Commit | Line | Data |
---|---|---|
184cd4a3 BH |
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 | ||
cee72d5b | 12 | #undef DEBUG |
184cd4a3 BH |
13 | |
14 | #include <linux/kernel.h> | |
15 | #include <linux/pci.h> | |
361f2a2a | 16 | #include <linux/crash_dump.h> |
37c367f2 | 17 | #include <linux/debugfs.h> |
184cd4a3 BH |
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> | |
cd15b048 | 25 | #include <linux/memblock.h> |
184cd4a3 BH |
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> | |
fb1b55d6 | 32 | #include <asm/msi_bitmap.h> |
184cd4a3 BH |
33 | #include <asm/ppc-pci.h> |
34 | #include <asm/opal.h> | |
35 | #include <asm/iommu.h> | |
36 | #include <asm/tce.h> | |
137436c9 | 37 | #include <asm/xics.h> |
37c367f2 | 38 | #include <asm/debug.h> |
262af557 | 39 | #include <asm/firmware.h> |
184cd4a3 BH |
40 | |
41 | #include "powernv.h" | |
42 | #include "pci.h" | |
43 | ||
184cd4a3 | 44 | #define define_pe_printk_level(func, kern_level) \ |
45eb4724 | 45 | static void func(const struct pnv_ioda_pe *pe, const char *fmt, ...) \ |
184cd4a3 BH |
46 | { \ |
47 | struct va_format vaf; \ | |
48 | va_list args; \ | |
490e078d | 49 | char pfix[32]; \ |
184cd4a3 BH |
50 | \ |
51 | va_start(args, fmt); \ | |
52 | \ | |
53 | vaf.fmt = fmt; \ | |
54 | vaf.va = &args; \ | |
55 | \ | |
490e078d GS |
56 | if (pe->pdev) \ |
57 | strlcpy(pfix, dev_name(&pe->pdev->dev), \ | |
58 | sizeof(pfix)); \ | |
59 | else \ | |
60 | sprintf(pfix, "%04x:%02x ", \ | |
61 | pci_domain_nr(pe->pbus), \ | |
62 | pe->pbus->number); \ | |
45eb4724 JP |
63 | printk(kern_level "pci %s: [PE# %.3d] %pV", \ |
64 | pfix, pe->pe_number, &vaf); \ | |
490e078d | 65 | \ |
184cd4a3 | 66 | va_end(args); \ |
184cd4a3 BH |
67 | } \ |
68 | ||
69 | define_pe_printk_level(pe_err, KERN_ERR); | |
70 | define_pe_printk_level(pe_warn, KERN_WARNING); | |
71 | define_pe_printk_level(pe_info, KERN_INFO); | |
72 | ||
8e0a1611 AK |
73 | /* |
74 | * stdcix is only supposed to be used in hypervisor real mode as per | |
75 | * the architecture spec | |
76 | */ | |
77 | static inline void __raw_rm_writeq(u64 val, volatile void __iomem *paddr) | |
78 | { | |
79 | __asm__ __volatile__("stdcix %0,0,%1" | |
80 | : : "r" (val), "r" (paddr) : "memory"); | |
81 | } | |
82 | ||
262af557 GC |
83 | static inline bool pnv_pci_is_mem_pref_64(unsigned long flags) |
84 | { | |
85 | return ((flags & (IORESOURCE_MEM_64 | IORESOURCE_PREFETCH)) == | |
86 | (IORESOURCE_MEM_64 | IORESOURCE_PREFETCH)); | |
87 | } | |
88 | ||
cad5cef6 | 89 | static int pnv_ioda_alloc_pe(struct pnv_phb *phb) |
184cd4a3 BH |
90 | { |
91 | unsigned long pe; | |
92 | ||
93 | do { | |
94 | pe = find_next_zero_bit(phb->ioda.pe_alloc, | |
95 | phb->ioda.total_pe, 0); | |
96 | if (pe >= phb->ioda.total_pe) | |
97 | return IODA_INVALID_PE; | |
98 | } while(test_and_set_bit(pe, phb->ioda.pe_alloc)); | |
99 | ||
4cce9550 | 100 | phb->ioda.pe_array[pe].phb = phb; |
184cd4a3 BH |
101 | phb->ioda.pe_array[pe].pe_number = pe; |
102 | return pe; | |
103 | } | |
104 | ||
cad5cef6 | 105 | static void pnv_ioda_free_pe(struct pnv_phb *phb, int pe) |
184cd4a3 BH |
106 | { |
107 | WARN_ON(phb->ioda.pe_array[pe].pdev); | |
108 | ||
109 | memset(&phb->ioda.pe_array[pe], 0, sizeof(struct pnv_ioda_pe)); | |
110 | clear_bit(pe, phb->ioda.pe_alloc); | |
111 | } | |
112 | ||
262af557 GC |
113 | /* The default M64 BAR is shared by all PEs */ |
114 | static int pnv_ioda2_init_m64(struct pnv_phb *phb) | |
115 | { | |
116 | const char *desc; | |
117 | struct resource *r; | |
118 | s64 rc; | |
119 | ||
120 | /* Configure the default M64 BAR */ | |
121 | rc = opal_pci_set_phb_mem_window(phb->opal_id, | |
122 | OPAL_M64_WINDOW_TYPE, | |
123 | phb->ioda.m64_bar_idx, | |
124 | phb->ioda.m64_base, | |
125 | 0, /* unused */ | |
126 | phb->ioda.m64_size); | |
127 | if (rc != OPAL_SUCCESS) { | |
128 | desc = "configuring"; | |
129 | goto fail; | |
130 | } | |
131 | ||
132 | /* Enable the default M64 BAR */ | |
133 | rc = opal_pci_phb_mmio_enable(phb->opal_id, | |
134 | OPAL_M64_WINDOW_TYPE, | |
135 | phb->ioda.m64_bar_idx, | |
136 | OPAL_ENABLE_M64_SPLIT); | |
137 | if (rc != OPAL_SUCCESS) { | |
138 | desc = "enabling"; | |
139 | goto fail; | |
140 | } | |
141 | ||
142 | /* Mark the M64 BAR assigned */ | |
143 | set_bit(phb->ioda.m64_bar_idx, &phb->ioda.m64_bar_alloc); | |
144 | ||
145 | /* | |
146 | * Strip off the segment used by the reserved PE, which is | |
147 | * expected to be 0 or last one of PE capabicity. | |
148 | */ | |
149 | r = &phb->hose->mem_resources[1]; | |
150 | if (phb->ioda.reserved_pe == 0) | |
151 | r->start += phb->ioda.m64_segsize; | |
152 | else if (phb->ioda.reserved_pe == (phb->ioda.total_pe - 1)) | |
153 | r->end -= phb->ioda.m64_segsize; | |
154 | else | |
155 | pr_warn(" Cannot strip M64 segment for reserved PE#%d\n", | |
156 | phb->ioda.reserved_pe); | |
157 | ||
158 | return 0; | |
159 | ||
160 | fail: | |
161 | pr_warn(" Failure %lld %s M64 BAR#%d\n", | |
162 | rc, desc, phb->ioda.m64_bar_idx); | |
163 | opal_pci_phb_mmio_enable(phb->opal_id, | |
164 | OPAL_M64_WINDOW_TYPE, | |
165 | phb->ioda.m64_bar_idx, | |
166 | OPAL_DISABLE_M64); | |
167 | return -EIO; | |
168 | } | |
169 | ||
170 | static void pnv_ioda2_alloc_m64_pe(struct pnv_phb *phb) | |
171 | { | |
172 | resource_size_t sgsz = phb->ioda.m64_segsize; | |
173 | struct pci_dev *pdev; | |
174 | struct resource *r; | |
175 | int base, step, i; | |
176 | ||
177 | /* | |
178 | * Root bus always has full M64 range and root port has | |
179 | * M64 range used in reality. So we're checking root port | |
180 | * instead of root bus. | |
181 | */ | |
182 | list_for_each_entry(pdev, &phb->hose->bus->devices, bus_list) { | |
183 | for (i = PCI_BRIDGE_RESOURCES; | |
184 | i <= PCI_BRIDGE_RESOURCE_END; i++) { | |
185 | r = &pdev->resource[i]; | |
186 | if (!r->parent || | |
187 | !pnv_pci_is_mem_pref_64(r->flags)) | |
188 | continue; | |
189 | ||
190 | base = (r->start - phb->ioda.m64_base) / sgsz; | |
191 | for (step = 0; step < resource_size(r) / sgsz; step++) | |
192 | set_bit(base + step, phb->ioda.pe_alloc); | |
193 | } | |
194 | } | |
195 | } | |
196 | ||
197 | static int pnv_ioda2_pick_m64_pe(struct pnv_phb *phb, | |
198 | struct pci_bus *bus, int all) | |
199 | { | |
200 | resource_size_t segsz = phb->ioda.m64_segsize; | |
201 | struct pci_dev *pdev; | |
202 | struct resource *r; | |
203 | struct pnv_ioda_pe *master_pe, *pe; | |
204 | unsigned long size, *pe_alloc; | |
205 | bool found; | |
206 | int start, i, j; | |
207 | ||
208 | /* Root bus shouldn't use M64 */ | |
209 | if (pci_is_root_bus(bus)) | |
210 | return IODA_INVALID_PE; | |
211 | ||
212 | /* We support only one M64 window on each bus */ | |
213 | found = false; | |
214 | pci_bus_for_each_resource(bus, r, i) { | |
215 | if (r && r->parent && | |
216 | pnv_pci_is_mem_pref_64(r->flags)) { | |
217 | found = true; | |
218 | break; | |
219 | } | |
220 | } | |
221 | ||
222 | /* No M64 window found ? */ | |
223 | if (!found) | |
224 | return IODA_INVALID_PE; | |
225 | ||
226 | /* Allocate bitmap */ | |
227 | size = _ALIGN_UP(phb->ioda.total_pe / 8, sizeof(unsigned long)); | |
228 | pe_alloc = kzalloc(size, GFP_KERNEL); | |
229 | if (!pe_alloc) { | |
230 | pr_warn("%s: Out of memory !\n", | |
231 | __func__); | |
232 | return IODA_INVALID_PE; | |
233 | } | |
234 | ||
235 | /* | |
236 | * Figure out reserved PE numbers by the PE | |
237 | * the its child PEs. | |
238 | */ | |
239 | start = (r->start - phb->ioda.m64_base) / segsz; | |
240 | for (i = 0; i < resource_size(r) / segsz; i++) | |
241 | set_bit(start + i, pe_alloc); | |
242 | ||
243 | if (all) | |
244 | goto done; | |
245 | ||
246 | /* | |
247 | * If the PE doesn't cover all subordinate buses, | |
248 | * we need subtract from reserved PEs for children. | |
249 | */ | |
250 | list_for_each_entry(pdev, &bus->devices, bus_list) { | |
251 | if (!pdev->subordinate) | |
252 | continue; | |
253 | ||
254 | pci_bus_for_each_resource(pdev->subordinate, r, i) { | |
255 | if (!r || !r->parent || | |
256 | !pnv_pci_is_mem_pref_64(r->flags)) | |
257 | continue; | |
258 | ||
259 | start = (r->start - phb->ioda.m64_base) / segsz; | |
260 | for (j = 0; j < resource_size(r) / segsz ; j++) | |
261 | clear_bit(start + j, pe_alloc); | |
262 | } | |
263 | } | |
264 | ||
265 | /* | |
266 | * the current bus might not own M64 window and that's all | |
267 | * contributed by its child buses. For the case, we needn't | |
268 | * pick M64 dependent PE#. | |
269 | */ | |
270 | if (bitmap_empty(pe_alloc, phb->ioda.total_pe)) { | |
271 | kfree(pe_alloc); | |
272 | return IODA_INVALID_PE; | |
273 | } | |
274 | ||
