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 main 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.. warning:: 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 156 DPDK IPC. 157 158+ Legacy memory mode 159 160This mode is enabled by specifying ``--legacy-mem`` command-line switch to the 161EAL. This switch will have no effect on FreeBSD as FreeBSD only supports 162legacy mode anyway. 163 164This mode mimics historical behavior of EAL. That is, EAL will reserve all 165memory at startup, sort all memory into large IOVA-contiguous chunks, and will 166not allow acquiring or releasing hugepages from the system at runtime. 167 168If neither ``-m`` nor ``--socket-mem`` were specified, the entire available 169hugepage memory will be preallocated. 170 171+ Hugepage allocation matching 172 173This behavior is enabled by specifying the ``--match-allocations`` command-line 174switch to the EAL. This switch is Linux-only and not supported with 175``--legacy-mem`` nor ``--no-huge``. 176 177Some applications using memory event callbacks may require that hugepages be 178freed exactly as they were allocated. These applications may also require 179that any allocation from the malloc heap not span across allocations 180associated with two different memory event callbacks. Hugepage allocation 181matching can be used by these types of applications to satisfy both of these 182requirements. This can result in some increased memory usage which is 183very dependent on the memory allocation patterns of the application. 184 185+ 32-bit support 186 187Additional restrictions are present when running in 32-bit mode. In dynamic 188memory mode, by default maximum of 2 gigabytes of VA space will be preallocated, 189and all of it will be on main lcore NUMA node unless ``--socket-mem`` flag is 190used. 191 192In legacy mode, VA space will only be preallocated for segments that were 193requested (plus padding, to keep IOVA-contiguousness). 194 195+ Maximum amount of memory 196 197All possible virtual memory space that can ever be used for hugepage mapping in 198a DPDK process is preallocated at startup, thereby placing an upper limit on how 199much memory a DPDK application can have. DPDK memory is stored in segment lists, 200each segment is strictly one physical page. It is possible to change the amount 201of virtual memory being preallocated at startup by editing the following config 202variables: 203 204* ``RTE_MAX_MEMSEG_LISTS`` controls how many segment lists can DPDK have 205* ``RTE_MAX_MEM_MB_PER_LIST`` controls how much megabytes of memory each 206 segment list can address 207* ``RTE_MAX_MEMSEG_PER_LIST`` controls how many segments each segment list 208 can have 209* ``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* ``RTE_MAX_MEM_MB_PER_TYPE`` controls how much megabytes of memory each 212 memory type can address 213* ``RTE_MAX_MEM_MB`` places a global maximum on the amount of memory 214 DPDK can reserve 215 216Normally, these options do not need to be changed. 217 218.. note:: 219 220 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. 224 225+ Segment file descriptors 226 227On Linux, in most cases, EAL will store segment file descriptors in EAL. This 228can become a problem when using smaller page sizes due to underlying limitations 229of ``glibc`` library. For example, Linux API calls such as ``select()`` may not 230work correctly because ``glibc`` does not support more than certain number of 231file descriptors. 232 233There are two possible solutions for this problem. The recommended solution is 234to use ``--single-file-segments`` mode, as that mode will not use a file 235descriptor per each page, and it will keep compatibility with Virtio with 236vhost-user backend. This option is not available when using ``--legacy-mem`` 237mode. 238 239Another option is to use bigger page sizes. Since fewer pages are required to 240cover the same memory area, fewer file descriptors will be stored internally 241by EAL. 242 243Support for Externally Allocated Memory 244~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 245 246It is possible to use externally allocated memory in DPDK. There are two ways in 247which using externally allocated memory can work: the malloc heap API's, and 248manual memory management. 