1
2
3
4
5
6
7
8#include "dmaengine.h"
9
10#include <linux/circ_buf.h>
11#include <linux/dmaengine.h>
12#include <linux/kref.h>
13#include <linux/list.h>
14#include <linux/module.h>
15#include <linux/pci.h>
16
17MODULE_DESCRIPTION("PLX ExpressLane PEX PCI Switch DMA Engine");
18MODULE_VERSION("0.1");
19MODULE_LICENSE("GPL");
20MODULE_AUTHOR("Logan Gunthorpe");
21
22#define PLX_REG_DESC_RING_ADDR 0x214
23#define PLX_REG_DESC_RING_ADDR_HI 0x218
24#define PLX_REG_DESC_RING_NEXT_ADDR 0x21C
25#define PLX_REG_DESC_RING_COUNT 0x220
26#define PLX_REG_DESC_RING_LAST_ADDR 0x224
27#define PLX_REG_DESC_RING_LAST_SIZE 0x228
28#define PLX_REG_PREF_LIMIT 0x234
29#define PLX_REG_CTRL 0x238
30#define PLX_REG_CTRL2 0x23A
31#define PLX_REG_INTR_CTRL 0x23C
32#define PLX_REG_INTR_STATUS 0x23E
33
34#define PLX_REG_PREF_LIMIT_PREF_FOUR 8
35
36#define PLX_REG_CTRL_GRACEFUL_PAUSE BIT(0)
37#define PLX_REG_CTRL_ABORT BIT(1)
38#define PLX_REG_CTRL_WRITE_BACK_EN BIT(2)
39#define PLX_REG_CTRL_START BIT(3)
40#define PLX_REG_CTRL_RING_STOP_MODE BIT(4)
41#define PLX_REG_CTRL_DESC_MODE_BLOCK (0 << 5)
42#define PLX_REG_CTRL_DESC_MODE_ON_CHIP (1 << 5)
43#define PLX_REG_CTRL_DESC_MODE_OFF_CHIP (2 << 5)
44#define PLX_REG_CTRL_DESC_INVALID BIT(8)
45#define PLX_REG_CTRL_GRACEFUL_PAUSE_DONE BIT(9)
46#define PLX_REG_CTRL_ABORT_DONE BIT(10)
47#define PLX_REG_CTRL_IMM_PAUSE_DONE BIT(12)
48#define PLX_REG_CTRL_IN_PROGRESS BIT(30)
49
50#define PLX_REG_CTRL_RESET_VAL (PLX_REG_CTRL_DESC_INVALID | \
51 PLX_REG_CTRL_GRACEFUL_PAUSE_DONE | \
52 PLX_REG_CTRL_ABORT_DONE | \
53 PLX_REG_CTRL_IMM_PAUSE_DONE)
54
55#define PLX_REG_CTRL_START_VAL (PLX_REG_CTRL_WRITE_BACK_EN | \
56 PLX_REG_CTRL_DESC_MODE_OFF_CHIP | \
57 PLX_REG_CTRL_START | \
58 PLX_REG_CTRL_RESET_VAL)
59
60#define PLX_REG_CTRL2_MAX_TXFR_SIZE_64B 0
61#define PLX_REG_CTRL2_MAX_TXFR_SIZE_128B 1
62#define PLX_REG_CTRL2_MAX_TXFR_SIZE_256B 2
63#define PLX_REG_CTRL2_MAX_TXFR_SIZE_512B 3
64#define PLX_REG_CTRL2_MAX_TXFR_SIZE_1KB 4
65#define PLX_REG_CTRL2_MAX_TXFR_SIZE_2KB 5
66#define PLX_REG_CTRL2_MAX_TXFR_SIZE_4B 7
67
68#define PLX_REG_INTR_CRTL_ERROR_EN BIT(0)
69#define PLX_REG_INTR_CRTL_INV_DESC_EN BIT(1)
70#define PLX_REG_INTR_CRTL_ABORT_DONE_EN BIT(3)
71#define PLX_REG_INTR_CRTL_PAUSE_DONE_EN BIT(4)
72#define PLX_REG_INTR_CRTL_IMM_PAUSE_DONE_EN BIT(5)
73
74#define PLX_REG_INTR_STATUS_ERROR BIT(0)
75#define PLX_REG_INTR_STATUS_INV_DESC BIT(1)
76#define PLX_REG_INTR_STATUS_DESC_DONE BIT(2)
77#define PLX_REG_INTR_CRTL_ABORT_DONE BIT(3)
78
79struct plx_dma_hw_std_desc {
80 __le32 flags_and_size;
81 __le16 dst_addr_hi;
82 __le16 src_addr_hi;
83 __le32 dst_addr_lo;
84 __le32 src_addr_lo;
85};
86
87#define PLX_DESC_SIZE_MASK 0x7ffffff
88#define PLX_DESC_FLAG_VALID BIT(31)
89#define PLX_DESC_FLAG_INT_WHEN_DONE BIT(30)
90
91#define PLX_DESC_WB_SUCCESS BIT(30)
