?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
cxl: Move include file cxl.h -> cxl-base.h
[linux/fpc-iii.git] / arch / powerpc / platforms / powernv / pci-ioda.c
blobb2841853550aa5abe4cb659b96067c139b4edfc9
1 /*
2 * Support PCI/PCIe on PowerNV platforms
3 *
4 * Copyright 2011 Benjamin Herrenschmidt, IBM Corp.
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 */
11
12 #undef DEBUG
13
14 #include <linux/kernel.h>
15 #include <linux/pci.h>
16 #include <linux/crash_dump.h>
17 #include <linux/debugfs.h>
18 #include <linux/delay.h>
19 #include <linux/string.h>
20 #include <linux/init.h>
21 #include <linux/bootmem.h>
22 #include <linux/irq.h>
23 #include <linux/io.h>
24 #include <linux/msi.h>
25 #include <linux/memblock.h>
26
27 #include <asm/sections.h>
28 #include <asm/io.h>
29 #include <asm/prom.h>
30 #include <asm/pci-bridge.h>
31 #include <asm/machdep.h>
32 #include <asm/msi_bitmap.h>
33 #include <asm/ppc-pci.h>
34 #include <asm/opal.h>
35 #include <asm/iommu.h>
36 #include <asm/tce.h>
37 #include <asm/xics.h>
38 #include <asm/debug.h>
39 #include <asm/firmware.h>
40 #include <asm/pnv-pci.h>
41
42 #include <misc/cxl-base.h>
43
44 #include "powernv.h"
45 #include "pci.h"
46
47 /* 256M DMA window, 4K TCE pages, 8 bytes TCE */
48 #define TCE32_TABLE_SIZE ((0x10000000 / 0x1000) * 8)
49
50 static void pe_level_printk(const struct pnv_ioda_pe *pe, const char *level,
51 const char *fmt, ...)
52 {
53 struct va_format vaf;
54 va_list args;
55 char pfix[32];
56
57 va_start(args, fmt);
58
59 vaf.fmt = fmt;
60 vaf.va = &args;
61
62 if (pe->flags & PNV_IODA_PE_DEV)
63 strlcpy(pfix, dev_name(&pe->pdev->dev), sizeof(pfix));
64 else if (pe->flags & (PNV_IODA_PE_BUS | PNV_IODA_PE_BUS_ALL))
65 sprintf(pfix, "%04x:%02x ",
66 pci_domain_nr(pe->pbus), pe->pbus->number);
67 #ifdef CONFIG_PCI_IOV
68 else if (pe->flags & PNV_IODA_PE_VF)
69 sprintf(pfix, "%04x:%02x:%2x.%d",
70 pci_domain_nr(pe->parent_dev->bus),
71 (pe->rid & 0xff00) >> 8,
72 PCI_SLOT(pe->rid), PCI_FUNC(pe->rid));
73 #endif /* CONFIG_PCI_IOV*/
74
75 printk("%spci %s: [PE# %.3d] %pV",
76 level, pfix, pe->pe_number, &vaf);
77
78 va_end(args);
79 }
80
81 #define pe_err(pe, fmt, ...) \
82 pe_level_printk(pe, KERN_ERR, fmt, ##__VA_ARGS__)
83 #define pe_warn(pe, fmt, ...) \
84 pe_level_printk(pe, KERN_WARNING, fmt, ##__VA_ARGS__)
85 #define pe_info(pe, fmt, ...) \
86 pe_level_printk(pe, KERN_INFO, fmt, ##__VA_ARGS__)
87
88 static bool pnv_iommu_bypass_disabled __read_mostly;
89
90 static int __init iommu_setup(char *str)
91 {
92 if (!str)
93 return -EINVAL;
94
95 while (*str) {
96 if (!strncmp(str, "nobypass", 8)) {
97 pnv_iommu_bypass_disabled = true;
98 pr_info("PowerNV: IOMMU bypass window disabled.\n");
99 break;
100 }
101 str += strcspn(str, ",");
102 if (*str == ',')
103 str++;
104 }
105
106 return 0;
107 }
108 early_param("iommu", iommu_setup);
109
110 /*
111 * stdcix is only supposed to be used in hypervisor real mode as per
112 * the architecture spec
113 */
114 static inline void __raw_rm_writeq(u64 val, volatile void __iomem *paddr)
115 {
116 __asm__ __volatile__("stdcix %0,0,%1"
117 : : "r" (val), "r" (paddr) : "memory");
118 }
119
120 static inline bool pnv_pci_is_mem_pref_64(unsigned long flags)
121 {
122 return ((flags & (IORESOURCE_MEM_64 | IORESOURCE_PREFETCH)) ==
123 (IORESOURCE_MEM_64 | IORESOURCE_PREFETCH));
124 }
125
126 static void pnv_ioda_reserve_pe(struct pnv_phb *phb, int pe_no)
127 {
128 if (!(pe_no >= 0 && pe_no < phb->ioda.total_pe)) {
129 pr_warn("%s: Invalid PE %d on PHB#%x\n",
130 __func__, pe_no, phb->hose->global_number);
131 return;
132 }
133
134 if (test_and_set_bit(pe_no, phb->ioda.pe_alloc)) {
135 pr_warn("%s: PE %d was assigned on PHB#%x\n",
136 __func__, pe_no, phb->hose->global_number);
137 return;
138 }
139
140 phb->ioda.pe_array[pe_no].phb = phb;
141 phb->ioda.pe_array[pe_no].pe_number = pe_no;
142 }
143
144 static int pnv_ioda_alloc_pe(struct pnv_phb *phb)
145 {
146 unsigned long pe;
147
148 do {
149 pe = find_next_zero_bit(phb->ioda.pe_alloc,
150 phb->ioda.total_pe, 0);
151 if (pe >= phb->ioda.total_pe)
152 return IODA_INVALID_PE;
153 } while(test_and_set_bit(pe, phb->ioda.pe_alloc));
154
155 phb->ioda.pe_array[pe].phb = phb;
156 phb->ioda.pe_array[pe].pe_number = pe;
157 return pe;
158 }
159
160 static void pnv_ioda_free_pe(struct pnv_phb *phb, int pe)
161 {
162 WARN_ON(phb->ioda.pe_array[pe].pdev);
163
164 memset(&phb->ioda.pe_array[pe], 0, sizeof(struct pnv_ioda_pe));
165 clear_bit(pe, phb->ioda.pe_alloc);
166 }
167
168 /* The default M64 BAR is shared by all PEs */
169 static int pnv_ioda2_init_m64(struct pnv_phb *phb)
170 {
171 const char *desc;
172 struct resource *r;
173 s64 rc;
174
175 /* Configure the default M64 BAR */
176 rc = opal_pci_set_phb_mem_window(phb->opal_id,
177 OPAL_M64_WINDOW_TYPE,
178 phb->ioda.m64_bar_idx,
179 phb->ioda.m64_base,
180 0, /* unused */
181 phb->ioda.m64_size);
182 if (rc != OPAL_SUCCESS) {
183 desc = "configuring";
184 goto fail;
185 }
186
187 /* Enable the default M64 BAR */
188 rc = opal_pci_phb_mmio_enable(phb->opal_id,
189 OPAL_M64_WINDOW_TYPE,
190 phb->ioda.m64_bar_idx,
191 OPAL_ENABLE_M64_SPLIT);
192 if (rc != OPAL_SUCCESS) {
193 desc = "enabling";
194 goto fail;
195 }
196
197 /* Mark the M64 BAR assigned */
198 set_bit(phb->ioda.m64_bar_idx, &phb->ioda.m64_bar_alloc);
199
200 /*
201 * Strip off the segment used by the reserved PE, which is
202 * expected to be 0 or last one of PE capabicity.
203 */
204 r = &phb->hose->mem_resources[1];
205 if (phb->ioda.reserved_pe == 0)
206 r->start += phb->ioda.m64_segsize;
207 else if (phb->ioda.reserved_pe == (phb->ioda.total_pe - 1))
208 r->end -= phb->ioda.m64_segsize;
209 else
210 pr_warn(" Cannot strip M64 segment for reserved PE#%d\n",
211 phb->ioda.reserved_pe);
212
213 return 0;
214
215 fail:
216 pr_warn(" Failure %lld %s M64 BAR#%d\n",
217 rc, desc, phb->ioda.m64_bar_idx);
218 opal_pci_phb_mmio_enable(phb->opal_id,
219 OPAL_M64_WINDOW_TYPE,
220 phb->ioda.m64_bar_idx,
221 OPAL_DISABLE_M64);
222 return -EIO;
223 }
224
225 static void pnv_ioda2_reserve_m64_pe(struct pnv_phb *phb)
226 {
227 resource_size_t sgsz = phb->ioda.m64_segsize;
228 struct pci_dev *pdev;
229 struct resource *r;
230 int base, step, i;
231
232 /*
233 * Root bus always has full M64 range and root port has
234 * M64 range used in reality. So we're checking root port
235 * instead of root bus.
236 */
237 list_for_each_entry(pdev, &phb->hose->bus->devices, bus_list) {
238 for (i = 0; i < PCI_BRIDGE_RESOURCE_NUM; i++) {
239 r = &pdev->resource[PCI_BRIDGE_RESOURCES + i];
240 if (!r->parent ||
241 !pnv_pci_is_mem_pref_64(r->flags))
242 continue;
243
244 base = (r->start - phb->ioda.m64_base) / sgsz;
245 for (step = 0; step < resource_size(r) / sgsz; step++)
246 pnv_ioda_reserve_pe(phb, base + step);
247 }
248 }
249 }
250
251 static int pnv_ioda2_pick_m64_pe(struct pnv_phb *phb,
252 struct pci_bus *bus, int all)
253 {
254 resource_size_t segsz = phb->ioda.m64_segsize;
255 struct pci_dev *pdev;
256 struct resource *r;
257 struct pnv_ioda_pe *master_pe, *pe;
258 unsigned long size, *pe_alloc;
259 bool found;
260 int start, i, j;
261
262 /* Root bus shouldn't use M64 */
263 if (pci_is_root_bus(bus))
264 return IODA_INVALID_PE;
265
266 /* We support only one M64 window on each bus */
267 found = false;
268 pci_bus_for_each_resource(bus, r, i) {
269 if (r && r->parent &&
270 pnv_pci_is_mem_pref_64(r->flags)) {
271 found = true;
272 break;
273 }
274 }
275
276 /* No M64 window found ? */
277 if (!found)
278 return IODA_INVALID_PE;
279
280 /* Allocate bitmap */
281 size = _ALIGN_UP(phb->ioda.total_pe / 8, sizeof(unsigned long));
282 pe_alloc = kzalloc(size, GFP_KERNEL);
283 if (!pe_alloc) {
284 pr_warn("%s: Out of memory !\n",
285 __func__);
286 return IODA_INVALID_PE;
287 }
288
289 /*
290 * Figure out reserved PE numbers by the PE
291 * the its child PEs.
292 */
293 start = (r->start - phb->ioda.m64_base) / segsz;
294 for (i = 0; i < resource_size(r) / segsz; i++)
295 set_bit(start + i, pe_alloc);
296
297 if (all)
298 goto done;
299
300 /*
301 * If the PE doesn't cover all subordinate buses,
302 * we need subtract from reserved PEs for children.
303 */
304 list_for_each_entry(pdev, &bus->devices, bus_list) {
305 if (!pdev->subordinate)
306 continue;
307
308 pci_bus_for_each_resource(pdev->subordinate, r, i) {
309 if (!r || !r->parent ||
310 !pnv_pci_is_mem_pref_64(r->flags))
311 continue;
312
313 start = (r->start - phb->ioda.m64_base) / segsz;
314 for (j = 0; j < resource_size(r) / segsz ; j++)
315 clear_bit(start + j, pe_alloc);
316 }
317 }
318
319 /*
320 * the current bus might not own M64 window and that's all
321 * contributed by its child buses. For the case, we needn't
322 * pick M64 dependent PE#.
323 */
324 if (bitmap_empty(pe_alloc, phb->ioda.total_pe)) {
325 kfree(pe_alloc);
326 return IODA_INVALID_PE;
327 }
328
329 /*
330 * Figure out the master PE and put all slave PEs to master
331 * PE's list to form compound PE.
