1 /*
2  * The file intends to implement the platform dependent EEH operations on
3  * powernv platform. Actually, the powernv was created in order to fully
4  * hypervisor support.
5  *
6  * Copyright Benjamin Herrenschmidt & Gavin Shan, IBM Corporation 2013.
7  *
8  * This program is free software; you can redistribute it and/or modify
9  * it under the terms of the GNU General Public License as published by
10  * the Free Software Foundation; either version 2 of the License, or
11  * (at your option) any later version.
12  */
13 
14 #include <linux/atomic.h>
15 #include <linux/delay.h>
16 #include <linux/export.h>
17 #include <linux/init.h>
18 #include <linux/list.h>
19 #include <linux/msi.h>
20 #include <linux/of.h>
21 #include <linux/pci.h>
22 #include <linux/proc_fs.h>
23 #include <linux/rbtree.h>
24 #include <linux/sched.h>
25 #include <linux/seq_file.h>
26 #include <linux/spinlock.h>
27 
28 #include <asm/eeh.h>
29 #include <asm/eeh_event.h>
30 #include <asm/firmware.h>
31 #include <asm/io.h>
32 #include <asm/iommu.h>
33 #include <asm/machdep.h>
34 #include <asm/msi_bitmap.h>
35 #include <asm/opal.h>
36 #include <asm/ppc-pci.h>
37 
38 #include "powernv.h"
39 #include "pci.h"
40 
41 /**
42  * powernv_eeh_init - EEH platform dependent initialization
43  *
44  * EEH platform dependent initialization on powernv
45  */
46 static int powernv_eeh_init(void)
47 {
48 	struct pci_controller *hose;
49 	struct pnv_phb *phb;
50 
51 	/* We require OPALv3 */
52 	if (!firmware_has_feature(FW_FEATURE_OPALv3)) {
53 		pr_warn("%s: OPALv3 is required !\n",
54 			__func__);
55 		return -EINVAL;
56 	}
57 
58 	/* Set probe mode */
59 	eeh_add_flag(EEH_PROBE_MODE_DEV);
60 
61 	/*
62 	 * P7IOC blocks PCI config access to frozen PE, but PHB3
63 	 * doesn't do that. So we have to selectively enable I/O
64 	 * prior to collecting error log.
65 	 */
66 	list_for_each_entry(hose, &hose_list, list_node) {
67 		phb = hose->private_data;
68 
69 		if (phb->model == PNV_PHB_MODEL_P7IOC)
70 			eeh_add_flag(EEH_ENABLE_IO_FOR_LOG);
71 		break;
72 	}
73 
74 	return 0;
75 }
76 
77 /**
78  * powernv_eeh_post_init - EEH platform dependent post initialization
79  *
80  * EEH platform dependent post initialization on powernv. When
81  * the function is called, the EEH PEs and devices should have
82  * been built. If the I/O cache staff has been built, EEH is
83  * ready to supply service.
84  */
85 static int powernv_eeh_post_init(void)
86 {
87 	struct pci_controller *hose;
88 	struct pnv_phb *phb;
89 	int ret = 0;
90 
91 	list_for_each_entry(hose, &hose_list, list_node) {
92 		phb = hose->private_data;
93 
94 		if (phb->eeh_ops && phb->eeh_ops->post_init) {
95 			ret = phb->eeh_ops->post_init(hose);
96 			if (ret)
97 				break;
98 		}
99 	}
100 
101 	return ret;
102 }
103 
104 /**
105  * powernv_eeh_dev_probe - Do probe on PCI device
106  * @dev: PCI device
107  * @flag: unused
108  *
109  * When EEH module is installed during system boot, all PCI devices
110  * are checked one by one to see if it supports EEH. The function
111  * is introduced for the purpose. By default, EEH has been enabled
112  * on all PCI devices. That's to say, we only need do necessary
113  * initialization on the corresponding eeh device and create PE
114  * accordingly.
