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/bootmem.h>
20 #include <linux/irq.h>
22 #include <linux/msi.h>
24 #include <asm/sections.h>
27 #include <asm/pci-bridge.h>
28 #include <asm/machdep.h>
29 #include <asm/ppc-pci.h>
31 #include <asm/iommu.h>
33 #include <asm/firmware.h>
39 #define PCI_RESET_DELAY_US 3000000
41 #define cfg_dbg(fmt...) do { } while(0)
42 //#define cfg_dbg(fmt...) printk(fmt)
45 static int pnv_msi_check_device(struct pci_dev* pdev, int nvec, int type)
47 struct pci_controller *hose = pci_bus_to_host(pdev->bus);
48 struct pnv_phb *phb = hose->private_data;
50 return (phb && phb->msi_map) ? 0 : -ENODEV;
53 static unsigned int pnv_get_one_msi(struct pnv_phb *phb)
58 spin_lock_irqsave(&phb->lock, flags);
60 id = find_next_zero_bit(phb->msi_map, phb->msi_count, phb->msi_next);
61 if (id >= phb->msi_count && phb->msi_next)
62 id = find_next_zero_bit(phb->msi_map, phb->msi_count, 0);
63 if (id >= phb->msi_count) {
67 __set_bit(id, phb->msi_map);
68 rc = id + phb->msi_base;
70 spin_unlock_irqrestore(&phb->lock, flags);
74 static void pnv_put_msi(struct pnv_phb *phb, unsigned int hwirq)
79 if (WARN_ON(hwirq < phb->msi_base ||
80 hwirq >= (phb->msi_base + phb->msi_count)))
82 id = hwirq - phb->msi_base;
84 spin_lock_irqsave(&phb->lock, flags);
85 __clear_bit(id, phb->msi_map);
86 spin_unlock_irqrestore(&phb->lock, flags);
89 static int pnv_setup_msi_irqs(struct pci_dev *pdev, int nvec, int type)
91 struct pci_controller *hose = pci_bus_to_host(pdev->bus);
92 struct pnv_phb *phb = hose->private_data;
93 struct msi_desc *entry;
95 unsigned int hwirq, virq;
101 list_for_each_entry(entry, &pdev->msi_list, list) {
102 if (!entry->msi_attrib.is_64 && !phb->msi32_support) {
103 pr_warn("%s: Supports only 64-bit MSIs\n",
107 hwirq = pnv_get_one_msi(phb);
109 pr_warn("%s: Failed to find a free MSI\n",
113 virq = irq_create_mapping(NULL, hwirq);
114 if (virq == NO_IRQ) {
115 pr_warn("%s: Failed to map MSI to linux irq\n",
117 pnv_put_msi(phb, hwirq);
120 rc = phb->msi_setup(phb, pdev, hwirq, entry->msi_attrib.is_64,
123 pr_warn("%s: Failed to setup MSI\n", pci_name(pdev));
124 irq_dispose_mapping(virq);
125 pnv_put_msi(phb, hwirq);
128 irq_set_msi_desc(virq, entry);
129 write_msi_msg(virq, &msg);
134 static void pnv_teardown_msi_irqs(struct pci_dev *pdev)
136 struct pci_controller *hose = pci_bus_to_host(pdev->bus);
137 struct pnv_phb *phb = hose->private_data;
138 struct msi_desc *entry;
143 list_for_each_entry(entry, &pdev->msi_list, list) {
144 if (entry->irq == NO_IRQ)
146 irq_set_msi_desc(entry->irq, NULL);
147 pnv_put_msi(phb, virq_to_hw(entry->irq));
148 irq_dispose_mapping(entry->irq);
151 #endif /* CONFIG_PCI_MSI */
153 static void pnv_pci_dump_p7ioc_diag_data(struct pnv_phb *phb)
155 struct OpalIoP7IOCPhbErrorData *data = &phb->diag.p7ioc;
158 pr_info("PHB %d diagnostic data:\n", phb->hose->global_number);
160 pr_info(" brdgCtl = 0x%08x\n", data->brdgCtl);
162 pr_info(" portStatusReg = 0x%08x\n", data->portStatusReg);
