1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Low-Level PCI Express Support for the SH7786
4 *
5 * Copyright (C) 2009 - 2011 Paul Mundt
6 */
7 #define pr_fmt(fmt) "PCI: " fmt
8
9 #include <linux/pci.h>
10 #include <linux/init.h>
11 #include <linux/kernel.h>
12 #include <linux/io.h>
13 #include <linux/async.h>
14 #include <linux/delay.h>
15 #include <linux/dma-map-ops.h>
16 #include <linux/slab.h>
17 #include <linux/clk.h>
18 #include <linux/sh_clk.h>
19 #include <linux/sh_intc.h>
20 #include <cpu/sh7786.h>
21 #include "pcie-sh7786.h"
22 #include <linux/sizes.h>
23
24 struct sh7786_pcie_port {
25 struct pci_channel *hose;
26 struct clk *fclk, phy_clk;
27 unsigned int index;
28 int endpoint;
29 int link;
30 };
31
32 static struct sh7786_pcie_port *sh7786_pcie_ports;
33 static unsigned int nr_ports;
34 size_t memsize;
35 u64 memstart;
36
37 static struct sh7786_pcie_hwops {
38 int (*core_init)(void);
39 async_func_t port_init_hw;
40 } *sh7786_pcie_hwops;
41
42 static struct resource sh7786_pci0_resources[] = {
43 {
44 .name = "PCIe0 MEM 0",
45 .start = 0xfd000000,
46 .end = 0xfd000000 + SZ_8M - 1,
47 .flags = IORESOURCE_MEM,
48 }, {
49 .name = "PCIe0 MEM 1",
50 .start = 0xc0000000,
51 .end = 0xc0000000 + SZ_512M - 1,
52 .flags = IORESOURCE_MEM | IORESOURCE_MEM_32BIT,
53 }, {
54 .name = "PCIe0 MEM 2",
55 .start = 0x10000000,
56 .end = 0x10000000 + SZ_64M - 1,
57 .flags = IORESOURCE_MEM,
58 }, {
59 .name = "PCIe0 IO",
60 .start = 0xfe100000,
61 .end = 0xfe100000 + SZ_1M - 1,
62 .flags = IORESOURCE_IO,
63 },
64 };
65
66 static struct resource sh7786_pci1_resources[] = {
67 {
68 .name = "PCIe1 MEM 0",
69 .start = 0xfd800000,
70 .end = 0xfd800000 + SZ_8M - 1,
71 .flags = IORESOURCE_MEM,
72 }, {
73 .name = "PCIe1 MEM 1",
74 .start = 0xa0000000,
75 .end = 0xa0000000 + SZ_512M - 1,
76 .flags = IORESOURCE_MEM | IORESOURCE_MEM_32BIT,
77 }, {
78 .name = "PCIe1 MEM 2",
79 .start = 0x30000000,
80 .end = 0x30000000 + SZ_256M - 1,
81 .flags = IORESOURCE_MEM | IORESOURCE_MEM_32BIT,
82 }, {
83 .name = "PCIe1 IO",
84 .start = 0xfe300000,
85 .end = 0xfe300000 + SZ_1M - 1,
86 .flags = IORESOURCE_IO,
87 },
88 };
89
90 static struct resource sh7786_pci2_resources[] = {
91 {
92 .name = "PCIe2 MEM 0",
93 .start = 0xfc800000,
94 .end = 0xfc800000 + SZ_4M - 1,
95 .flags = IORESOURCE_MEM,
96 }, {
97 .name = "PCIe2 MEM 1",
98 .start = 0x80000000,
99 .end = 0x80000000 + SZ_512M - 1,
100 .flags = IORESOURCE_MEM | IORESOURCE_MEM_32BIT,
101 }, {
102 .name = "PCIe2 MEM 2",
103 .start = 0x20000000,
104 .end = 0x20000000 + SZ_256M - 1,
105 .flags = IORESOURCE_MEM | IORESOURCE_MEM_32BIT,
106 }, {
107 .name = "PCIe2 IO",
108 .start = 0xfcd00000,
109 .end = 0xfcd00000 + SZ_1M - 1,
110 .flags = IORESOURCE_IO,
111 },
112 };
113
114 extern struct pci_ops sh7786_pci_ops;
115
116 #define DEFINE_CONTROLLER(start, idx) \
117 { \
118 .pci_ops = &sh7786_pci_ops, \
