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/msi_bitmap.h>
30 #include <asm/ppc-pci.h>
32 #include <asm/iommu.h>
34 #include <asm/firmware.h>
40 #define PCI_RESET_DELAY_US 3000000
42 #define cfg_dbg(fmt...) do { } while(0)
43 //#define cfg_dbg(fmt...) printk(fmt)
46 static int pnv_msi_check_device(struct pci_dev* pdev, int nvec, int type)
48 struct pci_controller *hose = pci_bus_to_host(pdev->bus);
49 struct pnv_phb *phb = hose->private_data;
51 return (phb && phb->msi_bmp.bitmap) ? 0 : -ENODEV;
54 static int pnv_setup_msi_irqs(struct pci_dev *pdev, int nvec, int type)
56 struct pci_controller *hose = pci_bus_to_host(pdev->bus);
57 struct pnv_phb *phb = hose->private_data;
58 struct msi_desc *entry;
67 list_for_each_entry(entry, &pdev->msi_list, list) {
68 if (!entry->msi_attrib.is_64 && !phb->msi32_support) {
69 pr_warn("%s: Supports only 64-bit MSIs\n",
73 hwirq = msi_bitmap_alloc_hwirqs(&phb->msi_bmp, 1);
75 pr_warn("%s: Failed to find a free MSI\n",
79 virq = irq_create_mapping(NULL, phb->msi_base + hwirq);
81 pr_warn("%s: Failed to map MSI to linux irq\n",
83 msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq, 1);
86 rc = phb->msi_setup(phb, pdev, phb->msi_base + hwirq,
87 virq, entry->msi_attrib.is_64, &msg);
89 pr_warn("%s: Failed to setup MSI\n", pci_name(pdev));
90 irq_dispose_mapping(virq);
91 msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq, 1);
94 irq_set_msi_desc(virq, entry);
95 write_msi_msg(virq, &msg);
100 static void pnv_teardown_msi_irqs(struct pci_dev *pdev)
102 struct pci_controller *hose = pci_bus_to_host(pdev->bus);
103 struct pnv_phb *phb = hose->private_data;
104 struct msi_desc *entry;
109 list_for_each_entry(entry, &pdev->msi_list, list) {
110 if (entry->irq == NO_IRQ)
112 irq_set_msi_desc(entry->irq, NULL);
113 msi_bitmap_free_hwirqs(&phb->msi_bmp,
114 virq_to_hw(entry->irq) - phb->msi_base, 1);
115 irq_dispose_mapping(entry->irq);
118 #endif /* CONFIG_PCI_MSI */
120 static void pnv_pci_dump_p7ioc_diag_data(struct pnv_phb *phb)
122 struct OpalIoP7IOCPhbErrorData *data = &phb->diag.p7ioc;
125 pr_info("PHB %d diagnostic data:\n", phb->hose->global_number);
127 pr_info(" brdgCtl = 0x%08x\n", data->brdgCtl);
129 pr_info(" portStatusReg = 0x%08x\n", data->portStatusReg);
130 pr_info(" rootCmplxStatus = 0x%08x\n", data->rootCmplxStatus);
131 pr_info(" busAgentStatus = 0x%08x\n", data->busAgentStatus);
133 pr_info(" deviceStatus = 0x%08x\n", data->deviceStatus);
134 pr_info(" slotStatus = 0x%08x\n", data->slotStatus);
135 pr_info(" linkStatus = 0x%08x\n", data->linkStatus);
136 pr_info(" devCmdStatus = 0x%08x\n", data->devCmdStatus);
137 pr_info(" devSecStatus = 0x%08x\n", data->devSecStatus);
139 pr_info(" rootErrorStatus = 0x%08x\n", data->rootErrorStatus);
140 pr_info(" uncorrErrorStatus = 0x%08x\n", data->uncorrErrorStatus);
141 pr_info(" corrErrorStatus = 0x%08x\n", data->corrErrorStatus);
142 pr_info(" tlpHdr1 = 0x%08x\n", data->tlpHdr1);
143 pr_info(" tlpHdr2 = 0x%08x\n", data->tlpHdr2);
144 pr_info(" tlpHdr3 = 0x%08x\n", data->tlpHdr3);
145 pr_info(" tlpHdr4 = 0x%08x\n", data->tlpHdr4);
146 pr_info(" sourceId = 0x%08x\n", data->sourceId);
148 pr_info(" errorClass = 0x%016llx\n", data->errorClass);
149 pr_info(" correlator = 0x%016llx\n", data->correlator);
151 pr_info(" p7iocPlssr = 0x%016llx\n", data->p7iocPlssr);
152 pr_info(" p7iocCsr = 0x%016llx\n", data->p7iocCsr);
153 pr_info(" lemFir = 0x%016llx\n", data->lemFir);