163 pr_info(" rootCmplxStatus = 0x%08x\n", data->rootCmplxStatus);
164 pr_info(" busAgentStatus = 0x%08x\n", data->busAgentStatus);
166 pr_info(" deviceStatus = 0x%08x\n", data->deviceStatus);
167 pr_info(" slotStatus = 0x%08x\n", data->slotStatus);
168 pr_info(" linkStatus = 0x%08x\n", data->linkStatus);
169 pr_info(" devCmdStatus = 0x%08x\n", data->devCmdStatus);
170 pr_info(" devSecStatus = 0x%08x\n", data->devSecStatus);
172 pr_info(" rootErrorStatus = 0x%08x\n", data->rootErrorStatus);
173 pr_info(" uncorrErrorStatus = 0x%08x\n", data->uncorrErrorStatus);
174 pr_info(" corrErrorStatus = 0x%08x\n", data->corrErrorStatus);
175 pr_info(" tlpHdr1 = 0x%08x\n", data->tlpHdr1);
176 pr_info(" tlpHdr2 = 0x%08x\n", data->tlpHdr2);
177 pr_info(" tlpHdr3 = 0x%08x\n", data->tlpHdr3);
178 pr_info(" tlpHdr4 = 0x%08x\n", data->tlpHdr4);
179 pr_info(" sourceId = 0x%08x\n", data->sourceId);
181 pr_info(" errorClass = 0x%016llx\n", data->errorClass);
182 pr_info(" correlator = 0x%016llx\n", data->correlator);
184 pr_info(" p7iocPlssr = 0x%016llx\n", data->p7iocPlssr);
185 pr_info(" p7iocCsr = 0x%016llx\n", data->p7iocCsr);
186 pr_info(" lemFir = 0x%016llx\n", data->lemFir);
187 pr_info(" lemErrorMask = 0x%016llx\n", data->lemErrorMask);
188 pr_info(" lemWOF = 0x%016llx\n", data->lemWOF);
189 pr_info(" phbErrorStatus = 0x%016llx\n", data->phbErrorStatus);
190 pr_info(" phbFirstErrorStatus = 0x%016llx\n", data->phbFirstErrorStatus);
191 pr_info(" phbErrorLog0 = 0x%016llx\n", data->phbErrorLog0);
192 pr_info(" phbErrorLog1 = 0x%016llx\n", data->phbErrorLog1);
193 pr_info(" mmioErrorStatus = 0x%016llx\n", data->mmioErrorStatus);
194 pr_info(" mmioFirstErrorStatus = 0x%016llx\n", data->mmioFirstErrorStatus);
195 pr_info(" mmioErrorLog0 = 0x%016llx\n", data->mmioErrorLog0);
196 pr_info(" mmioErrorLog1 = 0x%016llx\n", data->mmioErrorLog1);
197 pr_info(" dma0ErrorStatus = 0x%016llx\n", data->dma0ErrorStatus);
198 pr_info(" dma0FirstErrorStatus = 0x%016llx\n", data->dma0FirstErrorStatus);
199 pr_info(" dma0ErrorLog0 = 0x%016llx\n", data->dma0ErrorLog0);
200 pr_info(" dma0ErrorLog1 = 0x%016llx\n", data->dma0ErrorLog1);
201 pr_info(" dma1ErrorStatus = 0x%016llx\n", data->dma1ErrorStatus);
202 pr_info(" dma1FirstErrorStatus = 0x%016llx\n", data->dma1FirstErrorStatus);
203 pr_info(" dma1ErrorLog0 = 0x%016llx\n", data->dma1ErrorLog0);
204 pr_info(" dma1ErrorLog1 = 0x%016llx\n", data->dma1ErrorLog1);
206 for (i = 0; i < OPAL_P7IOC_NUM_PEST_REGS; i++) {
207 if ((data->pestA[i] >> 63) == 0 &&
208 (data->pestB[i] >> 63) == 0)
210 pr_info(" PE[%3d] PESTA = 0x%016llx\n", i, data->pestA[i]);
211 pr_info(" PESTB = 0x%016llx\n", data->pestB[i]);
215 static void pnv_pci_dump_phb_diag_data(struct pnv_phb *phb)
218 case PNV_PHB_MODEL_P7IOC:
219 pnv_pci_dump_p7ioc_diag_data(phb);
222 pr_warning("PCI %d: Can't decode this PHB diag data\n",
223 phb->hose->global_number);
227 static void pnv_pci_handle_eeh_config(struct pnv_phb *phb, u32 pe_no)
229 unsigned long flags, rc;
232 spin_lock_irqsave(&phb->lock, flags);