332 */
333 done:
334 master_pe = NULL;
335 i = -1;
336 while ((i = find_next_bit(pe_alloc, phb->ioda.total_pe, i + 1)) <
337 phb->ioda.total_pe) {
338 pe = &phb->ioda.pe_array[i];
339
340 if (!master_pe) {
341 pe->flags |= PNV_IODA_PE_MASTER;
342 INIT_LIST_HEAD(&pe->slaves);
343 master_pe = pe;
344 } else {
345 pe->flags |= PNV_IODA_PE_SLAVE;
346 pe->master = master_pe;
347 list_add_tail(&pe->list, &master_pe->slaves);
348 }
349 }
350
351 kfree(pe_alloc);
352 return master_pe->pe_number;
353 }
354
355 static void __init pnv_ioda_parse_m64_window(struct pnv_phb *phb)
356 {
357 struct pci_controller *hose = phb->hose;
358 struct device_node *dn = hose->dn;
359 struct resource *res;
360 const u32 *r;
361 u64 pci_addr;
362
363 /* FIXME: Support M64 for P7IOC */
364 if (phb->type != PNV_PHB_IODA2) {
365 pr_info(" Not support M64 window\n");
366 return;
367 }
368
369 if (!firmware_has_feature(FW_FEATURE_OPALv3)) {
370 pr_info(" Firmware too old to support M64 window\n");
371 return;
372 }
373
374 r = of_get_property(dn, "ibm,opal-m64-window", NULL);
375 if (!r) {
376 pr_info(" No <ibm,opal-m64-window> on %s\n",
377 dn->full_name);
378 return;
379 }
380
381 res = &hose->mem_resources[1];
382 res->start = of_translate_address(dn, r + 2);
383 res->end = res->start + of_read_number(r + 4, 2) - 1;
384 res->flags = (IORESOURCE_MEM | IORESOURCE_MEM_64 | IORESOURCE_PREFETCH);
385 pci_addr = of_read_number(r, 2);
386 hose->mem_offset[1] = res->start - pci_addr;
387
388 phb->ioda.m64_size = resource_size(res);
389 phb->ioda.m64_segsize = phb->ioda.m64_size / phb->ioda.total_pe;
390 phb->ioda.m64_base = pci_addr;
391
392 pr_info(" MEM64 0x%016llx..0x%016llx -> 0x%016llx\n",
393 res->start, res->end, pci_addr);
394
395 /* Use last M64 BAR to cover M64 window */
396 phb->ioda.m64_bar_idx = 15;
397 phb->init_m64 = pnv_ioda2_init_m64;
398 phb->reserve_m64_pe = pnv_ioda2_reserve_m64_pe;
399 phb->pick_m64_pe = pnv_ioda2_pick_m64_pe;
400 }
401
402 static void pnv_ioda_freeze_pe(struct pnv_phb *phb, int pe_no)
403 {
404 struct pnv_ioda_pe *pe = &phb->ioda.pe_array[pe_no];
405 struct pnv_ioda_pe *slave;
406 s64 rc;
407
408 /* Fetch master PE */
409 if (pe->flags & PNV_IODA_PE_SLAVE) {
410 pe = pe->master;
411 if (WARN_ON(!pe || !(pe->flags & PNV_IODA_PE_MASTER)))
412 return;
413
414 pe_no = pe->pe_number;
415 }
416
417 /* Freeze master PE */
418 rc = opal_pci_eeh_freeze_set(phb->opal_id,
419 pe_no,
420 OPAL_EEH_ACTION_SET_FREEZE_ALL);
421 if (rc != OPAL_SUCCESS) {
422 pr_warn("%s: Failure %lld freezing PHB#%x-PE#%x\n",
423 __func__, rc, phb->hose->global_number, pe_no);
424 return;
425 }
426
427 /* Freeze slave PEs */
428 if (!(pe->flags & PNV_IODA_PE_MASTER))
429 return;
430
431 list_for_each_entry(slave, &pe->slaves, list) {
432 rc = opal_pci_eeh_freeze_set(phb->opal_id,
433 slave->pe_number,
434 OPAL_EEH_ACTION_SET_FREEZE_ALL);
435 if (rc != OPAL_SUCCESS)
436 pr_warn("%s: Failure %lld freezing PHB#%x-PE#%x\n",
437 __func__, rc, phb->hose->global_number,
438 slave->pe_number);
439 }
440 }
441
442 static int pnv_ioda_unfreeze_pe(struct pnv_phb *phb, int pe_no, int opt)
443 {
444 struct pnv_ioda_pe *pe, *slave;
445 s64 rc;
446
447 /* Find master PE */
448 pe = &phb->ioda.pe_array[pe_no];
449 if (pe->flags & PNV_IODA_PE_SLAVE) {
450 pe = pe->master;
451 WARN_ON(!pe || !(pe->flags & PNV_IODA_PE_MASTER));
452 pe_no = pe->pe_number;
453 }
454
455 /* Clear frozen state for master PE */
456 rc = opal_pci_eeh_freeze_clear(phb->opal_id, pe_no, opt);
457 if (rc != OPAL_SUCCESS) {
458 pr_warn("%s: Failure %lld clear %d on PHB#%x-PE#%x\n",
459 __func__, rc, opt, phb->hose->global_number, pe_no);
460 return -EIO;
461 }
462
463 if (!(pe->flags & PNV_IODA_PE_MASTER))
464 return 0;
465
466 /* Clear frozen state for slave PEs */
467 list_for_each_entry(slave, &pe->slaves, list) {
468 rc = opal_pci_eeh_freeze_clear(phb->opal_id,
469 slave->pe_number,
470 opt);
471 if (rc != OPAL_SUCCESS) {
472 pr_warn("%s: Failure %lld clear %d on PHB#%x-PE#%x\n",
473 __func__, rc, opt, phb->hose->global_number,
474 slave->pe_number);
475 return -EIO;
476 }
477 }
478
479 return 0;
480 }
481
482 static int pnv_ioda_get_pe_state(struct pnv_phb *phb, int pe_no)
483 {
484 struct pnv_ioda_pe *slave, *pe;
485 u8 fstate, state;
486 __be16 pcierr;
487 s64 rc;
488
489 /* Sanity check on PE number */
490 if (pe_no < 0 || pe_no >= phb->ioda.total_pe)
491 return OPAL_EEH_STOPPED_PERM_UNAVAIL;
492
493 /*
494 * Fetch the master PE and the PE instance might be
495 * not initialized yet.
496 */
497 pe = &phb->ioda.pe_array[pe_no];
498 if (pe->flags & PNV_IODA_PE_SLAVE) {
499 pe = pe->master;
500 WARN_ON(!pe || !(pe->flags & PNV_IODA_PE_MASTER));
501 pe_no = pe->pe_number;
502 }
503
504 /* Check the master PE */
505 rc = opal_pci_eeh_freeze_status(phb->opal_id, pe_no,
506 &state, &pcierr, NULL);
507 if (rc != OPAL_SUCCESS) {
508 pr_warn("%s: Failure %lld getting "
509 "PHB#%x-PE#%x state\n",
510 __func__, rc,
511 phb->hose->global_number, pe_no);
512 return OPAL_EEH_STOPPED_TEMP_UNAVAIL;
513 }
514
515 /* Check the slave PE */
516 if (!(pe->flags & PNV_IODA_PE_MASTER))
517 return state;
518
519 list_for_each_entry(slave, &pe->slaves, list) {
520 rc = opal_pci_eeh_freeze_status(phb->opal_id,
521 slave->pe_number,
522 &fstate,
523 &pcierr,
524 NULL);
525 if (rc != OPAL_SUCCESS) {
526 pr_warn("%s: Failure %lld getting "
527 "PHB#%x-PE#%x state\n",
528 __func__, rc,
529 phb->hose->global_number, slave->pe_number);
530 return OPAL_EEH_STOPPED_TEMP_UNAVAIL;
531 }
532
533 /*
534 * Override the result based on the ascending
535 * priority.
536 */
537 if (fstate > state)
538 state = fstate;
539 }
540
541 return state;
542 }
543
544 /* Currently those 2 are only used when MSIs are enabled, this will change
545 * but in the meantime, we need to protect them to avoid warnings
546 */
547 #ifdef CONFIG_PCI_MSI
548 static struct pnv_ioda_pe *pnv_ioda_get_pe(struct pci_dev *dev)
549 {
550 struct pci_controller *hose = pci_bus_to_host(dev->bus);
551 struct pnv_phb *phb = hose->private_data;
552 struct pci_dn *pdn = pci_get_pdn(dev);
553
554 if (!pdn)
555 return NULL;
556 if (pdn->pe_number == IODA_INVALID_PE)
557 return NULL;
558 return &phb->ioda.pe_array[pdn->pe_number];
559 }
560 #endif /* CONFIG_PCI_MSI */
561
562 static int pnv_ioda_set_one_peltv(struct pnv_phb *phb,
563 struct pnv_ioda_pe *parent,
564 struct pnv_ioda_pe *child,
565 bool is_add)
566 {
567 const char *desc = is_add ? "adding" : "removing";
568 uint8_t op = is_add ? OPAL_ADD_PE_TO_DOMAIN :
569 OPAL_REMOVE_PE_FROM_DOMAIN;
570 struct pnv_ioda_pe *slave;
571 long rc;
572
573 /* Parent PE affects child PE */
574 rc = opal_pci_set_peltv(phb->opal_id, parent->pe_number,
575 child->pe_number, op);
576 if (rc != OPAL_SUCCESS) {
577 pe_warn(child, "OPAL error %ld %s to parent PELTV\n",
578 rc, desc);
579 return -ENXIO;
580 }
581
582 if (!(child->flags & PNV_IODA_PE_MASTER))
583 return 0;
584
585 /* Compound case: parent PE affects slave PEs */
586 list_for_each_entry(slave, &child->slaves, list) {
587 rc = opal_pci_set_peltv(phb->opal_id, parent->pe_number,
588 slave->pe_number, op);
589 if (rc != OPAL_SUCCESS) {
590 pe_warn(slave, "OPAL error %ld %s to parent PELTV\n",
591 rc, desc);
592 return -ENXIO;
593 }
594 }
595
596 return 0;
597 }
598
599 static int pnv_ioda_set_peltv(struct pnv_phb *phb,
600 struct pnv_ioda_pe *pe,
601 bool is_add)
602 {
603 struct pnv_ioda_pe *slave;
604 struct pci_dev *pdev = NULL;
605 int ret;
606
607 /*
608 * Clear PE frozen state. If it's master PE, we need
609 * clear slave PE frozen state as well.
610 */
611 if (is_add) {
612 opal_pci_eeh_freeze_clear(phb->opal_id, pe->pe_number,
613 OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
614 if (pe->flags & PNV_IODA_PE_MASTER) {
615 list_for_each_entry(slave, &pe->slaves, list)
616 opal_pci_eeh_freeze_clear(phb->opal_id,
617 slave->pe_number,
618 OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
619 }
620 }
621
622 /*
623 * Associate PE in PELT. We need add the PE into the
624 * corresponding PELT-V as well. Otherwise, the error
625 * originated from the PE might contribute to other
626 * PEs.
627 */
628 ret = pnv_ioda_set_one_peltv(phb, pe, pe, is_add);
629 if (ret)
630 return ret;
631
632 /* For compound PEs, any one affects all of them */
633 if (pe->flags & PNV_IODA_PE_MASTER) {
634 list_for_each_entry(slave, &pe->slaves, list) {
635 ret = pnv_ioda_set_one_peltv(phb, slave, pe, is_add);
636 if (ret)
637 return ret;
638 }
639 }
640
641 if (pe->flags & (PNV_IODA_PE_BUS_ALL | PNV_IODA_PE_BUS))
642 pdev = pe->pbus->self;
643 else if (pe->flags & PNV_IODA_PE_DEV)
644 pdev = pe->pdev->bus->self;
645 #ifdef CONFIG_PCI_IOV
646 else if (pe->flags & PNV_IODA_PE_VF)
647 pdev = pe->parent_dev->bus->self;
648 #endif /* CONFIG_PCI_IOV */
649 while (pdev) {
650 struct pci_dn *pdn = pci_get_pdn(pdev);
651 struct pnv_ioda_pe *parent;
652
653 if (pdn && pdn->pe_number != IODA_INVALID_PE) {
654 parent = &phb->ioda.pe_array[pdn->pe_number];
655 ret = pnv_ioda_set_one_peltv(phb, parent, pe, is_add);
656 if (ret)
657 return ret;
658 }
659
660 pdev = pdev->bus->self;
661 }
662
663 return 0;
664 }
665
666 #ifdef CONFIG_PCI_IOV
667 static int pnv_ioda_deconfigure_pe(struct pnv_phb *phb, struct pnv_ioda_pe *pe)
668 {
669 struct pci_dev *parent;
670 uint8_t bcomp, dcomp, fcomp;
671 int64_t rc;
672 long rid_end, rid;
673
674 /* Currently, we just deconfigure VF PE. Bus PE will always there.*/
675 if (pe->pbus) {
676 int count;
677
678 dcomp = OPAL_IGNORE_RID_DEVICE_NUMBER;
679 fcomp = OPAL_IGNORE_RID_FUNCTION_NUMBER;
680 parent = pe->pbus->self;
681 if (pe->flags & PNV_IODA_PE_BUS_ALL)
682 count = pe->pbus->busn_res.end - pe->pbus->busn_res.start + 1;
683 else
684 count = 1;
685
686 switch(count) {
687 case 1: bcomp = OpalPciBusAll; break;
688 case 2: bcomp = OpalPciBus7Bits; break;
689 case 4: bcomp = OpalPciBus6Bits; break;
690 case 8: bcomp = OpalPciBus5Bits; break;
691 case 16: bcomp = OpalPciBus4Bits; break;
692 case 32: bcomp = OpalPciBus3Bits; break;
693 default:
694 dev_err(&pe->pbus->dev, "Number of subordinate buses %d unsupported\n",
695 count);
696 /* Do an exact match only */
697 bcomp = OpalPciBusAll;
698 }
699 rid_end = pe->rid + (count << 8);
700 } else {
701 if (pe->flags & PNV_IODA_PE_VF)
702 parent = pe->parent_dev;
703 else
704 parent = pe->pdev->bus->self;
705 bcomp = OpalPciBusAll;
706 dcomp = OPAL_COMPARE_RID_DEVICE_NUMBER;
707 fcomp = OPAL_COMPARE_RID_FUNCTION_NUMBER;
708 rid_end = pe->rid + 1;
709 }
710
711 /* Clear the reverse map */
712 for (rid = pe->rid; rid < rid_end; rid++)
713 phb->ioda.pe_rmap[rid] = 0;
714
715 /* Release from all parents PELT-V */
716 while (parent) {
717 struct pci_dn *pdn = pci_get_pdn(parent);
718 if (pdn && pdn->pe_number != IODA_INVALID_PE) {
719 rc = opal_pci_set_peltv(phb->opal_id, pdn->pe_number,
720 pe->pe_number, OPAL_REMOVE_PE_FROM_DOMAIN);
721 /* XXX What to do in case of error ? */
722 }
723 parent = parent->bus->self;
724 }
725
726 opal_pci_eeh_freeze_set(phb->opal_id, pe->pe_number,
727 OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
728
729 /* Disassociate PE in PELT */
730 rc = opal_pci_set_peltv(phb->opal_id, pe->pe_number,
731 pe->pe_number, OPAL_REMOVE_PE_FROM_DOMAIN);
732 if (rc)
733 pe_warn(pe, "OPAL error %ld remove self from PELTV\n", rc);
734 rc = opal_pci_set_pe(phb->opal_id, pe->pe_number, pe->rid,
735 bcomp, dcomp, fcomp, OPAL_UNMAP_PE);
736 if (rc)
737 pe_err(pe, "OPAL error %ld trying to setup PELT table\n", rc);
738
739 pe->pbus = NULL;
740 pe->pdev = NULL;
741 pe->parent_dev = NULL;
742
743 return 0;
744 }
745 #endif /* CONFIG_PCI_IOV */
746
747 static int pnv_ioda_configure_pe(struct pnv_phb *phb, struct pnv_ioda_pe *pe)
748 {
749 struct pci_dev *parent;
750 uint8_t bcomp, dcomp, fcomp;
751 long rc, rid_end, rid;
752
753 /* Bus validation ? */
754 if (pe->pbus) {
755 int count;
756
757 dcomp = OPAL_IGNORE_RID_DEVICE_NUMBER;
758 fcomp = OPAL_IGNORE_RID_FUNCTION_NUMBER;
759 parent = pe->pbus->self;
760 if (pe->flags & PNV_IODA_PE_BUS_ALL)
761 count = pe->pbus->busn_res.end - pe->pbus->busn_res.start + 1;
762 else
763 count = 1;
764
765 switch(count) {
766 case 1: bcomp = OpalPciBusAll; break;
767 case 2: bcomp = OpalPciBus7Bits; break;
768 case 4: bcomp = OpalPciBus6Bits; break;
769 case 8: bcomp = OpalPciBus5Bits; break;
770 case 16: bcomp = OpalPciBus4Bits; break;
771 case 32: bcomp = OpalPciBus3Bits; break;
772 default:
773 dev_err(&pe->pbus->dev, "Number of subordinate buses %d unsupported\n",
774 count);
775 /* Do an exact match only */
776 bcomp = OpalPciBusAll;
777 }
778 rid_end = pe->rid + (count << 8);
779 } else {
780 #ifdef CONFIG_PCI_IOV
781 if (pe->flags & PNV_IODA_PE_VF)
782 parent = pe->parent_dev;
783 else
784 #endif /* CONFIG_PCI_IOV */
785 parent = pe->pdev->bus->self;
786 bcomp = OpalPciBusAll;
787 dcomp = OPAL_COMPARE_RID_DEVICE_NUMBER;
788 fcomp = OPAL_COMPARE_RID_FUNCTION_NUMBER;
789 rid_end = pe->rid + 1;
790 }
791
792 /*
793 * Associate PE in PELT. We need add the PE into the
794 * corresponding PELT-V as well. Otherwise, the error
795 * originated from the PE might contribute to other
796 * PEs.
