]>
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> |
80c49c7e IM |
40 | #include <asm/pnv-pci.h> |
41 | ||
ec249dd8 | 42 | #include <misc/cxl-base.h> |
184cd4a3 BH |
43 | |
44 | #include "powernv.h" | |
45 | #include "pci.h" | |
46 | ||
781a868f WY |
47 | /* 256M DMA window, 4K TCE pages, 8 bytes TCE */ |
48 | #define TCE32_TABLE_SIZE ((0x10000000 / 0x1000) * 8) | |
49 | ||
6d31c2fa JP |
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 | ||
781a868f | 62 | if (pe->flags & PNV_IODA_PE_DEV) |
6d31c2fa | 63 | strlcpy(pfix, dev_name(&pe->pdev->dev), sizeof(pfix)); |
781a868f | 64 | else if (pe->flags & (PNV_IODA_PE_BUS | PNV_IODA_PE_BUS_ALL)) |
6d31c2fa JP |
65 | sprintf(pfix, "%04x:%02x ", |
66 | pci_domain_nr(pe->pbus), pe->pbus->number); | |
781a868f WY |
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*/ | |
6d31c2fa JP |
74 | |
75 | printk("%spci %s: [PE# %.3d] %pV", | |
76 | level, pfix, pe->pe_number, &vaf); | |
77 | ||
78 | va_end(args); | |
79 | } | |
184cd4a3 | 80 | |
6d31c2fa JP |
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__) | |
184cd4a3 | 87 | |
4e287840 TLSC |
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 | ||
8e0a1611 AK |
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 | ||
262af557 GC |
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 | ||
4b82ab18 GS |
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 | ||
cad5cef6 | 144 | static int pnv_ioda_alloc_pe(struct pnv_phb *phb) |
184cd4a3 BH |
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 | ||
4cce9550 | 155 | phb->ioda.pe_array[pe].phb = phb; |
184cd4a3 BH |
156 | phb->ioda.pe_array[pe].pe_number = pe; |
157 | return pe; | |
158 | } | |
159 | ||
cad5cef6 | 160 | static void pnv_ioda_free_pe(struct pnv_phb *phb, int pe) |
184cd4a3 BH |
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 | ||
262af557 GC |
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 | ||
5ef73567 | 225 | static void pnv_ioda2_reserve_m64_pe(struct pnv_phb *phb) |
262af557 GC |
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) { | |
4b82ab18 GS |
238 | for (i = 0; i < PCI_BRIDGE_RESOURCE_NUM; i++) { |
239 | r = &pdev->resource[PCI_BRIDGE_RESOURCES + i]; | |
262af557 GC |
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++) | |
4b82ab18 | 246 | pnv_ioda_reserve_pe(phb, base + step); |
262af557 GC |
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]; | |
262af557 GC |
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 | ||
1665c4a8 GS |
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 | ||
262af557 GC |
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 | ||
262af557 GC |
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 | ||
e9863e68 WY |
392 | pr_info(" MEM64 0x%016llx..0x%016llx -> 0x%016llx\n", |
393 | res->start, res->end, pci_addr); | |
394 | ||
262af557 GC |
395 | /* Use last M64 BAR to cover M64 window */ |
396 | phb->ioda.m64_bar_idx = 15; | |
397 | phb->init_m64 = pnv_ioda2_init_m64; | |
5ef73567 | 398 | phb->reserve_m64_pe = pnv_ioda2_reserve_m64_pe; |
262af557 GC |
399 | phb->pick_m64_pe = pnv_ioda2_pick_m64_pe; |
400 | } | |
401 | ||
49dec922 GS |
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; | |
ec8e4e9d GS |
411 | if (WARN_ON(!pe || !(pe->flags & PNV_IODA_PE_MASTER))) |
412 | return; | |
413 | ||
49dec922 GS |
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 | ||
e51df2c1 | 442 | static int pnv_ioda_unfreeze_pe(struct pnv_phb *phb, int pe_no, int opt) |
49dec922 GS |
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 | ||
184cd4a3 BH |
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 | |
cad5cef6 | 548 | static struct pnv_ioda_pe *pnv_ioda_get_pe(struct pci_dev *dev) |
184cd4a3 BH |
549 | { |
550 | struct pci_controller *hose = pci_bus_to_host(dev->bus); | |
551 | struct pnv_phb *phb = hose->private_data; | |
b72c1f65 | 552 | struct pci_dn *pdn = pci_get_pdn(dev); |
184cd4a3 BH |
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 | } | |
184cd4a3 BH |
560 | #endif /* CONFIG_PCI_MSI */ |
561 | ||
b131a842 GS |
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; | |
781a868f | 604 | struct pci_dev *pdev = NULL; |
b131a842 GS |
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; | |
781a868f | 643 | else if (pe->flags & PNV_IODA_PE_DEV) |
b131a842 | 644 | pdev = pe->pdev->bus->self; |
781a868f WY |
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 */ | |
b131a842 GS |
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 | ||
781a868f WY |
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 | ||
cad5cef6 | 747 | static int pnv_ioda_configure_pe(struct pnv_phb *phb, struct pnv_ioda_pe *pe) |
184cd4a3 BH |
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; | |
fb446ad0 GS |
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 | ||
184cd4a3 BH |
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: | |
781a868f WY |
773 | dev_err(&pe->pbus->dev, "Number of subordinate buses %d unsupported\n", |
774 | count); | |
184cd4a3 BH |
775 | /* Do an exact match only */ |
776 | bcomp = OpalPciBusAll; | |
777 | } | |
778 | rid_end = pe->rid + (count << 8); | |
779 | } else { | |
781a868f WY |
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; | |
184cd4a3 BH |
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 | ||
631ad691 GS |
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 | */ | |
184cd4a3 BH |
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 | } | |
631ad691 | 804 | |
b131a842 GS |
805 | /* Configure PELTV */ |
806 | pnv_ioda_set_peltv(phb, pe, true); | |
184cd4a3 | 807 | |
184cd4a3 BH |
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 */ | |
4773f76b GS |
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); | |
184cd4a3 | 827 | if (rc) { |
4773f76b | 828 | pe_err(pe, "OPAL error %ld enabling MVE %d\n", |
184cd4a3 BH |
829 | rc, pe->mve_number); |
830 | pe->mve_number = -1; | |
184cd4a3 | 831 | } |
4773f76b | 832 | } |
184cd4a3 | 833 | |
4773f76b | 834 | out: |
184cd4a3 BH |
835 | return 0; |
836 | } | |
837 | ||
cad5cef6 GKH |
838 | static void pnv_ioda_link_pe_by_weight(struct pnv_phb *phb, |
839 | struct pnv_ioda_pe *pe) | |
184cd4a3 BH |
840 | { |
841 | struct pnv_ioda_pe *lpe; | |
842 | ||
7ebdf956 | 843 | list_for_each_entry(lpe, &phb->ioda.pe_dma_list, dma_link) { |
184cd4a3 | 844 | if (lpe->dma_weight < pe->dma_weight) { |
7ebdf956 | 845 | list_add_tail(&pe->dma_link, &lpe->dma_link); |
184cd4a3 BH |
846 | return; |
847 | } | |
848 | } | |
7ebdf956 | 849 | list_add_tail(&pe->dma_link, &phb->ioda.pe_dma_list); |
184cd4a3 BH |
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 | ||
781a868f WY |
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 | ||
fb446ad0 | 948 | #if 0 |
cad5cef6 | 949 | static struct pnv_ioda_pe *pnv_ioda_setup_dev_PE(struct pci_dev *dev) |
184cd4a3 BH |
950 | { |
951 | struct pci_controller *hose = pci_bus_to_host(dev->bus); | |
952 | struct pnv_phb *phb = hose->private_data; | |
b72c1f65 | 953 | struct pci_dn *pdn = pci_get_pdn(dev); |
184cd4a3 BH |
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 | } | |
fb446ad0 | 1017 | #endif /* Useful for SRIOV case */ |
184cd4a3 BH |
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) { | |
b72c1f65 | 1024 | struct pci_dn *pdn = pci_get_pdn(dev); |
184cd4a3 BH |
1025 | |
1026 | if (pdn == NULL) { | |
1027 | pr_warn("%s: No device node associated with device !\n", | |
1028 | pci_name(dev)); | |
1029 | continue; | |
1030 | } | |
184cd4a3 BH |
1031 | pdn->pe_number = pe->pe_number; |
1032 | pe->dma_weight += pnv_ioda_dma_weight(dev); | |
fb446ad0 | 1033 | if ((pe->flags & PNV_IODA_PE_BUS_ALL) && dev->subordinate) |
184cd4a3 BH |
1034 | pnv_ioda_setup_same_PE(dev->subordinate, pe); |
1035 | } | |
1036 | } | |
1037 | ||
fb446ad0 GS |
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 | */ | |
cad5cef6 | 1044 | static void pnv_ioda_setup_bus_PE(struct pci_bus *bus, int all) |
184cd4a3 | 1045 | { |
fb446ad0 | 1046 | struct pci_controller *hose = pci_bus_to_host(bus); |
184cd4a3 | 1047 | struct pnv_phb *phb = hose->private_data; |
184cd4a3 | 1048 | struct pnv_ioda_pe *pe; |
