[Xen-devel] [BUG] xenstat_vcpu_ns returns invalid value

[Xen-devel] [BUG] xenstat_vcpu_ns returns invalid value

[Richard Kojedzinszky]

Dear Xen Devs,

commit 2529c850ea48f036727ca2f148caed89391311b8 introduces the 
XEN_RUNSTATE_UPDATE marker bit, which is not handled in
vcpu_runstate_get() in xen/common/schedule.c. Relevant code:

     delta = NOW() - runstate->state_entry_time;
     if ( delta > 0 )
         runstate->time[runstate->state] += delta;

If state_entry_time has the bit set, it will be a negative value, so 
delta will be greater than 0, actually the marker bit
will appear in runstate->time too. This causes the MSB bit set in the 
return of xenstat_vcpu_ns(). Is it true, that when reading these values 
through xenstat, the user should take care of this bit, and reread, or 
is it a bug in the relevant code?

I am using xenstat.h interface, where I can only request statistics for 
the whole node. So, basically, walking through all vcpu of all domain, 
and checking that no one
contains this bit seems a bit ugly solution.

Thanks in advance,
Richard Kojedzinszky

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

Re: [Xen-devel] [BUG] xenstat_vcpu_ns returns invalid value

[Andrew Cooper]

On 09/01/2020 19:40, Richard Kojedzinszky wrote:
> Dear Xen Devs,
>
> commit 2529c850ea48f036727ca2f148caed89391311b8 introduces the
> XEN_RUNSTATE_UPDATE marker bit, which is not handled in
> vcpu_runstate_get() in xen/common/schedule.c. Relevant code:
>
>     delta = NOW() - runstate->state_entry_time;
>     if ( delta > 0 )
>         runstate->time[runstate->state] += delta;

That was found and fixed a while ago.  c/s f28c4c4c10 "sched: don't let
XEN_RUNSTATE_UPDATE leak into vcpu_runstate_get()".

No changes in userspace should be necessary, although you might need to
pester your distro for backports.

~Andrew

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

Re: [Xen-devel] [BUG] xenstat_vcpu_ns returns invalid value

[Richard Kojedzinszky]

2020-01-09 20:50 időpontban Andrew Cooper ezt írta:
> On 09/01/2020 19:40, Richard Kojedzinszky wrote:
>> Dear Xen Devs,
>> 
>> commit 2529c850ea48f036727ca2f148caed89391311b8 introduces the
>> XEN_RUNSTATE_UPDATE marker bit, which is not handled in
>> vcpu_runstate_get() in xen/common/schedule.c. Relevant code:
>> 
>>     delta = NOW() - runstate->state_entry_time;
>>     if ( delta > 0 )
>>         runstate->time[runstate->state] += delta;
> 
> That was found and fixed a while ago.  c/s f28c4c4c10 "sched: don't let
> XEN_RUNSTATE_UPDATE leak into vcpu_runstate_get()".
> 
> No changes in userspace should be necessary, although you might need to
> pester your distro for backports.
> 
> ~Andrew

Hi,

Thanks for the quick reply,

Then, as it is just a leak, until a backport arrives to my distro, I can 
mask out that bit from the results with no other side-effects, am I 
right?

Regards,
Richard Kojedzinszky

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

Re: [Xen-devel] [BUG] xenstat_vcpu_ns returns invalid value

[Andrew Cooper]

On 09/01/2020 20:09, Richard Kojedzinszky wrote:
> 2020-01-09 20:50 időpontban Andrew Cooper ezt írta:
>> On 09/01/2020 19:40, Richard Kojedzinszky wrote:
>>> Dear Xen Devs,
>>>
>>> commit 2529c850ea48f036727ca2f148caed89391311b8 introduces the
>>> XEN_RUNSTATE_UPDATE marker bit, which is not handled in
>>> vcpu_runstate_get() in xen/common/schedule.c. Relevant code:
>>>
>>>     delta = NOW() - runstate->state_entry_time;
>>>     if ( delta > 0 )
>>>         runstate->time[runstate->state] += delta;
>>
>> That was found and fixed a while ago.  c/s f28c4c4c10 "sched: don't let
>> XEN_RUNSTATE_UPDATE leak into vcpu_runstate_get()".
>>
>> No changes in userspace should be necessary, although you might need to
>> pester your distro for backports.
>>
>> ~Andrew
>
> Hi,
>
> Thanks for the quick reply,
>
> Then, as it is just a leak, until a backport arrives to my distro, I
> can mask out that bit from the results with no other side-effects, am
> I right?

Yes - that should be safe.

~Andrew

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

Re: [Xen-devel] [BUG] xenstat_vcpu_ns returns invalid value

[Richard Kojedzinszky]

2020-01-09 21:10 időpontban Andrew Cooper ezt írta:
> On 09/01/2020 20:09, Richard Kojedzinszky wrote:
>> 2020-01-09 20:50 időpontban Andrew Cooper ezt írta:
>>> On 09/01/2020 19:40, Richard Kojedzinszky wrote:
>>>> Dear Xen Devs,
>>>> 
>>>> commit 2529c850ea48f036727ca2f148caed89391311b8 introduces the
>>>> XEN_RUNSTATE_UPDATE marker bit, which is not handled in
>>>> vcpu_runstate_get() in xen/common/schedule.c. Relevant code:
>>>> 
>>>>     delta = NOW() - runstate->state_entry_time;
>>>>     if ( delta > 0 )
>>>>         runstate->time[runstate->state] += delta;
>>> 
>>> That was found and fixed a while ago.  c/s f28c4c4c10 "sched: don't 
>>> let
>>> XEN_RUNSTATE_UPDATE leak into vcpu_runstate_get()".
>>> 
>>> No changes in userspace should be necessary, although you might need 
>>> to
>>> pester your distro for backports.
>>> 
>>> ~Andrew
>> 
>> Hi,
>> 
>> Thanks for the quick reply,
>> 
>> Then, as it is just a leak, until a backport arrives to my distro, I
>> can mask out that bit from the results with no other side-effects, am
>> I right?
> 
> Yes - that should be safe.
> 
> ~Andrew

Hi,

Many thanks, I will do that.

Regards,
Richard

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