Commit | Line | Data |
---|---|---|
c2110923 DW |
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 | */ | |
49 | static 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 | ||
61 | static dma_cookie_t | |
62 | iop_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); | |
c2110923 DW |
66 | if (desc->async_tx.cookie > 0) { |
67 | cookie = desc->async_tx.cookie; | |
68 | desc->async_tx.cookie = 0; | |
69 | ||
70 | /* call the callback (must not sleep or submit new | |
71 | * operations to this channel) | |
72 | */ | |
73 | if (desc->async_tx.callback) | |
74 | desc->async_tx.callback( | |
75 | desc->async_tx.callback_param); | |
76 | ||
77 | /* unmap dma addresses | |
78 | * (unmap_single vs unmap_page?) | |
79 | */ | |
80 | if (desc->group_head && desc->unmap_len) { | |
81 | struct iop_adma_desc_slot *unmap = desc->group_head; | |
82 | struct device *dev = | |
83 | &iop_chan->device->pdev->dev; | |
84 | u32 len = unmap->unmap_len; | |
e1d181ef DW |
85 | enum dma_ctrl_flags flags = desc->async_tx.flags; |
86 | u32 src_cnt; | |
87 | dma_addr_t addr; | |
c2110923 | 88 | |
e1d181ef DW |
89 | if (!(flags & DMA_COMPL_SKIP_DEST_UNMAP)) { |
90 | addr = iop_desc_get_dest_addr(unmap, iop_chan); | |
91 | dma_unmap_page(dev, addr, len, DMA_FROM_DEVICE); | |
92 | } | |
93 | ||
94 | if (!(flags & DMA_COMPL_SKIP_SRC_UNMAP)) { | |
95 | src_cnt = unmap->unmap_src_cnt; | |
96 | while (src_cnt--) { | |
97 | addr = iop_desc_get_src_addr(unmap, | |
98 | iop_chan, | |
99 | src_cnt); | |
100 | dma_unmap_page(dev, addr, len, | |
101 | DMA_TO_DEVICE); | |
102 | } | |
c2110923 DW |
103 | } |
104 | desc->group_head = NULL; | |
105 | } | |
106 | } | |
107 | ||
108 | /* run dependent operations */ | |
109 | async_tx_run_dependencies(&desc->async_tx); | |
c2110923 DW |
110 | |
111 | return cookie; | |
112 | } | |
113 | ||
114 | static int | |
115 | iop_adma_clean_slot(struct iop_adma_desc_slot *desc, | |
116 | struct iop_adma_chan *iop_chan) | |
117 | { | |
118 | /* the client is allowed to attach dependent operations | |
119 | * until 'ack' is set | |
120 | */ | |
636bdeaa | 121 | if (!async_tx_test_ack(&desc->async_tx)) |
c2110923 DW |
122 | return 0; |
123 | ||
124 | /* leave the last descriptor in the chain | |
125 | * so we can append to it | |
126 | */ | |
127 | if (desc->chain_node.next == &iop_chan->chain) | |
128 | return 1; | |
129 | ||
130 | dev_dbg(iop_chan->device->common.dev, | |
131 | "\tfree slot: %d slots_per_op: %d\n", | |
132 | desc->idx, desc->slots_per_op); | |
133 | ||
134 | list_del(&desc->chain_node); | |
135 | iop_adma_free_slots(desc); | |
136 | ||
137 | return 0; | |
138 | } | |
139 | ||
140 | static void __iop_adma_slot_cleanup(struct iop_adma_chan *iop_chan) | |
141 | { | |
142 | struct iop_adma_desc_slot *iter, *_iter, *grp_start = NULL; | |
143 | dma_cookie_t cookie = 0; | |
144 | u32 current_desc = iop_chan_get_current_descriptor(iop_chan); | |
145 | int busy = iop_chan_is_busy(iop_chan); | |
146 | int seen_current = 0, slot_cnt = 0, slots_per_op = 0; | |
147 | ||
3d9b525b | 148 | dev_dbg(iop_chan->device->common.dev, "%s\n", __func__); |
c2110923 DW |
149 | /* free completed slots from the chain starting with |
150 | * the oldest descriptor | |
151 | */ | |
152 | list_for_each_entry_safe(iter, _iter, &iop_chan->chain, | |
153 | chain_node) { | |
154 | pr_debug("\tcookie: %d slot: %d busy: %d " | |
155 | "this_desc: %#x next_desc: %#x ack: %d\n", | |
156 | iter->async_tx.cookie, iter->idx, busy, | |
157 | iter->async_tx.phys, iop_desc_get_next_desc(iter), | |
636bdeaa | 158 | async_tx_test_ack(&iter->async_tx)); |
c2110923 DW |
159 | prefetch(_iter); |
160 | prefetch(&_iter->async_tx); | |
161 | ||
162 | /* do not advance past the current descriptor loaded into the | |
163 | * hardware channel, subsequent descriptors are either in | |
164 | * process or have not been submitted | |
165 | */ | |
166 | if (seen_current) | |
167 | break; | |
168 | ||
169 | /* stop the search if we reach the current descriptor and the | |
170 | * channel is busy, or if it appears that the current descriptor | |
171 | * needs to be re-read (i.e. has been appended to) | |
172 | */ | |
173 | if (iter->async_tx.phys == current_desc) { | |
174 | BUG_ON(seen_current++); | |
175 | if (busy || iop_desc_get_next_desc(iter)) | |
176 | break; | |
177 | } | |
178 | ||
179 | /* detect the start of a group transaction */ | |
180 | if (!slot_cnt && !slots_per_op) { | |
181 | slot_cnt = iter->slot_cnt; | |
182 | slots_per_op = iter->slots_per_op; | |
183 | if (slot_cnt <= slots_per_op) { | |
184 | slot_cnt = 0; | |
185 | slots_per_op = 0; | |
186 | } | |
187 | } | |
188 | ||
189 | if (slot_cnt) { | |
190 | pr_debug("\tgroup++\n"); | |
191 | if (!grp_start) | |
192 | grp_start = iter; | |
193 | slot_cnt -= slots_per_op; | |
194 | } | |
195 | ||
196 | /* all the members of a group are complete */ | |
197 | if (slots_per_op != 0 && slot_cnt == 0) { | |
198 | struct iop_adma_desc_slot *grp_iter, *_grp_iter; | |
199 | int end_of_chain = 0; | |
200 | pr_debug("\tgroup end\n"); | |
201 | ||
202 | /* collect the total results */ | |
203 | if (grp_start->xor_check_result) { | |
204 | u32 zero_sum_result = 0; | |
205 | slot_cnt = grp_start->slot_cnt; | |
206 | grp_iter = grp_start; | |
207 | ||
208 | list_for_each_entry_from(grp_iter, | |
209 | &iop_chan->chain, chain_node) { | |
210 | zero_sum_result |= | |
211 | iop_desc_get_zero_result(grp_iter); | |
212 | pr_debug("\titer%d result: %d\n", | |
213 | grp_iter->idx, zero_sum_result); | |
214 | slot_cnt -= slots_per_op; | |
215 | if (slot_cnt == 0) | |
216 | break; | |
217 | } | |
218 | pr_debug("\tgrp_start->xor_check_result: %p\n", | |
219 | grp_start->xor_check_result); | |
220 | *grp_start->xor_check_result = zero_sum_result; | |
221 | } | |
222 | ||
223 | /* clean up the group */ | |
224 | slot_cnt = grp_start->slot_cnt; | |
225 | grp_iter = grp_start; | |
226 | list_for_each_entry_safe_from(grp_iter, _grp_iter, | |
227 | &iop_chan->chain, chain_node) { | |
228 | cookie = iop_adma_run_tx_complete_actions( | |
229 | grp_iter, iop_chan, cookie); | |
230 | ||
231 | slot_cnt -= slots_per_op; | |
232 | end_of_chain = iop_adma_clean_slot(grp_iter, | |
233 | iop_chan); | |
234 | ||
235 | if (slot_cnt == 0 || end_of_chain) | |
236 | break; | |
237 | } | |
238 | ||
239 | /* the group should be complete at this point */ | |
240 | BUG_ON(slot_cnt); | |
241 | ||
242 | slots_per_op = 0; | |
243 | grp_start = NULL; | |
