[PATCH v2] core-parking: shrink core_parking_cpunum[]

[PATCH v2] core-parking: shrink core_parking_cpunum[]

[Jan Beulich]

This NR_CPUS-dimensioned array is likely unused on most installations.
Therefore it is especially wasteful for it to consume more space than
really needed. Allocate it dynamically ahead of registering the hooks.

Further the array having all fields set to -1 is actually useless. Nothing
relies on it, and core_parking_remove() doesn't restore the sentinel for
vacated slots. Drop the initializers altogether, rather than replacing
them.

Also take the opportunity and update an adjacent variable's type, where
a fixed-width type was pretty clearly inappropriate to use.

Finally drop the redundant initializer from core_parking_init().

Signed-off-by: Jan Beulich <jbeulich@suse.com>
---
I assume there is a reason this is acting (mostly) as a LIFO. Else a
simple cpumask_t would suffice.

num_possible_cpus() would be the more correct thing to use as array
dimension, yet we don't maintain cpu_possible_map on x86.
---
v2: Use dynamic allocation.

--- a/xen/common/core_parking.c
+++ b/xen/common/core_parking.c
@@ -20,6 +20,7 @@
 #include <xen/cpumask.h>
 #include <xen/init.h>
 #include <xen/param.h>
+#include <xen/xvmalloc.h>
 
 #include <asm/smp.h>
 
@@ -27,8 +28,8 @@
 #define CORE_PARKING_DECREMENT 2
 
 static DEFINE_SPINLOCK(accounting_lock);
-static uint32_t cur_idle_nums;
-static unsigned int core_parking_cpunum[NR_CPUS] = {[0 ... NR_CPUS-1] = -1};
+static unsigned int cur_idle_nums;
+static unsigned int *__ro_after_init core_parking_cpunum;
 
 struct cp_policy {
     char name[30];
@@ -188,7 +189,7 @@ long cf_check core_parking_helper(void *
             return ret;
 
         spin_lock(&accounting_lock);
-        BUG_ON(cur_idle_nums >= ARRAY_SIZE(core_parking_cpunum));
+        BUG_ON(cur_idle_nums >= nr_cpu_ids);
         core_parking_cpunum[cur_idle_nums++] = cpu;
         spin_unlock(&accounting_lock);
     }
@@ -263,9 +264,12 @@ static int __init register_core_parking_
 
 static int __init cf_check core_parking_init(void)
 {
-    int ret = 0;
+    int ret;
 
-    if ( core_parking_controller == PERFORMANCE_FIRST )
+    core_parking_cpunum = xvzalloc_array(unsigned int, nr_cpu_ids);
+    if ( !core_parking_cpunum )
+        ret = -ENOMEM;
+    else if ( core_parking_controller == PERFORMANCE_FIRST )
         ret = register_core_parking_policy(&performance_first);
     else
         ret = register_core_parking_policy(&power_first);

Re: [PATCH v2] core-parking: shrink core_parking_cpunum[]

[Roger Pau Monné]

You possibly want to adjust the subject, instead of shrink I would use
"dynamically allocate" or similar.

On Mon, Mar 09, 2026 at 04:35:34PM +0100, Jan Beulich wrote:
> This NR_CPUS-dimensioned array is likely unused on most installations.
> Therefore it is especially wasteful for it to consume more space than
> really needed. Allocate it dynamically ahead of registering the hooks.
> 
> Further the array having all fields set to -1 is actually useless. Nothing
> relies on it, and core_parking_remove() doesn't restore the sentinel for
> vacated slots. Drop the initializers altogether, rather than replacing
> them.
> 
> Also take the opportunity and update an adjacent variable's type, where
> a fixed-width type was pretty clearly inappropriate to use.
> 
> Finally drop the redundant initializer from core_parking_init().
> 
> Signed-off-by: Jan Beulich <jbeulich@suse.com>
> ---
> I assume there is a reason this is acting (mostly) as a LIFO. Else a
> simple cpumask_t would suffice.
> 
> num_possible_cpus() would be the more correct thing to use as array
> dimension, yet we don't maintain cpu_possible_map on x86.
> ---
> v2: Use dynamic allocation.
> 
> --- a/xen/common/core_parking.c
> +++ b/xen/common/core_parking.c
> @@ -20,6 +20,7 @@
>  #include <xen/cpumask.h>
>  #include <xen/init.h>
>  #include <xen/param.h>
> +#include <xen/xvmalloc.h>
>  
>  #include <asm/smp.h>
>  
> @@ -27,8 +28,8 @@
>  #define CORE_PARKING_DECREMENT 2
>  
>  static DEFINE_SPINLOCK(accounting_lock);
> -static uint32_t cur_idle_nums;
> -static unsigned int core_parking_cpunum[NR_CPUS] = {[0 ... NR_CPUS-1] = -1};
> +static unsigned int cur_idle_nums;
> +static unsigned int *__ro_after_init core_parking_cpunum;

Don't you need some kind of check in core_parking_remove() to prevent a
NULL pointer dereference if core_parking_cpunum hasn't been allocated?

