2 Copyright(c) 2010-2015 Intel Corporation. All rights reserved.
5 Redistribution and use in source and binary forms, with or without
6 modification, are permitted provided that the following conditions
9 * Redistributions of source code must retain the above copyright
10 notice, this list of conditions and the following disclaimer.
11 * Redistributions in binary form must reproduce the above copyright
12 notice, this list of conditions and the following disclaimer in
13 the documentation and/or other materials provided with the
15 * Neither the name of Intel Corporation nor the names of its
16 contributors may be used to endorse or promote products derived
17 from this software without specific prior written permission.
19 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
20 "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
21 LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
22 A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
23 OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
24 SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
25 LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
26 DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
27 THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
28 (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
29 OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
36 The DPDK includes 1 Gigabit, 10 Gigabit and 40 Gigabit and para virtualized virtio Poll Mode Drivers.
38 A Poll Mode Driver (PMD) consists of APIs, provided through the BSD driver running in user space,
39 to configure the devices and their respective queues.
40 In addition, a PMD accesses the RX and TX descriptors directly without any interrupts
41 (with the exception of Link Status Change interrupts) to quickly receive,
42 process and deliver packets in the user's application.
43 This section describes the requirements of the PMDs,
44 their global design principles and proposes a high-level architecture and a generic external API for the Ethernet PMDs.
46 Requirements and Assumptions
47 ----------------------------
49 The DPDK environment for packet processing applications allows for two models, run-to-completion and pipe-line:
51 * In the *run-to-completion* model, a specific port's RX descriptor ring is polled for packets through an API.
52 Packets are then processed on the same core and placed on a port's TX descriptor ring through an API for transmission.
54 * In the *pipe-line* model, one core polls one or more port's RX descriptor ring through an API.
55 Packets are received and passed to another core via a ring.
56 The other core continues to process the packet which then may be placed on a port's TX descriptor ring through an API for transmission.
58 In a synchronous run-to-completion model,
59 each logical core assigned to the DPDK executes a packet processing loop that includes the following steps:
61 * Retrieve input packets through the PMD receive API
63 * Process each received packet one at a time, up to its forwarding
65 * Send pending output packets through the PMD transmit API
67 Conversely, in an asynchronous pipe-line model, some logical cores may be dedicated to the retrieval of received packets and
68 other logical cores to the processing of previously received packets.
69 Received packets are exchanged between logical cores through rings.
70 The loop for packet retrieval includes the following steps:
72 * Retrieve input packets through the PMD receive API
74 * Provide received packets to processing lcores through packet queues
76 The loop for packet processing includes the following steps:
78 * Retrieve the received packet from the packet queue
80 * Process the received packet, up to its retransmission if forwarded
82 To avoid any unnecessary interrupt processing overhead, the execution environment must not use any asynchronous notification mechanisms.
83 Whenever needed and appropriate, asynchronous communication should be introduced as much as possible through the use of rings.
85 Avoiding lock contention is a key issue in a multi-core environment.
86 To address this issue, PMDs are designed to work with per-core private resources as much as possible.
87 For example, a PMD maintains a separate transmit queue per-core, per-port.
88 In the same way, every receive queue of a port is assigned to and polled by a single logical core (lcore).
90 To comply with Non-Uniform Memory Access (NUMA), memory management is designed to assign to each logical core
91 a private buffer pool in local memory to minimize remote memory access.
92 The configuration of packet buffer pools should take into account the underlying physical memory architecture in terms of DIMMS,
94 The application must ensure that appropriate parameters are given at memory pool creation time.
95 See :ref:`Mempool Library <Mempool_Library>`.
100 The API and architecture of the Ethernet* PMDs are designed with the following guidelines in mind.
102 PMDs must help global policy-oriented decisions to be enforced at the upper application level.
103 Conversely, NIC PMD functions should not impede the benefits expected by upper-level global policies,
104 or worse prevent such policies from being applied.
106 For instance, both the receive and transmit functions of a PMD have a maximum number of packets/descriptors to poll.
107 This allows a run-to-completion processing stack to statically fix or
108 to dynamically adapt its overall behavior through different global loop policies, such as:
110 * Receive, process immediately and transmit packets one at a time in a piecemeal fashion.
