.

But good to hear that it is working!

Quote:

Quote: Correct me, if I'm wrong. But I think there's an "exit" command in the middle of the postinstall.sh script. You're correct Looks like I killed the menu features with my IDE args addition. Oops.

Well, otherwise I wouldn't have played around with the script.

Quote:

Quote: BTW: I finally found a way to correctly sort the list of paritions. Oh I forgot to test your last suggestion. I also found here somewhere another option: blkid That works well to set out volumes with Linux filesystems. Which I need as for some reason I can't mount any volumes in the installer without specifying filesystem type.

True, but it looks like its output still needs to be sorted.

Quote:

Quote: Can you try to regenerate the initramfs and then the boot image? This works. I had to install mkimage and initramfs-tools but I can regenerate it. Forgot how to create the UBoot image so just kept using my old one!

That's why the installer should generate the boot image. Otherwise it never gets regenerated, when updates are installed. But I planned to work on your postinstall.sh script over the weekend so that it generates the boot image on the fly. Stay tuned!

Quote:

I also tested 2.6.38.3+. Mouse frozen. For some reason I had installed no modules. Copied them and it worked.

Yes, the mouse driver is compiled as a module. The weird thing is that keyboard support has to be compiled into the kernel for the A1, even if other Debian kernels deliver it as a module.

Quote:

For whatever reason the 2.6.32 kernel isn't working and crashes after loading.

Boot image too big?

Quote:

Quote: Is there a Xorg executable in /usr/bin? There is now! I ended up trying to install gdm which wanted to replace gdm3. Took 11MB. Then tried to install gnome and gmome-desktop. Wants 400MB about download. Noooo!

Actually gdm3 is the default login manager for Squeeze.

Quote:

Hypex wrote: @Geri Quote: Am I right that we didn't found the root cause for this? :oops You're right. I booted Debian today. Left it loading while I went to the loo. Came back and there were a host of USB errors about unable to add hub this or another. For some reason Linux gets stuck on my front USB ports until I plug a usb stick in! Then I realised it hadn't booted and pressed return to find I was in the initramfs. At this point it was waitign for /dev to populate but gave up. Usually I can press return and it finds it to continnue. But I tried to use the more command and got stuck as I couldn't break put of it. I then reset, reloaded and this time /dev populated in time!

Do you suspect that USB could have to do something with this strange behavior?

Quote:

Quote: I'm not sure, if only U-Boot is the only culprit here. Yes, the SLB is where the bug occurs.

What has the SLB to do with a non-working keyboard?

Quote:

Quote: How does the rev.63 card behave under Linux? The same! It's broken. Now I don't know how the 65 worked at all. As in OS4 UHCI or OHCI works fine but EHCI doesn't. We need the 6212 work arounnds I think. Check this source: http://sup.xenya.si/sup/info/linksys/Sources/wrt54g3g/WRT54G3G_1_99_7_0912_EURO/release/src/linux/linux/drivers/usb/host/ehci-hcd.c Here is a neater closeup. Search for 6212 in source: https://dev.openwrt.org/browser/branches/buildroot-ng/openwrt/target/linux/generic-2.4/patches/228-more_usb_fixes.patch?rev=4325&order=name

Thanks for the info! I see that not all of these workarounds are in the official and Debian kernel sources, so we have something to play around with.

Quote:

I can tell them what registers to poke with what values but I suspect that won't help. Might have worked once with the VIA DMA patch but now now in software: "This works but we don't know why." All I can find are sources. I can't find any docs on the workaround except for a note on timer delay . But the fact is the VIA EHCI chips cause problems in the Linux world and in the Amiga world. Works fine in Windows on x86. Can work fine with a compatible PCI bridge.

I also found the datasheet for this VIA USB controller and its PCI config register layout looks (unsurprisingly) similar to the one of the A1 southbridge's USB controllers.

Quote:

BTW Debian includes the aliaes.conf IIRC that tells to load the EHCI driver first. But it still doesn't work. EHCI is still oaded last! I've read this can break a 6212 on Linux and also that sometimes the UHCI or OHCI driver also needs to be disabled otherwise EHCI will not work.

Let's see how this works out with the 2.6.32 kernel, because I don't see this problem here anymore.

@Hypex, scabit

Quote:

Hypex wrote: @scabit If it has the 6212 work arounds it should. But I don'y know if they are in there. Given I see no evidence in 2.6.38.3+ I suspect not. But all we need to do is give Geri the USB2 patch so he can recompile the module.

