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",
135 		hose->global_number, be32_to_cpu(common->version));
136 
137 	if (data->brdgCtl)
138 		pr_info("brdgCtl:     %08x\n",
139 			be32_to_cpu(data->brdgCtl));
140 	if (data->portStatusReg || data->rootCmplxStatus ||
141 	    data->busAgentStatus)
142 		pr_info("UtlSts:      %08x %08x %08x\n",
143 			be32_to_cpu(data->portStatusReg),
144 			be32_to_cpu(data->rootCmplxStatus),
145 			be32_to_cpu(data->busAgentStatus));
146 	if (data->deviceStatus || data->slotStatus   ||
147 	    data->linkStatus   || data->devCmdStatus ||
148 	    data->devSecStatus)
149 		pr_info("RootSts:     %08x %08x %08x %08x %08x\n",
150 			be32_to_cpu(data->deviceStatus),
151 			be32_to_cpu(data->slotStatus),
152 			be32_to_cpu(data->linkStatus),
153 			be32_to_cpu(data->devCmdStatus),
154 			be32_to_cpu(data->devSecStatus));
155 	if (data->rootErrorStatus   || data->uncorrErrorStatus ||
156 	    data->corrErrorStatus)
157 		pr_info("RootErrSts:  %08x %08x %08x\n",
158 			be32_to_cpu(data->rootErrorStatus),
159 			be32_to_cpu(data->uncorrErrorStatus),
160 			be32_to_cpu(data->corrErrorStatus));
161 	if (data->tlpHdr1 || data->tlpHdr2 ||
162 	    data->tlpHdr3 || data->tlpHdr4)
163 		pr_info("RootErrLog:  %08x %08x %08x %08x\n",
164 			be32_to_cpu(data->tlpHdr1),
165 			be32_to_cpu(data->tlpHdr2),
166 			be32_to_cpu(data->tlpHdr3),
167 			be32_to_cpu(data->tlpHdr4));
168 	if (data->sourceId || data->errorClass ||
169 	    data->correlator)
170 		pr_info("RootErrLog1: %08x %016llx %016llx\n",
171 			be32_to_cpu(data->sourceId),
172 			be64_to_cpu(data->errorClass),
173 			be64_to_cpu(data->correlator));
174 	if (data->p7iocPlssr || data->p7iocCsr)
175 		pr_info("PhbSts:      %016llx %016llx\n",
176 			be64_to_cpu(data->p7iocPlssr),
177 			be64_to_cpu(data->p7iocCsr));
178 	if (data->lemFir)
179 		pr_info("Lem:         %016llx %016llx %016llx\n",
180 			be64_to_cpu(data->lemFir),
181 			be64_to_cpu(data->lemErrorMask),
182 			be64_to_cpu(data->lemWOF));
183 	if (data->phbErrorStatus)
184 		pr_info("PhbErr:      %016llx %016llx %016llx %016llx\n",
185 			be64_to_cpu(data->phbErrorStatus),
186 			be64_to_cpu(data->phbFirstErrorStatus),
187 			be64_to_cpu(data->phbErrorLog0),
188 			be64_to_cpu(data->phbErrorLog1));
189 	if (data->mmioErrorStatus)
190 		pr_info("OutErr:      %016llx %016llx %016llx %016llx\n",
191 			be64_to_cpu(data->mmioErrorStatus),
192 			be64_to_cpu(data->mmioFirstErrorStatus),
193 			be64_to_cpu(data->mmioErrorLog0),
194 			be64_to_cpu(data->mmioErrorLog1));
195 	if (data->dma0ErrorStatus)
196 		pr_info("InAErr:      %016llx %016llx %016llx %016llx\n",
197 			be64_to_cpu(data->dma0ErrorStatus),
198 			be64_to_cpu(data->dma0FirstErrorStatus),
199 			be64_to_cpu(data->dma0ErrorLog0),
200 			be64_to_cpu(data->dma0ErrorLog1));
201 	if (data->dma1ErrorStatus)
202 		pr_info("InBErr:      %016llx %016llx %016llx %016llx\n",
203 			be64_to_cpu(data->dma1ErrorStatus),
204 			be64_to_cpu(data->dma1FirstErrorStatus),
205 			be64_to_cpu(data->dma1ErrorLog0),
206 			be64_to_cpu(data->dma1ErrorLog1));
207 
208 	for (i = 0; i < OPAL_P7IOC_NUM_PEST_REGS; i++) {
209 		if ((data->pestA[i] >> 63) == 0 &&
