1 .. SPDX-License-Identifier: BSD-3-Clause
2 Copyright(c) 2019 Intel Corporation.
4 .. include:: <isonum.txt>
6 IOAT Rawdev Driver for Intel\ |reg| QuickData Technology
7 ======================================================================
9 The ``ioat`` rawdev driver provides a poll-mode driver (PMD) for Intel\ |reg|
10 QuickData Technology, part of Intel\ |reg| I/O Acceleration Technology
12 <https://www.intel.com/content/www/us/en/wireless-network/accel-technology.html>`_.
13 This PMD, when used on supported hardware, allows data copies, for example,
14 cloning packet data, to be accelerated by that hardware rather than having to
15 be done by software, freeing up CPU cycles for other tasks.
18 ----------------------
20 On Linux, the presence of an Intel\ |reg| QuickData Technology hardware can
21 be detected by checking the output of the ``lspci`` command, where the
22 hardware will be often listed as "Crystal Beach DMA" or "CBDMA". For
23 example, on a system with Intel\ |reg| Xeon\ |reg| CPU E5-2699 v4 @2.20GHz,
26 .. code-block:: console
29 00:04.0 System peripheral: Intel Corporation Xeon E7 v4/Xeon E5 v4/Xeon E3 v4/Xeon D Crystal Beach DMA Channel 0 (rev 01)
30 00:04.1 System peripheral: Intel Corporation Xeon E7 v4/Xeon E5 v4/Xeon E3 v4/Xeon D Crystal Beach DMA Channel 1 (rev 01)
31 00:04.2 System peripheral: Intel Corporation Xeon E7 v4/Xeon E5 v4/Xeon E3 v4/Xeon D Crystal Beach DMA Channel 2 (rev 01)
32 00:04.3 System peripheral: Intel Corporation Xeon E7 v4/Xeon E5 v4/Xeon E3 v4/Xeon D Crystal Beach DMA Channel 3 (rev 01)
33 00:04.4 System peripheral: Intel Corporation Xeon E7 v4/Xeon E5 v4/Xeon E3 v4/Xeon D Crystal Beach DMA Channel 4 (rev 01)
34 00:04.5 System peripheral: Intel Corporation Xeon E7 v4/Xeon E5 v4/Xeon E3 v4/Xeon D Crystal Beach DMA Channel 5 (rev 01)
35 00:04.6 System peripheral: Intel Corporation Xeon E7 v4/Xeon E5 v4/Xeon E3 v4/Xeon D Crystal Beach DMA Channel 6 (rev 01)
36 00:04.7 System peripheral: Intel Corporation Xeon E7 v4/Xeon E5 v4/Xeon E3 v4/Xeon D Crystal Beach DMA Channel 7 (rev 01)
38 On a system with Intel\ |reg| Xeon\ |reg| Gold 6154 CPU @ 3.00GHz, lspci
41 .. code-block:: console
44 00:04.0 System peripheral: Intel Corporation Sky Lake-E CBDMA Registers (rev 04)
45 00:04.1 System peripheral: Intel Corporation Sky Lake-E CBDMA Registers (rev 04)
46 00:04.2 System peripheral: Intel Corporation Sky Lake-E CBDMA Registers (rev 04)
47 00:04.3 System peripheral: Intel Corporation Sky Lake-E CBDMA Registers (rev 04)
48 00:04.4 System peripheral: Intel Corporation Sky Lake-E CBDMA Registers (rev 04)
49 00:04.5 System peripheral: Intel Corporation Sky Lake-E CBDMA Registers (rev 04)
50 00:04.6 System peripheral: Intel Corporation Sky Lake-E CBDMA Registers (rev 04)
51 00:04.7 System peripheral: Intel Corporation Sky Lake-E CBDMA Registers (rev 04)
57 For builds done with ``make``, the driver compilation is enabled by the
58 ``CONFIG_RTE_LIBRTE_PMD_IOAT_RAWDEV`` build configuration option. This is
59 enabled by default in builds for x86 platforms, and disabled in other
62 For builds using ``meson`` and ``ninja``, the driver will be built when the
63 target platform is x86-based.
