Re: pvh DOMU on NetBSD DOM0 crashes

[Manuel Bouyer <bouyer%antioche.eu.org@localhost>]

On Tue, Apr 16, 2024 at 10:45:39AM +0200, oskar%fessel.org@localhost wrote:
> 
> 
> > Am 15.04.2024 um 04:34 schrieb Mathew, Cherry G. <c%bow.st@localhost>:
> > 
> >>>>>> On Sun, 14 Apr 2024 19:15:09 +0200, Manuel Bouyer <bouyer%antioche.eu.org@localhost <mailto:bouyer%antioche.eu.org@localhost>> said:
> > 
> >> On Sun, Apr 14, 2024 at 06:23:02PM +0200, oskar%fessel.org@localhost wrote:
> >>> Hej,
> >>> 
> >>> I ran into a problem with one of my NetBSD DomUs, it always hits the PVH domu with the highest index.
> >>> 
> >>> Message in the DOM0:
> >>> (XEN) p2m_pod_demand_populate: Dom5 out of PoD memory! (tot=131016 ents=131072 dom5)
> >>> (XEN) domain_crash called from arch/x86/mm/p2m-pod.c:1288
> >>> (XEN) Domain 5 (vcpu#0) crashed on cpu#0:
> >>> (XEN) ----[ Xen-4.18.0_20231116nb0  x86_64  debug=n  Not tainted ]----
> >>> (XEN) CPU:    0
> >>> (XEN) RIP:    0008:[<ffffffff8023439f>]
> >>> (XEN) RFLAGS: 0000000000010206   CONTEXT: hvm guest (d5v0)
> >>> (XEN) rax: 0000000000002800   rbx: 0000000000002800   rcx: 0000000000000200
> >>> (XEN) rdx: 00007c1cee2ba2c0   rsi: 00007c1cee2b92c0   rdi: ffffc10066592000
> >>> (XEN) rbp: ffffc100726f2c50   rsp: ffffc100726f2c08   r8:  00007f7ffffff000
> >>> (XEN) r9:  ffffc100726f2ca0   r10: 000000000000001b   r11: 0000000000000000
> >>> (XEN) r12: 0000000000002800   r13: ffffc10066590800   r14: ffffc100726f2ee0
> >>> (XEN) r15: ffffc100726f2ef0   cr0: 0000000080050033   cr4: 00000000001406b0
> >>> (XEN) cr3: 00000000047d6000   cr2: ffffc10066591000
> >>> (XEN) fsb: 00007c1cee327850   gsb: ffffffff8183cb00   gss: 0000000000000000
> >>> (XEN) ds: 0023   es: 0023   fs: 0000   gs: 0000   ss: 0010   cs: 0008
> >>> (XEN) p2m_pod_demand_populate: Dom5 out of PoD memory! (tot=131016 ents=131072 dom5)
> >>> (XEN) domain_crash called from arch/x86/mm/p2m-pod.c:1288
> >>> (XEN) p2m_pod_demand_populate: Dom5 out of PoD memory! (tot=131016 ents=131072 dom5)
> >>> (XEN) domain_crash called from arch/x86/mm/p2m-pod.c:1288
> >>> (XEN) p2m_pod_demand_populate: Dom5 out of PoD memory! (tot=131016 ents=131072 dom5)
> >>> (XEN) domain_crash called from arch/x86/mm/p2m-pod.c:1288
> >>> 
> >>> This is XEN 4.18, it happened also on 4.15.  I upgraded to 4.18 because of this ;-)
> >>> 
> >>> The config for this DOMU is:
> >>> ??
> >>> type = pvh
> >>> name = "littlejoe"
> >>> boot = "c"
> >>> memory = 512
> >>> maxmem = 1024
> >>> vcpus = 1
> >>> vif = [ 'bridge=bridge1', 'bridge=bridge2' ]
> >>> kernel = "/netbsd-GENERIC.gz"
> >>> disk = [ '/hurz/joe/xendisklittlejoe,raw,xvda,rw? ]
> >>> ??
> >>> 
> >>> all my pvh domUs run amd64
> >>> NetBSD 10.99.10 (GENERIC) #2: Fri Dec  8 23:45:08 CET 2023
> >>> the DOM0 itself is older:
> >>> NetBSD 10.99.2 (XEN3_DOM0) #0: Tue Dec 27 03:19:40 EET 2022 
> >>> 
> >>> The boot.cfg has the following entry:
> >>> menu=Boot Xen com:load /netbsd-XEN3_DOM0 root=dk6
> >>> console=com1,115200;multiboot /xen418.gz dom0_mem=32768M
> >>> dom0_max_vcpus=8 console=com2 com2=115200,8n1,0x2f8
> >>> 
> >>> Anything missing there?
> > 
> >> Googling for "out of PoD memory" points to balooning. Using memory == maxmem
> >> may work around the issue (I set only maxmem in my config files, which
> >> may explain why I didn't hit the issue).
> > 
> > Is this a situation where, on boot, the balloon demands more memory than
> > the hypervisor has available ?
> > 
> 
> No, that happens after a day or so.  But since generic kernel does not initialize the balloon, that is supposedly when it runs out of memory and tries to access maxmem?

No, I think that works the other way round: the domU starts with maxmem
allocated and, if it has a balloon driver, it releases the difference
between memory and maxmem to the hypervisor.

-- 
Manuel Bouyer <bouyer%antioche.eu.org@localhost>
     NetBSD: 26 ans d'experience feront toujours la difference
--