blob: 2809c989528814d8dadf5e79d7aa75e32b6a888d [file] [log] [blame]
Gavin Shan8747f362013-06-20 13:21:06 +08001/*
2 * The file intends to implement the functions needed by EEH, which is
3 * built on IODA compliant chip. Actually, lots of functions related
4 * to EEH would be built based on the OPAL APIs.
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
Gavin Shan89988972013-06-20 18:13:26 +080014#include <linux/debugfs.h>
Gavin Shan8747f362013-06-20 13:21:06 +080015#include <linux/delay.h>
Gavin Shan8747f362013-06-20 13:21:06 +080016#include <linux/io.h>
17#include <linux/irq.h>
18#include <linux/kernel.h>
19#include <linux/msi.h>
Gavin Shan7cb9d932013-06-20 18:13:24 +080020#include <linux/notifier.h>
Gavin Shan8747f362013-06-20 13:21:06 +080021#include <linux/pci.h>
22#include <linux/string.h>
23
24#include <asm/eeh.h>
25#include <asm/eeh_event.h>
26#include <asm/io.h>
27#include <asm/iommu.h>
28#include <asm/msi_bitmap.h>
29#include <asm/opal.h>
30#include <asm/pci-bridge.h>
31#include <asm/ppc-pci.h>
32#include <asm/tce.h>
33
34#include "powernv.h"
35#include "pci.h"
36
Gavin Shan7cb9d932013-06-20 18:13:24 +080037static int ioda_eeh_nb_init = 0;
38
39static int ioda_eeh_event(struct notifier_block *nb,
40 unsigned long events, void *change)
41{
42 uint64_t changed_evts = (uint64_t)change;
43
Gavin Shan7f52a522014-04-24 18:00:18 +100044 /*
45 * We simply send special EEH event if EEH has
46 * been enabled, or clear pending events in
47 * case that we enable EEH soon
48 */
49 if (!(changed_evts & OPAL_EVENT_PCI_ERROR) ||
50 !(events & OPAL_EVENT_PCI_ERROR))
51 return 0;
52
53 if (eeh_enabled())
Gavin Shan7cb9d932013-06-20 18:13:24 +080054 eeh_send_failure_event(NULL);
Gavin Shan7f52a522014-04-24 18:00:18 +100055 else
56 opal_notifier_update_evt(OPAL_EVENT_PCI_ERROR, 0x0ul);
Gavin Shan7cb9d932013-06-20 18:13:24 +080057
58 return 0;
59}
60
61static struct notifier_block ioda_eeh_nb = {
62 .notifier_call = ioda_eeh_event,
63 .next = NULL,
64 .priority = 0
65};
Gavin Shan70f942d2013-06-20 13:21:12 +080066
Gavin Shan89988972013-06-20 18:13:26 +080067#ifdef CONFIG_DEBUG_FS
Mike Qiu7a062782014-09-30 12:38:57 +100068static ssize_t ioda_eeh_ei_write(struct file *filp,
69 const char __user *user_buf,
70 size_t count, loff_t *ppos)
71{
72 struct pci_controller *hose = filp->private_data;
73 struct pnv_phb *phb = hose->private_data;
74 struct eeh_dev *edev;
75 struct eeh_pe *pe;
76 int pe_no, type, func;
77 unsigned long addr, mask;
78 char buf[50];
79 int ret;
80
81 if (!phb->eeh_ops || !phb->eeh_ops->err_inject)
82 return -ENXIO;
83
84 ret = simple_write_to_buffer(buf, sizeof(buf), ppos, user_buf, count);
85 if (!ret)
86 return -EFAULT;
87
88 /* Retrieve parameters */
89 ret = sscanf(buf, "%x:%x:%x:%lx:%lx",
90 &pe_no, &type, &func, &addr, &mask);
91 if (ret != 5)
92 return -EINVAL;
93
94 /* Retrieve PE */