That shouldn't be necessary. We can also run some pci write commands in U-Boot.

Quote:

Quote: You guys are coming up with a 2.6.32 kernel for the AmigaOne sometime soon? Geri should have a package eventually. In the meantime I can rebuild my installer to use it with the latest Debian initrd and create the needed boot files.

Yes, the kernel package is already available, but I need to write the installation instructions resp. need to find a way to automate the installation somehow.

@damned

Quote:

damned wrote: @Geri i follow these istruction: http://amigaone-linux.sourceforge.net/kernel/2.6.38/INSTALL.txt so..is there any thing that i could try?

Yup, the INSTALL.txt only tells you half of the truth. I'll post updated installation instructions!

@scabit

Quote:

scabit wrote: @Geri Okay, after about a half hour of using google to look up such things as "download new debian libraries?" and "how do I upgrade debian libraries?" and getting nothing useful at all, I am led to ask...how exactly does one download and install a new library? apt-get does not recognize the libnss, libcairo or libpixman names and is for installing applications, not libraries. So what do I use to upgrade libraries?

It may be easier to resolve all dependency conflicts in aptitude. If you press enter over the iceweasel entry (or any other package name) in the list of packages, then aptitude will open a new tab with all dependencies of the package and the available package versions. From there you can select the 10.0.12 version (with the "+" key) and aptitude will mark all version conflicts in red. This way you can search for all libraries that need to be upgraded to a backports version (by going through the list of "Depends") until aptitude doesn't moan about broken package dependencies anymore.

Note that you can start from the beginning with "Cancel pending actions", if you selected a wrong package (version) and can't revert the selection anymore.


Quote:

scabit wrote: @olegil apt-get upgrade libcairo for example shows 0 upgraded, 0newly installed, 0 to remove and 0 not upgraded. Nothing happens. I changed my sources.list file as shown by Geri above and ran an apt-get update. Also, when I try to use aptitude to get some things like libcairo2, I get an error that I do not have a public key for deb-multimedia.org squeeze Release How can I get the required public key?

Install the "deb-multimedia-keyring" package. Afterwards aptitude will stay silent for any package installed from the deb-multimedia.org repository (however it will still warn you, when you install the keyring package).

_________________
A1SE: G3@600MHz, 2GB, 1GBit network card
A1XE: G4@933MHz, 2GB, refitted AC'97 codec
microA1: G3@800MHz, 1GB

- A1 Linux support -

@scabit

Quote:

scabit wrote: @hypex Oh, found the spec sheet for the 6212L here: http://www.equinoxefr.org/wp-content/uploads/2008/05/datasheet.pdf It shows that device specific ehci registers and the definitions of their bits are as follows: 0x49 miscellaneous control 5 default value 0x20 Offset 49 - Miscellaneous Control 5 (20h) ...................... RW 7 Enable MAC (provides more delay between transactions) 0 Disable................................................... default 1 Enable 6 Enable MAC (provides timeout to device when error is detected) 0 Disable................................................... default 1 Enable ->5 Clock Auto Stop ->0 Disable, no stop ->1 Enable, auto stop ................................... default 4-0 Reserved ........................................always reads 0

This workaround is not in the Debian or official Linux kernel source. So we could try to fix it up in U-Boot.

Quote:

0x4B MAC turn around time parameter default value 09 Offset 4B - MAC turn around time parameter (09h).... RW 7-6 Reserved ........................................always reads 0 ->5 EHCI sleep time select ->0 1 μs .................................................... default ->1 10 μs 4 Reserved ........................................always reads 0 3-0 USB 2.0 MAC TX time parameter .........default = 9

This is the only workaround in the Linux kernel.

Quote:

and in uhci: 0x41 miscellaneous control 2 default value 0x10 Offset 41 - Miscellaneous Control 2 (10h).......................RW 7-5 Reserved (Do Not Program) ................... default = 0 ->4 Hold PCI REQ_ for successive access ->0 Disable ->1 Enable 3 Reserved (Do Not Program) ................... default = 0 2 Trap Option 0 Set trap 60/64 status bits without checking enable bits ..............................................default 1 Set trap 60/64 status bits only when trap 60/64 enable bits are set 1 A20Gate Pass Through Option 0 Pass through A20Gate command sequence .def 1 Do not pass through write I/O port 64 0 Reserved (Do Not Program) ................... default = 0

AFAICT also this workaround is not implemented in the official Linux kernel source.

