powerpc/powernv: Discover IODA3 PHBs
[linux-2.6-block.git] / arch / powerpc / platforms / powernv / pci.c
... / ...
CommitLineData
1/*
2 * Support PCI/PCIe on PowerNV platforms
3 *
4 * Copyright 2011 Benjamin Herrenschmidt, IBM Corp.
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 */
11
12#include <linux/kernel.h>
13#include <linux/pci.h>
14#include <linux/delay.h>
15#include <linux/string.h>
16#include <linux/init.h>
17#include <linux/irq.h>
18#include <linux/io.h>
19#include <linux/msi.h>
20#include <linux/iommu.h>
21
22#include <asm/sections.h>
23#include <asm/io.h>
24#include <asm/prom.h>
25#include <asm/pci-bridge.h>
26#include <asm/machdep.h>
27#include <asm/msi_bitmap.h>
28#include <asm/ppc-pci.h>
29#include <asm/pnv-pci.h>
30#include <asm/opal.h>
31#include <asm/iommu.h>
32#include <asm/tce.h>
33#include <asm/firmware.h>
34#include <asm/eeh_event.h>
35#include <asm/eeh.h>
36
37#include "powernv.h"
38#include "pci.h"
39
40int pnv_pci_get_slot_id(struct device_node *np, uint64_t *id)
41{
42 struct device_node *parent = np;
43 u32 bdfn;
44 u64 phbid;
45 int ret;
46
47 ret = of_property_read_u32(np, "reg", &bdfn);
48 if (ret)
49 return -ENXIO;
50
51 bdfn = ((bdfn & 0x00ffff00) >> 8);
52 while ((parent = of_get_parent(parent))) {
53 if (!PCI_DN(parent)) {
54 of_node_put(parent);
55 break;
56 }
57
58 if (!of_device_is_compatible(parent, "ibm,ioda2-phb")) {
59 of_node_put(parent);
60 continue;
61 }
62
63 ret = of_property_read_u64(parent, "ibm,opal-phbid", &phbid);
64 if (ret) {
65 of_node_put(parent);
66 return -ENXIO;
67 }
68
69 *id = PCI_SLOT_ID(phbid, bdfn);
70 return 0;
71 }
72
73 return -ENODEV;
74}
75EXPORT_SYMBOL_GPL(pnv_pci_get_slot_id);
76
77int pnv_pci_get_device_tree(uint32_t phandle, void *buf, uint64_t len)
78{
79 int64_t rc;
80
81 if (!opal_check_token(OPAL_GET_DEVICE_TREE))
82 return -ENXIO;
83
84 rc = opal_get_device_tree(phandle, (uint64_t)buf, len);
85 if (rc < OPAL_SUCCESS)
86 return -EIO;
87
88 return rc;
89}
90EXPORT_SYMBOL_GPL(pnv_pci_get_device_tree);
91
92int pnv_pci_get_presence_state(uint64_t id, uint8_t *state)
93{
94 int64_t rc;
95
96 if (!opal_check_token(OPAL_PCI_GET_PRESENCE_STATE))
97 return -ENXIO;
98
99 rc = opal_pci_get_presence_state(id, (uint64_t)state);
100 if (rc != OPAL_SUCCESS)
101 return -EIO;
102
103 return 0;
104}
105EXPORT_SYMBOL_GPL(pnv_pci_get_presence_state);
106
107int pnv_pci_get_power_state(uint64_t id, uint8_t *state)
108{
109 int64_t rc;
110
111 if (!opal_check_token(OPAL_PCI_GET_POWER_STATE))
112 return -ENXIO;
113
114 rc = opal_pci_get_power_state(id, (uint64_t)state);
115 if (rc != OPAL_SUCCESS)
116 return -EIO;
117
118 return 0;
119}
120EXPORT_SYMBOL_GPL(pnv_pci_get_power_state);
121
122int pnv_pci_set_power_state(uint64_t id, uint8_t state, struct opal_msg *msg)
123{
124 struct opal_msg m;
125 int token, ret;
126 int64_t rc;
127
128 if (!opal_check_token(OPAL_PCI_SET_POWER_STATE))
129 return -ENXIO;
130
131 token = opal_async_get_token_interruptible();
132 if (unlikely(token < 0))
133 return token;
134
135 rc = opal_pci_set_power_state(token, id, (uint64_t)&state);
136 if (rc == OPAL_SUCCESS) {
