1 .. SPDX-License-Identifier: BSD-3-Clause
2 Copyright(c) 2010-2014 Intel Corporation.
4 .. _Environment_Abstraction_Layer:
6 Environment Abstraction Layer
7 =============================
9 The Environment Abstraction Layer (EAL) is responsible for gaining access to low-level resources such as hardware and memory space.
10 It provides a generic interface that hides the environment specifics from the applications and libraries.
11 It is the responsibility of the initialization routine to decide how to allocate these resources
12 (that is, memory space, PCI devices, timers, consoles, and so on).
14 Typical services expected from the EAL are:
16 * DPDK Loading and Launching:
17 The DPDK and its application are linked as a single application and must be loaded by some means.
19 * Core Affinity/Assignment Procedures:
20 The EAL provides mechanisms for assigning execution units to specific cores as well as creating execution instances.
22 * System Memory Reservation:
23 The EAL facilitates the reservation of different memory zones, for example, physical memory areas for device interactions.
25 * PCI Address Abstraction: The EAL provides an interface to access PCI address space.
27 * Trace and Debug Functions: Logs, dump_stack, panic and so on.
29 * Utility Functions: Spinlocks and atomic counters that are not provided in libc.
31 * CPU Feature Identification: Determine at runtime if a particular feature, for example, IntelĀ® AVX is supported.
32 Determine if the current CPU supports the feature set that the binary was compiled for.
34 * Interrupt Handling: Interfaces to register/unregister callbacks to specific interrupt sources.
36 * Alarm Functions: Interfaces to set/remove callbacks to be run at a specific time.
38 EAL in a Linux-userland Execution Environment
39 ---------------------------------------------
41 In a Linux user space environment, the DPDK application runs as a user-space application using the pthread library.
42 PCI information about devices and address space is discovered through the /sys kernel interface and through kernel modules such as uio_pci_generic, or igb_uio.
43 Refer to the UIO: User-space drivers documentation in the Linux kernel. This memory is mmap'd in the application.
45 The EAL performs physical memory allocation using mmap() in hugetlbfs (using huge page sizes to increase performance).
46 This memory is exposed to DPDK service layers such as the :ref:`Mempool Library <Mempool_Library>`.
48 At this point, the DPDK services layer will be initialized, then through pthread setaffinity calls,
49 each execution unit will be assigned to a specific logical core to run as a user-level thread.
51 The time reference is provided by the CPU Time-Stamp Counter (TSC) or by the HPET kernel API through a mmap() call.
53 Initialization and Core Launching
54 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
56 Part of the initialization is done by the start function of glibc.
57 A check is also performed at initialization time to ensure that the micro architecture type chosen in the config file is supported by the CPU.
58 Then, the main() function is called. The core initialization and launch is done in rte_eal_init() (see the API documentation).
59 It consist of calls to the pthread library (more specifically, pthread_self(), pthread_create(), and pthread_setaffinity_np()).
61 .. _figure_linuxapp_launch:
63 .. figure:: img/linuxapp_launch.*
65 EAL Initialization in a Linux Application Environment
70 Initialization of objects, such as memory zones, rings, memory pools, lpm tables and hash tables,
71 should be done as part of the overall application initialization on the master lcore.
72 The creation and initialization functions for these objects are not multi-thread safe.
73 However, once initialized, the objects themselves can safely be used in multiple threads simultaneously.
78 During the initialization of EAL resources such as hugepage backed memory can be
79 allocated by core components. The memory allocated during ``rte_eal_init()``
80 can be released by calling the ``rte_eal_cleanup()`` function. Refer to the
81 API documentation for details.
86 The Linuxapp EAL allows a multi-process as well as a multi-threaded (pthread) deployment model.
88 :ref:`Multi-process Support <Multi-process_Support>` for more details.
90 Memory Mapping Discovery and Memory Reservation
91 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
93 The allocation of large contiguous physical memory is done using the hugetlbfs kernel filesystem.
94 The EAL provides an API to reserve named memory zones in this contiguous memory.
95 The physical address of the reserved memory for that memory zone is also returned to the user by the memory zone reservation API.
97 There are two modes in which DPDK memory subsystem can operate: dynamic mode,
98 and legacy mode. Both modes are explained below.
