Pausing / unpausing a single domain VCPU via libxc

Pausing / unpausing a single domain VCPU via libxc

[Razvan Cojocaru]

Hello,

I see that there's currently no xc_vcpu_pause() (there's only
xc_domain_pause(), which freezes all the VCPUs), but there's
XEN_DOMCTL_gdbsx_pausevcpu, which is only used by
tools/debugger/gdbsx/xg/xg_main.c.

Wouldn't it make sense to have a "regular" DOMCTL in libxc that the
debugger, as well as other clients, would make use of (said
xc_vcpu_pause() / xc_vcpu_unpause())?


Thanks,
Razvan


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

Re: Pausing / unpausing a single domain VCPU via libxc

[Dario Faggioli]

[-- Attachment #1.1: Type: text/plain, Size: 1126 bytes --]

[Adding tools people, and Andrew's which, according to git history,
 has touched the involved (Xen) code for last :-) ]

On Thu, 2017-01-26 at 14:08 +0200, Razvan Cojocaru wrote:
> Hello,
> 
> I see that there's currently no xc_vcpu_pause() (there's only
> xc_domain_pause(), which freezes all the VCPUs), but there's
> XEN_DOMCTL_gdbsx_pausevcpu, which is only used by
> tools/debugger/gdbsx/xg/xg_main.c.
> 
> Wouldn't it make sense to have a "regular" DOMCTL in libxc that the
> debugger, as well as other clients, would make use of (said
> xc_vcpu_pause() / xc_vcpu_unpause())?
> 
You mean you'd want to implement xc_vcpu_pause() by means of
the XEN_DOMCTL_gdbsx_pausevcpu?

What's the use case for that, and does it fit with the implementation
of said hypercall (which, e.g., requires that the domain is already
paused)?

Regards,
Dario
-- 
<<This happens because I choose it to happen!>> (Raistlin Majere)
-----------------------------------------------------------------
Dario Faggioli, Ph.D, http://about.me/dario.faggioli
Senior Software Engineer, Citrix Systems R&D Ltd., Cambridge (UK)

[-- Attachment #1.2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

[-- Attachment #2: Type: text/plain, Size: 127 bytes --]

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

Re: Pausing / unpausing a single domain VCPU via libxc

[Razvan Cojocaru]

On 01/26/2017 05:14 PM, Dario Faggioli wrote:
> [Adding tools people, and Andrew's which, according to git history,
>  has touched the involved (Xen) code for last :-) ]
> 
> On Thu, 2017-01-26 at 14:08 +0200, Razvan Cojocaru wrote:
>> Hello,
>>
>> I see that there's currently no xc_vcpu_pause() (there's only
>> xc_domain_pause(), which freezes all the VCPUs), but there's
>> XEN_DOMCTL_gdbsx_pausevcpu, which is only used by
>> tools/debugger/gdbsx/xg/xg_main.c.
>>
>> Wouldn't it make sense to have a "regular" DOMCTL in libxc that the
>> debugger, as well as other clients, would make use of (said
>> xc_vcpu_pause() / xc_vcpu_unpause())?
>>
> You mean you'd want to implement xc_vcpu_pause() by means of
> the XEN_DOMCTL_gdbsx_pausevcpu?
> 
> What's the use case for that, and does it fit with the implementation
> of said hypercall (which, e.g., requires that the domain is already
> paused)?

Not necessarily implement xc_vcpu_pause() by means of
XEN_DOMCTL_gdbsx_pausevcpu, it's just that from a design perspective
this seems rather specialised - the domctl name suggests that this is
only useful for debugging, and it also hardcodes an implementation for
calling that hypercall in xg_main.c, whereas pausing a single VCPU is a
generic operation that other clients, as well as the debugger, might
find worthwhile to do.

As for a use case, an introspection application might be interested in
briefly pausing a VCPU while it inspects data related to it.

Does XEN_DOMCTL_gdbsx_pausevcpu require that the whole domain is paused
(for longer that the duration of the actual hypercall)?


Thanks,
Razvan

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

Re: Pausing / unpausing a single domain VCPU via libxc

[Jan Beulich]

>>> On 26.01.17 at 13:08, <rcojocaru@bitdefender.com> wrote:
> I see that there's currently no xc_vcpu_pause() (there's only
> xc_domain_pause(), which freezes all the VCPUs), but there's
> XEN_DOMCTL_gdbsx_pausevcpu, which is only used by
> tools/debugger/gdbsx/xg/xg_main.c.
> 
> Wouldn't it make sense to have a "regular" DOMCTL in libxc that the
> debugger, as well as other clients, would make use of (said
> xc_vcpu_pause() / xc_vcpu_unpause())?

I could imagine the main reason for it to not exist being that using
it may confuse the guest (seeing one or more of its vCPU-s no
longer respond for some arbitrary time period).

Jan


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

Re: Pausing / unpausing a single domain VCPU via libxc

[Dario Faggioli]

[-- Attachment #1.1: Type: text/plain, Size: 2657 bytes --]

On Thu, 2017-01-26 at 17:42 +0200, Razvan Cojocaru wrote:
> On 01/26/2017 05:14 PM, Dario Faggioli wrote:
> > You mean you'd want to implement xc_vcpu_pause() by means of
> > the XEN_DOMCTL_gdbsx_pausevcpu?
> > 
> > What's the use case for that, and does it fit with the
> > implementation
> > of said hypercall (which, e.g., requires that the domain is already
> > paused)?
> 
> Not necessarily implement xc_vcpu_pause() by means of
> XEN_DOMCTL_gdbsx_pausevcpu, it's just that from a design perspective
> this seems rather specialised - the domctl name suggests that this is
> only useful for debugging, and it also hardcodes an implementation
> for
> calling that hypercall in xg_main.c, whereas pausing a single VCPU is
> a
> generic operation that other clients, as well as the debugger, might
> find worthwhile to do.
> 
And, in principle, I agree that it would make sense to have a generic
vCPU pause/unpause API.