95 edev = kzalloc(sizeof(*edev), GFP_KERNEL);
96 if (!edev)
97 return -ENOMEM;
98 edev->phb = hose;
99 edev->pe_config_addr = pe_no;
100 pe = eeh_pe_get(edev);
101 kfree(edev);
102 if (!pe)
103 return -ENODEV;
104
105 /* Do error injection */
106 ret = phb->eeh_ops->err_inject(pe, type, func, addr, mask);
107 return ret < 0 ? ret : count;
108}
109
110static const struct file_operations ioda_eeh_ei_fops = {
111 .open = simple_open,
112 .llseek = no_llseek,
113 .write = ioda_eeh_ei_write,
114};
115
Gavin Shanff6bdcd2013-09-06 09:00:01 +0800116static int ioda_eeh_dbgfs_set(void *data, int offset, u64 val)
Gavin Shan89988972013-06-20 18:13:26 +0800117{
118 struct pci_controller *hose = data;
119 struct pnv_phb *phb = hose->private_data;
120
Gavin Shanff6bdcd2013-09-06 09:00:01 +0800121 out_be64(phb->regs + offset, val);
Gavin Shan89988972013-06-20 18:13:26 +0800122 return 0;
123}
124
Gavin Shanff6bdcd2013-09-06 09:00:01 +0800125static int ioda_eeh_dbgfs_get(void *data, int offset, u64 *val)
Gavin Shan89988972013-06-20 18:13:26 +0800126{
127 struct pci_controller *hose = data;
128 struct pnv_phb *phb = hose->private_data;
129
Gavin Shanff6bdcd2013-09-06 09:00:01 +0800130 *val = in_be64(phb->regs + offset);
Gavin Shan89988972013-06-20 18:13:26 +0800131 return 0;
132}
133
Gavin Shanff6bdcd2013-09-06 09:00:01 +0800134static int ioda_eeh_outb_dbgfs_set(void *data, u64 val)
135{
136 return ioda_eeh_dbgfs_set(data, 0xD10, val);
137}
138
139static int ioda_eeh_outb_dbgfs_get(void *data, u64 *val)
140{
141 return ioda_eeh_dbgfs_get(data, 0xD10, val);
142}
143
144static int ioda_eeh_inbA_dbgfs_set(void *data, u64 val)
145{
146 return ioda_eeh_dbgfs_set(data, 0xD90, val);
147}
148
149static int ioda_eeh_inbA_dbgfs_get(void *data, u64 *val)
150{
151 return ioda_eeh_dbgfs_get(data, 0xD90, val);
152}
153
154static int ioda_eeh_inbB_dbgfs_set(void *data, u64 val)
155{
156 return ioda_eeh_dbgfs_set(data, 0xE10, val);
157}
158
159static int ioda_eeh_inbB_dbgfs_get(void *data, u64 *val)
160{
161 return ioda_eeh_dbgfs_get(data, 0xE10, val);
162}
163
164DEFINE_SIMPLE_ATTRIBUTE(ioda_eeh_outb_dbgfs_ops, ioda_eeh_outb_dbgfs_get,
165 ioda_eeh_outb_dbgfs_set, "0x%llx\n");
166DEFINE_SIMPLE_ATTRIBUTE(ioda_eeh_inbA_dbgfs_ops, ioda_eeh_inbA_dbgfs_get,
167 ioda_eeh_inbA_dbgfs_set, "0x%llx\n");
168DEFINE_SIMPLE_ATTRIBUTE(ioda_eeh_inbB_dbgfs_ops, ioda_eeh_inbB_dbgfs_get,
169 ioda_eeh_inbB_dbgfs_set, "0x%llx\n");
Gavin Shan89988972013-06-20 18:13:26 +0800170#endif /* CONFIG_DEBUG_FS */
171
Gavin Shan94716602014-02-25 15:28:37 +0800172
Gavin Shan73370c62013-06-20 13:21:07 +0800173/**
174 * ioda_eeh_post_init - Chip dependent post initialization
175 * @hose: PCI controller
176 *
177 * The function will be called after eeh PEs and devices
178 * have been built. That means the EEH is ready to supply
179 * service with I/O cache.
