natty kernel does not boot on t1.micro in arch i386

Well after the normal 4 minutes, console data did appear for the crashed instance. It doesn't seem like its likely worth much, but its there.

This is the same as I got in the test environment. Seems a silent update of CONFIG_PHYSICAL_START to be 16M instead of 1M breaks boot with pv-grub. I can still boot the new kernel when I boot it directly. But pv-grub produces this dump. So for Natty, I am changing the physical start back to 1M for i386. In theory this is a pv-grub problem but not sure we get that fixed anytime (soon).

I'm not sure if this actually occurs on "normal grub" (0.97) or only on the grub version we get on amazon (pv-grub).

Note about 0.97 grub... I failed to realize that the failure in pv-grub is is limited to i386 kernels on t1.micro (ie, they boot fine in m1.small or c1.medium).

Are you using CONFIG_RELOCATABLE=y for your kernels? If so, CONFIG_PHYSICAL_START should not be a factor.

On Tue, 1 Feb 2011, Matt Wilson wrote:

> Are you using CONFIG_RELOCATABLE=y for your kernels? If so,
> CONFIG_PHYSICAL_START should not be a factor.

$ egrep "(CONFIG_RELOCATABLE|CONFIG_PHYSICAL_START)"
/boot/config-2.6.38-1-virtual
CONFIG_PHYSICAL_START=0x1000000
CONFIG_RELOCATABLE=y
$ uname -a
Linux ip-10-112-14-12 2.6.38-1-virtual #28-Ubuntu SMP Fri Jan 28 18:38:01
UTC 2011 i686 i686 i386 GNU/Linux

What is CONFIG_PHYSICAL_ALIGN?

-- Sent from my Palm Pre

________________________________
On Feb 1, 2011 12:16 PM, Scott Moser <email address hidden> wrote:

On Tue, 1 Feb 2011, Matt Wilson wrote:

> Are you using CONFIG_RELOCATABLE=y for your kernels? If so,
> CONFIG_PHYSICAL_START should not be a factor.

$ egrep "(CONFIG_RELOCATABLE|CONFIG_PHYSICAL_START)"
/boot/config-2.6.38-1-virtual
CONFIG_PHYSICAL_START=0x1000000
CONFIG_RELOCATABLE=y
$ uname -a
Linux ip-10-112-14-12 2.6.38-1-virtual #28-Ubuntu SMP Fri Jan 28 18:38:01
UTC 2011 i686 i686 i386 GNU/Linux

--
You received this bug notification because you are a direct subscriber
of the bug.
https://bugs.launchpad.net/bugs/710754

Title:
  natty kernel does not boot on t1.micro in arch i386

Status in “grub” package in Ubuntu:
  New
Status in “linux” package in Ubuntu:
  In Progress

Bug description:
  $ ec2-run-instances --region us-east-1 --instance-type t1.micro --key
  mykey ami-5c3fcf35

  That results in instance that has no console output and is not
  reachable.

  Note, that under bug 686692 the amd64 on t1.micro was fixed.

  ProblemType: Bug
  DistroRelease: Ubuntu 11.04
  Package: linux-image-2.6.38-1-virtual 2.6.38-1.28
  Regression: Yes
  Reproducible: Yes
  ProcVersionSignature: User Name 2.6.38-1.28-virtual 2.6.38-rc2
  Uname: Linux 2.6.38-1-virtual i686
  AlsaDevices:
   total 0
   crw------- 1 root root 116, 1 2011-01-31 16:40 seq
   crw------- 1 root root 116, 33 2011-01-31 16:40 timer
  AplayDevices: Error: [Errno 2] No such file or directory
  Architecture: i386
  ArecordDevices: Error: [Errno 2] No such file or directory
  CurrentDmesg:

  Date: Mon Jan 31 16:45:05 2011
  Ec2AMI: ami-5c3fcf35
  Ec2AMIManifest: (unknown)
  Ec2AvailabilityZone: us-east-1b
  Ec2InstanceType: t1.micro
  # above edited, originally reported on m1.small as t1.micro does not boot
  Ec2Kernel: aki-407d9529
  Ec2Ramdisk: unavailable
  Lspci:

  Lsusb: Error: command ['lsusb'] failed with exit code 1:
  ProcEnviron:
   PATH=(custom, user)
   LANG=en_US.UTF-8
   LC_MESSAGES=en_US.utf8
   SHELL=/bin/bash
  ProcKernelCmdLine: root=LABEL=uec-rootfs ro console=hvc0
  ProcModules: acpiphp 23425 0 - Live 0xedc10000
  SourcePackage: linux

To unsubscribe from this bug, go to:
https://bugs.launchpad.net/ubuntu/+source/grub/+bug/710754/+subscribe

> What is CONFIG_PHYSICAL_ALIGN?

