[PATCH] Adjust number of domains in cpupools when destroying domain

[PATCH] Adjust number of domains in cpupools when destroying domain

[Juergen Gross]

Commit bac6334b51d9bcfe57ecf4a4cb5288348fcf044a (move domain to
cpupool0 before destroying it) introduced an error in the accounting
of cpupools regarding the number of domains. The number of domains
is nor adjusted when a domain is moved to cpupool0 in kill_domain().

Correct this by introducing a cpupool function doing the move
instead of open coding it by calling sched_move_domain().

Signed-off-by: Juergen Gross <jgross@suse.com>
Tested-by: Dietmar Hahn <dietmar.hahn@ts.fujitsu.com>
Reviewed-by: Andrew Cooper <Andrew.Cooper3@citrix.com>
---
 xen/common/cpupool.c    | 47 +++++++++++++++++++++++++++++++++--------------
 xen/common/domain.c     |  2 +-
 xen/include/xen/sched.h |  1 +
 3 files changed, 35 insertions(+), 15 deletions(-)

diff --git a/xen/common/cpupool.c b/xen/common/cpupool.c
index 73249d3..a758a8b 100644
--- a/xen/common/cpupool.c
+++ b/xen/common/cpupool.c
@@ -225,6 +225,35 @@ static int cpupool_destroy(struct cpupool *c)
 }
 
 /*
+ * Move domain to another cpupool
+ */
+static int cpupool_move_domain_locked(struct domain *d, struct cpupool *c)
+{
+    int ret;
+
+    d->cpupool->n_dom--;
+    ret = sched_move_domain(d, c);
+    if ( ret )
+        d->cpupool->n_dom++;
+    else
+        c->n_dom++;
+
+    return ret;
+}
+int cpupool_move_domain(struct domain *d, struct cpupool *c)
+{
+    int ret;
+
+    spin_lock(&cpupool_lock);
+
+    ret = cpupool_move_domain_locked(d, c);
+
+    spin_unlock(&cpupool_lock);
+
+    return ret;
+}
+
+/*
  * assign a specific cpu to a cpupool
  * cpupool_lock must be held
  */
@@ -338,14 +367,9 @@ static int cpupool_unassign_cpu(struct cpupool *c, unsigned int cpu)
                 ret = -EBUSY;
                 break;
             }
-            c->n_dom--;
-            ret = sched_move_domain(d, cpupool0);
+            ret = cpupool_move_domain_locked(d, cpupool0);
             if ( ret )
-            {
-                c->n_dom++;
                 break;
-            }
-            cpupool0->n_dom++;
         }
         rcu_read_unlock(&domlist_read_lock);
         if ( ret )
@@ -613,16 +637,11 @@ int cpupool_do_sysctl(struct xen_sysctl_cpupool_op *op)
                         d->domain_id, op->cpupool_id);
         ret = -ENOENT;
         spin_lock(&cpupool_lock);
+
         c = cpupool_find_by_id(op->cpupool_id);
         if ( (c != NULL) && cpumask_weight(c->cpu_valid) )
-        {
-            d->cpupool->n_dom--;
-            ret = sched_move_domain(d, c);
-            if ( ret )
-                d->cpupool->n_dom++;
-            else
-                c->n_dom++;
-        }
+            ret = cpupool_move_domain_locked(d, c);
+
         spin_unlock(&cpupool_lock);
         cpupool_dprintk("cpupool move_domain(dom=%d)->pool=%d ret %d\n",
                         d->domain_id, op->cpupool_id, ret);
diff --git a/xen/common/domain.c b/xen/common/domain.c
index a3f51ec..4a62c1d 100644
--- a/xen/common/domain.c
+++ b/xen/common/domain.c
@@ -621,7 +621,7 @@ int domain_kill(struct domain *d)
                 rc = -EAGAIN;
             break;
         }
-        if ( sched_move_domain(d, cpupool0) )
+        if ( cpupool_move_domain(d, cpupool0) )
             return -EAGAIN;
         for_each_vcpu ( d, v )
             unmap_vcpu_info(v);
diff --git a/xen/include/xen/sched.h b/xen/include/xen/sched.h
index c5157e6..46fc6e3 100644
--- a/xen/include/xen/sched.h
+++ b/xen/include/xen/sched.h
@@ -871,6 +871,7 @@ struct cpupool *cpupool_get_by_id(int poolid);
 void cpupool_put(struct cpupool *pool);
 int cpupool_add_domain(struct domain *d, int poolid);
 void cpupool_rm_domain(struct domain *d);
+int cpupool_move_domain(struct domain *d, struct cpupool *c);
 int cpupool_do_sysctl(struct xen_sysctl_cpupool_op *op);
 void schedule_dump(struct cpupool *c);
 extern void dump_runq(unsigned char key);
-- 
2.1.2

