1 =================================================
2 FPGA Device Feature List (DFL) Framework Overview
3 =================================================
7 - Enno Luebbers <enno.luebbers@intel.com>
8 - Xiao Guangrong <guangrong.xiao@linux.intel.com>
9 - Wu Hao <hao.wu@intel.com>
11 The Device Feature List (DFL) FPGA framework (and drivers according to this
12 this framework) hides the very details of low layer hardwares and provides
13 unified interfaces to userspace. Applications could use these interfaces to
14 configure, enumerate, open and access FPGA accelerators on platforms which
15 implement the DFL in the device memory. Besides this, the DFL framework
16 enables system level management functions such as FPGA reconfiguration.
19 Device Feature List (DFL) Overview
20 ==================================
21 Device Feature List (DFL) defines a linked list of feature headers within the
22 device MMIO space to provide an extensible way of adding features. Software can
23 walk through these predefined data structures to enumerate FPGA features:
24 FPGA Interface Unit (FIU), Accelerated Function Unit (AFU) and Private Features,
25 as illustrated below::
27 Header Header Header Header
28 +----------+ +-->+----------+ +-->+----------+ +-->+----------+
29 | Type | | | Type | | | Type | | | Type |
30 | FIU | | | Private | | | Private | | | Private |
31 +----------+ | | Feature | | | Feature | | | Feature |
32 | Next_DFH |--+ +----------+ | +----------+ | +----------+
33 +----------+ | Next_DFH |--+ | Next_DFH |--+ | Next_DFH |--> NULL
34 | ID | +----------+ +----------+ +----------+
35 +----------+ | ID | | ID | | ID |
36 | Next_AFU |--+ +----------+ +----------+ +----------+
37 +----------+ | | Feature | | Feature | | Feature |
38 | Header | | | Register | | Register | | Register |
39 | Register | | | Set | | Set | | Set |
40 | Set | | +----------+ +----------+ +----------+
55 FPGA Interface Unit (FIU) represents a standalone functional unit for the
56 interface to FPGA, e.g. the FPGA Management Engine (FME) and Port (more
57 descriptions on FME and Port in later sections).
59 Accelerated Function Unit (AFU) represents a FPGA programmable region and
60 always connects to a FIU (e.g. a Port) as its child as illustrated above.
62 Private Features represent sub features of the FIU and AFU. They could be
63 various function blocks with different IDs, but all private features which
64 belong to the same FIU or AFU, must be linked to one list via the Next Device
65 Feature Header (Next_DFH) pointer.
67 Each FIU, AFU and Private Feature could implement its own functional registers.
68 The functional register set for FIU and AFU, is named as Header Register Set,
69 e.g. FME Header Register Set, and the one for Private Feature, is named as
70 Feature Register Set, e.g. FME Partial Reconfiguration Feature Register Set.
72 This Device Feature List provides a way of linking features together, it's
73 convenient for software to locate each feature by walking through this list,
74 and can be implemented in register regions of any FPGA device.
77 FIU - FME (FPGA Management Engine)
78 ==================================
79 The FPGA Management Engine performs reconfiguration and other infrastructure
80 functions. Each FPGA device only has one FME.
82 User-space applications can acquire exclusive access to the FME using open(),
83 and release it using close().
85 The following functions are exposed through ioctls:
87 - Get driver API version (DFL_FPGA_GET_API_VERSION)
88 - Check for extensions (DFL_FPGA_CHECK_EXTENSION)
89 - Program bitstream (DFL_FPGA_FME_PORT_PR)
90 - Assign port to PF (DFL_FPGA_FME_PORT_ASSIGN)
91 - Release port from PF (DFL_FPGA_FME_PORT_RELEASE)
93 More functions are exposed through sysfs
94 (/sys/class/fpga_region/regionX/dfl-fme.n/):
96 Read bitstream ID (bitstream_id)
97 bitstream_id indicates version of the static FPGA region.
