diff options
Diffstat (limited to 'drivers/dma')
-rw-r--r-- | drivers/dma/ioat/Makefile | 2 | ||||
-rw-r--r-- | drivers/dma/ioat/dma.c | 874 | ||||
-rw-r--r-- | drivers/dma/ioat/dma.h | 50 | ||||
-rw-r--r-- | drivers/dma/ioat/dma_v2.c | 750 | ||||
-rw-r--r-- | drivers/dma/ioat/dma_v2.h | 131 | ||||
-rw-r--r-- | drivers/dma/ioat/pci.c | 1 |
6 files changed, 1122 insertions, 686 deletions
diff --git a/drivers/dma/ioat/Makefile b/drivers/dma/ioat/Makefile index 2ce3d3a4270b..205a639e84df 100644 --- a/drivers/dma/ioat/Makefile +++ b/drivers/dma/ioat/Makefile | |||
@@ -1,2 +1,2 @@ | |||
1 | obj-$(CONFIG_INTEL_IOATDMA) += ioatdma.o | 1 | obj-$(CONFIG_INTEL_IOATDMA) += ioatdma.o |
2 | ioatdma-objs := pci.o dma.o dca.o | 2 | ioatdma-objs := pci.o dma.o dma_v2.o dca.o |
diff --git a/drivers/dma/ioat/dma.c b/drivers/dma/ioat/dma.c index 2e81e0c76e61..64b4d75059aa 100644 --- a/drivers/dma/ioat/dma.c +++ b/drivers/dma/ioat/dma.c | |||
@@ -38,28 +38,14 @@ | |||
38 | #include "registers.h" | 38 | #include "registers.h" |
39 | #include "hw.h" | 39 | #include "hw.h" |
40 | 40 | ||
41 | static int ioat_pending_level = 4; | 41 | int ioat_pending_level = 4; |
42 | module_param(ioat_pending_level, int, 0644); | 42 | module_param(ioat_pending_level, int, 0644); |
43 | MODULE_PARM_DESC(ioat_pending_level, | 43 | MODULE_PARM_DESC(ioat_pending_level, |
44 | "high-water mark for pushing ioat descriptors (default: 4)"); | 44 | "high-water mark for pushing ioat descriptors (default: 4)"); |
45 | 45 | ||
46 | static void ioat_dma_chan_reset_part2(struct work_struct *work); | ||
47 | static void ioat_dma_chan_watchdog(struct work_struct *work); | ||
48 | |||
49 | /* internal functions */ | 46 | /* internal functions */ |
50 | static void ioat_dma_start_null_desc(struct ioat_dma_chan *ioat); | 47 | static void ioat1_cleanup(struct ioat_dma_chan *ioat); |
51 | static void ioat_dma_memcpy_cleanup(struct ioat_dma_chan *ioat); | 48 | static void ioat1_dma_start_null_desc(struct ioat_dma_chan *ioat); |
52 | |||
53 | static struct ioat_desc_sw * | ||
54 | ioat1_dma_get_next_descriptor(struct ioat_dma_chan *ioat); | ||
55 | static struct ioat_desc_sw * | ||
56 | ioat2_dma_get_next_descriptor(struct ioat_dma_chan *ioat); | ||
57 | |||
58 | static inline struct ioat_chan_common * | ||
59 | ioat_chan_by_index(struct ioatdma_device *device, int index) | ||
60 | { | ||
61 | return device->idx[index]; | ||
62 | } | ||
63 | 49 | ||
64 | /** | 50 | /** |
65 | * ioat_dma_do_interrupt - handler used for single vector interrupt mode | 51 | * ioat_dma_do_interrupt - handler used for single vector interrupt mode |
@@ -108,18 +94,38 @@ static irqreturn_t ioat_dma_do_interrupt_msix(int irq, void *data) | |||
108 | return IRQ_HANDLED; | 94 | return IRQ_HANDLED; |
109 | } | 95 | } |
110 | 96 | ||
111 | static void ioat_dma_cleanup_tasklet(unsigned long data); | 97 | static void ioat1_cleanup_tasklet(unsigned long data); |
98 | |||
99 | /* common channel initialization */ | ||
100 | void ioat_init_channel(struct ioatdma_device *device, | ||
101 | struct ioat_chan_common *chan, int idx, | ||
102 | work_func_t work_fn, void (*tasklet)(unsigned long), | ||
103 | unsigned long tasklet_data) | ||
104 | { | ||
105 | struct dma_device *dma = &device->common; | ||
106 | |||
107 | chan->device = device; | ||
108 | chan->reg_base = device->reg_base + (0x80 * (idx + 1)); | ||
109 | INIT_DELAYED_WORK(&chan->work, work_fn); | ||
110 | spin_lock_init(&chan->cleanup_lock); | ||
111 | chan->common.device = dma; | ||
112 | list_add_tail(&chan->common.device_node, &dma->channels); | ||
113 | device->idx[idx] = chan; | ||
114 | tasklet_init(&chan->cleanup_task, tasklet, tasklet_data); | ||
115 | tasklet_disable(&chan->cleanup_task); | ||
116 | } | ||
117 | |||
118 | static void ioat1_reset_part2(struct work_struct *work); | ||
112 | 119 | ||
113 | /** | 120 | /** |
114 | * ioat_dma_enumerate_channels - find and initialize the device's channels | 121 | * ioat1_dma_enumerate_channels - find and initialize the device's channels |
115 | * @device: the device to be enumerated | 122 | * @device: the device to be enumerated |
116 | */ | 123 | */ |
117 | static int ioat_dma_enumerate_channels(struct ioatdma_device *device) | 124 | static int ioat1_enumerate_channels(struct ioatdma_device *device) |
118 | { | 125 | { |
119 | u8 xfercap_scale; | 126 | u8 xfercap_scale; |
120 | u32 xfercap; | 127 | u32 xfercap; |
121 | int i; | 128 | int i; |
122 | struct ioat_chan_common *chan; | ||
123 | struct ioat_dma_chan *ioat; | 129 | struct ioat_dma_chan *ioat; |
124 | struct device *dev = &device->pdev->dev; | 130 | struct device *dev = &device->pdev->dev; |
125 | struct dma_device *dma = &device->common; | 131 | struct dma_device *dma = &device->common; |
@@ -135,31 +141,20 @@ static int ioat_dma_enumerate_channels(struct ioatdma_device *device) | |||
135 | #endif | 141 | #endif |
136 | for (i = 0; i < dma->chancnt; i++) { | 142 | for (i = 0; i < dma->chancnt; i++) { |
137 | ioat = devm_kzalloc(dev, sizeof(*ioat), GFP_KERNEL); | 143 | ioat = devm_kzalloc(dev, sizeof(*ioat), GFP_KERNEL); |
138 | if (!ioat) { | 144 | if (!ioat) |
139 | dma->chancnt = i; | ||
140 | break; | 145 | break; |
141 | } | ||
142 | 146 | ||
143 | chan = &ioat->base; | 147 | ioat_init_channel(device, &ioat->base, i, |
144 | chan->device = device; | 148 | ioat1_reset_part2, |
145 | chan->reg_base = device->reg_base + (0x80 * (i + 1)); | 149 | ioat1_cleanup_tasklet, |
150 | (unsigned long) ioat); | ||
146 | ioat->xfercap = xfercap; | 151 | ioat->xfercap = xfercap; |
147 | ioat->desccount = 0; | ||
148 | INIT_DELAYED_WORK(&chan->work, ioat_dma_chan_reset_part2); | ||
149 | spin_lock_init(&chan->cleanup_lock); | ||
150 | spin_lock_init(&ioat->desc_lock); | 152 | spin_lock_init(&ioat->desc_lock); |
151 | INIT_LIST_HEAD(&ioat->free_desc); | 153 | INIT_LIST_HEAD(&ioat->free_desc); |
152 | INIT_LIST_HEAD(&ioat->used_desc); | 154 | INIT_LIST_HEAD(&ioat->used_desc); |
153 | /* This should be made common somewhere in dmaengine.c */ | ||
154 | chan->common.device = &device->common; | ||
155 | list_add_tail(&chan->common.device_node, &dma->channels); | ||
156 | device->idx[i] = chan; | ||
157 | tasklet_init(&chan->cleanup_task, | ||
158 | ioat_dma_cleanup_tasklet, | ||
159 | (unsigned long) ioat); | ||
160 | tasklet_disable(&chan->cleanup_task); | ||
161 | } | 155 | } |
162 | return dma->chancnt; | 156 | dma->chancnt = i; |
157 | return i; | ||
163 | } | 158 | } |
164 | 159 | ||
165 | /** | 160 | /** |
@@ -187,35 +182,16 @@ static void ioat1_dma_memcpy_issue_pending(struct dma_chan *chan) | |||
187 | } | 182 | } |
188 | } | 183 | } |
189 | 184 | ||
190 | static inline void | ||
191 | __ioat2_dma_memcpy_issue_pending(struct ioat_dma_chan *ioat) | ||
192 | { | ||
193 | void __iomem *reg_base = ioat->base.reg_base; | ||
194 | |||
195 | ioat->pending = 0; | ||
196 | writew(ioat->dmacount, reg_base + IOAT_CHAN_DMACOUNT_OFFSET); | ||
197 | } | ||
198 | |||
199 | static void ioat2_dma_memcpy_issue_pending(struct dma_chan *chan) | ||
200 | { | ||
201 | struct ioat_dma_chan *ioat = to_ioat_chan(chan); | ||
202 | |||
203 | if (ioat->pending > 0) { | ||
204 | spin_lock_bh(&ioat->desc_lock); | ||
205 | __ioat2_dma_memcpy_issue_pending(ioat); | ||
206 | spin_unlock_bh(&ioat->desc_lock); | ||
207 | } | ||
208 | } | ||
209 | |||
210 | |||
211 | /** | 185 | /** |
212 | * ioat_dma_chan_reset_part2 - reinit the channel after a reset | 186 | * ioat1_reset_part2 - reinit the channel after a reset |
213 | */ | 187 | */ |
214 | static void ioat_dma_chan_reset_part2(struct work_struct *work) | 188 | static void ioat1_reset_part2(struct work_struct *work) |
215 | { | 189 | { |
216 | struct ioat_chan_common *chan; | 190 | struct ioat_chan_common *chan; |
217 | struct ioat_dma_chan *ioat; | 191 | struct ioat_dma_chan *ioat; |
218 | struct ioat_desc_sw *desc; | 192 | struct ioat_desc_sw *desc; |
193 | int dmacount; | ||
194 | bool start_null = false; | ||
219 | 195 | ||
220 | chan = container_of(work, struct ioat_chan_common, work.work); | 196 | chan = container_of(work, struct ioat_chan_common, work.work); |
221 | ioat = container_of(chan, struct ioat_dma_chan, base); | 197 | ioat = container_of(chan, struct ioat_dma_chan, base); |
@@ -226,26 +202,22 @@ static void ioat_dma_chan_reset_part2(struct work_struct *work) | |||
226 | chan->completion_virt->high = 0; | 202 | chan->completion_virt->high = 0; |
227 | ioat->pending = 0; | 203 | ioat->pending = 0; |
228 | 204 | ||
229 | /* | 205 | /* count the descriptors waiting */ |
230 | * count the descriptors waiting, and be sure to do it | 206 | dmacount = 0; |
231 | * right for both the CB1 line and the CB2 ring | ||
232 | */ | ||
233 | ioat->dmacount = 0; | ||
234 | if (ioat->used_desc.prev) { | 207 | if (ioat->used_desc.prev) { |
235 | desc = to_ioat_desc(ioat->used_desc.prev); | 208 | desc = to_ioat_desc(ioat->used_desc.prev); |
236 | do { | 209 | do { |
237 | ioat->dmacount++; | 210 | dmacount++; |
238 | desc = to_ioat_desc(desc->node.next); | 211 | desc = to_ioat_desc(desc->node.next); |
239 | } while (&desc->node != ioat->used_desc.next); | 212 | } while (&desc->node != ioat->used_desc.next); |
240 | } | 213 | } |
241 | 214 | ||
242 | /* | 215 | if (dmacount) { |
243 | * write the new starting descriptor address | 216 | /* |
244 | * this puts channel engine into ARMED state | 217 | * write the new starting descriptor address |
245 | */ | 218 | * this puts channel engine into ARMED state |
246 | desc = to_ioat_desc(ioat->used_desc.prev); | 219 | */ |
247 | switch (chan->device->version) { | 220 | desc = to_ioat_desc(ioat->used_desc.prev); |
248 | case IOAT_VER_1_2: | ||
249 | writel(((u64) desc->txd.phys) & 0x00000000FFFFFFFF, | 221 | writel(((u64) desc->txd.phys) & 0x00000000FFFFFFFF, |
250 | chan->reg_base + IOAT1_CHAINADDR_OFFSET_LOW); | 222 | chan->reg_base + IOAT1_CHAINADDR_OFFSET_LOW); |
251 | writel(((u64) desc->txd.phys) >> 32, | 223 | writel(((u64) desc->txd.phys) >> 32, |
@@ -253,32 +225,24 @@ static void ioat_dma_chan_reset_part2(struct work_struct *work) | |||
253 | 225 | ||
254 | writeb(IOAT_CHANCMD_START, chan->reg_base | 226 | writeb(IOAT_CHANCMD_START, chan->reg_base |
255 | + IOAT_CHANCMD_OFFSET(chan->device->version)); | 227 | + IOAT_CHANCMD_OFFSET(chan->device->version)); |
256 | break; | 228 | } else |
257 | case IOAT_VER_2_0: | 229 | start_null = true; |
258 | writel(((u64) desc->txd.phys) & 0x00000000FFFFFFFF, | 230 | spin_unlock_bh(&ioat->desc_lock); |
259 | chan->reg_base + IOAT2_CHAINADDR_OFFSET_LOW); | 231 | spin_unlock_bh(&chan->cleanup_lock); |
260 | writel(((u64) desc->txd.phys) >> 32, | ||
261 | chan->reg_base + IOAT2_CHAINADDR_OFFSET_HIGH); | ||
262 | |||
263 | /* tell the engine to go with what's left to be done */ | ||
264 | writew(ioat->dmacount, | ||
265 | chan->reg_base + IOAT_CHAN_DMACOUNT_OFFSET); | ||
266 | 232 | ||
267 | break; | ||
268 | } | ||
269 | dev_err(to_dev(chan), | 233 | dev_err(to_dev(chan), |
270 | "chan%d reset - %d descs waiting, %d total desc\n", | 234 | "chan%d reset - %d descs waiting, %d total desc\n", |
271 | chan_num(chan), ioat->dmacount, ioat->desccount); | 235 | chan_num(chan), dmacount, ioat->desccount); |
272 | 236 | ||
273 | spin_unlock_bh(&ioat->desc_lock); | 237 | if (start_null) |
274 | spin_unlock_bh(&chan->cleanup_lock); | 238 | ioat1_dma_start_null_desc(ioat); |
275 | } | 239 | } |
276 | 240 | ||
277 | /** | 241 | /** |
278 | * ioat_dma_reset_channel - restart a channel | 242 | * ioat1_reset_channel - restart a channel |
279 | * @ioat: IOAT DMA channel handle | 243 | * @ioat: IOAT DMA channel handle |
280 | */ | 244 | */ |
281 | static void ioat_dma_reset_channel(struct ioat_dma_chan *ioat) | 245 | static void ioat1_reset_channel(struct ioat_dma_chan *ioat) |
