1 ==========================================
2 Xillybus driver for generic FPGA interface
3 ==========================================
4
5 :Author: Eli Billauer, Xillybus Ltd. (http://xillybus.com)
6 :Email: eli.billauer@gmail.com or as advertised on Xillybus' site.
7
8 .. Contents:
9
10 - Introduction
11 -- Background
12 -- Xillybus Overview
13
14 - Usage
15 -- User interface
16 -- Synchronization
17 -- Seekable pipes
18
19 - Internals
20 -- Source code organization
21 -- Pipe attributes
22 -- Host never reads from the FPGA
23 -- Channels, pipes, and the message channel
24 -- Data streaming
25 -- Data granularity
26 -- Probing
27 -- Buffer allocation
28 -- The "nonempty" message (supporting poll)
29
30
31 Introduction
32 ============
33
34 Background
35 ----------
36
37 An FPGA (Field Programmable Gate Array) is a piece of logic hardware, which
38 can be programmed to become virtually anything that is usually found as a
39 dedicated chipset: For instance, a display adapter, network interface card,
40 or even a processor with its peripherals. FPGAs are the LEGO of hardware:
41 Based upon certain building blocks, you make your own toys the way you like
42 them. It's usually pointless to reimplement something that is already
43 available on the market as a chipset, so FPGAs are mostly used when some
44 special functionality is needed, and the production volume is relatively low
45 (hence not justifying the development of an ASIC).
46
47 The challenge with FPGAs is that everything is implemented at a very low
48 level, even lower than assembly language. In order to allow FPGA designers to
49 focus on their specific project, and not reinvent the wheel over and over
50 again, pre-designed building blocks, IP cores, are often used. These are the
51 FPGA parallels of library functions. IP cores may implement certain
52 mathematical functions, a functional unit (e.g. a USB interface), an entire
53 processor (e.g. ARM) or anything that might come handy. Think of them as a
54 building block, with electrical wires dangling on the sides for connection to
55 other blocks.
56
57 One of the daunting tasks in FPGA design is communicating with a fullblown
58 operating system (actually, with the processor running it): Implementing the
59 low-level bus protocol and the somewhat higher-level interface with the host
60 (registers, interrupts, DMA etc.) is a project in itself. When the FPGA's
61 function is a well-known one (e.g. a video adapter card, or a NIC), it can
62 make sense to design the FPGA's interface logic specifically for the project.
63 A special driver is then written to present the FPGA as a well-known interface
64 to the kernel and/or user space. In that case, there is no reason to treat the
65 FPGA differently than any device on the bus.
66
67 It's however common that the desired data communication doesn't fit any well-
68 known peripheral function. Also, the effort of designing an elegant
69 abstraction for the data exchange is often considered too big. In those cases,
70 a quicker and possibly less elegant solution is sought: The driver is
71 effectively written as a user space program, leaving the kernel space part
72 with just elementary data transport. This still requires designing some
73 interface logic for the FPGA, and write a simple ad-hoc driver for the kernel.
74
75 Xillybus Overview
76 -----------------
77
78 Xillybus is an IP core and a Linux driver. Together, they form a kit for
79 elementary data transport between an FPGA and the host, providing pipe-like
80 data streams with a straightforward user interface. It's intended as a low-
81 effort solution for mixed FPGA-host projects, for which it makes sense to
82 have the project-specific part of the driver running in a user-space program.
83
84 Since the communication requirements may vary significantly from one FPGA
85 project to another (the number of data pipes needed in each direction and
86 their attributes), there isn't one specific chunk of logic being the Xillybus
87 IP core. Rather, the IP core is configured and built based upon a
88 specification given by its end user.
89
90 Xillybus presents independent data streams, which resemble pipes or TCP/IP
91 communication to the user. At the host side, a character device file is used
92 just like any pipe file. On the FPGA side, hardware FIFOs are used to stream
93 the data. This is contrary to a common method of communicating through fixed-
94 sized buffers (even though such buffers are used by Xillybus under the hood).
95 There may be more than a hundred of these streams on a single IP core, but
96 also no more than one, depending on the configuration.