249 250+ Using heap API's for externally allocated memory 251 252Using a set of malloc heap API's is the recommended way to use externally 253allocated memory in DPDK. In this way, support for externally allocated memory 254is implemented through overloading the socket ID - externally allocated heaps 255will have socket ID's that would be considered invalid under normal 256circumstances. Requesting an allocation to take place from a specified 257externally allocated memory is a matter of supplying the correct socket ID to 258DPDK allocator, either directly (e.g. through a call to ``rte_malloc``) or 259indirectly (through data structure-specific allocation API's such as 260``rte_ring_create``). Using these API's also ensures that mapping of externally 261allocated memory for DMA is also performed on any memory segment that is added 262to a DPDK malloc heap. 263 264Since there is no way DPDK can verify whether memory is available or valid, this 265responsibility falls on the shoulders of the user. All multiprocess 266synchronization is also user's responsibility, as well as ensuring that all 267calls to add/attach/detach/remove memory are done in the correct order. It is 268not required to attach to a memory area in all processes - only attach to memory 269areas as needed. 270 271The expected workflow is as follows: 272 273* 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 285 286For more information, please refer to ``rte_malloc`` API documentation, 287specifically the ``rte_malloc_heap_*`` family of function calls. 288 289+ Using externally allocated memory without DPDK API's 290 291While using heap API's is the recommended method of using externally allocated 292memory in DPDK, there are certain use cases where the overhead of DPDK heap API 293is undesirable - for example, when manual memory management is performed on an 294externally allocated area. To support use cases where externally allocated 295memory will not be used as part of normal DPDK workflow, there is also another 296set of API's under the ``rte_extmem_*`` namespace. 297 298These API's are (as their name implies) intended to allow registering or 299unregistering externally allocated memory to/from DPDK's internal page table, to 300allow API's like ``rte_mem_virt2memseg`` etc. to work with externally allocated 301memory. Memory added this way will not be available for any regular DPDK 302allocators; DPDK will leave this memory for the user application to manage. 303 304The expected workflow is as follows: 305 306* 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 309 unavailable 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 315 unregistering memory 316 - Other processes must detach from the memory area before it can be 317 unregistered 318 319Since these externally allocated memory areas will not be managed by DPDK, it is 320therefore up to the user application to decide how to use them and what to do 321with them once they're registered. 322 323Per-lcore and Shared Variables 324~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 325 326.. note:: 327 328 lcore refers to a logical execution unit of the processor, sometimes called a hardware *thread*. 329 330Shared variables are the default behavior. 331Per-lcore variables are implemented using *Thread Local Storage* (TLS) to provide per-thread local storage. 332 333Logs 334~~~~ 335 336A logging API is provided by EAL. 337By default, in a Linux application, logs are sent to syslog and also to the console. 338However, the log function can be overridden by the user to use a different logging mechanism. 339 340Trace and Debug Functions 341^^^^^^^^^^^^^^^^^^^^^^^^^ 342 343There are some debug functions to dump the stack in glibc. 344The rte_panic() function can voluntarily provoke a SIG_ABORT, 345which can trigger the generation of a core file, readable by gdb. 346 347CPU Feature Identification 348~~~~~~~~~~~~~~~~~~~~~~~~~~ 349 350The EAL can query the CPU at runtime (using the rte_cpu_get_features() function) to determine which CPU features are available. 351 352User Space Interrupt Event 353~~~~~~~~~~~~~~~~~~~~~~~~~~ 354 355+ User Space Interrupt and Alarm Handling in Host Thread 356 357The EAL creates a host thread to poll the UIO device file descriptors to detect the interrupts. 358Callbacks can be registered or unregistered by the EAL functions for a specific interrupt event 359and are called in the host thread asynchronously. 360The EAL also allows timed callbacks to be used in the same way as for NIC interrupts. 361 362.. note:: 363 364 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. 366 367 368+ RX Interrupt Event 369 370The receive and transmit routines provided by each PMD don't limit themselves to execute in polling thread mode. 371To ease the idle polling with tiny throughput, it's useful to pause the polling and wait until the wake-up event happens. 