2 * Support PCI/PCIe on PowerNV platforms
4 * Currently supports only P5IOC2
6 * Copyright 2011 Benjamin Herrenschmidt, IBM Corp.
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
14 #include <linux/kernel.h>
15 #include <linux/pci.h>
16 #include <linux/delay.h>
17 #include <linux/string.h>
18 #include <linux/init.h>
19 #include <linux/irq.h>
21 #include <linux/msi.h>
22 #include <linux/iommu.h>
24 #include <asm/sections.h>
27 #include <asm/pci-bridge.h>
28 #include <asm/machdep.h>
29 #include <asm/msi_bitmap.h>
30 #include <asm/ppc-pci.h>
32 #include <asm/iommu.h>
34 #include <asm/firmware.h>
35 #include <asm/eeh_event.h>
42 #define PCI_RESET_DELAY_US 3000000
44 #define cfg_dbg(fmt...) do { } while(0)
45 //#define cfg_dbg(fmt...) printk(fmt)
48 static int pnv_setup_msi_irqs(struct pci_dev *pdev, int nvec, int type)
50 struct pci_controller *hose = pci_bus_to_host(pdev->bus);
51 struct pnv_phb *phb = hose->private_data;
52 struct msi_desc *entry;
58 if (WARN_ON(!phb) || !phb->msi_bmp.bitmap)
61 if (pdev->no_64bit_msi && !phb->msi32_support)
64 list_for_each_entry(entry, &pdev->msi_list, list) {
65 if (!entry->msi_attrib.is_64 && !phb->msi32_support) {
66 pr_warn("%s: Supports only 64-bit MSIs\n",
70 hwirq = msi_bitmap_alloc_hwirqs(&phb->msi_bmp, 1);
72 pr_warn("%s: Failed to find a free MSI\n",
76 virq = irq_create_mapping(NULL, phb->msi_base + hwirq);
78 pr_warn("%s: Failed to map MSI to linux irq\n",
80 msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq, 1);
83 rc = phb->msi_setup(phb, pdev, phb->msi_base + hwirq,
84 virq, entry->msi_attrib.is_64, &msg);
86 pr_warn("%s: Failed to setup MSI\n", pci_name(pdev));
87 irq_dispose_mapping(virq);
88 msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq, 1);
91 irq_set_msi_desc(virq, entry);
92 pci_write_msi_msg(virq, &msg);
97 static void pnv_teardown_msi_irqs(struct pci_dev *pdev)
99 struct pci_controller *hose = pci_bus_to_host(pdev->bus);
100 struct pnv_phb *phb = hose->private_data;
101 struct msi_desc *entry;
106 list_for_each_entry(entry, &pdev->msi_list, list) {
107 if (entry->irq == NO_IRQ)
109 irq_set_msi_desc(entry->irq, NULL);
110 msi_bitmap_free_hwirqs(&phb->msi_bmp,
111 virq_to_hw(entry->irq) - phb->msi_base, 1);
112 irq_dispose_mapping(entry->irq);
115 #endif /* CONFIG_PCI_MSI */
117 static void pnv_pci_dump_p7ioc_diag_data(struct pci_controller *hose,
118 struct OpalIoPhbErrorCommon *common)
120 struct OpalIoP7IOCPhbErrorData *data;
123 data = (struct OpalIoP7IOCPhbErrorData *)common;
124 pr_info("P7IOC PHB#%d Diag-data (Version: %d)\n",
125 hose->global_number, be32_to_cpu(common->version));
128 pr_info("brdgCtl: %08x\n",
129 be32_to_cpu(data->brdgCtl));
130 if (data->portStatusReg || data->rootCmplxStatus ||
131 data->busAgentStatus)
132 pr_info("UtlSts: %08x %08x %08x\n",
133 be32_to_cpu(data->portStatusReg),
134 be32_to_cpu(data->rootCmplxStatus),
135 be32_to_cpu(data->busAgentStatus));
136 if (data->deviceStatus || data->slotStatus ||
137 data->linkStatus || data->devCmdStatus ||
139 pr_info("RootSts: %08x %08x %08x %08x %08x\n",
140 be32_to_cpu(data->deviceStatus),
