1 .. SPDX-License-Identifier: GPL-2.0
3 =====================================
4 Asynchronous Transfers/Transforms API
5 =====================================
14 3.1 General format of the API
15 3.2 Supported operations
16 3.3 Descriptor management
17 3.4 When does the operation execute?
18 3.5 When does the operation complete?
22 4 DMAENGINE DRIVER DEVELOPER NOTES
23 4.1 Conformance points
24 4.2 "My application needs exclusive control of hardware channels"
31 The async_tx API provides methods for describing a chain of asynchronous
32 bulk memory transfers/transforms with support for inter-transactional
33 dependencies. It is implemented as a dmaengine client that smooths over
34 the details of different hardware offload engine implementations. Code
35 that is written to the API can optimize for asynchronous operation and
36 the API will fit the chain of operations to the available offload
42 The API was initially designed to offload the memory copy and
43 xor-parity-calculations of the md-raid5 driver using the offload engines
44 present in the Intel(R) Xscale series of I/O processors. It also built
45 on the 'dmaengine' layer developed for offloading memory copies in the
46 network stack using Intel(R) I/OAT engines. The following design
47 features surfaced as a result:
49 1. implicit synchronous path: users of the API do not need to know if
50 the platform they are running on has offload capabilities. The
51 operation will be offloaded when an engine is available and carried out
52 in software otherwise.
53 2. cross channel dependency chains: the API allows a chain of dependent
54 operations to be submitted, like xor->copy->xor in the raid5 case. The
55 API automatically handles cases where the transition from one operation
56 to another implies a hardware channel switch.
57 3. dmaengine extensions to support multiple clients and operation types
63 3.1 General format of the API
64 -----------------------------
68 struct dma_async_tx_descriptor *
69 async_<operation>(<op specific parameters>, struct async_submit ctl *submit)
71 3.2 Supported operations
72 ------------------------
74 ======== ====================================================================
75 memcpy memory copy between a source and a destination buffer
76 memset fill a destination buffer with a byte value
77 xor xor a series of source buffers and write the result to a
79 xor_val xor a series of source buffers and set a flag if the
80 result is zero. The implementation attempts to prevent
82 pq generate the p+q (raid6 syndrome) from a series of source buffers
83 pq_val validate that a p and or q buffer are in sync with a given series of
85 datap (raid6_datap_recov) recover a raid6 data block and the p block
86 from the given sources
87 2data (raid6_2data_recov) recover 2 raid6 data blocks from the given
89 ======== ====================================================================
91 3.3 Descriptor management
92 -------------------------
94 The return value is non-NULL and points to a 'descriptor' when the operation
95 has been queued to execute asynchronously. Descriptors are recycled
96 resources, under control of the offload engine driver, to be reused as
97 operations complete. When an application needs to submit a chain of
98 operations it must guarantee that the descriptor is not automatically recycled
99 before the dependency is submitted. This requires that all descriptors be
100 acknowledged by the application before the offload engine driver is allowed to
101 recycle (or free) the descriptor. A descriptor can be acked by one of the
104 1. setting the ASYNC_TX_ACK flag if no child operations are to be submitted
105 2. submitting an unacknowledged descriptor as a dependency to another
106 async_tx call will implicitly set the acknowledged state.
107 3. calling async_tx_ack() on the descriptor.
109 3.4 When does the operation execute?
110 ------------------------------------
112 Operations do not immediately issue after return from the
113 async_<operation> call. Offload engine drivers batch operations to
114 improve performance by reducing the number of mmio cycles needed to
115 manage the channel. Once a driver-specific threshold is met the driver
116 automatically issues pending operations. An application can force this
117 event by calling async_tx_issue_pending_all(). This operates on all
118 channels since the application has no knowledge of channel to operation
121 3.5 When does the operation complete?
122 -------------------------------------
124 There are two methods for an application to learn about the completion
127 1. Call dma_wait_for_async_tx(). This call causes the CPU to spin while
128 it polls for the completion of the operation. It handles dependency
129 chains and issuing pending operations.
130 2. Specify a completion callback. The callback routine runs in tasklet
131 context if the offload engine driver supports interrupts, or it is
132 called in application context if the operation is carried out
133 synchronously in software. The callback can be set in the call to
134 async_<operation>, or when the application needs to submit a chain of
135 unknown length it can use the async_trigger_callback() routine to set a
136 completion interrupt/callback at the end of the chain.