180 */
181static int ioda_eeh_post_init(struct pci_controller *hose)
182{
183 struct pnv_phb *phb = hose->private_data;
Gavin Shan7cb9d932013-06-20 18:13:24 +0800184 int ret;
185
186 /* Register OPAL event notifier */
187 if (!ioda_eeh_nb_init) {
188 ret = opal_notifier_register(&ioda_eeh_nb);
189 if (ret) {
190 pr_err("%s: Can't register OPAL event notifier (%d)\n",
191 __func__, ret);
192 return ret;
193 }
194
195 ioda_eeh_nb_init = 1;
196 }
Gavin Shan73370c62013-06-20 13:21:07 +0800197
Gavin Shan89988972013-06-20 18:13:26 +0800198#ifdef CONFIG_DEBUG_FS
Gavin Shan7f52a522014-04-24 18:00:18 +1000199 if (!phb->has_dbgfs && phb->dbgfs) {
200 phb->has_dbgfs = 1;
201
Mike Qiu7a062782014-09-30 12:38:57 +1000202 debugfs_create_file("err_injct", 0200,
203 phb->dbgfs, hose,
204 &ioda_eeh_ei_fops);
205
Gavin Shanff6bdcd2013-09-06 09:00:01 +0800206 debugfs_create_file("err_injct_outbound", 0600,
Gavin Shan20bb8422013-09-06 09:00:00 +0800207 phb->dbgfs, hose,
Gavin Shanff6bdcd2013-09-06 09:00:01 +0800208 &ioda_eeh_outb_dbgfs_ops);
209 debugfs_create_file("err_injct_inboundA", 0600,
210 phb->dbgfs, hose,
211 &ioda_eeh_inbA_dbgfs_ops);
212 debugfs_create_file("err_injct_inboundB", 0600,
213 phb->dbgfs, hose,
214 &ioda_eeh_inbB_dbgfs_ops);
215 }
Gavin Shan89988972013-06-20 18:13:26 +0800216#endif
217
Gavin Shan7f52a522014-04-24 18:00:18 +1000218 /* If EEH is enabled, we're going to rely on that.
219 * Otherwise, we restore to conventional mechanism
220 * to clear frozen PE during PCI config access.
221 */
222 if (eeh_enabled())
223 phb->flags |= PNV_PHB_FLAG_EEH;
224 else
225 phb->flags &= ~PNV_PHB_FLAG_EEH;
Gavin Shan73370c62013-06-20 13:21:07 +0800226
227 return 0;
228}
229
Gavin Shaneb005982013-06-20 13:21:08 +0800230/**
231 * ioda_eeh_set_option - Set EEH operation or I/O setting
232 * @pe: EEH PE
233 * @option: options
234 *
235 * Enable or disable EEH option for the indicated PE. The
236 * function also can be used to enable I/O or DMA for the
237 * PE.