141 be32_to_cpu(data->slotStatus),
142 be32_to_cpu(data->linkStatus),
143 be32_to_cpu(data->devCmdStatus),
144 be32_to_cpu(data->devSecStatus));
145 if (data->rootErrorStatus || data->uncorrErrorStatus ||
146 data->corrErrorStatus)
147 pr_info("RootErrSts: %08x %08x %08x\n",
148 be32_to_cpu(data->rootErrorStatus),
149 be32_to_cpu(data->uncorrErrorStatus),
150 be32_to_cpu(data->corrErrorStatus));
151 if (data->tlpHdr1 || data->tlpHdr2 ||
152 data->tlpHdr3 || data->tlpHdr4)
153 pr_info("RootErrLog: %08x %08x %08x %08x\n",
154 be32_to_cpu(data->tlpHdr1),
155 be32_to_cpu(data->tlpHdr2),
156 be32_to_cpu(data->tlpHdr3),
157 be32_to_cpu(data->tlpHdr4));
158 if (data->sourceId || data->errorClass ||
160 pr_info("RootErrLog1: %08x %016llx %016llx\n",
161 be32_to_cpu(data->sourceId),
162 be64_to_cpu(data->errorClass),
163 be64_to_cpu(data->correlator));
164 if (data->p7iocPlssr || data->p7iocCsr)
165 pr_info("PhbSts: %016llx %016llx\n",
166 be64_to_cpu(data->p7iocPlssr),
167 be64_to_cpu(data->p7iocCsr));
169 pr_info("Lem: %016llx %016llx %016llx\n",
170 be64_to_cpu(data->lemFir),
171 be64_to_cpu(data->lemErrorMask),
172 be64_to_cpu(data->lemWOF));
173 if (data->phbErrorStatus)
174 pr_info("PhbErr: %016llx %016llx %016llx %016llx\n",
175 be64_to_cpu(data->phbErrorStatus),
176 be64_to_cpu(data->phbFirstErrorStatus),
177 be64_to_cpu(data->phbErrorLog0),
178 be64_to_cpu(data->phbErrorLog1));
179 if (data->mmioErrorStatus)
180 pr_info("OutErr: %016llx %016llx %016llx %016llx\n",
181 be64_to_cpu(data->mmioErrorStatus),
182 be64_to_cpu(data->mmioFirstErrorStatus),
183 be64_to_cpu(data->mmioErrorLog0),
184 be64_to_cpu(data->mmioErrorLog1));
185 if (data->dma0ErrorStatus)
186 pr_info("InAErr: %016llx %016llx %016llx %016llx\n",
187 be64_to_cpu(data->dma0ErrorStatus),
188 be64_to_cpu(data->dma0FirstErrorStatus),
189 be64_to_cpu(data->dma0ErrorLog0),
190 be64_to_cpu(data->dma0ErrorLog1));
191 if (data->dma1ErrorStatus)
192 pr_info("InBErr: %016llx %016llx %016llx %016llx\n",
193 be64_to_cpu(data->dma1ErrorStatus),
194 be64_to_cpu(data->dma1FirstErrorStatus),
195 be64_to_cpu(data->dma1ErrorLog0),
196 be64_to_cpu(data->dma1ErrorLog1));
198 for (i = 0; i < OPAL_P7IOC_NUM_PEST_REGS; i++) {
199 if ((data->pestA[i] >> 63) == 0 &&
200 (data->pestB[i] >> 63) == 0)
203 pr_info("PE[%3d] A/B: %016llx %016llx\n",
204 i, be64_to_cpu(data->pestA[i]),
205 be64_to_cpu(data->pestB[i]));
209 static void pnv_pci_dump_phb3_diag_data(struct pci_controller *hose,
210 struct OpalIoPhbErrorCommon *common)
212 struct OpalIoPhb3ErrorData *data;
215 data = (struct OpalIoPhb3ErrorData*)common;
216 pr_info("PHB3 PHB#%d Diag-data (Version: %d)\n",
217 hose->global_number, be32_to_cpu(common->version));
219 pr_info("brdgCtl: %08x\n",
220 be32_to_cpu(data->brdgCtl));
221 if (data->portStatusReg || data->rootCmplxStatus ||
222 data->busAgentStatus)
223 pr_info("UtlSts: %08x %08x %08x\n",
224 be32_to_cpu(data->portStatusReg),
225 be32_to_cpu(data->rootCmplxStatus),
226 be32_to_cpu(data->busAgentStatus));
227 if (data->deviceStatus || data->slotStatus ||