102 Memory reservations done using the APIs provided by rte_malloc are also backed by pages from the hugetlbfs filesystem.
104 + Dynamic memory mode
106 Currently, this mode is only supported on Linux.
108 In this mode, usage of hugepages by DPDK application will grow and shrink based
109 on application's requests. Any memory allocation through ``rte_malloc()``,
110 ``rte_memzone_reserve()`` or other methods, can potentially result in more
111 hugepages being reserved from the system. Similarly, any memory deallocation can
112 potentially result in hugepages being released back to the system.
114 Memory allocated in this mode is not guaranteed to be IOVA-contiguous. If large
115 chunks of IOVA-contiguous are required (with "large" defined as "more than one
116 page"), it is recommended to either use VFIO driver for all physical devices (so
117 that IOVA and VA addresses can be the same, thereby bypassing physical addresses
118 entirely), or use legacy memory mode.
120 For chunks of memory which must be IOVA-contiguous, it is recommended to use
121 ``rte_memzone_reserve()`` function with ``RTE_MEMZONE_IOVA_CONTIG`` flag
122 specified. This way, memory allocator will ensure that, whatever memory mode is
123 in use, either reserved memory will satisfy the requirements, or the allocation
126 There is no need to preallocate any memory at startup using ``-m`` or
127 ``--socket-mem`` command-line parameters, however it is still possible to do so,
128 in which case preallocate memory will be "pinned" (i.e. will never be released
129 by the application back to the system). It will be possible to allocate more
130 hugepages, and deallocate those, but any preallocated pages will not be freed.
131 If neither ``-m`` nor ``--socket-mem`` were specified, no memory will be
132 preallocated, and all memory will be allocated at runtime, as needed.
134 Another available option to use in dynamic memory mode is
135 ``--single-file-segments`` command-line option. This option will put pages in
136 single files (per memseg list), as opposed to creating a file per page. This is
137 normally not needed, but can be useful for use cases like userspace vhost, where
138 there is limited number of page file descriptors that can be passed to VirtIO.
140 If the application (or DPDK-internal code, such as device drivers) wishes to
141 receive notifications about newly allocated memory, it is possible to register
142 for memory event callbacks via ``rte_mem_event_callback_register()`` function.
143 This will call a callback function any time DPDK's memory map has changed.
145 If the application (or DPDK-internal code, such as device drivers) wishes to be
146 notified about memory allocations above specified threshold (and have a chance
147 to deny them), allocation validator callbacks are also available via
148 ``rte_mem_alloc_validator_callback_register()`` function.
150 A default validator callback is provided by EAL, which can be enabled with a
151 ``--socket-limit`` command-line option, for a simple way to limit maximum amount
152 of memory that can be used by DPDK application.
156 In multiprocess scenario, all related processes (i.e. primary process, and
157 secondary processes running with the same prefix) must be in the same memory
158 modes. That is, if primary process is run in dynamic memory mode, all of its
159 secondary processes must be run in the same mode. The same is applicable to
160 ``--single-file-segments`` command-line option - both primary and secondary
161 processes must shared this mode.
165 This mode is enabled by specifying ``--legacy-mem`` command-line switch to the
166 EAL. This switch will have no effect on FreeBSD as FreeBSD only supports
169 This mode mimics historical behavior of EAL. That is, EAL will reserve all
170 memory at startup, sort all memory into large IOVA-contiguous chunks, and will
171 not allow acquiring or releasing hugepages from the system at runtime.
173 If neither ``-m`` nor ``--socket-mem`` were specified, the entire available
174 hugepage memory will be preallocated.
178 Additional restrictions are present when running in 32-bit mode. In dynamic
179 memory mode, by default maximum of 2 gigabytes of VA space will be preallocated,
180 and all of it will be on master lcore NUMA node unless ``--socket-mem`` flag is
183 In legacy mode, VA space will only be preallocated for segments that were
184 requested (plus padding, to keep IOVA-contiguousness).
