Bug 242765

We need the exact messages from the screen. Please boot in 50-line mode (add the
kernel parameter "vga=ask" and select 80x50) then take a picture of the panic
with a digital camera. upload that as an attachment to this bugzilla.

Switching to vesa mode 80x50 doesnt help because
when INIT take the boot process they switch back
to, i dont know maybe 80x24 ?


Right now i haven't a camera here but i typed it down:
so no warranty 





R13: 0000000000000001 
R14: 0000000000000012 
R15: 0000000000001ac5
FS:   00002aaaab679710(0000) 
GS:   ffff81007fe8b940(0000) 
knlGS:0000000000000000

CS: 0010 DS: 0000 ES: 0000 CR0: 000000008005003b

CR2: 0000555555769e90 CR3: 000000007f310000 CR4: 00000000000006e0
Process udevd (pid: 496, threadinfo ffff81007fd6a000, task ffff81007ecff040)
Stack: ffffffff802582ac ffff810037c87f78 81007fea30000080 000000000000ffff
0000000000000000 ffffffff80221bf7 0000000000000000 0000000000000053
00000000000002ad ffffffff80257c76 ffff81007fd6bd60 <EOI> 0000000000000043

CallTrace:

<IRQ> [<ffffffff802582ac>] call_softirq+0x1c/0x28

<From me: Without [< >]

ffffffff80221bf7 smp_call_function_interrupt+0x40/0x5e
ffffffff80257c76 call_function_interrupt+0x66/0x70
<EOI> ffffffff8033190e __first_cpu+0x3/0x1f
ffffffff8023f929 show_stat+0x2f5/0x3a0
ffffffff8023c694 seq_read+0x105/0x28c
ffffffff8020af0a vfs_read+0xcb/0x173
ffffffff80210606 sys_read+0x45/0x6e
ffffffff8025711e system_call+0x7e/0x83


Code: 0f30488b5620488b0a48c1e933 48 89 c8 48 c1 e8 09

RIP ffffffff8823932a : kvm_amd:svm_hardware_enable+0xa0/0x141
RSP ffff810037c87f68
kernel panic - not syncing : aiee , killing interrupt handler!




JFYI: I can boot kernel-2.6.21-1.3xxx from FC7RC2 LiveCD !

I have tested this kernels:

kernel-2.6.21-1.3200.fc8 - doesnt work

kernel-2.6.21-1.3194.fc7 - doesnt work
i expected that at least this would work
because it is the same like the one on
FC7RC2-Live.iso

kernel-2.6.21-1.3116.fc7 - doesnt work
kernel-2.6.20-1.3104.fc7 - doesnt work

kernel-xen-2.6.20-2925.9.fc7 - does work

--

udev is on f7 and fc7rc2 the same (106-4.fc7)

--

my grub.conf

default=1
timeout=5
splashimage=(hd0,1)/boot/grub/splash.xpm.gz
hiddenmenu
password --md5 $1$nz2jA0d7$qZGgWbA8bqopvb3eQPZbJ.
title Fedora (2.6.21-1.3116.fc7)
        root (hd0,1)
        kernel /boot/vmlinuz-2.6.21-1.3116.fc7 ro root=LABEL=/
        initrd /boot/initrd-2.6.21-1.3116.fc7.img
title CasaRoot (2.6.20-2925.9.fc7xen)
        root (hd0,1)
        kernel /boot/xen.gz-2.6.20-2925.9.fc7 
        module /boot/vmlinuz-2.6.20-2925.9.fc7xen ro root=LABEL=/
        module /boot/initrd-2.6.20-2925.9.fc7xen.img

(In reply to comment #2)
> Switching to vesa mode 80x50 doesnt help because
> when INIT take the boot process they switch back
> to, i dont know maybe 80x24 ?
>

Yeah, it loads the stupid default font and that forces it back to 80x25.
That can be disabled (the easy way) by renaming /lib/kbd/consolefonts
to something else during testing.
 
> Right now i haven't a camera here but i typed it down:
> so no warranty 
> RIP ffffffff8823932a : kvm_amd:svm_hardware_enable+0xa0/0x141
> RSP ffff810037c87f68

You should be able to boot by blacklisting the kvm_amd module.
Edit /etc/modprobe.d/blacklist and add it.

