xref: /dpdk/doc/guides/prog_guide/env_abstraction_layer.rst (revision 5ecb687a5698d2d8ec1f3b3b5a7a16bceca3e29c)
1..  SPDX-License-Identifier: BSD-3-Clause
2    Copyright(c) 2010-2014 Intel Corporation.
3
4.. _Environment_Abstraction_Layer:
5
6Environment Abstraction Layer
7=============================
8
9The Environment Abstraction Layer (EAL) is responsible for gaining access to low-level resources such as hardware and memory space.
10It provides a generic interface that hides the environment specifics from the applications and libraries.
11It 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).
13
14Typical services expected from the EAL are:
15
16*   DPDK Loading and Launching:
17    The DPDK and its application are linked as a single application and must be loaded by some means.
18
19*   Core Affinity/Assignment Procedures:
20    The EAL provides mechanisms for assigning execution units to specific cores as well as creating execution instances.
21
22*   System Memory Reservation:
23    The EAL facilitates the reservation of different memory zones, for example, physical memory areas for device interactions.
24
25*   Trace and Debug Functions: Logs, dump_stack, panic and so on.
26
27*   Utility Functions: Spinlocks and atomic counters that are not provided in libc.
28
29*   CPU Feature Identification: Determine at runtime if a particular feature, for example, Intel® AVX is supported.
30    Determine if the current CPU supports the feature set that the binary was compiled for.
31
32*   Interrupt Handling: Interfaces to register/unregister callbacks to specific interrupt sources.
33
34*   Alarm Functions: Interfaces to set/remove callbacks to be run at a specific time.
35
36EAL in a Linux-userland Execution Environment
37---------------------------------------------
38
39In a Linux user space environment, the DPDK application runs as a user-space application using the pthread library.
40
41The EAL performs physical memory allocation using mmap() in hugetlbfs (using huge page sizes to increase performance).
42This memory is exposed to DPDK service layers such as the :ref:`Mempool Library <Mempool_Library>`.
43
44At this point, the DPDK services layer will be initialized, then through pthread setaffinity calls,
45each execution unit will be assigned to a specific logical core to run as a user-level thread.
46
47The time reference is provided by the CPU Time-Stamp Counter (TSC) or by the HPET kernel API through a mmap() call.
48
49Initialization and Core Launching
50~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
51
52Part of the initialization is done by the start function of glibc.
53A check is also performed at initialization time to ensure that the micro architecture type chosen in the config file is supported by the CPU.
54Then, the main() function is called. The core initialization and launch is done in rte_eal_init() (see the API documentation).
55It consist of calls to the pthread library (more specifically, pthread_self(), pthread_create(), and pthread_setaffinity_np()).
56
57.. _figure_linux_launch:
58
59.. figure:: img/linuxapp_launch.*
60
61   EAL Initialization in a Linux Application Environment
62
63
64.. note::
65
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.
70
71Shutdown and Cleanup
72~~~~~~~~~~~~~~~~~~~~
73
74During the initialization of EAL resources such as hugepage backed memory can be
75allocated by core components.  The memory allocated during ``rte_eal_init()``
76can be released by calling the ``rte_eal_cleanup()`` function. Refer to the
77API documentation for details.
78
79Multi-process Support
80~~~~~~~~~~~~~~~~~~~~~
81
82The Linux EAL allows a multi-process as well as a multi-threaded (pthread) deployment model.
83See chapter
84:ref:`Multi-process Support <Multi-process_Support>` for more details.
85
86Memory Mapping Discovery and Memory Reservation
87~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
88
89The allocation of large contiguous physical memory is done using the hugetlbfs kernel filesystem.
90The EAL provides an API to reserve named memory zones in this contiguous memory.
91The physical address of the reserved memory for that memory zone is also returned to the user by the memory zone reservation API.
92
93There are two modes in which DPDK memory subsystem can operate: dynamic mode,
94and legacy mode. Both modes are explained below.
95
96.. note::
97
98    Memory reservations done using the APIs provided by rte_malloc are also backed by pages from the hugetlbfs filesystem.
99
100+ Dynamic memory mode
101
102Currently, this mode is only supported on Linux.
103
104In this mode, usage of hugepages by DPDK application will grow and shrink based
105on application's requests. Any memory allocation through ``rte_malloc()``,
106``rte_memzone_reserve()`` or other methods, can potentially result in more
107hugepages being reserved from the system. Similarly, any memory deallocation can
108potentially result in hugepages being released back to the system.
109
110Memory allocated in this mode is not guaranteed to be IOVA-contiguous. If large
111chunks of IOVA-contiguous are required (with "large" defined as "more than one
112page"), it is recommended to either use VFIO driver for all physical devices (so
113that IOVA and VA addresses can be the same, thereby bypassing physical addresses
114entirely), or use legacy memory mode.