186 + Maximum amount of memory
188 All possible virtual memory space that can ever be used for hugepage mapping in
189 a DPDK process is preallocated at startup, thereby placing an upper limit on how
190 much memory a DPDK application can have. DPDK memory is stored in segment lists,
191 each segment is strictly one physical page. It is possible to change the amount
192 of virtual memory being preallocated at startup by editing the following config
195 * ``CONFIG_RTE_MAX_MEMSEG_LISTS`` controls how many segment lists can DPDK have
196 * ``CONFIG_RTE_MAX_MEM_MB_PER_LIST`` controls how much megabytes of memory each
197 segment list can address
198 * ``CONFIG_RTE_MAX_MEMSEG_PER_LIST`` controls how many segments each segment can
200 * ``CONFIG_RTE_MAX_MEMSEG_PER_TYPE`` controls how many segments each memory type
201 can have (where "type" is defined as "page size + NUMA node" combination)
202 * ``CONFIG_RTE_MAX_MEM_MB_PER_TYPE`` controls how much megabytes of memory each
203 memory type can address
204 * ``CONFIG_RTE_MAX_MEM_MB`` places a global maximum on the amount of memory
207 Normally, these options do not need to be changed.
211 Preallocated virtual memory is not to be confused with preallocated hugepage
212 memory! All DPDK processes preallocate virtual memory at startup. Hugepages
213 can later be mapped into that preallocated VA space (if dynamic memory mode
214 is enabled), and can optionally be mapped into it at startup.
216 Support for Externally Allocated Memory
217 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
219 It is possible to use externally allocated memory in DPDK, using a set of malloc
220 heap API's. Support for externally allocated memory is implemented through
221 overloading the socket ID - externally allocated heaps will have socket ID's
222 that would be considered invalid under normal circumstances. Requesting an
223 allocation to take place from a specified externally allocated memory is a
224 matter of supplying the correct socket ID to DPDK allocator, either directly
225 (e.g. through a call to ``rte_malloc``) or indirectly (through data
226 structure-specific allocation API's such as ``rte_ring_create``).
228 Since there is no way DPDK can verify whether memory are is available or valid,
229 this responsibility falls on the shoulders of the user. All multiprocess
230 synchronization is also user's responsibility, as well as ensuring that all
231 calls to add/attach/detach/remove memory are done in the correct order. It is
232 not required to attach to a memory area in all processes - only attach to memory
235 The expected workflow is as follows:
237 * Get a pointer to memory area
238 * Create a named heap
239 * Add memory area(s) to the heap
240 - If IOVA table is not specified, IOVA addresses will be assumed to be
241 unavailable, and DMA mappings will not be performed
242 - Other processes must attach to the memory area before they can use it
243 * Get socket ID used for the heap
244 * Use normal DPDK allocation procedures, using supplied socket ID
245 * If memory area is no longer needed, it can be removed from the heap
246 - Other processes must detach from this memory area before it can be removed
247 * If heap is no longer needed, remove it
248 - Socket ID will become invalid and will not be reused
250 For more information, please refer to ``rte_malloc`` API documentation,
251 specifically the ``rte_malloc_heap_*`` family of function calls.
256 The EAL uses the /sys/bus/pci utilities provided by the kernel to scan the content on the PCI bus.
257 To access PCI memory, a kernel module called uio_pci_generic provides a /dev/uioX device file
258 and resource files in /sys
259 that can be mmap'd to obtain access to PCI address space from the application.
260 The DPDK-specific igb_uio module can also be used for this. Both drivers use the uio kernel feature (userland driver).
262 Per-lcore and Shared Variables
263 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
267 lcore refers to a logical execution unit of the processor, sometimes called a hardware *thread*.
269 Shared variables are the default behavior.
270 Per-lcore variables are implemented using *Thread Local Storage* (TLS) to provide per-thread local storage.
275 A logging API is provided by EAL.
276 By default, in a Linux application, logs are sent to syslog and also to the console.
277 However, the log function can be overridden by the user to use a different logging mechanism.
279 Trace and Debug Functions
280 ^^^^^^^^^^^^^^^^^^^^^^^^^
282 There are some debug functions to dump the stack in glibc.
283 The rte_panic() function can voluntarily provoke a SIG_ABORT,
284 which can trigger the generation of a core file, readable by gdb.
286 CPU Feature Identification
287 ~~~~~~~~~~~~~~~~~~~~~~~~~~
289 The EAL can query the CPU at runtime (using the rte_cpu_get_features() function) to determine which CPU features are available.