282 | { | 246 | { |
283 | struct ioat_chan_common *chan = &ioat->base; | 247 | struct ioat_chan_common *chan = &ioat->base; |
284 | void __iomem *reg_base = chan->reg_base; | 248 | void __iomem *reg_base = chan->reg_base; |
@@ -316,9 +280,9 @@ static void ioat_dma_reset_channel(struct ioat_dma_chan *ioat) | |||
316 | } | 280 | } |
317 | 281 | ||
318 | /** | 282 | /** |
319 | * ioat_dma_chan_watchdog - watch for stuck channels | 283 | * ioat1_chan_watchdog - watch for stuck channels |
320 | */ | 284 | */ |
321 | static void ioat_dma_chan_watchdog(struct work_struct *work) | 285 | static void ioat1_chan_watchdog(struct work_struct *work) |
322 | { | 286 | { |
323 | struct ioatdma_device *device = | 287 | struct ioatdma_device *device = |
324 | container_of(work, struct ioatdma_device, work.work); | 288 | container_of(work, struct ioatdma_device, work.work); |
@@ -339,16 +303,15 @@ static void ioat_dma_chan_watchdog(struct work_struct *work) | |||
339 | chan = ioat_chan_by_index(device, i); | 303 | chan = ioat_chan_by_index(device, i); |
340 | ioat = container_of(chan, struct ioat_dma_chan, base); | 304 | ioat = container_of(chan, struct ioat_dma_chan, base); |
341 | 305 | ||
342 | if (chan->device->version == IOAT_VER_1_2 | 306 | if (/* have we started processing anything yet */ |
343 | /* have we started processing anything yet */ | 307 | chan->last_completion |
344 | && chan->last_completion | 308 | /* have we completed any since last watchdog cycle? */ |
345 | /* have we completed any since last watchdog cycle? */ | ||
346 | && (chan->last_completion == chan->watchdog_completion) | 309 | && (chan->last_completion == chan->watchdog_completion) |
347 | /* has TCP stuck on one cookie since last watchdog? */ | 310 | /* has TCP stuck on one cookie since last watchdog? */ |
348 | && (chan->watchdog_tcp_cookie == chan->watchdog_last_tcp_cookie) | 311 | && (chan->watchdog_tcp_cookie == chan->watchdog_last_tcp_cookie) |
349 | && (chan->watchdog_tcp_cookie != chan->completed_cookie) | 312 | && (chan->watchdog_tcp_cookie != chan->completed_cookie) |
350 | /* is there something in the chain to be processed? */ | 313 | /* is there something in the chain to be processed? */ |
351 | /* CB1 chain always has at least the last one processed */ | 314 | /* CB1 chain always has at least the last one processed */ |
352 | && (ioat->used_desc.prev != ioat->used_desc.next) | 315 | && (ioat->used_desc.prev != ioat->used_desc.next) |
353 | && ioat->pending == 0) { | 316 | && ioat->pending == 0) { |
354 | 317 | ||
@@ -387,34 +350,15 @@ static void ioat_dma_chan_watchdog(struct work_struct *work) | |||
387 | chan->completion_virt->low = completion_hw.low; | 350 | chan->completion_virt->low = completion_hw.low; |
388 | chan->completion_virt->high = completion_hw.high; | 351 | chan->completion_virt->high = completion_hw.high; |
389 | } else { | 352 | } else { |
390 | ioat_dma_reset_channel(ioat); | 353 | ioat1_reset_channel(ioat); |
391 | chan->watchdog_completion = 0; | 354 | chan->watchdog_completion = 0; |
392 | chan->last_compl_desc_addr_hw = 0; | 355 | chan->last_compl_desc_addr_hw = 0; |
393 | } | 356 | } |
394 | |||
395 | /* | ||
396 | * for version 2.0 if there are descriptors yet to be processed | ||
397 | * and the last completed hasn't changed since the last watchdog | ||
398 | * if they haven't hit the pending level | ||
399 | * issue the pending to push them through | ||
400 | * else | ||
401 | * try resetting the channel | ||
402 | */ | ||
403 | } else if (chan->device->version == IOAT_VER_2_0 | ||
404 | && ioat->used_desc.prev | ||
405 | && chan->last_completion | ||
406 | && chan->last_completion == chan->watchdog_completion) { | ||
407 | |||
408 | if (ioat->pending < ioat_pending_level) | ||
409 | ioat2_dma_memcpy_issue_pending(&chan->common); | ||
410 | else { | ||
411 | ioat_dma_reset_channel(ioat); | ||
412 | chan->watchdog_completion = 0; | ||
413 | } | ||
414 | } else { | 357 | } else { |
415 | chan->last_compl_desc_addr_hw = 0; | 358 | chan->last_compl_desc_addr_hw = 0; |
416 | chan->watchdog_completion = chan->last_completion; | 359 | chan->watchdog_completion = chan->last_completion; |
417 | } | 360 | } |
361 | |||
418 | chan->watchdog_last_tcp_cookie = chan->watchdog_tcp_cookie; | 362 | chan->watchdog_last_tcp_cookie = chan->watchdog_tcp_cookie; |
419 | } | 363 | } |
420 | 364 | ||
@@ -447,7 +391,6 @@ static dma_cookie_t ioat1_tx_submit(struct dma_async_tx_descriptor *tx) | |||
447 | chain_tail->hw->next = first->txd.phys; | 391 | chain_tail->hw->next = first->txd.phys; |
448 | list_splice_tail_init(&tx->tx_list, &ioat->used_desc); | 392 | list_splice_tail_init(&tx->tx_list, &ioat->used_desc); |
449 | 393 | ||
450 | ioat->dmacount += desc->tx_cnt; | ||
451 | ioat->pending += desc->tx_cnt; | 394 | ioat->pending += desc->tx_cnt; |
452 | if (ioat->pending >= ioat_pending_level) | 395 | if (ioat->pending >= ioat_pending_level) |
453 | __ioat1_dma_memcpy_issue_pending(ioat); | 396 | __ioat1_dma_memcpy_issue_pending(ioat); |
@@ -456,92 +399,6 @@ static dma_cookie_t ioat1_tx_submit(struct dma_async_tx_descriptor *tx) | |||
456 | return cookie; | 399 | return cookie; |
457 | } | 400 | } |
458 | 401 | ||
459 | static dma_cookie_t ioat2_tx_submit(struct dma_async_tx_descriptor *tx) | ||
460 | { | ||
461 | struct ioat_dma_chan *ioat = to_ioat_chan(tx->chan); | ||
462 | struct ioat_desc_sw *first = tx_to_ioat_desc(tx); | ||
463 | struct ioat_desc_sw *new; | ||
464 | struct ioat_dma_descriptor *hw; | ||
465 | dma_cookie_t cookie; | ||
466 | u32 copy; | ||
467 | size_t len; | ||
468 | dma_addr_t src, dst; | ||
469 | unsigned long orig_flags; | ||
470 | unsigned int desc_count = 0; | ||
471 | |||
472 | /* src and dest and len are stored in the initial descriptor */ | ||
473 | len = first->len; | ||
474 | src = first->src; | ||
475 | dst = first->dst; | ||
476 | orig_flags = first->txd.flags; | ||
477 | new = first; | ||
478 | |||
479 | /* | ||
480 | * ioat->desc_lock is still in force in version 2 path | ||
481 | * it gets unlocked at end of this function | ||
482 | */ | ||
483 | do { | ||
484 | copy = min_t(size_t, len, ioat->xfercap); | ||
485 | |||
486 | async_tx_ack(&new->txd); | ||
487 | |||
488 | hw = new->hw; | ||
489 | hw->size = copy; | ||
490 | hw->ctl = 0; | ||
491 | hw->src_addr = src; | ||
492 | hw->dst_addr = dst; | ||
493 | |||
494 | len -= copy; | ||
495 | dst += copy; | ||
496 | src += copy; | ||
497 | desc_count++; | ||
498 | } while (len && (new = ioat2_dma_get_next_descriptor(ioat))); | ||
499 | |||
500 | if (!new) { | ||
501 | dev_err(to_dev(&ioat->base), "tx submit failed\n"); | ||
502 | spin_unlock_bh(&ioat->desc_lock); | ||
503 | return -ENOMEM; | ||
504 | } | ||
505 | |||
506 | hw->ctl_f.compl_write = 1; | ||
507 | if (first->txd.callback) { | ||
508 | hw->ctl_f.int_en = 1; | ||
509 | if (first != new) { | ||
510 | /* move callback into to last desc */ | ||
511 | new->txd.callback = first->txd.callback; | ||
512 | new->txd.callback_param | ||
513 | = first->txd.callback_param; | ||
514 | first->txd.callback = NULL; | ||
515 | first->txd.callback_param = NULL; | ||
516 | } | ||
517 | } | ||
518 | |||
519 | new->tx_cnt = desc_count; | ||
520 | new->txd.flags = orig_flags; /* client is in control of this ack */ | ||
521 | |||
522 | /* store the original values for use in later cleanup */ | ||
523 | if (new != first) { | ||
524 | new->src = first->src; | ||
525 | new->dst = first->dst; | ||
526 | new->len = first->len; | ||
527 | } | ||
528 | |||
529 | /* cookie incr and addition to used_list must be atomic */ | ||
530 | cookie = ioat->base.common.cookie; | ||
531 | cookie++; | ||
532 | if (cookie < 0) | ||
533 | cookie = 1; | ||
534 | ioat->base.common.cookie = new->txd.cookie = cookie; | ||
535 | |||
536 | ioat->dmacount += desc_count; | ||
537 | ioat->pending += desc_count; | ||
538 | if (ioat->pending >= ioat_pending_level) | ||
539 | __ioat2_dma_memcpy_issue_pending(ioat); | ||
540 | spin_unlock_bh(&ioat->desc_lock); | ||
541 | |||
542 | return cookie; | ||
543 | } | ||
544 | |||
545 | /** | 402 | /** |
546 | * ioat_dma_alloc_descriptor - allocate and return a sw and hw descriptor pair | 403 | * ioat_dma_alloc_descriptor - allocate and return a sw and hw descriptor pair |
547 | * @ioat: the channel supplying the memory pool for the descriptors | 404 | * @ioat: the channel supplying the memory pool for the descriptors |
@@ -567,17 +424,9 @@ ioat_dma_alloc_descriptor(struct ioat_dma_chan *ioat, gfp_t flags) | |||
567 | } | 424 | } |
568 | 425 | ||
569 | memset(desc, 0, sizeof(*desc)); | 426 | memset(desc, 0, sizeof(*desc)); |
570 | dma_async_tx_descriptor_init(&desc_sw->txd, &ioat->base.common); | ||
571 | switch (ioatdma_device->version) { | ||
572 | case IOAT_VER_1_2: | ||
573 | desc_sw->txd.tx_submit = ioat1_tx_submit; | ||
574 | break; | ||
575 | case IOAT_VER_2_0: | ||
576 | case IOAT_VER_3_0: | ||
577 | desc_sw->txd.tx_submit = ioat2_tx_submit; | ||
578 | break; | ||
579 | } | ||
580 | 427 | ||
428 | dma_async_tx_descriptor_init(&desc_sw->txd, &ioat->base.common); | ||
429 | desc_sw->txd.tx_submit = ioat1_tx_submit; | ||
581 | desc_sw->hw = desc; | 430 | desc_sw->hw = desc; |
582 | desc_sw->txd.phys = phys; | 431 | desc_sw->txd.phys = phys; |
583 | 432 | ||
@@ -587,39 +436,12 @@ ioat_dma_alloc_descriptor(struct ioat_dma_chan *ioat, gfp_t flags) | |||
587 | static int ioat_initial_desc_count = 256; | 436 | static int ioat_initial_desc_count = 256; |
588 | module_param(ioat_initial_desc_count, int, 0644); | 437 | module_param(ioat_initial_desc_count, int, 0644); |
589 | MODULE_PARM_DESC(ioat_initial_desc_count, | 438 | MODULE_PARM_DESC(ioat_initial_desc_count, |
590 | "initial descriptors per channel (default: 256)"); | 439 | "ioat1: initial descriptors per channel (default: 256)"); |
591 | |||
592 | /** | ||
593 | * ioat2_dma_massage_chan_desc - link the descriptors into a circle | ||
594 | * @ioat: the channel to be massaged | ||
595 | */ | ||
596 | static void ioat2_dma_massage_chan_desc(struct ioat_dma_chan *ioat) | ||
597 | { | ||
598 | struct ioat_desc_sw *desc, *_desc; | ||
599 | |||
600 | /* setup used_desc */ | ||
601 | ioat->used_desc.next = ioat->free_desc.next; | ||
602 | ioat->used_desc.prev = NULL; | ||
603 | |||
604 | /* pull free_desc out of the circle so that every node is a hw | ||
605 | * descriptor, but leave it pointing to the list | ||
606 | */ | ||
607 | ioat->free_desc.prev->next = ioat->free_desc.next; | ||
608 | ioat->free_desc.next->prev = ioat->free_desc.prev; | ||
609 | |||
610 | /* circle link the hw descriptors */ | ||
611 | desc = to_ioat_desc(ioat->free_desc.next); | ||
612 | desc->hw->next = to_ioat_desc(desc->node.next)->txd.phys; | ||
613 | list_for_each_entry_safe(desc, _desc, ioat->free_desc.next, node) { | ||
614 | desc->hw->next = to_ioat_desc(desc->node.next)->txd.phys; | ||
615 | } | ||
616 | } | ||
617 | |||
618 | /** | 440 | /** |
619 | * ioat_dma_alloc_chan_resources - returns the number of allocated descriptors | 441 | * ioat1_dma_alloc_chan_resources - returns the number of allocated descriptors |
620 | * @chan: the channel to be filled out | 442 | * @chan: the channel to be filled out |
621 | */ | 443 | */ |
622 | static int ioat_dma_alloc_chan_resources(struct dma_chan *c) | 444 | static int ioat1_dma_alloc_chan_resources(struct dma_chan *c) |
623 | { | 445 | { |
624 | struct ioat_dma_chan *ioat = to_ioat_chan(c); | 446 | struct ioat_dma_chan *ioat = to_ioat_chan(c); |
625 | struct ioat_chan_common *chan = &ioat->base; | 447 | struct ioat_chan_common *chan = &ioat->base; |
@@ -657,8 +479,6 @@ static int ioat_dma_alloc_chan_resources(struct dma_chan *c) | |||
