1 /*
2  * Support PCI/PCIe on PowerNV platforms
3  *
4  * Currently supports only P5IOC2
5  *
6  * Copyright 2011 Benjamin Herrenschmidt, IBM Corp.
7  *
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.
12  */
13 
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>
21 #include <linux/io.h>
22 #include <linux/msi.h>
23 #include <linux/iommu.h>
24 
25 #include <asm/sections.h>
26 #include <asm/io.h>
27 #include <asm/prom.h>
28 #include <asm/pci-bridge.h>
29 #include <asm/machdep.h>
30 #include <asm/msi_bitmap.h>
31 #include <asm/ppc-pci.h>
32 #include <asm/opal.h>
33 #include <asm/iommu.h>
34 #include <asm/tce.h>
35 #include <asm/firmware.h>
36 #include <asm/eeh_event.h>
37 #include <asm/eeh.h>
38 
39 #include "powernv.h"
40 #include "pci.h"
41 
42 /* Delay in usec */
43 #define PCI_RESET_DELAY_US	3000000
44 
45 #define cfg_dbg(fmt...)	do { } while(0)
46 //#define cfg_dbg(fmt...)	printk(fmt)
47 
48 #ifdef CONFIG_PCI_MSI
49 static int pnv_msi_check_device(struct pci_dev* pdev, int nvec, int type)
50 {
51 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
52 	struct pnv_phb *phb = hose->private_data;
53 	struct pci_dn *pdn = pci_get_pdn(pdev);
54 
55 	if (pdn && pdn->force_32bit_msi && !phb->msi32_support)
56 		return -ENODEV;
57 
58 	return (phb && phb->msi_bmp.bitmap) ? 0 : -ENODEV;
59 }
60 
61 static int pnv_setup_msi_irqs(struct pci_dev *pdev, int nvec, int type)
62 {
63 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
64 	struct pnv_phb *phb = hose->private_data;
65 	struct msi_desc *entry;
66 	struct msi_msg msg;
67 	int hwirq;
68 	unsigned int virq;
69 	int rc;
70 
71 	if (WARN_ON(!phb))
72 		return -ENODEV;
73 
74 	list_for_each_entry(entry, &pdev->msi_list, list) {
75 		if (!entry->msi_attrib.is_64 && !phb->msi32_support) {
76 			pr_warn("%s: Supports only 64-bit MSIs\n",
77 				pci_name(pdev));
78 			return -ENXIO;
79 		}
80 		hwirq = msi_bitmap_alloc_hwirqs(&phb->msi_bmp, 1);
81 		if (hwirq < 0) {
82 			pr_warn("%s: Failed to find a free MSI\n",
83 				pci_name(pdev));
84 			return -ENOSPC;
85 		}
86 		virq = irq_create_mapping(NULL, phb->msi_base + hwirq);
87 		if (virq == NO_IRQ) {
88 			pr_warn("%s: Failed to map MSI to linux irq\n",
89 				pci_name(pdev));
90 			msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq, 1);
91 			return -ENOMEM;
92 		}
93 		rc = phb->msi_setup(phb, pdev, phb->msi_base + hwirq,
94 				    virq, entry->msi_attrib.is_64, &msg);
95 		if (rc) {
96 			pr_warn("%s: Failed to setup MSI\n", pci_name(pdev));
97 			irq_dispose_mapping(virq);
98 			msi_bitmap_free_hwirqs(&phb->msi_bmp, hwirq, 1);
99 			return rc;
100 		}
101 		irq_set_msi_desc(virq, entry);
102 		write_msi_msg(virq, &msg);
103 	}
104 	return 0;
105 }
106 
107 static void pnv_teardown_msi_irqs(struct pci_dev *pdev)
108 {
109 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
110 	struct pnv_phb *phb = hose->private_data;
111 	struct msi_desc *entry;
112 
113 	if (WARN_ON(!phb))
114 		return;
115 
116 	list_for_each_entry(entry, &pdev->msi_list, list) {
117 		if (entry->irq == NO_IRQ)
118 			continue;
119 		irq_set_msi_desc(entry->irq, NULL);
120 		msi_bitmap_free_hwirqs(&phb->msi_bmp,
