Root/Documentation/dmaengine.txt

1            DMA Engine API Guide
2            ====================
3
4         Vinod Koul <vinod dot koul at intel.com>
5
6NOTE: For DMA Engine usage in async_tx please see:
7    Documentation/crypto/async-tx-api.txt
8
9
10Below is a guide to device driver writers on how to use the Slave-DMA API of the
11DMA Engine. This is applicable only for slave DMA usage only.
12
13The slave DMA usage consists of following steps:
141. Allocate a DMA slave channel
152. Set slave and controller specific parameters
163. Get a descriptor for transaction
174. Submit the transaction
185. Issue pending requests and wait for callback notification
19
201. Allocate a DMA slave channel
21
22   Channel allocation is slightly different in the slave DMA context,
23   client drivers typically need a channel from a particular DMA
24   controller only and even in some cases a specific channel is desired.
25   To request a channel dma_request_channel() API is used.
26
27   Interface:
28    struct dma_chan *dma_request_channel(dma_cap_mask_t mask,
29            dma_filter_fn filter_fn,
30            void *filter_param);
31   where dma_filter_fn is defined as:
32    typedef bool (*dma_filter_fn)(struct dma_chan *chan, void *filter_param);
33
34   The 'filter_fn' parameter is optional, but highly recommended for
35   slave and cyclic channels as they typically need to obtain a specific
36   DMA channel.
37
38   When the optional 'filter_fn' parameter is NULL, dma_request_channel()
39   simply returns the first channel that satisfies the capability mask.
40
41   Otherwise, the 'filter_fn' routine will be called once for each free
42   channel which has a capability in 'mask'. 'filter_fn' is expected to
43   return 'true' when the desired DMA channel is found.
44
45   A channel allocated via this interface is exclusive to the caller,
46   until dma_release_channel() is called.
47
482. Set slave and controller specific parameters
49
50   Next step is always to pass some specific information to the DMA
51   driver. Most of the generic information which a slave DMA can use
52   is in struct dma_slave_config. This allows the clients to specify
53   DMA direction, DMA addresses, bus widths, DMA burst lengths etc
54   for the peripheral.
55
56   If some DMA controllers have more parameters to be sent then they
57   should try to embed struct dma_slave_config in their controller
58   specific structure. That gives flexibility to client to pass more
59   parameters, if required.
60
61   Interface:
62    int dmaengine_slave_config(struct dma_chan *chan,
63                  struct dma_slave_config *config)
64
65   Please see the dma_slave_config structure definition in dmaengine.h
66   for a detailed explaination of the struct members. Please note
67   that the 'direction' member will be going away as it duplicates the
68   direction given in the prepare call.
69
703. Get a descriptor for transaction
71
72   For slave usage the various modes of slave transfers supported by the
73   DMA-engine are:
74
75   slave_sg - DMA a list of scatter gather buffers from/to a peripheral
76   dma_cyclic - Perform a cyclic DMA operation from/to a peripheral till the
77          operation is explicitly stopped.
78
79   A non-NULL return of this transfer API represents a "descriptor" for
80   the given transaction.
81
82   Interface:
83    struct dma_async_tx_descriptor *(*chan->device->device_prep_slave_sg)(
84        struct dma_chan *chan, struct scatterlist *sgl,
85        unsigned int sg_len, enum dma_data_direction direction,
86        unsigned long flags);
87
88    struct dma_async_tx_descriptor *(*chan->device->device_prep_dma_cyclic)(
89        struct dma_chan *chan, dma_addr_t buf_addr, size_t buf_len,
90        size_t period_len, enum dma_data_direction direction);
91
92   The peripheral driver is expected to have mapped the scatterlist for
93   the DMA operation prior to calling device_prep_slave_sg, and must
94   keep the scatterlist mapped until the DMA operation has completed.
95   The scatterlist must be mapped using the DMA struct device. So,
96   normal setup should look like this:
97
98    nr_sg = dma_map_sg(chan->device->dev, sgl, sg_len);
99    if (nr_sg == 0)
100        /* error */
101
102    desc = chan->device->device_prep_slave_sg(chan, sgl, nr_sg,
103            direction, flags);
104
105   Once a descriptor has been obtained, the callback information can be
106   added and the descriptor must then be submitted. Some DMA engine
107   drivers may hold a spinlock between a successful preparation and
108   submission so it is important that these two operations are closely
109   paired.
110
111   Note:
112    Although the async_tx API specifies that completion callback
113    routines cannot submit any new operations, this is not the
114    case for slave/cyclic DMA.
115
116    For slave DMA, the subsequent transaction may not be available
117    for submission prior to callback function being invoked, so
118    slave DMA callbacks are permitted to prepare and submit a new
119    transaction.
120
121    For cyclic DMA, a callback function may wish to terminate the
122    DMA via dmaengine_terminate_all().
123
124    Therefore, it is important that DMA engine drivers drop any
125    locks before calling the callback function which may cause a
126    deadlock.
127
128    Note that callbacks will always be invoked from the DMA
129    engines tasklet, never from interrupt context.
130
1314. Submit the transaction
132
133   Once the descriptor has been prepared and the callback information
134   added, it must be placed on the DMA engine drivers pending queue.
135
136   Interface:
137    dma_cookie_t dmaengine_submit(struct dma_async_tx_descriptor *desc)
138
139   This returns a cookie can be used to check the progress of DMA engine
140   activity via other DMA engine calls not covered in this document.
141
142   dmaengine_submit() will not start the DMA operation, it merely adds
143   it to the pending queue. For this, see step 5, dma_async_issue_pending.
144
1455. Issue pending DMA requests and wait for callback notification
146
147   The transactions in the pending queue can be activated by calling the
148   issue_pending API. If channel is idle then the first transaction in
149   queue is started and subsequent ones queued up.
150
151   On completion of each DMA operation, the next in queue is started and
152   a tasklet triggered. The tasklet will then call the client driver
153   completion callback routine for notification, if set.
154
155   Interface:
156    void dma_async_issue_pending(struct dma_chan *chan);
157
158Further APIs:
159
1601. int dmaengine_terminate_all(struct dma_chan *chan)
161
162   This causes all activity for the DMA channel to be stopped, and may
163   discard data in the DMA FIFO which hasn't been fully transferred.
164   No callback functions will be called for any incomplete transfers.
165
1662. int dmaengine_pause(struct dma_chan *chan)
167
168   This pauses activity on the DMA channel without data loss.
169
1703. int dmaengine_resume(struct dma_chan *chan)
171
172   Resume a previously paused DMA channel. It is invalid to resume a
173   channel which is not currently paused.
174
1754. enum dma_status dma_async_is_tx_complete(struct dma_chan *chan,
176        dma_cookie_t cookie, dma_cookie_t *last, dma_cookie_t *used)
177
178   This can be used to check the status of the channel. Please see
179   the documentation in include/linux/dmaengine.h for a more complete
180   description of this API.
181
182   This can be used in conjunction with dma_async_is_complete() and
183   the cookie returned from 'descriptor->submit()' to check for
184   completion of a specific DMA transaction.
185
186   Note:
187    Not all DMA engine drivers can return reliable information for
188    a running DMA channel. It is recommended that DMA engine users
189    pause or stop (via dmaengine_terminate_all) the channel before
190    using this API.
191

Archive Download this file



interactive