Re: Dom0 ballooning: crash on guest start

[Christoph Egger <Christoph_Egger%gmx.de@localhost>]

On 04/29/11 14:37, Jean-Yves Migeon wrote:
> On Fri, 29 Apr 2011 14:08:54 +0200, Christoph Egger wrote:
> > balloon0: inflate 512 => inflated by 512
> > for about 20 seconds until the crash happens:
> >
> > balloon0: inflate 512 => inflated by 512
> > uvm_fault(0xffffffff80c6b220, 0xffffffff81400000, 1) -> e
> > fatal page fault in supervisor mode
> > trap type 6 code 0 rip ffffffff8054593d cs e030 rflags 10216 cr2
> > ffffffff81400000 cpl 0 rsp f
> > fffa00057ce3a40
> > kernel: page fault trap, code=0
> > Stopped in pid 0.58 (system) at netbsd:pmap_zero_page+0x41: movq 0(%rax,%
> > rbx,8),%rbx
> > pmap_zero_page() at netbsd:pmap_zero_page+0x41
> > uvm_pagealloc_strat() at netbsd:uvm_pagealloc_strat+0x22d
> > bpge_ctor() at netbsd:bpge_ctor+0x22
> > pool_cache_get_slow() at netbsd:pool_cache_get_slow+0x1df
> > pool_cache_get_paddr() at netbsd:pool_cache_get_paddr+0x1f7
> > balloon_thread() at netbsd:balloon_thread+0x3bc
>
> Okay, happens when balloon(4) has already been inflated by a good share.
>
> Out of interest, having XENDEBUG_SYNC enabled in
> arch/xen/x86/x86_xpmap.c does not change anything to the result?

No effect.

> > BTW: The HVM guest gets 4GB RAM assigned and Xen uses
> > 0x200 4KB pages, 0x3fb 2MB pages and 0x2 1GB pages for it.
>
> Interesting. I would have imagined a higher count for 4K pages relative
> to the 2MB one.

The reason is the x86 layout:
For the first 2MB you have to use 4KB pages.
For the range of 2MB - 1GB and for the PCI hole
you can use either 4KB or 2MB pages.

Christoph