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, 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 * Trace and Debug Functions: Logs, dump_stack, panic and so on.
27 * Utility Functions: Spinlocks and atomic counters that are not provided in libc.
29 * CPU Feature Identification: Determine at runtime if a particular feature, for example, IntelĀ® AVX is supported.
30 Determine if the current CPU supports the feature set that the binary was compiled for.
32 * Interrupt Handling: Interfaces to register/unregister callbacks to specific interrupt sources.
34 * Alarm Functions: Interfaces to set/remove callbacks to be run at a specific time.
36 EAL in a Linux-userland Execution Environment
37 ---------------------------------------------
39 In a Linux user space environment, the DPDK application runs as a user-space application using the pthread library.
41 The EAL performs physical memory allocation using mmap() in hugetlbfs (using huge page sizes to increase performance).
42 This memory is exposed to DPDK service layers such as the :ref:`Mempool Library <Mempool_Library>`.
44 At this point, the DPDK services layer will be initialized, then through pthread setaffinity calls,
45 each execution unit will be assigned to a specific logical core to run as a user-level thread.
47 The time reference is provided by the CPU Time-Stamp Counter (TSC) or by the HPET kernel API through a mmap() call.
49 Initialization and Core Launching
50 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
52 Part of the initialization is done by the start function of glibc.
53 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.
54 Then, the main() function is called. The core initialization and launch is done in rte_eal_init() (see the API documentation).
55 It consist of calls to the pthread library (more specifically, pthread_self(), pthread_create(), and pthread_setaffinity_np()).
57 .. _figure_linux_launch:
59 .. figure:: img/linuxapp_launch.*
61 EAL Initialization in a Linux Application Environment
66 Initialization of objects, such as memory zones, rings, memory pools, lpm tables and hash tables,
67 should be done as part of the overall application initialization on the master lcore.
68 The creation and initialization functions for these objects are not multi-thread safe.
69 However, once initialized, the objects themselves can safely be used in multiple threads simultaneously.
74 During the initialization of EAL resources such as hugepage backed memory can be
75 allocated by core components. The memory allocated during ``rte_eal_init()``
76 can be released by calling the ``rte_eal_cleanup()`` function. Refer to the
77 API documentation for details.
82 The Linux EAL allows a multi-process as well as a multi-threaded (pthread) deployment model.
84 :ref:`Multi-process Support <Multi-process_Support>` for more details.
86 Memory Mapping Discovery and Memory Reservation
87 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
89 The allocation of large contiguous physical memory is done using the hugetlbfs kernel filesystem.
90 The EAL provides an API to reserve named memory zones in this contiguous memory.
91 The physical address of the reserved memory for that memory zone is also returned to the user by the memory zone reservation API.
93 There are two modes in which DPDK memory subsystem can operate: dynamic mode,
94 and legacy mode. Both modes are explained below.
98 Memory reservations done using the APIs provided by rte_malloc are also backed by pages from the hugetlbfs filesystem.
100 + Dynamic memory mode
102 Currently, this mode is only supported on Linux.
104 In this mode, usage of hugepages by DPDK application will grow and shrink based
105 on application's requests. Any memory allocation through ``rte_malloc()``,
106 ``rte_memzone_reserve()`` or other methods, can potentially result in more
107 hugepages being reserved from the system. Similarly, any memory deallocation can
108 potentially result in hugepages being released back to the system.
110 Memory allocated in this mode is not guaranteed to be IOVA-contiguous. If large
111 chunks of IOVA-contiguous are required (with "large" defined as "more than one
112 page"), it is recommended to either use VFIO driver for all physical devices (so
113 that IOVA and VA addresses can be the same, thereby bypassing physical addresses
114 entirely), or use legacy memory mode.
116 For chunks of memory which must be IOVA-contiguous, it is recommended to use
117 ``rte_memzone_reserve()`` function with ``RTE_MEMZONE_IOVA_CONTIG`` flag
118 specified. This way, memory allocator will ensure that, whatever memory mode is
119 in use, either reserved memory will satisfy the requirements, or the allocation
122 There is no need to preallocate any memory at startup using ``-m`` or
123 ``--socket-mem`` command-line parameters, however it is still possible to do so,
124 in which case preallocate memory will be "pinned" (i.e. will never be released
125 by the application back to the system). It will be possible to allocate more
126 hugepages, and deallocate those, but any preallocated pages will not be freed.
127 If neither ``-m`` nor ``--socket-mem`` were specified, no memory will be
128 preallocated, and all memory will be allocated at runtime, as needed.
130 Another available option to use in dynamic memory mode is
131 ``--single-file-segments`` command-line option. This option will put pages in
132 single files (per memseg list), as opposed to creating a file per page. This is
133 normally not needed, but can be useful for use cases like userspace vhost, where
134 there is limited number of page file descriptors that can be passed to VirtIO.