112 * Receive as many packets as possible, then process all received packets, transmitting them immediately.
114 * Receive a given maximum number of packets, process the received packets, accumulate them and finally send all accumulated packets to transmit.
116 To achieve optimal performance, overall software design choices and pure software optimization techniques must be considered and
117 balanced against available low-level hardware-based optimization features (CPU cache properties, bus speed, NIC PCI bandwidth, and so on).
118 The case of packet transmission is an example of this software/hardware tradeoff issue when optimizing burst-oriented network packet processing engines.
119 In the initial case, the PMD could export only an rte_eth_tx_one function to transmit one packet at a time on a given queue.
120 On top of that, one can easily build an rte_eth_tx_burst function that loops invoking the rte_eth_tx_one function to transmit several packets at a time.
121 However, an rte_eth_tx_burst function is effectively implemented by the PMD to minimize the driver-level transmit cost per packet through the following optimizations:
123 * Share among multiple packets the un-amortized cost of invoking the rte_eth_tx_one function.
125 * Enable the rte_eth_tx_burst function to take advantage of burst-oriented hardware features (prefetch data in cache, use of NIC head/tail registers)
126 to minimize the number of CPU cycles per packet, for example by avoiding unnecessary read memory accesses to ring transmit descriptors,
127 or by systematically using arrays of pointers that exactly fit cache line boundaries and sizes.
129 * Apply burst-oriented software optimization techniques to remove operations that would otherwise be unavoidable, such as ring index wrap back management.
131 Burst-oriented functions are also introduced via the API for services that are intensively used by the PMD.
132 This applies in particular to buffer allocators used to populate NIC rings, which provide functions to allocate/free several buffers at a time.
133 For example, an mbuf_multiple_alloc function returning an array of pointers to rte_mbuf buffers which speeds up the receive poll function of the PMD when
134 replenishing multiple descriptors of the receive ring.
136 Logical Cores, Memory and NIC Queues Relationships
137 --------------------------------------------------
139 The DPDK supports NUMA allowing for better performance when a processor's logical cores and interfaces utilize its local memory.
140 Therefore, mbuf allocation associated with local PCIe* interfaces should be allocated from memory pools created in the local memory.
141 The buffers should, if possible, remain on the local processor to obtain the best performance results and RX and TX buffer descriptors
142 should be populated with mbufs allocated from a mempool allocated from local memory.
144 The run-to-completion model also performs better if packet or data manipulation is in local memory instead of a remote processors memory.
145 This is also true for the pipe-line model provided all logical cores used are located on the same processor.
147 Multiple logical cores should never share receive or transmit queues for interfaces since this would require global locks and hinder performance.
149 Device Identification and Configuration
150 ---------------------------------------
152 Device Identification
153 ~~~~~~~~~~~~~~~~~~~~~
155 Each NIC port is uniquely designated by its (bus/bridge, device, function) PCI
156 identifiers assigned by the PCI probing/enumeration function executed at DPDK initialization.
157 Based on their PCI identifier, NIC ports are assigned two other identifiers:
159 * A port index used to designate the NIC port in all functions exported by the PMD API.
161 * A port name used to designate the port in console messages, for administration or debugging purposes.
162 For ease of use, the port name includes the port index.
167 The configuration of each NIC port includes the following operations:
169 * Allocate PCI resources
171 * Reset the hardware (issue a Global Reset) to a well-known default state
173 * Set up the PHY and the link
175 * Initialize statistics counters
177 The PMD API must also export functions to start/stop the all-multicast feature of a port and functions to set/unset the port in promiscuous mode.
179 Some hardware offload features must be individually configured at port initialization through specific configuration parameters.
180 This is the case for the Receive Side Scaling (RSS) and Data Center Bridging (DCB) features for example.
182 On-the-Fly Configuration
183 ~~~~~~~~~~~~~~~~~~~~~~~~
185 All device features that can be started or stopped "on the fly" (that is, without stopping the device) do not require the PMD API to export dedicated functions for this purpose.