99 Read bitstream metadata (bitstream_metadata)
100 bitstream_metadata includes detailed information of static FPGA region,
101 e.g. synthesis date and seed.
103 Read number of ports (ports_num)
104 one FPGA device may have more than one port, this sysfs interface indicates
105 how many ports the FPGA device has.
107 Global error reporting management (errors/)
108 error reporting sysfs interfaces allow user to read errors detected by the
109 hardware, and clear the logged errors.
111 Power management (dfl_fme_power hwmon)
112 power management hwmon sysfs interfaces allow user to read power management
113 information (power consumption, thresholds, threshold status, limits, etc.)
114 and configure power thresholds for different throttling levels.
116 Thermal management (dfl_fme_thermal hwmon)
117 thermal management hwmon sysfs interfaces allow user to read thermal
118 management information (current temperature, thresholds, threshold status,
121 Performance reporting
122 performance counters are exposed through perf PMU APIs. Standard perf tool
123 can be used to monitor all available perf events. Please see performance
124 counter section below for more detailed information.
129 A port represents the interface between the static FPGA fabric and a partially
130 reconfigurable region containing an AFU. It controls the communication from SW
131 to the accelerator and exposes features such as reset and debug. Each FPGA
132 device may have more than one port, but always one AFU per port.
137 An AFU is attached to a port FIU and exposes a fixed length MMIO region to be
138 used for accelerator-specific control registers.
140 User-space applications can acquire exclusive access to an AFU attached to a
141 port by using open() on the port device node and release it using close().
143 The following functions are exposed through ioctls:
145 - Get driver API version (DFL_FPGA_GET_API_VERSION)
146 - Check for extensions (DFL_FPGA_CHECK_EXTENSION)
147 - Get port info (DFL_FPGA_PORT_GET_INFO)
148 - Get MMIO region info (DFL_FPGA_PORT_GET_REGION_INFO)
149 - Map DMA buffer (DFL_FPGA_PORT_DMA_MAP)
150 - Unmap DMA buffer (DFL_FPGA_PORT_DMA_UNMAP)
151 - Reset AFU (DFL_FPGA_PORT_RESET)
154 reset the FPGA Port and its AFU. Userspace can do Port
155 reset at any time, e.g. during DMA or Partial Reconfiguration. But it should
156 never cause any system level issue, only functional failure (e.g. DMA or PR
157 operation failure) and be recoverable from the failure.
159 User-space applications can also mmap() accelerator MMIO regions.
161 More functions are exposed through sysfs:
162 (/sys/class/fpga_region/<regionX>/<dfl-port.m>/):
164 Read Accelerator GUID (afu_id)
165 afu_id indicates which PR bitstream is programmed to this AFU.
167 Error reporting (errors/)
168 error reporting sysfs interfaces allow user to read port/afu errors
169 detected by the hardware, and clear the logged errors.
172 DFL Framework Overview
173 ======================
177 +----------+ +--------+ +--------+ +--------+
178 | FME | | AFU | | AFU | | AFU |
179 | Module | | Module | | Module | | Module |
180 +----------+ +--------+ +--------+ +--------+
181 +-----------------------+
182 | FPGA Container Device | Device Feature List
183 | (FPGA Base Region) | Framework
184 +-----------------------+
185 ------------------------------------------------------------------
186 +----------------------------+
187 | FPGA DFL Device Module |
188 | (e.g. PCIE/Platform Device)|
189 +----------------------------+
190 +------------------------+
191 | FPGA Hardware Device |
192 +------------------------+
194 DFL framework in kernel provides common interfaces to create container device
195 (FPGA base region), discover feature devices and their private features from the
196 given Device Feature Lists and create platform devices for feature devices
197 (e.g. FME, Port and AFU) with related resources under the container device. It
198 also abstracts operations for the private features and exposes common ops to
199 feature device drivers.