657 | spin_lock_bh(&ioat->desc_lock); | 479 | spin_lock_bh(&ioat->desc_lock); |
658 | ioat->desccount = i; | 480 | ioat->desccount = i; |
659 | list_splice(&tmp_list, &ioat->free_desc); | 481 | list_splice(&tmp_list, &ioat->free_desc); |
660 | if (chan->device->version != IOAT_VER_1_2) | ||
661 | ioat2_dma_massage_chan_desc(ioat); | ||
662 | spin_unlock_bh(&ioat->desc_lock); | 482 | spin_unlock_bh(&ioat->desc_lock); |
663 | 483 | ||
664 | /* allocate a completion writeback area */ | 484 | /* allocate a completion writeback area */ |
@@ -674,15 +494,15 @@ static int ioat_dma_alloc_chan_resources(struct dma_chan *c) | |||
674 | chan->reg_base + IOAT_CHANCMP_OFFSET_HIGH); | 494 | chan->reg_base + IOAT_CHANCMP_OFFSET_HIGH); |
675 | 495 | ||
676 | tasklet_enable(&chan->cleanup_task); | 496 | tasklet_enable(&chan->cleanup_task); |
677 | ioat_dma_start_null_desc(ioat); /* give chain to dma device */ | 497 | ioat1_dma_start_null_desc(ioat); /* give chain to dma device */ |
678 | return ioat->desccount; | 498 | return ioat->desccount; |
679 | } | 499 | } |
680 | 500 | ||
681 | /** | 501 | /** |
682 | * ioat_dma_free_chan_resources - release all the descriptors | 502 | * ioat1_dma_free_chan_resources - release all the descriptors |
683 | * @chan: the channel to be cleaned | 503 | * @chan: the channel to be cleaned |
684 | */ | 504 | */ |
685 | static void ioat_dma_free_chan_resources(struct dma_chan *c) | 505 | static void ioat1_dma_free_chan_resources(struct dma_chan *c) |
686 | { | 506 | { |
687 | struct ioat_dma_chan *ioat = to_ioat_chan(c); | 507 | struct ioat_dma_chan *ioat = to_ioat_chan(c); |
688 | struct ioat_chan_common *chan = &ioat->base; | 508 | struct ioat_chan_common *chan = &ioat->base; |
@@ -697,7 +517,7 @@ static void ioat_dma_free_chan_resources(struct dma_chan *c) | |||
697 | return; | 517 | return; |
698 | 518 | ||
699 | tasklet_disable(&chan->cleanup_task); | 519 | tasklet_disable(&chan->cleanup_task); |
700 | ioat_dma_memcpy_cleanup(ioat); | 520 | ioat1_cleanup(ioat); |
701 | 521 | ||
702 | /* Delay 100ms after reset to allow internal DMA logic to quiesce | 522 | /* Delay 100ms after reset to allow internal DMA logic to quiesce |
703 | * before removing DMA descriptor resources. | 523 | * before removing DMA descriptor resources. |
@@ -707,40 +527,20 @@ static void ioat_dma_free_chan_resources(struct dma_chan *c) | |||
707 | mdelay(100); | 527 | mdelay(100); |
708 | 528 | ||
709 | spin_lock_bh(&ioat->desc_lock); | 529 | spin_lock_bh(&ioat->desc_lock); |
710 | switch (chan->device->version) { | 530 | list_for_each_entry_safe(desc, _desc, |
711 | case IOAT_VER_1_2: | 531 | &ioat->used_desc, node) { |
712 | list_for_each_entry_safe(desc, _desc, | 532 | in_use_descs++; |
713 | &ioat->used_desc, node) { | 533 | list_del(&desc->node); |
714 | in_use_descs++; | 534 | pci_pool_free(ioatdma_device->dma_pool, desc->hw, |
715 | list_del(&desc->node); | 535 | desc->txd.phys); |
716 | pci_pool_free(ioatdma_device->dma_pool, desc->hw, | 536 | kfree(desc); |
717 | desc->txd.phys); | 537 | } |
718 | kfree(desc); | 538 | list_for_each_entry_safe(desc, _desc, |
719 | } | 539 | &ioat->free_desc, node) { |
720 | list_for_each_entry_safe(desc, _desc, | 540 | list_del(&desc->node); |
721 | &ioat->free_desc, node) { | ||
722 | list_del(&desc->node); | ||
723 | pci_pool_free(ioatdma_device->dma_pool, desc->hw, | ||
724 | desc->txd.phys); | ||
725 | kfree(desc); | ||
726 | } | ||
727 | break; | ||
728 | case IOAT_VER_2_0: | ||
729 | case IOAT_VER_3_0: | ||
730 | list_for_each_entry_safe(desc, _desc, | ||
731 | ioat->free_desc.next, node) { | ||
732 | list_del(&desc->node); | ||
733 | pci_pool_free(ioatdma_device->dma_pool, desc->hw, | ||
734 | desc->txd.phys); | ||
735 | kfree(desc); | ||
736 | } | ||
737 | desc = to_ioat_desc(ioat->free_desc.next); | ||
738 | pci_pool_free(ioatdma_device->dma_pool, desc->hw, | 541 | pci_pool_free(ioatdma_device->dma_pool, desc->hw, |
739 | desc->txd.phys); | 542 | desc->txd.phys); |
740 | kfree(desc); | 543 | kfree(desc); |
741 | INIT_LIST_HEAD(&ioat->free_desc); | ||
742 | INIT_LIST_HEAD(&ioat->used_desc); | ||
743 | break; | ||
744 | } | 544 | } |
745 | spin_unlock_bh(&ioat->desc_lock); | 545 | spin_unlock_bh(&ioat->desc_lock); |
746 | 546 | ||
@@ -758,7 +558,6 @@ static void ioat_dma_free_chan_resources(struct dma_chan *c) | |||
758 | chan->last_compl_desc_addr_hw = 0; | 558 | chan->last_compl_desc_addr_hw = 0; |
759 | chan->watchdog_tcp_cookie = chan->watchdog_last_tcp_cookie = 0; | 559 | chan->watchdog_tcp_cookie = chan->watchdog_last_tcp_cookie = 0; |
760 | ioat->pending = 0; | 560 | ioat->pending = 0; |
761 | ioat->dmacount = 0; | ||
762 | ioat->desccount = 0; | 561 | ioat->desccount = 0; |
763 | } | 562 | } |
764 | 563 | ||
@@ -791,86 +590,6 @@ ioat1_dma_get_next_descriptor(struct ioat_dma_chan *ioat) | |||
791 | return new; | 590 | return new; |
792 | } | 591 | } |
793 | 592 | ||
794 | static struct ioat_desc_sw * | ||
795 | ioat2_dma_get_next_descriptor(struct ioat_dma_chan *ioat) | ||
796 | { | ||
797 | struct ioat_desc_sw *new; | ||
798 | |||
799 | /* | ||
800 | * used.prev points to where to start processing | ||
801 | * used.next points to next free descriptor | ||
802 | * if used.prev == NULL, there are none waiting to be processed | ||
803 | * if used.next == used.prev.prev, there is only one free descriptor, | ||
804 | * and we need to use it to as a noop descriptor before | ||
805 | * linking in a new set of descriptors, since the device | ||
806 | * has probably already read the pointer to it | ||
807 | */ | ||
808 | if (ioat->used_desc.prev && | ||
809 | ioat->used_desc.next == ioat->used_desc.prev->prev) { | ||
810 | |||
811 | struct ioat_desc_sw *desc; | ||
812 | struct ioat_desc_sw *noop_desc; | ||
813 | int i; | ||
814 | |||
815 | /* set up the noop descriptor */ | ||
816 | noop_desc = to_ioat_desc(ioat->used_desc.next); | ||
817 | /* set size to non-zero value (channel returns error when size is 0) */ | ||
818 | noop_desc->hw->size = NULL_DESC_BUFFER_SIZE; | ||
819 | noop_desc->hw->ctl = 0; | ||
820 | noop_desc->hw->ctl_f.null = 1; | ||
821 | noop_desc->hw->src_addr = 0; | ||
822 | noop_desc->hw->dst_addr = 0; | ||
823 | |||
824 | ioat->used_desc.next = ioat->used_desc.next->next; | ||
825 | ioat->pending++; | ||
826 | ioat->dmacount++; | ||
827 | |||
828 | /* try to get a few more descriptors */ | ||
829 | for (i = 16; i; i--) { | ||
830 | desc = ioat_dma_alloc_descriptor(ioat, GFP_ATOMIC); | ||
831 | if (!desc) { | ||
832 | dev_err(to_dev(&ioat->base), | ||
833 | "alloc failed\n"); | ||
834 | break; | ||
835 | } | ||
836 | list_add_tail(&desc->node, ioat->used_desc.next); | ||
837 | |||
838 | desc->hw->next | ||
839 | = to_ioat_desc(desc->node.next)->txd.phys; | ||
840 | to_ioat_desc(desc->node.prev)->hw->next | ||
841 | = desc->txd.phys; | ||
842 | ioat->desccount++; | ||
843 | } | ||
844 | |||
845 | ioat->used_desc.next = noop_desc->node.next; | ||
846 | } | ||
847 | new = to_ioat_desc(ioat->used_desc.next); | ||
848 | prefetch(new); | ||
849 | ioat->used_desc.next = new->node.next; | ||
850 | |||
851 | if (ioat->used_desc.prev == NULL) | ||
852 | ioat->used_desc.prev = &new->node; | ||
853 | |||
854 | prefetch(new->hw); | ||
855 | return new; | ||
856 | } | ||
857 | |||
858 | static struct ioat_desc_sw * | ||
859 | ioat_dma_get_next_descriptor(struct ioat_dma_chan *ioat) | ||
860 | { | ||
861 | if (!ioat) | ||
862 | return NULL; | ||
863 | |||
864 | switch (ioat->base.device->version) { | ||
865 | case IOAT_VER_1_2: | ||
866 | return ioat1_dma_get_next_descriptor(ioat); | ||
867 | case IOAT_VER_2_0: | ||
868 | case IOAT_VER_3_0: | ||
869 | return ioat2_dma_get_next_descriptor(ioat); | ||
870 | } | ||
871 | return NULL; | ||
872 | } | ||
873 | |||
874 | static struct dma_async_tx_descriptor * | 593 | static struct dma_async_tx_descriptor * |
875 | ioat1_dma_prep_memcpy(struct dma_chan *c, dma_addr_t dma_dest, | 594 | ioat1_dma_prep_memcpy(struct dma_chan *c, dma_addr_t dma_dest, |
876 | dma_addr_t dma_src, size_t len, unsigned long flags) | 595 | dma_addr_t dma_src, size_t len, unsigned long flags) |
@@ -886,7 +605,7 @@ ioat1_dma_prep_memcpy(struct dma_chan *c, dma_addr_t dma_dest, | |||
886 | int tx_cnt = 0; | 605 | int tx_cnt = 0; |
887 | 606 | ||
888 | spin_lock_bh(&ioat->desc_lock); | 607 | spin_lock_bh(&ioat->desc_lock); |
889 | desc = ioat_dma_get_next_descriptor(ioat); | 608 | desc = ioat1_dma_get_next_descriptor(ioat); |
890 | do { | 609 | do { |
891 | if (!desc) | 610 | if (!desc) |
892 | break; | 611 | break; |
@@ -909,7 +628,7 @@ ioat1_dma_prep_memcpy(struct dma_chan *c, dma_addr_t dma_dest, | |||
909 | struct ioat_desc_sw *next; | 628 | struct ioat_desc_sw *next; |
910 | 629 | ||
911 | async_tx_ack(&desc->txd); | 630 | async_tx_ack(&desc->txd); |
912 | next = ioat_dma_get_next_descriptor(ioat); | 631 | next = ioat1_dma_get_next_descriptor(ioat); |
913 | hw->next = next ? next->txd.phys : 0; | 632 | hw->next = next ? next->txd.phys : 0; |
914 | desc = next; | 633 | desc = next; |
915 | } else | 634 | } else |
@@ -920,8 +639,7 @@ ioat1_dma_prep_memcpy(struct dma_chan *c, dma_addr_t dma_dest, | |||
920 | struct ioat_chan_common *chan = &ioat->base; | 639 | struct ioat_chan_common *chan = &ioat->base; |
921 | 640 | ||
922 | dev_err(to_dev(chan), | 641 | dev_err(to_dev(chan), |
923 | "chan%d - get_next_desc failed: %d descs waiting, %d total desc\n", | 642 | "chan%d - get_next_desc failed\n", chan_num(chan)); |
924 | chan_num(chan), ioat->dmacount, ioat->desccount); | ||
925 | list_splice(&chain, &ioat->free_desc); | 643 | list_splice(&chain, &ioat->free_desc); |
926 | spin_unlock_bh(&ioat->desc_lock); | 644 | spin_unlock_bh(&ioat->desc_lock); |
927 | return NULL; | 645 | return NULL; |
@@ -940,94 +658,43 @@ ioat1_dma_prep_memcpy(struct dma_chan *c, dma_addr_t dma_dest, | |||
940 | return &desc->txd; | 658 | return &desc->txd; |
941 | } | 659 | } |
942 | 660 | ||
943 | static struct dma_async_tx_descriptor * | 661 | static void ioat1_cleanup_tasklet(unsigned long data) |
944 | ioat2_dma_prep_memcpy(struct dma_chan *c, dma_addr_t dma_dest, | ||
945 | dma_addr_t dma_src, size_t len, unsigned long flags) | ||
946 | { | ||
947 | struct ioat_dma_chan *ioat = to_ioat_chan(c); | ||
948 | struct ioat_desc_sw *new; | ||
949 | |||
950 | spin_lock_bh(&ioat->desc_lock); | ||
951 | new = ioat2_dma_get_next_descriptor(ioat); | ||
952 | |||
953 | /* | ||
954 | * leave ioat->desc_lock set in ioat 2 path | ||
955 | * it will get unlocked at end of tx_submit | ||
956 | */ | ||
957 | |||
958 | if (new) { | ||
959 | new->len = len; | ||
960 | new->dst = dma_dest; | ||
961 | new->src = dma_src; | ||
962 | new->txd.flags = flags; | ||
963 | return &new->txd; | ||
964 | } else { | ||
965 | struct ioat_chan_common *chan = &ioat->base; | ||
966 | |||
967 | spin_unlock_bh(&ioat->desc_lock); | ||
968 | dev_err(to_dev(chan), | ||
969 | "chan%d - get_next_desc failed: %d descs waiting, %d total desc\n", | ||
970 | chan_num(chan), ioat->dmacount, ioat->desccount); | ||
971 | return NULL; | ||
972 | } | ||
973 | } | ||
974 | |||
975 | static void ioat_dma_cleanup_tasklet(unsigned long data) | ||
976 | { | 662 | { |
977 | struct ioat_dma_chan *chan = (void *)data; | 663 | struct ioat_dma_chan *chan = (void *)data; |
978 | ioat_dma_memcpy_cleanup(chan); | 664 | ioat1_cleanup(chan); |
979 | writew(IOAT_CHANCTRL_INT_DISABLE, | 665 | writew(IOAT_CHANCTRL_INT_DISABLE, |
980 | chan->base.reg_base + IOAT_CHANCTRL_OFFSET); | 666 | chan->base.reg_base + IOAT_CHANCTRL_OFFSET); |
981 | } | 667 | } |
982 | 668 | ||
983 | static void | 669 | static void ioat_unmap(struct pci_dev *pdev, dma_addr_t addr, size_t len, |
