1==========================
2Remote Processor Framework
3==========================
4
5Introduction
6============
7
8Modern SoCs typically have heterogeneous remote processor devices in asymmetric
9multiprocessing (AMP) configurations, which may be running different instances
10of operating system, whether it's Linux or any other flavor of real-time OS.
11
12OMAP4, for example, has dual Cortex-A9, dual Cortex-M3 and a C64x+ DSP.
13In a typical configuration, the dual cortex-A9 is running Linux in a SMP
14configuration, and each of the other three cores (two M3 cores and a DSP)
15is running its own instance of RTOS in an AMP configuration.
16
17The remoteproc framework allows different platforms/architectures to
18control (power on, load firmware, power off) those remote processors while
19abstracting the hardware differences, so the entire driver doesn't need to be
20duplicated. In addition, this framework also adds rpmsg virtio devices
21for remote processors that supports this kind of communication. This way,
22platform-specific remoteproc drivers only need to provide a few low-level
23handlers, and then all rpmsg drivers will then just work
24(for more information about the virtio-based rpmsg bus and its drivers,
25please read Documentation/staging/rpmsg.rst).
26Registration of other types of virtio devices is now also possible. Firmwares
27just need to publish what kind of virtio devices do they support, and then
28remoteproc will add those devices. This makes it possible to reuse the
29existing virtio drivers with remote processor backends at a minimal development
30cost.
31
32User API
33========
34
35::
36
37  int rproc_boot(struct rproc *rproc)
38
39Boot a remote processor (i.e. load its firmware, power it on, ...).
40
41If the remote processor is already powered on, this function immediately
42returns (successfully).
43
44Returns 0 on success, and an appropriate error value otherwise.
45Note: to use this function you should already have a valid rproc
46handle. There are several ways to achieve that cleanly (devres, pdata,
47the way remoteproc_rpmsg.c does this, or, if this becomes prevalent, we
48might also consider using dev_archdata for this).
49
50::
51
52  void rproc_shutdown(struct rproc *rproc)
53
54Power off a remote processor (previously booted with rproc_boot()).
55In case @rproc is still being used by an additional user(s), then
56this function will just decrement the power refcount and exit,
57without really powering off the device.
58
59Every call to rproc_boot() must (eventually) be accompanied by a call
60to rproc_shutdown(). Calling rproc_shutdown() redundantly is a bug.
61
62.. note::
63
64  we're not decrementing the rproc's refcount, only the power refcount.
65  which means that the @rproc handle stays valid even after
66  rproc_shutdown() returns, and users can still use it with a subsequent
67  rproc_boot(), if needed.
68
69::
70
71  struct rproc *rproc_get_by_phandle(phandle phandle)
72
73Find an rproc handle using a device tree phandle. Returns the rproc
74handle on success, and NULL on failure. This function increments
75the remote processor's refcount, so always use rproc_put() to
76decrement it back once rproc isn't needed anymore.
77
78Typical usage
79=============
80
81::
82
83  #include <linux/remoteproc.h>
84
85  /* in case we were given a valid 'rproc' handle */
86  int dummy_rproc_example(struct rproc *my_rproc)
87  {
88	int ret;
89
90	/* let's power on and boot our remote processor */
91	ret = rproc_boot(my_rproc);
92	if (ret) {
93		/*
94		 * something went wrong. handle it and leave.
95		 */
96	}
97
98	/*
99	 * our remote processor is now powered on... give it some work
100	 */
101
102	/* let's shut it down now */
103	rproc_shutdown(my_rproc);
104  }
105
106API for implementors
107====================
108
109::
110
111  struct rproc *rproc_alloc(struct device *dev, const char *name,
112				const struct rproc_ops *ops,
113				const char *firmware, int len)
114
115Allocate a new remote processor handle, but don't register
116it yet. Required parameters are the underlying device, the
117name of this remote processor, platform-specific ops handlers,
118the name of the firmware to boot this rproc with, and the
119length of private data needed by the allocating rproc driver (in bytes).
120
121This function should be used by rproc implementations during
122initialization of the remote processor.
123
124After creating an rproc handle using this function, and when ready,
125implementations should then call rproc_add() to complete
126the registration of the remote processor.
