Documentation/dmaengine/client.txt

   1                         DMA Engine API Guide
   2                         ====================
   3
   4                  Vinod Koul <vinod dot koul at intel.com>
   5
   6 NOTE: For DMA Engine usage in async_tx please see:
   7         Documentation/crypto/async-tx-api.txt
   8
   9
  10 Below is a guide to device driver writers on how to use the Slave-DMA API of the
  11 DMA Engine. This is applicable only for slave DMA usage only.
  12
  13 The slave DMA usage consists of following steps:
  14 1. Allocate a DMA slave channel
  15 2. Set slave and controller specific parameters
  16 3. Get a descriptor for transaction
  17 4. Submit the transaction
  18 5. Issue pending requests and wait for callback notification
  19
  20 1. 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_chan() API is used.
  26
  27    Interface:
  28         struct dma_chan *dma_request_chan(struct device *dev, const char *name);
  29
  30    Which will find and return the 'name' DMA channel associated with the 'dev'
  31    device. The association is done via DT, ACPI or board file based
  32    dma_slave_map matching table.
  33
  34    A channel allocated via this interface is exclusive to the caller,
  35    until dma_release_channel() is called.
  36
  37 2. Set slave and controller specific parameters
  38
  39    Next step is always to pass some specific information to the DMA
  40    driver. Most of the generic information which a slave DMA can use
  41    is in struct dma_slave_config. This allows the clients to specify
  42    DMA direction, DMA addresses, bus widths, DMA burst lengths etc
  43    for the peripheral.
  44
  45    If some DMA controllers have more parameters to be sent then they
  46    should try to embed struct dma_slave_config in their controller
  47    specific structure. That gives flexibility to client to pass more
  48    parameters, if required.
  49
  50    Interface:
  51         int dmaengine_slave_config(struct dma_chan *chan,
  52                                   struct dma_slave_config *config)
  53
  54    Please see the dma_slave_config structure definition in dmaengine.h
  55    for a detailed explanation of the struct members. Please note
  56    that the 'direction' member will be going away as it duplicates the
  57    direction given in the prepare call.
  58
  59 3. Get a descriptor for transaction
  60
  61    For slave usage the various modes of slave transfers supported by the
  62    DMA-engine are:
  63
  64    slave_sg     - DMA a list of scatter gather buffers from/to a peripheral
  65    dma_cyclic   - Perform a cyclic DMA operation from/to a peripheral till the
  66                   operation is explicitly stopped.
  67    interleaved_dma - This is common to Slave as well as M2M clients. For slave
  68                  address of devices' fifo could be already known to the driver.
  69                  Various types of operations could be expressed by setting
  70                  appropriate values to the 'dma_interleaved_template' members.
  71
  72    A non-NULL return of this transfer API represents a "descriptor" for
  73    the given transaction.
  74
  75    Interface:
  76         struct dma_async_tx_descriptor *dmaengine_prep_slave_sg(
  77                 struct dma_chan *chan, struct scatterlist *sgl,
  78                 unsigned int sg_len, enum dma_data_direction direction,
  79                 unsigned long flags);
  80
  81         struct dma_async_tx_descriptor *dmaengine_prep_dma_cyclic(
  82                 struct dma_chan *chan, dma_addr_t buf_addr, size_t buf_len,
  83                 size_t period_len, enum dma_data_direction direction);
  84
  85         struct dma_async_tx_descriptor *dmaengine_prep_interleaved_dma(
  86                 struct dma_chan *chan, struct dma_interleaved_template *xt,
  87                 unsigned long flags);
  88
  89    The peripheral driver is expected to have mapped the scatterlist for
  90    the DMA operation prior to calling dmaengine_prep_slave_sg(), and must
  91    keep the scatterlist mapped until the DMA operation has completed.
  92    The scatterlist must be mapped using the DMA struct device.
  93    If a mapping needs to be synchronized later, dma_sync_*_for_*() must be
  94    called using the DMA struct device, too.
  95    So, normal setup should look like this:
  96
  97         nr_sg = dma_map_sg(chan->device->dev, sgl, sg_len);
  98         if (nr_sg == 0)
  99                 /* error */
 100
 101         desc = dmaengine_prep_slave_sg(chan, sgl, nr_sg, direction, flags);
 102
 103    Once a descriptor has been obtained, the callback information can be
 104    added and the descriptor must then be submitted. Some DMA engine
 105    drivers may hold a spinlock between a successful preparation and
 106    submission so it is important that these two operations are closely
 107    paired.