154 pr_info(" lemErrorMask = 0x%016llx\n", data->lemErrorMask);
155 pr_info(" lemWOF = 0x%016llx\n", data->lemWOF);
156 pr_info(" phbErrorStatus = 0x%016llx\n", data->phbErrorStatus);
157 pr_info(" phbFirstErrorStatus = 0x%016llx\n", data->phbFirstErrorStatus);
158 pr_info(" phbErrorLog0 = 0x%016llx\n", data->phbErrorLog0);
159 pr_info(" phbErrorLog1 = 0x%016llx\n", data->phbErrorLog1);
160 pr_info(" mmioErrorStatus = 0x%016llx\n", data->mmioErrorStatus);
161 pr_info(" mmioFirstErrorStatus = 0x%016llx\n", data->mmioFirstErrorStatus);
162 pr_info(" mmioErrorLog0 = 0x%016llx\n", data->mmioErrorLog0);
163 pr_info(" mmioErrorLog1 = 0x%016llx\n", data->mmioErrorLog1);
164 pr_info(" dma0ErrorStatus = 0x%016llx\n", data->dma0ErrorStatus);
165 pr_info(" dma0FirstErrorStatus = 0x%016llx\n", data->dma0FirstErrorStatus);
166 pr_info(" dma0ErrorLog0 = 0x%016llx\n", data->dma0ErrorLog0);
167 pr_info(" dma0ErrorLog1 = 0x%016llx\n", data->dma0ErrorLog1);
168 pr_info(" dma1ErrorStatus = 0x%016llx\n", data->dma1ErrorStatus);
169 pr_info(" dma1FirstErrorStatus = 0x%016llx\n", data->dma1FirstErrorStatus);
170 pr_info(" dma1ErrorLog0 = 0x%016llx\n", data->dma1ErrorLog0);
171 pr_info(" dma1ErrorLog1 = 0x%016llx\n", data->dma1ErrorLog1);
173 for (i = 0; i < OPAL_P7IOC_NUM_PEST_REGS; i++) {
174 if ((data->pestA[i] >> 63) == 0 &&
175 (data->pestB[i] >> 63) == 0)
177 pr_info(" PE[%3d] PESTA = 0x%016llx\n", i, data->pestA[i]);
178 pr_info(" PESTB = 0x%016llx\n", data->pestB[i]);
182 static void pnv_pci_dump_phb_diag_data(struct pnv_phb *phb)
185 case PNV_PHB_MODEL_P7IOC:
186 pnv_pci_dump_p7ioc_diag_data(phb);
189 pr_warning("PCI %d: Can't decode this PHB diag data\n",
190 phb->hose->global_number);
194 static void pnv_pci_handle_eeh_config(struct pnv_phb *phb, u32 pe_no)
196 unsigned long flags, rc;
199 spin_lock_irqsave(&phb->lock, flags);
201 rc = opal_pci_get_phb_diag_data(phb->opal_id, phb->diag.blob, PNV_PCI_DIAG_BUF_SIZE);
202 has_diag = (rc == OPAL_SUCCESS);
204 rc = opal_pci_eeh_freeze_clear(phb->opal_id, pe_no,
205 OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
207 pr_warning("PCI %d: Failed to clear EEH freeze state"
208 " for PE#%d, err %ld\n",
209 phb->hose->global_number, pe_no, rc);
211 /* For now, let's only display the diag buffer when we fail to clear
212 * the EEH status. We'll do more sensible things later when we have
213 * proper EEH support. We need to make sure we don't pollute ourselves
214 * with the normal errors generated when probing empty slots
217 pnv_pci_dump_phb_diag_data(phb);
219 pr_warning("PCI %d: No diag data available\n",
220 phb->hose->global_number);
223 spin_unlock_irqrestore(&phb->lock, flags);
226 static void pnv_pci_config_check_eeh(struct pnv_phb *phb, struct pci_bus *bus,
234 /* Get PE# if we support IODA */
235 pe_no = phb->bdfn_to_pe ? phb->bdfn_to_pe(phb, bus, bdfn & 0xff) : 0;
237 /* Read freeze status */
238 rc = opal_pci_eeh_freeze_status(phb->opal_id, pe_no, &fstate, &pcierr,
241 pr_warning("PCI %d: Failed to read EEH status for PE#%d,"
242 " err %lld\n", phb->hose->global_number, pe_no, rc);
245 cfg_dbg(" -> EEH check, bdfn=%04x PE%d fstate=%x\n",
246 bdfn, pe_no, fstate);
248 pnv_pci_handle_eeh_config(phb, pe_no);
251 static int pnv_pci_read_config(struct pci_bus *bus,
253 int where, int size, u32 *val)
255 struct pci_controller *hose = pci_bus_to_host(bus);
256 struct pnv_phb *phb = hose->private_data;
257 u32 bdfn = (((uint64_t)bus->number) << 8) | devfn;