115
116For chunks of memory which must be IOVA-contiguous, it is recommended to use
117``rte_memzone_reserve()`` function with ``RTE_MEMZONE_IOVA_CONTIG`` flag
118specified. This way, memory allocator will ensure that, whatever memory mode is
119in use, either reserved memory will satisfy the requirements, or the allocation
120will fail.
121
122There 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,
124in which case preallocate memory will be "pinned" (i.e. will never be released
125by the application back to the system). It will be possible to allocate more
126hugepages, and deallocate those, but any preallocated pages will not be freed.
127If neither ``-m`` nor ``--socket-mem`` were specified, no memory will be
128preallocated, and all memory will be allocated at runtime, as needed.
129
130Another available option to use in dynamic memory mode is
131``--single-file-segments`` command-line option. This option will put pages in
132single files (per memseg list), as opposed to creating a file per page. This is
133normally not needed, but can be useful for use cases like userspace vhost, where
134there is limited number of page file descriptors that can be passed to VirtIO.
135
136If the application (or DPDK-internal code, such as device drivers) wishes to
137receive notifications about newly allocated memory, it is possible to register
138for memory event callbacks via ``rte_mem_event_callback_register()`` function.
139This will call a callback function any time DPDK's memory map has changed.
140
141If the application (or DPDK-internal code, such as device drivers) wishes to be
142notified about memory allocations above specified threshold (and have a chance
143to deny them), allocation validator callbacks are also available via
144``rte_mem_alloc_validator_callback_register()`` function.
145
146A 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
148of memory that can be used by DPDK application.
149
150+ Legacy memory mode
151
152This mode is enabled by specifying ``--legacy-mem`` command-line switch to the
153EAL. This switch will have no effect on FreeBSD as FreeBSD only supports
154legacy mode anyway.
155
156This mode mimics historical behavior of EAL. That is, EAL will reserve all
157memory at startup, sort all memory into large IOVA-contiguous chunks, and will
158not allow acquiring or releasing hugepages from the system at runtime.
159
160If neither ``-m`` nor ``--socket-mem`` were specified, the entire available
161hugepage memory will be preallocated.
162
163+ Hugepage allocation matching
164
165This behavior is enabled by specifying the ``--match-allocations`` command-line
166switch to the EAL. This switch is Linux-only and not supported with
167``--legacy-mem`` nor ``--no-huge``.
168
169Some applications using memory event callbacks may require that hugepages be
170freed exactly as they were allocated. These applications may also require
171that any allocation from the malloc heap not span across allocations
172associated with two different memory event callbacks. Hugepage allocation
173matching can be used by these types of applications to satisfy both of these
174requirements. This can result in some increased memory usage which is
175very dependent on the memory allocation patterns of the application.
176
177+ 32-bit support
178
179Additional restrictions are present when running in 32-bit mode. In dynamic
180memory mode, by default maximum of 2 gigabytes of VA space will be preallocated,
181and all of it will be on master lcore NUMA node unless ``--socket-mem`` flag is
182used.
183
184In legacy mode, VA space will only be preallocated for segments that were
185requested (plus padding, to keep IOVA-contiguousness).
186
187+ Maximum amount of memory
188
189All possible virtual memory space that can ever be used for hugepage mapping in
190a DPDK process is preallocated at startup, thereby placing an upper limit on how
191much memory a DPDK application can have. DPDK memory is stored in segment lists,
192each segment is strictly one physical page. It is possible to change the amount
193of virtual memory being preallocated at startup by editing the following config
194variables:
195
196* ``CONFIG_RTE_MAX_MEMSEG_LISTS`` controls how many segment lists can DPDK have
197* ``CONFIG_RTE_MAX_MEM_MB_PER_LIST`` controls how much megabytes of memory each
198  segment list can address
199* ``CONFIG_RTE_MAX_MEMSEG_PER_LIST`` controls how many segments each segment can
200  have
201* ``CONFIG_RTE_MAX_MEMSEG_PER_TYPE`` controls how many segments each memory type
202  can have (where "type" is defined as "page size + NUMA node" combination)
203* ``CONFIG_RTE_MAX_MEM_MB_PER_TYPE`` controls how much megabytes of memory each
204  memory type can address
205* ``CONFIG_RTE_MAX_MEM_MB`` places a global maximum on the amount of memory
206  DPDK can reserve
207
208Normally, these options do not need to be changed.
209
210.. note::
211
212    Preallocated virtual memory is not to be confused with preallocated hugepage
213    memory! All DPDK processes preallocate virtual memory at startup. Hugepages
214    can later be mapped into that preallocated VA space (if dynamic memory mode
215    is enabled), and can optionally be mapped into it at startup.
