2 * Support PCI/PCIe on PowerNV platforms
4 * Copyright 2011 Benjamin Herrenschmidt, IBM Corp.
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.
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>
19 #include <linux/msi.h>
20 #include <linux/iommu.h>
21 #include <linux/sched/mm.h>
23 #include <asm/sections.h>
26 #include <asm/pci-bridge.h>
27 #include <asm/machdep.h>
28 #include <asm/msi_bitmap.h>
29 #include <asm/ppc-pci.h>
30 #include <asm/pnv-pci.h>
32 #include <asm/iommu.h>
34 #include <asm/firmware.h>
35 #include <asm/eeh_event.h>
41 static DEFINE_MUTEX(p2p_mutex);
42 static DEFINE_MUTEX(tunnel_mutex);
44 int pnv_pci_get_slot_id(struct device_node *np, uint64_t *id)
46 struct device_node *parent = np;
51 ret = of_property_read_u32(np, "reg", &bdfn);
55 bdfn = ((bdfn & 0x00ffff00) >> 8);
56 while ((parent = of_get_parent(parent))) {
57 if (!PCI_DN(parent)) {
62 if (!of_device_is_compatible(parent, "ibm,ioda2-phb")) {
67 ret = of_property_read_u64(parent, "ibm,opal-phbid", &phbid);
73 *id = PCI_SLOT_ID(phbid, bdfn);
79 EXPORT_SYMBOL_GPL(pnv_pci_get_slot_id);
81 int pnv_pci_get_device_tree(uint32_t phandle, void *buf, uint64_t len)
85 if (!opal_check_token(OPAL_GET_DEVICE_TREE))
88 rc = opal_get_device_tree(phandle, (uint64_t)buf, len);
89 if (rc < OPAL_SUCCESS)
94 EXPORT_SYMBOL_GPL(pnv_pci_get_device_tree);
96 int pnv_pci_get_presence_state(uint64_t id, uint8_t *state)
100 if (!opal_check_token(OPAL_PCI_GET_PRESENCE_STATE))
103 rc = opal_pci_get_presence_state(id, (uint64_t)state);
104 if (rc != OPAL_SUCCESS)
109 EXPORT_SYMBOL_GPL(pnv_pci_get_presence_state);
111 int pnv_pci_get_power_state(uint64_t id, uint8_t *state)
115 if (!opal_check_token(OPAL_PCI_GET_POWER_STATE))
118 rc = opal_pci_get_power_state(id, (uint64_t)state);
119 if (rc != OPAL_SUCCESS)
124 EXPORT_SYMBOL_GPL(pnv_pci_get_power_state);
126 int pnv_pci_set_power_state(uint64_t id, uint8_t state, struct opal_msg *msg)
132 if (!opal_check_token(OPAL_PCI_SET_POWER_STATE))
135 token = opal_async_get_token_interruptible();
136 if (unlikely(token < 0))
139 rc = opal_pci_set_power_state(token, id, (uint64_t)&state);
140 if (rc == OPAL_SUCCESS) {
143 } else if (rc != OPAL_ASYNC_COMPLETION) {
148 ret = opal_async_wait_response(token, &m);
154 memcpy(msg, &m, sizeof(m));
158 opal_async_release_token(token);
161 EXPORT_SYMBOL_GPL(pnv_pci_set_power_state);
163 int pnv_setup_msi_irqs(struct pci_dev *pdev, int nvec, int type)
165 struct pci_controller *hose = pci_bus_to_host(pdev->bus);
166 struct pnv_phb *phb = hose->private_data;
167 struct msi_desc *entry;
173 if (WARN_ON(!phb) || !phb->msi_bmp.bitmap)
176 if (pdev->no_64bit_msi && !phb->msi32_support)
179 for_each_pci_msi_entry(entry, pdev) {
180 if (!entry->msi_attrib.is_64 && !phb->msi32_support) {
181 pr_warn("%s: Supports only 64-bit MSIs\n",
185 hwirq = msi_bitmap_alloc_hwirqs(&phb->msi_bmp, 1);
187 pr_warn("%s: Failed to find a free MSI\n",
191 virq = irq_create_mapping(NULL, phb->msi_base + hwirq);
193 pr_warn("%s: Failed to map MSI to linux irq\n",
195 msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq, 1);
198 rc = phb->msi_setup(phb, pdev, phb->msi_base + hwirq,
199 virq, entry->msi_attrib.is_64, &msg);
201 pr_warn("%s: Failed to setup MSI\n", pci_name(pdev));
202 irq_dispose_mapping(virq);
203 msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq, 1);