119 .resources = sh7786_pci##idx##_resources, \
120 .nr_resources = ARRAY_SIZE(sh7786_pci##idx##_resources), \
121 .reg_base = start, \
122 .mem_offset = 0, \
123 .io_offset = 0, \
124 }
125
126 static struct pci_channel sh7786_pci_channels[] = {
127 DEFINE_CONTROLLER(0xfe000000, 0),
128 DEFINE_CONTROLLER(0xfe200000, 1),
129 DEFINE_CONTROLLER(0xfcc00000, 2),
130 };
131
132 static struct clk fixed_pciexclkp = {
133 .rate = 100000000, /* 100 MHz reference clock */
134 };
135
sh7786_pci_fixup(struct pci_dev * dev)136 static void sh7786_pci_fixup(struct pci_dev *dev)
137 {
138 /*
139 * Prevent enumeration of root complex resources.
140 */
141 if (pci_is_root_bus(dev->bus) && dev->devfn == 0) {
142 struct resource *r;
143
144 pci_dev_for_each_resource(dev, r) {
145 r->start = 0;
146 r->end = 0;
147 r->flags = 0;
148 }
149 }
150 }
151 DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_RENESAS, PCI_DEVICE_ID_RENESAS_SH7786,
152 sh7786_pci_fixup);
153
phy_wait_for_ack(struct pci_channel * chan)154 static int __init phy_wait_for_ack(struct pci_channel *chan)
155 {
156 unsigned int timeout = 100;
157
158 while (timeout--) {
159 if (pci_read_reg(chan, SH4A_PCIEPHYADRR) & (1 << BITS_ACK))
160 return 0;
161
162 udelay(100);
163 }
164
165 return -ETIMEDOUT;
166 }
167
pci_wait_for_irq(struct pci_channel * chan,unsigned int mask)168 static int __init pci_wait_for_irq(struct pci_channel *chan, unsigned int mask)
169 {
170 unsigned int timeout = 100;
171
172 while (timeout--) {
173 if ((pci_read_reg(chan, SH4A_PCIEINTR) & mask) == mask)
174 return 0;
175
176 udelay(100);
177 }
178
179 return -ETIMEDOUT;
180 }
181
phy_write_reg(struct pci_channel * chan,unsigned int addr,unsigned int lane,unsigned int data)182 static void __init phy_write_reg(struct pci_channel *chan, unsigned int addr,
183 unsigned int lane, unsigned int data)
184 {
185 unsigned long phyaddr;
186
187 phyaddr = (1 << BITS_CMD) + ((lane & 0xf) << BITS_LANE) +
188 ((addr & 0xff) << BITS_ADR);
189
190 /* Set write data */
191 pci_write_reg(chan, data, SH4A_PCIEPHYDOUTR);
192 pci_write_reg(chan, phyaddr, SH4A_PCIEPHYADRR);
193
194 phy_wait_for_ack(chan);
195
196 /* Clear command */
197 pci_write_reg(chan, 0, SH4A_PCIEPHYDOUTR);
198 pci_write_reg(chan, 0, SH4A_PCIEPHYADRR);
199
200 phy_wait_for_ack(chan);
201 }
202
pcie_clk_init(struct sh7786_pcie_port * port)203 static int __init pcie_clk_init(struct sh7786_pcie_port *port)
204 {
205 struct pci_channel *chan = port->hose;
206 struct clk *clk;
207 char fclk_name[16];
208 int ret;
209
210 /*
211 * First register the fixed clock
212 */
213 ret = clk_register(&fixed_pciexclkp);
214 if (unlikely(ret != 0))
215 return ret;
216
217 /*
218 * Grab the port's function clock, which the PHY clock depends
219 * on. clock lookups don't help us much at this point, since no
220 * dev_id is available this early. Lame.