372The RX interrupt is the first choice to be such kind of wake-up event, but probably won't be the only one. 373 374EAL provides the event APIs for this event-driven thread mode. 375Taking Linux as an example, the implementation relies on epoll. Each thread can monitor an epoll instance 376in which all the wake-up events' file descriptors are added. The event file descriptors are created and mapped to 377the interrupt vectors according to the UIO/VFIO spec. 378From FreeBSD's perspective, kqueue is the alternative way, but not implemented yet. 379 380EAL initializes the mapping between event file descriptors and interrupt vectors, while each device initializes the mapping 381between interrupt vectors and queues. In this way, EAL actually is unaware of the interrupt cause on the specific vector. 382The eth_dev driver takes responsibility to program the latter mapping. 383 384.. note:: 385 386 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. 389 390The RX interrupt are controlled/enabled/disabled by ethdev APIs - 'rte_eth_dev_rx_intr_*'. They return failure if the PMD 391hasn't support them yet. The intr_conf.rxq flag is used to turn on the capability of RX interrupt per device. 392 393+ Device Removal Event 394 395This event is triggered by a device being removed at a bus level. Its 396underlying resources may have been made unavailable (i.e. PCI mappings 397unmapped). The PMD must make sure that on such occurrence, the application can 398still safely use its callbacks. 399 400This event can be subscribed to in the same way one would subscribe to a link 401status change event. The execution context is thus the same, i.e. it is the 402dedicated interrupt host thread. 403 404Considering this, it is likely that an application would want to close a 405device 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 407callback. Care must be taken not to close the device from the interrupt handler 408context. It is necessary to reschedule such closing operation. 409 410Block list 411~~~~~~~~~~ 412 413The EAL PCI device block list functionality can be used to mark certain NIC ports as unavailable, 414so they are ignored by the DPDK. 415The ports to be blocked are identified using the PCIe* description (Domain:Bus:Device.Function). 416 417Misc Functions 418~~~~~~~~~~~~~~ 419 420Locks and atomic operations are per-architecture (i686 and x86_64). 421 422IOVA Mode Detection 423~~~~~~~~~~~~~~~~~~~ 424 425IOVA Mode is selected by considering what the current usable Devices on the 426system require and/or support. 427 428On FreeBSD, RTE_IOVA_PA is always the default. On Linux, the IOVA mode is 429detected based on a 2-step heuristic detailed below. 430 431For the first step, EAL asks each bus its requirement in terms of IOVA mode 432and decides on a preferred IOVA mode. 433 434- 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), 441 442If the buses have expressed no preference on which IOVA mode to pick, then a 443default is selected using the following logic: 444 445- 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 448 449In the case when the buses had disagreed on their preferred IOVA mode, part of 450the buses won't work because of this decision. 451 452The second step checks if the preferred mode complies with the Physical 453Addresses availability since those are only available to root user in recent 454kernels. Namely, if the preferred mode is RTE_IOVA_PA but there is no access to 455Physical Addresses, then EAL init fails early, since later probing of the 456devices would fail anyway. 457 458.. note:: 459 460 The RTE_IOVA_VA mode is preferred as the default in most cases for the 461 following reasons: 462 463 - 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. 470 471 It is expected that all PCI drivers work in both RTE_IOVA_PA and 472 RTE_IOVA_VA modes. 473 474 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. 477 478 When the KNI kernel module is detected, RTE_IOVA_PA mode is preferred as a 479 performance penalty is expected in RTE_IOVA_VA mode. 480 481IOVA Mode Configuration 482~~~~~~~~~~~~~~~~~~~~~~~ 483 484Auto detection of the IOVA mode, based on probing the bus and IOMMU configuration, may not report 485the desired addressing mode when virtual devices that are not directly attached to the bus are present. 486To facilitate forcing the IOVA mode to a specific value the EAL command line option ``--iova-mode`` can 487be used to select either physical addressing('pa') or virtual addressing('va'). 488 489.. _max_simd_bitwidth: 490 491 492Max SIMD bitwidth 493~~~~~~~~~~~~~~~~~ 494 495The EAL provides a single setting to limit the max SIMD bitwidth used by DPDK, 496which is used in determining the vector path, if any, chosen by a component. 