216
217+ Segment file descriptors
218
219On Linux, in most cases, EAL will store segment file descriptors in EAL. This
220can become a problem when using smaller page sizes due to underlying limitations
221of ``glibc`` library. For example, Linux API calls such as ``select()`` may not
222work correctly because ``glibc`` does not support more than certain number of
223file descriptors.
224
225There are two possible solutions for this problem. The recommended solution is
226to use ``--single-file-segments`` mode, as that mode will not use a file
227descriptor per each page, and it will keep compatibility with Virtio with
228vhost-user backend. This option is not available when using ``--legacy-mem``
229mode.
230
231Another option is to use bigger page sizes. Since fewer pages are required to
232cover the same memory area, fewer file descriptors will be stored internally
233by EAL.
234
235Support for Externally Allocated Memory
236~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
237
238It is possible to use externally allocated memory in DPDK. There are two ways in
239which using externally allocated memory can work: the malloc heap API's, and
240manual memory management.
241
242+ Using heap API's for externally allocated memory
243
244Using using a set of malloc heap API's is the recommended way to use externally
245allocated memory in DPDK. In this way, support for externally allocated memory
246is implemented through overloading the socket ID - externally allocated heaps
247will have socket ID's that would be considered invalid under normal
248circumstances. Requesting an allocation to take place from a specified
249externally allocated memory is a matter of supplying the correct socket ID to
250DPDK allocator, either directly (e.g. through a call to ``rte_malloc``) or
251indirectly (through data structure-specific allocation API's such as
252``rte_ring_create``). Using these API's also ensures that mapping of externally
253allocated memory for DMA is also performed on any memory segment that is added
254to a DPDK malloc heap.
255
256Since there is no way DPDK can verify whether memory is available or valid, this
257responsibility falls on the shoulders of the user. All multiprocess
258synchronization is also user's responsibility, as well as ensuring  that all
259calls to add/attach/detach/remove memory are done in the correct order. It is
260not required to attach to a memory area in all processes - only attach to memory
261areas as needed.
262
263The expected workflow is as follows:
264
265* Get a pointer to memory area
266* Create a named heap
267* Add memory area(s) to the heap
268    - If IOVA table is not specified, IOVA addresses will be assumed to be
269      unavailable, and DMA mappings will not be performed
270    - Other processes must attach to the memory area before they can use it
271* Get socket ID used for the heap
272* Use normal DPDK allocation procedures, using supplied socket ID
273* If memory area is no longer needed, it can be removed from the heap
274    - Other processes must detach from this memory area before it can be removed
275* If heap is no longer needed, remove it
276    - Socket ID will become invalid and will not be reused
277
278For more information, please refer to ``rte_malloc`` API documentation,
279specifically the ``rte_malloc_heap_*`` family of function calls.
280
281+ Using externally allocated memory without DPDK API's
282
283While using heap API's is the recommended method of using externally allocated
284memory in DPDK, there are certain use cases where the overhead of DPDK heap API
285is undesirable - for example, when manual memory management is performed on an
286externally allocated area. To support use cases where externally allocated
287memory will not be used as part of normal DPDK workflow, there is also another
288set of API's under the ``rte_extmem_*`` namespace.
289
290These API's are (as their name implies) intended to allow registering or
291unregistering externally allocated memory to/from DPDK's internal page table, to
292allow API's like ``rte_virt2memseg`` etc. to work with externally allocated
293memory. Memory added this way will not be available for any regular DPDK
294allocators; DPDK will leave this memory for the user application to manage.
295
296The expected workflow is as follows:
297
298* Get a pointer to memory area
299* Register memory within DPDK
300    - If IOVA table is not specified, IOVA addresses will be assumed to be
301      unavailable
302    - Other processes must attach to the memory area before they can use it
303* Perform DMA mapping with ``rte_dev_dma_map`` if needed
304* Use the memory area in your application
305* If memory area is no longer needed, it can be unregistered
306    - If the area was mapped for DMA, unmapping must be performed before
307      unregistering memory
308    - Other processes must detach from the memory area before it can be
309      unregistered
310
311Since these externally allocated memory areas will not be managed by DPDK, it is
312therefore up to the user application to decide how to use them and what to do
313with them once they're registered.
314
315Per-lcore and Shared Variables
316~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
317
318.. note::
319
320    lcore refers to a logical execution unit of the processor, sometimes called a hardware *thread*.
321
322Shared variables are the default behavior.
323Per-lcore variables are implemented using *Thread Local Storage* (TLS) to provide per-thread local storage.
324
325Logs
326~~~~
327
328A logging API is provided by EAL.
329By default, in a Linux application, logs are sent to syslog and also to the console.
330However, the log function can be overridden by the user to use a different logging mechanism.