221 */
222 snprintf(fclk_name, sizeof(fclk_name), "pcie%d_fck", port->index);
223
224 port->fclk = clk_get(NULL, fclk_name);
225 if (IS_ERR(port->fclk)) {
226 ret = PTR_ERR(port->fclk);
227 goto err_fclk;
228 }
229
230 clk_enable(port->fclk);
231
232 /*
233 * And now, set up the PHY clock
234 */
235 clk = &port->phy_clk;
236
237 memset(clk, 0, sizeof(struct clk));
238
239 clk->parent = &fixed_pciexclkp;
240 clk->enable_reg = (void __iomem *)(chan->reg_base + SH4A_PCIEPHYCTLR);
241 clk->enable_bit = BITS_CKE;
242
243 ret = sh_clk_mstp_register(clk, 1);
244 if (unlikely(ret < 0))
245 goto err_phy;
246
247 return 0;
248
249 err_phy:
250 clk_disable(port->fclk);
251 clk_put(port->fclk);
252 err_fclk:
253 clk_unregister(&fixed_pciexclkp);
254
255 return ret;
256 }
257
phy_init(struct sh7786_pcie_port * port)258 static int __init phy_init(struct sh7786_pcie_port *port)
259 {
260 struct pci_channel *chan = port->hose;
261 unsigned int timeout = 100;
262
263 clk_enable(&port->phy_clk);
264
265 /* Initialize the phy */
266 phy_write_reg(chan, 0x60, 0xf, 0x004b008b);
267 phy_write_reg(chan, 0x61, 0xf, 0x00007b41);
268 phy_write_reg(chan, 0x64, 0xf, 0x00ff4f00);
269 phy_write_reg(chan, 0x65, 0xf, 0x09070907);
270 phy_write_reg(chan, 0x66, 0xf, 0x00000010);
271 phy_write_reg(chan, 0x74, 0xf, 0x0007001c);
272 phy_write_reg(chan, 0x79, 0xf, 0x01fc000d);
273 phy_write_reg(chan, 0xb0, 0xf, 0x00000610);
274
275 /* Deassert Standby */
276 phy_write_reg(chan, 0x67, 0x1, 0x00000400);
277
278 /* Disable clock */
279 clk_disable(&port->phy_clk);
280
281 while (timeout--) {
282 if (pci_read_reg(chan, SH4A_PCIEPHYSR))
283 return 0;
284
285 udelay(100);
286 }
287
288 return -ETIMEDOUT;
289 }
290
pcie_reset(struct sh7786_pcie_port * port)291 static void __init pcie_reset(struct sh7786_pcie_port *port)
292 {
293 struct pci_channel *chan = port->hose;
294
295 pci_write_reg(chan, 1, SH4A_PCIESRSTR);
296 pci_write_reg(chan, 0, SH4A_PCIETCTLR);
297 pci_write_reg(chan, 0, SH4A_PCIESRSTR);
298 pci_write_reg(chan, 0, SH4A_PCIETXVC0SR);
299 }
300
pcie_init(struct sh7786_pcie_port * port)301 static int __init pcie_init(struct sh7786_pcie_port *port)
302 {
303 struct pci_channel *chan = port->hose;
304 unsigned int data;
305 phys_addr_t memstart, memend;
306 int ret, i, win;
307
308 /* Begin initialization */
309 pcie_reset(port);
310
311 /*
312 * Initial header for port config space is type 1, set the device
313 * class to match. Hardware takes care of propagating the IDSETR
314 * settings, so there is no need to bother with a quirk.