210 		    (data->pestB[i] >> 63) == 0)
211 			continue;
212 
213 		pr_info("PE[%3d] A/B: %016llx %016llx\n",
214 			i, be64_to_cpu(data->pestA[i]),
215 			be64_to_cpu(data->pestB[i]));
216 	}
217 }
218 
219 static void pnv_pci_dump_phb3_diag_data(struct pci_controller *hose,
220 					struct OpalIoPhbErrorCommon *common)
221 {
222 	struct OpalIoPhb3ErrorData *data;
223 	int i;
224 
225 	data = (struct OpalIoPhb3ErrorData*)common;
226 	pr_info("PHB3 PHB#%d Diag-data (Version: %d)\n",
227 		hose->global_number, be32_to_cpu(common->version));
228 	if (data->brdgCtl)
229 		pr_info("brdgCtl:     %08x\n",
230 			be32_to_cpu(data->brdgCtl));
231 	if (data->portStatusReg || data->rootCmplxStatus ||
232 	    data->busAgentStatus)
233 		pr_info("UtlSts:      %08x %08x %08x\n",
234 			be32_to_cpu(data->portStatusReg),
235 			be32_to_cpu(data->rootCmplxStatus),
236 			be32_to_cpu(data->busAgentStatus));
237 	if (data->deviceStatus || data->slotStatus   ||
238 	    data->linkStatus   || data->devCmdStatus ||
239 	    data->devSecStatus)
240 		pr_info("RootSts:     %08x %08x %08x %08x %08x\n",
241 			be32_to_cpu(data->deviceStatus),
242 			be32_to_cpu(data->slotStatus),
243 			be32_to_cpu(data->linkStatus),
244 			be32_to_cpu(data->devCmdStatus),
245 			be32_to_cpu(data->devSecStatus));
246 	if (data->rootErrorStatus || data->uncorrErrorStatus ||
247 	    data->corrErrorStatus)
248 		pr_info("RootErrSts:  %08x %08x %08x\n",
249 			be32_to_cpu(data->rootErrorStatus),
250 			be32_to_cpu(data->uncorrErrorStatus),
251 			be32_to_cpu(data->corrErrorStatus));
252 	if (data->tlpHdr1 || data->tlpHdr2 ||
253 	    data->tlpHdr3 || data->tlpHdr4)
254 		pr_info("RootErrLog:  %08x %08x %08x %08x\n",
255 			be32_to_cpu(data->tlpHdr1),
256 			be32_to_cpu(data->tlpHdr2),
257 			be32_to_cpu(data->tlpHdr3),
258 			be32_to_cpu(data->tlpHdr4));
259 	if (data->sourceId || data->errorClass ||
260 	    data->correlator)
261 		pr_info("RootErrLog1: %08x %016llx %016llx\n",
262 			be32_to_cpu(data->sourceId),
263 			be64_to_cpu(data->errorClass),
264 			be64_to_cpu(data->correlator));
265 	if (data->nFir)
266 		pr_info("nFir:        %016llx %016llx %016llx\n",
267 			be64_to_cpu(data->nFir),
268 			be64_to_cpu(data->nFirMask),
269 			be64_to_cpu(data->nFirWOF));
270 	if (data->phbPlssr || data->phbCsr)
271 		pr_info("PhbSts:      %016llx %016llx\n",
272 			be64_to_cpu(data->phbPlssr),
273 			be64_to_cpu(data->phbCsr));
274 	if (data->lemFir)
275 		pr_info("Lem:         %016llx %016llx %016llx\n",
276 			be64_to_cpu(data->lemFir),
277 			be64_to_cpu(data->lemErrorMask),
278 			be64_to_cpu(data->lemWOF));
279 	if (data->phbErrorStatus)
280 		pr_info("PhbErr:      %016llx %016llx %016llx %016llx\n",
281 			be64_to_cpu(data->phbErrorStatus),
282 			be64_to_cpu(data->phbFirstErrorStatus),
283 			be64_to_cpu(data->phbErrorLog0),
284 			be64_to_cpu(data->phbErrorLog1));
285 	if (data->mmioErrorStatus)
286 		pr_info("OutErr:      %016llx %016llx %016llx %016llx\n",
287 			be64_to_cpu(data->mmioErrorStatus),
288 			be64_to_cpu(data->mmioFirstErrorStatus),
289 			be64_to_cpu(data->mmioErrorLog0),
290 			be64_to_cpu(data->mmioErrorLog1));
291 	if (data->dma0ErrorStatus)
