[syslinux] PXE boot error on a thin client

[Gene Cumm]

On Wed, Dec 17, 2014 at 5:54 AM, Willem Bargeman <willem at bargeman.eu> wrote:
> On Wed, Dec 17, 2014 at 2:55 AM, Gene Cumm <gene.cumm at gmail.com> wrote:
>> On Tue, Dec 16, 2014 at 4:22 PM, Geert Stappers <stappers at stappers.nl> wrote:
>>> On Tue, Dec 16, 2014 at 08:17:32PM +0100, Willem Bargeman wrote:
>>>> >> Hi,
>>>> >>
>>>> >> We have some problems while network booting thin clients. The type is T5540.
>>>> >>
>>>> >> At this moment we are using pxe linux 6.03.
>>>> >>
>>>> >> The systems are booting correctly with pxe linux version 6.03-pre14.
>>>> >> We have seen the same error with version 6.03-pre9.
>>>> >>
>>>> >> The error message:
>>>> >>
>>>> >> Failed to load ldlinux.c32
>>>> >> Boot failed: press a key to retry, or wait for reset...
>>>> >>
>>>> >> Any idea?
>>>> >>
>>>> >>
>>>> >
>>>> > _ When you update the version of pxelinux.0, are you also updating the
>>>> > version of ldlinux.c32?
>>>>
>>>> Yes, I've updated both files.
>>>
>>>> > _ The version of all the c32 files in use shall match the version of
>>>> > the boot loader.
>>>>
>>>> Yes, I've copied all the files which we are using. The version of all
>>>> the files is the same.
>>>
>>>> > _ When you mention the error message, "Failed to load...", is always
>>>> > the same file being mentioned? Or are you seeing instead other files
>>>> > mentioned in the message in same cases?
>>>>
>>>> It's always the same file in the error message.
>>>
>>>> > _ Have you tested with the official upstream pre-compiled binaries from
>>>> > version 6.03 final downloaded from kernel.org (as oppose to using your
>>>> > own compiled binaries, or some package from some distro)?
>>>>
>>>> Yes, I've used the official binaries. Not some own compiled binaries.
>>>
>>>> > _ Are the same binaries working correctly in some other hardware while
>>>> > failing in specific models only?
>>>>
>>>> At this moment we have issues with one specific hardware model. It's a
>>>> HP T5540 Thin client. We have many other hardware that is working
>>>> great with this version.
>>>>
>>>> Version 6.03-pre9 > not working
>>>> Version 6.03-pre14 > working
>>>> Version 6.03 final > not working
>>>
>>>
>>>> > There are additional potential issues / questions, but let's start with
>>>> > those basic ones.
>>>> >
>>>> > Regards,
>>>> > Ady.
>>>>
>>>> Hi,
>>>>
>>>> The answers again and now in 'normal' order.
>>>
>>> Thank you.  Please use next time "quote characters"
>>>
>>>
>>> So now a summary:
>>> * Error only with  one specific hardware model, a HP T5540 Thin client.
>>> * Version 6.03-pre9 > not working
>>> * Version 6.03-pre14 > working
>>> * Version 6.03 final > not working
>>> * Error message is
>>>      Failed to load ldlinux.c32
>>>      Boot failed: press a key to retry, or wait for reset...
>>>
>>>
>>> What is in the logging of the (TFTP) server?
>>> Do you see the request for 'ldlinux.c32'?
>>
>> Though a seemingly unusual version set, this sounds a lot like some of
>> the deaf clients I've seen on lpxelinux.0 (which I haven't seen with
>> pxelinux.0).  Are you using lpxelinux.0 or pxelinux.0 (regardless of
>> how you may have renamed it)?  lpxelinux.0 will show "lwIP" in the
>> banner while pxelinux.0 will show "PXE", indicating what network stack
>> is compiled into the binary.
>>
>> Did you try any of the prebuilt versions between 6.03-pre9 and
>> 6.03-pre14 or 6.03-pre14 and 6.03 to narrow down within which pair of
>> prereleases it starts working then stops working?  The deaf clients
>> would send the request, the tftpd sends the reply but the client acts
>> as if it was never received.
>>
>> --
>> -Gene
>
> Hi,
>
> I've collected more information.
>
> * USB boot is working with the 6.03 final release
> * we have tried multiple 6.03 pre releases. Below the results
>       pre9 - pre11        not working
>       pre12                  working
>       pre13                  not working
>       pre14 - pre16      working
>       pre17 - final         not working
> * I've created a pcap file while booting the HP T5540 client. The tftp
> server will deliver the pxelinux.0 file. After this tftp transfer I
> will see a error in the pcap file.
>        Unknown (0xcc32) [ETHERNET FRAME CHECK SEQUENCE INCORRECT]
>   I see also a TFTP unknow opcode. 52274. This is the last message in
> the TFTP transfer.
> * We are using pxelinux.0 and not lpxelinux.0
> * Till now, only the HP T5540 is not working.
>
> Regards, Willem

Comparing commits from -pre16 to -pre17, there's nothing that should
have changed yet the files are noticeably different in size (46419
versus 46895).  The _trend_ is that the ones that don't work seem
larger while the ones that work seem smaller (though not strictly).
This has me suspecting _something_ with how your one system and
specific compiles of PXELINUX use memory.  If I recall correctly, the
decompression does examine the various memory reports (free spaces and
the proper memory map) before decompressing though if they were
incongruent (ie the PXE OROM updated the lowest call but not the map),
I'm not sure if the decompression code would utilize that.

Is there any chance you could use 6.03-pre16 and collect a sysdump?
http://www.syslinux.org/wiki/index.php/Sysdump.c32

-- 
-Gene