262af557 GC |
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); | |
184cd4a3 | 1058 | |
184cd4a3 | 1059 | if (pe_num == IODA_INVALID_PE) { |
fb446ad0 GS |
1060 | pr_warning("%s: Not enough PE# available for PCI bus %04x:%02x\n", |
1061 | __func__, pci_domain_nr(bus), bus->number); | |
184cd4a3 BH |
1062 | return; |
1063 | } | |
1064 | ||
1065 | pe = &phb->ioda.pe_array[pe_num]; | |
262af557 | 1066 | pe->flags |= (all ? PNV_IODA_PE_BUS_ALL : PNV_IODA_PE_BUS); |
184cd4a3 BH |
1067 | pe->pbus = bus; |
1068 | pe->pdev = NULL; | |
1069 | pe->tce32_seg = -1; | |
1070 | pe->mve_number = -1; | |
b918c62e | 1071 | pe->rid = bus->busn_res.start << 8; |
184cd4a3 BH |
1072 | pe->dma_weight = 0; |
1073 | ||
fb446ad0 GS |
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); | |
184cd4a3 BH |
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 | ||
9e8d4a19 WY |
1089 | pe->tce32_table = kzalloc_node(sizeof(struct iommu_table), |
1090 | GFP_KERNEL, hose->node); | |
1091 | pe->tce32_table->data = pe; | |
1092 | ||
184cd4a3 BH |
1093 | /* Associate it with all child devices */ |
1094 | pnv_ioda_setup_same_PE(bus, pe); | |
1095 | ||
7ebdf956 GS |
1096 | /* Put PE to the list */ |
1097 | list_add_tail(&pe->list, &phb->ioda.pe_list); | |
1098 | ||
184cd4a3 BH |
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 | ||
cad5cef6 | 1111 | static void pnv_ioda_setup_PEs(struct pci_bus *bus) |
184cd4a3 BH |
1112 | { |
1113 | struct pci_dev *dev; | |
fb446ad0 GS |
1114 | |
1115 | pnv_ioda_setup_bus_PE(bus, 0); | |
184cd4a3 BH |
1116 | |
1117 | list_for_each_entry(dev, &bus->devices, bus_list) { | |
fb446ad0 GS |
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 | */ | |
cad5cef6 | 1135 | static void pnv_pci_ioda_setup_PEs(void) |
fb446ad0 GS |
1136 | { |
1137 | struct pci_controller *hose, *tmp; | |
262af557 | 1138 | struct pnv_phb *phb; |
fb446ad0 GS |
1139 | |
1140 | list_for_each_entry_safe(hose, tmp, &hose_list, list_node) { | |
262af557 GC |
1141 | phb = hose->private_data; |
1142 | ||
1143 | /* M64 layout might affect PE allocation */ | |
5ef73567 GS |
1144 | if (phb->reserve_m64_pe) |
1145 | phb->reserve_m64_pe(phb); | |
262af557 | 1146 | |
fb446ad0 | 1147 | pnv_ioda_setup_PEs(hose->bus); |
184cd4a3 BH |
1148 | } |
1149 | } | |
1150 | ||
a8b2f828 | 1151 | #ifdef CONFIG_PCI_IOV |
781a868f WY |
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; | |
02639b0e | 1158 | int i, j; |
781a868f WY |
1159 | |
1160 | bus = pdev->bus; | |
1161 | hose = pci_bus_to_host(bus); | |
1162 | phb = hose->private_data; | |
1163 | pdn = pci_get_pdn(pdev); | |
1164 | ||
02639b0e WY |
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 | } | |
781a868f WY |
1174 | |
1175 | return 0; | |
1176 | } | |
1177 | ||
02639b0e | 1178 | static int pnv_pci_vf_assign_m64(struct pci_dev *pdev, u16 num_vfs) |
781a868f WY |
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; | |
02639b0e | 1186 | int i, j; |
781a868f | 1187 | int64_t rc; |
02639b0e WY |
1188 | int total_vfs; |
1189 | resource_size_t size, start; | |
1190 | int pe_num; | |
1191 | int vf_groups; | |
1192 | int vf_per_group; | |
781a868f WY |
1193 | |
1194 | bus = pdev->bus; | |
1195 | hose = pci_bus_to_host(bus); | |
1196 | phb = hose->private_data; | |
1197 | pdn = pci_get_pdn(pdev); | |
02639b0e | 1198 | total_vfs = pci_sriov_get_totalvfs(pdev); |
781a868f WY |
1199 | |
1200 | /* Initialize the m64_wins to IODA_INVALID_M64 */ | |
1201 | for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) | |
02639b0e WY |
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 | } | |
781a868f WY |
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 | ||
02639b0e WY |
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); | |
781a868f | 1257 | |
781a868f | 1258 | |
02639b0e WY |
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 | } | |
781a868f | 1264 | |
02639b0e WY |
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); | |
781a868f | 1271 | |
02639b0e WY |
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 | } | |
781a868f WY |
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 | ||
02639b0e | 1318 | static void pnv_ioda_release_vf_PE(struct pci_dev *pdev, u16 num_vfs) |
781a868f WY |
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; | |
02639b0e WY |
1325 | u16 vf_index; |
1326 | int64_t rc; | |
781a868f WY |
1327 | |
1328 | bus = pdev->bus; | |
1329 | hose = pci_bus_to_host(bus); | |
1330 | phb = hose->private_data; | |
02639b0e | 1331 | pdn = pci_get_pdn(pdev); |
781a868f WY |
1332 | |
1333 | if (!pdev->is_physfn) | |
1334 | return; | |
1335 | ||
02639b0e WY |
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 | ||
781a868f WY |
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 */ | |
02639b0e | 1399 | pnv_ioda_release_vf_PE(pdev, num_vfs); |
781a868f WY |
1400 | |
1401 | if (phb->type == PNV_PHB_IODA2) { | |
02639b0e WY |
1402 | if (pdn->m64_per_iov == 1) |
1403 | pnv_pci_vf_resource_shift(pdev, -pdn->offset); | |
781a868f WY |
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; | |
02639b0e | 1425 | int64_t rc; |
781a868f WY |
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 | } | |
02639b0e WY |
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 | } | |
781a868f WY |
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 */ | |
02639b0e | 1537 | ret = pnv_pci_vf_assign_m64(pdev, num_vfs); |
781a868f WY |
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 | */ | |
02639b0e WY |
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 | } | |
781a868f WY |
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 | ||
a8b2f828 GS |
1567 | int pcibios_sriov_disable(struct pci_dev *pdev) |
1568 | { | |
781a868f WY |
1569 | pnv_pci_sriov_disable(pdev); |
1570 | ||
a8b2f828 GS |
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); | |
781a868f WY |
1580 | |
1581 | pnv_pci_sriov_enable(pdev, num_vfs); | |
a8b2f828 GS |
1582 | return 0; |
1583 | } | |
1584 | #endif /* CONFIG_PCI_IOV */ | |
1585 | ||
959c9bdd | 1586 | static void pnv_pci_ioda_dma_dev_setup(struct pnv_phb *phb, struct pci_dev *pdev) |
184cd4a3 | 1587 | { |
b72c1f65 | 1588 | struct pci_dn *pdn = pci_get_pdn(pdev); |
959c9bdd | 1589 | struct pnv_ioda_pe *pe; |
184cd4a3 | 1590 | |
959c9bdd GS |
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; | |
184cd4a3 | 1598 | |
959c9bdd | 1599 | pe = &phb->ioda.pe_array[pdn->pe_number]; |
cd15b048 | 1600 | WARN_ON(get_dma_ops(&pdev->dev) != &dma_iommu_ops); |
4617082e AK |
1601 | set_iommu_table_base(&pdev->dev, pe->tce32_table); |
1602 | /* | |
1603 | * Note: iommu_add_device() will fail here as | |
1604 | * for physical PE: the device is already added by now; | |
1605 | * for virtual PE: sysfs entries are not ready yet and | |
1606 | * tce_iommu_bus_notifier will add the device to a group later. | |
1607 | */ | |
184cd4a3 BH |
1608 | } |
1609 | ||
763d2d8d | 1610 | static int pnv_pci_ioda_dma_set_mask(struct pci_dev *pdev, u64 dma_mask) |
cd15b048 | 1611 | { |
763d2d8d DA |
1612 | struct pci_controller *hose = pci_bus_to_host(pdev->bus); |
1613 | struct pnv_phb *phb = hose->private_data; | |
cd15b048 BH |
1614 | struct pci_dn *pdn = pci_get_pdn(pdev); |
1615 | struct pnv_ioda_pe *pe; | |
1616 | uint64_t top; | |
1617 | bool bypass = false; | |
1618 | ||
1619 | if (WARN_ON(!pdn || pdn->pe_number == IODA_INVALID_PE)) | |
1620 | return -ENODEV;; | |
1621 | ||
1622 | pe = &phb->ioda.pe_array[pdn->pe_number]; | |
1623 | if (pe->tce_bypass_enabled) { | |
1624 | top = pe->tce_bypass_base + memblock_end_of_DRAM() - 1; | |
1625 | bypass = (dma_mask >= top); | |
1626 | } | |
1627 | ||
1628 | if (bypass) { | |
1629 | dev_info(&pdev->dev, "Using 64-bit DMA iommu bypass\n"); | |
1630 | set_dma_ops(&pdev->dev, &dma_direct_ops); | |
1631 | set_dma_offset(&pdev->dev, pe->tce_bypass_base); | |
1632 | } else { | |
1633 | dev_info(&pdev->dev, "Using 32-bit DMA via iommu\n"); | |
1634 | set_dma_ops(&pdev->dev, &dma_iommu_ops); | |
9e8d4a19 | 1635 | set_iommu_table_base(&pdev->dev, pe->tce32_table); |
cd15b048 | 1636 | } |
a32305bf | 1637 | *pdev->dev.dma_mask = dma_mask; |
cd15b048 BH |