228 data->linkStatus || data->devCmdStatus ||
230 pr_info("RootSts: %08x %08x %08x %08x %08x\n",
231 be32_to_cpu(data->deviceStatus),
232 be32_to_cpu(data->slotStatus),
233 be32_to_cpu(data->linkStatus),
234 be32_to_cpu(data->devCmdStatus),
235 be32_to_cpu(data->devSecStatus));
236 if (data->rootErrorStatus || data->uncorrErrorStatus ||
237 data->corrErrorStatus)
238 pr_info("RootErrSts: %08x %08x %08x\n",
239 be32_to_cpu(data->rootErrorStatus),
240 be32_to_cpu(data->uncorrErrorStatus),
241 be32_to_cpu(data->corrErrorStatus));
242 if (data->tlpHdr1 || data->tlpHdr2 ||
243 data->tlpHdr3 || data->tlpHdr4)
244 pr_info("RootErrLog: %08x %08x %08x %08x\n",
245 be32_to_cpu(data->tlpHdr1),
246 be32_to_cpu(data->tlpHdr2),
247 be32_to_cpu(data->tlpHdr3),
248 be32_to_cpu(data->tlpHdr4));
249 if (data->sourceId || data->errorClass ||
251 pr_info("RootErrLog1: %08x %016llx %016llx\n",
252 be32_to_cpu(data->sourceId),
253 be64_to_cpu(data->errorClass),
254 be64_to_cpu(data->correlator));
256 pr_info("nFir: %016llx %016llx %016llx\n",
257 be64_to_cpu(data->nFir),
258 be64_to_cpu(data->nFirMask),
259 be64_to_cpu(data->nFirWOF));
260 if (data->phbPlssr || data->phbCsr)
261 pr_info("PhbSts: %016llx %016llx\n",
262 be64_to_cpu(data->phbPlssr),
263 be64_to_cpu(data->phbCsr));
265 pr_info("Lem: %016llx %016llx %016llx\n",
266 be64_to_cpu(data->lemFir),
267 be64_to_cpu(data->lemErrorMask),
268 be64_to_cpu(data->lemWOF));
269 if (data->phbErrorStatus)
270 pr_info("PhbErr: %016llx %016llx %016llx %016llx\n",
271 be64_to_cpu(data->phbErrorStatus),
272 be64_to_cpu(data->phbFirstErrorStatus),
273 be64_to_cpu(data->phbErrorLog0),
274 be64_to_cpu(data->phbErrorLog1));
275 if (data->mmioErrorStatus)
276 pr_info("OutErr: %016llx %016llx %016llx %016llx\n",
277 be64_to_cpu(data->mmioErrorStatus),
278 be64_to_cpu(data->mmioFirstErrorStatus),
279 be64_to_cpu(data->mmioErrorLog0),
280 be64_to_cpu(data->mmioErrorLog1));
281 if (data->dma0ErrorStatus)
282 pr_info("InAErr: %016llx %016llx %016llx %016llx\n",
283 be64_to_cpu(data->dma0ErrorStatus),
284 be64_to_cpu(data->dma0FirstErrorStatus),
285 be64_to_cpu(data->dma0ErrorLog0),
286 be64_to_cpu(data->dma0ErrorLog1));
287 if (data->dma1ErrorStatus)
288 pr_info("InBErr: %016llx %016llx %016llx %016llx\n",
289 be64_to_cpu(data->dma1ErrorStatus),
290 be64_to_cpu(data->dma1FirstErrorStatus),
291 be64_to_cpu(data->dma1ErrorLog0),
292 be64_to_cpu(data->dma1ErrorLog1));
294 for (i = 0; i < OPAL_PHB3_NUM_PEST_REGS; i++) {
295 if ((be64_to_cpu(data->pestA[i]) >> 63) == 0 &&
296 (be64_to_cpu(data->pestB[i]) >> 63) == 0)
299 pr_info("PE[%3d] A/B: %016llx %016llx\n",
300 i, be64_to_cpu(data->pestA[i]),
301 be64_to_cpu(data->pestB[i]));
305 void pnv_pci_dump_phb_diag_data(struct pci_controller *hose,
306 unsigned char *log_buff)
308 struct OpalIoPhbErrorCommon *common;
310 if (!hose || !log_buff)
313 common = (struct OpalIoPhbErrorCommon *)log_buff;
314 switch (be32_to_cpu(common->ioType)) {
315 case OPAL_PHB_ERROR_DATA_TYPE_P7IOC:
316 pnv_pci_dump_p7ioc_diag_data(hose, common);
318 case OPAL_PHB_ERROR_DATA_TYPE_PHB3:
319 pnv_pci_dump_phb3_diag_data(hose, common);