Re: [PATCH] Adjust number of domains in cpupools when destroying domain

[George Dunlap]

On Wed, Nov 12, 2014 at 11:10 AM, Juergen Gross <jgross@suse.com> wrote:
> Commit bac6334b51d9bcfe57ecf4a4cb5288348fcf044a (move domain to
> cpupool0 before destroying it) introduced an error in the accounting
> of cpupools regarding the number of domains. The number of domains
> is nor adjusted when a domain is moved to cpupool0 in kill_domain().
>
> Correct this by introducing a cpupool function doing the move
> instead of open coding it by calling sched_move_domain().
>
> Signed-off-by: Juergen Gross <jgross@suse.com>
> Tested-by: Dietmar Hahn <dietmar.hahn@ts.fujitsu.com>
> Reviewed-by: Andrew Cooper <Andrew.Cooper3@citrix.com>

Acked-by: George Dunlap <george.dunlap@eu.citrix.com>

Re: [PATCH] Adjust number of domains in cpupools when destroying domain

[Konrad Rzeszutek Wilk]

On Wed, Nov 12, 2014 at 11:11:15AM +0000, George Dunlap wrote:
> On Wed, Nov 12, 2014 at 11:10 AM, Juergen Gross <jgross@suse.com> wrote:
> > Commit bac6334b51d9bcfe57ecf4a4cb5288348fcf044a (move domain to
> > cpupool0 before destroying it) introduced an error in the accounting
> > of cpupools regarding the number of domains. The number of domains
> > is nor adjusted when a domain is moved to cpupool0 in kill_domain().
> >
> > Correct this by introducing a cpupool function doing the move
> > instead of open coding it by calling sched_move_domain().
> >
> > Signed-off-by: Juergen Gross <jgross@suse.com>
> > Tested-by: Dietmar Hahn <dietmar.hahn@ts.fujitsu.com>
> > Reviewed-by: Andrew Cooper <Andrew.Cooper3@citrix.com>
> 
> Acked-by: George Dunlap <george.dunlap@eu.citrix.com>

Excellent!

Release-Acked-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>

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

Re: [PATCH] Adjust number of domains in cpupools when destroying domain

[Konrad Rzeszutek Wilk]

On Wed, Nov 12, 2014 at 12:10:02PM +0100, Juergen Gross wrote:
> Commit bac6334b51d9bcfe57ecf4a4cb5288348fcf044a (move domain to
> cpupool0 before destroying it) introduced an error in the accounting
> of cpupools regarding the number of domains. The number of domains
> is nor adjusted when a domain is moved to cpupool0 in kill_domain().