137 ret = 0;
138 goto exit;
139 } else if (rc != OPAL_ASYNC_COMPLETION) {
140 ret = -EIO;
141 goto exit;
142 }
143
144 ret = opal_async_wait_response(token, &m);
145 if (ret < 0)
146 goto exit;
147
148 if (msg) {
149 ret = 1;
150 memcpy(msg, &m, sizeof(m));
151 }
152
153exit:
154 opal_async_release_token(token);
155 return ret;
156}
157EXPORT_SYMBOL_GPL(pnv_pci_set_power_state);
158
159#ifdef CONFIG_PCI_MSI
160int pnv_setup_msi_irqs(struct pci_dev *pdev, int nvec, int type)
161{
162 struct pci_controller *hose = pci_bus_to_host(pdev->bus);
163 struct pnv_phb *phb = hose->private_data;
164 struct msi_desc *entry;
165 struct msi_msg msg;
166 int hwirq;
167 unsigned int virq;
168 int rc;
169
170 if (WARN_ON(!phb) || !phb->msi_bmp.bitmap)
171 return -ENODEV;
172
173 if (pdev->no_64bit_msi && !phb->msi32_support)
174 return -ENODEV;
175
176 for_each_pci_msi_entry(entry, pdev) {
177 if (!entry->msi_attrib.is_64 && !phb->msi32_support) {
178 pr_warn("%s: Supports only 64-bit MSIs\n",
179 pci_name(pdev));
180 return -ENXIO;
181 }
182 hwirq = msi_bitmap_alloc_hwirqs(&phb->msi_bmp, 1);
183 if (hwirq < 0) {
184 pr_warn("%s: Failed to find a free MSI\n",
185 pci_name(pdev));
186 return -ENOSPC;
187 }
188 virq = irq_create_mapping(NULL, phb->msi_base + hwirq);
189 if (virq == NO_IRQ) {
190 pr_warn("%s: Failed to map MSI to linux irq\n",
191 pci_name(pdev));
192 msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq, 1);
193 return -ENOMEM;
194 }
195 rc = phb->msi_setup(phb, pdev, phb->msi_base + hwirq,
196 virq, entry->msi_attrib.is_64, &msg);
197 if (rc) {
198 pr_warn("%s: Failed to setup MSI\n", pci_name(pdev));
199 irq_dispose_mapping(virq);
200 msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq, 1);
201 return rc;
202 }
203 irq_set_msi_desc(virq, entry);
204 pci_write_msi_msg(virq, &msg);
205 }
206 return 0;
207}
208
209void pnv_teardown_msi_irqs(struct pci_dev *pdev)
210{
211 struct pci_controller *hose = pci_bus_to_host(pdev->bus);
212 struct pnv_phb *phb = hose->private_data;
213 struct msi_desc *entry;
214 irq_hw_number_t hwirq;
215
216 if (WARN_ON(!phb))
217 return;
218
219 for_each_pci_msi_entry(entry, pdev) {
220 if (entry->irq == NO_IRQ)
221 continue;
222 hwirq = virq_to_hw(entry->irq);
223 irq_set_msi_desc(entry->irq, NULL);
224 irq_dispose_mapping(entry->irq);
225 msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq - phb->msi_base, 1);
226 }
227}
228#endif /* CONFIG_PCI_MSI */
229
230static void pnv_pci_dump_p7ioc_diag_data(struct pci_controller *hose,
231 struct OpalIoPhbErrorCommon *common)
232{
233 struct OpalIoP7IOCPhbErrorData *data;
234 int i;
235
236 data = (struct OpalIoP7IOCPhbErrorData *)common;
237 pr_info("P7IOC PHB#%d Diag-data (Version: %d)\n",
238 hose->global_number, be32_to_cpu(common->version));
239
240 if (data->brdgCtl)
241 pr_info("brdgCtl: %08x\n",
242 be32_to_cpu(data->brdgCtl));
243 if (data->portStatusReg || data->rootCmplxStatus ||
244 data->busAgentStatus)
245 pr_info("UtlSts: %08x %08x %08x\n",
246 be32_to_cpu(data->portStatusReg),
247 be32_to_cpu(data->rootCmplxStatus),
248 be32_to_cpu(data->busAgentStatus));