322 pr_warn("%s: Unrecognized ioType %d\n",
323 __func__, be32_to_cpu(common->ioType));
327 static void pnv_pci_handle_eeh_config(struct pnv_phb *phb, u32 pe_no)
329 unsigned long flags, rc;
330 int has_diag, ret = 0;
332 spin_lock_irqsave(&phb->lock, flags);
334 /* Fetch PHB diag-data */
335 rc = opal_pci_get_phb_diag_data2(phb->opal_id, phb->diag.blob,
336 PNV_PCI_DIAG_BUF_SIZE);
337 has_diag = (rc == OPAL_SUCCESS);
339 /* If PHB supports compound PE, to handle it */
340 if (phb->unfreeze_pe) {
341 ret = phb->unfreeze_pe(phb,
343 OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
345 rc = opal_pci_eeh_freeze_clear(phb->opal_id,
347 OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
349 pr_warn("%s: Failure %ld clearing frozen "
351 __func__, rc, phb->hose->global_number,
358 * For now, let's only display the diag buffer when we fail to clear
359 * the EEH status. We'll do more sensible things later when we have
360 * proper EEH support. We need to make sure we don't pollute ourselves
361 * with the normal errors generated when probing empty slots
364 pnv_pci_dump_phb_diag_data(phb->hose, phb->diag.blob);
366 spin_unlock_irqrestore(&phb->lock, flags);
369 static void pnv_pci_config_check_eeh(struct pnv_phb *phb,
370 struct device_node *dn)
378 * Get the PE#. During the PCI probe stage, we might not
379 * setup that yet. So all ER errors should be mapped to
382 pe_no = PCI_DN(dn)->pe_number;
383 if (pe_no == IODA_INVALID_PE) {
384 if (phb->type == PNV_PHB_P5IOC2)
387 pe_no = phb->ioda.reserved_pe;
391 * Fetch frozen state. If the PHB support compound PE,
392 * we need handle that case.
394 if (phb->get_pe_state) {
395 fstate = phb->get_pe_state(phb, pe_no);
397 rc = opal_pci_eeh_freeze_status(phb->opal_id,
403 pr_warn("%s: Failure %lld getting PHB#%x-PE#%x state\n",
404 __func__, rc, phb->hose->global_number, pe_no);
409 cfg_dbg(" -> EEH check, bdfn=%04x PE#%d fstate=%x\n",
410 (PCI_DN(dn)->busno << 8) | (PCI_DN(dn)->devfn),
413 /* Clear the frozen state if applicable */
414 if (fstate == OPAL_EEH_STOPPED_MMIO_FREEZE ||
415 fstate == OPAL_EEH_STOPPED_DMA_FREEZE ||
416 fstate == OPAL_EEH_STOPPED_MMIO_DMA_FREEZE) {
418 * If PHB supports compound PE, freeze it for
422 phb->freeze_pe(phb, pe_no);
424 pnv_pci_handle_eeh_config(phb, pe_no);
428 int pnv_pci_cfg_read(struct device_node *dn,
429 int where, int size, u32 *val)
431 struct pci_dn *pdn = PCI_DN(dn);
432 struct pnv_phb *phb = pdn->phb->private_data;
433 u32 bdfn = (pdn->busno << 8) | pdn->devfn;
439 rc = opal_pci_config_read_byte(phb->opal_id, bdfn, where, &v8);
440 *val = (rc == OPAL_SUCCESS) ? v8 : 0xff;
445 rc = opal_pci_config_read_half_word(phb->opal_id, bdfn, where,
447 *val = (rc == OPAL_SUCCESS) ? be16_to_cpu(v16) : 0xffff;
452 rc = opal_pci_config_read_word(phb->opal_id, bdfn, where, &v32);
453 *val = (rc == OPAL_SUCCESS) ? be32_to_cpu(v32) : 0xffffffff;
457 return PCIBIOS_FUNC_NOT_SUPPORTED;
460 cfg_dbg("%s: bus: %x devfn: %x +%x/%x -> %08x\n",
461 __func__, pdn->busno, pdn->devfn, where, size, *val);
462 return PCIBIOS_SUCCESSFUL;
