1 // SPDX-License-Identifier: GPL-2.0
3 * Microsemi Switchtec(tm) PCIe Management Driver
4 * Copyright (c) 2019, Logan Gunthorpe <logang@deltatee.com>
5 * Copyright (c) 2019, GigaIO Networks, Inc
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>
17 MODULE_DESCRIPTION("PLX ExpressLane PEX PCI Switch DMA Engine");
18 MODULE_VERSION("0.1");
19 MODULE_LICENSE("GPL");
20 MODULE_AUTHOR("Logan Gunthorpe");
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
34 #define PLX_REG_PREF_LIMIT_PREF_FOUR 8
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)
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)
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)
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
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)
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)
79 struct plx_dma_hw_std_desc {
80 __le32 flags_and_size;
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)
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)
95 #define PLX_DMA_RING_COUNT 2048
98 struct dma_async_tx_descriptor txd;
99 struct plx_dma_hw_std_desc *hw;
104 struct dma_device dma_dev;
105 struct dma_chan dma_chan;
106 struct pci_dev __rcu *pdev;
108 struct tasklet_struct desc_task;
110 spinlock_t ring_lock;
114 struct plx_dma_hw_std_desc *hw_ring;
115 dma_addr_t hw_ring_dma;
116 struct plx_dma_desc **desc_ring;
119 static struct plx_dma_dev *chan_to_plx_dma_dev(struct dma_chan *c)
121 return container_of(c, struct plx_dma_dev, dma_chan);
124 static struct plx_dma_desc *to_plx_desc(struct dma_async_tx_descriptor *txd)
126 return container_of(txd, struct plx_dma_desc, txd);
129 static struct plx_dma_desc *plx_dma_get_desc(struct plx_dma_dev *plxdev, int i)
131 return plxdev->desc_ring[i & (PLX_DMA_RING_COUNT - 1)];
134 static void plx_dma_process_desc(struct plx_dma_dev *plxdev)
136 struct dmaengine_result res;
137 struct plx_dma_desc *desc;
140 spin_lock(&plxdev->ring_lock);
142 while (plxdev->tail != plxdev->head) {
143 desc = plx_dma_get_desc(plxdev, plxdev->tail);
145 flags = le32_to_cpu(READ_ONCE(desc->hw->flags_and_size));
147 if (flags & PLX_DESC_FLAG_VALID)
150 res.residue = desc->orig_size - (flags & PLX_DESC_SIZE_MASK);
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;
157 res.result = DMA_TRANS_READ_FAILED;
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;
168 spin_unlock(&plxdev->ring_lock);
171 static void plx_dma_abort_desc(struct plx_dma_dev *plxdev)
173 struct dmaengine_result res;
174 struct plx_dma_desc *desc;
176 plx_dma_process_desc(plxdev);
178 spin_lock_bh(&plxdev->ring_lock);
180 while (plxdev->tail != plxdev->head) {
181 desc = plx_dma_get_desc(plxdev, plxdev->tail);
183 res.residue = desc->orig_size;
184 res.result = DMA_TRANS_ABORTED;
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;
195 spin_unlock_bh(&plxdev->ring_lock);
198 static void __plx_dma_stop(struct plx_dma_dev *plxdev)
200 unsigned long timeout = jiffies + msecs_to_jiffies(1000);
203 val = readl(plxdev->bar + PLX_REG_CTRL);
204 if (!(val & ~PLX_REG_CTRL_GRACEFUL_PAUSE))
207 writel(PLX_REG_CTRL_RESET_VAL | PLX_REG_CTRL_GRACEFUL_PAUSE,
208 plxdev->bar + PLX_REG_CTRL);
210 while (!time_after(jiffies, timeout)) {
211 val = readl(plxdev->bar + PLX_REG_CTRL);
212 if (val & PLX_REG_CTRL_GRACEFUL_PAUSE_DONE)
218 if (!(val & PLX_REG_CTRL_GRACEFUL_PAUSE_DONE))
219 dev_err(plxdev->dma_dev.dev,
220 "Timeout waiting for graceful pause!\n");
222 writel(PLX_REG_CTRL_RESET_VAL | PLX_REG_CTRL_GRACEFUL_PAUSE,
223 plxdev->bar + PLX_REG_CTRL);
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);
231 static void plx_dma_stop(struct plx_dma_dev *plxdev)
