Re: [PATCH] tools/ocaml/xc: Fix xc_physinfo() bindings

[Andrew Cooper <Andrew.Cooper3@xxxxxxxxxx>]

On 09/06/2023 9:17 am, Christian Lindig wrote:
>> On 8 Jun 2023, at 20:33, Andrew Cooper <Andrew.Cooper3@xxxxxxxxxx> wrote:
>>
>> +type arm_physinfo_caps =
>> +  {
>> +    sve_vl: int;
>> +  }
>> +
>
> Does the OCaml side need to know about the structure of this value or is it 
> enough to pass it around as an abstract value because all logic is on the C 
> side? I assume the OCaml side needs at least a way to persist the value and 
> hence needs to know some representation.

Yes, Ocaml does need to know about the structure.

info->arch_capabilities is a collection of misc info.  It was intended
to be just a bitmap of things, hence why it's a list on the x86 side (We
added fields, then had to revert and I haven't got back around to
reworking that yet).

But now ARM have put a non-bit field field into it, where a toolstack
needs to select a domain_create sve_vl setting between min(0,
physinfo.sve_vl)

The Ocaml code doesn't have an #include <public/sysctl.h> so doesn't
have the masks/constants required to decompose the integer into it's
constitute parts.

>
>>      Store_field(arch_obj, 0,
>> +                Val_int(MASK_EXTR(info->arch_capabilities,
>> +                                  XEN_SYSCTL_PHYSCAP_ARM_SVE_MASK) * 128));
>> +
> What is the “* 128” achieving as part of this encoding?

No functional change from before...  except it was previously hidden in
a different header file which I'm in the process of deleting.

The vector length is a multiple of 128.  This value is more amenable to
being understood in a log file by a human.

~Andrew