121 			virq_to_hw(entry->irq) - phb->msi_base, 1);
122 		irq_dispose_mapping(entry->irq);
123 	}
124 }
125 #endif /* CONFIG_PCI_MSI */
126 
127 static void pnv_pci_dump_p7ioc_diag_data(struct pci_controller *hose,
128 					 struct OpalIoPhbErrorCommon *common)
129 {
130 	struct OpalIoP7IOCPhbErrorData *data;
131 	int i;
132 
133 	data = (struct OpalIoP7IOCPhbErrorData *)common;
134 	pr_info("P7IOC PHB#%d Diag-data (Version: %d)\n\n",
135 		hose->global_number, common->version);
136 
137 	pr_info("  brdgCtl:              %08x\n", data->brdgCtl);
138 
139 	pr_info("  portStatusReg:        %08x\n", data->portStatusReg);
140 	pr_info("  rootCmplxStatus:      %08x\n", data->rootCmplxStatus);
141 	pr_info("  busAgentStatus:       %08x\n", data->busAgentStatus);
142 
143 	pr_info("  deviceStatus:         %08x\n", data->deviceStatus);
144 	pr_info("  slotStatus:           %08x\n", data->slotStatus);
145 	pr_info("  linkStatus:           %08x\n", data->linkStatus);
146 	pr_info("  devCmdStatus:         %08x\n", data->devCmdStatus);
147 	pr_info("  devSecStatus:         %08x\n", data->devSecStatus);
148 
149 	pr_info("  rootErrorStatus:      %08x\n", data->rootErrorStatus);
150 	pr_info("  uncorrErrorStatus:    %08x\n", data->uncorrErrorStatus);
151 	pr_info("  corrErrorStatus:      %08x\n", data->corrErrorStatus);
152 	pr_info("  tlpHdr1:              %08x\n", data->tlpHdr1);
153 	pr_info("  tlpHdr2:              %08x\n", data->tlpHdr2);
154 	pr_info("  tlpHdr3:              %08x\n", data->tlpHdr3);
155 	pr_info("  tlpHdr4:              %08x\n", data->tlpHdr4);
156 	pr_info("  sourceId:             %08x\n", data->sourceId);
157 	pr_info("  errorClass:           %016llx\n", data->errorClass);
158 	pr_info("  correlator:           %016llx\n", data->correlator);
159 	pr_info("  p7iocPlssr:           %016llx\n", data->p7iocPlssr);
160 	pr_info("  p7iocCsr:             %016llx\n", data->p7iocCsr);
161 	pr_info("  lemFir:               %016llx\n", data->lemFir);
162 	pr_info("  lemErrorMask:         %016llx\n", data->lemErrorMask);
163 	pr_info("  lemWOF:               %016llx\n", data->lemWOF);
164 	pr_info("  phbErrorStatus:       %016llx\n", data->phbErrorStatus);
165 	pr_info("  phbFirstErrorStatus:  %016llx\n", data->phbFirstErrorStatus);
166 	pr_info("  phbErrorLog0:         %016llx\n", data->phbErrorLog0);
167 	pr_info("  phbErrorLog1:         %016llx\n", data->phbErrorLog1);
168 	pr_info("  mmioErrorStatus:      %016llx\n", data->mmioErrorStatus);
169 	pr_info("  mmioFirstErrorStatus: %016llx\n", data->mmioFirstErrorStatus);
170 	pr_info("  mmioErrorLog0:        %016llx\n", data->mmioErrorLog0);
171 	pr_info("  mmioErrorLog1:        %016llx\n", data->mmioErrorLog1);
172 	pr_info("  dma0ErrorStatus:      %016llx\n", data->dma0ErrorStatus);
173 	pr_info("  dma0FirstErrorStatus: %016llx\n", data->dma0FirstErrorStatus);
174 	pr_info("  dma0ErrorLog0:        %016llx\n", data->dma0ErrorLog0);
175 	pr_info("  dma0ErrorLog1:        %016llx\n", data->dma0ErrorLog1);
176 	pr_info("  dma1ErrorStatus:      %016llx\n", data->dma1ErrorStatus);