136 If the application (or DPDK-internal code, such as device drivers) wishes to
137 receive notifications about newly allocated memory, it is possible to register
138 for memory event callbacks via ``rte_mem_event_callback_register()`` function.
139 This will call a callback function any time DPDK's memory map has changed.
141 If the application (or DPDK-internal code, such as device drivers) wishes to be
142 notified about memory allocations above specified threshold (and have a chance
143 to deny them), allocation validator callbacks are also available via
144 ``rte_mem_alloc_validator_callback_register()`` function.
146 A default validator callback is provided by EAL, which can be enabled with a
147 ``--socket-limit`` command-line option, for a simple way to limit maximum amount
148 of memory that can be used by DPDK application.
151 Memory subsystem uses DPDK IPC internally, so memory allocations/callbacks
152 and IPC must not be mixed: it is not safe to allocate/free memory inside
153 memory-related or IPC callbacks, and it is not safe to use IPC inside
154 memory-related callbacks. See chapter
155 :ref:`Multi-process Support <Multi-process_Support>` for more details about
160 This mode is enabled by specifying ``--legacy-mem`` command-line switch to the
161 EAL. This switch will have no effect on FreeBSD as FreeBSD only supports
164 This mode mimics historical behavior of EAL. That is, EAL will reserve all
165 memory at startup, sort all memory into large IOVA-contiguous chunks, and will
166 not allow acquiring or releasing hugepages from the system at runtime.
168 If neither ``-m`` nor ``--socket-mem`` were specified, the entire available
169 hugepage memory will be preallocated.
171 + Hugepage allocation matching
173 This behavior is enabled by specifying the ``--match-allocations`` command-line
174 switch to the EAL. This switch is Linux-only and not supported with
175 ``--legacy-mem`` nor ``--no-huge``.
177 Some applications using memory event callbacks may require that hugepages be
178 freed exactly as they were allocated. These applications may also require
179 that any allocation from the malloc heap not span across allocations
180 associated with two different memory event callbacks. Hugepage allocation
181 matching can be used by these types of applications to satisfy both of these
182 requirements. This can result in some increased memory usage which is
183 very dependent on the memory allocation patterns of the application.
187 Additional restrictions are present when running in 32-bit mode. In dynamic
188 memory mode, by default maximum of 2 gigabytes of VA space will be preallocated,
189 and all of it will be on master lcore NUMA node unless ``--socket-mem`` flag is
192 In legacy mode, VA space will only be preallocated for segments that were
193 requested (plus padding, to keep IOVA-contiguousness).
195 + Maximum amount of memory
197 All possible virtual memory space that can ever be used for hugepage mapping in
198 a DPDK process is preallocated at startup, thereby placing an upper limit on how
199 much memory a DPDK application can have. DPDK memory is stored in segment lists,
200 each segment is strictly one physical page. It is possible to change the amount
201 of virtual memory being preallocated at startup by editing the following config
204 * ``CONFIG_RTE_MAX_MEMSEG_LISTS`` controls how many segment lists can DPDK have
205 * ``CONFIG_RTE_MAX_MEM_MB_PER_LIST`` controls how much megabytes of memory each
206 segment list can address
207 * ``CONFIG_RTE_MAX_MEMSEG_PER_LIST`` controls how many segments each segment can
209 * ``CONFIG_RTE_MAX_MEMSEG_PER_TYPE`` controls how many segments each memory type
210 can have (where "type" is defined as "page size + NUMA node" combination)
211 * ``CONFIG_RTE_MAX_MEM_MB_PER_TYPE`` controls how much megabytes of memory each
212 memory type can address
213 * ``CONFIG_RTE_MAX_MEM_MB`` places a global maximum on the amount of memory
216 Normally, these options do not need to be changed.
220 Preallocated virtual memory is not to be confused with preallocated hugepage
221 memory! All DPDK processes preallocate virtual memory at startup. Hugepages
222 can later be mapped into that preallocated VA space (if dynamic memory mode
223 is enabled), and can optionally be mapped into it at startup.
225 + Segment file descriptors
227 On Linux, in most cases, EAL will store segment file descriptors in EAL. This
228 can become a problem when using smaller page sizes due to underlying limitations
229 of ``glibc`` library. For example, Linux API calls such as ``select()`` may not
230 work correctly because ``glibc`` does not support more than certain number of
233 There are two possible solutions for this problem. The recommended solution is
234 to use ``--single-file-segments`` mode, as that mode will not use a file
235 descriptor per each page, and it will keep compatibility with Virtio with
236 vhost-user backend. This option is not available when using ``--legacy-mem``
239 Another option is to use bigger page sizes. Since fewer pages are required to
240 cover the same memory area, fewer file descriptors will be stored internally
243 Support for Externally Allocated Memory
244 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
246 It is possible to use externally allocated memory in DPDK. There are two ways in
247 which using externally allocated memory can work: the malloc heap API's, and
248 manual memory management.