201 The FPGA DFL Device could be different hardwares, e.g. PCIe device, platform
202 device and etc. Its driver module is always loaded first once the device is
203 created by the system. This driver plays an infrastructural role in the
204 driver architecture. It locates the DFLs in the device memory, handles them
205 and related resources to common interfaces from DFL framework for enumeration.
206 (Please refer to drivers/fpga/dfl.c for detailed enumeration APIs).
208 The FPGA Management Engine (FME) driver is a platform driver which is loaded
209 automatically after FME platform device creation from the DFL device module. It
210 provides the key features for FPGA management, including:
212 a) Expose static FPGA region information, e.g. version and metadata.
213 Users can read related information via sysfs interfaces exposed
216 b) Partial Reconfiguration. The FME driver creates FPGA manager, FPGA
217 bridges and FPGA regions during PR sub feature initialization. Once
218 it receives a DFL_FPGA_FME_PORT_PR ioctl from user, it invokes the
219 common interface function from FPGA Region to complete the partial
220 reconfiguration of the PR bitstream to the given port.
222 Similar to the FME driver, the FPGA Accelerated Function Unit (AFU) driver is
223 probed once the AFU platform device is created. The main function of this module
224 is to provide an interface for userspace applications to access the individual
225 accelerators, including basic reset control on port, AFU MMIO region export, dma
226 buffer mapping service functions.
228 After feature platform devices creation, matched platform drivers will be loaded
229 automatically to handle different functionalities. Please refer to next sections
230 for detailed information on functional units which have been already implemented
231 under this DFL framework.
234 Partial Reconfiguration
235 =======================
236 As mentioned above, accelerators can be reconfigured through partial
237 reconfiguration of a PR bitstream file. The PR bitstream file must have been
238 generated for the exact static FPGA region and targeted reconfigurable region
239 (port) of the FPGA, otherwise, the reconfiguration operation will fail and
240 possibly cause system instability. This compatibility can be checked by
241 comparing the compatibility ID noted in the header of PR bitstream file against
242 the compat_id exposed by the target FPGA region. This check is usually done by
243 userspace before calling the reconfiguration IOCTL.
246 FPGA virtualization - PCIe SRIOV
247 ================================
248 This section describes the virtualization support on DFL based FPGA device to
249 enable accessing an accelerator from applications running in a virtual machine
250 (VM). This section only describes the PCIe based FPGA device with SRIOV support.
252 Features supported by the particular FPGA device are exposed through Device
253 Feature Lists, as illustrated below:
257 +-------------------------------+ +-------------+
259 +-------------------------------+ +-------------+
262 +-----|------------|---------|--------------|-------+
264 | +-----+ +-------+ +-------+ +-------+ |
265 | | FME | | Port0 | | Port1 | | Port2 | |
266 | +-----+ +-------+ +-------+ +-------+ |
269 | +-------+ +------+ +-------+ |
270 | | AFU | | AFU | | AFU | |
271 | +-------+ +------+ +-------+ |
273 | DFL based FPGA PCIe Device |
274 +---------------------------------------------------+
276 FME is always accessed through the physical function (PF).
278 Ports (and related AFUs) are accessed via PF by default, but could be exposed
279 through virtual function (VF) devices via PCIe SRIOV. Each VF only contains
280 1 Port and 1 AFU for isolation. Users could assign individual VFs (accelerators)
281 created via PCIe SRIOV interface, to virtual machines.
