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.
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>
39 #define define_pe_printk_level(func, kern_level) \
40 static int func(const struct pnv_ioda_pe *pe, const char *fmt, ...) \
42 struct va_format vaf; \
47 va_start(args, fmt); \
53 strlcpy(pfix, dev_name(&pe->pdev->dev), \
56 sprintf(pfix, "%04x:%02x ", \
57 pci_domain_nr(pe->pbus), \
59 r = printk(kern_level "pci %s: [PE# %.3d] %pV", \
60 pfix, pe->pe_number, &vaf); \
67 define_pe_printk_level(pe_err, KERN_ERR);
68 define_pe_printk_level(pe_warn, KERN_WARNING);
69 define_pe_printk_level(pe_info, KERN_INFO);
71 static int pnv_ioda_alloc_pe(struct pnv_phb *phb)
76 pe = find_next_zero_bit(phb->ioda.pe_alloc,
77 phb->ioda.total_pe, 0);
78 if (pe >= phb->ioda.total_pe)
79 return IODA_INVALID_PE;
80 } while(test_and_set_bit(pe, phb->ioda.pe_alloc));
82 phb->ioda.pe_array[pe].phb = phb;
83 phb->ioda.pe_array[pe].pe_number = pe;
87 static void pnv_ioda_free_pe(struct pnv_phb *phb, int pe)
89 WARN_ON(phb->ioda.pe_array[pe].pdev);
91 memset(&phb->ioda.pe_array[pe], 0, sizeof(struct pnv_ioda_pe));
92 clear_bit(pe, phb->ioda.pe_alloc);
95 /* Currently those 2 are only used when MSIs are enabled, this will change
96 * but in the meantime, we need to protect them to avoid warnings
99 static struct pnv_ioda_pe *pnv_ioda_get_pe(struct pci_dev *dev)
101 struct pci_controller *hose = pci_bus_to_host(dev->bus);
102 struct pnv_phb *phb = hose->private_data;
103 struct pci_dn *pdn = pci_get_pdn(dev);
107 if (pdn->pe_number == IODA_INVALID_PE)
109 return &phb->ioda.pe_array[pdn->pe_number];
111 #endif /* CONFIG_PCI_MSI */
113 static int pnv_ioda_configure_pe(struct pnv_phb *phb, struct pnv_ioda_pe *pe)
115 struct pci_dev *parent;
116 uint8_t bcomp, dcomp, fcomp;
117 long rc, rid_end, rid;
119 /* Bus validation ? */
123 dcomp = OPAL_IGNORE_RID_DEVICE_NUMBER;
124 fcomp = OPAL_IGNORE_RID_FUNCTION_NUMBER;
125 parent = pe->pbus->self;
126 if (pe->flags & PNV_IODA_PE_BUS_ALL)
127 count = pe->pbus->busn_res.end - pe->pbus->busn_res.start + 1;
132 case 1: bcomp = OpalPciBusAll; break;
133 case 2: bcomp = OpalPciBus7Bits; break;
134 case 4: bcomp = OpalPciBus6Bits; break;
135 case 8: bcomp = OpalPciBus5Bits; break;
136 case 16: bcomp = OpalPciBus4Bits; break;
137 case 32: bcomp = OpalPciBus3Bits; break;
139 pr_err("%s: Number of subordinate busses %d"
141 pci_name(pe->pbus->self), count);
142 /* Do an exact match only */
143 bcomp = OpalPciBusAll;
145 rid_end = pe->rid + (count << 8);
147 parent = pe->pdev->bus->self;
148 bcomp = OpalPciBusAll;
149 dcomp = OPAL_COMPARE_RID_DEVICE_NUMBER;
150 fcomp = OPAL_COMPARE_RID_FUNCTION_NUMBER;
151 rid_end = pe->rid + 1;
155 * Associate PE in PELT. We need add the PE into the
156 * corresponding PELT-V as well. Otherwise, the error
157 * originated from the PE might contribute to other
160 rc = opal_pci_set_pe(phb->opal_id, pe->pe_number, pe->rid,
161 bcomp, dcomp, fcomp, OPAL_MAP_PE);
163 pe_err(pe, "OPAL error %ld trying to setup PELT table\n", rc);
167 rc = opal_pci_set_peltv(phb->opal_id, pe->pe_number,
168 pe->pe_number, OPAL_ADD_PE_TO_DOMAIN);
170 pe_warn(pe, "OPAL error %d adding self to PELTV\n", rc);
171 opal_pci_eeh_freeze_clear(phb->opal_id, pe->pe_number,
172 OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
174 /* Add to all parents PELT-V */
176 struct pci_dn *pdn = pci_get_pdn(parent);
177 if (pdn && pdn->pe_number != IODA_INVALID_PE) {
178 rc = opal_pci_set_peltv(phb->opal_id, pdn->pe_number,
179 pe->pe_number, OPAL_ADD_PE_TO_DOMAIN);
180 /* XXX What to do in case of error ? */
182 parent = parent->bus->self;
184 /* Setup reverse map */
185 for (rid = pe->rid; rid < rid_end; rid++)
186 phb->ioda.pe_rmap[rid] = pe->pe_number;
188 /* Setup one MVTs on IODA1 */
189 if (phb->type == PNV_PHB_IODA1) {
190 pe->mve_number = pe->pe_number;
191 rc = opal_pci_set_mve(phb->opal_id, pe->mve_number,
194 pe_err(pe, "OPAL error %ld setting up MVE %d\n",
198 rc = opal_pci_set_mve_enable(phb->opal_id,
199 pe->mve_number, OPAL_ENABLE_MVE);
201 pe_err(pe, "OPAL error %ld enabling MVE %d\n",
206 } else if (phb->type == PNV_PHB_IODA2)
212 static void pnv_ioda_link_pe_by_weight(struct pnv_phb *phb,
213 struct pnv_ioda_pe *pe)
215 struct pnv_ioda_pe *lpe;
217 list_for_each_entry(lpe, &phb->ioda.pe_dma_list, dma_link) {
218 if (lpe->dma_weight < pe->dma_weight) {
219 list_add_tail(&pe->dma_link, &lpe->dma_link);
223 list_add_tail(&pe->dma_link, &phb->ioda.pe_dma_list);
226 static unsigned int pnv_ioda_dma_weight(struct pci_dev *dev)
228 /* This is quite simplistic. The "base" weight of a device
229 * is 10. 0 means no DMA is to be accounted for it.