92#define PLX_DESC_WB_RD_FAIL BIT(29)
93#define PLX_DESC_WB_WR_FAIL BIT(28)
94
95#define PLX_DMA_RING_COUNT 2048
96
97struct plx_dma_desc {
98 struct dma_async_tx_descriptor txd;
99 struct plx_dma_hw_std_desc *hw;
100 u32 orig_size;
101};
102
103struct plx_dma_dev {
104 struct dma_device dma_dev;
105 struct dma_chan dma_chan;
106 struct pci_dev __rcu *pdev;
107 void __iomem *bar;
108 struct tasklet_struct desc_task;
109
110 spinlock_t ring_lock;
111 bool ring_active;
112 int head;
113 int tail;
114 struct plx_dma_hw_std_desc *hw_ring;
115 dma_addr_t hw_ring_dma;
116 struct plx_dma_desc **desc_ring;
117};
118
119static struct plx_dma_dev *chan_to_plx_dma_dev(struct dma_chan *c)
120{
121 return container_of(c, struct plx_dma_dev, dma_chan);
122}
123
124static struct plx_dma_desc *to_plx_desc(struct dma_async_tx_descriptor *txd)
125{
126 return container_of(txd, struct plx_dma_desc, txd);
127}
128
129static struct plx_dma_desc *plx_dma_get_desc(struct plx_dma_dev *plxdev, int i)
130{
131 return plxdev->desc_ring[i & (PLX_DMA_RING_COUNT - 1)];
132}
133
134static void plx_dma_process_desc(struct plx_dma_dev *plxdev)
135{
136 struct dmaengine_result res;
137 struct plx_dma_desc *desc;
138 u32 flags;
139
140 spin_lock_bh(&plxdev->ring_lock);
141
142 while (plxdev->tail != plxdev->head) {
143 desc = plx_dma_get_desc(plxdev, plxdev->tail);
144
145 flags = le32_to_cpu(READ_ONCE(desc->hw->flags_and_size));
146
147 if (flags & PLX_DESC_FLAG_VALID)
148 break;
149
150 res.residue = desc->orig_size - (flags & PLX_DESC_SIZE_MASK);
151
152 if (flags & PLX_DESC_WB_SUCCESS)
153 res.result = DMA_TRANS_NOERROR;
154 else if (flags & PLX_DESC_WB_WR_FAIL)
155 res.result = DMA_TRANS_WRITE_FAILED;
156 else
157 res.result = DMA_TRANS_READ_FAILED;
158
159 dma_cookie_complete(&desc->txd);
160 dma_descriptor_unmap(&desc->txd);
161 dmaengine_desc_get_callback_invoke(&desc->txd, &res);
162 desc->txd.callback = NULL;
163 desc->txd.callback_result = NULL;
164
165 plxdev->tail++;
166 }
167
168 spin_unlock_bh(&plxdev->ring_lock);
169}
170
171static void plx_dma_abort_desc(struct plx_dma_dev *plxdev)
172{
173 struct dmaengine_result res;
174 struct plx_dma_desc *desc;
175
176 plx_dma_process_desc(plxdev);
177
178 spin_lock_bh(&plxdev->ring_lock);
179
180 while (plxdev->tail != plxdev->head) {
181 desc = plx_dma_get_desc(plxdev, plxdev->tail);
182
183 res.residue = desc->orig_size;
184 res.result = DMA_TRANS_ABORTED;
185
186 dma_cookie_complete(&desc->txd);
187 dma_descriptor_unmap(&desc->txd);
188 dmaengine_desc_get_callback_invoke(&desc->txd, &res);
189 desc->txd.callback = NULL;
190 desc->txd.callback_result = NULL;
191
192 plxdev->tail++;
193 }
194
195 spin_unlock_bh(&plxdev->ring_lock);
196}
197
198static void __plx_dma_stop(struct plx_dma_dev *plxdev)
199{