797 */
798 rc = opal_pci_set_pe(phb->opal_id, pe->pe_number, pe->rid,
799 bcomp, dcomp, fcomp, OPAL_MAP_PE);
800 if (rc) {
801 pe_err(pe, "OPAL error %ld trying to setup PELT table\n", rc);
802 return -ENXIO;
803 }
804
805 /* Configure PELTV */
806 pnv_ioda_set_peltv(phb, pe, true);
807
808 /* Setup reverse map */
809 for (rid = pe->rid; rid < rid_end; rid++)
810 phb->ioda.pe_rmap[rid] = pe->pe_number;
811
812 /* Setup one MVTs on IODA1 */
813 if (phb->type != PNV_PHB_IODA1) {
814 pe->mve_number = 0;
815 goto out;
816 }
817
818 pe->mve_number = pe->pe_number;
819 rc = opal_pci_set_mve(phb->opal_id, pe->mve_number, pe->pe_number);
820 if (rc != OPAL_SUCCESS) {
821 pe_err(pe, "OPAL error %ld setting up MVE %d\n",
822 rc, pe->mve_number);
823 pe->mve_number = -1;
824 } else {
825 rc = opal_pci_set_mve_enable(phb->opal_id,
826 pe->mve_number, OPAL_ENABLE_MVE);
827 if (rc) {
828 pe_err(pe, "OPAL error %ld enabling MVE %d\n",
829 rc, pe->mve_number);
830 pe->mve_number = -1;
831 }
832 }
833
834 out:
835 return 0;
836 }
837
838 static void pnv_ioda_link_pe_by_weight(struct pnv_phb *phb,
839 struct pnv_ioda_pe *pe)
840 {
841 struct pnv_ioda_pe *lpe;
842
843 list_for_each_entry(lpe, &phb->ioda.pe_dma_list, dma_link) {
844 if (lpe->dma_weight < pe->dma_weight) {
845 list_add_tail(&pe->dma_link, &lpe->dma_link);
846 return;
847 }
848 }
849 list_add_tail(&pe->dma_link, &phb->ioda.pe_dma_list);
850 }
851
852 static unsigned int pnv_ioda_dma_weight(struct pci_dev *dev)
853 {
854 /* This is quite simplistic. The "base" weight of a device
855 * is 10. 0 means no DMA is to be accounted for it.
856 */
857
858 /* If it's a bridge, no DMA */
859 if (dev->hdr_type != PCI_HEADER_TYPE_NORMAL)
860 return 0;
861
862 /* Reduce the weight of slow USB controllers */
863 if (dev->class == PCI_CLASS_SERIAL_USB_UHCI ||
864 dev->class == PCI_CLASS_SERIAL_USB_OHCI ||
865 dev->class == PCI_CLASS_SERIAL_USB_EHCI)
866 return 3;
867
868 /* Increase the weight of RAID (includes Obsidian) */
869 if ((dev->class >> 8) == PCI_CLASS_STORAGE_RAID)
870 return 15;
871
872 /* Default */
873 return 10;
874 }
875
876 #ifdef CONFIG_PCI_IOV
877 static int pnv_pci_vf_resource_shift(struct pci_dev *dev, int offset)
878 {
879 struct pci_dn *pdn = pci_get_pdn(dev);
880 int i;
881 struct resource *res, res2;
882 resource_size_t size;
883 u16 num_vfs;
884
885 if (!dev->is_physfn)
886 return -EINVAL;
887
888 /*
889 * "offset" is in VFs. The M64 windows are sized so that when they
890 * are segmented, each segment is the same size as the IOV BAR.
891 * Each segment is in a separate PE, and the high order bits of the
892 * address are the PE number. Therefore, each VF's BAR is in a
893 * separate PE, and changing the IOV BAR start address changes the
894 * range of PEs the VFs are in.
895 */
896 num_vfs = pdn->num_vfs;
897 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
898 res = &dev->resource[i + PCI_IOV_RESOURCES];
899 if (!res->flags || !res->parent)
900 continue;
901
902 if (!pnv_pci_is_mem_pref_64(res->flags))
903 continue;
904
905 /*
906 * The actual IOV BAR range is determined by the start address
907 * and the actual size for num_vfs VFs BAR. This check is to
908 * make sure that after shifting, the range will not overlap
909 * with another device.
910 */
911 size = pci_iov_resource_size(dev, i + PCI_IOV_RESOURCES);
912 res2.flags = res->flags;
913 res2.start = res->start + (size * offset);
914 res2.end = res2.start + (size * num_vfs) - 1;
915
916 if (res2.end > res->end) {
917 dev_err(&dev->dev, "VF BAR%d: %pR would extend past %pR (trying to enable %d VFs shifted by %d)\n",
918 i, &res2, res, num_vfs, offset);
919 return -EBUSY;
920 }
921 }
922
923 /*
924 * After doing so, there would be a "hole" in the /proc/iomem when
925 * offset is a positive value. It looks like the device return some
926 * mmio back to the system, which actually no one could use it.
927 */
928 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
929 res = &dev->resource[i + PCI_IOV_RESOURCES];
930 if (!res->flags || !res->parent)
931 continue;
932
933 if (!pnv_pci_is_mem_pref_64(res->flags))
934 continue;
935
936 size = pci_iov_resource_size(dev, i + PCI_IOV_RESOURCES);
937 res2 = *res;
938 res->start += size * offset;
939
940 dev_info(&dev->dev, "VF BAR%d: %pR shifted to %pR (enabling %d VFs shifted by %d)\n",
941 i, &res2, res, num_vfs, offset);
942 pci_update_resource(dev, i + PCI_IOV_RESOURCES);
943 }
944 return 0;
945 }
946 #endif /* CONFIG_PCI_IOV */
947
948 #if 0
949 static struct pnv_ioda_pe *pnv_ioda_setup_dev_PE(struct pci_dev *dev)
950 {
951 struct pci_controller *hose = pci_bus_to_host(dev->bus);
952 struct pnv_phb *phb = hose->private_data;
953 struct pci_dn *pdn = pci_get_pdn(dev);
954 struct pnv_ioda_pe *pe;
955 int pe_num;
956
957 if (!pdn) {
958 pr_err("%s: Device tree node not associated properly\n",
959 pci_name(dev));
960 return NULL;
961 }
962 if (pdn->pe_number != IODA_INVALID_PE)
963 return NULL;
964
965 /* PE#0 has been pre-set */
966 if (dev->bus->number == 0)
967 pe_num = 0;
968 else
969 pe_num = pnv_ioda_alloc_pe(phb);
970 if (pe_num == IODA_INVALID_PE) {
971 pr_warning("%s: Not enough PE# available, disabling device\n",
972 pci_name(dev));
973 return NULL;
974 }
975
976 /* NOTE: We get only one ref to the pci_dev for the pdn, not for the
977 * pointer in the PE data structure, both should be destroyed at the
978 * same time. However, this needs to be looked at more closely again
979 * once we actually start removing things (Hotplug, SR-IOV, ...)
980 *
981 * At some point we want to remove the PDN completely anyways
982 */
983 pe = &phb->ioda.pe_array[pe_num];
984 pci_dev_get(dev);
985 pdn->pcidev = dev;
986 pdn->pe_number = pe_num;
987 pe->pdev = dev;
988 pe->pbus = NULL;
989 pe->tce32_seg = -1;
990 pe->mve_number = -1;
991 pe->rid = dev->bus->number << 8 | pdn->devfn;
992
993 pe_info(pe, "Associated device to PE\n");
994
995 if (pnv_ioda_configure_pe(phb, pe)) {
996 /* XXX What do we do here ? */
997 if (pe_num)
998 pnv_ioda_free_pe(phb, pe_num);
999 pdn->pe_number = IODA_INVALID_PE;
1000 pe->pdev = NULL;
1001 pci_dev_put(dev);
1002 return NULL;
1003 }
1004
1005 /* Assign a DMA weight to the device */
1006 pe->dma_weight = pnv_ioda_dma_weight(dev);
1007 if (pe->dma_weight != 0) {
1008 phb->ioda.dma_weight += pe->dma_weight;
1009 phb->ioda.dma_pe_count++;
1010 }
1011
1012 /* Link the PE */
1013 pnv_ioda_link_pe_by_weight(phb, pe);
1014
1015 return pe;
1016 }
1017 #endif /* Useful for SRIOV case */
1018
1019 static void pnv_ioda_setup_same_PE(struct pci_bus *bus, struct pnv_ioda_pe *pe)
1020 {
1021 struct pci_dev *dev;
1022
1023 list_for_each_entry(dev, &bus->devices, bus_list) {
1024 struct pci_dn *pdn = pci_get_pdn(dev);
1025
1026 if (pdn == NULL) {
1027 pr_warn("%s: No device node associated with device !\n",
1028 pci_name(dev));
1029 continue;
1030 }
1031 pdn->pe_number = pe->pe_number;
1032 pe->dma_weight += pnv_ioda_dma_weight(dev);
1033 if ((pe->flags & PNV_IODA_PE_BUS_ALL) && dev->subordinate)
1034 pnv_ioda_setup_same_PE(dev->subordinate, pe);
1035 }
1036 }
1037
1038 /*
1039 * There're 2 types of PCI bus sensitive PEs: One that is compromised of
1040 * single PCI bus. Another one that contains the primary PCI bus and its
1041 * subordinate PCI devices and buses. The second type of PE is normally
1042 * orgiriated by PCIe-to-PCI bridge or PLX switch downstream ports.
1043 */
1044 static void pnv_ioda_setup_bus_PE(struct pci_bus *bus, int all)
1045 {
1046 struct pci_controller *hose = pci_bus_to_host(bus);
1047 struct pnv_phb *phb = hose->private_data;
1048 struct pnv_ioda_pe *pe;
1049 int pe_num = IODA_INVALID_PE;
1050
1051 /* Check if PE is determined by M64 */
1052 if (phb->pick_m64_pe)
1053 pe_num = phb->pick_m64_pe(phb, bus, all);
1054
1055 /* The PE number isn't pinned by M64 */
1056 if (pe_num == IODA_INVALID_PE)
1057 pe_num = pnv_ioda_alloc_pe(phb);
1058
1059 if (pe_num == IODA_INVALID_PE) {
1060 pr_warning("%s: Not enough PE# available for PCI bus %04x:%02x\n",
1061 __func__, pci_domain_nr(bus), bus->number);
1062 return;
1063 }
1064
1065 pe = &phb->ioda.pe_array[pe_num];
1066 pe->flags |= (all ? PNV_IODA_PE_BUS_ALL : PNV_IODA_PE_BUS);
1067 pe->pbus = bus;
1068 pe->pdev = NULL;
1069 pe->tce32_seg = -1;
1070 pe->mve_number = -1;
1071 pe->rid = bus->busn_res.start << 8;
1072 pe->dma_weight = 0;
1073
1074 if (all)
1075 pe_info(pe, "Secondary bus %d..%d associated with PE#%d\n",
1076 bus->busn_res.start, bus->busn_res.end, pe_num);
1077 else
1078 pe_info(pe, "Secondary bus %d associated with PE#%d\n",
1079 bus->busn_res.start, pe_num);
1080
1081 if (pnv_ioda_configure_pe(phb, pe)) {
1082 /* XXX What do we do here ? */
1083 if (pe_num)
1084 pnv_ioda_free_pe(phb, pe_num);
1085 pe->pbus = NULL;
1086 return;
1087 }
1088
1089 pe->tce32_table = kzalloc_node(sizeof(struct iommu_table),
1090 GFP_KERNEL, hose->node);
1091 pe->tce32_table->data = pe;
1092
1093 /* Associate it with all child devices */
1094 pnv_ioda_setup_same_PE(bus, pe);
1095
1096 /* Put PE to the list */
1097 list_add_tail(&pe->list, &phb->ioda.pe_list);
1098
1099 /* Account for one DMA PE if at least one DMA capable device exist
1100 * below the bridge
1101 */
1102 if (pe->dma_weight != 0) {
1103 phb->ioda.dma_weight += pe->dma_weight;
1104 phb->ioda.dma_pe_count++;
1105 }
1106
1107 /* Link the PE */
1108 pnv_ioda_link_pe_by_weight(phb, pe);
1109 }
1110
1111 static void pnv_ioda_setup_PEs(struct pci_bus *bus)
1112 {
1113 struct pci_dev *dev;
1114
1115 pnv_ioda_setup_bus_PE(bus, 0);
1116
1117 list_for_each_entry(dev, &bus->devices, bus_list) {
1118 if (dev->subordinate) {
1119 if (pci_pcie_type(dev) == PCI_EXP_TYPE_PCI_BRIDGE)
1120 pnv_ioda_setup_bus_PE(dev->subordinate, 1);
1121 else
1122 pnv_ioda_setup_PEs(dev->subordinate);
1123 }
1124 }
1125 }
1126
1127 /*
1128 * Configure PEs so that the downstream PCI buses and devices
1129 * could have their associated PE#. Unfortunately, we didn't
1130 * figure out the way to identify the PLX bridge yet. So we
1131 * simply put the PCI bus and the subordinate behind the root
1132 * port to PE# here. The game rule here is expected to be changed
1133 * as soon as we can detected PLX bridge correctly.
