[gPXE] wrong mac with rtl8139 nic on qemu-kvm

[Marty Connor]

Hi James,

Please do enjoy the rest of your Sunday.  I am watching some football 
while doing a bit of light hacking :)

I just instrumented rtl8139 a bit more.  When you have a moment would 
you try again with:

     http://etherboot.org/share/mdc/gpxe.git/contrib/rom-o-matic/

It should print out an indication of whether reset completed 
successfully.  This happens just before the NVS reading code executes.

If it isn't a timing issue with reset, we may have to look at the NVS 
code to look at each bit as it is read.

This is a very curious issue since rtl8139 is so widely used, I would 
have expected to have heard about this problem previously.

Can you please describe your booting environment?  Is this a regular PC 
BIOS or Coreboot+SeaBIOS.  Any details on your environment may help us 
to reproduce and address this problem.

Many thanks for your help with debugging this.

Marty

James wrote on 12/20/09 3:58 PM:
> Thanks Marty,
>
> I've generated a rom from your link but it hasn't helped
>
> gPXE 0.9.9+ -- Open Source Boot Firmware -- http://etherboot.org
> Features: FTP HTTP HTTPS DNS TFTP bzImage COMBOOT ELF Multiboot NBI PXE
> PXEXT
>
> net0: 00:00:d8:ab:e2:7e on PCI00:03.0 (open)
> [Link:up, TX:0 TXE:0 RX:0 RXE:0]
> DHCP (net0 00:00:d8:ab:e2:7e)..............
>
> Besides the features listed above I also selected
> CONSOLE_SERIAL
> COMPRESERVE
>
> I've gone ahead and also generated a rom without serial console options
> and have also seen it fail.
>
> I'm gonna have to sign off for today and get on with my Sunday.
> I'll follow-up with any other emails late tonight or Monday.
>
> Thanks,
> James
>
>
>
>
> On 20-Dec-09 11:47 Marty Connor wrote:
>> Hi James,
>>
>> I have patched my testing rom-o-matic.net instance with the patch I
>> suggested earlier. You can build with the patch from here:
>>
>> http://etherboot.org/share/mdc/gpxe.git/contrib/rom-o-matic/
>>
>> Thanks for your testing!
>>
>> Marty
>>
>> James wrote on 12/20/09 2:27 PM:
>>> Since I've been getting my roms from rom-o-matic it might
>>> take me a while to get around to trying to build my own.
>>> I'll look into doing this on Monday.
>>>
>>> Thanks,
>>> James
>>>
>>>
>>> On 20-Dec-09 11:09 Marty Connor wrote:
>>>> Hi James,
>>>>
>>>> I think this is a timing issue.
>>>>
>>>> Can you please find this line (around line 313):
>>>>
>>>> mdelay ( 10 );
>>>>
>>>> and change it to:
>>>>
>>>> mdelay ( 100 );
>>>>
>>>> If that fixes the problem I may suggest another patch that actually
>>>> tests that reset has completed rather than simply waiting the required
>>>> 10ms for the reset to complete.
>>>>
>>>> Please let us know know if this helps.
>>>>
>>>> Regards,
>>>>
>>>> Marty
>>>>
>>>> James wrote on 12/20/09 1:42 PM:
>>>>> Hello Shao,
>>>>>
>>>>> gPXE is loaded as an option rom in qemu-kvm. My qemu-kvm version is
>>>>> 0.9.1
>>>>> from RHEL 5.4.9.
>>>>>
>>>>>
>>>>> For first pass it means first DHCP discover after initial start or
>>>>> after
>>>>> a qemu system_reset in the monitor. It does appear to be random but
>>>>> it is
>>>>> happening more often than not. Here's a capture from the console.
>>>>>
>>>>> gPXE 0.9.9 -- Open Source Boot Firmware --http://etherboot.org
>>>>> Features: FTP HTTP HTTPS DNS TFTP bzImage COMBOOT ELF Multiboot NBI
>>>>> PXE PXEXT
>>>>>
>>>>> net0: 00:00:d8:ab:e2:7e on PCI00:03.0 (open)
>>>>> [Link:up, TX:0 TXE:0 RX:0 RXE:0]
>>>>> DHCP (net0 00:00:d8:ab:e2:7e)................ Connection timed out
>>>>> (0x4c106035)
>>>>> No more network devices
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> gPXE 0.9.9 -- Open Source Boot Firmware --http://etherboot.org
>>>>> Features: FTP HTTP HTTPS DNS TFTP bzImage COMBOOT ELF Multiboot NBI
>>>>> PXE PXEXT
>>>>>
>>>>> net0: 00:21:d8:ab:e2:7e on PCI00:03.0 (open)
>>>>> [Link:up, TX:0 TXE:0 RX:0 RXE:0]
>>>>> DHCP (net0 00:21:d8:ab:e2:7e).... ok
>>>>> net0: 2.43.241.156/255.255.255.240 gw 2.43.241.145
>>>>> Booting from filename "pxelinux.0"
>>>>> tftp://2.43.241.155/pxelinux.0.. ok
>>>>>
>>>>> James
>>>>>
>>>>>
>>>>>
>>>>> On 19-Dec-09 14:29 Shao Miller wrote:
>>>>>> James wrote:
>>>>>>> hi... i have a problem with gPXE 0.9.9 for rtl8139 10ec:8139 and
>>>>>>> qemu-kvm
>>>>>>>
>>>>>>> The first pass of the boot the device mac is sometimes incorrect...
>>>>>>> I see 00:00:d8:ab:e2:7e instead of 00:21:d8:ab:e2:7e.
>>>>>>> The second octet is 00 instead of 21 in the mac.
>>>>>>
>>>>>> How is gPXE booting? A floppy disk for the VM? A ROM for the VM?
>>>>>> Chained from another boot-loader in the VM? When you say "first
>>>>>> pass,"
>>>>>> do you mean that rebooting resolves the issue or that you perform a
>>>>>> DHCP
>>>>>> twice and it's been corrected by the second time? Is it random?
>>>>>>
>>>>>> - Shao Miller
>>>>> _______________________________________________
>>>>> gPXE mailing list
>>>>> gPXE at etherboot.org
>>>>> http://etherboot.org/mailman/listinfo/gpxe
>>
>>