On Fri, Jun 7, 2019 at 2:12 PM Dave Hansen <dave.hansen@xxxxxxxxx> wrote: > > On 6/7/19 1:03 PM, Dan Williams wrote: > >> Separate from these patches, should we have a runtime file that dumps > >> out the same info? dmesg isn't always available, and hotplug could > >> change this too, I'd imagine. > > Perhaps, but I thought /proc/iomem was that runtime file. Given that > > x86/Linux only seems to care about the the EFI to E820 translation of > > the map and the E820 map is directly reflected in /proc/iomem, do we > > need another file? > > Probably not. > > I'm just trying to think of ways that we can debug systems where someone > "loses" a bunch of memory, especially if they're moving from an old > kernel to a new one with these patches. From their perspective, they > just lost a bunch of expensive memory. > > Do we owe a pr_info(), perhaps? Or even a /proc/meminfo entry for how > much memory these devices own? We have this existing print when this bit is found: [ 0.023650] e820: update [mem 0x240000000-0x43fffffff] usable ==> application reserved ...but perhaps /proc/meminfo could grow: ApplicationReservedOffline ApplicationReservedOnline ...to show the relative amount of this memory that has been routed to device-dax and how much has been returned to the core-mm?
