aboutsummaryrefslogtreecommitdiffstats
path: root/drivers/dma/iop-adma.c
diff options
context:
space:
mode:
authorDan Williams <dan.j.williams@intel.com>2007-01-02 15:52:26 -0500
committerDan Williams <dan.j.williams@intel.com>2007-07-13 11:06:18 -0400
commitc211092313b90f898dec61f35207fc282d1eadc3 (patch)
tree30df0c81f207d0babb3fe56a17419f37e71e973a /drivers/dma/iop-adma.c
parentf6dff381af01006ffae3c23cd2e07e30584de0ec (diff)
dmaengine: driver for the iop32x, iop33x, and iop13xx raid engines
The Intel(R) IOP series of i/o processors integrate an Xscale core with raid acceleration engines. The capabilities per platform are: iop219: (2) copy engines iop321: (2) copy engines (1) xor and block fill engine iop33x: (2) copy and crc32c engines (1) xor, xor zero sum, pq, pq zero sum, and block fill engine iop34x (iop13xx): (2) copy, crc32c, xor, xor zero sum, and block fill engines (1) copy, crc32c, xor, xor zero sum, pq, pq zero sum, and block fill engine The driver supports the features of the async_tx api: * asynchronous notification of operation completion * implicit (interupt triggered) handling of inter-channel transaction dependencies The driver adapts to the platform it is running by two methods. 1/ #include <asm/arch/adma.h> which defines the hardware specific iop_chan_* and iop_desc_* routines as a series of static inline functions 2/ The private platform data attached to the platform_device defines the capabilities of the channels 20070626: Callbacks are run in a tasklet. Given the recent discussion on LKML about killing tasklets in favor of workqueues I did a quick conversion of the driver. Raid5 resync performance dropped from 50MB/s to 30MB/s, so the tasklet implementation remains until a generic softirq interface is available. Changelog: * fixed a slot allocation bug in do_iop13xx_adma_xor that caused too few slots to be requested eventually leading to data corruption * enabled the slot allocation routine to attempt to free slots before returning -ENOMEM * switched the cleanup routine to solely use the software chain and the status register to determine if a descriptor is complete. This is necessary to support other IOP engines that do not have status writeback capability * make the driver iop generic * modified the allocation routines to understand allocating a group of slots for a single operation * added a null xor initialization operation for the xor only channel on iop3xx * support xor operations on buffers larger than the hardware maximum * split the do_* routines into separate prep, src/dest set, submit stages * added async_tx support (dependent operations initiation at cleanup time) * simplified group handling * added interrupt support (callbacks via tasklets) * brought the pending depth inline with ioat (i.e. 4 descriptors) * drop dma mapping methods, suggested by Chris Leech * don't use inline in C files, Adrian Bunk * remove static tasklet declarations * make iop_adma_alloc_slots easier to read and remove chances for a corrupted descriptor chain * fix locking bug in iop_adma_alloc_chan_resources, Benjamin Herrenschmidt * convert capabilities over to dma_cap_mask_t * fixup sparse warnings * add descriptor flush before iop_chan_enable * checkpatch.pl fixes * gpl v2 only correction * move set_src, set_dest, submit to async_tx methods * move group_list and phys to async_tx Cc: Russell King <rmk@arm.linux.org.uk> Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Diffstat (limited to 'drivers/dma/iop-adma.c')
-rw-r--r--drivers/dma/iop-adma.c1467
1 files changed, 1467 insertions, 0 deletions
diff --git a/drivers/dma/iop-adma.c b/drivers/dma/iop-adma.c
new file mode 100644
index 000000000000..5a1d426744d6
--- /dev/null
+++ b/drivers/dma/iop-adma.c
@@ -0,0 +1,1467 @@
1/*
2 * offload engine driver for the Intel Xscale series of i/o processors
3 * Copyright © 2006, Intel Corporation.
4 *
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
12 * more details.
13 *
14 * You should have received a copy of the GNU General Public License along with
15 * this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
17 *
18 */
19
20/*
21 * This driver supports the asynchrounous DMA copy and RAID engines available
22 * on the Intel Xscale(R) family of I/O Processors (IOP 32x, 33x, 134x)
23 */
24
25#include <linux/init.h>
26#include <linux/module.h>
27#include <linux/async_tx.h>
28#include <linux/delay.h>
29#include <linux/dma-mapping.h>
30#include <linux/spinlock.h>
31#include <linux/interrupt.h>
32#include <linux/platform_device.h>
33#include <linux/memory.h>
34#include <linux/ioport.h>
35
36#include <asm/arch/adma.h>
37
38#define to_iop_adma_chan(chan) container_of(chan, struct iop_adma_chan, common)
39#define to_iop_adma_device(dev) \
40 container_of(dev, struct iop_adma_device, common)
41#define tx_to_iop_adma_slot(tx) \
42 container_of(tx, struct iop_adma_desc_slot, async_tx)
43
44/**
45 * iop_adma_free_slots - flags descriptor slots for reuse
46 * @slot: Slot to free
47 * Caller must hold &iop_chan->lock while calling this function
48 */
49static void iop_adma_free_slots(struct iop_adma_desc_slot *slot)
50{
51 int stride = slot->slots_per_op;
52
53 while (stride--) {
54 slot->slots_per_op = 0;
55 slot = list_entry(slot->slot_node.next,
56 struct iop_adma_desc_slot,
57 slot_node);
58 }
59}
60
61static dma_cookie_t
62iop_adma_run_tx_complete_actions(struct iop_adma_desc_slot *desc,
63 struct iop_adma_chan *iop_chan, dma_cookie_t cookie)
64{
65 BUG_ON(desc->async_tx.cookie < 0);
66 spin_lock_bh(&desc->async_tx.lock);
67 if (desc->async_tx.cookie > 0) {
68 cookie = desc->async_tx.cookie;
69 desc->async_tx.cookie = 0;
70
71 /* call the callback (must not sleep or submit new
72 * operations to this channel)
73 */
74 if (desc->async_tx.callback)
75 desc->async_tx.callback(
76 desc->async_tx.callback_param);
77
78 /* unmap dma addresses
79 * (unmap_single vs unmap_page?)