206 irq_set_msi_desc(virq, entry);
207 pci_write_msi_msg(virq, &msg);
212 void pnv_teardown_msi_irqs(struct pci_dev *pdev)
214 struct pci_controller *hose = pci_bus_to_host(pdev->bus);
215 struct pnv_phb *phb = hose->private_data;
216 struct msi_desc *entry;
217 irq_hw_number_t hwirq;
222 for_each_pci_msi_entry(entry, pdev) {
225 hwirq = virq_to_hw(entry->irq);
226 irq_set_msi_desc(entry->irq, NULL);
227 irq_dispose_mapping(entry->irq);
228 msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq - phb->msi_base, 1);
232 /* Nicely print the contents of the PE State Tables (PEST). */
233 static void pnv_pci_dump_pest(__be64 pestA[], __be64 pestB[], int pest_size)
235 __be64 prevA = ULONG_MAX, prevB = ULONG_MAX;
239 for (i = 0; i < pest_size; i++) {
240 __be64 peA = be64_to_cpu(pestA[i]);
241 __be64 peB = be64_to_cpu(pestB[i]);
243 if (peA != prevA || peB != prevB) {
245 pr_info("PE[..%03x] A/B: as above\n", i-1);
250 if (peA & PNV_IODA_STOPPED_STATE ||
251 peB & PNV_IODA_STOPPED_STATE)
252 pr_info("PE[%03x] A/B: %016llx %016llx\n",
254 } else if (!dup && (peA & PNV_IODA_STOPPED_STATE ||
255 peB & PNV_IODA_STOPPED_STATE)) {
261 static void pnv_pci_dump_p7ioc_diag_data(struct pci_controller *hose,
262 struct OpalIoPhbErrorCommon *common)
264 struct OpalIoP7IOCPhbErrorData *data;
266 data = (struct OpalIoP7IOCPhbErrorData *)common;
267 pr_info("P7IOC PHB#%x Diag-data (Version: %d)\n",
268 hose->global_number, be32_to_cpu(common->version));
271 pr_info("brdgCtl: %08x\n",
272 be32_to_cpu(data->brdgCtl));
273 if (data->portStatusReg || data->rootCmplxStatus ||
274 data->busAgentStatus)
275 pr_info("UtlSts: %08x %08x %08x\n",
276 be32_to_cpu(data->portStatusReg),
277 be32_to_cpu(data->rootCmplxStatus),
278 be32_to_cpu(data->busAgentStatus));
279 if (data->deviceStatus || data->slotStatus ||
280 data->linkStatus || data->devCmdStatus ||
282 pr_info("RootSts: %08x %08x %08x %08x %08x\n",
283 be32_to_cpu(data->deviceStatus),
284 be32_to_cpu(data->slotStatus),
285 be32_to_cpu(data->linkStatus),
286 be32_to_cpu(data->devCmdStatus),
287 be32_to_cpu(data->devSecStatus));
288 if (data->rootErrorStatus || data->uncorrErrorStatus ||
289 data->corrErrorStatus)
290 pr_info("RootErrSts: %08x %08x %08x\n",
291 be32_to_cpu(data->rootErrorStatus),
292 be32_to_cpu(data->uncorrErrorStatus),
293 be32_to_cpu(data->corrErrorStatus));
294 if (data->tlpHdr1 || data->tlpHdr2 ||
295 data->tlpHdr3 || data->tlpHdr4)
296 pr_info("RootErrLog: %08x %08x %08x %08x\n",
297 be32_to_cpu(data->tlpHdr1),
298 be32_to_cpu(data->tlpHdr2),
299 be32_to_cpu(data->tlpHdr3),
300 be32_to_cpu(data->tlpHdr4));
301 if (data->sourceId || data->errorClass ||
303 pr_info("RootErrLog1: %08x %016llx %016llx\n",
304 be32_to_cpu(data->sourceId),
305 be64_to_cpu(data->errorClass),
306 be64_to_cpu(data->correlator));
307 if (data->p7iocPlssr || data->p7iocCsr)
308 pr_info("PhbSts: %016llx %016llx\n",
309 be64_to_cpu(data->p7iocPlssr),
310 be64_to_cpu(data->p7iocCsr));
312 pr_info("Lem: %016llx %016llx %016llx\n",
313 be64_to_cpu(data->lemFir),
314 be64_to_cpu(data->lemErrorMask),
315 be64_to_cpu(data->lemWOF));
316 if (data->phbErrorStatus)
317 pr_info("PhbErr: %016llx %016llx %016llx %016llx\n",
318 be64_to_cpu(data->phbErrorStatus),
319 be64_to_cpu(data->phbFirstErrorStatus),
320 be64_to_cpu(data->phbErrorLog0),