291 User Space Interrupt Event
292 ~~~~~~~~~~~~~~~~~~~~~~~~~~
294 + User Space Interrupt and Alarm Handling in Host Thread
296 The EAL creates a host thread to poll the UIO device file descriptors to detect the interrupts.
297 Callbacks can be registered or unregistered by the EAL functions for a specific interrupt event
298 and are called in the host thread asynchronously.
299 The EAL also allows timed callbacks to be used in the same way as for NIC interrupts.
303 In DPDK PMD, the only interrupts handled by the dedicated host thread are those for link status change
304 (link up and link down notification) and for sudden device removal.
309 The receive and transmit routines provided by each PMD don't limit themselves to execute in polling thread mode.
310 To ease the idle polling with tiny throughput, it's useful to pause the polling and wait until the wake-up event happens.
311 The RX interrupt is the first choice to be such kind of wake-up event, but probably won't be the only one.
313 EAL provides the event APIs for this event-driven thread mode.
314 Taking linuxapp as an example, the implementation relies on epoll. Each thread can monitor an epoll instance
315 in which all the wake-up events' file descriptors are added. The event file descriptors are created and mapped to
316 the interrupt vectors according to the UIO/VFIO spec.
317 From bsdapp's perspective, kqueue is the alternative way, but not implemented yet.
319 EAL initializes the mapping between event file descriptors and interrupt vectors, while each device initializes the mapping
320 between interrupt vectors and queues. In this way, EAL actually is unaware of the interrupt cause on the specific vector.
321 The eth_dev driver takes responsibility to program the latter mapping.
325 Per queue RX interrupt event is only allowed in VFIO which supports multiple MSI-X vector. In UIO, the RX interrupt
326 together with other interrupt causes shares the same vector. In this case, when RX interrupt and LSC(link status change)
327 interrupt are both enabled(intr_conf.lsc == 1 && intr_conf.rxq == 1), only the former is capable.
329 The RX interrupt are controlled/enabled/disabled by ethdev APIs - 'rte_eth_dev_rx_intr_*'. They return failure if the PMD
330 hasn't support them yet. The intr_conf.rxq flag is used to turn on the capability of RX interrupt per device.
332 + Device Removal Event
334 This event is triggered by a device being removed at a bus level. Its
335 underlying resources may have been made unavailable (i.e. PCI mappings
336 unmapped). The PMD must make sure that on such occurrence, the application can
337 still safely use its callbacks.
339 This event can be subscribed to in the same way one would subscribe to a link
340 status change event. The execution context is thus the same, i.e. it is the
341 dedicated interrupt host thread.
343 Considering this, it is likely that an application would want to close a
344 device having emitted a Device Removal Event. In such case, calling
345 ``rte_eth_dev_close()`` can trigger it to unregister its own Device Removal Event
346 callback. Care must be taken not to close the device from the interrupt handler
347 context. It is necessary to reschedule such closing operation.
352 The EAL PCI device blacklist functionality can be used to mark certain NIC ports as blacklisted,
353 so they are ignored by the DPDK.
354 The ports to be blacklisted are identified using the PCIe* description (Domain:Bus:Device.Function).
359 Locks and atomic operations are per-architecture (i686 and x86_64).
361 Memory Segments and Memory Zones (memzone)
362 ------------------------------------------
364 The mapping of physical memory is provided by this feature in the EAL.
365 As physical memory can have gaps, the memory is described in a table of descriptors,
366 and each descriptor (called rte_memseg ) describes a physical page.
368 On top of this, the memzone allocator's role is to reserve contiguous portions of physical memory.
369 These zones are identified by a unique name when the memory is reserved.
371 The rte_memzone descriptors are also located in the configuration structure.
372 This structure is accessed using rte_eal_get_configuration().
373 The lookup (by name) of a memory zone returns a descriptor containing the physical address of the memory zone.
375 Memory zones can be reserved with specific start address alignment by supplying the align parameter
376 (by default, they are aligned to cache line size).
377 The alignment value should be a power of two and not less than the cache line size (64 bytes).
378 Memory zones can also be reserved from either 2 MB or 1 GB hugepages, provided that both are available on the system.