249 if (data->deviceStatus || data->slotStatus ||
250 data->linkStatus || data->devCmdStatus ||
251 data->devSecStatus)
252 pr_info("RootSts: %08x %08x %08x %08x %08x\n",
253 be32_to_cpu(data->deviceStatus),
254 be32_to_cpu(data->slotStatus),
255 be32_to_cpu(data->linkStatus),
256 be32_to_cpu(data->devCmdStatus),
257 be32_to_cpu(data->devSecStatus));
258 if (data->rootErrorStatus || data->uncorrErrorStatus ||
259 data->corrErrorStatus)
260 pr_info("RootErrSts: %08x %08x %08x\n",
261 be32_to_cpu(data->rootErrorStatus),
262 be32_to_cpu(data->uncorrErrorStatus),
263 be32_to_cpu(data->corrErrorStatus));
264 if (data->tlpHdr1 || data->tlpHdr2 ||
265 data->tlpHdr3 || data->tlpHdr4)
266 pr_info("RootErrLog: %08x %08x %08x %08x\n",
267 be32_to_cpu(data->tlpHdr1),
268 be32_to_cpu(data->tlpHdr2),
269 be32_to_cpu(data->tlpHdr3),
270 be32_to_cpu(data->tlpHdr4));
271 if (data->sourceId || data->errorClass ||
272 data->correlator)
273 pr_info("RootErrLog1: %08x %016llx %016llx\n",
274 be32_to_cpu(data->sourceId),
275 be64_to_cpu(data->errorClass),
276 be64_to_cpu(data->correlator));
277 if (data->p7iocPlssr || data->p7iocCsr)
278 pr_info("PhbSts: %016llx %016llx\n",
279 be64_to_cpu(data->p7iocPlssr),
280 be64_to_cpu(data->p7iocCsr));
281 if (data->lemFir)
282 pr_info("Lem: %016llx %016llx %016llx\n",
283 be64_to_cpu(data->lemFir),
284 be64_to_cpu(data->lemErrorMask),
285 be64_to_cpu(data->lemWOF));
286 if (data->phbErrorStatus)
287 pr_info("PhbErr: %016llx %016llx %016llx %016llx\n",
288 be64_to_cpu(data->phbErrorStatus),
289 be64_to_cpu(data->phbFirstErrorStatus),
290 be64_to_cpu(data->phbErrorLog0),
291 be64_to_cpu(data->phbErrorLog1));
292 if (data->mmioErrorStatus)
293 pr_info("OutErr: %016llx %016llx %016llx %016llx\n",
294 be64_to_cpu(data->mmioErrorStatus),
295 be64_to_cpu(data->mmioFirstErrorStatus),
296 be64_to_cpu(data->mmioErrorLog0),
297 be64_to_cpu(data->mmioErrorLog1));
298 if (data->dma0ErrorStatus)
299 pr_info("InAErr: %016llx %016llx %016llx %016llx\n",
300 be64_to_cpu(data->dma0ErrorStatus),
301 be64_to_cpu(data->dma0FirstErrorStatus),
302 be64_to_cpu(data->dma0ErrorLog0),
303 be64_to_cpu(data->dma0ErrorLog1));
304 if (data->dma1ErrorStatus)
305 pr_info("InBErr: %016llx %016llx %016llx %016llx\n",
306 be64_to_cpu(data->dma1ErrorStatus),
307 be64_to_cpu(data->dma1FirstErrorStatus),
308 be64_to_cpu(data->dma1ErrorLog0),
309 be64_to_cpu(data->dma1ErrorLog1));
310
311 for (i = 0; i < OPAL_P7IOC_NUM_PEST_REGS; i++) {
312 if ((data->pestA[i] >> 63) == 0 &&
313 (data->pestB[i] >> 63) == 0)
314 continue;
315
316 pr_info("PE[%3d] A/B: %016llx %016llx\n",
317 i, be64_to_cpu(data->pestA[i]),
318 be64_to_cpu(data->pestB[i]));
319 }
320}
321
322static void pnv_pci_dump_phb3_diag_data(struct pci_controller *hose,
323 struct OpalIoPhbErrorCommon *common)
324{
325 struct OpalIoPhb3ErrorData *data;
326 int i;
327
328 data = (struct OpalIoPhb3ErrorData*)common;
329 pr_info("PHB3 PHB#%d Diag-data (Version: %d)\n",
330 hose->global_number, be32_to_cpu(common->version));
331 if (data->brdgCtl)
332 pr_info("brdgCtl: %08x\n",
333 be32_to_cpu(data->brdgCtl));