283 The driver organization in virtualization case is illustrated below:
286 +-------++------++------+ |
287 | FME || FME || FME | |
288 | FPGA || FPGA || FPGA | |
289 |Manager||Bridge||Region| |
290 +-------++------++------+ |
291 +-----------------------+ +--------+ | +--------+
292 | FME | | AFU | | | AFU |
293 | Module | | Module | | | Module |
294 +-----------------------+ +--------+ | +--------+
295 +-----------------------+ | +-----------------------+
296 | FPGA Container Device | | | FPGA Container Device |
297 | (FPGA Base Region) | | | (FPGA Base Region) |
298 +-----------------------+ | +-----------------------+
299 +------------------+ | +------------------+
300 | FPGA PCIE Module | | Virtual | FPGA PCIE Module |
301 +------------------+ Host | Machine +------------------+
302 -------------------------------------- | ------------------------------
303 +---------------+ | +---------------+
304 | PCI PF Device | | | PCI VF Device |
305 +---------------+ | +---------------+
307 FPGA PCIe device driver is always loaded first once a FPGA PCIe PF or VF device
310 * Finishes enumeration on both FPGA PCIe PF and VF device using common
311 interfaces from DFL framework.
314 The FME device driver plays a management role in this driver architecture, it
315 provides ioctls to release Port from PF and assign Port to PF. After release
316 a port from PF, then it's safe to expose this port through a VF via PCIe SRIOV
319 To enable accessing an accelerator from applications running in a VM, the
320 respective AFU's port needs to be assigned to a VF using the following steps:
322 #. The PF owns all AFU ports by default. Any port that needs to be
323 reassigned to a VF must first be released through the
324 DFL_FPGA_FME_PORT_RELEASE ioctl on the FME device.
326 #. Once N ports are released from PF, then user can use command below
327 to enable SRIOV and VFs. Each VF owns only one Port with AFU.
331 echo N > $PCI_DEVICE_PATH/sriov_numvfs
333 #. Pass through the VFs to VMs
335 #. The AFU under VF is accessible from applications in VM (using the
336 same driver inside the VF).
338 Note that an FME can't be assigned to a VF, thus PR and other management
339 functions are only available via the PF.
343 This section introduces how applications enumerate the fpga device from
344 the sysfs hierarchy under /sys/class/fpga_region.
346 In the example below, two DFL based FPGA devices are installed in the host. Each
347 fpga device has one FME and two ports (AFUs).
349 FPGA regions are created under /sys/class/fpga_region/::
351 /sys/class/fpga_region/region0
352 /sys/class/fpga_region/region1
353 /sys/class/fpga_region/region2
356 Application needs to search each regionX folder, if feature device is found,
357 (e.g. "dfl-port.n" or "dfl-fme.m" is found), then it's the base
358 fpga region which represents the FPGA device.
360 Each base region has one FME and two ports (AFUs) as child devices::
362 /sys/class/fpga_region/region0/dfl-fme.0
363 /sys/class/fpga_region/region0/dfl-port.0
364 /sys/class/fpga_region/region0/dfl-port.1
367 /sys/class/fpga_region/region3/dfl-fme.1
368 /sys/class/fpga_region/region3/dfl-port.2
369 /sys/class/fpga_region/region3/dfl-port.3
372 In general, the FME/AFU sysfs interfaces are named as follows::
374 /sys/class/fpga_region/<regionX>/<dfl-fme.n>/
375 /sys/class/fpga_region/<regionX>/<dfl-port.m>/
377 with 'n' consecutively numbering all FMEs and 'm' consecutively numbering all
380 The device nodes used for ioctl() or mmap() can be referenced through::
382 /sys/class/fpga_region/<regionX>/<dfl-fme.n>/dev
383 /sys/class/fpga_region/<regionX>/<dfl-port.n>/dev
388 Performance reporting is one private feature implemented in FME. It could
389 supports several independent, system-wide, device counter sets in hardware to
390 monitor and count for performance events, including "basic", "cache", "fabric",
391 "vtd" and "vtd_sip" counters. Users could use standard perf tool to monitor
392 FPGA cache hit/miss rate, transaction number, interface clock counter of AFU
393 and other FPGA performance events.
395 Different FPGA devices may have different counter sets, depending on hardware
396 implementation. E.g., some discrete FPGA cards don't have any cache. User could
397 use "perf list" to check which perf events are supported by target hardware.