275 | /* | |
276 | * Figure out the master PE and put all slave PEs to master | |
277 | * PE's list to form compound PE. | |
278 | */ | |
279 | done: | |
280 | master_pe = NULL; | |
281 | i = -1; | |
282 | while ((i = find_next_bit(pe_alloc, phb->ioda.total_pe, i + 1)) < | |
283 | phb->ioda.total_pe) { | |
284 | pe = &phb->ioda.pe_array[i]; | |
285 | pe->phb = phb; | |
286 | pe->pe_number = i; | |
287 | ||
288 | if (!master_pe) { | |
289 | pe->flags |= PNV_IODA_PE_MASTER; | |
290 | INIT_LIST_HEAD(&pe->slaves); | |
291 | master_pe = pe; | |
292 | } else { | |
293 | pe->flags |= PNV_IODA_PE_SLAVE; | |
294 | pe->master = master_pe; | |
295 | list_add_tail(&pe->list, &master_pe->slaves); | |
296 | } | |
297 | } | |
298 | ||
299 | kfree(pe_alloc); | |
300 | return master_pe->pe_number; | |
301 | } | |
302 | ||
303 | static void __init pnv_ioda_parse_m64_window(struct pnv_phb *phb) | |
304 | { | |
305 | struct pci_controller *hose = phb->hose; | |
306 | struct device_node *dn = hose->dn; | |
307 | struct resource *res; | |
308 | const u32 *r; | |
309 | u64 pci_addr; | |
310 | ||
311 | if (!firmware_has_feature(FW_FEATURE_OPALv3)) { | |
312 | pr_info(" Firmware too old to support M64 window\n"); | |
313 | return; | |
314 | } | |
315 | ||
316 | r = of_get_property(dn, "ibm,opal-m64-window", NULL); | |
317 | if (!r) { | |
318 | pr_info(" No <ibm,opal-m64-window> on %s\n", | |
319 | dn->full_name); | |
320 | return; | |
321 | } | |
322 | ||
323 | /* FIXME: Support M64 for P7IOC */ | |
324 | if (phb->type != PNV_PHB_IODA2) { | |
325 | pr_info(" Not support M64 window\n"); | |
326 | return; | |
327 | } | |
328 | ||
329 | res = &hose->mem_resources[1]; | |
330 | res->start = of_translate_address(dn, r + 2); | |
331 | res->end = res->start + of_read_number(r + 4, 2) - 1; | |
332 | res->flags = (IORESOURCE_MEM | IORESOURCE_MEM_64 | IORESOURCE_PREFETCH); | |
333 | pci_addr = of_read_number(r, 2); | |
334 | hose->mem_offset[1] = res->start - pci_addr; | |
335 | ||
336 | phb->ioda.m64_size = resource_size(res); | |
337 | phb->ioda.m64_segsize = phb->ioda.m64_size / phb->ioda.total_pe; | |
338 | phb->ioda.m64_base = pci_addr; | |
339 | ||
340 | /* Use last M64 BAR to cover M64 window */ | |
341 | phb->ioda.m64_bar_idx = 15; | |
342 | phb->init_m64 = pnv_ioda2_init_m64; | |
343 | phb->alloc_m64_pe = pnv_ioda2_alloc_m64_pe; | |
344 | phb->pick_m64_pe = pnv_ioda2_pick_m64_pe; | |
345 | } | |
346 | ||
49dec922 GS |
347 | static void pnv_ioda_freeze_pe(struct pnv_phb *phb, int pe_no) |
348 | { | |
349 | struct pnv_ioda_pe *pe = &phb->ioda.pe_array[pe_no]; | |
350 | struct pnv_ioda_pe *slave; | |
351 | s64 rc; | |
352 | ||
353 | /* Fetch master PE */ | |
354 | if (pe->flags & PNV_IODA_PE_SLAVE) { | |
355 | pe = pe->master; | |
356 | WARN_ON(!pe || !(pe->flags & PNV_IODA_PE_MASTER)); | |
357 | pe_no = pe->pe_number; | |
358 | } | |
359 | ||
360 | /* Freeze master PE */ | |
361 | rc = opal_pci_eeh_freeze_set(phb->opal_id, | |
362 | pe_no, | |
363 | OPAL_EEH_ACTION_SET_FREEZE_ALL); | |
364 | if (rc != OPAL_SUCCESS) { | |
365 | pr_warn("%s: Failure %lld freezing PHB#%x-PE#%x\n", | |
366 | __func__, rc, phb->hose->global_number, pe_no); | |
367 | return; | |
368 | } | |
369 | ||
370 | /* Freeze slave PEs */ | |
371 | if (!(pe->flags & PNV_IODA_PE_MASTER)) | |
372 | return; | |
373 | ||
374 | list_for_each_entry(slave, &pe->slaves, list) { | |
375 | rc = opal_pci_eeh_freeze_set(phb->opal_id, | |
376 | slave->pe_number, | |
377 | OPAL_EEH_ACTION_SET_FREEZE_ALL); | |
378 | if (rc != OPAL_SUCCESS) | |
379 | pr_warn("%s: Failure %lld freezing PHB#%x-PE#%x\n", | |
380 | __func__, rc, phb->hose->global_number, | |
381 | slave->pe_number); | |
382 | } | |
383 | } | |
384 | ||
e51df2c1 | 385 | static int pnv_ioda_unfreeze_pe(struct pnv_phb *phb, int pe_no, int opt) |
49dec922 GS |
386 | { |
387 | struct pnv_ioda_pe *pe, *slave; | |
388 | s64 rc; | |
389 | ||
390 | /* Find master PE */ | |
391 | pe = &phb->ioda.pe_array[pe_no]; | |
392 | if (pe->flags & PNV_IODA_PE_SLAVE) { | |
393 | pe = pe->master; | |
394 | WARN_ON(!pe || !(pe->flags & PNV_IODA_PE_MASTER)); | |
395 | pe_no = pe->pe_number; | |
396 | } | |
397 | ||
398 | /* Clear frozen state for master PE */ | |
399 | rc = opal_pci_eeh_freeze_clear(phb->opal_id, pe_no, opt); | |
400 | if (rc != OPAL_SUCCESS) { | |
401 | pr_warn("%s: Failure %lld clear %d on PHB#%x-PE#%x\n", | |
402 | __func__, rc, opt, phb->hose->global_number, pe_no); | |
403 | return -EIO; | |
404 | } | |
405 | ||
406 | if (!(pe->flags & PNV_IODA_PE_MASTER)) | |
407 | return 0; | |
408 | ||
409 | /* Clear frozen state for slave PEs */ | |
410 | list_for_each_entry(slave, &pe->slaves, list) { | |
411 | rc = opal_pci_eeh_freeze_clear(phb->opal_id, | |
412 | slave->pe_number, | |
413 | opt); | |
414 | if (rc != OPAL_SUCCESS) { | |
415 | pr_warn("%s: Failure %lld clear %d on PHB#%x-PE#%x\n", | |
416 | __func__, rc, opt, phb->hose->global_number, | |
417 | slave->pe_number); | |
418 | return -EIO; | |
419 | } | |
420 | } | |
421 | ||
422 | return 0; | |
423 | } | |
424 | ||
425 | static int pnv_ioda_get_pe_state(struct pnv_phb *phb, int pe_no) | |
426 | { | |
427 | struct pnv_ioda_pe *slave, *pe; | |
428 | u8 fstate, state; | |
429 | __be16 pcierr; | |
430 | s64 rc; | |
431 | ||
432 | /* Sanity check on PE number */ | |
433 | if (pe_no < 0 || pe_no >= phb->ioda.total_pe) | |
434 | return OPAL_EEH_STOPPED_PERM_UNAVAIL; | |
435 | ||
436 | /* | |
437 | * Fetch the master PE and the PE instance might be | |
438 | * not initialized yet. | |
439 | */ | |
440 | pe = &phb->ioda.pe_array[pe_no]; | |
441 | if (pe->flags & PNV_IODA_PE_SLAVE) { | |
442 | pe = pe->master; | |
443 | WARN_ON(!pe || !(pe->flags & PNV_IODA_PE_MASTER)); | |
444 | pe_no = pe->pe_number; | |
445 | } | |
446 | ||
447 | /* Check the master PE */ | |
448 | rc = opal_pci_eeh_freeze_status(phb->opal_id, pe_no, | |
449 | &state, &pcierr, NULL); | |
450 | if (rc != OPAL_SUCCESS) { | |
451 | pr_warn("%s: Failure %lld getting " | |
452 | "PHB#%x-PE#%x state\n", | |
453 | __func__, rc, | |
454 | phb->hose->global_number, pe_no); | |
455 | return OPAL_EEH_STOPPED_TEMP_UNAVAIL; | |
456 | } | |
457 | ||
458 | /* Check the slave PE */ | |
459 | if (!(pe->flags & PNV_IODA_PE_MASTER)) | |
460 | return state; | |
461 | ||
462 | list_for_each_entry(slave, &pe->slaves, list) { | |
463 | rc = opal_pci_eeh_freeze_status(phb->opal_id, | |
464 | slave->pe_number, | |
465 | &fstate, | |
466 | &pcierr, | |
467 | NULL); | |
468 | if (rc != OPAL_SUCCESS) { | |
469 | pr_warn("%s: Failure %lld getting " | |
470 | "PHB#%x-PE#%x state\n", | |
471 | __func__, rc, | |
472 | phb->hose->global_number, slave->pe_number); | |
473 | return OPAL_EEH_STOPPED_TEMP_UNAVAIL; | |
474 | } | |
475 | ||
476 | /* | |
477 | * Override the result based on the ascending | |
478 | * priority. | |
479 | */ | |
480 | if (fstate > state) | |
481 | state = fstate; | |
482 | } | |
483 | ||
484 | return state; | |
485 | } | |
486 | ||
184cd4a3 BH |
487 | /* Currently those 2 are only used when MSIs are enabled, this will change |
488 | * but in the meantime, we need to protect them to avoid warnings | |
489 | */ | |
490 | #ifdef CONFIG_PCI_MSI | |
cad5cef6 | 491 | static struct pnv_ioda_pe *pnv_ioda_get_pe(struct pci_dev *dev) |
184cd4a3 BH |
492 | { |
493 | struct pci_controller *hose = pci_bus_to_host(dev->bus); | |
494 | struct pnv_phb *phb = hose->private_data; | |
b72c1f65 | 495 | struct pci_dn *pdn = pci_get_pdn(dev); |
184cd4a3 BH |
496 | |
497 | if (!pdn) | |
498 | return NULL; | |
499 | if (pdn->pe_number == IODA_INVALID_PE) | |
500 | return NULL; | |
501 | return &phb->ioda.pe_array[pdn->pe_number]; | |
502 | } | |
184cd4a3 BH |
503 | #endif /* CONFIG_PCI_MSI */ |
504 | ||
cad5cef6 | 505 | static int pnv_ioda_configure_pe(struct pnv_phb *phb, struct pnv_ioda_pe *pe) |
184cd4a3 BH |
506 | { |
507 | struct pci_dev *parent; | |
508 | uint8_t bcomp, dcomp, fcomp; | |
509 | long rc, rid_end, rid; | |
510 | ||
511 | /* Bus validation ? */ | |
512 | if (pe->pbus) { | |
513 | int count; | |
514 | ||
515 | dcomp = OPAL_IGNORE_RID_DEVICE_NUMBER; | |
516 | fcomp = OPAL_IGNORE_RID_FUNCTION_NUMBER; | |
517 | parent = pe->pbus->self; | |
fb446ad0 GS |
518 | if (pe->flags & PNV_IODA_PE_BUS_ALL) |
519 | count = pe->pbus->busn_res.end - pe->pbus->busn_res.start + 1; | |
520 | else | |
521 | count = 1; | |
522 | ||
184cd4a3 BH |
523 | switch(count) { |
524 | case 1: bcomp = OpalPciBusAll; break; | |
525 | case 2: bcomp = OpalPciBus7Bits; break; | |
526 | case 4: bcomp = OpalPciBus6Bits; break; | |
527 | case 8: bcomp = OpalPciBus5Bits; break; | |
528 | case 16: bcomp = OpalPciBus4Bits; break; | |