80 */
81 if (desc->group_head && desc->unmap_len) {
82 struct iop_adma_desc_slot *unmap = desc->group_head;
83 struct device *dev =
84 &iop_chan->device->pdev->dev;
85 u32 len = unmap->unmap_len;
86 u32 src_cnt = unmap->unmap_src_cnt;
87 dma_addr_t addr = iop_desc_get_dest_addr(unmap,
88 iop_chan);
89
90 dma_unmap_page(dev, addr, len, DMA_FROM_DEVICE);
91 while (src_cnt--) {
92 addr = iop_desc_get_src_addr(unmap,
93 iop_chan,
94 src_cnt);
95 dma_unmap_page(dev, addr, len,
96 DMA_TO_DEVICE);
97 }
98 desc->group_head = NULL;
99 }
100 }
101
102 /* run dependent operations */
103 async_tx_run_dependencies(&desc->async_tx);
104 spin_unlock_bh(&desc->async_tx.lock);
105
106 return cookie;
107}
108
109static int
110iop_adma_clean_slot(struct iop_adma_desc_slot *desc,
111 struct iop_adma_chan *iop_chan)
112{
113 /* the client is allowed to attach dependent operations
114 * until 'ack' is set
115 */
116 if (!desc->async_tx.ack)
117 return 0;
118
119 /* leave the last descriptor in the chain
120 * so we can append to it
121 */
122 if (desc->chain_node.next == &iop_chan->chain)
123 return 1;
124
125 dev_dbg(iop_chan->device->common.dev,
126 "\tfree slot: %d slots_per_op: %d\n",
127 desc->idx, desc->slots_per_op);
128
129 list_del(&desc->chain_node);
130 iop_adma_free_slots(desc);
131
132 return 0;
133}
134
135static void __iop_adma_slot_cleanup(struct iop_adma_chan *iop_chan)
136{
137 struct iop_adma_desc_slot *iter, *_iter, *grp_start = NULL;
138 dma_cookie_t cookie = 0;
139 u32 current_desc = iop_chan_get_current_descriptor(iop_chan);
140 int busy = iop_chan_is_busy(iop_chan);
141 int seen_current = 0, slot_cnt = 0, slots_per_op = 0;
142
143 dev_dbg(iop_chan->device->common.dev, "%s\n", __FUNCTION__);
144 /* free completed slots from the chain starting with
145 * the oldest descriptor
146 */
147 list_for_each_entry_safe(iter, _iter, &iop_chan->chain,
148 chain_node) {
149 pr_debug("\tcookie: %d slot: %d busy: %d "
150 "this_desc: %#x next_desc: %#x ack: %d\n",
151 iter->async_tx.cookie, iter->idx, busy,
152 iter->async_tx.phys, iop_desc_get_next_desc(iter),
153 iter->async_tx.ack);
154 prefetch(_iter);
155 prefetch(&_iter->async_tx);
156
157 /* do not advance past the current descriptor loaded into the
158 * hardware channel, subsequent descriptors are either in
159 * process or have not been submitted
160 */
161 if (seen_current)
162 break;
163
164 /* stop the search if we reach the current descriptor and the
165 * channel is busy, or if it appears that the current descriptor
166 * needs to be re-read (i.e. has been appended to)
167 */
168 if (iter->async_tx.phys == current_desc) {
169 BUG_ON(seen_current++);
170 if (busy || iop_desc_get_next_desc(iter))
171 break;
172 }
173
174 /* detect the start of a group transaction */
175 if (!slot_cnt && !slots_per_op) {
176 slot_cnt = iter->slot_cnt;
177 slots_per_op = iter->slots_per_op;
178 if (slot_cnt <= slots_per_op) {
179 slot_cnt = 0;
180 slots_per_op = 0;
181 }
182 }
183
184 if (slot_cnt) {
185 pr_debug("\tgroup++\n");
186 if (!grp_start)
187 grp_start = iter;
188 slot_cnt -= slots_per_op;
189 }
190
191 /* all the members of a group are complete */
192 if (slots_per_op != 0 && slot_cnt == 0) {
193 struct iop_adma_desc_slot *grp_iter, *_grp_iter;
194 int end_of_chain = 0;
195 pr_debug("\tgroup end\n");
196
197 /* collect the total results */
198 if (grp_start->xor_check_result) {
199 u32 zero_sum_result = 0;
200 slot_cnt = grp_start->slot_cnt;
201 grp_iter = grp_start;
202
203 list_for_each_entry_from(grp_iter,
204 &iop_chan->chain, chain_node) {
205 zero_sum_result |=
206 iop_desc_get_zero_result(grp_iter);
207 pr_debug("\titer%d result: %d\n",
208 grp_iter->idx, zero_sum_result);
209 slot_cnt -= slots_per_op;
210 if (slot_cnt == 0)
211 break;
212 }
213 pr_debug("\tgrp_start->xor_check_result: %p\n",
214 grp_start->xor_check_result);
215 *grp_start->xor_check_result = zero_sum_result;
216 }
217
218 /* clean up the group */
219 slot_cnt = grp_start->slot_cnt;
220 grp_iter = grp_start;
221 list_for_each_entry_safe_from(grp_iter, _grp_iter,
222 &iop_chan->chain, chain_node) {
223 cookie = iop_adma_run_tx_complete_actions(
224 grp_iter, iop_chan, cookie);
225
226 slot_cnt -= slots_per_op;
227 end_of_chain = iop_adma_clean_slot(grp_iter,
228 iop_chan);
229
230 if (slot_cnt == 0 || end_of_chain)
231 break;
232 }
233
234 /* the group should be complete at this point */
235 BUG_ON(slot_cnt);
236
237 slots_per_op = 0;
238 grp_start = NULL;
239 if (end_of_chain)
240 break;
241 else
242 continue;
243 } else if (slots_per_op) /* wait for group completion */
244 continue;
245
246 /* write back zero sum results (single descriptor case) */
247 if (iter->xor_check_result && iter->async_tx.cookie)
248 *iter->xor_check_result =
249 iop_desc_get_zero_result(iter);
250
251 cookie = iop_adma_run_tx_complete_actions(
252 iter, iop_chan, cookie);
253
254 if (iop_adma_clean_slot(iter, iop_chan))
255 break;
256 }
257
258 BUG_ON(!seen_current);
259
260 iop_chan_idle(busy, iop_chan);
261
262 if (cookie > 0) {
263 iop_chan->completed_cookie = cookie;
264 pr_debug("\tcompleted cookie %d\n", cookie);
265 }
266}
267
268static void
269iop_adma_slot_cleanup(struct iop_adma_chan *iop_chan)
270{
271 spin_lock_bh(&iop_chan->lock);
272 __iop_adma_slot_cleanup(iop_chan);
273 spin_unlock_bh(&iop_chan->lock);
274}
275
276static void iop_adma_tasklet(unsigned long data)
277{
278 struct iop_adma_chan *chan = (struct iop_adma_chan *) data;
279 __iop_adma_slot_cleanup(chan);
280}
281
282static struct iop_adma_desc_slot *
283iop_adma_alloc_slots(struct iop_adma_chan *iop_chan, int num_slots,
284 int slots_per_op)
285{
286 struct iop_adma_desc_slot *iter, *_iter, *alloc_start = NULL;
287 struct list_head chain = LIST_HEAD_INIT(chain);
288 int slots_found, retry = 0;
289
290 /* start search from the last allocated descrtiptor
291 * if a contiguous allocation can not be found start searching
292 * from the beginning of the list
293 */
294retry:
295 slots_found = 0;
296 if (retry == 0)
297 iter = iop_chan->last_used;
298 else
299 iter = list_entry(&iop_chan->all_slots,