97
98 In order to ease the deployment of the Xillybus IP core, it contains a simple
99 data structure which completely defines the core's configuration. The Linux
100 driver fetches this data structure during its initialization process, and sets
101 up the DMA buffers and character devices accordingly. As a result, a single
102 driver is used to work out of the box with any Xillybus IP core.
103
104 The data structure just mentioned should not be confused with PCI's
105 configuration space or the Flattened Device Tree.
106
107 Usage
108 =====
109
110 User interface
111 --------------
112
113 On the host, all interface with Xillybus is done through /dev/xillybus_*
114 device files, which are generated automatically as the drivers loads. The
115 names of these files depend on the IP core that is loaded in the FPGA (see
116 Probing below). To communicate with the FPGA, open the device file that
117 corresponds to the hardware FIFO you want to send data or receive data from,
118 and use plain write() or read() calls, just like with a regular pipe. In
119 particular, it makes perfect sense to go::
120
121 $ cat mydata > /dev/xillybus_thisfifo
122
123 $ cat /dev/xillybus_thatfifo > hisdata
124
125 possibly pressing CTRL-C as some stage, even though the xillybus_* pipes have
126 the capability to send an EOF (but may not use it).
127
128 The driver and hardware are designed to behave sensibly as pipes, including:
129
130 * Supporting non-blocking I/O (by setting O_NONBLOCK on open() ).
131
132 * Supporting poll() and select().
133
134 * Being bandwidth efficient under load (using DMA) but also handle small
135 pieces of data sent across (like TCP/IP) by autoflushing.
136
137 A device file can be read only, write only or bidirectional. Bidirectional
138 device files are treated like two independent pipes (except for sharing a
139 "channel" structure in the implementation code).
140
141 Synchronization
142 ---------------
143
144 Xillybus pipes are configured (on the IP core) to be either synchronous or
145 asynchronous. For a synchronous pipe, write() returns successfully only after
146 some data has been submitted and acknowledged by the FPGA. This slows down
147 bulk data transfers, and is nearly impossible for use with streams that
148 require data at a constant rate: There is no data transmitted to the FPGA
149 between write() calls, in particular when the process loses the CPU.
150
151 When a pipe is configured asynchronous, write() returns if there was enough
152 room in the buffers to store any of the data in the buffers.
153
154 For FPGA to host pipes, asynchronous pipes allow data transfer from the FPGA
155 as soon as the respective device file is opened, regardless of if the data
156 has been requested by a read() call. On synchronous pipes, only the amount
157 of data requested by a read() call is transmitted.
158
159 In summary, for synchronous pipes, data between the host and FPGA is
160 transmitted only to satisfy the read() or write() call currently handled
161 by the driver, and those calls wait for the transmission to complete before
162 returning.
163
164 Note that the synchronization attribute has nothing to do with the possibility
165 that read() or write() completes less bytes than requested. There is a
166 separate configuration flag ("allowpartial") that determines whether such a
167 partial completion is allowed.
168
169 Seekable pipes
170 --------------
171
172 A synchronous pipe can be configured to have the stream's position exposed
173 to the user logic at the FPGA. Such a pipe is also seekable on the host API.
174 With this feature, a memory or register interface can be attached on the
175 FPGA side to the seekable stream. Reading or writing to a certain address in
176 the attached memory is done by seeking to the desired address, and calling
177 read() or write() as required.
178
179
180 Internals
181 =========
182
183 Source code organization
184 ------------------------
185
186 The Xillybus driver consists of a core module, xillybus_core.c, and modules
187 that depend on the specific bus interface (xillybus_of.c and xillybus_pcie.c).
188
189 The bus specific modules are those probed when a suitable device is found by
190 the kernel. Since the DMA mapping and synchronization functions, which are bus
191 dependent by their nature, are used by the core module, a
192 xilly_endpoint_hardware structure is passed to the core module on
193 initialization. This structure is populated with pointers to wrapper functions
194 which execute the DMA-related operations on the bus.
195
196 Pipe attributes
197 ---------------
198
199 Each pipe has a number of attributes which are set when the FPGA component
200 (IP core) is built. They are fetched from the IDT (the data structure which
201 defines the core's configuration, see Probing below) by xilly_setupchannels()
202 in xillybus_core.c as follows:
203
204 * is_writebuf: The pipe's direction. A non-zero value means it's an FPGA to
205 host pipe (the FPGA "writes").
