Re: [PATCH v3 07/18] machine: Add a new function init_apicid_fn in MachineClass

[Igor Mammedov]

On Mon, 3 Feb 2020 15:49:31 -0600
Babu Moger <address@hidden> wrote:

> On 2/3/20 9:17 AM, Igor Mammedov wrote:
> > On Wed, 29 Jan 2020 10:17:11 -0600
> > Babu Moger <address@hidden> wrote:
> >   
> >> On 1/29/20 3:14 AM, Igor Mammedov wrote:  
> >>> On Tue, 28 Jan 2020 13:45:31 -0600
> >>> Babu Moger <address@hidden> wrote:
> >>>     
> >>>> On 1/28/20 10:29 AM, Igor Mammedov wrote:    
> >>>>> On Tue, 03 Dec 2019 18:37:42 -0600
> >>>>> Babu Moger <address@hidden> wrote:
> >>>>>       
> >>>>>> Add a new function init_apicid_fn in MachineClass to initialize the 
> >>>>>> mode
> >>>>>> specific handlers to decode the apic ids.
> >>>>>>
> >>>>>> Signed-off-by: Babu Moger <address@hidden>
> >>>>>> ---
> >>>>>>  include/hw/boards.h |    1 +
> >>>>>>  vl.c                |    3 +++
> >>>>>>  2 files changed, 4 insertions(+)
> >>>>>>
> >>>>>> diff --git a/include/hw/boards.h b/include/hw/boards.h
> >>>>>> index d4fab218e6..ce5aa365cb 100644
> >>>>>> --- a/include/hw/boards.h
> >>>>>> +++ b/include/hw/boards.h
> >>>>>> @@ -238,6 +238,7 @@ struct MachineClass {
> >>>>>>                                                           unsigned 
> >>>>>> cpu_index);
> >>>>>>      const CPUArchIdList *(*possible_cpu_arch_ids)(MachineState 
> >>>>>> *machine);
> >>>>>>      int64_t (*get_default_cpu_node_id)(const MachineState *ms, int 
> >>>>>> idx);
> >>>>>> +    void (*init_apicid_fn)(MachineState *ms);      
> >>>>> it's x86 specific, so why it wasn put into PCMachineClass?      
> >>>>
> >>>> Yes. It is x86 specific for now. I tried to make it generic function so
> >>>> other OSes can use it if required(like we have done in
> >>>> possible_cpu_arch_ids). It initializes functions required to build the
> >>>> apicid for each CPUs. We need these functions much early in the
> >>>> initialization. It should be initialized before parse_numa_opts or
> >>>> machine_run_board_init(in v1.c) which are called from generic context. We
> >>>> cannot use PCMachineClass at this time.    
> >>>
> >>> could you point to specific patches in this series that require
> >>> apic ids being initialized before parse_numa_opts and elaborate why?
> >>>
> >>> we already have possible_cpu_arch_ids() which could be called very
> >>> early and calculates APIC IDs in x86 case, so why not reuse it?    
> >>
> >>
> >> The current code(before this series) parses the numa information and then
> >> sequentially builds the apicid. Both are done together.
> >>
> >> But this series separates the numa parsing and apicid generation. Numa
> >> parsing is done first and after that the apicid is generated. Reason is we
> >> need to know the number of numa nodes in advance to decode the apicid.
> >>
> >> Look at this patch.
> >> https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Flore.kernel.org%2Fqemu-devel%2F157541988471.46157.6587693720990965800.stgit%40naples-babu.amd.com%2F&amp;data=02%7C01%7Cbabu.moger%40amd.com%7C0a643dd978f149acf9d108d7a8bc487a%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C637163398941923379&amp;sdata=sP2TnNaqNXRGEeQNhJMna3wyeBqN0XbNKqgsCTVDaOQ%3D&amp;reserved=0
> >>
> >> static inline apic_id_t apicid_from_topo_ids_epyc(X86CPUTopoInfo 
> >> *topo_info,
> >> +                                                  const X86CPUTopoIDs
> >> *topo_ids)
> >> +{
> >> +    return (topo_ids->pkg_id  << apicid_pkg_offset_epyc(topo_info)) |
> >> +           (topo_ids->llc_id << apicid_llc_offset_epyc(topo_info)) |
> >> +           (topo_ids->die_id  << apicid_die_offset(topo_info)) |
> >> +           (topo_ids->core_id << apicid_core_offset(topo_info)) |
> >> +           topo_ids->smt_id;
> >> +}
> >>
> >>
> >> The function apicid_from_topo_ids_epyc builds the apicid. New decode adds
> >> llc_id(which is numa id here) to the current decoding. Other fields are
> >> mostly remains same.  
> > 
> > If llc_id is the same as numa id, why not reuse 
> > CpuInstanceProperties::node-id
> > instead of llc_id you are adding in previous patch 6/18?
> >   
> I tried to use that earlier. But dropped the idea as it required some
> changes. Don't remember exactly now. I am going to investigate again if we
> can use the node_id for our purpose here. Will let you know if I have any
> issues.
The reason I'm asking to not add new properties here is that it
expands interface visible/used by management tools and it's maintenance
burden not only on QEMU but on engagement side as well. So if yo can reuse
node-id, it will work out of box with existing users.

It should also be less confusing for us since we don't have to keep in mind
(or figure out) that llc_id is the same as node id and wonder why the later
wasn't used in the first place.