244 | if (end_of_chain) | |
245 | break; | |
246 | else | |
247 | continue; | |
248 | } else if (slots_per_op) /* wait for group completion */ | |
249 | continue; | |
250 | ||
251 | /* write back zero sum results (single descriptor case) */ | |
252 | if (iter->xor_check_result && iter->async_tx.cookie) | |
253 | *iter->xor_check_result = | |
254 | iop_desc_get_zero_result(iter); | |
255 | ||
256 | cookie = iop_adma_run_tx_complete_actions( | |
257 | iter, iop_chan, cookie); | |
258 | ||
259 | if (iop_adma_clean_slot(iter, iop_chan)) | |
260 | break; | |
261 | } | |
262 | ||
263 | BUG_ON(!seen_current); | |
264 | ||
c2110923 DW |
265 | if (cookie > 0) { |
266 | iop_chan->completed_cookie = cookie; | |
267 | pr_debug("\tcompleted cookie %d\n", cookie); | |
268 | } | |
269 | } | |
270 | ||
271 | static void | |
272 | iop_adma_slot_cleanup(struct iop_adma_chan *iop_chan) | |
273 | { | |
274 | spin_lock_bh(&iop_chan->lock); | |
275 | __iop_adma_slot_cleanup(iop_chan); | |
276 | spin_unlock_bh(&iop_chan->lock); | |
277 | } | |
278 | ||
279 | static void iop_adma_tasklet(unsigned long data) | |
280 | { | |
19242d72 DW |
281 | struct iop_adma_chan *iop_chan = (struct iop_adma_chan *) data; |
282 | ||
283 | spin_lock(&iop_chan->lock); | |
284 | __iop_adma_slot_cleanup(iop_chan); | |
285 | spin_unlock(&iop_chan->lock); | |
c2110923 DW |
286 | } |
287 | ||
288 | static struct iop_adma_desc_slot * | |
289 | iop_adma_alloc_slots(struct iop_adma_chan *iop_chan, int num_slots, | |
290 | int slots_per_op) | |
291 | { | |
292 | struct iop_adma_desc_slot *iter, *_iter, *alloc_start = NULL; | |
e73ef9ac | 293 | LIST_HEAD(chain); |
c2110923 DW |
294 | int slots_found, retry = 0; |
295 | ||
296 | /* start search from the last allocated descrtiptor | |
297 | * if a contiguous allocation can not be found start searching | |
298 | * from the beginning of the list | |
299 | */ | |
300 | retry: | |
301 | slots_found = 0; | |
302 | if (retry == 0) | |
303 | iter = iop_chan->last_used; | |
304 | else | |
305 | iter = list_entry(&iop_chan->all_slots, | |
306 | struct iop_adma_desc_slot, | |
307 | slot_node); | |
308 | ||
309 | list_for_each_entry_safe_continue( | |
310 | iter, _iter, &iop_chan->all_slots, slot_node) { | |
311 | prefetch(_iter); | |
312 | prefetch(&_iter->async_tx); | |
313 | if (iter->slots_per_op) { | |
314 | /* give up after finding the first busy slot | |
315 | * on the second pass through the list | |
316 | */ | |
317 | if (retry) | |
318 | break; | |
319 | ||
320 | slots_found = 0; | |
321 | continue; | |
322 | } | |
323 | ||
324 | /* start the allocation if the slot is correctly aligned */ | |
325 | if (!slots_found++) { | |
326 | if (iop_desc_is_aligned(iter, slots_per_op)) | |
327 | alloc_start = iter; | |
328 | else { | |
329 | slots_found = 0; | |
330 | continue; | |
331 | } | |
332 | } | |
333 | ||
334 | if (slots_found == num_slots) { | |
335 | struct iop_adma_desc_slot *alloc_tail = NULL; | |
336 | struct iop_adma_desc_slot *last_used = NULL; | |
337 | iter = alloc_start; | |
338 | while (num_slots) { | |
339 | int i; | |
340 | dev_dbg(iop_chan->device->common.dev, | |
341 | "allocated slot: %d " | |
342 | "(desc %p phys: %#x) slots_per_op %d\n", | |
343 | iter->idx, iter->hw_desc, | |
344 | iter->async_tx.phys, slots_per_op); | |
345 | ||
346 | /* pre-ack all but the last descriptor */ | |
347 | if (num_slots != slots_per_op) | |
636bdeaa | 348 | async_tx_ack(&iter->async_tx); |
c2110923 DW |
349 | |
350 | list_add_tail(&iter->chain_node, &chain); | |
351 | alloc_tail = iter; | |
352 | iter->async_tx.cookie = 0; | |
353 | iter->slot_cnt = num_slots; | |
354 | iter->xor_check_result = NULL; | |
355 | for (i = 0; i < slots_per_op; i++) { | |
356 | iter->slots_per_op = slots_per_op - i; | |
357 | last_used = iter; | |
358 | iter = list_entry(iter->slot_node.next, | |
359 | struct iop_adma_desc_slot, | |
360 | slot_node); | |
361 | } | |
362 | num_slots -= slots_per_op; | |
363 | } | |
364 | alloc_tail->group_head = alloc_start; | |
365 | alloc_tail->async_tx.cookie = -EBUSY; | |
366 | list_splice(&chain, &alloc_tail->async_tx.tx_list); | |
367 | iop_chan->last_used = last_used; | |
368 | iop_desc_clear_next_desc(alloc_start); | |
369 | iop_desc_clear_next_desc(alloc_tail); | |
370 | return alloc_tail; | |
371 | } | |
372 | } | |
373 | if (!retry++) | |
374 | goto retry; | |
375 | ||
376 | /* try to free some slots if the allocation fails */ | |
377 | tasklet_schedule(&iop_chan->irq_tasklet); | |
378 | ||
379 | return NULL; | |
380 | } | |
381 | ||
382 | static dma_cookie_t | |
383 | iop_desc_assign_cookie(struct iop_adma_chan *iop_chan, | |
384 | struct iop_adma_desc_slot *desc) | |
385 | { | |
386 | dma_cookie_t cookie = iop_chan->common.cookie; | |
387 | cookie++; | |
388 | if (cookie < 0) | |
389 | cookie = 1; | |
390 | iop_chan->common.cookie = desc->async_tx.cookie = cookie; | |
391 | return cookie; | |
392 | } | |
393 | ||
394 | static void iop_adma_check_threshold(struct iop_adma_chan *iop_chan) | |
395 | { | |
396 | dev_dbg(iop_chan->device->common.dev, "pending: %d\n", | |
397 | iop_chan->pending); | |
398 | ||
399 | if (iop_chan->pending >= IOP_ADMA_THRESHOLD) { | |
400 | iop_chan->pending = 0; | |
401 | iop_chan_append(iop_chan); | |
402 | } | |
403 | } | |
404 | ||
405 | static dma_cookie_t | |
406 | iop_adma_tx_submit(struct dma_async_tx_descriptor *tx) | |
407 | { | |
408 | struct iop_adma_desc_slot *sw_desc = tx_to_iop_adma_slot(tx); | |
409 | struct iop_adma_chan *iop_chan = to_iop_adma_chan(tx->chan); | |
410 | struct iop_adma_desc_slot *grp_start, *old_chain_tail; | |
411 | int slot_cnt; | |
412 | int slots_per_op; | |
413 | dma_cookie_t cookie; | |
414 | ||
415 | grp_start = sw_desc->group_head; | |
416 | slot_cnt = grp_start->slot_cnt; | |
417 | slots_per_op = grp_start->slots_per_op; | |
418 | ||
419 | spin_lock_bh(&iop_chan->lock); | |
420 | cookie = iop_desc_assign_cookie(iop_chan, sw_desc); | |
421 | ||
422 | old_chain_tail = list_entry(iop_chan->chain.prev, | |
423 | struct iop_adma_desc_slot, chain_node); | |
424 | list_splice_init(&sw_desc->async_tx.tx_list, | |
425 | &old_chain_tail->chain_node); | |
426 | ||
427 | /* fix up the hardware chain */ | |
428 | iop_desc_set_next_desc(old_chain_tail, grp_start->async_tx.phys); | |
429 | ||
430 | /* 1/ don't add pre-chained descriptors | |
431 | * 2/ dummy read to flush next_desc write | |
432 | */ | |
433 | BUG_ON(iop_desc_get_next_desc(sw_desc)); | |
434 | ||