334 if (data->portStatusReg || data->rootCmplxStatus ||
335 data->busAgentStatus)
336 pr_info("UtlSts: %08x %08x %08x\n",
337 be32_to_cpu(data->portStatusReg),
338 be32_to_cpu(data->rootCmplxStatus),
339 be32_to_cpu(data->busAgentStatus));
340 if (data->deviceStatus || data->slotStatus ||
341 data->linkStatus || data->devCmdStatus ||
342 data->devSecStatus)
343 pr_info("RootSts: %08x %08x %08x %08x %08x\n",
344 be32_to_cpu(data->deviceStatus),
345 be32_to_cpu(data->slotStatus),
346 be32_to_cpu(data->linkStatus),
347 be32_to_cpu(data->devCmdStatus),
348 be32_to_cpu(data->devSecStatus));
349 if (data->rootErrorStatus || data->uncorrErrorStatus ||
350 data->corrErrorStatus)
351 pr_info("RootErrSts: %08x %08x %08x\n",
352 be32_to_cpu(data->rootErrorStatus),
353 be32_to_cpu(data->uncorrErrorStatus),
354 be32_to_cpu(data->corrErrorStatus));
355 if (data->tlpHdr1 || data->tlpHdr2 ||
356 data->tlpHdr3 || data->tlpHdr4)
357 pr_info("RootErrLog: %08x %08x %08x %08x\n",
358 be32_to_cpu(data->tlpHdr1),
359 be32_to_cpu(data->tlpHdr2),
360 be32_to_cpu(data->tlpHdr3),
361 be32_to_cpu(data->tlpHdr4));
362 if (data->sourceId || data->errorClass ||
363 data->correlator)
364 pr_info("RootErrLog1: %08x %016llx %016llx\n",
365 be32_to_cpu(data->sourceId),
366 be64_to_cpu(data->errorClass),
367 be64_to_cpu(data->correlator));
368 if (data->nFir)
369 pr_info("nFir: %016llx %016llx %016llx\n",
370 be64_to_cpu(data->nFir),
371 be64_to_cpu(data->nFirMask),
372 be64_to_cpu(data->nFirWOF));
373 if (data->phbPlssr || data->phbCsr)
374 pr_info("PhbSts: %016llx %016llx\n",
375 be64_to_cpu(data->phbPlssr),
376 be64_to_cpu(data->phbCsr));
377 if (data->lemFir)
378 pr_info("Lem: %016llx %016llx %016llx\n",
379 be64_to_cpu(data->lemFir),
380 be64_to_cpu(data->lemErrorMask),
381 be64_to_cpu(data->lemWOF));
382 if (data->phbErrorStatus)
383 pr_info("PhbErr: %016llx %016llx %016llx %016llx\n",
384 be64_to_cpu(data->phbErrorStatus),
385 be64_to_cpu(data->phbFirstErrorStatus),
386 be64_to_cpu(data->phbErrorLog0),
387 be64_to_cpu(data->phbErrorLog1));
388 if (data->mmioErrorStatus)
389 pr_info("OutErr: %016llx %016llx %016llx %016llx\n",
390 be64_to_cpu(data->mmioErrorStatus),
391 be64_to_cpu(data->mmioFirstErrorStatus),
392 be64_to_cpu(data->mmioErrorLog0),
393 be64_to_cpu(data->mmioErrorLog1));
394 if (data->dma0ErrorStatus)
395 pr_info("InAErr: %016llx %016llx %016llx %016llx\n",
396 be64_to_cpu(data->dma0ErrorStatus),
397 be64_to_cpu(data->dma0FirstErrorStatus),
398 be64_to_cpu(data->dma0ErrorLog0),
399 be64_to_cpu(data->dma0ErrorLog1));
400 if (data->dma1ErrorStatus)
401 pr_info("InBErr: %016llx %016llx %016llx %016llx\n",
402 be64_to_cpu(data->dma1ErrorStatus),
403 be64_to_cpu(data->dma1FirstErrorStatus),
404 be64_to_cpu(data->dma1ErrorLog0),
405 be64_to_cpu(data->dma1ErrorLog1));
406
407 for (i = 0; i < OPAL_PHB3_NUM_PEST_REGS; i++) {
408 if ((be64_to_cpu(data->pestA[i]) >> 63) == 0 &&
409 (be64_to_cpu(data->pestB[i]) >> 63) == 0)
410 continue;
411
412 pr_info("PE[%3d] A/B: %016llx %016llx\n",
413 i, be64_to_cpu(data->pestA[i]),
414 be64_to_cpu(data->pestB[i]));
415 }
416}
417
418void pnv_pci_dump_phb_diag_data(struct pci_controller *hose,