292 		pr_info("InAErr:      %016llx %016llx %016llx %016llx\n",
293 			be64_to_cpu(data->dma0ErrorStatus),
294 			be64_to_cpu(data->dma0FirstErrorStatus),
295 			be64_to_cpu(data->dma0ErrorLog0),
296 			be64_to_cpu(data->dma0ErrorLog1));
297 	if (data->dma1ErrorStatus)
298 		pr_info("InBErr:      %016llx %016llx %016llx %016llx\n",
299 			be64_to_cpu(data->dma1ErrorStatus),
300 			be64_to_cpu(data->dma1FirstErrorStatus),
301 			be64_to_cpu(data->dma1ErrorLog0),
302 			be64_to_cpu(data->dma1ErrorLog1));
303 
304 	for (i = 0; i < OPAL_PHB3_NUM_PEST_REGS; i++) {
305 		if ((be64_to_cpu(data->pestA[i]) >> 63) == 0 &&
306 		    (be64_to_cpu(data->pestB[i]) >> 63) == 0)
307 			continue;
308 
309 		pr_info("PE[%3d] A/B: %016llx %016llx\n",
310 				i, be64_to_cpu(data->pestA[i]),
311 				be64_to_cpu(data->pestB[i]));
312 	}
313 }
314 
315 void pnv_pci_dump_phb_diag_data(struct pci_controller *hose,
316 				unsigned char *log_buff)
317 {
318 	struct OpalIoPhbErrorCommon *common;
319 
320 	if (!hose || !log_buff)
321 		return;
322 
323 	common = (struct OpalIoPhbErrorCommon *)log_buff;
324 	switch (be32_to_cpu(common->ioType)) {
325 	case OPAL_PHB_ERROR_DATA_TYPE_P7IOC:
326 		pnv_pci_dump_p7ioc_diag_data(hose, common);
327 		break;
328 	case OPAL_PHB_ERROR_DATA_TYPE_PHB3:
329 		pnv_pci_dump_phb3_diag_data(hose, common);
330 		break;
331 	default:
332 		pr_warn("%s: Unrecognized ioType %d\n",
333 			__func__, be32_to_cpu(common->ioType));
334 	}
335 }
336 
337 static void pnv_pci_handle_eeh_config(struct pnv_phb *phb, u32 pe_no)
338 {
339 	unsigned long flags, rc;
340 	int has_diag, ret = 0;
341 
342 	spin_lock_irqsave(&phb->lock, flags);
343 
344 	/* Fetch PHB diag-data */
345 	rc = opal_pci_get_phb_diag_data2(phb->opal_id, phb->diag.blob,
346 					 PNV_PCI_DIAG_BUF_SIZE);
347 	has_diag = (rc == OPAL_SUCCESS);
348 
349 	/* If PHB supports compound PE, to handle it */
350 	if (phb->unfreeze_pe) {
351 		ret = phb->unfreeze_pe(phb,
352 				       pe_no,
353 				       OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
354 	} else {
355 		rc = opal_pci_eeh_freeze_clear(phb->opal_id,
356 					     pe_no,
357 					     OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
358 		if (rc) {
359 			pr_warn("%s: Failure %ld clearing frozen "
360 				"PHB#%x-PE#%x\n",
361 				__func__, rc, phb->hose->global_number,
362 				pe_no);
363 			ret = -EIO;
364 		}
365 	}
366 
367 	/*
368 	 * For now, let's only display the diag buffer when we fail to clear
369 	 * the EEH status. We'll do more sensible things later when we have
370 	 * proper EEH support. We need to make sure we don't pollute ourselves
371 	 * with the normal errors generated when probing empty slots
372 	 */
373 	if (has_diag && ret)
374 		pnv_pci_dump_phb_diag_data(phb->hose, phb->diag.blob);
375 
376 	spin_unlock_irqrestore(&phb->lock, flags);
377 }
378 
379 static void pnv_pci_config_check_eeh(struct pnv_phb *phb,
380 				     struct device_node *dn)
381 {
382 	u8	fstate;
383 	__be16	pcierr;
384 	int	pe_no;
385 	s64	rc;
386 
387 	/*
388 	 * Get the PE#. During the PCI probe stage, we might not
389 	 * setup that yet. So all ER errors should be mapped to
390 	 * reserved PE.