Outside of principle --a.k.a., in practise-- I'm not sure whether it
really makes sense, and whether it would actually work.

I'm thinking of how happy a guest would be to be running with one of
it's (v)CPU frozen, and am wondering whether this could cause it to
panic, or behave funnily.

But I haven't tried, of course.

> Does XEN_DOMCTL_gdbsx_pausevcpu require that the whole domain is
> paused
> (for longer that the duration of the actual hypercall)?
> 
AFAICT, it does. See the implementation of the hypercall:

        if ( !d->controller_pause_count )
            break;
        ret = -EINVAL;
        if ( domctl->u.gdbsx_pauseunp_vcpu.vcpu >= d->max_vcpus ||
             (v = d->vcpu[domctl->u.gdbsx_pauseunp_vcpu.vcpu]) == NULL )
            break;
        ret = vcpu_pause_by_systemcontroller(v);

See commit 680d79f10 ("x86/gdbsx: invert preconditions for
XEN_DOMCTL_gdbsx_{,un}pausevcpu hypercalls":

    Revert back to how it was originally, i.e. the XEN_DOMCTL_gdbsx_{,un}pausevcpu
    hypercalls are only valid for a domain already paused by the system controller.

And see how it's used, e.g., as described in the comment above
xg_step() (or even in its actual code):

/*
 * Single step the given vcpu. This is achieved by pausing all but given vcpus,
 * setting the TF flag, let the domain run and pause, unpause all vcpus, and
 * clear TF flag on given vcpu.
 * Returns: 0 success
 */

Dario
-- 
<<This happens because I choose it to happen!>> (Raistlin Majere)
-----------------------------------------------------------------
Dario Faggioli, Ph.D, http://about.me/dario.faggioli
Senior Software Engineer, Citrix Systems R&D Ltd., Cambridge (UK)

[-- Attachment #1.2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

[-- Attachment #2: Type: text/plain, Size: 127 bytes --]

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

Re: Pausing / unpausing a single domain VCPU via libxc

[Razvan Cojocaru]

On 01/26/2017 06:00 PM, Dario Faggioli wrote:
> On Thu, 2017-01-26 at 17:42 +0200, Razvan Cojocaru wrote:
>> On 01/26/2017 05:14 PM, Dario Faggioli wrote:
>>> You mean you'd want to implement xc_vcpu_pause() by means of
>>> the XEN_DOMCTL_gdbsx_pausevcpu?
>>>
>>> What's the use case for that, and does it fit with the
>>> implementation
>>> of said hypercall (which, e.g., requires that the domain is already
>>> paused)?
>>
>> Not necessarily implement xc_vcpu_pause() by means of
>> XEN_DOMCTL_gdbsx_pausevcpu, it's just that from a design perspective
>> this seems rather specialised - the domctl name suggests that this is
>> only useful for debugging, and it also hardcodes an implementation
>> for
>> calling that hypercall in xg_main.c, whereas pausing a single VCPU is
>> a
>> generic operation that other clients, as well as the debugger, might
>> find worthwhile to do.
>>
> And, in principle, I agree that it would make sense to have a generic
> vCPU pause/unpause API.
> 
> Outside of principle --a.k.a., in practise-- I'm not sure whether it
> really makes sense, and whether it would actually work.
> 
> I'm thinking of how happy a guest would be to be running with one of
> it's (v)CPU frozen, and am wondering whether this could cause it to
> panic, or behave funnily.
> 
> But I haven't tried, of course.
> 
>> Does XEN_DOMCTL_gdbsx_pausevcpu require that the whole domain is
>> paused
>> (for longer that the duration of the actual hypercall)?
>>
> AFAICT, it does. See the implementation of the hypercall:
> 
>         if ( !d->controller_pause_count )
>             break;
>         ret = -EINVAL;
>         if ( domctl->u.gdbsx_pauseunp_vcpu.vcpu >= d->max_vcpus ||
>              (v = d->vcpu[domctl->u.gdbsx_pauseunp_vcpu.vcpu]) == NULL )
>             break;
>         ret = vcpu_pause_by_systemcontroller(v);
> 
> See commit 680d79f10 ("x86/gdbsx: invert preconditions for
> XEN_DOMCTL_gdbsx_{,un}pausevcpu hypercalls":
> 
>     Revert back to how it was originally, i.e. the XEN_DOMCTL_gdbsx_{,un}pausevcpu
>     hypercalls are only valid for a domain already paused by the system controller.
> 
> And see how it's used, e.g., as described in the comment above
> xg_step() (or even in its actual code):
> 
> /*
>  * Single step the given vcpu. This is achieved by pausing all but given vcpus,
>  * setting the TF flag, let the domain run and pause, unpause all vcpus, and
>  * clear TF flag on given vcpu.
>  * Returns: 0 success
>  */

Right. Thank you for clearing that up!


Thanks,
Razvan

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