IIRC I tried to disable "Hold PCI REQ_ for successive access" for the onboard USB controllers on my A1XE a long time ago, but it didn't have any visible effect - probably because the southbridge already has priority over any other PCI master for A1s with the DMA fix. But now it would be good to know what it does on the A1SE, which doesn't have any DMA fix...

Note that this bit is also defined for the EHCI function of the VT6212!

@scabit

So how do we get your Linux system running on the onboard IDE controller in order to run some tests with these workarounds enabled?

_________________
A1SE: G3@600MHz, 2GB, 1GBit network card
A1XE: G4@933MHz, 2GB, refitted AC'97 codec
microA1: G3@800MHz, 1GB

- A1 Linux support -

@Geri

About Iceweasel:

I did an install of "deb-multimedia-keyring - that went okay
Did an apt-get update - that went okay
Did an aptitude update - that went okay
If I do an apt-get install iceweasel it installs version 3.5.16-20
If i run aptitude and select web/main/iceweasel and scroll all the way down, I see iceweasel 10.0.12esr~1^bpo60+1
I press + and it turns red, changes to "pB" and highlights 4 other packages. I move to select each of the other red highlighted packages and press + on each one, then press "g" and aptitude downloads and installs all the packages and they all also deletes iceweasel (my current version). But it does NOT install any other version.
I have tried numerous things to install the 10.0.12 version but nothings works.

What obvious thing am I missing?


and yes...until we can get my Squeeze partition on my internal drive to boot (be recognized by uboot!) by itself, we can;t test my SYBA usb/sata card.

Any ideas on that ?

Scott

_________________
AmigaOne uA1-c 512M RAM - Only Amiga Makes It Possible!
Check my blog AmigaOne Computing

@scabit

Quote:

scabit wrote: @Geri About Iceweasel: I did an install of "deb-multimedia-keyring - that went okay Did an apt-get update - that went okay Did an aptitude update - that went okay If I do an apt-get install iceweasel it installs version 3.5.16-20

That's because "backports" packages have a lower priority their "stable" counterparts.

Quote:

If i run aptitude and select web/main/iceweasel and scroll all the way down, I see iceweasel 10.0.12esr~1^bpo60+1 I press + and it turns red, changes to "pB" and highlights 4 other packages. I move to select each of the other red highlighted packages and press + on each one, then press "g" and aptitude downloads and installs all the packages and they all also deletes iceweasel (my current version). But it does NOT install any other version. I have tried numerous things to install the 10.0.12 version but nothings works.

If you look at the bottom, then you will see another red line, where aptitude offers some package changes to resolve the dependency errors (with the "e" key). Unfortunately most of aptitude's solutions involve the removal of a lot of packages, when you try to install a package's backports version. Thus it's better to resolve the dependencies manually until the bottom red line disappears. I think I had to install the following packages:

- iceweasel: 10.0.12esr-1~bpo60+1
- xulrunner: 10.0.12esr-1~bpo60+1
- libnspr4: 2:4.9.2-1~bpo60+1
- libnspr4-0d: 2:4.9.2-1~bpo60+1
- libsqlite3-0: 3.7.13-1~bpo60+1
- libcairo2: 1.10.2-7~bpo60+1
- libmozjs10d: 10.0.12esr-1~bpo60+1
- libnss3: 2:3.13.6-2~bpo60+1
- libnss3-1d: 2:3.13.6-2~bpo60+1
- libpixman-1-0: 0.24.0-1~bpo60+1
- libvpx1: 1.0.0-2~bpo60+1

Quote:

and yes...until we can get my Squeeze partition on my internal drive to boot (be recognized by uboot!) by itself, we can;t test my SYBA usb/sata card Any ideas on that ?

Not really. I have the same problem on my A1SE, where I have to boot Linux over TFTP.

_________________
A1SE: G3@600MHz, 2GB, 1GBit network card
A1XE: G4@933MHz, 2GB, refitted AC'97 codec
microA1: G3@800MHz, 1GB

- A1 Linux support -

@Geri

Okay, got it all worked out. I didn;t realize how many levels deep you need to go to be able to get things resolved. However, as I type this in the new 10.0.12 version of Iceweasel I notice that my on screen typing is NOT lagging behind. The slowdown problem is fixed!
Hurray!
This version also seems less prone to simply vanish when I scroll too fast, etc.
Thanks for walking me through that. Not sure if I could figure it out again....but I'm slowly learning how this stuff works.