391 	 */
392 	pe_no = PCI_DN(dn)->pe_number;
393 	if (pe_no == IODA_INVALID_PE) {
394 		if (phb->type == PNV_PHB_P5IOC2)
395 			pe_no = 0;
396 		else
397 			pe_no = phb->ioda.reserved_pe;
398 	}
399 
400 	/*
401 	 * Fetch frozen state. If the PHB support compound PE,
402 	 * we need handle that case.
403 	 */
404 	if (phb->get_pe_state) {
405 		fstate = phb->get_pe_state(phb, pe_no);
406 	} else {
407 		rc = opal_pci_eeh_freeze_status(phb->opal_id,
408 						pe_no,
409 						&fstate,
410 						&pcierr,
411 						NULL);
412 		if (rc) {
413 			pr_warn("%s: Failure %lld getting PHB#%x-PE#%x state\n",
414 				__func__, rc, phb->hose->global_number, pe_no);
415 			return;
416 		}
417 	}
418 
419 	cfg_dbg(" -> EEH check, bdfn=%04x PE#%d fstate=%x\n",
420 		(PCI_DN(dn)->busno << 8) | (PCI_DN(dn)->devfn),
421 		pe_no, fstate);
422 
423 	/* Clear the frozen state if applicable */
424 	if (fstate == OPAL_EEH_STOPPED_MMIO_FREEZE ||
425 	    fstate == OPAL_EEH_STOPPED_DMA_FREEZE  ||
426 	    fstate == OPAL_EEH_STOPPED_MMIO_DMA_FREEZE) {
427 		/*
428 		 * If PHB supports compound PE, freeze it for
429 		 * consistency.
430 		 */
431 		if (phb->freeze_pe)
432 			phb->freeze_pe(phb, pe_no);
433 
434 		pnv_pci_handle_eeh_config(phb, pe_no);
435 	}
436 }
437 
438 int pnv_pci_cfg_read(struct device_node *dn,
439 		     int where, int size, u32 *val)
440 {
441 	struct pci_dn *pdn = PCI_DN(dn);
442 	struct pnv_phb *phb = pdn->phb->private_data;
443 	u32 bdfn = (pdn->busno << 8) | pdn->devfn;
444 	s64 rc;
445 
446 	switch (size) {
447 	case 1: {
448 		u8 v8;
449 		rc = opal_pci_config_read_byte(phb->opal_id, bdfn, where, &v8);
450 		*val = (rc == OPAL_SUCCESS) ? v8 : 0xff;
451 		break;
452 	}
453 	case 2: {
454 		__be16 v16;
455 		rc = opal_pci_config_read_half_word(phb->opal_id, bdfn, where,
456 						   &v16);
457 		*val = (rc == OPAL_SUCCESS) ? be16_to_cpu(v16) : 0xffff;
458 		break;
459 	}
460 	case 4: {
461 		__be32 v32;
462 		rc = opal_pci_config_read_word(phb->opal_id, bdfn, where, &v32);
463 		*val = (rc == OPAL_SUCCESS) ? be32_to_cpu(v32) : 0xffffffff;
464 		break;
465 	}
466 	default:
467 		return PCIBIOS_FUNC_NOT_SUPPORTED;
468 	}
469 
470 	cfg_dbg("%s: bus: %x devfn: %x +%x/%x -> %08x\n",
471 		__func__, pdn->busno, pdn->devfn, where, size, *val);
472 	return PCIBIOS_SUCCESSFUL;
473 }
474 
475 int pnv_pci_cfg_write(struct device_node *dn,
476 		      int where, int size, u32 val)
477 {
478 	struct pci_dn *pdn = PCI_DN(dn);
479 	struct pnv_phb *phb = pdn->phb->private_data;
480 	u32 bdfn = (pdn->busno << 8) | pdn->devfn;
481 
482 	cfg_dbg("%s: bus: %x devfn: %x +%x/%x -> %08x\n",
483 		pdn->busno, pdn->devfn, where, size, val);
484 	switch (size) {
485 	case 1:
486 		opal_pci_config_write_byte(phb->opal_id, bdfn, where, val);
487 		break;
488 	case 2:
489 		opal_pci_config_write_half_word(phb->opal_id, bdfn, where, val);
490 		break;
491 	case 4:
492 		opal_pci_config_write_word(phb->opal_id, bdfn, where, val);
493 		break;