115  *
116  * It's notable that's unsafe to retrieve the EEH device through
117  * the corresponding PCI device. During the PCI device hotplug, which
118  * was possiblly triggered by EEH core, the binding between EEH device
119  * and the PCI device isn't built yet.
120  */
121 static int powernv_eeh_dev_probe(struct pci_dev *dev, void *flag)
122 {
123 	struct pci_controller *hose = pci_bus_to_host(dev->bus);
124 	struct pnv_phb *phb = hose->private_data;
125 	struct device_node *dn = pci_device_to_OF_node(dev);
126 	struct eeh_dev *edev = of_node_to_eeh_dev(dn);
127 	int ret;
128 
129 	/*
130 	 * When probing the root bridge, which doesn't have any
131 	 * subordinate PCI devices. We don't have OF node for
132 	 * the root bridge. So it's not reasonable to continue
133 	 * the probing.
134 	 */
135 	if (!dn || !edev || edev->pe)
136 		return 0;
137 
138 	/* Skip for PCI-ISA bridge */
139 	if ((dev->class >> 8) == PCI_CLASS_BRIDGE_ISA)
140 		return 0;
141 
142 	/* Initialize eeh device */
143 	edev->class_code = dev->class;
144 	edev->mode	&= 0xFFFFFF00;
145 	if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE)
146 		edev->mode |= EEH_DEV_BRIDGE;
147 	edev->pcix_cap = pci_find_capability(dev, PCI_CAP_ID_PCIX);
148 	if (pci_is_pcie(dev)) {
149 		edev->pcie_cap = pci_pcie_cap(dev);
150 
151 		if (pci_pcie_type(dev) == PCI_EXP_TYPE_ROOT_PORT)
152 			edev->mode |= EEH_DEV_ROOT_PORT;
153 		else if (pci_pcie_type(dev) == PCI_EXP_TYPE_DOWNSTREAM)
154 			edev->mode |= EEH_DEV_DS_PORT;
155 
156 		edev->aer_cap = pci_find_ext_capability(dev,
157 							PCI_EXT_CAP_ID_ERR);
158 	}
159 
160 	edev->config_addr	= ((dev->bus->number << 8) | dev->devfn);
161 	edev->pe_config_addr	= phb->bdfn_to_pe(phb, dev->bus, dev->devfn & 0xff);
162 
163 	/* Create PE */
164 	ret = eeh_add_to_parent_pe(edev);
165 	if (ret) {
166 		pr_warn("%s: Can't add PCI dev %s to parent PE (%d)\n",
167 			__func__, pci_name(dev), ret);
168 		return ret;
169 	}
170 
171 	/*
172 	 * If the PE contains any one of following adapters, the
173 	 * PCI config space can't be accessed when dumping EEH log.
174 	 * Otherwise, we will run into fenced PHB caused by shortage
175 	 * of outbound credits in the adapter. The PCI config access
176 	 * should be blocked until PE reset. MMIO access is dropped
177 	 * by hardware certainly. In order to drop PCI config requests,
178 	 * one more flag (EEH_PE_CFG_RESTRICTED) is introduced, which
179 	 * will be checked in the backend for PE state retrival. If
180 	 * the PE becomes frozen for the first time and the flag has
181 	 * been set for the PE, we will set EEH_PE_CFG_BLOCKED for
182 	 * that PE to block its config space.
183 	 *
184 	 * Broadcom Austin 4-ports NICs (14e4:1657)
185 	 * Broadcom Shiner 2-ports 10G NICs (14e4:168e)
186 	 */
187 	if ((dev->vendor == PCI_VENDOR_ID_BROADCOM && dev->device == 0x1657) ||
188 	    (dev->vendor == PCI_VENDOR_ID_BROADCOM && dev->device == 0x168e))
189 		edev->pe->state |= EEH_PE_CFG_RESTRICTED;
190 
191 	/*
192 	 * Cache the PE primary bus, which can't be fetched when
193 	 * full hotplug is in progress. In that case, all child
194 	 * PCI devices of the PE are expected to be removed prior
195 	 * to PE reset.
