On 17.09.19 19:04, Waiman Long wrote: > On 9/17/19 12:21 PM, Qian Cai wrote: >> On Tue, 2019-09-17 at 11:49 -0400, Waiman Long wrote: >>> On 9/17/19 3:13 AM, David Hildenbrand wrote: >>>> On 17.09.19 04:34, Toshiki Fukasawa wrote: >>>>> On 2019/09/09 16:46, David Hildenbrand wrote: >>>>>> Let's take a step back here to understand the issues I am aware of. I >>>>>> think we should solve this for good now: >>>>>> >>>>>> A PFN walker takes a look at a random PFN at a random point in time. It >>>>>> finds a PFN with SECTION_MARKED_PRESENT && !SECTION_IS_ONLINE. The >>>>>> options are: >>>>>> >>>>>> 1. It is buddy memory (add_memory()) that has not been online yet. The >>>>>> memmap contains garbage. Don't access. >>>>>> >>>>>> 2. It is ZONE_DEVICE memory with a valid memmap. Access it. >>>>>> >>>>>> 3. It is ZONE_DEVICE memory with an invalid memmap, because the section >>>>>> is only partially present: E.g., device starts at offset 64MB within a >>>>>> section or the device ends at offset 64MB within a section. Don't access it. >>>>> I don't agree with case #3. In the case, struct page area is not allocated on >>>>> ZONE_DEVICE, but is allocated on system memory. So I think we can access the >>>>> struct pages. What do you mean "invalid memmap"? >>>> No, that's not the case. There is no memory, especially not system >>>> memory. We only allow partially present sections (sub-section memory >>>> hotplug) for ZONE_DEVICE. >>>> >>>> invalid memmap == memmap was not initialized == struct pages contains >>>> garbage. There is a memmap, but accessing it (e.g., pfn_to_nid()) will >>>> trigger a BUG. >>>> >>> As long as the page structures exist, they should be initialized to some >>> known state. We could set PagePoison for those invalid memmap. It is the >> Sounds like you want to run page_init_poison() by default. > > Yes for those pages that are not initialized otherwise. I don't want to > run page_init_poison() for the whole ZONE_DEVICE memory range as it can > take a while if we are talking about TBs of persistent memory. Also most > of the pages will be reinitialized anyway in the init process. So it is > mostly a wasted effort. However, for those reserved pages that are not > being exported to the memory management layer, having them initialized > to a known state will cause less problem down the road. > No hacks please. There has to be a proper way to identify if a memmap was initialized or not. Fake-initializing a memmap is *not* the answer. -- Thanks, David / dhildenb