s/nor/not/
> 
> Correct this by introducing a cpupool function doing the move
> instead of open coding it by calling sched_move_domain().
> 
> Signed-off-by: Juergen Gross <jgross@suse.com>
> Tested-by: Dietmar Hahn <dietmar.hahn@ts.fujitsu.com>
> Reviewed-by: Andrew Cooper <Andrew.Cooper3@citrix.com>
> ---
>  xen/common/cpupool.c    | 47 +++++++++++++++++++++++++++++++++--------------
>  xen/common/domain.c     |  2 +-
>  xen/include/xen/sched.h |  1 +
>  3 files changed, 35 insertions(+), 15 deletions(-)
> 
> diff --git a/xen/common/cpupool.c b/xen/common/cpupool.c
> index 73249d3..a758a8b 100644
> --- a/xen/common/cpupool.c
> +++ b/xen/common/cpupool.c
> @@ -225,6 +225,35 @@ static int cpupool_destroy(struct cpupool *c)
>  }
>  
>  /*
> + * Move domain to another cpupool
> + */
> +static int cpupool_move_domain_locked(struct domain *d, struct cpupool *c)
> +{
> +    int ret;
> +
> +    d->cpupool->n_dom--;
> +    ret = sched_move_domain(d, c);
> +    if ( ret )
> +        d->cpupool->n_dom++;
> +    else
> +        c->n_dom++;
> +
> +    return ret;
> +}

\n ?

> +int cpupool_move_domain(struct domain *d, struct cpupool *c)
> +{
> +    int ret;
> +
> +    spin_lock(&cpupool_lock);
> +
> +    ret = cpupool_move_domain_locked(d, c);
> +
> +    spin_unlock(&cpupool_lock);
> +
> +    return ret;
> +}
> +
> +/*
>   * assign a specific cpu to a cpupool
>   * cpupool_lock must be held
>   */
> @@ -338,14 +367,9 @@ static int cpupool_unassign_cpu(struct cpupool *c, unsigned int cpu)
>                  ret = -EBUSY;
>                  break;
>              }
> -            c->n_dom--;
> -            ret = sched_move_domain(d, cpupool0);
> +            ret = cpupool_move_domain_locked(d, cpupool0);
>              if ( ret )
> -            {
> -                c->n_dom++;
>                  break;
> -            }
> -            cpupool0->n_dom++;
>          }
>          rcu_read_unlock(&domlist_read_lock);
>          if ( ret )
> @@ -613,16 +637,11 @@ int cpupool_do_sysctl(struct xen_sysctl_cpupool_op *op)
>                          d->domain_id, op->cpupool_id);
>          ret = -ENOENT;
>          spin_lock(&cpupool_lock);
> +
>          c = cpupool_find_by_id(op->cpupool_id);
>          if ( (c != NULL) && cpumask_weight(c->cpu_valid) )
> -        {
> -            d->cpupool->n_dom--;
> -            ret = sched_move_domain(d, c);
> -            if ( ret )
> -                d->cpupool->n_dom++;
> -            else
> -                c->n_dom++;
> -        }
> +            ret = cpupool_move_domain_locked(d, c);
> +
>          spin_unlock(&cpupool_lock);
>          cpupool_dprintk("cpupool move_domain(dom=%d)->pool=%d ret %d\n",
>                          d->domain_id, op->cpupool_id, ret);
> diff --git a/xen/common/domain.c b/xen/common/domain.c
> index a3f51ec..4a62c1d 100644
> --- a/xen/common/domain.c
> +++ b/xen/common/domain.c
> @@ -621,7 +621,7 @@ int domain_kill(struct domain *d)
>                  rc = -EAGAIN;
>              break;
>          }
> -        if ( sched_move_domain(d, cpupool0) )
> +        if ( cpupool_move_domain(d, cpupool0) )
>              return -EAGAIN;
>          for_each_vcpu ( d, v )
>              unmap_vcpu_info(v);
> diff --git a/xen/include/xen/sched.h b/xen/include/xen/sched.h
> index c5157e6..46fc6e3 100644
> --- a/xen/include/xen/sched.h
> +++ b/xen/include/xen/sched.h
> @@ -871,6 +871,7 @@ struct cpupool *cpupool_get_by_id(int poolid);
>  void cpupool_put(struct cpupool *pool);
>  int cpupool_add_domain(struct domain *d, int poolid);
>  void cpupool_rm_domain(struct domain *d);
> +int cpupool_move_domain(struct domain *d, struct cpupool *c);
>  int cpupool_do_sysctl(struct xen_sysctl_cpupool_op *op);
>  void schedule_dump(struct cpupool *c);
>  extern void dump_runq(unsigned char key);
> -- 
> 2.1.2
> 
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xen.org
> http://lists.xen.org/xen-devel

