[Xen-users] Xen 3.0.0 32bit-pae (testing changeset 8270) crashes (pgtable.c:284, kernel bug?)

[Ralph Passgang <ralph@xxxxxxxxxxxxx>]

Hi,

I've got two identical system with the following specs:
- Dual Xeon 2,8GHz (HT enabled)
- 4GB RAM
- 3x160GB SATA HDD
- 3ware 9000 controller (with 4 sata-ports)
-  Intel Board
- 2x Intel GBit NICs (onboard)
- Rage XL VGA (onboard)

Both systems were running debian sarge (3.1) with xen 2.0.7 for a long time 
without any problems at all. Xen 3.0.0 also seems to work fine, as long as 
the normal i386 mode is used. With pae both systems crashes everytime I want 
to create a domainU that needs to use some ram of last GB available. I 
haven't tested x86_64 on this machines at all.

With xen 2.0.7 I had only 3GB of the complete 4GB available, and with 
xen3.0.0-pae it seems so that I can use the same amount of ram safely (about 
3GB) and if I want to assign memory from the last GB to new domainUs, then 
the whole system crashes. Without using the last gb of ram the system seems 
to run very stable.

The following is shown (when the flag "noreboot" was used at boot time or if 
the output was grabed over a serial line). Without "noreboot" the whole 
system will reboot immidently:

kernel BUG at arch/xen/i386/mm/pgtable.c:284 (pgd_ctor)!
 [<c0118272>] pgd_ctor+0x52/0x60
 [<c014fab1>] cache_init_objs+0x71/0x80
 [<c014fc7f>] cache_grow+0x10f/0x1e0
 [<c014fecb>] cache_alloc_refill+0x17b/0x250
 [<c01bbf26>] cramfs_readpage+0xd6/0x170
 [<c01501bf>] kmem_cache_alloc+0x7f/0x90
 [<c01182fd>] pgd_alloc+0x1d/0x310
 [<c014b270>] prep_new_page+0x60/0x70
 [<c014b778>] buffered_rmqueue+0x128/0x260
 [<c0121c4e>] mm_init+0xce/0x100
 [<c0121fa4>] copy_mm+0xd4/0x3d0
 [<c0122b9f>] copy_process+0x3df/0xd20
 [<c01235d9>] do_fork+0x69/0x18f
 [<c0108491>] sys_fork+0x31/0x40
 [<c0109b8d>] syscall_call+0x7/0xb
Kernel panic - not syncing: BUG!

the interessting lines in xen's dmesg log regarding the memory layout are 
these:

(XEN) PHYSICAL MEMORY ARRANGEMENT:
(XEN)  Dom0 alloc.:   0000000007000000->0000000008000000 (999392 pages to be 
allocated)
(XEN) VIRTUAL MEMORY ARRANGEMENT:
(XEN)  Loaded kernel: c0100000->c048db00
(XEN)  Init. ramdisk: c048e000->c09cd000
(XEN)  Phys-Mach map: c09cd000->c0da0f80
(XEN)  Start info:    c0da1000->c0da2000
(XEN)  Page tables:   c0da2000->c0daf000
(XEN)  Boot stack:    c0daf000->c0db0000
(XEN)  TOTAL:         c0000000->c1000000
(XEN)  ENTRY ADDRESS: c0100000
(XEN) Initrd len 0x53f000, start at 0xc048e000

I already checked the bios settings and changed all settings that could have 
something to do with pae. it doesn't make any difference at all.

Has somebody a clue what's happening here? Is it really a xen bug? If so, why 
is nobody else having this problem? Is it a problem with the hardware? at 
least windows with pae and native linux seems to work without a problem on 
the very same hardware.

If a developer wants to take a closer look, ssh access should be possible.

--Ralph

_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users