250 + Using heap API's for externally allocated memory
252 Using a set of malloc heap API's is the recommended way to use externally
253 allocated memory in DPDK. In this way, support for externally allocated memory
254 is implemented through overloading the socket ID - externally allocated heaps
255 will have socket ID's that would be considered invalid under normal
256 circumstances. Requesting an allocation to take place from a specified
257 externally allocated memory is a matter of supplying the correct socket ID to
258 DPDK allocator, either directly (e.g. through a call to ``rte_malloc``) or
259 indirectly (through data structure-specific allocation API's such as
260 ``rte_ring_create``). Using these API's also ensures that mapping of externally
261 allocated memory for DMA is also performed on any memory segment that is added
262 to a DPDK malloc heap.
264 Since there is no way DPDK can verify whether memory is available or valid, this
265 responsibility falls on the shoulders of the user. All multiprocess
266 synchronization is also user's responsibility, as well as ensuring that all
267 calls to add/attach/detach/remove memory are done in the correct order. It is
268 not required to attach to a memory area in all processes - only attach to memory
271 The expected workflow is as follows:
273 * Get a pointer to memory area
274 * Create a named heap
275 * Add memory area(s) to the heap
276 - If IOVA table is not specified, IOVA addresses will be assumed to be
277 unavailable, and DMA mappings will not be performed
278 - Other processes must attach to the memory area before they can use it
279 * Get socket ID used for the heap
280 * Use normal DPDK allocation procedures, using supplied socket ID
281 * If memory area is no longer needed, it can be removed from the heap
282 - Other processes must detach from this memory area before it can be removed
283 * If heap is no longer needed, remove it
284 - Socket ID will become invalid and will not be reused
286 For more information, please refer to ``rte_malloc`` API documentation,
287 specifically the ``rte_malloc_heap_*`` family of function calls.
289 + Using externally allocated memory without DPDK API's
291 While using heap API's is the recommended method of using externally allocated
292 memory in DPDK, there are certain use cases where the overhead of DPDK heap API
293 is undesirable - for example, when manual memory management is performed on an
294 externally allocated area. To support use cases where externally allocated
295 memory will not be used as part of normal DPDK workflow, there is also another
296 set of API's under the ``rte_extmem_*`` namespace.
298 These API's are (as their name implies) intended to allow registering or
299 unregistering externally allocated memory to/from DPDK's internal page table, to
300 allow API's like ``rte_virt2memseg`` etc. to work with externally allocated
301 memory. Memory added this way will not be available for any regular DPDK
302 allocators; DPDK will leave this memory for the user application to manage.
304 The expected workflow is as follows:
306 * Get a pointer to memory area
307 * Register memory within DPDK
308 - If IOVA table is not specified, IOVA addresses will be assumed to be
310 - Other processes must attach to the memory area before they can use it
311 * Perform DMA mapping with ``rte_dev_dma_map`` if needed
312 * Use the memory area in your application
313 * If memory area is no longer needed, it can be unregistered
314 - If the area was mapped for DMA, unmapping must be performed before
316 - Other processes must detach from the memory area before it can be
319 Since these externally allocated memory areas will not be managed by DPDK, it is
320 therefore up to the user application to decide how to use them and what to do
321 with them once they're registered.
323 Per-lcore and Shared Variables
324 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
328 lcore refers to a logical execution unit of the processor, sometimes called a hardware *thread*.
330 Shared variables are the default behavior.
331 Per-lcore variables are implemented using *Thread Local Storage* (TLS) to provide per-thread local storage.
336 A logging API is provided by EAL.
337 By default, in a Linux application, logs are sent to syslog and also to the console.
338 However, the log function can be overridden by the user to use a different logging mechanism.
340 Trace and Debug Functions
341 ^^^^^^^^^^^^^^^^^^^^^^^^^
343 There are some debug functions to dump the stack in glibc.
344 The rte_panic() function can voluntarily provoke a SIG_ABORT,
345 which can trigger the generation of a core file, readable by gdb.
347 CPU Feature Identification
348 ~~~~~~~~~~~~~~~~~~~~~~~~~~
350 The EAL can query the CPU at runtime (using the rte_cpu_get_features() function) to determine which CPU features are available.
352 User Space Interrupt Event
353 ~~~~~~~~~~~~~~~~~~~~~~~~~~
355 + User Space Interrupt and Alarm Handling in Host Thread
357 The EAL creates a host thread to poll the UIO device file descriptors to detect the interrupts.
358 Callbacks can be registered or unregistered by the EAL functions for a specific interrupt event
359 and are called in the host thread asynchronously.
