Re: [PATCH] of: introduce event tracepoints for dynamic device_node lifecyle

[Oliver O'Halloran]

On Wed, Apr 19, 2017 at 2:46 AM, Rob Herring <robh+dt@xxxxxxxxxx> wrote:
> On Mon, Apr 17, 2017 at 7:32 PM, Tyrel Datwyler
> <tyreld@xxxxxxxxxxxxxxxxxx> wrote:
>> This patch introduces event tracepoints for tracking a device_nodes
>> reference cycle as well as reconfig notifications generated in response
>> to node/property manipulations.
>>
>> With the recent upstreaming of the refcount API several device_node
>> underflows and leaks have come to my attention in the pseries (DLPAR) dynamic
>> logical partitioning code (ie. POWER speak for hotplugging virtual and physcial
>> resources at runtime such as cpus or IOAs). These tracepoints provide a
>> easy and quick mechanism for validating the reference counting of
>> device_nodes during their lifetime.
>
> Not really relevant for this patch, but since you are looking at
> pseries and refcounting, the refcounting largely exists for pseries.
> It's also hard to get right as this type of fix is fairly common. It's
> now used for overlays, but we really probably only need to refcount
> the overlays or changesets as a whole, not at a node level. If you
> have any thoughts on how a different model of refcounting could work
> for pseries, I'd like to discuss it.

One idea I've been kicking around is differentiating short and long
term references to a node. I figure most leaks are due to a missing
of_node_put() within a stack frame so it might be possible to use the
ftrace infrastructure to detect and emit warnings if a short term
reference is leaked. Long term references are slightly harder to deal
with, but they're less common so we can add more detailed reference
tracking there (devm_of_get_node?).

Oliver