419 unsigned char *log_buff)
420{
421 struct OpalIoPhbErrorCommon *common;
422
423 if (!hose || !log_buff)
424 return;
425
426 common = (struct OpalIoPhbErrorCommon *)log_buff;
427 switch (be32_to_cpu(common->ioType)) {
428 case OPAL_PHB_ERROR_DATA_TYPE_P7IOC:
429 pnv_pci_dump_p7ioc_diag_data(hose, common);
430 break;
431 case OPAL_PHB_ERROR_DATA_TYPE_PHB3:
432 pnv_pci_dump_phb3_diag_data(hose, common);
433 break;
434 default:
435 pr_warn("%s: Unrecognized ioType %d\n",
436 __func__, be32_to_cpu(common->ioType));
437 }
438}
439
440static void pnv_pci_handle_eeh_config(struct pnv_phb *phb, u32 pe_no)
441{
442 unsigned long flags, rc;
443 int has_diag, ret = 0;
444
445 spin_lock_irqsave(&phb->lock, flags);
446
447 /* Fetch PHB diag-data */
448 rc = opal_pci_get_phb_diag_data2(phb->opal_id, phb->diag.blob,
449 PNV_PCI_DIAG_BUF_SIZE);
450 has_diag = (rc == OPAL_SUCCESS);
451
452 /* If PHB supports compound PE, to handle it */
453 if (phb->unfreeze_pe) {
454 ret = phb->unfreeze_pe(phb,
455 pe_no,
456 OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
457 } else {
458 rc = opal_pci_eeh_freeze_clear(phb->opal_id,
459 pe_no,
460 OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
461 if (rc) {
462 pr_warn("%s: Failure %ld clearing frozen "
463 "PHB#%x-PE#%x\n",
464 __func__, rc, phb->hose->global_number,
465 pe_no);
466 ret = -EIO;
467 }
468 }
469
470 /*
471 * For now, let's only display the diag buffer when we fail to clear
472 * the EEH status. We'll do more sensible things later when we have
473 * proper EEH support. We need to make sure we don't pollute ourselves
474 * with the normal errors generated when probing empty slots
475 */
476 if (has_diag && ret)
477 pnv_pci_dump_phb_diag_data(phb->hose, phb->diag.blob);
478
479 spin_unlock_irqrestore(&phb->lock, flags);
480}
481
482static void pnv_pci_config_check_eeh(struct pci_dn *pdn)
483{
484 struct pnv_phb *phb = pdn->phb->private_data;
485 u8 fstate;
486 __be16 pcierr;
487 unsigned int pe_no;
488 s64 rc;
489
490 /*
491 * Get the PE#. During the PCI probe stage, we might not
492 * setup that yet. So all ER errors should be mapped to
493 * reserved PE.
494 */
495 pe_no = pdn->pe_number;
496 if (pe_no == IODA_INVALID_PE) {
497 pe_no = phb->ioda.reserved_pe_idx;
498 }
499
500 /*
501 * Fetch frozen state. If the PHB support compound PE,
502 * we need handle that case.
503 */
504 if (phb->get_pe_state) {
505 fstate = phb->get_pe_state(phb, pe_no);
506 } else {
507 rc = opal_pci_eeh_freeze_status(phb->opal_id,
508 pe_no,
509 &fstate,
510 &pcierr,
511 NULL);
512 if (rc) {
513 pr_warn("%s: Failure %lld getting PHB#%x-PE#%x state\n",
514 __func__, rc, phb->hose->global_number, pe_no);
515 return;
516 }
517 }
518
519 pr_devel(" -> EEH check, bdfn=%04x PE#%d fstate=%x\n",
520 (pdn->busno << 8) | (pdn->devfn), pe_no, fstate);
521
522 /* Clear the frozen state if applicable */
523 if (fstate == OPAL_EEH_STOPPED_MMIO_FREEZE ||
524 fstate == OPAL_EEH_STOPPED_DMA_FREEZE ||
525 fstate == OPAL_EEH_STOPPED_MMIO_DMA_FREEZE) {
526 /*
527 * If PHB supports compound PE, freeze it for
528 * consistency.