1638 | return 0; |
1639 | } | |
1640 | ||
fe7e85c6 GS |
1641 | static u64 pnv_pci_ioda_dma_get_required_mask(struct pnv_phb *phb, |
1642 | struct pci_dev *pdev) | |
1643 | { | |
1644 | struct pci_dn *pdn = pci_get_pdn(pdev); | |
1645 | struct pnv_ioda_pe *pe; | |
1646 | u64 end, mask; | |
1647 | ||
1648 | if (WARN_ON(!pdn || pdn->pe_number == IODA_INVALID_PE)) | |
1649 | return 0; | |
1650 | ||
1651 | pe = &phb->ioda.pe_array[pdn->pe_number]; | |
1652 | if (!pe->tce_bypass_enabled) | |
1653 | return __dma_get_required_mask(&pdev->dev); | |
1654 | ||
1655 | ||
1656 | end = pe->tce_bypass_base + memblock_end_of_DRAM(); | |
1657 | mask = 1ULL << (fls64(end) - 1); | |
1658 | mask += mask - 1; | |
1659 | ||
1660 | return mask; | |
1661 | } | |
1662 | ||
dff4a39e | 1663 | static void pnv_ioda_setup_bus_dma(struct pnv_ioda_pe *pe, |
ea30e99e | 1664 | struct pci_bus *bus) |
74251fe2 BH |
1665 | { |
1666 | struct pci_dev *dev; | |
1667 | ||
1668 | list_for_each_entry(dev, &bus->devices, bus_list) { | |
4617082e AK |
1669 | set_iommu_table_base(&dev->dev, pe->tce32_table); |
1670 | iommu_add_device(&dev->dev); | |
dff4a39e | 1671 | |
74251fe2 | 1672 | if (dev->subordinate) |
ea30e99e | 1673 | pnv_ioda_setup_bus_dma(pe, dev->subordinate); |
74251fe2 BH |
1674 | } |
1675 | } | |
1676 | ||
8e0a1611 AK |
1677 | static void pnv_pci_ioda1_tce_invalidate(struct pnv_ioda_pe *pe, |
1678 | struct iommu_table *tbl, | |
3ad26e5c | 1679 | __be64 *startp, __be64 *endp, bool rm) |
4cce9550 | 1680 | { |
3ad26e5c BH |
1681 | __be64 __iomem *invalidate = rm ? |
1682 | (__be64 __iomem *)pe->tce_inval_reg_phys : | |
1683 | (__be64 __iomem *)tbl->it_index; | |
4cce9550 | 1684 | unsigned long start, end, inc; |
b0376c9b | 1685 | const unsigned shift = tbl->it_page_shift; |
4cce9550 GS |
1686 | |
1687 | start = __pa(startp); | |
1688 | end = __pa(endp); | |
1689 | ||
1690 | /* BML uses this case for p6/p7/galaxy2: Shift addr and put in node */ | |
1691 | if (tbl->it_busno) { | |
b0376c9b AK |
1692 | start <<= shift; |
1693 | end <<= shift; | |
1694 | inc = 128ull << shift; | |
4cce9550 GS |
1695 | start |= tbl->it_busno; |
1696 | end |= tbl->it_busno; | |
1697 | } else if (tbl->it_type & TCE_PCI_SWINV_PAIR) { | |
1698 | /* p7ioc-style invalidation, 2 TCEs per write */ | |
1699 | start |= (1ull << 63); | |
1700 | end |= (1ull << 63); | |
1701 | inc = 16; | |
1702 | } else { | |
1703 | /* Default (older HW) */ | |
1704 | inc = 128; | |
1705 | } | |
1706 | ||
1707 | end |= inc - 1; /* round up end to be different than start */ | |
1708 | ||
1709 | mb(); /* Ensure above stores are visible */ | |
1710 | while (start <= end) { | |
8e0a1611 | 1711 | if (rm) |
3ad26e5c | 1712 | __raw_rm_writeq(cpu_to_be64(start), invalidate); |
8e0a1611 | 1713 | else |
3ad26e5c | 1714 | __raw_writeq(cpu_to_be64(start), invalidate); |
4cce9550 GS |
1715 | start += inc; |
1716 | } | |
1717 | ||
1718 | /* | |
1719 | * The iommu layer will do another mb() for us on build() | |
1720 | * and we don't care on free() | |
1721 | */ | |
1722 | } | |
1723 | ||
1724 | static void pnv_pci_ioda2_tce_invalidate(struct pnv_ioda_pe *pe, | |
1725 | struct iommu_table *tbl, | |
3ad26e5c | 1726 | __be64 *startp, __be64 *endp, bool rm) |
4cce9550 GS |
1727 | { |
1728 | unsigned long start, end, inc; | |
3ad26e5c BH |
1729 | __be64 __iomem *invalidate = rm ? |
1730 | (__be64 __iomem *)pe->tce_inval_reg_phys : | |
1731 | (__be64 __iomem *)tbl->it_index; | |
b0376c9b | 1732 | const unsigned shift = tbl->it_page_shift; |
4cce9550 GS |
1733 | |
1734 | /* We'll invalidate DMA address in PE scope */ | |
b0376c9b | 1735 | start = 0x2ull << 60; |
4cce9550 GS |
1736 | start |= (pe->pe_number & 0xFF); |
1737 | end = start; | |
1738 | ||
1739 | /* Figure out the start, end and step */ | |
1740 | inc = tbl->it_offset + (((u64)startp - tbl->it_base) / sizeof(u64)); | |
b0376c9b | 1741 | start |= (inc << shift); |
4cce9550 | 1742 | inc = tbl->it_offset + (((u64)endp - tbl->it_base) / sizeof(u64)); |
b0376c9b AK |
1743 | end |= (inc << shift); |
1744 | inc = (0x1ull << shift); | |
4cce9550 GS |
1745 | mb(); |
1746 | ||
1747 | while (start <= end) { | |
8e0a1611 | 1748 | if (rm) |
3ad26e5c | 1749 | __raw_rm_writeq(cpu_to_be64(start), invalidate); |
8e0a1611 | 1750 | else |
3ad26e5c | 1751 | __raw_writeq(cpu_to_be64(start), invalidate); |
4cce9550 GS |
1752 | start += inc; |
1753 | } | |
1754 | } | |
1755 | ||
1756 | void pnv_pci_ioda_tce_invalidate(struct iommu_table *tbl, | |
3ad26e5c | 1757 | __be64 *startp, __be64 *endp, bool rm) |
4cce9550 | 1758 | { |
9e8d4a19 | 1759 | struct pnv_ioda_pe *pe = tbl->data; |
4cce9550 GS |
1760 | struct pnv_phb *phb = pe->phb; |
1761 | ||
1762 | if (phb->type == PNV_PHB_IODA1) | |
8e0a1611 | 1763 | pnv_pci_ioda1_tce_invalidate(pe, tbl, startp, endp, rm); |
4cce9550 | 1764 | else |
8e0a1611 | 1765 | pnv_pci_ioda2_tce_invalidate(pe, tbl, startp, endp, rm); |
4cce9550 GS |
1766 | } |
1767 | ||
cad5cef6 GKH |
1768 | static void pnv_pci_ioda_setup_dma_pe(struct pnv_phb *phb, |
1769 | struct pnv_ioda_pe *pe, unsigned int base, | |
1770 | unsigned int segs) | |
184cd4a3 BH |
1771 | { |
1772 | ||
1773 | struct page *tce_mem = NULL; | |
1774 | const __be64 *swinvp; | |
1775 | struct iommu_table *tbl; | |
1776 | unsigned int i; | |
1777 | int64_t rc; | |
1778 | void *addr; | |
1779 | ||
184cd4a3 BH |
1780 | /* XXX FIXME: Handle 64-bit only DMA devices */ |
1781 | /* XXX FIXME: Provide 64-bit DMA facilities & non-4K TCE tables etc.. */ | |
1782 | /* XXX FIXME: Allocate multi-level tables on PHB3 */ | |
1783 | ||
1784 | /* We shouldn't already have a 32-bit DMA associated */ | |
1785 | if (WARN_ON(pe->tce32_seg >= 0)) | |
1786 | return; | |
1787 | ||
c5773822 AK |
1788 | tbl = pe->tce32_table; |
1789 | iommu_register_group(tbl, phb->hose->global_number, pe->pe_number); | |
1790 | ||
184cd4a3 BH |
1791 | /* Grab a 32-bit TCE table */ |
1792 | pe->tce32_seg = base; | |
1793 | pe_info(pe, " Setting up 32-bit TCE table at %08x..%08x\n", | |
1794 | (base << 28), ((base + segs) << 28) - 1); | |
1795 | ||
1796 | /* XXX Currently, we allocate one big contiguous table for the | |
1797 | * TCEs. We only really need one chunk per 256M of TCE space | |
1798 | * (ie per segment) but that's an optimization for later, it | |
1799 | * requires some added smarts with our get/put_tce implementation | |
1800 | */ | |
1801 | tce_mem = alloc_pages_node(phb->hose->node, GFP_KERNEL, | |
1802 | get_order(TCE32_TABLE_SIZE * segs)); | |
1803 | if (!tce_mem) { | |
1804 | pe_err(pe, " Failed to allocate a 32-bit TCE memory\n"); | |
1805 | goto fail; | |
1806 | } | |
1807 | addr = page_address(tce_mem); | |
1808 | memset(addr, 0, TCE32_TABLE_SIZE * segs); | |
1809 | ||
1810 | /* Configure HW */ | |
1811 | for (i = 0; i < segs; i++) { | |
1812 | rc = opal_pci_map_pe_dma_window(phb->opal_id, | |
1813 | pe->pe_number, | |
1814 | base + i, 1, | |
1815 | __pa(addr) + TCE32_TABLE_SIZE * i, | |
1816 | TCE32_TABLE_SIZE, 0x1000); | |
1817 | if (rc) { | |
1818 | pe_err(pe, " Failed to configure 32-bit TCE table," | |
1819 | " err %ld\n", rc); | |
1820 | goto fail; | |
1821 | } | |
1822 | } | |
1823 | ||
1824 | /* Setup linux iommu table */ | |
184cd4a3 | 1825 | pnv_pci_setup_iommu_table(tbl, addr, TCE32_TABLE_SIZE * segs, |
8fa5d454 | 1826 | base << 28, IOMMU_PAGE_SHIFT_4K); |
184cd4a3 BH |
1827 | |
1828 | /* OPAL variant of P7IOC SW invalidated TCEs */ | |
1829 | swinvp = of_get_property(phb->hose->dn, "ibm,opal-tce-kill", NULL); | |
1830 | if (swinvp) { | |
1831 | /* We need a couple more fields -- an address and a data | |
1832 | * to or. Since the bus is only printed out on table free | |
1833 | * errors, and on the first pass the data will be a relative | |
1834 | * bus number, print that out instead. | |
1835 | */ | |
8e0a1611 AK |
1836 | pe->tce_inval_reg_phys = be64_to_cpup(swinvp); |
1837 | tbl->it_index = (unsigned long)ioremap(pe->tce_inval_reg_phys, | |
1838 | 8); | |
65fd766b GS |
1839 | tbl->it_type |= (TCE_PCI_SWINV_CREATE | |
1840 | TCE_PCI_SWINV_FREE | | |
1841 | TCE_PCI_SWINV_PAIR); | |
184cd4a3 BH |
1842 | } |