[PATCH] Adjust number of domains in cpupools when destroying domain

[Juergen Gross]

Commit bac6334b51d9bcfe57ecf4a4cb5288348fcf044a (move domain to
cpupool0 before destroying it) introduced an error in the accounting
of cpupools regarding the number of domains. The number of domains
is nor adjusted when a domain is moved to cpupool0 in kill_domain().

Correct this by introducing a cpupool function doing the move
instead of open coding it by calling sched_move_domain().

Signed-off-by: Juergen Gross <jgross@suse.com>
Tested-by: Dietmar Hahn <dietmar.hahn@ts.fujitsu.com>
---
 xen/common/cpupool.c    | 47 +++++++++++++++++++++++++++++++++--------------
 xen/common/domain.c     |  2 +-
 xen/include/xen/sched.h |  1 +
 3 files changed, 35 insertions(+), 15 deletions(-)

diff --git a/xen/common/cpupool.c b/xen/common/cpupool.c
index 73249d3..c6e3869 100644
--- a/xen/common/cpupool.c
+++ b/xen/common/cpupool.c
@@ -225,6 +225,35 @@ static int cpupool_destroy(struct cpupool *c)
 }
 
 /*
+ * Move domain to another cpupool
+ */
+static int cpupool_move_domain_unlocked(struct domain *d, struct cpupool *c)
+{
+    int ret;
+
+    d->cpupool->n_dom--;
+    ret = sched_move_domain(d, c);
+    if ( ret )
+        d->cpupool->n_dom++;
+    else
+        c->n_dom++;
+
+    return ret;
+}
+int cpupool_move_domain(struct domain *d, struct cpupool *c)
+{
+    int ret;
+
+    spin_lock(&cpupool_lock);
+
+    ret = cpupool_move_domain_unlocked(d, c);
+
+    spin_unlock(&cpupool_lock);
+
+    return ret;
+}
+
+/*
  * assign a specific cpu to a cpupool
  * cpupool_lock must be held
  */
@@ -338,14 +367,9 @@ static int cpupool_unassign_cpu(struct cpupool *c, unsigned int cpu)
                 ret = -EBUSY;
                 break;
             }
-            c->n_dom--;
-            ret = sched_move_domain(d, cpupool0);
+            ret = cpupool_move_domain_unlocked(d, cpupool0);
             if ( ret )
-            {
-                c->n_dom++;
                 break;
-            }
-            cpupool0->n_dom++;
         }
         rcu_read_unlock(&domlist_read_lock);
         if ( ret )
@@ -613,16 +637,11 @@ int cpupool_do_sysctl(struct xen_sysctl_cpupool_op *op)
                         d->domain_id, op->cpupool_id);
         ret = -ENOENT;
         spin_lock(&cpupool_lock);
+
         c = cpupool_find_by_id(op->cpupool_id);
         if ( (c != NULL) && cpumask_weight(c->cpu_valid) )
-        {
-            d->cpupool->n_dom--;
-            ret = sched_move_domain(d, c);
-            if ( ret )
-                d->cpupool->n_dom++;
-            else
-                c->n_dom++;
-        }
+            ret = cpupool_move_domain_unlocked(d, c);
+
         spin_unlock(&cpupool_lock);
         cpupool_dprintk("cpupool move_domain(dom=%d)->pool=%d ret %d\n",
                         d->domain_id, op->cpupool_id, ret);
diff --git a/xen/common/domain.c b/xen/common/domain.c
index a3f51ec..4a62c1d 100644
--- a/xen/common/domain.c
+++ b/xen/common/domain.c
@@ -621,7 +621,7 @@ int domain_kill(struct domain *d)
                 rc = -EAGAIN;
             break;
         }
-        if ( sched_move_domain(d, cpupool0) )
+        if ( cpupool_move_domain(d, cpupool0) )
             return -EAGAIN;
         for_each_vcpu ( d, v )
             unmap_vcpu_info(v);
diff --git a/xen/include/xen/sched.h b/xen/include/xen/sched.h
index c5157e6..46fc6e3 100644
--- a/xen/include/xen/sched.h
+++ b/xen/include/xen/sched.h
@@ -871,6 +871,7 @@ struct cpupool *cpupool_get_by_id(int poolid);
 void cpupool_put(struct cpupool *pool);
 int cpupool_add_domain(struct domain *d, int poolid);
 void cpupool_rm_domain(struct domain *d);
+int cpupool_move_domain(struct domain *d, struct cpupool *c);
 int cpupool_do_sysctl(struct xen_sysctl_cpupool_op *op);
 void schedule_dump(struct cpupool *c);
 extern void dump_runq(unsigned char key);
-- 
2.1.2