1134 */
1135 static void pnv_pci_ioda_setup_PEs(void)
1136 {
1137 struct pci_controller *hose, *tmp;
1138 struct pnv_phb *phb;
1139
1140 list_for_each_entry_safe(hose, tmp, &hose_list, list_node) {
1141 phb = hose->private_data;
1142
1143 /* M64 layout might affect PE allocation */
1144 if (phb->reserve_m64_pe)
1145 phb->reserve_m64_pe(phb);
1146
1147 pnv_ioda_setup_PEs(hose->bus);
1148 }
1149 }
1150
1151 #ifdef CONFIG_PCI_IOV
1152 static int pnv_pci_vf_release_m64(struct pci_dev *pdev)
1153 {
1154 struct pci_bus *bus;
1155 struct pci_controller *hose;
1156 struct pnv_phb *phb;
1157 struct pci_dn *pdn;
1158 int i, j;
1159
1160 bus = pdev->bus;
1161 hose = pci_bus_to_host(bus);
1162 phb = hose->private_data;
1163 pdn = pci_get_pdn(pdev);
1164
1165 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++)
1166 for (j = 0; j < M64_PER_IOV; j++) {
1167 if (pdn->m64_wins[i][j] == IODA_INVALID_M64)
1168 continue;
1169 opal_pci_phb_mmio_enable(phb->opal_id,
1170 OPAL_M64_WINDOW_TYPE, pdn->m64_wins[i][j], 0);
1171 clear_bit(pdn->m64_wins[i][j], &phb->ioda.m64_bar_alloc);
1172 pdn->m64_wins[i][j] = IODA_INVALID_M64;
1173 }
1174
1175 return 0;
1176 }
1177
1178 static int pnv_pci_vf_assign_m64(struct pci_dev *pdev, u16 num_vfs)
1179 {
1180 struct pci_bus *bus;
1181 struct pci_controller *hose;
1182 struct pnv_phb *phb;
1183 struct pci_dn *pdn;
1184 unsigned int win;
1185 struct resource *res;
1186 int i, j;
1187 int64_t rc;
1188 int total_vfs;
1189 resource_size_t size, start;
1190 int pe_num;
1191 int vf_groups;
1192 int vf_per_group;
1193
1194 bus = pdev->bus;
1195 hose = pci_bus_to_host(bus);
1196 phb = hose->private_data;
1197 pdn = pci_get_pdn(pdev);
1198 total_vfs = pci_sriov_get_totalvfs(pdev);
1199
1200 /* Initialize the m64_wins to IODA_INVALID_M64 */
1201 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++)
1202 for (j = 0; j < M64_PER_IOV; j++)
1203 pdn->m64_wins[i][j] = IODA_INVALID_M64;
1204
1205 if (pdn->m64_per_iov == M64_PER_IOV) {
1206 vf_groups = (num_vfs <= M64_PER_IOV) ? num_vfs: M64_PER_IOV;
1207 vf_per_group = (num_vfs <= M64_PER_IOV)? 1:
1208 roundup_pow_of_two(num_vfs) / pdn->m64_per_iov;
1209 } else {
1210 vf_groups = 1;
1211 vf_per_group = 1;
1212 }
1213
1214 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
1215 res = &pdev->resource[i + PCI_IOV_RESOURCES];
1216 if (!res->flags || !res->parent)
1217 continue;
1218
1219 if (!pnv_pci_is_mem_pref_64(res->flags))
1220 continue;
1221
1222 for (j = 0; j < vf_groups; j++) {
1223 do {
1224 win = find_next_zero_bit(&phb->ioda.m64_bar_alloc,
1225 phb->ioda.m64_bar_idx + 1, 0);
1226
1227 if (win >= phb->ioda.m64_bar_idx + 1)
1228 goto m64_failed;
1229 } while (test_and_set_bit(win, &phb->ioda.m64_bar_alloc));
1230
1231 pdn->m64_wins[i][j] = win;
1232
1233 if (pdn->m64_per_iov == M64_PER_IOV) {
1234 size = pci_iov_resource_size(pdev,
1235 PCI_IOV_RESOURCES + i);
1236 size = size * vf_per_group;
1237 start = res->start + size * j;
1238 } else {
1239 size = resource_size(res);
1240 start = res->start;
1241 }
1242
1243 /* Map the M64 here */
1244 if (pdn->m64_per_iov == M64_PER_IOV) {
1245 pe_num = pdn->offset + j;
1246 rc = opal_pci_map_pe_mmio_window(phb->opal_id,
1247 pe_num, OPAL_M64_WINDOW_TYPE,
1248 pdn->m64_wins[i][j], 0);
1249 }
1250
1251 rc = opal_pci_set_phb_mem_window(phb->opal_id,
1252 OPAL_M64_WINDOW_TYPE,
1253 pdn->m64_wins[i][j],
1254 start,
1255 0, /* unused */
1256 size);
1257
1258
1259 if (rc != OPAL_SUCCESS) {
1260 dev_err(&pdev->dev, "Failed to map M64 window #%d: %lld\n",
1261 win, rc);
1262 goto m64_failed;
1263 }
1264
1265 if (pdn->m64_per_iov == M64_PER_IOV)
1266 rc = opal_pci_phb_mmio_enable(phb->opal_id,
1267 OPAL_M64_WINDOW_TYPE, pdn->m64_wins[i][j], 2);
1268 else
1269 rc = opal_pci_phb_mmio_enable(phb->opal_id,
1270 OPAL_M64_WINDOW_TYPE, pdn->m64_wins[i][j], 1);
1271
1272 if (rc != OPAL_SUCCESS) {
1273 dev_err(&pdev->dev, "Failed to enable M64 window #%d: %llx\n",
1274 win, rc);
1275 goto m64_failed;
1276 }
1277 }
1278 }
1279 return 0;
1280
1281 m64_failed:
1282 pnv_pci_vf_release_m64(pdev);
1283 return -EBUSY;
1284 }
1285
1286 static void pnv_pci_ioda2_release_dma_pe(struct pci_dev *dev, struct pnv_ioda_pe *pe)
1287 {
1288 struct pci_bus *bus;
1289 struct pci_controller *hose;
1290 struct pnv_phb *phb;
1291 struct iommu_table *tbl;
1292 unsigned long addr;
1293 int64_t rc;
1294
1295 bus = dev->bus;
1296 hose = pci_bus_to_host(bus);
1297 phb = hose->private_data;
1298 tbl = pe->tce32_table;
1299 addr = tbl->it_base;
1300
1301 opal_pci_map_pe_dma_window(phb->opal_id, pe->pe_number,
1302 pe->pe_number << 1, 1, __pa(addr),
1303 0, 0x1000);
1304
1305 rc = opal_pci_map_pe_dma_window_real(pe->phb->opal_id,
1306 pe->pe_number,
1307 (pe->pe_number << 1) + 1,
1308 pe->tce_bypass_base,
1309 0);
1310 if (rc)
1311 pe_warn(pe, "OPAL error %ld release DMA window\n", rc);
1312
1313 iommu_free_table(tbl, of_node_full_name(dev->dev.of_node));
1314 free_pages(addr, get_order(TCE32_TABLE_SIZE));
1315 pe->tce32_table = NULL;
1316 }
1317
1318 static void pnv_ioda_release_vf_PE(struct pci_dev *pdev, u16 num_vfs)
1319 {
1320 struct pci_bus *bus;
1321 struct pci_controller *hose;
1322 struct pnv_phb *phb;
1323 struct pnv_ioda_pe *pe, *pe_n;
1324 struct pci_dn *pdn;
1325 u16 vf_index;
1326 int64_t rc;
1327
1328 bus = pdev->bus;
1329 hose = pci_bus_to_host(bus);
1330 phb = hose->private_data;
1331 pdn = pci_get_pdn(pdev);
1332
1333 if (!pdev->is_physfn)
1334 return;
1335
1336 if (pdn->m64_per_iov == M64_PER_IOV && num_vfs > M64_PER_IOV) {
1337 int vf_group;
1338 int vf_per_group;
1339 int vf_index1;
1340
1341 vf_per_group = roundup_pow_of_two(num_vfs) / pdn->m64_per_iov;
1342
1343 for (vf_group = 0; vf_group < M64_PER_IOV; vf_group++)
1344 for (vf_index = vf_group * vf_per_group;
1345 vf_index < (vf_group + 1) * vf_per_group &&
1346 vf_index < num_vfs;
1347 vf_index++)
1348 for (vf_index1 = vf_group * vf_per_group;
1349 vf_index1 < (vf_group + 1) * vf_per_group &&
1350 vf_index1 < num_vfs;
1351 vf_index1++){
1352
1353 rc = opal_pci_set_peltv(phb->opal_id,
1354 pdn->offset + vf_index,
1355 pdn->offset + vf_index1,
1356 OPAL_REMOVE_PE_FROM_DOMAIN);
1357
1358 if (rc)
1359 dev_warn(&pdev->dev, "%s: Failed to unlink same group PE#%d(%lld)\n",
1360 __func__,
1361 pdn->offset + vf_index1, rc);
1362 }
1363 }
1364
1365 list_for_each_entry_safe(pe, pe_n, &phb->ioda.pe_list, list) {
1366 if (pe->parent_dev != pdev)
1367 continue;
1368
1369 pnv_pci_ioda2_release_dma_pe(pdev, pe);
1370
1371 /* Remove from list */
1372 mutex_lock(&phb->ioda.pe_list_mutex);
1373 list_del(&pe->list);
1374 mutex_unlock(&phb->ioda.pe_list_mutex);
1375
1376 pnv_ioda_deconfigure_pe(phb, pe);
1377
1378 pnv_ioda_free_pe(phb, pe->pe_number);
1379 }
1380 }
1381
1382 void pnv_pci_sriov_disable(struct pci_dev *pdev)
1383 {
1384 struct pci_bus *bus;
1385 struct pci_controller *hose;
1386 struct pnv_phb *phb;
1387 struct pci_dn *pdn;
1388 struct pci_sriov *iov;
1389 u16 num_vfs;
1390
1391 bus = pdev->bus;
1392 hose = pci_bus_to_host(bus);
1393 phb = hose->private_data;
1394 pdn = pci_get_pdn(pdev);
1395 iov = pdev->sriov;
1396 num_vfs = pdn->num_vfs;
1397
1398 /* Release VF PEs */
1399 pnv_ioda_release_vf_PE(pdev, num_vfs);
1400
1401 if (phb->type == PNV_PHB_IODA2) {
1402 if (pdn->m64_per_iov == 1)
1403 pnv_pci_vf_resource_shift(pdev, -pdn->offset);
1404
1405 /* Release M64 windows */
1406 pnv_pci_vf_release_m64(pdev);
1407
1408 /* Release PE numbers */
1409 bitmap_clear(phb->ioda.pe_alloc, pdn->offset, num_vfs);
1410 pdn->offset = 0;
1411 }
1412 }
1413
1414 static void pnv_pci_ioda2_setup_dma_pe(struct pnv_phb *phb,
1415 struct pnv_ioda_pe *pe);
1416 static void pnv_ioda_setup_vf_PE(struct pci_dev *pdev, u16 num_vfs)
1417 {
1418 struct pci_bus *bus;
1419 struct pci_controller *hose;
1420 struct pnv_phb *phb;
1421 struct pnv_ioda_pe *pe;
1422 int pe_num;
1423 u16 vf_index;
1424 struct pci_dn *pdn;
1425 int64_t rc;
1426
1427 bus = pdev->bus;
1428 hose = pci_bus_to_host(bus);
1429 phb = hose->private_data;
1430 pdn = pci_get_pdn(pdev);
1431
1432 if (!pdev->is_physfn)
1433 return;
1434
1435 /* Reserve PE for each VF */
1436 for (vf_index = 0; vf_index < num_vfs; vf_index++) {
1437 pe_num = pdn->offset + vf_index;
1438
1439 pe = &phb->ioda.pe_array[pe_num];
1440 pe->pe_number = pe_num;
1441 pe->phb = phb;
1442 pe->flags = PNV_IODA_PE_VF;
1443 pe->pbus = NULL;
1444 pe->parent_dev = pdev;
1445 pe->tce32_seg = -1;
1446 pe->mve_number = -1;
1447 pe->rid = (pci_iov_virtfn_bus(pdev, vf_index) << 8) |
1448 pci_iov_virtfn_devfn(pdev, vf_index);
1449
1450 pe_info(pe, "VF %04d:%02d:%02d.%d associated with PE#%d\n",
1451 hose->global_number, pdev->bus->number,
1452 PCI_SLOT(pci_iov_virtfn_devfn(pdev, vf_index)),
1453 PCI_FUNC(pci_iov_virtfn_devfn(pdev, vf_index)), pe_num);
1454
1455 if (pnv_ioda_configure_pe(phb, pe)) {
1456 /* XXX What do we do here ? */
1457 if (pe_num)
1458 pnv_ioda_free_pe(phb, pe_num);
1459 pe->pdev = NULL;
1460 continue;
1461 }
1462
1463 pe->tce32_table = kzalloc_node(sizeof(struct iommu_table),
1464 GFP_KERNEL, hose->node);
1465 pe->tce32_table->data = pe;
1466
1467 /* Put PE to the list */
1468 mutex_lock(&phb->ioda.pe_list_mutex);
1469 list_add_tail(&pe->list, &phb->ioda.pe_list);
1470 mutex_unlock(&phb->ioda.pe_list_mutex);
1471
1472 pnv_pci_ioda2_setup_dma_pe(phb, pe);
1473 }
1474
1475 if (pdn->m64_per_iov == M64_PER_IOV && num_vfs > M64_PER_IOV) {
1476 int vf_group;
1477 int vf_per_group;
1478 int vf_index1;
1479
1480 vf_per_group = roundup_pow_of_two(num_vfs) / pdn->m64_per_iov;
1481
1482 for (vf_group = 0; vf_group < M64_PER_IOV; vf_group++) {
1483 for (vf_index = vf_group * vf_per_group;
1484 vf_index < (vf_group + 1) * vf_per_group &&
1485 vf_index < num_vfs;
1486 vf_index++) {
1487 for (vf_index1 = vf_group * vf_per_group;
1488 vf_index1 < (vf_group + 1) * vf_per_group &&
1489 vf_index1 < num_vfs;
1490 vf_index1++) {
1491
1492 rc = opal_pci_set_peltv(phb->opal_id,
1493 pdn->offset + vf_index,
1494 pdn->offset + vf_index1,
1495 OPAL_ADD_PE_TO_DOMAIN);
1496
1497 if (rc)
1498 dev_warn(&pdev->dev, "%s: Failed to link same group PE#%d(%lld)\n",
1499 __func__,
1500 pdn->offset + vf_index1, rc);
1501 }
1502 }
1503 }
1504 }
1505 }
1506
1507 int pnv_pci_sriov_enable(struct pci_dev *pdev, u16 num_vfs)
1508 {
1509 struct pci_bus *bus;
1510 struct pci_controller *hose;
1511 struct pnv_phb *phb;
1512 struct pci_dn *pdn;
1513 int ret;
1514
1515 bus = pdev->bus;
1516 hose = pci_bus_to_host(bus);
1517 phb = hose->private_data;
1518 pdn = pci_get_pdn(pdev);
1519
1520 if (phb->type == PNV_PHB_IODA2) {
1521 /* Calculate available PE for required VFs */
1522 mutex_lock(&phb->ioda.pe_alloc_mutex);
1523 pdn->offset = bitmap_find_next_zero_area(
1524 phb->ioda.pe_alloc, phb->ioda.total_pe,
1525 0, num_vfs, 0);
1526 if (pdn->offset >= phb->ioda.total_pe) {
1527 mutex_unlock(&phb->ioda.pe_alloc_mutex);
1528 dev_info(&pdev->dev, "Failed to enable VF%d\n", num_vfs);
1529 pdn->offset = 0;
1530 return -EBUSY;
1531 }
1532 bitmap_set(phb->ioda.pe_alloc, pdn->offset, num_vfs);
1533 pdn->num_vfs = num_vfs;
1534 mutex_unlock(&phb->ioda.pe_alloc_mutex);
1535
1536 /* Assign M64 window accordingly */
1537 ret = pnv_pci_vf_assign_m64(pdev, num_vfs);
1538 if (ret) {
1539 dev_info(&pdev->dev, "Not enough M64 window resources\n");
1540 goto m64_failed;
1541 }
1542
1543 /*
1544 * When using one M64 BAR to map one IOV BAR, we need to shift
1545 * the IOV BAR according to the PE# allocated to the VFs.