331
332Trace and Debug Functions
333^^^^^^^^^^^^^^^^^^^^^^^^^
334
335There are some debug functions to dump the stack in glibc.
336The rte_panic() function can voluntarily provoke a SIG_ABORT,
337which can trigger the generation of a core file, readable by gdb.
338
339CPU Feature Identification
340~~~~~~~~~~~~~~~~~~~~~~~~~~
341
342The EAL can query the CPU at runtime (using the rte_cpu_get_features() function) to determine which CPU features are available.
343
344User Space Interrupt Event
345~~~~~~~~~~~~~~~~~~~~~~~~~~
346
347+ User Space Interrupt and Alarm Handling in Host Thread
348
349The EAL creates a host thread to poll the UIO device file descriptors to detect the interrupts.
350Callbacks can be registered or unregistered by the EAL functions for a specific interrupt event
351and are called in the host thread asynchronously.
352The EAL also allows timed callbacks to be used in the same way as for NIC interrupts.
353
354.. note::
355
356    In DPDK PMD, the only interrupts handled by the dedicated host thread are those for link status change
357    (link up and link down notification) and for sudden device removal.
358
359
360+ RX Interrupt Event
361
362The receive and transmit routines provided by each PMD don't limit themselves to execute in polling thread mode.
363To ease the idle polling with tiny throughput, it's useful to pause the polling and wait until the wake-up event happens.
364The RX interrupt is the first choice to be such kind of wake-up event, but probably won't be the only one.
365
366EAL provides the event APIs for this event-driven thread mode.
367Taking Linux as an example, the implementation relies on epoll. Each thread can monitor an epoll instance
368in which all the wake-up events' file descriptors are added. The event file descriptors are created and mapped to
369the interrupt vectors according to the UIO/VFIO spec.
370From FreeBSD's perspective, kqueue is the alternative way, but not implemented yet.
371
372EAL initializes the mapping between event file descriptors and interrupt vectors, while each device initializes the mapping
373between interrupt vectors and queues. In this way, EAL actually is unaware of the interrupt cause on the specific vector.
374The eth_dev driver takes responsibility to program the latter mapping.
375
376.. note::
377
378    Per queue RX interrupt event is only allowed in VFIO which supports multiple MSI-X vector. In UIO, the RX interrupt
379    together with other interrupt causes shares the same vector. In this case, when RX interrupt and LSC(link status change)
380    interrupt are both enabled(intr_conf.lsc == 1 && intr_conf.rxq == 1), only the former is capable.
381
382The RX interrupt are controlled/enabled/disabled by ethdev APIs - 'rte_eth_dev_rx_intr_*'. They return failure if the PMD
383hasn't support them yet. The intr_conf.rxq flag is used to turn on the capability of RX interrupt per device.
384
385+ Device Removal Event
386
387This event is triggered by a device being removed at a bus level. Its
388underlying resources may have been made unavailable (i.e. PCI mappings
389unmapped). The PMD must make sure that on such occurrence, the application can
390still safely use its callbacks.
391
392This event can be subscribed to in the same way one would subscribe to a link
393status change event. The execution context is thus the same, i.e. it is the
394dedicated interrupt host thread.
395
396Considering this, it is likely that an application would want to close a
397device having emitted a Device Removal Event. In such case, calling
398``rte_eth_dev_close()`` can trigger it to unregister its own Device Removal Event
399callback. Care must be taken not to close the device from the interrupt handler
400context. It is necessary to reschedule such closing operation.
401
402Blacklisting
403~~~~~~~~~~~~
404
405The EAL PCI device blacklist functionality can be used to mark certain NIC ports as blacklisted,
406so they are ignored by the DPDK.
407The ports to be blacklisted are identified using the PCIe* description (Domain:Bus:Device.Function).
408
409Misc Functions
410~~~~~~~~~~~~~~
411
412Locks and atomic operations are per-architecture (i686 and x86_64).
413
414IOVA Mode Configuration
415~~~~~~~~~~~~~~~~~~~~~~~
416
417Auto detection of the IOVA mode, based on probing the bus and IOMMU configuration, may not report
418the desired addressing mode when virtual devices that are not directly attached to the bus are present.
419To facilitate forcing the IOVA mode to a specific value the EAL command line option ``--iova-mode`` can
420be used to select either physical addressing('pa') or virtual addressing('va').
421
422Memory Segments and Memory Zones (memzone)
423------------------------------------------
424
425The mapping of physical memory is provided by this feature in the EAL.
426As physical memory can have gaps, the memory is described in a table of descriptors,
427and each descriptor (called rte_memseg ) describes a physical page.
428
429On top of this, the memzone allocator's role is to reserve contiguous portions of physical memory.