238 */
239static int ioda_eeh_set_option(struct eeh_pe *pe, int option)
240{
Gavin Shaneb005982013-06-20 13:21:08 +0800241 struct pci_controller *hose = pe->phb;
242 struct pnv_phb *phb = hose->private_data;
Gavin Shan0d5ee522014-09-30 12:38:52 +1000243 bool freeze_pe = false;
Gavin Shan58287902014-07-21 14:42:34 +1000244 int enable, ret = 0;
245 s64 rc;
Gavin Shaneb005982013-06-20 13:21:08 +0800246
247 /* Check on PE number */
248 if (pe->addr < 0 || pe->addr >= phb->ioda.total_pe) {
249 pr_err("%s: PE address %x out of range [0, %x] "
250 "on PHB#%x\n",
251 __func__, pe->addr, phb->ioda.total_pe,
252 hose->global_number);
253 return -EINVAL;
254 }
255
Gavin Shaneb005982013-06-20 13:21:08 +0800256 switch (option) {
257 case EEH_OPT_DISABLE:
Gavin Shan58287902014-07-21 14:42:34 +1000258 return -EPERM;
Gavin Shaneb005982013-06-20 13:21:08 +0800259 case EEH_OPT_ENABLE:
Gavin Shan58287902014-07-21 14:42:34 +1000260 return 0;
Gavin Shaneb005982013-06-20 13:21:08 +0800261 case EEH_OPT_THAW_MMIO:
Gavin Shan58287902014-07-21 14:42:34 +1000262 enable = OPAL_EEH_ACTION_CLEAR_FREEZE_MMIO;
Gavin Shaneb005982013-06-20 13:21:08 +0800263 break;
264 case EEH_OPT_THAW_DMA:
Gavin Shan58287902014-07-21 14:42:34 +1000265 enable = OPAL_EEH_ACTION_CLEAR_FREEZE_DMA;
Gavin Shaneb005982013-06-20 13:21:08 +0800266 break;
Gavin Shan0d5ee522014-09-30 12:38:52 +1000267 case EEH_OPT_FREEZE_PE:
268 freeze_pe = true;
269 enable = OPAL_EEH_ACTION_SET_FREEZE_ALL;
270 break;
Gavin Shaneb005982013-06-20 13:21:08 +0800271 default:
Gavin Shan58287902014-07-21 14:42:34 +1000272 pr_warn("%s: Invalid option %d\n",
273 __func__, option);
Gavin Shaneb005982013-06-20 13:21:08 +0800274 return -EINVAL;
275 }
276
Gavin Shan58287902014-07-21 14:42:34 +1000277 /* If PHB supports compound PE, to handle it */
Gavin Shan0d5ee522014-09-30 12:38:52 +1000278 if (freeze_pe) {
279 if (phb->freeze_pe) {
280 phb->freeze_pe(phb, pe->addr);
281 } else {
282 rc = opal_pci_eeh_freeze_set(phb->opal_id,
283 pe->addr,
284 enable);
285 if (rc != OPAL_SUCCESS) {
286 pr_warn("%s: Failure %lld freezing "
287 "PHB#%x-PE#%x\n",
288 __func__, rc,
289 phb->hose->global_number, pe->addr);
290 ret = -EIO;
291 }
292 }
Gavin Shan58287902014-07-21 14:42:34 +1000293 } else {
Gavin Shan0d5ee522014-09-30 12:38:52 +1000294 if (phb->unfreeze_pe) {
295 ret = phb->unfreeze_pe(phb, pe->addr, enable);
296 } else {
297 rc = opal_pci_eeh_freeze_clear(phb->opal_id,
298 pe->addr,
299 enable);
300 if (rc != OPAL_SUCCESS) {
301 pr_warn("%s: Failure %lld enable %d "
302 "for PHB#%x-PE#%x\n",
303 __func__, rc, option,
304 phb->hose->global_number, pe->addr);
305 ret = -EIO;
306 }
Gavin Shan58287902014-07-21 14:42:34 +1000307 }
308 }
309
Gavin Shaneb005982013-06-20 13:21:08 +0800310 return ret;
