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.

comment: <> (SPDX-License-Identifier: BSD-3-Clause comment: <> (Copyright 2020, Intel Corporation)

NAME
SYNOPSIS
DESCRIPTION
RETURN VALUE
SEE ALSO

NAME

pmem2_get_memmove_fn(), pmem2_get_memset_fn(), pmem2_get_memcpy_fn() - get a function that provides optimized copying to persistent memory

SYNOPSIS

#include <libpmem2.h>

typedef void *(*pmem2_memmove_fn)(void *pmemdest, const void *src, size_t len,
		unsigned flags);
typedef void *(*pmem2_memcpy_fn)(void *pmemdest, const void *src, size_t len,
		unsigned flags);
typedef void *(*pmem2_memset_fn)(void *pmemdest, int c, size_t len,
		unsigned flags);

struct pmem2_map;

pmem2_memmove_fn pmem2_get_memmove_fn(struct pmem2_map *map);
pmem2_memset_fn pmem2_get_memset_fn(struct pmem2_map *map);
pmem2_memcpy_fn pmem2_get_memcpy_fn(struct pmem2_map *map);

DESCRIPTION

The pmem2_get_memmove_fn(), pmem2_get_memset_fn(), pmem2_get_memcpy_fn() functions return a pointer to a function responsible for efficient storing and flushing of data for mapping map.

pmem2_memmove_fn(), pmem2_memset_fn() and pmem2_memcpy_fn() functions provide the same memory copying functionalities as their namesakes memmove(3), memcpy(3) and memset(3), and ensure that the result has been flushed to persistence before returning (unless PMEM2_F_MEM_NOFLUSH flag was used).

For example, the following code:

        memmove(dest, src, len);
        pmem2_persist_fn persist_fn = pmem2_get_persist_fn(map);
        persist_fn(dest, len);

is functionally equivalent to:

        pmem2_memmove_fn memmove_fn = pmem2_get_memmove_fn(map);
        memmove_fn(dest, src, len, 0);

Unlike libc implementation, libpmem2 functions guarantee that if destination buffer address and length are 8 byte aligned then all stores will be performed using at least 8 byte store instructions. This means that a series of 8 byte stores followed by persist_fn can be safely replaced by a single memmove_fn call.

The flags argument of all of the above functions has the same meaning. It can be 0 or a bitwise OR of one or more of the following flags:

  • PMEM2_F_MEM_NODRAIN - modifies the behavior to skip the final pmem2_drain_fn step. This allows applications to optimize cases where several ranges are being copied to persistent memory, followed by a single call to pmem2_drain_fn. The following example illustrates how this flag might be used to avoid multiple calls to pmem2_drain_fn when copying several ranges of memory to pmem:
pmem2_memcpy_fn memcpy_fn = pmem2_get_memcpy_fn(map);
pmem2_drain_fn drain_fn = pmem2_get_drain_fn(map);

/* ... write several ranges to pmem ... */
memcpy_fn(pmemdest1, src1, len1, PMEM2_F_MEM_NODRAIN);
memcpy_fn(pmemdest2, src2, len2, PMEM2_F_MEM_NODRAIN);

/* ... */

/* wait for any pmem stores to drain from HW buffers */
drain_fn();
  • PMEM2_F_MEM_NOFLUSH - Don’t flush anything. This implies PMEM2_F_MEM_NODRAIN. Using this flag only makes sense when it’s followed by any function that flushes data.

The remaining flags say how the operation should be done, and are merely hints.

  • PMEM2_F_MEM_NONTEMPORAL - Use non-temporal instructions. This flag is mutually exclusive with PMEM2_F_MEM_TEMPORAL. On x86_64 this flag is mutually exclusive with PMEM2_F_MEM_NOFLUSH.

  • PMEM2_F_MEM_TEMPORAL - Use temporal instructions. This flag is mutually exclusive with PMEM2_F_MEM_NONTEMPORAL.

  • PMEM2_F_MEM_WC - Use write combining mode. This flag is mutually exclusive with PMEM2_F_MEM_WB. On x86_64 this flag is mutually exclusive with PMEM2_F_MEM_NOFLUSH.

  • PMEM2_F_MEM_WB - Use write back mode. This flag is mutually exclusive with PMEM2_F_MEM_WC. On x86_64 this is an alias for PMEM2_F_MEM_TEMPORAL.

Using an invalid combination of flags has undefined behavior.

Without any of the above flags libpmem2 will try to guess the best strategy based on the data size. See PMEM_MOVNT_THRESHOLD description in libpmem2(7) for details.

RETURN VALUE

The pmem2_get_memmove_fn(), pmem2_get_memset_fn(), pmem2_get_memcpy_fn() functions never return NULL.

They return the same function for the same mapping.

This means that it’s safe to cache their return values. However, these functions are very cheap (because their return values are precomputed), so caching may not be necessary.

If two (or more) mappings share the same pmem2_memmove_fn, pmem2_memset_fn, pmem2_memcpy_fn and they are adjacent to each other, it is safe to call these functions for a range spanning those mappings.

SEE ALSO

memcpy(3), memmove(3), memset(3), pmem2_get_drain_fn(3), pmem2_get_memcpy_fn(3), pmem2_get_memset_fn(3), pmem2_map_new(3), pmem2_get_persist_fn(3), libpmem2(7) and https://pmem.io

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