300 struct iop_adma_desc_slot,
301 slot_node);
302
303 list_for_each_entry_safe_continue(
304 iter, _iter, &iop_chan->all_slots, slot_node) {
305 prefetch(_iter);
306 prefetch(&_iter->async_tx);
307 if (iter->slots_per_op) {
308 /* give up after finding the first busy slot
309 * on the second pass through the list
310 */
311 if (retry)
312 break;
313
314 slots_found = 0;
315 continue;
316 }
317
318 /* start the allocation if the slot is correctly aligned */
319 if (!slots_found++) {
320 if (iop_desc_is_aligned(iter, slots_per_op))
321 alloc_start = iter;
322 else {
323 slots_found = 0;
324 continue;
325 }
326 }
327
328 if (slots_found == num_slots) {
329 struct iop_adma_desc_slot *alloc_tail = NULL;
330 struct iop_adma_desc_slot *last_used = NULL;
331 iter = alloc_start;
332 while (num_slots) {
333 int i;
334 dev_dbg(iop_chan->device->common.dev,
335 "allocated slot: %d "
336 "(desc %p phys: %#x) slots_per_op %d\n",
337 iter->idx, iter->hw_desc,
338 iter->async_tx.phys, slots_per_op);
339
340 /* pre-ack all but the last descriptor */
341 if (num_slots != slots_per_op)
342 iter->async_tx.ack = 1;
343 else
344 iter->async_tx.ack = 0;
345
346 list_add_tail(&iter->chain_node, &chain);
347 alloc_tail = iter;
348 iter->async_tx.cookie = 0;
349 iter->slot_cnt = num_slots;
350 iter->xor_check_result = NULL;
351 for (i = 0; i < slots_per_op; i++) {
352 iter->slots_per_op = slots_per_op - i;
353 last_used = iter;
354 iter = list_entry(iter->slot_node.next,
355 struct iop_adma_desc_slot,
356 slot_node);
357 }
358 num_slots -= slots_per_op;
359 }
360 alloc_tail->group_head = alloc_start;
361 alloc_tail->async_tx.cookie = -EBUSY;
362 list_splice(&chain, &alloc_tail->async_tx.tx_list);
363 iop_chan->last_used = last_used;
364 iop_desc_clear_next_desc(alloc_start);
365 iop_desc_clear_next_desc(alloc_tail);
366 return alloc_tail;
367 }
368 }
369 if (!retry++)
370 goto retry;
371
372 /* try to free some slots if the allocation fails */
373 tasklet_schedule(&iop_chan->irq_tasklet);
374
375 return NULL;
376}
377
378static dma_cookie_t
379iop_desc_assign_cookie(struct iop_adma_chan *iop_chan,
380 struct iop_adma_desc_slot *desc)
381{
382 dma_cookie_t cookie = iop_chan->common.cookie;
383 cookie++;
384 if (cookie < 0)
385 cookie = 1;
386 iop_chan->common.cookie = desc->async_tx.cookie = cookie;
387 return cookie;
388}
389
390static void iop_adma_check_threshold(struct iop_adma_chan *iop_chan)
391{
392 dev_dbg(iop_chan->device->common.dev, "pending: %d\n",
393 iop_chan->pending);
394
395 if (iop_chan->pending >= IOP_ADMA_THRESHOLD) {
396 iop_chan->pending = 0;
397 iop_chan_append(iop_chan);
398 }
399}
400
401static dma_cookie_t
402iop_adma_tx_submit(struct dma_async_tx_descriptor *tx)
403{
404 struct iop_adma_desc_slot *sw_desc = tx_to_iop_adma_slot(tx);
405 struct iop_adma_chan *iop_chan = to_iop_adma_chan(tx->chan);
406 struct iop_adma_desc_slot *grp_start, *old_chain_tail;
407 int slot_cnt;
408 int slots_per_op;
409 dma_cookie_t cookie;
410
411 grp_start = sw_desc->group_head;
412 slot_cnt = grp_start->slot_cnt;
413 slots_per_op = grp_start->slots_per_op;
414
415 spin_lock_bh(&iop_chan->lock);
416 cookie = iop_desc_assign_cookie(iop_chan, sw_desc);
417
418 old_chain_tail = list_entry(iop_chan->chain.prev,
419 struct iop_adma_desc_slot, chain_node);
420 list_splice_init(&sw_desc->async_tx.tx_list,
421 &old_chain_tail->chain_node);
422
423 /* fix up the hardware chain */
424 iop_desc_set_next_desc(old_chain_tail, grp_start->async_tx.phys);
425
426 /* 1/ don't add pre-chained descriptors
427 * 2/ dummy read to flush next_desc write
428 */
429 BUG_ON(iop_desc_get_next_desc(sw_desc));
430
431 /* increment the pending count by the number of slots
432 * memcpy operations have a 1:1 (slot:operation) relation
433 * other operations are heavier and will pop the threshold
434 * more often.
435 */
436 iop_chan->pending += slot_cnt;
437 iop_adma_check_threshold(iop_chan);
438 spin_unlock_bh(&iop_chan->lock);
439
440 dev_dbg(iop_chan->device->common.dev, "%s cookie: %d slot: %d\n",
441 __FUNCTION__, sw_desc->async_tx.cookie, sw_desc->idx);
442
443 return cookie;
444}
445
446static void
447iop_adma_set_dest(dma_addr_t addr, struct dma_async_tx_descriptor *tx,
448 int index)
449{
450 struct iop_adma_desc_slot *sw_desc = tx_to_iop_adma_slot(tx);
451 struct iop_adma_chan *iop_chan = to_iop_adma_chan(tx->chan);
452
453 /* to do: support transfers lengths > IOP_ADMA_MAX_BYTE_COUNT */
454 iop_desc_set_dest_addr(sw_desc->group_head, iop_chan, addr);
455}
456
457static void iop_chan_start_null_memcpy(struct iop_adma_chan *iop_chan);
458static void iop_chan_start_null_xor(struct iop_adma_chan *iop_chan);
459
460/* returns the number of allocated descriptors */
461static int iop_adma_alloc_chan_resources(struct dma_chan *chan)
462{
463 char *hw_desc;
464 int idx;
465 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
466 struct iop_adma_desc_slot *slot = NULL;
467 int init = iop_chan->slots_allocated ? 0 : 1;
468 struct iop_adma_platform_data *plat_data =
469 iop_chan->device->pdev->dev.platform_data;
470 int num_descs_in_pool = plat_data->pool_size/IOP_ADMA_SLOT_SIZE;
471
472 /* Allocate descriptor slots */
473 do {
474 idx = iop_chan->slots_allocated;
475 if (idx == num_descs_in_pool)
476 break;
477
478 slot = kzalloc(sizeof(*slot), GFP_KERNEL);
479 if (!slot) {
480 printk(KERN_INFO "IOP ADMA Channel only initialized"
481 " %d descriptor slots", idx);
482 break;
483 }
484 hw_desc = (char *) iop_chan->device->dma_desc_pool_virt;
485 slot->hw_desc = (void *) &hw_desc[idx * IOP_ADMA_SLOT_SIZE];
486
487 dma_async_tx_descriptor_init(&slot->async_tx, chan);
488 slot->async_tx.tx_submit = iop_adma_tx_submit;
489 slot->async_tx.tx_set_dest = iop_adma_set_dest;
490 INIT_LIST_HEAD(&slot->chain_node);
491 INIT_LIST_HEAD(&slot->slot_node);
492 INIT_LIST_HEAD(&slot->async_tx.tx_list);
493 hw_desc = (char *) iop_chan->device->dma_desc_pool;
494 slot->async_tx.phys =
495 (dma_addr_t) &hw_desc[idx * IOP_ADMA_SLOT_SIZE];
496 slot->idx = idx;
497
498 spin_lock_bh(&iop_chan->lock);
499 iop_chan->slots_allocated++;
