Re: [LinuxBIOS] CONFIG_LB_MEM_TOPK (and a patch)

[Alan Mimms]

One side comment from a lurker on this list.  Your experience with the
error hang after your login prompt (prints message like the one below)

LinuxBIOS-1.1.8_s2881_Fallback Fri Apr  7 18:06:33 EDT 2006
starting...
  (0,1) link=00
  (1,0) link=00
  02 nodes initialized.
  SBLink=02
  NC node|link=02

reminded me of something I found and fixed a couple of weeks ago.

The board hangs when it does a warm reset on a bus error condition.
This is due to a typo in the code in
src/northbridge/amdk8/incoherent_ht.c in the function ht_setup_chainx
(base indentation removed for clarity):

if (ctrl & ((1 << 4) | (1 << 8))) {
                   /*
    * Either the link has failed, or we have
    * a CRC error.
    * Sometimes this can happen due to link
    * retrain, so lets knock it down and see
    * if its transient
    */
-->  ctrl |= ((1 << 6) | (1 <<8)); // Link fail + Crc
    pci_write_config16(udev, upos + LINK_CTRL(uoffs), ctrl);
    ctrl = pci_read_config16(udev, upos + LINK_CTRL(uoffs));
    if (ctrl & ((1 << 4) | (1 << 8))) {
    	print_err("Detected error on Hypertransport Link\n");
    	break;
    }
}

The line (marked with "-->" above) where the error bits to be cleared
are set to '1'
    ctrl |= ((1 << 6) | (1 <<8)); // Link fail + Crc
SHOULD be
    ctrl |= ((1 << 4) | (1 <<8)); // Link fail + Crc

The code should be setting the 1<<4 bit (so as to clear one of the error
bits) instead of 1<<6 (the read only bit that signals end of HT bus
chain).

After making this change, your board won't hang when the BIOS or Linux
detects a need to reboot due to some fatal error.

This is a pretty trivial change to an error case that hardly ever occurs
but was crucial for us.  I claim this change is safe and correct and
should be merged into the main line code.

Cheers, and thanks to all you who work so hard for all of the good and
useful LinuxBIOS code.  We have been very successful with it thus far in
our project.

Alan Mimms, Senior Architect
F5 Networks, Inc.  Spokane Development Center
1322 North Whitman Lane
Liberty Lake, Washington   99019
v: 509-343-3524   f: 509-343-3501

-----Original Message-----
From: [e-mail]
[mailto:[e-mail]] On Behalf Of Ward Vandewege
Sent: Friday, April 07, 2006 3:15 PM
To: Lu, Yinghai
Cc: [e-mail]
Subject: Re: [LinuxBIOS] CONFIG_LB_MEM_TOPK

On Fri, Apr 07, 2006 at 02:02:47PM -0700, Lu, Yinghai wrote:
> Can you make one elf image with mkelfImage ( combine the kernel and
>
> initrd)?

OK; this is what I've done:

1. Retrieved the mkelfImage program (v2.7) from

ftp://ftp.lnxi.com

2.

mkelfImage --command-line="ro root=/dev/md3 quiet splash console=tty0
console=ttyS0,115200n8" \
            --kernel="/boot/vmlinuz-2.6.12-9-amd64-generic" \
            --initrd="/boot/initrd.img-2.6.12-9-amd64-generic" \
            --output="/boot/linuxbios_2.6.12-9-amd64-generic.elf"

3. Adjusted etherboot-5.4.1/file/Config

AUTOBOOT_FILE = "hde1:/linuxbios_2.6.12-9-amd64-generic.elf"

4.

cd etherboot-5.4.1/src
 make bin/tg3--filo.elf

5. Adjusted targets/tyan/s2881/Config.lb to point to the new payload
(tg3--filo.elf)

6. Tried booting - it actually got to a login prompt this time, but as I
hit
enter on the kbd, it crashed again. Log is attached.

So that did not help. I'm going to try your next suggestion now.

Thanks,
Ward.
> use latest kernel + suse rescue disk to make on elf to check your
>
> /dev/md3...
>
> YH
>
> -----Original Message-----
> From: Ward Vandewege [mailto:[e-mail]] 
> Sent: Friday, April 07, 2006 1:55 PM
> To: Lu, Yinghai
> Cc: [e-mail]
> Subject: Re: [LinuxBIOS] CONFIG_LB_MEM_TOPK
>
> On Fri, Apr 07, 2006 at 01:37:06PM -0700, Lu, Yinghai wrote:
> > I fixed one merge typo in
> > src/southbride/amd/amd8111/amd8111_early_ctrl.c
> >
> > Please change 15 to 11 at line 13.
>
> OK; did that, but it did not help. Same problem still.
>
> Ward.
>
> !DSPAM:4436d470278011542430122!

[LinuxBIOS] CONFIG_LB_MEM_TOPK (and a patch)

[Ward Vandewege]

On Fri, Apr 07, 2006 at 04:00:28PM -0700, Alan Mimms wrote:
> The code should be setting the 1<<4 bit (so as to clear one of the error
> bits) instead of 1<<6 (the read only bit that signals end of HT bus
> chain).
>
> After making this change, your board won't hang when the BIOS or Linux
> detects a need to reboot due to some fatal error.
>
> This is a pretty trivial change to an error case that hardly ever occurs
> but was crucial for us.  I claim this change is safe and correct and
> should be merged into the main line code.

Yes, please merge that change!

Our box now keeps rebooting itself instead of crashing - have a look at the
attached boot log. Maybe this will be more of a clue as to what exactly is
happening? Anyone got anymore ideas?

Thanks, Alan!!
Ward.