177 	pr_info("  dma1FirstErrorStatus: %016llx\n", data->dma1FirstErrorStatus);
178 	pr_info("  dma1ErrorLog0:        %016llx\n", data->dma1ErrorLog0);
179 	pr_info("  dma1ErrorLog1:        %016llx\n", data->dma1ErrorLog1);
180 
181 	for (i = 0; i < OPAL_P7IOC_NUM_PEST_REGS; i++) {
182 		if ((data->pestA[i] >> 63) == 0 &&
183 		    (data->pestB[i] >> 63) == 0)
184 			continue;
185 
186 		pr_info("  PE[%3d] PESTA:        %016llx\n", i, data->pestA[i]);
187 		pr_info("          PESTB:        %016llx\n", data->pestB[i]);
188 	}
189 }
190 
191 static void pnv_pci_dump_phb3_diag_data(struct pci_controller *hose,
192 					struct OpalIoPhbErrorCommon *common)
193 {
194 	struct OpalIoPhb3ErrorData *data;
195 	int i;
196 
197 	data = (struct OpalIoPhb3ErrorData*)common;
198 	pr_info("PHB3 PHB#%d Diag-data (Version: %d)\n\n",
199 		hose->global_number, common->version);
200 
201 	pr_info("  brdgCtl:              %08x\n", data->brdgCtl);
202 
203 	pr_info("  portStatusReg:        %08x\n", data->portStatusReg);
204 	pr_info("  rootCmplxStatus:      %08x\n", data->rootCmplxStatus);
205 	pr_info("  busAgentStatus:       %08x\n", data->busAgentStatus);
206 
207 	pr_info("  deviceStatus:         %08x\n", data->deviceStatus);
208 	pr_info("  slotStatus:           %08x\n", data->slotStatus);
209 	pr_info("  linkStatus:           %08x\n", data->linkStatus);
210 	pr_info("  devCmdStatus:         %08x\n", data->devCmdStatus);
211 	pr_info("  devSecStatus:         %08x\n", data->devSecStatus);
212 
213 	pr_info("  rootErrorStatus:      %08x\n", data->rootErrorStatus);
214 	pr_info("  uncorrErrorStatus:    %08x\n", data->uncorrErrorStatus);
215 	pr_info("  corrErrorStatus:      %08x\n", data->corrErrorStatus);
216 	pr_info("  tlpHdr1:              %08x\n", data->tlpHdr1);
217 	pr_info("  tlpHdr2:              %08x\n", data->tlpHdr2);
218 	pr_info("  tlpHdr3:              %08x\n", data->tlpHdr3);
219 	pr_info("  tlpHdr4:              %08x\n", data->tlpHdr4);
220 	pr_info("  sourceId:             %08x\n", data->sourceId);
221 	pr_info("  errorClass:           %016llx\n", data->errorClass);
222 	pr_info("  correlator:           %016llx\n", data->correlator);
223 
224 	pr_info("  nFir:                 %016llx\n", data->nFir);
225 	pr_info("  nFirMask:             %016llx\n", data->nFirMask);
226 	pr_info("  nFirWOF:              %016llx\n", data->nFirWOF);
227 	pr_info("  PhbPlssr:             %016llx\n", data->phbPlssr);
228 	pr_info("  PhbCsr:               %016llx\n", data->phbCsr);
229 	pr_info("  lemFir:               %016llx\n", data->lemFir);
230 	pr_info("  lemErrorMask:         %016llx\n", data->lemErrorMask);
231 	pr_info("  lemWOF:               %016llx\n", data->lemWOF);
232 	pr_info("  phbErrorStatus:       %016llx\n", data->phbErrorStatus);
233 	pr_info("  phbFirstErrorStatus:  %016llx\n", data->phbFirstErrorStatus);
234 	pr_info("  phbErrorLog0:         %016llx\n", data->phbErrorLog0);
235 	pr_info("  phbErrorLog1:         %016llx\n", data->phbErrorLog1);
236 	pr_info("  mmioErrorStatus:      %016llx\n", data->mmioErrorStatus);
237 	pr_info("  mmioFirstErrorStatus: %016llx\n", data->mmioFirstErrorStatus);