232 /* If it's a bridge, no DMA */
233 if (dev->hdr_type != PCI_HEADER_TYPE_NORMAL)
236 /* Reduce the weight of slow USB controllers */
237 if (dev->class == PCI_CLASS_SERIAL_USB_UHCI ||
238 dev->class == PCI_CLASS_SERIAL_USB_OHCI ||
239 dev->class == PCI_CLASS_SERIAL_USB_EHCI)
242 /* Increase the weight of RAID (includes Obsidian) */
243 if ((dev->class >> 8) == PCI_CLASS_STORAGE_RAID)
251 static struct pnv_ioda_pe *pnv_ioda_setup_dev_PE(struct pci_dev *dev)
253 struct pci_controller *hose = pci_bus_to_host(dev->bus);
254 struct pnv_phb *phb = hose->private_data;
255 struct pci_dn *pdn = pci_get_pdn(dev);
256 struct pnv_ioda_pe *pe;
260 pr_err("%s: Device tree node not associated properly\n",
264 if (pdn->pe_number != IODA_INVALID_PE)
267 /* PE#0 has been pre-set */
268 if (dev->bus->number == 0)
271 pe_num = pnv_ioda_alloc_pe(phb);
272 if (pe_num == IODA_INVALID_PE) {
273 pr_warning("%s: Not enough PE# available, disabling device\n",
278 /* NOTE: We get only one ref to the pci_dev for the pdn, not for the
279 * pointer in the PE data structure, both should be destroyed at the
280 * same time. However, this needs to be looked at more closely again
281 * once we actually start removing things (Hotplug, SR-IOV, ...)
283 * At some point we want to remove the PDN completely anyways
285 pe = &phb->ioda.pe_array[pe_num];
288 pdn->pe_number = pe_num;
293 pe->rid = dev->bus->number << 8 | pdn->devfn;
295 pe_info(pe, "Associated device to PE\n");
297 if (pnv_ioda_configure_pe(phb, pe)) {
298 /* XXX What do we do here ? */
300 pnv_ioda_free_pe(phb, pe_num);
301 pdn->pe_number = IODA_INVALID_PE;
307 /* Assign a DMA weight to the device */
308 pe->dma_weight = pnv_ioda_dma_weight(dev);
309 if (pe->dma_weight != 0) {
310 phb->ioda.dma_weight += pe->dma_weight;
311 phb->ioda.dma_pe_count++;
315 pnv_ioda_link_pe_by_weight(phb, pe);
319 #endif /* Useful for SRIOV case */
321 static void pnv_ioda_setup_same_PE(struct pci_bus *bus, struct pnv_ioda_pe *pe)
325 list_for_each_entry(dev, &bus->devices, bus_list) {
326 struct pci_dn *pdn = pci_get_pdn(dev);
329 pr_warn("%s: No device node associated with device !\n",
335 pdn->pe_number = pe->pe_number;
336 pe->dma_weight += pnv_ioda_dma_weight(dev);
337 if ((pe->flags & PNV_IODA_PE_BUS_ALL) && dev->subordinate)
338 pnv_ioda_setup_same_PE(dev->subordinate, pe);
343 * There're 2 types of PCI bus sensitive PEs: One that is compromised of
344 * single PCI bus. Another one that contains the primary PCI bus and its
345 * subordinate PCI devices and buses. The second type of PE is normally
346 * orgiriated by PCIe-to-PCI bridge or PLX switch downstream ports.