360 The EAL also allows timed callbacks to be used in the same way as for NIC interrupts.
364 In DPDK PMD, the only interrupts handled by the dedicated host thread are those for link status change
365 (link up and link down notification) and for sudden device removal.
370 The receive and transmit routines provided by each PMD don't limit themselves to execute in polling thread mode.
371 To ease the idle polling with tiny throughput, it's useful to pause the polling and wait until the wake-up event happens.
372 The RX interrupt is the first choice to be such kind of wake-up event, but probably won't be the only one.
374 EAL provides the event APIs for this event-driven thread mode.
375 Taking Linux as an example, the implementation relies on epoll. Each thread can monitor an epoll instance
376 in which all the wake-up events' file descriptors are added. The event file descriptors are created and mapped to
377 the interrupt vectors according to the UIO/VFIO spec.
378 From FreeBSD's perspective, kqueue is the alternative way, but not implemented yet.
380 EAL initializes the mapping between event file descriptors and interrupt vectors, while each device initializes the mapping
381 between interrupt vectors and queues. In this way, EAL actually is unaware of the interrupt cause on the specific vector.
382 The eth_dev driver takes responsibility to program the latter mapping.
386 Per queue RX interrupt event is only allowed in VFIO which supports multiple MSI-X vector. In UIO, the RX interrupt
387 together with other interrupt causes shares the same vector. In this case, when RX interrupt and LSC(link status change)
388 interrupt are both enabled(intr_conf.lsc == 1 && intr_conf.rxq == 1), only the former is capable.
390 The RX interrupt are controlled/enabled/disabled by ethdev APIs - 'rte_eth_dev_rx_intr_*'. They return failure if the PMD
391 hasn't support them yet. The intr_conf.rxq flag is used to turn on the capability of RX interrupt per device.
393 + Device Removal Event
395 This event is triggered by a device being removed at a bus level. Its
396 underlying resources may have been made unavailable (i.e. PCI mappings
397 unmapped). The PMD must make sure that on such occurrence, the application can
398 still safely use its callbacks.
400 This event can be subscribed to in the same way one would subscribe to a link
401 status change event. The execution context is thus the same, i.e. it is the
402 dedicated interrupt host thread.
404 Considering this, it is likely that an application would want to close a
405 device having emitted a Device Removal Event. In such case, calling
406 ``rte_eth_dev_close()`` can trigger it to unregister its own Device Removal Event
407 callback. Care must be taken not to close the device from the interrupt handler
408 context. It is necessary to reschedule such closing operation.
413 The EAL PCI device blacklist functionality can be used to mark certain NIC ports as blacklisted,
414 so they are ignored by the DPDK.
415 The ports to be blacklisted are identified using the PCIe* description (Domain:Bus:Device.Function).
420 Locks and atomic operations are per-architecture (i686 and x86_64).
425 IOVA Mode is selected by considering what the current usable Devices on the
426 system require and/or support.
428 On FreeBSD, RTE_IOVA_PA is always the default. On Linux, the IOVA mode is
429 detected based on a 2-step heuristic detailed below.
431 For the first step, EAL asks each bus its requirement in terms of IOVA mode
432 and decides on a preferred IOVA mode.
434 - if all buses report RTE_IOVA_PA, then the preferred IOVA mode is RTE_IOVA_PA,
435 - if all buses report RTE_IOVA_VA, then the preferred IOVA mode is RTE_IOVA_VA,
436 - if all buses report RTE_IOVA_DC, no bus expressed a preferrence, then the
437 preferred mode is RTE_IOVA_DC,
438 - if the buses disagree (at least one wants RTE_IOVA_PA and at least one wants
439 RTE_IOVA_VA), then the preferred IOVA mode is RTE_IOVA_DC (see below with the
440 check on Physical Addresses availability),
442 If the buses have expressed no preference on which IOVA mode to pick, then a
443 default is selected using the following logic:
445 - if physical addresses are not available, RTE_IOVA_VA mode is used
446 - if /sys/kernel/iommu_groups is not empty, RTE_IOVA_VA mode is used
447 - otherwise, RTE_IOVA_PA mode is used
449 In the case when the buses had disagreed on their preferred IOVA mode, part of
450 the buses won't work because of this decision.
452 The second step checks if the preferred mode complies with the Physical
453 Addresses availability since those are only available to root user in recent
454 kernels. Namely, if the preferred mode is RTE_IOVA_PA but there is no access to
455 Physical Addresses, then EAL init fails early, since later probing of the
456 devices would fail anyway.
460 The RTE_IOVA_VA mode is preferred as the default in most cases for the
463 - All drivers are expected to work in RTE_IOVA_VA mode, irrespective of
464 physical address availability.
