diff options
author | Dan Williams <dan.j.williams@intel.com> | 2009-08-26 16:01:44 -0400 |
---|---|---|
committer | Dan Williams <dan.j.williams@intel.com> | 2009-09-08 20:29:55 -0400 |
commit | 5cbafa65b92ee4f5b8ba915cddf94b91f186b989 (patch) | |
tree | f074c9dbcdedf05c5567a4e456a15120895363a6 /drivers/dma/ioat/dma.c | |
parent | dcbc853af6f0c056088e4df0794d9bf36184809e (diff) |
ioat2,3: convert to a true ring buffer
Replace the current linked list munged into a ring with a native ring
buffer implementation. The benefit of this approach is reduced overhead
as many parameters can be derived from ring position with simple pointer
comparisons and descriptor allocation/freeing becomes just a
manipulation of head/tail pointers.
It requires a contiguous allocation for the software descriptor
information.
Since this arrangement is significantly different from the ioat1 chain,
move ioat2,3 support into its own file and header. Common routines are
exported from driver/dma/ioat/dma.[ch].
Signed-off-by: Maciej Sosnowski <maciej.sosnowski@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Diffstat (limited to 'drivers/dma/ioat/dma.c')
-rw-r--r-- | drivers/dma/ioat/dma.c | 874 |
1 files changed, 194 insertions, 680 deletions
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 | |||