196 	 */
197 	if (!edev->pe->bus)
198 		edev->pe->bus = dev->bus;
199 
200 	/*
201 	 * Enable EEH explicitly so that we will do EEH check
202 	 * while accessing I/O stuff
203 	 */
204 	eeh_add_flag(EEH_ENABLED);
205 
206 	/* Save memory bars */
207 	eeh_save_bars(edev);
208 
209 	return 0;
210 }
211 
212 /**
213  * powernv_eeh_set_option - Initialize EEH or MMIO/DMA reenable
214  * @pe: EEH PE
215  * @option: operation to be issued
216  *
217  * The function is used to control the EEH functionality globally.
218  * Currently, following options are support according to PAPR:
219  * Enable EEH, Disable EEH, Enable MMIO and Enable DMA
220  */
221 static int powernv_eeh_set_option(struct eeh_pe *pe, int option)
222 {
223 	struct pci_controller *hose = pe->phb;
224 	struct pnv_phb *phb = hose->private_data;
225 	int ret = -EEXIST;
226 
227 	/*
228 	 * What we need do is pass it down for hardware
229 	 * implementation to handle it.
230 	 */
231 	if (phb->eeh_ops && phb->eeh_ops->set_option)
232 		ret = phb->eeh_ops->set_option(pe, option);
233 
234 	return ret;
235 }
236 
237 /**
238  * powernv_eeh_get_pe_addr - Retrieve PE address
239  * @pe: EEH PE
240  *
241  * Retrieve the PE address according to the given tranditional
242  * PCI BDF (Bus/Device/Function) address.
243  */
244 static int powernv_eeh_get_pe_addr(struct eeh_pe *pe)
245 {
246 	return pe->addr;
247 }
248 
249 /**
250  * powernv_eeh_get_state - Retrieve PE state
251  * @pe: EEH PE
252  * @delay: delay while PE state is temporarily unavailable
253  *
254  * Retrieve the state of the specified PE. For IODA-compitable
255  * platform, it should be retrieved from IODA table. Therefore,
256  * we prefer passing down to hardware implementation to handle
257  * it.
258  */
259 static int powernv_eeh_get_state(struct eeh_pe *pe, int *delay)
260 {
261 	struct pci_controller *hose = pe->phb;
262 	struct pnv_phb *phb = hose->private_data;
263 	int ret = EEH_STATE_NOT_SUPPORT;
264 
265 	if (phb->eeh_ops && phb->eeh_ops->get_state) {
266 		ret = phb->eeh_ops->get_state(pe);
267 
268 		/*
269 		 * If the PE state is temporarily unavailable,
270 		 * to inform the EEH core delay for default
271 		 * period (1 second)
272 		 */
273 		if (delay) {
274 			*delay = 0;
275 			if (ret & EEH_STATE_UNAVAILABLE)
276 				*delay = 1000;
277 		}
278 	}
279 
280 	return ret;
281 }
282 
283 /**
284  * powernv_eeh_reset - Reset the specified PE
285  * @pe: EEH PE
286  * @option: reset option
287  *
288  * Reset the specified PE
289  */
290 static int powernv_eeh_reset(struct eeh_pe *pe, int option)
291 {
292 	struct pci_controller *hose = pe->phb;
293 	struct pnv_phb *phb = hose->private_data;
294 	int ret = -EEXIST;
295 
296 	if (phb->eeh_ops && phb->eeh_ops->reset)
297 		ret = phb->eeh_ops->reset(pe, option);
298 
299 	return ret;
300 }
301 
302 /**
303  * powernv_eeh_wait_state - Wait for PE state
304  * @pe: EEH PE
305  * @max_wait: maximal period in microsecond
306  *
307  * Wait for the state of associated PE. It might take some time
308  * to retrieve the PE's state.