465 - By default, the mempool, first asks for IOVA-contiguous memory using
466 ``RTE_MEMZONE_IOVA_CONTIG``. This is slow in RTE_IOVA_PA mode and it may
467 affect the application boot time.
468 - It is easy to enable large amount of IOVA-contiguous memory use-cases
469 with IOVA in VA mode.
471 It is expected that all PCI drivers work in both RTE_IOVA_PA and
474 If a PCI driver does not support RTE_IOVA_PA mode, the
475 ``RTE_PCI_DRV_NEED_IOVA_AS_VA`` flag is used to dictate that this PCI
476 driver can only work in RTE_IOVA_VA mode.
478 IOVA Mode Configuration
479 ~~~~~~~~~~~~~~~~~~~~~~~
481 Auto detection of the IOVA mode, based on probing the bus and IOMMU configuration, may not report
482 the desired addressing mode when virtual devices that are not directly attached to the bus are present.
483 To facilitate forcing the IOVA mode to a specific value the EAL command line option ``--iova-mode`` can
484 be used to select either physical addressing('pa') or virtual addressing('va').
486 Memory Segments and Memory Zones (memzone)
487 ------------------------------------------
489 The mapping of physical memory is provided by this feature in the EAL.
490 As physical memory can have gaps, the memory is described in a table of descriptors,
491 and each descriptor (called rte_memseg ) describes a physical page.
493 On top of this, the memzone allocator's role is to reserve contiguous portions of physical memory.
494 These zones are identified by a unique name when the memory is reserved.
496 The rte_memzone descriptors are also located in the configuration structure.
497 This structure is accessed using rte_eal_get_configuration().
498 The lookup (by name) of a memory zone returns a descriptor containing the physical address of the memory zone.
500 Memory zones can be reserved with specific start address alignment by supplying the align parameter
501 (by default, they are aligned to cache line size).
502 The alignment value should be a power of two and not less than the cache line size (64 bytes).
503 Memory zones can also be reserved from either 2 MB or 1 GB hugepages, provided that both are available on the system.
505 Both memsegs and memzones are stored using ``rte_fbarray`` structures. Please
506 refer to *DPDK API Reference* for more information.
512 DPDK usually pins one pthread per core to avoid the overhead of task switching.
513 This allows for significant performance gains, but lacks flexibility and is not always efficient.
515 Power management helps to improve the CPU efficiency by limiting the CPU runtime frequency.
516 However, alternately it is possible to utilize the idle cycles available to take advantage of
517 the full capability of the CPU.
519 By taking advantage of cgroup, the CPU utilization quota can be simply assigned.
520 This gives another way to improve the CPU efficiency, however, there is a prerequisite;
521 DPDK must handle the context switching between multiple pthreads per core.
523 For further flexibility, it is useful to set pthread affinity not only to a CPU but to a CPU set.
525 EAL pthread and lcore Affinity
526 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
528 The term "lcore" refers to an EAL thread, which is really a Linux/FreeBSD pthread.
529 "EAL pthreads" are created and managed by EAL and execute the tasks issued by *remote_launch*.
530 In each EAL pthread, there is a TLS (Thread Local Storage) called *_lcore_id* for unique identification.
531 As EAL pthreads usually bind 1:1 to the physical CPU, the *_lcore_id* is typically equal to the CPU ID.
533 When using multiple pthreads, however, the binding is no longer always 1:1 between an EAL pthread and a specified physical CPU.
534 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.
535 For this reason, there is an EAL long option '--lcores' defined to assign the CPU affinity of lcores.
536 For a specified lcore ID or ID group, the option allows setting the CPU set for that EAL pthread.
539 --lcores='<lcore_set>[@cpu_set][,<lcore_set>[@cpu_set],...]'
541 'lcore_set' and 'cpu_set' can be a single number, range or a group.
543 A number is a "digit([0-9]+)"; a range is "<number>-<number>"; a group is "(<number|range>[,<number|range>,...])".
545 If a '\@cpu_set' value is not supplied, the value of 'cpu_set' will default to the value of 'lcore_set'.
549 For example, "--lcores='1,2@(5-7),(3-5)@(0,2),(0,6),7-8'" which means start 9 EAL thread;
550 lcore 0 runs on cpuset 0x41 (cpu 0,6);
551 lcore 1 runs on cpuset 0x2 (cpu 1);
552 lcore 2 runs on cpuset 0xe0 (cpu 5,6,7);
553 lcore 3,4,5 runs on cpuset 0x5 (cpu 0,2);
554 lcore 6 runs on cpuset 0x41 (cpu 0,6);
555 lcore 7 runs on cpuset 0x80 (cpu 7);
556 lcore 8 runs on cpuset 0x100 (cpu 8).
