With Dave Hansen's patches merged into Linus's tree https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=c221c0b0308fd01d9fb33a16f64d2fd95f8830a4 PMEM could be hot plugged as NUMA node now. But, how to use PMEM as NUMA node effectively and efficiently is still a question. There have been a couple of proposals posted on the mailing list [1] [2]. The patchset is aimed to try a different approach from this proposal [1] to use PMEM as NUMA nodes. The approach is designed to follow the below principles: 1. Use PMEM as normal NUMA node, no special gfp flag, zone, zonelist, etc. 2. DRAM first/by default. No surprise to existing applications and default running. PMEM will not be allocated unless its node is specified explicitly by NUMA policy. Some applications may be not very sensitive to memory latency, so they could be placed on PMEM nodes then have hot pages promote to DRAM gradually. 3. Compatible with current NUMA policy semantics. 4. Don't assume hardware topology. But, the patchset still assumes two tier heterogeneous memory system. I understood generalizing multi tier heterogeneous memory had been discussed before. I do agree that is preferred eventually. However, currently kernel doesn't have such capability yet. When HMAT is fully ready we definitely could extract NUMA topology from it. 5. Control memory allocation and hot/cold pages promotion/demotion on per VMA basis. To achieve the above principles, the design can be summarized by the following points: 1. Per node global fallback zonelists (include both DRAM and PMEM), use def_alloc_nodemask to exclude non-DRAM nodes from default allocation unless they are specified by mempolicy. Currently kernel just can distinguish volatile and non-volatile memory. So, just build the nodemask by SRAT flag. In the future it may be better to build the nodemask with more exposed hardware information, i.e. HMAT attributes so that it could be extended to multi tier memory system easily. 2. Introduce a new mempolicy, called MPOL_HYBRID to keep other mempolicy semantics intact. We would like to have memory placement control on per process or even per VMA granularity. So, mempolicy sounds more reasonable than madvise. The new mempolicy is mainly used for launching processes on PMEM nodes then migrate hot pages to DRAM nodes via NUMA balancing. MPOL_BIND could bind to PMEM nodes too, but migrating to DRAM nodes would just break the semantic of it. MPOL_PREFERRED can't constraint the allocation to PMEM nodes. So, it sounds a new mempolicy is needed to fulfill the usecase. 3. The new mempolicy would promote pages to DRAM via NUMA balancing. IMHO, I don't think kernel is a good place to implement sophisticated hot/cold page distinguish algorithm due to the complexity and overhead. But, kernel should have such capability. NUMA balancing sounds like a good start point. 4. Promote twice faulted page. Use PG_promote to track if a page is faulted twice. This is an optimization to NUMA balancing to reduce the migration thrashing and overhead for migrating from PMEM. 5. When DRAM has memory pressure, demote page to PMEM via page reclaim path. This is quite similar to other proposals. Then NUMA balancing will promote page to DRAM as long as the page is referenced again. But, the promotion/demotion still assumes two tier main memory. And, the demotion may break mempolicy. 6. Anonymous page only for the time being since NUMA balancing can't promote unmapped page cache. The patchset still misses a some pieces and is pre-mature, but I would like to post to LKML to gather more feedback and comments and have more eyes on it to make sure I'm on the right track. Any comment is welcome. TODO: 1. Promote page cache. There are a couple of ways to handle this in kernel, i.e. promote via active LRU in reclaim path on PMEM node, or promote in mark_page_accessed(). 2. Promote/demote HugeTLB. Now HugeTLB is not on LRU and NUMA balancing just skips it. 3. May place kernel pages (i.e. page table, slabs, etc) on DRAM only. 4. Support the new mempolicy in userspace tools, i.e. numactl. [1]: https://lore.kernel.org/linux-mm/20181226131446.330864849@xxxxxxxxx/ [2]: https://lore.kernel.org/linux-mm/20190321200157.29678-1-keith.busch@xxxxxxxxx/ Yang Shi (10): mm: control memory placement by nodemask for two tier main memory mm: mempolicy: introduce MPOL_HYBRID policy mm: mempolicy: promote page to DRAM for MPOL_HYBRID mm: numa: promote pages to DRAM when it is accessed twice mm: page_alloc: make find_next_best_node could skip DRAM node mm: vmscan: demote anon DRAM pages to PMEM node mm: vmscan: add page demotion counter mm: numa: add page promotion counter doc: add description for MPOL_HYBRID mode doc: elaborate the PMEM allocation rule Documentation/admin-guide/mm/numa_memory_policy.rst | 10 ++++ Documentation/vm/numa.rst | 7 ++- arch/x86/mm/numa.c | 1 + drivers/acpi/numa.c | 8 +++ include/linux/migrate.h | 1 + include/linux/mmzone.h | 3 ++ include/linux/page-flags.h | 4 ++ include/linux/vm_event_item.h | 3 ++ include/linux/vmstat.h | 1 + include/trace/events/migrate.h | 3 +- include/trace/events/mmflags.h | 3 +- include/uapi/linux/mempolicy.h | 1 + mm/debug.c | 1 + mm/huge_memory.c | 14 ++++++ mm/internal.h | 33 ++++++++++++ mm/memory.c | 12 +++++ mm/mempolicy.c | 74 ++++++++++++++++++++++++--- mm/page_alloc.c | 33 +++++++++--- mm/vmscan.c | 113 +++++++++++++++++++++++++++++++++++------- mm/vmstat.c | 3 ++ 20 files changed, 295 insertions(+), 33 deletions(-)