348 static void pnv_ioda_setup_bus_PE(struct pci_bus *bus, int all)
350 struct pci_controller *hose = pci_bus_to_host(bus);
351 struct pnv_phb *phb = hose->private_data;
352 struct pnv_ioda_pe *pe;
355 pe_num = pnv_ioda_alloc_pe(phb);
356 if (pe_num == IODA_INVALID_PE) {
357 pr_warning("%s: Not enough PE# available for PCI bus %04x:%02x\n",
358 __func__, pci_domain_nr(bus), bus->number);
362 pe = &phb->ioda.pe_array[pe_num];
363 pe->flags = (all ? PNV_IODA_PE_BUS_ALL : PNV_IODA_PE_BUS);
368 pe->rid = bus->busn_res.start << 8;
372 pe_info(pe, "Secondary bus %d..%d associated with PE#%d\n",
373 bus->busn_res.start, bus->busn_res.end, pe_num);
375 pe_info(pe, "Secondary bus %d associated with PE#%d\n",
376 bus->busn_res.start, pe_num);
378 if (pnv_ioda_configure_pe(phb, pe)) {
379 /* XXX What do we do here ? */
381 pnv_ioda_free_pe(phb, pe_num);
386 /* Associate it with all child devices */
387 pnv_ioda_setup_same_PE(bus, pe);
389 /* Put PE to the list */
390 list_add_tail(&pe->list, &phb->ioda.pe_list);
392 /* Account for one DMA PE if at least one DMA capable device exist
395 if (pe->dma_weight != 0) {
396 phb->ioda.dma_weight += pe->dma_weight;
397 phb->ioda.dma_pe_count++;
401 pnv_ioda_link_pe_by_weight(phb, pe);
404 static void pnv_ioda_setup_PEs(struct pci_bus *bus)
408 pnv_ioda_setup_bus_PE(bus, 0);
410 list_for_each_entry(dev, &bus->devices, bus_list) {
411 if (dev->subordinate) {
412 if (pci_pcie_type(dev) == PCI_EXP_TYPE_PCI_BRIDGE)
413 pnv_ioda_setup_bus_PE(dev->subordinate, 1);
415 pnv_ioda_setup_PEs(dev->subordinate);
421 * Configure PEs so that the downstream PCI buses and devices
422 * could have their associated PE#. Unfortunately, we didn't
423 * figure out the way to identify the PLX bridge yet. So we
424 * simply put the PCI bus and the subordinate behind the root
425 * port to PE# here. The game rule here is expected to be changed
426 * as soon as we can detected PLX bridge correctly.
428 static void pnv_pci_ioda_setup_PEs(void)
430 struct pci_controller *hose, *tmp;
432 list_for_each_entry_safe(hose, tmp, &hose_list, list_node) {
433 pnv_ioda_setup_PEs(hose->bus);
437 static void pnv_pci_ioda_dma_dev_setup(struct pnv_phb *phb, struct pci_dev *pdev)
439 struct pci_dn *pdn = pci_get_pdn(pdev);
440 struct pnv_ioda_pe *pe;
443 * The function can be called while the PE#
444 * hasn't been assigned. Do nothing for the
447 if (!pdn || pdn->pe_number == IODA_INVALID_PE)
450 pe = &phb->ioda.pe_array[pdn->pe_number];
451 set_iommu_table_base(&pdev->dev, &pe->tce32_table);
454 static void pnv_ioda_setup_bus_dma(struct pnv_ioda_pe *pe, struct pci_bus *bus)
458 list_for_each_entry(dev, &bus->devices, bus_list) {
459 set_iommu_table_base(&dev->dev, &pe->tce32_table);
460 if (dev->subordinate)
461 pnv_ioda_setup_bus_dma(pe, dev->subordinate);
465 static void pnv_pci_ioda1_tce_invalidate(struct iommu_table *tbl,
466 u64 *startp, u64 *endp)
468 u64 __iomem *invalidate = (u64 __iomem *)tbl->it_index;
469 unsigned long start, end, inc;
471 start = __pa(startp);
474 /* BML uses this case for p6/p7/galaxy2: Shift addr and put in node */
479 start |= tbl->it_busno;
480 end |= tbl->it_busno;
481 } else if (tbl->it_type & TCE_PCI_SWINV_PAIR) {
482 /* p7ioc-style invalidation, 2 TCEs per write */
483 start |= (1ull << 63);
487 /* Default (older HW) */
491 end |= inc - 1; /* round up end to be different than start */
493 mb(); /* Ensure above stores are visible */
494 while (start <= end) {
495 __raw_writeq(start, invalidate);
500 * The iommu layer will do another mb() for us on build()
501 * and we don't care on free()
505 static void pnv_pci_ioda2_tce_invalidate(struct pnv_ioda_pe *pe,
506 struct iommu_table *tbl,
507 u64 *startp, u64 *endp)
509 unsigned long start, end, inc;
510 u64 __iomem *invalidate = (u64 __iomem *)tbl->it_index;
512 /* We'll invalidate DMA address in PE scope */
514 start |= (pe->pe_number & 0xFF);