Back to the issue: what is required for uboot to recognize and present the options from the /boot/a1boot.conf file during AmigaOne bootup?

1) the linux partition with the a1boot.conf MUST be on the drive booted from (check)
2) the booted drive must have a valid SLB2, in my case 1.19 IIRC that supports linux (check)
3) The a1boot.conf file must have valid entries (I am using the same a1boot.conf that works on my lenny partition, so I know it works already) (check)
4) THe /boot/a1boot.conf file must be on a bootable linux partition - i.e. EXT3 (check)

What else? There is something else required, because I meet all of these conditions and DO NOT see any linux boot menu options available when booting from my internal IDE port on the microA1.

Here's a question - even though I unplug the DRIVES from the SIL0680 card, do I also have to unplug the SIL0680 itself for the internal partition to show the configurations in my a1boot.conf file? That would be crazy....but I guess crazier things have happened. I am setting my boot1 uboot variable to IDE, the internal drive, and OS4.1 boots properly from that internal drive. Why do the linux partitions listed in a1boot.conf on that same drive NOT appear?

More thoughts on this?

Thanks!

SCott

Last edited by scabit on 17-Feb-2013 at 02:28 PM.

_________________
AmigaOne uA1-c 512M RAM - Only Amiga Makes It Possible!
Check my blog AmigaOne Computing

@scabit

Quote:

scabit wrote: @Geri Okay, got it all worked out. I didn;t realize how many levels deep you need to go to be able to get things resolved. However, as I type this in the new 10.0.12 version of Iceweasel I notice that my on screen typing is NOT lagging behind. The slowdown problem is fixed! Hurray!

Quote:

This version also seems less prone to simply vanish when I scroll too fast, etc.

I know this problem, as you can read above.

Quote:

Thanks for walking me through that. Not sure if I could figure it out again....but I'm slowly learning how this stuff works.

Your welcome.

Quote:

Back to the issue: what is required for uboot to recognize and present the options from the /boot/a1boot.conf file during AmigaOne bootup? 1) the linux partition with the a1boot.conf MUST be on the drive booted from (check) 2) the booted drive must have a valid SLB2, in my case 1.19 IIRC that supports linux (check) 3) The a1boot.conf file must have valid entries (I am using the same a1boot.conf that works on my lenny partition, so I know it works already) (check) 4) THe /boot/a1boot.conf file must be on a bootable linux partition - i.e. EXT3 (check) What else? There is something else required, because I meet all of these conditions and DO NOT see any linux boot menu options available when booting from my internal IDE port on the microA1.

Good question. I'll tried out several setups:
- moved the Linux partition directly after the OS4 partition (the same setup as I have on my A1XE),
- used an EXT2 filesystem for the boot partition (it works with EXT3 on my A1XE)
- set the exact block size in the RDB as reported by "tune2fs -l partition"
- tried it with an older version of the slb (1.16) and
- played around with the boot priorities (I use -127 as boot priority for both the OS4 and Linux boot partition on my A1XE).

Nothing worked so far on my A1SE.

Quote:

Here's a question - even though I unplug the DRIVES from the SIL0680 card, do I also have to unplug the SIL0680 itself for the internal partition to show the configurations in my a1boot.conf file? That would be crazy....but I guess crazier things have happened. I am setting my boot1 uboot variable to IDE, the internal drive, and OS4.1 boots properly from that internal drive. Why do the linux partitions listed in a1boot.conf on that same drive NOT appear?

I also have an unused Sii3114 controller in my A1XE and the slb correctly detects the Linux boot partition on the drive connected to the onboard controller. I guess we can rule this out as the root cause.

_________________
A1SE: G3@600MHz, 2GB, 1GBit network card
A1XE: G4@933MHz, 2GB, refitted AC'97 codec
microA1: G3@800MHz, 1GB

- A1 Linux support -

@scabit

Quote:

From what I can recall, this card works just fine as USB 1.1 in amigaOS4 if I disable the EHCI kickstart module and reboot without it.

Yep, same here!

Quote:

Oh, great info on the 6212 EHCI driver work arounds by the way.

Yeah, I did a bit of surfing on the web.