311}
312
Gavin Shanbb593c02014-07-17 14:41:43 +1000313static void ioda_eeh_phb_diag(struct eeh_pe *pe)
Gavin Shan94716602014-02-25 15:28:37 +0800314{
Gavin Shanbb593c02014-07-17 14:41:43 +1000315 struct pnv_phb *phb = pe->phb->private_data;
Gavin Shan94716602014-02-25 15:28:37 +0800316 long rc;
317
Gavin Shanbb593c02014-07-17 14:41:43 +1000318 rc = opal_pci_get_phb_diag_data2(phb->opal_id, pe->data,
Gavin Shan94716602014-02-25 15:28:37 +0800319 PNV_PCI_DIAG_BUF_SIZE);
Gavin Shanbb593c02014-07-17 14:41:43 +1000320 if (rc != OPAL_SUCCESS)
Gavin Shan0dae2742014-07-17 14:41:41 +1000321 pr_warn("%s: Failed to get diag-data for PHB#%x (%ld)\n",
Gavin Shanbb593c02014-07-17 14:41:43 +1000322 __func__, pe->phb->global_number, rc);
Gavin Shan94716602014-02-25 15:28:37 +0800323}
324
Gavin Shanc979c702014-07-21 14:42:32 +1000325static int ioda_eeh_get_phb_state(struct eeh_pe *pe)
326{
327 struct pnv_phb *phb = pe->phb->private_data;
328 u8 fstate;
329 __be16 pcierr;
330 s64 rc;
331 int result = 0;
332
333 rc = opal_pci_eeh_freeze_status(phb->opal_id,
334 pe->addr,
335 &fstate,
336 &pcierr,
337 NULL);
338 if (rc != OPAL_SUCCESS) {
339 pr_warn("%s: Failure %lld getting PHB#%x state\n",
340 __func__, rc, phb->hose->global_number);
341 return EEH_STATE_NOT_SUPPORT;
342 }
343
344 /*
345 * Check PHB state. If the PHB is frozen for the
346 * first time, to dump the PHB diag-data.
347 */
348 if (be16_to_cpu(pcierr) != OPAL_EEH_PHB_ERROR) {
349 result = (EEH_STATE_MMIO_ACTIVE |
350 EEH_STATE_DMA_ACTIVE |
351 EEH_STATE_MMIO_ENABLED |
352 EEH_STATE_DMA_ENABLED);
353 } else if (!(pe->state & EEH_PE_ISOLATED)) {
354 eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
355 ioda_eeh_phb_diag(pe);
Gavin Shana450e8f2014-11-22 21:58:09 +1100356
357 if (eeh_has_flag(EEH_EARLY_DUMP_LOG))
358 pnv_pci_dump_phb_diag_data(pe->phb, pe->data);
Gavin Shanc979c702014-07-21 14:42:32 +1000359 }
360
361 return result;
362}
363
364static int ioda_eeh_get_pe_state(struct eeh_pe *pe)
365{
366 struct pnv_phb *phb = pe->phb->private_data;
367 u8 fstate;
368 __be16 pcierr;
369 s64 rc;
370 int result;
371
372 /*
373 * We don't clobber hardware frozen state until PE
374 * reset is completed. In order to keep EEH core
375 * moving forward, we have to return operational
376 * state during PE reset.
377 */
Gavin Shan28bf36f2014-11-14 10:47:29 +1100378 if (pe->state & EEH_PE_RESET) {
Gavin Shanc979c702014-07-21 14:42:32 +1000379 result = (EEH_STATE_MMIO_ACTIVE |
380 EEH_STATE_DMA_ACTIVE |
381 EEH_STATE_MMIO_ENABLED |
382 EEH_STATE_DMA_ENABLED);
383 return result;
384 }
385
Gavin Shan58287902014-07-21 14:42:34 +1000386 /*
387 * Fetch PE state from hardware. If the PHB
388 * supports compound PE, let it handle that.
