pmem2 API version 1.0

The PMDK repository on GitHub is the ultimate source of information on PMDK from release 2.0! For all questions and to submit eventual issues please follow to that repository. The PMDK documentation collected here should be valid up to the 1.13.1 release but is maintained only on a best-effort basis and may not reflect the latest state of the art.



libpmem2 - persistent memory support library

NOTE: Support for Windows and FreeBSD deprecated since PMDK 1.13.0 release and will be removed in the PMDK 2.0.0 release.


#include <libpmem2.h>
cc ... -lpmem2


libpmem2 provides low-level persistent memory (pmem) support for applications using direct access storage (DAX), which is storage that supports load/store access without paging blocks from a block storage device. Some types of non-volatile memory DIMMs (NVDIMMs) provide this type of byte addressable access to storage. A persistent memory aware file system is typically used to expose the direct access to applications. Memory mapping a file from this type of file system results in the load/store, non-paged access to pmem.

This library is for applications that use persistent memory directly, without the help of any library-supplied transactions or memory allocation. Higher-level libraries that currently build on libpmem (previous variation of libpmem2) are available and are recommended for most applications, see:

  • libpmemobj(7), a general use persistent memory API, providing memory allocation and transactional operations on variable-sized objects.

  • libpmemblk(7), providing pmem-resident arrays of fixed-sized blocks with atomic updates.

  • libpmemlog(7), providing a pmem-resident log file.

The libpmem2 library provides a comprehensive set of functions for robust use of Persistent Memory. It relies on three core concepts: struct pmem2_src source, struct pmem2_config config and struct pmem2_map map:

  • source - an object describing the data source for mapping. The data source can be a file descriptor, a file handle, or an anonymous mapping. APIs dedicated for creating source are: pmem2_source_from_fd(3), pmem2_source_from_handle(3), pmem2_source_from_anon(3).

  • config - an object containing parameters that are used to create a mapping from a source. The configuration structure must always be provided to create a mapping, but the only required parameter to set in the config is granularity. The granularity should by set using dedicated libpmem2 function pmem2_config_set_required_store_granularity(3) which defines a maximum permitted granularity requested by the user. For more information about the granularity concept read GRANULARITY section below.

In addition to the granularity setting, libpmem2 provides multiple optional functions to configure target mapping, e.g., pmem2_config_set_length(3) to set length which will be used for mapping, or pmem2_config_set_offset(3) which will be used to map the contents from the specified location of the source, pmem2_config_set_sharing(3) which defines the behavior and visibility of writes to the mapping’s pages.

  • map - an object created by pmem2_map_new(3) using source and config as an input parameters. The map structure can be then used to directly operate on the created mapping through the use of its associated set of functions: pmem2_map_get_address(3), pmem2_map_get_size(3), pmem2_map_get_store_granularity(3) - for getting address, size and effective mapping granularity.

In addition to the basic functionality of managing the virtual address mapping, libpmem2 also provides optimized functions for modifying the mapped data. This includes data flushing as well as memory copying.

To get proper function for data flushing use: pmem2_get_flush_fn(3), pmem2_get_persist_fn(3) or pmem2_get_drain_fn(3). To get proper function for copying to persistent memory, use map getters: pmem2_get_memcpy_fn(3), pmem2_get_memset_fn(3), pmem2_get_memmove_fn(3).

The libpmem2 API also provides support for the badblock and unsafe shutdown state handling.

To read or clear badblocks, the following functions are provided: pmem2_badblock_context_new(3), pmem2_badblock_context_delete(3), pmem2_badblock_next(3) and pmem2_badblock_clear(3).

To handle unsafe shutdown in the application, the following functions are provided: pmem2_source_device_id(3), pmem2_source_device_usc(3). More detailed information about unsafe shutdown detection and unsafe shutdown count can be found in the libpmem2_unsafe_shutdown(7) man page.


The libpmem2 library introduces the concept of granularity through which you may easily distinguish between different levels of storage performance capabilities available to the application as related to power-fail protected domain. The way data reaches this protected domain differs based on the platform and storage device capabilities.

Traditional block storage devices (SSD, HDD) must use system API calls such as msync(), fsync() on Linux, or FlushFileBuffers(),FlushViewOfFile() on Windows to write data reliably. Invoking these functions flushes the data to the medium with page granularity. In the libpmem2 library, this type of flushing behavior is called PMEM2_GRANULARITY_PAGE.

In systems with persistent memory support, a power-fail protected domain may cover different sets of resources: either the memory controller or the memory controller and CPU caches. For this reason, libpmem2 distinguishes two types of granularity for persistent memory: PMEM2_GRANULARITY_CACHE_LINE and PMEM2_GRANULARITY_BYTE.

If the power-fail protected domain covers only the memory controller, the CPU appropriate cache lines must be flushed for the data to be considered persistent. This granularity type is called PMEM2_GRANULARITY_CACHE_LINE. Depending on the architecture, there are different types of machine instructions for flushing cache lines (e.g., CLWB, CLFLUSHOPT, CLFLUSH for Intel x86_64 architecture). Usually, to ensure the ordering of stores, such instructions must be followed by a barrier (e.g., SFENCE).

The third type of granularity PMEM2_GRANULARITY_BYTE applies to platforms where power-fail protected domain covers both the memory controller and CPU caches. In such cases, cache flush instructions are no longer needed, and the platform itself guarantees the persistence of data. But barriers might still be required for ordering.