1843 | iommu_init_table(tbl, phb->hose->node); | |
1844 | ||
781a868f | 1845 | if (pe->flags & PNV_IODA_PE_DEV) { |
4617082e AK |
1846 | /* |
1847 | * Setting table base here only for carrying iommu_group | |
1848 | * further down to let iommu_add_device() do the job. | |
1849 | * pnv_pci_ioda_dma_dev_setup will override it later anyway. | |
1850 | */ | |
1851 | set_iommu_table_base(&pe->pdev->dev, tbl); | |
1852 | iommu_add_device(&pe->pdev->dev); | |
c5773822 | 1853 | } else if (pe->flags & (PNV_IODA_PE_BUS | PNV_IODA_PE_BUS_ALL)) |
ea30e99e | 1854 | pnv_ioda_setup_bus_dma(pe, pe->pbus); |
74251fe2 | 1855 | |
184cd4a3 BH |
1856 | return; |
1857 | fail: | |
1858 | /* XXX Failure: Try to fallback to 64-bit only ? */ | |
1859 | if (pe->tce32_seg >= 0) | |
1860 | pe->tce32_seg = -1; | |
1861 | if (tce_mem) | |
1862 | __free_pages(tce_mem, get_order(TCE32_TABLE_SIZE * segs)); | |
1863 | } | |
1864 | ||
cd15b048 BH |
1865 | static void pnv_pci_ioda2_set_bypass(struct iommu_table *tbl, bool enable) |
1866 | { | |
9e8d4a19 | 1867 | struct pnv_ioda_pe *pe = tbl->data; |
cd15b048 BH |
1868 | uint16_t window_id = (pe->pe_number << 1 ) + 1; |
1869 | int64_t rc; | |
1870 | ||
1871 | pe_info(pe, "%sabling 64-bit DMA bypass\n", enable ? "En" : "Dis"); | |
1872 | if (enable) { | |
1873 | phys_addr_t top = memblock_end_of_DRAM(); | |
1874 | ||
1875 | top = roundup_pow_of_two(top); | |
1876 | rc = opal_pci_map_pe_dma_window_real(pe->phb->opal_id, | |
1877 | pe->pe_number, | |
1878 | window_id, | |
1879 | pe->tce_bypass_base, | |
1880 | top); | |
1881 | } else { | |
1882 | rc = opal_pci_map_pe_dma_window_real(pe->phb->opal_id, | |
1883 | pe->pe_number, | |
1884 | window_id, | |
1885 | pe->tce_bypass_base, | |
1886 | 0); | |
cd15b048 BH |
1887 | } |
1888 | if (rc) | |
1889 | pe_err(pe, "OPAL error %lld configuring bypass window\n", rc); | |
1890 | else | |
1891 | pe->tce_bypass_enabled = enable; | |
1892 | } | |
1893 | ||
1894 | static void pnv_pci_ioda2_setup_bypass_pe(struct pnv_phb *phb, | |
1895 | struct pnv_ioda_pe *pe) | |
1896 | { | |
1897 | /* TVE #1 is selected by PCI address bit 59 */ | |
1898 | pe->tce_bypass_base = 1ull << 59; | |
1899 | ||
1900 | /* Install set_bypass callback for VFIO */ | |
9e8d4a19 | 1901 | pe->tce32_table->set_bypass = pnv_pci_ioda2_set_bypass; |
cd15b048 BH |
1902 | |
1903 | /* Enable bypass by default */ | |
9e8d4a19 | 1904 | pnv_pci_ioda2_set_bypass(pe->tce32_table, true); |
cd15b048 BH |
1905 | } |
1906 | ||
373f5657 GS |
1907 | static void pnv_pci_ioda2_setup_dma_pe(struct pnv_phb *phb, |
1908 | struct pnv_ioda_pe *pe) | |
1909 | { | |
1910 | struct page *tce_mem = NULL; | |
1911 | void *addr; | |
1912 | const __be64 *swinvp; | |
1913 | struct iommu_table *tbl; | |
1914 | unsigned int tce_table_size, end; | |
1915 | int64_t rc; | |
1916 | ||
1917 | /* We shouldn't already have a 32-bit DMA associated */ | |
1918 | if (WARN_ON(pe->tce32_seg >= 0)) | |
1919 | return; | |
1920 | ||
c5773822 AK |
1921 | tbl = pe->tce32_table; |
1922 | iommu_register_group(tbl, phb->hose->global_number, pe->pe_number); | |
1923 | ||
373f5657 GS |
1924 | /* The PE will reserve all possible 32-bits space */ |
1925 | pe->tce32_seg = 0; | |
1926 | end = (1 << ilog2(phb->ioda.m32_pci_base)); | |
1927 | tce_table_size = (end / 0x1000) * 8; | |
1928 | pe_info(pe, "Setting up 32-bit TCE table at 0..%08x\n", | |
1929 | end); | |
1930 | ||
1931 | /* Allocate TCE table */ | |
1932 | tce_mem = alloc_pages_node(phb->hose->node, GFP_KERNEL, | |
1933 | get_order(tce_table_size)); | |
1934 | if (!tce_mem) { | |
1935 | pe_err(pe, "Failed to allocate a 32-bit TCE memory\n"); | |
1936 | goto fail; | |
1937 | } | |
1938 | addr = page_address(tce_mem); | |
1939 | memset(addr, 0, tce_table_size); | |
1940 | ||
1941 | /* | |
1942 | * Map TCE table through TVT. The TVE index is the PE number | |
1943 | * shifted by 1 bit for 32-bits DMA space. | |
1944 | */ | |
1945 | rc = opal_pci_map_pe_dma_window(phb->opal_id, pe->pe_number, | |
1946 | pe->pe_number << 1, 1, __pa(addr), | |
1947 | tce_table_size, 0x1000); | |
1948 | if (rc) { | |
1949 | pe_err(pe, "Failed to configure 32-bit TCE table," | |
1950 | " err %ld\n", rc); | |
1951 | goto fail; | |
1952 | } | |
1953 | ||
1954 | /* Setup linux iommu table */ | |
8fa5d454 AK |
1955 | pnv_pci_setup_iommu_table(tbl, addr, tce_table_size, 0, |
1956 | IOMMU_PAGE_SHIFT_4K); | |
373f5657 GS |
1957 | |
1958 | /* OPAL variant of PHB3 invalidated TCEs */ | |
1959 | swinvp = of_get_property(phb->hose->dn, "ibm,opal-tce-kill", NULL); | |
1960 | if (swinvp) { | |
1961 | /* We need a couple more fields -- an address and a data | |
1962 | * to or. Since the bus is only printed out on table free | |
1963 | * errors, and on the first pass the data will be a relative | |
1964 | * bus number, print that out instead. | |
1965 | */ | |
8e0a1611 AK |
1966 | pe->tce_inval_reg_phys = be64_to_cpup(swinvp); |
1967 | tbl->it_index = (unsigned long)ioremap(pe->tce_inval_reg_phys, | |
1968 | 8); | |
65fd766b | 1969 | tbl->it_type |= (TCE_PCI_SWINV_CREATE | TCE_PCI_SWINV_FREE); |
373f5657 GS |
1970 | } |
1971 | iommu_init_table(tbl, phb->hose->node); | |
1972 | ||
781a868f | 1973 | if (pe->flags & PNV_IODA_PE_DEV) { |
4617082e AK |
1974 | /* |
1975 | * Setting table base here only for carrying iommu_group | |
1976 | * further down to let iommu_add_device() do the job. | |
1977 | * pnv_pci_ioda_dma_dev_setup will override it later anyway. | |
1978 | */ | |
1979 | set_iommu_table_base(&pe->pdev->dev, tbl); | |
1980 | iommu_add_device(&pe->pdev->dev); | |
c5773822 | 1981 | } else if (pe->flags & (PNV_IODA_PE_BUS | PNV_IODA_PE_BUS_ALL)) |
ea30e99e | 1982 | pnv_ioda_setup_bus_dma(pe, pe->pbus); |
74251fe2 | 1983 | |
cd15b048 | 1984 | /* Also create a bypass window */ |
4e287840 TLSC |
1985 | if (!pnv_iommu_bypass_disabled) |
1986 | pnv_pci_ioda2_setup_bypass_pe(phb, pe); | |
1987 | ||
373f5657 GS |
1988 | return; |
1989 | fail: | |
1990 | if (pe->tce32_seg >= 0) | |
1991 | pe->tce32_seg = -1; | |
1992 | if (tce_mem) | |
1993 | __free_pages(tce_mem, get_order(tce_table_size)); | |
1994 | } | |
1995 | ||
cad5cef6 | 1996 | static void pnv_ioda_setup_dma(struct pnv_phb *phb) |
184cd4a3 BH |
1997 | { |
1998 | struct pci_controller *hose = phb->hose; | |
1999 | unsigned int residual, remaining, segs, tw, base; | |
2000 | struct pnv_ioda_pe *pe; | |
2001 | ||
2002 | /* If we have more PE# than segments available, hand out one | |
2003 | * per PE until we run out and let the rest fail. If not, | |
2004 | * then we assign at least one segment per PE, plus more based | |
2005 | * on the amount of devices under that PE | |
2006 | */ | |
2007 | if (phb->ioda.dma_pe_count > phb->ioda.tce32_count) | |
2008 | residual = 0; | |
2009 | else | |
2010 | residual = phb->ioda.tce32_count - | |
2011 | phb->ioda.dma_pe_count; | |
2012 | ||
2013 | pr_info("PCI: Domain %04x has %ld available 32-bit DMA segments\n", | |
2014 | hose->global_number, phb->ioda.tce32_count); | |
2015 | pr_info("PCI: %d PE# for a total weight of %d\n", | |
2016 | phb->ioda.dma_pe_count, phb->ioda.dma_weight); | |
2017 | ||
2018 | /* Walk our PE list and configure their DMA segments, hand them | |
2019 | * out one base segment plus any residual segments based on | |
2020 | * weight | |
2021 | */ | |
2022 | remaining = phb->ioda.tce32_count; | |
2023 | tw = phb->ioda.dma_weight; | |
2024 | base = 0; | |
7ebdf956 | 2025 | list_for_each_entry(pe, &phb->ioda.pe_dma_list, dma_link) { |
184cd4a3 BH |
2026 | if (!pe->dma_weight) |
2027 | continue; | |
2028 | if (!remaining) { | |
2029 | pe_warn(pe, "No DMA32 resources available\n"); | |
2030 | continue; | |
2031 | } | |
2032 | segs = 1; | |
2033 | if (residual) { | |
2034 | segs += ((pe->dma_weight * residual) + (tw / 2)) / tw; | |
2035 | if (segs > remaining) | |
2036 | segs = remaining; | |
2037 | } | |
373f5657 GS |
2038 | |
2039 | /* | |
2040 | * For IODA2 compliant PHB3, we needn't care about the weight. | |
2041 | * The all available 32-bits DMA space will be assigned to | |
2042 | * the specific PE. | |
2043 | */ | |
2044 | if (phb->type == PNV_PHB_IODA1) { | |
