1====================
2DMA Engine API Guide
3====================
4
5Vinod Koul <vinod dot koul at intel.com>
6
7.. note:: For DMA Engine usage in async_tx please see:
8          ``Documentation/crypto/async-tx-api.rst``
9
10
11Below is a guide to device driver writers on how to use the Slave-DMA API of the
12DMA Engine. This is applicable only for slave DMA usage only.
13
14DMA usage
15=========
16
17The slave DMA usage consists of following steps:
18
19- Allocate a DMA slave channel
20
21- Set slave and controller specific parameters
22
23- Get a descriptor for transaction
24
25- Submit the transaction
26
27- Issue pending requests and wait for callback notification
28
29The details of these operations are:
30
311. Allocate a DMA slave channel
32
33   Channel allocation is slightly different in the slave DMA context,
34   client drivers typically need a channel from a particular DMA
35   controller only and even in some cases a specific channel is desired.
36   To request a channel dma_request_chan() API is used.
37
38   Interface:
39
40   .. code-block:: c
41
42      struct dma_chan *dma_request_chan(struct device *dev, const char *name);
43
44   Which will find and return the ``name`` DMA channel associated with the 'dev'
45   device. The association is done via DT, ACPI or board file based
46   dma_slave_map matching table.
47
48   A channel allocated via this interface is exclusive to the caller,
49   until dma_release_channel() is called.
50
512. Set slave and controller specific parameters
52
53   Next step is always to pass some specific information to the DMA
54   driver. Most of the generic information which a slave DMA can use
55   is in struct dma_slave_config. This allows the clients to specify
56   DMA direction, DMA addresses, bus widths, DMA burst lengths etc
57   for the peripheral.
58
59   If some DMA controllers have more parameters to be sent then they
60   should try to embed struct dma_slave_config in their controller
61   specific structure. That gives flexibility to client to pass more
62   parameters, if required.
63
64   Interface:
65
66   .. code-block:: c
67
68      int dmaengine_slave_config(struct dma_chan *chan,
69			struct dma_slave_config *config)
70
71   Please see the dma_slave_config structure definition in dmaengine.h
72   for a detailed explanation of the struct members. Please note
73   that the 'direction' member will be going away as it duplicates the
74   direction given in the prepare call.
75
763. Get a descriptor for transaction
77
78  For slave usage the various modes of slave transfers supported by the
79  DMA-engine are:
80
81  - slave_sg: DMA a list of scatter gather buffers from/to a peripheral
82
83  - dma_cyclic: Perform a cyclic DMA operation from/to a peripheral till the
84    operation is explicitly stopped.
85
86  - interleaved_dma: This is common to Slave as well as M2M clients. For slave
87    address of devices' fifo could be already known to the driver.
88    Various types of operations could be expressed by setting
89    appropriate values to the 'dma_interleaved_template' members. Cyclic
90    interleaved DMA transfers are also possible if supported by the channel by
91    setting the DMA_PREP_REPEAT transfer flag.
92
93  A non-NULL return of this transfer API represents a "descriptor" for
94  the given transaction.
95
96  Interface:
97
98  .. code-block:: c
99
100     struct dma_async_tx_descriptor *dmaengine_prep_slave_sg(
101		struct dma_chan *chan, struct scatterlist *sgl,
102		unsigned int sg_len, enum dma_data_direction direction,
103		unsigned long flags);
104
105     struct dma_async_tx_descriptor *dmaengine_prep_dma_cyclic(
106		struct dma_chan *chan, dma_addr_t buf_addr, size_t buf_len,
107		size_t period_len, enum dma_data_direction direction);
108
109     struct dma_async_tx_descriptor *dmaengine_prep_interleaved_dma(
110		struct dma_chan *chan, struct dma_interleaved_template *xt,
111		unsigned long flags);
112
113  The peripheral driver is expected to have mapped the scatterlist for
114  the DMA operation prior to calling dmaengine_prep_slave_sg(), and must
115  keep the scatterlist mapped until the DMA operation has completed.
116  The scatterlist must be mapped using the DMA struct device.
117  If a mapping needs to be synchronized later, dma_sync_*_for_*() must be
118  called using the DMA struct device, too.
119  So, normal setup should look like this:
120
121  .. code-block:: c
122
123     struct device *dma_dev = dmaengine_get_dma_device(chan);
124
125     nr_sg = dma_map_sg(dma_dev, sgl, sg_len);
126	if (nr_sg == 0)
127		/* error */
128
129	desc = dmaengine_prep_slave_sg(chan, sgl, nr_sg, direction, flags);
130
131  Once a descriptor has been obtained, the callback information can be
132  added and the descriptor must then be submitted. Some DMA engine
133  drivers may hold a spinlock between a successful preparation and
134  submission so it is important that these two operations are closely
135  paired.