435 | /* increment the pending count by the number of slots | |
436 | * memcpy operations have a 1:1 (slot:operation) relation | |
437 | * other operations are heavier and will pop the threshold | |
438 | * more often. | |
439 | */ | |
440 | iop_chan->pending += slot_cnt; | |
441 | iop_adma_check_threshold(iop_chan); | |
442 | spin_unlock_bh(&iop_chan->lock); | |
443 | ||
444 | dev_dbg(iop_chan->device->common.dev, "%s cookie: %d slot: %d\n", | |
3d9b525b | 445 | __func__, sw_desc->async_tx.cookie, sw_desc->idx); |
c2110923 DW |
446 | |
447 | return cookie; | |
448 | } | |
449 | ||
c2110923 DW |
450 | static void iop_chan_start_null_memcpy(struct iop_adma_chan *iop_chan); |
451 | static void iop_chan_start_null_xor(struct iop_adma_chan *iop_chan); | |
452 | ||
453 | /* returns the number of allocated descriptors */ | |
848c536a HS |
454 | static int iop_adma_alloc_chan_resources(struct dma_chan *chan, |
455 | struct dma_client *client) | |
c2110923 DW |
456 | { |
457 | char *hw_desc; | |
458 | int idx; | |
459 | struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan); | |
460 | struct iop_adma_desc_slot *slot = NULL; | |
461 | int init = iop_chan->slots_allocated ? 0 : 1; | |
462 | struct iop_adma_platform_data *plat_data = | |
463 | iop_chan->device->pdev->dev.platform_data; | |
464 | int num_descs_in_pool = plat_data->pool_size/IOP_ADMA_SLOT_SIZE; | |
465 | ||
466 | /* Allocate descriptor slots */ | |
467 | do { | |
468 | idx = iop_chan->slots_allocated; | |
469 | if (idx == num_descs_in_pool) | |
470 | break; | |
471 | ||
472 | slot = kzalloc(sizeof(*slot), GFP_KERNEL); | |
473 | if (!slot) { | |
474 | printk(KERN_INFO "IOP ADMA Channel only initialized" | |
475 | " %d descriptor slots", idx); | |
476 | break; | |
477 | } | |
478 | hw_desc = (char *) iop_chan->device->dma_desc_pool_virt; | |
479 | slot->hw_desc = (void *) &hw_desc[idx * IOP_ADMA_SLOT_SIZE]; | |
480 | ||
481 | dma_async_tx_descriptor_init(&slot->async_tx, chan); | |
482 | slot->async_tx.tx_submit = iop_adma_tx_submit; | |
c2110923 DW |
483 | INIT_LIST_HEAD(&slot->chain_node); |
484 | INIT_LIST_HEAD(&slot->slot_node); | |
485 | INIT_LIST_HEAD(&slot->async_tx.tx_list); | |
486 | hw_desc = (char *) iop_chan->device->dma_desc_pool; | |
487 | slot->async_tx.phys = | |
488 | (dma_addr_t) &hw_desc[idx * IOP_ADMA_SLOT_SIZE]; | |
489 | slot->idx = idx; | |
490 | ||
491 | spin_lock_bh(&iop_chan->lock); | |
492 | iop_chan->slots_allocated++; | |
493 | list_add_tail(&slot->slot_node, &iop_chan->all_slots); | |
494 | spin_unlock_bh(&iop_chan->lock); | |
495 | } while (iop_chan->slots_allocated < num_descs_in_pool); | |
496 | ||
497 | if (idx && !iop_chan->last_used) | |
498 | iop_chan->last_used = list_entry(iop_chan->all_slots.next, | |
499 | struct iop_adma_desc_slot, | |
500 | slot_node); | |
501 | ||
502 | dev_dbg(iop_chan->device->common.dev, | |
503 | "allocated %d descriptor slots last_used: %p\n", | |
504 | iop_chan->slots_allocated, iop_chan->last_used); | |
505 | ||
506 | /* initialize the channel and the chain with a null operation */ | |
507 | if (init) { | |
508 | if (dma_has_cap(DMA_MEMCPY, | |
509 | iop_chan->device->common.cap_mask)) | |
510 | iop_chan_start_null_memcpy(iop_chan); | |
511 | else if (dma_has_cap(DMA_XOR, | |
512 | iop_chan->device->common.cap_mask)) | |
513 | iop_chan_start_null_xor(iop_chan); | |
514 | else | |
515 | BUG(); | |
516 | } | |
517 | ||
518 | return (idx > 0) ? idx : -ENOMEM; | |
519 | } | |
520 | ||
521 | static struct dma_async_tx_descriptor * | |
636bdeaa | 522 | iop_adma_prep_dma_interrupt(struct dma_chan *chan, unsigned long flags) |
c2110923 DW |
523 | { |
524 | struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan); | |
525 | struct iop_adma_desc_slot *sw_desc, *grp_start; | |
526 | int slot_cnt, slots_per_op; | |
527 | ||
3d9b525b | 528 | dev_dbg(iop_chan->device->common.dev, "%s\n", __func__); |
c2110923 DW |
529 | |
530 | spin_lock_bh(&iop_chan->lock); | |
531 | slot_cnt = iop_chan_interrupt_slot_count(&slots_per_op, iop_chan); | |
532 | sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op); | |
533 | if (sw_desc) { | |
534 | grp_start = sw_desc->group_head; | |
535 | iop_desc_init_interrupt(grp_start, iop_chan); | |
536 | grp_start->unmap_len = 0; | |
636bdeaa | 537 | sw_desc->async_tx.flags = flags; |
c2110923 DW |
538 | } |
539 | spin_unlock_bh(&iop_chan->lock); | |
540 | ||
541 | return sw_desc ? &sw_desc->async_tx : NULL; | |
542 | } | |
543 | ||
c2110923 | 544 | static struct dma_async_tx_descriptor * |
0036731c | 545 | iop_adma_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dma_dest, |
d4c56f97 | 546 | dma_addr_t dma_src, size_t len, unsigned long flags) |
c2110923 DW |
547 | { |
548 | struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan); | |
549 | struct iop_adma_desc_slot *sw_desc, *grp_start; | |
550 | int slot_cnt, slots_per_op; | |
551 | ||
552 | if (unlikely(!len)) | |
553 | return NULL; | |
554 | BUG_ON(unlikely(len > IOP_ADMA_MAX_BYTE_COUNT)); | |
555 | ||
556 | dev_dbg(iop_chan->device->common.dev, "%s len: %u\n", | |
3d9b525b | 557 | __func__, len); |
c2110923 DW |
558 | |
559 | spin_lock_bh(&iop_chan->lock); | |
560 | slot_cnt = iop_chan_memcpy_slot_count(len, &slots_per_op); | |
561 | sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op); | |
562 | if (sw_desc) { | |
563 | grp_start = sw_desc->group_head; | |
d4c56f97 | 564 | iop_desc_init_memcpy(grp_start, flags); |
c2110923 | 565 | iop_desc_set_byte_count(grp_start, iop_chan, len); |
0036731c DW |
566 | iop_desc_set_dest_addr(grp_start, iop_chan, dma_dest); |
567 | iop_desc_set_memcpy_src_addr(grp_start, dma_src); | |
c2110923 DW |
568 | sw_desc->unmap_src_cnt = 1; |
569 | sw_desc->unmap_len = len; | |
636bdeaa | 570 | sw_desc->async_tx.flags = flags; |
c2110923 DW |
571 | } |
572 | spin_unlock_bh(&iop_chan->lock); | |
573 | ||
574 | return sw_desc ? &sw_desc->async_tx : NULL; | |
575 | } | |
576 | ||
577 | static struct dma_async_tx_descriptor * | |
0036731c | 578 | iop_adma_prep_dma_memset(struct dma_chan *chan, dma_addr_t dma_dest, |
d4c56f97 | 579 | int value, size_t len, unsigned long flags) |
c2110923 DW |
580 | { |
581 | struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan); | |
582 | struct iop_adma_desc_slot *sw_desc, *grp_start; | |
583 | int slot_cnt, slots_per_op; | |
584 | ||
585 | if (unlikely(!len)) | |
586 | return NULL; | |
587 | BUG_ON(unlikely(len > IOP_ADMA_MAX_BYTE_COUNT)); | |
588 | ||