I had kernel-2.6.21-1.3116.fc7 installed 

I.
i first modified /etc/modprobe.d/blacklist 
then updated the kernel to kernel-2.6.21-1.3194.fc7
and rebooted - blacklisting kvm_amd doenst help.
 
II.
i removed drivers/kvm dir, then i rebuild
the kernel dependencies and rebooted. This worked !!!


Being in 2.6.21-1.3194 i restored the drivers/kvm dir
and rebuilded the kernel dependencies again.

so - doing manualy modprobe kvm_amd killed the box.

for now - kvm_amd triggers these freezes.

so far - thank you - but one eye is crying because
i was planning to setup this box as a virt server ,)
 
Good Night (here is midnight)

The latest dev kernel (2.6.21-1.3218.fc8.x86_64)
also gets panic if i load the kvm-amd modul.

If virtualization is enabled in the BIOS, disable is.
If it is enabled, disable it :)

And let us know what happens.

About the BIOS: There is no possibility for change any 
virtualization option. it dosnt offer it. 

By the way - should we change the bug title - now where we 
known what it caused?

Created attachment 158069 [details]
detect svm disabled by bios

The attached patch fixes this.

Patch went into kernel 3237

I tried kernel-2.6.21-1.3240.fc8.x86_64 and kernel-2.6.21-1.3228.fc7.x86_64.

The difference to the other tested kernels - it passes the point where init starts udev.

After booting - no kvm* module is loaded.

Loading kvm-amd manually ends in a kernel panic.

Patch is in kernel 1.3237.fc7

fc8 is following upstream and still doesn't have this fix.

Where to find 1.3237.fc7 kernel?

I searched in development/x86_64/os/Fedora/, updates/testing/7/x86_64/, 
updates/7/x86_64/ and http://people.redhat.com/cebbert/kernels/!

(In reply to comment #13)
> Where to find 1.3237.fc7 kernel?

Not released, some kernel after 3237 will be released soon.

(In reply to comment #9)
> Created an attachment (id=158069) [edit]
> detect svm disabled by bios
> 
> The attached patch fixes this.

Avi, this patch is broken on 32-bit.

See bug #246250

Patch is in kernel 2.6.21-1.3255.fc7, in updates-testing now.

# uname -a
Linux casa.localdomain 2.6.21-1.3255.fc7 #1 SMP Thu Jul 5 18:03:06 EDT 2007 x86_64 x86_64 
x86_64 GNU/Linux

# modprobe kvm

# modprobe kvm-amd
kvm: disabled by bios
FATAL: Error inserting kvm_amd (/lib/modules/2.6.21-1.3255.fc7/kernel/drivers/kvm/kvm-amd.ko): 
Operation not supported

# lsmod |head
Module                  Size  Used by
kvm                    75969  0 
autofs4                32841  2 
sunrpc                188329  1 
ipv6                  340289  16 
cpufreq_ondemand       17489  1 
dm_mirror              30209  0 
dm_mod                 69713  1 dm_mirror
video                  28113  0 
sbs                    25729  0 


Well as the message above stated - kvm stuff is unusable and no 
bios options exists for enabling it. 

Is it  possible to bypass the bios and enabling it while booting?

There is no way to override the BIOS settings for svm. The patch is doing its
job and preventing a kernel crash when svm is disabled -- that's all it can do.

This appears to be still broken on kernel-2.6.22.4-65.fc7 - is this a new branch
that doesn't include the above fix?

2.6.21-1.3228.fc7 (which is before the fix) does boot, usually, but always
outputs a kernel traceback while starting udev.  The newer kernel panics every time.

For anyone reading this who has an HP machine (dc5750 here with BIOS 2.20), I
found the BIOS option that controls virtualization technology - it's in Security
-> OS Security.  Setting this to Enable fixes the oops.

There's a kvm-devel thread about the same problem (no patch, though):
http://www.mail-archive.com/kvm-devel@lists.sourceforge.net/msg03890.html