497The value can be set at runtime by an application using the 498'rte_vect_set_max_simd_bitwidth(uint16_t bitwidth)' function, 499which should only be called once at initialization, before EAL init. 500The value can be overridden by the user using the EAL command-line option '--force-max-simd-bitwidth'. 501 502When choosing a vector path, along with checking the CPU feature support, 503the value of the max SIMD bitwidth must also be checked, and can be retrieved using the 504'rte_vect_get_max_simd_bitwidth()' function. 505The value should be compared against the enum values for accepted max SIMD bitwidths: 506 507.. code-block:: c 508 509 enum rte_vect_max_simd { 510 RTE_VECT_SIMD_DISABLED = 64, 511 RTE_VECT_SIMD_128 = 128, 512 RTE_VECT_SIMD_256 = 256, 513 RTE_VECT_SIMD_512 = 512, 514 RTE_VECT_SIMD_MAX = INT16_MAX + 1, 515 }; 516 517 if (rte_vect_get_max_simd_bitwidth() >= RTE_VECT_SIMD_512) 518 /* Take AVX-512 vector path */ 519 else if (rte_vect_get_max_simd_bitwidth() >= RTE_VECT_SIMD_256) 520 /* Take AVX2 vector path */ 521 522 523Memory Segments and Memory Zones (memzone) 524------------------------------------------ 525 526The mapping of physical memory is provided by this feature in the EAL. 527As physical memory can have gaps, the memory is described in a table of descriptors, 528and each descriptor (called rte_memseg ) describes a physical page. 529 530On top of this, the memzone allocator's role is to reserve contiguous portions of physical memory. 531These zones are identified by a unique name when the memory is reserved. 532 533The rte_memzone descriptors are also located in the configuration structure. 534This structure is accessed using rte_eal_get_configuration(). 535The lookup (by name) of a memory zone returns a descriptor containing the physical address of the memory zone. 536 537Memory zones can be reserved with specific start address alignment by supplying the align parameter 538(by default, they are aligned to cache line size). 539The alignment value should be a power of two and not less than the cache line size (64 bytes). 540Memory zones can also be reserved from either 2 MB or 1 GB hugepages, provided that both are available on the system. 541 542Both memsegs and memzones are stored using ``rte_fbarray`` structures. Please 543refer to *DPDK API Reference* for more information. 544 545 546Multiple pthread 547---------------- 548 549DPDK usually pins one pthread per core to avoid the overhead of task switching. 550This allows for significant performance gains, but lacks flexibility and is not always efficient. 551 552Power management helps to improve the CPU efficiency by limiting the CPU runtime frequency. 553However, alternately it is possible to utilize the idle cycles available to take advantage of 554the full capability of the CPU. 555 556By taking advantage of cgroup, the CPU utilization quota can be simply assigned. 557This gives another way to improve the CPU efficiency, however, there is a prerequisite; 558DPDK must handle the context switching between multiple pthreads per core. 559 560For further flexibility, it is useful to set pthread affinity not only to a CPU but to a CPU set. 561 562EAL pthread and lcore Affinity 563~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 564 565The term "lcore" refers to an EAL thread, which is really a Linux/FreeBSD pthread. 566"EAL pthreads" are created and managed by EAL and execute the tasks issued by *remote_launch*. 567In each EAL pthread, there is a TLS (Thread Local Storage) called *_lcore_id* for unique identification. 568As EAL pthreads usually bind 1:1 to the physical CPU, the *_lcore_id* is typically equal to the CPU ID. 569 570When using multiple pthreads, however, the binding is no longer always 1:1 between an EAL pthread and a specified physical CPU. 571The EAL pthread may have affinity to a CPU set, and as such the *_lcore_id* will not be the same as the CPU ID. 572For this reason, there is an EAL long option '--lcores' defined to assign the CPU affinity of lcores. 573For a specified lcore ID or ID group, the option allows setting the CPU set for that EAL pthread. 574 575The format pattern: 576 --lcores='<lcore_set>[@cpu_set][,<lcore_set>[@cpu_set],...]' 577 578'lcore_set' and 'cpu_set' can be a single number, range or a group. 579 580A number is a "digit([0-9]+)"; a range is "<number>-<number>"; a group is "(<number|range>[,<number|range>,...])". 581 582If a '\@cpu_set' value is not supplied, the value of 'cpu_set' will default to the value of 'lcore_set'. 