321 be64_to_cpu(data->phbErrorLog1));
322 if (data->mmioErrorStatus)
323 pr_info("OutErr: %016llx %016llx %016llx %016llx\n",
324 be64_to_cpu(data->mmioErrorStatus),
325 be64_to_cpu(data->mmioFirstErrorStatus),
326 be64_to_cpu(data->mmioErrorLog0),
327 be64_to_cpu(data->mmioErrorLog1));
328 if (data->dma0ErrorStatus)
329 pr_info("InAErr: %016llx %016llx %016llx %016llx\n",
330 be64_to_cpu(data->dma0ErrorStatus),
331 be64_to_cpu(data->dma0FirstErrorStatus),
332 be64_to_cpu(data->dma0ErrorLog0),
333 be64_to_cpu(data->dma0ErrorLog1));
334 if (data->dma1ErrorStatus)
335 pr_info("InBErr: %016llx %016llx %016llx %016llx\n",
336 be64_to_cpu(data->dma1ErrorStatus),
337 be64_to_cpu(data->dma1FirstErrorStatus),
338 be64_to_cpu(data->dma1ErrorLog0),
339 be64_to_cpu(data->dma1ErrorLog1));
341 pnv_pci_dump_pest(data->pestA, data->pestB, OPAL_P7IOC_NUM_PEST_REGS);
344 static void pnv_pci_dump_phb3_diag_data(struct pci_controller *hose,
345 struct OpalIoPhbErrorCommon *common)
347 struct OpalIoPhb3ErrorData *data;
349 data = (struct OpalIoPhb3ErrorData*)common;
350 pr_info("PHB3 PHB#%x Diag-data (Version: %d)\n",
351 hose->global_number, be32_to_cpu(common->version));
353 pr_info("brdgCtl: %08x\n",
354 be32_to_cpu(data->brdgCtl));
355 if (data->portStatusReg || data->rootCmplxStatus ||
356 data->busAgentStatus)
357 pr_info("UtlSts: %08x %08x %08x\n",
358 be32_to_cpu(data->portStatusReg),
359 be32_to_cpu(data->rootCmplxStatus),
360 be32_to_cpu(data->busAgentStatus));
361 if (data->deviceStatus || data->slotStatus ||
362 data->linkStatus || data->devCmdStatus ||
364 pr_info("RootSts: %08x %08x %08x %08x %08x\n",
365 be32_to_cpu(data->deviceStatus),
366 be32_to_cpu(data->slotStatus),
367 be32_to_cpu(data->linkStatus),
368 be32_to_cpu(data->devCmdStatus),
369 be32_to_cpu(data->devSecStatus));
370 if (data->rootErrorStatus || data->uncorrErrorStatus ||
371 data->corrErrorStatus)
372 pr_info("RootErrSts: %08x %08x %08x\n",
373 be32_to_cpu(data->rootErrorStatus),
374 be32_to_cpu(data->uncorrErrorStatus),
375 be32_to_cpu(data->corrErrorStatus));
376 if (data->tlpHdr1 || data->tlpHdr2 ||
377 data->tlpHdr3 || data->tlpHdr4)
378 pr_info("RootErrLog: %08x %08x %08x %08x\n",
379 be32_to_cpu(data->tlpHdr1),
380 be32_to_cpu(data->tlpHdr2),
381 be32_to_cpu(data->tlpHdr3),
382 be32_to_cpu(data->tlpHdr4));
383 if (data->sourceId || data->errorClass ||
385 pr_info("RootErrLog1: %08x %016llx %016llx\n",
386 be32_to_cpu(data->sourceId),
387 be64_to_cpu(data->errorClass),
388 be64_to_cpu(data->correlator));
390 pr_info("nFir: %016llx %016llx %016llx\n",
391 be64_to_cpu(data->nFir),
392 be64_to_cpu(data->nFirMask),
393 be64_to_cpu(data->nFirWOF));
394 if (data->phbPlssr || data->phbCsr)
395 pr_info("PhbSts: %016llx %016llx\n",
396 be64_to_cpu(data->phbPlssr),
397 be64_to_cpu(data->phbCsr));
399 pr_info("Lem: %016llx %016llx %016llx\n",
400 be64_to_cpu(data->lemFir),
401 be64_to_cpu(data->lemErrorMask),
402 be64_to_cpu(data->lemWOF));
403 if (data->phbErrorStatus)
404 pr_info("PhbErr: %016llx %016llx %016llx %016llx\n",
405 be64_to_cpu(data->phbErrorStatus),
406 be64_to_cpu(data->phbFirstErrorStatus),
407 be64_to_cpu(data->phbErrorLog0),
408 be64_to_cpu(data->phbErrorLog1));
409 if (data->mmioErrorStatus)
410 pr_info("OutErr: %016llx %016llx %016llx %016llx\n",
411 be64_to_cpu(data->mmioErrorStatus),