1546 * Otherwise, the PE# for the VF will conflict with others.
1547 */
1548 if (pdn->m64_per_iov == 1) {
1549 ret = pnv_pci_vf_resource_shift(pdev, pdn->offset);
1550 if (ret)
1551 goto m64_failed;
1552 }
1553 }
1554
1555 /* Setup VF PEs */
1556 pnv_ioda_setup_vf_PE(pdev, num_vfs);
1557
1558 return 0;
1559
1560 m64_failed:
1561 bitmap_clear(phb->ioda.pe_alloc, pdn->offset, num_vfs);
1562 pdn->offset = 0;
1563
1564 return ret;
1565 }
1566
1567 int pcibios_sriov_disable(struct pci_dev *pdev)
1568 {
1569 pnv_pci_sriov_disable(pdev);
1570
1571 /* Release PCI data */
1572 remove_dev_pci_data(pdev);
1573 return 0;
1574 }
1575
1576 int pcibios_sriov_enable(struct pci_dev *pdev, u16 num_vfs)
1577 {
1578 /* Allocate PCI data */
1579 add_dev_pci_data(pdev);
1580
1581 pnv_pci_sriov_enable(pdev, num_vfs);
1582 return 0;
1583 }
1584 #endif /* CONFIG_PCI_IOV */
1585
1586 static void pnv_pci_ioda_dma_dev_setup(struct pnv_phb *phb, struct pci_dev *pdev)
1587 {
1588 struct pci_dn *pdn = pci_get_pdn(pdev);
1589 struct pnv_ioda_pe *pe;
1590
1591 /*
1592 * The function can be called while the PE#
1593 * hasn't been assigned. Do nothing for the
1594 * case.
1595 */
1596 if (!pdn || pdn->pe_number == IODA_INVALID_PE)
1597 return;
1598
1599 pe = &phb->ioda.pe_array[pdn->pe_number];
1600 WARN_ON(get_dma_ops(&pdev->dev) != &dma_iommu_ops);
1601 set_iommu_table_base_and_group(&pdev->dev, pe->tce32_table);
1602 }
1603
1604 static int pnv_pci_ioda_dma_set_mask(struct pci_dev *pdev, u64 dma_mask)
1605 {
1606 struct pci_controller *hose = pci_bus_to_host(pdev->bus);
1607 struct pnv_phb *phb = hose->private_data;
1608 struct pci_dn *pdn = pci_get_pdn(pdev);
1609 struct pnv_ioda_pe *pe;
1610 uint64_t top;
1611 bool bypass = false;
1612
1613 if (WARN_ON(!pdn || pdn->pe_number == IODA_INVALID_PE))
1614 return -ENODEV;;
1615
1616 pe = &phb->ioda.pe_array[pdn->pe_number];
1617 if (pe->tce_bypass_enabled) {
1618 top = pe->tce_bypass_base + memblock_end_of_DRAM() - 1;
1619 bypass = (dma_mask >= top);
1620 }
1621
1622 if (bypass) {
1623 dev_info(&pdev->dev, "Using 64-bit DMA iommu bypass\n");
1624 set_dma_ops(&pdev->dev, &dma_direct_ops);
1625 set_dma_offset(&pdev->dev, pe->tce_bypass_base);
1626 } else {
1627 dev_info(&pdev->dev, "Using 32-bit DMA via iommu\n");
1628 set_dma_ops(&pdev->dev, &dma_iommu_ops);
1629 set_iommu_table_base(&pdev->dev, pe->tce32_table);
1630 }
1631 *pdev->dev.dma_mask = dma_mask;
1632 return 0;
1633 }
1634
1635 static u64 pnv_pci_ioda_dma_get_required_mask(struct pnv_phb *phb,
1636 struct pci_dev *pdev)
1637 {
1638 struct pci_dn *pdn = pci_get_pdn(pdev);
1639 struct pnv_ioda_pe *pe;
1640 u64 end, mask;
1641
1642 if (WARN_ON(!pdn || pdn->pe_number == IODA_INVALID_PE))
1643 return 0;
1644
1645 pe = &phb->ioda.pe_array[pdn->pe_number];
1646 if (!pe->tce_bypass_enabled)
1647 return __dma_get_required_mask(&pdev->dev);
1648
1649
1650 end = pe->tce_bypass_base + memblock_end_of_DRAM();
1651 mask = 1ULL << (fls64(end) - 1);
1652 mask += mask - 1;
1653
1654 return mask;
1655 }
1656
1657 static void pnv_ioda_setup_bus_dma(struct pnv_ioda_pe *pe,
1658 struct pci_bus *bus,
1659 bool add_to_iommu_group)
1660 {
1661 struct pci_dev *dev;
1662
1663 list_for_each_entry(dev, &bus->devices, bus_list) {
1664 if (add_to_iommu_group)
1665 set_iommu_table_base_and_group(&dev->dev,
1666 pe->tce32_table);
1667 else
1668 set_iommu_table_base(&dev->dev, pe->tce32_table);
1669
1670 if (dev->subordinate)
1671 pnv_ioda_setup_bus_dma(pe, dev->subordinate,
1672 add_to_iommu_group);
1673 }
1674 }
1675
1676 static void pnv_pci_ioda1_tce_invalidate(struct pnv_ioda_pe *pe,
1677 struct iommu_table *tbl,
1678 __be64 *startp, __be64 *endp, bool rm)
1679 {
1680 __be64 __iomem *invalidate = rm ?
1681 (__be64 __iomem *)pe->tce_inval_reg_phys :
1682 (__be64 __iomem *)tbl->it_index;
1683 unsigned long start, end, inc;
1684 const unsigned shift = tbl->it_page_shift;
1685
1686 start = __pa(startp);
1687 end = __pa(endp);
1688
1689 /* BML uses this case for p6/p7/galaxy2: Shift addr and put in node */
1690 if (tbl->it_busno) {
1691 start <<= shift;
1692 end <<= shift;
1693 inc = 128ull << shift;
1694 start |= tbl->it_busno;
1695 end |= tbl->it_busno;
1696 } else if (tbl->it_type & TCE_PCI_SWINV_PAIR) {
1697 /* p7ioc-style invalidation, 2 TCEs per write */
1698 start |= (1ull << 63);
1699 end |= (1ull << 63);
1700 inc = 16;
1701 } else {
1702 /* Default (older HW) */
1703 inc = 128;
1704 }
1705
1706 end |= inc - 1; /* round up end to be different than start */
1707
1708 mb(); /* Ensure above stores are visible */
1709 while (start <= end) {
1710 if (rm)
1711 __raw_rm_writeq(cpu_to_be64(start), invalidate);
1712 else
1713 __raw_writeq(cpu_to_be64(start), invalidate);
1714 start += inc;
1715 }
1716
1717 /*
1718 * The iommu layer will do another mb() for us on build()
1719 * and we don't care on free()
1720 */
1721 }
1722
1723 static void pnv_pci_ioda2_tce_invalidate(struct pnv_ioda_pe *pe,
1724 struct iommu_table *tbl,
1725 __be64 *startp, __be64 *endp, bool rm)
1726 {
1727 unsigned long start, end, inc;
1728 __be64 __iomem *invalidate = rm ?
1729 (__be64 __iomem *)pe->tce_inval_reg_phys :
1730 (__be64 __iomem *)tbl->it_index;
1731 const unsigned shift = tbl->it_page_shift;
1732
1733 /* We'll invalidate DMA address in PE scope */
1734 start = 0x2ull << 60;
1735 start |= (pe->pe_number & 0xFF);
1736 end = start;
1737
1738 /* Figure out the start, end and step */
1739 inc = tbl->it_offset + (((u64)startp - tbl->it_base) / sizeof(u64));
1740 start |= (inc << shift);
1741 inc = tbl->it_offset + (((u64)endp - tbl->it_base) / sizeof(u64));
1742 end |= (inc << shift);
1743 inc = (0x1ull << shift);
1744 mb();
1745
1746 while (start <= end) {
1747 if (rm)
1748 __raw_rm_writeq(cpu_to_be64(start), invalidate);
1749 else
1750 __raw_writeq(cpu_to_be64(start), invalidate);
1751 start += inc;
1752 }
1753 }
1754
1755 void pnv_pci_ioda_tce_invalidate(struct iommu_table *tbl,
1756 __be64 *startp, __be64 *endp, bool rm)
1757 {
1758 struct pnv_ioda_pe *pe = tbl->data;
1759 struct pnv_phb *phb = pe->phb;
1760
1761 if (phb->type == PNV_PHB_IODA1)
1762 pnv_pci_ioda1_tce_invalidate(pe, tbl, startp, endp, rm);
1763 else
1764 pnv_pci_ioda2_tce_invalidate(pe, tbl, startp, endp, rm);
1765 }
1766
1767 static void pnv_pci_ioda_setup_dma_pe(struct pnv_phb *phb,
1768 struct pnv_ioda_pe *pe, unsigned int base,
1769 unsigned int segs)
1770 {
1771
1772 struct page *tce_mem = NULL;
1773 const __be64 *swinvp;
1774 struct iommu_table *tbl;
1775 unsigned int i;
1776 int64_t rc;
1777 void *addr;
1778
1779 /* XXX FIXME: Handle 64-bit only DMA devices */
1780 /* XXX FIXME: Provide 64-bit DMA facilities & non-4K TCE tables etc.. */
1781 /* XXX FIXME: Allocate multi-level tables on PHB3 */
1782
1783 /* We shouldn't already have a 32-bit DMA associated */
1784 if (WARN_ON(pe->tce32_seg >= 0))
1785 return;
1786
1787 /* Grab a 32-bit TCE table */
1788 pe->tce32_seg = base;
1789 pe_info(pe, " Setting up 32-bit TCE table at %08x..%08x\n",
1790 (base << 28), ((base + segs) << 28) - 1);
1791
1792 /* XXX Currently, we allocate one big contiguous table for the
1793 * TCEs. We only really need one chunk per 256M of TCE space
1794 * (ie per segment) but that's an optimization for later, it
1795 * requires some added smarts with our get/put_tce implementation
1796 */
1797 tce_mem = alloc_pages_node(phb->hose->node, GFP_KERNEL,
1798 get_order(TCE32_TABLE_SIZE * segs));
1799 if (!tce_mem) {
1800 pe_err(pe, " Failed to allocate a 32-bit TCE memory\n");
1801 goto fail;
1802 }
1803 addr = page_address(tce_mem);
1804 memset(addr, 0, TCE32_TABLE_SIZE * segs);
1805
1806 /* Configure HW */
1807 for (i = 0; i < segs; i++) {
1808 rc = opal_pci_map_pe_dma_window(phb->opal_id,
1809 pe->pe_number,
1810 base + i, 1,
1811 __pa(addr) + TCE32_TABLE_SIZE * i,
1812 TCE32_TABLE_SIZE, 0x1000);
1813 if (rc) {
1814 pe_err(pe, " Failed to configure 32-bit TCE table,"
1815 " err %ld\n", rc);
1816 goto fail;
1817 }
1818 }
1819
1820 /* Setup linux iommu table */
1821 tbl = pe->tce32_table;
1822 pnv_pci_setup_iommu_table(tbl, addr, TCE32_TABLE_SIZE * segs,
1823 base << 28, IOMMU_PAGE_SHIFT_4K);
1824
1825 /* OPAL variant of P7IOC SW invalidated TCEs */
1826 swinvp = of_get_property(phb->hose->dn, "ibm,opal-tce-kill", NULL);
1827 if (swinvp) {
1828 /* We need a couple more fields -- an address and a data
1829 * to or. Since the bus is only printed out on table free
1830 * errors, and on the first pass the data will be a relative
1831 * bus number, print that out instead.