200 unsigned long timeout = jiffies + msecs_to_jiffies(1000);
201 u32 val;
202
203 val = readl(plxdev->bar + PLX_REG_CTRL);
204 if (!(val & ~PLX_REG_CTRL_GRACEFUL_PAUSE))
205 return;
206
207 writel(PLX_REG_CTRL_RESET_VAL | PLX_REG_CTRL_GRACEFUL_PAUSE,
208 plxdev->bar + PLX_REG_CTRL);
209
210 while (!time_after(jiffies, timeout)) {
211 val = readl(plxdev->bar + PLX_REG_CTRL);
212 if (val & PLX_REG_CTRL_GRACEFUL_PAUSE_DONE)
213 break;
214
215 cpu_relax();
216 }
217
218 if (!(val & PLX_REG_CTRL_GRACEFUL_PAUSE_DONE))
219 dev_err(plxdev->dma_dev.dev,
220 "Timeout waiting for graceful pause!\n");
221
222 writel(PLX_REG_CTRL_RESET_VAL | PLX_REG_CTRL_GRACEFUL_PAUSE,
223 plxdev->bar + PLX_REG_CTRL);
224
225 writel(0, plxdev->bar + PLX_REG_DESC_RING_COUNT);
226 writel(0, plxdev->bar + PLX_REG_DESC_RING_ADDR);
227 writel(0, plxdev->bar + PLX_REG_DESC_RING_ADDR_HI);
228 writel(0, plxdev->bar + PLX_REG_DESC_RING_NEXT_ADDR);
229}
230
231static void plx_dma_stop(struct plx_dma_dev *plxdev)
232{
233 rcu_read_lock();
234 if (!rcu_dereference(plxdev->pdev)) {
235 rcu_read_unlock();
236 return;
237 }
238
239 __plx_dma_stop(plxdev);
240
241 rcu_read_unlock();
242}
243
244static void plx_dma_desc_task(struct tasklet_struct *t)
245{
246 struct plx_dma_dev *plxdev = from_tasklet(plxdev, t, desc_task);
247
248 plx_dma_process_desc(plxdev);
249}
250
251static struct dma_async_tx_descriptor *plx_dma_prep_memcpy(struct dma_chan *c,
252 dma_addr_t dma_dst, dma_addr_t dma_src, size_t len,
253 unsigned long flags)
254 __acquires(plxdev->ring_lock)
255{
256 struct plx_dma_dev *plxdev = chan_to_plx_dma_dev(c);
257 struct plx_dma_desc *plxdesc;
258
259 spin_lock_bh(&plxdev->ring_lock);
260 if (!plxdev->ring_active)
261 goto err_unlock;
262
263 if (!CIRC_SPACE(plxdev->head, plxdev->tail, PLX_DMA_RING_COUNT))
264 goto err_unlock;
265
266 if (len > PLX_DESC_SIZE_MASK)
267 goto err_unlock;
268
269 plxdesc = plx_dma_get_desc(plxdev, plxdev->head);
270 plxdev->head++;
271
272 plxdesc->hw->dst_addr_lo = cpu_to_le32(lower_32_bits(dma_dst));
273 plxdesc->hw->dst_addr_hi = cpu_to_le16(upper_32_bits(dma_dst));
274 plxdesc->hw->src_addr_lo = cpu_to_le32(lower_32_bits(dma_src));
275 plxdesc->hw->src_addr_hi = cpu_to_le16(upper_32_bits(dma_src));
276
277 plxdesc->orig_size = len;
278
279 if (flags & DMA_PREP_INTERRUPT)
280 len |= PLX_DESC_FLAG_INT_WHEN_DONE;
281
282 plxdesc->hw->flags_and_size = cpu_to_le32(len);
283 plxdesc->txd.flags = flags;
284
285
286
287 return &plxdesc->txd;
288
289err_unlock:
290
291
292
293
294 __acquire(plxdev->ring_lock);
295
296 spin_unlock_bh(&plxdev->ring_lock);
297 return NULL;
298}
299
300static dma_cookie_t plx_dma_tx_submit(struct dma_async_tx_descriptor *desc)
301 __releases(plxdev->ring_lock)
302{
303 struct plx_dma_dev *plxdev = chan_to_plx_dma_dev(desc->chan);