430These zones are identified by a unique name when the memory is reserved.
431
432The rte_memzone descriptors are also located in the configuration structure.
433This structure is accessed using rte_eal_get_configuration().
434The lookup (by name) of a memory zone returns a descriptor containing the physical address of the memory zone.
435
436Memory zones can be reserved with specific start address alignment by supplying the align parameter
437(by default, they are aligned to cache line size).
438The alignment value should be a power of two and not less than the cache line size (64 bytes).
439Memory zones can also be reserved from either 2 MB or 1 GB hugepages, provided that both are available on the system.
440
441Both memsegs and memzones are stored using ``rte_fbarray`` structures. Please
442refer to *DPDK API Reference* for more information.
443
444
445Multiple pthread
446----------------
447
448DPDK usually pins one pthread per core to avoid the overhead of task switching.
449This allows for significant performance gains, but lacks flexibility and is not always efficient.
450
451Power management helps to improve the CPU efficiency by limiting the CPU runtime frequency.
452However, alternately it is possible to utilize the idle cycles available to take advantage of
453the full capability of the CPU.
454
455By taking advantage of cgroup, the CPU utilization quota can be simply assigned.
456This gives another way to improve the CPU efficiency, however, there is a prerequisite;
457DPDK must handle the context switching between multiple pthreads per core.
458
459For further flexibility, it is useful to set pthread affinity not only to a CPU but to a CPU set.
460
461EAL pthread and lcore Affinity
462~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
463
464The term "lcore" refers to an EAL thread, which is really a Linux/FreeBSD pthread.
465"EAL pthreads"  are created and managed by EAL and execute the tasks issued by *remote_launch*.
466In each EAL pthread, there is a TLS (Thread Local Storage) called *_lcore_id* for unique identification.
467As EAL pthreads usually bind 1:1 to the physical CPU, the *_lcore_id* is typically equal to the CPU ID.
468
469When using multiple pthreads, however, the binding is no longer always 1:1 between an EAL pthread and a specified physical CPU.
470The EAL pthread may have affinity to a CPU set, and as such the *_lcore_id* will not be the same as the CPU ID.
471For this reason, there is an EAL long option '--lcores' defined to assign the CPU affinity of lcores.
472For a specified lcore ID or ID group, the option allows setting the CPU set for that EAL pthread.
473
474The format pattern:
475	--lcores='<lcore_set>[@cpu_set][,<lcore_set>[@cpu_set],...]'
476
477'lcore_set' and 'cpu_set' can be a single number, range or a group.
478
479A number is a "digit([0-9]+)"; a range is "<number>-<number>"; a group is "(<number|range>[,<number|range>,...])".
480
481If a '\@cpu_set' value is not supplied, the value of 'cpu_set' will default to the value of 'lcore_set'.
482
483    ::
484
485    	For example, "--lcores='1,2@(5-7),(3-5)@(0,2),(0,6),7-8'" which means start 9 EAL thread;
486    	    lcore 0 runs on cpuset 0x41 (cpu 0,6);
487    	    lcore 1 runs on cpuset 0x2 (cpu 1);
488    	    lcore 2 runs on cpuset 0xe0 (cpu 5,6,7);
489    	    lcore 3,4,5 runs on cpuset 0x5 (cpu 0,2);
490    	    lcore 6 runs on cpuset 0x41 (cpu 0,6);
491    	    lcore 7 runs on cpuset 0x80 (cpu 7);
492    	    lcore 8 runs on cpuset 0x100 (cpu 8).
493
494Using this option, for each given lcore ID, the associated CPUs can be assigned.
495It's also compatible with the pattern of corelist('-l') option.
496
497non-EAL pthread support
498~~~~~~~~~~~~~~~~~~~~~~~
499
500It is possible to use the DPDK execution context with any user pthread (aka. Non-EAL pthreads).
501In 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*.
502Some 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).
503
504All these impacts are mentioned in :ref:`known_issue_label` section.
505
506Public Thread API
507~~~~~~~~~~~~~~~~~
508
509There are two public APIs ``rte_thread_set_affinity()`` and ``rte_thread_get_affinity()`` introduced for threads.
510When they're used in any pthread context, the Thread Local Storage(TLS) will be set/get.
511
512Those TLS include *_cpuset* and *_socket_id*:
513
514*	*_cpuset* stores the CPUs bitmap to which the pthread is affinitized.
515
516*	*_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.
517
518
519Control Thread API
520~~~~~~~~~~~~~~~~~~
521
522It is possible to create Control Threads using the public API
523``rte_ctrl_thread_create()``.
524Those threads can be used for management/infrastructure tasks and are used
525internally by DPDK for multi process support and interrupt handling.
526
527Those threads will be scheduled on CPUs part of the original process CPU
528affinity from which the dataplane and service lcores are excluded.