500 list_add_tail(&slot->slot_node, &iop_chan->all_slots);
501 spin_unlock_bh(&iop_chan->lock);
502 } while (iop_chan->slots_allocated < num_descs_in_pool);
503
504 if (idx && !iop_chan->last_used)
505 iop_chan->last_used = list_entry(iop_chan->all_slots.next,
506 struct iop_adma_desc_slot,
507 slot_node);
508
509 dev_dbg(iop_chan->device->common.dev,
510 "allocated %d descriptor slots last_used: %p\n",
511 iop_chan->slots_allocated, iop_chan->last_used);
512
513 /* initialize the channel and the chain with a null operation */
514 if (init) {
515 if (dma_has_cap(DMA_MEMCPY,
516 iop_chan->device->common.cap_mask))
517 iop_chan_start_null_memcpy(iop_chan);
518 else if (dma_has_cap(DMA_XOR,
519 iop_chan->device->common.cap_mask))
520 iop_chan_start_null_xor(iop_chan);
521 else
522 BUG();
523 }
524
525 return (idx > 0) ? idx : -ENOMEM;
526}
527
528static struct dma_async_tx_descriptor *
529iop_adma_prep_dma_interrupt(struct dma_chan *chan)
530{
531 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
532 struct iop_adma_desc_slot *sw_desc, *grp_start;
533 int slot_cnt, slots_per_op;
534
535 dev_dbg(iop_chan->device->common.dev, "%s\n", __FUNCTION__);
536
537 spin_lock_bh(&iop_chan->lock);
538 slot_cnt = iop_chan_interrupt_slot_count(&slots_per_op, iop_chan);
539 sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
540 if (sw_desc) {
541 grp_start = sw_desc->group_head;
542 iop_desc_init_interrupt(grp_start, iop_chan);
543 grp_start->unmap_len = 0;
544 }
545 spin_unlock_bh(&iop_chan->lock);
546
547 return sw_desc ? &sw_desc->async_tx : NULL;
548}
549
550static void
551iop_adma_memcpy_set_src(dma_addr_t addr, struct dma_async_tx_descriptor *tx,
552 int index)
553{
554 struct iop_adma_desc_slot *sw_desc = tx_to_iop_adma_slot(tx);
555 struct iop_adma_desc_slot *grp_start = sw_desc->group_head;
556
557 iop_desc_set_memcpy_src_addr(grp_start, addr);
558}
559
560static struct dma_async_tx_descriptor *
561iop_adma_prep_dma_memcpy(struct dma_chan *chan, size_t len, int int_en)
562{
563 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
564 struct iop_adma_desc_slot *sw_desc, *grp_start;
565 int slot_cnt, slots_per_op;
566
567 if (unlikely(!len))
568 return NULL;
569 BUG_ON(unlikely(len > IOP_ADMA_MAX_BYTE_COUNT));
570
571 dev_dbg(iop_chan->device->common.dev, "%s len: %u\n",
572 __FUNCTION__, len);
573
574 spin_lock_bh(&iop_chan->lock);
575 slot_cnt = iop_chan_memcpy_slot_count(len, &slots_per_op);
576 sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
577 if (sw_desc) {
578 grp_start = sw_desc->group_head;
579 iop_desc_init_memcpy(grp_start, int_en);
580 iop_desc_set_byte_count(grp_start, iop_chan, len);
581 sw_desc->unmap_src_cnt = 1;
582 sw_desc->unmap_len = len;
583 sw_desc->async_tx.tx_set_src = iop_adma_memcpy_set_src;
584 }
585 spin_unlock_bh(&iop_chan->lock);
586
587 return sw_desc ? &sw_desc->async_tx : NULL;
588}
589
590static struct dma_async_tx_descriptor *
591iop_adma_prep_dma_memset(struct dma_chan *chan, int value, size_t len,
592 int int_en)
593{
594 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
595 struct iop_adma_desc_slot *sw_desc, *grp_start;
596 int slot_cnt, slots_per_op;
597
598 if (unlikely(!len))
599 return NULL;
600 BUG_ON(unlikely(len > IOP_ADMA_MAX_BYTE_COUNT));
601
602 dev_dbg(iop_chan->device->common.dev, "%s len: %u\n",
603 __FUNCTION__, len);
604
605 spin_lock_bh(&iop_chan->lock);
606 slot_cnt = iop_chan_memset_slot_count(len, &slots_per_op);
607 sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
608 if (sw_desc) {
609 grp_start = sw_desc->group_head;
610 iop_desc_init_memset(grp_start, int_en);
611 iop_desc_set_byte_count(grp_start, iop_chan, len);
612 iop_desc_set_block_fill_val(grp_start, value);
613 sw_desc->unmap_src_cnt = 1;
614 sw_desc->unmap_len = len;
615 }
616 spin_unlock_bh(&iop_chan->lock);
617
618 return sw_desc ? &sw_desc->async_tx : NULL;
619}
620
621static void
622iop_adma_xor_set_src(dma_addr_t addr, struct dma_async_tx_descriptor *tx,
623 int index)
624{
625 struct iop_adma_desc_slot *sw_desc = tx_to_iop_adma_slot(tx);
626 struct iop_adma_desc_slot *grp_start = sw_desc->group_head;
627
628 iop_desc_set_xor_src_addr(grp_start, index, addr);
629}
630
631static struct dma_async_tx_descriptor *
632iop_adma_prep_dma_xor(struct dma_chan *chan, unsigned int src_cnt, size_t len,
633 int int_en)
634{
635 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
636 struct iop_adma_desc_slot *sw_desc, *grp_start;
637 int slot_cnt, slots_per_op;
638
639 if (unlikely(!len))
640 return NULL;
641 BUG_ON(unlikely(len > IOP_ADMA_XOR_MAX_BYTE_COUNT));
642
643 dev_dbg(iop_chan->device->common.dev,
644 "%s src_cnt: %d len: %u int_en: %d\n",
645 __FUNCTION__, src_cnt, len, int_en);
646
647 spin_lock_bh(&iop_chan->lock);
648 slot_cnt = iop_chan_xor_slot_count(len, src_cnt, &slots_per_op);
649 sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
650 if (sw_desc) {
651 grp_start = sw_desc->group_head;
652 iop_desc_init_xor(grp_start, src_cnt, int_en);
653 iop_desc_set_byte_count(grp_start, iop_chan, len);
654 sw_desc->unmap_src_cnt = src_cnt;
655 sw_desc->unmap_len = len;
656 sw_desc->async_tx.tx_set_src = iop_adma_xor_set_src;
657 }
658 spin_unlock_bh(&iop_chan->lock);
659
660 return sw_desc ? &sw_desc->async_tx : NULL;
661}
662
663static void
664iop_adma_xor_zero_sum_set_src(dma_addr_t addr,
665 struct dma_async_tx_descriptor *tx,
666 int index)
667{
668 struct iop_adma_desc_slot *sw_desc = tx_to_iop_adma_slot(tx);
669 struct iop_adma_desc_slot *grp_start = sw_desc->group_head;
670
671 iop_desc_set_zero_sum_src_addr(grp_start, index, addr);
672}
673
674static struct dma_async_tx_descriptor *
675iop_adma_prep_dma_zero_sum(struct dma_chan *chan, unsigned int src_cnt,
676 size_t len, u32 *result, int int_en)
677{
678 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
679 struct iop_adma_desc_slot *sw_desc, *grp_start;
680 int slot_cnt, slots_per_op;
681
682 if (unlikely(!len))
683 return NULL;
684
685 dev_dbg(iop_chan->device->common.dev, "%s src_cnt: %d len: %u\n",
686 __FUNCTION__, src_cnt, len);
687
688 spin_lock_bh(&iop_chan->lock);