494 	default:
495 		return PCIBIOS_FUNC_NOT_SUPPORTED;
496 	}
497 
498 	return PCIBIOS_SUCCESSFUL;
499 }
500 
501 #if CONFIG_EEH
502 static bool pnv_pci_cfg_check(struct pci_controller *hose,
503 			      struct device_node *dn)
504 {
505 	struct eeh_dev *edev = NULL;
506 	struct pnv_phb *phb = hose->private_data;
507 
508 	/* EEH not enabled ? */
509 	if (!(phb->flags & PNV_PHB_FLAG_EEH))
510 		return true;
511 
512 	/* PE reset or device removed ? */
513 	edev = of_node_to_eeh_dev(dn);
514 	if (edev) {
515 		if (edev->pe &&
516 		    (edev->pe->state & EEH_PE_RESET))
517 			return false;
518 
519 		if (edev->mode & EEH_DEV_REMOVED)
520 			return false;
521 	}
522 
523 	return true;
524 }
525 #else
526 static inline pnv_pci_cfg_check(struct pci_controller *hose,
527 				struct device_node *dn)
528 {
529 	return true;
530 }
531 #endif /* CONFIG_EEH */
532 
533 static int pnv_pci_read_config(struct pci_bus *bus,
534 			       unsigned int devfn,
535 			       int where, int size, u32 *val)
536 {
537 	struct device_node *dn, *busdn = pci_bus_to_OF_node(bus);
538 	struct pci_dn *pdn;
539 	struct pnv_phb *phb;
540 	bool found = false;
541 	int ret;
542 
543 	*val = 0xFFFFFFFF;
544 	for (dn = busdn->child; dn; dn = dn->sibling) {
545 		pdn = PCI_DN(dn);
546 		if (pdn && pdn->devfn == devfn) {
547 			phb = pdn->phb->private_data;
548 			found = true;
549 			break;
550 		}
551 	}
552 
553 	if (!found || !pnv_pci_cfg_check(pdn->phb, dn))
554 		return PCIBIOS_DEVICE_NOT_FOUND;
555 
556 	ret = pnv_pci_cfg_read(dn, where, size, val);
557 	if (phb->flags & PNV_PHB_FLAG_EEH) {
558 		if (*val == EEH_IO_ERROR_VALUE(size) &&
559 		    eeh_dev_check_failure(of_node_to_eeh_dev(dn)))
560                         return PCIBIOS_DEVICE_NOT_FOUND;
561 	} else {
562 		pnv_pci_config_check_eeh(phb, dn);
563 	}
564 
565 	return ret;
566 }
567 
568 static int pnv_pci_write_config(struct pci_bus *bus,
569 				unsigned int devfn,
570 				int where, int size, u32 val)
571 {
572 	struct device_node *dn, *busdn = pci_bus_to_OF_node(bus);
573 	struct pci_dn *pdn;
574 	struct pnv_phb *phb;
575 	bool found = false;
576 	int ret;
577 
578 	for (dn = busdn->child; dn; dn = dn->sibling) {
579 		pdn = PCI_DN(dn);
580 		if (pdn && pdn->devfn == devfn) {
581 			phb = pdn->phb->private_data;
582 			found = true;
583 			break;
584 		}
585 	}
586 
587 	if (!found || !pnv_pci_cfg_check(pdn->phb, dn))
588 		return PCIBIOS_DEVICE_NOT_FOUND;
589 
590 	ret = pnv_pci_cfg_write(dn, where, size, val);
591 	if (!(phb->flags & PNV_PHB_FLAG_EEH))
592 		pnv_pci_config_check_eeh(phb, dn);
593 
594 	return ret;
595 }
596 
597 struct pci_ops pnv_pci_ops = {
598 	.read  = pnv_pci_read_config,
599 	.write = pnv_pci_write_config,
600 };
601 
602 static int pnv_tce_build(struct iommu_table *tbl, long index, long npages,
603 			 unsigned long uaddr, enum dma_data_direction direction,
604 			 struct dma_attrs *attrs, bool rm)
605 {
606 	u64 proto_tce;
607 	__be64 *tcep, *tces;