984 | ioat_dma_unmap(struct ioat_chan_common *chan, struct ioat_desc_sw *desc) | 670 | int direction, enum dma_ctrl_flags flags, bool dst) |
985 | { | 671 | { |
986 | if (!(desc->txd.flags & DMA_COMPL_SKIP_DEST_UNMAP)) { | 672 | if ((dst && (flags & DMA_COMPL_DEST_UNMAP_SINGLE)) || |
987 | if (desc->txd.flags & DMA_COMPL_DEST_UNMAP_SINGLE) | 673 | (!dst && (flags & DMA_COMPL_SRC_UNMAP_SINGLE))) |
988 | pci_unmap_single(chan->device->pdev, | 674 | pci_unmap_single(pdev, addr, len, direction); |
989 | pci_unmap_addr(desc, dst), | 675 | else |
990 | pci_unmap_len(desc, len), | 676 | pci_unmap_page(pdev, addr, len, direction); |
991 | PCI_DMA_FROMDEVICE); | ||
992 | else | ||
993 | pci_unmap_page(chan->device->pdev, | ||
994 | pci_unmap_addr(desc, dst), | ||
995 | pci_unmap_len(desc, len), | ||
996 | PCI_DMA_FROMDEVICE); | ||
997 | } | ||
998 | |||
999 | if (!(desc->txd.flags & DMA_COMPL_SKIP_SRC_UNMAP)) { | ||
1000 | if (desc->txd.flags & DMA_COMPL_SRC_UNMAP_SINGLE) | ||
1001 | pci_unmap_single(chan->device->pdev, | ||
1002 | pci_unmap_addr(desc, src), | ||
1003 | pci_unmap_len(desc, len), | ||
1004 | PCI_DMA_TODEVICE); | ||
1005 | else | ||
1006 | pci_unmap_page(chan->device->pdev, | ||
1007 | pci_unmap_addr(desc, src), | ||
1008 | pci_unmap_len(desc, len), | ||
1009 | PCI_DMA_TODEVICE); | ||
1010 | } | ||
1011 | } | 677 | } |
1012 | 678 | ||
1013 | /** | 679 | |
1014 | * ioat_dma_memcpy_cleanup - cleanup up finished descriptors | 680 | void ioat_dma_unmap(struct ioat_chan_common *chan, enum dma_ctrl_flags flags, |
1015 | * @chan: ioat channel to be cleaned up | 681 | size_t len, struct ioat_dma_descriptor *hw) |
1016 | */ | ||
1017 | static void ioat_dma_memcpy_cleanup(struct ioat_dma_chan *ioat) | ||
1018 | { | 682 | { |
1019 | struct ioat_chan_common *chan = &ioat->base; | 683 | struct pci_dev *pdev = chan->device->pdev; |
1020 | unsigned long phys_complete; | 684 | size_t offset = len - hw->size; |
1021 | struct ioat_desc_sw *desc, *_desc; | ||
1022 | dma_cookie_t cookie = 0; | ||
1023 | unsigned long desc_phys; | ||
1024 | struct ioat_desc_sw *latest_desc; | ||
1025 | struct dma_async_tx_descriptor *tx; | ||
1026 | 685 | ||
1027 | prefetch(chan->completion_virt); | 686 | if (!(flags & DMA_COMPL_SKIP_DEST_UNMAP)) |
687 | ioat_unmap(pdev, hw->dst_addr - offset, len, | ||
688 | PCI_DMA_FROMDEVICE, flags, 1); | ||
1028 | 689 | ||
1029 | if (!spin_trylock_bh(&chan->cleanup_lock)) | 690 | if (!(flags & DMA_COMPL_SKIP_SRC_UNMAP)) |
1030 | return; | 691 | ioat_unmap(pdev, hw->src_addr - offset, len, |
692 | PCI_DMA_TODEVICE, flags, 0); | ||
693 | } | ||
694 | |||
695 | unsigned long ioat_get_current_completion(struct ioat_chan_common *chan) | ||
696 | { | ||
697 | unsigned long phys_complete; | ||
1031 | 698 | ||
1032 | /* The completion writeback can happen at any time, | 699 | /* The completion writeback can happen at any time, |
1033 | so reads by the driver need to be atomic operations | 700 | so reads by the driver need to be atomic operations |
@@ -1051,18 +718,37 @@ static void ioat_dma_memcpy_cleanup(struct ioat_dma_chan *ioat) | |||
1051 | /* TODO do something to salvage the situation */ | 718 | /* TODO do something to salvage the situation */ |
1052 | } | 719 | } |
1053 | 720 | ||
721 | return phys_complete; | ||
722 | } | ||
723 | |||
724 | /** | ||
725 | * ioat1_cleanup - cleanup up finished descriptors | ||
726 | * @chan: ioat channel to be cleaned up | ||
727 | */ | ||
728 | static void ioat1_cleanup(struct ioat_dma_chan *ioat) | ||
729 | { | ||
730 | struct ioat_chan_common *chan = &ioat->base; | ||
731 | unsigned long phys_complete; | ||
732 | struct ioat_desc_sw *desc, *_desc; | ||
733 | dma_cookie_t cookie = 0; | ||
734 | struct dma_async_tx_descriptor *tx; | ||
735 | |||
736 | prefetch(chan->completion_virt); | ||
737 | |||
738 | if (!spin_trylock_bh(&chan->cleanup_lock)) | ||
739 | return; | ||
740 | |||
741 | phys_complete = ioat_get_current_completion(chan); | ||
1054 | if (phys_complete == chan->last_completion) { | 742 | if (phys_complete == chan->last_completion) { |
1055 | spin_unlock_bh(&chan->cleanup_lock); | 743 | spin_unlock_bh(&chan->cleanup_lock); |
1056 | /* | 744 | /* |
1057 | * perhaps we're stuck so hard that the watchdog can't go off? | 745 | * perhaps we're stuck so hard that the watchdog can't go off? |
1058 | * try to catch it after 2 seconds | 746 | * try to catch it after 2 seconds |
1059 | */ | 747 | */ |
1060 | if (chan->device->version != IOAT_VER_3_0) { | 748 | if (time_after(jiffies, |
1061 | if (time_after(jiffies, | 749 | chan->last_completion_time + HZ*WATCHDOG_DELAY)) { |
1062 | chan->last_completion_time + HZ*WATCHDOG_DELAY)) { | 750 | ioat1_chan_watchdog(&(chan->device->work.work)); |
1063 | ioat_dma_chan_watchdog(&(chan->device->work.work)); | 751 | chan->last_completion_time = jiffies; |
1064 | chan->last_completion_time = jiffies; | ||
1065 | } | ||
1066 | } | 752 | } |
1067 | return; | 753 | return; |
1068 | } | 754 | } |
@@ -1074,91 +760,42 @@ static void ioat_dma_memcpy_cleanup(struct ioat_dma_chan *ioat) | |||
1074 | return; | 760 | return; |
1075 | } | 761 | } |
1076 | 762 | ||
1077 | switch (chan->device->version) { | 763 | list_for_each_entry_safe(desc, _desc, &ioat->used_desc, node) { |
1078 | case IOAT_VER_1_2: | 764 | tx = &desc->txd; |
1079 | list_for_each_entry_safe(desc, _desc, &ioat->used_desc, node) { | 765 | /* |
1080 | tx = &desc->txd; | 766 | * Incoming DMA requests may use multiple descriptors, |
1081 | /* | 767 | * due to exceeding xfercap, perhaps. If so, only the |
1082 | * Incoming DMA requests may use multiple descriptors, | 768 | * last one will have a cookie, and require unmapping. |
1083 | * due to exceeding xfercap, perhaps. If so, only the | 769 | */ |
1084 | * last one will have a cookie, and require unmapping. | 770 | if (tx->cookie) { |
1085 | */ | 771 | cookie = tx->cookie; |
1086 | if (tx->cookie) { | 772 | ioat_dma_unmap(chan, tx->flags, desc->len, desc->hw); |
1087 | cookie = tx->cookie; | 773 | if (tx->callback) { |
1088 | ioat_dma_unmap(chan, desc); | 774 | tx->callback(tx->callback_param); |
1089 | if (tx->callback) { | 775 | tx->callback = NULL; |
1090 | tx->callback(tx->callback_param); | ||
1091 | tx->callback = NULL; | ||
1092 | } | ||
1093 | } | 776 | } |
777 | } | ||
1094 | 778 | ||
1095 | if (tx->phys != phys_complete) { | 779 | if (tx->phys != phys_complete) { |
1096 | /* | 780 | /* |
1097 | * a completed entry, but not the last, so clean | 781 | * a completed entry, but not the last, so clean |
1098 | * up if the client is done with the descriptor | 782 | * up if the client is done with the descriptor |
1099 | */ | 783 | */ |
1100 | if (async_tx_test_ack(tx)) { | 784 | if (async_tx_test_ack(tx)) |
1101 | list_move_tail(&desc->node, | 785 | list_move_tail(&desc->node, &ioat->free_desc); |
1102 | &ioat->free_desc); | 786 | else |
1103 | } else | ||
1104 | tx->cookie = 0; | ||
1105 | } else { | ||
1106 | /* | ||
1107 | * last used desc. Do not remove, so we can | ||
1108 | * append from it, but don't look at it next | ||
1109 | * time, either | ||
1110 | */ | ||
1111 | tx->cookie = 0; | 787 | tx->cookie = 0; |
788 | } else { | ||
789 | /* | ||
790 | * last used desc. Do not remove, so we can | ||
791 | * append from it, but don't look at it next | ||
792 | * time, either | ||
793 | */ | ||
794 | tx->cookie = 0; | ||
1112 | 795 | ||
1113 | /* TODO check status bits? */ | 796 | /* TODO check status bits? */ |
1114 | break; | ||
1115 | } | ||
1116 | } | ||
1117 | break; | ||
1118 | case IOAT_VER_2_0: | ||
1119 | case IOAT_VER_3_0: | ||
1120 | /* has some other thread has already cleaned up? */ | ||
1121 | if (ioat->used_desc.prev == NULL) | ||
1122 | break; | 797 | break; |
1123 | |||
1124 | /* work backwards to find latest finished desc */ | ||
1125 | desc = to_ioat_desc(ioat->used_desc.next); | ||
1126 | tx = &desc->txd; | ||
1127 | latest_desc = NULL; | ||
1128 | do { | ||
1129 | desc = to_ioat_desc(desc->node.prev); | ||
1130 | desc_phys = (unsigned long)tx->phys | ||
1131 | & IOAT_CHANSTS_COMPLETED_DESCRIPTOR_ADDR; | ||
1132 | if (desc_phys == phys_complete) { | ||
1133 | latest_desc = desc; | ||
1134 | break; | ||
1135 | } | ||
1136 | } while (&desc->node != ioat->used_desc.prev); | ||
1137 | |||
1138 | if (latest_desc != NULL) { | ||
1139 | /* work forwards to clear finished descriptors */ | ||
1140 | for (desc = to_ioat_desc(ioat->used_desc.prev); | ||
1141 | &desc->node != latest_desc->node.next && | ||
1142 | &desc->node != ioat->used_desc.next; | ||
1143 | desc = to_ioat_desc(desc->node.next)) { | ||
1144 | if (tx->cookie) { | ||
1145 | cookie = tx->cookie; | ||
1146 | tx->cookie = 0; | ||
1147 | ioat_dma_unmap(chan, desc); | ||
1148 | if (tx->callback) { | ||
1149 | tx->callback(tx->callback_param); | ||
1150 | tx->callback = NULL; | ||
1151 | } | ||
1152 | } | ||
1153 | } | ||
1154 | |||
1155 | /* move used.prev up beyond those that are finished */ | ||
1156 | if (&desc->node == ioat->used_desc.next) | ||
1157 | ioat->used_desc.prev = NULL; | ||
1158 | else | ||
1159 | ioat->used_desc.prev = &desc->node; | ||
1160 | } | 798 | } |
1161 | break; | ||
1162 | } | 799 | } |
1163 | 800 | ||
1164 | spin_unlock_bh(&ioat->desc_lock); | 801 | spin_unlock_bh(&ioat->desc_lock); |
@@ -1170,50 +807,21 @@ static void ioat_dma_memcpy_cleanup(struct ioat_dma_chan *ioat) | |||
1170 | spin_unlock_bh(&chan->cleanup_lock); | 807 | spin_unlock_bh(&chan->cleanup_lock); |
1171 | } | 808 | } |
1172 | 809 | ||
1173 | /** | ||
1174 | * ioat_dma_is_complete - poll the status of a IOAT DMA transaction | ||
1175 | * @chan: IOAT DMA channel handle | ||
1176 | * @cookie: DMA transaction identifier | ||
1177 | * @done: if not %NULL, updated with last completed transaction | ||
1178 | * @used: if not %NULL, updated with last used transaction | ||
1179 | */ | ||
1180 | static enum dma_status | 810 | static enum dma_status |
1181 | ioat_dma_is_complete(struct dma_chan *c, dma_cookie_t cookie, | 811 | ioat1_dma_is_complete(struct dma_chan *c, dma_cookie_t cookie, |
1182 | dma_cookie_t *done, dma_cookie_t *used) | 812 | dma_cookie_t *done, dma_cookie_t *used) |
1183 | { | 813 | { |
1184 | struct ioat_dma_chan *ioat = to_ioat_chan(c); | 814 | struct ioat_dma_chan *ioat = to_ioat_chan(c); |
1185 | struct ioat_chan_common *chan = &ioat->base; | ||
1186 | dma_cookie_t last_used; | ||
1187 | dma_cookie_t last_complete; | ||
1188 | enum dma_status ret; | ||
1189 | |||
1190 | last_used = c->cookie; | ||
1191 | last_complete = chan->completed_cookie; | ||
1192 | chan->watchdog_tcp_cookie = cookie; | ||
1193 | |||
1194 | if (done) | ||
1195 | *done = last_complete; | ||
1196 | if (used) | ||
1197 | *used = last_used; | ||
1198 | |||
1199 | ret = dma_async_is_complete(cookie, last_complete, last_used); | ||
1200 | if (ret == DMA_SUCCESS) | ||
1201 | return ret; | ||
1202 | 815 | ||
1203 | ioat_dma_memcpy_cleanup(ioat); | 816 | if (ioat_is_complete(c, cookie, done, used) == DMA_SUCCESS) |
817 | return DMA_SUCCESS; | ||
1204 | 818 | ||
1205 | last_used = c->cookie; | 819 | ioat1_cleanup(ioat); |
1206 | last_complete = chan->completed_cookie; | ||
1207 | 820 | ||
1208 | if (done) | 821 | return ioat_is_complete(c, cookie, done, used); |
1209 | *done = last_complete; | ||
1210 | if (used) | ||
1211 | *used = last_used; | ||
1212 | |||
1213 | return dma_async_is_complete(cookie, last_complete, last_used); | ||
1214 | } | 822 | } |
1215 | 823 | ||
1216 | static void ioat_dma_start_null_desc(struct ioat_dma_chan *ioat) | 824 | static void ioat1_dma_start_null_desc(struct ioat_dma_chan *ioat) |
1217 | { | 825 | { |
1218 | struct ioat_chan_common *chan = &ioat->base; | 826 | struct ioat_chan_common *chan = &ioat->base; |
1219 | struct ioat_desc_sw *desc; | 827 | struct ioat_desc_sw *desc; |