380 Both memsegs and memzones are stored using ``rte_fbarray`` structures. Please
381 refer to *DPDK API Reference* for more information.
387 DPDK usually pins one pthread per core to avoid the overhead of task switching.
388 This allows for significant performance gains, but lacks flexibility and is not always efficient.
390 Power management helps to improve the CPU efficiency by limiting the CPU runtime frequency.
391 However, alternately it is possible to utilize the idle cycles available to take advantage of
392 the full capability of the CPU.
394 By taking advantage of cgroup, the CPU utilization quota can be simply assigned.
395 This gives another way to improve the CPU efficiency, however, there is a prerequisite;
396 DPDK must handle the context switching between multiple pthreads per core.
398 For further flexibility, it is useful to set pthread affinity not only to a CPU but to a CPU set.
400 EAL pthread and lcore Affinity
401 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
403 The term "lcore" refers to an EAL thread, which is really a Linux/FreeBSD pthread.
404 "EAL pthreads" are created and managed by EAL and execute the tasks issued by *remote_launch*.
405 In each EAL pthread, there is a TLS (Thread Local Storage) called *_lcore_id* for unique identification.
406 As EAL pthreads usually bind 1:1 to the physical CPU, the *_lcore_id* is typically equal to the CPU ID.
408 When using multiple pthreads, however, the binding is no longer always 1:1 between an EAL pthread and a specified physical CPU.
409 The EAL pthread may have affinity to a CPU set, and as such the *_lcore_id* will not be the same as the CPU ID.
410 For this reason, there is an EAL long option '--lcores' defined to assign the CPU affinity of lcores.
411 For a specified lcore ID or ID group, the option allows setting the CPU set for that EAL pthread.
414 --lcores='<lcore_set>[@cpu_set][,<lcore_set>[@cpu_set],...]'
416 'lcore_set' and 'cpu_set' can be a single number, range or a group.
418 A number is a "digit([0-9]+)"; a range is "<number>-<number>"; a group is "(<number|range>[,<number|range>,...])".
420 If a '\@cpu_set' value is not supplied, the value of 'cpu_set' will default to the value of 'lcore_set'.
424 For example, "--lcores='1,2@(5-7),(3-5)@(0,2),(0,6),7-8'" which means start 9 EAL thread;
425 lcore 0 runs on cpuset 0x41 (cpu 0,6);
426 lcore 1 runs on cpuset 0x2 (cpu 1);
427 lcore 2 runs on cpuset 0xe0 (cpu 5,6,7);
428 lcore 3,4,5 runs on cpuset 0x5 (cpu 0,2);
429 lcore 6 runs on cpuset 0x41 (cpu 0,6);
430 lcore 7 runs on cpuset 0x80 (cpu 7);
431 lcore 8 runs on cpuset 0x100 (cpu 8).
433 Using this option, for each given lcore ID, the associated CPUs can be assigned.
434 It's also compatible with the pattern of corelist('-l') option.
436 non-EAL pthread support
437 ~~~~~~~~~~~~~~~~~~~~~~~
439 It is possible to use the DPDK execution context with any user pthread (aka. Non-EAL pthreads).
440 In a non-EAL pthread, the *_lcore_id* is always LCORE_ID_ANY which identifies that it is not an EAL thread with a valid, unique, *_lcore_id*.
441 Some libraries will use an alternative unique ID (e.g. TID), some will not be impacted at all, and some will work but with limitations (e.g. timer and mempool libraries).
443 All these impacts are mentioned in :ref:`known_issue_label` section.
448 There are two public APIs ``rte_thread_set_affinity()`` and ``rte_thread_get_affinity()`` introduced for threads.
449 When they're used in any pthread context, the Thread Local Storage(TLS) will be set/get.
451 Those TLS include *_cpuset* and *_socket_id*:
453 * *_cpuset* stores the CPUs bitmap to which the pthread is affinitized.
455 * *_socket_id* stores the NUMA node of the CPU set. If the CPUs in CPU set belong to different NUMA node, the *_socket_id* will be set to SOCKET_ID_ANY.
458 .. _known_issue_label:
465 The rte_mempool uses a per-lcore cache inside the mempool.
466 For non-EAL pthreads, ``rte_lcore_id()`` will not return a valid number.
467 So for now, when rte_mempool is used with non-EAL pthreads, the put/get operations will bypass the default mempool cache and there is a performance penalty because of this bypass.