Re: [PATCH] Adjust number of domains in cpupools when destroying domain

[Andrew Cooper]

On 12/11/14 10:40, Juergen Gross wrote:
> Commit bac6334b51d9bcfe57ecf4a4cb5288348fcf044a (move domain to
> cpupool0 before destroying it) introduced an error in the accounting
> of cpupools regarding the number of domains. The number of domains
> is nor adjusted when a domain is moved to cpupool0 in kill_domain().
>
> Correct this by introducing a cpupool function doing the move
> instead of open coding it by calling sched_move_domain().
>
> Signed-off-by: Juergen Gross <jgross@suse.com>
> Tested-by: Dietmar Hahn <dietmar.hahn@ts.fujitsu.com>
> ---
>  xen/common/cpupool.c    | 47 +++++++++++++++++++++++++++++++++--------------
>  xen/common/domain.c     |  2 +-
>  xen/include/xen/sched.h |  1 +
>  3 files changed, 35 insertions(+), 15 deletions(-)
>
> diff --git a/xen/common/cpupool.c b/xen/common/cpupool.c
> index 73249d3..c6e3869 100644
> --- a/xen/common/cpupool.c
> +++ b/xen/common/cpupool.c
> @@ -225,6 +225,35 @@ static int cpupool_destroy(struct cpupool *c)
>  }
>  
>  /*
> + * Move domain to another cpupool
> + */
> +static int cpupool_move_domain_unlocked(struct domain *d, struct cpupool *c)

This isn't an unlocked function.  It is strictly called with the
cpupool_lock held.  Per prevailing style, it should be named
"__cpupool_move_domain()".

> +{
> +    int ret;
> +
> +    d->cpupool->n_dom--;
> +    ret = sched_move_domain(d, c);
> +    if ( ret )
> +        d->cpupool->n_dom++;
> +    else
> +        c->n_dom++;
> +
> +    return ret;
> +}

Newline here please.

Once these two issues are fixed, content Reviewed-by: Andrew Cooper
<andrew.cooper3@citrix.com>