583 584 :: 585 586 For example, "--lcores='1,2@(5-7),(3-5)@(0,2),(0,6),7-8'" which means start 9 EAL thread; 587 lcore 0 runs on cpuset 0x41 (cpu 0,6); 588 lcore 1 runs on cpuset 0x2 (cpu 1); 589 lcore 2 runs on cpuset 0xe0 (cpu 5,6,7); 590 lcore 3,4,5 runs on cpuset 0x5 (cpu 0,2); 591 lcore 6 runs on cpuset 0x41 (cpu 0,6); 592 lcore 7 runs on cpuset 0x80 (cpu 7); 593 lcore 8 runs on cpuset 0x100 (cpu 8). 594 595Using this option, for each given lcore ID, the associated CPUs can be assigned. 596It's also compatible with the pattern of corelist('-l') option. 597 598non-EAL pthread support 599~~~~~~~~~~~~~~~~~~~~~~~ 600 601It is possible to use the DPDK execution context with any user pthread (aka. non-EAL pthreads). 602There are two kinds of non-EAL pthreads: 603 604- a registered non-EAL pthread with a valid *_lcore_id* that was successfully assigned by calling ``rte_thread_register()``, 605- a non registered non-EAL pthread with a LCORE_ID_ANY, 606 607For non registered non-EAL pthread (with a LCORE_ID_ANY *_lcore_id*), 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). 608 609All these impacts are mentioned in :ref:`known_issue_label` section. 610 611Public Thread API 612~~~~~~~~~~~~~~~~~ 613 614There are two public APIs ``rte_thread_set_affinity()`` and ``rte_thread_get_affinity()`` introduced for threads. 615When they're used in any pthread context, the Thread Local Storage(TLS) will be set/get. 616 617Those TLS include *_cpuset* and *_socket_id*: 618 619* *_cpuset* stores the CPUs bitmap to which the pthread is affinitized. 620 621* *_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. 622 623 624Control Thread API 625~~~~~~~~~~~~~~~~~~ 626 627It is possible to create Control Threads using the public API 628``rte_ctrl_thread_create()``. 629Those threads can be used for management/infrastructure tasks and are used 630internally by DPDK for multi process support and interrupt handling. 631 632Those threads will be scheduled on CPUs part of the original process CPU 633affinity from which the dataplane and service lcores are excluded. 634 635For example, on a 8 CPUs system, starting a dpdk application with -l 2,3 636(dataplane cores), then depending on the affinity configuration which can be 637controlled with tools like taskset (Linux) or cpuset (FreeBSD), 638 639- with no affinity configuration, the Control Threads will end up on 640 0-1,4-7 CPUs. 641- with affinity restricted to 2-4, the Control Threads will end up on 642 CPU 4. 643- with affinity restricted to 2-3, the Control Threads will end up on 644 CPU 2 (main lcore, which is the default when no CPU is available). 645 646.. _known_issue_label: 647 648Known Issues 649~~~~~~~~~~~~ 650 651+ rte_mempool 652 653 The rte_mempool uses a per-lcore cache inside the mempool. 654 For unregistered non-EAL pthreads, ``rte_lcore_id()`` will not return a valid number. 655 So for now, when rte_mempool is used with unregistered non-EAL pthreads, the put/get operations will bypass the default mempool cache and there is a performance penalty because of this bypass. 656 Only user-owned external caches can be used in an unregistered non-EAL context in conjunction with ``rte_mempool_generic_put()`` and ``rte_mempool_generic_get()`` that accept an explicit cache parameter. 657 658+ rte_ring 659 660 rte_ring supports multi-producer enqueue and multi-consumer dequeue. 661 However, it is non-preemptive, this has a knock on effect of making rte_mempool non-preemptable. 662 663 .. note:: 664 665 The "non-preemptive" constraint means: 666 667 - a pthread doing multi-producers enqueues on a given ring must not 668 be preempted by another pthread doing a multi-producer enqueue on 669 the same ring. 670 - a pthread doing multi-consumers dequeues on a given ring must not 671 be preempted by another pthread doing a multi-consumer dequeue on 672 the same ring. 673 674 Bypassing this constraint may cause the 2nd pthread to spin until the 1st one is scheduled again. 675 Moreover, if the 1st pthread is preempted by a context that has an higher priority, it may even cause a dead lock. 676 677 This means, use cases involving preemptible pthreads should consider using rte_ring carefully. 678 679 1. It CAN be used for preemptible single-producer and single-consumer use case. 680 681 2. It CAN be used for non-preemptible multi-producer and preemptible single-consumer use case. 682 683 3. It CAN be used for preemptible single-producer and non-preemptible multi-consumer use case. 684 685 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. 