Callers of XEN_SYSCTL_cpu_hotplug can set fn = smt_up_down_helper, and
that would call core_parking_remove().  core_parking_helper() already
contains a check that prevents accessing core_parking_cpunum if no
policy has been registered.

Thanks, Roger.

Re: [PATCH v2] core-parking: shrink core_parking_cpunum[]

[Jan Beulich]

On 10.03.2026 10:07, Roger Pau Monné wrote:
> You possibly want to adjust the subject, instead of shrink I would use
> "dynamically allocate" or similar.

I've changed it, albeit the goal really is the shrinking.

> On Mon, Mar 09, 2026 at 04:35:34PM +0100, Jan Beulich wrote:
>> --- a/xen/common/core_parking.c
>> +++ b/xen/common/core_parking.c
>> @@ -20,6 +20,7 @@
>>  #include <xen/cpumask.h>
>>  #include <xen/init.h>
>>  #include <xen/param.h>
>> +#include <xen/xvmalloc.h>
>>  
>>  #include <asm/smp.h>
>>  
>> @@ -27,8 +28,8 @@
>>  #define CORE_PARKING_DECREMENT 2
>>  
>>  static DEFINE_SPINLOCK(accounting_lock);
>> -static uint32_t cur_idle_nums;
>> -static unsigned int core_parking_cpunum[NR_CPUS] = {[0 ... NR_CPUS-1] = -1};
>> +static unsigned int cur_idle_nums;
>> +static unsigned int *__ro_after_init core_parking_cpunum;
> 
> Don't you need some kind of check in core_parking_remove() to prevent a
> NULL pointer dereference if core_parking_cpunum hasn't been allocated?
> 
> Callers of XEN_SYSCTL_cpu_hotplug can set fn = smt_up_down_helper, and
> that would call core_parking_remove().  core_parking_helper() already
> contains a check that prevents accessing core_parking_cpunum if no
> policy has been registered.

Because of this check, cur_idle_nums can never become non-zero when the
array couldn't be allocated. Hence core_parking_remove() will be a
somewhat expensive no-op in that case, with no deref of core_parking_cpunum.

Jan

Re: [PATCH v2] core-parking: shrink core_parking_cpunum[]

[Roger Pau Monné]

On Tue, Mar 10, 2026 at 10:20:53AM +0100, Jan Beulich wrote:
> On 10.03.2026 10:07, Roger Pau Monné wrote:
> > You possibly want to adjust the subject, instead of shrink I would use
> > "dynamically allocate" or similar.
> 
> I've changed it, albeit the goal really is the shrinking.
> 
> > On Mon, Mar 09, 2026 at 04:35:34PM +0100, Jan Beulich wrote:
> >> --- a/xen/common/core_parking.c
> >> +++ b/xen/common/core_parking.c
> >> @@ -20,6 +20,7 @@
> >>  #include <xen/cpumask.h>
> >>  #include <xen/init.h>
> >>  #include <xen/param.h>
> >> +#include <xen/xvmalloc.h>
> >>  
> >>  #include <asm/smp.h>
> >>  
> >> @@ -27,8 +28,8 @@
> >>  #define CORE_PARKING_DECREMENT 2
> >>  
> >>  static DEFINE_SPINLOCK(accounting_lock);
> >> -static uint32_t cur_idle_nums;
> >> -static unsigned int core_parking_cpunum[NR_CPUS] = {[0 ... NR_CPUS-1] = -1};
> >> +static unsigned int cur_idle_nums;
> >> +static unsigned int *__ro_after_init core_parking_cpunum;
> > 
> > Don't you need some kind of check in core_parking_remove() to prevent a
> > NULL pointer dereference if core_parking_cpunum hasn't been allocated?
> > 
> > Callers of XEN_SYSCTL_cpu_hotplug can set fn = smt_up_down_helper, and
> > that would call core_parking_remove().  core_parking_helper() already
> > contains a check that prevents accessing core_parking_cpunum if no
> > policy has been registered.
> 
> Because of this check, cur_idle_nums can never become non-zero when the
> array couldn't be allocated. Hence core_parking_remove() will be a
> somewhat expensive no-op in that case, with no deref of core_parking_cpunum.

Oh, I see, both loops in core_parking_remove() are bounded to
cur_idle_nums.

Reviewed-by: Roger Pau Monné <roger.pau@citrix.com>

Thanks, Roger.

Re: [PATCH v2] core-parking: shrink core_parking_cpunum[]

[Jan Beulich]

On 10.03.2026 10:20, Jan Beulich wrote:
> On 10.03.2026 10:07, Roger Pau Monné wrote:
>> You possibly want to adjust the subject, instead of shrink I would use
>> "dynamically allocate" or similar.
> 
> I've changed it, albeit the goal really is the shrinking.

And it's only now that I notice that I committed the patch with its old
title, sorry. I failed to properly propagate the adjustment.

Jan