136
137  .. note::
138
139     Although the async_tx API specifies that completion callback
140     routines cannot submit any new operations, this is not the
141     case for slave/cyclic DMA.
142
143     For slave DMA, the subsequent transaction may not be available
144     for submission prior to callback function being invoked, so
145     slave DMA callbacks are permitted to prepare and submit a new
146     transaction.
147
148     For cyclic DMA, a callback function may wish to terminate the
149     DMA via dmaengine_terminate_async().
150
151     Therefore, it is important that DMA engine drivers drop any
152     locks before calling the callback function which may cause a
153     deadlock.
154
155     Note that callbacks will always be invoked from the DMA
156     engines tasklet, never from interrupt context.
157
158  **Optional: per descriptor metadata**
159
160  DMAengine provides two ways for metadata support.
161
162  DESC_METADATA_CLIENT
163
164    The metadata buffer is allocated/provided by the client driver and it is
165    attached to the descriptor.
166
167  .. code-block:: c
168
169     int dmaengine_desc_attach_metadata(struct dma_async_tx_descriptor *desc,
170				   void *data, size_t len);
171
172  DESC_METADATA_ENGINE
173
174    The metadata buffer is allocated/managed by the DMA driver. The client
175    driver can ask for the pointer, maximum size and the currently used size of
176    the metadata and can directly update or read it.
177
178    Becasue the DMA driver manages the memory area containing the metadata,
179    clients must make sure that they do not try to access or get the pointer
180    after their transfer completion callback has run for the descriptor.
181    If no completion callback has been defined for the transfer, then the
182    metadata must not be accessed after issue_pending.
183    In other words: if the aim is to read back metadata after the transfer is
184    completed, then the client must use completion callback.
185
186  .. code-block:: c
187
188     void *dmaengine_desc_get_metadata_ptr(struct dma_async_tx_descriptor *desc,
189		size_t *payload_len, size_t *max_len);
190
191     int dmaengine_desc_set_metadata_len(struct dma_async_tx_descriptor *desc,
192		size_t payload_len);
193
194  Client drivers can query if a given mode is supported with:
195
196  .. code-block:: c
197
198     bool dmaengine_is_metadata_mode_supported(struct dma_chan *chan,
199		enum dma_desc_metadata_mode mode);
200
201  Depending on the used mode client drivers must follow different flow.
202
203  DESC_METADATA_CLIENT
204
205    - DMA_MEM_TO_DEV / DEV_MEM_TO_MEM:
206
207      1. prepare the descriptor (dmaengine_prep_*)
208         construct the metadata in the client's buffer
209      2. use dmaengine_desc_attach_metadata() to attach the buffer to the
210         descriptor
211      3. submit the transfer
212
213    - DMA_DEV_TO_MEM:
214
215      1. prepare the descriptor (dmaengine_prep_*)
216      2. use dmaengine_desc_attach_metadata() to attach the buffer to the
217         descriptor
218      3. submit the transfer
219      4. when the transfer is completed, the metadata should be available in the
220         attached buffer
221
222  DESC_METADATA_ENGINE
223
224    - DMA_MEM_TO_DEV / DEV_MEM_TO_MEM:
225
226      1. prepare the descriptor (dmaengine_prep_*)
227      2. use dmaengine_desc_get_metadata_ptr() to get the pointer to the
228         engine's metadata area
229      3. update the metadata at the pointer
230      4. use dmaengine_desc_set_metadata_len()  to tell the DMA engine the
231         amount of data the client has placed into the metadata buffer
232      5. submit the transfer
233
234    - DMA_DEV_TO_MEM:
235
236      1. prepare the descriptor (dmaengine_prep_*)
237      2. submit the transfer
238      3. on transfer completion, use dmaengine_desc_get_metadata_ptr() to get
239         the pointer to the engine's metadata area
240      4. read out the metadata from the pointer
241
242  .. note::
243
244     When DESC_METADATA_ENGINE mode is used the metadata area for the descriptor
245     is no longer valid after the transfer has been completed (valid up to the
246     point when the completion callback returns if used).
247
248     Mixed use of DESC_METADATA_CLIENT / DESC_METADATA_ENGINE is not allowed,
249     client drivers must use either of the modes per descriptor.