686 687 5. It MUST not be used by multi-producer/consumer pthreads, whose scheduling policies are SCHED_FIFO or SCHED_RR. 688 689 Alternatively, applications can use the lock-free stack mempool handler. When 690 considering this handler, note that: 691 692 - It is currently limited to the aarch64 and x86_64 platforms, because it uses 693 an instruction (16-byte compare-and-swap) that is not yet available on other 694 platforms. 695 - It has worse average-case performance than the non-preemptive rte_ring, but 696 software caching (e.g. the mempool cache) can mitigate this by reducing the 697 number of stack accesses. 698 699+ rte_timer 700 701 Running ``rte_timer_manage()`` on an unregistered non-EAL pthread is not allowed. However, resetting/stopping the timer from a non-EAL pthread is allowed. 702 703+ rte_log 704 705 In unregistered non-EAL pthreads, there is no per thread loglevel and logtype, global loglevels are used. 706 707+ misc 708 709 The debug statistics of rte_ring, rte_mempool and rte_timer are not supported in an unregistered non-EAL pthread. 710 711cgroup control 712~~~~~~~~~~~~~~ 713 714The 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). 715We expect only 50% of CPU spend on packet IO. 716 717 .. code-block:: console 718 719 mkdir /sys/fs/cgroup/cpu/pkt_io 720 mkdir /sys/fs/cgroup/cpuset/pkt_io 721 722 echo $cpu > /sys/fs/cgroup/cpuset/cpuset.cpus 723 724 echo $t0 > /sys/fs/cgroup/cpu/pkt_io/tasks 725 echo $t0 > /sys/fs/cgroup/cpuset/pkt_io/tasks 726 727 echo $t1 > /sys/fs/cgroup/cpu/pkt_io/tasks 728 echo $t1 > /sys/fs/cgroup/cpuset/pkt_io/tasks 729 730 cd /sys/fs/cgroup/cpu/pkt_io 731 echo 100000 > pkt_io/cpu.cfs_period_us 732 echo 50000 > pkt_io/cpu.cfs_quota_us 733 734 735Malloc 736------ 737 738The EAL provides a malloc API to allocate any-sized memory. 739 740The objective of this API is to provide malloc-like functions to allow 741allocation from hugepage memory and to facilitate application porting. 742The *DPDK API Reference* manual describes the available functions. 743 744Typically, these kinds of allocations should not be done in data plane 745processing because they are slower than pool-based allocation and make 746use of locks within the allocation and free paths. 747However, they can be used in configuration code. 748 749Refer to the rte_malloc() function description in the *DPDK API Reference* 750manual for more information. 751 752 753Alignment and NUMA Constraints 754~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 755 756The rte_malloc() takes an align argument that can be used to request a memory 757area that is aligned on a multiple of this value (which must be a power of two). 758 759On systems with NUMA support, a call to the rte_malloc() function will return 760memory that has been allocated on the NUMA socket of the core which made the call. 761A set of APIs is also provided, to allow memory to be explicitly allocated on a 762NUMA socket directly, or by allocated on the NUMA socket where another core is 763located, in the case where the memory is to be used by a logical core other than 764on the one doing the memory allocation. 765 766Use Cases 767~~~~~~~~~ 768 769This API is meant to be used by an application that requires malloc-like 770functions at initialization time. 771 772For allocating/freeing data at runtime, in the fast-path of an application, 773the memory pool library should be used instead. 774 775Internal Implementation 776~~~~~~~~~~~~~~~~~~~~~~~ 777 778Data Structures 779^^^^^^^^^^^^^^^ 780 781There are two data structure types used internally in the malloc library: 782 783* struct malloc_heap - used to track free space on a per-socket basis 784 785* struct malloc_elem - the basic element of allocation and free-space 786 tracking inside the library. 787 788Structure: malloc_heap 789"""""""""""""""""""""" 790 791The malloc_heap structure is used to manage free space on a per-socket basis. 792Internally, there is one heap structure per NUMA node, which allows us to 793allocate memory to a thread based on the NUMA node on which this thread runs. 794While this does not guarantee that the memory will be used on that NUMA node, 795it is no worse than a scheme where the memory is always allocated on a fixed 796or random node. 797 798The key fields of the heap structure and their function are described below 799(see also diagram above): 800 801* lock - the lock field is needed to synchronize access to the heap. 802 Given that the free space in the heap is tracked using a linked list, 803 we need a lock to prevent two threads manipulating the list at the same time. 804 805* free_head - this points to the first element in the list of free nodes for 806 this malloc heap. 807 808* first - this points to the first element in the heap. 809 810* last - this points to the last element in the heap. 811 812.. _figure_malloc_heap: 813 814.. figure:: img/malloc_heap.