529 */
530 if (phb->freeze_pe)
531 phb->freeze_pe(phb, pe_no);
532
533 pnv_pci_handle_eeh_config(phb, pe_no);
534 }
535}
536
537int pnv_pci_cfg_read(struct pci_dn *pdn,
538 int where, int size, u32 *val)
539{
540 struct pnv_phb *phb = pdn->phb->private_data;
541 u32 bdfn = (pdn->busno << 8) | pdn->devfn;
542 s64 rc;
543
544 switch (size) {
545 case 1: {
546 u8 v8;
547 rc = opal_pci_config_read_byte(phb->opal_id, bdfn, where, &v8);
548 *val = (rc == OPAL_SUCCESS) ? v8 : 0xff;
549 break;
550 }
551 case 2: {
552 __be16 v16;
553 rc = opal_pci_config_read_half_word(phb->opal_id, bdfn, where,
554 &v16);
555 *val = (rc == OPAL_SUCCESS) ? be16_to_cpu(v16) : 0xffff;
556 break;
557 }
558 case 4: {
559 __be32 v32;
560 rc = opal_pci_config_read_word(phb->opal_id, bdfn, where, &v32);
561 *val = (rc == OPAL_SUCCESS) ? be32_to_cpu(v32) : 0xffffffff;
562 break;
563 }
564 default:
565 return PCIBIOS_FUNC_NOT_SUPPORTED;
566 }
567
568 pr_devel("%s: bus: %x devfn: %x +%x/%x -> %08x\n",
569 __func__, pdn->busno, pdn->devfn, where, size, *val);
570 return PCIBIOS_SUCCESSFUL;
571}
572
573int pnv_pci_cfg_write(struct pci_dn *pdn,
574 int where, int size, u32 val)
575{
576 struct pnv_phb *phb = pdn->phb->private_data;
577 u32 bdfn = (pdn->busno << 8) | pdn->devfn;
578
579 pr_devel("%s: bus: %x devfn: %x +%x/%x -> %08x\n",
580 __func__, pdn->busno, pdn->devfn, where, size, val);
581 switch (size) {
582 case 1:
583 opal_pci_config_write_byte(phb->opal_id, bdfn, where, val);
584 break;
585 case 2:
586 opal_pci_config_write_half_word(phb->opal_id, bdfn, where, val);
587 break;
588 case 4:
589 opal_pci_config_write_word(phb->opal_id, bdfn, where, val);
590 break;
591 default:
592 return PCIBIOS_FUNC_NOT_SUPPORTED;
593 }
594
595 return PCIBIOS_SUCCESSFUL;
596}
597
598#if CONFIG_EEH
599static bool pnv_pci_cfg_check(struct pci_dn *pdn)
600{
601 struct eeh_dev *edev = NULL;
602 struct pnv_phb *phb = pdn->phb->private_data;
603
604 /* EEH not enabled ? */
605 if (!(phb->flags & PNV_PHB_FLAG_EEH))
606 return true;
607
608 /* PE reset or device removed ? */
609 edev = pdn->edev;
610 if (edev) {
611 if (edev->pe &&
612 (edev->pe->state & EEH_PE_CFG_BLOCKED))
613 return false;
614
615 if (edev->mode & EEH_DEV_REMOVED)
616 return false;
617 }
618
619 return true;
620}
621#else
622static inline pnv_pci_cfg_check(struct pci_dn *pdn)
623{
624 return true;
625}
626#endif /* CONFIG_EEH */
627
628static int pnv_pci_read_config(struct pci_bus *bus,
629 unsigned int devfn,
630 int where, int size, u32 *val)
631{
632 struct pci_dn *pdn;
633 struct pnv_phb *phb;
634 int ret;
635
636 *val = 0xFFFFFFFF;
637 pdn = pci_get_pdn_by_devfn(bus, devfn);
638 if (!pdn)
639 return PCIBIOS_DEVICE_NOT_FOUND;
640
641 if (!pnv_pci_cfg_check(pdn))
642 return PCIBIOS_DEVICE_NOT_FOUND;
643
644 ret = pnv_pci_cfg_read(pdn, where, size, val);
645 phb = pdn->phb->private_data;
646 if (phb->flags & PNV_PHB_FLAG_EEH && pdn->edev) {
647 if (*val == EEH_IO_ERROR_VALUE(size) &&
648 eeh_dev_check_failure(pdn->edev))
649 return PCIBIOS_DEVICE_NOT_FOUND;
650 } else {
651 pnv_pci_config_check_eeh(pdn);
652 }
653
654 return ret;
655}
656
657static int pnv_pci_write_config(struct pci_bus *bus,
658 unsigned int devfn,
659 int where, int size, u32 val)
660{
661 struct pci_dn *pdn;
662 struct pnv_phb *phb;
663 int ret;
664
665 pdn = pci_get_pdn_by_devfn(bus, devfn);
666 if (!pdn)
667 return PCIBIOS_DEVICE_NOT_FOUND;
668
669 if (!pnv_pci_cfg_check(pdn))
670 return PCIBIOS_DEVICE_NOT_FOUND;
671
672 ret = pnv_pci_cfg_write(pdn, where, size, val);
673 phb = pdn->phb->private_data;
674 if (!(phb->flags & PNV_PHB_FLAG_EEH))