558 Using this option, for each given lcore ID, the associated CPUs can be assigned.
559 It's also compatible with the pattern of corelist('-l') option.
561 non-EAL pthread support
562 ~~~~~~~~~~~~~~~~~~~~~~~
564 It is possible to use the DPDK execution context with any user pthread (aka. Non-EAL pthreads).
565 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*.
566 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).
568 All these impacts are mentioned in :ref:`known_issue_label` section.
573 There are two public APIs ``rte_thread_set_affinity()`` and ``rte_thread_get_affinity()`` introduced for threads.
574 When they're used in any pthread context, the Thread Local Storage(TLS) will be set/get.
576 Those TLS include *_cpuset* and *_socket_id*:
578 * *_cpuset* stores the CPUs bitmap to which the pthread is affinitized.
580 * *_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.
586 It is possible to create Control Threads using the public API
587 ``rte_ctrl_thread_create()``.
588 Those threads can be used for management/infrastructure tasks and are used
589 internally by DPDK for multi process support and interrupt handling.
591 Those threads will be scheduled on CPUs part of the original process CPU
592 affinity from which the dataplane and service lcores are excluded.
594 For example, on a 8 CPUs system, starting a dpdk application with -l 2,3
595 (dataplane cores), then depending on the affinity configuration which can be
596 controlled with tools like taskset (Linux) or cpuset (FreeBSD),
598 - with no affinity configuration, the Control Threads will end up on
600 - with affinity restricted to 2-4, the Control Threads will end up on
602 - with affinity restricted to 2-3, the Control Threads will end up on
603 CPU 2 (master lcore, which is the default when no CPU is available).
605 .. _known_issue_label:
612 The rte_mempool uses a per-lcore cache inside the mempool.
613 For non-EAL pthreads, ``rte_lcore_id()`` will not return a valid number.
614 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.
615 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.
619 rte_ring supports multi-producer enqueue and multi-consumer dequeue.
620 However, it is non-preemptive, this has a knock on effect of making rte_mempool non-preemptable.
624 The "non-preemptive" constraint means:
626 - a pthread doing multi-producers enqueues on a given ring must not
627 be preempted by another pthread doing a multi-producer enqueue on
629 - a pthread doing multi-consumers dequeues on a given ring must not
630 be preempted by another pthread doing a multi-consumer dequeue on
633 Bypassing this constraint may cause the 2nd pthread to spin until the 1st one is scheduled again.
634 Moreover, if the 1st pthread is preempted by a context that has an higher priority, it may even cause a dead lock.
636 This means, use cases involving preemptible pthreads should consider using rte_ring carefully.
638 1. It CAN be used for preemptible single-producer and single-consumer use case.
640 2. It CAN be used for non-preemptible multi-producer and preemptible single-consumer use case.
642 3. It CAN be used for preemptible single-producer and non-preemptible multi-consumer use case.
644 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.
646 5. It MUST not be used by multi-producer/consumer pthreads, whose scheduling policies are SCHED_FIFO or SCHED_RR.
648 Alternatively, applications can use the lock-free stack mempool handler. When
649 considering this handler, note that:
651 - It is currently limited to the aarch64 and x86_64 platforms, because it uses
652 an instruction (16-byte compare-and-swap) that is not yet available on other
654 - It has worse average-case performance than the non-preemptive rte_ring, but
655 software caching (e.g. the mempool cache) can mitigate this by reducing the
656 number of stack accesses.
660 Running ``rte_timer_manage()`` on a non-EAL pthread is not allowed. However, resetting/stopping the timer from a non-EAL pthread is allowed.
664 In non-EAL pthreads, there is no per thread loglevel and logtype, global loglevels are used.
668 The debug statistics of rte_ring, rte_mempool and rte_timer are not supported in a non-EAL pthread.
673 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).
674 We expect only 50% of CPU spend on packet IO.
676 .. code-block:: console
678 mkdir /sys/fs/cgroup/cpu/pkt_io
679 mkdir /sys/fs/cgroup/cpuset/pkt_io
681 echo $cpu > /sys/fs/cgroup/cpuset/cpuset.cpus
683 echo $t0 > /sys/fs/cgroup/cpu/pkt_io/tasks
684 echo $t0 > /sys/fs/cgroup/cpuset/pkt_io/tasks
686 echo $t1 > /sys/fs/cgroup/cpu/pkt_io/tasks
687 echo $t1 > /sys/fs/cgroup/cpuset/pkt_io/tasks
689 cd /sys/fs/cgroup/cpu/pkt_io
690 echo 100000 > pkt_io/cpu.cfs_period_us
691 echo 50000 > pkt_io/cpu.cfs_quota_us
697 The EAL provides a malloc API to allocate any-sized memory.
699 The objective of this API is to provide malloc-like functions to allow
700 allocation from hugepage memory and to facilitate application porting.