517 /* Figure out the start, end and step */
518 inc = tbl->it_offset + (((u64)startp - tbl->it_base) / sizeof(u64));
519 start |= (inc << 12);
520 inc = tbl->it_offset + (((u64)endp - tbl->it_base) / sizeof(u64));
525 while (start <= end) {
526 __raw_writeq(start, invalidate);
531 void pnv_pci_ioda_tce_invalidate(struct iommu_table *tbl,
532 u64 *startp, u64 *endp)
534 struct pnv_ioda_pe *pe = container_of(tbl, struct pnv_ioda_pe,
536 struct pnv_phb *phb = pe->phb;
538 if (phb->type == PNV_PHB_IODA1)
539 pnv_pci_ioda1_tce_invalidate(tbl, startp, endp);
541 pnv_pci_ioda2_tce_invalidate(pe, tbl, startp, endp);
544 static void pnv_pci_ioda_setup_dma_pe(struct pnv_phb *phb,
545 struct pnv_ioda_pe *pe, unsigned int base,
549 struct page *tce_mem = NULL;
550 const __be64 *swinvp;
551 struct iommu_table *tbl;
556 /* 256M DMA window, 4K TCE pages, 8 bytes TCE */
557 #define TCE32_TABLE_SIZE ((0x10000000 / 0x1000) * 8)
559 /* XXX FIXME: Handle 64-bit only DMA devices */
560 /* XXX FIXME: Provide 64-bit DMA facilities & non-4K TCE tables etc.. */
561 /* XXX FIXME: Allocate multi-level tables on PHB3 */
563 /* We shouldn't already have a 32-bit DMA associated */
564 if (WARN_ON(pe->tce32_seg >= 0))
567 /* Grab a 32-bit TCE table */
568 pe->tce32_seg = base;
569 pe_info(pe, " Setting up 32-bit TCE table at %08x..%08x\n",
570 (base << 28), ((base + segs) << 28) - 1);
572 /* XXX Currently, we allocate one big contiguous table for the
573 * TCEs. We only really need one chunk per 256M of TCE space
574 * (ie per segment) but that's an optimization for later, it
575 * requires some added smarts with our get/put_tce implementation
577 tce_mem = alloc_pages_node(phb->hose->node, GFP_KERNEL,
578 get_order(TCE32_TABLE_SIZE * segs));
580 pe_err(pe, " Failed to allocate a 32-bit TCE memory\n");
583 addr = page_address(tce_mem);
584 memset(addr, 0, TCE32_TABLE_SIZE * segs);
587 for (i = 0; i < segs; i++) {
588 rc = opal_pci_map_pe_dma_window(phb->opal_id,
591 __pa(addr) + TCE32_TABLE_SIZE * i,
592 TCE32_TABLE_SIZE, 0x1000);
594 pe_err(pe, " Failed to configure 32-bit TCE table,"
600 /* Setup linux iommu table */
601 tbl = &pe->tce32_table;
602 pnv_pci_setup_iommu_table(tbl, addr, TCE32_TABLE_SIZE * segs,
605 /* OPAL variant of P7IOC SW invalidated TCEs */
606 swinvp = of_get_property(phb->hose->dn, "ibm,opal-tce-kill", NULL);
608 /* We need a couple more fields -- an address and a data
609 * to or. Since the bus is only printed out on table free
610 * errors, and on the first pass the data will be a relative
611 * bus number, print that out instead.
614 tbl->it_index = (unsigned long)ioremap(be64_to_cpup(swinvp), 8);
615 tbl->it_type = TCE_PCI_SWINV_CREATE | TCE_PCI_SWINV_FREE |
618 iommu_init_table(tbl, phb->hose->node);
621 set_iommu_table_base(&pe->pdev->dev, tbl);
623 pnv_ioda_setup_bus_dma(pe, pe->pbus);
627 /* XXX Failure: Try to fallback to 64-bit only ? */
628 if (pe->tce32_seg >= 0)
631 __free_pages(tce_mem, get_order(TCE32_TABLE_SIZE * segs));
634 static void pnv_pci_ioda2_setup_dma_pe(struct pnv_phb *phb,
635 struct pnv_ioda_pe *pe)
637 struct page *tce_mem = NULL;
639 const __be64 *swinvp;
640 struct iommu_table *tbl;
641 unsigned int tce_table_size, end;
644 /* We shouldn't already have a 32-bit DMA associated */
645 if (WARN_ON(pe->tce32_seg >= 0))
648 /* The PE will reserve all possible 32-bits space */
650 end = (1 << ilog2(phb->ioda.m32_pci_base));
651 tce_table_size = (end / 0x1000) * 8;
652 pe_info(pe, "Setting up 32-bit TCE table at 0..%08x\n",
655 /* Allocate TCE table */
656 tce_mem = alloc_pages_node(phb->hose->node, GFP_KERNEL,
657 get_order(tce_table_size));
659 pe_err(pe, "Failed to allocate a 32-bit TCE memory\n");
662 addr = page_address(tce_mem);
663 memset(addr, 0, tce_table_size);
666 * Map TCE table through TVT. The TVE index is the PE number
667 * shifted by 1 bit for 32-bits DMA space.