315 */
316 pci_write_reg(chan, PCI_CLASS_BRIDGE_PCI_NORMAL << 8, SH4A_PCIEIDSETR1);
317
318 /* Initialize default capabilities. */
319 data = pci_read_reg(chan, SH4A_PCIEEXPCAP0);
320 data &= ~(PCI_EXP_FLAGS_TYPE << 16);
321
322 if (port->endpoint)
323 data |= PCI_EXP_TYPE_ENDPOINT << 20;
324 else
325 data |= PCI_EXP_TYPE_ROOT_PORT << 20;
326
327 data |= PCI_CAP_ID_EXP;
328 pci_write_reg(chan, data, SH4A_PCIEEXPCAP0);
329
330 /* Enable data link layer active state reporting */
331 pci_write_reg(chan, PCI_EXP_LNKCAP_DLLLARC, SH4A_PCIEEXPCAP3);
332
333 /* Enable extended sync and ASPM L0s support */
334 data = pci_read_reg(chan, SH4A_PCIEEXPCAP4);
335 data &= ~PCI_EXP_LNKCTL_ASPMC;
336 data |= PCI_EXP_LNKCTL_ES | 1;
337 pci_write_reg(chan, data, SH4A_PCIEEXPCAP4);
338
339 /* Write out the physical slot number */
340 data = pci_read_reg(chan, SH4A_PCIEEXPCAP5);
341 data &= ~PCI_EXP_SLTCAP_PSN;
342 data |= (port->index + 1) << 19;
343 pci_write_reg(chan, data, SH4A_PCIEEXPCAP5);
344
345 /* Set the completion timer timeout to the maximum 32ms. */
346 data = pci_read_reg(chan, SH4A_PCIETLCTLR);
347 data &= ~0x3f00;
348 data |= 0x32 << 8;
349 pci_write_reg(chan, data, SH4A_PCIETLCTLR);
350
351 /*
352 * Set fast training sequences to the maximum 255,
353 * and enable MAC data scrambling.
354 */
355 data = pci_read_reg(chan, SH4A_PCIEMACCTLR);
356 data &= ~PCIEMACCTLR_SCR_DIS;
357 data |= (0xff << 16);
358 pci_write_reg(chan, data, SH4A_PCIEMACCTLR);
359
360 memstart = __pa(memory_start);
361 memend = __pa(memory_end);
362 memsize = roundup_pow_of_two(memend - memstart);
363
364 /*
365 * The start address must be aligned on its size. So we round
366 * it down, and then recalculate the size so that it covers
367 * the entire memory.
368 */
369 memstart = ALIGN_DOWN(memstart, memsize);
370 memsize = roundup_pow_of_two(memend - memstart);
371
372 /*
373 * If there's more than 512MB of memory, we need to roll over to
374 * LAR1/LAMR1.
375 */
376 if (memsize > SZ_512M) {
377 pci_write_reg(chan, memstart + SZ_512M, SH4A_PCIELAR1);
378 pci_write_reg(chan, ((memsize - SZ_512M) - SZ_256) | 1,
379 SH4A_PCIELAMR1);
380 memsize = SZ_512M;
381 } else {
382 /*
383 * Otherwise just zero it out and disable it.
384 */
385 pci_write_reg(chan, 0, SH4A_PCIELAR1);
386 pci_write_reg(chan, 0, SH4A_PCIELAMR1);
387 }
388
389 /*
390 * LAR0/LAMR0 covers up to the first 512MB, which is enough to
391 * cover all of lowmem on most platforms.
392 */
393 pci_write_reg(chan, memstart, SH4A_PCIELAR0);
394 pci_write_reg(chan, (memsize - SZ_256) | 1, SH4A_PCIELAMR0);
395
396 /* Finish initialization */
397 data = pci_read_reg(chan, SH4A_PCIETCTLR);
398 data |= 0x1;
399 pci_write_reg(chan, data, SH4A_PCIETCTLR);
400
401 /* Let things settle down a bit.. */
402 mdelay(100);
403
404 /* Enable DL_Active Interrupt generation */
405 data = pci_read_reg(chan, SH4A_PCIEDLINTENR);
406 data |= PCIEDLINTENR_DLL_ACT_ENABLE;
407 pci_write_reg(chan, data, SH4A_PCIEDLINTENR);
408
409 /* Disable MAC data scrambling. */
410 data = pci_read_reg(chan, SH4A_PCIEMACCTLR);
411 data |= PCIEMACCTLR_SCR_DIS | (0xff << 16);
412 pci_write_reg(chan, data, SH4A_PCIEMACCTLR);
413
414 /*
415 * This will timeout if we don't have a link, but we permit the
416 * port to register anyways in order to support hotplug on future
417 * hardware.