468 Only user-owned external caches can be used in a non-EAL context in conjunction with ``rte_mempool_generic_put()`` and ``rte_mempool_generic_get()`` that accept an explicit cache parameter.
472 rte_ring supports multi-producer enqueue and multi-consumer dequeue.
473 However, it is non-preemptive, this has a knock on effect of making rte_mempool non-preemptable.
477 The "non-preemptive" constraint means:
479 - a pthread doing multi-producers enqueues on a given ring must not
480 be preempted by another pthread doing a multi-producer enqueue on
482 - a pthread doing multi-consumers dequeues on a given ring must not
483 be preempted by another pthread doing a multi-consumer dequeue on
486 Bypassing this constraint may cause the 2nd pthread to spin until the 1st one is scheduled again.
487 Moreover, if the 1st pthread is preempted by a context that has an higher priority, it may even cause a dead lock.
489 This means, use cases involving preemptible pthreads should consider using rte_ring carefully.
491 1. It CAN be used for preemptible single-producer and single-consumer use case.
493 2. It CAN be used for non-preemptible multi-producer and preemptible single-consumer use case.
495 3. It CAN be used for preemptible single-producer and non-preemptible multi-consumer use case.
497 4. It MAY be used by preemptible multi-producer and/or preemptible multi-consumer pthreads whose scheduling policy are all SCHED_OTHER(cfs), SCHED_IDLE or SCHED_BATCH. User SHOULD be aware of the performance penalty before using it.
499 5. It MUST not be used by multi-producer/consumer pthreads, whose scheduling policies are SCHED_FIFO or SCHED_RR.
503 Running ``rte_timer_manage()`` on a non-EAL pthread is not allowed. However, resetting/stopping the timer from a non-EAL pthread is allowed.
507 In non-EAL pthreads, there is no per thread loglevel and logtype, global loglevels are used.
511 The debug statistics of rte_ring, rte_mempool and rte_timer are not supported in a non-EAL pthread.
516 The following is a simple example of cgroup control usage, there are two pthreads(t0 and t1) doing packet I/O on the same core ($CPU).
517 We expect only 50% of CPU spend on packet IO.
519 .. code-block:: console
521 mkdir /sys/fs/cgroup/cpu/pkt_io
522 mkdir /sys/fs/cgroup/cpuset/pkt_io
524 echo $cpu > /sys/fs/cgroup/cpuset/cpuset.cpus
526 echo $t0 > /sys/fs/cgroup/cpu/pkt_io/tasks
527 echo $t0 > /sys/fs/cgroup/cpuset/pkt_io/tasks
529 echo $t1 > /sys/fs/cgroup/cpu/pkt_io/tasks
530 echo $t1 > /sys/fs/cgroup/cpuset/pkt_io/tasks
532 cd /sys/fs/cgroup/cpu/pkt_io
533 echo 100000 > pkt_io/cpu.cfs_period_us
534 echo 50000 > pkt_io/cpu.cfs_quota_us
540 The EAL provides a malloc API to allocate any-sized memory.
542 The objective of this API is to provide malloc-like functions to allow
543 allocation from hugepage memory and to facilitate application porting.
544 The *DPDK API Reference* manual describes the available functions.
546 Typically, these kinds of allocations should not be done in data plane
547 processing because they are slower than pool-based allocation and make
548 use of locks within the allocation and free paths.
549 However, they can be used in configuration code.
551 Refer to the rte_malloc() function description in the *DPDK API Reference*
552 manual for more information.
557 When CONFIG_RTE_MALLOC_DEBUG is enabled, the allocated memory contains
558 overwrite protection fields to help identify buffer overflows.
560 Alignment and NUMA Constraints
561 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
563 The rte_malloc() takes an align argument that can be used to request a memory
564 area that is aligned on a multiple of this value (which must be a power of two).
566 On systems with NUMA support, a call to the rte_malloc() function will return
567 memory that has been allocated on the NUMA socket of the core which made the call.