675 pnv_pci_config_check_eeh(pdn);
676
677 return ret;
678}
679
680struct pci_ops pnv_pci_ops = {
681 .read = pnv_pci_read_config,
682 .write = pnv_pci_write_config,
683};
684
685static __be64 *pnv_tce(struct iommu_table *tbl, long idx)
686{
687 __be64 *tmp = ((__be64 *)tbl->it_base);
688 int level = tbl->it_indirect_levels;
689 const long shift = ilog2(tbl->it_level_size);
690 unsigned long mask = (tbl->it_level_size - 1) << (level * shift);
691
692 while (level) {
693 int n = (idx & mask) >> (level * shift);
694 unsigned long tce = be64_to_cpu(tmp[n]);
695
696 tmp = __va(tce & ~(TCE_PCI_READ | TCE_PCI_WRITE));
697 idx &= ~mask;
698 mask >>= shift;
699 --level;
700 }
701
702 return tmp + idx;
703}
704
705int pnv_tce_build(struct iommu_table *tbl, long index, long npages,
706 unsigned long uaddr, enum dma_data_direction direction,
707 struct dma_attrs *attrs)
708{
709 u64 proto_tce = iommu_direction_to_tce_perm(direction);
710 u64 rpn = __pa(uaddr) >> tbl->it_page_shift;
711 long i;
712
713 if (proto_tce & TCE_PCI_WRITE)
714 proto_tce |= TCE_PCI_READ;
715
716 for (i = 0; i < npages; i++) {
717 unsigned long newtce = proto_tce |
718 ((rpn + i) << tbl->it_page_shift);
719 unsigned long idx = index - tbl->it_offset + i;
720
721 *(pnv_tce(tbl, idx)) = cpu_to_be64(newtce);
722 }
723
724 return 0;
725}
726
727#ifdef CONFIG_IOMMU_API
728int pnv_tce_xchg(struct iommu_table *tbl, long index,
729 unsigned long *hpa, enum dma_data_direction *direction)
730{
731 u64 proto_tce = iommu_direction_to_tce_perm(*direction);
732 unsigned long newtce = *hpa | proto_tce, oldtce;
733 unsigned long idx = index - tbl->it_offset;
734
735 BUG_ON(*hpa & ~IOMMU_PAGE_MASK(tbl));
736
737 if (newtce & TCE_PCI_WRITE)
738 newtce |= TCE_PCI_READ;
739
740 oldtce = xchg(pnv_tce(tbl, idx), cpu_to_be64(newtce));
741 *hpa = be64_to_cpu(oldtce) & ~(TCE_PCI_READ | TCE_PCI_WRITE);
742 *direction = iommu_tce_direction(oldtce);
743
744 return 0;
745}
746#endif
747
748void pnv_tce_free(struct iommu_table *tbl, long index, long npages)
749{
750 long i;
751
752 for (i = 0; i < npages; i++) {
753 unsigned long idx = index - tbl->it_offset + i;
754
755 *(pnv_tce(tbl, idx)) = cpu_to_be64(0);
756 }
757}
758
759unsigned long pnv_tce_get(struct iommu_table *tbl, long index)
760{
761 return *(pnv_tce(tbl, index - tbl->it_offset));
762}
763
764struct iommu_table *pnv_pci_table_alloc(int nid)
765{
766 struct iommu_table *tbl;
767
768 tbl = kzalloc_node(sizeof(struct iommu_table), GFP_KERNEL, nid);
769 INIT_LIST_HEAD_RCU(&tbl->it_group_list);
770
771 return tbl;
772}
773
774long pnv_pci_link_table_and_group(int node, int num,
775 struct iommu_table *tbl,
776 struct iommu_table_group *table_group)
777{
778 struct iommu_table_group_link *tgl = NULL;
779
780 if (WARN_ON(!tbl || !table_group))
781 return -EINVAL;
782
783 tgl = kzalloc_node(sizeof(struct iommu_table_group_link), GFP_KERNEL,
784 node);
785 if (!tgl)
786 return -ENOMEM;
787
788 tgl->table_group = table_group;
789 list_add_rcu(&tgl->next, &tbl->it_group_list);
790
791 table_group->tables[num] = tbl;
792
793 return 0;
794}
795
796static void pnv_iommu_table_group_link_free(struct rcu_head *head)
797{
798 struct iommu_table_group_link *tgl = container_of(head,
799 struct iommu_table_group_link, rcu);
800
801 kfree(tgl);
802}
803
804void pnv_pci_unlink_table_and_group(struct iommu_table *tbl,
805 struct iommu_table_group *table_group)
806{
807 long i;
808 bool found;
809 struct iommu_table_group_link *tgl;
810
811 if (!tbl || !table_group)
812 return;
813
814 /* Remove link to a group from table's list of attached groups */
815 found = false;
