Re: making identify_hypervisor arch independent

> I briefly thought of dev/smbios, BUT, that brings up another issue, I think we should deorbit dev/smbios, and move smbios.h to dev/ipmi, because as far as I can tell, the device does nothing useful. It allocates the resource, but nothing that I can find in the tree attempts to use/access the device, and it doesn't expose any interface. The header is only used by dev/ipmi for the structure, but not the device, as ipmi does it's own parsing of the table.

Once upon a time, Panasas had a much more involved smbios(4). It parsed out a bunch of the same SMBIOS strings as the loaders do, but also decoded a bunch of information about the DIMMs (Table 17) and exposed them via sysctls. However, when we subsequently rebased onto a more modern version of FreeBSD, expediency came down on the side of screen-scraping `dmidecode' output in a script, and having userland tools exec that rather than reading sysctls.

At one point, ambrisko@ and I had talked about moving generic smbios-walking code into smbios(4), exposing that to the kernel through an smbios_if.m, then things like ipmi(4) could leverage that. (And, for example, if a driver wanted to be able to map between physical address and DIMM, it could use that interface to find Table 17, Table 19, Table 20, etc.) Alas, neither of us has had the time to work on that.

In a later message in this thread, Warner suggested consolidating that code into subr_smbios.c rather than smbios(4). I think that would be functionally equivalent.

Thanks,

Ravi

-----Original Message-----
From: <owner-freebsd-arch@freebsd.org <mailto:owner-freebsd-arch@freebsd.org>> on behalf of John-Mark Gurney <jmg@funkthat.com <mailto:jmg@funkthat.com>>
Date: 2023-02-21, Tuesday at 20:05
To: <freebsd-arch@FreeBSD.org <mailto:freebsd-arch@FreeBSD.org>>
Subject: making identify_hypervisor arch independent


Hello,


I have a pending diff (https://reviews.freebsd.org/D38721 <https://reviews.freebsd.org/D38721>) that will make
SMBIOS work on arm64 systems (specifically under qemu, but likely it may
add support for other EFI arm64 systems that have SMBIOS as well).


The goal is to support identifying that we are running as a guest under
QEMU so that we automatically switch to hz=100 on arm64.


Currently there is code in x86/x86/identcpu.c that has code to identify
hypervisers via SMBIOS, so I'd like to move most of identify_hypervisor
to a new location so that arm64 code can call it as well.


Where should I put it? kern/subr_identsmbios.c? And make it optional
on EFIRT for arm64 and standard on x86?


I briefly thought of dev/smbios, BUT, that brings up another issue,
I think we should deorbit dev/smbios, and move smbios.h to dev/ipmi,
because as far as I can tell, the device does nothing useful. It
allocates the resource, but nothing that I can find in the tree attempts
to use/access the device, and it doesn't expose any interface. The
header is only used by dev/ipmi for the structure, but not the device,
as ipmi does it's own parsing of the table.


All of the uses of smbios data is getting it via kenv which is populated
via loader (libsa).


Thanks.


-- 
John-Mark Gurney Voice: +1 415 225 5579


"All that I will do, has been done, All that I have, has not."