 108
 109    Note:
 110         Although the async_tx API specifies that completion callback
 111         routines cannot submit any new operations, this is not the
 112         case for slave/cyclic DMA.
 113
 114         For slave DMA, the subsequent transaction may not be available
 115         for submission prior to callback function being invoked, so
 116         slave DMA callbacks are permitted to prepare and submit a new
 117         transaction.
 118
 119         For cyclic DMA, a callback function may wish to terminate the
 120         DMA via dmaengine_terminate_async().
 121
 122         Therefore, it is important that DMA engine drivers drop any
 123         locks before calling the callback function which may cause a
 124         deadlock.
 125
 126         Note that callbacks will always be invoked from the DMA
 127         engines tasklet, never from interrupt context.
 128
 129 4. Submit the transaction
 130
 131    Once the descriptor has been prepared and the callback information
 132    added, it must be placed on the DMA engine drivers pending queue.
 133
 134    Interface:
 135         dma_cookie_t dmaengine_submit(struct dma_async_tx_descriptor *desc)
 136
 137    This returns a cookie can be used to check the progress of DMA engine
 138    activity via other DMA engine calls not covered in this document.
 139
 140    dmaengine_submit() will not start the DMA operation, it merely adds
 141    it to the pending queue. For this, see step 5, dma_async_issue_pending.
 142
 143 5. Issue pending DMA requests and wait for callback notification
 144
 145    The transactions in the pending queue can be activated by calling the
 146    issue_pending API. If channel is idle then the first transaction in
 147    queue is started and subsequent ones queued up.
 148
 149    On completion of each DMA operation, the next in queue is started and
 150    a tasklet triggered. The tasklet will then call the client driver
 151    completion callback routine for notification, if set.
 152
 153    Interface:
 154         void dma_async_issue_pending(struct dma_chan *chan);
 155
 156 Further APIs:
 157
 158 1. int dmaengine_terminate_sync(struct dma_chan *chan)
 159    int dmaengine_terminate_async(struct dma_chan *chan)
 160    int dmaengine_terminate_all(struct dma_chan *chan) /* DEPRECATED */
 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
 166    Two variants of this function are available.
 167
 168    dmaengine_terminate_async() might not wait until the DMA has been fully
 169    stopped or until any running complete callbacks have finished. But it is
 170    possible to call dmaengine_terminate_async() from atomic context or from
 171    within a complete callback. dmaengine_synchronize() must be called before it
 172    is safe to free the memory accessed by the DMA transfer or free resources
 173    accessed from within the complete callback.
 174
 175    dmaengine_terminate_sync() will wait for the transfer and any running
 176    complete callbacks to finish before it returns. But the function must not be
 177    called from atomic context or from within a complete callback.
 178
 179    dmaengine_terminate_all() is deprecated and should not be used in new code.
 180
 181 2. int dmaengine_pause(struct dma_chan *chan)
 182
 183    This pauses activity on the DMA channel without data loss.
 184
 185 3. int dmaengine_resume(struct dma_chan *chan)
 186
 187    Resume a previously paused DMA channel. It is invalid to resume a
 188    channel which is not currently paused.
 189
 190 4. enum dma_status dma_async_is_tx_complete(struct dma_chan *chan,
 191         dma_cookie_t cookie, dma_cookie_t *last, dma_cookie_t *used)
 192
 193    This can be used to check the status of the channel. Please see
 194    the documentation in include/linux/dmaengine.h for a more complete
 195    description of this API.
 196
 197    This can be used in conjunction with dma_async_is_complete() and
 198    the cookie returned from dmaengine_submit() to check for
 199    completion of a specific DMA transaction.
 200
 201    Note:
 202         Not all DMA engine drivers can return reliable information for
 203         a running DMA channel. It is recommended that DMA engine users
 204         pause or stop (via dmaengine_terminate_all()) the channel before
 205         using this API.
 206
 207 5. void dmaengine_synchronize(struct dma_chan *chan)
 208
 209   Synchronize the termination of the DMA channel to the current context.
 210
 211   This function should be used after dmaengine_terminate_async() to synchronize
 212   the termination of the DMA channel to the current context. The function will
 213   wait for the transfer and any running complete callbacks to finish before it
 214   returns.
 215
 216   If dmaengine_terminate_async() is used to stop the DMA channel this function
 217   must be called before it is safe to free memory accessed by previously
 218   submitted descriptors or to free any resources accessed within the complete
 219   callback of previously submitted descriptors.
 220
 221   The behavior of this function is undefined if dma_async_issue_pending() has
 222   been called between dmaengine_terminate_async() and this function.