Quote:

0x41 is changed in usb-uhci.c while 0x49 and 0x4b are changed in ehci-hcd.c. Interesting....

I've done some tests in Linux 2.6.31-1. Looks like 0x4B was set correctly. And also 0x49 IIRC. In any case I used a pci modifier to ensure the registers were correct including the UHCI one. It made no change so I wonder if revision 0x63 can work now. The drive light flashing, stopping then the same process repeated, with read errors in the kernel log. Of course I was modifying things outside the driver code sop that may affect it.

I also did the same in OS4 to no change. I even ran just the EHCI driver. Still get the ongoing flashing and read errors reported in the USB log.

@Hypex, scabit

Quote:

Hypex wrote: @scabit Quote: 0x41 is changed in usb-uhci.c while 0x49 and 0x4b are changed in ehci-hcd.c. Interesting.... I've done some tests in Linux 2.6.31-1. Looks like 0x4B was set correctly. And also 0x49 IIRC.

It's likely that 0x49 was already set correctly on power-up.

Quote:

In any case I used a pci modifier to ensure the registers were correct including the UHCI one. It made no change so I wonder if revision 0x63 can work now. The drive light flashing, stopping then the same process repeated, with read errors in the kernel log. Of course I was modifying things outside the driver code sop that may affect it.[quote] You just modified the PCI config space. The driver usually accesses the device's PCI I/O or MMIO address space for the real USB functionality. [quote]I also did the same in OS4 to no change. I even ran just the EHCI driver. Still get the ongoing flashing and read errors reported in the USB log.

There are some other interesting PCI config bits. Maybe one of you could do a dump of the complete PCI config space (with lspci -xxx -s bus:device.function under Linux for example) for the UHCI and EHCI controllers of the VT6212 (best with the uhci/ehci_hcd modules blacklisted or with "nousb" in the kernel options, so that we know which bits the kernel really modifies).

_________________
A1SE: G3@600MHz, 2GB, 1GBit network card
A1XE: G4@933MHz, 2GB, refitted AC'97 codec
microA1: G3@800MHz, 1GB

- A1 Linux support -

@Geri

Quote:

Then I wonder why it works without the xserver-xorg-input-kbd package here!?

Perhaps it was a driver issue. I had missed some files like depmod in /boot so that mayve affected it.

Quote:

Maybe it can be installed within the installer by running "anna-install e2fsprogs-udeb"!? I didn't try it out however.

Possibly. I found fsck-ext2 worksed.

Quote:

But good to hear that it is working!

Yes. Like that Root Terminal menu option. But wish they had all resolutions for LCD to pick from. Can't be bothered finding out what you told me to change which works in Ubuntu.

Quote:

True, but it looks like its output still needs to be sorted.

IIRC the drives were in order here. Sorting is nice yes, but unsorted is okay if it finds the boot volume.

Quote:

That's why the installer should generate the boot image.

Should be easy enough as I can put the tools into the installer. Or even download them to the target.

Quote:

But I planned to work on your postinstall.sh script

Oh no! It's going to branch off and split down the middle!

Knew an open source script should been closed up so no one could see.

Quote:

The weird thing is that keyboard support has to be compiled into the kernel for the A1, even if other Debian kernels deliver it as a module.

In all my cases keyboard as well as mouse was dead.

Quote:

Boot image too big?

No that's fine. I went back and checked. It wasn't exaclty a crash. For some reason it can't mount root and panics. Root fs on unknown block. Set from label.

Quote:

Actually gdm3 is the default login manager for Squeeze.

Then gdm should have resolved to gdm3 or installe dthat I think but for some reason doesn't.

Quote:

Do you suspect that USB could have to do something with this strange behavior?

I have wondered that. The USB is acting strangely. But it is always able to boot even spewing out annoying messages. They just clutter up the kernel log until XOrg opens. And then secretly clutter up the on disk logs.

Quote:

What has the SLB to do with a non-working keyboard?

Because I have only noticed keys going dead within the SLB when trying to pick a selection!

Perhaps it is the UBoot code that is at fault. Don't remember if fault showed up in UBoot menu. Nor if a corruped keyboard fault appears in UBoot CLI.

Quote:

Thanks for the info! I see that not all of these workarounds are in the official and Debian kernel sources

I checked in the kernel source I compiled on the X1000 and it looks like they have been moved to pci-quirks.c I think it was in the usb folder. Also, given that 0x4b is set to 0x29 here in Linux they still could be activated.

