xref: /openbmc/linux/Documentation/fpga/dfl.rst (revision ca48739e) 
1=================================================
2FPGA Device Feature List (DFL) Framework Overview
3=================================================
4
5Authors:
6
7- Enno Luebbers <enno.luebbers@intel.com>
8- Xiao Guangrong <guangrong.xiao@linux.intel.com>
9- Wu Hao <hao.wu@intel.com>
10
11The Device Feature List (DFL) FPGA framework (and drivers according to
12this framework) hides the very details of low layer hardwares and provides
13unified interfaces to userspace. Applications could use these interfaces to
14configure, enumerate, open and access FPGA accelerators on platforms which
15implement the DFL in the device memory. Besides this, the DFL framework
16enables system level management functions such as FPGA reconfiguration.
17
18
19Device Feature List (DFL) Overview
20==================================
21Device Feature List (DFL) defines a linked list of feature headers within the
22device MMIO space to provide an extensible way of adding features. Software can
23walk through these predefined data structures to enumerate FPGA features:
24FPGA Interface Unit (FIU), Accelerated Function Unit (AFU) and Private Features,
25as illustrated below::
26
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    |  |   +----------+      +----------+      +----------+
41 +----------+  |      Header
42               +-->+----------+
43                   |   Type   |
44                   |   AFU    |
45                   +----------+
46                   | Next_DFH |--> NULL
47                   +----------+
48                   |   GUID   |
49                   +----------+
50                   |  Header  |
51                   | Register |
52                   |   Set    |
53                   +----------+
54
55FPGA Interface Unit (FIU) represents a standalone functional unit for the
56interface to FPGA, e.g. the FPGA Management Engine (FME) and Port (more
57descriptions on FME and Port in later sections).
58
59Accelerated Function Unit (AFU) represents a FPGA programmable region and
60always connects to a FIU (e.g. a Port) as its child as illustrated above.
61
62Private Features represent sub features of the FIU and AFU. They could be
63various function blocks with different IDs, but all private features which
64belong to the same FIU or AFU, must be linked to one list via the Next Device
65Feature Header (Next_DFH) pointer.
66
67Each FIU, AFU and Private Feature could implement its own functional registers.
68The functional register set for FIU and AFU, is named as Header Register Set,
69e.g. FME Header Register Set, and the one for Private Feature, is named as
70Feature Register Set, e.g. FME Partial Reconfiguration Feature Register Set.
71
72This Device Feature List provides a way of linking features together, it's
73convenient for software to locate each feature by walking through this list,
74and can be implemented in register regions of any FPGA device.
75
76
77FIU - FME (FPGA Management Engine)
78==================================
79The FPGA Management Engine performs reconfiguration and other infrastructure
80functions. Each FPGA device only has one FME.
81
82User-space applications can acquire exclusive access to the FME using open(),
83and release it using close().
84
85The following functions are exposed through ioctls:
86
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)
92- Get number of irqs of FME global error (DFL_FPGA_FME_ERR_GET_IRQ_NUM)
93- Set interrupt trigger for FME error (DFL_FPGA_FME_ERR_SET_IRQ)
94
95More functions are exposed through sysfs
96(/sys/class/fpga_region/regionX/dfl-fme.n/):
97
98 Read bitstream ID (bitstream_id)
99     bitstream_id indicates version of the static FPGA region.
100
101 Read bitstream metadata (bitstream_metadata)
102     bitstream_metadata includes detailed information of static FPGA region,
103     e.g. synthesis date and seed.
104
105 Read number of ports (ports_num)
106     one FPGA device may have more than one port, this sysfs interface indicates
107     how many ports the FPGA device has.
108
109 Global error reporting management (errors/)
110     error reporting sysfs interfaces allow user to read errors detected by the
111     hardware, and clear the logged errors.
112
113 Power management (dfl_fme_power hwmon)
114     power management hwmon sysfs interfaces allow user to read power management
115     information (power consumption, thresholds, threshold status, limits, etc.)
116     and configure power thresholds for different throttling levels.
117
118 Thermal management (dfl_fme_thermal hwmon)
119     thermal management hwmon sysfs interfaces allow user to read thermal
120     management information (current temperature, thresholds, threshold status,
121     etc.).
122
123 Performance reporting
124     performance counters are exposed through perf PMU APIs. Standard perf tool
125     can be used to monitor all available perf events. Please see performance
126     counter section below for more detailed information.