2045 | pe_info(pe, "DMA weight %d, assigned %d DMA32 segments\n", | |
2046 | pe->dma_weight, segs); | |
2047 | pnv_pci_ioda_setup_dma_pe(phb, pe, base, segs); | |
2048 | } else { | |
2049 | pe_info(pe, "Assign DMA32 space\n"); | |
2050 | segs = 0; | |
2051 | pnv_pci_ioda2_setup_dma_pe(phb, pe); | |
2052 | } | |
2053 | ||
184cd4a3 BH |
2054 | remaining -= segs; |
2055 | base += segs; | |
2056 | } | |
2057 | } | |
2058 | ||
2059 | #ifdef CONFIG_PCI_MSI | |
137436c9 GS |
2060 | static void pnv_ioda2_msi_eoi(struct irq_data *d) |
2061 | { | |
2062 | unsigned int hw_irq = (unsigned int)irqd_to_hwirq(d); | |
2063 | struct irq_chip *chip = irq_data_get_irq_chip(d); | |
2064 | struct pnv_phb *phb = container_of(chip, struct pnv_phb, | |
2065 | ioda.irq_chip); | |
2066 | int64_t rc; | |
2067 | ||
2068 | rc = opal_pci_msi_eoi(phb->opal_id, hw_irq); | |
2069 | WARN_ON_ONCE(rc); | |
2070 | ||
2071 | icp_native_eoi(d); | |
2072 | } | |
2073 | ||
fd9a1c26 IM |
2074 | |
2075 | static void set_msi_irq_chip(struct pnv_phb *phb, unsigned int virq) | |
2076 | { | |
2077 | struct irq_data *idata; | |
2078 | struct irq_chip *ichip; | |
2079 | ||
2080 | if (phb->type != PNV_PHB_IODA2) | |
2081 | return; | |
2082 | ||
2083 | if (!phb->ioda.irq_chip_init) { | |
2084 | /* | |
2085 | * First time we setup an MSI IRQ, we need to setup the | |
2086 | * corresponding IRQ chip to route correctly. | |
2087 | */ | |
2088 | idata = irq_get_irq_data(virq); | |
2089 | ichip = irq_data_get_irq_chip(idata); | |
2090 | phb->ioda.irq_chip_init = 1; | |
2091 | phb->ioda.irq_chip = *ichip; | |
2092 | phb->ioda.irq_chip.irq_eoi = pnv_ioda2_msi_eoi; | |
2093 | } | |
2094 | irq_set_chip(virq, &phb->ioda.irq_chip); | |
2095 | } | |
2096 | ||
80c49c7e IM |
2097 | #ifdef CONFIG_CXL_BASE |
2098 | ||
6f963ec2 | 2099 | struct device_node *pnv_pci_get_phb_node(struct pci_dev *dev) |
80c49c7e IM |
2100 | { |
2101 | struct pci_controller *hose = pci_bus_to_host(dev->bus); | |
2102 | ||
6f963ec2 | 2103 | return of_node_get(hose->dn); |
80c49c7e | 2104 | } |
6f963ec2 | 2105 | EXPORT_SYMBOL(pnv_pci_get_phb_node); |
80c49c7e | 2106 | |
1212aa1c | 2107 | int pnv_phb_to_cxl_mode(struct pci_dev *dev, uint64_t mode) |
80c49c7e IM |
2108 | { |
2109 | struct pci_controller *hose = pci_bus_to_host(dev->bus); | |
2110 | struct pnv_phb *phb = hose->private_data; | |
2111 | struct pnv_ioda_pe *pe; | |
2112 | int rc; | |
2113 | ||
2114 | pe = pnv_ioda_get_pe(dev); | |
2115 | if (!pe) | |
2116 | return -ENODEV; | |
2117 | ||
2118 | pe_info(pe, "Switching PHB to CXL\n"); | |
2119 | ||
1212aa1c | 2120 | rc = opal_pci_set_phb_cxl_mode(phb->opal_id, mode, pe->pe_number); |
80c49c7e IM |
2121 | if (rc) |
2122 | dev_err(&dev->dev, "opal_pci_set_phb_cxl_mode failed: %i\n", rc); | |
2123 | ||
2124 | return rc; | |
2125 | } | |
1212aa1c | 2126 | EXPORT_SYMBOL(pnv_phb_to_cxl_mode); |
80c49c7e IM |
2127 | |
2128 | /* Find PHB for cxl dev and allocate MSI hwirqs? | |
2129 | * Returns the absolute hardware IRQ number | |
2130 | */ | |
2131 | int pnv_cxl_alloc_hwirqs(struct pci_dev *dev, int num) | |
2132 | { | |
2133 | struct pci_controller *hose = pci_bus_to_host(dev->bus); | |
2134 | struct pnv_phb *phb = hose->private_data; | |
2135 | int hwirq = msi_bitmap_alloc_hwirqs(&phb->msi_bmp, num); | |
2136 | ||
2137 | if (hwirq < 0) { | |
2138 | dev_warn(&dev->dev, "Failed to find a free MSI\n"); | |
2139 | return -ENOSPC; | |
2140 | } | |
2141 | ||
2142 | return phb->msi_base + hwirq; | |
2143 | } | |
2144 | EXPORT_SYMBOL(pnv_cxl_alloc_hwirqs); | |
2145 | ||
2146 | void pnv_cxl_release_hwirqs(struct pci_dev *dev, int hwirq, int num) | |
2147 | { | |
2148 | struct pci_controller *hose = pci_bus_to_host(dev->bus); | |
2149 | struct pnv_phb *phb = hose->private_data; | |
2150 | ||
2151 | msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq - phb->msi_base, num); | |
2152 | } | |
2153 | EXPORT_SYMBOL(pnv_cxl_release_hwirqs); | |
2154 | ||
2155 | void pnv_cxl_release_hwirq_ranges(struct cxl_irq_ranges *irqs, | |
2156 | struct pci_dev *dev) | |
2157 | { | |
2158 | struct pci_controller *hose = pci_bus_to_host(dev->bus); | |
2159 | struct pnv_phb *phb = hose->private_data; | |
2160 | int i, hwirq; | |
2161 | ||
2162 | for (i = 1; i < CXL_IRQ_RANGES; i++) { | |
2163 | if (!irqs->range[i]) | |
2164 | continue; | |
2165 | pr_devel("cxl release irq range 0x%x: offset: 0x%lx limit: %ld\n", | |
2166 | i, irqs->offset[i], | |
2167 | irqs->range[i]); | |
2168 | hwirq = irqs->offset[i] - phb->msi_base; | |
2169 | msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq, | |
2170 | irqs->range[i]); | |
2171 | } | |
2172 | } | |
2173 | EXPORT_SYMBOL(pnv_cxl_release_hwirq_ranges); | |
2174 | ||
2175 | int pnv_cxl_alloc_hwirq_ranges(struct cxl_irq_ranges *irqs, | |
2176 | struct pci_dev *dev, int num) | |
2177 | { | |
2178 | struct pci_controller *hose = pci_bus_to_host(dev->bus); | |
2179 | struct pnv_phb *phb = hose->private_data; | |
2180 | int i, hwirq, try; | |
2181 | ||
2182 | memset(irqs, 0, sizeof(struct cxl_irq_ranges)); | |
2183 | ||
2184 | /* 0 is reserved for the multiplexed PSL DSI interrupt */ | |
2185 | for (i = 1; i < CXL_IRQ_RANGES && num; i++) { | |
2186 | try = num; | |
2187 | while (try) { | |
2188 | hwirq = msi_bitmap_alloc_hwirqs(&phb->msi_bmp, try); | |
2189 | if (hwirq >= 0) | |
2190 | break; | |
2191 | try /= 2; | |
2192 | } | |
2193 | if (!try) | |
2194 | goto fail; | |
2195 | ||
2196 | irqs->offset[i] = phb->msi_base + hwirq; | |
2197 | irqs->range[i] = try; | |
2198 | pr_devel("cxl alloc irq range 0x%x: offset: 0x%lx limit: %li\n", | |
2199 | i, irqs->offset[i], irqs->range[i]); | |
2200 | num -= try; | |
2201 | } | |
2202 | if (num) | |
2203 | goto fail; | |
2204 | ||
2205 | return 0; | |
2206 | fail: | |
2207 | pnv_cxl_release_hwirq_ranges(irqs, dev); | |
2208 | return -ENOSPC; | |
2209 | } | |
2210 | EXPORT_SYMBOL(pnv_cxl_alloc_hwirq_ranges); | |
2211 | ||
2212 | int pnv_cxl_get_irq_count(struct pci_dev *dev) | |
2213 | { | |
2214 | struct pci_controller *hose = pci_bus_to_host(dev->bus); | |
2215 | struct pnv_phb *phb = hose->private_data; | |
2216 | ||
2217 | return phb->msi_bmp.irq_count; | |
2218 | } | |
2219 | EXPORT_SYMBOL(pnv_cxl_get_irq_count); | |
2220 | ||
2221 | int pnv_cxl_ioda_msi_setup(struct pci_dev *dev, unsigned int hwirq, | |
2222 | unsigned int virq) | |
2223 | { | |
2224 | struct pci_controller *hose = pci_bus_to_host(dev->bus); | |
2225 | struct pnv_phb *phb = hose->private_data; | |
2226 | unsigned int xive_num = hwirq - phb->msi_base; | |
2227 | struct pnv_ioda_pe *pe; | |
2228 | int rc; | |
2229 | ||
2230 | if (!(pe = pnv_ioda_get_pe(dev))) | |
2231 | return -ENODEV; | |
2232 | ||
2233 | /* Assign XIVE to PE */ | |
2234 | rc = opal_pci_set_xive_pe(phb->opal_id, pe->pe_number, xive_num); | |
2235 | if (rc) { | |
2236 | pe_warn(pe, "%s: OPAL error %d setting msi_base 0x%x " | |
2237 | "hwirq 0x%x XIVE 0x%x PE\n", | |
2238 | pci_name(dev), rc, phb->msi_base, hwirq, xive_num); | |
2239 | return -EIO; | |
2240 | } | |
2241 | set_msi_irq_chip(phb, virq); | |
2242 | ||
2243 | return 0; | |
2244 | } | |
2245 | EXPORT_SYMBOL(pnv_cxl_ioda_msi_setup); | |
2246 | #endif | |
2247 | ||
184cd4a3 | 2248 | static int pnv_pci_ioda_msi_setup(struct pnv_phb *phb, struct pci_dev *dev, |
137436c9 GS |
2249 | unsigned int hwirq, unsigned int virq, |
2250 | unsigned int is_64, struct msi_msg *msg) | |
184cd4a3 BH |
2251 | { |
2252 | struct pnv_ioda_pe *pe = pnv_ioda_get_pe(dev); | |
2253 | unsigned int xive_num = hwirq - phb->msi_base; | |
3a1a4661 | 2254 | __be32 data; |
184cd4a3 BH |
2255 | int rc; |
2256 | ||
2257 | /* No PE assigned ? bail out ... no MSI for you ! */ | |
2258 | if (pe == NULL) | |
2259 | return -ENXIO; | |
2260 | ||
2261 | /* Check if we have an MVE */ | |
2262 | if (pe->mve_number < 0) | |
2263 | return -ENXIO; | |
2264 | ||
b72c1f65 | 2265 | /* Force 32-bit MSI on some broken devices */ |
36074381 | 2266 | if (dev->no_64bit_msi) |
b72c1f65 BH |
2267 | is_64 = 0; |
2268 | ||
184cd4a3 BH |
2269 | /* Assign XIVE to PE */ |
2270 | rc = opal_pci_set_xive_pe(phb->opal_id, pe->pe_number, xive_num); | |
