Re: [Xen-devel] domU jiffies not incrementing - timer issue? - Kernel 3.18.10 on Xen 4.5.0

[Mark Chambers <mark@xxxxxxxxxxxxxxx>]

On 31 March 2015 at 17:31, Mark Chambers <mark@xxxxxxxxxxxxxxx> wrote:

On 31 March 2015 at 11:56, Mark Chambers <mark@xxxxxxxxxxxxxxx> wrote:

It's nested under Hyper-V in the same manner as the problematic install. I was deliberately trying to replicate the issue, but the problem doesn't manifest.

Mark

Hi,

I've got it booting.

The machine without boot problems reports the use of emulated TSC:

(XEN) TSC not marked as either constant or reliable, warp=575 (count=2)

(XEN) dom109: mode=0,ofs=0x417376aa9c8c,khz=2633032,inc=1,vtsc count: 3576850 kernel, 9534 user

The machine with problems reports no domains having emulated TSC:

(XEN) TSC has constant rate, deep Cstates possible, so not reliable, warp=0 (count=3)

(XEN) dom23: mode=0,ofs=0x41dc316839ac,khz=2208968,inc=1

(XEN) No domains have emulated TSC

I have nothing specified in the xl config for tsc_mode. If I set tsc_mode='native'Â

and restart the DomU it boots without any problems.

If I explicitly specify any of the other tsc_mode it gets stuck with jiffies not

incrementing as before.

Mark

Hi all,Â

As Xen is reporting that "TSC has constant rate, deep Cstates possible, so not

reliable" it would seem risky to use native mode on the domU and I would prefer

to use emulated mode.

I added debug code to xen to help understand why jiffies increment correctly in

a DomU on one system but not at all on another system with an identical software

setup.

I don't understand the mechanism for updating jiffies in a Linux PV under Xen

but I have been looking at the x86 time and trap code inside Xen and have gained

a little insight.

System 1 and system 2 have identical software configurations. Both are running

windows 2012 RC2 hyper-V, running a VM which contains Xen 4.5.0 running a linux

3.18.10 dom0 running a PV domU. The biggest difference are their CPUs.

System 1 has a AMD Athlon(tm) 7750 Dual-Core Processor

System 2 has a Intel(R) Xeon(R) CPU E5520Â

From what I can deduce when using emulated TSC xen should receive lots of RDTSC

traps (opcode 0x31). The system that isn't working doesn't receive any RDTSC

traps. I suspect this may be a bug in Xen or the PV code in the linux kernel.

i.e:

System 1 tsc_mode='always_emulate' Â- xen receives RDTSC traps

System 1 tsc_mode='native' Â Â Â Â Â- xen doesn't receive RDTSC traps

System 2 tsc_mode='always_emulate' - Âxen doesn't receive RDTSC traps and

                   DomU's jiffies do not increment.

System 2 tsc_mode='native' Â Â Â Â - Âworks, xen doesn't receive RDTSC traps

I don't know if this is useful but the tsc lines from cpuid on system 1 report:

   RDTSCP                Â= false

   TscInvariant          = false

   MSR based TSC rate control  Â= false

while on system 2:

   IA32_TSC_ADJUST MSR supported      Â= false

   RDTSCP                 = false

   TscInvariant          = false

I'm trying to understand how the jiffies are updated in a PV DomU when the TSC

is emulated.Â

If anyone can point me in the right direction it'd be much appreciated.

Thanks for your time,Â

Mark

Â

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel