1 ==============================
2 PXA/MMP - DMA Slave controller
3 ==============================
8 a) Transfers hot queuing
9 A driver submitting a transfer and issuing it should be granted the transfer
10 is queued even on a running DMA channel.
11 This implies that the queuing doesn't wait for the previous transfer end,
12 and that the descriptor chaining is not only done in the irq/tasklet code
13 triggered by the end of the transfer.
14 A transfer which is submitted and issued on a phy doesn't wait for a phy to
15 stop and restart, but is submitted on a "running channel". The other
16 drivers, especially mmp_pdma waited for the phy to stop before relaunching
19 b) All transfers having asked for confirmation should be signaled
20 Any issued transfer with DMA_PREP_INTERRUPT should trigger a callback call.
21 This implies that even if an irq/tasklet is triggered by end of tx1, but
22 at the time of irq/dma tx2 is already finished, tx1->complete() and
23 tx2->complete() should be called.
25 c) Channel running state
26 A driver should be able to query if a channel is running or not. For the
27 multimedia case, such as video capture, if a transfer is submitted and then
28 a check of the DMA channel reports a "stopped channel", the transfer should
29 not be issued until the next "start of frame interrupt", hence the need to
30 know if a channel is in running or stopped state.
32 d) Bandwidth guarantee
33 The PXA architecture has 4 levels of DMAs priorities : high, normal, low.
34 The high priorities get twice as much bandwidth as the normal, which get twice
35 as much as the low priorities.
36 A driver should be able to request a priority, especially the real-time
37 ones such as pxa_camera with (big) throughputs.
42 Same concept as in sa11x0 driver, ie. a driver was assigned a "virtual
43 channel" linked to the requestor line, and the physical DMA channel is
44 assigned on the fly when the transfer is issued.
46 b) Transfer anatomy for a scatter-gather transfer
50 +------------+-----+---------------+----------------+-----------------+
51 | desc-sg[0] | ... | desc-sg[last] | status updater | finisher/linker |
52 +------------+-----+---------------+----------------+-----------------+
54 This structure is pointed by dma->sg_cpu.
55 The descriptors are used as follows :
57 - desc-sg[i]: i-th descriptor, transferring the i-th sg
58 element to the video buffer scatter gather
61 Transfers a single u32 to a well known dma coherent memory to leave
62 a trace that this transfer is done. The "well known" is unique per
63 physical channel, meaning that a read of this value will tell which
64 is the last finished transfer at that point in time.
66 - finisher: has ddadr=DADDR_STOP, dcmd=ENDIRQEN
68 - linker: has ddadr= desc-sg[0] of next transfer, dcmd=0
70 c) Transfers hot-chaining
71 Suppose the running chain is:
76 +---------+----+---+ +----+----+----+---+
77 | d0 | .. | dN | l | | d0 | .. | dN | f |
78 +---------+----+-|-+ ^----+----+----+---+
82 After a call to dmaengine_submit(b3), the chain will look like:
86 Buffer 1 Buffer 2 Buffer 3
87 +---------+----+---+ +----+----+----+---+ +----+----+----+---+
88 | d0 | .. | dN | l | | d0 | .. | dN | l | | d0 | .. | dN | f |
89 +---------+----+-|-+ ^----+----+----+-|-+ ^----+----+----+---+
94 If while new_link was created the DMA channel stopped, it is _not_
95 restarted. Hot-chaining doesn't break the assumption that
96 dma_async_issue_pending() is to be used to ensure the transfer is actually started.
98 One exception to this rule :
100 - if Buffer1 and Buffer2 had all their addresses 8 bytes aligned
102 - and if Buffer3 has at least one address not 4 bytes aligned
104 - then hot-chaining cannot happen, as the channel must be stopped, the
105 "align bit" must be set, and the channel restarted As a consequence,
106 such a transfer tx_submit() will be queued on the submitted queue, and
107 this specific case if the DMA is already running in aligned mode.
109 d) Transfers completion updater
110 Each time a transfer is completed on a channel, an interrupt might be
111 generated or not, up to the client's request. But in each case, the last
112 descriptor of a transfer, the "status updater", will write the latest
113 transfer being completed into the physical channel's completion mark.
115 This will speed up residue calculation, for large transfers such as video
116 buffers which hold around 6k descriptors or more. This also allows without
117 any lock to find out what is the latest completed transfer in a running
120 e) Transfers completion, irq and tasklet
121 When a transfer flagged as "DMA_PREP_INTERRUPT" is finished, the dma irq
122 is raised. Upon this interrupt, a tasklet is scheduled for the physical
125 The tasklet is responsible for :
127 - reading the physical channel last updater mark
129 - calling all the transfer callbacks of finished transfers, based on
130 that mark, and each transfer flags.
132 If a transfer is completed while this handling is done, a dma irq will
133 be raised, and the tasklet will be scheduled once again, having a new
137 Residue granularity will be descriptor based. The issued but not completed
138 transfers will be scanned for all of their descriptors against the
139 currently running descriptor.
141 g) Most complicated case of driver's tx queues
142 The most tricky situation is when :
144 - there are not "acked" transfers (tx0)
146 - a driver submitted an aligned tx1, not chained
148 - a driver submitted an aligned tx2 => tx2 is cold chained to tx1
150 - a driver issued tx1+tx2 => channel is running in aligned mode
152 - a driver submitted an aligned tx3 => tx3 is hot-chained
154 - a driver submitted an unaligned tx4 => tx4 is put in submitted queue,
157 - a driver issued tx4 => tx4 is put in issued queue, not chained
159 - a driver submitted an aligned tx5 => tx5 is put in submitted queue, not
162 - a driver submitted an aligned tx6 => tx6 is put in submitted queue,
165 This translates into (after tx4 is issued) :
171 +-----+ +-----+ +-----+ +-----+
172 | tx1 | | tx2 | | tx3 | | tx4 |
173 +---|-+ ^---|-+ ^-----+ +-----+
183 - completed queue : empty
185 - allocated queue : tx0
187 It should be noted that after tx3 is completed, the channel is stopped, and
188 restarted in "unaligned mode" to handle tx4.
190 Author: Robert Jarzmik <robert.jarzmik@free.fr>