669 rc = opal_pci_map_pe_dma_window(phb->opal_id, pe->pe_number,
670 pe->pe_number << 1, 1, __pa(addr),
671 tce_table_size, 0x1000);
673 pe_err(pe, "Failed to configure 32-bit TCE table,"
678 /* Setup linux iommu table */
679 tbl = &pe->tce32_table;
680 pnv_pci_setup_iommu_table(tbl, addr, tce_table_size, 0);
682 /* OPAL variant of PHB3 invalidated TCEs */
683 swinvp = of_get_property(phb->hose->dn, "ibm,opal-tce-kill", NULL);
685 /* We need a couple more fields -- an address and a data
686 * to or. Since the bus is only printed out on table free
687 * errors, and on the first pass the data will be a relative
688 * bus number, print that out instead.
691 tbl->it_index = (unsigned long)ioremap(be64_to_cpup(swinvp), 8);
692 tbl->it_type = TCE_PCI_SWINV_CREATE | TCE_PCI_SWINV_FREE;
694 iommu_init_table(tbl, phb->hose->node);
697 set_iommu_table_base(&pe->pdev->dev, tbl);
699 pnv_ioda_setup_bus_dma(pe, pe->pbus);
703 if (pe->tce32_seg >= 0)
706 __free_pages(tce_mem, get_order(tce_table_size));
709 static void pnv_ioda_setup_dma(struct pnv_phb *phb)
711 struct pci_controller *hose = phb->hose;
712 unsigned int residual, remaining, segs, tw, base;
713 struct pnv_ioda_pe *pe;
715 /* If we have more PE# than segments available, hand out one
716 * per PE until we run out and let the rest fail. If not,
717 * then we assign at least one segment per PE, plus more based
718 * on the amount of devices under that PE
720 if (phb->ioda.dma_pe_count > phb->ioda.tce32_count)
723 residual = phb->ioda.tce32_count -
724 phb->ioda.dma_pe_count;
726 pr_info("PCI: Domain %04x has %ld available 32-bit DMA segments\n",
727 hose->global_number, phb->ioda.tce32_count);
728 pr_info("PCI: %d PE# for a total weight of %d\n",
729 phb->ioda.dma_pe_count, phb->ioda.dma_weight);
731 /* Walk our PE list and configure their DMA segments, hand them
732 * out one base segment plus any residual segments based on
735 remaining = phb->ioda.tce32_count;
736 tw = phb->ioda.dma_weight;
738 list_for_each_entry(pe, &phb->ioda.pe_dma_list, dma_link) {
742 pe_warn(pe, "No DMA32 resources available\n");
747 segs += ((pe->dma_weight * residual) + (tw / 2)) / tw;
748 if (segs > remaining)
753 * For IODA2 compliant PHB3, we needn't care about the weight.
754 * The all available 32-bits DMA space will be assigned to
757 if (phb->type == PNV_PHB_IODA1) {
758 pe_info(pe, "DMA weight %d, assigned %d DMA32 segments\n",
759 pe->dma_weight, segs);
760 pnv_pci_ioda_setup_dma_pe(phb, pe, base, segs);
762 pe_info(pe, "Assign DMA32 space\n");
764 pnv_pci_ioda2_setup_dma_pe(phb, pe);
772 #ifdef CONFIG_PCI_MSI
773 static void pnv_ioda2_msi_eoi(struct irq_data *d)
775 unsigned int hw_irq = (unsigned int)irqd_to_hwirq(d);
776 struct irq_chip *chip = irq_data_get_irq_chip(d);
777 struct pnv_phb *phb = container_of(chip, struct pnv_phb,
781 rc = opal_pci_msi_eoi(phb->opal_id, hw_irq);
787 static int pnv_pci_ioda_msi_setup(struct pnv_phb *phb, struct pci_dev *dev,
788 unsigned int hwirq, unsigned int virq,
789 unsigned int is_64, struct msi_msg *msg)
791 struct pnv_ioda_pe *pe = pnv_ioda_get_pe(dev);
792 struct pci_dn *pdn = pci_get_pdn(dev);
793 struct irq_data *idata;
794 struct irq_chip *ichip;
795 unsigned int xive_num = hwirq - phb->msi_base;
797 uint32_t addr32, data;
800 /* No PE assigned ? bail out ... no MSI for you ! */
804 /* Check if we have an MVE */
805 if (pe->mve_number < 0)
808 /* Force 32-bit MSI on some broken devices */
809 if (pdn && pdn->force_32bit_msi)
812 /* Assign XIVE to PE */
813 rc = opal_pci_set_xive_pe(phb->opal_id, pe->pe_number, xive_num);
815 pr_warn("%s: OPAL error %d setting XIVE %d PE\n",
816 pci_name(dev), rc, xive_num);
821 rc = opal_get_msi_64(phb->opal_id, pe->mve_number, xive_num, 1,
824 pr_warn("%s: OPAL error %d getting 64-bit MSI data\n",
828 msg->address_hi = addr64 >> 32;
829 msg->address_lo = addr64 & 0xfffffffful;
831 rc = opal_get_msi_32(phb->opal_id, pe->mve_number, xive_num, 1,
834 pr_warn("%s: OPAL error %d getting 32-bit MSI data\n",
839 msg->address_lo = addr32;
844 * Change the IRQ chip for the MSI interrupts on PHB3.