465 int pnv_pci_cfg_write(struct device_node *dn,
466 int where, int size, u32 val)
468 struct pci_dn *pdn = PCI_DN(dn);
469 struct pnv_phb *phb = pdn->phb->private_data;
470 u32 bdfn = (pdn->busno << 8) | pdn->devfn;
472 cfg_dbg("%s: bus: %x devfn: %x +%x/%x -> %08x\n",
473 pdn->busno, pdn->devfn, where, size, val);
476 opal_pci_config_write_byte(phb->opal_id, bdfn, where, val);
479 opal_pci_config_write_half_word(phb->opal_id, bdfn, where, val);
482 opal_pci_config_write_word(phb->opal_id, bdfn, where, val);
485 return PCIBIOS_FUNC_NOT_SUPPORTED;
488 return PCIBIOS_SUCCESSFUL;
492 static bool pnv_pci_cfg_check(struct pci_controller *hose,
493 struct device_node *dn)
495 struct eeh_dev *edev = NULL;
496 struct pnv_phb *phb = hose->private_data;
498 /* EEH not enabled ? */
499 if (!(phb->flags & PNV_PHB_FLAG_EEH))
502 /* PE reset or device removed ? */
503 edev = of_node_to_eeh_dev(dn);
506 (edev->pe->state & EEH_PE_CFG_BLOCKED))
509 if (edev->mode & EEH_DEV_REMOVED)
516 static inline pnv_pci_cfg_check(struct pci_controller *hose,
517 struct device_node *dn)
521 #endif /* CONFIG_EEH */
523 static int pnv_pci_read_config(struct pci_bus *bus,
525 int where, int size, u32 *val)
527 struct device_node *dn, *busdn = pci_bus_to_OF_node(bus);
534 for (dn = busdn->child; dn; dn = dn->sibling) {
536 if (pdn && pdn->devfn == devfn) {
537 phb = pdn->phb->private_data;
543 if (!found || !pnv_pci_cfg_check(pdn->phb, dn))
544 return PCIBIOS_DEVICE_NOT_FOUND;
546 ret = pnv_pci_cfg_read(dn, where, size, val);
547 if (phb->flags & PNV_PHB_FLAG_EEH) {
548 if (*val == EEH_IO_ERROR_VALUE(size) &&
549 eeh_dev_check_failure(of_node_to_eeh_dev(dn)))
550 return PCIBIOS_DEVICE_NOT_FOUND;
552 pnv_pci_config_check_eeh(phb, dn);
558 static int pnv_pci_write_config(struct pci_bus *bus,
560 int where, int size, u32 val)
562 struct device_node *dn, *busdn = pci_bus_to_OF_node(bus);
568 for (dn = busdn->child; dn; dn = dn->sibling) {
570 if (pdn && pdn->devfn == devfn) {
571 phb = pdn->phb->private_data;
577 if (!found || !pnv_pci_cfg_check(pdn->phb, dn))
578 return PCIBIOS_DEVICE_NOT_FOUND;
580 ret = pnv_pci_cfg_write(dn, where, size, val);
581 if (!(phb->flags & PNV_PHB_FLAG_EEH))
582 pnv_pci_config_check_eeh(phb, dn);
587 struct pci_ops pnv_pci_ops = {
588 .read = pnv_pci_read_config,
589 .write = pnv_pci_write_config,
592 static int pnv_tce_build(struct iommu_table *tbl, long index, long npages,
593 unsigned long uaddr, enum dma_data_direction direction,
594 struct dma_attrs *attrs, bool rm)
600 proto_tce = TCE_PCI_READ; // Read allowed
602 if (direction != DMA_TO_DEVICE)
603 proto_tce |= TCE_PCI_WRITE;
605 tces = tcep = ((__be64 *)tbl->it_base) + index - tbl->it_offset;
606 rpn = __pa(uaddr) >> tbl->it_page_shift;
609 *(tcep++) = cpu_to_be64(proto_tce |
610 (rpn++ << tbl->it_page_shift));
612 /* Some implementations won't cache invalid TCEs and thus may not
613 * need that flush. We'll probably turn it_type into a bit mask
614 * of flags if that becomes the case
616 if (tbl->it_type & TCE_PCI_SWINV_CREATE)
617 pnv_pci_ioda_tce_invalidate(tbl, tces, tcep - 1, rm);