412 be64_to_cpu(data->mmioFirstErrorStatus),
413 be64_to_cpu(data->mmioErrorLog0),
414 be64_to_cpu(data->mmioErrorLog1));
415 if (data->dma0ErrorStatus)
416 pr_info("InAErr: %016llx %016llx %016llx %016llx\n",
417 be64_to_cpu(data->dma0ErrorStatus),
418 be64_to_cpu(data->dma0FirstErrorStatus),
419 be64_to_cpu(data->dma0ErrorLog0),
420 be64_to_cpu(data->dma0ErrorLog1));
421 if (data->dma1ErrorStatus)
422 pr_info("InBErr: %016llx %016llx %016llx %016llx\n",
423 be64_to_cpu(data->dma1ErrorStatus),
424 be64_to_cpu(data->dma1FirstErrorStatus),
425 be64_to_cpu(data->dma1ErrorLog0),
426 be64_to_cpu(data->dma1ErrorLog1));
428 pnv_pci_dump_pest(data->pestA, data->pestB, OPAL_PHB3_NUM_PEST_REGS);
431 static void pnv_pci_dump_phb4_diag_data(struct pci_controller *hose,
432 struct OpalIoPhbErrorCommon *common)
434 struct OpalIoPhb4ErrorData *data;
436 data = (struct OpalIoPhb4ErrorData*)common;
437 pr_info("PHB4 PHB#%d Diag-data (Version: %d)\n",
438 hose->global_number, be32_to_cpu(common->version));
440 pr_info("brdgCtl: %08x\n",
441 be32_to_cpu(data->brdgCtl));
442 if (data->deviceStatus || data->slotStatus ||
443 data->linkStatus || data->devCmdStatus ||
445 pr_info("RootSts: %08x %08x %08x %08x %08x\n",
446 be32_to_cpu(data->deviceStatus),
447 be32_to_cpu(data->slotStatus),
448 be32_to_cpu(data->linkStatus),
449 be32_to_cpu(data->devCmdStatus),
450 be32_to_cpu(data->devSecStatus));
451 if (data->rootErrorStatus || data->uncorrErrorStatus ||
452 data->corrErrorStatus)
453 pr_info("RootErrSts: %08x %08x %08x\n",
454 be32_to_cpu(data->rootErrorStatus),
455 be32_to_cpu(data->uncorrErrorStatus),
456 be32_to_cpu(data->corrErrorStatus));
457 if (data->tlpHdr1 || data->tlpHdr2 ||
458 data->tlpHdr3 || data->tlpHdr4)
459 pr_info("RootErrLog: %08x %08x %08x %08x\n",
460 be32_to_cpu(data->tlpHdr1),
461 be32_to_cpu(data->tlpHdr2),
462 be32_to_cpu(data->tlpHdr3),
463 be32_to_cpu(data->tlpHdr4));
465 pr_info("sourceId: %08x\n", be32_to_cpu(data->sourceId));
467 pr_info("nFir: %016llx %016llx %016llx\n",
468 be64_to_cpu(data->nFir),
469 be64_to_cpu(data->nFirMask),
470 be64_to_cpu(data->nFirWOF));
471 if (data->phbPlssr || data->phbCsr)
472 pr_info("PhbSts: %016llx %016llx\n",
473 be64_to_cpu(data->phbPlssr),
474 be64_to_cpu(data->phbCsr));
476 pr_info("Lem: %016llx %016llx %016llx\n",
477 be64_to_cpu(data->lemFir),
478 be64_to_cpu(data->lemErrorMask),
479 be64_to_cpu(data->lemWOF));
480 if (data->phbErrorStatus)
481 pr_info("PhbErr: %016llx %016llx %016llx %016llx\n",
482 be64_to_cpu(data->phbErrorStatus),
483 be64_to_cpu(data->phbFirstErrorStatus),
484 be64_to_cpu(data->phbErrorLog0),
485 be64_to_cpu(data->phbErrorLog1));
486 if (data->phbTxeErrorStatus)
487 pr_info("PhbTxeErr: %016llx %016llx %016llx %016llx\n",
488 be64_to_cpu(data->phbTxeErrorStatus),
489 be64_to_cpu(data->phbTxeFirstErrorStatus),
490 be64_to_cpu(data->phbTxeErrorLog0),
491 be64_to_cpu(data->phbTxeErrorLog1));
492 if (data->phbRxeArbErrorStatus)
493 pr_info("RxeArbErr: %016llx %016llx %016llx %016llx\n",
494 be64_to_cpu(data->phbRxeArbErrorStatus),
495 be64_to_cpu(data->phbRxeArbFirstErrorStatus),
496 be64_to_cpu(data->phbRxeArbErrorLog0),
497 be64_to_cpu(data->phbRxeArbErrorLog1));
498 if (data->phbRxeMrgErrorStatus)
499 pr_info("RxeMrgErr: %016llx %016llx %016llx %016llx\n",
500 be64_to_cpu(data->phbRxeMrgErrorStatus),
501 be64_to_cpu(data->phbRxeMrgFirstErrorStatus),
