Xen devel - Dec 2008 - [PATCH] vmx: Fix single step on debugger

The hvm domain which is being debugged sometimes crashes with the
following message:

(XEN) Failed vm entry (exit reason 0x80000021) caused by invalid guest state
(0).
(XEN) ************* VMCS Area **************
(XEN) *** Guest State ***
(XEN) CR0: actual=0x000000008005003b, shadow=0x000000008005003b,
gh_mask=ffffffffffffffff
...[snip]...
(XEN) DebugCtl=0000000000000000 DebugExceptions=0000000000000000
(XEN) Interruptibility=0001 ActivityState=0000
(XEN) *** Host State ***
(XEN) RSP = 0xffff828c8024ffa0  RIP = 0xffff828c801885b0
(XEN) CS=e008 DS=0000 ES=0000 FS=0000 GS=0000 SS=0000 TR=e040
(XEN) FSBase=0000000000000000 GSBase=0000000000000000 TRBase=ffff828c8028a200
(XEN) GDTBase=ffff828c800f3000 IDTBase=ffff828c8028db20
(XEN) CR0=000000008005003b CR3=00000001315b9000 CR4=00000000000026b0
(XEN) Sysenter RSP=ffff828c8024ffd0 CS:RIP=e008:ffff828c801af290
(XEN) *** Control State ***
(XEN) PinBased=0000003f CPUBased=b6a1e7fe SecondaryExec=00000041
(XEN) EntryControls=000013ff ExitControls=0003efff
(XEN) ExceptionBitmap=0004400a
(XEN) VMEntry: intr_info=00000031 errcode=00000004 ilen=00000000
(XEN) VMExit: intr_info=80000301 errcode=00000400 ilen=00000000
(XEN)         reason=80000021 qualification=00000000
(XEN) IDTVectoring: info=00000000 errcode=00000000
(XEN) TPR Threshold = 0x00
(XEN) EPT pointer = 0x0000000000000000
(XEN) Virtual processor ID = 0x0000
(XEN) **************************************
(XEN) domain_crash called from vmx.c:2207
(XEN) Domain 14 (vcpu#0) crashed on cpu#0:


Signed-off-by: Kouya Shimura <kouya@jp.fujitsu.com>



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

On 19/12/2008 09:48, "Kouya Shimura" <kouya@jp.fujitsu.com>
wrote:
> The hvm domain which is being debugged sometimes crashes with the
> following message:
The fix looks like a dubious hack to me. Wouldn''t it at least be a bit
cleaner to mess with DR6.BS (the flag that''s actually being checked by
the
CPU) rather than messing with the DEBUGCTL.BTF control flag?

 -- Keir



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

Indeed, this fix is hacky but it works well for me.

At first, I tried setting GUEST_PENDING_DBG_EXCEPTIONS.BS
according to the SDM. As a result, the guest doesn''t be killed
but never go progress after the STI instruction
(i.e. VM exit immediately after VM entry) as long as RFLAGS.TF=1.
That is inconvenient for debugging. 

Is there any other way?

Thanks,
Kouya

Keir Fraser writes:
> On 19/12/2008 09:48, "Kouya Shimura" <kouya@jp.fujitsu.com>
wrote:
> 
> > The hvm domain which is being debugged sometimes crashes with the
> > following message:
> 
> The fix looks like a dubious hack to me. Wouldn''t it at least be a
bit
> cleaner to mess with DR6.BS (the flag that''s actually being
checked by the
> CPU) rather than messing with the DEBUGCTL.BTF control flag?
> 
>  -- Keir
> 
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On 22/12/2008 02:34, "Kouya Shimura" <kouya@jp.fujitsu.com>
wrote:
> Indeed, this fix is hacky but it works well for me.
> 
> At first, I tried setting GUEST_PENDING_DBG_EXCEPTIONS.BS
> according to the SDM. As a result, the guest doesn''t be killed
> but never go progress after the STI instruction
> (i.e. VM exit immediately after VM entry) as long as RFLAGS.TF=1.
> That is inconvenient for debugging.
> 
> Is there any other way?
If you make it conditional on debugger_attached then I guess you can do what
you like. I strongly dislike it though.

 -- Keir



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

Keir Fraser writes:
> On 22/12/2008 02:34, "Kouya Shimura" <kouya@jp.fujitsu.com>
wrote:
> 
> > Indeed, this fix is hacky but it works well for me.
> > 
> > At first, I tried setting GUEST_PENDING_DBG_EXCEPTIONS.BS
> > according to the SDM. As a result, the guest doesn''t be
killed
> > but never go progress after the STI instruction
> > (i.e. VM exit immediately after VM entry) as long as RFLAGS.TF=1.
> > That is inconvenient for debugging.
> > 
> > Is there any other way?
> 
> If you make it conditional on debugger_attached then I guess you can do
what
> you like. I strongly dislike it though.
> 
>  -- Keir
> 
I read the SDM carefully, however, I can''t find any solution for this.
Although the monitor trap flag might help, my machine doesn''t have such
a feature.

At least, crashing the guest should be avoided.
Attached is the updated patch. I think it might become a little better.

Thanks,
Kouya

Signed-off-by: Kouya Shimura <kouya@jp.fujitsu.com>



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