@@ -1221,7 +829,7 @@ static void ioat_dma_start_null_desc(struct ioat_dma_chan *ioat) | |||
1221 | 829 | ||
1222 | spin_lock_bh(&ioat->desc_lock); | 830 | spin_lock_bh(&ioat->desc_lock); |
1223 | 831 | ||
1224 | desc = ioat_dma_get_next_descriptor(ioat); | 832 | desc = ioat1_dma_get_next_descriptor(ioat); |
1225 | 833 | ||
1226 | if (!desc) { | 834 | if (!desc) { |
1227 | dev_err(to_dev(chan), | 835 | dev_err(to_dev(chan), |
@@ -1240,30 +848,16 @@ static void ioat_dma_start_null_desc(struct ioat_dma_chan *ioat) | |||
1240 | hw->src_addr = 0; | 848 | hw->src_addr = 0; |
1241 | hw->dst_addr = 0; | 849 | hw->dst_addr = 0; |
1242 | async_tx_ack(&desc->txd); | 850 | async_tx_ack(&desc->txd); |
1243 | switch (chan->device->version) { | 851 | hw->next = 0; |
1244 | case IOAT_VER_1_2: | 852 | list_add_tail(&desc->node, &ioat->used_desc); |
1245 | hw->next = 0; | ||
1246 | list_add_tail(&desc->node, &ioat->used_desc); | ||
1247 | 853 | ||
1248 | writel(((u64) desc->txd.phys) & 0x00000000FFFFFFFF, | 854 | writel(((u64) desc->txd.phys) & 0x00000000FFFFFFFF, |
1249 | chan->reg_base + IOAT1_CHAINADDR_OFFSET_LOW); | 855 | chan->reg_base + IOAT1_CHAINADDR_OFFSET_LOW); |
1250 | writel(((u64) desc->txd.phys) >> 32, | 856 | writel(((u64) desc->txd.phys) >> 32, |
1251 | chan->reg_base + IOAT1_CHAINADDR_OFFSET_HIGH); | 857 | chan->reg_base + IOAT1_CHAINADDR_OFFSET_HIGH); |
1252 | |||
1253 | writeb(IOAT_CHANCMD_START, chan->reg_base | ||
1254 | + IOAT_CHANCMD_OFFSET(chan->device->version)); | ||
1255 | break; | ||
1256 | case IOAT_VER_2_0: | ||
1257 | case IOAT_VER_3_0: | ||
1258 | writel(((u64) desc->txd.phys) & 0x00000000FFFFFFFF, | ||
1259 | chan->reg_base + IOAT2_CHAINADDR_OFFSET_LOW); | ||
1260 | writel(((u64) desc->txd.phys) >> 32, | ||
1261 | chan->reg_base + IOAT2_CHAINADDR_OFFSET_HIGH); | ||
1262 | 858 | ||
1263 | ioat->dmacount++; | 859 | writeb(IOAT_CHANCMD_START, chan->reg_base |
1264 | __ioat2_dma_memcpy_issue_pending(ioat); | 860 | + IOAT_CHANCMD_OFFSET(chan->device->version)); |
1265 | break; | ||
1266 | } | ||
1267 | spin_unlock_bh(&ioat->desc_lock); | 861 | spin_unlock_bh(&ioat->desc_lock); |
1268 | } | 862 | } |
1269 | 863 | ||
@@ -1484,7 +1078,7 @@ static void ioat_disable_interrupts(struct ioatdma_device *device) | |||
1484 | writeb(0, device->reg_base + IOAT_INTRCTRL_OFFSET); | 1078 | writeb(0, device->reg_base + IOAT_INTRCTRL_OFFSET); |
1485 | } | 1079 | } |
1486 | 1080 | ||
1487 | static int ioat_probe(struct ioatdma_device *device) | 1081 | int ioat_probe(struct ioatdma_device *device) |
1488 | { | 1082 | { |
1489 | int err = -ENODEV; | 1083 | int err = -ENODEV; |
1490 | struct dma_device *dma = &device->common; | 1084 | struct dma_device *dma = &device->common; |
@@ -1503,17 +1097,15 @@ static int ioat_probe(struct ioatdma_device *device) | |||
1503 | device->completion_pool = pci_pool_create("completion_pool", pdev, | 1097 | device->completion_pool = pci_pool_create("completion_pool", pdev, |
1504 | sizeof(u64), SMP_CACHE_BYTES, | 1098 | sizeof(u64), SMP_CACHE_BYTES, |
1505 | SMP_CACHE_BYTES); | 1099 | SMP_CACHE_BYTES); |
1100 | |||
1506 | if (!device->completion_pool) { | 1101 | if (!device->completion_pool) { |
1507 | err = -ENOMEM; | 1102 | err = -ENOMEM; |
1508 | goto err_completion_pool; | 1103 | goto err_completion_pool; |
1509 | } | 1104 | } |
1510 | 1105 | ||
1511 | ioat_dma_enumerate_channels(device); | 1106 | device->enumerate_channels(device); |
1512 | 1107 | ||
1513 | dma_cap_set(DMA_MEMCPY, dma->cap_mask); | 1108 | dma_cap_set(DMA_MEMCPY, dma->cap_mask); |
1514 | dma->device_alloc_chan_resources = ioat_dma_alloc_chan_resources; | ||
1515 | dma->device_free_chan_resources = ioat_dma_free_chan_resources; | ||
1516 | dma->device_is_tx_complete = ioat_dma_is_complete; | ||
1517 | dma->dev = &pdev->dev; | 1109 | dma->dev = &pdev->dev; |
1518 | 1110 | ||
1519 | dev_err(dev, "Intel(R) I/OAT DMA Engine found," | 1111 | dev_err(dev, "Intel(R) I/OAT DMA Engine found," |
@@ -1546,7 +1138,7 @@ err_dma_pool: | |||
1546 | return err; | 1138 | return err; |
1547 | } | 1139 | } |
1548 | 1140 | ||
1549 | static int ioat_register(struct ioatdma_device *device) | 1141 | int ioat_register(struct ioatdma_device *device) |
1550 | { | 1142 | { |
1551 | int err = dma_async_device_register(&device->common); | 1143 | int err = dma_async_device_register(&device->common); |
1552 | 1144 | ||
@@ -1580,9 +1172,13 @@ int ioat1_dma_probe(struct ioatdma_device *device, int dca) | |||
1580 | int err; | 1172 | int err; |
1581 | 1173 | ||
1582 | device->intr_quirk = ioat1_intr_quirk; | 1174 | device->intr_quirk = ioat1_intr_quirk; |
1175 | device->enumerate_channels = ioat1_enumerate_channels; | ||
1583 | dma = &device->common; | 1176 | dma = &device->common; |
1584 | dma->device_prep_dma_memcpy = ioat1_dma_prep_memcpy; | 1177 | dma->device_prep_dma_memcpy = ioat1_dma_prep_memcpy; |
1585 | dma->device_issue_pending = ioat1_dma_memcpy_issue_pending; | 1178 | dma->device_issue_pending = ioat1_dma_memcpy_issue_pending; |
1179 | dma->device_alloc_chan_resources = ioat1_dma_alloc_chan_resources; | ||
1180 | dma->device_free_chan_resources = ioat1_dma_free_chan_resources; | ||
1181 | dma->device_is_tx_complete = ioat1_dma_is_complete; | ||
1586 | 1182 | ||
1587 | err = ioat_probe(device); | 1183 | err = ioat_probe(device); |
1588 | if (err) | 1184 | if (err) |
@@ -1594,93 +1190,12 @@ int ioat1_dma_probe(struct ioatdma_device *device, int dca) | |||
1594 | if (dca) | 1190 | if (dca) |
1595 | device->dca = ioat_dca_init(pdev, device->reg_base); | 1191 | device->dca = ioat_dca_init(pdev, device->reg_base); |
1596 | 1192 | ||
1597 | INIT_DELAYED_WORK(&device->work, ioat_dma_chan_watchdog); | 1193 | INIT_DELAYED_WORK(&device->work, ioat1_chan_watchdog); |
1598 | schedule_delayed_work(&device->work, WATCHDOG_DELAY); | ||
1599 | |||
1600 | return err; | ||
1601 | } | ||
1602 | |||
1603 | int ioat2_dma_probe(struct ioatdma_device *device, int dca) | ||
1604 | { | ||
1605 | struct pci_dev *pdev = device->pdev; | ||
1606 | struct dma_device *dma; | ||
1607 | struct dma_chan *c; | ||
1608 | struct ioat_chan_common *chan; | ||
1609 | int err; | ||
1610 | |||
1611 | dma = &device->common; | ||
1612 | dma->device_prep_dma_memcpy = ioat2_dma_prep_memcpy; | ||
1613 | dma->device_issue_pending = ioat2_dma_memcpy_issue_pending; | ||
1614 | |||
1615 | err = ioat_probe(device); | ||
1616 | if (err) | ||
1617 | return err; | ||
1618 | ioat_set_tcp_copy_break(2048); | ||
1619 | |||
1620 | list_for_each_entry(c, &dma->channels, device_node) { | ||
1621 | chan = to_chan_common(c); | ||
1622 | writel(IOAT_DCACTRL_CMPL_WRITE_ENABLE | IOAT_DMA_DCA_ANY_CPU, | ||
1623 | chan->reg_base + IOAT_DCACTRL_OFFSET); | ||
1624 | } | ||
1625 | |||
1626 | err = ioat_register(device); | ||
1627 | if (err) | ||
1628 | return err; | ||
1629 | if (dca) | ||
1630 | device->dca = ioat2_dca_init(pdev, device->reg_base); | ||
1631 | |||
1632 | INIT_DELAYED_WORK(&device->work, ioat_dma_chan_watchdog); | ||
1633 | schedule_delayed_work(&device->work, WATCHDOG_DELAY); | 1194 | schedule_delayed_work(&device->work, WATCHDOG_DELAY); |
1634 | 1195 | ||
1635 | return err; | 1196 | return err; |
1636 | } | 1197 | } |
1637 | 1198 | ||
1638 | int ioat3_dma_probe(struct ioatdma_device *device, int dca) | ||
1639 | { | ||
1640 | struct pci_dev *pdev = device->pdev; | ||
1641 | struct dma_device *dma; | ||
1642 | struct dma_chan *c; | ||
1643 | struct ioat_chan_common *chan; | ||
1644 | int err; | ||
1645 | u16 dev_id; | ||
1646 | |||
1647 | dma = &device->common; | ||
1648 | dma->device_prep_dma_memcpy = ioat2_dma_prep_memcpy; | ||
1649 | dma->device_issue_pending = ioat2_dma_memcpy_issue_pending; | ||
1650 | |||
1651 | /* -= IOAT ver.3 workarounds =- */ | ||
1652 | /* Write CHANERRMSK_INT with 3E07h to mask out the errors | ||
1653 | * that can cause stability issues for IOAT ver.3 | ||
1654 | */ | ||
1655 | pci_write_config_dword(pdev, IOAT_PCI_CHANERRMASK_INT_OFFSET, 0x3e07); | ||
1656 | |||
1657 | /* Clear DMAUNCERRSTS Cfg-Reg Parity Error status bit | ||
1658 | * (workaround for spurious config parity error after restart) | ||
1659 | */ | ||
1660 | pci_read_config_word(pdev, IOAT_PCI_DEVICE_ID_OFFSET, &dev_id); | ||
1661 | if (dev_id == PCI_DEVICE_ID_INTEL_IOAT_TBG0) | ||
1662 | pci_write_config_dword(pdev, IOAT_PCI_DMAUNCERRSTS_OFFSET, 0x10); | ||
1663 | |||
1664 | err = ioat_probe(device); | ||
1665 | if (err) | ||
1666 | return err; | ||
1667 | ioat_set_tcp_copy_break(262144); | ||
1668 | |||
1669 | list_for_each_entry(c, &dma->channels, device_node) { | ||
1670 | chan = to_chan_common(c); | ||
1671 | writel(IOAT_DMA_DCA_ANY_CPU, | ||
1672 | chan->reg_base + IOAT_DCACTRL_OFFSET); | ||
1673 | } | ||
1674 | |||
1675 | err = ioat_register(device); | ||
1676 | if (err) | ||
1677 | return err; | ||
1678 | if (dca) | ||
1679 | device->dca = ioat3_dca_init(pdev, device->reg_base); | ||
1680 | |||
1681 | return err; | ||
1682 | } | ||
1683 | |||
1684 | void ioat_dma_remove(struct ioatdma_device *device) | 1199 | void ioat_dma_remove(struct ioatdma_device *device) |
1685 | { | 1200 | { |
1686 | struct dma_device *dma = &device->common; | 1201 | struct dma_device *dma = &device->common; |
@@ -1697,4 +1212,3 @@ void ioat_dma_remove(struct ioatdma_device *device) | |||
1697 | 1212 | ||
1698 | INIT_LIST_HEAD(&dma->channels); | 1213 | INIT_LIST_HEAD(&dma->channels); |
1699 | } | 1214 | } |
1700 | |||
diff --git a/drivers/dma/ioat/dma.h b/drivers/dma/ioat/dma.h index 5b31db73ad8e..84065dfa4d40 100644 --- a/drivers/dma/ioat/dma.h +++ b/drivers/dma/ioat/dma.h | |||
@@ -62,6 +62,7 @@ | |||
62 | * @idx: per channel data | 62 | * @idx: per channel data |
63 | * @dca: direct cache access context | 63 | * @dca: direct cache access context |
64 | * @intr_quirk: interrupt setup quirk (for ioat_v1 devices) | 64 | * @intr_quirk: interrupt setup quirk (for ioat_v1 devices) |
65 | * @enumerate_channels: hw version specific channel enumeration | ||
65 | */ | 66 | */ |
66 | 67 | ||
67 | struct ioatdma_device { | 68 | struct ioatdma_device { |
@@ -76,6 +77,7 @@ struct ioatdma_device { | |||
76 | struct ioat_chan_common *idx[4]; | 77 | struct ioat_chan_common *idx[4]; |
77 | struct dca_provider *dca; | 78 | struct dca_provider *dca; |
78 | void (*intr_quirk)(struct ioatdma_device *device); | 79 | void (*intr_quirk)(struct ioatdma_device *device); |
80 | int (*enumerate_channels)(struct ioatdma_device *device); | ||
79 | }; | 81 | }; |
80 | 82 | ||
81 | struct ioat_chan_common { | 83 | struct ioat_chan_common { |
@@ -106,6 +108,7 @@ struct ioat_chan_common { | |||
106 | struct tasklet_struct cleanup_task; | 108 | struct tasklet_struct cleanup_task; |
107 | }; | 109 | }; |
108 | 110 | ||
111 | |||
109 | /** | 112 | /** |
110 | * struct ioat_dma_chan - internal representation of a DMA channel | 113 | * struct ioat_dma_chan - internal representation of a DMA channel |
111 | */ | 114 | */ |
@@ -119,7 +122,6 @@ struct ioat_dma_chan { | |||
119 | struct list_head used_desc; | 122 | struct list_head used_desc; |
120 | 123 | ||
121 | int pending; | 124 | int pending; |
122 | u16 dmacount; | ||
123 | u16 desccount; | 125 | u16 desccount; |
124 | }; | 126 | }; |
125 | 127 | ||
@@ -135,6 +137,33 @@ static inline struct ioat_dma_chan *to_ioat_chan(struct dma_chan *c) | |||
135 | return container_of(chan, struct ioat_dma_chan, base); | 137 | return container_of(chan, struct ioat_dma_chan, base); |
136 | } | 138 | } |
137 | 139 | ||
140 | /** | ||
141 | * ioat_is_complete - poll the status of an ioat transaction | ||
142 | * @c: channel handle | ||
143 | * @cookie: transaction identifier | ||
144 | * @done: if set, updated with last completed transaction | ||
145 | * @used: if set, updated with last used transaction | ||