238 	pr_info("  mmioErrorLog0:        %016llx\n", data->mmioErrorLog0);
239 	pr_info("  mmioErrorLog1:        %016llx\n", data->mmioErrorLog1);
240 	pr_info("  dma0ErrorStatus:      %016llx\n", data->dma0ErrorStatus);
241 	pr_info("  dma0FirstErrorStatus: %016llx\n", data->dma0FirstErrorStatus);
242 	pr_info("  dma0ErrorLog0:        %016llx\n", data->dma0ErrorLog0);
243 	pr_info("  dma0ErrorLog1:        %016llx\n", data->dma0ErrorLog1);
244 	pr_info("  dma1ErrorStatus:      %016llx\n", data->dma1ErrorStatus);
245 	pr_info("  dma1FirstErrorStatus: %016llx\n", data->dma1FirstErrorStatus);
246 	pr_info("  dma1ErrorLog0:        %016llx\n", data->dma1ErrorLog0);
247 	pr_info("  dma1ErrorLog1:        %016llx\n", data->dma1ErrorLog1);
248 
249 	for (i = 0; i < OPAL_PHB3_NUM_PEST_REGS; i++) {
250 		if ((data->pestA[i] >> 63) == 0 &&
251 		    (data->pestB[i] >> 63) == 0)
252 			continue;
253 
254 		pr_info("  PE[%3d] PESTA:        %016llx\n", i, data->pestA[i]);
255 		pr_info("          PESTB:        %016llx\n", data->pestB[i]);
256 	}
257 }
258 
259 void pnv_pci_dump_phb_diag_data(struct pci_controller *hose,
260 				unsigned char *log_buff)
261 {
262 	struct OpalIoPhbErrorCommon *common;
263 
264 	if (!hose || !log_buff)
265 		return;
266 
267 	common = (struct OpalIoPhbErrorCommon *)log_buff;
268 	switch (common->ioType) {
269 	case OPAL_PHB_ERROR_DATA_TYPE_P7IOC:
270 		pnv_pci_dump_p7ioc_diag_data(hose, common);
271 		break;
272 	case OPAL_PHB_ERROR_DATA_TYPE_PHB3:
273 		pnv_pci_dump_phb3_diag_data(hose, common);
274 		break;
275 	default:
276 		pr_warn("%s: Unrecognized ioType %d\n",
277 			__func__, common->ioType);
278 	}
279 }
280 
281 static void pnv_pci_handle_eeh_config(struct pnv_phb *phb, u32 pe_no)
282 {
283 	unsigned long flags, rc;
284 	int has_diag;
285 
286 	spin_lock_irqsave(&phb->lock, flags);
287 
288 	rc = opal_pci_get_phb_diag_data2(phb->opal_id, phb->diag.blob,
289 					 PNV_PCI_DIAG_BUF_SIZE);
290 	has_diag = (rc == OPAL_SUCCESS);
291 
292 	rc = opal_pci_eeh_freeze_clear(phb->opal_id, pe_no,
293 				       OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
294 	if (rc) {
295 		pr_warning("PCI %d: Failed to clear EEH freeze state"
296 			   " for PE#%d, err %ld\n",
297 			   phb->hose->global_number, pe_no, rc);
298 
299 		/* For now, let's only display the diag buffer when we fail to clear
300 		 * the EEH status. We'll do more sensible things later when we have
301 		 * proper EEH support. We need to make sure we don't pollute ourselves
302 		 * with the normal errors generated when probing empty slots
303 		 */
304 		if (has_diag)
305 			pnv_pci_dump_phb_diag_data(phb->hose, phb->diag.blob);
306 		else
307 			pr_warning("PCI %d: No diag data available\n",
308 				   phb->hose->global_number);
309 	}
310 
311 	spin_unlock_irqrestore(&phb->lock, flags);
312 }
313 
314 static void pnv_pci_config_check_eeh(struct pnv_phb *phb,
315 				     struct device_node *dn)
316 {
317 	s64	rc;
318 	u8	fstate;
319 	__be16	pcierr;
320 	u32	pe_no;
321 
322 	/*
323 	 * Get the PE#. During the PCI probe stage, we might not
324 	 * setup that yet. So all ER errors should be mapped to
325 	 * reserved PE.
