Message ID | 20220812055710.357820-1-aneesh.kumar@linux.ibm.com (mailing list archive) |
---|---|
Headers | show |
Series | mm/demotion: Memory tiers and demotion | expand |
"Aneesh Kumar K.V" <aneesh.kumar@linux.ibm.com> writes: > The current kernel has the basic memory tiering support: Inactive pages on a > higher tier NUMA node can be migrated (demoted) to a lower tier NUMA node to > make room for new allocations on the higher tier NUMA node. Frequently accessed > pages on a lower tier NUMA node can be migrated (promoted) to a higher tier NUMA > node to improve the performance. > > In the current kernel, memory tiers are defined implicitly via a demotion path > relationship between NUMA nodes, which is created during the kernel > initialization and updated when a NUMA node is hot-added or hot-removed. The > current implementation puts all nodes with CPU into the highest tier, and builds the > tier hierarchy tier-by-tier by establishing the per-node demotion targets based > on the distances between nodes. > > This current memory tier kernel implementation needs to be improved for several > important use cases: > > * The current tier initialization code always initializes each memory-only NUMA > node into a lower tier. But a memory-only NUMA node may have a high > performance memory device (e.g. a DRAM-backed memory-only node on a virtual > machine) and that should be put into a higher tier. > > * The current tier hierarchy always puts CPU nodes into the top tier. But on a > system with HBM (e.g. GPU memory) devices, these memory-only HBM NUMA nodes > should be in the top tier, and DRAM nodes with CPUs are better to be placed > into the next lower tier. > > * Also because the current tier hierarchy always puts CPU nodes into the top > tier, when a CPU is hot-added (or hot-removed) and triggers a memory node from > CPU-less into a CPU node (or vice versa), the memory tier hierarchy gets > changed, even though no memory node is added or removed. This can make the > tier hierarchy unstable and make it difficult to support tier-based memory > accounting. > > * A higher tier node can only be demoted to nodes with shortest distance on the > next lower tier as defined by the demotion path, not any other node from any > lower tier. This strict, demotion order does not work in all use > cases (e.g. some use cases may want to allow cross-socket demotion to another > node in the same demotion tier as a fallback when the preferred demotion node > is out of space), and has resulted in the feature request for an interface to > override the system-wide, per-node demotion order from the userspace. This > demotion order is also inconsistent with the page allocation fallback order > when all the nodes in a higher tier are out of space: The page allocation can > fall back to any node from any lower tier, whereas the demotion order doesn't > allow that. > > This patch series make the creation of memory tiers explicit under > the control of device driver. > > Memory Tier Initialization > ========================== > > Linux kernel presents memory devices as NUMA nodes and each memory device is of > a specific type. The memory type of a device is represented by its abstract > distance. A memory tier corresponds to a range of abstract distance. This allows > for classifying memory devices with a specific performance range into a memory > tier. > > By default, all memory nodes are assigned to the default tier with > abstract distance 512. > > A device driver can move its memory nodes from the default tier. For example, > PMEM can move its memory nodes below the default tier, whereas GPU can move its > memory nodes above the default tier. > > The kernel initialization code makes the decision on which exact tier a memory > node should be assigned to based on the requests from the device drivers as well > as the memory device hardware information provided by the firmware. > > Hot-adding/removing CPUs doesn't affect memory tier hierarchy. > > Changes from v13 > * Address review feedback. > * Add path dropping memtier from struct memory_dev_type > > Changes from v12 > * Fix kernel crash on module unload > * Address review feedback. > * Add node_random patch to this series based on review feedback > > Changes from v11: > * smaller abstract distance imply faster(higher) memory tier. > > Changes from v10: > * rename performance level to abstract distance > * Thanks to all the good feedback from Huang, Ying <ying.huang@intel.com>. > Updated the patchset to cover most of the review feedback. > > Changes from v9: > * Use performance level for initializing memory tiers. > > Changes from v8: > * Drop the sysfs interface patches and related documentation changes. > > Changes from v7: > * Fix kernel crash with demotion. > * Improve documentation. > > Changes from v6: > * Drop the usage of rank. > * Address other review feedback. > > Changes from v5: > * Remove patch supporting N_MEMORY node removal from memory tiers. memory tiers > are going to be used for features other than demotion. Hence keep all N_MEMORY > nodes in memory tiers irrespective of whether they want to participate in promotion or demotion. > * Add NODE_DATA->memtier > * Rearrage patches to add sysfs files later. > * Add support to create memory tiers from userspace. > * Address other review feedback. > > > Changes from v4: > * Address review feedback. > * Reverse the meaning of "rank": higher rank value means higher tier. > * Add "/sys/devices/system/memtier/default_tier". > * Add node_is_toptier > > v4: > Add support for explicit memory tiers and ranks. > > v3: > - Modify patch 1 subject to make it more specific > - Remove /sys/kernel/mm/numa/demotion_targets interface, use > /sys/devices/system/node/demotion_targets instead and make > it writable to override node_states[N_DEMOTION_TARGETS]. > - Add support to view per node demotion targets via sysfs > > v2: > In v1, only 1st patch of this patch series was sent, which was > implemented to avoid some of the limitations on the demotion > target sharing, however for certain numa topology, the demotion > targets found by that patch was not most optimal, so 1st patch > in this series is modified according to suggestions from Huang > and Baolin. Different examples of demotion list comparasion > between existing implementation and changed implementation can > be found in the commit message of 1st patch. > > > > Aneesh Kumar K.V (9): > mm/demotion: Add support for explicit memory tiers > mm/demotion: Move memory demotion related code > mm/demotion: Add hotplug callbacks to handle new numa node onlined > mm/demotion/dax/kmem: Set node's abstract distance to > MEMTIER_DEFAULT_DAX_ADISTANCE > mm/demotion: Build demotion targets based on explicit memory tiers > mm/demotion: Add pg_data_t member to track node memory tier details > mm/demotion: Drop memtier from memtype > mm/demotion: Update node_is_toptier to work with memory tiers > lib/nodemask: Optimize node_random for nodemask with single NUMA node > > Jagdish Gediya (1): > mm/demotion: Demote pages according to allocation fallback order > > drivers/dax/kmem.c | 42 ++- > include/linux/memory-tiers.h | 99 ++++++ > include/linux/migrate.h | 15 - > include/linux/mmzone.h | 3 + > include/linux/node.h | 5 - > include/linux/nodemask.h | 15 +- > mm/Makefile | 1 + > mm/huge_memory.c | 1 + > mm/memory-tiers.c | 645 +++++++++++++++++++++++++++++++++++ > mm/migrate.c | 453 +----------------------- > mm/mprotect.c | 1 + > mm/vmscan.c | 59 +++- > mm/vmstat.c | 4 - > 13 files changed, 846 insertions(+), 497 deletions(-) > create mode 100644 include/linux/memory-tiers.h > create mode 100644 mm/memory-tiers.c Except some minor comments, the series looks good to me. Thanks! Feel free to add Reviewed-by: "Huang, Ying" <ying.huang@intel.com> for the whole series. Best Regards, Huang, Ying