234 if (!rcu_dereference(plxdev->pdev)) {
239 __plx_dma_stop(plxdev);
244 static void plx_dma_desc_task(struct tasklet_struct *t)
246 struct plx_dma_dev *plxdev = from_tasklet(plxdev, t, desc_task);
248 plx_dma_process_desc(plxdev);
251 static 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,
254 __acquires(plxdev->ring_lock)
256 struct plx_dma_dev *plxdev = chan_to_plx_dma_dev(c);
257 struct plx_dma_desc *plxdesc;
259 spin_lock_bh(&plxdev->ring_lock);
260 if (!plxdev->ring_active)
263 if (!CIRC_SPACE(plxdev->head, plxdev->tail, PLX_DMA_RING_COUNT))
266 if (len > PLX_DESC_SIZE_MASK)
269 plxdesc = plx_dma_get_desc(plxdev, plxdev->head);
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));
277 plxdesc->orig_size = len;
279 if (flags & DMA_PREP_INTERRUPT)
280 len |= PLX_DESC_FLAG_INT_WHEN_DONE;
282 plxdesc->hw->flags_and_size = cpu_to_le32(len);
283 plxdesc->txd.flags = flags;
285 /* return with the lock held, it will be released in tx_submit */
287 return &plxdesc->txd;
291 * Keep sparse happy by restoring an even lock count on
294 __acquire(plxdev->ring_lock);
296 spin_unlock_bh(&plxdev->ring_lock);
300 static dma_cookie_t plx_dma_tx_submit(struct dma_async_tx_descriptor *desc)
301 __releases(plxdev->ring_lock)
303 struct plx_dma_dev *plxdev = chan_to_plx_dma_dev(desc->chan);
304 struct plx_dma_desc *plxdesc = to_plx_desc(desc);
307 cookie = dma_cookie_assign(desc);
310 * Ensure the descriptor updates are visible to the dma device
311 * before setting the valid bit.
315 plxdesc->hw->flags_and_size |= cpu_to_le32(PLX_DESC_FLAG_VALID);
317 spin_unlock_bh(&plxdev->ring_lock);
322 static enum dma_status plx_dma_tx_status(struct dma_chan *chan,
323 dma_cookie_t cookie, struct dma_tx_state *txstate)
325 struct plx_dma_dev *plxdev = chan_to_plx_dma_dev(chan);
328 ret = dma_cookie_status(chan, cookie, txstate);
329 if (ret == DMA_COMPLETE)
332 plx_dma_process_desc(plxdev);
334 return dma_cookie_status(chan, cookie, txstate);
337 static void plx_dma_issue_pending(struct dma_chan *chan)
339 struct plx_dma_dev *plxdev = chan_to_plx_dma_dev(chan);
342 if (!rcu_dereference(plxdev->pdev)) {
348 * Ensure the valid bits are visible before starting the
353 writew(PLX_REG_CTRL_START_VAL, plxdev->bar + PLX_REG_CTRL);
358 static irqreturn_t plx_dma_isr(int irq, void *devid)
360 struct plx_dma_dev *plxdev = devid;
363 status = readw(plxdev->bar + PLX_REG_INTR_STATUS);
368 if (status & PLX_REG_INTR_STATUS_DESC_DONE && plxdev->ring_active)
369 tasklet_schedule(&plxdev->desc_task);
371 writew(status, plxdev->bar + PLX_REG_INTR_STATUS);
376 static int plx_dma_alloc_desc(struct plx_dma_dev *plxdev)
378 struct plx_dma_desc *desc;
381 plxdev->desc_ring = kcalloc(PLX_DMA_RING_COUNT,
382 sizeof(*plxdev->desc_ring), GFP_KERNEL);
383 if (!plxdev->desc_ring)
386 for (i = 0; i < PLX_DMA_RING_COUNT; i++) {
387 desc = kzalloc(sizeof(*desc), GFP_KERNEL);
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];
395 plxdev->desc_ring[i] = desc;
401 for (i = 0; i < PLX_DMA_RING_COUNT; i++)
402 kfree(plxdev->desc_ring[i]);
403 kfree(plxdev->desc_ring);
407 static int plx_dma_alloc_chan_resources(struct dma_chan *chan)
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);
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)
419 rc = plx_dma_alloc_desc(plxdev);
421 goto out_free_hw_ring;
424 if (!rcu_dereference(plxdev->pdev)) {
427 goto out_free_hw_ring;
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);
440 plxdev->ring_active = true;
444 return PLX_DMA_RING_COUNT;
447 dma_free_coherent(plxdev->dma_dev.dev, ring_sz, plxdev->hw_ring,
448 plxdev->hw_ring_dma);