> +int cpupool_move_domain(struct domain *d, struct cpupool *c)
> +{
> +    int ret;
> +
> +    spin_lock(&cpupool_lock);
> +
> +    ret = cpupool_move_domain_unlocked(d, c);
> +
> +    spin_unlock(&cpupool_lock);
> +
> +    return ret;
> +}
> +
> +/*
>   * assign a specific cpu to a cpupool
>   * cpupool_lock must be held
>   */
> @@ -338,14 +367,9 @@ static int cpupool_unassign_cpu(struct cpupool *c, unsigned int cpu)
>                  ret = -EBUSY;
>                  break;
>              }
> -            c->n_dom--;
> -            ret = sched_move_domain(d, cpupool0);
> +            ret = cpupool_move_domain_unlocked(d, cpupool0);
>              if ( ret )
> -            {
> -                c->n_dom++;
>                  break;
> -            }
> -            cpupool0->n_dom++;
>          }
>          rcu_read_unlock(&domlist_read_lock);
>          if ( ret )
> @@ -613,16 +637,11 @@ int cpupool_do_sysctl(struct xen_sysctl_cpupool_op *op)
>                          d->domain_id, op->cpupool_id);
>          ret = -ENOENT;
>          spin_lock(&cpupool_lock);
> +
>          c = cpupool_find_by_id(op->cpupool_id);
>          if ( (c != NULL) && cpumask_weight(c->cpu_valid) )
> -        {
> -            d->cpupool->n_dom--;
> -            ret = sched_move_domain(d, c);
> -            if ( ret )
> -                d->cpupool->n_dom++;
> -            else
> -                c->n_dom++;
> -        }
> +            ret = cpupool_move_domain_unlocked(d, c);
> +
>          spin_unlock(&cpupool_lock);
>          cpupool_dprintk("cpupool move_domain(dom=%d)->pool=%d ret %d\n",
>                          d->domain_id, op->cpupool_id, ret);
> diff --git a/xen/common/domain.c b/xen/common/domain.c
> index a3f51ec..4a62c1d 100644
> --- a/xen/common/domain.c
> +++ b/xen/common/domain.c
> @@ -621,7 +621,7 @@ int domain_kill(struct domain *d)
>                  rc = -EAGAIN;
>              break;
>          }
> -        if ( sched_move_domain(d, cpupool0) )
> +        if ( cpupool_move_domain(d, cpupool0) )
>              return -EAGAIN;
>          for_each_vcpu ( d, v )
>              unmap_vcpu_info(v);
> diff --git a/xen/include/xen/sched.h b/xen/include/xen/sched.h
> index c5157e6..46fc6e3 100644
> --- a/xen/include/xen/sched.h
> +++ b/xen/include/xen/sched.h
> @@ -871,6 +871,7 @@ struct cpupool *cpupool_get_by_id(int poolid);
>  void cpupool_put(struct cpupool *pool);
>  int cpupool_add_domain(struct domain *d, int poolid);
>  void cpupool_rm_domain(struct domain *d);
> +int cpupool_move_domain(struct domain *d, struct cpupool *c);
>  int cpupool_do_sysctl(struct xen_sysctl_cpupool_op *op);
>  void schedule_dump(struct cpupool *c);
>  extern void dump_runq(unsigned char key);

Re: [PATCH] Adjust number of domains in cpupools when destroying domain

[Jan Beulich]

>>> On 12.11.14 at 11:46, <andrew.cooper3@citrix.com> wrote:
> On 12/11/14 10:40, Juergen Gross wrote:
>> --- a/xen/common/cpupool.c
>> +++ b/xen/common/cpupool.c
>> @@ -225,6 +225,35 @@ static int cpupool_destroy(struct cpupool *c)
>>  }
>>  
>>  /*
>> + * Move domain to another cpupool
>> + */
>> +static int cpupool_move_domain_unlocked(struct domain *d, struct cpupool *c)
> 
> This isn't an unlocked function.  It is strictly called with the
> cpupool_lock held.  Per prevailing style, it should be named
> "__cpupool_move_domain()".

I generally disagree to this, even if this is the prevailing style.
Double-underscore prefixed names shouldn't be used at all in our
code, as they're being reserved by the C library standard (and
the compiler is free to introduce library calls named such). But
the question of course is valid why the function name says
"unlocked" when it's always being called with the lock held -
"locked" would seem more natural in this case. But in the end
Jürgen is the maintainer of that code, so he decides.