568 A set of APIs is also provided, to allow memory to be explicitly allocated on a
569 NUMA socket directly, or by allocated on the NUMA socket where another core is
570 located, in the case where the memory is to be used by a logical core other than
571 on the one doing the memory allocation.
576 This API is meant to be used by an application that requires malloc-like
577 functions at initialization time.
579 For allocating/freeing data at runtime, in the fast-path of an application,
580 the memory pool library should be used instead.
582 Internal Implementation
583 ~~~~~~~~~~~~~~~~~~~~~~~
588 There are two data structure types used internally in the malloc library:
590 * struct malloc_heap - used to track free space on a per-socket basis
592 * struct malloc_elem - the basic element of allocation and free-space
593 tracking inside the library.
595 Structure: malloc_heap
596 """"""""""""""""""""""
598 The malloc_heap structure is used to manage free space on a per-socket basis.
599 Internally, there is one heap structure per NUMA node, which allows us to
600 allocate memory to a thread based on the NUMA node on which this thread runs.
601 While this does not guarantee that the memory will be used on that NUMA node,
602 it is no worse than a scheme where the memory is always allocated on a fixed
605 The key fields of the heap structure and their function are described below
606 (see also diagram above):
608 * lock - the lock field is needed to synchronize access to the heap.
609 Given that the free space in the heap is tracked using a linked list,
610 we need a lock to prevent two threads manipulating the list at the same time.
612 * free_head - this points to the first element in the list of free nodes for
615 * first - this points to the first element in the heap.
617 * last - this points to the last element in the heap.
619 .. _figure_malloc_heap:
621 .. figure:: img/malloc_heap.*
623 Example of a malloc heap and malloc elements within the malloc library
628 Structure: malloc_elem
629 """"""""""""""""""""""
631 The malloc_elem structure is used as a generic header structure for various
633 It is used in two different ways - all shown in the diagram above:
635 #. As a header on a block of free or allocated memory - normal case
637 #. As a padding header inside a block of memory
639 The most important fields in the structure and how they are used are described below.
641 Malloc heap is a doubly-linked list, where each element keeps track of its
642 previous and next elements. Due to the fact that hugepage memory can come and
643 go, neighbouring malloc elements may not necessarily be adjacent in memory.
644 Also, since a malloc element may span multiple pages, its contents may not
645 necessarily be IOVA-contiguous either - each malloc element is only guaranteed
646 to be virtually contiguous.
650 If the usage of a particular field in one of the above three usages is not
651 described, the field can be assumed to have an undefined value in that
652 situation, for example, for padding headers only the "state" and "pad"
653 fields have valid values.
655 * heap - this pointer is a reference back to the heap structure from which
656 this block was allocated.
657 It is used for normal memory blocks when they are being freed, to add the
658 newly-freed block to the heap's free-list.
660 * prev - this pointer points to previous header element/block in memory. When
661 freeing a block, this pointer is used to reference the previous block to
662 check if that block is also free. If so, and the two blocks are immediately
663 adjacent to each other, then the two free blocks are merged to form a single
666 * next - this pointer points to next header element/block in memory. When
667 freeing a block, this pointer is used to reference the next block to check
668 if that block is also free. If so, and the two blocks are immediately
669 adjacent to each other, then the two free blocks are merged to form a single
672 * free_list - this is a structure pointing to previous and next elements in
673 this heap's free list.
674 It is only used in normal memory blocks; on ``malloc()`` to find a suitable
675 free block to allocate and on ``free()`` to add the newly freed element to
678 * state - This field can have one of three values: ``FREE``, ``BUSY`` or
680 The former two are to indicate the allocation state of a normal memory block
681 and the latter is to indicate that the element structure is a dummy structure
682 at the end of the start-of-block padding, i.e. where the start of the data
683 within a block is not at the start of the block itself, due to alignment
685 In that case, the pad header is used to locate the actual malloc element
686 header for the block.
688 * pad - this holds the length of the padding present at the start of the block.
689 In the case of a normal block header, it is added to the address of the end
690 of the header to give the address of the start of the data area, i.e. the
691 value passed back to the application on a malloc.
692 Within a dummy header inside the padding, this same value is stored, and is
693 subtracted from the address of the dummy header to yield the address of the
696 * size - the size of the data block, including the header itself.