304 struct plx_dma_desc *plxdesc = to_plx_desc(desc);
305 dma_cookie_t cookie;
306
307 cookie = dma_cookie_assign(desc);
308
309
310
311
312
313 wmb();
314
315 plxdesc->hw->flags_and_size |= cpu_to_le32(PLX_DESC_FLAG_VALID);
316
317 spin_unlock_bh(&plxdev->ring_lock);
318
319 return cookie;
320}
321
322static enum dma_status plx_dma_tx_status(struct dma_chan *chan,
323 dma_cookie_t cookie, struct dma_tx_state *txstate)
324{
325 struct plx_dma_dev *plxdev = chan_to_plx_dma_dev(chan);
326 enum dma_status ret;
327
328 ret = dma_cookie_status(chan, cookie, txstate);
329 if (ret == DMA_COMPLETE)
330 return ret;
331
332 plx_dma_process_desc(plxdev);
333
334 return dma_cookie_status(chan, cookie, txstate);
335}
336
337static void plx_dma_issue_pending(struct dma_chan *chan)
338{
339 struct plx_dma_dev *plxdev = chan_to_plx_dma_dev(chan);
340
341 rcu_read_lock();
342 if (!rcu_dereference(plxdev->pdev)) {
343 rcu_read_unlock();
344 return;
345 }
346
347
348
349
350
351 wmb();
352
353 writew(PLX_REG_CTRL_START_VAL, plxdev->bar + PLX_REG_CTRL);
354
355 rcu_read_unlock();
356}
357
358static irqreturn_t plx_dma_isr(int irq, void *devid)
359{
360 struct plx_dma_dev *plxdev = devid;
361 u32 status;
362
363 status = readw(plxdev->bar + PLX_REG_INTR_STATUS);
364
365 if (!status)
366 return IRQ_NONE;
367
368 if (status & PLX_REG_INTR_STATUS_DESC_DONE && plxdev->ring_active)
369 tasklet_schedule(&plxdev->desc_task);
370
371 writew(status, plxdev->bar + PLX_REG_INTR_STATUS);
372
373 return IRQ_HANDLED;
374}
375
376static int plx_dma_alloc_desc(struct plx_dma_dev *plxdev)
377{
378 struct plx_dma_desc *desc;
379 int i;
380
381 plxdev->desc_ring = kcalloc(PLX_DMA_RING_COUNT,
382 sizeof(*plxdev->desc_ring), GFP_KERNEL);
383 if (!plxdev->desc_ring)
384 return -ENOMEM;
385
386 for (i = 0; i < PLX_DMA_RING_COUNT; i++) {
387 desc = kzalloc(sizeof(*desc), GFP_KERNEL);
388 if (!desc)
389 goto free_and_exit;
390
391 dma_async_tx_descriptor_init(&desc->txd, &plxdev->dma_chan);
392 desc->txd.tx_submit = plx_dma_tx_submit;
393 desc->hw = &plxdev->hw_ring[i];
394
395 plxdev->desc_ring[i] = desc;
396 }
397
398 return 0;
399
400free_and_exit:
401 for (i = 0; i < PLX_DMA_RING_COUNT; i++)
402 kfree(plxdev->desc_ring[i]);
403 kfree(plxdev->desc_ring);
404 return -ENOMEM;
405}
406
407static int plx_dma_alloc_chan_resources(struct dma_chan *chan)
408{
409 struct plx_dma_dev *plxdev = chan_to_plx_dma_dev(chan);
410 size_t ring_sz = PLX_DMA_RING_COUNT * sizeof(*plxdev->hw_ring);
411 int rc;
412
413 plxdev->head = plxdev->tail = 0;
414 plxdev->hw_ring = dma_alloc_coherent(plxdev->dma_dev.dev, ring_sz,
415 &plxdev->hw_ring_dma, GFP_KERNEL);
416 if (!plxdev->hw_ring)
417 return -ENOMEM;
418
419 rc = plx_dma_alloc_desc(plxdev);
420 if (rc)
421 goto out_free_hw_ring;
422
423 rcu_read_lock();