452 static void plx_dma_free_chan_resources(struct dma_chan *chan)
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;
460 spin_lock_bh(&plxdev->ring_lock);
461 plxdev->ring_active = false;
462 spin_unlock_bh(&plxdev->ring_lock);
464 plx_dma_stop(plxdev);
467 pdev = rcu_dereference(plxdev->pdev);
469 irq = pci_irq_vector(pdev, 0);
473 synchronize_irq(irq);
475 tasklet_kill(&plxdev->desc_task);
477 plx_dma_abort_desc(plxdev);
479 for (i = 0; i < PLX_DMA_RING_COUNT; i++)
480 kfree(plxdev->desc_ring[i]);
482 kfree(plxdev->desc_ring);
483 dma_free_coherent(plxdev->dma_dev.dev, ring_sz, plxdev->hw_ring,
484 plxdev->hw_ring_dma);
488 static void plx_dma_release(struct dma_device *dma_dev)
490 struct plx_dma_dev *plxdev =
491 container_of(dma_dev, struct plx_dma_dev, dma_dev);
493 put_device(dma_dev->dev);
497 static int plx_dma_create(struct pci_dev *pdev)
499 struct plx_dma_dev *plxdev;
500 struct dma_device *dma;
501 struct dma_chan *chan;
504 plxdev = kzalloc(sizeof(*plxdev), GFP_KERNEL);
508 rc = request_irq(pci_irq_vector(pdev, 0), plx_dma_isr, 0,
509 KBUILD_MODNAME, plxdev);
513 spin_lock_init(&plxdev->ring_lock);
514 tasklet_setup(&plxdev->desc_task, plx_dma_desc_task);
516 RCU_INIT_POINTER(plxdev->pdev, pdev);
517 plxdev->bar = pcim_iomap_table(pdev)[0];
519 dma = &plxdev->dma_dev;
520 INIT_LIST_HEAD(&dma->channels);
521 dma_cap_set(DMA_MEMCPY, dma->cap_mask);
522 dma->copy_align = DMAENGINE_ALIGN_1_BYTE;
523 dma->dev = get_device(&pdev->dev);
525 dma->device_alloc_chan_resources = plx_dma_alloc_chan_resources;
526 dma->device_free_chan_resources = plx_dma_free_chan_resources;
527 dma->device_prep_dma_memcpy = plx_dma_prep_memcpy;
528 dma->device_issue_pending = plx_dma_issue_pending;
529 dma->device_tx_status = plx_dma_tx_status;
530 dma->device_release = plx_dma_release;
532 chan = &plxdev->dma_chan;
534 dma_cookie_init(chan);
535 list_add_tail(&chan->device_node, &dma->channels);
537 rc = dma_async_device_register(dma);
539 pci_err(pdev, "Failed to register dma device: %d\n", rc);
543 pci_set_drvdata(pdev, plxdev);
548 put_device(&pdev->dev);
549 free_irq(pci_irq_vector(pdev, 0), plxdev);
556 static int plx_dma_probe(struct pci_dev *pdev,
557 const struct pci_device_id *id)
561 rc = pcim_enable_device(pdev);
565 rc = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(48));
567 rc = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
571 rc = pcim_iomap_regions(pdev, 1, KBUILD_MODNAME);
575 rc = pci_alloc_irq_vectors(pdev, 1, 1, PCI_IRQ_ALL_TYPES);
579 pci_set_master(pdev);
581 rc = plx_dma_create(pdev);
583 goto err_free_irq_vectors;
585 pci_info(pdev, "PLX DMA Channel Registered\n");
589 err_free_irq_vectors:
590 pci_free_irq_vectors(pdev);
594 static void plx_dma_remove(struct pci_dev *pdev)
596 struct plx_dma_dev *plxdev = pci_get_drvdata(pdev);
598 free_irq(pci_irq_vector(pdev, 0), plxdev);
600 rcu_assign_pointer(plxdev->pdev, NULL);
603 spin_lock_bh(&plxdev->ring_lock);
604 plxdev->ring_active = false;
605 spin_unlock_bh(&plxdev->ring_lock);
607 __plx_dma_stop(plxdev);
608 plx_dma_abort_desc(plxdev);
611 dma_async_device_unregister(&plxdev->dma_dev);
613 pci_free_irq_vectors(pdev);
616 static const struct pci_device_id plx_dma_pci_tbl[] = {
618 .vendor = PCI_VENDOR_ID_PLX,
620 .subvendor = PCI_ANY_ID,
621 .subdevice = PCI_ANY_ID,
622 .class = PCI_CLASS_SYSTEM_OTHER << 8,
623 .class_mask = 0xFFFFFFFF,
627 MODULE_DEVICE_TABLE(pci, plx_dma_pci_tbl);
629 static struct pci_driver plx_dma_pci_driver = {
630 .name = KBUILD_MODNAME,
631 .id_table = plx_dma_pci_tbl,
632 .probe = plx_dma_probe,
633 .remove = plx_dma_remove,
635 module_pci_driver(plx_dma_pci_driver);