234 rc = opal_pci_get_phb_diag_data(phb->opal_id, phb->diag.blob, PNV_PCI_DIAG_BUF_SIZE);
235 has_diag = (rc == OPAL_SUCCESS);
237 rc = opal_pci_eeh_freeze_clear(phb->opal_id, pe_no,
238 OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
240 pr_warning("PCI %d: Failed to clear EEH freeze state"
241 " for PE#%d, err %ld\n",
242 phb->hose->global_number, pe_no, rc);
244 /* For now, let's only display the diag buffer when we fail to clear
245 * the EEH status. We'll do more sensible things later when we have
246 * proper EEH support. We need to make sure we don't pollute ourselves
247 * with the normal errors generated when probing empty slots
250 pnv_pci_dump_phb_diag_data(phb);
252 pr_warning("PCI %d: No diag data available\n",
253 phb->hose->global_number);
256 spin_unlock_irqrestore(&phb->lock, flags);
259 static void pnv_pci_config_check_eeh(struct pnv_phb *phb, struct pci_bus *bus,
267 /* Get PE# if we support IODA */
268 pe_no = phb->bdfn_to_pe ? phb->bdfn_to_pe(phb, bus, bdfn & 0xff) : 0;
270 /* Read freeze status */
271 rc = opal_pci_eeh_freeze_status(phb->opal_id, pe_no, &fstate, &pcierr,
274 pr_warning("PCI %d: Failed to read EEH status for PE#%d,"
275 " err %lld\n", phb->hose->global_number, pe_no, rc);
278 cfg_dbg(" -> EEH check, bdfn=%04x PE%d fstate=%x\n",
279 bdfn, pe_no, fstate);
281 pnv_pci_handle_eeh_config(phb, pe_no);
284 static int pnv_pci_read_config(struct pci_bus *bus,
286 int where, int size, u32 *val)
288 struct pci_controller *hose = pci_bus_to_host(bus);
289 struct pnv_phb *phb = hose->private_data;
290 u32 bdfn = (((uint64_t)bus->number) << 8) | devfn;
294 return PCIBIOS_DEVICE_NOT_FOUND;
299 rc = opal_pci_config_read_byte(phb->opal_id, bdfn, where, &v8);
300 *val = (rc == OPAL_SUCCESS) ? v8 : 0xff;
305 rc = opal_pci_config_read_half_word(phb->opal_id, bdfn, where,
307 *val = (rc == OPAL_SUCCESS) ? v16 : 0xffff;
312 rc = opal_pci_config_read_word(phb->opal_id, bdfn, where, &v32);
313 *val = (rc == OPAL_SUCCESS) ? v32 : 0xffffffff;
317 return PCIBIOS_FUNC_NOT_SUPPORTED;
319 cfg_dbg("pnv_pci_read_config bus: %x devfn: %x +%x/%x -> %08x\n",
320 bus->number, devfn, where, size, *val);
322 /* Check if the PHB got frozen due to an error (no response) */
323 pnv_pci_config_check_eeh(phb, bus, bdfn);
325 return PCIBIOS_SUCCESSFUL;
328 static int pnv_pci_write_config(struct pci_bus *bus,
330 int where, int size, u32 val)
332 struct pci_controller *hose = pci_bus_to_host(bus);
333 struct pnv_phb *phb = hose->private_data;
334 u32 bdfn = (((uint64_t)bus->number) << 8) | devfn;
337 return PCIBIOS_DEVICE_NOT_FOUND;
339 cfg_dbg("pnv_pci_write_config bus: %x devfn: %x +%x/%x -> %08x\n",
340 bus->number, devfn, where, size, val);
343 opal_pci_config_write_byte(phb->opal_id, bdfn, where, val);
346 opal_pci_config_write_half_word(phb->opal_id, bdfn, where, val);
349 opal_pci_config_write_word(phb->opal_id, bdfn, where, val);
352 return PCIBIOS_FUNC_NOT_SUPPORTED;
354 /* Check if the PHB got frozen due to an error (no response) */
355 pnv_pci_config_check_eeh(phb, bus, bdfn);