424 if (!rcu_dereference(plxdev->pdev)) {
425 rcu_read_unlock();
426 rc = -ENODEV;
427 goto out_free_hw_ring;
428 }
429
430 writel(PLX_REG_CTRL_RESET_VAL, plxdev->bar + PLX_REG_CTRL);
431 writel(lower_32_bits(plxdev->hw_ring_dma),
432 plxdev->bar + PLX_REG_DESC_RING_ADDR);
433 writel(upper_32_bits(plxdev->hw_ring_dma),
434 plxdev->bar + PLX_REG_DESC_RING_ADDR_HI);
435 writel(lower_32_bits(plxdev->hw_ring_dma),
436 plxdev->bar + PLX_REG_DESC_RING_NEXT_ADDR);
437 writel(PLX_DMA_RING_COUNT, plxdev->bar + PLX_REG_DESC_RING_COUNT);
438 writel(PLX_REG_PREF_LIMIT_PREF_FOUR, plxdev->bar + PLX_REG_PREF_LIMIT);
439
440 plxdev->ring_active = true;
441
442 rcu_read_unlock();
443
444 return PLX_DMA_RING_COUNT;
445
446out_free_hw_ring:
447 dma_free_coherent(plxdev->dma_dev.dev, ring_sz, plxdev->hw_ring,
448 plxdev->hw_ring_dma);
449 return rc;
450}
451
452static void plx_dma_free_chan_resources(struct dma_chan *chan)
453{
454 struct plx_dma_dev *plxdev = chan_to_plx_dma_dev(chan);
455 size_t ring_sz = PLX_DMA_RING_COUNT * sizeof(*plxdev->hw_ring);
456 struct pci_dev *pdev;
457 int irq = -1;
458 int i;
459
460 spin_lock_bh(&plxdev->ring_lock);
461 plxdev->ring_active = false;
462 spin_unlock_bh(&plxdev->ring_lock);
463
464 plx_dma_stop(plxdev);
465
466 rcu_read_lock();
467 pdev = rcu_dereference(plxdev->pdev);
468 if (pdev)
469 irq = pci_irq_vector(pdev, 0);
470 rcu_read_unlock();
471
472 if (irq > 0)
473 synchronize_irq(irq);
474
475 tasklet_kill(&plxdev->desc_task);
476
477 plx_dma_abort_desc(plxdev);
478
479 for (i = 0; i < PLX_DMA_RING_COUNT; i++)
480 kfree(plxdev->desc_ring[i]);
481
482 kfree(plxdev->desc_ring);
483 dma_free_coherent(plxdev->dma_dev.dev, ring_sz, plxdev->hw_ring,
484 plxdev->hw_ring_dma);
485
486}
487
488static void plx_dma_release(struct dma_device *dma_dev)
489{
490 struct plx_dma_dev *plxdev =
491 container_of(dma_dev, struct plx_dma_dev, dma_dev);
492
493 put_device(dma_dev->dev);
494 kfree(plxdev);
495}
496
497static int plx_dma_create(struct pci_dev *pdev)
498{
499 struct plx_dma_dev *plxdev;
500 struct dma_device *dma;
501 struct dma_chan *chan;
502 int rc;
503
504 plxdev = kzalloc(sizeof(*plxdev), GFP_KERNEL);
505 if (!plxdev)
506 return -ENOMEM;
507
508 rc = request_irq(pci_irq_vector(pdev, 0), plx_dma_isr, 0,
509 KBUILD_MODNAME, plxdev);
510 if (rc)
511 goto free_plx;
512
513 spin_lock_init(&plxdev->ring_lock);
514 tasklet_setup(&plxdev->desc_task, plx_dma_desc_task);
515
516 RCU_INIT_POINTER(plxdev->pdev, pdev);
517 plxdev->bar = pcim_iomap_table(pdev)[0];
518
519 dma = &plxdev->dma_dev;
520 dma->chancnt = 1;
521 INIT_LIST_HEAD(&dma->channels);
522 dma_cap_set(DMA_MEMCPY, dma->cap_mask);
523 dma->copy_align = DMAENGINE_ALIGN_1_BYTE;
524 dma->dev = get_device(&pdev->dev);
525
526 dma->device_alloc_chan_resources = plx_dma_alloc_chan_resources;