529 | case 32: bcomp = OpalPciBus3Bits; break; | |
530 | default: | |
531 | pr_err("%s: Number of subordinate busses %d" | |
532 | " unsupported\n", | |
533 | pci_name(pe->pbus->self), count); | |
534 | /* Do an exact match only */ | |
535 | bcomp = OpalPciBusAll; | |
536 | } | |
537 | rid_end = pe->rid + (count << 8); | |
538 | } else { | |
539 | parent = pe->pdev->bus->self; | |
540 | bcomp = OpalPciBusAll; | |
541 | dcomp = OPAL_COMPARE_RID_DEVICE_NUMBER; | |
542 | fcomp = OPAL_COMPARE_RID_FUNCTION_NUMBER; | |
543 | rid_end = pe->rid + 1; | |
544 | } | |
545 | ||
631ad691 GS |
546 | /* |
547 | * Associate PE in PELT. We need add the PE into the | |
548 | * corresponding PELT-V as well. Otherwise, the error | |
549 | * originated from the PE might contribute to other | |
550 | * PEs. | |
551 | */ | |
184cd4a3 BH |
552 | rc = opal_pci_set_pe(phb->opal_id, pe->pe_number, pe->rid, |
553 | bcomp, dcomp, fcomp, OPAL_MAP_PE); | |
554 | if (rc) { | |
555 | pe_err(pe, "OPAL error %ld trying to setup PELT table\n", rc); | |
556 | return -ENXIO; | |
557 | } | |
631ad691 GS |
558 | |
559 | rc = opal_pci_set_peltv(phb->opal_id, pe->pe_number, | |
560 | pe->pe_number, OPAL_ADD_PE_TO_DOMAIN); | |
561 | if (rc) | |
562 | pe_warn(pe, "OPAL error %d adding self to PELTV\n", rc); | |
184cd4a3 BH |
563 | opal_pci_eeh_freeze_clear(phb->opal_id, pe->pe_number, |
564 | OPAL_EEH_ACTION_CLEAR_FREEZE_ALL); | |
565 | ||
566 | /* Add to all parents PELT-V */ | |
567 | while (parent) { | |
b72c1f65 | 568 | struct pci_dn *pdn = pci_get_pdn(parent); |
184cd4a3 BH |
569 | if (pdn && pdn->pe_number != IODA_INVALID_PE) { |
570 | rc = opal_pci_set_peltv(phb->opal_id, pdn->pe_number, | |
cee72d5b | 571 | pe->pe_number, OPAL_ADD_PE_TO_DOMAIN); |
184cd4a3 BH |
572 | /* XXX What to do in case of error ? */ |
573 | } | |
574 | parent = parent->bus->self; | |
575 | } | |
576 | /* Setup reverse map */ | |
577 | for (rid = pe->rid; rid < rid_end; rid++) | |
578 | phb->ioda.pe_rmap[rid] = pe->pe_number; | |
579 | ||
580 | /* Setup one MVTs on IODA1 */ | |
581 | if (phb->type == PNV_PHB_IODA1) { | |
582 | pe->mve_number = pe->pe_number; | |
583 | rc = opal_pci_set_mve(phb->opal_id, pe->mve_number, | |
584 | pe->pe_number); | |
585 | if (rc) { | |
586 | pe_err(pe, "OPAL error %ld setting up MVE %d\n", | |
587 | rc, pe->mve_number); | |
588 | pe->mve_number = -1; | |
589 | } else { | |
590 | rc = opal_pci_set_mve_enable(phb->opal_id, | |
cee72d5b | 591 | pe->mve_number, OPAL_ENABLE_MVE); |
184cd4a3 BH |
592 | if (rc) { |
593 | pe_err(pe, "OPAL error %ld enabling MVE %d\n", | |
594 | rc, pe->mve_number); | |
595 | pe->mve_number = -1; | |
596 | } | |
597 | } | |
598 | } else if (phb->type == PNV_PHB_IODA2) | |
599 | pe->mve_number = 0; | |
600 | ||
601 | return 0; | |
602 | } | |
603 | ||
cad5cef6 GKH |
604 | static void pnv_ioda_link_pe_by_weight(struct pnv_phb *phb, |
605 | struct pnv_ioda_pe *pe) | |
184cd4a3 BH |
606 | { |
607 | struct pnv_ioda_pe *lpe; | |
608 | ||
7ebdf956 | 609 | list_for_each_entry(lpe, &phb->ioda.pe_dma_list, dma_link) { |
184cd4a3 | 610 | if (lpe->dma_weight < pe->dma_weight) { |
7ebdf956 | 611 | list_add_tail(&pe->dma_link, &lpe->dma_link); |
184cd4a3 BH |
612 | return; |
613 | } | |
614 | } | |
7ebdf956 | 615 | list_add_tail(&pe->dma_link, &phb->ioda.pe_dma_list); |
184cd4a3 BH |
616 | } |
617 | ||
618 | static unsigned int pnv_ioda_dma_weight(struct pci_dev *dev) | |
619 | { | |
620 | /* This is quite simplistic. The "base" weight of a device | |
621 | * is 10. 0 means no DMA is to be accounted for it. | |
622 | */ | |
623 | ||
624 | /* If it's a bridge, no DMA */ | |
625 | if (dev->hdr_type != PCI_HEADER_TYPE_NORMAL) | |
626 | return 0; | |
627 | ||
628 | /* Reduce the weight of slow USB controllers */ | |
629 | if (dev->class == PCI_CLASS_SERIAL_USB_UHCI || | |
630 | dev->class == PCI_CLASS_SERIAL_USB_OHCI || | |
631 | dev->class == PCI_CLASS_SERIAL_USB_EHCI) | |
632 | return 3; | |
633 | ||
634 | /* Increase the weight of RAID (includes Obsidian) */ | |
635 | if ((dev->class >> 8) == PCI_CLASS_STORAGE_RAID) | |
636 | return 15; | |
637 | ||
638 | /* Default */ | |
639 | return 10; | |
640 | } | |
641 | ||
fb446ad0 | 642 | #if 0 |
cad5cef6 | 643 | static struct pnv_ioda_pe *pnv_ioda_setup_dev_PE(struct pci_dev *dev) |
184cd4a3 BH |
644 | { |
645 | struct pci_controller *hose = pci_bus_to_host(dev->bus); | |
646 | struct pnv_phb *phb = hose->private_data; | |
b72c1f65 | 647 | struct pci_dn *pdn = pci_get_pdn(dev); |
184cd4a3 BH |
648 | struct pnv_ioda_pe *pe; |
649 | int pe_num; | |
650 | ||
651 | if (!pdn) { | |
652 | pr_err("%s: Device tree node not associated properly\n", | |
653 | pci_name(dev)); | |
654 | return NULL; | |
655 | } | |
656 | if (pdn->pe_number != IODA_INVALID_PE) | |
657 | return NULL; | |
658 | ||
659 | /* PE#0 has been pre-set */ | |
660 | if (dev->bus->number == 0) | |
661 | pe_num = 0; | |
662 | else | |
663 | pe_num = pnv_ioda_alloc_pe(phb); | |
664 | if (pe_num == IODA_INVALID_PE) { | |
665 | pr_warning("%s: Not enough PE# available, disabling device\n", | |
666 | pci_name(dev)); | |
667 | return NULL; | |
668 | } | |
669 | ||
670 | /* NOTE: We get only one ref to the pci_dev for the pdn, not for the | |
671 | * pointer in the PE data structure, both should be destroyed at the | |
672 | * same time. However, this needs to be looked at more closely again | |
673 | * once we actually start removing things (Hotplug, SR-IOV, ...) | |
674 | * | |
675 | * At some point we want to remove the PDN completely anyways | |
676 | */ | |
677 | pe = &phb->ioda.pe_array[pe_num]; | |
678 | pci_dev_get(dev); | |
679 | pdn->pcidev = dev; | |
680 | pdn->pe_number = pe_num; | |
681 | pe->pdev = dev; | |
682 | pe->pbus = NULL; | |
683 | pe->tce32_seg = -1; | |
684 | pe->mve_number = -1; | |
685 | pe->rid = dev->bus->number << 8 | pdn->devfn; | |
686 | ||
687 | pe_info(pe, "Associated device to PE\n"); | |
688 | ||
689 | if (pnv_ioda_configure_pe(phb, pe)) { | |
690 | /* XXX What do we do here ? */ | |
691 | if (pe_num) | |
692 | pnv_ioda_free_pe(phb, pe_num); | |
693 | pdn->pe_number = IODA_INVALID_PE; | |
694 | pe->pdev = NULL; | |
695 | pci_dev_put(dev); | |
696 | return NULL; | |
697 | } | |
698 | ||
699 | /* Assign a DMA weight to the device */ | |
700 | pe->dma_weight = pnv_ioda_dma_weight(dev); | |
701 | if (pe->dma_weight != 0) { | |
702 | phb->ioda.dma_weight += pe->dma_weight; | |
703 | phb->ioda.dma_pe_count++; | |
704 | } | |
705 | ||
706 | /* Link the PE */ | |
707 | pnv_ioda_link_pe_by_weight(phb, pe); | |
708 | ||
709 | return pe; | |
710 | } | |
fb446ad0 | 711 | #endif /* Useful for SRIOV case */ |
184cd4a3 BH |
712 | |
713 | static void pnv_ioda_setup_same_PE(struct pci_bus *bus, struct pnv_ioda_pe *pe) | |
714 | { | |
715 | struct pci_dev *dev; | |
716 | ||
717 | list_for_each_entry(dev, &bus->devices, bus_list) { | |
b72c1f65 | 718 | struct pci_dn *pdn = pci_get_pdn(dev); |
184cd4a3 BH |
719 | |
720 | if (pdn == NULL) { | |
721 | pr_warn("%s: No device node associated with device !\n", | |
722 | pci_name(dev)); | |
723 | continue; | |
724 | } | |
184cd4a3 BH |
725 | pdn->pcidev = dev; |
726 | pdn->pe_number = pe->pe_number; | |
727 | pe->dma_weight += pnv_ioda_dma_weight(dev); | |
fb446ad0 | 728 | if ((pe->flags & PNV_IODA_PE_BUS_ALL) && dev->subordinate) |
184cd4a3 BH |
729 | pnv_ioda_setup_same_PE(dev->subordinate, pe); |
730 | } | |
731 | } | |
732 | ||
fb446ad0 GS |
733 | /* |
734 | * There're 2 types of PCI bus sensitive PEs: One that is compromised of | |
735 | * single PCI bus. Another one that contains the primary PCI bus and its | |
736 | * subordinate PCI devices and buses. The second type of PE is normally | |
737 | * orgiriated by PCIe-to-PCI bridge or PLX switch downstream ports. | |
738 | */ | |
cad5cef6 | 739 | static void pnv_ioda_setup_bus_PE(struct pci_bus *bus, int all) |
184cd4a3 | 740 | { |
fb446ad0 | 741 | struct pci_controller *hose = pci_bus_to_host(bus); |
184cd4a3 | 742 | struct pnv_phb *phb = hose->private_data; |
184cd4a3 | 743 | struct pnv_ioda_pe *pe; |
262af557 GC |
744 | int pe_num = IODA_INVALID_PE; |
745 | ||
746 | /* Check if PE is determined by M64 */ | |
747 | if (phb->pick_m64_pe) | |
748 | pe_num = phb->pick_m64_pe(phb, bus, all); | |
749 | ||
750 | /* The PE number isn't pinned by M64 */ | |
751 | if (pe_num == IODA_INVALID_PE) | |
752 | pe_num = pnv_ioda_alloc_pe(phb); | |
184cd4a3 | 753 | |
184cd4a3 | 754 | if (pe_num == IODA_INVALID_PE) { |
fb446ad0 GS |
755 | pr_warning("%s: Not enough PE# available for PCI bus %04x:%02x\n", |
756 | __func__, pci_domain_nr(bus), bus->number); | |
