On Thu, Nov 9, 2017 at 10:51 AM, Mikulas Patocka <mpatocka@xxxxxxxxxx> wrote: [..] >> The drivers don't need to react, once the pages are pinned for dma the >> hot-unplug will not progress until all those page references are >> dropped. > > I am not talking about moving pages here, I'm talking about possible > hardware errors in persistent memory. In this situation, the storage > controller receives an error on the bus - and the question is, how will it > react. Ideally, it should abort just this transfer and return an error > that the driver will propagate up. But I'm skeptical that someone is > really testing the controllers and drivers for this possiblity. This is something that drive controllers already need to deal with today on DRAM, but I suspect you are right because in general error-path testing in drivers is rare to non-existent in Linux. We can endeavor to do better with persistent memory where we have some explicit error injection facilities defined in ACPI that might enjoy wider support than the existing EINJ facility. -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>