622 static int pnv_tce_build_vm(struct iommu_table *tbl, long index, long npages,
624 enum dma_data_direction direction,
625 struct dma_attrs *attrs)
627 return pnv_tce_build(tbl, index, npages, uaddr, direction, attrs,
631 static void pnv_tce_free(struct iommu_table *tbl, long index, long npages,
636 tces = tcep = ((__be64 *)tbl->it_base) + index - tbl->it_offset;
639 *(tcep++) = cpu_to_be64(0);
641 if (tbl->it_type & TCE_PCI_SWINV_FREE)
642 pnv_pci_ioda_tce_invalidate(tbl, tces, tcep - 1, rm);
645 static void pnv_tce_free_vm(struct iommu_table *tbl, long index, long npages)
647 pnv_tce_free(tbl, index, npages, false);
650 static unsigned long pnv_tce_get(struct iommu_table *tbl, long index)
652 return ((u64 *)tbl->it_base)[index - tbl->it_offset];
655 static int pnv_tce_build_rm(struct iommu_table *tbl, long index, long npages,
657 enum dma_data_direction direction,
658 struct dma_attrs *attrs)
660 return pnv_tce_build(tbl, index, npages, uaddr, direction, attrs, true);
663 static void pnv_tce_free_rm(struct iommu_table *tbl, long index, long npages)
665 pnv_tce_free(tbl, index, npages, true);
668 void pnv_pci_setup_iommu_table(struct iommu_table *tbl,
669 void *tce_mem, u64 tce_size,
670 u64 dma_offset, unsigned page_shift)
672 tbl->it_blocksize = 16;
673 tbl->it_base = (unsigned long)tce_mem;
674 tbl->it_page_shift = page_shift;
675 tbl->it_offset = dma_offset >> tbl->it_page_shift;
677 tbl->it_size = tce_size >> 3;
679 tbl->it_type = TCE_PCI;
682 static struct iommu_table *pnv_pci_setup_bml_iommu(struct pci_controller *hose)
684 struct iommu_table *tbl;
685 const __be64 *basep, *swinvp;
688 basep = of_get_property(hose->dn, "linux,tce-base", NULL);
689 sizep = of_get_property(hose->dn, "linux,tce-size", NULL);
690 if (basep == NULL || sizep == NULL) {
691 pr_err("PCI: %s has missing tce entries !\n",
692 hose->dn->full_name);
695 tbl = kzalloc_node(sizeof(struct iommu_table), GFP_KERNEL, hose->node);
698 pnv_pci_setup_iommu_table(tbl, __va(be64_to_cpup(basep)),
699 be32_to_cpup(sizep), 0, IOMMU_PAGE_SHIFT_4K);
700 iommu_init_table(tbl, hose->node);
701 iommu_register_group(tbl, pci_domain_nr(hose->bus), 0);
703 /* Deal with SW invalidated TCEs when needed (BML way) */
704 swinvp = of_get_property(hose->dn, "linux,tce-sw-invalidate-info",
707 tbl->it_busno = be64_to_cpu(swinvp[1]);
708 tbl->it_index = (unsigned long)ioremap(be64_to_cpup(swinvp), 8);
709 tbl->it_type = TCE_PCI_SWINV_CREATE | TCE_PCI_SWINV_FREE;
714 static void pnv_pci_dma_fallback_setup(struct pci_controller *hose,
715 struct pci_dev *pdev)
717 struct device_node *np = pci_bus_to_OF_node(hose->bus);
723 if (!pdn->iommu_table)
724 pdn->iommu_table = pnv_pci_setup_bml_iommu(hose);
725 if (!pdn->iommu_table)
727 set_iommu_table_base_and_group(&pdev->dev, pdn->iommu_table);
730 static void pnv_pci_dma_dev_setup(struct pci_dev *pdev)
732 struct pci_controller *hose = pci_bus_to_host(pdev->bus);
733 struct pnv_phb *phb = hose->private_data;
735 /* If we have no phb structure, try to setup a fallback based on
736 * the device-tree (RTAS PCI for example)
738 if (phb && phb->dma_dev_setup)