357 return PCIBIOS_SUCCESSFUL;
360 struct pci_ops pnv_pci_ops = {
361 .read = pnv_pci_read_config,
362 .write = pnv_pci_write_config,
366 static void pnv_tce_invalidate(struct iommu_table *tbl,
367 u64 *startp, u64 *endp)
369 u64 __iomem *invalidate = (u64 __iomem *)tbl->it_index;
370 unsigned long start, end, inc;
372 start = __pa(startp);
376 /* BML uses this case for p6/p7/galaxy2: Shift addr and put in node */
381 start |= tbl->it_busno;
382 end |= tbl->it_busno;
384 /* p7ioc-style invalidation, 2 TCEs per write */
385 else if (tbl->it_type & TCE_PCI_SWINV_PAIR) {
386 start |= (1ull << 63);
390 /* Default (older HW) */
394 end |= inc - 1; /* round up end to be different than start */
396 mb(); /* Ensure above stores are visible */
397 while (start <= end) {
398 __raw_writeq(start, invalidate);
401 /* The iommu layer will do another mb() for us on build() and
402 * we don't care on free()
407 static int pnv_tce_build(struct iommu_table *tbl, long index, long npages,
408 unsigned long uaddr, enum dma_data_direction direction,
409 struct dma_attrs *attrs)
415 proto_tce = TCE_PCI_READ; // Read allowed
417 if (direction != DMA_TO_DEVICE)
418 proto_tce |= TCE_PCI_WRITE;
420 tces = tcep = ((u64 *)tbl->it_base) + index - tbl->it_offset;
421 rpn = __pa(uaddr) >> TCE_SHIFT;
424 *(tcep++) = proto_tce | (rpn++ << TCE_RPN_SHIFT);
426 /* Some implementations won't cache invalid TCEs and thus may not
427 * need that flush. We'll probably turn it_type into a bit mask
428 * of flags if that becomes the case
430 if (tbl->it_type & TCE_PCI_SWINV_CREATE)
431 pnv_tce_invalidate(tbl, tces, tcep - 1);
436 static void pnv_tce_free(struct iommu_table *tbl, long index, long npages)
440 tces = tcep = ((u64 *)tbl->it_base) + index - tbl->it_offset;
445 if (tbl->it_type & TCE_PCI_SWINV_FREE)
446 pnv_tce_invalidate(tbl, tces, tcep - 1);
449 static unsigned long pnv_tce_get(struct iommu_table *tbl, long index)
451 return ((u64 *)tbl->it_base)[index - tbl->it_offset];
454 void pnv_pci_setup_iommu_table(struct iommu_table *tbl,
455 void *tce_mem, u64 tce_size,
458 tbl->it_blocksize = 16;
459 tbl->it_base = (unsigned long)tce_mem;
460 tbl->it_offset = dma_offset >> IOMMU_PAGE_SHIFT;
462 tbl->it_size = tce_size >> 3;
464 tbl->it_type = TCE_PCI;
467 static struct iommu_table *pnv_pci_setup_bml_iommu(struct pci_controller *hose)
469 struct iommu_table *tbl;
470 const __be64 *basep, *swinvp;
473 basep = of_get_property(hose->dn, "linux,tce-base", NULL);
474 sizep = of_get_property(hose->dn, "linux,tce-size", NULL);
475 if (basep == NULL || sizep == NULL) {
476 pr_err("PCI: %s has missing tce entries !\n",
477 hose->dn->full_name);
480 tbl = kzalloc_node(sizeof(struct iommu_table), GFP_KERNEL, hose->node);
483 pnv_pci_setup_iommu_table(tbl, __va(be64_to_cpup(basep)),
484 be32_to_cpup(sizep), 0);
485 iommu_init_table(tbl, hose->node);
487 /* Deal with SW invalidated TCEs when needed (BML way) */
488 swinvp = of_get_property(hose->dn, "linux,tce-sw-invalidate-info",
491 tbl->it_busno = swinvp[1];
