kernel/linux.git/include/linux/memory-tiers.h, branch v6.19.11

mm/memory-tier: fix abstract distance calculation overflow

2025-07-10T05:42:06+00:00

In mt_perf_to_adistance(), the calculation of abstract distance (adist) involves multiplying several int values including MEMTIER_ADISTANCE_DRAM. *adist = MEMTIER_ADISTANCE_DRAM * (perf->read_latency + perf->write_latency) / (default_dram_perf.read_latency + default_dram_perf.write_latency) * (default_dram_perf.read_bandwidth + default_dram_perf.write_bandwidth) / (perf->read_bandwidth + perf->write_bandwidth); Since these values can be large, the multiplication may exceed the maximum value of an int (INT_MAX) and overflow (Our platform did), leading to an incorrect adist. User-visible impact: The memory tiering subsystem will misinterpret slow memory (like CXL) as faster than DRAM, causing inappropriate demotion of pages from CXL (slow memory) to DRAM (fast memory). For example, we will see the following demotion chains from the dmesg, where Node0,1 are DRAM, and Node2,3 are CXL node: Demotion targets for Node 0: null Demotion targets for Node 1: null Demotion targets for Node 2: preferred: 0-1, fallback: 0-1 Demotion targets for Node 3: preferred: 0-1, fallback: 0-1 Change MEMTIER_ADISTANCE_DRAM to be a long constant by writing it with the 'L' suffix. This prevents the overflow because the multiplication will then be done in the long type which has a larger range. Link: https://lkml.kernel.org/r/20250611023439.2845785-1-lizhijian@fujitsu.com Link: https://lkml.kernel.org/r/20250610062751.2365436-1-lizhijian@fujitsu.com Fixes: 3718c02dbd4c ("acpi, hmat: calculate abstract distance with HMAT") Signed-off-by: Li Zhijian Reviewed-by: Huang Ying Acked-by: Balbir Singh Reviewed-by: Donet Tom Reviewed-by: Oscar Salvador Cc: Signed-off-by: Andrew Morton

memory tier: consolidate the initialization of memory tiers

2024-07-12T22:52:20+00:00

The current memory tier initialization process is distributed across two different functions, memory_tier_init() and memory_tier_late_init(). This design is hard to maintain. Thus, this patch is proposed to reduce the possible code paths by consolidating different initialization patches into one. The earlier discussion with Jonathan and Ying is listed here: https://lore.kernel.org/lkml/20240405150244.00004b49@Huawei.com/ If we want to put these two initializations together, they must be placed together in the later function. Because only at that time, the HMAT information will be ready, adist between nodes can be calculated, and memory tiering can be established based on the adist. So we position the initialization at memory_tier_init() to the memory_tier_late_init() call. Moreover, it's natural to keep memory_tier initialization in drivers at device_initcall() level. If we simply move the set_node_memory_tier() from memory_tier_init() to late_initcall(), it will result in HMAT not registering the mt_adistance_algorithm callback function, because set_node_memory_tier() is not performed during the memory tiering initialization phase, leading to a lack of correct default_dram information. Therefore, we introduced a nodemask to pass the information of the default DRAM nodes. The reason for not choosing to reuse default_dram_type->nodes is that it is not clean enough. So in the end, we use a __initdata variable, which is a variable that is released once initialization is complete, including both CPU and memory nodes for HMAT to iterate through. Link: https://lkml.kernel.org/r/20240704072646.437579-1-horen.chuang@linux.dev Signed-off-by: Ho-Ren (Jack) Chuang Suggested-by: Jonathan Cameron Reviewed-by: "Huang, Ying" Reviewed-by: Jonathan Cameron Cc: Alistair Popple Cc: Aneesh Kumar K.V Cc: Dan Williams Cc: Dave Jiang Cc: Gregory Price Cc: Len Brown Cc: Michal Hocko Cc: Rafael J. Wysocki Cc: Ravi Jonnalagadda Cc: SeongJae Park Cc: Tejun Heo Signed-off-by: Andrew Morton

memory tier: dax/kmem: introduce an abstract layer for finding, allocating, and putting memory types