589 | dev_dbg(iop_chan->device->common.dev, "%s len: %u\n", | |
3d9b525b | 590 | __func__, len); |
c2110923 DW |
591 | |
592 | spin_lock_bh(&iop_chan->lock); | |
593 | slot_cnt = iop_chan_memset_slot_count(len, &slots_per_op); | |
594 | sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op); | |
595 | if (sw_desc) { | |
596 | grp_start = sw_desc->group_head; | |
d4c56f97 | 597 | iop_desc_init_memset(grp_start, flags); |
c2110923 DW |
598 | iop_desc_set_byte_count(grp_start, iop_chan, len); |
599 | iop_desc_set_block_fill_val(grp_start, value); | |
0036731c | 600 | iop_desc_set_dest_addr(grp_start, iop_chan, dma_dest); |
c2110923 DW |
601 | sw_desc->unmap_src_cnt = 1; |
602 | sw_desc->unmap_len = len; | |
636bdeaa | 603 | sw_desc->async_tx.flags = flags; |
c2110923 DW |
604 | } |
605 | spin_unlock_bh(&iop_chan->lock); | |
606 | ||
607 | return sw_desc ? &sw_desc->async_tx : NULL; | |
608 | } | |
609 | ||
c2110923 | 610 | static struct dma_async_tx_descriptor * |
0036731c DW |
611 | iop_adma_prep_dma_xor(struct dma_chan *chan, dma_addr_t dma_dest, |
612 | dma_addr_t *dma_src, unsigned int src_cnt, size_t len, | |
d4c56f97 | 613 | unsigned long flags) |
c2110923 DW |
614 | { |
615 | struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan); | |
616 | struct iop_adma_desc_slot *sw_desc, *grp_start; | |
617 | int slot_cnt, slots_per_op; | |
618 | ||
619 | if (unlikely(!len)) | |
620 | return NULL; | |
621 | BUG_ON(unlikely(len > IOP_ADMA_XOR_MAX_BYTE_COUNT)); | |
622 | ||
623 | dev_dbg(iop_chan->device->common.dev, | |
d4c56f97 | 624 | "%s src_cnt: %d len: %u flags: %lx\n", |
3d9b525b | 625 | __func__, src_cnt, len, flags); |
c2110923 DW |
626 | |
627 | spin_lock_bh(&iop_chan->lock); | |
628 | slot_cnt = iop_chan_xor_slot_count(len, src_cnt, &slots_per_op); | |
629 | sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op); | |
630 | if (sw_desc) { | |
631 | grp_start = sw_desc->group_head; | |
d4c56f97 | 632 | iop_desc_init_xor(grp_start, src_cnt, flags); |
c2110923 | 633 | iop_desc_set_byte_count(grp_start, iop_chan, len); |
0036731c | 634 | iop_desc_set_dest_addr(grp_start, iop_chan, dma_dest); |
c2110923 DW |
635 | sw_desc->unmap_src_cnt = src_cnt; |
636 | sw_desc->unmap_len = len; | |
636bdeaa | 637 | sw_desc->async_tx.flags = flags; |
0036731c DW |
638 | while (src_cnt--) |
639 | iop_desc_set_xor_src_addr(grp_start, src_cnt, | |
640 | dma_src[src_cnt]); | |
c2110923 DW |
641 | } |
642 | spin_unlock_bh(&iop_chan->lock); | |
643 | ||
644 | return sw_desc ? &sw_desc->async_tx : NULL; | |
645 | } | |
646 | ||
c2110923 | 647 | static struct dma_async_tx_descriptor * |
0036731c DW |
648 | iop_adma_prep_dma_zero_sum(struct dma_chan *chan, dma_addr_t *dma_src, |
649 | unsigned int src_cnt, size_t len, u32 *result, | |
d4c56f97 | 650 | unsigned long flags) |
c2110923 DW |
651 | { |
652 | struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan); | |
653 | struct iop_adma_desc_slot *sw_desc, *grp_start; | |
654 | int slot_cnt, slots_per_op; | |
655 | ||
656 | if (unlikely(!len)) | |
657 | return NULL; | |
658 | ||
659 | dev_dbg(iop_chan->device->common.dev, "%s src_cnt: %d len: %u\n", | |
3d9b525b | 660 | __func__, src_cnt, len); |
c2110923 DW |
661 | |
662 | spin_lock_bh(&iop_chan->lock); | |
663 | slot_cnt = iop_chan_zero_sum_slot_count(len, src_cnt, &slots_per_op); | |
664 | sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op); | |
665 | if (sw_desc) { | |
666 | grp_start = sw_desc->group_head; | |
d4c56f97 | 667 | iop_desc_init_zero_sum(grp_start, src_cnt, flags); |
c2110923 DW |
668 | iop_desc_set_zero_sum_byte_count(grp_start, len); |
669 | grp_start->xor_check_result = result; | |
670 | pr_debug("\t%s: grp_start->xor_check_result: %p\n", | |
3d9b525b | 671 | __func__, grp_start->xor_check_result); |
c2110923 DW |
672 | sw_desc->unmap_src_cnt = src_cnt; |
673 | sw_desc->unmap_len = len; | |
636bdeaa | 674 | sw_desc->async_tx.flags = flags; |
0036731c DW |
675 | while (src_cnt--) |
676 | iop_desc_set_zero_sum_src_addr(grp_start, src_cnt, | |
677 | dma_src[src_cnt]); | |
c2110923 DW |
678 | } |
679 | spin_unlock_bh(&iop_chan->lock); | |
680 | ||
681 | return sw_desc ? &sw_desc->async_tx : NULL; | |
682 | } | |
683 | ||
c2110923 DW |
684 | static void iop_adma_free_chan_resources(struct dma_chan *chan) |
685 | { | |
686 | struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan); | |
687 | struct iop_adma_desc_slot *iter, *_iter; | |
688 | int in_use_descs = 0; | |
689 | ||
690 | iop_adma_slot_cleanup(iop_chan); | |
691 | ||
692 | spin_lock_bh(&iop_chan->lock); | |
693 | list_for_each_entry_safe(iter, _iter, &iop_chan->chain, | |
694 | chain_node) { | |
695 | in_use_descs++; | |
696 | list_del(&iter->chain_node); | |
697 | } | |
698 | list_for_each_entry_safe_reverse( | |
699 | iter, _iter, &iop_chan->all_slots, slot_node) { | |
700 | list_del(&iter->slot_node); | |
701 | kfree(iter); | |
702 | iop_chan->slots_allocated--; | |
703 | } | |
704 | iop_chan->last_used = NULL; | |
705 | ||
706 | dev_dbg(iop_chan->device->common.dev, "%s slots_allocated %d\n", | |
3d9b525b | 707 | __func__, iop_chan->slots_allocated); |
c2110923 DW |
708 | spin_unlock_bh(&iop_chan->lock); |
709 | ||
710 | /* one is ok since we left it on there on purpose */ | |
711 | if (in_use_descs > 1) | |
712 | printk(KERN_ERR "IOP: Freeing %d in use descriptors!\n", | |
713 | in_use_descs - 1); | |
714 | } | |
715 | ||
716 | /** | |
717 | * iop_adma_is_complete - poll the status of an ADMA transaction | |
718 | * @chan: ADMA channel handle | |
719 | * @cookie: ADMA transaction identifier | |
720 | */ | |
721 | static enum dma_status iop_adma_is_complete(struct dma_chan *chan, | |
722 | dma_cookie_t cookie, | |
723 | dma_cookie_t *done, | |
724 | dma_cookie_t *used) | |
725 | { | |
726 | struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan); | |
727 | dma_cookie_t last_used; | |
728 | dma_cookie_t last_complete; | |
729 | enum dma_status ret; | |
730 | ||
731 | last_used = chan->cookie; | |
732 | last_complete = iop_chan->completed_cookie; | |
733 | ||
734 | if (done) | |
735 | *done = last_complete; | |
736 | if (used) | |
737 | *used = last_used; | |
738 | ||
739 | ret = dma_async_is_complete(cookie, last_complete, last_used); | |
740 | if (ret == DMA_SUCCESS) | |
741 | return ret; | |
742 | ||
743 | iop_adma_slot_cleanup(iop_chan); | |
744 | ||
745 | last_used = chan->cookie; | |
746 | last_complete = iop_chan->completed_cookie; | |
747 | ||
