On 3/29/19 11:12 AM, Jan Beulich wrote:
On 29.03.19 at 14:42, wrote:
>> On 3/29/19 4:54 AM, Jan Beulich wrote:
>> On 28.03.19 at 17:50, wrote:
Given especially xen_pv_smp_prepare_cpus(), I think re-working proper
setting of present/possible masks is well beyond the scope of you
>>> On 29.03.19 at 14:42, wrote:
> On 3/29/19 4:54 AM, Jan Beulich wrote:
> On 28.03.19 at 17:50, wrote:
>>>
>>> Given especially xen_pv_smp_prepare_cpus(), I think re-working proper
>>> setting of present/possible masks is well beyond the scope of your
>>> original patch.
>> Well, then the q
On 3/29/19 4:54 AM, Jan Beulich wrote:
On 28.03.19 at 17:50, wrote:
>>
>> Given especially xen_pv_smp_prepare_cpus(), I think re-working proper
>> setting of present/possible masks is well beyond the scope of your
>> original patch.
> Well, then the question is, what (if any) changes are you
>>> On 28.03.19 at 17:50, wrote:
> On 3/28/19 5:03 AM, Jan Beulich wrote:
> On 27.03.19 at 18:07, wrote:
>>> On 3/27/19 11:12 AM, Jan Beulich wrote:
-
-static void __init xen_filter_cpu_maps(void)
+static void __init _get_smp_config(unsigned int early)
{
int i, rc
On 3/28/19 5:03 AM, Jan Beulich wrote:
On 27.03.19 at 18:07, wrote:
>> On 3/27/19 11:12 AM, Jan Beulich wrote:
>>> -
>>> -static void __init xen_filter_cpu_maps(void)
>>> +static void __init _get_smp_config(unsigned int early)
>>> {
>>> int i, rc;
>>> unsigned int subtract = 0;
>>>
>>> On 27.03.19 at 18:07, wrote:
> On 3/27/19 11:12 AM, Jan Beulich wrote:
>> -
>> -static void __init xen_filter_cpu_maps(void)
>> +static void __init _get_smp_config(unsigned int early)
>> {
>> int i, rc;
>> unsigned int subtract = 0;
>>
>> -if (!xen_initial_domain())
>> +if
On 3/27/19 11:12 AM, Jan Beulich wrote:
> -
> -static void __init xen_filter_cpu_maps(void)
> +static void __init _get_smp_config(unsigned int early)
> {
> int i, rc;
> unsigned int subtract = 0;
>
> - if (!xen_initial_domain())
> + if (early)
> return;
>
>
This started out with me noticing that "dom0_max_vcpus=" with
larger than the number of physical CPUs reported through ACPI tables
would not bring up the "excess" vCPU-s.
Noticing that xen_fill_possible_map() gets called way too early, whereas
xen_filter_cpu_maps() gets called too late (after per