529
530For example, on a 8 CPUs system, starting a dpdk application with -l 2,3
531(dataplane cores), then depending on the affinity configuration which can be
532controlled with tools like taskset (Linux) or cpuset (FreeBSD),
533
534- with no affinity configuration, the Control Threads will end up on
535  0-1,4-7 CPUs.
536- with affinity restricted to 2-4, the Control Threads will end up on
537  CPU 4.
538- with affinity restricted to 2-3, the Control Threads will end up on
539  CPU 2 (master lcore, which is the default when no CPU is available).
540
541.. _known_issue_label:
542
543Known Issues
544~~~~~~~~~~~~
545
546+ rte_mempool
547
548  The rte_mempool uses a per-lcore cache inside the mempool.
549  For non-EAL pthreads, ``rte_lcore_id()`` will not return a valid number.
550  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.
551  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.
552
553+ rte_ring
554
555  rte_ring supports multi-producer enqueue and multi-consumer dequeue.
556  However, it is non-preemptive, this has a knock on effect of making rte_mempool non-preemptable.
557
558  .. note::
559
560    The "non-preemptive" constraint means:
561
562    - a pthread doing multi-producers enqueues on a given ring must not
563      be preempted by another pthread doing a multi-producer enqueue on
564      the same ring.
565    - a pthread doing multi-consumers dequeues on a given ring must not
566      be preempted by another pthread doing a multi-consumer dequeue on
567      the same ring.
568
569    Bypassing this constraint may cause the 2nd pthread to spin until the 1st one is scheduled again.
570    Moreover, if the 1st pthread is preempted by a context that has an higher priority, it may even cause a dead lock.
571
572  This means, use cases involving preemptible pthreads should consider using rte_ring carefully.
573
574  1. It CAN be used for preemptible single-producer and single-consumer use case.
575
576  2. It CAN be used for non-preemptible multi-producer and preemptible single-consumer use case.
577
578  3. It CAN be used for preemptible single-producer and non-preemptible multi-consumer use case.
579
580  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.
581
582  5. It MUST not be used by multi-producer/consumer pthreads, whose scheduling policies are SCHED_FIFO or SCHED_RR.
583
584  Alternatively, applications can use the lock-free stack mempool handler. When
585  considering this handler, note that:
586
587  - It is currently limited to the x86_64 platform, because it uses an
588    instruction (16-byte compare-and-swap) that is not yet available on other
589    platforms.
590  - It has worse average-case performance than the non-preemptive rte_ring, but
591    software caching (e.g. the mempool cache) can mitigate this by reducing the
592    number of stack accesses.
593
594+ rte_timer
595
596  Running  ``rte_timer_manage()`` on a non-EAL pthread is not allowed. However, resetting/stopping the timer from a non-EAL pthread is allowed.
597
598+ rte_log
599
600  In non-EAL pthreads, there is no per thread loglevel and logtype, global loglevels are used.
601
602+ misc
603
604  The debug statistics of rte_ring, rte_mempool and rte_timer are not supported in a non-EAL pthread.
605
606cgroup control
607~~~~~~~~~~~~~~
608
609The 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).
610We expect only 50% of CPU spend on packet IO.
611
612  .. code-block:: console
613
614    mkdir /sys/fs/cgroup/cpu/pkt_io
615    mkdir /sys/fs/cgroup/cpuset/pkt_io
616
617    echo $cpu > /sys/fs/cgroup/cpuset/cpuset.cpus
618
619    echo $t0 > /sys/fs/cgroup/cpu/pkt_io/tasks
620    echo $t0 > /sys/fs/cgroup/cpuset/pkt_io/tasks
621
622    echo $t1 > /sys/fs/cgroup/cpu/pkt_io/tasks
623    echo $t1 > /sys/fs/cgroup/cpuset/pkt_io/tasks
624
625    cd /sys/fs/cgroup/cpu/pkt_io
626    echo 100000 > pkt_io/cpu.cfs_period_us
627    echo  50000 > pkt_io/cpu.cfs_quota_us
628
629
630Malloc
631------
632
633The EAL provides a malloc API to allocate any-sized memory.
634
635The objective of this API is to provide malloc-like functions to allow
636allocation from hugepage memory and to facilitate application porting.
637The *DPDK API Reference* manual describes the available functions.
638
639Typically, these kinds of allocations should not be done in data plane
640processing because they are slower than pool-based allocation and make
641use of locks within the allocation and free paths.
642However, they can be used in configuration code.
643
644Refer to the rte_malloc() function description in the *DPDK API Reference*
645manual for more information.
646
647Cookies
648~~~~~~~
649
650When CONFIG_RTE_MALLOC_DEBUG is enabled, the allocated memory contains
651overwrite protection fields to help identify buffer overflows.