748 | if (done) | |
749 | *done = last_complete; | |
750 | if (used) | |
751 | *used = last_used; | |
752 | ||
753 | return dma_async_is_complete(cookie, last_complete, last_used); | |
754 | } | |
755 | ||
756 | static irqreturn_t iop_adma_eot_handler(int irq, void *data) | |
757 | { | |
758 | struct iop_adma_chan *chan = data; | |
759 | ||
3d9b525b | 760 | dev_dbg(chan->device->common.dev, "%s\n", __func__); |
c2110923 DW |
761 | |
762 | tasklet_schedule(&chan->irq_tasklet); | |
763 | ||
764 | iop_adma_device_clear_eot_status(chan); | |
765 | ||
766 | return IRQ_HANDLED; | |
767 | } | |
768 | ||
769 | static irqreturn_t iop_adma_eoc_handler(int irq, void *data) | |
770 | { | |
771 | struct iop_adma_chan *chan = data; | |
772 | ||
3d9b525b | 773 | dev_dbg(chan->device->common.dev, "%s\n", __func__); |
c2110923 DW |
774 | |
775 | tasklet_schedule(&chan->irq_tasklet); | |
776 | ||
777 | iop_adma_device_clear_eoc_status(chan); | |
778 | ||
779 | return IRQ_HANDLED; | |
780 | } | |
781 | ||
782 | static irqreturn_t iop_adma_err_handler(int irq, void *data) | |
783 | { | |
784 | struct iop_adma_chan *chan = data; | |
785 | unsigned long status = iop_chan_get_status(chan); | |
786 | ||
787 | dev_printk(KERN_ERR, chan->device->common.dev, | |
788 | "error ( %s%s%s%s%s%s%s)\n", | |
789 | iop_is_err_int_parity(status, chan) ? "int_parity " : "", | |
790 | iop_is_err_mcu_abort(status, chan) ? "mcu_abort " : "", | |
791 | iop_is_err_int_tabort(status, chan) ? "int_tabort " : "", | |
792 | iop_is_err_int_mabort(status, chan) ? "int_mabort " : "", | |
793 | iop_is_err_pci_tabort(status, chan) ? "pci_tabort " : "", | |
794 | iop_is_err_pci_mabort(status, chan) ? "pci_mabort " : "", | |
795 | iop_is_err_split_tx(status, chan) ? "split_tx " : ""); | |
796 | ||
797 | iop_adma_device_clear_err_status(chan); | |
798 | ||
799 | BUG(); | |
800 | ||
801 | return IRQ_HANDLED; | |
802 | } | |
803 | ||
804 | static void iop_adma_issue_pending(struct dma_chan *chan) | |
805 | { | |
806 | struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan); | |
807 | ||
808 | if (iop_chan->pending) { | |
809 | iop_chan->pending = 0; | |
810 | iop_chan_append(iop_chan); | |
811 | } | |
812 | } | |
813 | ||
814 | /* | |
815 | * Perform a transaction to verify the HW works. | |
816 | */ | |
817 | #define IOP_ADMA_TEST_SIZE 2000 | |
818 | ||
819 | static int __devinit iop_adma_memcpy_self_test(struct iop_adma_device *device) | |
820 | { | |
821 | int i; | |
822 | void *src, *dest; | |
823 | dma_addr_t src_dma, dest_dma; | |
824 | struct dma_chan *dma_chan; | |
825 | dma_cookie_t cookie; | |
826 | struct dma_async_tx_descriptor *tx; | |
827 | int err = 0; | |
828 | struct iop_adma_chan *iop_chan; | |
829 | ||
3d9b525b | 830 | dev_dbg(device->common.dev, "%s\n", __func__); |
c2110923 | 831 | |
eccf2144 | 832 | src = kmalloc(IOP_ADMA_TEST_SIZE, GFP_KERNEL); |
c2110923 DW |
833 | if (!src) |
834 | return -ENOMEM; | |
eccf2144 | 835 | dest = kzalloc(IOP_ADMA_TEST_SIZE, GFP_KERNEL); |
c2110923 DW |
836 | if (!dest) { |
837 | kfree(src); | |
838 | return -ENOMEM; | |
839 | } | |
840 | ||
841 | /* Fill in src buffer */ | |
842 | for (i = 0; i < IOP_ADMA_TEST_SIZE; i++) | |
843 | ((u8 *) src)[i] = (u8)i; | |
844 | ||
c2110923 DW |
845 | /* Start copy, using first DMA channel */ |
846 | dma_chan = container_of(device->common.channels.next, | |
847 | struct dma_chan, | |
848 | device_node); | |
848c536a | 849 | if (iop_adma_alloc_chan_resources(dma_chan, NULL) < 1) { |
c2110923 DW |
850 | err = -ENODEV; |
851 | goto out; | |
852 | } | |
853 | ||
c2110923 DW |
854 | dest_dma = dma_map_single(dma_chan->device->dev, dest, |
855 | IOP_ADMA_TEST_SIZE, DMA_FROM_DEVICE); | |
c2110923 DW |
856 | src_dma = dma_map_single(dma_chan->device->dev, src, |
857 | IOP_ADMA_TEST_SIZE, DMA_TO_DEVICE); | |
0036731c | 858 | tx = iop_adma_prep_dma_memcpy(dma_chan, dest_dma, src_dma, |
636bdeaa DW |
859 | IOP_ADMA_TEST_SIZE, |
860 | DMA_PREP_INTERRUPT | DMA_CTRL_ACK); | |
c2110923 DW |
861 | |
862 | cookie = iop_adma_tx_submit(tx); | |
863 | iop_adma_issue_pending(dma_chan); | |
c2110923 DW |
864 | msleep(1); |
865 | ||
866 | if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) != | |
867 | DMA_SUCCESS) { | |
868 | dev_printk(KERN_ERR, dma_chan->device->dev, | |
869 | "Self-test copy timed out, disabling\n"); | |
870 | err = -ENODEV; | |
871 | goto free_resources; | |
872 | } | |
873 | ||
874 | iop_chan = to_iop_adma_chan(dma_chan); | |
875 | dma_sync_single_for_cpu(&iop_chan->device->pdev->dev, dest_dma, | |
876 | IOP_ADMA_TEST_SIZE, DMA_FROM_DEVICE); | |
877 | if (memcmp(src, dest, IOP_ADMA_TEST_SIZE)) { | |
878 | dev_printk(KERN_ERR, dma_chan->device->dev, | |
879 | "Self-test copy failed compare, disabling\n"); | |
880 | err = -ENODEV; | |
881 | goto free_resources; | |
882 | } | |
883 | ||
884 | free_resources: | |
885 | iop_adma_free_chan_resources(dma_chan); | |
886 | out: | |
887 | kfree(src); | |
888 | kfree(dest); | |
889 | return err; | |
890 | } | |
891 | ||
892 | #define IOP_ADMA_NUM_SRC_TEST 4 /* must be <= 15 */ | |
893 | static int __devinit | |
894 | iop_adma_xor_zero_sum_self_test(struct iop_adma_device *device) | |
895 | { | |
896 | int i, src_idx; | |
897 | struct page *dest; | |
898 | struct page *xor_srcs[IOP_ADMA_NUM_SRC_TEST]; | |
899 | struct page *zero_sum_srcs[IOP_ADMA_NUM_SRC_TEST + 1]; | |
0036731c | 900 | dma_addr_t dma_srcs[IOP_ADMA_NUM_SRC_TEST + 1]; |
c2110923 DW |
901 | dma_addr_t dma_addr, dest_dma; |
902 | struct dma_async_tx_descriptor *tx; | |
903 | struct dma_chan *dma_chan; | |
904 | dma_cookie_t cookie; | |
905 | u8 cmp_byte = 0; | |
906 | u32 cmp_word; | |
907 | u32 zero_sum_result; | |
908 | int err = 0; | |
909 | struct iop_adma_chan *iop_chan; | |
910 | ||
3d9b525b | 911 | dev_dbg(device->common.dev, "%s\n", __func__); |
c2110923 DW |
912 | |
913 | for (src_idx = 0; src_idx < IOP_ADMA_NUM_SRC_TEST; src_idx++) { | |
914 | xor_srcs[src_idx] = alloc_page(GFP_KERNEL); | |
915 | if (!xor_srcs[src_idx]) | |
916 | while (src_idx--) { | |
917 | __free_page(xor_srcs[src_idx]); | |
918 | return -ENOMEM; | |
919 | } | |
920 | } | |
921 | ||
922 | dest = alloc_page(GFP_KERNEL); | |
923 | if (!dest) | |
924 | while (src_idx--) { | |
925 | __free_page(xor_srcs[src_idx]); | |
926 | return -ENOMEM; | |
927 | } | |
928 | ||
929 | /* Fill in src buffers */ | |
930 | for (src_idx = 0; src_idx < IOP_ADMA_NUM_SRC_TEST; src_idx++) { | |
931 | u8 *ptr = page_address(xor_srcs[src_idx]); | |