2271 | if (rc) { | |
2272 | pr_warn("%s: OPAL error %d setting XIVE %d PE\n", | |
2273 | pci_name(dev), rc, xive_num); | |
2274 | return -EIO; | |
2275 | } | |
2276 | ||
2277 | if (is_64) { | |
3a1a4661 BH |
2278 | __be64 addr64; |
2279 | ||
184cd4a3 BH |
2280 | rc = opal_get_msi_64(phb->opal_id, pe->mve_number, xive_num, 1, |
2281 | &addr64, &data); | |
2282 | if (rc) { | |
2283 | pr_warn("%s: OPAL error %d getting 64-bit MSI data\n", | |
2284 | pci_name(dev), rc); | |
2285 | return -EIO; | |
2286 | } | |
3a1a4661 BH |
2287 | msg->address_hi = be64_to_cpu(addr64) >> 32; |
2288 | msg->address_lo = be64_to_cpu(addr64) & 0xfffffffful; | |
184cd4a3 | 2289 | } else { |
3a1a4661 BH |
2290 | __be32 addr32; |
2291 | ||
184cd4a3 BH |
2292 | rc = opal_get_msi_32(phb->opal_id, pe->mve_number, xive_num, 1, |
2293 | &addr32, &data); | |
2294 | if (rc) { | |
2295 | pr_warn("%s: OPAL error %d getting 32-bit MSI data\n", | |
2296 | pci_name(dev), rc); | |
2297 | return -EIO; | |
2298 | } | |
2299 | msg->address_hi = 0; | |
3a1a4661 | 2300 | msg->address_lo = be32_to_cpu(addr32); |
184cd4a3 | 2301 | } |
3a1a4661 | 2302 | msg->data = be32_to_cpu(data); |
184cd4a3 | 2303 | |
fd9a1c26 | 2304 | set_msi_irq_chip(phb, virq); |
137436c9 | 2305 | |
184cd4a3 BH |
2306 | pr_devel("%s: %s-bit MSI on hwirq %x (xive #%d)," |
2307 | " address=%x_%08x data=%x PE# %d\n", | |
2308 | pci_name(dev), is_64 ? "64" : "32", hwirq, xive_num, | |
2309 | msg->address_hi, msg->address_lo, data, pe->pe_number); | |
2310 | ||
2311 | return 0; | |
2312 | } | |
2313 | ||
2314 | static void pnv_pci_init_ioda_msis(struct pnv_phb *phb) | |
2315 | { | |
fb1b55d6 | 2316 | unsigned int count; |
184cd4a3 BH |
2317 | const __be32 *prop = of_get_property(phb->hose->dn, |
2318 | "ibm,opal-msi-ranges", NULL); | |
2319 | if (!prop) { | |
2320 | /* BML Fallback */ | |
2321 | prop = of_get_property(phb->hose->dn, "msi-ranges", NULL); | |
2322 | } | |
2323 | if (!prop) | |
2324 | return; | |
2325 | ||
2326 | phb->msi_base = be32_to_cpup(prop); | |
fb1b55d6 GS |
2327 | count = be32_to_cpup(prop + 1); |
2328 | if (msi_bitmap_alloc(&phb->msi_bmp, count, phb->hose->dn)) { | |
184cd4a3 BH |
2329 | pr_err("PCI %d: Failed to allocate MSI bitmap !\n", |
2330 | phb->hose->global_number); | |
2331 | return; | |
2332 | } | |
fb1b55d6 | 2333 | |
184cd4a3 BH |
2334 | phb->msi_setup = pnv_pci_ioda_msi_setup; |
2335 | phb->msi32_support = 1; | |
2336 | pr_info(" Allocated bitmap for %d MSIs (base IRQ 0x%x)\n", | |
fb1b55d6 | 2337 | count, phb->msi_base); |
184cd4a3 BH |
2338 | } |
2339 | #else | |
2340 | static void pnv_pci_init_ioda_msis(struct pnv_phb *phb) { } | |
2341 | #endif /* CONFIG_PCI_MSI */ | |
2342 | ||
6e628c7d WY |
2343 | #ifdef CONFIG_PCI_IOV |
2344 | static void pnv_pci_ioda_fixup_iov_resources(struct pci_dev *pdev) | |
2345 | { | |
2346 | struct pci_controller *hose; | |
2347 | struct pnv_phb *phb; | |
2348 | struct resource *res; | |
2349 | int i; | |
2350 | resource_size_t size; | |
2351 | struct pci_dn *pdn; | |
5b88ec22 | 2352 | int mul, total_vfs; |
6e628c7d WY |
2353 | |
2354 | if (!pdev->is_physfn || pdev->is_added) | |
2355 | return; | |
2356 | ||
2357 | hose = pci_bus_to_host(pdev->bus); | |
2358 | phb = hose->private_data; | |
2359 | ||
2360 | pdn = pci_get_pdn(pdev); | |
2361 | pdn->vfs_expanded = 0; | |
2362 | ||
5b88ec22 WY |
2363 | total_vfs = pci_sriov_get_totalvfs(pdev); |
2364 | pdn->m64_per_iov = 1; | |
2365 | mul = phb->ioda.total_pe; | |
2366 | ||
2367 | for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) { | |
2368 | res = &pdev->resource[i + PCI_IOV_RESOURCES]; | |
2369 | if (!res->flags || res->parent) | |
2370 | continue; | |
2371 | if (!pnv_pci_is_mem_pref_64(res->flags)) { | |
2372 | dev_warn(&pdev->dev, " non M64 VF BAR%d: %pR\n", | |
2373 | i, res); | |
2374 | continue; | |
2375 | } | |
2376 | ||
2377 | size = pci_iov_resource_size(pdev, i + PCI_IOV_RESOURCES); | |
2378 | ||
2379 | /* bigger than 64M */ | |
2380 | if (size > (1 << 26)) { | |
2381 | dev_info(&pdev->dev, "PowerNV: VF BAR%d: %pR IOV size is bigger than 64M, roundup power2\n", | |
2382 | i, res); | |
2383 | pdn->m64_per_iov = M64_PER_IOV; | |
2384 | mul = roundup_pow_of_two(total_vfs); | |
2385 | break; | |
2386 | } | |
2387 | } | |
2388 | ||
6e628c7d WY |
2389 | for (i = 0; i < PCI_SRIOV_NUM_BARS; i++) { |
2390 | res = &pdev->resource[i + PCI_IOV_RESOURCES]; | |
2391 | if (!res->flags || res->parent) | |
2392 | continue; | |
2393 | if (!pnv_pci_is_mem_pref_64(res->flags)) { | |
2394 | dev_warn(&pdev->dev, "Skipping expanding VF BAR%d: %pR\n", | |
2395 | i, res); | |
2396 | continue; | |
2397 | } | |
2398 | ||
2399 | dev_dbg(&pdev->dev, " Fixing VF BAR%d: %pR to\n", i, res); | |
2400 | size = pci_iov_resource_size(pdev, i + PCI_IOV_RESOURCES); | |
5b88ec22 | 2401 | res->end = res->start + size * mul - 1; |
6e628c7d WY |
2402 | dev_dbg(&pdev->dev, " %pR\n", res); |
2403 | dev_info(&pdev->dev, "VF BAR%d: %pR (expanded to %d VFs for PE alignment)", | |
5b88ec22 | 2404 | i, res, mul); |
6e628c7d | 2405 | } |
5b88ec22 | 2406 | pdn->vfs_expanded = mul; |
6e628c7d WY |
2407 | } |
2408 | #endif /* CONFIG_PCI_IOV */ | |
2409 | ||
11685bec GS |
2410 | /* |
2411 | * This function is supposed to be called on basis of PE from top | |
2412 | * to bottom style. So the the I/O or MMIO segment assigned to | |
2413 | * parent PE could be overrided by its child PEs if necessary. | |
2414 | */ | |
cad5cef6 GKH |
2415 | static void pnv_ioda_setup_pe_seg(struct pci_controller *hose, |
2416 | struct pnv_ioda_pe *pe) | |
11685bec GS |
2417 | { |
2418 | struct pnv_phb *phb = hose->private_data; | |
2419 | struct pci_bus_region region; | |
2420 | struct resource *res; | |
2421 | int i, index; | |
2422 | int rc; | |
2423 | ||
2424 | /* | |
2425 | * NOTE: We only care PCI bus based PE for now. For PCI | |
2426 | * device based PE, for example SRIOV sensitive VF should | |
2427 | * be figured out later. | |
2428 | */ | |
2429 | BUG_ON(!(pe->flags & (PNV_IODA_PE_BUS | PNV_IODA_PE_BUS_ALL))); | |
2430 | ||
2431 | pci_bus_for_each_resource(pe->pbus, res, i) { | |
2432 | if (!res || !res->flags || | |
2433 | res->start > res->end) | |
2434 | continue; | |
2435 | ||
2436 | if (res->flags & IORESOURCE_IO) { | |
2437 | region.start = res->start - phb->ioda.io_pci_base; | |
2438 | region.end = res->end - phb->ioda.io_pci_base; | |
2439 | index = region.start / phb->ioda.io_segsize; | |
2440 | ||
2441 | while (index < phb->ioda.total_pe && | |
2442 | region.start <= region.end) { | |
2443 | phb->ioda.io_segmap[index] = pe->pe_number; | |
2444 | rc = opal_pci_map_pe_mmio_window(phb->opal_id, | |
2445 | pe->pe_number, OPAL_IO_WINDOW_TYPE, 0, index); | |
2446 | if (rc != OPAL_SUCCESS) { | |
2447 | pr_err("%s: OPAL error %d when mapping IO " | |
2448 | "segment #%d to PE#%d\n", | |
2449 | __func__, rc, index, pe->pe_number); | |
2450 | break; | |
2451 | } | |
2452 | ||
2453 | region.start += phb->ioda.io_segsize; | |
2454 | index++; | |
2455 | } | |
027fa02f GS |
2456 | } else if ((res->flags & IORESOURCE_MEM) && |
2457 | !pnv_pci_is_mem_pref_64(res->flags)) { | |
11685bec | 2458 | region.start = res->start - |
3fd47f06 | 2459 | hose->mem_offset[0] - |
11685bec GS |
2460 | phb->ioda.m32_pci_base; |
2461 | region.end = res->end - | |
3fd47f06 | 2462 | hose->mem_offset[0] - |
11685bec GS |
2463 | phb->ioda.m32_pci_base; |
2464 | index = region.start / phb->ioda.m32_segsize; | |
2465 | ||
2466 | while (index < phb->ioda.total_pe && | |
2467 | region.start <= region.end) { | |
2468 | phb->ioda.m32_segmap[index] = pe->pe_number; | |
2469 | rc = opal_pci_map_pe_mmio_window(phb->opal_id, | |
2470 | pe->pe_number, OPAL_M32_WINDOW_TYPE, 0, index); | |
2471 | if (rc != OPAL_SUCCESS) { | |
2472 | pr_err("%s: OPAL error %d when mapping M32 " | |
2473 | "segment#%d to PE#%d", | |
2474 | __func__, rc, index, pe->pe_number); | |
2475 | break; | |
2476 | } | |
2477 | ||
2478 | region.start += phb->ioda.m32_segsize; | |
2479 | index++; | |
2480 | } | |
2481 | } | |
2482 | } | |
2483 | } | |
2484 | ||
cad5cef6 | 2485 | static void pnv_pci_ioda_setup_seg(void) |
