Syslinux - Sep 2015 - BUG: confusion between next-server and gateway

Hello,

I'am trying to efi-netboot a HP ProLiant DL120 Gen9[1], that has
multiple NICs, one embedded 2-port and an extra 4-port, so I got to the
HP/EFI/multinic problem.

I read through the multinic mailthreads and using current master[2], it
booted successfully to syslinux but then another problem manifested:
Syslinux for some reason asks the gateway and not next-server about
ldlinux.e64 (I got this through tcpdumping the gw)

Our setup is the following:

10.1.10.1: gw
10.1.10.105: client-to-be-booted
192.168.1.2: dhcp next-server/tftp, routed through gw

To make sure that this is a syslinux bug, I disabled the extra nics
through HP's firmware (so that the multinic problem would not manifest)
and booted successfully to debian installer with stable syslinux 6.03.

Some bisecting (from 6.03 onwards) and reboots later, the first commit
that exhibits this behaviour seems to be 'f858a54 efi: Hunt for service
binding handle if needed'

Sadly I'm not that familiar with syslinux/EFI internals to help you
more, but I can test anything you throw at me :)

Regards,
Costas

[1] Upgraded to latest firmware/drivers as of yesterday
[2] Current master now is on d70168d

On Wed, Sep 2, 2015 at 5:54 AM, Costas Drogos via Syslinux
<syslinux at zytor.com> wrote:
> Hello,
>
> I'am trying to efi-netboot a HP ProLiant DL120 Gen9[1], that has
> multiple NICs, one embedded 2-port and an extra 4-port, so I got to the
> HP/EFI/multinic problem.
HP DL120 G9 versus HP DL160 G9.  Sounds like they'll be awfully similar.
> I read through the multinic mailthreads and using current master[2], it
> booted successfully to syslinux but then another problem manifested:
> Syslinux for some reason asks the gateway and not next-server about
> ldlinux.e64 (I got this through tcpdumping the gw)
OH?  This has me thinking.  Please look closer.  Is it targetting the
gateway's MAC address but the next-server's IPv4 address?  Or is it
targetting the gateway's MAC address and IPv4 address?
> Our setup is the following:
>
> 10.1.10.1: gw
> 10.1.10.105: client-to-be-booted
> 192.168.1.2: dhcp next-server/tftp, routed through gw
>
> To make sure that this is a syslinux bug, I disabled the extra nics
> through HP's firmware (so that the multinic problem would not manifest)
> and booted successfully to debian installer with stable syslinux 6.03.
>
> Some bisecting (from 6.03 onwards) and reboots later, the first commit
> that exhibits this behaviour seems to be 'f858a54 efi: Hunt for service
> binding handle if needed'
What behavior do you see before this?
> Sadly I'm not that familiar with syslinux/EFI internals to help you
> more, but I can test anything you throw at me :)
>
> Regards,
> Costas
>
> [1] Upgraded to latest firmware/drivers as of yesterday
> [2] Current master now is on d70168d
Digging like this is both helpful and attracts my interest.

-- 
-Gene

On 09/02/2015 01:33 PM, Gene Cumm wrote:
> On Wed, Sep 2, 2015 at 5:54 AM, Costas Drogos via Syslinux
> <syslinux at zytor.com> wrote:
>> Hello,
>>
>> I'am trying to efi-netboot a HP ProLiant DL120 Gen9[1], that has
>> multiple NICs, one embedded 2-port and an extra 4-port, so I got to the
>> HP/EFI/multinic problem.
> 
> HP DL120 G9 versus HP DL160 G9.  Sounds like they'll be awfully
similar.
Yes, they are quite similar.
>> I read through the multinic mailthreads and using current master[2], it
>> booted successfully to syslinux but then another problem manifested:
>> Syslinux for some reason asks the gateway and not next-server about
>> ldlinux.e64 (I got this through tcpdumping the gw)
> 
> OH?  This has me thinking.  Please look closer.  Is it targetting the
> gateway's MAC address but the next-server's IPv4 address?  Or is it
> targetting the gateway's MAC address and IPv4 address?
So I did a packet capture in our gateway and it goes like this:
- client (.105) does arp, get the gateway and gateway mac
- client gets /syslinux/syslinux.efi from next-server
- boots it
- client tries to get /syslinux/ldlinux.e64 from gateway's ip (with
gateway's mac) many times, then /syslinux//ldlinux.e64 and so on.

All MAC and IPs on packets are the correct ones.
>> Some bisecting (from 6.03 onwards) and reboots later, the first commit
>> that exhibits this behaviour seems to be 'f858a54 efi: Hunt for
service
>> binding handle if needed'
> 
> What behavior do you see before this?
- 6.03 boots nice
Then either:
 - failing to boot at all
 - booting and working
until f858a54 which boots but exhibits the problem.

As I was after the gateway/next-server problem, I didn't examine further
the boot failures, could do if you need me to

The setup during all these tests has 1 bootable card, the others are
fully disabled through HP's firmware.