On Wed, Feb 16, 2022 at 10:41:43AM +0100, David Marchand wrote: > On Wed, Feb 16, 2022 at 10:31 AM Bruce Richardson > <bruce.richard...@intel.com> wrote: > > > > I see that we have other telemetry callbacks for stats, link status > > > > which might already have the issue. > > > > > > You mean there is no lock protection? Neither in the API, nor in > > > telemetry? > > > > > For reporting out stats or link status, I'm not sure a lock should ever > > be needed since both are just read-only operations. Therefore, I don't > > believe we have a general issue here. > > Ok, if you are sure about this for all drivers for stats and link status, > you can discard my comment. > No, I'm not 100% sure about it, but if any driver has issues with multiple threads reading stats simultaneously I would hope that that should be resolved at the driver level. Normally, one would expect read-only operations to be implicitly multi-thread safe, and I also see that the documentation on e.g. rte_eth_stats_get() makes no reference to needing locks or not being thread-safe. I note too, that any such issues would not be limited to telemetry only, but would also apply to pdump or any mechanism we would want to have for reading information about a running process.
/Bruce