206
207 * channelnum: The pipe's identification number in communication between the
208 host and FPGA.
209
210 * format: The underlying data width. See Data Granularity below.
211
212 * allowpartial: A non-zero value means that a read() or write() (whichever
213 applies) may return with less than the requested number of bytes. The common
214 choice is a non-zero value, to match standard UNIX behavior.
215
216 * synchronous: A non-zero value means that the pipe is synchronous. See
217 Synchronization above.
218
219 * bufsize: Each DMA buffer's size. Always a power of two.
220
221 * bufnum: The number of buffers allocated for this pipe. Always a power of two.
222
223 * exclusive_open: A non-zero value forces exclusive opening of the associated
224 device file. If the device file is bidirectional, and already opened only in
225 one direction, the opposite direction may be opened once.
226
227 * seekable: A non-zero value indicates that the pipe is seekable. See
228 Seekable pipes above.
229
230 * supports_nonempty: A non-zero value (which is typical) indicates that the
231 hardware will send the messages that are necessary to support select() and
232 poll() for this pipe.
233
234 Host never reads from the FPGA
235 ------------------------------
236
237 Even though PCI Express is hotpluggable in general, a typical motherboard
238 doesn't expect a card to go away all of the sudden. But since the PCIe card
239 is based upon reprogrammable logic, a sudden disappearance from the bus is
240 quite likely as a result of an accidental reprogramming of the FPGA while the
241 host is up. In practice, nothing happens immediately in such a situation. But
242 if the host attempts to read from an address that is mapped to the PCI Express
243 device, that leads to an immediate freeze of the system on some motherboards,
244 even though the PCIe standard requires a graceful recovery.
245
246 In order to avoid these freezes, the Xillybus driver refrains completely from
247 reading from the device's register space. All communication from the FPGA to
248 the host is done through DMA. In particular, the Interrupt Service Routine
249 doesn't follow the common practice of checking a status register when it's
250 invoked. Rather, the FPGA prepares a small buffer which contains short
251 messages, which inform the host what the interrupt was about.
252
253 This mechanism is used on non-PCIe buses as well for the sake of uniformity.
254
255
256 Channels, pipes, and the message channel
257 ----------------------------------------
258
259 Each of the (possibly bidirectional) pipes presented to the user is allocated
260 a data channel between the FPGA and the host. The distinction between channels
261 and pipes is necessary only because of channel 0, which is used for interrupt-
262 related messages from the FPGA, and has no pipe attached to it.
263
264 Data streaming
265 --------------
266
267 Even though a non-segmented data stream is presented to the user at both
268 sides, the implementation relies on a set of DMA buffers which is allocated
269 for each channel. For the sake of illustration, let's take the FPGA to host
270 direction: As data streams into the respective channel's interface in the
271 FPGA, the Xillybus IP core writes it to one of the DMA buffers. When the
272 buffer is full, the FPGA informs the host about that (appending a
273 XILLYMSG_OPCODE_RELEASEBUF message channel 0 and sending an interrupt if
274 necessary). The host responds by making the data available for reading through
275 the character device. When all data has been read, the host writes on the
276 FPGA's buffer control register, allowing the buffer's overwriting. Flow
277 control mechanisms exist on both sides to prevent underflows and overflows.
278
279 This is not good enough for creating a TCP/IP-like stream: If the data flow
280 stops momentarily before a DMA buffer is filled, the intuitive expectation is
281 that the partial data in buffer will arrive anyhow, despite the buffer not
282 being completed. This is implemented by adding a field in the
283 XILLYMSG_OPCODE_RELEASEBUF message, through which the FPGA informs not just
284 which buffer is submitted, but how much data it contains.
285
286 But the FPGA will submit a partially filled buffer only if directed to do so
287 by the host. This situation occurs when the read() method has been blocking
288 for XILLY_RX_TIMEOUT jiffies (currently 10 ms), after which the host commands
289 the FPGA to submit a DMA buffer as soon as it can. This timeout mechanism
290 balances between bus bandwidth efficiency (preventing a lot of partially
291 filled buffers being sent) and a latency held fairly low for tails of data.