689 slot_cnt = iop_chan_zero_sum_slot_count(len, src_cnt, &slots_per_op);
690 sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
691 if (sw_desc) {
692 grp_start = sw_desc->group_head;
693 iop_desc_init_zero_sum(grp_start, src_cnt, int_en);
694 iop_desc_set_zero_sum_byte_count(grp_start, len);
695 grp_start->xor_check_result = result;
696 pr_debug("\t%s: grp_start->xor_check_result: %p\n",
697 __FUNCTION__, grp_start->xor_check_result);
698 sw_desc->unmap_src_cnt = src_cnt;
699 sw_desc->unmap_len = len;
700 sw_desc->async_tx.tx_set_src = iop_adma_xor_zero_sum_set_src;
701 }
702 spin_unlock_bh(&iop_chan->lock);
703
704 return sw_desc ? &sw_desc->async_tx : NULL;
705}
706
707static void iop_adma_dependency_added(struct dma_chan *chan)
708{
709 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
710 tasklet_schedule(&iop_chan->irq_tasklet);
711}
712
713static void iop_adma_free_chan_resources(struct dma_chan *chan)
714{
715 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
716 struct iop_adma_desc_slot *iter, *_iter;
717 int in_use_descs = 0;
718
719 iop_adma_slot_cleanup(iop_chan);
720
721 spin_lock_bh(&iop_chan->lock);
722 list_for_each_entry_safe(iter, _iter, &iop_chan->chain,
723 chain_node) {
724 in_use_descs++;
725 list_del(&iter->chain_node);
726 }
727 list_for_each_entry_safe_reverse(
728 iter, _iter, &iop_chan->all_slots, slot_node) {
729 list_del(&iter->slot_node);
730 kfree(iter);
731 iop_chan->slots_allocated--;
732 }
733 iop_chan->last_used = NULL;
734
735 dev_dbg(iop_chan->device->common.dev, "%s slots_allocated %d\n",
736 __FUNCTION__, iop_chan->slots_allocated);
737 spin_unlock_bh(&iop_chan->lock);
738
739 /* one is ok since we left it on there on purpose */
740 if (in_use_descs > 1)
741 printk(KERN_ERR "IOP: Freeing %d in use descriptors!\n",
742 in_use_descs - 1);
743}
744
745/**
746 * iop_adma_is_complete - poll the status of an ADMA transaction
747 * @chan: ADMA channel handle
748 * @cookie: ADMA transaction identifier
749 */
750static enum dma_status iop_adma_is_complete(struct dma_chan *chan,
751 dma_cookie_t cookie,
752 dma_cookie_t *done,
753 dma_cookie_t *used)
754{
755 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
756 dma_cookie_t last_used;
757 dma_cookie_t last_complete;
758 enum dma_status ret;
759
760 last_used = chan->cookie;
761 last_complete = iop_chan->completed_cookie;
762
763 if (done)
764 *done = last_complete;
765 if (used)
766 *used = last_used;
767
768 ret = dma_async_is_complete(cookie, last_complete, last_used);
769 if (ret == DMA_SUCCESS)
770 return ret;
771
772 iop_adma_slot_cleanup(iop_chan);
773
774 last_used = chan->cookie;
775 last_complete = iop_chan->completed_cookie;
776
777 if (done)
778 *done = last_complete;
779 if (used)
780 *used = last_used;
781
782 return dma_async_is_complete(cookie, last_complete, last_used);
783}
784
785static irqreturn_t iop_adma_eot_handler(int irq, void *data)
786{
787 struct iop_adma_chan *chan = data;
788
789 dev_dbg(chan->device->common.dev, "%s\n", __FUNCTION__);
790
791 tasklet_schedule(&chan->irq_tasklet);
792
793 iop_adma_device_clear_eot_status(chan);
794
795 return IRQ_HANDLED;
796}
797
798static irqreturn_t iop_adma_eoc_handler(int irq, void *data)
799{
800 struct iop_adma_chan *chan = data;
801
802 dev_dbg(chan->device->common.dev, "%s\n", __FUNCTION__);
803
804 tasklet_schedule(&chan->irq_tasklet);
805
806 iop_adma_device_clear_eoc_status(chan);
807
808 return IRQ_HANDLED;
809}
810
811static irqreturn_t iop_adma_err_handler(int irq, void *data)
812{
813 struct iop_adma_chan *chan = data;
814 unsigned long status = iop_chan_get_status(chan);
815
816 dev_printk(KERN_ERR, chan->device->common.dev,
817 "error ( %s%s%s%s%s%s%s)\n",
818 iop_is_err_int_parity(status, chan) ? "int_parity " : "",
819 iop_is_err_mcu_abort(status, chan) ? "mcu_abort " : "",
820 iop_is_err_int_tabort(status, chan) ? "int_tabort " : "",
821 iop_is_err_int_mabort(status, chan) ? "int_mabort " : "",
822 iop_is_err_pci_tabort(status, chan) ? "pci_tabort " : "",
823 iop_is_err_pci_mabort(status, chan) ? "pci_mabort " : "",
824 iop_is_err_split_tx(status, chan) ? "split_tx " : "");
825
826 iop_adma_device_clear_err_status(chan);
827
828 BUG();
829
830 return IRQ_HANDLED;
831}
832
833static void iop_adma_issue_pending(struct dma_chan *chan)
834{
835 struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
836
837 if (iop_chan->pending) {
838 iop_chan->pending = 0;
839 iop_chan_append(iop_chan);
840 }
841}
842
843/*
844 * Perform a transaction to verify the HW works.
845 */
846#define IOP_ADMA_TEST_SIZE 2000
847
848static int __devinit iop_adma_memcpy_self_test(struct iop_adma_device *device)
849{
850 int i;
851 void *src, *dest;
852 dma_addr_t src_dma, dest_dma;
853 struct dma_chan *dma_chan;
854 dma_cookie_t cookie;
855 struct dma_async_tx_descriptor *tx;
856 int err = 0;
857 struct iop_adma_chan *iop_chan;
858
859 dev_dbg(device->common.dev, "%s\n", __FUNCTION__);
860
861 src = kzalloc(sizeof(u8) * IOP_ADMA_TEST_SIZE, GFP_KERNEL);
862 if (!src)
863 return -ENOMEM;
864 dest = kzalloc(sizeof(u8) * IOP_ADMA_TEST_SIZE, GFP_KERNEL);
865 if (!dest) {
866 kfree(src);
867 return -ENOMEM;
868 }
869
870 /* Fill in src buffer */
871 for (i = 0; i < IOP_ADMA_TEST_SIZE; i++)
872 ((u8 *) src)[i] = (u8)i;
873
874 memset(dest, 0, IOP_ADMA_TEST_SIZE);
875
876 /* Start copy, using first DMA channel */
877 dma_chan = container_of(device->common.channels.next,
878 struct dma_chan,
879 device_node);
880 if (iop_adma_alloc_chan_resources(dma_chan) < 1) {
881 err = -ENODEV;
882 goto out;
883 }
884
885 tx = iop_adma_prep_dma_memcpy(dma_chan, IOP_ADMA_TEST_SIZE, 1);
886 dest_dma = dma_map_single(dma_chan->device->dev, dest,
887 IOP_ADMA_TEST_SIZE, DMA_FROM_DEVICE);
888 iop_adma_set_dest(dest_dma, tx, 0);
889 src_dma = dma_map_single(dma_chan->device->dev, src,
890 IOP_ADMA_TEST_SIZE, DMA_TO_DEVICE);
891 iop_adma_memcpy_set_src(src_dma, tx, 0);
892
893 cookie = iop_adma_tx_submit(tx);
894 iop_adma_issue_pending(dma_chan);
895 async_tx_ack(tx);
896 msleep(1);
897
898 if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) !=