1832 */
1833 pe->tce_inval_reg_phys = be64_to_cpup(swinvp);
1834 tbl->it_index = (unsigned long)ioremap(pe->tce_inval_reg_phys,
1835 8);
1836 tbl->it_type |= (TCE_PCI_SWINV_CREATE |
1837 TCE_PCI_SWINV_FREE |
1838 TCE_PCI_SWINV_PAIR);
1839 }
1840 iommu_init_table(tbl, phb->hose->node);
1841
1842 if (pe->flags & PNV_IODA_PE_DEV) {
1843 iommu_register_group(tbl, phb->hose->global_number,
1844 pe->pe_number);
1845 set_iommu_table_base_and_group(&pe->pdev->dev, tbl);
1846 } else if (pe->flags & (PNV_IODA_PE_BUS | PNV_IODA_PE_BUS_ALL)) {
1847 iommu_register_group(tbl, phb->hose->global_number,
1848 pe->pe_number);
1849 pnv_ioda_setup_bus_dma(pe, pe->pbus, true);
1850 } else if (pe->flags & PNV_IODA_PE_VF) {
1851 iommu_register_group(tbl, phb->hose->global_number,
1852 pe->pe_number);
1853 }
1854
1855 return;
1856 fail:
1857 /* XXX Failure: Try to fallback to 64-bit only ? */
1858 if (pe->tce32_seg >= 0)
1859 pe->tce32_seg = -1;
1860 if (tce_mem)
1861 __free_pages(tce_mem, get_order(TCE32_TABLE_SIZE * segs));
1862 }
1863
1864 static void pnv_pci_ioda2_set_bypass(struct iommu_table *tbl, bool enable)
1865 {
1866 struct pnv_ioda_pe *pe = tbl->data;
1867 uint16_t window_id = (pe->pe_number << 1 ) + 1;
1868 int64_t rc;
1869
1870 pe_info(pe, "%sabling 64-bit DMA bypass\n", enable ? "En" : "Dis");
1871 if (enable) {
1872 phys_addr_t top = memblock_end_of_DRAM();
1873
1874 top = roundup_pow_of_two(top);
1875 rc = opal_pci_map_pe_dma_window_real(pe->phb->opal_id,
1876 pe->pe_number,
1877 window_id,
1878 pe->tce_bypass_base,
1879 top);
1880 } else {
1881 rc = opal_pci_map_pe_dma_window_real(pe->phb->opal_id,
1882 pe->pe_number,
1883 window_id,
1884 pe->tce_bypass_base,
1885 0);
1886
1887 /*
1888 * EEH needs the mapping between IOMMU table and group
1889 * of those VFIO/KVM pass-through devices. We can postpone
1890 * resetting DMA ops until the DMA mask is configured in
1891 * host side.
1892 */
1893 if (pe->pdev)
1894 set_iommu_table_base(&pe->pdev->dev, tbl);
1895 else
1896 pnv_ioda_setup_bus_dma(pe, pe->pbus, false);
1897 }
1898 if (rc)
1899 pe_err(pe, "OPAL error %lld configuring bypass window\n", rc);
1900 else
1901 pe->tce_bypass_enabled = enable;
1902 }
1903
1904 static void pnv_pci_ioda2_setup_bypass_pe(struct pnv_phb *phb,
1905 struct pnv_ioda_pe *pe)
1906 {
1907 /* TVE #1 is selected by PCI address bit 59 */
1908 pe->tce_bypass_base = 1ull << 59;
1909
1910 /* Install set_bypass callback for VFIO */
1911 pe->tce32_table->set_bypass = pnv_pci_ioda2_set_bypass;
1912
1913 /* Enable bypass by default */
1914 pnv_pci_ioda2_set_bypass(pe->tce32_table, true);
1915 }
1916
1917 static void pnv_pci_ioda2_setup_dma_pe(struct pnv_phb *phb,
1918 struct pnv_ioda_pe *pe)
1919 {
1920 struct page *tce_mem = NULL;
1921 void *addr;
1922 const __be64 *swinvp;
1923 struct iommu_table *tbl;
1924 unsigned int tce_table_size, end;
1925 int64_t rc;
1926
1927 /* We shouldn't already have a 32-bit DMA associated */
1928 if (WARN_ON(pe->tce32_seg >= 0))
1929 return;
1930
1931 /* The PE will reserve all possible 32-bits space */
1932 pe->tce32_seg = 0;
1933 end = (1 << ilog2(phb->ioda.m32_pci_base));
1934 tce_table_size = (end / 0x1000) * 8;
1935 pe_info(pe, "Setting up 32-bit TCE table at 0..%08x\n",
1936 end);
1937
1938 /* Allocate TCE table */
1939 tce_mem = alloc_pages_node(phb->hose->node, GFP_KERNEL,
1940 get_order(tce_table_size));
1941 if (!tce_mem) {
1942 pe_err(pe, "Failed to allocate a 32-bit TCE memory\n");
1943 goto fail;
1944 }
1945 addr = page_address(tce_mem);
1946 memset(addr, 0, tce_table_size);
1947
1948 /*
1949 * Map TCE table through TVT. The TVE index is the PE number
1950 * shifted by 1 bit for 32-bits DMA space.
1951 */
1952 rc = opal_pci_map_pe_dma_window(phb->opal_id, pe->pe_number,
1953 pe->pe_number << 1, 1, __pa(addr),
1954 tce_table_size, 0x1000);
1955 if (rc) {
1956 pe_err(pe, "Failed to configure 32-bit TCE table,"
1957 " err %ld\n", rc);
1958 goto fail;
1959 }
1960
1961 /* Setup linux iommu table */
1962 tbl = pe->tce32_table;
1963 pnv_pci_setup_iommu_table(tbl, addr, tce_table_size, 0,
1964 IOMMU_PAGE_SHIFT_4K);
1965
1966 /* OPAL variant of PHB3 invalidated TCEs */
1967 swinvp = of_get_property(phb->hose->dn, "ibm,opal-tce-kill", NULL);
1968 if (swinvp) {
1969 /* We need a couple more fields -- an address and a data
1970 * to or. Since the bus is only printed out on table free
1971 * errors, and on the first pass the data will be a relative
1972 * bus number, print that out instead.
1973 */
1974 pe->tce_inval_reg_phys = be64_to_cpup(swinvp);
1975 tbl->it_index = (unsigned long)ioremap(pe->tce_inval_reg_phys,
1976 8);
1977 tbl->it_type |= (TCE_PCI_SWINV_CREATE | TCE_PCI_SWINV_FREE);
1978 }
1979 iommu_init_table(tbl, phb->hose->node);
1980
1981 if (pe->flags & PNV_IODA_PE_DEV) {
1982 iommu_register_group(tbl, phb->hose->global_number,
1983 pe->pe_number);
1984 set_iommu_table_base_and_group(&pe->pdev->dev, tbl);
1985 } else if (pe->flags & (PNV_IODA_PE_BUS | PNV_IODA_PE_BUS_ALL)) {
1986 iommu_register_group(tbl, phb->hose->global_number,
1987 pe->pe_number);
1988 pnv_ioda_setup_bus_dma(pe, pe->pbus, true);
1989 } else if (pe->flags & PNV_IODA_PE_VF) {
1990 iommu_register_group(tbl, phb->hose->global_number,
1991 pe->pe_number);
1992 }
1993
1994 /* Also create a bypass window */
1995 if (!pnv_iommu_bypass_disabled)
1996 pnv_pci_ioda2_setup_bypass_pe(phb, pe);
1997
1998 return;
1999 fail:
2000 if (pe->tce32_seg >= 0)
2001 pe->tce32_seg = -1;
2002 if (tce_mem)
2003 __free_pages(tce_mem, get_order(tce_table_size));
2004 }
2005
2006 static void pnv_ioda_setup_dma(struct pnv_phb *phb)
2007 {
2008 struct pci_controller *hose = phb->hose;
2009 unsigned int residual, remaining, segs, tw, base;
2010 struct pnv_ioda_pe *pe;
2011
2012 /* If we have more PE# than segments available, hand out one
2013 * per PE until we run out and let the rest fail. If not,
2014 * then we assign at least one segment per PE, plus more based
2015 * on the amount of devices under that PE
2016 */
2017 if (phb->ioda.dma_pe_count > phb->ioda.tce32_count)
2018 residual = 0;
2019 else
2020 residual = phb->ioda.tce32_count -
2021 phb->ioda.dma_pe_count;
2022
2023 pr_info("PCI: Domain %04x has %ld available 32-bit DMA segments\n",
2024 hose->global_number, phb->ioda.tce32_count);
2025 pr_info("PCI: %d PE# for a total weight of %d\n",
2026 phb->ioda.dma_pe_count, phb->ioda.dma_weight);
2027
2028 /* Walk our PE list and configure their DMA segments, hand them
2029 * out one base segment plus any residual segments based on
2030 * weight
2031 */
2032 remaining = phb->ioda.tce32_count;
2033 tw = phb->ioda.dma_weight;
2034 base = 0;
2035 list_for_each_entry(pe, &phb->ioda.pe_dma_list, dma_link) {
2036 if (!pe->dma_weight)
2037 continue;
2038 if (!remaining) {
2039 pe_warn(pe, "No DMA32 resources available\n");
2040 continue;
2041 }
2042 segs = 1;
2043 if (residual) {
2044 segs += ((pe->dma_weight * residual) + (tw / 2)) / tw;
2045 if (segs > remaining)
2046 segs = remaining;
2047 }
2048
2049 /*
2050 * For IODA2 compliant PHB3, we needn't care about the weight.
2051 * The all available 32-bits DMA space will be assigned to
2052 * the specific PE.
2053 */
2054 if (phb->type == PNV_PHB_IODA1) {
2055 pe_info(pe, "DMA weight %d, assigned %d DMA32 segments\n",
2056 pe->dma_weight, segs);
2057 pnv_pci_ioda_setup_dma_pe(phb, pe, base, segs);
2058 } else {
2059 pe_info(pe, "Assign DMA32 space\n");
2060 segs = 0;
2061 pnv_pci_ioda2_setup_dma_pe(phb, pe);
2062 }
2063
2064 remaining -= segs;
2065 base += segs;
2066 }
2067 }
2068
2069 #ifdef CONFIG_PCI_MSI
2070 static void pnv_ioda2_msi_eoi(struct irq_data *d)
2071 {
2072 unsigned int hw_irq = (unsigned int)irqd_to_hwirq(d);
2073 struct irq_chip *chip = irq_data_get_irq_chip(d);
2074 struct pnv_phb *phb = container_of(chip, struct pnv_phb,
2075 ioda.irq_chip);
2076 int64_t rc;
2077
2078 rc = opal_pci_msi_eoi(phb->opal_id, hw_irq);
2079 WARN_ON_ONCE(rc);
2080
2081 icp_native_eoi(d);
2082 }
2083
2084
2085 static void set_msi_irq_chip(struct pnv_phb *phb, unsigned int virq)
2086 {
2087 struct irq_data *idata;
2088 struct irq_chip *ichip;
2089
2090 if (phb->type != PNV_PHB_IODA2)
2091 return;
2092
2093 if (!phb->ioda.irq_chip_init) {
2094 /*
2095 * First time we setup an MSI IRQ, we need to setup the
2096 * corresponding IRQ chip to route correctly.
2097 */
2098 idata = irq_get_irq_data(virq);
2099 ichip = irq_data_get_irq_chip(idata);
2100 phb->ioda.irq_chip_init = 1;
2101 phb->ioda.irq_chip = *ichip;
2102 phb->ioda.irq_chip.irq_eoi = pnv_ioda2_msi_eoi;
2103 }
2104 irq_set_chip(virq, &phb->ioda.irq_chip);
2105 }
2106
2107 #ifdef CONFIG_CXL_BASE
2108
2109 struct device_node *pnv_pci_get_phb_node(struct pci_dev *dev)
2110 {
2111 struct pci_controller *hose = pci_bus_to_host(dev->bus);
2112
2113 return of_node_get(hose->dn);
2114 }
2115 EXPORT_SYMBOL(pnv_pci_get_phb_node);
2116
2117 int pnv_phb_to_cxl_mode(struct pci_dev *dev, uint64_t mode)
2118 {
2119 struct pci_controller *hose = pci_bus_to_host(dev->bus);
2120 struct pnv_phb *phb = hose->private_data;
2121 struct pnv_ioda_pe *pe;
2122 int rc;
2123
2124 pe = pnv_ioda_get_pe(dev);
2125 if (!pe)
2126 return -ENODEV;
2127
2128 pe_info(pe, "Switching PHB to CXL\n");
2129
2130 rc = opal_pci_set_phb_cxl_mode(phb->opal_id, mode, pe->pe_number);
2131 if (rc)
2132 dev_err(&dev->dev, "opal_pci_set_phb_cxl_mode failed: %i\n", rc);
2133
2134 return rc;
2135 }
2136 EXPORT_SYMBOL(pnv_phb_to_cxl_mode);
2137
2138 /* Find PHB for cxl dev and allocate MSI hwirqs?