502 be64_to_cpu(data->phbRxeMrgErrorLog0),
503 be64_to_cpu(data->phbRxeMrgErrorLog1));
504 if (data->phbRxeTceErrorStatus)
505 pr_info("RxeTceErr: %016llx %016llx %016llx %016llx\n",
506 be64_to_cpu(data->phbRxeTceErrorStatus),
507 be64_to_cpu(data->phbRxeTceFirstErrorStatus),
508 be64_to_cpu(data->phbRxeTceErrorLog0),
509 be64_to_cpu(data->phbRxeTceErrorLog1));
511 if (data->phbPblErrorStatus)
512 pr_info("PblErr: %016llx %016llx %016llx %016llx\n",
513 be64_to_cpu(data->phbPblErrorStatus),
514 be64_to_cpu(data->phbPblFirstErrorStatus),
515 be64_to_cpu(data->phbPblErrorLog0),
516 be64_to_cpu(data->phbPblErrorLog1));
517 if (data->phbPcieDlpErrorStatus)
518 pr_info("PcieDlp: %016llx %016llx %016llx\n",
519 be64_to_cpu(data->phbPcieDlpErrorLog1),
520 be64_to_cpu(data->phbPcieDlpErrorLog2),
521 be64_to_cpu(data->phbPcieDlpErrorStatus));
522 if (data->phbRegbErrorStatus)
523 pr_info("RegbErr: %016llx %016llx %016llx %016llx\n",
524 be64_to_cpu(data->phbRegbErrorStatus),
525 be64_to_cpu(data->phbRegbFirstErrorStatus),
526 be64_to_cpu(data->phbRegbErrorLog0),
527 be64_to_cpu(data->phbRegbErrorLog1));
530 pnv_pci_dump_pest(data->pestA, data->pestB, OPAL_PHB4_NUM_PEST_REGS);
533 void pnv_pci_dump_phb_diag_data(struct pci_controller *hose,
534 unsigned char *log_buff)
536 struct OpalIoPhbErrorCommon *common;
538 if (!hose || !log_buff)
541 common = (struct OpalIoPhbErrorCommon *)log_buff;
542 switch (be32_to_cpu(common->ioType)) {
543 case OPAL_PHB_ERROR_DATA_TYPE_P7IOC:
544 pnv_pci_dump_p7ioc_diag_data(hose, common);
546 case OPAL_PHB_ERROR_DATA_TYPE_PHB3:
547 pnv_pci_dump_phb3_diag_data(hose, common);
549 case OPAL_PHB_ERROR_DATA_TYPE_PHB4:
550 pnv_pci_dump_phb4_diag_data(hose, common);
553 pr_warn("%s: Unrecognized ioType %d\n",
554 __func__, be32_to_cpu(common->ioType));
558 static void pnv_pci_handle_eeh_config(struct pnv_phb *phb, u32 pe_no)
560 unsigned long flags, rc;
561 int has_diag, ret = 0;
563 spin_lock_irqsave(&phb->lock, flags);
565 /* Fetch PHB diag-data */
566 rc = opal_pci_get_phb_diag_data2(phb->opal_id, phb->diag_data,
567 phb->diag_data_size);
568 has_diag = (rc == OPAL_SUCCESS);
570 /* If PHB supports compound PE, to handle it */
571 if (phb->unfreeze_pe) {
572 ret = phb->unfreeze_pe(phb,
574 OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
576 rc = opal_pci_eeh_freeze_clear(phb->opal_id,
578 OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
580 pr_warn("%s: Failure %ld clearing frozen "
582 __func__, rc, phb->hose->global_number,
589 * For now, let's only display the diag buffer when we fail to clear
590 * the EEH status. We'll do more sensible things later when we have
591 * proper EEH support. We need to make sure we don't pollute ourselves
592 * with the normal errors generated when probing empty slots
595 pnv_pci_dump_phb_diag_data(phb->hose, phb->diag_data);
597 spin_unlock_irqrestore(&phb->lock, flags);
600 static void pnv_pci_config_check_eeh(struct pci_dn *pdn)
602 struct pnv_phb *phb = pdn->phb->private_data;
609 * Get the PE#. During the PCI probe stage, we might not
610 * setup that yet. So all ER errors should be mapped to
613 pe_no = pdn->pe_number;
614 if (pe_no == IODA_INVALID_PE) {
615 pe_no = phb->ioda.reserved_pe_idx;
619 * Fetch frozen state. If the PHB support compound PE,
620 * we need handle that case.