652
653Alignment and NUMA Constraints
654~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
655
656The rte_malloc() takes an align argument that can be used to request a memory
657area that is aligned on a multiple of this value (which must be a power of two).
658
659On systems with NUMA support, a call to the rte_malloc() function will return
660memory that has been allocated on the NUMA socket of the core which made the call.
661A set of APIs is also provided, to allow memory to be explicitly allocated on a
662NUMA socket directly, or by allocated on the NUMA socket where another core is
663located, in the case where the memory is to be used by a logical core other than
664on the one doing the memory allocation.
665
666Use Cases
667~~~~~~~~~
668
669This API is meant to be used by an application that requires malloc-like
670functions at initialization time.
671
672For allocating/freeing data at runtime, in the fast-path of an application,
673the memory pool library should be used instead.
674
675Internal Implementation
676~~~~~~~~~~~~~~~~~~~~~~~
677
678Data Structures
679^^^^^^^^^^^^^^^
680
681There are two data structure types used internally in the malloc library:
682
683*   struct malloc_heap - used to track free space on a per-socket basis
684
685*   struct malloc_elem - the basic element of allocation and free-space
686    tracking inside the library.
687
688Structure: malloc_heap
689""""""""""""""""""""""
690
691The malloc_heap structure is used to manage free space on a per-socket basis.
692Internally, there is one heap structure per NUMA node, which allows us to
693allocate memory to a thread based on the NUMA node on which this thread runs.
694While this does not guarantee that the memory will be used on that NUMA node,
695it is no worse than a scheme where the memory is always allocated on a fixed
696or random node.
697
698The key fields of the heap structure and their function are described below
699(see also diagram above):
700
701*   lock - the lock field is needed to synchronize access to the heap.
702    Given that the free space in the heap is tracked using a linked list,
703    we need a lock to prevent two threads manipulating the list at the same time.
704
705*   free_head - this points to the first element in the list of free nodes for
706    this malloc heap.
707
708*   first - this points to the first element in the heap.
709
710*   last - this points to the last element in the heap.
711
712.. _figure_malloc_heap:
713
714.. figure:: img/malloc_heap.*
715
716   Example of a malloc heap and malloc elements within the malloc library
717
718
719.. _malloc_elem:
720
721Structure: malloc_elem
722""""""""""""""""""""""
723
724The malloc_elem structure is used as a generic header structure for various
725blocks of memory.
726It is used in two different ways - all shown in the diagram above:
727
728#.  As a header on a block of free or allocated memory - normal case
729
730#.  As a padding header inside a block of memory
731
732The most important fields in the structure and how they are used are described below.
733
734Malloc heap is a doubly-linked list, where each element keeps track of its
735previous and next elements. Due to the fact that hugepage memory can come and
736go, neighbouring malloc elements may not necessarily be adjacent in memory.
737Also, since a malloc element may span multiple pages, its contents may not
738necessarily be IOVA-contiguous either - each malloc element is only guaranteed
739to be virtually contiguous.
740
741.. note::
742
743    If the usage of a particular field in one of the above three usages is not
744    described, the field can be assumed to have an undefined value in that
745    situation, for example, for padding headers only the "state" and "pad"
746    fields have valid values.
747
748*   heap - this pointer is a reference back to the heap structure from which
749    this block was allocated.
750    It is used for normal memory blocks when they are being freed, to add the
751    newly-freed block to the heap's free-list.
752
753*   prev - this pointer points to previous header element/block in memory. When
754    freeing a block, this pointer is used to reference the previous block to
755    check if that block is also free. If so, and the two blocks are immediately
756    adjacent to each other, then the two free blocks are merged to form a single
757    larger block.
758
759*   next - this pointer points to next header element/block in memory. When
760    freeing a block, this pointer is used to reference the next block to check
761    if that block is also free. If so, and the two blocks are immediately
762    adjacent to each other, then the two free blocks are merged to form a single
763    larger block.
764
765*   free_list - this is a structure pointing to previous and next elements in
766    this heap's free list.
767    It is only used in normal memory blocks; on ``malloc()`` to find a suitable
768    free block to allocate and on ``free()`` to add the newly freed element to
769    the free-list.
770
771*   state - This field can have one of three values: ``FREE``, ``BUSY`` or
772    ``PAD``.
773    The former two are to indicate the allocation state of a normal memory block
774    and the latter is to indicate that the element structure is a dummy structure
775    at the end of the start-of-block padding, i.e. where the start of the data
776    within a block is not at the start of the block itself, due to alignment
777    constraints.
778    In that case, the pad header is used to locate the actual malloc element
779    header for the block.
780
781*   pad - this holds the length of the padding present at the start of the block.
