Re: [RFC] libibverbs IB device hotplug support

[Leon Romanovsky <leon-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org>]

[-- Attachment #1: Type: text/plain, Size: 1096 bytes --]

On Wed, Mar 01, 2017 at 04:07:43PM -0700, Jason Gunthorpe wrote:
> On Wed, Mar 01, 2017 at 10:15:31PM +0000, Weiny, Ira wrote:
> > > On the call we discussed that today there is no 'incarnation'
> > > value exposed for the uverbs%.  You suggested an optimization by
> > > using the cdev's file creation timestamp as option for libibverbs
> > > to understand that the driver did or did-not restart since mapping
> > > of that ibv_device.
>
> Not the cdev, read the timestamp from /sys/class/infiniband/mlx4_0/hca_type
>
> > > Main pitfalls with this is if system time is being modified (admin
> > > or ntp/ptp). In the rare case that we have a hoplug event and time
> > > is rewinded we can accidentally reach a very hard debugging
> > > exercise.
>
> You are only using it for direct equality comparision, so someone
> would have to roll back the clock and hotplug the driver at the exact
> same nano-second resolution timestamp. That is deeply unlikely to ever
> happen by accident.

Jason,

I think that we are over engineering here, but it can occur with NTP clock hopping.

Thanks

>
> Jason

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]