739 phb->dma_dev_setup(phb, pdev);
741 pnv_pci_dma_fallback_setup(hose, pdev);
744 int pnv_pci_dma_set_mask(struct pci_dev *pdev, u64 dma_mask)
746 struct pci_controller *hose = pci_bus_to_host(pdev->bus);
747 struct pnv_phb *phb = hose->private_data;
749 if (phb && phb->dma_set_mask)
750 return phb->dma_set_mask(phb, pdev, dma_mask);
751 return __dma_set_mask(&pdev->dev, dma_mask);
754 u64 pnv_pci_dma_get_required_mask(struct pci_dev *pdev)
756 struct pci_controller *hose = pci_bus_to_host(pdev->bus);
757 struct pnv_phb *phb = hose->private_data;
759 if (phb && phb->dma_get_required_mask)
760 return phb->dma_get_required_mask(phb, pdev);
762 return __dma_get_required_mask(&pdev->dev);
765 void pnv_pci_shutdown(void)
767 struct pci_controller *hose;
769 list_for_each_entry(hose, &hose_list, list_node) {
770 struct pnv_phb *phb = hose->private_data;
772 if (phb && phb->shutdown)
777 /* Fixup wrong class code in p7ioc and p8 root complex */
778 static void pnv_p7ioc_rc_quirk(struct pci_dev *dev)
780 dev->class = PCI_CLASS_BRIDGE_PCI << 8;
782 DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_IBM, 0x3b9, pnv_p7ioc_rc_quirk);
784 void __init pnv_pci_init(void)
786 struct device_node *np;
788 pci_add_flags(PCI_CAN_SKIP_ISA_ALIGN);
790 /* OPAL absent, try POPAL first then RTAS detection of PHBs */
791 if (!firmware_has_feature(FW_FEATURE_OPAL)) {
792 #ifdef CONFIG_PPC_POWERNV_RTAS
793 init_pci_config_tokens();
794 find_and_init_phbs();
795 #endif /* CONFIG_PPC_POWERNV_RTAS */
797 /* OPAL is here, do our normal stuff */
801 /* Look for IODA IO-Hubs. We don't support mixing IODA
802 * and p5ioc2 due to the need to change some global
805 for_each_compatible_node(np, NULL, "ibm,ioda-hub") {
806 pnv_pci_init_ioda_hub(np);
810 /* Look for p5ioc2 IO-Hubs */
812 for_each_compatible_node(np, NULL, "ibm,p5ioc2")
813 pnv_pci_init_p5ioc2_hub(np);
815 /* Look for ioda2 built-in PHB3's */
816 for_each_compatible_node(np, NULL, "ibm,ioda2-phb")
817 pnv_pci_init_ioda2_phb(np);
820 /* Setup the linkage between OF nodes and PHBs */
823 /* Configure IOMMU DMA hooks */
824 ppc_md.pci_dma_dev_setup = pnv_pci_dma_dev_setup;
825 ppc_md.tce_build = pnv_tce_build_vm;
826 ppc_md.tce_free = pnv_tce_free_vm;
827 ppc_md.tce_build_rm = pnv_tce_build_rm;
828 ppc_md.tce_free_rm = pnv_tce_free_rm;
829 ppc_md.tce_get = pnv_tce_get;
830 set_pci_dma_ops(&dma_iommu_ops);
833 #ifdef CONFIG_PCI_MSI
834 ppc_md.setup_msi_irqs = pnv_setup_msi_irqs;
835 ppc_md.teardown_msi_irqs = pnv_teardown_msi_irqs;
839 static int tce_iommu_bus_notifier(struct notifier_block *nb,
840 unsigned long action, void *data)
842 struct device *dev = data;
845 case BUS_NOTIFY_ADD_DEVICE:
846 return iommu_add_device(dev);
847 case BUS_NOTIFY_DEL_DEVICE:
848 if (dev->iommu_group)
849 iommu_del_device(dev);
856 static struct notifier_block tce_iommu_bus_nb = {
857 .notifier_call = tce_iommu_bus_notifier,
860 static int __init tce_iommu_bus_notifier_init(void)
862 bus_register_notifier(&pci_bus_type, &tce_iommu_bus_nb);
865 machine_subsys_initcall_sync(powernv, tce_iommu_bus_notifier_init);