Quote:

Let's see how this works out with the 2.6.32 kernel, because I don't see this problem here anymore.

Got also a VIA 6212 to check?

Quote:

We can also run some pci write commands in U-Boot.

A driver is the proper way. But I did some tests in UBoot. I don't know if it can be used to modify the registers. I dumped the config space and it doesn't look right. UBoot converts it from little endian to big endian so it is readable. But the 0x49 and 0x4B registers aren't right! IIRC one was 0x00 and the other 0xA0. When they should be 0x20 and 0x09 as default.

Quote:

Yes, the kernel package is already available

I could just stick your install instructions into a script.

@Geri

Quote:

It's likely that 0x49 was already set correctly on power-up.

Okay just checked again. And 0x49 is at default 0x20. 0x4b is set to 0x29.

Quote:

You just modified the PCI config space. The driver usually accesses the device's PCI I/O or MMIO address space for the real USB functionality.

Yes I did only the config as that's all I know! And the only thing I noticed to be changed.

Quote:

There are some other interesting PCI config bits. Maybe one of you could do a dump of the complete PCI config space

Well for starters I disabled USB in OS4, shut down and powered on. The result was from idetool:

----------Vendor 1106 / Device 3104 ConfigSpace Dump -------
00 01 02 03 04 05 06 07 08 09 0A 0B 0C 0D 0E 0F
+ -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
00 | 06 11 04 31 06 00 10 02 63 20 03 0C 08 80 80 00
10 | 00 02 10 98 00 00 00 00 00 00 00 00 00 00 00 00
20 | 00 00 00 00 00 00 00 00 00 00 00 00 06 11 04 31
30 | 00 00 00 00 80 00 00 00 00 00 00 00 07 03 00 00
40 | 00 00 0B 00 00 00 00 00 A0 20 00 09 00 00 FF FF
50 | 00 5A 00 80 00 00 00 00 04 0B 88 88 33 00 00 00
60 | 20 20 01 00 00 00 00 00 01 00 00 00 00 00 00 C0
70 | 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
80 | 01 00 C2 FF 00 00 00 00 00 00 00 00 00 00 00 00
90 | 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
A0 | 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
B0 | 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
C0 | 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
D0 | 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
E0 | 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
F0 | 00 00 00 00 00 00 00 03 00 00 00 00 00 00 00 00

And from Linux after bootup:

root@ubuntu:/home/damien# lspci -d 1106:3104 -xxx
00:09.2 USB Controller: VIA Technologies, Inc. USB 2.0 (rev 63)
00: 06 11 04 31 16 00 10 02 63 20 03 0c 08 80 80 00
10: 00 02 10 98 00 00 00 00 00 00 00 00 00 00 00 00
20: 00 00 00 00 00 00 00 00 00 00 00 00 06 11 04 31
30: 00 00 00 00 80 00 00 00 00 00 00 00 07 03 00 00
40: 00 00 0b 00 00 00 00 00 a0 20 00 29 00 00 ff ff
50: 00 5a 00 80 00 00 00 00 04 0b 88 88 33 00 00 00
60: 20 20 01 00 00 00 00 00 01 00 00 00 00 00 08 c0
70: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
80: 01 00 c2 ff 00 00 00 00 00 00 00 00 00 00 00 00
90: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
a0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
b0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
c0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
d0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
e0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
f0: 00 00 00 00 00 00 00 03 00 00 00 00 00 00 00 00

If we really need UHCI I can grab that later. BTW in the kernel log I see all these VIA quirks applied. Expecially things like via_vlink. It appplies a lot to the devices.

Hey here's an interesting thread for some amusement. It seems that we are not alone and even x86 Windows users are affected also unlike I previously thought. So it's not just Linux that is affected.

http://www.viaarena.com/forums/showthread.php?t=32588

@scabit

Quote:

1) the linux partition with the a1boot.conf MUST be on the drive booted from (check) 2) the booted drive must have a valid SLB 3) The a1boot.conf file must have valid entries (I am using the same a1boot.conf that works on my lenny partition, so I know it works already) (check) 4) THe /boot/a1boot.conf file must be on a bootable linux partition - i.e. EXT3 (check)

5) Create the linux partitions from the os4 side (otherwise the a1boot.conf don't work)

_________________
--------------------------------------
Search and Destroy
---------------------------------------

@damned

Good point. I womder how it differs if OS4 can still see it?