261 return PCIBIOS_DEVICE_NOT_FOUND;
266 rc = opal_pci_config_read_byte(phb->opal_id, bdfn, where, &v8);
267 *val = (rc == OPAL_SUCCESS) ? v8 : 0xff;
272 rc = opal_pci_config_read_half_word(phb->opal_id, bdfn, where,
274 *val = (rc == OPAL_SUCCESS) ? v16 : 0xffff;
279 rc = opal_pci_config_read_word(phb->opal_id, bdfn, where, &v32);
280 *val = (rc == OPAL_SUCCESS) ? v32 : 0xffffffff;
284 return PCIBIOS_FUNC_NOT_SUPPORTED;
286 cfg_dbg("pnv_pci_read_config bus: %x devfn: %x +%x/%x -> %08x\n",
287 bus->number, devfn, where, size, *val);
289 /* Check if the PHB got frozen due to an error (no response) */
290 pnv_pci_config_check_eeh(phb, bus, bdfn);
292 return PCIBIOS_SUCCESSFUL;
295 static int pnv_pci_write_config(struct pci_bus *bus,
297 int where, int size, u32 val)
299 struct pci_controller *hose = pci_bus_to_host(bus);
300 struct pnv_phb *phb = hose->private_data;
301 u32 bdfn = (((uint64_t)bus->number) << 8) | devfn;
304 return PCIBIOS_DEVICE_NOT_FOUND;
306 cfg_dbg("pnv_pci_write_config bus: %x devfn: %x +%x/%x -> %08x\n",
307 bus->number, devfn, where, size, val);
310 opal_pci_config_write_byte(phb->opal_id, bdfn, where, val);
313 opal_pci_config_write_half_word(phb->opal_id, bdfn, where, val);
316 opal_pci_config_write_word(phb->opal_id, bdfn, where, val);
319 return PCIBIOS_FUNC_NOT_SUPPORTED;
321 /* Check if the PHB got frozen due to an error (no response) */
322 pnv_pci_config_check_eeh(phb, bus, bdfn);
324 return PCIBIOS_SUCCESSFUL;
327 struct pci_ops pnv_pci_ops = {
328 .read = pnv_pci_read_config,
329 .write = pnv_pci_write_config,
332 static int pnv_tce_build(struct iommu_table *tbl, long index, long npages,
333 unsigned long uaddr, enum dma_data_direction direction,
334 struct dma_attrs *attrs)
340 proto_tce = TCE_PCI_READ; // Read allowed
342 if (direction != DMA_TO_DEVICE)
343 proto_tce |= TCE_PCI_WRITE;
345 tces = tcep = ((u64 *)tbl->it_base) + index - tbl->it_offset;
346 rpn = __pa(uaddr) >> TCE_SHIFT;
349 *(tcep++) = proto_tce | (rpn++ << TCE_RPN_SHIFT);
351 /* Some implementations won't cache invalid TCEs and thus may not
352 * need that flush. We'll probably turn it_type into a bit mask
353 * of flags if that becomes the case
355 if (tbl->it_type & TCE_PCI_SWINV_CREATE)
356 pnv_pci_ioda_tce_invalidate(tbl, tces, tcep - 1);
361 static void pnv_tce_free(struct iommu_table *tbl, long index, long npages)
365 tces = tcep = ((u64 *)tbl->it_base) + index - tbl->it_offset;
370 if (tbl->it_type & TCE_PCI_SWINV_CREATE)
371 pnv_pci_ioda_tce_invalidate(tbl, tces, tcep - 1);
374 static unsigned long pnv_tce_get(struct iommu_table *tbl, long index)
376 return ((u64 *)tbl->it_base)[index - tbl->it_offset];
379 void pnv_pci_setup_iommu_table(struct iommu_table *tbl,
380 void *tce_mem, u64 tce_size,
383 tbl->it_blocksize = 16;
384 tbl->it_base = (unsigned long)tce_mem;
385 tbl->it_offset = dma_offset >> IOMMU_PAGE_SHIFT;
387 tbl->it_size = tce_size >> 3;
389 tbl->it_type = TCE_PCI;
392 static struct iommu_table *pnv_pci_setup_bml_iommu(struct pci_controller *hose)
394 struct iommu_table *tbl;
395 const __be64 *basep, *swinvp;
398 basep = of_get_property(hose->dn, "linux,tce-base", NULL);
399 sizep = of_get_property(hose->dn, "linux,tce-size", NULL);
400 if (basep == NULL || sizep == NULL) {
401 pr_err("PCI: %s has missing tce entries !\n",
402 hose->dn->full_name);
405 tbl = kzalloc_node(sizeof(struct iommu_table), GFP_KERNEL, hose->node);
408 pnv_pci_setup_iommu_table(tbl, __va(be64_to_cpup(basep)),