418 */
419 ret = pci_wait_for_irq(chan, MASK_INT_TX_CTRL);
420
421 data = pci_read_reg(chan, SH4A_PCIEPCICONF1);
422 data &= ~(PCI_STATUS_DEVSEL_MASK << 16);
423 data |= PCI_COMMAND_IO | PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER |
424 (PCI_STATUS_CAP_LIST | PCI_STATUS_DEVSEL_FAST) << 16;
425 pci_write_reg(chan, data, SH4A_PCIEPCICONF1);
426
427 pci_write_reg(chan, 0x80888000, SH4A_PCIETXVC0DCTLR);
428 pci_write_reg(chan, 0x00222000, SH4A_PCIERXVC0DCTLR);
429
430 wmb();
431
432 if (ret == 0) {
433 data = pci_read_reg(chan, SH4A_PCIEMACSR);
434 printk(KERN_NOTICE "PCI: PCIe#%d x%d link detected\n",
435 port->index, (data >> 20) & 0x3f);
436 } else
437 printk(KERN_NOTICE "PCI: PCIe#%d link down\n",
438 port->index);
439
440 for (i = win = 0; i < chan->nr_resources; i++) {
441 struct resource *res = chan->resources + i;
442 resource_size_t size;
443 u32 mask;
444
445 /*
446 * We can't use the 32-bit mode windows in legacy 29-bit
447 * mode, so just skip them entirely.
448 */
449 if ((res->flags & IORESOURCE_MEM_32BIT) && __in_29bit_mode())
450 res->flags |= IORESOURCE_DISABLED;
451
452 if (res->flags & IORESOURCE_DISABLED)
453 continue;
454
455 pci_write_reg(chan, 0x00000000, SH4A_PCIEPTCTLR(win));
456
457 /*
458 * The PAMR mask is calculated in units of 256kB, which
459 * keeps things pretty simple.
460 */
461 size = resource_size(res);
462 mask = (roundup_pow_of_two(size) / SZ_256K) - 1;
463 pci_write_reg(chan, mask << 18, SH4A_PCIEPAMR(win));
464
465 pci_write_reg(chan, upper_32_bits(res->start),
466 SH4A_PCIEPARH(win));
467 pci_write_reg(chan, lower_32_bits(res->start),
468 SH4A_PCIEPARL(win));
469
470 mask = MASK_PARE;
471 if (res->flags & IORESOURCE_IO)
472 mask |= MASK_SPC;
473
474 pci_write_reg(chan, mask, SH4A_PCIEPTCTLR(win));
475
476 win++;
477 }
478
479 return 0;
480 }
481
pcibios_map_platform_irq(const struct pci_dev * pdev,u8 slot,u8 pin)482 int pcibios_map_platform_irq(const struct pci_dev *pdev, u8 slot, u8 pin)
483 {
484 return evt2irq(0xae0);
485 }
486
pcibios_bus_add_device(struct pci_dev * pdev)487 void pcibios_bus_add_device(struct pci_dev *pdev)
488 {
489 dma_direct_set_offset(&pdev->dev, __pa(memory_start),
490 __pa(memory_start) - memstart, memsize);
491 }
492
sh7786_pcie_core_init(void)493 static int __init sh7786_pcie_core_init(void)
494 {
495 /* Return the number of ports */
496 return test_mode_pin(MODE_PIN12) ? 3 : 2;
497 }
498
sh7786_pcie_init_hw(void * data,async_cookie_t cookie)499 static void __init sh7786_pcie_init_hw(void *data, async_cookie_t cookie)
500 {
501 struct sh7786_pcie_port *port = data;
502 int ret;
503
504 /*
505 * Check if we are configured in endpoint or root complex mode,
506 * this is a fixed pin setting that applies to all PCIe ports.