184cd4a3 BH |
757 | return; |
758 | } | |
759 | ||
760 | pe = &phb->ioda.pe_array[pe_num]; | |
262af557 | 761 | pe->flags |= (all ? PNV_IODA_PE_BUS_ALL : PNV_IODA_PE_BUS); |
184cd4a3 BH |
762 | pe->pbus = bus; |
763 | pe->pdev = NULL; | |
764 | pe->tce32_seg = -1; | |
765 | pe->mve_number = -1; | |
b918c62e | 766 | pe->rid = bus->busn_res.start << 8; |
184cd4a3 BH |
767 | pe->dma_weight = 0; |
768 | ||
fb446ad0 GS |
769 | if (all) |
770 | pe_info(pe, "Secondary bus %d..%d associated with PE#%d\n", | |
771 | bus->busn_res.start, bus->busn_res.end, pe_num); | |
772 | else | |
773 | pe_info(pe, "Secondary bus %d associated with PE#%d\n", | |
774 | bus->busn_res.start, pe_num); | |
184cd4a3 BH |
775 | |
776 | if (pnv_ioda_configure_pe(phb, pe)) { | |
777 | /* XXX What do we do here ? */ | |
778 | if (pe_num) | |
779 | pnv_ioda_free_pe(phb, pe_num); | |
780 | pe->pbus = NULL; | |
781 | return; | |
782 | } | |
783 | ||
784 | /* Associate it with all child devices */ | |
785 | pnv_ioda_setup_same_PE(bus, pe); | |
786 | ||
7ebdf956 GS |
787 | /* Put PE to the list */ |
788 | list_add_tail(&pe->list, &phb->ioda.pe_list); | |
789 | ||
184cd4a3 BH |
790 | /* Account for one DMA PE if at least one DMA capable device exist |
791 | * below the bridge | |
792 | */ | |
793 | if (pe->dma_weight != 0) { | |
794 | phb->ioda.dma_weight += pe->dma_weight; | |
795 | phb->ioda.dma_pe_count++; | |
796 | } | |
797 | ||
798 | /* Link the PE */ | |
799 | pnv_ioda_link_pe_by_weight(phb, pe); | |
800 | } | |
801 | ||
cad5cef6 | 802 | static void pnv_ioda_setup_PEs(struct pci_bus *bus) |
184cd4a3 BH |
803 | { |
804 | struct pci_dev *dev; | |
fb446ad0 GS |
805 | |
806 | pnv_ioda_setup_bus_PE(bus, 0); | |
184cd4a3 BH |
807 | |
808 | list_for_each_entry(dev, &bus->devices, bus_list) { | |
fb446ad0 GS |
809 | if (dev->subordinate) { |
810 | if (pci_pcie_type(dev) == PCI_EXP_TYPE_PCI_BRIDGE) | |
811 | pnv_ioda_setup_bus_PE(dev->subordinate, 1); | |
812 | else | |
813 | pnv_ioda_setup_PEs(dev->subordinate); | |
814 | } | |
815 | } | |
816 | } | |
817 | ||
818 | /* | |
819 | * Configure PEs so that the downstream PCI buses and devices | |
820 | * could have their associated PE#. Unfortunately, we didn't | |
821 | * figure out the way to identify the PLX bridge yet. So we | |
822 | * simply put the PCI bus and the subordinate behind the root | |
823 | * port to PE# here. The game rule here is expected to be changed | |
824 | * as soon as we can detected PLX bridge correctly. | |
825 | */ | |
cad5cef6 | 826 | static void pnv_pci_ioda_setup_PEs(void) |
fb446ad0 GS |
827 | { |
828 | struct pci_controller *hose, *tmp; | |
262af557 | 829 | struct pnv_phb *phb; |
fb446ad0 GS |
830 | |
831 | list_for_each_entry_safe(hose, tmp, &hose_list, list_node) { | |
262af557 GC |
832 | phb = hose->private_data; |
833 | ||
834 | /* M64 layout might affect PE allocation */ | |
835 | if (phb->alloc_m64_pe) | |
836 | phb->alloc_m64_pe(phb); | |
837 | ||
fb446ad0 | 838 | pnv_ioda_setup_PEs(hose->bus); |
184cd4a3 BH |
839 | } |
840 | } | |
841 | ||
959c9bdd | 842 | static void pnv_pci_ioda_dma_dev_setup(struct pnv_phb *phb, struct pci_dev *pdev) |
184cd4a3 | 843 | { |
b72c1f65 | 844 | struct pci_dn *pdn = pci_get_pdn(pdev); |
959c9bdd | 845 | struct pnv_ioda_pe *pe; |
184cd4a3 | 846 | |
959c9bdd GS |
847 | /* |
848 | * The function can be called while the PE# | |
849 | * hasn't been assigned. Do nothing for the | |
850 | * case. | |
851 | */ | |
852 | if (!pdn || pdn->pe_number == IODA_INVALID_PE) | |
853 | return; | |
184cd4a3 | 854 | |
959c9bdd | 855 | pe = &phb->ioda.pe_array[pdn->pe_number]; |
cd15b048 | 856 | WARN_ON(get_dma_ops(&pdev->dev) != &dma_iommu_ops); |
763fe0ad | 857 | set_iommu_table_base_and_group(&pdev->dev, &pe->tce32_table); |
184cd4a3 BH |
858 | } |
859 | ||
cd15b048 BH |
860 | static int pnv_pci_ioda_dma_set_mask(struct pnv_phb *phb, |
861 | struct pci_dev *pdev, u64 dma_mask) | |
862 | { | |
863 | struct pci_dn *pdn = pci_get_pdn(pdev); | |
864 | struct pnv_ioda_pe *pe; | |
865 | uint64_t top; | |
866 | bool bypass = false; | |
867 | ||
868 | if (WARN_ON(!pdn || pdn->pe_number == IODA_INVALID_PE)) | |
869 | return -ENODEV;; | |
870 | ||
871 | pe = &phb->ioda.pe_array[pdn->pe_number]; | |
872 | if (pe->tce_bypass_enabled) { | |
873 | top = pe->tce_bypass_base + memblock_end_of_DRAM() - 1; | |
874 | bypass = (dma_mask >= top); | |
875 | } | |
876 | ||
877 | if (bypass) { | |
878 | dev_info(&pdev->dev, "Using 64-bit DMA iommu bypass\n"); | |
879 | set_dma_ops(&pdev->dev, &dma_direct_ops); | |
880 | set_dma_offset(&pdev->dev, pe->tce_bypass_base); | |
881 | } else { | |
882 | dev_info(&pdev->dev, "Using 32-bit DMA via iommu\n"); | |
883 | set_dma_ops(&pdev->dev, &dma_iommu_ops); | |
884 | set_iommu_table_base(&pdev->dev, &pe->tce32_table); | |
885 | } | |
a32305bf | 886 | *pdev->dev.dma_mask = dma_mask; |
cd15b048 BH |
887 | return 0; |
888 | } | |
889 | ||
dff4a39e GS |
890 | static void pnv_ioda_setup_bus_dma(struct pnv_ioda_pe *pe, |
891 | struct pci_bus *bus, | |
892 | bool add_to_iommu_group) | |
74251fe2 BH |
893 | { |
894 | struct pci_dev *dev; | |
895 | ||
896 | list_for_each_entry(dev, &bus->devices, bus_list) { | |
dff4a39e GS |
897 | if (add_to_iommu_group) |
898 | set_iommu_table_base_and_group(&dev->dev, | |
899 | &pe->tce32_table); | |
900 | else | |
901 | set_iommu_table_base(&dev->dev, &pe->tce32_table); | |
902 | ||
74251fe2 | 903 | if (dev->subordinate) |
dff4a39e GS |
904 | pnv_ioda_setup_bus_dma(pe, dev->subordinate, |
905 | add_to_iommu_group); | |
74251fe2 BH |
906 | } |
907 | } | |
908 | ||
8e0a1611 AK |
909 | static void pnv_pci_ioda1_tce_invalidate(struct pnv_ioda_pe *pe, |
910 | struct iommu_table *tbl, | |
3ad26e5c | 911 | __be64 *startp, __be64 *endp, bool rm) |
4cce9550 | 912 | { |
3ad26e5c BH |
913 | __be64 __iomem *invalidate = rm ? |
914 | (__be64 __iomem *)pe->tce_inval_reg_phys : | |
915 | (__be64 __iomem *)tbl->it_index; | |
4cce9550 | 916 | unsigned long start, end, inc; |
b0376c9b | 917 | const unsigned shift = tbl->it_page_shift; |
4cce9550 GS |
918 | |
919 | start = __pa(startp); | |
920 | end = __pa(endp); | |
921 | ||
922 | /* BML uses this case for p6/p7/galaxy2: Shift addr and put in node */ | |
923 | if (tbl->it_busno) { | |
b0376c9b AK |
924 | start <<= shift; |
925 | end <<= shift; | |
926 | inc = 128ull << shift; | |
4cce9550 GS |
927 | start |= tbl->it_busno; |
928 | end |= tbl->it_busno; | |
929 | } else if (tbl->it_type & TCE_PCI_SWINV_PAIR) { | |
930 | /* p7ioc-style invalidation, 2 TCEs per write */ | |
931 | start |= (1ull << 63); | |
932 | end |= (1ull << 63); | |
933 | inc = 16; | |
934 | } else { | |
935 | /* Default (older HW) */ | |
936 | inc = 128; | |
937 | } | |
938 | ||
939 | end |= inc - 1; /* round up end to be different than start */ | |
940 | ||
941 | mb(); /* Ensure above stores are visible */ | |
942 | while (start <= end) { | |
8e0a1611 | 943 | if (rm) |
3ad26e5c | 944 | __raw_rm_writeq(cpu_to_be64(start), invalidate); |
8e0a1611 | 945 | else |
3ad26e5c | 946 | __raw_writeq(cpu_to_be64(start), invalidate); |
4cce9550 GS |
947 | start += inc; |
948 | } | |
949 | ||
950 | /* | |
951 | * The iommu layer will do another mb() for us on build() | |
952 | * and we don't care on free() | |
953 | */ | |
954 | } | |
955 | ||
956 | static void pnv_pci_ioda2_tce_invalidate(struct pnv_ioda_pe *pe, | |
957 | struct iommu_table *tbl, | |
3ad26e5c | 958 | __be64 *startp, __be64 *endp, bool rm) |
4cce9550 GS |
959 | { |
960 | unsigned long start, end, inc; | |
3ad26e5c BH |
961 | __be64 __iomem *invalidate = rm ? |
962 | (__be64 __iomem *)pe->tce_inval_reg_phys : | |
963 | (__be64 __iomem *)tbl->it_index; | |
b0376c9b | 964 | const unsigned shift = tbl->it_page_shift; |
4cce9550 GS |
965 | |
966 | /* We'll invalidate DMA address in PE scope */ | |
b0376c9b | 967 | start = 0x2ull << 60; |
4cce9550 GS |
968 | start |= (pe->pe_number & 0xFF); |
969 | end = start; | |
970 | ||
971 | /* Figure out the start, end and step */ | |
972 | inc = tbl->it_offset + (((u64)startp - tbl->it_base) / sizeof(u64)); | |
b0376c9b | 973 | start |= (inc << shift); |
4cce9550 | 974 | inc = tbl->it_offset + (((u64)endp - tbl->it_base) / sizeof(u64)); |
b0376c9b AK |
975 | end |= (inc << shift); |
976 | inc = (0x1ull << shift); | |
4cce9550 GS |
977 | mb(); |
978 | ||
979 | while (start <= end) { | |
8e0a1611 | 980 | if (rm) |
3ad26e5c | 981 | __raw_rm_writeq(cpu_to_be64(start), invalidate); |
8e0a1611 | 982 | else |
3ad26e5c | 983 | __raw_writeq(cpu_to_be64(start), invalidate); |
4cce9550 GS |
984 | start += inc; |
985 | } | |
986 | } | |
987 | ||