932 | for (i = 0; i < PAGE_SIZE; i++) | |
933 | ptr[i] = (1 << src_idx); | |
934 | } | |
935 | ||
936 | for (src_idx = 0; src_idx < IOP_ADMA_NUM_SRC_TEST; src_idx++) | |
937 | cmp_byte ^= (u8) (1 << src_idx); | |
938 | ||
939 | cmp_word = (cmp_byte << 24) | (cmp_byte << 16) | | |
940 | (cmp_byte << 8) | cmp_byte; | |
941 | ||
942 | memset(page_address(dest), 0, PAGE_SIZE); | |
943 | ||
944 | dma_chan = container_of(device->common.channels.next, | |
945 | struct dma_chan, | |
946 | device_node); | |
848c536a | 947 | if (iop_adma_alloc_chan_resources(dma_chan, NULL) < 1) { |
c2110923 DW |
948 | err = -ENODEV; |
949 | goto out; | |
950 | } | |
951 | ||
952 | /* test xor */ | |
c2110923 DW |
953 | dest_dma = dma_map_page(dma_chan->device->dev, dest, 0, |
954 | PAGE_SIZE, DMA_FROM_DEVICE); | |
0036731c DW |
955 | for (i = 0; i < IOP_ADMA_NUM_SRC_TEST; i++) |
956 | dma_srcs[i] = dma_map_page(dma_chan->device->dev, xor_srcs[i], | |
957 | 0, PAGE_SIZE, DMA_TO_DEVICE); | |
958 | tx = iop_adma_prep_dma_xor(dma_chan, dest_dma, dma_srcs, | |
636bdeaa DW |
959 | IOP_ADMA_NUM_SRC_TEST, PAGE_SIZE, |
960 | DMA_PREP_INTERRUPT | DMA_CTRL_ACK); | |
c2110923 DW |
961 | |
962 | cookie = iop_adma_tx_submit(tx); | |
963 | iop_adma_issue_pending(dma_chan); | |
c2110923 DW |
964 | msleep(8); |
965 | ||
966 | if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) != | |
967 | DMA_SUCCESS) { | |
968 | dev_printk(KERN_ERR, dma_chan->device->dev, | |
969 | "Self-test xor timed out, disabling\n"); | |
970 | err = -ENODEV; | |
971 | goto free_resources; | |
972 | } | |
973 | ||
974 | iop_chan = to_iop_adma_chan(dma_chan); | |
975 | dma_sync_single_for_cpu(&iop_chan->device->pdev->dev, dest_dma, | |
976 | PAGE_SIZE, DMA_FROM_DEVICE); | |
977 | for (i = 0; i < (PAGE_SIZE / sizeof(u32)); i++) { | |
978 | u32 *ptr = page_address(dest); | |
979 | if (ptr[i] != cmp_word) { | |
980 | dev_printk(KERN_ERR, dma_chan->device->dev, | |
981 | "Self-test xor failed compare, disabling\n"); | |
982 | err = -ENODEV; | |
983 | goto free_resources; | |
984 | } | |
985 | } | |
986 | dma_sync_single_for_device(&iop_chan->device->pdev->dev, dest_dma, | |
987 | PAGE_SIZE, DMA_TO_DEVICE); | |
988 | ||
989 | /* skip zero sum if the capability is not present */ | |
990 | if (!dma_has_cap(DMA_ZERO_SUM, dma_chan->device->cap_mask)) | |
991 | goto free_resources; | |
992 | ||
993 | /* zero sum the sources with the destintation page */ | |
994 | for (i = 0; i < IOP_ADMA_NUM_SRC_TEST; i++) | |
995 | zero_sum_srcs[i] = xor_srcs[i]; | |
996 | zero_sum_srcs[i] = dest; | |
997 | ||
998 | zero_sum_result = 1; | |
999 | ||
0036731c DW |
1000 | for (i = 0; i < IOP_ADMA_NUM_SRC_TEST + 1; i++) |
1001 | dma_srcs[i] = dma_map_page(dma_chan->device->dev, | |
1002 | zero_sum_srcs[i], 0, PAGE_SIZE, | |
1003 | DMA_TO_DEVICE); | |
1004 | tx = iop_adma_prep_dma_zero_sum(dma_chan, dma_srcs, | |
1005 | IOP_ADMA_NUM_SRC_TEST + 1, PAGE_SIZE, | |
636bdeaa DW |
1006 | &zero_sum_result, |
1007 | DMA_PREP_INTERRUPT | DMA_CTRL_ACK); | |
c2110923 DW |
1008 | |
1009 | cookie = iop_adma_tx_submit(tx); | |
1010 | iop_adma_issue_pending(dma_chan); | |
c2110923 DW |
1011 | msleep(8); |
1012 | ||
1013 | if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) != DMA_SUCCESS) { | |
1014 | dev_printk(KERN_ERR, dma_chan->device->dev, | |
1015 | "Self-test zero sum timed out, disabling\n"); | |
1016 | err = -ENODEV; | |
1017 | goto free_resources; | |
1018 | } | |
1019 | ||
1020 | if (zero_sum_result != 0) { | |
1021 | dev_printk(KERN_ERR, dma_chan->device->dev, | |
1022 | "Self-test zero sum failed compare, disabling\n"); | |
1023 | err = -ENODEV; | |
1024 | goto free_resources; | |
1025 | } | |
1026 | ||
1027 | /* test memset */ | |
c2110923 DW |
1028 | dma_addr = dma_map_page(dma_chan->device->dev, dest, 0, |
1029 | PAGE_SIZE, DMA_FROM_DEVICE); | |
636bdeaa DW |
1030 | tx = iop_adma_prep_dma_memset(dma_chan, dma_addr, 0, PAGE_SIZE, |
1031 | DMA_PREP_INTERRUPT | DMA_CTRL_ACK); | |
c2110923 DW |
1032 | |
1033 | cookie = iop_adma_tx_submit(tx); | |
1034 | iop_adma_issue_pending(dma_chan); | |
c2110923 DW |
1035 | msleep(8); |
1036 | ||
1037 | if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) != DMA_SUCCESS) { | |
1038 | dev_printk(KERN_ERR, dma_chan->device->dev, | |
1039 | "Self-test memset timed out, disabling\n"); | |
1040 | err = -ENODEV; | |
1041 | goto free_resources; | |
1042 | } | |
1043 | ||
1044 | for (i = 0; i < PAGE_SIZE/sizeof(u32); i++) { | |
1045 | u32 *ptr = page_address(dest); | |
1046 | if (ptr[i]) { | |
1047 | dev_printk(KERN_ERR, dma_chan->device->dev, | |
1048 | "Self-test memset failed compare, disabling\n"); | |
1049 | err = -ENODEV; | |
1050 | goto free_resources; | |
1051 | } | |
1052 | } | |
1053 | ||
1054 | /* test for non-zero parity sum */ | |
1055 | zero_sum_result = 0; | |
0036731c DW |
1056 | for (i = 0; i < IOP_ADMA_NUM_SRC_TEST + 1; i++) |
1057 | dma_srcs[i] = dma_map_page(dma_chan->device->dev, | |
1058 | zero_sum_srcs[i], 0, PAGE_SIZE, | |
1059 | DMA_TO_DEVICE); | |
1060 | tx = iop_adma_prep_dma_zero_sum(dma_chan, dma_srcs, | |
1061 | IOP_ADMA_NUM_SRC_TEST + 1, PAGE_SIZE, | |
636bdeaa DW |
1062 | &zero_sum_result, |
1063 | DMA_PREP_INTERRUPT | DMA_CTRL_ACK); | |
c2110923 DW |
1064 | |
1065 | cookie = iop_adma_tx_submit(tx); | |
1066 | iop_adma_issue_pending(dma_chan); | |
c2110923 DW |
1067 | msleep(8); |
1068 | ||
1069 | if (iop_adma_is_complete(dma_chan, cookie, NULL, NULL) != DMA_SUCCESS) { | |
1070 | dev_printk(KERN_ERR, dma_chan->device->dev, | |
1071 | "Self-test non-zero sum timed out, disabling\n"); | |
1072 | err = -ENODEV; | |
1073 | goto free_resources; | |
1074 | } | |
1075 | ||
1076 | if (zero_sum_result != 1) { | |
1077 | dev_printk(KERN_ERR, dma_chan->device->dev, | |
1078 | "Self-test non-zero sum failed compare, disabling\n"); | |
1079 | err = -ENODEV; | |
1080 | goto free_resources; | |
1081 | } | |
1082 | ||
1083 | free_resources: | |
1084 | iop_adma_free_chan_resources(dma_chan); | |
1085 | out: | |
1086 | src_idx = IOP_ADMA_NUM_SRC_TEST; | |
1087 | while (src_idx--) | |
1088 | __free_page(xor_srcs[src_idx]); | |
1089 | __free_page(dest); | |
1090 | return err; | |
1091 | } | |
1092 | ||
1093 | static int __devexit iop_adma_remove(struct platform_device *dev) | |
1094 | { | |
1095 | struct iop_adma_device *device = platform_get_drvdata(dev); | |
1096 | struct dma_chan *chan, *_chan; | |
1097 | struct iop_adma_chan *iop_chan; | |
1098 | int i; | |