622 if (phb->get_pe_state) {
623 fstate = phb->get_pe_state(phb, pe_no);
625 rc = opal_pci_eeh_freeze_status(phb->opal_id,
631 pr_warn("%s: Failure %lld getting PHB#%x-PE#%x state\n",
632 __func__, rc, phb->hose->global_number, pe_no);
637 pr_devel(" -> EEH check, bdfn=%04x PE#%x fstate=%x\n",
638 (pdn->busno << 8) | (pdn->devfn), pe_no, fstate);
640 /* Clear the frozen state if applicable */
641 if (fstate == OPAL_EEH_STOPPED_MMIO_FREEZE ||
642 fstate == OPAL_EEH_STOPPED_DMA_FREEZE ||
643 fstate == OPAL_EEH_STOPPED_MMIO_DMA_FREEZE) {
645 * If PHB supports compound PE, freeze it for
649 phb->freeze_pe(phb, pe_no);
651 pnv_pci_handle_eeh_config(phb, pe_no);
655 int pnv_pci_cfg_read(struct pci_dn *pdn,
656 int where, int size, u32 *val)
658 struct pnv_phb *phb = pdn->phb->private_data;
659 u32 bdfn = (pdn->busno << 8) | pdn->devfn;
665 rc = opal_pci_config_read_byte(phb->opal_id, bdfn, where, &v8);
666 *val = (rc == OPAL_SUCCESS) ? v8 : 0xff;
671 rc = opal_pci_config_read_half_word(phb->opal_id, bdfn, where,
673 *val = (rc == OPAL_SUCCESS) ? be16_to_cpu(v16) : 0xffff;
678 rc = opal_pci_config_read_word(phb->opal_id, bdfn, where, &v32);
679 *val = (rc == OPAL_SUCCESS) ? be32_to_cpu(v32) : 0xffffffff;
683 return PCIBIOS_FUNC_NOT_SUPPORTED;
686 pr_devel("%s: bus: %x devfn: %x +%x/%x -> %08x\n",
687 __func__, pdn->busno, pdn->devfn, where, size, *val);
688 return PCIBIOS_SUCCESSFUL;
691 int pnv_pci_cfg_write(struct pci_dn *pdn,
692 int where, int size, u32 val)
694 struct pnv_phb *phb = pdn->phb->private_data;
695 u32 bdfn = (pdn->busno << 8) | pdn->devfn;
697 pr_devel("%s: bus: %x devfn: %x +%x/%x -> %08x\n",
698 __func__, pdn->busno, pdn->devfn, where, size, val);
701 opal_pci_config_write_byte(phb->opal_id, bdfn, where, val);
704 opal_pci_config_write_half_word(phb->opal_id, bdfn, where, val);
707 opal_pci_config_write_word(phb->opal_id, bdfn, where, val);
710 return PCIBIOS_FUNC_NOT_SUPPORTED;
713 return PCIBIOS_SUCCESSFUL;
717 static bool pnv_pci_cfg_check(struct pci_dn *pdn)
719 struct eeh_dev *edev = NULL;
720 struct pnv_phb *phb = pdn->phb->private_data;
722 /* EEH not enabled ? */
723 if (!(phb->flags & PNV_PHB_FLAG_EEH))
726 /* PE reset or device removed ? */
730 (edev->pe->state & EEH_PE_CFG_BLOCKED))
733 if (edev->mode & EEH_DEV_REMOVED)
740 static inline pnv_pci_cfg_check(struct pci_dn *pdn)
744 #endif /* CONFIG_EEH */
746 static int pnv_pci_read_config(struct pci_bus *bus,
748 int where, int size, u32 *val)
755 pdn = pci_get_pdn_by_devfn(bus, devfn);
757 return PCIBIOS_DEVICE_NOT_FOUND;
759 if (!pnv_pci_cfg_check(pdn))
760 return PCIBIOS_DEVICE_NOT_FOUND;
762 ret = pnv_pci_cfg_read(pdn, where, size, val);
763 phb = pdn->phb->private_data;
764 if (phb->flags & PNV_PHB_FLAG_EEH && pdn->edev) {
765 if (*val == EEH_IO_ERROR_VALUE(size) &&
766 eeh_dev_check_failure(pdn->edev))
767 return PCIBIOS_DEVICE_NOT_FOUND;
769 pnv_pci_config_check_eeh(pdn);
775 static int pnv_pci_write_config(struct pci_bus *bus,
777 int where, int size, u32 val)
783 pdn = pci_get_pdn_by_devfn(bus, devfn);
785 return PCIBIOS_DEVICE_NOT_FOUND;
787 if (!pnv_pci_cfg_check(pdn))
788 return PCIBIOS_DEVICE_NOT_FOUND;
790 ret = pnv_pci_cfg_write(pdn, where, size, val);
791 phb = pdn->phb->private_data;
792 if (!(phb->flags & PNV_PHB_FLAG_EEH))
793 pnv_pci_config_check_eeh(pdn);
798 struct pci_ops pnv_pci_ops = {
799 .read = pnv_pci_read_config,
800 .write = pnv_pci_write_config,
803 struct iommu_table *pnv_pci_table_alloc(int nid)
805 struct iommu_table *tbl;
807 tbl = kzalloc_node(sizeof(struct iommu_table), GFP_KERNEL, nid);
811 INIT_LIST_HEAD_RCU(&tbl->it_group_list);
812 kref_init(&tbl->it_kref);
817 void pnv_pci_dma_dev_setup(struct pci_dev *pdev)
819 struct pci_controller *hose = pci_bus_to_host(pdev->bus);
820 struct pnv_phb *phb = hose->private_data;
821 #ifdef CONFIG_PCI_IOV
822 struct pnv_ioda_pe *pe;
825 /* Fix the VF pdn PE number */
826 if (pdev->is_virtfn) {