988 | void pnv_pci_ioda_tce_invalidate(struct iommu_table *tbl, | |
3ad26e5c | 989 | __be64 *startp, __be64 *endp, bool rm) |
4cce9550 GS |
990 | { |
991 | struct pnv_ioda_pe *pe = container_of(tbl, struct pnv_ioda_pe, | |
992 | tce32_table); | |
993 | struct pnv_phb *phb = pe->phb; | |
994 | ||
995 | if (phb->type == PNV_PHB_IODA1) | |
8e0a1611 | 996 | pnv_pci_ioda1_tce_invalidate(pe, tbl, startp, endp, rm); |
4cce9550 | 997 | else |
8e0a1611 | 998 | pnv_pci_ioda2_tce_invalidate(pe, tbl, startp, endp, rm); |
4cce9550 GS |
999 | } |
1000 | ||
cad5cef6 GKH |
1001 | static void pnv_pci_ioda_setup_dma_pe(struct pnv_phb *phb, |
1002 | struct pnv_ioda_pe *pe, unsigned int base, | |
1003 | unsigned int segs) | |
184cd4a3 BH |
1004 | { |
1005 | ||
1006 | struct page *tce_mem = NULL; | |
1007 | const __be64 *swinvp; | |
1008 | struct iommu_table *tbl; | |
1009 | unsigned int i; | |
1010 | int64_t rc; | |
1011 | void *addr; | |
1012 | ||
1013 | /* 256M DMA window, 4K TCE pages, 8 bytes TCE */ | |
1014 | #define TCE32_TABLE_SIZE ((0x10000000 / 0x1000) * 8) | |
1015 | ||
1016 | /* XXX FIXME: Handle 64-bit only DMA devices */ | |
1017 | /* XXX FIXME: Provide 64-bit DMA facilities & non-4K TCE tables etc.. */ | |
1018 | /* XXX FIXME: Allocate multi-level tables on PHB3 */ | |
1019 | ||
1020 | /* We shouldn't already have a 32-bit DMA associated */ | |
1021 | if (WARN_ON(pe->tce32_seg >= 0)) | |
1022 | return; | |
1023 | ||
1024 | /* Grab a 32-bit TCE table */ | |
1025 | pe->tce32_seg = base; | |
1026 | pe_info(pe, " Setting up 32-bit TCE table at %08x..%08x\n", | |
1027 | (base << 28), ((base + segs) << 28) - 1); | |
1028 | ||
1029 | /* XXX Currently, we allocate one big contiguous table for the | |
1030 | * TCEs. We only really need one chunk per 256M of TCE space | |
1031 | * (ie per segment) but that's an optimization for later, it | |
1032 | * requires some added smarts with our get/put_tce implementation | |
1033 | */ | |
1034 | tce_mem = alloc_pages_node(phb->hose->node, GFP_KERNEL, | |
1035 | get_order(TCE32_TABLE_SIZE * segs)); | |
1036 | if (!tce_mem) { | |
1037 | pe_err(pe, " Failed to allocate a 32-bit TCE memory\n"); | |
1038 | goto fail; | |
1039 | } | |
1040 | addr = page_address(tce_mem); | |
1041 | memset(addr, 0, TCE32_TABLE_SIZE * segs); | |
1042 | ||
1043 | /* Configure HW */ | |
1044 | for (i = 0; i < segs; i++) { | |
1045 | rc = opal_pci_map_pe_dma_window(phb->opal_id, | |
1046 | pe->pe_number, | |
1047 | base + i, 1, | |
1048 | __pa(addr) + TCE32_TABLE_SIZE * i, | |
1049 | TCE32_TABLE_SIZE, 0x1000); | |
1050 | if (rc) { | |
1051 | pe_err(pe, " Failed to configure 32-bit TCE table," | |
1052 | " err %ld\n", rc); | |
1053 | goto fail; | |
1054 | } | |
1055 | } | |
1056 | ||
1057 | /* Setup linux iommu table */ | |
1058 | tbl = &pe->tce32_table; | |
1059 | pnv_pci_setup_iommu_table(tbl, addr, TCE32_TABLE_SIZE * segs, | |
8fa5d454 | 1060 | base << 28, IOMMU_PAGE_SHIFT_4K); |
184cd4a3 BH |
1061 | |
1062 | /* OPAL variant of P7IOC SW invalidated TCEs */ | |
1063 | swinvp = of_get_property(phb->hose->dn, "ibm,opal-tce-kill", NULL); | |
1064 | if (swinvp) { | |
1065 | /* We need a couple more fields -- an address and a data | |
1066 | * to or. Since the bus is only printed out on table free | |
1067 | * errors, and on the first pass the data will be a relative | |
1068 | * bus number, print that out instead. | |
1069 | */ | |
8e0a1611 AK |
1070 | pe->tce_inval_reg_phys = be64_to_cpup(swinvp); |
1071 | tbl->it_index = (unsigned long)ioremap(pe->tce_inval_reg_phys, | |
1072 | 8); | |
65fd766b GS |
1073 | tbl->it_type |= (TCE_PCI_SWINV_CREATE | |
1074 | TCE_PCI_SWINV_FREE | | |
1075 | TCE_PCI_SWINV_PAIR); | |
184cd4a3 BH |
1076 | } |
1077 | iommu_init_table(tbl, phb->hose->node); | |
e9bc03fe | 1078 | iommu_register_group(tbl, phb->hose->global_number, pe->pe_number); |
184cd4a3 | 1079 | |
74251fe2 | 1080 | if (pe->pdev) |
d905c5df | 1081 | set_iommu_table_base_and_group(&pe->pdev->dev, tbl); |
74251fe2 | 1082 | else |
dff4a39e | 1083 | pnv_ioda_setup_bus_dma(pe, pe->pbus, true); |
74251fe2 | 1084 | |
184cd4a3 BH |
1085 | return; |
1086 | fail: | |
1087 | /* XXX Failure: Try to fallback to 64-bit only ? */ | |
1088 | if (pe->tce32_seg >= 0) | |
1089 | pe->tce32_seg = -1; | |
1090 | if (tce_mem) | |
1091 | __free_pages(tce_mem, get_order(TCE32_TABLE_SIZE * segs)); | |
1092 | } | |
1093 | ||
cd15b048 BH |
1094 | static void pnv_pci_ioda2_set_bypass(struct iommu_table *tbl, bool enable) |
1095 | { | |
1096 | struct pnv_ioda_pe *pe = container_of(tbl, struct pnv_ioda_pe, | |
1097 | tce32_table); | |
1098 | uint16_t window_id = (pe->pe_number << 1 ) + 1; | |
1099 | int64_t rc; | |
1100 | ||
1101 | pe_info(pe, "%sabling 64-bit DMA bypass\n", enable ? "En" : "Dis"); | |
1102 | if (enable) { | |
1103 | phys_addr_t top = memblock_end_of_DRAM(); | |
1104 | ||
1105 | top = roundup_pow_of_two(top); | |
1106 | rc = opal_pci_map_pe_dma_window_real(pe->phb->opal_id, | |
1107 | pe->pe_number, | |
1108 | window_id, | |
1109 | pe->tce_bypass_base, | |
1110 | top); | |
1111 | } else { | |
1112 | rc = opal_pci_map_pe_dma_window_real(pe->phb->opal_id, | |
1113 | pe->pe_number, | |
1114 | window_id, | |
1115 | pe->tce_bypass_base, | |
1116 | 0); | |
1117 | ||
1118 | /* | |
dff4a39e GS |
1119 | * EEH needs the mapping between IOMMU table and group |
1120 | * of those VFIO/KVM pass-through devices. We can postpone | |
1121 | * resetting DMA ops until the DMA mask is configured in | |
1122 | * host side. | |
cd15b048 | 1123 | */ |
dff4a39e GS |
1124 | if (pe->pdev) |
1125 | set_iommu_table_base(&pe->pdev->dev, tbl); | |
1126 | else | |
1127 | pnv_ioda_setup_bus_dma(pe, pe->pbus, false); | |
cd15b048 BH |
1128 | } |
1129 | if (rc) | |
1130 | pe_err(pe, "OPAL error %lld configuring bypass window\n", rc); | |
1131 | else | |
1132 | pe->tce_bypass_enabled = enable; | |
1133 | } | |
1134 | ||
1135 | static void pnv_pci_ioda2_setup_bypass_pe(struct pnv_phb *phb, | |
1136 | struct pnv_ioda_pe *pe) | |
1137 | { | |
1138 | /* TVE #1 is selected by PCI address bit 59 */ | |
1139 | pe->tce_bypass_base = 1ull << 59; | |
1140 | ||
1141 | /* Install set_bypass callback for VFIO */ | |
1142 | pe->tce32_table.set_bypass = pnv_pci_ioda2_set_bypass; | |
1143 | ||
1144 | /* Enable bypass by default */ | |
1145 | pnv_pci_ioda2_set_bypass(&pe->tce32_table, true); | |
1146 | } | |
1147 | ||
373f5657 GS |
1148 | static void pnv_pci_ioda2_setup_dma_pe(struct pnv_phb *phb, |
1149 | struct pnv_ioda_pe *pe) | |
1150 | { | |
1151 | struct page *tce_mem = NULL; | |
1152 | void *addr; | |
1153 | const __be64 *swinvp; | |
1154 | struct iommu_table *tbl; | |
1155 | unsigned int tce_table_size, end; | |
1156 | int64_t rc; | |
1157 | ||
1158 | /* We shouldn't already have a 32-bit DMA associated */ | |
1159 | if (WARN_ON(pe->tce32_seg >= 0)) | |
1160 | return; | |
1161 | ||
1162 | /* The PE will reserve all possible 32-bits space */ | |
1163 | pe->tce32_seg = 0; | |
1164 | end = (1 << ilog2(phb->ioda.m32_pci_base)); | |
1165 | tce_table_size = (end / 0x1000) * 8; | |
1166 | pe_info(pe, "Setting up 32-bit TCE table at 0..%08x\n", | |
1167 | end); | |
1168 | ||
1169 | /* Allocate TCE table */ | |
1170 | tce_mem = alloc_pages_node(phb->hose->node, GFP_KERNEL, | |
1171 | get_order(tce_table_size)); | |
1172 | if (!tce_mem) { | |
1173 | pe_err(pe, "Failed to allocate a 32-bit TCE memory\n"); | |
1174 | goto fail; | |
1175 | } | |
1176 | addr = page_address(tce_mem); | |
1177 | memset(addr, 0, tce_table_size); | |
1178 | ||
1179 | /* | |
1180 | * Map TCE table through TVT. The TVE index is the PE number | |
1181 | * shifted by 1 bit for 32-bits DMA space. | |
1182 | */ | |
1183 | rc = opal_pci_map_pe_dma_window(phb->opal_id, pe->pe_number, | |
1184 | pe->pe_number << 1, 1, __pa(addr), | |
1185 | tce_table_size, 0x1000); | |
1186 | if (rc) { | |
1187 | pe_err(pe, "Failed to configure 32-bit TCE table," | |
1188 | " err %ld\n", rc); | |
1189 | goto fail; | |
1190 | } | |
1191 | ||
1192 | /* Setup linux iommu table */ | |
1193 | tbl = &pe->tce32_table; | |
8fa5d454 AK |
1194 | pnv_pci_setup_iommu_table(tbl, addr, tce_table_size, 0, |
1195 | IOMMU_PAGE_SHIFT_4K); | |
373f5657 GS |
1196 | |
1197 | /* OPAL variant of PHB3 invalidated TCEs */ | |
1198 | swinvp = of_get_property(phb->hose->dn, "ibm,opal-tce-kill", NULL); | |
1199 | if (swinvp) { | |
1200 | /* We need a couple more fields -- an address and a data | |
1201 | * to or. Since the bus is only printed out on table free | |
1202 | * errors, and on the first pass the data will be a relative | |
1203 | * bus number, print that out instead. | |
1204 | */ | |
8e0a1611 AK |
