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 explanation 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 interleaved_dma - This is common to Slave as well as M2M clients. For slave 79 address of devices' fifo could be already known to the driver. 80 Various types of operations could be expressed by setting 81 appropriate values to the 'dma_interleaved_template' members. 82 83 A non-NULL return of this transfer API represents a "descriptor" for 84 the given transaction. 85 86 Interface: 87 struct dma_async_tx_descriptor *dmaengine_prep_slave_sg( 88 struct dma_chan *chan, struct scatterlist *sgl, 89 unsigned int sg_len, enum dma_data_direction direction, 90 unsigned long flags); 91 92 struct dma_async_tx_descriptor *dmaengine_prep_dma_cyclic( 93 struct dma_chan *chan, dma_addr_t buf_addr, size_t buf_len, 94 size_t period_len, enum dma_data_direction direction); 95 96 struct dma_async_tx_descriptor *dmaengine_prep_interleaved_dma( 97 struct dma_chan *chan, struct dma_interleaved_template *xt, 98 unsigned long flags); 99 100 The peripheral driver is expected to have mapped the scatterlist for 101 the DMA operation prior to calling dmaengine_prep_slave_sg(), and must 102 keep the scatterlist mapped until the DMA operation has completed. 103 The scatterlist must be mapped using the DMA struct device. 104 If a mapping needs to be synchronized later, dma_sync_*_for_*() must be 105 called using the DMA struct device, too. 106 So, normal setup should look like this: 107 108 nr_sg = dma_map_sg(chan->device->dev, sgl, sg_len); 109 if (nr_sg == 0) 110 /* error */ 111 112 desc = dmaengine_prep_slave_sg(chan, sgl, nr_sg, direction, flags); 113 114 Once a descriptor has been obtained, the callback information can be 115 added and the descriptor must then be submitted. Some DMA engine 116 drivers may hold a spinlock between a successful preparation and 117 submission so it is important that these two operations are closely 118 paired. 119 120 Note: 121 Although the async_tx API specifies that completion callback 122 routines cannot submit any new operations, this is not the 123 case for slave/cyclic DMA. 124 125 For slave DMA, the subsequent transaction may not be available 126 for submission prior to callback function being invoked, so 127 slave DMA callbacks are permitted to prepare and submit a new 128 transaction. 129 130 For cyclic DMA, a callback function may wish to terminate the 131 DMA via dmaengine_terminate_all(). 132 133 Therefore, it is important that DMA engine drivers drop any 134 locks before calling the callback function which may cause a 135 deadlock. 136 137 Note that callbacks will always be invoked from the DMA 138 engines tasklet, never from interrupt context. 139 1404. Submit the transaction 141 142 Once the descriptor has been prepared and the callback information 143 added, it must be placed on the DMA engine drivers pending queue. 144 145 Interface: 146 dma_cookie_t dmaengine_submit(struct dma_async_tx_descriptor *desc) 147 148 This returns a cookie can be used to check the progress of DMA engine 149 activity via other DMA engine calls not covered in this document. 150 151 dmaengine_submit() will not start the DMA operation, it merely adds 152 it to the pending queue. For this, see step 5, dma_async_issue_pending. 153 1545. Issue pending DMA requests and wait for callback notification 155 156 The transactions in the pending queue can be activated by calling the 157 issue_pending API. If channel is idle then the first transaction in 158 queue is started and subsequent ones queued up. 159 160 On completion of each DMA operation, the next in queue is started and 161 a tasklet triggered. The tasklet will then call the client driver 162 completion callback routine for notification, if set. 163 164 Interface: 165 void dma_async_issue_pending(struct dma_chan *chan); 166 167Further APIs: 168 1691. int dmaengine_terminate_all(struct dma_chan *chan) 170 171 This causes all activity for the DMA channel to be stopped, and may 172 discard data in the DMA FIFO which hasn't been fully transferred. 173 No callback functions will be called for any incomplete transfers. 174 1752. int dmaengine_pause(struct dma_chan *chan) 176 177 This pauses activity on the DMA channel without data loss. 178 1793. int dmaengine_resume(struct dma_chan *chan) 180 181 Resume a previously paused DMA channel. It is invalid to resume a 182 channel which is not currently paused. 183 1844. enum dma_status dma_async_is_tx_complete(struct dma_chan *chan, 185 dma_cookie_t cookie, dma_cookie_t *last, dma_cookie_t *used) 186 187 This can be used to check the status of the channel. Please see 188 the documentation in include/linux/dmaengine.h for a more complete 189 description of this API. 190 191 This can be used in conjunction with dma_async_is_complete() and 192 the cookie returned from dmaengine_submit() to check for 193 completion of a specific DMA transaction. 194 195 Note: 196 Not all DMA engine drivers can return reliable information for 197 a running DMA channel. It is recommended that DMA engine users 198 pause or stop (via dmaengine_terminate_all()) the channel before 199 using this API. 200