Re: [PATCH] arm64: acpi: reenumerate topology ids

[Andrew Jones]

On Fri, Jun 29, 2018 at 01:42:27PM +0200, Andrew Jones wrote:
> On Fri, Jun 29, 2018 at 11:53:34AM +0100, Sudeep Holla wrote:
> > On Thu, Jun 28, 2018 at 12:12:00PM -0500, Jeremy Linton wrote:
> > > Hi,
> > > 
> > > On 06/28/2018 11:30 AM, Sudeep Holla wrote:
> > 
> > [...]
> > 
> > > >I am not sure if we can ever guarantee that DT and ACPI will get the
> > > >same ids whatever counter we use as it depends on the order presented in
> > > >the firmware(DT or ACPI). So I am not for generating ids for core and
> > > >threads in that way.
> > > >
> > > >So I would like to keep it simple and just have this counters for
> > > >package ids as demonstrated in Shunyong's patch.
> > >
> > > So, currently on a non threaded system, the core id's look nice because they
> > > are just the ACPI ids. Its the package id's that look strange, we could just
> > > fix the package ids, but on threaded machines the threads have the nice acpi
> > > ids, and the core ids are then funny numbers. So, I suspect that is driving
> > > this as much as the strange package ids.
> > >
> > 
> > Yes, I know that and that's what made be look at topology_get_acpi_cpu_tag
> > For me, if the PPTT has valid ID, we should use that. Just becuase DT lacks
> > it and uses counter doesn't mean ACPI also needs to follow that.
> 
> AFAIK, a valid ACPI UID doesn't need to be something derivable directly
> from the hardware, so it's just as arbitrary as the CPU phandle that is
> in the DT cpu-map, i.e. DT *does* have an analogous leaf node integer.
> 
> > 
> > I am sure some vendor will put valid UID and expect that to be in the
> > sysfs.
> 
> I can't think of any reason that would be useful, especially when the
> UID is for a thread, which isn't even displayed by sysfs.
> 
> > 
> > > (and as a side, I actually like the PE has a acpi id behavior, but for
> > > threads its being lost with this patch...)
> > > 
> > > Given i've seen odd package/core ids on x86s a few years ago, it never
> 
> So this inspired me to grep some x86 topology code. I found
> arch/x86/kernel/smpboot.c:topology_update_package_map(), which uses
> a counter to set the logical package id and Documentation/x86/topology.txt
> states
> 
> """
>   - cpuinfo_x86.logical_id:
> 
>     The logical ID of the package. As we do not trust BIOSes to enumerate the
>     packages in a consistent way, we introduced the concept of logical package
>     ID so we can sanely calculate the number of maximum possible packages in
>     the system and have the packages enumerated linearly.
> """

Eh, x86 does seem to display the physical, rather than logical (linear)
IDs in sysfs though,

arch/x86/include/asm/topology.h:#define topology_physical_package_id(cpu)       (cpu_data(cpu).phys_proc_id)

"""
  - cpuinfo_x86.phys_proc_id:

    The physical ID of the package. This information is retrieved via CPUID
    and deduced from the APIC IDs of the cores in the package.
"""

So, hmmm...

But, I think we should either be looking for a hardware derived ID to use
(like x86), or remap to counters. I don't believe the current scheme of
using ACPI offsets can be better than counters, and it has consistency and
human readability issues.

Thanks,
drew

> 
> Which I see as x86 precedent for the consistency argument I made in my
> other reply.
> 
> Thanks,
> drew