1205 | pe->tce_inval_reg_phys = be64_to_cpup(swinvp); |
1206 | tbl->it_index = (unsigned long)ioremap(pe->tce_inval_reg_phys, | |
1207 | 8); | |
65fd766b | 1208 | tbl->it_type |= (TCE_PCI_SWINV_CREATE | TCE_PCI_SWINV_FREE); |
373f5657 GS |
1209 | } |
1210 | iommu_init_table(tbl, phb->hose->node); | |
e9bc03fe | 1211 | iommu_register_group(tbl, phb->hose->global_number, pe->pe_number); |
373f5657 | 1212 | |
74251fe2 | 1213 | if (pe->pdev) |
d905c5df | 1214 | set_iommu_table_base_and_group(&pe->pdev->dev, tbl); |
74251fe2 | 1215 | else |
dff4a39e | 1216 | pnv_ioda_setup_bus_dma(pe, pe->pbus, true); |
74251fe2 | 1217 | |
cd15b048 BH |
1218 | /* Also create a bypass window */ |
1219 | pnv_pci_ioda2_setup_bypass_pe(phb, pe); | |
373f5657 GS |
1220 | return; |
1221 | fail: | |
1222 | if (pe->tce32_seg >= 0) | |
1223 | pe->tce32_seg = -1; | |
1224 | if (tce_mem) | |
1225 | __free_pages(tce_mem, get_order(tce_table_size)); | |
1226 | } | |
1227 | ||
cad5cef6 | 1228 | static void pnv_ioda_setup_dma(struct pnv_phb *phb) |
184cd4a3 BH |
1229 | { |
1230 | struct pci_controller *hose = phb->hose; | |
1231 | unsigned int residual, remaining, segs, tw, base; | |
1232 | struct pnv_ioda_pe *pe; | |
1233 | ||
1234 | /* If we have more PE# than segments available, hand out one | |
1235 | * per PE until we run out and let the rest fail. If not, | |
1236 | * then we assign at least one segment per PE, plus more based | |
1237 | * on the amount of devices under that PE | |
1238 | */ | |
1239 | if (phb->ioda.dma_pe_count > phb->ioda.tce32_count) | |
1240 | residual = 0; | |
1241 | else | |
1242 | residual = phb->ioda.tce32_count - | |
1243 | phb->ioda.dma_pe_count; | |
1244 | ||
1245 | pr_info("PCI: Domain %04x has %ld available 32-bit DMA segments\n", | |
1246 | hose->global_number, phb->ioda.tce32_count); | |
1247 | pr_info("PCI: %d PE# for a total weight of %d\n", | |
1248 | phb->ioda.dma_pe_count, phb->ioda.dma_weight); | |
1249 | ||
1250 | /* Walk our PE list and configure their DMA segments, hand them | |
1251 | * out one base segment plus any residual segments based on | |
1252 | * weight | |
1253 | */ | |
1254 | remaining = phb->ioda.tce32_count; | |
1255 | tw = phb->ioda.dma_weight; | |
1256 | base = 0; | |
7ebdf956 | 1257 | list_for_each_entry(pe, &phb->ioda.pe_dma_list, dma_link) { |
184cd4a3 BH |
1258 | if (!pe->dma_weight) |
1259 | continue; | |
1260 | if (!remaining) { | |
1261 | pe_warn(pe, "No DMA32 resources available\n"); | |
1262 | continue; | |
1263 | } | |
1264 | segs = 1; | |
1265 | if (residual) { | |
1266 | segs += ((pe->dma_weight * residual) + (tw / 2)) / tw; | |
1267 | if (segs > remaining) | |
1268 | segs = remaining; | |
1269 | } | |
373f5657 GS |
1270 | |
1271 | /* | |
1272 | * For IODA2 compliant PHB3, we needn't care about the weight. | |
1273 | * The all available 32-bits DMA space will be assigned to | |
1274 | * the specific PE. | |
1275 | */ | |
1276 | if (phb->type == PNV_PHB_IODA1) { | |
1277 | pe_info(pe, "DMA weight %d, assigned %d DMA32 segments\n", | |
1278 | pe->dma_weight, segs); | |
1279 | pnv_pci_ioda_setup_dma_pe(phb, pe, base, segs); | |
1280 | } else { | |
1281 | pe_info(pe, "Assign DMA32 space\n"); | |
1282 | segs = 0; | |
1283 | pnv_pci_ioda2_setup_dma_pe(phb, pe); | |
1284 | } | |
1285 | ||
184cd4a3 BH |
1286 | remaining -= segs; |
1287 | base += segs; | |
1288 | } | |
1289 | } | |
1290 | ||
1291 | #ifdef CONFIG_PCI_MSI | |
137436c9 GS |
1292 | static void pnv_ioda2_msi_eoi(struct irq_data *d) |
1293 | { | |
1294 | unsigned int hw_irq = (unsigned int)irqd_to_hwirq(d); | |
1295 | struct irq_chip *chip = irq_data_get_irq_chip(d); | |
1296 | struct pnv_phb *phb = container_of(chip, struct pnv_phb, | |
1297 | ioda.irq_chip); | |
1298 | int64_t rc; | |
1299 | ||
1300 | rc = opal_pci_msi_eoi(phb->opal_id, hw_irq); | |
1301 | WARN_ON_ONCE(rc); | |
1302 | ||
1303 | icp_native_eoi(d); | |
1304 | } | |
1305 | ||
184cd4a3 | 1306 | static int pnv_pci_ioda_msi_setup(struct pnv_phb *phb, struct pci_dev *dev, |
137436c9 GS |
1307 | unsigned int hwirq, unsigned int virq, |
1308 | unsigned int is_64, struct msi_msg *msg) | |
184cd4a3 BH |
1309 | { |
1310 | struct pnv_ioda_pe *pe = pnv_ioda_get_pe(dev); | |
b72c1f65 | 1311 | struct pci_dn *pdn = pci_get_pdn(dev); |
137436c9 GS |
1312 | struct irq_data *idata; |
1313 | struct irq_chip *ichip; | |
184cd4a3 | 1314 | unsigned int xive_num = hwirq - phb->msi_base; |
3a1a4661 | 1315 | __be32 data; |
184cd4a3 BH |
1316 | int rc; |
1317 | ||
1318 | /* No PE assigned ? bail out ... no MSI for you ! */ | |
1319 | if (pe == NULL) | |
1320 | return -ENXIO; | |
1321 | ||
1322 | /* Check if we have an MVE */ | |
1323 | if (pe->mve_number < 0) | |
1324 | return -ENXIO; | |
1325 | ||
b72c1f65 BH |
1326 | /* Force 32-bit MSI on some broken devices */ |
1327 | if (pdn && pdn->force_32bit_msi) | |
1328 | is_64 = 0; | |
1329 | ||
184cd4a3 BH |
1330 | /* Assign XIVE to PE */ |
1331 | rc = opal_pci_set_xive_pe(phb->opal_id, pe->pe_number, xive_num); | |
1332 | if (rc) { | |
1333 | pr_warn("%s: OPAL error %d setting XIVE %d PE\n", | |
1334 | pci_name(dev), rc, xive_num); | |
1335 | return -EIO; | |
1336 | } | |
1337 | ||
1338 | if (is_64) { | |
3a1a4661 BH |
1339 | __be64 addr64; |
1340 | ||
184cd4a3 BH |
1341 | rc = opal_get_msi_64(phb->opal_id, pe->mve_number, xive_num, 1, |
1342 | &addr64, &data); | |
1343 | if (rc) { | |
1344 | pr_warn("%s: OPAL error %d getting 64-bit MSI data\n", | |
1345 | pci_name(dev), rc); | |
1346 | return -EIO; | |
1347 | } | |
3a1a4661 BH |
1348 | msg->address_hi = be64_to_cpu(addr64) >> 32; |
1349 | msg->address_lo = be64_to_cpu(addr64) & 0xfffffffful; | |
184cd4a3 | 1350 | } else { |
3a1a4661 BH |
1351 | __be32 addr32; |
1352 | ||
184cd4a3 BH |
1353 | rc = opal_get_msi_32(phb->opal_id, pe->mve_number, xive_num, 1, |
1354 | &addr32, &data); | |
1355 | if (rc) { | |
1356 | pr_warn("%s: OPAL error %d getting 32-bit MSI data\n", | |
1357 | pci_name(dev), rc); | |
1358 | return -EIO; | |
1359 | } | |
1360 | msg->address_hi = 0; | |
3a1a4661 | 1361 | msg->address_lo = be32_to_cpu(addr32); |
184cd4a3 | 1362 | } |
3a1a4661 | 1363 | msg->data = be32_to_cpu(data); |
184cd4a3 | 1364 | |
137436c9 GS |
1365 | /* |
1366 | * Change the IRQ chip for the MSI interrupts on PHB3. | |
1367 | * The corresponding IRQ chip should be populated for | |
1368 | * the first time. | |
1369 | */ | |
1370 | if (phb->type == PNV_PHB_IODA2) { | |
1371 | if (!phb->ioda.irq_chip_init) { | |
1372 | idata = irq_get_irq_data(virq); | |
1373 | ichip = irq_data_get_irq_chip(idata); | |
1374 | phb->ioda.irq_chip_init = 1; | |
1375 | phb->ioda.irq_chip = *ichip; | |
1376 | phb->ioda.irq_chip.irq_eoi = pnv_ioda2_msi_eoi; | |
1377 | } | |
1378 | ||
1379 | irq_set_chip(virq, &phb->ioda.irq_chip); | |
1380 | } | |
1381 | ||
184cd4a3 BH |
1382 | pr_devel("%s: %s-bit MSI on hwirq %x (xive #%d)," |
1383 | " address=%x_%08x data=%x PE# %d\n", | |
1384 | pci_name(dev), is_64 ? "64" : "32", hwirq, xive_num, | |
1385 | msg->address_hi, msg->address_lo, data, pe->pe_number); | |
1386 | ||
1387 | return 0; | |
1388 | } | |
1389 | ||
1390 | static void pnv_pci_init_ioda_msis(struct pnv_phb *phb) | |
1391 | { | |
fb1b55d6 | 1392 | unsigned int count; |
184cd4a3 BH |
1393 | const __be32 *prop = of_get_property(phb->hose->dn, |
1394 | "ibm,opal-msi-ranges", NULL); | |
1395 | if (!prop) { | |
1396 | /* BML Fallback */ | |
1397 | prop = of_get_property(phb->hose->dn, "msi-ranges", NULL); | |
1398 | } | |
1399 | if (!prop) | |
1400 | return; | |
1401 | ||
1402 | phb->msi_base = be32_to_cpup(prop); | |
fb1b55d6 GS |
1403 | count = be32_to_cpup(prop + 1); |
1404 | if (msi_bitmap_alloc(&phb->msi_bmp, count, phb->hose->dn)) { | |
184cd4a3 BH |
1405 | pr_err("PCI %d: Failed to allocate MSI bitmap !\n", |
1406 | phb->hose->global_number); | |
1407 | return; | |
1408 | } | |
fb1b55d6 | 1409 | |
184cd4a3 BH |
1410 | phb->msi_setup = pnv_pci_ioda_msi_setup; |
1411 | phb->msi32_support = 1; | |
1412 | pr_info(" Allocated bitmap for %d MSIs (base IRQ 0x%x)\n", | |
fb1b55d6 | 1413 | count, phb->msi_base); |
184cd4a3 BH |
1414 | } |
1415 | #else | |
1416 | static void pnv_pci_init_ioda_msis(struct pnv_phb *phb) { } | |
1417 | #endif /* CONFIG_PCI_MSI */ | |
1418 | ||
11685bec GS |
1419 | /* |
1420 | * This function is supposed to be called on basis of PE from top | |
1421 | * to bottom style. So the the I/O or MMIO segment assigned to | |
1422 | * parent PE could be overrided by its child PEs if necessary. | |
1423 | */ | |
cad5cef6 GKH |
1424 | static void pnv_ioda_setup_pe_seg(struct pci_controller *hose, |
1425 | struct pnv_ioda_pe *pe) | |
11685bec GS |
1426 | { |