827 pdn = pci_get_pdn(pdev);
828 WARN_ON(pdn->pe_number != IODA_INVALID_PE);
829 list_for_each_entry(pe, &phb->ioda.pe_list, list) {
830 if (pe->rid == ((pdev->bus->number << 8) |
831 (pdev->devfn & 0xff))) {
832 pdn->pe_number = pe->pe_number;
838 #endif /* CONFIG_PCI_IOV */
840 if (phb && phb->dma_dev_setup)
841 phb->dma_dev_setup(phb, pdev);
844 void pnv_pci_dma_bus_setup(struct pci_bus *bus)
846 struct pci_controller *hose = bus->sysdata;
847 struct pnv_phb *phb = hose->private_data;
848 struct pnv_ioda_pe *pe;
850 list_for_each_entry(pe, &phb->ioda.pe_list, list) {
851 if (!(pe->flags & (PNV_IODA_PE_BUS | PNV_IODA_PE_BUS_ALL)))
857 if (bus->number == ((pe->rid >> 8) & 0xFF)) {
864 int pnv_pci_set_p2p(struct pci_dev *initiator, struct pci_dev *target, u64 desc)
866 struct pci_controller *hose;
867 struct pnv_phb *phb_init, *phb_target;
868 struct pnv_ioda_pe *pe_init;
871 if (!opal_check_token(OPAL_PCI_SET_P2P))
874 hose = pci_bus_to_host(initiator->bus);
875 phb_init = hose->private_data;
877 hose = pci_bus_to_host(target->bus);
878 phb_target = hose->private_data;
880 pe_init = pnv_ioda_get_pe(initiator);
885 * Configuring the initiator's PHB requires to adjust its
886 * TVE#1 setting. Since the same device can be an initiator
887 * several times for different target devices, we need to keep
888 * a reference count to know when we can restore the default
889 * bypass setting on its TVE#1 when disabling. Opal is not
890 * tracking PE states, so we add a reference count on the PE
893 * For the target, the configuration is per PHB, so we keep a
894 * target reference count on the PHB.
896 mutex_lock(&p2p_mutex);
898 if (desc & OPAL_PCI_P2P_ENABLE) {
899 /* always go to opal to validate the configuration */
900 rc = opal_pci_set_p2p(phb_init->opal_id, phb_target->opal_id,
901 desc, pe_init->pe_number);
903 if (rc != OPAL_SUCCESS) {
908 pe_init->p2p_initiator_count++;
909 phb_target->p2p_target_count++;
911 if (!pe_init->p2p_initiator_count ||
912 !phb_target->p2p_target_count) {
917 if (--pe_init->p2p_initiator_count == 0)
918 pnv_pci_ioda2_set_bypass(pe_init, true);
920 if (--phb_target->p2p_target_count == 0) {
921 rc = opal_pci_set_p2p(phb_init->opal_id,
922 phb_target->opal_id, desc,
924 if (rc != OPAL_SUCCESS) {
932 mutex_unlock(&p2p_mutex);
935 EXPORT_SYMBOL_GPL(pnv_pci_set_p2p);
937 struct device_node *pnv_pci_get_phb_node(struct pci_dev *dev)
939 struct pci_controller *hose = pci_bus_to_host(dev->bus);
941 return of_node_get(hose->dn);
943 EXPORT_SYMBOL(pnv_pci_get_phb_node);
945 int pnv_pci_enable_tunnel(struct pci_dev *dev, u64 *asnind)
947 struct device_node *np;
949 struct pnv_ioda_pe *pe;
953 if (!radix_enabled())
956 if (!(np = pnv_pci_get_phb_node(dev)))
959 prop = of_get_property(np, "ibm,phb-indications", NULL);
962 if (!prop || !prop[1])
965 *asnind = (u64)be32_to_cpu(prop[1]);
966 pe = pnv_ioda_get_pe(dev);
970 /* Increase real window size to accept as_notify messages. */
971 window_id = (pe->pe_number << 1 ) + 1;
972 rc = opal_pci_map_pe_dma_window_real(pe->phb->opal_id, pe->pe_number,
973 window_id, pe->tce_bypass_base,
975 return opal_error_code(rc);
977 EXPORT_SYMBOL_GPL(pnv_pci_enable_tunnel);
979 int pnv_pci_disable_tunnel(struct pci_dev *dev)
981 struct pnv_ioda_pe *pe;
983 pe = pnv_ioda_get_pe(dev);
987 /* Restore default real window size. */
988 pnv_pci_ioda2_set_bypass(pe, true);
991 EXPORT_SYMBOL_GPL(pnv_pci_disable_tunnel);
993 int pnv_pci_set_tunnel_bar(struct pci_dev *dev, u64 addr, int enable)
996 struct pci_controller *hose;
1001 if (!opal_check_token(OPAL_PCI_GET_PBCQ_TUNNEL_BAR))
1003 if (!opal_check_token(OPAL_PCI_SET_PBCQ_TUNNEL_BAR))
1006 hose = pci_bus_to_host(dev->bus);
1007 phb = hose->private_data;
1009 mutex_lock(&tunnel_mutex);
1010 rc = opal_pci_get_pbcq_tunnel_bar(phb->opal_id, &val);
1011 if (rc != OPAL_SUCCESS) {
1015 tunnel_bar = be64_to_cpu(val);
1018 * Only one device per PHB can use atomics.