What else I wonder is if the permissions matter. Does one need to do a chmod 777 on the menu file? Given SLB uses UBoot routine to read Ext2 AFAIK it may not read a root only file. Of course that sounds silly but the situation is getting desperate here.

But I think I change permission cause I got sick of not being able to edit a file on the desktop. I hate that in Linux. You can't edit some file as you don't have permission and it doesn't even ask you for a password to unlock it! I OWN the file damn it! It annoys me more on other volumes I mount but can't freely copy/to with. Restricting root access is all very good but with these reasons alone Linux forces you to becoem super user just to copy damn files to your own volumes! Argh!

I edit my menu file in OS4 so that may have modified something. Also my filesystem is Ext3 where as the SLB/UBoot must read it as Ext2. If that makes any difference.

Last edited by Hypex on 20-Feb-2013 at 03:09 PM.

@Hypex

I don't knok why but it's like this..some years ago i used hdtoolbox to make the os4 partitions and gparted for the linux ones...i went crazy before i discover that particular..so i maked ad image of linux partition with partimage, recreate the partitions with hdtoolbox, reverse the linux image and finally...a1boot.conf worked..

_________________
--------------------------------------
Search and Destroy
---------------------------------------

@Hypex

Quote:

Hypex wrote: @Geri Perhaps it was a driver issue. I had missed some files like depmod in /boot so that mayve affected it.

Okay.

Quote:

Yes. Like that Root Terminal menu option. But wish they had all resolutions for LCD to pick from. Can't be bothered finding out what you told me to change which works in Ubuntu.

Well, if it has something todo with X11, then it's likely in /etc/X11/xorg.conf[.d/].

Quote:

IIRC the drives were in order here. Sorting is nice yes, but unsorted is okay if it finds the boot volume.

Not here. But the script anyway has to check for an existing a1boot.conf, so sorting the output may not be that important. It just has to make sure that it finds the right a1boot.conf file.

Quote:

Should be easy enough as I can put the tools into the installer. Or even download them to the target.

The script also has to install uboot-mkimage. The initramfs-tools package should always be installed, IIRC.

Quote:

Quote: But I planned to work on your postinstall.sh script Oh no! It's going to branch off and split down the middle! Knew an open source script should been closed up so no one could see.

Setup a GIT repository for distributed development!

Anyway, I wanted to try out the following:
1. copy all installation files to the /target directory (kernel package, postinstall.sh script or what's needed else)
2. bind-mount /dev, /proc, /sys on their /target/ equivalents
3. chroot into /target by executing a sub-postinstall.sh script. This way we have access to all the commands we need!
4. directly use dpkg to install the kernel package
5. install uboot-mkimage using apt-get
6. run update-initramfs -u -k "kernelversion"
7. run mkimage to create the final boot image
8. assemble the command line
9. search for an a1boot.conf script or generate a local one
10. exit the sub-postinstall.sh script and thus exit the chroot environment
11. bind-unmount the /dev, /proc, /sys directories
12. done!

Furthermore I would like to create a Debian package that handles items 7, 8 and 9 automatically on every update (I'm working on it...)

Quote:

Quote: The weird thing is that keyboard support has to be compiled into the kernel for the A1, even if other Debian kernels deliver it as a module. In all my cases keyboard as well as mouse was dead.

But hopefully not on a virtual console!?

Quote:

Quote: Boot image too big? No that's fine. I went back and checked. It wasn't exaclty a crash. For some reason it can't mount root and panics. Root fs on unknown block. Set from label.

A complete boot log could be helpful here. However I wonder why the slb_v2 can boot a full initramfs image with the 2.6.32 kernel on your system, but not on my machine!? Maybe I should upgrade to the latest slb_v2 version on my XE, if this is really the case.

Quote:

Quote: Actually gdm3 is the default login manager for Squeeze. Then gdm should have resolved to gdm3 or installed that I think but for some reason doesn't.

But you installed gdm manually, right? If so, you picked the wrong one, as a new Squeeze installation comes up with gdm3.

Quote:

Quote: Do you suspect that USB could have to do something with this strange behavior? I have wondered that. The USB is acting strangely. But it is always able to boot even spewing out annoying messages. They just clutter up the kernel log until XOrg opens. And then secretly clutter up the on disk logs.

Did you ever try, if the boot fails when you completely disable USB (with the "nousb" options").