1427 | struct pnv_phb *phb = hose->private_data; | |
1428 | struct pci_bus_region region; | |
1429 | struct resource *res; | |
1430 | int i, index; | |
1431 | int rc; | |
1432 | ||
1433 | /* | |
1434 | * NOTE: We only care PCI bus based PE for now. For PCI | |
1435 | * device based PE, for example SRIOV sensitive VF should | |
1436 | * be figured out later. | |
1437 | */ | |
1438 | BUG_ON(!(pe->flags & (PNV_IODA_PE_BUS | PNV_IODA_PE_BUS_ALL))); | |
1439 | ||
1440 | pci_bus_for_each_resource(pe->pbus, res, i) { | |
1441 | if (!res || !res->flags || | |
1442 | res->start > res->end) | |
1443 | continue; | |
1444 | ||
1445 | if (res->flags & IORESOURCE_IO) { | |
1446 | region.start = res->start - phb->ioda.io_pci_base; | |
1447 | region.end = res->end - phb->ioda.io_pci_base; | |
1448 | index = region.start / phb->ioda.io_segsize; | |
1449 | ||
1450 | while (index < phb->ioda.total_pe && | |
1451 | region.start <= region.end) { | |
1452 | phb->ioda.io_segmap[index] = pe->pe_number; | |
1453 | rc = opal_pci_map_pe_mmio_window(phb->opal_id, | |
1454 | pe->pe_number, OPAL_IO_WINDOW_TYPE, 0, index); | |
1455 | if (rc != OPAL_SUCCESS) { | |
1456 | pr_err("%s: OPAL error %d when mapping IO " | |
1457 | "segment #%d to PE#%d\n", | |
1458 | __func__, rc, index, pe->pe_number); | |
1459 | break; | |
1460 | } | |
1461 | ||
1462 | region.start += phb->ioda.io_segsize; | |
1463 | index++; | |
1464 | } | |
1465 | } else if (res->flags & IORESOURCE_MEM) { | |
1466 | region.start = res->start - | |
3fd47f06 | 1467 | hose->mem_offset[0] - |
11685bec GS |
1468 | phb->ioda.m32_pci_base; |
1469 | region.end = res->end - | |
3fd47f06 | 1470 | hose->mem_offset[0] - |
11685bec GS |
1471 | phb->ioda.m32_pci_base; |
1472 | index = region.start / phb->ioda.m32_segsize; | |
1473 | ||
1474 | while (index < phb->ioda.total_pe && | |
1475 | region.start <= region.end) { | |
1476 | phb->ioda.m32_segmap[index] = pe->pe_number; | |
1477 | rc = opal_pci_map_pe_mmio_window(phb->opal_id, | |
1478 | pe->pe_number, OPAL_M32_WINDOW_TYPE, 0, index); | |
1479 | if (rc != OPAL_SUCCESS) { | |
1480 | pr_err("%s: OPAL error %d when mapping M32 " | |
1481 | "segment#%d to PE#%d", | |
1482 | __func__, rc, index, pe->pe_number); | |
1483 | break; | |
1484 | } | |
1485 | ||
1486 | region.start += phb->ioda.m32_segsize; | |
1487 | index++; | |
1488 | } | |
1489 | } | |
1490 | } | |
1491 | } | |
1492 | ||
cad5cef6 | 1493 | static void pnv_pci_ioda_setup_seg(void) |
11685bec GS |
1494 | { |
1495 | struct pci_controller *tmp, *hose; | |
1496 | struct pnv_phb *phb; | |
1497 | struct pnv_ioda_pe *pe; | |
1498 | ||
1499 | list_for_each_entry_safe(hose, tmp, &hose_list, list_node) { | |
1500 | phb = hose->private_data; | |
1501 | list_for_each_entry(pe, &phb->ioda.pe_list, list) { | |
1502 | pnv_ioda_setup_pe_seg(hose, pe); | |
1503 | } | |
1504 | } | |
1505 | } | |
1506 | ||
cad5cef6 | 1507 | static void pnv_pci_ioda_setup_DMA(void) |
13395c48 GS |
1508 | { |
1509 | struct pci_controller *hose, *tmp; | |
db1266c8 | 1510 | struct pnv_phb *phb; |
13395c48 GS |
1511 | |
1512 | list_for_each_entry_safe(hose, tmp, &hose_list, list_node) { | |
1513 | pnv_ioda_setup_dma(hose->private_data); | |
db1266c8 GS |
1514 | |
1515 | /* Mark the PHB initialization done */ | |
1516 | phb = hose->private_data; | |
1517 | phb->initialized = 1; | |
13395c48 GS |
1518 | } |
1519 | } | |
1520 | ||
37c367f2 GS |
1521 | static void pnv_pci_ioda_create_dbgfs(void) |
1522 | { | |
1523 | #ifdef CONFIG_DEBUG_FS | |
1524 | struct pci_controller *hose, *tmp; | |
1525 | struct pnv_phb *phb; | |
1526 | char name[16]; | |
1527 | ||
1528 | list_for_each_entry_safe(hose, tmp, &hose_list, list_node) { | |
1529 | phb = hose->private_data; | |
1530 | ||
1531 | sprintf(name, "PCI%04x", hose->global_number); | |
1532 | phb->dbgfs = debugfs_create_dir(name, powerpc_debugfs_root); | |
1533 | if (!phb->dbgfs) | |
1534 | pr_warning("%s: Error on creating debugfs on PHB#%x\n", | |
1535 | __func__, hose->global_number); | |
1536 | } | |
1537 | #endif /* CONFIG_DEBUG_FS */ | |
1538 | } | |
1539 | ||
cad5cef6 | 1540 | static void pnv_pci_ioda_fixup(void) |
fb446ad0 GS |
1541 | { |
1542 | pnv_pci_ioda_setup_PEs(); | |
11685bec | 1543 | pnv_pci_ioda_setup_seg(); |
13395c48 | 1544 | pnv_pci_ioda_setup_DMA(); |
e9cc17d4 | 1545 | |
37c367f2 GS |
1546 | pnv_pci_ioda_create_dbgfs(); |
1547 | ||
e9cc17d4 | 1548 | #ifdef CONFIG_EEH |
e9cc17d4 | 1549 | eeh_init(); |
dadcd6d6 | 1550 | eeh_addr_cache_build(); |
e9cc17d4 | 1551 | #endif |
fb446ad0 GS |
1552 | } |
1553 | ||
271fd03a GS |
1554 | /* |
1555 | * Returns the alignment for I/O or memory windows for P2P | |
1556 | * bridges. That actually depends on how PEs are segmented. | |
1557 | * For now, we return I/O or M32 segment size for PE sensitive | |
1558 | * P2P bridges. Otherwise, the default values (4KiB for I/O, | |
1559 | * 1MiB for memory) will be returned. | |
1560 | * | |
1561 | * The current PCI bus might be put into one PE, which was | |
1562 | * create against the parent PCI bridge. For that case, we | |
1563 | * needn't enlarge the alignment so that we can save some | |
1564 | * resources. | |
1565 | */ | |
1566 | static resource_size_t pnv_pci_window_alignment(struct pci_bus *bus, | |
1567 | unsigned long type) | |
1568 | { | |
1569 | struct pci_dev *bridge; | |
1570 | struct pci_controller *hose = pci_bus_to_host(bus); | |
1571 | struct pnv_phb *phb = hose->private_data; | |
1572 | int num_pci_bridges = 0; | |
1573 | ||
1574 | bridge = bus->self; | |
1575 | while (bridge) { | |
1576 | if (pci_pcie_type(bridge) == PCI_EXP_TYPE_PCI_BRIDGE) { | |
1577 | num_pci_bridges++; | |
1578 | if (num_pci_bridges >= 2) | |
1579 | return 1; | |
1580 | } | |
1581 | ||
1582 | bridge = bridge->bus->self; | |
1583 | } | |
1584 | ||
262af557 GC |
1585 | /* We fail back to M32 if M64 isn't supported */ |
1586 | if (phb->ioda.m64_segsize && | |
1587 | pnv_pci_is_mem_pref_64(type)) | |
1588 | return phb->ioda.m64_segsize; | |
271fd03a GS |
1589 | if (type & IORESOURCE_MEM) |
1590 | return phb->ioda.m32_segsize; | |
1591 | ||
1592 | return phb->ioda.io_segsize; | |
1593 | } | |
1594 | ||
184cd4a3 BH |
1595 | /* Prevent enabling devices for which we couldn't properly |
1596 | * assign a PE | |
1597 | */ | |
cad5cef6 | 1598 | static int pnv_pci_enable_device_hook(struct pci_dev *dev) |
184cd4a3 | 1599 | { |
db1266c8 GS |
1600 | struct pci_controller *hose = pci_bus_to_host(dev->bus); |
1601 | struct pnv_phb *phb = hose->private_data; | |
1602 | struct pci_dn *pdn; | |
184cd4a3 | 1603 | |
db1266c8 GS |
1604 | /* The function is probably called while the PEs have |
1605 | * not be created yet. For example, resource reassignment | |
1606 | * during PCI probe period. We just skip the check if | |
1607 | * PEs isn't ready. | |
1608 | */ | |
1609 | if (!phb->initialized) | |
1610 | return 0; | |
1611 | ||
b72c1f65 | 1612 | pdn = pci_get_pdn(dev); |
184cd4a3 BH |
1613 | if (!pdn || pdn->pe_number == IODA_INVALID_PE) |
1614 | return -EINVAL; | |
db1266c8 | 1615 | |
184cd4a3 BH |
1616 | return 0; |
1617 | } | |
1618 | ||
1619 | static u32 pnv_ioda_bdfn_to_pe(struct pnv_phb *phb, struct pci_bus *bus, | |
1620 | u32 devfn) | |
1621 | { | |
1622 | return phb->ioda.pe_rmap[(bus->number << 8) | devfn]; | |
1623 | } | |
1624 | ||
73ed148a BH |
1625 | static void pnv_pci_ioda_shutdown(struct pnv_phb *phb) |
1626 | { | |
1627 | opal_pci_reset(phb->opal_id, OPAL_PCI_IODA_TABLE_RESET, | |
1628 | OPAL_ASSERT_RESET); | |
1629 | } | |
1630 | ||
e51df2c1 AB |
1631 | static void __init pnv_pci_init_ioda_phb(struct device_node *np, |
1632 | u64 hub_id, int ioda_type) | |
184cd4a3 BH |
1633 | { |
1634 | struct pci_controller *hose; | |
184cd4a3 | 1635 | struct pnv_phb *phb; |
8184616f | 1636 | unsigned long size, m32map_off, pemap_off, iomap_off = 0; |
c681b93c | 1637 | const __be64 *prop64; |
3a1a4661 | 1638 | const __be32 *prop32; |
f1b7cc3e | 1639 | int len; |
184cd4a3 BH |
1640 | u64 phb_id; |
1641 | void *aux; | |
1642 | long rc; | |
1643 | ||
58d714ec | 1644 | pr_info("Initializing IODA%d OPAL PHB %s\n", ioda_type, np->full_name); |
184cd4a3 BH |
1645 | |
1646 | prop64 = of_get_property(np, "ibm,opal-phbid", NULL); | |
1647 | if (!prop64) { | |
1648 | pr_err(" Missing \"ibm,opal-phbid\" property !\n"); | |
1649 | return; | |
1650 | } | |
1651 | phb_id = be64_to_cpup(prop64); | |
1652 | pr_debug(" PHB-ID : 0x%016llx\n", phb_id); | |
1653 | ||
1654 | phb = alloc_bootmem(sizeof(struct pnv_phb)); | |
58d714ec GS |
1655 | if (!phb) { |
1656 | pr_err(" Out of memory !\n"); | |
1657 | return; | |
184cd4a3 | 1658 | } |
58d714ec GS |
1659 | |
1660 | /* Allocate PCI controller */ | |