845 * The corresponding IRQ chip should be populated for
848 if (phb->type == PNV_PHB_IODA2) {
849 if (!phb->ioda.irq_chip_init) {
850 idata = irq_get_irq_data(virq);
851 ichip = irq_data_get_irq_chip(idata);
852 phb->ioda.irq_chip_init = 1;
853 phb->ioda.irq_chip = *ichip;
854 phb->ioda.irq_chip.irq_eoi = pnv_ioda2_msi_eoi;
857 irq_set_chip(virq, &phb->ioda.irq_chip);
860 pr_devel("%s: %s-bit MSI on hwirq %x (xive #%d),"
861 " address=%x_%08x data=%x PE# %d\n",
862 pci_name(dev), is_64 ? "64" : "32", hwirq, xive_num,
863 msg->address_hi, msg->address_lo, data, pe->pe_number);
868 static void pnv_pci_init_ioda_msis(struct pnv_phb *phb)
871 const __be32 *prop = of_get_property(phb->hose->dn,
872 "ibm,opal-msi-ranges", NULL);
875 prop = of_get_property(phb->hose->dn, "msi-ranges", NULL);
880 phb->msi_base = be32_to_cpup(prop);
881 count = be32_to_cpup(prop + 1);
882 if (msi_bitmap_alloc(&phb->msi_bmp, count, phb->hose->dn)) {
883 pr_err("PCI %d: Failed to allocate MSI bitmap !\n",
884 phb->hose->global_number);
888 phb->msi_setup = pnv_pci_ioda_msi_setup;
889 phb->msi32_support = 1;
890 pr_info(" Allocated bitmap for %d MSIs (base IRQ 0x%x)\n",
891 count, phb->msi_base);
894 static void pnv_pci_init_ioda_msis(struct pnv_phb *phb) { }
895 #endif /* CONFIG_PCI_MSI */
898 * This function is supposed to be called on basis of PE from top
899 * to bottom style. So the the I/O or MMIO segment assigned to
900 * parent PE could be overrided by its child PEs if necessary.
902 static void pnv_ioda_setup_pe_seg(struct pci_controller *hose,
903 struct pnv_ioda_pe *pe)
905 struct pnv_phb *phb = hose->private_data;
906 struct pci_bus_region region;
907 struct resource *res;
912 * NOTE: We only care PCI bus based PE for now. For PCI
913 * device based PE, for example SRIOV sensitive VF should
914 * be figured out later.
916 BUG_ON(!(pe->flags & (PNV_IODA_PE_BUS | PNV_IODA_PE_BUS_ALL)));
918 pci_bus_for_each_resource(pe->pbus, res, i) {
919 if (!res || !res->flags ||
920 res->start > res->end)
923 if (res->flags & IORESOURCE_IO) {
924 region.start = res->start - phb->ioda.io_pci_base;
925 region.end = res->end - phb->ioda.io_pci_base;
926 index = region.start / phb->ioda.io_segsize;
928 while (index < phb->ioda.total_pe &&
929 region.start <= region.end) {
930 phb->ioda.io_segmap[index] = pe->pe_number;
931 rc = opal_pci_map_pe_mmio_window(phb->opal_id,
932 pe->pe_number, OPAL_IO_WINDOW_TYPE, 0, index);
933 if (rc != OPAL_SUCCESS) {
934 pr_err("%s: OPAL error %d when mapping IO "
935 "segment #%d to PE#%d\n",
936 __func__, rc, index, pe->pe_number);
940 region.start += phb->ioda.io_segsize;
943 } else if (res->flags & IORESOURCE_MEM) {
944 /* WARNING: Assumes M32 is mem region 0 in PHB. We need to
945 * harden that algorithm when we start supporting M64
947 region.start = res->start -
948 hose->mem_offset[0] -
949 phb->ioda.m32_pci_base;
950 region.end = res->end -
951 hose->mem_offset[0] -
952 phb->ioda.m32_pci_base;
953 index = region.start / phb->ioda.m32_segsize;
955 while (index < phb->ioda.total_pe &&
956 region.start <= region.end) {
957 phb->ioda.m32_segmap[index] = pe->pe_number;
958 rc = opal_pci_map_pe_mmio_window(phb->opal_id,
959 pe->pe_number, OPAL_M32_WINDOW_TYPE, 0, index);
960 if (rc != OPAL_SUCCESS) {
961 pr_err("%s: OPAL error %d when mapping M32 "
962 "segment#%d to PE#%d",
963 __func__, rc, index, pe->pe_number);
967 region.start += phb->ioda.m32_segsize;
974 static void pnv_pci_ioda_setup_seg(void)
976 struct pci_controller *tmp, *hose;
978 struct pnv_ioda_pe *pe;
980 list_for_each_entry_safe(hose, tmp, &hose_list, list_node) {
981 phb = hose->private_data;
982 list_for_each_entry(pe, &phb->ioda.pe_list, list) {
983 pnv_ioda_setup_pe_seg(hose, pe);
988 static void pnv_pci_ioda_setup_DMA(void)
990 struct pci_controller *hose, *tmp;
993 list_for_each_entry_safe(hose, tmp, &hose_list, list_node) {
994 pnv_ioda_setup_dma(hose->private_data);
996 /* Mark the PHB initialization done */
997 phb = hose->private_data;
998 phb->initialized = 1;
1002 static void pnv_pci_ioda_fixup(void)
1004 pnv_pci_ioda_setup_PEs();
1005 pnv_pci_ioda_setup_seg();
1006 pnv_pci_ioda_setup_DMA();
1010 * Returns the alignment for I/O or memory windows for P2P
1011 * bridges. That actually depends on how PEs are segmented.