68 The Intel\ |reg| QuickData Technology HW devices will need to be bound to a
69 user-space IO driver for use. The script ``dpdk-devbind.py`` script
70 included with DPDK can be used to view the state of the devices and to bind
71 them to a suitable DPDK-supported kernel driver. When querying the status
72 of the devices, they will appear under the category of "Misc (rawdev)
73 devices", i.e. the command ``dpdk-devbind.py --status-dev misc`` can be
74 used to see the state of those devices alone.
76 Device Probing and Initialization
77 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
79 Once bound to a suitable kernel device driver, the HW devices will be found
80 as part of the PCI scan done at application initialization time. No vdev
81 parameters need to be passed to create or initialize the device.
83 Once probed successfully, the device will appear as a ``rawdev``, that is a
84 "raw device type" inside DPDK, and can be accessed using APIs from the
85 ``rte_rawdev`` library.
87 Using IOAT Rawdev Devices
88 --------------------------
90 To use the devices from an application, the rawdev API can be used, along
91 with definitions taken from the device-specific header file
92 ``rte_ioat_rawdev.h``. This header is needed to get the definition of
93 structure parameters used by some of the rawdev APIs for IOAT rawdev
94 devices, as well as providing key functions for using the device for memory
97 Getting Device Information
98 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
100 Basic information about each rawdev device can be queried using the
101 ``rte_rawdev_info_get()`` API. For most applications, this API will be
102 needed to verify that the rawdev in question is of the expected type. For
103 example, the following code snippet can be used to identify an IOAT
104 rawdev device for use by an application:
108 for (i = 0; i < count && !found; i++) {
109 struct rte_rawdev_info info = { .dev_private = NULL };
110 found = (rte_rawdev_info_get(i, &info) == 0 &&
111 strcmp(info.driver_name,
112 IOAT_PMD_RAWDEV_NAME_STR) == 0);
115 When calling the ``rte_rawdev_info_get()`` API for an IOAT rawdev device,
116 the ``dev_private`` field in the ``rte_rawdev_info`` struct should either
117 be NULL, or else be set to point to a structure of type
118 ``rte_ioat_rawdev_config``, in which case the size of the configured device
119 input ring will be returned in that structure.
122 ~~~~~~~~~~~~~~~~~~~~~
124 Configuring an IOAT rawdev device is done using the
125 ``rte_rawdev_configure()`` API, which takes the same structure parameters
126 as the, previously referenced, ``rte_rawdev_info_get()`` API. The main
127 difference is that, because the parameter is used as input rather than
128 output, the ``dev_private`` structure element cannot be NULL, and must
129 point to a valid ``rte_ioat_rawdev_config`` structure, containing the ring
130 size to be used by the device. The ring size must be a power of two,
133 The following code shows how the device is configured in
134 ``test_ioat_rawdev.c``:
138 #define IOAT_TEST_RINGSIZE 512
139 struct rte_ioat_rawdev_config p = { .ring_size = -1 };
140 struct rte_rawdev_info info = { .dev_private = &p };
144 p.ring_size = IOAT_TEST_RINGSIZE;
145 if (rte_rawdev_configure(dev_id, &info) != 0) {
146 printf("Error with rte_rawdev_configure()\n");
150 Once configured, the device can then be made ready for use by calling the
151 ``rte_rawdev_start()`` API.
153 Performing Data Copies
154 ~~~~~~~~~~~~~~~~~~~~~~~
156 To perform data copies using IOAT rawdev devices, the functions
157 ``rte_ioat_enqueue_copy()`` and ``rte_ioat_do_copies()`` should be used.
