On Thu, Jul 9, 2020 at 7:57 AM Rafael J. Wysocki <rjw@xxxxxxxxxxxxx> wrote: > > On Tuesday, July 7, 2020 3:59:32 AM CEST Dan Williams wrote: > > The runtime firmware activation capability of Intel NVDIMM devices > > requires memory transactions to be disabled for 100s of microseconds. > > This timeout is large enough to cause in-flight DMA to fail and other > > application detectable timeouts. Arrange for firmware activation to be > > executed while the system is "quiesced", all processes and device-DMA > > frozen. > > > > It is already required that invoking device ->freeze() callbacks is > > sufficient to cease DMA. A device that continues memory writes outside > > of user-direction violates expectations of the PM core to be to > > establish a coherent hibernation image. > > > > That said, RDMA devices are an example of a device that access memory > > outside of user process direction. RDMA drivers also typically assume > > the system they are operating in will never be hibernated. A solution > > for RDMA collisions with firmware activation is outside the scope of > > this change and may need to rely on being able to survive the platform > > imposed memory controller quiesce period. > > Thanks for following my suggestion to use the hibernation infrastructure > rather than the suspend one, but I think it would be better to go a bit > further with that. > > Namely, after thinking about this a bit more I have come to the conclusion > that what is needed is an ability to execute a function, inside of the > kernel, in a "quiet" environment in which memory updates are unlikely. > > While the hibernation infrastructure as is can be used for that, kind of, IMO > it would be cleaner to introduce a helper for that, like in the (untested) > patch below, so if the "quiet execution environment" is needed, whoever > needs it may simply pass a function to hibernate_quiet_exec() and provide > whatever user-space I/F is suitable on top of that. > > Please let me know what you think. This looks good to me in concept. Would you expect that I trigger this from libnvdimm sysfs, or any future users of this functionality to trigger it through their own subsystem specific mechanisms? I have a place for it in libvdimm and could specify the activation method directly as "suspend" vs "live" activation.
