Re: [PATCH v2 1/2] xen: fix a (latent) cpupool-related race during domain destroy

[Juergen Gross <jgross@suse.com>]

On 15/07/16 12:14, Dario Faggioli wrote:
> On Fri, 2016-07-15 at 11:38 +0200, Juergen Gross wrote:
>> Hmm, are you aware of commit bac6334b51d9bcfe57ecf4a4cb5288348fcf044a
>> which explicitly moved cpupool_rm_domain() at the place where you are
>> removing it again? Please verify that the scenario mentioned in the
>> description of that commit is still working with your patch.
>>
> Sorry, but I only partly see the problem.
> 
> In particular, I'm probably not fully understanding, from that commit
> changelog, what is the set of operations/command that I should run to
> check whether or not I reintroduced the issue back.

You need to create a domain in a cpupool and destroy it again while
some dom0 process still is holding a reference to it (resulting in a
zombie domain). Then try to destroy the cpupool.

> 
> What I did so far is as follows:
> 
> root@Zhaman:~# xl cpupool-list 
> Name               CPUs   Sched     Active   Domain count
> Pool-0              12    credit       y          1
> Pool-credit          4    credit       y          1
> root@Zhaman:~# xl list -c
> Name                                        ID   Mem VCPUs	State	Time(s)         Cpupool
> Domain-0                                     0  1019    16     r-----      34.5          Pool-0
> vm1                                          1  4096     4     -b----       9.7     Pool-credit
> root@Zhaman:~# xl cpupool-cpu-remove Pool-credit all
> libxl: error: libxl.c:6998:libxl_cpupool_cpuremove: Error removing cpu 9 from cpupool: Device or resource busy
> Some cpus may have not or only partially been removed from 'Pool-credit'.
> If a cpu can't be added to another cpupool, add it to 'Pool-credit' again and retry.
> root@Zhaman:~# xl cpupool-list -c
> Name               CPU list
> Pool-0             0,1,2,3,4,5,10,11,12,13,14,15
> Pool-credit        9
> root@Zhaman:~# xl shutdown vm1
> Shutting down domain 1
> root@Zhaman:~# xl cpupool-cpu-remove Pool-credit all
> root@Zhaman:~# xl cpupool-list -c
> Name               CPU list
> Pool-0             0,1,2,3,4,5,10,11,12,13,14,15
> Pool-credit
> 
> If (with vm1 still in Pool-credit), I do this, it indeed fails:
> 
> root@Zhaman:~# xl shutdown vm1 & xl cpupool-cpu-remove Pool-credit all
> [1] 3275
> Shutting down domain 2
> libxl: error: libxl.c:6998:libxl_cpupool_cpuremove: Error removing cpu 9 from cpupool: Device or resource busy
> Some cpus may have not or only partially been removed from 'Pool-credit'.
> If a cpu can't be added to another cpupool, add it to 'Pool-credit' again and retry.
> [1]+  Done                    xl shutdown vm1
> root@Zhaman:~# xl cpupool-list -c
> Name               CPU list
> Pool-0             0,1,2,3,4,5,10,11,12,13,14,15
> Pool-credit        9
> 
> But that does not look too strange to me, as it's entirely possible
> that the domain has not been moved yet, when we try to remove the last
> cpu. And in fact, after the domain has properly shutdown:
> 
> root@Zhaman:~# xl cpupool-cpu-remove Pool-credit all
> root@Zhaman:~# xl cpupool-list 
> Name               CPUs   Sched     Active   Domain count
> Pool-0              12    credit       y          1
> Pool-credit          0    credit       y          0
> 
> And in fact, looking at the code introduced by that commit, the
> important part, to me, seems to be the moving of the domain to
> cpupool0, which is indeed the right thing to do. OTOH, what I am seeing
> and fixing, happens (well, could happen) all the times, even when the
> domain being shutdown is already in cpupool0, and (as you say yourself
> in your changelog) there not such issue as removing the last cpu of
> cpupool0.
> 
> What am I missing?

The domain being a zombie domain might change the picture. Moving it to
cpupool0 was failing before my patch and it might do so again with your
patch applied.


Juergen

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