141 1. Calls to async_<operation> are not permitted in IRQ context. Other
142 contexts are permitted provided constraint #2 is not violated.
143 2. Completion callback routines cannot submit new operations. This
144 results in recursion in the synchronous case and spin_locks being
145 acquired twice in the asynchronous case.
150 Perform a xor->copy->xor operation where each operation depends on the
151 result from the previous operation::
153 void callback(void *param)
155 struct completion *cmp = param;
160 void run_xor_copy_xor(struct page **xor_srcs,
162 struct page *xor_dest,
164 struct page *copy_src,
165 struct page *copy_dest,
168 struct dma_async_tx_descriptor *tx;
169 addr_conv_t addr_conv[xor_src_cnt];
170 struct async_submit_ctl submit;
171 addr_conv_t addr_conv[NDISKS];
172 struct completion cmp;
174 init_async_submit(&submit, ASYNC_TX_XOR_DROP_DST, NULL, NULL, NULL,
176 tx = async_xor(xor_dest, xor_srcs, 0, xor_src_cnt, xor_len, &submit)
178 submit->depend_tx = tx;
179 tx = async_memcpy(copy_dest, copy_src, 0, 0, copy_len, &submit);
181 init_completion(&cmp);
182 init_async_submit(&submit, ASYNC_TX_XOR_DROP_DST | ASYNC_TX_ACK, tx,
183 callback, &cmp, addr_conv);
184 tx = async_xor(xor_dest, xor_srcs, 0, xor_src_cnt, xor_len, &submit);
186 async_tx_issue_pending_all();
188 wait_for_completion(&cmp);
191 See include/linux/async_tx.h for more information on the flags. See the
192 ops_run_* and ops_complete_* routines in drivers/md/raid5.c for more
193 implementation examples.
195 4. Driver Development Notes
196 ===========================
198 4.1 Conformance points
199 ----------------------
201 There are a few conformance points required in dmaengine drivers to
202 accommodate assumptions made by applications using the async_tx API:
204 1. Completion callbacks are expected to happen in tasklet context
205 2. dma_async_tx_descriptor fields are never manipulated in IRQ context
206 3. Use async_tx_run_dependencies() in the descriptor clean up path to
207 handle submission of dependent operations
209 4.2 "My application needs exclusive control of hardware channels"
210 -----------------------------------------------------------------
212 Primarily this requirement arises from cases where a DMA engine driver
213 is being used to support device-to-memory operations. A channel that is
214 performing these operations cannot, for many platform specific reasons,
215 be shared. For these cases the dma_request_channel() interface is
220 struct dma_chan *dma_request_channel(dma_cap_mask_t mask,
221 dma_filter_fn filter_fn,
224 Where dma_filter_fn is defined as::
226 typedef bool (*dma_filter_fn)(struct dma_chan *chan, void *filter_param);
228 When the optional 'filter_fn' parameter is set to NULL
229 dma_request_channel simply returns the first channel that satisfies the
230 capability mask. Otherwise, when the mask parameter is insufficient for
231 specifying the necessary channel, the filter_fn routine can be used to
232 disposition the available channels in the system. The filter_fn routine
233 is called once for each free channel in the system. Upon seeing a
234 suitable channel filter_fn returns DMA_ACK which flags that channel to
235 be the return value from dma_request_channel. A channel allocated via
236 this interface is exclusive to the caller, until dma_release_channel()
239 The DMA_PRIVATE capability flag is used to tag dma devices that should
240 not be used by the general-purpose allocator. It can be set at
241 initialization time if it is known that a channel will always be
242 private. Alternatively, it is set when dma_request_channel() finds an
243 unused "public" channel.
245 A couple caveats to note when implementing a driver and consumer:
247 1. Once a channel has been privately allocated it will no longer be
248 considered by the general-purpose allocator even after a call to
249 dma_release_channel().
250 2. Since capabilities are specified at the device level a dma_device
251 with multiple channels will either have all channels public, or all
257 include/linux/dmaengine.h:
258 core header file for DMA drivers and api users
259 drivers/dma/dmaengine.c:
260 offload engine channel management routines
262 location for offload engine drivers
263 include/linux/async_tx.h:
264 core header file for the async_tx api
265 crypto/async_tx/async_tx.c:
266 async_tx interface to dmaengine and common code
267 crypto/async_tx/async_memcpy.c:
269 crypto/async_tx/async_xor.c:
270 xor and xor zero sum offload