127
128On success, the new rproc is returned, and on failure, NULL.
129
130.. note::
131
132  **never** directly deallocate @rproc, even if it was not registered
133  yet. Instead, when you need to unroll rproc_alloc(), use rproc_free().
134
135::
136
137  void rproc_free(struct rproc *rproc)
138
139Free an rproc handle that was allocated by rproc_alloc.
140
141This function essentially unrolls rproc_alloc(), by decrementing the
142rproc's refcount. It doesn't directly free rproc; that would happen
143only if there are no other references to rproc and its refcount now
144dropped to zero.
145
146::
147
148  int rproc_add(struct rproc *rproc)
149
150Register @rproc with the remoteproc framework, after it has been
151allocated with rproc_alloc().
152
153This is called by the platform-specific rproc implementation, whenever
154a new remote processor device is probed.
155
156Returns 0 on success and an appropriate error code otherwise.
157Note: this function initiates an asynchronous firmware loading
158context, which will look for virtio devices supported by the rproc's
159firmware.
160
161If found, those virtio devices will be created and added, so as a result
162of registering this remote processor, additional virtio drivers might get
163probed.
164
165::
166
167  int rproc_del(struct rproc *rproc)
168
169Unroll rproc_add().
170
171This function should be called when the platform specific rproc
172implementation decides to remove the rproc device. it should
173_only_ be called if a previous invocation of rproc_add()
174has completed successfully.
175
176After rproc_del() returns, @rproc is still valid, and its
177last refcount should be decremented by calling rproc_free().
178
179Returns 0 on success and -EINVAL if @rproc isn't valid.
180
181::
182
183  void rproc_report_crash(struct rproc *rproc, enum rproc_crash_type type)
184
185Report a crash in a remoteproc
186
187This function must be called every time a crash is detected by the
188platform specific rproc implementation. This should not be called from a
189non-remoteproc driver. This function can be called from atomic/interrupt
190context.
191
192Implementation callbacks
193========================
194
195These callbacks should be provided by platform-specific remoteproc
196drivers::
197
198  /**
199   * struct rproc_ops - platform-specific device handlers
200   * @start:	power on the device and boot it
201   * @stop:	power off the device
202   * @kick:	kick a virtqueue (virtqueue id given as a parameter)
203   */
204  struct rproc_ops {
205	int (*start)(struct rproc *rproc);
206	int (*stop)(struct rproc *rproc);
207	void (*kick)(struct rproc *rproc, int vqid);
208  };
209
210Every remoteproc implementation should at least provide the ->start and ->stop
211handlers. If rpmsg/virtio functionality is also desired, then the ->kick handler
212should be provided as well.
213
214The ->start() handler takes an rproc handle and should then power on the
215device and boot it (use rproc->priv to access platform-specific private data).
216The boot address, in case needed, can be found in rproc->bootaddr (remoteproc
217core puts there the ELF entry point).
218On success, 0 should be returned, and on failure, an appropriate error code.
219
220The ->stop() handler takes an rproc handle and powers the device down.
221On success, 0 is returned, and on failure, an appropriate error code.
222
223The ->kick() handler takes an rproc handle, and an index of a virtqueue
224where new message was placed in. Implementations should interrupt the remote
225processor and let it know it has pending messages. Notifying remote processors
226the exact virtqueue index to look in is optional: it is easy (and not
227too expensive) to go through the existing virtqueues and look for new buffers
228in the used rings.
229
230Binary Firmware Structure
231=========================
232
233At this point remoteproc supports ELF32 and ELF64 firmware binaries. However,
234it is quite expected that other platforms/devices which we'd want to
235support with this framework will be based on different binary formats.
236
237When those use cases show up, we will have to decouple the binary format
238from the framework core, so we can support several binary formats without
239duplicating common code.
240
241When the firmware is parsed, its various segments are loaded to memory
242according to the specified device address (might be a physical address
243if the remote processor is accessing memory directly).
244
245In addition to the standard ELF segments, most remote processors would
246also include a special section which we call "the resource table".
247
248The resource table contains system resources that the remote processor
249requires before it should be powered on, such as allocation of physically
250contiguous memory, or iommu mapping of certain on-chip peripherals.