127
128
129FIU - PORT
130==========
131A port represents the interface between the static FPGA fabric and a partially
132reconfigurable region containing an AFU. It controls the communication from SW
133to the accelerator and exposes features such as reset and debug. Each FPGA
134device may have more than one port, but always one AFU per port.
135
136
137AFU
138===
139An AFU is attached to a port FIU and exposes a fixed length MMIO region to be
140used for accelerator-specific control registers.
141
142User-space applications can acquire exclusive access to an AFU attached to a
143port by using open() on the port device node and release it using close().
144
145The following functions are exposed through ioctls:
146
147- Get driver API version (DFL_FPGA_GET_API_VERSION)
148- Check for extensions (DFL_FPGA_CHECK_EXTENSION)
149- Get port info (DFL_FPGA_PORT_GET_INFO)
150- Get MMIO region info (DFL_FPGA_PORT_GET_REGION_INFO)
151- Map DMA buffer (DFL_FPGA_PORT_DMA_MAP)
152- Unmap DMA buffer (DFL_FPGA_PORT_DMA_UNMAP)
153- Reset AFU (DFL_FPGA_PORT_RESET)
154- Get number of irqs of port error (DFL_FPGA_PORT_ERR_GET_IRQ_NUM)
155- Set interrupt trigger for port error (DFL_FPGA_PORT_ERR_SET_IRQ)
156- Get number of irqs of UINT (DFL_FPGA_PORT_UINT_GET_IRQ_NUM)
157- Set interrupt trigger for UINT (DFL_FPGA_PORT_UINT_SET_IRQ)
158
159DFL_FPGA_PORT_RESET:
160  reset the FPGA Port and its AFU. Userspace can do Port
161  reset at any time, e.g. during DMA or Partial Reconfiguration. But it should
162  never cause any system level issue, only functional failure (e.g. DMA or PR
163  operation failure) and be recoverable from the failure.
164
165User-space applications can also mmap() accelerator MMIO regions.
166
167More functions are exposed through sysfs:
168(/sys/class/fpga_region/<regionX>/<dfl-port.m>/):
169
170 Read Accelerator GUID (afu_id)
171     afu_id indicates which PR bitstream is programmed to this AFU.
172
173 Error reporting (errors/)
174     error reporting sysfs interfaces allow user to read port/afu errors
175     detected by the hardware, and clear the logged errors.
176
177
178DFL Framework Overview
179======================
180
181::
182
183         +----------+    +--------+ +--------+ +--------+
184         |   FME    |    |  AFU   | |  AFU   | |  AFU   |
185         |  Module  |    | Module | | Module | | Module |
186         +----------+    +--------+ +--------+ +--------+
187                 +-----------------------+
188                 | FPGA Container Device |    Device Feature List
189                 |  (FPGA Base Region)   |         Framework
190                 +-----------------------+
191  ------------------------------------------------------------------
192               +----------------------------+
193               |   FPGA DFL Device Module   |
194               | (e.g. PCIE/Platform Device)|
195               +----------------------------+
196                 +------------------------+
197                 |  FPGA Hardware Device  |
198                 +------------------------+
199
200DFL framework in kernel provides common interfaces to create container device
201(FPGA base region), discover feature devices and their private features from the
202given Device Feature Lists and create platform devices for feature devices
203(e.g. FME, Port and AFU) with related resources under the container device. It
204also abstracts operations for the private features and exposes common ops to
205feature device drivers.
206
207The FPGA DFL Device could be different hardwares, e.g. PCIe device, platform
208device and etc. Its driver module is always loaded first once the device is
209created by the system. This driver plays an infrastructural role in the
210driver architecture. It locates the DFLs in the device memory, handles them
211and related resources to common interfaces from DFL framework for enumeration.
212(Please refer to drivers/fpga/dfl.c for detailed enumeration APIs).
213
214The FPGA Management Engine (FME) driver is a platform driver which is loaded
215automatically after FME platform device creation from the DFL device module. It
216provides the key features for FPGA management, including:
217
218	a) Expose static FPGA region information, e.g. version and metadata.
219	   Users can read related information via sysfs interfaces exposed
220	   by FME driver.
221
222	b) Partial Reconfiguration. The FME driver creates FPGA manager, FPGA
223	   bridges and FPGA regions during PR sub feature initialization. Once
224	   it receives a DFL_FPGA_FME_PORT_PR ioctl from user, it invokes the
225	   common interface function from FPGA Region to complete the partial
226	   reconfiguration of the PR bitstream to the given port.