399 In order to allow user to use standard perf API to access these performance
400 counters, driver creates a perf PMU, and related sysfs interfaces in
401 /sys/bus/event_source/devices/dfl_fme* to describe available perf events and
402 configuration options.
404 The "format" directory describes the format of the config field of struct
405 perf_event_attr. There are 3 bitfields for config: "evtype" defines which type
406 the perf event belongs to; "event" is the identity of the event within its
407 category; "portid" is introduced to decide counters set to monitor on FPGA
408 overall data or a specific port.
410 The "events" directory describes the configuration templates for all available
411 events which can be used with perf tool directly. For example, fab_mmio_read
412 has the configuration "event=0x06,evtype=0x02,portid=0xff", which shows this
413 event belongs to fabric type (0x02), the local event id is 0x06 and it is for
414 overall monitoring (portid=0xff).
416 Example usage of perf::
418 $# perf list |grep dfl_fme
420 dfl_fme0/fab_mmio_read/ [Kernel PMU event]
422 dfl_fme0/fab_port_mmio_read,portid=?/ [Kernel PMU event]
425 $# perf stat -a -e dfl_fme0/fab_mmio_read/ <command>
427 $# perf stat -a -e dfl_fme0/event=0x06,evtype=0x02,portid=0xff/ <command>
429 $# perf stat -a -e dfl_fme0/config=0xff2006/ <command>
431 Another example, fab_port_mmio_read monitors mmio read of a specific port. So
432 its configuration template is "event=0x06,evtype=0x01,portid=?". The portid
433 should be explicitly set.
437 $# perf stat -a -e dfl_fme0/fab_port_mmio_read,portid=0x0/ <command>
439 $# perf stat -a -e dfl_fme0/event=0x06,evtype=0x02,portid=0x0/ <command>
441 $# perf stat -a -e dfl_fme0/config=0x2006/ <command>
443 Please note for fabric counters, overall perf events (fab_*) and port perf
444 events (fab_port_*) actually share one set of counters in hardware, so it can't
445 monitor both at the same time. If this set of counters is configured to monitor
446 overall data, then per port perf data is not supported. See below example::
448 $# perf stat -e dfl_fme0/fab_mmio_read/,dfl_fme0/fab_port_mmio_write,\
451 Performance counter stats for 'system wide':
453 3 dfl_fme0/fab_mmio_read/
454 <not supported> dfl_fme0/fab_port_mmio_write,portid=0x0/
456 1.001750904 seconds time elapsed
458 The driver also provides a "cpumask" sysfs attribute, which contains only one
459 CPU id used to access these perf events. Counting on multiple CPU is not allowed
460 since they are system-wide counters on FPGA device.
462 The current driver does not support sampling. So "perf record" is unsupported.
467 It's possible that developers made some new function blocks (FIUs) under this
468 DFL framework, then new platform device driver needs to be developed for the
469 new feature dev (FIU) following the same way as existing feature dev drivers
470 (e.g. FME and Port/AFU platform device driver). Besides that, it requires
471 modification on DFL framework enumeration code too, for new FIU type detection
472 and related platform devices creation.
475 Add new private features support
476 ================================
477 In some cases, we may need to add some new private features to existing FIUs
478 (e.g. FME or Port). Developers don't need to touch enumeration code in DFL
479 framework, as each private feature will be parsed automatically and related
480 mmio resources can be found under FIU platform device created by DFL framework.
481 Developer only needs to provide a sub feature driver with matched feature id.
482 FME Partial Reconfiguration Sub Feature driver (see drivers/fpga/dfl-fme-pr.c)
483 could be a reference.
488 FME driver exports one ioctl (DFL_FPGA_FME_PORT_PR) for partial reconfiguration
489 to user now. In the future, if unified user interfaces for reconfiguration are
490 added, FME driver should switch to them from ioctl interface.