608 	u64 rpn;
609 
610 	proto_tce = TCE_PCI_READ; // Read allowed
611 
612 	if (direction != DMA_TO_DEVICE)
613 		proto_tce |= TCE_PCI_WRITE;
614 
615 	tces = tcep = ((__be64 *)tbl->it_base) + index - tbl->it_offset;
616 	rpn = __pa(uaddr) >> tbl->it_page_shift;
617 
618 	while (npages--)
619 		*(tcep++) = cpu_to_be64(proto_tce |
620 				(rpn++ << tbl->it_page_shift));
621 
622 	/* Some implementations won't cache invalid TCEs and thus may not
623 	 * need that flush. We'll probably turn it_type into a bit mask
624 	 * of flags if that becomes the case
625 	 */
626 	if (tbl->it_type & TCE_PCI_SWINV_CREATE)
627 		pnv_pci_ioda_tce_invalidate(tbl, tces, tcep - 1, rm);
628 
629 	return 0;
630 }
631 
632 static int pnv_tce_build_vm(struct iommu_table *tbl, long index, long npages,
633 			    unsigned long uaddr,
634 			    enum dma_data_direction direction,
635 			    struct dma_attrs *attrs)
636 {
637 	return pnv_tce_build(tbl, index, npages, uaddr, direction, attrs,
638 			false);
639 }
640 
641 static void pnv_tce_free(struct iommu_table *tbl, long index, long npages,
642 		bool rm)
643 {
644 	__be64 *tcep, *tces;
645 
646 	tces = tcep = ((__be64 *)tbl->it_base) + index - tbl->it_offset;
647 
648 	while (npages--)
649 		*(tcep++) = cpu_to_be64(0);
650 
651 	if (tbl->it_type & TCE_PCI_SWINV_FREE)
652 		pnv_pci_ioda_tce_invalidate(tbl, tces, tcep - 1, rm);
653 }
654 
655 static void pnv_tce_free_vm(struct iommu_table *tbl, long index, long npages)
656 {
657 	pnv_tce_free(tbl, index, npages, false);
658 }
659 
660 static unsigned long pnv_tce_get(struct iommu_table *tbl, long index)
661 {
662 	return ((u64 *)tbl->it_base)[index - tbl->it_offset];
663 }
664 
665 static int pnv_tce_build_rm(struct iommu_table *tbl, long index, long npages,
666 			    unsigned long uaddr,
667 			    enum dma_data_direction direction,
668 			    struct dma_attrs *attrs)
669 {
670 	return pnv_tce_build(tbl, index, npages, uaddr, direction, attrs, true);
671 }
672 
673 static void pnv_tce_free_rm(struct iommu_table *tbl, long index, long npages)
674 {
675 	pnv_tce_free(tbl, index, npages, true);
676 }
677 
678 void pnv_pci_setup_iommu_table(struct iommu_table *tbl,
679 			       void *tce_mem, u64 tce_size,
680 			       u64 dma_offset, unsigned page_shift)
681 {
682 	tbl->it_blocksize = 16;
683 	tbl->it_base = (unsigned long)tce_mem;
684 	tbl->it_page_shift = page_shift;
685 	tbl->it_offset = dma_offset >> tbl->it_page_shift;
686 	tbl->it_index = 0;
687 	tbl->it_size = tce_size >> 3;
688 	tbl->it_busno = 0;
689 	tbl->it_type = TCE_PCI;
690 }
691 
692 static struct iommu_table *pnv_pci_setup_bml_iommu(struct pci_controller *hose)
693 {
694 	struct iommu_table *tbl;
695 	const __be64 *basep, *swinvp;
696 	const __be32 *sizep;
697 
698 	basep = of_get_property(hose->dn, "linux,tce-base", NULL);
699 	sizep = of_get_property(hose->dn, "linux,tce-size", NULL);
700 	if (basep == NULL || sizep == NULL) {
701 		pr_err("PCI: %s has missing tce entries !\n",
