Re: [PATCH v6 5/7] ACPI/NUMA: Return memblk modification state from numa_fill_memblks()

From: Robert Richter
Date: Thu May 02 2024 - 08:45:44 EST


On 30.04.24 10:05:39, Dan Williams wrote:
> Robert Richter wrote:
> > When registering a memory range a possibly overlapping memory block
> > will be extended instead of creating a new one. If both ranges exactly
> > overlap, the blocks remain unchanged and are just reused. The
> > information if a memblock was extended is useful for diagnostics.
>
> What diagnostic flow is this useful for?
>
> I feel like this post-hoc debug prints for problems we never expect to
> have again, or is there an enduring need for this?

As we are already targeting -rc7, I am going to drop the printout
patches to not block the first patches in this series and will post
them again after the next merge window.

The SRAT logging is very useful to get a picture of the firmware
provided mem blocks. For NUMA debugging these are very helpful, esp.
when working with unmodified generic or distribution kernels. As CFMWS
entries add additional similar information as SRAT, that information
is no longer complete on systems with a CEDT.

The patches just enable the same level of logging for CFMWS as for
SRAT which just adds a single line (info level) per CFMWS entry (in
total just a few). The table printouts are at debug level already. Of
course, there are always other ways to get this information from, but
a grep of dmesg makes things very easy to check things.

Thanks,

-Robert