187 All that is required is the mapping address of the device PCI registers to implement the configuration of these features in specific functions outside of the drivers.
190 the PMD API exports a function that provides all the information associated with a device that can be used to set up a given device feature outside of the driver.
191 This includes the PCI vendor identifier, the PCI device identifier, the mapping address of the PCI device registers, and the name of the driver.
193 The main advantage of this approach is that it gives complete freedom on the choice of the API used to configure, to start, and to stop such features.
195 As an example, refer to the configuration of the IEEE1588 feature for the Intel® 82576 Gigabit Ethernet Controller and
196 the Intel® 82599 10 Gigabit Ethernet Controller controllers in the testpmd application.
198 Other features such as the L3/L4 5-Tuple packet filtering feature of a port can be configured in the same way.
199 Ethernet* flow control (pause frame) can be configured on the individual port.
200 Refer to the testpmd source code for details.
201 Also, L4 (UDP/TCP/ SCTP) checksum offload by the NIC can be enabled for an individual packet as long as the packet mbuf is set up correctly. See `Hardware Offload`_ for details.
203 Configuration of Transmit and Receive Queues
204 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
206 Each transmit queue is independently configured with the following information:
208 * The number of descriptors of the transmit ring
210 * The socket identifier used to identify the appropriate DMA memory zone from which to allocate the transmit ring in NUMA architectures
212 * The values of the Prefetch, Host and Write-Back threshold registers of the transmit queue
214 * The *minimum* transmit packets to free threshold (tx_free_thresh).
215 When the number of descriptors used to transmit packets exceeds this threshold, the network adaptor should be checked to see if it has written back descriptors.
216 A value of 0 can be passed during the TX queue configuration to indicate the default value should be used.
217 The default value for tx_free_thresh is 32.
218 This ensures that the PMD does not search for completed descriptors until at least 32 have been processed by the NIC for this queue.
220 * The *minimum* RS bit threshold. The minimum number of transmit descriptors to use before setting the Report Status (RS) bit in the transmit descriptor.
221 Note that this parameter may only be valid for Intel 10 GbE network adapters.
222 The RS bit is set on the last descriptor used to transmit a packet if the number of descriptors used since the last RS bit setting,
223 up to the first descriptor used to transmit the packet, exceeds the transmit RS bit threshold (tx_rs_thresh).
224 In short, this parameter controls which transmit descriptors are written back to host memory by the network adapter.
225 A value of 0 can be passed during the TX queue configuration to indicate that the default value should be used.
226 The default value for tx_rs_thresh is 32.
227 This ensures that at least 32 descriptors are used before the network adapter writes back the most recently used descriptor.
228 This saves upstream PCIe* bandwidth resulting from TX descriptor write-backs.
229 It is important to note that the TX Write-back threshold (TX wthresh) should be set to 0 when tx_rs_thresh is greater than 1.
230 Refer to the Intel® 82599 10 Gigabit Ethernet Controller Datasheet for more details.
232 The following constraints must be satisfied for tx_free_thresh and tx_rs_thresh:
234 * tx_rs_thresh must be greater than 0.
236 * tx_rs_thresh must be less than the size of the ring minus 2.
238 * tx_rs_thresh must be less than or equal to tx_free_thresh.
240 * tx_free_thresh must be greater than 0.
242 * tx_free_thresh must be less than the size of the ring minus 3.
244 * For optimal performance, TX wthresh should be set to 0 when tx_rs_thresh is greater than 1.
246 One descriptor in the TX ring is used as a sentinel to avoid a hardware race condition, hence the maximum threshold constraints.
250 When configuring for DCB operation, at port initialization, both the number of transmit queues and the number of receive queues must be set to 128.
255 Depending on driver capabilities advertised by
256 ``rte_eth_dev_info_get()``, the PMD may support hardware offloading
257 feature like checksumming, TCP segmentation or VLAN insertion.
259 The support of these offload features implies the addition of dedicated
260 status bit(s) and value field(s) into the rte_mbuf data structure, along
261 with their appropriate handling by the receive/transmit functions
262 exported by each PMD. The list of flags and their precise meaning is
263 described in the mbuf API documentation and in the in :ref:`Mbuf Library
264 <Mbuf_Library>`, section "Meta Information".