* 815 816 Example of a malloc heap and malloc elements within the malloc library 817 818 819.. _malloc_elem: 820 821Structure: malloc_elem 822"""""""""""""""""""""" 823 824The malloc_elem structure is used as a generic header structure for various 825blocks of memory. 826It is used in two different ways - all shown in the diagram above: 827 828#. As a header on a block of free or allocated memory - normal case 829 830#. As a padding header inside a block of memory 831 832The most important fields in the structure and how they are used are described below. 833 834Malloc heap is a doubly-linked list, where each element keeps track of its 835previous and next elements. Due to the fact that hugepage memory can come and 836go, neighboring malloc elements may not necessarily be adjacent in memory. 837Also, since a malloc element may span multiple pages, its contents may not 838necessarily be IOVA-contiguous either - each malloc element is only guaranteed 839to be virtually contiguous. 840 841.. note:: 842 843 If the usage of a particular field in one of the above three usages is not 844 described, the field can be assumed to have an undefined value in that 845 situation, for example, for padding headers only the "state" and "pad" 846 fields have valid values. 847 848* heap - this pointer is a reference back to the heap structure from which 849 this block was allocated. 850 It is used for normal memory blocks when they are being freed, to add the 851 newly-freed block to the heap's free-list. 852 853* prev - this pointer points to previous header element/block in memory. When 854 freeing a block, this pointer is used to reference the previous block to 855 check if that block is also free. If so, and the two blocks are immediately 856 adjacent to each other, then the two free blocks are merged to form a single 857 larger block. 858 859* next - this pointer points to next header element/block in memory. When 860 freeing a block, this pointer is used to reference the next block to check 861 if that block is also free. If so, and the two blocks are immediately 862 adjacent to each other, then the two free blocks are merged to form a single 863 larger block. 864 865* free_list - this is a structure pointing to previous and next elements in 866 this heap's free list. 867 It is only used in normal memory blocks; on ``malloc()`` to find a suitable 868 free block to allocate and on ``free()`` to add the newly freed element to 869 the free-list. 870 871* state - This field can have one of three values: ``FREE``, ``BUSY`` or 872 ``PAD``. 873 The former two are to indicate the allocation state of a normal memory block 874 and the latter is to indicate that the element structure is a dummy structure 875 at the end of the start-of-block padding, i.e. where the start of the data 876 within a block is not at the start of the block itself, due to alignment 877 constraints. 878 In that case, the pad header is used to locate the actual malloc element 879 header for the block. 880 881* pad - this holds the length of the padding present at the start of the block. 882 In the case of a normal block header, it is added to the address of the end 883 of the header to give the address of the start of the data area, i.e. the 884 value passed back to the application on a malloc. 885 Within a dummy header inside the padding, this same value is stored, and is 886 subtracted from the address of the dummy header to yield the address of the 887 actual block header. 888 889* size - the size of the data block, including the header itself. 890 891Memory Allocation 892^^^^^^^^^^^^^^^^^ 893 894On EAL initialization, all preallocated memory segments are setup as part of the 895malloc heap. This setup involves placing an :ref:`element header<malloc_elem>` 896with ``FREE`` at the start of each virtually contiguous segment of memory. 897The ``FREE`` element is then added to the ``free_list`` for the malloc heap. 898 899This setup also happens whenever memory is allocated at runtime (if supported), 900in which case newly allocated pages are also added to the heap, merging with any 901adjacent free segments if there are any. 902 903When an application makes a call to a malloc-like function, the malloc function 904will first index the ``lcore_config`` structure for the calling thread, and 905determine the NUMA node of that thread. 