899 DMA_SUCCESS) {
900 dev_printk(KERN_ERR, dma_chan->device->dev,
901 "Self-test copy timed out, disabling\n");
902 err = -ENODEV;
903 goto free_resources;
904 }
905
906 iop_chan = to_iop_adma_chan(dma_chan);
907 dma_sync_single_for_cpu(&iop_chan->device->pdev->dev, dest_dma,
908 IOP_ADMA_TEST_SIZE, DMA_FROM_DEVICE);
909 if (memcmp(src, dest, IOP_ADMA_TEST_SIZE)) {
910 dev_printk(KERN_ERR, dma_chan->device->dev,
911 "Self-test copy failed compare, disabling\n");
912 err = -ENODEV;
913 goto free_resources;
914 }
915
916free_resources:
917 iop_adma_free_chan_resources(dma_chan);
918out:
919 kfree(src);
920 kfree(dest);
921 return err;
922}
923
924#define IOP_ADMA_NUM_SRC_TEST 4 /* must be <= 15 */
925static int __devinit
926iop_adma_xor_zero_sum_self_test(struct iop_adma_device *device)
927{
928 int i, src_idx;
929 struct page *dest;
930 struct page *xor_srcs[IOP_ADMA_NUM_SRC_TEST];
931 struct page *zero_sum_srcs[IOP_ADMA_NUM_SRC_TEST + 1];
932 dma_addr_t dma_addr, dest_dma;
933 struct dma_async_tx_descriptor *tx;
934 struct dma_chan *dma_chan;
935 dma_cookie_t cookie;
936 u8 cmp_byte = 0;
937 u32 cmp_word;
938 u32 zero_sum_result;
939 int err = 0;
940 struct iop_adma_chan *iop_chan;
941
942 dev_dbg(device->common.dev, "%s\n", __FUNCTION__);
943
944 for (src_idx = 0; src_idx < IOP_ADMA_NUM_SRC_TEST; src_idx++) {
945 xor_srcs[src_idx] = alloc_page(GFP_KERNEL);
946 if (!xor_srcs[src_idx])
947 while (src_idx--) {
948 __free_page(xor_srcs[src_idx]);
949 return -ENOMEM;
950 }
951 }
952
953 dest = alloc_page(GFP_KERNEL);
954 if (!dest)
955 while (src_idx--) {
956 __free_page(xor_srcs[src_idx]);
957 return -ENOMEM;
958 }
959
960 /* Fill in src buffers */
961 for (src_idx = 0; src_idx < IOP_ADMA_NUM_SRC_TEST; src_idx++) {
962 u8 *ptr = page_address(xor_srcs[src_idx]);
963 for (i = 0; i < PAGE_SIZE; i++)
964 ptr[i] = (1 << src_idx);
965 }
966
967 for (src_idx = 0; src_idx < IOP_ADMA_NUM_SRC_TEST; src_idx++)
968 cmp_byte ^= (u8) (1 << src_idx);
969
970 cmp_word = (cmp_byte << 24) | (cmp_byte << 16) |
971 (cmp_byte << 8) | cmp_byte;
972
973 memset(page_address(dest), 0, PAGE_SIZE);
974
975 dma_chan = container_of(device->common.channels.next,
976 struct dma_chan,
977 device_node);
978 if (iop_adma_alloc_chan_resources(dma_chan) < 1) {
979 err = -ENODEV;
980 goto out;
981 }
982
983 /* test xor */
984 tx = iop_adma_prep_dma_xor(dma_chan, IOP_ADMA_NUM_SRC_TEST,
985 PAGE_SIZE, 1);
986 dest_dma = dma_map_page(dma_chan->device->dev, dest, 0,
987 PAGE_SIZE, DMA_FROM_DEVICE);
988 iop_adma_set_dest(dest_dma, tx, 0);
989
990 for (i = 0; i < IOP_ADMA_NUM_SRC_TEST; i++) {
991 dma_addr = dma_map_page(dma_chan->device->dev, xor_srcs[i], 0,
992 PAGE_SIZE, DMA_TO_DEVICE);
993 iop_adma_xor_set_src(dma_addr, tx, i);
994 }
995
996 cookie = iop_adma_tx_submit(tx);
997 iop_adma_issue_pending(dma_chan);
998 async_tx_ack(tx);
999 msleep(8);
1000
1001 if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) !=
1002 DMA_SUCCESS) {
1003 dev_printk(KERN_ERR, dma_chan->device->dev,
1004 "Self-test xor timed out, disabling\n");
1005 err = -ENODEV;
1006 goto free_resources;
1007 }
1008
1009 iop_chan = to_iop_adma_chan(dma_chan);
1010 dma_sync_single_for_cpu(&iop_chan->device->pdev->dev, dest_dma,
1011 PAGE_SIZE, DMA_FROM_DEVICE);
1012 for (i = 0; i < (PAGE_SIZE / sizeof(u32)); i++) {
1013 u32 *ptr = page_address(dest);
1014 if (ptr[i] != cmp_word) {
1015 dev_printk(KERN_ERR, dma_chan->device->dev,
1016 "Self-test xor failed compare, disabling\n");
1017 err = -ENODEV;
1018 goto free_resources;
1019 }
1020 }
1021 dma_sync_single_for_device(&iop_chan->device->pdev->dev, dest_dma,
1022 PAGE_SIZE, DMA_TO_DEVICE);
1023
1024 /* skip zero sum if the capability is not present */
1025 if (!dma_has_cap(DMA_ZERO_SUM, dma_chan->device->cap_mask))
1026 goto free_resources;
1027
1028 /* zero sum the sources with the destintation page */
1029 for (i = 0; i < IOP_ADMA_NUM_SRC_TEST; i++)
1030 zero_sum_srcs[i] = xor_srcs[i];
1031 zero_sum_srcs[i] = dest;
1032
1033 zero_sum_result = 1;
1034
1035 tx = iop_adma_prep_dma_zero_sum(dma_chan, IOP_ADMA_NUM_SRC_TEST + 1,
1036 PAGE_SIZE, &zero_sum_result, 1);
1037 for (i = 0; i < IOP_ADMA_NUM_SRC_TEST + 1; i++) {
1038 dma_addr = dma_map_page(dma_chan->device->dev, zero_sum_srcs[i],
1039 0, PAGE_SIZE, DMA_TO_DEVICE);
1040 iop_adma_xor_zero_sum_set_src(dma_addr, tx, i);
1041 }
1042
1043 cookie = iop_adma_tx_submit(tx);
1044 iop_adma_issue_pending(dma_chan);
1045 async_tx_ack(tx);
1046 msleep(8);
1047
1048 if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) != DMA_SUCCESS) {
1049 dev_printk(KERN_ERR, dma_chan->device->dev,
1050 "Self-test zero sum timed out, disabling\n");
1051 err = -ENODEV;
1052 goto free_resources;
1053 }
1054
1055 if (zero_sum_result != 0) {
1056 dev_printk(KERN_ERR, dma_chan->device->dev,
1057 "Self-test zero sum failed compare, disabling\n");
1058 err = -ENODEV;
1059 goto free_resources;
1060 }
1061
1062 /* test memset */
1063 tx = iop_adma_prep_dma_memset(dma_chan, 0, PAGE_SIZE, 1);
1064 dma_addr = dma_map_page(dma_chan->device->dev, dest, 0,
1065 PAGE_SIZE, DMA_FROM_DEVICE);
1066 iop_adma_set_dest(dma_addr, tx, 0);
1067
1068 cookie = iop_adma_tx_submit(tx);
1069 iop_adma_issue_pending(dma_chan);
1070 async_tx_ack(tx);
1071 msleep(8);
1072
1073 if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) != DMA_SUCCESS) {
1074 dev_printk(KERN_ERR, dma_chan->device->dev,
1075 "Self-test memset timed out, disabling\n");
1076 err = -ENODEV;
1077 goto free_resources;
1078 }
1079
1080 for (i = 0; i < PAGE_SIZE/sizeof(u32); i++) {
1081 u32 *ptr = page_address(dest);
1082 if (ptr[i]) {
1083 dev_printk(KERN_ERR, dma_chan->device->dev,
1084 "Self-test memset failed compare, disabling\n");
1085 err = -ENODEV;
1086 goto free_resources;
1087 }
1088 }
1089
1090 /* test for non-zero parity sum */
1091 zero_sum_result = 0;
1092 tx = iop_adma_prep_dma_zero_sum(dma_chan, IOP_ADMA_NUM_SRC_TEST + 1,
1093 PAGE_SIZE, &zero_sum_result, 1);
1094 for (i = 0; i < IOP_ADMA_NUM_SRC_TEST + 1; i++) {
1095 dma_addr = dma_map_page(dma_chan->device->dev, zero_sum_srcs[i],