2024-05-06T00:53:25+00:00

Patch series "Improved Memory Tier Creation for CPUless NUMA Nodes", v11. When a memory device, such as CXL1.1 type3 memory, is emulated as normal memory (E820_TYPE_RAM), the memory device is indistinguishable from normal DRAM in terms of memory tiering with the current implementation. The current memory tiering assigns all detected normal memory nodes to the same DRAM tier. This results in normal memory devices with different attributions being unable to be assigned to the correct memory tier, leading to the inability to migrate pages between different types of memory. https://lore.kernel.org/linux-mm/PH0PR08MB7955E9F08CCB64F23963B5C3A860A@PH0PR08MB7955.namprd08.prod.outlook.com/T/ This patchset automatically resolves the issues. It delays the initialization of memory tiers for CPUless NUMA nodes until they obtain HMAT information and after all devices are initialized at boot time, eliminating the need for user intervention. If no HMAT is specified, it falls back to using `default_dram_type`. Example usecase: We have CXL memory on the host, and we create VMs with a new system memory device backed by host CXL memory. We inject CXL memory performance attributes through QEMU, and the guest now sees memory nodes with performance attributes in HMAT. With this change, we enable the guest kernel to construct the correct memory tiering for the memory nodes. This patch (of 2): Since different memory devices require finding, allocating, and putting memory types, these common steps are abstracted in this patch, enhancing the scalability and conciseness of the code. Link: https://lkml.kernel.org/r/20240405000707.2670063-1-horenchuang@bytedance.com Link: https://lkml.kernel.org/r/20240405000707.2670063-2-horenchuang@bytedance.com Signed-off-by: Ho-Ren (Jack) Chuang Reviewed-by: "Huang, Ying" Reviewed-by: Jonathan Cameron Cc: Alistair Popple Cc: Aneesh Kumar K.V Cc: Dan Williams Cc: Dave Jiang Cc: Gregory Price Cc: Hao Xiang Cc: Jonathan Cameron Cc: Michal Hocko Cc: Ravi Jonnalagadda Cc: SeongJae Park Cc: Tejun Heo Cc: Vishal Verma Signed-off-by: Andrew Morton

base/node / acpi: Change 'node_hmem_attrs' to 'access_coordinates'

2023-12-22T22:23:13+00:00

Dan Williams suggested changing the struct 'node_hmem_attrs' to 'access_coordinates' [1]. The struct is a container of r/w-latency and r/w-bandwidth numbers. Moving forward, this container will also be used by CXL to store the performance characteristics of each link hop in the PCIE/CXL topology. So, where node_hmem_attrs is just the access parameters of a memory-node, access_coordinates applies more broadly to hardware topology characteristics. The observation is that seemed like an exercise in having the application identify "where" it falls on a spectrum of bandwidth and latency needs. For the tuple of read/write-latency and read/write-bandwidth, "coordinates" is not a perfect fit. Sometimes it is just conveying values in isolation and not a "location" relative to other performance points, but in the end this data is used to identify the performance operation point of a given memory-node. [2] Link: http://lore.kernel.org/r/64471313421f7_1b66294d5@dwillia2-xfh.jf.intel.com.notmuch/ Link: https://lore.kernel.org/linux-cxl/645e6215ee0de_1e6f2945e@dwillia2-xfh.jf.intel.com.notmuch/ Suggested-by: Dan Williams Reviewed-by: Dan Williams Reviewed-by: Jonathan Cameron Signed-off-by: Dave Jiang Acked-by: Greg Kroah-Hartman Link: https://lore.kernel.org/r/170319615734.2212653.15319394025985499185.stgit@djiang5-mobl3 Signed-off-by: Dan Williams

dax, kmem: calculate abstract distance with general interface

2023-10-16T22:44:39+00:00

Previously, a fixed abstract distance MEMTIER_DEFAULT_DAX_ADISTANCE is used for slow memory type in kmem driver. This limits the usage of kmem driver, for example, it cannot be used for HBM (high bandwidth memory). So, we use the general abstract distance calculation mechanism in kmem drivers to get more accurate abstract distance on systems with proper support. The original MEMTIER_DEFAULT_DAX_ADISTANCE is used as fallback only. Now, multiple memory types may be managed by kmem. These memory types are put into the "kmem_memory_types" list and protected by kmem_memory_type_lock. Link: https://lkml.kernel.org/r/20230926060628.265989-5-ying.huang@intel.com Signed-off-by: "Huang, Ying" Tested-by: Bharata B Rao Reviewed-by: Dave Jiang Reviewed-by: Alistair Popple Cc: Aneesh Kumar K.V Cc: Wei Xu Cc: Dan Williams Cc: Dave Hansen Cc: Davidlohr Bueso Cc: Johannes Weiner Cc: Jonathan Cameron Cc: Michal Hocko Cc: Yang Shi Cc: Rafael J Wysocki Signed-off-by: Andrew Morton