527 dma->device_free_chan_resources = plx_dma_free_chan_resources;
528 dma->device_prep_dma_memcpy = plx_dma_prep_memcpy;
529 dma->device_issue_pending = plx_dma_issue_pending;
530 dma->device_tx_status = plx_dma_tx_status;
531 dma->device_release = plx_dma_release;
532
533 chan = &plxdev->dma_chan;
534 chan->device = dma;
535 dma_cookie_init(chan);
536 list_add_tail(&chan->device_node, &dma->channels);
537
538 rc = dma_async_device_register(dma);
539 if (rc) {
540 pci_err(pdev, "Failed to register dma device: %d\n", rc);
541 goto put_device;
542 }
543
544 pci_set_drvdata(pdev, plxdev);
545
546 return 0;
547
548put_device:
549 put_device(&pdev->dev);
550 free_irq(pci_irq_vector(pdev, 0), plxdev);
551free_plx:
552 kfree(plxdev);
553
554 return rc;
555}
556
557static int plx_dma_probe(struct pci_dev *pdev,
558 const struct pci_device_id *id)
559{
560 int rc;
561
562 rc = pcim_enable_device(pdev);
563 if (rc)
564 return rc;
565
566 rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(48));
567 if (rc)
568 rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
569 if (rc)
570 return rc;
571
572 rc = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(48));
573 if (rc)
574 rc = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
575 if (rc)
576 return rc;
577
578 rc = pcim_iomap_regions(pdev, 1, KBUILD_MODNAME);
579 if (rc)
580 return rc;
581
582 rc = pci_alloc_irq_vectors(pdev, 1, 1, PCI_IRQ_ALL_TYPES);
583 if (rc <= 0)
584 return rc;
585
586 pci_set_master(pdev);
587
588 rc = plx_dma_create(pdev);
589 if (rc)
590 goto err_free_irq_vectors;
591
592 pci_info(pdev, "PLX DMA Channel Registered\n");
593
594 return 0;
595
596err_free_irq_vectors:
597 pci_free_irq_vectors(pdev);
598 return rc;
599}
600
601static void plx_dma_remove(struct pci_dev *pdev)
602{
603 struct plx_dma_dev *plxdev = pci_get_drvdata(pdev);
604
605 free_irq(pci_irq_vector(pdev, 0), plxdev);
606
607 rcu_assign_pointer(plxdev->pdev, NULL);
608 synchronize_rcu();
609
610 spin_lock_bh(&plxdev->ring_lock);
611 plxdev->ring_active = false;
612 spin_unlock_bh(&plxdev->ring_lock);
613
614 __plx_dma_stop(plxdev);
615 plx_dma_abort_desc(plxdev);
616
617 plxdev->bar = NULL;
618 dma_async_device_unregister(&plxdev->dma_dev);
619
620 pci_free_irq_vectors(pdev);
621}
622
623static const struct pci_device_id plx_dma_pci_tbl[] = {
624 {
625 .vendor = PCI_VENDOR_ID_PLX,
626 .device = 0x87D0,
627 .subvendor = PCI_ANY_ID,
628 .subdevice = PCI_ANY_ID,
629 .class = PCI_CLASS_SYSTEM_OTHER << 8,
630 .class_mask = 0xFFFFFFFF,
631 },
632 {0}
633};
634MODULE_DEVICE_TABLE(pci, plx_dma_pci_tbl);
635
636static struct pci_driver plx_dma_pci_driver = {
637 .name = KBUILD_MODNAME,
638 .id_table = plx_dma_pci_tbl,
639 .probe = plx_dma_probe,
640 .remove = plx_dma_remove,
641};
642module_pci_driver(plx_dma_pci_driver);
643