251Remotecore will only power up the device after all the resource table's
252requirement are met.
253
254In addition to system resources, the resource table may also contain
255resource entries that publish the existence of supported features
256or configurations by the remote processor, such as trace buffers and
257supported virtio devices (and their configurations).
258
259The resource table begins with this header::
260
261  /**
262   * struct resource_table - firmware resource table header
263   * @ver: version number
264   * @num: number of resource entries
265   * @reserved: reserved (must be zero)
266   * @offset: array of offsets pointing at the various resource entries
267   *
268   * The header of the resource table, as expressed by this structure,
269   * contains a version number (should we need to change this format in the
270   * future), the number of available resource entries, and their offsets
271   * in the table.
272   */
273  struct resource_table {
274	u32 ver;
275	u32 num;
276	u32 reserved[2];
277	u32 offset[0];
278  } __packed;
279
280Immediately following this header are the resource entries themselves,
281each of which begins with the following resource entry header::
282
283  /**
284   * struct fw_rsc_hdr - firmware resource entry header
285   * @type: resource type
286   * @data: resource data
287   *
288   * Every resource entry begins with a 'struct fw_rsc_hdr' header providing
289   * its @type. The content of the entry itself will immediately follow
290   * this header, and it should be parsed according to the resource type.
291   */
292  struct fw_rsc_hdr {
293	u32 type;
294	u8 data[0];
295  } __packed;
296
297Some resources entries are mere announcements, where the host is informed
298of specific remoteproc configuration. Other entries require the host to
299do something (e.g. allocate a system resource). Sometimes a negotiation
300is expected, where the firmware requests a resource, and once allocated,
301the host should provide back its details (e.g. address of an allocated
302memory region).
303
304Here are the various resource types that are currently supported::
305
306  /**
307   * enum fw_resource_type - types of resource entries
308   *
309   * @RSC_CARVEOUT:   request for allocation of a physically contiguous
310   *		    memory region.
311   * @RSC_DEVMEM:     request to iommu_map a memory-based peripheral.
312   * @RSC_TRACE:	    announces the availability of a trace buffer into which
313   *		    the remote processor will be writing logs.
314   * @RSC_VDEV:       declare support for a virtio device, and serve as its
315   *		    virtio header.
316   * @RSC_LAST:       just keep this one at the end
317   * @RSC_VENDOR_START:	start of the vendor specific resource types range
318   * @RSC_VENDOR_END:	end of the vendor specific resource types range
319   *
320   * Please note that these values are used as indices to the rproc_handle_rsc
321   * lookup table, so please keep them sane. Moreover, @RSC_LAST is used to
322   * check the validity of an index before the lookup table is accessed, so
323   * please update it as needed.
324   */
325  enum fw_resource_type {
326	RSC_CARVEOUT		= 0,
327	RSC_DEVMEM		= 1,
328	RSC_TRACE		= 2,
329	RSC_VDEV		= 3,
330	RSC_LAST		= 4,
331	RSC_VENDOR_START	= 128,
332	RSC_VENDOR_END		= 512,
333  };
334
335For more details regarding a specific resource type, please see its
336dedicated structure in include/linux/remoteproc.h.
337
338We also expect that platform-specific resource entries will show up
339at some point. When that happens, we could easily add a new RSC_PLATFORM
340type, and hand those resources to the platform-specific rproc driver to handle.
341
342Virtio and remoteproc
343=====================
344
345The firmware should provide remoteproc information about virtio devices
346that it supports, and their configurations: a RSC_VDEV resource entry
347should specify the virtio device id (as in virtio_ids.h), virtio features,
348virtio config space, vrings information, etc.
349
350When a new remote processor is registered, the remoteproc framework
351will look for its resource table and will register the virtio devices
352it supports. A firmware may support any number of virtio devices, and
353of any type (a single remote processor can also easily support several
354rpmsg virtio devices this way, if desired).
355
356Of course, RSC_VDEV resource entries are only good enough for static
357allocation of virtio devices. Dynamic allocations will also be made possible
358using the rpmsg bus (similar to how we already do dynamic allocations of
359rpmsg channels; read more about it in rpmsg.txt).
360