409 be32_to_cpup(sizep), 0);
410 iommu_init_table(tbl, hose->node);
412 /* Deal with SW invalidated TCEs when needed (BML way) */
413 swinvp = of_get_property(hose->dn, "linux,tce-sw-invalidate-info",
416 tbl->it_busno = swinvp[1];
417 tbl->it_index = (unsigned long)ioremap(swinvp[0], 8);
418 tbl->it_type = TCE_PCI_SWINV_CREATE | TCE_PCI_SWINV_FREE;
423 static void pnv_pci_dma_fallback_setup(struct pci_controller *hose,
424 struct pci_dev *pdev)
426 struct device_node *np = pci_bus_to_OF_node(hose->bus);
432 if (!pdn->iommu_table)
433 pdn->iommu_table = pnv_pci_setup_bml_iommu(hose);
434 if (!pdn->iommu_table)
436 set_iommu_table_base(&pdev->dev, pdn->iommu_table);
439 static void pnv_pci_dma_dev_setup(struct pci_dev *pdev)
441 struct pci_controller *hose = pci_bus_to_host(pdev->bus);
442 struct pnv_phb *phb = hose->private_data;
444 /* If we have no phb structure, try to setup a fallback based on
445 * the device-tree (RTAS PCI for example)
447 if (phb && phb->dma_dev_setup)
448 phb->dma_dev_setup(phb, pdev);
450 pnv_pci_dma_fallback_setup(hose, pdev);
453 /* Fixup wrong class code in p7ioc and p8 root complex */
454 static void pnv_p7ioc_rc_quirk(struct pci_dev *dev)
456 dev->class = PCI_CLASS_BRIDGE_PCI << 8;
458 DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_IBM, 0x3b9, pnv_p7ioc_rc_quirk);
460 static int pnv_pci_probe_mode(struct pci_bus *bus)
462 struct pci_controller *hose = pci_bus_to_host(bus);
463 const __be64 *tstamp;
467 /* We hijack this as a way to ensure we have waited long
468 * enough since the reset was lifted on the PCI bus
470 if (bus != hose->bus)
471 return PCI_PROBE_NORMAL;
472 tstamp = of_get_property(hose->dn, "reset-clear-timestamp", NULL);
473 if (!tstamp || !*tstamp)
474 return PCI_PROBE_NORMAL;
476 now = mftb() / tb_ticks_per_usec;
477 target = (be64_to_cpup(tstamp) / tb_ticks_per_usec)
478 + PCI_RESET_DELAY_US;
480 pr_devel("pci %04d: Reset target: 0x%llx now: 0x%llx\n",
481 hose->global_number, target, now);
484 msleep((target - now + 999) / 1000);
486 return PCI_PROBE_NORMAL;
489 void __init pnv_pci_init(void)
491 struct device_node *np;
493 pci_add_flags(PCI_CAN_SKIP_ISA_ALIGN);
495 /* OPAL absent, try POPAL first then RTAS detection of PHBs */
496 if (!firmware_has_feature(FW_FEATURE_OPAL)) {
497 #ifdef CONFIG_PPC_POWERNV_RTAS
498 init_pci_config_tokens();
499 find_and_init_phbs();
500 #endif /* CONFIG_PPC_POWERNV_RTAS */
502 /* OPAL is here, do our normal stuff */
506 /* Look for IODA IO-Hubs. We don't support mixing IODA
507 * and p5ioc2 due to the need to change some global
510 for_each_compatible_node(np, NULL, "ibm,ioda-hub") {
511 pnv_pci_init_ioda_hub(np);
515 /* Look for p5ioc2 IO-Hubs */
517 for_each_compatible_node(np, NULL, "ibm,p5ioc2")
518 pnv_pci_init_p5ioc2_hub(np);
520 /* Look for ioda2 built-in PHB3's */
521 for_each_compatible_node(np, NULL, "ibm,ioda2-phb")
522 pnv_pci_init_ioda2_phb(np);
525 /* Setup the linkage between OF nodes and PHBs */
528 /* Configure IOMMU DMA hooks */
529 ppc_md.pci_dma_dev_setup = pnv_pci_dma_dev_setup;
530 ppc_md.tce_build = pnv_tce_build;
531 ppc_md.tce_free = pnv_tce_free;
532 ppc_md.tce_get = pnv_tce_get;
533 ppc_md.pci_probe_mode = pnv_pci_probe_mode;
534 set_pci_dma_ops(&dma_iommu_ops);
537 #ifdef CONFIG_PCI_MSI
538 ppc_md.msi_check_device = pnv_msi_check_device;
539 ppc_md.setup_msi_irqs = pnv_setup_msi_irqs;
540 ppc_md.teardown_msi_irqs = pnv_teardown_msi_irqs;