309  */
310 static int powernv_eeh_wait_state(struct eeh_pe *pe, int max_wait)
311 {
312 	int ret;
313 	int mwait;
314 
315 	while (1) {
316 		ret = powernv_eeh_get_state(pe, &mwait);
317 
318 		/*
319 		 * If the PE's state is temporarily unavailable,
320 		 * we have to wait for the specified time. Otherwise,
321 		 * the PE's state will be returned immediately.
322 		 */
323 		if (ret != EEH_STATE_UNAVAILABLE)
324 			return ret;
325 
326 		max_wait -= mwait;
327 		if (max_wait <= 0) {
328 			pr_warn("%s: Timeout getting PE#%x's state (%d)\n",
329 				__func__, pe->addr, max_wait);
330 			return EEH_STATE_NOT_SUPPORT;
331 		}
332 
333 		msleep(mwait);
334 	}
335 
336 	return EEH_STATE_NOT_SUPPORT;
337 }
338 
339 /**
340  * powernv_eeh_get_log - Retrieve error log
341  * @pe: EEH PE
342  * @severity: temporary or permanent error log
343  * @drv_log: driver log to be combined with retrieved error log
344  * @len: length of driver log
345  *
346  * Retrieve the temporary or permanent error from the PE.
347  */
348 static int powernv_eeh_get_log(struct eeh_pe *pe, int severity,
349 			       char *drv_log, unsigned long len)
350 {
351 	struct pci_controller *hose = pe->phb;
352 	struct pnv_phb *phb = hose->private_data;
353 	int ret = -EEXIST;
354 
355 	if (phb->eeh_ops && phb->eeh_ops->get_log)
356 		ret = phb->eeh_ops->get_log(pe, severity, drv_log, len);
357 
358 	return ret;
359 }
360 
361 /**
362  * powernv_eeh_configure_bridge - Configure PCI bridges in the indicated PE
363  * @pe: EEH PE
364  *
365  * The function will be called to reconfigure the bridges included
366  * in the specified PE so that the mulfunctional PE would be recovered
367  * again.
368  */
369 static int powernv_eeh_configure_bridge(struct eeh_pe *pe)
370 {
371 	struct pci_controller *hose = pe->phb;
372 	struct pnv_phb *phb = hose->private_data;
373 	int ret = 0;
374 
375 	if (phb->eeh_ops && phb->eeh_ops->configure_bridge)
376 		ret = phb->eeh_ops->configure_bridge(pe);
377 
378 	return ret;
379 }
380 
381 /**
382  * powernv_pe_err_inject - Inject specified error to the indicated PE
383  * @pe: the indicated PE
384  * @type: error type
385  * @func: specific error type
386  * @addr: address
387  * @mask: address mask
388  *
389  * The routine is called to inject specified error, which is
390  * determined by @type and @func, to the indicated PE for
391  * testing purpose.