782    In the case of a normal block header, it is added to the address of the end
783    of the header to give the address of the start of the data area, i.e. the
784    value passed back to the application on a malloc.
785    Within a dummy header inside the padding, this same value is stored, and is
786    subtracted from the address of the dummy header to yield the address of the
787    actual block header.
788
789*   size - the size of the data block, including the header itself.
790
791Memory Allocation
792^^^^^^^^^^^^^^^^^
793
794On EAL initialization, all preallocated memory segments are setup as part of the
795malloc heap. This setup involves placing an :ref:`element header<malloc_elem>`
796with ``FREE`` at the start of each virtually contiguous segment of memory.
797The ``FREE`` element is then added to the ``free_list`` for the malloc heap.
798
799This setup also happens whenever memory is allocated at runtime (if supported),
800in which case newly allocated pages are also added to the heap, merging with any
801adjacent free segments if there are any.
802
803When an application makes a call to a malloc-like function, the malloc function
804will first index the ``lcore_config`` structure for the calling thread, and
805determine the NUMA node of that thread.
806The NUMA node is used to index the array of ``malloc_heap`` structures which is
807passed as a parameter to the ``heap_alloc()`` function, along with the
808requested size, type, alignment and boundary parameters.
809
810The ``heap_alloc()`` function will scan the free_list of the heap, and attempt
811to find a free block suitable for storing data of the requested size, with the
812requested alignment and boundary constraints.
813
814When a suitable free element has been identified, the pointer to be returned
815to the user is calculated.
816The cache-line of memory immediately preceding this pointer is filled with a
817struct malloc_elem header.
818Because of alignment and boundary constraints, there could be free space at
819the start and/or end of the element, resulting in the following behavior:
820
821#. Check for trailing space.
822   If the trailing space is big enough, i.e. > 128 bytes, then the free element
823   is split.
824   If it is not, then we just ignore it (wasted space).
825
826#. Check for space at the start of the element.
827   If the space at the start is small, i.e. <=128 bytes, then a pad header is
828   used, and the remaining space is wasted.
829   If, however, the remaining space is greater, then the free element is split.
830
831The advantage of allocating the memory from the end of the existing element is
832that no adjustment of the free list needs to take place - the existing element
833on the free list just has its size value adjusted, and the next/previous elements
834have their "prev"/"next" pointers redirected to the newly created element.
835
836In case when there is not enough memory in the heap to satisfy allocation
837request, EAL will attempt to allocate more memory from the system (if supported)
838and, following successful allocation, will retry reserving the memory again. In
839a multiprocessing scenario, all primary and secondary processes will synchronize
840their memory maps to ensure that any valid pointer to DPDK memory is guaranteed
841to be valid at all times in all currently running processes.
842
843Failure to synchronize memory maps in one of the processes will cause allocation
844to fail, even though some of the processes may have allocated the memory
845successfully. The memory is not added to the malloc heap unless primary process
846has ensured that all other processes have mapped this memory successfully.
847
848Any successful allocation event will trigger a callback, for which user
849applications and other DPDK subsystems can register. Additionally, validation
850callbacks will be triggered before allocation if the newly allocated memory will
851exceed threshold set by the user, giving a chance to allow or deny allocation.
852
853.. note::
854
855    Any allocation of new pages has to go through primary process. If the
856    primary process is not active, no memory will be allocated even if it was
857    theoretically possible to do so. This is because primary's process map acts
858    as an authority on what should or should not be mapped, while each secondary
859    process has its own, local memory map. Secondary processes do not update the
860    shared memory map, they only copy its contents to their local memory map.
861
862Freeing Memory
863^^^^^^^^^^^^^^
864
865To free an area of memory, the pointer to the start of the data area is passed
866to the free function.
867The size of the ``malloc_elem`` structure is subtracted from this pointer to get
868the element header for the block.
869If this header is of type ``PAD`` then the pad length is further subtracted from
870the pointer to get the proper element header for the entire block.
871
872From this element header, we get pointers to the heap from which the block was
873allocated and to where it must be freed, as well as the pointer to the previous
874and next elements. These next and previous elements are then checked to see if
875they are also ``FREE`` and are immediately adjacent to the current one, and if
876so, they are merged with the current element. This means that we can never have
877two ``FREE`` memory blocks adjacent to one another, as they are always merged
878into a single block.
879
880If deallocating pages at runtime is supported, and the free element encloses
881one or more pages, those pages can be deallocated and be removed from the heap.
882If DPDK was started with command-line parameters for preallocating memory
883(``-m`` or ``--socket-mem``), then those pages that were allocated at startup
884will not be deallocated.
885
886Any successful deallocation event will trigger a callback, for which user
887applications and other DPDK subsystems can register.
888