Jan

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

Re: [PATCH] Adjust number of domains in cpupools when destroying domain

[Jürgen Groß]

On 11/12/2014 11:46 AM, Andrew Cooper wrote:
> On 12/11/14 10:40, Juergen Gross wrote:
>> Commit bac6334b51d9bcfe57ecf4a4cb5288348fcf044a (move domain to
>> cpupool0 before destroying it) introduced an error in the accounting
>> of cpupools regarding the number of domains. The number of domains
>> is nor adjusted when a domain is moved to cpupool0 in kill_domain().
>>
>> Correct this by introducing a cpupool function doing the move
>> instead of open coding it by calling sched_move_domain().
>>
>> Signed-off-by: Juergen Gross <jgross@suse.com>
>> Tested-by: Dietmar Hahn <dietmar.hahn@ts.fujitsu.com>
>> ---
>>   xen/common/cpupool.c    | 47 +++++++++++++++++++++++++++++++++--------------
>>   xen/common/domain.c     |  2 +-
>>   xen/include/xen/sched.h |  1 +
>>   3 files changed, 35 insertions(+), 15 deletions(-)
>>
>> diff --git a/xen/common/cpupool.c b/xen/common/cpupool.c
>> index 73249d3..c6e3869 100644
>> --- a/xen/common/cpupool.c
>> +++ b/xen/common/cpupool.c
>> @@ -225,6 +225,35 @@ static int cpupool_destroy(struct cpupool *c)
>>   }
>>
>>   /*
>> + * Move domain to another cpupool
>> + */
>> +static int cpupool_move_domain_unlocked(struct domain *d, struct cpupool *c)
>
> This isn't an unlocked function.  It is strictly called with the
> cpupool_lock held.  Per prevailing style, it should be named
> "__cpupool_move_domain()".

Umpf. Fingers faster than brain. :-)

>
>> +{
>> +    int ret;
>> +
>> +    d->cpupool->n_dom--;
>> +    ret = sched_move_domain(d, c);
>> +    if ( ret )
>> +        d->cpupool->n_dom++;
>> +    else
>> +        c->n_dom++;
>> +
>> +    return ret;
>> +}
>
> Newline here please.
>
> Once these two issues are fixed, content Reviewed-by: Andrew Cooper
> <andrew.cooper3@citrix.com>
>
>> +int cpupool_move_domain(struct domain *d, struct cpupool *c)
>> +{
>> +    int ret;
>> +
>> +    spin_lock(&cpupool_lock);
>> +
>> +    ret = cpupool_move_domain_unlocked(d, c);
>> +
>> +    spin_unlock(&cpupool_lock);
>> +
>> +    return ret;
>> +}
>> +
>> +/*
>>    * assign a specific cpu to a cpupool
>>    * cpupool_lock must be held
>>    */
>> @@ -338,14 +367,9 @@ static int cpupool_unassign_cpu(struct cpupool *c, unsigned int cpu)
>>                   ret = -EBUSY;
>>                   break;
>>               }
>> -            c->n_dom--;
>> -            ret = sched_move_domain(d, cpupool0);
>> +            ret = cpupool_move_domain_unlocked(d, cpupool0);
>>               if ( ret )
>> -            {
>> -                c->n_dom++;
>>                   break;
>> -            }
>> -            cpupool0->n_dom++;
>>           }
>>           rcu_read_unlock(&domlist_read_lock);
>>           if ( ret )
>> @@ -613,16 +637,11 @@ int cpupool_do_sysctl(struct xen_sysctl_cpupool_op *op)
>>                           d->domain_id, op->cpupool_id);
>>           ret = -ENOENT;
>>           spin_lock(&cpupool_lock);
>> +
>>           c = cpupool_find_by_id(op->cpupool_id);
>>           if ( (c != NULL) && cpumask_weight(c->cpu_valid) )
>> -        {
>> -            d->cpupool->n_dom--;
>> -            ret = sched_move_domain(d, c);
>> -            if ( ret )
>> -                d->cpupool->n_dom++;
>> -            else
>> -                c->n_dom++;
>> -        }
>> +            ret = cpupool_move_domain_unlocked(d, c);
>> +
>>           spin_unlock(&cpupool_lock);
>>           cpupool_dprintk("cpupool move_domain(dom=%d)->pool=%d ret %d\n",
>>                           d->domain_id, op->cpupool_id, ret);
>> diff --git a/xen/common/domain.c b/xen/common/domain.c
>> index a3f51ec..4a62c1d 100644
>> --- a/xen/common/domain.c
>> +++ b/xen/common/domain.c
>> @@ -621,7 +621,7 @@ int domain_kill(struct domain *d)
>>                   rc = -EAGAIN;
>>               break;
>>           }
>> -        if ( sched_move_domain(d, cpupool0) )
>> +        if ( cpupool_move_domain(d, cpupool0) )
>>               return -EAGAIN;
>>           for_each_vcpu ( d, v )
>>               unmap_vcpu_info(v);
>> diff --git a/xen/include/xen/sched.h b/xen/include/xen/sched.h
>> index c5157e6..46fc6e3 100644
>> --- a/xen/include/xen/sched.h
>> +++ b/xen/include/xen/sched.h
>> @@ -871,6 +871,7 @@ struct cpupool *cpupool_get_by_id(int poolid);
>>   void cpupool_put(struct cpupool *pool);
>>   int cpupool_add_domain(struct domain *d, int poolid);
>>   void cpupool_rm_domain(struct domain *d);
>> +int cpupool_move_domain(struct domain *d, struct cpupool *c);
>>   int cpupool_do_sysctl(struct xen_sysctl_cpupool_op *op);
>>   void schedule_dump(struct cpupool *c);
>>   extern void dump_runq(unsigned char key);
>
>
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xen.org
> http://lists.xen.org/xen-devel
>