326 	 */
327 	pe_no = PCI_DN(dn)->pe_number;
328 	if (pe_no == IODA_INVALID_PE) {
329 		if (phb->type == PNV_PHB_P5IOC2)
330 			pe_no = 0;
331 		else
332 			pe_no = phb->ioda.reserved_pe;
333 	}
334 
335 	/* Read freeze status */
336 	rc = opal_pci_eeh_freeze_status(phb->opal_id, pe_no, &fstate, &pcierr,
337 					NULL);
338 	if (rc) {
339 		pr_warning("%s: Can't read EEH status (PE#%d) for "
340 			   "%s, err %lld\n",
341 			   __func__, pe_no, dn->full_name, rc);
342 		return;
343 	}
344 	cfg_dbg(" -> EEH check, bdfn=%04x PE#%d fstate=%x\n",
345 		(PCI_DN(dn)->busno << 8) | (PCI_DN(dn)->devfn),
346 		pe_no, fstate);
347 	if (fstate != 0)
348 		pnv_pci_handle_eeh_config(phb, pe_no);
349 }
350 
351 int pnv_pci_cfg_read(struct device_node *dn,
352 		     int where, int size, u32 *val)
353 {
354 	struct pci_dn *pdn = PCI_DN(dn);
355 	struct pnv_phb *phb = pdn->phb->private_data;
356 	u32 bdfn = (pdn->busno << 8) | pdn->devfn;
357 #ifdef CONFIG_EEH
358 	struct eeh_pe *phb_pe = NULL;
359 #endif
360 	s64 rc;
361 
362 	switch (size) {
363 	case 1: {
364 		u8 v8;
365 		rc = opal_pci_config_read_byte(phb->opal_id, bdfn, where, &v8);
366 		*val = (rc == OPAL_SUCCESS) ? v8 : 0xff;
367 		break;
368 	}
369 	case 2: {
370 		__be16 v16;
371 		rc = opal_pci_config_read_half_word(phb->opal_id, bdfn, where,
372 						   &v16);
373 		*val = (rc == OPAL_SUCCESS) ? be16_to_cpu(v16) : 0xffff;
374 		break;
375 	}
376 	case 4: {
377 		__be32 v32;
378 		rc = opal_pci_config_read_word(phb->opal_id, bdfn, where, &v32);
379 		*val = (rc == OPAL_SUCCESS) ? be32_to_cpu(v32) : 0xffffffff;
380 		break;
381 	}
382 	default:
383 		return PCIBIOS_FUNC_NOT_SUPPORTED;
384 	}
385 	cfg_dbg("%s: bus: %x devfn: %x +%x/%x -> %08x\n",
386 		__func__, pdn->busno, pdn->devfn, where, size, *val);
387 
388 	/*
389 	 * Check if the specified PE has been put into frozen
390 	 * state. On the other hand, we needn't do that while
391 	 * the PHB has been put into frozen state because of
392 	 * PHB-fatal errors.
393 	 */
394 #ifdef CONFIG_EEH
395 	phb_pe = eeh_phb_pe_get(pdn->phb);
396 	if (phb_pe && (phb_pe->state & EEH_PE_ISOLATED))
397 		return PCIBIOS_SUCCESSFUL;
398 
399 	if (phb->eeh_state & PNV_EEH_STATE_ENABLED) {
400 		if (*val == EEH_IO_ERROR_VALUE(size) &&
401 		    eeh_dev_check_failure(of_node_to_eeh_dev(dn)))
402 			return PCIBIOS_DEVICE_NOT_FOUND;
403 	} else {
404 		pnv_pci_config_check_eeh(phb, dn);
405 	}
406 #else
407 	pnv_pci_config_check_eeh(phb, dn);
408 #endif
409 
410 	return PCIBIOS_SUCCESSFUL;
411 }
412 
413 int pnv_pci_cfg_write(struct device_node *dn,
414 		      int where, int size, u32 val)
415 {
416 	struct pci_dn *pdn = PCI_DN(dn);
417 	struct pnv_phb *phb = pdn->phb->private_data;
418 	u32 bdfn = (pdn->busno << 8) | pdn->devfn;
419 
420 	cfg_dbg("%s: bus: %x devfn: %x +%x/%x -> %08x\n",
421 		pdn->busno, pdn->devfn, where, size, val);
422 	switch (size) {
423 	case 1:
424 		opal_pci_config_write_byte(phb->opal_id, bdfn, where, val);
425 		break;
426 	case 2:
427 		opal_pci_config_write_half_word(phb->opal_id, bdfn, where, val);
428 		break;
429 	case 4:
430 		opal_pci_config_write_word(phb->opal_id, bdfn, where, val);
431 		break;
432 	default:
433 		return PCIBIOS_FUNC_NOT_SUPPORTED;
434 	}
435 
436 	/* Check if the PHB got frozen due to an error (no response) */
437 #ifdef CONFIG_EEH