1019 * Our policy is first-come, first-served.
1022 if (tunnel_bar != addr)
1025 rc = 0; /* Setting same address twice is ok */
1030 * The device that owns atomics and wants to release
1031 * them must pass the same address with enable == 0.
1033 if (tunnel_bar != addr) {
1039 rc = opal_pci_set_pbcq_tunnel_bar(phb->opal_id, addr);
1040 rc = opal_error_code(rc);
1042 mutex_unlock(&tunnel_mutex);
1045 EXPORT_SYMBOL_GPL(pnv_pci_set_tunnel_bar);
1047 #ifdef CONFIG_PPC64 /* for thread.tidr */
1048 int pnv_pci_get_as_notify_info(struct task_struct *task, u32 *lpid, u32 *pid,
1051 struct mm_struct *mm = NULL;
1056 mm = get_task_mm(task);
1060 *pid = mm->context.id;
1063 *tid = task->thread.tidr;
1064 *lpid = mfspr(SPRN_LPID);
1067 EXPORT_SYMBOL_GPL(pnv_pci_get_as_notify_info);
1070 void pnv_pci_shutdown(void)
1072 struct pci_controller *hose;
1074 list_for_each_entry(hose, &hose_list, list_node)
1075 if (hose->controller_ops.shutdown)
1076 hose->controller_ops.shutdown(hose);
1079 /* Fixup wrong class code in p7ioc and p8 root complex */
1080 static void pnv_p7ioc_rc_quirk(struct pci_dev *dev)
1082 dev->class = PCI_CLASS_BRIDGE_PCI << 8;
1084 DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_IBM, 0x3b9, pnv_p7ioc_rc_quirk);
1086 void __init pnv_pci_init(void)
1088 struct device_node *np;
1090 pci_add_flags(PCI_CAN_SKIP_ISA_ALIGN);
1092 /* If we don't have OPAL, eg. in sim, just skip PCI probe */
1093 if (!firmware_has_feature(FW_FEATURE_OPAL))
1096 /* Look for IODA IO-Hubs. */
1097 for_each_compatible_node(np, NULL, "ibm,ioda-hub") {
1098 pnv_pci_init_ioda_hub(np);
1101 /* Look for ioda2 built-in PHB3's */
1102 for_each_compatible_node(np, NULL, "ibm,ioda2-phb")
1103 pnv_pci_init_ioda2_phb(np);
1105 /* Look for ioda3 built-in PHB4's, we treat them as IODA2 */
1106 for_each_compatible_node(np, NULL, "ibm,ioda3-phb")
1107 pnv_pci_init_ioda2_phb(np);
1109 /* Look for NPU PHBs */
1110 for_each_compatible_node(np, NULL, "ibm,ioda2-npu-phb")
1111 pnv_pci_init_npu_phb(np);
1114 * Look for NPU2 PHBs which we treat mostly as NPU PHBs with
1115 * the exception of TCE kill which requires an OPAL call.
1117 for_each_compatible_node(np, NULL, "ibm,ioda2-npu2-phb")
1118 pnv_pci_init_npu_phb(np);
1120 /* Look for NPU2 OpenCAPI PHBs */
1121 for_each_compatible_node(np, NULL, "ibm,ioda2-npu2-opencapi-phb")
1122 pnv_pci_init_npu2_opencapi_phb(np);
1124 /* Configure IOMMU DMA hooks */
1125 set_pci_dma_ops(&dma_iommu_ops);
1128 static int pnv_tce_iommu_bus_notifier(struct notifier_block *nb,
1129 unsigned long action, void *data)
1131 struct device *dev = data;
1132 struct pci_dev *pdev;
1134 struct pnv_ioda_pe *pe;
1135 struct pci_controller *hose;
1136 struct pnv_phb *phb;
1139 case BUS_NOTIFY_ADD_DEVICE:
1140 pdev = to_pci_dev(dev);
1141 pdn = pci_get_pdn(pdev);
1142 hose = pci_bus_to_host(pdev->bus);
1143 phb = hose->private_data;
1146 if (!pdn || pdn->pe_number == IODA_INVALID_PE || !phb)
1149 pe = &phb->ioda.pe_array[pdn->pe_number];
1150 iommu_add_device(&pe->table_group, dev);
1152 case BUS_NOTIFY_DEL_DEVICE:
1153 iommu_del_device(dev);
1160 static struct notifier_block pnv_tce_iommu_bus_nb = {
1161 .notifier_call = pnv_tce_iommu_bus_notifier,
1164 static int __init pnv_tce_iommu_bus_notifier_init(void)
1166 bus_register_notifier(&pci_bus_type, &pnv_tce_iommu_bus_nb);
1169 machine_subsys_initcall_sync(powernv, pnv_tce_iommu_bus_notifier_init);