Hello Phillipe, On 03/22/2011 05:36 PM, Seewer Philippe wrote: > Hello Gert > > On 03/22/2011 05:08 PM, Gert Huisman wrote: >> Hello Philippe, >> >> Thanks for the quick response. >> >> I have tested some other cmdline variations, and it makes a difference. >> I fail to see the logics behind it but here are the results: >> >> - several runs with our original settings: >> about 25% of the runs fails >> >> - I removed "bootdev=eth0" as it is not needed with only 1 ip= entry: >> same result as the original settings > > Explanation: Not providing an interface name for an ip line means > "interface is unspecified". Hence dracut doesn't really know which > interface to use, so the ip line should be applied to all interfaces > that pop up in sequence (If one mount on an iface fails, the next > should be tried). At least by design, I'll have to dig a bit to find > out why it failed. OK, we can specify the device name in the ip line. I expected the selection from bootdev=eth0. > > Did you see any 'dracut Warning: Mount root via '...' failed' messages? No > >> >> - I changed the ip entry to "ip=192.168.4.73:::255.255.255.0::eth0:none" >> (option 1) >> booting always failed (tried 20 times) > > This sounds like eth0 isn't at all the device you want to boot from. > > Have you tried using eth1 or eth2? Yes, whatever interface I use it fails. (see cmdline-3.console-trace-eth0?) > >> >> - I added the "ifname=eth0:<mac-address>" entry (option 2) >> mutch more reliable (1 failure out of 32 runs) >> >> Here are the last lines of the bootrace if the failure: >> -------- >> [ 0.193542] SCSI subsystem initialized >> [ 0.195914] Loading iSCSI transport class v2.0-870. >> [ 0.201704] iscsi: registered transport (tcp) >> Running in PV context on Xen v4.0. >> [ 0.211742] udev[68]: starting version 164 >> [ 0.241931] Initialising Xen virtual ethernet driver. >> [ 0.244882] udev[70]: renamed network interface eth0 to eth0-renamed >> [ 0.357445] udev[70]: renamed network interface eth0 to >> eth0-eth0-renam >> udevd-work[70]: error changing net interface name eth0-eth0-renam to >> eth0-renamed: File exists > > This confirms my suspicion that the Xen driver doesn't name interfaces > in an expected way. But the one failure is strange. > > If it isn't too much to ask: please boot with rdinitdebug and > rdnetdebug for all the above cases and provide dmesg (or /init.log) as > well as /tmp/*.out? This would help a lot.
I attaches a set of console traces with the debugging options that you asked for, cmdline-1.console-trace-failed our original cmdline: failed boot session cmdline-1.console-trace-ok our original cmdline: succesfull boot session cmdline-2.console-trace-failed bootdev=eth0 removed: failed boot session cmdline-2.console-trace-ok bootdev=eth0 removed: succesfull boot session cmdline-3.console-trace-eth0 eth0 in the ip line: failed boot session cmdline-3.console-trace-eth1 eth1 in the ip line: failed boot session cmdline-3.console-trace-eth2 eth2 in the ip line: failed boot session cmdline-4.console-trace-ok ifname=eth0:<mac-address> added: succesfull boot session. I don't have a "cmdline-4.console-trace-failed" version yet. I hope this provides the information you are looking for. We willing to send more information or run more test if needed. Thank you, Gert -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