1099 | struct iop_adma_platform_data *plat_data = dev->dev.platform_data; | |
1100 | ||
1101 | dma_async_device_unregister(&device->common); | |
1102 | ||
1103 | for (i = 0; i < 3; i++) { | |
1104 | unsigned int irq; | |
1105 | irq = platform_get_irq(dev, i); | |
1106 | free_irq(irq, device); | |
1107 | } | |
1108 | ||
1109 | dma_free_coherent(&dev->dev, plat_data->pool_size, | |
1110 | device->dma_desc_pool_virt, device->dma_desc_pool); | |
1111 | ||
1112 | do { | |
1113 | struct resource *res; | |
1114 | res = platform_get_resource(dev, IORESOURCE_MEM, 0); | |
1115 | release_mem_region(res->start, res->end - res->start); | |
1116 | } while (0); | |
1117 | ||
1118 | list_for_each_entry_safe(chan, _chan, &device->common.channels, | |
1119 | device_node) { | |
1120 | iop_chan = to_iop_adma_chan(chan); | |
1121 | list_del(&chan->device_node); | |
1122 | kfree(iop_chan); | |
1123 | } | |
1124 | kfree(device); | |
1125 | ||
1126 | return 0; | |
1127 | } | |
1128 | ||
1129 | static int __devinit iop_adma_probe(struct platform_device *pdev) | |
1130 | { | |
1131 | struct resource *res; | |
1132 | int ret = 0, i; | |
1133 | struct iop_adma_device *adev; | |
1134 | struct iop_adma_chan *iop_chan; | |
1135 | struct dma_device *dma_dev; | |
1136 | struct iop_adma_platform_data *plat_data = pdev->dev.platform_data; | |
1137 | ||
1138 | res = platform_get_resource(pdev, IORESOURCE_MEM, 0); | |
1139 | if (!res) | |
1140 | return -ENODEV; | |
1141 | ||
1142 | if (!devm_request_mem_region(&pdev->dev, res->start, | |
1143 | res->end - res->start, pdev->name)) | |
1144 | return -EBUSY; | |
1145 | ||
1146 | adev = kzalloc(sizeof(*adev), GFP_KERNEL); | |
1147 | if (!adev) | |
1148 | return -ENOMEM; | |
1149 | dma_dev = &adev->common; | |
1150 | ||
1151 | /* allocate coherent memory for hardware descriptors | |
1152 | * note: writecombine gives slightly better performance, but | |
1153 | * requires that we explicitly flush the writes | |
1154 | */ | |
1155 | if ((adev->dma_desc_pool_virt = dma_alloc_writecombine(&pdev->dev, | |
1156 | plat_data->pool_size, | |
1157 | &adev->dma_desc_pool, | |
1158 | GFP_KERNEL)) == NULL) { | |
1159 | ret = -ENOMEM; | |
1160 | goto err_free_adev; | |
1161 | } | |
1162 | ||
1163 | dev_dbg(&pdev->dev, "%s: allocted descriptor pool virt %p phys %p\n", | |
3d9b525b | 1164 | __func__, adev->dma_desc_pool_virt, |
c2110923 DW |
1165 | (void *) adev->dma_desc_pool); |
1166 | ||
1167 | adev->id = plat_data->hw_id; | |
1168 | ||
1169 | /* discover transaction capabilites from the platform data */ | |
1170 | dma_dev->cap_mask = plat_data->cap_mask; | |
1171 | ||
1172 | adev->pdev = pdev; | |
1173 | platform_set_drvdata(pdev, adev); | |
1174 | ||
1175 | INIT_LIST_HEAD(&dma_dev->channels); | |
1176 | ||
1177 | /* set base routines */ | |
1178 | dma_dev->device_alloc_chan_resources = iop_adma_alloc_chan_resources; | |
1179 | dma_dev->device_free_chan_resources = iop_adma_free_chan_resources; | |
1180 | dma_dev->device_is_tx_complete = iop_adma_is_complete; | |
1181 | dma_dev->device_issue_pending = iop_adma_issue_pending; | |
c2110923 DW |
1182 | dma_dev->dev = &pdev->dev; |
1183 | ||
1184 | /* set prep routines based on capability */ | |
1185 | if (dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask)) | |
1186 | dma_dev->device_prep_dma_memcpy = iop_adma_prep_dma_memcpy; | |
1187 | if (dma_has_cap(DMA_MEMSET, dma_dev->cap_mask)) | |
1188 | dma_dev->device_prep_dma_memset = iop_adma_prep_dma_memset; | |
1189 | if (dma_has_cap(DMA_XOR, dma_dev->cap_mask)) { | |
1190 | dma_dev->max_xor = iop_adma_get_max_xor(); | |
1191 | dma_dev->device_prep_dma_xor = iop_adma_prep_dma_xor; | |
1192 | } | |
1193 | if (dma_has_cap(DMA_ZERO_SUM, dma_dev->cap_mask)) | |
1194 | dma_dev->device_prep_dma_zero_sum = | |
1195 | iop_adma_prep_dma_zero_sum; | |
1196 | if (dma_has_cap(DMA_INTERRUPT, dma_dev->cap_mask)) | |
1197 | dma_dev->device_prep_dma_interrupt = | |
1198 | iop_adma_prep_dma_interrupt; | |
1199 | ||
1200 | iop_chan = kzalloc(sizeof(*iop_chan), GFP_KERNEL); | |
1201 | if (!iop_chan) { | |
1202 | ret = -ENOMEM; | |
1203 | goto err_free_dma; | |
1204 | } | |
1205 | iop_chan->device = adev; | |
1206 | ||
1207 | iop_chan->mmr_base = devm_ioremap(&pdev->dev, res->start, | |
1208 | res->end - res->start); | |
1209 | if (!iop_chan->mmr_base) { | |
1210 | ret = -ENOMEM; | |
1211 | goto err_free_iop_chan; | |
1212 | } | |
1213 | tasklet_init(&iop_chan->irq_tasklet, iop_adma_tasklet, (unsigned long) | |
1214 | iop_chan); | |
1215 | ||
1216 | /* clear errors before enabling interrupts */ | |
1217 | iop_adma_device_clear_err_status(iop_chan); | |
1218 | ||
1219 | for (i = 0; i < 3; i++) { | |
1220 | irq_handler_t handler[] = { iop_adma_eot_handler, | |
1221 | iop_adma_eoc_handler, | |
1222 | iop_adma_err_handler }; | |
1223 | int irq = platform_get_irq(pdev, i); | |
1224 | if (irq < 0) { | |
1225 | ret = -ENXIO; | |
1226 | goto err_free_iop_chan; | |
1227 | } else { | |
1228 | ret = devm_request_irq(&pdev->dev, irq, | |
1229 | handler[i], 0, pdev->name, iop_chan); | |
1230 | if (ret) | |
1231 | goto err_free_iop_chan; | |
1232 | } | |
1233 | } | |
1234 | ||
1235 | spin_lock_init(&iop_chan->lock); | |
c2110923 DW |
1236 | INIT_LIST_HEAD(&iop_chan->chain); |
1237 | INIT_LIST_HEAD(&iop_chan->all_slots); | |
1238 | INIT_RCU_HEAD(&iop_chan->common.rcu); | |
1239 | iop_chan->common.device = dma_dev; | |
1240 | list_add_tail(&iop_chan->common.device_node, &dma_dev->channels); | |
1241 | ||
1242 | if (dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask)) { | |
1243 | ret = iop_adma_memcpy_self_test(adev); | |
1244 | dev_dbg(&pdev->dev, "memcpy self test returned %d\n", ret); | |
1245 | if (ret) | |
1246 | goto err_free_iop_chan; | |
1247 | } | |
1248 | ||
1249 | if (dma_has_cap(DMA_XOR, dma_dev->cap_mask) || | |
1250 | dma_has_cap(DMA_MEMSET, dma_dev->cap_mask)) { | |
1251 | ret = iop_adma_xor_zero_sum_self_test(adev); | |
1252 | dev_dbg(&pdev->dev, "xor self test returned %d\n", ret); | |
1253 | if (ret) | |
1254 | goto err_free_iop_chan; | |
1255 | } | |
1256 | ||
1257 | dev_printk(KERN_INFO, &pdev->dev, "Intel(R) IOP: " | |
1258 | "( %s%s%s%s%s%s%s%s%s%s)\n", | |
1259 | dma_has_cap(DMA_PQ_XOR, dma_dev->cap_mask) ? "pq_xor " : "", | |
1260 | dma_has_cap(DMA_PQ_UPDATE, dma_dev->cap_mask) ? "pq_update " : "", | |
1261 | dma_has_cap(DMA_PQ_ZERO_SUM, dma_dev->cap_mask) ? "pq_zero_sum " : "", | |
1262 | dma_has_cap(DMA_XOR, dma_dev->cap_mask) ? "xor " : "", | |