492 tbl->it_index = (unsigned long)ioremap(swinvp[0], 8);
493 tbl->it_type = TCE_PCI_SWINV_CREATE | TCE_PCI_SWINV_FREE;
498 static void pnv_pci_dma_fallback_setup(struct pci_controller *hose,
499 struct pci_dev *pdev)
501 struct device_node *np = pci_bus_to_OF_node(hose->bus);
507 if (!pdn->iommu_table)
508 pdn->iommu_table = pnv_pci_setup_bml_iommu(hose);
509 if (!pdn->iommu_table)
511 set_iommu_table_base(&pdev->dev, pdn->iommu_table);
514 static void pnv_pci_dma_dev_setup(struct pci_dev *pdev)
516 struct pci_controller *hose = pci_bus_to_host(pdev->bus);
517 struct pnv_phb *phb = hose->private_data;
519 /* If we have no phb structure, try to setup a fallback based on
520 * the device-tree (RTAS PCI for example)
522 if (phb && phb->dma_dev_setup)
523 phb->dma_dev_setup(phb, pdev);
525 pnv_pci_dma_fallback_setup(hose, pdev);
528 /* Fixup wrong class code in p7ioc root complex */
529 static void pnv_p7ioc_rc_quirk(struct pci_dev *dev)
531 dev->class = PCI_CLASS_BRIDGE_PCI << 8;
533 DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_IBM, 0x3b9, pnv_p7ioc_rc_quirk);
535 static int pnv_pci_probe_mode(struct pci_bus *bus)
537 struct pci_controller *hose = pci_bus_to_host(bus);
538 const __be64 *tstamp;
542 /* We hijack this as a way to ensure we have waited long
543 * enough since the reset was lifted on the PCI bus
545 if (bus != hose->bus)
546 return PCI_PROBE_NORMAL;
547 tstamp = of_get_property(hose->dn, "reset-clear-timestamp", NULL);
548 if (!tstamp || !*tstamp)
549 return PCI_PROBE_NORMAL;
551 now = mftb() / tb_ticks_per_usec;
552 target = (be64_to_cpup(tstamp) / tb_ticks_per_usec)
553 + PCI_RESET_DELAY_US;
555 pr_devel("pci %04d: Reset target: 0x%llx now: 0x%llx\n",
556 hose->global_number, target, now);
559 msleep((target - now + 999) / 1000);
561 return PCI_PROBE_NORMAL;
564 void __init pnv_pci_init(void)
566 struct device_node *np;
568 pci_add_flags(PCI_CAN_SKIP_ISA_ALIGN);
570 /* OPAL absent, try POPAL first then RTAS detection of PHBs */
571 if (!firmware_has_feature(FW_FEATURE_OPAL)) {
572 #ifdef CONFIG_PPC_POWERNV_RTAS
573 init_pci_config_tokens();
574 find_and_init_phbs();
575 #endif /* CONFIG_PPC_POWERNV_RTAS */
577 /* OPAL is here, do our normal stuff */
581 /* Look for IODA IO-Hubs. We don't support mixing IODA
582 * and p5ioc2 due to the need to change some global
585 for_each_compatible_node(np, NULL, "ibm,ioda-hub") {
586 pnv_pci_init_ioda_hub(np);
590 /* Look for p5ioc2 IO-Hubs */
592 for_each_compatible_node(np, NULL, "ibm,p5ioc2")
593 pnv_pci_init_p5ioc2_hub(np);
596 /* Setup the linkage between OF nodes and PHBs */
599 /* Configure IOMMU DMA hooks */
600 ppc_md.pci_dma_dev_setup = pnv_pci_dma_dev_setup;
601 ppc_md.tce_build = pnv_tce_build;
602 ppc_md.tce_free = pnv_tce_free;
603 ppc_md.tce_get = pnv_tce_get;
604 ppc_md.pci_probe_mode = pnv_pci_probe_mode;
605 set_pci_dma_ops(&dma_iommu_ops);
608 #ifdef CONFIG_PCI_MSI
609 ppc_md.msi_check_device = pnv_msi_check_device;
610 ppc_md.setup_msi_irqs = pnv_setup_msi_irqs;
611 ppc_md.teardown_msi_irqs = pnv_teardown_msi_irqs;