1012 * For now, we return I/O or M32 segment size for PE sensitive
1013 * P2P bridges. Otherwise, the default values (4KiB for I/O,
1014 * 1MiB for memory) will be returned.
1016 * The current PCI bus might be put into one PE, which was
1017 * create against the parent PCI bridge. For that case, we
1018 * needn't enlarge the alignment so that we can save some
1021 static resource_size_t pnv_pci_window_alignment(struct pci_bus *bus,
1024 struct pci_dev *bridge;
1025 struct pci_controller *hose = pci_bus_to_host(bus);
1026 struct pnv_phb *phb = hose->private_data;
1027 int num_pci_bridges = 0;
1031 if (pci_pcie_type(bridge) == PCI_EXP_TYPE_PCI_BRIDGE) {
1033 if (num_pci_bridges >= 2)
1037 bridge = bridge->bus->self;
1040 /* We need support prefetchable memory window later */
1041 if (type & IORESOURCE_MEM)
1042 return phb->ioda.m32_segsize;
1044 return phb->ioda.io_segsize;
1047 /* Prevent enabling devices for which we couldn't properly
1050 static int pnv_pci_enable_device_hook(struct pci_dev *dev)
1052 struct pci_controller *hose = pci_bus_to_host(dev->bus);
1053 struct pnv_phb *phb = hose->private_data;
1056 /* The function is probably called while the PEs have
1057 * not be created yet. For example, resource reassignment
1058 * during PCI probe period. We just skip the check if
1061 if (!phb->initialized)
1064 pdn = pci_get_pdn(dev);
1065 if (!pdn || pdn->pe_number == IODA_INVALID_PE)
1071 static u32 pnv_ioda_bdfn_to_pe(struct pnv_phb *phb, struct pci_bus *bus,
1074 return phb->ioda.pe_rmap[(bus->number << 8) | devfn];
1077 static void pnv_pci_ioda_shutdown(struct pnv_phb *phb)
1079 opal_pci_reset(phb->opal_id, OPAL_PCI_IODA_TABLE_RESET,
1083 void __init pnv_pci_init_ioda_phb(struct device_node *np, int ioda_type)
1085 struct pci_controller *hose;
1086 static int primary = 1;
1087 struct pnv_phb *phb;
1088 unsigned long size, m32map_off, iomap_off, pemap_off;
1095 pr_info(" Initializing IODA%d OPAL PHB %s\n", ioda_type, np->full_name);
1097 prop64 = of_get_property(np, "ibm,opal-phbid", NULL);
1099 pr_err(" Missing \"ibm,opal-phbid\" property !\n");
1102 phb_id = be64_to_cpup(prop64);
1103 pr_debug(" PHB-ID : 0x%016llx\n", phb_id);
1105 phb = alloc_bootmem(sizeof(struct pnv_phb));
1107 memset(phb, 0, sizeof(struct pnv_phb));
1108 phb->hose = hose = pcibios_alloc_controller(np);
1110 if (!phb || !phb->hose) {
1111 pr_err("PCI: Failed to allocate PCI controller for %s\n",
1116 spin_lock_init(&phb->lock);
1117 /* XXX Use device-tree */
1118 hose->first_busno = 0;
1119 hose->last_busno = 0xff;
1120 hose->private_data = phb;
1121 phb->opal_id = phb_id;
1122 phb->type = ioda_type;
1124 /* Detect specific models for error handling */
1125 if (of_device_is_compatible(np, "ibm,p7ioc-pciex"))
1126 phb->model = PNV_PHB_MODEL_P7IOC;
1127 else if (of_device_is_compatible(np, "ibm,power8-pciex"))
1128 phb->model = PNV_PHB_MODEL_PHB3;
1130 phb->model = PNV_PHB_MODEL_UNKNOWN;
1132 /* Parse 32-bit and IO ranges (if any) */
1133 pci_process_bridge_OF_ranges(phb->hose, np, primary);
1137 phb->regs = of_iomap(np, 0);
1138 if (phb->regs == NULL)
1139 pr_err(" Failed to map registers !\n");
1141 /* Initialize more IODA stuff */
1142 prop32 = of_get_property(np, "ibm,opal-num-pes", NULL);
1144 phb->ioda.total_pe = 1;
1146 phb->ioda.total_pe = *prop32;
1148 phb->ioda.m32_size = resource_size(&hose->mem_resources[0]);
1149 /* FW Has already off top 64k of M32 space (MSI space) */
1150 phb->ioda.m32_size += 0x10000;
1152 phb->ioda.m32_segsize = phb->ioda.m32_size / phb->ioda.total_pe;