438 	if (!(phb->eeh_state & PNV_EEH_STATE_ENABLED))
439 		pnv_pci_config_check_eeh(phb, dn);
440 #else
441 	pnv_pci_config_check_eeh(phb, dn);
442 #endif
443 
444 	return PCIBIOS_SUCCESSFUL;
445 }
446 
447 static int pnv_pci_read_config(struct pci_bus *bus,
448 			       unsigned int devfn,
449 			       int where, int size, u32 *val)
450 {
451 	struct device_node *dn, *busdn = pci_bus_to_OF_node(bus);
452 	struct pci_dn *pdn;
453 
454 	for (dn = busdn->child; dn; dn = dn->sibling) {
455 		pdn = PCI_DN(dn);
456 		if (pdn && pdn->devfn == devfn)
457 			return pnv_pci_cfg_read(dn, where, size, val);
458 	}
459 
460 	*val = 0xFFFFFFFF;
461 	return PCIBIOS_DEVICE_NOT_FOUND;
462 
463 }
464 
465 static int pnv_pci_write_config(struct pci_bus *bus,
466 				unsigned int devfn,
467 				int where, int size, u32 val)
468 {
469 	struct device_node *dn, *busdn = pci_bus_to_OF_node(bus);
470 	struct pci_dn *pdn;
471 
472 	for (dn = busdn->child; dn; dn = dn->sibling) {
473 		pdn = PCI_DN(dn);
474 		if (pdn && pdn->devfn == devfn)
475 			return pnv_pci_cfg_write(dn, where, size, val);
476 	}
477 
478 	return PCIBIOS_DEVICE_NOT_FOUND;
479 }
480 
481 struct pci_ops pnv_pci_ops = {
482 	.read  = pnv_pci_read_config,
483 	.write = pnv_pci_write_config,
484 };
485 
486 static int pnv_tce_build(struct iommu_table *tbl, long index, long npages,
487 			 unsigned long uaddr, enum dma_data_direction direction,
488 			 struct dma_attrs *attrs, bool rm)
489 {
490 	u64 proto_tce;
491 	__be64 *tcep, *tces;
492 	u64 rpn;
493 
494 	proto_tce = TCE_PCI_READ; // Read allowed
495 
496 	if (direction != DMA_TO_DEVICE)
497 		proto_tce |= TCE_PCI_WRITE;
498 
499 	tces = tcep = ((__be64 *)tbl->it_base) + index - tbl->it_offset;
500 	rpn = __pa(uaddr) >> TCE_SHIFT;
501 
502 	while (npages--)
503 		*(tcep++) = cpu_to_be64(proto_tce | (rpn++ << TCE_RPN_SHIFT));
504 
505 	/* Some implementations won't cache invalid TCEs and thus may not
506 	 * need that flush. We'll probably turn it_type into a bit mask
507 	 * of flags if that becomes the case
508 	 */
509 	if (tbl->it_type & TCE_PCI_SWINV_CREATE)
510 		pnv_pci_ioda_tce_invalidate(tbl, tces, tcep - 1, rm);
511 
512 	return 0;
513 }
514 
515 static int pnv_tce_build_vm(struct iommu_table *tbl, long index, long npages,
516 			    unsigned long uaddr,
517 			    enum dma_data_direction direction,
518 			    struct dma_attrs *attrs)
519 {
520 	return pnv_tce_build(tbl, index, npages, uaddr, direction, attrs,
521 			false);
522 }
523 
524 static void pnv_tce_free(struct iommu_table *tbl, long index, long npages,
525 		bool rm)
526 {
527 	__be64 *tcep, *tces;
528 
529 	tces = tcep = ((__be64 *)tbl->it_base) + index - tbl->it_offset;
530 
531 	while (npages--)
532 		*(tcep++) = cpu_to_be64(0);
533 
534 	if (tbl->it_type & TCE_PCI_SWINV_FREE)
535 		pnv_pci_ioda_tce_invalidate(tbl, tces, tcep - 1, rm);
536 }
537 
538 static void pnv_tce_free_vm(struct iommu_table *tbl, long index, long npages)
539 {
540 	pnv_tce_free(tbl, index, npages, false);
541 }
542 
543 static unsigned long pnv_tce_get(struct iommu_table *tbl, long index)
544 {
545 	return ((u64 *)tbl->it_base)[index - tbl->it_offset];
546 }
547 
548 static int pnv_tce_build_rm(struct iommu_table *tbl, long index, long npages,
549 			    unsigned long uaddr,
550 			    enum dma_data_direction direction,