acpi, hmat: calculate abstract distance with HMAT

2023-10-16T22:44:39+00:00

A memory tiering abstract distance calculation algorithm based on ACPI HMAT is implemented. The basic idea is as follows. The performance attributes of system default DRAM nodes are recorded as the base line. Whose abstract distance is MEMTIER_ADISTANCE_DRAM. Then, the ratio of the abstract distance of a memory node (target) to MEMTIER_ADISTANCE_DRAM is scaled based on the ratio of the performance attributes of the node to that of the default DRAM nodes. The functions to record the read/write latency/bandwidth of the default DRAM nodes and calculate abstract distance according to read/write latency/bandwidth ratio will be used by CXL CDAT (Coherent Device Attribute Table) and other memory device drivers. So, they are put in memory-tiers.c. Link: https://lkml.kernel.org/r/20230926060628.265989-4-ying.huang@intel.com Signed-off-by: "Huang, Ying" Tested-by: Bharata B Rao Reviewed-by: Dave Jiang Reviewed-by: Alistair Popple Cc: Aneesh Kumar K.V Cc: Wei Xu Cc: Dan Williams Cc: Dave Hansen Cc: Davidlohr Bueso Cc: Johannes Weiner Cc: Jonathan Cameron Cc: Michal Hocko Cc: Yang Shi Cc: Rafael J Wysocki Signed-off-by: Andrew Morton

memory tiering: add abstract distance calculation algorithms management

2023-10-16T22:44:38+00:00

Patch series "memory tiering: calculate abstract distance based on ACPI HMAT", v4. We have the explicit memory tiers framework to manage systems with multiple types of memory, e.g., DRAM in DIMM slots and CXL memory devices. Where, same kind of memory devices will be grouped into memory types, then put into memory tiers. To describe the performance of a memory type, abstract distance is defined. Which is in direct proportion to the memory latency and inversely proportional to the memory bandwidth. To keep the code as simple as possible, fixed abstract distance is used in dax/kmem to describe slow memory such as Optane DCPMM. To support more memory types, in this series, we added the abstract distance calculation algorithm management mechanism, provided a algorithm implementation based on ACPI HMAT, and used the general abstract distance calculation interface in dax/kmem driver. So, dax/kmem can support HBM (high bandwidth memory) in addition to the original Optane DCPMM. This patch (of 4): The abstract distance may be calculated by various drivers, such as ACPI HMAT, CXL CDAT, etc. While it may be used by various code which hot-add memory node, such as dax/kmem etc. To decouple the algorithm users and the providers, the abstract distance calculation algorithms management mechanism is implemented in this patch. It provides interface for the providers to register the implementation, and interface for the users. Multiple algorithm implementations can cooperate via calculating abstract distance for different memory nodes. The preference of algorithm implementations can be specified via priority (notifier_block.priority). Link: https://lkml.kernel.org/r/20230926060628.265989-1-ying.huang@intel.com Link: https://lkml.kernel.org/r/20230926060628.265989-2-ying.huang@intel.com Signed-off-by: "Huang, Ying" Tested-by: Bharata B Rao Reviewed-by: Alistair Popple Reviewed-by: Dave Jiang Cc: Aneesh Kumar K.V Cc: Wei Xu Cc: Dan Williams Cc: Dave Hansen Cc: Davidlohr Bueso Cc: Johannes Weiner Cc: Jonathan Cameron Cc: Michal Hocko Cc: Yang Shi Cc: Rafael J Wysocki Signed-off-by: Andrew Morton

acpi,mm: fix typo sibiling -> sibling

2023-10-06T21:44:10+00:00

First found this typo as reviewing memory tier code. Fix it by sed like: $ sed -i 's/sibiling/sibling/g' $(git grep -l sibiling) so the acpi one will be corrected as well. Link: https://lkml.kernel.org/r/20230802092856.819328-1-lizhijian@cn.fujitsu.com Signed-off-by: Li Zhijian Cc: Aneesh Kumar K.V Cc: Huang, Ying Cc: Len Brown Cc: "Rafael J. Wysocki" Signed-off-by: Andrew Morton

memory tier: rename destroy_memory_type() to put_memory_type()

2023-08-18T17:12:11+00:00

It appears that destroy_memory_type() isn't a very good name because we usually will not free the memory_type here. So rename it to a more appropriate name i.e. put_memory_type(). Link: https://lkml.kernel.org/r/20230706063905.543800-1-linmiaohe@huawei.com Signed-off-by: Miaohe Lin Suggested-by: Huang, Ying Reviewed-by: "Huang, Ying" Reviewed-by: Xiao Yang Cc: Dan Williams Cc: Dave Jiang Cc: Vishal Verma Signed-off-by: Andrew Morton

memory: move hotplug memory notifier priority to same file for easy sorting

2022-11-09T01:37:17+00:00

The priority of hotplug memory callback is defined in a different file. And there are some callers using numbers directly. Collect them together into include/linux/memory.h for easy reading. This allows us to sort their priorities more intuitively without additional comments. Link: https://lkml.kernel.org/r/20220923033347.3935160-9-liushixin2@huawei.com Signed-off-by: Liu Shixin Cc: Christoph Lameter Cc: David Hildenbrand Cc: Kefeng Wang Cc: Waiman Long Cc: zefan li Signed-off-by: Andrew Morton