816 list_for_each_entry_rcu(tgl, &tbl->it_group_list, next) {
817 if (tgl->table_group == table_group) {
818 list_del_rcu(&tgl->next);
819 call_rcu(&tgl->rcu, pnv_iommu_table_group_link_free);
820 found = true;
821 break;
822 }
823 }
824 if (WARN_ON(!found))
825 return;
826
827 /* Clean a pointer to iommu_table in iommu_table_group::tables[] */
828 found = false;
829 for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) {
830 if (table_group->tables[i] == tbl) {
831 table_group->tables[i] = NULL;
832 found = true;
833 break;
834 }
835 }
836 WARN_ON(!found);
837}
838
839void pnv_pci_setup_iommu_table(struct iommu_table *tbl,
840 void *tce_mem, u64 tce_size,
841 u64 dma_offset, unsigned page_shift)
842{
843 tbl->it_blocksize = 16;
844 tbl->it_base = (unsigned long)tce_mem;
845 tbl->it_page_shift = page_shift;
846 tbl->it_offset = dma_offset >> tbl->it_page_shift;
847 tbl->it_index = 0;
848 tbl->it_size = tce_size >> 3;
849 tbl->it_busno = 0;
850 tbl->it_type = TCE_PCI;
851}
852
853void pnv_pci_dma_dev_setup(struct pci_dev *pdev)
854{
855 struct pci_controller *hose = pci_bus_to_host(pdev->bus);
856 struct pnv_phb *phb = hose->private_data;
857#ifdef CONFIG_PCI_IOV
858 struct pnv_ioda_pe *pe;
859 struct pci_dn *pdn;
860
861 /* Fix the VF pdn PE number */
862 if (pdev->is_virtfn) {
863 pdn = pci_get_pdn(pdev);
864 WARN_ON(pdn->pe_number != IODA_INVALID_PE);
865 list_for_each_entry(pe, &phb->ioda.pe_list, list) {
866 if (pe->rid == ((pdev->bus->number << 8) |
867 (pdev->devfn & 0xff))) {
868 pdn->pe_number = pe->pe_number;
869 pe->pdev = pdev;
870 break;
871 }
872 }
873 }
874#endif /* CONFIG_PCI_IOV */
875
876 if (phb && phb->dma_dev_setup)
877 phb->dma_dev_setup(phb, pdev);
878}
879
880void pnv_pci_dma_bus_setup(struct pci_bus *bus)
881{
882 struct pci_controller *hose = bus->sysdata;
883 struct pnv_phb *phb = hose->private_data;
884 struct pnv_ioda_pe *pe;
885
886 list_for_each_entry(pe, &phb->ioda.pe_list, list) {
887 if (!(pe->flags & (PNV_IODA_PE_BUS | PNV_IODA_PE_BUS_ALL)))
888 continue;
889
890 if (!pe->pbus)
891 continue;
892
893 if (bus->number == ((pe->rid >> 8) & 0xFF)) {
894 pe->pbus = bus;
895 break;
896 }
897 }
898}
899
900void pnv_pci_shutdown(void)
901{
902 struct pci_controller *hose;
903
904 list_for_each_entry(hose, &hose_list, list_node)
905 if (hose->controller_ops.shutdown)
906 hose->controller_ops.shutdown(hose);
907}
908
909/* Fixup wrong class code in p7ioc and p8 root complex */
910static void pnv_p7ioc_rc_quirk(struct pci_dev *dev)
911{
912 dev->class = PCI_CLASS_BRIDGE_PCI << 8;
913}
914DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_IBM, 0x3b9, pnv_p7ioc_rc_quirk);
915
916void __init pnv_pci_init(void)
917{
918 struct device_node *np;
919
920 pci_add_flags(PCI_CAN_SKIP_ISA_ALIGN);
921
922 /* If we don't have OPAL, eg. in sim, just skip PCI probe */
923 if (!firmware_has_feature(FW_FEATURE_OPAL))
924 return;
925
926 /* Look for IODA IO-Hubs. */
927 for_each_compatible_node(np, NULL, "ibm,ioda-hub") {
928 pnv_pci_init_ioda_hub(np);
929 }
930
931 /* Look for ioda2 built-in PHB3's */
932 for_each_compatible_node(np, NULL, "ibm,ioda2-phb")
933 pnv_pci_init_ioda2_phb(np);
934
935 /* Look for ioda3 built-in PHB4's, we treat them as IODA2 */
936 for_each_compatible_node(np, NULL, "ibm,ioda3-phb")
937 pnv_pci_init_ioda2_phb(np);
938
939 /* Look for NPU PHBs */
940 for_each_compatible_node(np, NULL, "ibm,ioda2-npu-phb")
941 pnv_pci_init_npu_phb(np);
942
943 /* Configure IOMMU DMA hooks */
944 set_pci_dma_ops(&dma_iommu_ops);
945}
946
947machine_subsys_initcall_sync(powernv, tce_iommu_bus_notifier_init);