701 On EAL initialization, all preallocated memory segments are setup as part of the
702 malloc heap. This setup involves placing an :ref:`element header<malloc_elem>`
703 with ``FREE`` at the start of each virtually contiguous segment of memory.
704 The ``FREE`` element is then added to the ``free_list`` for the malloc heap.
706 This setup also happens whenever memory is allocated at runtime (if supported),
707 in which case newly allocated pages are also added to the heap, merging with any
708 adjacent free segments if there are any.
710 When an application makes a call to a malloc-like function, the malloc function
711 will first index the ``lcore_config`` structure for the calling thread, and
712 determine the NUMA node of that thread.
713 The NUMA node is used to index the array of ``malloc_heap`` structures which is
714 passed as a parameter to the ``heap_alloc()`` function, along with the
715 requested size, type, alignment and boundary parameters.
717 The ``heap_alloc()`` function will scan the free_list of the heap, and attempt
718 to find a free block suitable for storing data of the requested size, with the
719 requested alignment and boundary constraints.
721 When a suitable free element has been identified, the pointer to be returned
722 to the user is calculated.
723 The cache-line of memory immediately preceding this pointer is filled with a
724 struct malloc_elem header.
725 Because of alignment and boundary constraints, there could be free space at
726 the start and/or end of the element, resulting in the following behavior:
728 #. Check for trailing space.
729 If the trailing space is big enough, i.e. > 128 bytes, then the free element
731 If it is not, then we just ignore it (wasted space).
733 #. Check for space at the start of the element.
734 If the space at the start is small, i.e. <=128 bytes, then a pad header is
735 used, and the remaining space is wasted.
736 If, however, the remaining space is greater, then the free element is split.
738 The advantage of allocating the memory from the end of the existing element is
739 that no adjustment of the free list needs to take place - the existing element
740 on the free list just has its size value adjusted, and the next/previous elements
741 have their "prev"/"next" pointers redirected to the newly created element.
743 In case when there is not enough memory in the heap to satisfy allocation
744 request, EAL will attempt to allocate more memory from the system (if supported)
745 and, following successful allocation, will retry reserving the memory again. In
746 a multiprocessing scenario, all primary and secondary processes will synchronize
747 their memory maps to ensure that any valid pointer to DPDK memory is guaranteed
748 to be valid at all times in all currently running processes.
750 Failure to synchronize memory maps in one of the processes will cause allocation
751 to fail, even though some of the processes may have allocated the memory
752 successfully. The memory is not added to the malloc heap unless primary process
753 has ensured that all other processes have mapped this memory successfully.
755 Any successful allocation event will trigger a callback, for which user
756 applications and other DPDK subsystems can register. Additionally, validation
757 callbacks will be triggered before allocation if the newly allocated memory will
758 exceed threshold set by the user, giving a chance to allow or deny allocation.
762 Any allocation of new pages has to go through primary process. If the
763 primary process is not active, no memory will be allocated even if it was
764 theoretically possible to do so. This is because primary's process map acts
765 as an authority on what should or should not be mapped, while each secondary
766 process has its own, local memory map. Secondary processes do not update the
767 shared memory map, they only copy its contents to their local memory map.
772 To free an area of memory, the pointer to the start of the data area is passed
773 to the free function.
774 The size of the ``malloc_elem`` structure is subtracted from this pointer to get
775 the element header for the block.
776 If this header is of type ``PAD`` then the pad length is further subtracted from
777 the pointer to get the proper element header for the entire block.
779 From this element header, we get pointers to the heap from which the block was
780 allocated and to where it must be freed, as well as the pointer to the previous
781 and next elements. These next and previous elements are then checked to see if
782 they are also ``FREE`` and are immediately adjacent to the current one, and if
783 so, they are merged with the current element. This means that we can never have
784 two ``FREE`` memory blocks adjacent to one another, as they are always merged
787 If deallocating pages at runtime is supported, and the free element encloses
788 one or more pages, those pages can be deallocated and be removed from the heap.
789 If DPDK was started with command-line parameters for preallocating memory
790 (``-m`` or ``--socket-mem``), then those pages that were allocated at startup
791 will not be deallocated.
793 Any successful deallocation event will trigger a callback, for which user
794 applications and other DPDK subsystems can register.