146 | */ | ||
147 | static inline enum dma_status | ||
148 | ioat_is_complete(struct dma_chan *c, dma_cookie_t cookie, | ||
149 | dma_cookie_t *done, dma_cookie_t *used) | ||
150 | { | ||
151 | struct ioat_chan_common *chan = to_chan_common(c); | ||
152 | dma_cookie_t last_used; | ||
153 | dma_cookie_t last_complete; | ||
154 | |||
155 | last_used = c->cookie; | ||
156 | last_complete = chan->completed_cookie; | ||
157 | chan->watchdog_tcp_cookie = cookie; | ||
158 | |||
159 | if (done) | ||
160 | *done = last_complete; | ||
161 | if (used) | ||
162 | *used = last_used; | ||
163 | |||
164 | return dma_async_is_complete(cookie, last_complete, last_used); | ||
165 | } | ||
166 | |||
138 | /* wrapper around hardware descriptor format + additional software fields */ | 167 | /* wrapper around hardware descriptor format + additional software fields */ |
139 | 168 | ||
140 | /** | 169 | /** |
@@ -162,11 +191,22 @@ static inline void ioat_set_tcp_copy_break(unsigned long copybreak) | |||
162 | #endif | 191 | #endif |
163 | } | 192 | } |
164 | 193 | ||
194 | static inline struct ioat_chan_common * | ||
195 | ioat_chan_by_index(struct ioatdma_device *device, int index) | ||
196 | { | ||
197 | return device->idx[index]; | ||
198 | } | ||
199 | |||
200 | int ioat_probe(struct ioatdma_device *device); | ||
201 | int ioat_register(struct ioatdma_device *device); | ||
165 | int ioat1_dma_probe(struct ioatdma_device *dev, int dca); | 202 | int ioat1_dma_probe(struct ioatdma_device *dev, int dca); |
166 | int ioat2_dma_probe(struct ioatdma_device *dev, int dca); | ||
167 | int ioat3_dma_probe(struct ioatdma_device *dev, int dca); | ||
168 | void ioat_dma_remove(struct ioatdma_device *device); | 203 | void ioat_dma_remove(struct ioatdma_device *device); |
169 | struct dca_provider *ioat_dca_init(struct pci_dev *pdev, void __iomem *iobase); | 204 | struct dca_provider *ioat_dca_init(struct pci_dev *pdev, void __iomem *iobase); |
170 | struct dca_provider *ioat2_dca_init(struct pci_dev *pdev, void __iomem *iobase); | 205 | unsigned long ioat_get_current_completion(struct ioat_chan_common *chan); |
171 | struct dca_provider *ioat3_dca_init(struct pci_dev *pdev, void __iomem *iobase); | 206 | void ioat_init_channel(struct ioatdma_device *device, |
207 | struct ioat_chan_common *chan, int idx, | ||
208 | work_func_t work_fn, void (*tasklet)(unsigned long), | ||
209 | unsigned long tasklet_data); | ||
210 | void ioat_dma_unmap(struct ioat_chan_common *chan, enum dma_ctrl_flags flags, | ||
211 | size_t len, struct ioat_dma_descriptor *hw); | ||
172 | #endif /* IOATDMA_H */ | 212 | #endif /* IOATDMA_H */ |
diff --git a/drivers/dma/ioat/dma_v2.c b/drivers/dma/ioat/dma_v2.c new file mode 100644 index 000000000000..49ba1c73d95e --- /dev/null +++ b/drivers/dma/ioat/dma_v2.c | |||
@@ -0,0 +1,750 @@ | |||
1 | /* | ||
2 | * Intel I/OAT DMA Linux driver | ||
3 | * Copyright(c) 2004 - 2009 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 that 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 | * The full GNU General Public License is included in this distribution in | ||
19 | * the file called "COPYING". | ||
20 | * | ||
21 | */ | ||
22 | |||
23 | /* | ||
24 | * This driver supports an Intel I/OAT DMA engine (versions >= 2), which | ||
25 | * does asynchronous data movement and checksumming operations. | ||
26 | */ | ||
27 | |||
28 | #include <linux/init.h> | ||
29 | #include <linux/module.h> | ||
30 | #include <linux/pci.h> | ||
31 | #include <linux/interrupt.h> | ||
32 | #include <linux/dmaengine.h> | ||
33 | #include <linux/delay.h> | ||
34 | #include <linux/dma-mapping.h> | ||
35 | #include <linux/workqueue.h> | ||
36 | #include <linux/i7300_idle.h> | ||
37 | #include "dma.h" | ||
38 | #include "dma_v2.h" | ||
39 | #include "registers.h" | ||
40 | #include "hw.h" | ||
41 | |||
42 | static int ioat_ring_alloc_order = 8; | ||
43 | module_param(ioat_ring_alloc_order, int, 0644); | ||
44 | MODULE_PARM_DESC(ioat_ring_alloc_order, | ||
45 | "ioat2+: allocate 2^n descriptors per channel (default: n=8)"); | ||
46 | |||
47 | static void __ioat2_issue_pending(struct ioat2_dma_chan *ioat) | ||
48 | { | ||
49 | void * __iomem reg_base = ioat->base.reg_base; | ||
50 | |||
51 | ioat->pending = 0; | ||
52 | ioat->dmacount += ioat2_ring_pending(ioat); | ||
53 | ioat->issued = ioat->head; | ||
54 | /* make descriptor updates globally visible before notifying channel */ | ||
55 | wmb(); | ||
56 | writew(ioat->dmacount, reg_base + IOAT_CHAN_DMACOUNT_OFFSET); | ||
57 | |||
58 | } | ||
59 | |||
60 | static void ioat2_issue_pending(struct dma_chan *chan) | ||
61 | { | ||
62 | struct ioat2_dma_chan *ioat = to_ioat2_chan(chan); | ||
63 | |||
64 | spin_lock_bh(&ioat->ring_lock); | ||
65 | if (ioat->pending == 1) | ||
66 | __ioat2_issue_pending(ioat); | ||
67 | spin_unlock_bh(&ioat->ring_lock); | ||
68 | } | ||
69 | |||
70 | /** | ||
71 | * ioat2_update_pending - log pending descriptors | ||
72 | * @ioat: ioat2+ channel | ||
73 | * | ||
74 | * set pending to '1' unless pending is already set to '2', pending == 2 | ||
75 | * indicates that submission is temporarily blocked due to an in-flight | ||
76 | * reset. If we are already above the ioat_pending_level threshold then | ||
77 | * just issue pending. | ||
78 | * | ||
79 | * called with ring_lock held | ||
80 | */ | ||
81 | static void ioat2_update_pending(struct ioat2_dma_chan *ioat) | ||
82 | { | ||
83 | if (unlikely(ioat->pending == 2)) | ||
84 | return; | ||
85 | else if (ioat2_ring_pending(ioat) > ioat_pending_level) | ||
86 | __ioat2_issue_pending(ioat); | ||
87 | else | ||
88 | ioat->pending = 1; | ||
89 | } | ||
90 | |||
91 | static void __ioat2_start_null_desc(struct ioat2_dma_chan *ioat) | ||
92 | { | ||
93 | void __iomem *reg_base = ioat->base.reg_base; | ||
94 | struct ioat_ring_ent *desc; | ||
95 | struct ioat_dma_descriptor *hw; | ||
96 | int idx; | ||
97 | |||
98 | if (ioat2_ring_space(ioat) < 1) { | ||
99 | dev_err(to_dev(&ioat->base), | ||
100 | "Unable to start null desc - ring full\n"); | ||
101 | return; | ||
102 | } | ||
103 | |||
104 | idx = ioat2_desc_alloc(ioat, 1); | ||
105 | desc = ioat2_get_ring_ent(ioat, idx); | ||
106 | |||
107 | hw = desc->hw; | ||
108 | hw->ctl = 0; | ||
109 | hw->ctl_f.null = 1; | ||
110 | hw->ctl_f.int_en = 1; | ||
111 | hw->ctl_f.compl_write = 1; | ||
112 | /* set size to non-zero value (channel returns error when size is 0) */ | ||
113 | hw->size = NULL_DESC_BUFFER_SIZE; | ||
114 | hw->src_addr = 0; | ||
115 | hw->dst_addr = 0; | ||
116 | async_tx_ack(&desc->txd); | ||
117 | writel(((u64) desc->txd.phys) & 0x00000000FFFFFFFF, | ||
118 | reg_base + IOAT2_CHAINADDR_OFFSET_LOW); | ||
119 | writel(((u64) desc->txd.phys) >> 32, | ||
120 | reg_base + IOAT2_CHAINADDR_OFFSET_HIGH); | ||
121 | __ioat2_issue_pending(ioat); | ||
122 | } | ||
123 | |||
124 | static void ioat2_start_null_desc(struct ioat2_dma_chan *ioat) | ||
125 | { | ||
126 | spin_lock_bh(&ioat->ring_lock); | ||
127 | __ioat2_start_null_desc(ioat); | ||
128 | spin_unlock_bh(&ioat->ring_lock); | ||
129 | } | ||
130 | |||
131 | static void ioat2_cleanup(struct ioat2_dma_chan *ioat); | ||
132 | |||
133 | /** | ||
134 | * ioat2_reset_part2 - reinit the channel after a reset | ||
135 | */ | ||
136 | static void ioat2_reset_part2(struct work_struct *work) | ||
137 | { | ||
138 | struct ioat_chan_common *chan; | ||
139 | struct ioat2_dma_chan *ioat; | ||
140 | |||
141 | chan = container_of(work, struct ioat_chan_common, work.work); | ||
142 | ioat = container_of(chan, struct ioat2_dma_chan, base); | ||
143 | |||
144 | /* ensure that ->tail points to the stalled descriptor | ||
145 | * (ioat->pending is set to 2 at this point so no new | ||
146 | * descriptors will be issued while we perform this cleanup) | ||
147 | */ | ||
148 | ioat2_cleanup(ioat); | ||
149 | |||
150 | spin_lock_bh(&chan->cleanup_lock); | ||
151 | spin_lock_bh(&ioat->ring_lock); | ||
152 | |||
153 | /* set the tail to be re-issued */ | ||
154 | ioat->issued = ioat->tail; | ||
155 | ioat->dmacount = 0; | ||
156 | |||
157 | if (ioat2_ring_pending(ioat)) { | ||
158 | struct ioat_ring_ent *desc; | ||
159 | |||
160 | desc = ioat2_get_ring_ent(ioat, ioat->tail); | ||
161 | writel(((u64) desc->txd.phys) & 0x00000000FFFFFFFF, | ||
162 | chan->reg_base + IOAT2_CHAINADDR_OFFSET_LOW); | ||
163 | writel(((u64) desc->txd.phys) >> 32, | ||
164 | chan->reg_base + IOAT2_CHAINADDR_OFFSET_HIGH); | ||
165 | __ioat2_issue_pending(ioat); | ||
166 | } else | ||
167 | __ioat2_start_null_desc(ioat); | ||
168 | |||
169 | spin_unlock_bh(&ioat->ring_lock); | ||
170 | spin_unlock_bh(&chan->cleanup_lock); | ||
171 | |||
172 | dev_info(to_dev(chan), | ||
173 | "chan%d reset - %d descs waiting, %d total desc\n", | ||
174 | chan_num(chan), ioat->dmacount, 1 << ioat->alloc_order); | ||
175 | } | ||
176 | |||
177 | /** | ||
178 | * ioat2_reset_channel - restart a channel | ||
179 | * @ioat: IOAT DMA channel handle | ||
180 | */ | ||
181 | static void ioat2_reset_channel(struct ioat2_dma_chan *ioat) | ||
182 | { | ||
183 | u32 chansts, chanerr; | ||
184 | struct ioat_chan_common *chan = &ioat->base; | ||
185 | u16 active; | ||
186 | |||
187 | spin_lock_bh(&ioat->ring_lock); | ||
188 | active = ioat2_ring_active(ioat); | ||
189 | spin_unlock_bh(&ioat->ring_lock); | ||
190 | if (!active) | ||
191 | return; | ||
192 | |||
193 | chanerr = readl(chan->reg_base + IOAT_CHANERR_OFFSET); | ||
194 | chansts = (chan->completion_virt->low | ||
195 | & IOAT_CHANSTS_DMA_TRANSFER_STATUS); | ||
196 | if (chanerr) { | ||
197 | dev_err(to_dev(chan), | ||
198 | "chan%d, CHANSTS = 0x%08x CHANERR = 0x%04x, clearing\n", | ||
199 | chan_num(chan), chansts, chanerr); | ||
200 | writel(chanerr, chan->reg_base + IOAT_CHANERR_OFFSET); | ||
201 | } | ||
202 | |||
203 | spin_lock_bh(&ioat->ring_lock); | ||
204 | ioat->pending = 2; | ||
205 | writeb(IOAT_CHANCMD_RESET, | ||
206 | chan->reg_base | ||
207 | + IOAT_CHANCMD_OFFSET(chan->device->version)); | ||
208 | spin_unlock_bh(&ioat->ring_lock); | ||
209 | schedule_delayed_work(&chan->work, RESET_DELAY); | ||
210 | } | ||
211 | |||
212 | /** | ||
213 | * ioat2_chan_watchdog - watch for stuck channels | ||
214 | */ | ||
215 | static void ioat2_chan_watchdog(struct work_struct *work) | ||
216 | { | ||
217 | struct ioatdma_device *device = | ||
218 | container_of(work, struct ioatdma_device, work.work); | ||
219 | struct ioat2_dma_chan *ioat; | ||
220 | struct ioat_chan_common *chan; | ||
221 | u16 active; | ||
222 | int i; | ||
223 | |||
224 | for (i = 0; i < device->common.chancnt; i++) { | ||
225 | chan = ioat_chan_by_index(device, i); | ||
226 | ioat = container_of(chan, struct ioat2_dma_chan, base); | ||
227 | |||
228 | /* | ||
229 | * for version 2.0 if there are descriptors yet to be processed | ||
230 | * and the last completed hasn't changed since the last watchdog | ||
231 | * if they haven't hit the pending level | ||
232 | * issue the pending to push them through | ||
233 | * else | ||
234 | * try resetting the channel | ||
235 | */ | ||
236 | spin_lock_bh(&ioat->ring_lock); | ||
237 | active = ioat2_ring_active(ioat); | ||
238 | spin_unlock_bh(&ioat->ring_lock); | ||
239 | |||
240 | if (active && | ||
241 | chan->last_completion && | ||
242 | chan->last_completion == chan->watchdog_completion) { | ||
243 | |||
244 | if (ioat->pending == 1) | ||
245 | ioat2_issue_pending(&chan->common); | ||
246 | else { | ||
247 | ioat2_reset_channel(ioat); | ||
248 | chan->watchdog_completion = 0; | ||
249 | } | ||
250 | } else { | ||
251 | chan->last_compl_desc_addr_hw = 0; | ||
252 | chan->watchdog_completion = chan->last_completion; | ||
253 | } | ||
254 | chan->watchdog_last_tcp_cookie = chan->watchdog_tcp_cookie; | ||