227
228Similar to the FME driver, the FPGA Accelerated Function Unit (AFU) driver is
229probed once the AFU platform device is created. The main function of this module
230is to provide an interface for userspace applications to access the individual
231accelerators, including basic reset control on port, AFU MMIO region export, dma
232buffer mapping service functions.
233
234After feature platform devices creation, matched platform drivers will be loaded
235automatically to handle different functionalities. Please refer to next sections
236for detailed information on functional units which have been already implemented
237under this DFL framework.
238
239
240Partial Reconfiguration
241=======================
242As mentioned above, accelerators can be reconfigured through partial
243reconfiguration of a PR bitstream file. The PR bitstream file must have been
244generated for the exact static FPGA region and targeted reconfigurable region
245(port) of the FPGA, otherwise, the reconfiguration operation will fail and
246possibly cause system instability. This compatibility can be checked by
247comparing the compatibility ID noted in the header of PR bitstream file against
248the compat_id exposed by the target FPGA region. This check is usually done by
249userspace before calling the reconfiguration IOCTL.
250
251
252FPGA virtualization - PCIe SRIOV
253================================
254This section describes the virtualization support on DFL based FPGA device to
255enable accessing an accelerator from applications running in a virtual machine
256(VM). This section only describes the PCIe based FPGA device with SRIOV support.
257
258Features supported by the particular FPGA device are exposed through Device
259Feature Lists, as illustrated below:
260
261::
262
263    +-------------------------------+  +-------------+
264    |              PF               |  |     VF      |
265    +-------------------------------+  +-------------+
266        ^            ^         ^              ^
267        |            |         |              |
268  +-----|------------|---------|--------------|-------+
269  |     |            |         |              |       |
270  |  +-----+     +-------+ +-------+      +-------+   |
271  |  | FME |     | Port0 | | Port1 |      | Port2 |   |
272  |  +-----+     +-------+ +-------+      +-------+   |
273  |                  ^         ^              ^       |
274  |                  |         |              |       |
275  |              +-------+ +------+       +-------+   |
276  |              |  AFU  | |  AFU |       |  AFU  |   |
277  |              +-------+ +------+       +-------+   |
278  |                                                   |
279  |            DFL based FPGA PCIe Device             |
280  +---------------------------------------------------+
281
282FME is always accessed through the physical function (PF).
283
284Ports (and related AFUs) are accessed via PF by default, but could be exposed
285through virtual function (VF) devices via PCIe SRIOV. Each VF only contains
2861 Port and 1 AFU for isolation. Users could assign individual VFs (accelerators)
287created via PCIe SRIOV interface, to virtual machines.
288
289The driver organization in virtualization case is illustrated below:
290::
291
292    +-------++------++------+             |
293    | FME   || FME  || FME  |             |
294    | FPGA  || FPGA || FPGA |             |
295    |Manager||Bridge||Region|             |
296    +-------++------++------+             |
297    +-----------------------+  +--------+ |             +--------+
298    |          FME          |  |  AFU   | |             |  AFU   |
299    |         Module        |  | Module | |             | Module |
300    +-----------------------+  +--------+ |             +--------+
301          +-----------------------+       |       +-----------------------+
302          | FPGA Container Device |       |       | FPGA Container Device |
303          |  (FPGA Base Region)   |       |       |  (FPGA Base Region)   |
304          +-----------------------+       |       +-----------------------+
305            +------------------+          |         +------------------+
306            | FPGA PCIE Module |          | Virtual | FPGA PCIE Module |
307            +------------------+   Host   | Machine +------------------+
308   -------------------------------------- | ------------------------------
309             +---------------+            |          +---------------+
310             | PCI PF Device |            |          | PCI VF Device |
311             +---------------+            |          +---------------+
312
313FPGA PCIe device driver is always loaded first once a FPGA PCIe PF or VF device
314is detected. It:
315
316* Finishes enumeration on both FPGA PCIe PF and VF device using common
317  interfaces from DFL framework.
318* Supports SRIOV.
319
320The FME device driver plays a management role in this driver architecture, it
321provides ioctls to release Port from PF and assign Port to PF. After release
322a port from PF, then it's safe to expose this port through a VF via PCIe SRIOV
323sysfs interface.