551 			    struct dma_attrs *attrs)
552 {
553 	return pnv_tce_build(tbl, index, npages, uaddr, direction, attrs, true);
554 }
555 
556 static void pnv_tce_free_rm(struct iommu_table *tbl, long index, long npages)
557 {
558 	pnv_tce_free(tbl, index, npages, true);
559 }
560 
561 void pnv_pci_setup_iommu_table(struct iommu_table *tbl,
562 			       void *tce_mem, u64 tce_size,
563 			       u64 dma_offset)
564 {
565 	tbl->it_blocksize = 16;
566 	tbl->it_base = (unsigned long)tce_mem;
567 	tbl->it_page_shift = IOMMU_PAGE_SHIFT_4K;
568 	tbl->it_offset = dma_offset >> tbl->it_page_shift;
569 	tbl->it_index = 0;
570 	tbl->it_size = tce_size >> 3;
571 	tbl->it_busno = 0;
572 	tbl->it_type = TCE_PCI;
573 }
574 
575 static struct iommu_table *pnv_pci_setup_bml_iommu(struct pci_controller *hose)
576 {
577 	struct iommu_table *tbl;
578 	const __be64 *basep, *swinvp;
579 	const __be32 *sizep;
580 
581 	basep = of_get_property(hose->dn, "linux,tce-base", NULL);
582 	sizep = of_get_property(hose->dn, "linux,tce-size", NULL);
583 	if (basep == NULL || sizep == NULL) {
584 		pr_err("PCI: %s has missing tce entries !\n",
585 		       hose->dn->full_name);
586 		return NULL;
587 	}
588 	tbl = kzalloc_node(sizeof(struct iommu_table), GFP_KERNEL, hose->node);
589 	if (WARN_ON(!tbl))
590 		return NULL;
591 	pnv_pci_setup_iommu_table(tbl, __va(be64_to_cpup(basep)),
592 				  be32_to_cpup(sizep), 0);
593 	iommu_init_table(tbl, hose->node);
594 	iommu_register_group(tbl, pci_domain_nr(hose->bus), 0);
595 
596 	/* Deal with SW invalidated TCEs when needed (BML way) */
597 	swinvp = of_get_property(hose->dn, "linux,tce-sw-invalidate-info",
598 				 NULL);
599 	if (swinvp) {
600 		tbl->it_busno = be64_to_cpu(swinvp[1]);
601 		tbl->it_index = (unsigned long)ioremap(be64_to_cpup(swinvp), 8);
602 		tbl->it_type = TCE_PCI_SWINV_CREATE | TCE_PCI_SWINV_FREE;
603 	}
604 	return tbl;
605 }
606 
607 static void pnv_pci_dma_fallback_setup(struct pci_controller *hose,
608 				       struct pci_dev *pdev)
609 {
610 	struct device_node *np = pci_bus_to_OF_node(hose->bus);
611 	struct pci_dn *pdn;
612 
613 	if (np == NULL)
614 		return;
615 	pdn = PCI_DN(np);
616 	if (!pdn->iommu_table)
617 		pdn->iommu_table = pnv_pci_setup_bml_iommu(hose);
618 	if (!pdn->iommu_table)
619 		return;
620 	set_iommu_table_base_and_group(&pdev->dev, pdn->iommu_table);
621 }
622 
623 static void pnv_pci_dma_dev_setup(struct pci_dev *pdev)
624 {
625 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
626 	struct pnv_phb *phb = hose->private_data;
627 
628 	/* If we have no phb structure, try to setup a fallback based on
629 	 * the device-tree (RTAS PCI for example)
630 	 */
631 	if (phb && phb->dma_dev_setup)
632 		phb->dma_dev_setup(phb, pdev);
633 	else
634 		pnv_pci_dma_fallback_setup(hose, pdev);
635 }
636 
637 void pnv_pci_shutdown(void)
638 {
639 	struct pci_controller *hose;
640 
641 	list_for_each_entry(hose, &hose_list, list_node) {
642 		struct pnv_phb *phb = hose->private_data;
643 
644 		if (phb && phb->shutdown)
645 			phb->shutdown(phb);
646 	}
647 }
648 
649 /* Fixup wrong class code in p7ioc and p8 root complex */
650 static void pnv_p7ioc_rc_quirk(struct pci_dev *dev)
651 {
652 	dev->class = PCI_CLASS_BRIDGE_PCI << 8;
653 }
654 DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_IBM, 0x3b9, pnv_p7ioc_rc_quirk);
655 