255 | } | ||
256 | schedule_delayed_work(&device->work, WATCHDOG_DELAY); | ||
257 | } | ||
258 | |||
259 | /** | ||
260 | * ioat2_cleanup - clean finished descriptors (advance tail pointer) | ||
261 | * @chan: ioat channel to be cleaned up | ||
262 | */ | ||
263 | static void ioat2_cleanup(struct ioat2_dma_chan *ioat) | ||
264 | { | ||
265 | struct ioat_chan_common *chan = &ioat->base; | ||
266 | unsigned long phys_complete; | ||
267 | struct ioat_ring_ent *desc; | ||
268 | bool seen_current = false; | ||
269 | u16 active; | ||
270 | int i; | ||
271 | struct dma_async_tx_descriptor *tx; | ||
272 | |||
273 | prefetch(chan->completion_virt); | ||
274 | |||
275 | spin_lock_bh(&chan->cleanup_lock); | ||
276 | phys_complete = ioat_get_current_completion(chan); | ||
277 | if (phys_complete == chan->last_completion) { | ||
278 | spin_unlock_bh(&chan->cleanup_lock); | ||
279 | /* | ||
280 | * perhaps we're stuck so hard that the watchdog can't go off? | ||
281 | * try to catch it after WATCHDOG_DELAY seconds | ||
282 | */ | ||
283 | if (chan->device->version < IOAT_VER_3_0) { | ||
284 | unsigned long tmo; | ||
285 | |||
286 | tmo = chan->last_completion_time + HZ*WATCHDOG_DELAY; | ||
287 | if (time_after(jiffies, tmo)) { | ||
288 | ioat2_chan_watchdog(&(chan->device->work.work)); | ||
289 | chan->last_completion_time = jiffies; | ||
290 | } | ||
291 | } | ||
292 | return; | ||
293 | } | ||
294 | chan->last_completion_time = jiffies; | ||
295 | |||
296 | spin_lock_bh(&ioat->ring_lock); | ||
297 | |||
298 | active = ioat2_ring_active(ioat); | ||
299 | for (i = 0; i < active && !seen_current; i++) { | ||
300 | prefetch(ioat2_get_ring_ent(ioat, ioat->tail + i + 1)); | ||
301 | desc = ioat2_get_ring_ent(ioat, ioat->tail + i); | ||
302 | tx = &desc->txd; | ||
303 | if (tx->cookie) { | ||
304 | ioat_dma_unmap(chan, tx->flags, desc->len, desc->hw); | ||
305 | chan->completed_cookie = tx->cookie; | ||
306 | tx->cookie = 0; | ||
307 | if (tx->callback) { | ||
308 | tx->callback(tx->callback_param); | ||
309 | tx->callback = NULL; | ||
310 | } | ||
311 | } | ||
312 | |||
313 | if (tx->phys == phys_complete) | ||
314 | seen_current = true; | ||
315 | } | ||
316 | ioat->tail += i; | ||
317 | BUG_ON(!seen_current); /* no active descs have written a completion? */ | ||
318 | spin_unlock_bh(&ioat->ring_lock); | ||
319 | |||
320 | chan->last_completion = phys_complete; | ||
321 | |||
322 | spin_unlock_bh(&chan->cleanup_lock); | ||
323 | } | ||
324 | |||
325 | static void ioat2_cleanup_tasklet(unsigned long data) | ||
326 | { | ||
327 | struct ioat2_dma_chan *ioat = (void *) data; | ||
328 | |||
329 | ioat2_cleanup(ioat); | ||
330 | writew(IOAT_CHANCTRL_INT_DISABLE, | ||
331 | ioat->base.reg_base + IOAT_CHANCTRL_OFFSET); | ||
332 | } | ||
333 | |||
334 | /** | ||
335 | * ioat2_enumerate_channels - find and initialize the device's channels | ||
336 | * @device: the device to be enumerated | ||
337 | */ | ||
338 | static int ioat2_enumerate_channels(struct ioatdma_device *device) | ||
339 | { | ||
340 | struct ioat2_dma_chan *ioat; | ||
341 | struct device *dev = &device->pdev->dev; | ||
342 | struct dma_device *dma = &device->common; | ||
343 | u8 xfercap_log; | ||
344 | int i; | ||
345 | |||
346 | INIT_LIST_HEAD(&dma->channels); | ||
347 | dma->chancnt = readb(device->reg_base + IOAT_CHANCNT_OFFSET); | ||
348 | xfercap_log = readb(device->reg_base + IOAT_XFERCAP_OFFSET); | ||
349 | if (xfercap_log == 0) | ||
350 | return 0; | ||
351 | |||
352 | /* FIXME which i/oat version is i7300? */ | ||
353 | #ifdef CONFIG_I7300_IDLE_IOAT_CHANNEL | ||
354 | if (i7300_idle_platform_probe(NULL, NULL, 1) == 0) | ||
355 | dma->chancnt--; | ||
356 | #endif | ||
357 | for (i = 0; i < dma->chancnt; i++) { | ||
358 | ioat = devm_kzalloc(dev, sizeof(*ioat), GFP_KERNEL); | ||
359 | if (!ioat) | ||
360 | break; | ||
361 | |||
362 | ioat_init_channel(device, &ioat->base, i, | ||
363 | ioat2_reset_part2, | ||
364 | ioat2_cleanup_tasklet, | ||
365 | (unsigned long) ioat); | ||
366 | ioat->xfercap_log = xfercap_log; | ||
367 | spin_lock_init(&ioat->ring_lock); | ||
368 | } | ||
369 | dma->chancnt = i; | ||
370 | return i; | ||
371 | } | ||
372 | |||
373 | static dma_cookie_t ioat2_tx_submit_unlock(struct dma_async_tx_descriptor *tx) | ||
374 | { | ||
375 | struct dma_chan *c = tx->chan; | ||
376 | struct ioat2_dma_chan *ioat = to_ioat2_chan(c); | ||
377 | dma_cookie_t cookie = c->cookie; | ||
378 | |||
379 | cookie++; | ||
380 | if (cookie < 0) | ||
381 | cookie = 1; | ||
382 | tx->cookie = cookie; | ||
383 | c->cookie = cookie; | ||
384 | ioat2_update_pending(ioat); | ||
385 | spin_unlock_bh(&ioat->ring_lock); | ||
386 | |||
387 | return cookie; | ||
388 | } | ||
389 | |||
390 | static struct ioat_ring_ent *ioat2_alloc_ring_ent(struct dma_chan *chan) | ||
391 | { | ||
392 | struct ioat_dma_descriptor *hw; | ||
393 | struct ioat_ring_ent *desc; | ||
394 | struct ioatdma_device *dma; | ||
395 | dma_addr_t phys; | ||
396 | |||
397 | dma = to_ioatdma_device(chan->device); | ||
398 | hw = pci_pool_alloc(dma->dma_pool, GFP_KERNEL, &phys); | ||
399 | if (!hw) | ||
400 | return NULL; | ||
401 | memset(hw, 0, sizeof(*hw)); | ||
402 | |||
403 | desc = kzalloc(sizeof(*desc), GFP_KERNEL); | ||
404 | if (!desc) { | ||
405 | pci_pool_free(dma->dma_pool, hw, phys); | ||
406 | return NULL; | ||
407 | } | ||
408 | |||
409 | dma_async_tx_descriptor_init(&desc->txd, chan); | ||
410 | desc->txd.tx_submit = ioat2_tx_submit_unlock; | ||
411 | desc->hw = hw; | ||
412 | desc->txd.phys = phys; | ||
413 | return desc; | ||
414 | } | ||
415 | |||
416 | static void ioat2_free_ring_ent(struct ioat_ring_ent *desc, struct dma_chan *chan) | ||
417 | { | ||
418 | struct ioatdma_device *dma; | ||
419 | |||
420 | dma = to_ioatdma_device(chan->device); | ||
421 | pci_pool_free(dma->dma_pool, desc->hw, desc->txd.phys); | ||
422 | kfree(desc); | ||
423 | } | ||
424 | |||
425 | /* ioat2_alloc_chan_resources - allocate/initialize ioat2 descriptor ring | ||
426 | * @chan: channel to be initialized | ||
427 | */ | ||
428 | static int ioat2_alloc_chan_resources(struct dma_chan *c) | ||
429 | { | ||
430 | struct ioat2_dma_chan *ioat = to_ioat2_chan(c); | ||
431 | struct ioat_chan_common *chan = &ioat->base; | ||
432 | struct ioat_ring_ent **ring; | ||
433 | u16 chanctrl; | ||
434 | u32 chanerr; | ||
435 | int descs; | ||
436 | int i; | ||
437 | |||
438 | /* have we already been set up? */ | ||
439 | if (ioat->ring) | ||
440 | return 1 << ioat->alloc_order; | ||
441 | |||
442 | /* Setup register to interrupt and write completion status on error */ | ||
443 | chanctrl = IOAT_CHANCTRL_ERR_INT_EN | IOAT_CHANCTRL_ANY_ERR_ABORT_EN | | ||
444 | IOAT_CHANCTRL_ERR_COMPLETION_EN; | ||
445 | writew(chanctrl, chan->reg_base + IOAT_CHANCTRL_OFFSET); | ||
446 | |||
447 | chanerr = readl(chan->reg_base + IOAT_CHANERR_OFFSET); | ||
448 | if (chanerr) { | ||
449 | dev_err(to_dev(chan), "CHANERR = %x, clearing\n", chanerr); | ||
450 | writel(chanerr, chan->reg_base + IOAT_CHANERR_OFFSET); | ||
451 | } | ||
452 | |||
453 | /* allocate a completion writeback area */ | ||
454 | /* doing 2 32bit writes to mmio since 1 64b write doesn't work */ | ||
455 | chan->completion_virt = pci_pool_alloc(chan->device->completion_pool, | ||
456 | GFP_KERNEL, | ||
457 | &chan->completion_addr); | ||
458 | if (!chan->completion_virt) | ||
459 | return -ENOMEM; | ||
460 | |||
461 | memset(chan->completion_virt, 0, | ||
462 | sizeof(*chan->completion_virt)); | ||
463 | writel(((u64) chan->completion_addr) & 0x00000000FFFFFFFF, | ||
464 | chan->reg_base + IOAT_CHANCMP_OFFSET_LOW); | ||
465 | writel(((u64) chan->completion_addr) >> 32, | ||
466 | chan->reg_base + IOAT_CHANCMP_OFFSET_HIGH); | ||
467 | |||
468 | ioat->alloc_order = ioat_get_alloc_order(); | ||
469 | descs = 1 << ioat->alloc_order; | ||
470 | |||
471 | /* allocate the array to hold the software ring */ | ||
472 | ring = kcalloc(descs, sizeof(*ring), GFP_KERNEL); | ||
473 | if (!ring) | ||
474 | return -ENOMEM; | ||
475 | for (i = 0; i < descs; i++) { | ||
476 | ring[i] = ioat2_alloc_ring_ent(c); | ||
477 | if (!ring[i]) { | ||
478 | while (i--) | ||
479 | ioat2_free_ring_ent(ring[i], c); | ||
480 | kfree(ring); | ||
481 | return -ENOMEM; | ||
482 | } | ||
483 | } | ||
484 | |||
485 | /* link descs */ | ||
486 | for (i = 0; i < descs-1; i++) { | ||
487 | struct ioat_ring_ent *next = ring[i+1]; | ||
488 | struct ioat_dma_descriptor *hw = ring[i]->hw; | ||
489 | |||
490 | hw->next = next->txd.phys; | ||
491 | } | ||
492 | ring[i]->hw->next = ring[0]->txd.phys; | ||
493 | |||
494 | spin_lock_bh(&ioat->ring_lock); | ||
495 | ioat->ring = ring; | ||
496 | ioat->head = 0; | ||
497 | ioat->issued = 0; | ||
498 | ioat->tail = 0; | ||
499 | ioat->pending = 0; | ||
500 | spin_unlock_bh(&ioat->ring_lock); | ||
501 | |||
502 | tasklet_enable(&chan->cleanup_task); | ||
503 | ioat2_start_null_desc(ioat); | ||
504 | |||
505 | return descs; | ||
506 | } | ||
507 | |||
508 | /** | ||
509 | * ioat2_alloc_and_lock - common descriptor alloc boilerplate for ioat2,3 ops | ||
510 | * @idx: gets starting descriptor index on successful allocation | ||
511 | * @ioat: ioat2,3 channel (ring) to operate on | ||
512 | * @num_descs: allocation length | ||
513 | */ | ||
514 | static int ioat2_alloc_and_lock(u16 *idx, struct ioat2_dma_chan *ioat, int num_descs) | ||
515 | { | ||
516 | struct ioat_chan_common *chan = &ioat->base; | ||
517 | |||
518 | spin_lock_bh(&ioat->ring_lock); | ||
519 | if (unlikely(ioat2_ring_space(ioat) < num_descs)) { | ||
520 | if (printk_ratelimit()) | ||
521 | dev_dbg(to_dev(chan), | ||
522 | "%s: ring full! num_descs: %d (%x:%x:%x)\n", | ||
523 | __func__, num_descs, ioat->head, ioat->tail, | ||
524 | ioat->issued); | ||
525 | spin_unlock_bh(&ioat->ring_lock); | ||
526 | |||
527 | /* do direct reclaim in the allocation failure case */ | ||
528 | ioat2_cleanup(ioat); | ||
529 | |||
530 | return -ENOMEM; | ||
531 | } | ||
532 | |||
533 | dev_dbg(to_dev(chan), "%s: num_descs: %d (%x:%x:%x)\n", | ||
534 | __func__, num_descs, ioat->head, ioat->tail, ioat->issued); | ||
535 | |||
536 | *idx = ioat2_desc_alloc(ioat, num_descs); | ||
537 | return 0; /* with ioat->ring_lock held */ | ||
538 | } | ||
539 | |||
540 | static struct dma_async_tx_descriptor * | ||
541 | ioat2_dma_prep_memcpy_lock(struct dma_chan *c, dma_addr_t dma_dest, | ||
542 | dma_addr_t dma_src, size_t len, unsigned long flags) | ||
543 | { | ||
544 | struct ioat2_dma_chan *ioat = to_ioat2_chan(c); | ||
545 | struct ioat_dma_descriptor *hw; | ||
546 | struct ioat_ring_ent *desc; | ||
547 | dma_addr_t dst = dma_dest; | ||
548 | dma_addr_t src = dma_src; | ||
549 | size_t total_len = len; | ||
550 | int num_descs; | ||
551 | u16 idx; | ||
552 | int i; | ||
553 | |||
554 | num_descs = ioat2_xferlen_to_descs(ioat, len); | ||
555 | if (likely(num_descs) && | ||
556 | ioat2_alloc_and_lock(&idx, ioat, num_descs) == 0) | ||
557 | /* pass */; | ||
558 | else | ||
559 | return NULL; | ||
560 | for (i = 0; i < num_descs; i++) { | ||
561 | size_t copy = min_t(size_t, len, 1 << ioat->xfercap_log); | ||
562 | |||
563 | desc = ioat2_get_ring_ent(ioat, idx + i); | ||
564 | hw = desc->hw; | ||
565 | |||
566 | hw->size = copy; | ||
567 | hw->ctl = 0; | ||
568 | hw->src_addr = src; | ||
569 | hw->dst_addr = dst; | ||
570 | |||
571 | len -= copy; | ||
572 | dst += copy; | ||
573 | src += copy; | ||
574 | } | ||
575 | |||
576 | desc->txd.flags = flags; | ||
577 | desc->len = total_len; | ||
578 | hw->ctl_f.int_en = !!(flags & DMA_PREP_INTERRUPT); | ||
579 | hw->ctl_f.compl_write = 1; | ||
580 | /* we leave the channel locked to ensure in order submission */ | ||
581 | |||
582 | return &desc->txd; | ||