1263 | dma_has_cap(DMA_DUAL_XOR, dma_dev->cap_mask) ? "dual_xor " : "", | |
1264 | dma_has_cap(DMA_ZERO_SUM, dma_dev->cap_mask) ? "xor_zero_sum " : "", | |
1265 | dma_has_cap(DMA_MEMSET, dma_dev->cap_mask) ? "fill " : "", | |
1266 | dma_has_cap(DMA_MEMCPY_CRC32C, dma_dev->cap_mask) ? "cpy+crc " : "", | |
1267 | dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask) ? "cpy " : "", | |
1268 | dma_has_cap(DMA_INTERRUPT, dma_dev->cap_mask) ? "intr " : ""); | |
1269 | ||
1270 | dma_async_device_register(dma_dev); | |
1271 | goto out; | |
1272 | ||
1273 | err_free_iop_chan: | |
1274 | kfree(iop_chan); | |
1275 | err_free_dma: | |
1276 | dma_free_coherent(&adev->pdev->dev, plat_data->pool_size, | |
1277 | adev->dma_desc_pool_virt, adev->dma_desc_pool); | |
1278 | err_free_adev: | |
1279 | kfree(adev); | |
1280 | out: | |
1281 | return ret; | |
1282 | } | |
1283 | ||
1284 | static void iop_chan_start_null_memcpy(struct iop_adma_chan *iop_chan) | |
1285 | { | |
1286 | struct iop_adma_desc_slot *sw_desc, *grp_start; | |
1287 | dma_cookie_t cookie; | |
1288 | int slot_cnt, slots_per_op; | |
1289 | ||
3d9b525b | 1290 | dev_dbg(iop_chan->device->common.dev, "%s\n", __func__); |
c2110923 DW |
1291 | |
1292 | spin_lock_bh(&iop_chan->lock); | |
1293 | slot_cnt = iop_chan_memcpy_slot_count(0, &slots_per_op); | |
1294 | sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op); | |
1295 | if (sw_desc) { | |
1296 | grp_start = sw_desc->group_head; | |
1297 | ||
1298 | list_splice_init(&sw_desc->async_tx.tx_list, &iop_chan->chain); | |
636bdeaa | 1299 | async_tx_ack(&sw_desc->async_tx); |
c2110923 DW |
1300 | iop_desc_init_memcpy(grp_start, 0); |
1301 | iop_desc_set_byte_count(grp_start, iop_chan, 0); | |
1302 | iop_desc_set_dest_addr(grp_start, iop_chan, 0); | |
1303 | iop_desc_set_memcpy_src_addr(grp_start, 0); | |
1304 | ||
1305 | cookie = iop_chan->common.cookie; | |
1306 | cookie++; | |
1307 | if (cookie <= 1) | |
1308 | cookie = 2; | |
1309 | ||
1310 | /* initialize the completed cookie to be less than | |
1311 | * the most recently used cookie | |
1312 | */ | |
1313 | iop_chan->completed_cookie = cookie - 1; | |
1314 | iop_chan->common.cookie = sw_desc->async_tx.cookie = cookie; | |
1315 | ||
1316 | /* channel should not be busy */ | |
1317 | BUG_ON(iop_chan_is_busy(iop_chan)); | |
1318 | ||
1319 | /* clear any prior error-status bits */ | |
1320 | iop_adma_device_clear_err_status(iop_chan); | |
1321 | ||
1322 | /* disable operation */ | |
1323 | iop_chan_disable(iop_chan); | |
1324 | ||
1325 | /* set the descriptor address */ | |
1326 | iop_chan_set_next_descriptor(iop_chan, sw_desc->async_tx.phys); | |
1327 | ||
1328 | /* 1/ don't add pre-chained descriptors | |
1329 | * 2/ dummy read to flush next_desc write | |
1330 | */ | |
1331 | BUG_ON(iop_desc_get_next_desc(sw_desc)); | |
1332 | ||
1333 | /* run the descriptor */ | |
1334 | iop_chan_enable(iop_chan); | |
1335 | } else | |
1336 | dev_printk(KERN_ERR, iop_chan->device->common.dev, | |
1337 | "failed to allocate null descriptor\n"); | |
1338 | spin_unlock_bh(&iop_chan->lock); | |
1339 | } | |
1340 | ||
1341 | static void iop_chan_start_null_xor(struct iop_adma_chan *iop_chan) | |
1342 | { | |
1343 | struct iop_adma_desc_slot *sw_desc, *grp_start; | |
1344 | dma_cookie_t cookie; | |
1345 | int slot_cnt, slots_per_op; | |
1346 | ||
3d9b525b | 1347 | dev_dbg(iop_chan->device->common.dev, "%s\n", __func__); |
c2110923 DW |
1348 | |
1349 | spin_lock_bh(&iop_chan->lock); | |
1350 | slot_cnt = iop_chan_xor_slot_count(0, 2, &slots_per_op); | |
1351 | sw_desc = iop_adma_alloc_slots(iop_chan, slot_cnt, slots_per_op); | |
1352 | if (sw_desc) { | |
1353 | grp_start = sw_desc->group_head; | |
1354 | list_splice_init(&sw_desc->async_tx.tx_list, &iop_chan->chain); | |
636bdeaa | 1355 | async_tx_ack(&sw_desc->async_tx); |
c2110923 DW |
1356 | iop_desc_init_null_xor(grp_start, 2, 0); |
1357 | iop_desc_set_byte_count(grp_start, iop_chan, 0); | |
1358 | iop_desc_set_dest_addr(grp_start, iop_chan, 0); | |
1359 | iop_desc_set_xor_src_addr(grp_start, 0, 0); | |
1360 | iop_desc_set_xor_src_addr(grp_start, 1, 0); | |
1361 | ||
1362 | cookie = iop_chan->common.cookie; | |
1363 | cookie++; | |
1364 | if (cookie <= 1) | |
1365 | cookie = 2; | |
1366 | ||
1367 | /* initialize the completed cookie to be less than | |
1368 | * the most recently used cookie | |
1369 | */ | |
1370 | iop_chan->completed_cookie = cookie - 1; | |
1371 | iop_chan->common.cookie = sw_desc->async_tx.cookie = cookie; | |
1372 | ||
1373 | /* channel should not be busy */ | |
1374 | BUG_ON(iop_chan_is_busy(iop_chan)); | |
1375 | ||
1376 | /* clear any prior error-status bits */ | |
1377 | iop_adma_device_clear_err_status(iop_chan); | |
1378 | ||
1379 | /* disable operation */ | |
1380 | iop_chan_disable(iop_chan); | |
1381 | ||
1382 | /* set the descriptor address */ | |
1383 | iop_chan_set_next_descriptor(iop_chan, sw_desc->async_tx.phys); | |
1384 | ||
1385 | /* 1/ don't add pre-chained descriptors | |
1386 | * 2/ dummy read to flush next_desc write | |
1387 | */ | |
1388 | BUG_ON(iop_desc_get_next_desc(sw_desc)); | |
1389 | ||
1390 | /* run the descriptor */ | |
1391 | iop_chan_enable(iop_chan); | |
1392 | } else | |
1393 | dev_printk(KERN_ERR, iop_chan->device->common.dev, | |
1394 | "failed to allocate null descriptor\n"); | |
1395 | spin_unlock_bh(&iop_chan->lock); | |
1396 | } | |
1397 | ||
ebabe276 KS |
1398 | MODULE_ALIAS("platform:iop-adma"); |
1399 | ||
c2110923 DW |
1400 | static struct platform_driver iop_adma_driver = { |
1401 | .probe = iop_adma_probe, | |
1402 | .remove = iop_adma_remove, | |
1403 | .driver = { | |
1404 | .owner = THIS_MODULE, | |
1405 | .name = "iop-adma", | |
1406 | }, | |
1407 | }; | |
1408 | ||
1409 | static int __init iop_adma_init (void) | |
1410 | { | |
c2110923 DW |
1411 | return platform_driver_register(&iop_adma_driver); |
1412 | } | |
1413 | ||
af49d924 RR |
1414 | /* it's currently unsafe to unload this module */ |
1415 | #if 0 | |
c2110923 DW |
1416 | static void __exit iop_adma_exit (void) |
1417 | { | |
1418 | platform_driver_unregister(&iop_adma_driver); | |
1419 | return; | |
1420 | } | |
af49d924 RR |
1421 | module_exit(iop_adma_exit); |
1422 | #endif | |
c2110923 DW |
1423 | |
1424 | module_init(iop_adma_init); | |
c2110923 DW |
1425 | |
1426 | MODULE_AUTHOR("Intel Corporation"); | |
1427 | MODULE_DESCRIPTION("IOP ADMA Engine Driver"); | |
1428 | MODULE_LICENSE("GPL"); |