2139 * Returns the absolute hardware IRQ number
2140 */
2141 int pnv_cxl_alloc_hwirqs(struct pci_dev *dev, int num)
2142 {
2143 struct pci_controller *hose = pci_bus_to_host(dev->bus);
2144 struct pnv_phb *phb = hose->private_data;
2145 int hwirq = msi_bitmap_alloc_hwirqs(&phb->msi_bmp, num);
2146
2147 if (hwirq < 0) {
2148 dev_warn(&dev->dev, "Failed to find a free MSI\n");
2149 return -ENOSPC;
2150 }
2151
2152 return phb->msi_base + hwirq;
2153 }
2154 EXPORT_SYMBOL(pnv_cxl_alloc_hwirqs);
2155
2156 void pnv_cxl_release_hwirqs(struct pci_dev *dev, int hwirq, int num)
2157 {
2158 struct pci_controller *hose = pci_bus_to_host(dev->bus);
2159 struct pnv_phb *phb = hose->private_data;
2160
2161 msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq - phb->msi_base, num);
2162 }
2163 EXPORT_SYMBOL(pnv_cxl_release_hwirqs);
2164
2165 void pnv_cxl_release_hwirq_ranges(struct cxl_irq_ranges *irqs,
2166 struct pci_dev *dev)
2167 {
2168 struct pci_controller *hose = pci_bus_to_host(dev->bus);
2169 struct pnv_phb *phb = hose->private_data;
2170 int i, hwirq;
2171
2172 for (i = 1; i < CXL_IRQ_RANGES; i++) {
2173 if (!irqs->range[i])
2174 continue;
2175 pr_devel("cxl release irq range 0x%x: offset: 0x%lx limit: %ld\n",
2176 i, irqs->offset[i],
2177 irqs->range[i]);
2178 hwirq = irqs->offset[i] - phb->msi_base;
2179 msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq,
2180 irqs->range[i]);
2181 }
2182 }
2183 EXPORT_SYMBOL(pnv_cxl_release_hwirq_ranges);
2184
2185 int pnv_cxl_alloc_hwirq_ranges(struct cxl_irq_ranges *irqs,
2186 struct pci_dev *dev, int num)
2187 {
2188 struct pci_controller *hose = pci_bus_to_host(dev->bus);
2189 struct pnv_phb *phb = hose->private_data;
2190 int i, hwirq, try;
2191
2192 memset(irqs, 0, sizeof(struct cxl_irq_ranges));
2193
2194 /* 0 is reserved for the multiplexed PSL DSI interrupt */
2195 for (i = 1; i < CXL_IRQ_RANGES && num; i++) {
2196 try = num;
2197 while (try) {
2198 hwirq = msi_bitmap_alloc_hwirqs(&phb->msi_bmp, try);
2199 if (hwirq >= 0)
2200 break;
2201 try /= 2;
2202 }
2203 if (!try)
2204 goto fail;
2205
2206 irqs->offset[i] = phb->msi_base + hwirq;
2207 irqs->range[i] = try;
2208 pr_devel("cxl alloc irq range 0x%x: offset: 0x%lx limit: %li\n",
2209 i, irqs->offset[i], irqs->range[i]);
2210 num -= try;
2211 }
2212 if (num)
2213 goto fail;
2214
2215 return 0;
2216 fail:
2217 pnv_cxl_release_hwirq_ranges(irqs, dev);
2218 return -ENOSPC;
2219 }
2220 EXPORT_SYMBOL(pnv_cxl_alloc_hwirq_ranges);
2221
2222 int pnv_cxl_get_irq_count(struct pci_dev *dev)
2223 {
2224 struct pci_controller *hose = pci_bus_to_host(dev->bus);
2225 struct pnv_phb *phb = hose->private_data;
2226
2227 return phb->msi_bmp.irq_count;
2228 }
2229 EXPORT_SYMBOL(pnv_cxl_get_irq_count);
2230
2231 int pnv_cxl_ioda_msi_setup(struct pci_dev *dev, unsigned int hwirq,
2232 unsigned int virq)
2233 {
2234 struct pci_controller *hose = pci_bus_to_host(dev->bus);
2235 struct pnv_phb *phb = hose->private_data;
2236 unsigned int xive_num = hwirq - phb->msi_base;
2237 struct pnv_ioda_pe *pe;
2238 int rc;
2239
2240 if (!(pe = pnv_ioda_get_pe(dev)))
2241 return -ENODEV;
2242
2243 /* Assign XIVE to PE */
2244 rc = opal_pci_set_xive_pe(phb->opal_id, pe->pe_number, xive_num);
2245 if (rc) {
2246 pe_warn(pe, "%s: OPAL error %d setting msi_base 0x%x "
2247 "hwirq 0x%x XIVE 0x%x PE\n",
2248 pci_name(dev), rc, phb->msi_base, hwirq, xive_num);
2249 return -EIO;
2250 }
2251 set_msi_irq_chip(phb, virq);
2252
2253 return 0;
2254 }
2255 EXPORT_SYMBOL(pnv_cxl_ioda_msi_setup);
2256 #endif
2257
2258 static int pnv_pci_ioda_msi_setup(struct pnv_phb *phb, struct pci_dev *dev,
2259 unsigned int hwirq, unsigned int virq,
2260 unsigned int is_64, struct msi_msg *msg)
2261 {
2262 struct pnv_ioda_pe *pe = pnv_ioda_get_pe(dev);
2263 unsigned int xive_num = hwirq - phb->msi_base;
2264 __be32 data;
2265 int rc;
2266
2267 /* No PE assigned ? bail out ... no MSI for you ! */
2268 if (pe == NULL)
2269 return -ENXIO;
2270
2271 /* Check if we have an MVE */
2272 if (pe->mve_number < 0)
2273 return -ENXIO;
2274
2275 /* Force 32-bit MSI on some broken devices */
2276 if (dev->no_64bit_msi)
2277 is_64 = 0;
2278
2279 /* Assign XIVE to PE */
2280 rc = opal_pci_set_xive_pe(phb->opal_id, pe->pe_number, xive_num);
2281 if (rc) {
2282 pr_warn("%s: OPAL error %d setting XIVE %d PE\n",
2283 pci_name(dev), rc, xive_num);
2284 return -EIO;
2285 }
2286
2287 if (is_64) {
2288 __be64 addr64;
2289
2290 rc = opal_get_msi_64(phb->opal_id, pe->mve_number, xive_num, 1,
2291 &addr64, &data);
2292 if (rc) {
2293 pr_warn("%s: OPAL error %d getting 64-bit MSI data\n",
2294 pci_name(dev), rc);
2295 return -EIO;
2296 }
2297 msg->address_hi = be64_to_cpu(addr64) >> 32;
2298 msg->address_lo = be64_to_cpu(addr64) & 0xfffffffful;
2299 } else {
2300 __be32 addr32;
2301
2302 rc = opal_get_msi_32(phb->opal_id, pe->mve_number, xive_num, 1,
2303 &addr32, &data);
2304 if (rc) {
2305 pr_warn("%s: OPAL error %d getting 32-bit MSI data\n",
2306 pci_name(dev), rc);
2307 return -EIO;
2308 }
2309 msg->address_hi = 0;
2310 msg->address_lo = be32_to_cpu(addr32);
2311 }
2312 msg->data = be32_to_cpu(data);
2313
2314 set_msi_irq_chip(phb, virq);
2315
2316 pr_devel("%s: %s-bit MSI on hwirq %x (xive #%d),"
2317 " address=%x_%08x data=%x PE# %d\n",
2318 pci_name(dev), is_64 ? "64" : "32", hwirq, xive_num,
2319 msg->address_hi, msg->address_lo, data, pe->pe_number);
2320
2321 return 0;
2322 }
2323
2324 static void pnv_pci_init_ioda_msis(struct pnv_phb *phb)
2325 {
2326 unsigned int count;
2327 const __be32 *prop = of_get_property(phb->hose->dn,
2328 "ibm,opal-msi-ranges", NULL);
2329 if (!prop) {
2330 /* BML Fallback */
2331 prop = of_get_property(phb->hose->dn, "msi-ranges", NULL);
2332 }
2333 if (!prop)
2334 return;
2335
2336 phb->msi_base = be32_to_cpup(prop);
2337 count = be32_to_cpup(prop + 1);
2338 if (msi_bitmap_alloc(&phb->msi_bmp, count, phb->hose->dn)) {
2339 pr_err("PCI %d: Failed to allocate MSI bitmap !\n",
2340 phb->hose->global_number);
2341 return;
2342 }
2343
2344 phb->msi_setup = pnv_pci_ioda_msi_setup;
2345 phb->msi32_support = 1;
2346 pr_info(" Allocated bitmap for %d MSIs (base IRQ 0x%x)\n",
2347 count, phb->msi_base);
2348 }
2349 #else
2350 static void pnv_pci_init_ioda_msis(struct pnv_phb *phb) { }
2351 #endif /* CONFIG_PCI_MSI */
2352
2353 #ifdef CONFIG_PCI_IOV
2354 static void pnv_pci_ioda_fixup_iov_resources(struct pci_dev *pdev)
2355 {
2356 struct pci_controller *hose;
2357 struct pnv_phb *phb;
2358 struct resource *res;
2359 int i;
2360 resource_size_t size;
2361 struct pci_dn *pdn;
2362 int mul, total_vfs;
2363
2364 if (!pdev->is_physfn || pdev->is_added)
2365 return;
2366
2367 hose = pci_bus_to_host(pdev->bus);
2368 phb = hose->private_data;
2369
2370 pdn = pci_get_pdn(pdev);
2371 pdn->vfs_expanded = 0;
2372
2373 total_vfs = pci_sriov_get_totalvfs(pdev);
2374 pdn->m64_per_iov = 1;
2375 mul = phb->ioda.total_pe;
2376
2377 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
2378 res = &pdev->resource[i + PCI_IOV_RESOURCES];
2379 if (!res->flags || res->parent)
2380 continue;
2381 if (!pnv_pci_is_mem_pref_64(res->flags)) {
2382 dev_warn(&pdev->dev, " non M64 VF BAR%d: %pR\n",
2383 i, res);
2384 continue;
2385 }
2386
2387 size = pci_iov_resource_size(pdev, i + PCI_IOV_RESOURCES);
2388
2389 /* bigger than 64M */
2390 if (size > (1 << 26)) {
2391 dev_info(&pdev->dev, "PowerNV: VF BAR%d: %pR IOV size is bigger than 64M, roundup power2\n",
2392 i, res);
2393 pdn->m64_per_iov = M64_PER_IOV;
2394 mul = roundup_pow_of_two(total_vfs);
2395 break;
2396 }
2397 }
2398
2399 for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) {
2400 res = &pdev->resource[i + PCI_IOV_RESOURCES];
2401 if (!res->flags || res->parent)
2402 continue;
2403 if (!pnv_pci_is_mem_pref_64(res->flags)) {
2404 dev_warn(&pdev->dev, "Skipping expanding VF BAR%d: %pR\n",
2405 i, res);
2406 continue;
2407 }
2408
2409 dev_dbg(&pdev->dev, " Fixing VF BAR%d: %pR to\n", i, res);
2410 size = pci_iov_resource_size(pdev, i + PCI_IOV_RESOURCES);
2411 res->end = res->start + size * mul - 1;
2412 dev_dbg(&pdev->dev, " %pR\n", res);
2413 dev_info(&pdev->dev, "VF BAR%d: %pR (expanded to %d VFs for PE alignment)",
2414 i, res, mul);
2415 }
2416 pdn->vfs_expanded = mul;
2417 }
2418 #endif /* CONFIG_PCI_IOV */
2419
2420 /*
2421 * This function is supposed to be called on basis of PE from top
2422 * to bottom style. So the the I/O or MMIO segment assigned to
2423 * parent PE could be overrided by its child PEs if necessary.
2424 */
2425 static void pnv_ioda_setup_pe_seg(struct pci_controller *hose,
2426 struct pnv_ioda_pe *pe)
2427 {
2428 struct pnv_phb *phb = hose->private_data;
2429 struct pci_bus_region region;
2430 struct resource *res;
2431 int i, index;
2432 int rc;
2433
2434 /*
2435 * NOTE: We only care PCI bus based PE for now. For PCI
2436 * device based PE, for example SRIOV sensitive VF should
2437 * be figured out later.
2438 */
2439 BUG_ON(!(pe->flags & (PNV_IODA_PE_BUS | PNV_IODA_PE_BUS_ALL)));
2440
2441 pci_bus_for_each_resource(pe->pbus, res, i) {
2442 if (!res || !res->flags ||
2443 res->start > res->end)
2444 continue;
2445
2446 if (res->flags & IORESOURCE_IO) {
2447 region.start = res->start - phb->ioda.io_pci_base;
2448 region.end = res->end - phb->ioda.io_pci_base;
2449 index = region.start / phb->ioda.io_segsize;
2450
2451 while (index < phb->ioda.total_pe &&
2452 region.start <= region.end) {
2453 phb->ioda.io_segmap[index] = pe->pe_number;
2454 rc = opal_pci_map_pe_mmio_window(phb->opal_id,
2455 pe->pe_number, OPAL_IO_WINDOW_TYPE, 0, index);
2456 if (rc != OPAL_SUCCESS) {
2457 pr_err("%s: OPAL error %d when mapping IO "
2458 "segment #%d to PE#%d\n",
2459 __func__, rc, index, pe->pe_number);
2460 break;
2461 }
2462
2463 region.start += phb->ioda.io_segsize;
2464 index++;
2465 }
2466 } else if ((res->flags & IORESOURCE_MEM) &&
2467 !pnv_pci_is_mem_pref_64(res->flags)) {
2468 region.start = res->start -
2469 hose->mem_offset[0] -
2470 phb->ioda.m32_pci_base;
2471 region.end = res->end -
2472 hose->mem_offset[0] -
2473 phb->ioda.m32_pci_base;
2474 index = region.start / phb->ioda.m32_segsize;
2475
2476 while (index < phb->ioda.total_pe &&
2477 region.start <= region.end) {
2478 phb->ioda.m32_segmap[index] = pe->pe_number;
2479 rc = opal_pci_map_pe_mmio_window(phb->opal_id,
2480 pe->pe_number, OPAL_M32_WINDOW_TYPE, 0, index);
2481 if (rc != OPAL_SUCCESS) {
2482 pr_err("%s: OPAL error %d when mapping M32 "
2483 "segment#%d to PE#%d",
2484 __func__, rc, index, pe->pe_number);
2485 break;
2486 }
2487
2488 region.start += phb->ioda.m32_segsize;
2489 index++;
2490 }
2491 }
2492 }
2493 }
2494
2495 static void pnv_pci_ioda_setup_seg(void)
2496 {
2497 struct pci_controller *tmp, *hose;
2498 struct pnv_phb *phb;
2499 struct pnv_ioda_pe *pe;
2500
2501 list_for_each_entry_safe(hose, tmp, &hose_list, list_node) {
2502 phb = hose->private_data;
2503 list_for_each_entry(pe, &phb->ioda.pe_list, list) {
2504 pnv_ioda_setup_pe_seg(hose, pe);
2505 }
2506 }
2507 }
2508
2509 static void pnv_pci_ioda_setup_DMA(void)
2510 {
2511 struct pci_controller *hose, *tmp;
2512 struct pnv_phb *phb;
2513
2514 list_for_each_entry_safe(hose, tmp, &hose_list, list_node) {
2515 pnv_ioda_setup_dma(hose->private_data);
2516
2517 /* Mark the PHB initialization done */
2518 phb = hose->private_data;
2519 phb->initialized = 1;
2520 }
2521 }
2522
2523 static void pnv_pci_ioda_create_dbgfs(void)
2524 {
2525 #ifdef CONFIG_DEBUG_FS
2526 struct pci_controller *hose, *tmp;
2527 struct pnv_phb *phb;
2528 char name[16];
2529
2530 list_for_each_entry_safe(hose, tmp, &hose_list, list_node) {
2531 phb = hose->private_data;
2532
2533 sprintf(name, "PCI%04x", hose->global_number);
2534 phb->dbgfs = debugfs_create_dir(name, powerpc_debugfs_root);
2535 if (!phb->dbgfs)
2536 pr_warning("%s: Error on creating debugfs on PHB#%x\n",
2537 __func__, hose->global_number);
2538 }
2539 #endif /* CONFIG_DEBUG_FS */
2540 }
2541
2542 static void pnv_pci_ioda_fixup(void)
2543 {
2544 pnv_pci_ioda_setup_PEs();
2545 pnv_pci_ioda_setup_seg();
2546 pnv_pci_ioda_setup_DMA();
2547
2548 pnv_pci_ioda_create_dbgfs();
2549
2550 #ifdef CONFIG_EEH
2551 eeh_init();
2552 eeh_addr_cache_build();
2553 #endif
2554 }
2555
2556 /*
2557 * Returns the alignment for I/O or memory windows for P2P
2558 * bridges. That actually depends on how PEs are segmented.
2559 * For now, we return I/O or M32 segment size for PE sensitive
2560 * P2P bridges. Otherwise, the default values (4KiB for I/O,
2561 * 1MiB for memory) will be returned.