$ egrep "(CONFIG_RELOCATABLE|CONFIG_PHYSICAL)" /boot/config-2.6.38-1-virtual
CONFIG_PHYSICAL_START=0x1000000
CONFIG_RELOCATABLE=y
CONFIG_PHYSICAL_ALIGN=0x100000

We use:
CONFIG_PHYSICAL_START=0x1000000
CONFIG_PHYSICAL_ALIGN=0x1000000

It sounds like that works for you too?

right. Stefan is going to move that back.

No, we go back to 0x100000 (5 zeros ;)) for start and alignment. I have tried in the test environment with both align and start being 0x1000000 (16M) which did *not* work when using pv-grub for the boot process. At that point I wondered whether this is kernel code itself or something going along with pv-grub. So I modified the xen config to load the kernel directly. This works.

Note that the other configuration works as well when using mem=512 for the DomU, just not with mem=615. I did not exhaustively try other memory layouts but at leas 1024 and whatever is used for the m1.medium and c1.small seem to work as well.

In theory the physical (according to the help text) start should not matter when relocatable is set. Though still there is LOAD_PHYSICAL_ADDR calculated from PHYSICAL_START and PHYSICAL_ALIGN and that seems to be used at least in some places even with relocatable turned on (to calculate the delta between compile and load addresses for example).

We probably would have run into this before. Though we carried the defaults which were 0x200000 for amd64 and 0x100000 for i386 until somewhere around 2.6.31. The change to physical start sort of sneaked in when upstream changed the settings this value depends on.

This bug was fixed in the package linux - 2.6.38-2.29

---------------
linux (2.6.38-2.29) natty; urgency=low

  [ Andy Whitcroft ]

  * rebase to 1f0324caefd39985e9fe052fac97da31694db31e
  * [Config] updateconfigs following rebase to
    1f0324caefd39985e9fe052fac97da31694db31e
  * rebase to 70d1f365568e0cdbc9f4ab92428e1830fdb09ab0
  * [Config] reenable HIBERNATE
    - LP: #710877
  * rebase to v2.6.38-rc3
  * [Config] reenable CONFIG_CRASH_DUMP

  [ Kamal Mostafa ]

  * SAUCE: rtl8192se: fix source file perms
  * SAUCE: rtl8192se: fix source file newline
  * SAUCE: omnibook: fix source file newline

  [ Kees Cook ]

  * [Config] packaging: really make System.map mode 0600

  [ Ricardo Salveti de Araujo ]

  * SAUCE: OMAP3630: PM: don't warn the user with a trace in case of
    PM34XX_ERRATUM

  [ Soren Hansen ]

  * SAUCE: nbd: Remove module-level ioctl mutex

  [ Tim Gardner ]

  * SAUCE: Disable building the ACPI debugfs source

  [ Upstream Kernel Changes ]

  * Set physical start and alignment 1M for virtual i386
    - LP: #710754

  [ Upstream Kernel Changes ]

  * rebase from v2.6.38-rc2 + c723fdab8aa728dc2bf0da6a0de8bb9c3f588d84
    to v2.6.38-rc3
 -- Andy Whitcroft <email address hidden> Fri, 28 Jan 2011 16:30:32 +0000

Verified fixed.
us-east-1 ami-0e6b9b67 canonical ebs/ubuntu-natty-daily-i386-server-20110207
$ uname -a
Linux ip-10-112-58-95 2.6.38-2-virtual #29-Ubuntu SMP Fri Feb 4 16:24:23 UTC 2011 i686 i686 i386 GNU/Linux

The permanent fix for this is likely in PV-GRUB. See: https://patchwork.kernel.org/patch/727511/

Natty is no more a supported version; and grub legacy upstream is also stopped, only receiving possible random fixes locally