Re: [PATCH] Adjust number of domains in cpupools when destroying domain

[George Dunlap]

On Wed, Nov 12, 2014 at 10:40 AM, Juergen Gross <jgross@suse.com> wrote:
> Commit bac6334b51d9bcfe57ecf4a4cb5288348fcf044a (move domain to
> cpupool0 before destroying it) introduced an error in the accounting
> of cpupools regarding the number of domains. The number of domains
> is nor adjusted when a domain is moved to cpupool0 in kill_domain().
>
> Correct this by introducing a cpupool function doing the move
> instead of open coding it by calling sched_move_domain().
>
> Signed-off-by: Juergen Gross <jgross@suse.com>
> Tested-by: Dietmar Hahn <dietmar.hahn@ts.fujitsu.com>

Juergen / Dietmar -- do either of you have a reasonably complete set
of tests for cpupools?  It seems like even basic corner cases (like
shutting down a domain in a pool and then destroying a pool) aren't
being tested.

It would be really good if someone could try to do a more thorough
test before the 4.5 release.  It shouldn't be too hard to write a
script to test a lot of this functionality programmatically.

 -George

Re: [PATCH] Adjust number of domains in cpupools when destroying domain

[Juergen Gross]

On 11/12/2014 12:10 PM, George Dunlap wrote:
> On Wed, Nov 12, 2014 at 10:40 AM, Juergen Gross <jgross@suse.com> wrote:
>> Commit bac6334b51d9bcfe57ecf4a4cb5288348fcf044a (move domain to
>> cpupool0 before destroying it) introduced an error in the accounting
>> of cpupools regarding the number of domains. The number of domains
>> is nor adjusted when a domain is moved to cpupool0 in kill_domain().
>>
>> Correct this by introducing a cpupool function doing the move
>> instead of open coding it by calling sched_move_domain().
>>
>> Signed-off-by: Juergen Gross <jgross@suse.com>
>> Tested-by: Dietmar Hahn <dietmar.hahn@ts.fujitsu.com>
>
> Juergen / Dietmar -- do either of you have a reasonably complete set
> of tests for cpupools?  It seems like even basic corner cases (like
> shutting down a domain in a pool and then destroying a pool) aren't
> being tested.
>
> It would be really good if someone could try to do a more thorough
> test before the 4.5 release.  It shouldn't be too hard to write a
> script to test a lot of this functionality programmatically.

For the xm toolstack we had some tests at Fujitsu. Dietmar, you could
ask Lutz for advice. He might still have the scripts somewhere. They
should be easily adaptable to xl. In case you don't have time to try
them would you send them to me?

Juergen