Question about using NVDIMM for Linux Kernel debugging

Question about using NVDIMM for Linux Kernel debugging

[Dru Nelson]

Hello,

I am doing Linux kernel work on an AMD Ryzen desktop PC.
In particular, I'm trying to debug some hardware driver issues.
The challenging aspect is that they that pop up daily and (of course...)
randomly.
As you are well aware, this can be painful in a production environment on
actual hardware.
Sometimes the computer just reboots or locks up, so having something
persistent would be ideal.
I would like to track down the issue by having some basic instrumentation
of the kernel.

After doing a little research, this looks like a great application for an
NVDIMM.

I would probably divide the nvdimm with ring buffers for each core of the
CPU to avoid locks. The logging would just be simple fixed size records
with a timestamp.

Does this look doable for a reasonably priced NVDIMM?
Or is there a better solution?

Dru, Redwood City, California
_______________________________________________
Linux-nvdimm mailing list
Linux-nvdimm@lists.01.org
https://lists.01.org/mailman/listinfo/linux-nvdimm

Re: Question about using NVDIMM for Linux Kernel debugging

[Dan Williams]

On Mon, Mar 25, 2019 at 1:36 PM Dru Nelson <drudru@gmail.com> wrote:
>
> Hello,
>
> I am doing Linux kernel work on an AMD Ryzen desktop PC.
> In particular, I'm trying to debug some hardware driver issues.
> The challenging aspect is that they that pop up daily and (of course...)
> randomly.
> As you are well aware, this can be painful in a production environment on
> actual hardware.
> Sometimes the computer just reboots or locks up, so having something
> persistent would be ideal.
> I would like to track down the issue by having some basic instrumentation
> of the kernel.
>
> After doing a little research, this looks like a great application for an
> NVDIMM.
>
> I would probably divide the nvdimm with ring buffers for each core of the
> CPU to avoid locks. The logging would just be simple fixed size records
> with a timestamp.
>
> Does this look doable for a reasonably priced NVDIMM?
> Or is there a better solution?

I'd recommend taking a look at ramoops, but that said I've not tried
hooking it up to an NVDIMM backend.

https://github.com/torvalds/linux/blob/master/Documentation/admin-guide/ramoops.rst
_______________________________________________
Linux-nvdimm mailing list
Linux-nvdimm@lists.01.org
https://lists.01.org/mailman/listinfo/linux-nvdimm