2562 *
2563 * The current PCI bus might be put into one PE, which was
2564 * create against the parent PCI bridge. For that case, we
2565 * needn't enlarge the alignment so that we can save some
2566 * resources.
2567 */
2568 static resource_size_t pnv_pci_window_alignment(struct pci_bus *bus,
2569 unsigned long type)
2570 {
2571 struct pci_dev *bridge;
2572 struct pci_controller *hose = pci_bus_to_host(bus);
2573 struct pnv_phb *phb = hose->private_data;
2574 int num_pci_bridges = 0;
2575
2576 bridge = bus->self;
2577 while (bridge) {
2578 if (pci_pcie_type(bridge) == PCI_EXP_TYPE_PCI_BRIDGE) {
2579 num_pci_bridges++;
2580 if (num_pci_bridges >= 2)
2581 return 1;
2582 }
2583
2584 bridge = bridge->bus->self;
2585 }
2586
2587 /* We fail back to M32 if M64 isn't supported */
2588 if (phb->ioda.m64_segsize &&
2589 pnv_pci_is_mem_pref_64(type))
2590 return phb->ioda.m64_segsize;
2591 if (type & IORESOURCE_MEM)
2592 return phb->ioda.m32_segsize;
2593
2594 return phb->ioda.io_segsize;
2595 }
2596
2597 #ifdef CONFIG_PCI_IOV
2598 static resource_size_t pnv_pci_iov_resource_alignment(struct pci_dev *pdev,
2599 int resno)
2600 {
2601 struct pci_dn *pdn = pci_get_pdn(pdev);
2602 resource_size_t align, iov_align;
2603
2604 iov_align = resource_size(&pdev->resource[resno]);
2605 if (iov_align)
2606 return iov_align;
2607
2608 align = pci_iov_resource_size(pdev, resno);
2609 if (pdn->vfs_expanded)
2610 return pdn->vfs_expanded * align;
2611
2612 return align;
2613 }
2614 #endif /* CONFIG_PCI_IOV */
2615
2616 /* Prevent enabling devices for which we couldn't properly
2617 * assign a PE
2618 */
2619 static bool pnv_pci_enable_device_hook(struct pci_dev *dev)
2620 {
2621 struct pci_controller *hose = pci_bus_to_host(dev->bus);
2622 struct pnv_phb *phb = hose->private_data;
2623 struct pci_dn *pdn;
2624
2625 /* The function is probably called while the PEs have
2626 * not be created yet. For example, resource reassignment
2627 * during PCI probe period. We just skip the check if
2628 * PEs isn't ready.
2629 */
2630 if (!phb->initialized)
2631 return true;
2632
2633 pdn = pci_get_pdn(dev);
2634 if (!pdn || pdn->pe_number == IODA_INVALID_PE)
2635 return false;
2636
2637 return true;
2638 }
2639
2640 static u32 pnv_ioda_bdfn_to_pe(struct pnv_phb *phb, struct pci_bus *bus,
2641 u32 devfn)
2642 {
2643 return phb->ioda.pe_rmap[(bus->number << 8) | devfn];
2644 }
2645
2646 static void pnv_pci_ioda_shutdown(struct pci_controller *hose)
2647 {
2648 struct pnv_phb *phb = hose->private_data;
2649
2650 opal_pci_reset(phb->opal_id, OPAL_RESET_PCI_IODA_TABLE,
2651 OPAL_ASSERT_RESET);
2652 }
2653
2654 static const struct pci_controller_ops pnv_pci_ioda_controller_ops = {
2655 .dma_dev_setup = pnv_pci_dma_dev_setup,
2656 #ifdef CONFIG_PCI_MSI
2657 .setup_msi_irqs = pnv_setup_msi_irqs,
2658 .teardown_msi_irqs = pnv_teardown_msi_irqs,
2659 #endif
2660 .enable_device_hook = pnv_pci_enable_device_hook,
2661 .window_alignment = pnv_pci_window_alignment,
2662 .reset_secondary_bus = pnv_pci_reset_secondary_bus,
2663 .dma_set_mask = pnv_pci_ioda_dma_set_mask,
2664 .shutdown = pnv_pci_ioda_shutdown,
2665 };
2666
2667 static void __init pnv_pci_init_ioda_phb(struct device_node *np,
2668 u64 hub_id, int ioda_type)
2669 {
2670 struct pci_controller *hose;
2671 struct pnv_phb *phb;
2672 unsigned long size, m32map_off, pemap_off, iomap_off = 0;
2673 const __be64 *prop64;
2674 const __be32 *prop32;
2675 int len;
2676 u64 phb_id;
2677 void *aux;
2678 long rc;
2679
2680 pr_info("Initializing IODA%d OPAL PHB %s\n", ioda_type, np->full_name);
2681
2682 prop64 = of_get_property(np, "ibm,opal-phbid", NULL);
2683 if (!prop64) {
2684 pr_err(" Missing \"ibm,opal-phbid\" property !\n");
2685 return;
2686 }
2687 phb_id = be64_to_cpup(prop64);
2688 pr_debug(" PHB-ID : 0x%016llx\n", phb_id);
2689
2690 phb = memblock_virt_alloc(sizeof(struct pnv_phb), 0);
2691
2692 /* Allocate PCI controller */
2693 phb->hose = hose = pcibios_alloc_controller(np);
2694 if (!phb->hose) {
2695 pr_err(" Can't allocate PCI controller for %s\n",
2696 np->full_name);
2697 memblock_free(__pa(phb), sizeof(struct pnv_phb));
2698 return;
2699 }
2700
2701 spin_lock_init(&phb->lock);
2702 prop32 = of_get_property(np, "bus-range", &len);
2703 if (prop32 && len == 8) {
2704 hose->first_busno = be32_to_cpu(prop32[0]);
2705 hose->last_busno = be32_to_cpu(prop32[1]);
2706 } else {
2707 pr_warn(" Broken <bus-range> on %s\n", np->full_name);
2708 hose->first_busno = 0;
2709 hose->last_busno = 0xff;
2710 }
2711 hose->private_data = phb;
2712 phb->hub_id = hub_id;
2713 phb->opal_id = phb_id;
2714 phb->type = ioda_type;
2715 mutex_init(&phb->ioda.pe_alloc_mutex);
2716
2717 /* Detect specific models for error handling */
2718 if (of_device_is_compatible(np, "ibm,p7ioc-pciex"))
2719 phb->model = PNV_PHB_MODEL_P7IOC;
2720 else if (of_device_is_compatible(np, "ibm,power8-pciex"))
2721 phb->model = PNV_PHB_MODEL_PHB3;
2722 else
2723 phb->model = PNV_PHB_MODEL_UNKNOWN;
2724
2725 /* Parse 32-bit and IO ranges (if any) */
2726 pci_process_bridge_OF_ranges(hose, np, !hose->global_number);
2727
2728 /* Get registers */
2729 phb->regs = of_iomap(np, 0);
2730 if (phb->regs == NULL)
2731 pr_err(" Failed to map registers !\n");
2732
2733 /* Initialize more IODA stuff */
2734 phb->ioda.total_pe = 1;
2735 prop32 = of_get_property(np, "ibm,opal-num-pes", NULL);
2736 if (prop32)
2737 phb->ioda.total_pe = be32_to_cpup(prop32);
2738 prop32 = of_get_property(np, "ibm,opal-reserved-pe", NULL);
2739 if (prop32)
2740 phb->ioda.reserved_pe = be32_to_cpup(prop32);
2741
2742 /* Parse 64-bit MMIO range */
2743 pnv_ioda_parse_m64_window(phb);
2744
2745 phb->ioda.m32_size = resource_size(&hose->mem_resources[0]);
2746 /* FW Has already off top 64k of M32 space (MSI space) */
2747 phb->ioda.m32_size += 0x10000;
2748
2749 phb->ioda.m32_segsize = phb->ioda.m32_size / phb->ioda.total_pe;
2750 phb->ioda.m32_pci_base = hose->mem_resources[0].start - hose->mem_offset[0];
2751 phb->ioda.io_size = hose->pci_io_size;
2752 phb->ioda.io_segsize = phb->ioda.io_size / phb->ioda.total_pe;
2753 phb->ioda.io_pci_base = 0; /* XXX calculate this ? */
2754
2755 /* Allocate aux data & arrays. We don't have IO ports on PHB3 */
2756 size = _ALIGN_UP(phb->ioda.total_pe / 8, sizeof(unsigned long));
2757 m32map_off = size;
2758 size += phb->ioda.total_pe * sizeof(phb->ioda.m32_segmap[0]);
2759 if (phb->type == PNV_PHB_IODA1) {
2760 iomap_off = size;
2761 size += phb->ioda.total_pe * sizeof(phb->ioda.io_segmap[0]);
2762 }
2763 pemap_off = size;
2764 size += phb->ioda.total_pe * sizeof(struct pnv_ioda_pe);
2765 aux = memblock_virt_alloc(size, 0);
2766 phb->ioda.pe_alloc = aux;
2767 phb->ioda.m32_segmap = aux + m32map_off;
2768 if (phb->type == PNV_PHB_IODA1)
2769 phb->ioda.io_segmap = aux + iomap_off;
2770 phb->ioda.pe_array = aux + pemap_off;
2771 set_bit(phb->ioda.reserved_pe, phb->ioda.pe_alloc);
2772
2773 INIT_LIST_HEAD(&phb->ioda.pe_dma_list);
2774 INIT_LIST_HEAD(&phb->ioda.pe_list);
2775 mutex_init(&phb->ioda.pe_list_mutex);
2776
2777 /* Calculate how many 32-bit TCE segments we have */
2778 phb->ioda.tce32_count = phb->ioda.m32_pci_base >> 28;
2779
2780 #if 0 /* We should really do that ... */
2781 rc = opal_pci_set_phb_mem_window(opal->phb_id,
2782 window_type,
2783 window_num,
2784 starting_real_address,
2785 starting_pci_address,
2786 segment_size);
2787 #endif
2788
2789 pr_info(" %03d (%03d) PE's M32: 0x%x [segment=0x%x]\n",
2790 phb->ioda.total_pe, phb->ioda.reserved_pe,
2791 phb->ioda.m32_size, phb->ioda.m32_segsize);
2792 if (phb->ioda.m64_size)
2793 pr_info(" M64: 0x%lx [segment=0x%lx]\n",
2794 phb->ioda.m64_size, phb->ioda.m64_segsize);
2795 if (phb->ioda.io_size)
2796 pr_info(" IO: 0x%x [segment=0x%x]\n",
2797 phb->ioda.io_size, phb->ioda.io_segsize);
2798
2799
2800 phb->hose->ops = &pnv_pci_ops;
2801 phb->get_pe_state = pnv_ioda_get_pe_state;
2802 phb->freeze_pe = pnv_ioda_freeze_pe;
2803 phb->unfreeze_pe = pnv_ioda_unfreeze_pe;
2804
2805 /* Setup RID -> PE mapping function */
2806 phb->bdfn_to_pe = pnv_ioda_bdfn_to_pe;
2807
2808 /* Setup TCEs */
2809 phb->dma_dev_setup = pnv_pci_ioda_dma_dev_setup;
2810 phb->dma_get_required_mask = pnv_pci_ioda_dma_get_required_mask;
2811
2812 /* Setup MSI support */
2813 pnv_pci_init_ioda_msis(phb);
2814
2815 /*
2816 * We pass the PCI probe flag PCI_REASSIGN_ALL_RSRC here
2817 * to let the PCI core do resource assignment. It's supposed
2818 * that the PCI core will do correct I/O and MMIO alignment
2819 * for the P2P bridge bars so that each PCI bus (excluding
2820 * the child P2P bridges) can form individual PE.
2821 */
2822 ppc_md.pcibios_fixup = pnv_pci_ioda_fixup;
2823 hose->controller_ops = pnv_pci_ioda_controller_ops;
2824
2825 #ifdef CONFIG_PCI_IOV
2826 ppc_md.pcibios_fixup_sriov = pnv_pci_ioda_fixup_iov_resources;
2827 ppc_md.pcibios_iov_resource_alignment = pnv_pci_iov_resource_alignment;
2828 #endif
2829
2830 pci_add_flags(PCI_REASSIGN_ALL_RSRC);
2831
2832 /* Reset IODA tables to a clean state */
2833 rc = opal_pci_reset(phb_id, OPAL_RESET_PCI_IODA_TABLE, OPAL_ASSERT_RESET);
2834 if (rc)
2835 pr_warning(" OPAL Error %ld performing IODA table reset !\n", rc);
2836
2837 /* If we're running in kdump kerenl, the previous kerenl never
2838 * shutdown PCI devices correctly. We already got IODA table
2839 * cleaned out. So we have to issue PHB reset to stop all PCI
2840 * transactions from previous kerenl.
2841 */
2842 if (is_kdump_kernel()) {
2843 pr_info(" Issue PHB reset ...\n");
2844 pnv_eeh_phb_reset(hose, EEH_RESET_FUNDAMENTAL);
2845 pnv_eeh_phb_reset(hose, EEH_RESET_DEACTIVATE);
2846 }
2847
2848 /* Remove M64 resource if we can't configure it successfully */
2849 if (!phb->init_m64 || phb->init_m64(phb))
2850 hose->mem_resources[1].flags = 0;
2851 }
2852
2853 void __init pnv_pci_init_ioda2_phb(struct device_node *np)
2854 {
2855 pnv_pci_init_ioda_phb(np, 0, PNV_PHB_IODA2);
2856 }
2857
2858 void __init pnv_pci_init_ioda_hub(struct device_node *np)
2859 {
2860 struct device_node *phbn;
2861 const __be64 *prop64;
2862 u64 hub_id;
2863
2864 pr_info("Probing IODA IO-Hub %s\n", np->full_name);
2865
2866 prop64 = of_get_property(np, "ibm,opal-hubid", NULL);
2867 if (!prop64) {
2868 pr_err(" Missing \"ibm,opal-hubid\" property !\n");
2869 return;
2870 }
2871 hub_id = be64_to_cpup(prop64);
2872 pr_devel(" HUB-ID : 0x%016llx\n", hub_id);
2873
2874 /* Count child PHBs */
2875 for_each_child_of_node(np, phbn) {
2876 /* Look for IODA1 PHBs */
2877 if (of_device_is_compatible(phbn, "ibm,ioda-phb"))
2878 pnv_pci_init_ioda_phb(phbn, hub_id, PNV_PHB_IODA1);
2879 }
2880 }
Linux with added FPC-III support