Slow FTP/'Try to free memory without signature'

[alex_knox]

Hi! This is a bit of a weird one. I have two Dell PowerEdge servers before me, an R210 II and an R720, and I'm trying to boot them both into a WinPE image. wimboot works wonderfully and saves me from the elaborate process I had to go through with gpxe in order to boot different WinPE images. Sadly, I'm running into two problems:

1. If I use rom-o-matic to build ipxe.kkpxe with undionly, the R210 II won't boot. It gives the 'Try to free memory without signature' error that comes up in some google results without any resolution that I saw. The R720, on the other hand, soars like a mighty eagle into WinPE.

2. So if I forge bravely forward and try to make my own undionly.kkpxe by cloning the git repository and running 'make bin/undionly.kkpxe', then the R210 II works, but FTP goes extremely slowly for both servers.

I realize, or I guess assume, that rom-o-matic isn't an official ipxe thing, but isn't it pulling from the same git repository I am? As far as I can tell I'm building with the same options - kkpxe, 'undi only', FTP flipped on. Is there any way I can achieve both these things - swift FTP and booting R210s?

Thanks!

[robinsmidsrod]

http://rom-o-matic.eu/ is building from a mirror of the official iPXE git repository. You can check that out with the commit id it presents where you want to choose version to build. It should be the same commit id as you can find on git.ipxe.org.

From what I can understand, the UNDI driver (undionly.kpxe) is having an issue with the memory map on one of the Dell servers, but the native driver build (ipxe.pxe) doesn't have that issue, but is apparently quite slow. My guess is that this is because of the buggy Broadcom driver being used by ipxe.pxe.

You should verify behavior on both drivers with the test suite at http://ipxe.org/dev/driver and report back what you find out.

[alex_knox]

Hi Robin,

I guess that setting 'NIC Type' to 'Undionly' in Rom-o-Matic isn't the same as 'make bin/undionly.kkpxe'? I noticed that the rom-o-matic rom is listing the chip by its proper name, whereas the rom built from source is just giving undionly.

I went through and did that test suite as best I could. There were a few issues. Let me know if I can be of any further assistance.

undionly built from source

link detection:
- only detects net0 (not net1)
- does not change link state to down after removing cable
- does not change link state after ifclose/ifopen, either

dhcp: good

close and reopen: good

large file transfer correctness: good

large file transfer speed: 4 times, average 300s

link interruption: good

all loopback: net1 not listed/recognized?

chip: undionly; version: 1.0.0+ (bf15)




rom built from rom-o-matic

link detection: good

dhcp: good

close and reopen: good

large file transfer correctness: good

large file transfer speed: 4 times, average 4.6s

link interruption: does NOT resume (killed after 5 minutes)

loopback:
- lotest net0 net7: counter does not increase
- lotest net7 not0: counter does not increase

chip: 14e4-1657; version: 1.0.0+