702 		       hose->dn->full_name);
703 		return NULL;
704 	}
705 	tbl = kzalloc_node(sizeof(struct iommu_table), GFP_KERNEL, hose->node);
706 	if (WARN_ON(!tbl))
707 		return NULL;
708 	pnv_pci_setup_iommu_table(tbl, __va(be64_to_cpup(basep)),
709 				  be32_to_cpup(sizep), 0, IOMMU_PAGE_SHIFT_4K);
710 	iommu_init_table(tbl, hose->node);
711 	iommu_register_group(tbl, pci_domain_nr(hose->bus), 0);
712 
713 	/* Deal with SW invalidated TCEs when needed (BML way) */
714 	swinvp = of_get_property(hose->dn, "linux,tce-sw-invalidate-info",
715 				 NULL);
716 	if (swinvp) {
717 		tbl->it_busno = be64_to_cpu(swinvp[1]);
718 		tbl->it_index = (unsigned long)ioremap(be64_to_cpup(swinvp), 8);
719 		tbl->it_type = TCE_PCI_SWINV_CREATE | TCE_PCI_SWINV_FREE;
720 	}
721 	return tbl;
722 }
723 
724 static void pnv_pci_dma_fallback_setup(struct pci_controller *hose,
725 				       struct pci_dev *pdev)
726 {
727 	struct device_node *np = pci_bus_to_OF_node(hose->bus);
728 	struct pci_dn *pdn;
729 
730 	if (np == NULL)
731 		return;
732 	pdn = PCI_DN(np);
733 	if (!pdn->iommu_table)
734 		pdn->iommu_table = pnv_pci_setup_bml_iommu(hose);
735 	if (!pdn->iommu_table)
736 		return;
737 	set_iommu_table_base_and_group(&pdev->dev, pdn->iommu_table);
738 }
739 
740 static void pnv_pci_dma_dev_setup(struct pci_dev *pdev)
741 {
742 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
743 	struct pnv_phb *phb = hose->private_data;
744 
745 	/* If we have no phb structure, try to setup a fallback based on
746 	 * the device-tree (RTAS PCI for example)
747 	 */
748 	if (phb && phb->dma_dev_setup)
749 		phb->dma_dev_setup(phb, pdev);
750 	else
751 		pnv_pci_dma_fallback_setup(hose, pdev);
752 }
753 
754 int pnv_pci_dma_set_mask(struct pci_dev *pdev, u64 dma_mask)
755 {
756 	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
757 	struct pnv_phb *phb = hose->private_data;
758 
759 	if (phb && phb->dma_set_mask)
760 		return phb->dma_set_mask(phb, pdev, dma_mask);
761 	return __dma_set_mask(&pdev->dev, dma_mask);
762 }
763 
764 void pnv_pci_shutdown(void)
765 {
766 	struct pci_controller *hose;
767 
768 	list_for_each_entry(hose, &hose_list, list_node) {
769 		struct pnv_phb *phb = hose->private_data;
770 
771 		if (phb && phb->shutdown)
772 			phb->shutdown(phb);
773 	}
774 }
775 
776 /* Fixup wrong class code in p7ioc and p8 root complex */
777 static void pnv_p7ioc_rc_quirk(struct pci_dev *dev)
778 {
779 	dev->class = PCI_CLASS_BRIDGE_PCI << 8;
780 }
781 DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_IBM, 0x3b9, pnv_p7ioc_rc_quirk);
782 
783 static int pnv_pci_probe_mode(struct pci_bus *bus)
784 {
785 	struct pci_controller *hose = pci_bus_to_host(bus);
786 	const __be64 *tstamp;
787 	u64 now, target;
788 
789 
790 	/* We hijack this as a way to ensure we have waited long
791 	 * enough since the reset was lifted on the PCI bus
792 	 */
793 	if (bus != hose->bus)
794 		return PCI_PROBE_NORMAL;