250
2514. Submit the transaction
252
253   Once the descriptor has been prepared and the callback information
254   added, it must be placed on the DMA engine drivers pending queue.
255
256   Interface:
257
258   .. code-block:: c
259
260      dma_cookie_t dmaengine_submit(struct dma_async_tx_descriptor *desc)
261
262   This returns a cookie can be used to check the progress of DMA engine
263   activity via other DMA engine calls not covered in this document.
264
265   dmaengine_submit() will not start the DMA operation, it merely adds
266   it to the pending queue. For this, see step 5, dma_async_issue_pending.
267
268   .. note::
269
270      After calling ``dmaengine_submit()`` the submitted transfer descriptor
271      (``struct dma_async_tx_descriptor``) belongs to the DMA engine.
272      Consequently, the client must consider invalid the pointer to that
273      descriptor.
274
2755. Issue pending DMA requests and wait for callback notification
276
277   The transactions in the pending queue can be activated by calling the
278   issue_pending API. If channel is idle then the first transaction in
279   queue is started and subsequent ones queued up.
280
281   On completion of each DMA operation, the next in queue is started and
282   a tasklet triggered. The tasklet will then call the client driver
283   completion callback routine for notification, if set.
284
285   Interface:
286
287   .. code-block:: c
288
289      void dma_async_issue_pending(struct dma_chan *chan);
290
291Further APIs
292------------
293
2941. Terminate APIs
295
296   .. code-block:: c
297
298      int dmaengine_terminate_sync(struct dma_chan *chan)
299      int dmaengine_terminate_async(struct dma_chan *chan)
300      int dmaengine_terminate_all(struct dma_chan *chan) /* DEPRECATED */
301
302   This causes all activity for the DMA channel to be stopped, and may
303   discard data in the DMA FIFO which hasn't been fully transferred.
304   No callback functions will be called for any incomplete transfers.
305
306   Two variants of this function are available.
307
308   dmaengine_terminate_async() might not wait until the DMA has been fully
309   stopped or until any running complete callbacks have finished. But it is
310   possible to call dmaengine_terminate_async() from atomic context or from
311   within a complete callback. dmaengine_synchronize() must be called before it
312   is safe to free the memory accessed by the DMA transfer or free resources
313   accessed from within the complete callback.
314
315   dmaengine_terminate_sync() will wait for the transfer and any running
316   complete callbacks to finish before it returns. But the function must not be
317   called from atomic context or from within a complete callback.
318
319   dmaengine_terminate_all() is deprecated and should not be used in new code.
320
3212. Pause API
322
323   .. code-block:: c
324
325      int dmaengine_pause(struct dma_chan *chan)
326
327   This pauses activity on the DMA channel without data loss.
328
3293. Resume API
330
331   .. code-block:: c
332
333       int dmaengine_resume(struct dma_chan *chan)
334
335   Resume a previously paused DMA channel. It is invalid to resume a
336   channel which is not currently paused.
337
3384. Check Txn complete
339
340   .. code-block:: c
341
342      enum dma_status dma_async_is_tx_complete(struct dma_chan *chan,
343		dma_cookie_t cookie, dma_cookie_t *last, dma_cookie_t *used)
344
345   This can be used to check the status of the channel. Please see
346   the documentation in include/linux/dmaengine.h for a more complete
347   description of this API.
348
349   This can be used in conjunction with dma_async_is_complete() and
350   the cookie returned from dmaengine_submit() to check for
351   completion of a specific DMA transaction.
352
353   .. note::
354
355      Not all DMA engine drivers can return reliable information for
356      a running DMA channel. It is recommended that DMA engine users
357      pause or stop (via dmaengine_terminate_all()) the channel before
358      using this API.
359
3605. Synchronize termination API
361
362   .. code-block:: c
363
364      void dmaengine_synchronize(struct dma_chan *chan)
365
366   Synchronize the termination of the DMA channel to the current context.
367
368   This function should be used after dmaengine_terminate_async() to synchronize
369   the termination of the DMA channel to the current context. The function will
370   wait for the transfer and any running complete callbacks to finish before it
371   returns.
372
373   If dmaengine_terminate_async() is used to stop the DMA channel this function
374   must be called before it is safe to free memory accessed by previously
375   submitted descriptors or to free any resources accessed within the complete
376   callback of previously submitted descriptors.
377
378   The behavior of this function is undefined if dma_async_issue_pending() has
379   been called between dmaengine_terminate_async() and this function.
380