Re: [PATCH v12 for-xen-4.5 11/20] x86/VPMU: Interface for setting PMU mode and flags

[Boris Ostrovsky <boris.ostrovsky@oracle.com>]

On 09/29/2014 09:25 AM, Dietmar Hahn wrote:
> +static int vpmu_force_context_switch(void)
> +{
> +    unsigned i, j, allbutself_num, mycpu;
> +    static s_time_t start, now;
> +    struct tasklet **sync_task;
> +    struct vcpu *curr_vcpu = current;
> +    int ret = 0;
> +
> +    allbutself_num = num_online_cpus() - 1;
> +
> +    sync_task = xzalloc_array(struct tasklet *, allbutself_num);
> +    if ( !sync_task )
> +    {
> +        printk(XENLOG_WARNING "vpmu_force_context_switch: out of memory\n");
> +        return -ENOMEM;
> +    }
> +
> +    for ( i = 0; i < allbutself_num; i++ )
> +    {
> +        sync_task[i] = xmalloc(struct tasklet);
> +        if ( sync_task[i] == NULL )
> +        {
> +            printk(XENLOG_WARNING "vpmu_force_context_switch: out of memory\n");
> +            ret = -ENOMEM;
> +            goto out;
> +        }
> +        tasklet_init(sync_task[i], vpmu_sched_checkin, 0);
> Only a question of understanding.
> Is there a special reason not to use a single memory allocation
> except for memory fragmentation on systems with a large number of cpus?
>
>       struct tasklet *sync_task;
>       sync_task = xmalloc(sizeof(struct tasklet) * allbutself_num);


Exactly because of fragmentation -- this will avoid asking for more than 
a page during runtime. I, in fact, originally had it allocated as a 
single chunk, just as you suggested above, but Jan asked this to be 
split into smaller, sub-page pieces.

-boris