701 The *DPDK API Reference* manual describes the available functions.
703 Typically, these kinds of allocations should not be done in data plane
704 processing because they are slower than pool-based allocation and make
705 use of locks within the allocation and free paths.
706 However, they can be used in configuration code.
708 Refer to the rte_malloc() function description in the *DPDK API Reference*
709 manual for more information.
714 When CONFIG_RTE_MALLOC_DEBUG is enabled, the allocated memory contains
715 overwrite protection fields to help identify buffer overflows.
717 Alignment and NUMA Constraints
718 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
720 The rte_malloc() takes an align argument that can be used to request a memory
721 area that is aligned on a multiple of this value (which must be a power of two).
723 On systems with NUMA support, a call to the rte_malloc() function will return
724 memory that has been allocated on the NUMA socket of the core which made the call.
725 A set of APIs is also provided, to allow memory to be explicitly allocated on a
726 NUMA socket directly, or by allocated on the NUMA socket where another core is
727 located, in the case where the memory is to be used by a logical core other than
728 on the one doing the memory allocation.
733 This API is meant to be used by an application that requires malloc-like
734 functions at initialization time.
736 For allocating/freeing data at runtime, in the fast-path of an application,
737 the memory pool library should be used instead.
739 Internal Implementation
740 ~~~~~~~~~~~~~~~~~~~~~~~
745 There are two data structure types used internally in the malloc library:
747 * struct malloc_heap - used to track free space on a per-socket basis
749 * struct malloc_elem - the basic element of allocation and free-space
750 tracking inside the library.
752 Structure: malloc_heap
753 """"""""""""""""""""""
755 The malloc_heap structure is used to manage free space on a per-socket basis.
756 Internally, there is one heap structure per NUMA node, which allows us to
757 allocate memory to a thread based on the NUMA node on which this thread runs.
758 While this does not guarantee that the memory will be used on that NUMA node,
759 it is no worse than a scheme where the memory is always allocated on a fixed
762 The key fields of the heap structure and their function are described below
763 (see also diagram above):
765 * lock - the lock field is needed to synchronize access to the heap.
766 Given that the free space in the heap is tracked using a linked list,
767 we need a lock to prevent two threads manipulating the list at the same time.
769 * free_head - this points to the first element in the list of free nodes for
772 * first - this points to the first element in the heap.
774 * last - this points to the last element in the heap.
776 .. _figure_malloc_heap:
778 .. figure:: img/malloc_heap.*
780 Example of a malloc heap and malloc elements within the malloc library
785 Structure: malloc_elem
786 """"""""""""""""""""""
788 The malloc_elem structure is used as a generic header structure for various
790 It is used in two different ways - all shown in the diagram above:
792 #. As a header on a block of free or allocated memory - normal case
794 #. As a padding header inside a block of memory
796 The most important fields in the structure and how they are used are described below.
798 Malloc heap is a doubly-linked list, where each element keeps track of its
799 previous and next elements. Due to the fact that hugepage memory can come and
800 go, neighboring malloc elements may not necessarily be adjacent in memory.
801 Also, since a malloc element may span multiple pages, its contents may not
802 necessarily be IOVA-contiguous either - each malloc element is only guaranteed
803 to be virtually contiguous.
807 If the usage of a particular field in one of the above three usages is not
808 described, the field can be assumed to have an undefined value in that
809 situation, for example, for padding headers only the "state" and "pad"
810 fields have valid values.
812 * heap - this pointer is a reference back to the heap structure from which
813 this block was allocated.
814 It is used for normal memory blocks when they are being freed, to add the
815 newly-freed block to the heap's free-list.
817 * prev - this pointer points to previous header element/block in memory. When
818 freeing a block, this pointer is used to reference the previous block to
819 check if that block is also free. If so, and the two blocks are immediately
820 adjacent to each other, then the two free blocks are merged to form a single
823 * next - this pointer points to next header element/block in memory. When
824 freeing a block, this pointer is used to reference the next block to check
825 if that block is also free. If so, and the two blocks are immediately
826 adjacent to each other, then the two free blocks are merged to form a single
829 * free_list - this is a structure pointing to previous and next elements in
830 this heap's free list.
831 It is only used in normal memory blocks; on ``malloc()`` to find a suitable
832 free block to allocate and on ``free()`` to add the newly freed element to
835 * state - This field can have one of three values: ``FREE``, ``BUSY`` or
837 The former two are to indicate the allocation state of a normal memory block
838 and the latter is to indicate that the element structure is a dummy structure
839 at the end of the start-of-block padding, i.e. where the start of the data
840 within a block is not at the start of the block itself, due to alignment
842 In that case, the pad header is used to locate the actual malloc element
843 header for the block.