11685bec GS |
2486 | { |
2487 | struct pci_controller *tmp, *hose; | |
2488 | struct pnv_phb *phb; | |
2489 | struct pnv_ioda_pe *pe; | |
2490 | ||
2491 | list_for_each_entry_safe(hose, tmp, &hose_list, list_node) { | |
2492 | phb = hose->private_data; | |
2493 | list_for_each_entry(pe, &phb->ioda.pe_list, list) { | |
2494 | pnv_ioda_setup_pe_seg(hose, pe); | |
2495 | } | |
2496 | } | |
2497 | } | |
2498 | ||
cad5cef6 | 2499 | static void pnv_pci_ioda_setup_DMA(void) |
13395c48 GS |
2500 | { |
2501 | struct pci_controller *hose, *tmp; | |
db1266c8 | 2502 | struct pnv_phb *phb; |
13395c48 GS |
2503 | |
2504 | list_for_each_entry_safe(hose, tmp, &hose_list, list_node) { | |
2505 | pnv_ioda_setup_dma(hose->private_data); | |
db1266c8 GS |
2506 | |
2507 | /* Mark the PHB initialization done */ | |
2508 | phb = hose->private_data; | |
2509 | phb->initialized = 1; | |
13395c48 GS |
2510 | } |
2511 | } | |
2512 | ||
37c367f2 GS |
2513 | static void pnv_pci_ioda_create_dbgfs(void) |
2514 | { | |
2515 | #ifdef CONFIG_DEBUG_FS | |
2516 | struct pci_controller *hose, *tmp; | |
2517 | struct pnv_phb *phb; | |
2518 | char name[16]; | |
2519 | ||
2520 | list_for_each_entry_safe(hose, tmp, &hose_list, list_node) { | |
2521 | phb = hose->private_data; | |
2522 | ||
2523 | sprintf(name, "PCI%04x", hose->global_number); | |
2524 | phb->dbgfs = debugfs_create_dir(name, powerpc_debugfs_root); | |
2525 | if (!phb->dbgfs) | |
2526 | pr_warning("%s: Error on creating debugfs on PHB#%x\n", | |
2527 | __func__, hose->global_number); | |
2528 | } | |
2529 | #endif /* CONFIG_DEBUG_FS */ | |
2530 | } | |
2531 | ||
cad5cef6 | 2532 | static void pnv_pci_ioda_fixup(void) |
fb446ad0 GS |
2533 | { |
2534 | pnv_pci_ioda_setup_PEs(); | |
11685bec | 2535 | pnv_pci_ioda_setup_seg(); |
13395c48 | 2536 | pnv_pci_ioda_setup_DMA(); |
e9cc17d4 | 2537 | |
37c367f2 GS |
2538 | pnv_pci_ioda_create_dbgfs(); |
2539 | ||
e9cc17d4 | 2540 | #ifdef CONFIG_EEH |
e9cc17d4 | 2541 | eeh_init(); |
dadcd6d6 | 2542 | eeh_addr_cache_build(); |
e9cc17d4 | 2543 | #endif |
fb446ad0 GS |
2544 | } |
2545 | ||
271fd03a GS |
2546 | /* |
2547 | * Returns the alignment for I/O or memory windows for P2P | |
2548 | * bridges. That actually depends on how PEs are segmented. | |
2549 | * For now, we return I/O or M32 segment size for PE sensitive | |
2550 | * P2P bridges. Otherwise, the default values (4KiB for I/O, | |
2551 | * 1MiB for memory) will be returned. | |
2552 | * | |
2553 | * The current PCI bus might be put into one PE, which was | |
2554 | * create against the parent PCI bridge. For that case, we | |
2555 | * needn't enlarge the alignment so that we can save some | |
2556 | * resources. | |
2557 | */ | |
2558 | static resource_size_t pnv_pci_window_alignment(struct pci_bus *bus, | |
2559 | unsigned long type) | |
2560 | { | |
2561 | struct pci_dev *bridge; | |
2562 | struct pci_controller *hose = pci_bus_to_host(bus); | |
2563 | struct pnv_phb *phb = hose->private_data; | |
2564 | int num_pci_bridges = 0; | |
2565 | ||
2566 | bridge = bus->self; | |
2567 | while (bridge) { | |
2568 | if (pci_pcie_type(bridge) == PCI_EXP_TYPE_PCI_BRIDGE) { | |
2569 | num_pci_bridges++; | |
2570 | if (num_pci_bridges >= 2) | |
2571 | return 1; | |
2572 | } | |
2573 | ||
2574 | bridge = bridge->bus->self; | |
2575 | } | |
2576 | ||
262af557 GC |
2577 | /* We fail back to M32 if M64 isn't supported */ |
2578 | if (phb->ioda.m64_segsize && | |
2579 | pnv_pci_is_mem_pref_64(type)) | |
2580 | return phb->ioda.m64_segsize; | |
271fd03a GS |
2581 | if (type & IORESOURCE_MEM) |
2582 | return phb->ioda.m32_segsize; | |
2583 | ||
2584 | return phb->ioda.io_segsize; | |
2585 | } | |
2586 | ||
5350ab3f WY |
2587 | #ifdef CONFIG_PCI_IOV |
2588 | static resource_size_t pnv_pci_iov_resource_alignment(struct pci_dev *pdev, | |
2589 | int resno) | |
2590 | { | |
2591 | struct pci_dn *pdn = pci_get_pdn(pdev); | |
2592 | resource_size_t align, iov_align; | |
2593 | ||
2594 | iov_align = resource_size(&pdev->resource[resno]); | |
2595 | if (iov_align) | |
2596 | return iov_align; | |
2597 | ||
2598 | align = pci_iov_resource_size(pdev, resno); | |
2599 | if (pdn->vfs_expanded) | |
2600 | return pdn->vfs_expanded * align; | |
2601 | ||
2602 | return align; | |
2603 | } | |
2604 | #endif /* CONFIG_PCI_IOV */ | |
2605 | ||
184cd4a3 BH |
2606 | /* Prevent enabling devices for which we couldn't properly |
2607 | * assign a PE | |
2608 | */ | |
c88c2a18 | 2609 | static bool pnv_pci_enable_device_hook(struct pci_dev *dev) |
184cd4a3 | 2610 | { |
db1266c8 GS |
2611 | struct pci_controller *hose = pci_bus_to_host(dev->bus); |
2612 | struct pnv_phb *phb = hose->private_data; | |
2613 | struct pci_dn *pdn; | |
184cd4a3 | 2614 | |
db1266c8 GS |
2615 | /* The function is probably called while the PEs have |
2616 | * not be created yet. For example, resource reassignment | |
2617 | * during PCI probe period. We just skip the check if | |
2618 | * PEs isn't ready. | |
2619 | */ | |
2620 | if (!phb->initialized) | |
c88c2a18 | 2621 | return true; |
db1266c8 | 2622 | |
b72c1f65 | 2623 | pdn = pci_get_pdn(dev); |
184cd4a3 | 2624 | if (!pdn || pdn->pe_number == IODA_INVALID_PE) |
c88c2a18 | 2625 | return false; |
db1266c8 | 2626 | |
c88c2a18 | 2627 | return true; |
184cd4a3 BH |
2628 | } |
2629 | ||
2630 | static u32 pnv_ioda_bdfn_to_pe(struct pnv_phb *phb, struct pci_bus *bus, | |
2631 | u32 devfn) | |
2632 | { | |
2633 | return phb->ioda.pe_rmap[(bus->number << 8) | devfn]; | |
2634 | } | |
2635 | ||
7a8e6bbf | 2636 | static void pnv_pci_ioda_shutdown(struct pci_controller *hose) |
73ed148a | 2637 | { |
7a8e6bbf MN |
2638 | struct pnv_phb *phb = hose->private_data; |
2639 | ||
d1a85eee | 2640 | opal_pci_reset(phb->opal_id, OPAL_RESET_PCI_IODA_TABLE, |
73ed148a BH |
2641 | OPAL_ASSERT_RESET); |
2642 | } | |
2643 | ||
92ae0353 DA |
2644 | static const struct pci_controller_ops pnv_pci_ioda_controller_ops = { |
2645 | .dma_dev_setup = pnv_pci_dma_dev_setup, | |
2646 | #ifdef CONFIG_PCI_MSI | |
2647 | .setup_msi_irqs = pnv_setup_msi_irqs, | |
2648 | .teardown_msi_irqs = pnv_teardown_msi_irqs, | |
2649 | #endif | |
2650 | .enable_device_hook = pnv_pci_enable_device_hook, | |
2651 | .window_alignment = pnv_pci_window_alignment, | |
2652 | .reset_secondary_bus = pnv_pci_reset_secondary_bus, | |
763d2d8d | 2653 | .dma_set_mask = pnv_pci_ioda_dma_set_mask, |
7a8e6bbf | 2654 | .shutdown = pnv_pci_ioda_shutdown, |
92ae0353 DA |
2655 | }; |
2656 | ||
e51df2c1 AB |
2657 | static void __init pnv_pci_init_ioda_phb(struct device_node *np, |
2658 | u64 hub_id, int ioda_type) | |
184cd4a3 BH |
2659 | { |
2660 | struct pci_controller *hose; | |
184cd4a3 | 2661 | struct pnv_phb *phb; |
8184616f | 2662 | unsigned long size, m32map_off, pemap_off, iomap_off = 0; |
c681b93c | 2663 | const __be64 *prop64; |
3a1a4661 | 2664 | const __be32 *prop32; |
f1b7cc3e | 2665 | int len; |
184cd4a3 BH |
2666 | u64 phb_id; |
2667 | void *aux; | |
2668 | long rc; | |
2669 | ||
58d714ec | 2670 | pr_info("Initializing IODA%d OPAL PHB %s\n", ioda_type, np->full_name); |
184cd4a3 BH |
2671 | |
2672 | prop64 = of_get_property(np, "ibm,opal-phbid", NULL); | |
2673 | if (!prop64) { | |
2674 | pr_err(" Missing \"ibm,opal-phbid\" property !\n"); | |
2675 | return; | |
2676 | } | |
2677 | phb_id = be64_to_cpup(prop64); | |
2678 | pr_debug(" PHB-ID : 0x%016llx\n", phb_id); | |
2679 | ||
e39f223f | 2680 | phb = memblock_virt_alloc(sizeof(struct pnv_phb), 0); |
58d714ec GS |
2681 | |
2682 | /* Allocate PCI controller */ | |
58d714ec GS |
2683 | phb->hose = hose = pcibios_alloc_controller(np); |
2684 | if (!phb->hose) { | |
2685 | pr_err(" Can't allocate PCI controller for %s\n", | |
184cd4a3 | 2686 | np->full_name); |
e39f223f | 2687 | memblock_free(__pa(phb), sizeof(struct pnv_phb)); |
184cd4a3 BH |
2688 | return; |
2689 | } | |
2690 | ||
2691 | spin_lock_init(&phb->lock); | |
f1b7cc3e GS |
2692 | prop32 = of_get_property(np, "bus-range", &len); |
2693 | if (prop32 && len == 8) { | |
3a1a4661 BH |
2694 | hose->first_busno = be32_to_cpu(prop32[0]); |