292
293 A similar setting is used in the host to FPGA direction. The handling of
294 partial DMA buffers is somewhat different, though. The user can tell the
295 driver to submit all data it has in the buffers to the FPGA, by issuing a
296 write() with the byte count set to zero. This is similar to a flush request,
297 but it doesn't block. There is also an autoflushing mechanism, which triggers
298 an equivalent flush roughly XILLY_RX_TIMEOUT jiffies after the last write().
299 This allows the user to be oblivious about the underlying buffering mechanism
300 and yet enjoy a stream-like interface.
301
302 Note that the issue of partial buffer flushing is irrelevant for pipes having
303 the "synchronous" attribute nonzero, since synchronous pipes don't allow data
304 to lay around in the DMA buffers between read() and write() anyhow.
305
306 Data granularity
307 ----------------
308
309 The data arrives or is sent at the FPGA as 8, 16 or 32 bit wide words, as
310 configured by the "format" attribute. Whenever possible, the driver attempts
311 to hide this when the pipe is accessed differently from its natural alignment.
312 For example, reading single bytes from a pipe with 32 bit granularity works
313 with no issues. Writing single bytes to pipes with 16 or 32 bit granularity
314 will also work, but the driver can't send partially completed words to the
315 FPGA, so the transmission of up to one word may be held until it's fully
316 occupied with user data.
317
318 This somewhat complicates the handling of host to FPGA streams, because
319 when a buffer is flushed, it may contain up to 3 bytes don't form a word in
320 the FPGA, and hence can't be sent. To prevent loss of data, these leftover
321 bytes need to be moved to the next buffer. The parts in xillybus_core.c
322 that mention "leftovers" in some way are related to this complication.
323
324 Probing
325 -------
326
327 As mentioned earlier, the number of pipes that are created when the driver
328 loads and their attributes depend on the Xillybus IP core in the FPGA. During
329 the driver's initialization, a blob containing configuration info, the
330 Interface Description Table (IDT), is sent from the FPGA to the host. The
331 bootstrap process is done in three phases:
332
333 1. Acquire the length of the IDT, so a buffer can be allocated for it. This
334 is done by sending a quiesce command to the device, since the acknowledge
335 for this command contains the IDT's buffer length.
336
337 2. Acquire the IDT itself.
338
339 3. Create the interfaces according to the IDT.
340
341 Buffer allocation
342 -----------------
343
344 In order to simplify the logic that prevents illegal boundary crossings of
345 PCIe packets, the following rule applies: If a buffer is smaller than 4kB,
346 it must not cross a 4kB boundary. Otherwise, it must be 4kB aligned. The
347 xilly_setupchannels() functions allocates these buffers by requesting whole
348 pages from the kernel, and diving them into DMA buffers as necessary. Since
349 all buffers' sizes are powers of two, it's possible to pack any set of such
350 buffers, with a maximal waste of one page of memory.
351
352 All buffers are allocated when the driver is loaded. This is necessary,
353 since large continuous physical memory segments are sometimes requested,
354 which are more likely to be available when the system is freshly booted.
355
356 The allocation of buffer memory takes place in the same order they appear in
357 the IDT. The driver relies on a rule that the pipes are sorted with decreasing
358 buffer size in the IDT. If a requested buffer is larger or equal to a page,
359 the necessary number of pages is requested from the kernel, and these are
360 used for this buffer. If the requested buffer is smaller than a page, one
361 single page is requested from the kernel, and that page is partially used.
362 Or, if there already is a partially used page at hand, the buffer is packed
363 into that page. It can be shown that all pages requested from the kernel
364 (except possibly for the last) are 100% utilized this way.
365
366 The "nonempty" message (supporting poll)
367 ----------------------------------------
368
369 In order to support the "poll" method (and hence select() ), there is a small
370 catch regarding the FPGA to host direction: The FPGA may have filled a DMA
371 buffer with some data, but not submitted that buffer. If the host waited for
372 the buffer's submission by the FPGA, there would be a possibility that the
373 FPGA side has sent data, but a select() call would still block, because the
374 host has not received any notification about this. This is solved with
375 XILLYMSG_OPCODE_NONEMPTY messages sent by the FPGA when a channel goes from
376 completely empty to containing some data.
377
378 These messages are used only to support poll() and select(). The IP core can
379 be configured not to send them for a slight reduction of bandwidth.
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.