272 By default, all functions exported by a PMD are lock-free functions that are assumed
273 not to be invoked in parallel on different logical cores to work on the same target object.
274 For instance, a PMD receive function cannot be invoked in parallel on two logical cores to poll the same RX queue of the same port.
275 Of course, this function can be invoked in parallel by different logical cores on different RX queues.
276 It is the responsibility of the upper-level application to enforce this rule.
278 If needed, parallel accesses by multiple logical cores to shared queues can be explicitly protected by dedicated inline lock-aware functions
279 built on top of their corresponding lock-free functions of the PMD API.
281 Generic Packet Representation
282 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
284 A packet is represented by an rte_mbuf structure, which is a generic metadata structure containing all necessary housekeeping information.
285 This includes fields and status bits corresponding to offload hardware features, such as checksum computation of IP headers or VLAN tags.
287 The rte_mbuf data structure includes specific fields to represent, in a generic way, the offload features provided by network controllers.
288 For an input packet, most fields of the rte_mbuf structure are filled in by the PMD receive function with the information contained in the receive descriptor.
289 Conversely, for output packets, most fields of rte_mbuf structures are used by the PMD transmit function to initialize transmit descriptors.
291 The mbuf structure is fully described in the :ref:`Mbuf Library <Mbuf_Library>` chapter.
296 The Ethernet device API exported by the Ethernet PMDs is described in the *DPDK API Reference*.
298 Extended Statistics API
299 ~~~~~~~~~~~~~~~~~~~~~~~
301 The extended statistics API allows each individual PMD to expose a unique set
302 of statistics. Accessing these from application programs is done via two
305 * ``rte_eth_xstats_get``: Fills in an array of ``struct rte_eth_xstat``
306 with extended statistics.
307 * ``rte_eth_xstats_get_names``: Fills in an array of
308 ``struct rte_eth_xstat_name`` with extended statistic name lookup
311 Each ``struct rte_eth_xstat`` contains an identifier and value pair, and
312 each ``struct rte_eth_xstat_name`` contains an identifier and string pair.
313 Each identifier within ``struct rte_eth_xstat`` must have a corresponding
314 entry in ``struct rte_eth_xstat_name`` with a matching identifier. These
315 identifiers, as well as the number of extended statistic exposed, must
316 remain constant during runtime.
318 Note that extended statistic identifiers are driver-specific, and hence
319 might not be the same for different ports. Although it is expected that
320 drivers will make the identifiers used within ``struct rte_eth_xstat`` and
321 ``struct rte_eth_xstat_name`` entries match the entries' array index, this
322 property should not be relied on by applications for lookups.
324 A naming scheme exists for the strings exposed to clients of the API. This is
325 to allow scraping of the API for statistics of interest. The naming scheme uses
326 strings split by a single underscore ``_``. The scheme is as follows:
334 Examples of common statistics xstats strings, formatted to comply to the scheme
339 * ``tx_multicast_packets``
341 The scheme, although quite simple, allows flexibility in presenting and reading
342 information from the statistic strings. The following example illustrates the
343 naming scheme:``rx_packets``. In this example, the string is split into two
344 components. The first component ``rx`` indicates that the statistic is
345 associated with the receive side of the NIC. The second component ``packets``
346 indicates that the unit of measure is packets.
348 A more complicated example: ``tx_size_128_to_255_packets``. In this example,
349 ``tx`` indicates transmission, ``size`` is the first detail, ``128`` etc are
350 more details, and ``packets`` indicates that this is a packet counter.
352 Some additions in the metadata scheme are as follows:
354 * If the first part does not match ``rx`` or ``tx``, the statistic does not
355 have an affinity with either receive of transmit.
357 * If the first letter of the second part is ``q`` and this ``q`` is followed
358 by a number, this statistic is part of a specific queue.
360 An example where queue numbers are used is as follows: ``tx_q7_bytes`` which
361 indicates this statistic applies to queue number 7, and represents the number
362 of transmitted bytes on that queue.
git.droids-corp.org / dpdk.git / blob