845 * pad - this holds the length of the padding present at the start of the block.
846 In the case of a normal block header, it is added to the address of the end
847 of the header to give the address of the start of the data area, i.e. the
848 value passed back to the application on a malloc.
849 Within a dummy header inside the padding, this same value is stored, and is
850 subtracted from the address of the dummy header to yield the address of the
853 * size - the size of the data block, including the header itself.
858 On EAL initialization, all preallocated memory segments are setup as part of the
859 malloc heap. This setup involves placing an :ref:`element header<malloc_elem>`
860 with ``FREE`` at the start of each virtually contiguous segment of memory.
861 The ``FREE`` element is then added to the ``free_list`` for the malloc heap.
863 This setup also happens whenever memory is allocated at runtime (if supported),
864 in which case newly allocated pages are also added to the heap, merging with any
865 adjacent free segments if there are any.
867 When an application makes a call to a malloc-like function, the malloc function
868 will first index the ``lcore_config`` structure for the calling thread, and
869 determine the NUMA node of that thread.
870 The NUMA node is used to index the array of ``malloc_heap`` structures which is
871 passed as a parameter to the ``heap_alloc()`` function, along with the
872 requested size, type, alignment and boundary parameters.
874 The ``heap_alloc()`` function will scan the free_list of the heap, and attempt
875 to find a free block suitable for storing data of the requested size, with the
876 requested alignment and boundary constraints.
878 When a suitable free element has been identified, the pointer to be returned
879 to the user is calculated.
880 The cache-line of memory immediately preceding this pointer is filled with a
881 struct malloc_elem header.
882 Because of alignment and boundary constraints, there could be free space at
883 the start and/or end of the element, resulting in the following behavior:
885 #. Check for trailing space.
886 If the trailing space is big enough, i.e. > 128 bytes, then the free element
888 If it is not, then we just ignore it (wasted space).
890 #. Check for space at the start of the element.
891 If the space at the start is small, i.e. <=128 bytes, then a pad header is
892 used, and the remaining space is wasted.
893 If, however, the remaining space is greater, then the free element is split.
895 The advantage of allocating the memory from the end of the existing element is
896 that no adjustment of the free list needs to take place - the existing element
897 on the free list just has its size value adjusted, and the next/previous elements
898 have their "prev"/"next" pointers redirected to the newly created element.
900 In case when there is not enough memory in the heap to satisfy allocation
901 request, EAL will attempt to allocate more memory from the system (if supported)
902 and, following successful allocation, will retry reserving the memory again. In
903 a multiprocessing scenario, all primary and secondary processes will synchronize
904 their memory maps to ensure that any valid pointer to DPDK memory is guaranteed
905 to be valid at all times in all currently running processes.
907 Failure to synchronize memory maps in one of the processes will cause allocation
908 to fail, even though some of the processes may have allocated the memory
909 successfully. The memory is not added to the malloc heap unless primary process
910 has ensured that all other processes have mapped this memory successfully.
912 Any successful allocation event will trigger a callback, for which user
913 applications and other DPDK subsystems can register. Additionally, validation
914 callbacks will be triggered before allocation if the newly allocated memory will
915 exceed threshold set by the user, giving a chance to allow or deny allocation.
919 Any allocation of new pages has to go through primary process. If the
920 primary process is not active, no memory will be allocated even if it was
921 theoretically possible to do so. This is because primary's process map acts
922 as an authority on what should or should not be mapped, while each secondary
923 process has its own, local memory map. Secondary processes do not update the
924 shared memory map, they only copy its contents to their local memory map.
929 To free an area of memory, the pointer to the start of the data area is passed
930 to the free function.
931 The size of the ``malloc_elem`` structure is subtracted from this pointer to get
932 the element header for the block.
933 If this header is of type ``PAD`` then the pad length is further subtracted from
934 the pointer to get the proper element header for the entire block.
936 From this element header, we get pointers to the heap from which the block was
937 allocated and to where it must be freed, as well as the pointer to the previous
938 and next elements. These next and previous elements are then checked to see if
939 they are also ``FREE`` and are immediately adjacent to the current one, and if
940 so, they are merged with the current element. This means that we can never have
941 two ``FREE`` memory blocks adjacent to one another, as they are always merged
944 If deallocating pages at runtime is supported, and the free element encloses
945 one or more pages, those pages can be deallocated and be removed from the heap.
946 If DPDK was started with command-line parameters for preallocating memory
947 (``-m`` or ``--socket-mem``), then those pages that were allocated at startup
948 will not be deallocated.
950 Any successful deallocation event will trigger a callback, for which user
951 applications and other DPDK subsystems can register.