1153 phb->ioda.m32_pci_base = hose->mem_resources[0].start - hose->mem_offset[0];
1154 phb->ioda.io_size = hose->pci_io_size;
1155 phb->ioda.io_segsize = phb->ioda.io_size / phb->ioda.total_pe;
1156 phb->ioda.io_pci_base = 0; /* XXX calculate this ? */
1158 /* Allocate aux data & arrays
1160 * XXX TODO: Don't allocate io segmap on PHB3
1162 size = _ALIGN_UP(phb->ioda.total_pe / 8, sizeof(unsigned long));
1164 size += phb->ioda.total_pe * sizeof(phb->ioda.m32_segmap[0]);
1166 size += phb->ioda.total_pe * sizeof(phb->ioda.io_segmap[0]);
1168 size += phb->ioda.total_pe * sizeof(struct pnv_ioda_pe);
1169 aux = alloc_bootmem(size);
1170 memset(aux, 0, size);
1171 phb->ioda.pe_alloc = aux;
1172 phb->ioda.m32_segmap = aux + m32map_off;
1173 phb->ioda.io_segmap = aux + iomap_off;
1174 phb->ioda.pe_array = aux + pemap_off;
1175 set_bit(0, phb->ioda.pe_alloc);
1177 INIT_LIST_HEAD(&phb->ioda.pe_dma_list);
1178 INIT_LIST_HEAD(&phb->ioda.pe_list);
1180 /* Calculate how many 32-bit TCE segments we have */
1181 phb->ioda.tce32_count = phb->ioda.m32_pci_base >> 28;
1183 /* Clear unusable m64 */
1184 hose->mem_resources[1].flags = 0;
1185 hose->mem_resources[1].start = 0;
1186 hose->mem_resources[1].end = 0;
1187 hose->mem_resources[2].flags = 0;
1188 hose->mem_resources[2].start = 0;
1189 hose->mem_resources[2].end = 0;
1191 #if 0 /* We should really do that ... */
1192 rc = opal_pci_set_phb_mem_window(opal->phb_id,
1195 starting_real_address,
1196 starting_pci_address,
1200 pr_info(" %d PE's M32: 0x%x [segment=0x%x] IO: 0x%x [segment=0x%x]\n",
1202 phb->ioda.m32_size, phb->ioda.m32_segsize,
1203 phb->ioda.io_size, phb->ioda.io_segsize);
1205 phb->hose->ops = &pnv_pci_ops;
1207 /* Setup RID -> PE mapping function */
1208 phb->bdfn_to_pe = pnv_ioda_bdfn_to_pe;
1211 phb->dma_dev_setup = pnv_pci_ioda_dma_dev_setup;
1213 /* Setup shutdown function for kexec */
1214 phb->shutdown = pnv_pci_ioda_shutdown;
1216 /* Setup MSI support */
1217 pnv_pci_init_ioda_msis(phb);
1220 * We pass the PCI probe flag PCI_REASSIGN_ALL_RSRC here
1221 * to let the PCI core do resource assignment. It's supposed
1222 * that the PCI core will do correct I/O and MMIO alignment
1223 * for the P2P bridge bars so that each PCI bus (excluding
1224 * the child P2P bridges) can form individual PE.
1226 ppc_md.pcibios_fixup = pnv_pci_ioda_fixup;
1227 ppc_md.pcibios_enable_device_hook = pnv_pci_enable_device_hook;
1228 ppc_md.pcibios_window_alignment = pnv_pci_window_alignment;
1229 pci_add_flags(PCI_REASSIGN_ALL_RSRC);
1231 /* Reset IODA tables to a clean state */
1232 rc = opal_pci_reset(phb_id, OPAL_PCI_IODA_TABLE_RESET, OPAL_ASSERT_RESET);
1234 pr_warning(" OPAL Error %ld performing IODA table reset !\n", rc);
1237 * On IODA1 map everything to PE#0, on IODA2 we assume the IODA reset
1238 * has cleared the RTT which has the same effect
1240 if (ioda_type == PNV_PHB_IODA1)
1241 opal_pci_set_pe(phb_id, 0, 0, 7, 1, 1 , OPAL_MAP_PE);
1244 void pnv_pci_init_ioda2_phb(struct device_node *np)
1246 pnv_pci_init_ioda_phb(np, PNV_PHB_IODA2);
1249 void __init pnv_pci_init_ioda_hub(struct device_node *np)
1251 struct device_node *phbn;
1255 pr_info("Probing IODA IO-Hub %s\n", np->full_name);
1257 prop64 = of_get_property(np, "ibm,opal-hubid", NULL);
1259 pr_err(" Missing \"ibm,opal-hubid\" property !\n");
1262 hub_id = be64_to_cpup(prop64);
1263 pr_devel(" HUB-ID : 0x%016llx\n", hub_id);
1265 /* Count child PHBs */
1266 for_each_child_of_node(np, phbn) {
1267 /* Look for IODA1 PHBs */
1268 if (of_device_is_compatible(phbn, "ibm,ioda-phb"))
1269 pnv_pci_init_ioda_phb(phbn, PNV_PHB_IODA1);