324
325To enable accessing an accelerator from applications running in a VM, the
326respective AFU's port needs to be assigned to a VF using the following steps:
327
328#. The PF owns all AFU ports by default. Any port that needs to be
329   reassigned to a VF must first be released through the
330   DFL_FPGA_FME_PORT_RELEASE ioctl on the FME device.
331
332#. Once N ports are released from PF, then user can use command below
333   to enable SRIOV and VFs. Each VF owns only one Port with AFU.
334
335   ::
336
337      echo N > $PCI_DEVICE_PATH/sriov_numvfs
338
339#. Pass through the VFs to VMs
340
341#. The AFU under VF is accessible from applications in VM (using the
342   same driver inside the VF).
343
344Note that an FME can't be assigned to a VF, thus PR and other management
345functions are only available via the PF.
346
347Device enumeration
348==================
349This section introduces how applications enumerate the fpga device from
350the sysfs hierarchy under /sys/class/fpga_region.
351
352In the example below, two DFL based FPGA devices are installed in the host. Each
353fpga device has one FME and two ports (AFUs).
354
355FPGA regions are created under /sys/class/fpga_region/::
356
357	/sys/class/fpga_region/region0
358	/sys/class/fpga_region/region1
359	/sys/class/fpga_region/region2
360	...
361
362Application needs to search each regionX folder, if feature device is found,
363(e.g. "dfl-port.n" or "dfl-fme.m" is found), then it's the base
364fpga region which represents the FPGA device.
365
366Each base region has one FME and two ports (AFUs) as child devices::
367
368	/sys/class/fpga_region/region0/dfl-fme.0
369	/sys/class/fpga_region/region0/dfl-port.0
370	/sys/class/fpga_region/region0/dfl-port.1
371	...
372
373	/sys/class/fpga_region/region3/dfl-fme.1
374	/sys/class/fpga_region/region3/dfl-port.2
375	/sys/class/fpga_region/region3/dfl-port.3
376	...
377
378In general, the FME/AFU sysfs interfaces are named as follows::
379
380	/sys/class/fpga_region/<regionX>/<dfl-fme.n>/
381	/sys/class/fpga_region/<regionX>/<dfl-port.m>/
382
383with 'n' consecutively numbering all FMEs and 'm' consecutively numbering all
384ports.
385
386The device nodes used for ioctl() or mmap() can be referenced through::
387
388	/sys/class/fpga_region/<regionX>/<dfl-fme.n>/dev
389	/sys/class/fpga_region/<regionX>/<dfl-port.n>/dev
390
391
392Performance Counters
393====================
394Performance reporting is one private feature implemented in FME. It could
395supports several independent, system-wide, device counter sets in hardware to
396monitor and count for performance events, including "basic", "cache", "fabric",
397"vtd" and "vtd_sip" counters. Users could use standard perf tool to monitor
398FPGA cache hit/miss rate, transaction number, interface clock counter of AFU
399and other FPGA performance events.
400
401Different FPGA devices may have different counter sets, depending on hardware
402implementation. E.g., some discrete FPGA cards don't have any cache. User could
403use "perf list" to check which perf events are supported by target hardware.
404
405In order to allow user to use standard perf API to access these performance
406counters, driver creates a perf PMU, and related sysfs interfaces in
407/sys/bus/event_source/devices/dfl_fme* to describe available perf events and
408configuration options.
409
410The "format" directory describes the format of the config field of struct
411perf_event_attr. There are 3 bitfields for config: "evtype" defines which type
412the perf event belongs to; "event" is the identity of the event within its
413category; "portid" is introduced to decide counters set to monitor on FPGA
414overall data or a specific port.
415
416The "events" directory describes the configuration templates for all available
417events which can be used with perf tool directly. For example, fab_mmio_read
418has the configuration "event=0x06,evtype=0x02,portid=0xff", which shows this
419event belongs to fabric type (0x02), the local event id is 0x06 and it is for
420overall monitoring (portid=0xff).
421
422Example usage of perf::
423
424  $# perf list |grep dfl_fme
425
426  dfl_fme0/fab_mmio_read/                              [Kernel PMU event]
427  <...>
428  dfl_fme0/fab_port_mmio_read,portid=?/                [Kernel PMU event]
429  <...>
430
431  $# perf stat -a -e dfl_fme0/fab_mmio_read/ <command>
432  or
433  $# perf stat -a -e dfl_fme0/event=0x06,evtype=0x02,portid=0xff/ <command>
434  or
435  $# perf stat -a -e dfl_fme0/config=0xff2006/ <command>
436
437Another example, fab_port_mmio_read monitors mmio read of a specific port. So
438its configuration template is "event=0x06,evtype=0x01,portid=?". The portid
439should be explicitly set.