906The NUMA node is used to index the array of ``malloc_heap`` structures which is 907passed as a parameter to the ``heap_alloc()`` function, along with the 908requested size, type, alignment and boundary parameters. 909 910The ``heap_alloc()`` function will scan the free_list of the heap, and attempt 911to find a free block suitable for storing data of the requested size, with the 912requested alignment and boundary constraints. 913 914When a suitable free element has been identified, the pointer to be returned 915to the user is calculated. 916The cache-line of memory immediately preceding this pointer is filled with a 917struct malloc_elem header. 918Because of alignment and boundary constraints, there could be free space at 919the start and/or end of the element, resulting in the following behavior: 920 921#. Check for trailing space. 922 If the trailing space is big enough, i.e. > 128 bytes, then the free element 923 is split. 924 If it is not, then we just ignore it (wasted space). 925 926#. Check for space at the start of the element. 927 If the space at the start is small, i.e. <=128 bytes, then a pad header is 928 used, and the remaining space is wasted. 929 If, however, the remaining space is greater, then the free element is split. 930 931The advantage of allocating the memory from the end of the existing element is 932that no adjustment of the free list needs to take place - the existing element 933on the free list just has its size value adjusted, and the next/previous elements 934have their "prev"/"next" pointers redirected to the newly created element. 935 936In case when there is not enough memory in the heap to satisfy allocation 937request, EAL will attempt to allocate more memory from the system (if supported) 938and, following successful allocation, will retry reserving the memory again. In 939a multiprocessing scenario, all primary and secondary processes will synchronize 940their memory maps to ensure that any valid pointer to DPDK memory is guaranteed 941to be valid at all times in all currently running processes. 942 943Failure to synchronize memory maps in one of the processes will cause allocation 944to fail, even though some of the processes may have allocated the memory 945successfully. The memory is not added to the malloc heap unless primary process 946has ensured that all other processes have mapped this memory successfully. 947 948Any successful allocation event will trigger a callback, for which user 949applications and other DPDK subsystems can register. Additionally, validation 950callbacks will be triggered before allocation if the newly allocated memory will 951exceed threshold set by the user, giving a chance to allow or deny allocation. 952 953.. note:: 954 955 Any allocation of new pages has to go through primary process. If the 956 primary process is not active, no memory will be allocated even if it was 957 theoretically possible to do so. This is because primary's process map acts 958 as an authority on what should or should not be mapped, while each secondary 959 process has its own, local memory map. Secondary processes do not update the 960 shared memory map, they only copy its contents to their local memory map. 961 962Freeing Memory 963^^^^^^^^^^^^^^ 964 965To free an area of memory, the pointer to the start of the data area is passed 966to the free function. 967The size of the ``malloc_elem`` structure is subtracted from this pointer to get 968the element header for the block. 969If this header is of type ``PAD`` then the pad length is further subtracted from 970the pointer to get the proper element header for the entire block. 971 972From this element header, we get pointers to the heap from which the block was 973allocated and to where it must be freed, as well as the pointer to the previous 974and next elements. These next and previous elements are then checked to see if 975they are also ``FREE`` and are immediately adjacent to the current one, and if 976so, they are merged with the current element. This means that we can never have 977two ``FREE`` memory blocks adjacent to one another, as they are always merged 978into a single block. 979 980If deallocating pages at runtime is supported, and the free element encloses 981one or more pages, those pages can be deallocated and be removed from the heap. 982If DPDK was started with command-line parameters for preallocating memory 983(``-m`` or ``--socket-mem``), then those pages that were allocated at startup 984will not be deallocated. 985 986Any successful deallocation event will trigger a callback, for which user 987applications and other DPDK subsystems can register. 988