1096 0, PAGE_SIZE, DMA_TO_DEVICE);
1097 iop_adma_xor_zero_sum_set_src(dma_addr, tx, i);
1098 }
1099
1100 cookie = iop_adma_tx_submit(tx);
1101 iop_adma_issue_pending(dma_chan);
1102 async_tx_ack(tx);
1103 msleep(8);
1104
1105 if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) != DMA_SUCCESS) {
1106 dev_printk(KERN_ERR, dma_chan->device->dev,
1107 "Self-test non-zero sum timed out, disabling\n");
1108 err = -ENODEV;
1109 goto free_resources;
1110 }
1111
1112 if (zero_sum_result != 1) {
1113 dev_printk(KERN_ERR, dma_chan->device->dev,
1114 "Self-test non-zero sum failed compare, disabling\n");
1115 err = -ENODEV;
1116 goto free_resources;
1117 }
1118
1119free_resources:
1120 iop_adma_free_chan_resources(dma_chan);
1121out:
1122 src_idx = IOP_ADMA_NUM_SRC_TEST;
1123 while (src_idx--)
1124 __free_page(xor_srcs[src_idx]);
1125 __free_page(dest);
1126 return err;
1127}
1128
1129static int __devexit iop_adma_remove(struct platform_device *dev)
1130{
1131 struct iop_adma_device *device = platform_get_drvdata(dev);
1132 struct dma_chan *chan, *_chan;
1133 struct iop_adma_chan *iop_chan;
1134 int i;
1135 struct iop_adma_platform_data *plat_data = dev->dev.platform_data;
1136
1137 dma_async_device_unregister(&device->common);
1138
1139 for (i = 0; i < 3; i++) {
1140 unsigned int irq;
1141 irq = platform_get_irq(dev, i);
1142 free_irq(irq, device);
1143 }
1144
1145 dma_free_coherent(&dev->dev, plat_data->pool_size,
1146 device->dma_desc_pool_virt, device->dma_desc_pool);
1147
1148 do {
1149 struct resource *res;
1150 res = platform_get_resource(dev, IORESOURCE_MEM, 0);
1151 release_mem_region(res->start, res->end - res->start);
1152 } while (0);
1153
1154 list_for_each_entry_safe(chan, _chan, &device->common.channels,
1155 device_node) {
1156 iop_chan = to_iop_adma_chan(chan);
1157 list_del(&chan->device_node);
1158 kfree(iop_chan);
1159 }
1160 kfree(device);
1161
1162 return 0;
1163}
1164
1165static int __devinit iop_adma_probe(struct platform_device *pdev)
1166{
1167 struct resource *res;
1168 int ret = 0, i;
1169 struct iop_adma_device *adev;
1170 struct iop_adma_chan *iop_chan;
1171 struct dma_device *dma_dev;
1172 struct iop_adma_platform_data *plat_data = pdev->dev.platform_data;
1173
1174 res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1175 if (!res)
1176 return -ENODEV;
1177
1178 if (!devm_request_mem_region(&pdev->dev, res->start,
1179 res->end - res->start, pdev->name))
1180 return -EBUSY;
1181
1182 adev = kzalloc(sizeof(*adev), GFP_KERNEL);
1183 if (!adev)
1184 return -ENOMEM;
1185 dma_dev = &adev->common;
1186
1187 /* allocate coherent memory for hardware descriptors
1188 * note: writecombine gives slightly better performance, but
1189 * requires that we explicitly flush the writes
1190 */
1191 if ((adev->dma_desc_pool_virt = dma_alloc_writecombine(&pdev->dev,
1192 plat_data->pool_size,
1193 &adev->dma_desc_pool,
1194 GFP_KERNEL)) == NULL) {
1195 ret = -ENOMEM;
1196 goto err_free_adev;
1197 }
1198
1199 dev_dbg(&pdev->dev, "%s: allocted descriptor pool virt %p phys %p\n",
1200 __FUNCTION__, adev->dma_desc_pool_virt,
1201 (void *) adev->dma_desc_pool);
1202
1203 adev->id = plat_data->hw_id;
1204
1205 /* discover transaction capabilites from the platform data */
1206 dma_dev->cap_mask = plat_data->cap_mask;
1207
1208 adev->pdev = pdev;
1209 platform_set_drvdata(pdev, adev);
1210
1211 INIT_LIST_HEAD(&dma_dev->channels);
1212
1213 /* set base routines */
1214 dma_dev->device_alloc_chan_resources = iop_adma_alloc_chan_resources;
1215 dma_dev->device_free_chan_resources = iop_adma_free_chan_resources;
1216 dma_dev->device_is_tx_complete = iop_adma_is_complete;
1217 dma_dev->device_issue_pending = iop_adma_issue_pending;
1218 dma_dev->device_dependency_added = iop_adma_dependency_added;
1219 dma_dev->dev = &pdev->dev;
1220
1221 /* set prep routines based on capability */
1222 if (dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask))
1223 dma_dev->device_prep_dma_memcpy = iop_adma_prep_dma_memcpy;
1224 if (dma_has_cap(DMA_MEMSET, dma_dev->cap_mask))
1225 dma_dev->device_prep_dma_memset = iop_adma_prep_dma_memset;
1226 if (dma_has_cap(DMA_XOR, dma_dev->cap_mask)) {
1227 dma_dev->max_xor = iop_adma_get_max_xor();
1228 dma_dev->device_prep_dma_xor = iop_adma_prep_dma_xor;
1229 }
1230 if (dma_has_cap(DMA_ZERO_SUM, dma_dev->cap_mask))
1231 dma_dev->device_prep_dma_zero_sum =
1232 iop_adma_prep_dma_zero_sum;
1233 if (dma_has_cap(DMA_INTERRUPT, dma_dev->cap_mask))
1234 dma_dev->device_prep_dma_interrupt =
1235 iop_adma_prep_dma_interrupt;
1236
1237 iop_chan = kzalloc(sizeof(*iop_chan), GFP_KERNEL);
1238 if (!iop_chan) {
1239 ret = -ENOMEM;
1240 goto err_free_dma;
1241 }
1242 iop_chan->device = adev;
1243
1244 iop_chan->mmr_base = devm_ioremap(&pdev->dev, res->start,
1245 res->end - res->start);
1246 if (!iop_chan->mmr_base) {
1247 ret = -ENOMEM;
1248 goto err_free_iop_chan;
1249 }
1250 tasklet_init(&iop_chan->irq_tasklet, iop_adma_tasklet, (unsigned long)
1251 iop_chan);
1252
1253 /* clear errors before enabling interrupts */
1254 iop_adma_device_clear_err_status(iop_chan);
1255
1256 for (i = 0; i < 3; i++) {
1257 irq_handler_t handler[] = { iop_adma_eot_handler,
1258 iop_adma_eoc_handler,
1259 iop_adma_err_handler };
1260 int irq = platform_get_irq(pdev, i);
1261 if (irq < 0) {
1262 ret = -ENXIO;
1263 goto err_free_iop_chan;
1264 } else {
1265 ret = devm_request_irq(&pdev->dev, irq,
1266 handler[i], 0, pdev->name, iop_chan);
1267 if (ret)
1268 goto err_free_iop_chan;
1269 }
1270 }
1271
1272 spin_lock_init(&iop_chan->lock);
1273 init_timer(&iop_chan->cleanup_watchdog);
1274 iop_chan->cleanup_watchdog.data = (unsigned long) iop_chan;
1275 iop_chan->cleanup_watchdog.function = iop_adma_tasklet;
1276 INIT_LIST_HEAD(&iop_chan->chain);
1277 INIT_LIST_HEAD(&iop_chan->all_slots);
1278 INIT_RCU_HEAD(&iop_chan->common.rcu);
1279 iop_chan->common.device = dma_dev;
1280 list_add_tail(&iop_chan->common.device_node, &dma_dev->channels);
1281
1282 if (dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask)) {
1283 ret = iop_adma_memcpy_self_test(adev);