Quote:

Quote: What has the SLB to do with a non-working keyboard? Because I have only noticed keys going dead within the SLB when trying to pick a selection!

Okay, I didn't stumble over this problem yet. But I suppose this only happens by accident in the SLB menu. My understand is that the SLB can rely on U-Boot to correctly initialize the keyboard for it.

Quote:

Perhaps it is the UBoot code that is at fault. Don't remember if fault showed up in UBoot menu. Nor if a corruped keyboard fault appears in UBoot CLI.

BTW: Are you using a USB keyboard? Maybe that makes the difference here, as I'm still using PS2 keyboards on my A1s. I don't think USB support is well tested in old U-Boot versions...

Quote:

Quote: Thanks for the info! I see that not all of these workarounds are in the official and Debian kernel sources I checked in the kernel source I compiled on the X1000 and it looks like they have been moved to pci-quirks.c I think it was in the usb folder. Also, given that 0x4b is set to 0x29 here in Linux they still could be activated.

I only see the workaround for register 0x4b in the Debian kernel sources (usb/host/ehci-pci.c), which seems to be working (0x4b is set to 0x29). But there's no VIA USB related workaround in usb/host/pci-quirks.c nor in pci/quirks.c (expect for a VIA CX700 northbridge quirk).

Quote:

Quote: Let's see how this works out with the 2.6.32 kernel, because I don't see this problem here anymore. Got also a VIA 6212 to check?

I wish I had one. Would also be helpful to understand the onboard USB controllers.

On the other side: where should I put it in? My A1XE is already fully loaded and for my A1SE PCI cards only work correctly on PCI bus 1, which is also fully loaded.

Quote:

A driver is the proper way. But I did some tests in UBoot. I don't know if it can be used to modify the registers.

To be honest, I hate to clutter up the kernel code with workarounds for buggy hardware, when U-Boot could fix the problem for both Linux and OS4. But I know that a new U-Boot version will not materialize any time soon.

However you can use U-Boot to modify PCI config registers! "pci write.[b|w|l] bus.device.function address value" is the right command for this.

Quote:

Quote: Yes, the kernel package is already available I could just stick your install instructions into a script.

True, but that's not enough to get the image regenerated on an upgrade.

Last edited by Geri on 21-Feb-2013 at 08:09 PM.

_________________
A1SE: G3@600MHz, 2GB, 1GBit network card
A1XE: G4@933MHz, 2GB, refitted AC'97 codec
microA1: G3@800MHz, 1GB

- A1 Linux support -

@Hypex

Quote:

Hypex wrote: @Geri Quote: There are some other interesting PCI config bits. Maybe one of you could do a dump of the complete PCI config space Well for starters I disabled USB in OS4, shut down and powered on. The result was from idetool: ...

I would play with bit 7 in offset 40 (PCI Memory Command Options), bit 4 in offset 41 (Hold PCI_REQ# for successive access - as suggested by scabit) and bits 5 to 7 in offset 49 (bit 5 is also modified in one of these inofficial kernel sources).

So for example you could run combinations of the following U-Boot command before booting Linux (replace xx with the device number, function .02 should be the EHCI controller):

Quote:

pci write.b 00.xx.02 40 80 # Only support memory read and memory write commands pci write.b 00.xx.[00|01|02] 41 [00|10] # Disable/enable hold of PCI_REQ# on successive access. You could actually try to disable it for UHCI and enable it for EHCI. I wonder, if that makes a difference. pci write.b 00.xx.02 49 [00|40|80|C0] # Enable MACs and disable stop clock.

There were also some EHCI related patches in the latest Linux kernel that should fix long oustanding bugs. However I don't know, if they can be easily backported. Eventually I would have to compile a 3.8 Linux kernel for testing.

Quote:

If we really need UHCI I can grab that later. BTW in the kernel log I see all these VIA quirks applied. Expecially things like via_vlink. It appplies a lot to the devices.

These quirks are only for interrupt routing.

Quote:

Hey here's an interesting thread for some amusement. It seems that we are not alone and even x86 Windows users are affected also unlike I previously thought. So it's not just Linux that is affected. http://www.viaarena.com/forums/showthread.php?t=32588

But no solution.

_________________
A1SE: G3@600MHz, 2GB, 1GBit network card
A1XE: G4@933MHz, 2GB, refitted AC'97 codec
microA1: G3@800MHz, 1GB

- A1 Linux support -