2695 | hose->last_busno = be32_to_cpu(prop32[1]); | |
f1b7cc3e GS |
2696 | } else { |
2697 | pr_warn(" Broken <bus-range> on %s\n", np->full_name); | |
2698 | hose->first_busno = 0; | |
2699 | hose->last_busno = 0xff; | |
2700 | } | |
184cd4a3 | 2701 | hose->private_data = phb; |
e9cc17d4 | 2702 | phb->hub_id = hub_id; |
184cd4a3 | 2703 | phb->opal_id = phb_id; |
aa0c033f | 2704 | phb->type = ioda_type; |
781a868f | 2705 | mutex_init(&phb->ioda.pe_alloc_mutex); |
184cd4a3 | 2706 | |
cee72d5b BH |
2707 | /* Detect specific models for error handling */ |
2708 | if (of_device_is_compatible(np, "ibm,p7ioc-pciex")) | |
2709 | phb->model = PNV_PHB_MODEL_P7IOC; | |
f3d40c25 | 2710 | else if (of_device_is_compatible(np, "ibm,power8-pciex")) |
aa0c033f | 2711 | phb->model = PNV_PHB_MODEL_PHB3; |
cee72d5b BH |
2712 | else |
2713 | phb->model = PNV_PHB_MODEL_UNKNOWN; | |
2714 | ||
aa0c033f | 2715 | /* Parse 32-bit and IO ranges (if any) */ |
2f1ec02e | 2716 | pci_process_bridge_OF_ranges(hose, np, !hose->global_number); |
184cd4a3 | 2717 | |
aa0c033f | 2718 | /* Get registers */ |
184cd4a3 BH |
2719 | phb->regs = of_iomap(np, 0); |
2720 | if (phb->regs == NULL) | |
2721 | pr_err(" Failed to map registers !\n"); | |
2722 | ||
184cd4a3 | 2723 | /* Initialize more IODA stuff */ |
36954dc7 | 2724 | phb->ioda.total_pe = 1; |
aa0c033f | 2725 | prop32 = of_get_property(np, "ibm,opal-num-pes", NULL); |
36954dc7 | 2726 | if (prop32) |
3a1a4661 | 2727 | phb->ioda.total_pe = be32_to_cpup(prop32); |
36954dc7 GS |
2728 | prop32 = of_get_property(np, "ibm,opal-reserved-pe", NULL); |
2729 | if (prop32) | |
2730 | phb->ioda.reserved_pe = be32_to_cpup(prop32); | |
262af557 GC |
2731 | |
2732 | /* Parse 64-bit MMIO range */ | |
2733 | pnv_ioda_parse_m64_window(phb); | |
2734 | ||
184cd4a3 | 2735 | phb->ioda.m32_size = resource_size(&hose->mem_resources[0]); |
aa0c033f | 2736 | /* FW Has already off top 64k of M32 space (MSI space) */ |
184cd4a3 BH |
2737 | phb->ioda.m32_size += 0x10000; |
2738 | ||
2739 | phb->ioda.m32_segsize = phb->ioda.m32_size / phb->ioda.total_pe; | |
3fd47f06 | 2740 | phb->ioda.m32_pci_base = hose->mem_resources[0].start - hose->mem_offset[0]; |
184cd4a3 BH |
2741 | phb->ioda.io_size = hose->pci_io_size; |
2742 | phb->ioda.io_segsize = phb->ioda.io_size / phb->ioda.total_pe; | |
2743 | phb->ioda.io_pci_base = 0; /* XXX calculate this ? */ | |
2744 | ||
c35d2a8c | 2745 | /* Allocate aux data & arrays. We don't have IO ports on PHB3 */ |
184cd4a3 BH |
2746 | size = _ALIGN_UP(phb->ioda.total_pe / 8, sizeof(unsigned long)); |
2747 | m32map_off = size; | |
e47747f4 | 2748 | size += phb->ioda.total_pe * sizeof(phb->ioda.m32_segmap[0]); |
c35d2a8c GS |
2749 | if (phb->type == PNV_PHB_IODA1) { |
2750 | iomap_off = size; | |
2751 | size += phb->ioda.total_pe * sizeof(phb->ioda.io_segmap[0]); | |
2752 | } | |
184cd4a3 BH |
2753 | pemap_off = size; |
2754 | size += phb->ioda.total_pe * sizeof(struct pnv_ioda_pe); | |
e39f223f | 2755 | aux = memblock_virt_alloc(size, 0); |
184cd4a3 BH |
2756 | phb->ioda.pe_alloc = aux; |
2757 | phb->ioda.m32_segmap = aux + m32map_off; | |
c35d2a8c GS |
2758 | if (phb->type == PNV_PHB_IODA1) |
2759 | phb->ioda.io_segmap = aux + iomap_off; | |
184cd4a3 | 2760 | phb->ioda.pe_array = aux + pemap_off; |
36954dc7 | 2761 | set_bit(phb->ioda.reserved_pe, phb->ioda.pe_alloc); |
184cd4a3 | 2762 | |
7ebdf956 | 2763 | INIT_LIST_HEAD(&phb->ioda.pe_dma_list); |
184cd4a3 | 2764 | INIT_LIST_HEAD(&phb->ioda.pe_list); |
781a868f | 2765 | mutex_init(&phb->ioda.pe_list_mutex); |
184cd4a3 BH |
2766 | |
2767 | /* Calculate how many 32-bit TCE segments we have */ | |
2768 | phb->ioda.tce32_count = phb->ioda.m32_pci_base >> 28; | |
2769 | ||
aa0c033f | 2770 | #if 0 /* We should really do that ... */ |
184cd4a3 BH |
2771 | rc = opal_pci_set_phb_mem_window(opal->phb_id, |
2772 | window_type, | |
2773 | window_num, | |
2774 | starting_real_address, | |
2775 | starting_pci_address, | |
2776 | segment_size); | |
2777 | #endif | |
2778 | ||
262af557 GC |
2779 | pr_info(" %03d (%03d) PE's M32: 0x%x [segment=0x%x]\n", |
2780 | phb->ioda.total_pe, phb->ioda.reserved_pe, | |
2781 | phb->ioda.m32_size, phb->ioda.m32_segsize); | |
2782 | if (phb->ioda.m64_size) | |
2783 | pr_info(" M64: 0x%lx [segment=0x%lx]\n", | |
2784 | phb->ioda.m64_size, phb->ioda.m64_segsize); | |
2785 | if (phb->ioda.io_size) | |
2786 | pr_info(" IO: 0x%x [segment=0x%x]\n", | |
2787 | phb->ioda.io_size, phb->ioda.io_segsize); | |
2788 | ||
184cd4a3 | 2789 | |
184cd4a3 | 2790 | phb->hose->ops = &pnv_pci_ops; |
49dec922 GS |
2791 | phb->get_pe_state = pnv_ioda_get_pe_state; |
2792 | phb->freeze_pe = pnv_ioda_freeze_pe; | |
2793 | phb->unfreeze_pe = pnv_ioda_unfreeze_pe; | |
184cd4a3 BH |
2794 | |
2795 | /* Setup RID -> PE mapping function */ | |
2796 | phb->bdfn_to_pe = pnv_ioda_bdfn_to_pe; | |
2797 | ||
2798 | /* Setup TCEs */ | |
2799 | phb->dma_dev_setup = pnv_pci_ioda_dma_dev_setup; | |
fe7e85c6 | 2800 | phb->dma_get_required_mask = pnv_pci_ioda_dma_get_required_mask; |
184cd4a3 BH |
2801 | |
2802 | /* Setup MSI support */ | |
2803 | pnv_pci_init_ioda_msis(phb); | |
2804 | ||
c40a4210 GS |
2805 | /* |
2806 | * We pass the PCI probe flag PCI_REASSIGN_ALL_RSRC here | |
2807 | * to let the PCI core do resource assignment. It's supposed | |
2808 | * that the PCI core will do correct I/O and MMIO alignment | |
2809 | * for the P2P bridge bars so that each PCI bus (excluding | |
2810 | * the child P2P bridges) can form individual PE. | |
184cd4a3 | 2811 | */ |
fb446ad0 | 2812 | ppc_md.pcibios_fixup = pnv_pci_ioda_fixup; |
92ae0353 | 2813 | hose->controller_ops = pnv_pci_ioda_controller_ops; |
ad30cb99 | 2814 | |
6e628c7d WY |
2815 | #ifdef CONFIG_PCI_IOV |
2816 | ppc_md.pcibios_fixup_sriov = pnv_pci_ioda_fixup_iov_resources; | |
5350ab3f | 2817 | ppc_md.pcibios_iov_resource_alignment = pnv_pci_iov_resource_alignment; |
ad30cb99 ME |
2818 | #endif |
2819 | ||
c40a4210 | 2820 | pci_add_flags(PCI_REASSIGN_ALL_RSRC); |
184cd4a3 BH |
2821 | |
2822 | /* Reset IODA tables to a clean state */ | |
d1a85eee | 2823 | rc = opal_pci_reset(phb_id, OPAL_RESET_PCI_IODA_TABLE, OPAL_ASSERT_RESET); |
184cd4a3 | 2824 | if (rc) |
f11fe552 | 2825 | pr_warning(" OPAL Error %ld performing IODA table reset !\n", rc); |
361f2a2a GS |
2826 | |
2827 | /* If we're running in kdump kerenl, the previous kerenl never | |
2828 | * shutdown PCI devices correctly. We already got IODA table | |
2829 | * cleaned out. So we have to issue PHB reset to stop all PCI | |
2830 | * transactions from previous kerenl. | |
2831 | */ | |
2832 | if (is_kdump_kernel()) { | |
2833 | pr_info(" Issue PHB reset ...\n"); | |
cadf364d GS |
2834 | pnv_eeh_phb_reset(hose, EEH_RESET_FUNDAMENTAL); |
2835 | pnv_eeh_phb_reset(hose, EEH_RESET_DEACTIVATE); | |
361f2a2a | 2836 | } |
262af557 | 2837 | |
9e9e8935 GS |
2838 | /* Remove M64 resource if we can't configure it successfully */ |
2839 | if (!phb->init_m64 || phb->init_m64(phb)) | |
262af557 | 2840 | hose->mem_resources[1].flags = 0; |
aa0c033f GS |
2841 | } |
2842 | ||
67975005 | 2843 | void __init pnv_pci_init_ioda2_phb(struct device_node *np) |
aa0c033f | 2844 | { |
e9cc17d4 | 2845 | pnv_pci_init_ioda_phb(np, 0, PNV_PHB_IODA2); |
184cd4a3 BH |
2846 | } |
2847 | ||
2848 | void __init pnv_pci_init_ioda_hub(struct device_node *np) | |
2849 | { | |
2850 | struct device_node *phbn; | |
c681b93c | 2851 | const __be64 *prop64; |
184cd4a3 BH |
2852 | u64 hub_id; |
2853 | ||
2854 | pr_info("Probing IODA IO-Hub %s\n", np->full_name); | |
2855 | ||
2856 | prop64 = of_get_property(np, "ibm,opal-hubid", NULL); | |
2857 | if (!prop64) { | |
2858 | pr_err(" Missing \"ibm,opal-hubid\" property !\n"); | |
2859 | return; | |
2860 | } | |
2861 | hub_id = be64_to_cpup(prop64); | |
2862 | pr_devel(" HUB-ID : 0x%016llx\n", hub_id); | |
2863 | ||
2864 | /* Count child PHBs */ | |
2865 | for_each_child_of_node(np, phbn) { | |
2866 | /* Look for IODA1 PHBs */ | |
2867 | if (of_device_is_compatible(phbn, "ibm,ioda-phb")) | |
e9cc17d4 | 2868 | pnv_pci_init_ioda_phb(phbn, hub_id, PNV_PHB_IODA1); |
184cd4a3 BH |
2869 | } |
2870 | } |