583 | } | ||
584 | |||
585 | /** | ||
586 | * ioat2_free_chan_resources - release all the descriptors | ||
587 | * @chan: the channel to be cleaned | ||
588 | */ | ||
589 | static void ioat2_free_chan_resources(struct dma_chan *c) | ||
590 | { | ||
591 | struct ioat2_dma_chan *ioat = to_ioat2_chan(c); | ||
592 | struct ioat_chan_common *chan = &ioat->base; | ||
593 | struct ioatdma_device *ioatdma_device = chan->device; | ||
594 | struct ioat_ring_ent *desc; | ||
595 | const u16 total_descs = 1 << ioat->alloc_order; | ||
596 | int descs; | ||
597 | int i; | ||
598 | |||
599 | /* Before freeing channel resources first check | ||
600 | * if they have been previously allocated for this channel. | ||
601 | */ | ||
602 | if (!ioat->ring) | ||
603 | return; | ||
604 | |||
605 | tasklet_disable(&chan->cleanup_task); | ||
606 | ioat2_cleanup(ioat); | ||
607 | |||
608 | /* Delay 100ms after reset to allow internal DMA logic to quiesce | ||
609 | * before removing DMA descriptor resources. | ||
610 | */ | ||
611 | writeb(IOAT_CHANCMD_RESET, | ||
612 | chan->reg_base + IOAT_CHANCMD_OFFSET(chan->device->version)); | ||
613 | mdelay(100); | ||
614 | |||
615 | spin_lock_bh(&ioat->ring_lock); | ||
616 | descs = ioat2_ring_space(ioat); | ||
617 | for (i = 0; i < descs; i++) { | ||
618 | desc = ioat2_get_ring_ent(ioat, ioat->head + i); | ||
619 | ioat2_free_ring_ent(desc, c); | ||
620 | } | ||
621 | |||
622 | if (descs < total_descs) | ||
623 | dev_err(to_dev(chan), "Freeing %d in use descriptors!\n", | ||
624 | total_descs - descs); | ||
625 | |||
626 | for (i = 0; i < total_descs - descs; i++) { | ||
627 | desc = ioat2_get_ring_ent(ioat, ioat->tail + i); | ||
628 | ioat2_free_ring_ent(desc, c); | ||
629 | } | ||
630 | |||
631 | kfree(ioat->ring); | ||
632 | ioat->ring = NULL; | ||
633 | ioat->alloc_order = 0; | ||
634 | pci_pool_free(ioatdma_device->completion_pool, | ||
635 | chan->completion_virt, | ||
636 | chan->completion_addr); | ||
637 | spin_unlock_bh(&ioat->ring_lock); | ||
638 | |||
639 | chan->last_completion = 0; | ||
640 | chan->completion_addr = 0; | ||
641 | ioat->pending = 0; | ||
642 | ioat->dmacount = 0; | ||
643 | chan->watchdog_completion = 0; | ||
644 | chan->last_compl_desc_addr_hw = 0; | ||
645 | chan->watchdog_tcp_cookie = 0; | ||
646 | chan->watchdog_last_tcp_cookie = 0; | ||
647 | } | ||
648 | |||
649 | static enum dma_status | ||
650 | ioat2_is_complete(struct dma_chan *c, dma_cookie_t cookie, | ||
651 | dma_cookie_t *done, dma_cookie_t *used) | ||
652 | { | ||
653 | struct ioat2_dma_chan *ioat = to_ioat2_chan(c); | ||
654 | |||
655 | if (ioat_is_complete(c, cookie, done, used) == DMA_SUCCESS) | ||
656 | return DMA_SUCCESS; | ||
657 | |||
658 | ioat2_cleanup(ioat); | ||
659 | |||
660 | return ioat_is_complete(c, cookie, done, used); | ||
661 | } | ||
662 | |||
663 | int ioat2_dma_probe(struct ioatdma_device *device, int dca) | ||
664 | { | ||
665 | struct pci_dev *pdev = device->pdev; | ||
666 | struct dma_device *dma; | ||
667 | struct dma_chan *c; | ||
668 | struct ioat_chan_common *chan; | ||
669 | int err; | ||
670 | |||
671 | device->enumerate_channels = ioat2_enumerate_channels; | ||
672 | dma = &device->common; | ||
673 | dma->device_prep_dma_memcpy = ioat2_dma_prep_memcpy_lock; | ||
674 | dma->device_issue_pending = ioat2_issue_pending; | ||
675 | dma->device_alloc_chan_resources = ioat2_alloc_chan_resources; | ||
676 | dma->device_free_chan_resources = ioat2_free_chan_resources; | ||
677 | dma->device_is_tx_complete = ioat2_is_complete; | ||
678 | |||
679 | err = ioat_probe(device); | ||
680 | if (err) | ||
681 | return err; | ||
682 | ioat_set_tcp_copy_break(2048); | ||
683 | |||
684 | list_for_each_entry(c, &dma->channels, device_node) { | ||
685 | chan = to_chan_common(c); | ||
686 | writel(IOAT_DCACTRL_CMPL_WRITE_ENABLE | IOAT_DMA_DCA_ANY_CPU, | ||
687 | chan->reg_base + IOAT_DCACTRL_OFFSET); | ||
688 | } | ||
689 | |||
690 | err = ioat_register(device); | ||
691 | if (err) | ||
692 | return err; | ||
693 | if (dca) | ||
694 | device->dca = ioat2_dca_init(pdev, device->reg_base); | ||
695 | |||
696 | INIT_DELAYED_WORK(&device->work, ioat2_chan_watchdog); | ||
697 | schedule_delayed_work(&device->work, WATCHDOG_DELAY); | ||
698 | |||
699 | return err; | ||
700 | } | ||
701 | |||
702 | int ioat3_dma_probe(struct ioatdma_device *device, int dca) | ||
703 | { | ||
704 | struct pci_dev *pdev = device->pdev; | ||
705 | struct dma_device *dma; | ||
706 | struct dma_chan *c; | ||
707 | struct ioat_chan_common *chan; | ||
708 | int err; | ||
709 | u16 dev_id; | ||
710 | |||
711 | device->enumerate_channels = ioat2_enumerate_channels; | ||
712 | dma = &device->common; | ||
713 | dma->device_prep_dma_memcpy = ioat2_dma_prep_memcpy_lock; | ||
714 | dma->device_issue_pending = ioat2_issue_pending; | ||
715 | dma->device_alloc_chan_resources = ioat2_alloc_chan_resources; | ||
716 | dma->device_free_chan_resources = ioat2_free_chan_resources; | ||
717 | dma->device_is_tx_complete = ioat2_is_complete; | ||
718 | |||
719 | /* -= IOAT ver.3 workarounds =- */ | ||
720 | /* Write CHANERRMSK_INT with 3E07h to mask out the errors | ||
721 | * that can cause stability issues for IOAT ver.3 | ||
722 | */ | ||
723 | pci_write_config_dword(pdev, IOAT_PCI_CHANERRMASK_INT_OFFSET, 0x3e07); | ||
724 | |||
725 | /* Clear DMAUNCERRSTS Cfg-Reg Parity Error status bit | ||
726 | * (workaround for spurious config parity error after restart) | ||
727 | */ | ||
728 | pci_read_config_word(pdev, IOAT_PCI_DEVICE_ID_OFFSET, &dev_id); | ||
729 | if (dev_id == PCI_DEVICE_ID_INTEL_IOAT_TBG0) | ||
730 | pci_write_config_dword(pdev, IOAT_PCI_DMAUNCERRSTS_OFFSET, 0x10); | ||
731 | |||
732 | err = ioat_probe(device); | ||
733 | if (err) | ||
734 | return err; | ||
735 | ioat_set_tcp_copy_break(262144); | ||
736 | |||
737 | list_for_each_entry(c, &dma->channels, device_node) { | ||
738 | chan = to_chan_common(c); | ||
739 | writel(IOAT_DMA_DCA_ANY_CPU, | ||
740 | chan->reg_base + IOAT_DCACTRL_OFFSET); | ||
741 | } | ||
742 | |||
743 | err = ioat_register(device); | ||
744 | if (err) | ||
745 | return err; | ||
746 | if (dca) | ||
747 | device->dca = ioat3_dca_init(pdev, device->reg_base); | ||
748 | |||
749 | return err; | ||
750 | } | ||
diff --git a/drivers/dma/ioat/dma_v2.h b/drivers/dma/ioat/dma_v2.h new file mode 100644 index 000000000000..94a553eacdbd --- /dev/null +++ b/drivers/dma/ioat/dma_v2.h | |||
@@ -0,0 +1,131 @@ | |||
1 | /* | ||
2 | * Copyright(c) 2004 - 2009 Intel Corporation. All rights reserved. | ||
3 | * | ||
4 | * This program is free software; you can redistribute it and/or modify it | ||
5 | * under the terms of the GNU General Public License as published by the Free | ||
6 | * Software Foundation; either version 2 of the License, or (at your option) | ||
7 | * any later version. | ||
8 | * | ||
9 | * This program is distributed in the hope that 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., 59 | ||
16 | * Temple Place - Suite 330, Boston, MA 02111-1307, USA. | ||
17 | * | ||
18 | * The full GNU General Public License is included in this distribution in the | ||
19 | * file called COPYING. | ||
20 | */ | ||
21 | #ifndef IOATDMA_V2_H | ||
22 | #define IOATDMA_V2_H | ||
23 | |||
24 | #include <linux/dmaengine.h> | ||
25 | #include "dma.h" | ||
26 | #include "hw.h" | ||
27 | |||
28 | |||
29 | extern int ioat_pending_level; | ||
30 | |||
31 | /* | ||
32 | * workaround for IOAT ver.3.0 null descriptor issue | ||
33 | * (channel returns error when size is 0) | ||
34 | */ | ||
35 | #define NULL_DESC_BUFFER_SIZE 1 | ||
36 | |||
37 | #define IOAT_MAX_ORDER 16 | ||
38 | #define ioat_get_alloc_order() \ | ||
39 | (min(ioat_ring_alloc_order, IOAT_MAX_ORDER)) | ||
40 | |||
41 | /* struct ioat2_dma_chan - ioat v2 / v3 channel attributes | ||
42 | * @base: common ioat channel parameters | ||
43 | * @xfercap_log; log2 of channel max transfer length (for fast division) | ||
44 | * @head: allocated index | ||
45 | * @issued: hardware notification point | ||
46 | * @tail: cleanup index | ||
47 | * @pending: lock free indicator for issued != head | ||
48 | * @dmacount: identical to 'head' except for occasionally resetting to zero | ||
49 | * @alloc_order: log2 of the number of allocated descriptors | ||
50 | * @ring: software ring buffer implementation of hardware ring | ||
51 | * @ring_lock: protects ring attributes | ||
52 | */ | ||
53 | struct ioat2_dma_chan { | ||
54 | struct ioat_chan_common base; | ||
55 | size_t xfercap_log; | ||
56 | u16 head; | ||
57 | u16 issued; | ||
58 | u16 tail; | ||
59 | u16 dmacount; | ||
60 | u16 alloc_order; | ||
61 | int pending; | ||
62 | struct ioat_ring_ent **ring; | ||
63 | spinlock_t ring_lock; | ||
64 | }; | ||
65 | |||
66 | static inline struct ioat2_dma_chan *to_ioat2_chan(struct dma_chan *c) | ||
67 | { | ||
68 | struct ioat_chan_common *chan = to_chan_common(c); | ||
69 | |||
70 | return container_of(chan, struct ioat2_dma_chan, base); | ||
71 | } | ||
72 | |||
73 | static inline u16 ioat2_ring_mask(struct ioat2_dma_chan *ioat) | ||
74 | { | ||
75 | return (1 << ioat->alloc_order) - 1; | ||
76 | } | ||
77 | |||
78 | /* count of descriptors in flight with the engine */ | ||
79 | static inline u16 ioat2_ring_active(struct ioat2_dma_chan *ioat) | ||
80 | { | ||
81 | return (ioat->head - ioat->tail) & ioat2_ring_mask(ioat); | ||
82 | } | ||
83 | |||
84 | /* count of descriptors pending submission to hardware */ | ||
85 | static inline u16 ioat2_ring_pending(struct ioat2_dma_chan *ioat) | ||
86 | { | ||
87 | return (ioat->head - ioat->issued) & ioat2_ring_mask(ioat); | ||
88 | } | ||
89 | |||
90 | static inline u16 ioat2_ring_space(struct ioat2_dma_chan *ioat) | ||
91 | { | ||
92 | u16 num_descs = ioat2_ring_mask(ioat) + 1; | ||
93 | u16 active = ioat2_ring_active(ioat); | ||
94 | |||
95 | BUG_ON(active > num_descs); | ||
96 | |||
97 | return num_descs - active; | ||
98 | } | ||
99 | |||
100 | /* assumes caller already checked space */ | ||
101 | static inline u16 ioat2_desc_alloc(struct ioat2_dma_chan *ioat, u16 len) | ||
102 | { | ||
103 | ioat->head += len; | ||
104 | return ioat->head - len; | ||
105 | } | ||
106 | |||
107 | static inline u16 ioat2_xferlen_to_descs(struct ioat2_dma_chan *ioat, size_t len) | ||
108 | { | ||
109 | u16 num_descs = len >> ioat->xfercap_log; | ||
110 | |||
111 | num_descs += !!(len & ((1 << ioat->xfercap_log) - 1)); | ||
112 | return num_descs; | ||
113 | } | ||
114 | |||
115 | struct ioat_ring_ent { | ||
116 | struct ioat_dma_descriptor *hw; | ||
117 | struct dma_async_tx_descriptor txd; | ||
118 | size_t len; | ||
119 | }; | ||
120 | |||
121 | static inline struct ioat_ring_ent * | ||
122 | ioat2_get_ring_ent(struct ioat2_dma_chan *ioat, u16 idx) | ||
123 | { | ||
124 | return ioat->ring[idx & ioat2_ring_mask(ioat)]; | ||
125 | } | ||
126 | |||
127 | int ioat2_dma_probe(struct ioatdma_device *dev, int dca); | ||
128 | int ioat3_dma_probe(struct ioatdma_device *dev, int dca); | ||
129 | struct dca_provider *ioat2_dca_init(struct pci_dev *pdev, void __iomem *iobase); | ||
130 | struct dca_provider *ioat3_dca_init(struct pci_dev *pdev, void __iomem *iobase); | ||
131 | #endif /* IOATDMA_V2_H */ | ||
diff --git a/drivers/dma/ioat/pci.c b/drivers/dma/ioat/pci.c index 55414d88ac1b..c4e432269252 100644 --- a/drivers/dma/ioat/pci.c +++ b/drivers/dma/ioat/pci.c | |||
@@ -31,6 +31,7 @@ | |||
31 | #include <linux/interrupt.h> | 31 | #include <linux/interrupt.h> |
32 | #include <linux/dca.h> | 32 | #include <linux/dca.h> |
33 | #include "dma.h" | 33 | #include "dma.h" |
34 | #include "dma_v2.h" | ||
34 | #include "registers.h" | 35 | #include "registers.h" |
35 | #include "hw.h" | 36 | #include "hw.h" |
36 | 37 | ||