NVM Library

The libpmem library

libpmem provides low level persistent memory support. In particular, support for the persistent memory instructions for flushing changes to pmem is provided.

This library is provided for software which tracks every store to pmem and needs to flush those changes to durability. Most developers will find higher level libraries like libpmemobj to be much more convenient.

Man pages that contains a list of the interfaces provided:

libpmem Examples

The Basics

If you’ve decided to handle persistent memory allocation and consistency across program interruption yourself, you will find the functions in libpmem useful. It is important to understand that programming to raw pmem means you must create your own transactions or convince yourself you don’t care if a system or program crash leaves your pmem files in an inconsistent state. Libraries like libpmemobj provide transactional interfaces by building on these libpmem functions, but the interfaces in libpmem are non-transactional.

To illustrate the basics, let’s walk through the man page example first:

The example starts, as shown above, by including the necessary headers. Line 45 (the highlighted line) shows the header file you need to include to use libpmem: libpmem.h.

For this simple example, we’re just going to hard code a pmem file size of 4 kilobytes.

The lines above create the file, make sure 4k is allocated, and map the file into memory. This illustrates one of the helper functions in libpmem: pmem_map() which takes a file descriptor and calls mmap(2) to memory map the entire file. Calling mmap() directly will work just fine – the main advantage of pmem_map() is that it tries to find an address where mapping is likely to use large page mappings, for better performance when using large ranges of pmem.

Note that once the pmem file is mapped into memory, it is not necessary to keep the file descriptor open.

Since the system calls for memory mapping persistent memory are the same as the POSIX calls for memory mapping any file, you may want to write your code to run correctly when given either a pmem file or a file on a traditional file system. For many decades it has been the case that changes written to a memory mapped range of a file may not be persistent until flushed to the media. One common way to do this is using the POSIX call msync(2). If you write your program to memory map a file and use msync() every time you want to flush the changes media, it will work correctly for pmem as well as files on a traditional file system. However, you may find your program performs better if you detect pmem explicitly and use libpmem to flush changes in that case.

The libpmem function pmem_is_pmem() can be used to determine if the memory in the given range is really persistent memory or if it is just a memory mapped file on a traditional file system. Using this call in your program will allow you to decide what to do when given a non-pmem file. Your program could decide to print an error message and exit (for example: “ERROR: This program only works on pmem”). But it seems more likely you will want to save the result of pmem_is_pmem() as shown above, and then use that flag to decide what to do when flushing changes to persistence as later in this example program.

The novel thing about pmem is you can copy to it directly, like any memory. The strcpy() call shown on line 80 above is just the usual libc function that stores a string to memory. If this example program were to be interrupted either during or just after the strcpy() call, you can’t be sure which parts of the string made it all the way to the media. It might be none of the string, all of the string, or somewhere in-between. In addition, there’s no guarantee the string will make it to the media in the order it was stored! For longer ranges, it is just as likely that portions copied later make it to the media before earlier portions. (So don’t write code like the example above and then expect to check for zeros to see how much of the string was written.)

How can a string get stored in seemingly random order? The reason is that until a flush function like msync() has returned successfully, the normal cache pressure that happens on an active system can push changes out to the media at any time in any order. Most processors have barrier instructions (like SFENCE on the Intel platform) but those instructions deal with ordering in the visibility of stores to other threads, not with the order that changes reach persistence. The only barriers for flushing to persistence are functions like msync() or pmem_persist() as shown below.

As shown above, this example uses the is_pmem flag it saved from the previous call to pmem_is_pmem(). This is the recommended way to use this information rather than calling pmem_is_pmem() each time you want to make changes durable. That’s because pmem_is_pmem() can have a high overhead, having to search through data structures to ensure the entire range is really persistent memory.

For true pmem, the highlighted line 84 above is the most optimal way to flush changes to persistence. pmem_persist() will, if possible, perform the flush directly from user space, without calling into the OS. This is made possible on the Intel platform using instructions like CLWB and CLFLUSHOPT which are described in Intel’s manuals. Of course you are free to use these instructions directly in your program, but the program will crash with an undefined opcode if you try to use the instructions on a platform that doesn’t support them. This is where libpmem helps you out, by checking the platform capabilities on start-up and choosing the best instructions for each operation it supports.

The above example also uses pmem_msync() for the non-pmem case instead of calling msync(2) directly. For convenience, the pmem_msync() call is a small wrapper around msync() that ensures the arguments are aligned, as requirement of POSIX.

Buildable source for the libpmem manpage.c example above is available in the NVML repository.

Copying to Persistent Memory

Another feature of libpmem is a set of routines for optimally copying to persistent memory. These functions perform the same functions as the libc functions memcpy(), memset(), and memmove(), but they are optimized for copying to pmem. On the Intel platform, this is done using the non-temporal store instructions which bypass the processor caches (eliminating the need to flush that portion of the data path).

The first copy example, called simple_copy, illustrates how pmem_memcpy() is used.

The highlighted line, line 105 above, shows how pmem_memcpy() is used just like memcpy(3) except that when the destination is pmem, libpmem handles flushing the data to persistence as part of the copy.

Buildable source for the libpmem simple_copy.c example above is available in the NVML repository.

Separating the Flush Steps

There are two steps in flushing to persistence. The first step is to flush the processor caches, or bypass them entirely as explained in the previous example. The second step is to wait for any hardware buffers to drain, to ensure writes have reached the media. These steps are performed together when pmem_persist() is called, or they can be called individually by calling pmem_flush() for the first step and pmem_drain() for the second. Note that either of these steps may be unnecessary on a given platform, and the library knows how to check for that and do the right thing. For example, on Intel platforms, pmem_drain() is an empty function.

When does it make sense to break flushing into steps? This example, called full_copy illustrates one reason you might do this. Since the example copies data using multiple calls to memcpy(), it uses the version of libpmem copy that only performs the flush, postponing the final drain step to the end. This works because unlike the flush step, the drain step does not take an address range – it is a system-wide drain operation so can happen at the end of the loop that copies individual blocks of data.

As each block is copied, line 65 in the above example copies a block of data to pmem, effectively flushing it from the processor caches. But rather than waiting for the hardware queues to drain each time, that step is saved until the end, as shown on line 75 above.

Buildable source for the libpmem full_copy.c example above is available in the NVML repository.