507 */
508 port->endpoint = test_mode_pin(MODE_PIN11);
509
510 /*
511 * Setup clocks, needed both for PHY and PCIe registers.
512 */
513 ret = pcie_clk_init(port);
514 if (unlikely(ret < 0)) {
515 pr_err("clock initialization failed for port#%d\n",
516 port->index);
517 return;
518 }
519
520 ret = phy_init(port);
521 if (unlikely(ret < 0)) {
522 pr_err("phy initialization failed for port#%d\n",
523 port->index);
524 return;
525 }
526
527 ret = pcie_init(port);
528 if (unlikely(ret < 0)) {
529 pr_err("core initialization failed for port#%d\n",
530 port->index);
531 return;
532 }
533
534 /* In the interest of preserving device ordering, synchronize */
535 async_synchronize_cookie(cookie);
536
537 register_pci_controller(port->hose);
538 }
539
540 static struct sh7786_pcie_hwops sh7786_65nm_pcie_hwops __initdata = {
541 .core_init = sh7786_pcie_core_init,
542 .port_init_hw = sh7786_pcie_init_hw,
543 };
544
sh7786_pcie_init(void)545 static int __init sh7786_pcie_init(void)
546 {
547 struct clk *platclk;
548 u32 mm_sel;
549 int i;
550
551 printk(KERN_NOTICE "PCI: Starting initialization.\n");
552
553 sh7786_pcie_hwops = &sh7786_65nm_pcie_hwops;
554
555 nr_ports = sh7786_pcie_hwops->core_init();
556 BUG_ON(nr_ports > ARRAY_SIZE(sh7786_pci_channels));
557
558 if (unlikely(nr_ports == 0))
559 return -ENODEV;
560
561 sh7786_pcie_ports = kcalloc(nr_ports, sizeof(struct sh7786_pcie_port),
562 GFP_KERNEL);
563 if (unlikely(!sh7786_pcie_ports))
564 return -ENOMEM;
565
566 /*
567 * Fetch any optional platform clock associated with this block.
568 *
569 * This is a rather nasty hack for boards with spec-mocking FPGAs
570 * that have a secondary set of clocks outside of the on-chip
571 * ones that need to be accounted for before there is any chance
572 * of touching the existing MSTP bits or CPG clocks.
573 */
574 platclk = clk_get(NULL, "pcie_plat_clk");
575 if (IS_ERR(platclk)) {
576 /* Sane hardware should probably get a WARN_ON.. */
577 platclk = NULL;
578 }
579
580 clk_enable(platclk);
581
582 mm_sel = sh7786_mm_sel();
583
584 /*
585 * Depending on the MMSELR register value, the PCIe0 MEM 1
586 * area may not be available. See Table 13.11 of the SH7786
587 * datasheet.
588 */
589 if (mm_sel != 1 && mm_sel != 2 && mm_sel != 5 && mm_sel != 6)
590 sh7786_pci0_resources[2].flags |= IORESOURCE_DISABLED;
591
592 printk(KERN_NOTICE "PCI: probing %d ports.\n", nr_ports);
593
594 for (i = 0; i < nr_ports; i++) {
595 struct sh7786_pcie_port *port = sh7786_pcie_ports + i;
596
597 port->index = i;
598 port->hose = sh7786_pci_channels + i;
599 port->hose->io_map_base = port->hose->resources[0].start;
600
601 async_schedule(sh7786_pcie_hwops->port_init_hw, port);
602 }
603
604 async_synchronize_full();
605
606 return 0;
607 }
608 arch_initcall(sh7786_pcie_init);
609