1284 dev_dbg(&pdev->dev, "memcpy self test returned %d\n", ret);
1285 if (ret)
1286 goto err_free_iop_chan;
1287 }
1288
1289 if (dma_has_cap(DMA_XOR, dma_dev->cap_mask) ||
1290 dma_has_cap(DMA_MEMSET, dma_dev->cap_mask)) {
1291 ret = iop_adma_xor_zero_sum_self_test(adev);
1292 dev_dbg(&pdev->dev, "xor self test returned %d\n", ret);
1293 if (ret)
1294 goto err_free_iop_chan;
1295 }
1296
1297 dev_printk(KERN_INFO, &pdev->dev, "Intel(R) IOP: "
1298 "( %s%s%s%s%s%s%s%s%s%s)\n",
1299 dma_has_cap(DMA_PQ_XOR, dma_dev->cap_mask) ? "pq_xor " : "",
1300 dma_has_cap(DMA_PQ_UPDATE, dma_dev->cap_mask) ? "pq_update " : "",
1301 dma_has_cap(DMA_PQ_ZERO_SUM, dma_dev->cap_mask) ? "pq_zero_sum " : "",
1302 dma_has_cap(DMA_XOR, dma_dev->cap_mask) ? "xor " : "",
1303 dma_has_cap(DMA_DUAL_XOR, dma_dev->cap_mask) ? "dual_xor " : "",
1304 dma_has_cap(DMA_ZERO_SUM, dma_dev->cap_mask) ? "xor_zero_sum " : "",
1305 dma_has_cap(DMA_MEMSET, dma_dev->cap_mask) ? "fill " : "",
1306 dma_has_cap(DMA_MEMCPY_CRC32C, dma_dev->cap_mask) ? "cpy+crc " : "",
1307 dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask) ? "cpy " : "",
1308 dma_has_cap(DMA_INTERRUPT, dma_dev->cap_mask) ? "intr " : "");
1309
1310 dma_async_device_register(dma_dev);
1311 goto out;
1312
1313 err_free_iop_chan:
1314 kfree(iop_chan);
1315 err_free_dma:
1316 dma_free_coherent(&adev->pdev->dev, plat_data->pool_size,
1317 adev->dma_desc_pool_virt, adev->dma_desc_pool);
1318 err_free_adev:
1319 kfree(adev);
1320 out:
1321 return ret;
1322}
1323
1324static void iop_chan_start_null_memcpy(struct iop_adma_chan *iop_chan)
1325{
1326 struct iop_adma_desc_slot *sw_desc, *grp_start;
1327 dma_cookie_t cookie;
1328 int slot_cnt, slots_per_op;
1329
1330 dev_dbg(iop_chan->device->common.dev, "%s\n", __FUNCTION__);
1331
1332 spin_lock_bh(&iop_chan->lock);
1333 slot_cnt = iop_chan_memcpy_slot_count(0, &slots_per_op);
1334 sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
1335 if (sw_desc) {
1336 grp_start = sw_desc->group_head;
1337
1338 list_splice_init(&sw_desc->async_tx.tx_list, &iop_chan->chain);
1339 sw_desc->async_tx.ack = 1;
1340 iop_desc_init_memcpy(grp_start, 0);
1341 iop_desc_set_byte_count(grp_start, iop_chan, 0);
1342 iop_desc_set_dest_addr(grp_start, iop_chan, 0);
1343 iop_desc_set_memcpy_src_addr(grp_start, 0);
1344
1345 cookie = iop_chan->common.cookie;
1346 cookie++;
1347 if (cookie <= 1)
1348 cookie = 2;
1349
1350 /* initialize the completed cookie to be less than
1351 * the most recently used cookie
1352 */
1353 iop_chan->completed_cookie = cookie - 1;
1354 iop_chan->common.cookie = sw_desc->async_tx.cookie = cookie;
1355
1356 /* channel should not be busy */
1357 BUG_ON(iop_chan_is_busy(iop_chan));
1358
1359 /* clear any prior error-status bits */
1360 iop_adma_device_clear_err_status(iop_chan);
1361
1362 /* disable operation */
1363 iop_chan_disable(iop_chan);
1364
1365 /* set the descriptor address */
1366 iop_chan_set_next_descriptor(iop_chan, sw_desc->async_tx.phys);
1367
1368 /* 1/ don't add pre-chained descriptors
1369 * 2/ dummy read to flush next_desc write
1370 */
1371 BUG_ON(iop_desc_get_next_desc(sw_desc));
1372
1373 /* run the descriptor */
1374 iop_chan_enable(iop_chan);
1375 } else
1376 dev_printk(KERN_ERR, iop_chan->device->common.dev,
1377 "failed to allocate null descriptor\n");
1378 spin_unlock_bh(&iop_chan->lock);
1379}
1380
1381static void iop_chan_start_null_xor(struct iop_adma_chan *iop_chan)
1382{
1383 struct iop_adma_desc_slot *sw_desc, *grp_start;
1384 dma_cookie_t cookie;
1385 int slot_cnt, slots_per_op;
1386
1387 dev_dbg(iop_chan->device->common.dev, "%s\n", __FUNCTION__);
1388
1389 spin_lock_bh(&iop_chan->lock);
1390 slot_cnt = iop_chan_xor_slot_count(0, 2, &slots_per_op);
1391 sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op);
1392 if (sw_desc) {
1393 grp_start = sw_desc->group_head;
1394 list_splice_init(&sw_desc->async_tx.tx_list, &iop_chan->chain);
1395 sw_desc->async_tx.ack = 1;
1396 iop_desc_init_null_xor(grp_start, 2, 0);
1397 iop_desc_set_byte_count(grp_start, iop_chan, 0);
1398 iop_desc_set_dest_addr(grp_start, iop_chan, 0);
1399 iop_desc_set_xor_src_addr(grp_start, 0, 0);
1400 iop_desc_set_xor_src_addr(grp_start, 1, 0);
1401
1402 cookie = iop_chan->common.cookie;
1403 cookie++;
1404 if (cookie <= 1)
1405 cookie = 2;
1406
1407 /* initialize the completed cookie to be less than
1408 * the most recently used cookie
1409 */
1410 iop_chan->completed_cookie = cookie - 1;
1411 iop_chan->common.cookie = sw_desc->async_tx.cookie = cookie;
1412
1413 /* channel should not be busy */
1414 BUG_ON(iop_chan_is_busy(iop_chan));
1415
1416 /* clear any prior error-status bits */
1417 iop_adma_device_clear_err_status(iop_chan);
1418
1419 /* disable operation */
1420 iop_chan_disable(iop_chan);
1421
1422 /* set the descriptor address */
1423 iop_chan_set_next_descriptor(iop_chan, sw_desc->async_tx.phys);
1424
1425 /* 1/ don't add pre-chained descriptors
1426 * 2/ dummy read to flush next_desc write
1427 */
1428 BUG_ON(iop_desc_get_next_desc(sw_desc));
1429
1430 /* run the descriptor */
1431 iop_chan_enable(iop_chan);
1432 } else
1433 dev_printk(KERN_ERR, iop_chan->device->common.dev,
1434 "failed to allocate null descriptor\n");
1435 spin_unlock_bh(&iop_chan->lock);
1436}
1437
1438static struct platform_driver iop_adma_driver = {
1439 .probe = iop_adma_probe,
1440 .remove = iop_adma_remove,
1441 .driver = {
1442 .owner = THIS_MODULE,
1443 .name = "iop-adma",
1444 },
1445};
1446
1447static int __init iop_adma_init (void)
1448{
1449 /* it's currently unsafe to unload this module */
1450 /* if forced, worst case is that rmmod hangs */
1451 __unsafe(THIS_MODULE);
1452
1453 return platform_driver_register(&iop_adma_driver);
1454}
1455
1456static void __exit iop_adma_exit (void)
1457{
1458 platform_driver_unregister(&iop_adma_driver);
1459 return;
1460}
1461
1462module_init(iop_adma_init);
1463module_exit(iop_adma_exit);
1464
1465MODULE_AUTHOR("Intel Corporation");
1466MODULE_DESCRIPTION("IOP ADMA Engine Driver");
1467MODULE_LICENSE("GPL");
generated by cgit v1.2.2 (git 2.25.0) at 2024-12-12 14:07:57 -0500