392  */
393 static int powernv_eeh_err_inject(struct eeh_pe *pe, int type, int func,
394 				  unsigned long addr, unsigned long mask)
395 {
396 	struct pci_controller *hose = pe->phb;
397 	struct pnv_phb *phb = hose->private_data;
398 	int ret = -EEXIST;
399 
400 	if (phb->eeh_ops && phb->eeh_ops->err_inject)
401 		ret = phb->eeh_ops->err_inject(pe, type, func, addr, mask);
402 
403 	return ret;
404 }
405 
406 static inline bool powernv_eeh_cfg_blocked(struct device_node *dn)
407 {
408 	struct eeh_dev *edev = of_node_to_eeh_dev(dn);
409 
410 	if (!edev || !edev->pe)
411 		return false;
412 
413 	if (edev->pe->state & EEH_PE_CFG_BLOCKED)
414 		return true;
415 
416 	return false;
417 }
418 
419 static int powernv_eeh_read_config(struct device_node *dn,
420 				   int where, int size, u32 *val)
421 {
422 	if (powernv_eeh_cfg_blocked(dn)) {
423 		*val = 0xFFFFFFFF;
424 		return PCIBIOS_SET_FAILED;
425 	}
426 
427 	return pnv_pci_cfg_read(dn, where, size, val);
428 }
429 
430 static int powernv_eeh_write_config(struct device_node *dn,
431 				    int where, int size, u32 val)
432 {
433 	if (powernv_eeh_cfg_blocked(dn))
434 		return PCIBIOS_SET_FAILED;
435 
436 	return pnv_pci_cfg_write(dn, where, size, val);
437 }
438 
439 /**
440  * powernv_eeh_next_error - Retrieve next EEH error to handle
441  * @pe: Affected PE
442  *
443  * Using OPAL API, to retrieve next EEH error for EEH core to handle
444  */
445 static int powernv_eeh_next_error(struct eeh_pe **pe)
446 {
447 	struct pci_controller *hose;
448 	struct pnv_phb *phb = NULL;
449 
450 	list_for_each_entry(hose, &hose_list, list_node) {
451 		phb = hose->private_data;
452 		break;
453 	}
454 
455 	if (phb && phb->eeh_ops->next_error)
456 		return phb->eeh_ops->next_error(pe);
457 
458 	return -EEXIST;
459 }
460 
461 static int powernv_eeh_restore_config(struct device_node *dn)
462 {
463 	struct eeh_dev *edev = of_node_to_eeh_dev(dn);
464 	struct pnv_phb *phb;
465 	s64 ret;
466 
467 	if (!edev)
468 		return -EEXIST;
469 
470 	phb = edev->phb->private_data;
471 	ret = opal_pci_reinit(phb->opal_id,
472 			      OPAL_REINIT_PCI_DEV, edev->config_addr);
473 	if (ret) {
474 		pr_warn("%s: Can't reinit PCI dev 0x%x (%lld)\n",
475 			__func__, edev->config_addr, ret);
476 		return -EIO;
477 	}
478 
479 	return 0;
480 }
481 
482 static struct eeh_ops powernv_eeh_ops = {
483 	.name                   = "powernv",
484 	.init                   = powernv_eeh_init,
485 	.post_init              = powernv_eeh_post_init,
486 	.of_probe               = NULL,
487 	.dev_probe              = powernv_eeh_dev_probe,
488 	.set_option             = powernv_eeh_set_option,
489 	.get_pe_addr            = powernv_eeh_get_pe_addr,
490 	.get_state              = powernv_eeh_get_state,
491 	.reset                  = powernv_eeh_reset,
492 	.wait_state             = powernv_eeh_wait_state,
493 	.get_log                = powernv_eeh_get_log,
494 	.configure_bridge       = powernv_eeh_configure_bridge,
495 	.err_inject		= powernv_eeh_err_inject,
496 	.read_config            = powernv_eeh_read_config,
497 	.write_config           = powernv_eeh_write_config,
498 	.next_error		= powernv_eeh_next_error,
499 	.restore_config		= powernv_eeh_restore_config
500 };
501 
502 /**
503  * eeh_powernv_init - Register platform dependent EEH operations
504  *
505  * EEH initialization on powernv platform. This function should be
506  * called before any EEH related functions.
507  */
508 static int __init eeh_powernv_init(void)
509 {
510 	int ret = -EINVAL;
511 
512 	eeh_set_pe_aux_size(PNV_PCI_DIAG_BUF_SIZE);
513 	ret = eeh_ops_register(&powernv_eeh_ops);
514 	if (!ret)
515 		pr_info("EEH: PowerNV platform initialized\n");
516 	else
517 		pr_info("EEH: Failed to initialize PowerNV platform (%d)\n", ret);
518 
519 	return ret;
520 }
521 machine_early_initcall(powernv, eeh_powernv_init);
522