Syslinux - Jun 2019 - lpxelinux.0 problems with CentOS 7.6

Hello list.

This issue seems like it may be related to one described in a post in
May: https://www.syslinux.org/archives/2019-May/026424.html.  I've
only just subscribed, so I was unable to reply directly to that
thread.

I am having a similar issue when I attempt to PXE boot CentOS 7.6.1810
using lpxelinux.0.  The files are fully downloaded from the server and
the kernel starts booting but panics when it attempts to mount the
init ramdisk as its root filesystem.  This occurs when using either
HTTP or TFTP to retrieve the files.

If I switch to pxelinux.0 the kernel boots successfully.

I tested a few other versions with lpxelinux.0 and HTTP:
CentOS 7.5.1804: success
CentOS 7.4.1708: success
CentOS 6.10: success

I am using the official pre-compiled binaries from the syslinux 6.03
tarball downloaded directly from kernel.org, with this configuration:

LABEL centos7-ks
  MENU LABEL CentOS 7.x kickstart
  KERNEL http://filer/repo/centos/7/os/x86_64/images/pxeboot/vmlinuz
  INITRD http://filer/repo/centos/7/os/x86_64/images/pxeboot/initrd.img
  APPEND ip=dhcp inst.ks=http://filer/repo/kickstart/centos7-ks.cfg inst.gpt

In the older thread, there was some talk of the increasing size of the
initrd.img file possibly being an issue, but in my testing, the 7.5
initrd.img is larger than 7.6's but still boots fine

# unxz -l /pool2/repo/centos/7.?.????/os/x86_64/images/pxeboot/initrd.img
Strms  Blocks   Compressed Uncompressed  Ratio  Check   Filename
    1       1     46.2 MiB    130.1 MiB  0.355  CRC32
/pool2/repo/centos/7.4.1708/os/x86_64/images/pxeboot/initrd.img
    1       1     50.4 MiB    133.6 MiB  0.377  CRC32
/pool2/repo/centos/7.5.1804/os/x86_64/images/pxeboot/initrd.img
    1       1     50.1 MiB    132.7 MiB  0.378  CRC32
/pool2/repo/centos/7.6.1810/os/x86_64/images/pxeboot/initrd.img

The fact that the older versions boot fine suggests that something may
have changed on the CentOS side, but conversely, the fact that 7.6
fails to boot with lpxelinux.0 but boots fine with pxelinux.0 suggests
an issue in lpxelinux.0.  I don't know much about these things, but I
would assume that pxelinux.0 and lpxelinux.0 load the vmlinuz and
initrd.img files into memory and execute the kernel in the same way?

What can I do to further diagnose this issue?

JM

> What can I do to further diagnose this issue?
 
 
The email thread you linked to (still) continues; it does not finish in 
May 2019.

Reading the whole set of emails in that email thread, you will notice 
that the other email thread is slightly different than yours (and 
actually that email thread has not finished yet). In spite of the 
differences, the suggestions that were already given in that other 
email thread are also relevant for the symptoms you are seeing (e.g. 
test Syslinux-related packages from current Debian Sid; use "LINUX" 
rather than "KERNEL", CSM vs. UEFI mode...).

I would suggest (re-)reading the entire email thread (sorted by date), 
not only the emails from May 2019. Hopefully sharing the results of 
your tests would help narrowing down the problem(s).

FWIW, there are even more tests that I could suggest (for both 
problems); I can only hope that at some point Peter (hpa) and/or Gene 
will get involved.


@hpa, the problem described in the other email thread is in Syslinux, 
not in Cisco's hardware/firmware.

Regards,
Ady.

Thanks for your response Ady.

I have now tested with the prebuilt binaries from
https://www.zytor.com/pub/syslinux/Testing/6.04/syslinux-6.04-pre1.tar.xz

Using this version of lpxelinux.0, with either TFTP or HTTP transfers,
results in a successful boot.

The test system is a QEMU+KVM virtual machine emulating a BIOS-based
system btw (not UEFI CSM).

Kind regards,
Jason

On Fri, 28 Jun 2019 at 16:35, Ady Ady via Syslinux
<syslinux at syslinux.org> wrote:
>
> > What can I do to further diagnose this issue?
>
>
> The email thread you linked to (still) continues; it does not finish in
> May 2019.
>
> Reading the whole set of emails in that email thread, you will notice
> that the other email thread is slightly different than yours (and
> actually that email thread has not finished yet). In spite of the
> differences, the suggestions that were already given in that other
> email thread are also relevant for the symptoms you are seeing (e.g.
> test Syslinux-related packages from current Debian Sid; use
"LINUX"
> rather than "KERNEL", CSM vs. UEFI mode...).
>
> I would suggest (re-)reading the entire email thread (sorted by date),
> not only the emails from May 2019. Hopefully sharing the results of
> your tests would help narrowing down the problem(s).
>
> FWIW, there are even more tests that I could suggest (for both
> problems); I can only hope that at some point Peter (hpa) and/or Gene
> will get involved.
>
>
> @hpa, the problem described in the other email thread is in Syslinux,
> not in Cisco's hardware/firmware.
>
> Regards,
> Ady.
>
> _______________________________________________
> Syslinux mailing list
> Submissions to Syslinux at syslinux.org
> Unsubscribe or set options at:
> https://lists.syslinux.org/syslinux