1661 | memset(phb, 0, sizeof(struct pnv_phb)); | |
1662 | phb->hose = hose = pcibios_alloc_controller(np); | |
1663 | if (!phb->hose) { | |
1664 | pr_err(" Can't allocate PCI controller for %s\n", | |
184cd4a3 | 1665 | np->full_name); |
58d714ec | 1666 | free_bootmem((unsigned long)phb, sizeof(struct pnv_phb)); |
184cd4a3 BH |
1667 | return; |
1668 | } | |
1669 | ||
1670 | spin_lock_init(&phb->lock); | |
f1b7cc3e GS |
1671 | prop32 = of_get_property(np, "bus-range", &len); |
1672 | if (prop32 && len == 8) { | |
3a1a4661 BH |
1673 | hose->first_busno = be32_to_cpu(prop32[0]); |
1674 | hose->last_busno = be32_to_cpu(prop32[1]); | |
f1b7cc3e GS |
1675 | } else { |
1676 | pr_warn(" Broken <bus-range> on %s\n", np->full_name); | |
1677 | hose->first_busno = 0; | |
1678 | hose->last_busno = 0xff; | |
1679 | } | |
184cd4a3 | 1680 | hose->private_data = phb; |
e9cc17d4 | 1681 | phb->hub_id = hub_id; |
184cd4a3 | 1682 | phb->opal_id = phb_id; |
aa0c033f | 1683 | phb->type = ioda_type; |
184cd4a3 | 1684 | |
cee72d5b BH |
1685 | /* Detect specific models for error handling */ |
1686 | if (of_device_is_compatible(np, "ibm,p7ioc-pciex")) | |
1687 | phb->model = PNV_PHB_MODEL_P7IOC; | |
f3d40c25 | 1688 | else if (of_device_is_compatible(np, "ibm,power8-pciex")) |
aa0c033f | 1689 | phb->model = PNV_PHB_MODEL_PHB3; |
cee72d5b BH |
1690 | else |
1691 | phb->model = PNV_PHB_MODEL_UNKNOWN; | |
1692 | ||
aa0c033f | 1693 | /* Parse 32-bit and IO ranges (if any) */ |
2f1ec02e | 1694 | pci_process_bridge_OF_ranges(hose, np, !hose->global_number); |
184cd4a3 | 1695 | |
aa0c033f | 1696 | /* Get registers */ |
184cd4a3 BH |
1697 | phb->regs = of_iomap(np, 0); |
1698 | if (phb->regs == NULL) | |
1699 | pr_err(" Failed to map registers !\n"); | |
1700 | ||
184cd4a3 | 1701 | /* Initialize more IODA stuff */ |
36954dc7 | 1702 | phb->ioda.total_pe = 1; |
aa0c033f | 1703 | prop32 = of_get_property(np, "ibm,opal-num-pes", NULL); |
36954dc7 | 1704 | if (prop32) |
3a1a4661 | 1705 | phb->ioda.total_pe = be32_to_cpup(prop32); |
36954dc7 GS |
1706 | prop32 = of_get_property(np, "ibm,opal-reserved-pe", NULL); |
1707 | if (prop32) | |
1708 | phb->ioda.reserved_pe = be32_to_cpup(prop32); | |
262af557 GC |
1709 | |
1710 | /* Parse 64-bit MMIO range */ | |
1711 | pnv_ioda_parse_m64_window(phb); | |
1712 | ||
184cd4a3 | 1713 | phb->ioda.m32_size = resource_size(&hose->mem_resources[0]); |
aa0c033f | 1714 | /* FW Has already off top 64k of M32 space (MSI space) */ |
184cd4a3 BH |
1715 | phb->ioda.m32_size += 0x10000; |
1716 | ||
1717 | phb->ioda.m32_segsize = phb->ioda.m32_size / phb->ioda.total_pe; | |
3fd47f06 | 1718 | phb->ioda.m32_pci_base = hose->mem_resources[0].start - hose->mem_offset[0]; |
184cd4a3 BH |
1719 | phb->ioda.io_size = hose->pci_io_size; |
1720 | phb->ioda.io_segsize = phb->ioda.io_size / phb->ioda.total_pe; | |
1721 | phb->ioda.io_pci_base = 0; /* XXX calculate this ? */ | |
1722 | ||
c35d2a8c | 1723 | /* Allocate aux data & arrays. We don't have IO ports on PHB3 */ |
184cd4a3 BH |
1724 | size = _ALIGN_UP(phb->ioda.total_pe / 8, sizeof(unsigned long)); |
1725 | m32map_off = size; | |
e47747f4 | 1726 | size += phb->ioda.total_pe * sizeof(phb->ioda.m32_segmap[0]); |
c35d2a8c GS |
1727 | if (phb->type == PNV_PHB_IODA1) { |
1728 | iomap_off = size; | |
1729 | size += phb->ioda.total_pe * sizeof(phb->ioda.io_segmap[0]); | |
1730 | } | |
184cd4a3 BH |
1731 | pemap_off = size; |
1732 | size += phb->ioda.total_pe * sizeof(struct pnv_ioda_pe); | |
1733 | aux = alloc_bootmem(size); | |
1734 | memset(aux, 0, size); | |
1735 | phb->ioda.pe_alloc = aux; | |
1736 | phb->ioda.m32_segmap = aux + m32map_off; | |
c35d2a8c GS |
1737 | if (phb->type == PNV_PHB_IODA1) |
1738 | phb->ioda.io_segmap = aux + iomap_off; | |
184cd4a3 | 1739 | phb->ioda.pe_array = aux + pemap_off; |
36954dc7 | 1740 | set_bit(phb->ioda.reserved_pe, phb->ioda.pe_alloc); |
184cd4a3 | 1741 | |
7ebdf956 | 1742 | INIT_LIST_HEAD(&phb->ioda.pe_dma_list); |
184cd4a3 BH |
1743 | INIT_LIST_HEAD(&phb->ioda.pe_list); |
1744 | ||
1745 | /* Calculate how many 32-bit TCE segments we have */ | |
1746 | phb->ioda.tce32_count = phb->ioda.m32_pci_base >> 28; | |
1747 | ||
aa0c033f | 1748 | #if 0 /* We should really do that ... */ |
184cd4a3 BH |
1749 | rc = opal_pci_set_phb_mem_window(opal->phb_id, |
1750 | window_type, | |
1751 | window_num, | |
1752 | starting_real_address, | |
1753 | starting_pci_address, | |
1754 | segment_size); | |
1755 | #endif | |
1756 | ||
262af557 GC |
1757 | pr_info(" %03d (%03d) PE's M32: 0x%x [segment=0x%x]\n", |
1758 | phb->ioda.total_pe, phb->ioda.reserved_pe, | |
1759 | phb->ioda.m32_size, phb->ioda.m32_segsize); | |
1760 | if (phb->ioda.m64_size) | |
1761 | pr_info(" M64: 0x%lx [segment=0x%lx]\n", | |
1762 | phb->ioda.m64_size, phb->ioda.m64_segsize); | |
1763 | if (phb->ioda.io_size) | |
1764 | pr_info(" IO: 0x%x [segment=0x%x]\n", | |
1765 | phb->ioda.io_size, phb->ioda.io_segsize); | |
1766 | ||
184cd4a3 | 1767 | |
184cd4a3 | 1768 | phb->hose->ops = &pnv_pci_ops; |
49dec922 GS |
1769 | phb->get_pe_state = pnv_ioda_get_pe_state; |
1770 | phb->freeze_pe = pnv_ioda_freeze_pe; | |
1771 | phb->unfreeze_pe = pnv_ioda_unfreeze_pe; | |
e9cc17d4 GS |
1772 | #ifdef CONFIG_EEH |
1773 | phb->eeh_ops = &ioda_eeh_ops; | |
1774 | #endif | |
184cd4a3 BH |
1775 | |
1776 | /* Setup RID -> PE mapping function */ | |
1777 | phb->bdfn_to_pe = pnv_ioda_bdfn_to_pe; | |
1778 | ||
1779 | /* Setup TCEs */ | |
1780 | phb->dma_dev_setup = pnv_pci_ioda_dma_dev_setup; | |
cd15b048 | 1781 | phb->dma_set_mask = pnv_pci_ioda_dma_set_mask; |
184cd4a3 | 1782 | |
73ed148a BH |
1783 | /* Setup shutdown function for kexec */ |
1784 | phb->shutdown = pnv_pci_ioda_shutdown; | |
1785 | ||
184cd4a3 BH |
1786 | /* Setup MSI support */ |
1787 | pnv_pci_init_ioda_msis(phb); | |
1788 | ||
c40a4210 GS |
1789 | /* |
1790 | * We pass the PCI probe flag PCI_REASSIGN_ALL_RSRC here | |
1791 | * to let the PCI core do resource assignment. It's supposed | |
1792 | * that the PCI core will do correct I/O and MMIO alignment | |
1793 | * for the P2P bridge bars so that each PCI bus (excluding | |
1794 | * the child P2P bridges) can form individual PE. | |
184cd4a3 | 1795 | */ |
fb446ad0 | 1796 | ppc_md.pcibios_fixup = pnv_pci_ioda_fixup; |
184cd4a3 | 1797 | ppc_md.pcibios_enable_device_hook = pnv_pci_enable_device_hook; |
271fd03a | 1798 | ppc_md.pcibios_window_alignment = pnv_pci_window_alignment; |
d92a208d | 1799 | ppc_md.pcibios_reset_secondary_bus = pnv_pci_reset_secondary_bus; |
c40a4210 | 1800 | pci_add_flags(PCI_REASSIGN_ALL_RSRC); |
184cd4a3 BH |
1801 | |
1802 | /* Reset IODA tables to a clean state */ | |
f11fe552 | 1803 | rc = opal_pci_reset(phb_id, OPAL_PCI_IODA_TABLE_RESET, OPAL_ASSERT_RESET); |
184cd4a3 | 1804 | if (rc) |
f11fe552 | 1805 | pr_warning(" OPAL Error %ld performing IODA table reset !\n", rc); |
361f2a2a GS |
1806 | |
1807 | /* If we're running in kdump kerenl, the previous kerenl never | |
1808 | * shutdown PCI devices correctly. We already got IODA table | |
1809 | * cleaned out. So we have to issue PHB reset to stop all PCI | |
1810 | * transactions from previous kerenl. | |
1811 | */ | |
1812 | if (is_kdump_kernel()) { | |
1813 | pr_info(" Issue PHB reset ...\n"); | |
1814 | ioda_eeh_phb_reset(hose, EEH_RESET_FUNDAMENTAL); | |
1815 | ioda_eeh_phb_reset(hose, OPAL_DEASSERT_RESET); | |
1816 | } | |
262af557 GC |
1817 | |
1818 | /* Configure M64 window */ | |
1819 | if (phb->init_m64 && phb->init_m64(phb)) | |
1820 | hose->mem_resources[1].flags = 0; | |
aa0c033f GS |
1821 | } |
1822 | ||
67975005 | 1823 | void __init pnv_pci_init_ioda2_phb(struct device_node *np) |
aa0c033f | 1824 | { |
e9cc17d4 | 1825 | pnv_pci_init_ioda_phb(np, 0, PNV_PHB_IODA2); |
184cd4a3 BH |
1826 | } |
1827 | ||
1828 | void __init pnv_pci_init_ioda_hub(struct device_node *np) | |
1829 | { | |
1830 | struct device_node *phbn; | |
c681b93c | 1831 | const __be64 *prop64; |
184cd4a3 BH |
1832 | u64 hub_id; |
1833 | ||
1834 | pr_info("Probing IODA IO-Hub %s\n", np->full_name); | |
1835 | ||
1836 | prop64 = of_get_property(np, "ibm,opal-hubid", NULL); | |
1837 | if (!prop64) { | |
1838 | pr_err(" Missing \"ibm,opal-hubid\" property !\n"); | |
1839 | return; | |
1840 | } | |
1841 | hub_id = be64_to_cpup(prop64); | |
1842 | pr_devel(" HUB-ID : 0x%016llx\n", hub_id); | |
1843 | ||
1844 | /* Count child PHBs */ | |
1845 | for_each_child_of_node(np, phbn) { | |
1846 | /* Look for IODA1 PHBs */ | |
1847 | if (of_device_is_compatible(phbn, "ibm,ioda-phb")) | |
e9cc17d4 | 1848 | pnv_pci_init_ioda_phb(phbn, hub_id, PNV_PHB_IODA1); |
184cd4a3 BH |
1849 | } |
1850 | } |