795 	tstamp = of_get_property(hose->dn, "reset-clear-timestamp", NULL);
796 	if (!tstamp || !*tstamp)
797 		return PCI_PROBE_NORMAL;
798 
799 	now = mftb() / tb_ticks_per_usec;
800 	target = (be64_to_cpup(tstamp) / tb_ticks_per_usec)
801 		+ PCI_RESET_DELAY_US;
802 
803 	pr_devel("pci %04d: Reset target: 0x%llx now: 0x%llx\n",
804 		 hose->global_number, target, now);
805 
806 	if (now < target)
807 		msleep((target - now + 999) / 1000);
808 
809 	return PCI_PROBE_NORMAL;
810 }
811 
812 void __init pnv_pci_init(void)
813 {
814 	struct device_node *np;
815 
816 	pci_add_flags(PCI_CAN_SKIP_ISA_ALIGN);
817 
818 	/* OPAL absent, try POPAL first then RTAS detection of PHBs */
819 	if (!firmware_has_feature(FW_FEATURE_OPAL)) {
820 #ifdef CONFIG_PPC_POWERNV_RTAS
821 		init_pci_config_tokens();
822 		find_and_init_phbs();
823 #endif /* CONFIG_PPC_POWERNV_RTAS */
824 	}
825 	/* OPAL is here, do our normal stuff */
826 	else {
827 		int found_ioda = 0;
828 
829 		/* Look for IODA IO-Hubs. We don't support mixing IODA
830 		 * and p5ioc2 due to the need to change some global
831 		 * probing flags
832 		 */
833 		for_each_compatible_node(np, NULL, "ibm,ioda-hub") {
834 			pnv_pci_init_ioda_hub(np);
835 			found_ioda = 1;
836 		}
837 
838 		/* Look for p5ioc2 IO-Hubs */
839 		if (!found_ioda)
840 			for_each_compatible_node(np, NULL, "ibm,p5ioc2")
841 				pnv_pci_init_p5ioc2_hub(np);
842 
843 		/* Look for ioda2 built-in PHB3's */
844 		for_each_compatible_node(np, NULL, "ibm,ioda2-phb")
845 			pnv_pci_init_ioda2_phb(np);
846 	}
847 
848 	/* Setup the linkage between OF nodes and PHBs */
849 	pci_devs_phb_init();
850 
851 	/* Configure IOMMU DMA hooks */
852 	ppc_md.pci_dma_dev_setup = pnv_pci_dma_dev_setup;
853 	ppc_md.tce_build = pnv_tce_build_vm;
854 	ppc_md.tce_free = pnv_tce_free_vm;
855 	ppc_md.tce_build_rm = pnv_tce_build_rm;
856 	ppc_md.tce_free_rm = pnv_tce_free_rm;
857 	ppc_md.tce_get = pnv_tce_get;
858 	ppc_md.pci_probe_mode = pnv_pci_probe_mode;
859 	set_pci_dma_ops(&dma_iommu_ops);
860 
861 	/* Configure MSIs */
862 #ifdef CONFIG_PCI_MSI
863 	ppc_md.msi_check_device = pnv_msi_check_device;
864 	ppc_md.setup_msi_irqs = pnv_setup_msi_irqs;
865 	ppc_md.teardown_msi_irqs = pnv_teardown_msi_irqs;
866 #endif
867 }
868 
869 static int tce_iommu_bus_notifier(struct notifier_block *nb,
870 		unsigned long action, void *data)
871 {
872 	struct device *dev = data;
873 
874 	switch (action) {
875 	case BUS_NOTIFY_ADD_DEVICE:
876 		return iommu_add_device(dev);
877 	case BUS_NOTIFY_DEL_DEVICE:
878 		if (dev->iommu_group)
879 			iommu_del_device(dev);
880 		return 0;
881 	default:
882 		return 0;
883 	}
884 }
885 
886 static struct notifier_block tce_iommu_bus_nb = {
887 	.notifier_call = tce_iommu_bus_notifier,
888 };
889 
890 static int __init tce_iommu_bus_notifier_init(void)
891 {
892 	bus_register_notifier(&pci_bus_type, &tce_iommu_bus_nb);
893 	return 0;
894 }
895 machine_subsys_initcall_sync(powernv, tce_iommu_bus_notifier_init);
896