440
441Its usage of perf::
442
443  $# perf stat -a -e dfl_fme0/fab_port_mmio_read,portid=0x0/ <command>
444  or
445  $# perf stat -a -e dfl_fme0/event=0x06,evtype=0x02,portid=0x0/ <command>
446  or
447  $# perf stat -a -e dfl_fme0/config=0x2006/ <command>
448
449Please note for fabric counters, overall perf events (fab_*) and port perf
450events (fab_port_*) actually share one set of counters in hardware, so it can't
451monitor both at the same time. If this set of counters is configured to monitor
452overall data, then per port perf data is not supported. See below example::
453
454  $# perf stat -e dfl_fme0/fab_mmio_read/,dfl_fme0/fab_port_mmio_write,\
455                                                    portid=0/ sleep 1
456
457  Performance counter stats for 'system wide':
458
459                 3      dfl_fme0/fab_mmio_read/
460   <not supported>      dfl_fme0/fab_port_mmio_write,portid=0x0/
461
462       1.001750904 seconds time elapsed
463
464The driver also provides a "cpumask" sysfs attribute, which contains only one
465CPU id used to access these perf events. Counting on multiple CPU is not allowed
466since they are system-wide counters on FPGA device.
467
468The current driver does not support sampling. So "perf record" is unsupported.
469
470
471Interrupt support
472=================
473Some FME and AFU private features are able to generate interrupts. As mentioned
474above, users could call ioctl (DFL_FPGA_*_GET_IRQ_NUM) to know whether or how
475many interrupts are supported for this private feature. Drivers also implement
476an eventfd based interrupt handling mechanism for users to get notified when
477interrupt happens. Users could set eventfds to driver via
478ioctl (DFL_FPGA_*_SET_IRQ), and then poll/select on these eventfds waiting for
479notification.
480In Current DFL, 3 sub features (Port error, FME global error and AFU interrupt)
481support interrupts.
482
483
484Add new FIUs support
485====================
486It's possible that developers made some new function blocks (FIUs) under this
487DFL framework, then new platform device driver needs to be developed for the
488new feature dev (FIU) following the same way as existing feature dev drivers
489(e.g. FME and Port/AFU platform device driver). Besides that, it requires
490modification on DFL framework enumeration code too, for new FIU type detection
491and related platform devices creation.
492
493
494Add new private features support
495================================
496In some cases, we may need to add some new private features to existing FIUs
497(e.g. FME or Port). Developers don't need to touch enumeration code in DFL
498framework, as each private feature will be parsed automatically and related
499mmio resources can be found under FIU platform device created by DFL framework.
500Developer only needs to provide a sub feature driver with matched feature id.
501FME Partial Reconfiguration Sub Feature driver (see drivers/fpga/dfl-fme-pr.c)
502could be a reference.
503
504Location of DFLs on a PCI Device
505================================
506The original method for finding a DFL on a PCI device assumed the start of the
507first DFL to offset 0 of bar 0.  If the first node of the DFL is an FME,
508then further DFLs in the port(s) are specified in FME header registers.
509Alternatively, a PCIe vendor specific capability structure can be used to
510specify the location of all the DFLs on the device, providing flexibility
511for the type of starting node in the DFL.  Intel has reserved the
512VSEC ID of 0x43 for this purpose.  The vendor specific
513data begins with a 4 byte vendor specific register for the number of DFLs followed 4 byte
514Offset/BIR vendor specific registers for each DFL. Bits 2:0 of Offset/BIR register
515indicates the BAR, and bits 31:3 form the 8 byte aligned offset where bits 2:0 are
516zero.
517::
518
519        +----------------------------+
520        |31     Number of DFLS      0|
521        +----------------------------+
522        |31     Offset     3|2 BIR  0|
523        +----------------------------+
524                      . . .
525        +----------------------------+
526        |31     Offset     3|2 BIR  0|
527        +----------------------------+
528
529Being able to specify more than one DFL per BAR has been considered, but it
530was determined the use case did not provide value.  Specifying a single DFL
531per BAR simplifies the implementation and allows for extra error checking.
532
533Open discussion
534===============
535FME driver exports one ioctl (DFL_FPGA_FME_PORT_PR) for partial reconfiguration
536to user now. In the future, if unified user interfaces for reconfiguration are
537added, FME driver should switch to them from ioctl interface.
538