Re: [Patch v2 2/2] x86/tsc: use logical_packages as a better estimation of socket numbers

[Feng Tang]

On Fri, Jun 16, 2023 at 10:02:31AM +0200, Peter Zijlstra wrote:
> On Fri, Jun 16, 2023 at 06:53:21AM +0000, Zhang, Rui wrote:
> > On Thu, 2023-06-15 at 11:20 +0200, Peter Zijlstra wrote:
> 
> > > So I have at least two machines where I boot with 'possible_cpus=#'
> > > because the BIOS MADT is reporting a stupid number of CPUs that
> > > aren't
> > > actually there.
> > 
> > Does the MADT report those CPUs as disabled but online capable?
> > can you send me a copy of the acpidmp?
> 
> Sent privately, it's a bit big.
> 
> > I had a patch to parse MADT and count the number of physical packages
> > by decoding all the valid APICIDs in MADT.
> > I'm wondering if the patch still works on this machine.
> 
> I can certainly give it a spin; it has IPMI serial-over-ethernet that
> works. Brilliant dev machine.
> 
> > > So I think I'm lucky and side-stepped this nonsense, but if someone
> > > were
> > > to use nr_cpus= for this same purpose, they get screwed over and get
> > > the
> > > watchdog. Sad day for them I suppose.
> > 
> > what if using package_count_from_MADT?
> 
> So I'm thinking that if you cap possible_mask the actual logical
> packages is the right number.
> 
> Suppose you have a machine with 8 sockets, but limit possible_mask to
> only 1 socket. Then TSC will actually be stable, it doesn't matter you
> have 7 idle sockets that are not synchronized.
> 
> Then again, perhaps if you limit it to 2 sockets you're still in
> trouble, I'm not entirely sure how the TSC sync stuff comes apart on
> these large systems.

I had the similar thought. For this case, the defensive way is to keep
the watchdog for 'nr_cpus=' and 'possible_cpus=' setup, and if the
specific setup has no TSC sync issue, people can add one more parameter
'tsc=reliable' to skip the watchdog, while aggressive way is to ignore
the 2 cmdline parameters as the above case is really rare.

Again, as you mentioned, I can't find a perfect solution to cover all
kinds of setup and broken firmware. But at least 'logical_packages' is
much better than 'nr_online_nodes' :)

Thanks,
Feng