158 Once copies have been completed, the completion will be reported back when
159 the application calls ``rte_ioat_completed_copies()``.
161 The ``rte_ioat_enqueue_copy()`` function enqueues a single copy to the
162 device ring for copying at a later point. The parameters to that function
163 include the IOVA addresses of both the source and destination buffers,
164 as well as two "handles" to be returned to the user when the copy is
165 completed. These handles can be arbitrary values, but two are provided so
166 that the library can track handles for both source and destination on
167 behalf of the user, e.g. virtual addresses for the buffers, or mbuf
168 pointers if packet data is being copied.
170 While the ``rte_ioat_enqueue_copy()`` function enqueues a copy operation on
171 the device ring, the copy will not actually be performed until after the
172 application calls the ``rte_ioat_do_copies()`` function. This function
173 informs the device hardware of the elements enqueued on the ring, and the
174 device will begin to process them. It is expected that, for efficiency
175 reasons, a burst of operations will be enqueued to the device via multiple
176 enqueue calls between calls to the ``rte_ioat_do_copies()`` function.
178 The following code from ``test_ioat_rawdev.c`` demonstrates how to enqueue
179 a burst of copies to the device and start the hardware processing of them:
183 struct rte_mbuf *srcs[32], *dsts[32];
186 for (i = 0; i < RTE_DIM(srcs); i++) {
189 srcs[i] = rte_pktmbuf_alloc(pool);
190 dsts[i] = rte_pktmbuf_alloc(pool);
191 srcs[i]->data_len = srcs[i]->pkt_len = length;
192 dsts[i]->data_len = dsts[i]->pkt_len = length;
193 src_data = rte_pktmbuf_mtod(srcs[i], char *);
195 for (j = 0; j < length; j++)
196 src_data[j] = rand() & 0xFF;
198 if (rte_ioat_enqueue_copy(dev_id,
199 srcs[i]->buf_iova + srcs[i]->data_off,
200 dsts[i]->buf_iova + dsts[i]->data_off,
204 0 /* nofence */) != 1) {
205 printf("Error with rte_ioat_enqueue_copy for buffer %u\n",
210 rte_ioat_do_copies(dev_id);
212 To retrieve information about completed copies, the API
213 ``rte_ioat_completed_copies()`` should be used. This API will return to the
214 application a set of completion handles passed in when the relevant copies
217 The following code from ``test_ioat_rawdev.c`` shows the test code
218 retrieving information about the completed copies and validating the data
219 is correct before freeing the data buffers using the returned handles:
223 if (rte_ioat_completed_copies(dev_id, 64, (void *)completed_src,
224 (void *)completed_dst) != RTE_DIM(srcs)) {
225 printf("Error with rte_ioat_completed_copies\n");
228 for (i = 0; i < RTE_DIM(srcs); i++) {
229 char *src_data, *dst_data;
231 if (completed_src[i] != srcs[i]) {
232 printf("Error with source pointer %u\n", i);
235 if (completed_dst[i] != dsts[i]) {
236 printf("Error with dest pointer %u\n", i);
240 src_data = rte_pktmbuf_mtod(srcs[i], char *);
241 dst_data = rte_pktmbuf_mtod(dsts[i], char *);
242 for (j = 0; j < length; j++)
243 if (src_data[j] != dst_data[j]) {
244 printf("Error with copy of packet %u, byte %u\n",
248 rte_pktmbuf_free(srcs[i]);
249 rte_pktmbuf_free(dsts[i]);
253 Querying Device Statistics
254 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
256 The statistics from the IOAT rawdev device can be got via the xstats
257 functions in the ``rte_rawdev`` library, i.e.
258 ``rte_rawdev_xstats_names_get()``, ``rte_rawdev_xstats_get()`` and
259 ``rte_rawdev_xstats_by_name_get``. The statistics returned for each device
262 * ``failed_enqueues``
263 * ``successful_enqueues``
265 * ``copies_completed``