The library declares these granularity level in pmem2_granularity enum, which the application must set in pmem2_config to the appropriate level for a mapping to succeed. The software should set this config parameter to a value that most accurately represents the target hardware characteristics and the storage patterns of the application. For example, a database storage engine that operates on large logical pages that reside either on SSDs or PMEM should set this value to PMEM2_GRANULARITY_PAGE. The library will create mappings where the new map granularity is lower or equal to the requested one. For example, a mapping with PMEM2_GRANULARITY_CACHE_LINE can be created for the required granularity PMEM2_GRANULARITY_PAGE, but not vice versa.


libpmem2 relies on the library destructor being called from the main thread. For this reason, all functions that might trigger destruction (e.g. dlclose(3)) should be called in the main thread. Otherwise some of the resources associated with that thread might not be cleaned up properly.


libpmem2 can change its default behavior based on the following environment variables. These are primarily intended for testing and are generally not required.


Setting this environment variable to val forces libpmem2 to use persist method specific for forced granularity and skip granularity autodetecting mechanism. The concept of the granularity is described in GRANULARITY section above. This variable is intended for use during library testing.

The val argument accepts following text values:

  • BYTE - force byte granularity.
  • CACHE_LINE - force cache line granularity.
  • PAGE - force page granularity.

Granularity values listed above are case-insensitive.

NOTE: The value of PMEM2_FORCE_GRANULARITY is not queried (and cached) at library initialization time, but read during each pmem2_map_new(3) call.

This means that PMEM2_FORCE_GRANULARITY may still be set or modified by the program until the first attempt to map a file.


Setting this environment variable to 1 forces libpmem2 to never issue the CLWB instruction on Intel hardware, falling back to other cache flush instructions on that hardware instead (CLFLUSHOPT or CLFLUSH). Without this setting, libpmem2 will always use the CLWB instruction for flushing processor caches on platforms that support this instruction. This variable is intended for use during library testing, but may be required for some rare cases when using CLWB has a negative impact on performance.


Setting this environment variable to 1 forces libpmem2 to never issue the CLFLUSHOPT instruction on Intel hardware, falling back to the CLFLUSH instructions instead. Without this environment variable, libpmem2 will always use the CLFLUSHOPT instruction for flushing processor caches on platforms that support the instruction, but where CLWB is not available. This variable is intended for use during library testing.


Setting this environment variable to 1 forces libpmem2 to never use the non-temporal move instructions on Intel hardware. Without this environment variable, libpmem2 will use the non-temporal instructions for copying larger ranges to persistent memory on platforms that support these instructions. This variable is intended for use during library testing.


This environment variable allows overriding the minimum length of the pmem2_memmove_fn operations, for which libpmem2 uses non-temporal move instructions. Setting this environment variable to 0 forces libpmem2 to always use the non-temporal move instructions if available. It has no effect if PMEM_NO_MOVNT is set to 1. This variable is intended for use during library testing.


Two versions of libpmem2 are typically available on a development system. The normal version, accessed when a program is linked using the -lpmem2 option, is optimized for performance. That version skips checks that impact performance and never logs any trace information or performs any run-time assertions.

A second version of libpmem2, accessed when a program uses the libraries under /pmdk/src/x64/Debug, contains run-time assertions and trace points. The typical way to access the debug version is to set the environment variable LD_LIBRARY_PATH to /pmdk/src/x64/Debug. Debugging output is controlled using the following environment variables. These variables have no effect on the non-debug version of the library.


The value of PMEM2_LOG_LEVEL enables trace points in the debug version of the library, as follows:

  • 0 - This is the default level when PMEM2_LOG_LEVEL is not set. No log messages are emitted at this level.

  • 1 - Additional details on any errors detected are logged, in addition to returning the errno-based errors as usual. The same information may be retrieved using pmem2_errormsgU()/pmem2_errormsgW().

  • 2 - A trace of basic operations is logged.

  • 3 - Enables a very verbose amount of function call tracing in the library.

  • 4 - Enables voluminous and fairly obscure tracing information that is likely only useful to the libpmem2 developers.

Unless PMEM2_LOG_FILE is set, debugging output is written to stderr.


Specifies the name of a file where all logging information should be written. If the last character in the name is “-”, the PID of the current process will be appended to the file name when the log file is created. If PMEM2_LOG_FILE is not set, output is written to stderr.


The following example uses libpmem2 to flush changes made to raw, memory-mapped persistent memory.

WARNING: There is nothing transactional about the persist from pmem2_get_persist_fn(3) call in this example. Interrupting the program may result in a partial write to pmem. Use a transactional library such as libpmemobj(7) to avoid torn updates.

The basic example can be found in the repository under path src/examples/libpmem2/basic/basic.c. It is described in detail here.


libpmem2 builds on the persistent memory programming model recommended by the SNIA NVM Programming Technical Work Group:


FlushFileBuffers(), fsync(2), msync(2), pmem2_config_set_length(3), pmem2_config_set_offset(3), pmem2_config_set_required_store_granularity(3), pmem2_config_set_sharing(3),pmem2_get_drain_fn(3), pmem2_get_flush_fn(3), pmem2_get_memcpy_fn(3), pmem2_get_memmove_fn(3), pmem2_get_memset_fn(3), pmem2_get_persist_fn(3),pmem2_map_get_store_granularity(3), pmem2_map_new(3), pmem2_source_from_anon(3), pmem2_source_from_fd(3), pmem2_source_from_handle(3), libpmem2_unsafe_shutdown(7), libpmemblk(7), libpmemlog(7), libpmemobj(7) and

The contents of this web site and the associated GitHub repositories are BSD-licensed open source.