656 static int pnv_pci_probe_mode(struct pci_bus *bus)
657 {
658 	struct pci_controller *hose = pci_bus_to_host(bus);
659 	const __be64 *tstamp;
660 	u64 now, target;
661 
662 
663 	/* We hijack this as a way to ensure we have waited long
664 	 * enough since the reset was lifted on the PCI bus
665 	 */
666 	if (bus != hose->bus)
667 		return PCI_PROBE_NORMAL;
668 	tstamp = of_get_property(hose->dn, "reset-clear-timestamp", NULL);
669 	if (!tstamp || !*tstamp)
670 		return PCI_PROBE_NORMAL;
671 
672 	now = mftb() / tb_ticks_per_usec;
673 	target = (be64_to_cpup(tstamp) / tb_ticks_per_usec)
674 		+ PCI_RESET_DELAY_US;
675 
676 	pr_devel("pci %04d: Reset target: 0x%llx now: 0x%llx\n",
677 		 hose->global_number, target, now);
678 
679 	if (now < target)
680 		msleep((target - now + 999) / 1000);
681 
682 	return PCI_PROBE_NORMAL;
683 }
684 
685 void __init pnv_pci_init(void)
686 {
687 	struct device_node *np;
688 
689 	pci_add_flags(PCI_CAN_SKIP_ISA_ALIGN);
690 
691 	/* OPAL absent, try POPAL first then RTAS detection of PHBs */
692 	if (!firmware_has_feature(FW_FEATURE_OPAL)) {
693 #ifdef CONFIG_PPC_POWERNV_RTAS
694 		init_pci_config_tokens();
695 		find_and_init_phbs();
696 #endif /* CONFIG_PPC_POWERNV_RTAS */
697 	}
698 	/* OPAL is here, do our normal stuff */
699 	else {
700 		int found_ioda = 0;
701 
702 		/* Look for IODA IO-Hubs. We don't support mixing IODA
703 		 * and p5ioc2 due to the need to change some global
704 		 * probing flags
705 		 */
706 		for_each_compatible_node(np, NULL, "ibm,ioda-hub") {
707 			pnv_pci_init_ioda_hub(np);
708 			found_ioda = 1;
709 		}
710 
711 		/* Look for p5ioc2 IO-Hubs */
712 		if (!found_ioda)
713 			for_each_compatible_node(np, NULL, "ibm,p5ioc2")
714 				pnv_pci_init_p5ioc2_hub(np);
715 
716 		/* Look for ioda2 built-in PHB3's */
717 		for_each_compatible_node(np, NULL, "ibm,ioda2-phb")
718 			pnv_pci_init_ioda2_phb(np);
719 	}
720 
721 	/* Setup the linkage between OF nodes and PHBs */
722 	pci_devs_phb_init();
723 
724 	/* Configure IOMMU DMA hooks */
725 	ppc_md.pci_dma_dev_setup = pnv_pci_dma_dev_setup;
726 	ppc_md.tce_build = pnv_tce_build_vm;
727 	ppc_md.tce_free = pnv_tce_free_vm;
728 	ppc_md.tce_build_rm = pnv_tce_build_rm;
729 	ppc_md.tce_free_rm = pnv_tce_free_rm;
730 	ppc_md.tce_get = pnv_tce_get;
731 	ppc_md.pci_probe_mode = pnv_pci_probe_mode;
732 	set_pci_dma_ops(&dma_iommu_ops);
733 
734 	/* Configure MSIs */
735 #ifdef CONFIG_PCI_MSI
736 	ppc_md.msi_check_device = pnv_msi_check_device;
737 	ppc_md.setup_msi_irqs = pnv_setup_msi_irqs;
738 	ppc_md.teardown_msi_irqs = pnv_teardown_msi_irqs;
739 #endif
740 }
741 
742 static int tce_iommu_bus_notifier(struct notifier_block *nb,
743 		unsigned long action, void *data)
744 {
745 	struct device *dev = data;
746 
747 	switch (action) {
748 	case BUS_NOTIFY_ADD_DEVICE:
749 		return iommu_add_device(dev);
750 	case BUS_NOTIFY_DEL_DEVICE:
751 		if (dev->iommu_group)
752 			iommu_del_device(dev);
753 		return 0;
754 	default:
755 		return 0;
756 	}
757 }
758 
759 static struct notifier_block tce_iommu_bus_nb = {
760 	.notifier_call = tce_iommu_bus_notifier,
761 };
762 
763 static int __init tce_iommu_bus_notifier_init(void)
764 {
765 	bus_register_notifier(&pci_bus_type, &tce_iommu_bus_nb);
766 	return 0;
767 }
768 
769 subsys_initcall_sync(tce_iommu_bus_notifier_init);
770