389 */
390 if (phb->get_pe_state) {
391 fstate = phb->get_pe_state(phb, pe->addr);
392 } else {
393 rc = opal_pci_eeh_freeze_status(phb->opal_id,
394 pe->addr,
395 &fstate,
396 &pcierr,
397 NULL);
398 if (rc != OPAL_SUCCESS) {
399 pr_warn("%s: Failure %lld getting PHB#%x-PE%x state\n",
400 __func__, rc, phb->hose->global_number, pe->addr);
401 return EEH_STATE_NOT_SUPPORT;
402 }
Gavin Shanc979c702014-07-21 14:42:32 +1000403 }
404
405 /* Figure out state */
406 switch (fstate) {
407 case OPAL_EEH_STOPPED_NOT_FROZEN:
408 result = (EEH_STATE_MMIO_ACTIVE |
409 EEH_STATE_DMA_ACTIVE |
410 EEH_STATE_MMIO_ENABLED |
411 EEH_STATE_DMA_ENABLED);
412 break;
413 case OPAL_EEH_STOPPED_MMIO_FREEZE:
414 result = (EEH_STATE_DMA_ACTIVE |
415 EEH_STATE_DMA_ENABLED);
416 break;
417 case OPAL_EEH_STOPPED_DMA_FREEZE:
418 result = (EEH_STATE_MMIO_ACTIVE |
419 EEH_STATE_MMIO_ENABLED);
420 break;
421 case OPAL_EEH_STOPPED_MMIO_DMA_FREEZE:
422 result = 0;
423 break;
424 case OPAL_EEH_STOPPED_RESET:
425 result = EEH_STATE_RESET_ACTIVE;
426 break;
427 case OPAL_EEH_STOPPED_TEMP_UNAVAIL:
428 result = EEH_STATE_UNAVAILABLE;
429 break;
430 case OPAL_EEH_STOPPED_PERM_UNAVAIL:
431 result = EEH_STATE_NOT_SUPPORT;
432 break;
433 default:
434 result = EEH_STATE_NOT_SUPPORT;
435 pr_warn("%s: Invalid PHB#%x-PE#%x state %x\n",
436 __func__, phb->hose->global_number,
437 pe->addr, fstate);
438 }
439
440 /*
Gavin Shan58287902014-07-21 14:42:34 +1000441 * If PHB supports compound PE, to freeze all
442 * slave PEs for consistency.
443 *
Gavin Shanc979c702014-07-21 14:42:32 +1000444 * If the PE is switching to frozen state for the
445 * first time, to dump the PHB diag-data.
446 */
447 if (!(result & EEH_STATE_NOT_SUPPORT) &&
448 !(result & EEH_STATE_UNAVAILABLE) &&
449 !(result & EEH_STATE_MMIO_ACTIVE) &&
450 !(result & EEH_STATE_DMA_ACTIVE) &&
451 !(pe->state & EEH_PE_ISOLATED)) {
Gavin Shan58287902014-07-21 14:42:34 +1000452 if (phb->freeze_pe)
453 phb->freeze_pe(phb, pe->addr);
454
Gavin Shanc979c702014-07-21 14:42:32 +1000455 eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
456 ioda_eeh_phb_diag(pe);
Gavin Shana450e8f2014-11-22 21:58:09 +1100457
458 if (eeh_has_flag(EEH_EARLY_DUMP_LOG))
459 pnv_pci_dump_phb_diag_data(pe->phb, pe->data);
Gavin Shanc979c702014-07-21 14:42:32 +1000460 }
461
462 return result;
463}
464
Gavin Shan8c41a7f2013-06-20 13:21:09 +0800465/**
466 * ioda_eeh_get_state - Retrieve the state of PE
467 * @pe: EEH PE
468 *
469 * The PE's state should be retrieved from the PEEV, PEST
470 * IODA tables. Since the OPAL has exported the function
471 * to do it, it'd better to use that.
472 */
473static int ioda_eeh_get_state(struct eeh_pe *pe)
474{
Gavin Shanc979c702014-07-21 14:42:32 +1000475 struct pnv_phb *phb = pe->phb->private_data;
Gavin Shan8c41a7f2013-06-20 13:21:09 +0800476
Gavin Shanc979c702014-07-21 14:42:32 +1000477 /* Sanity check on PE number. PHB PE should have 0 */
478 if (pe->addr < 0 ||
479 pe->addr >= phb->ioda.total_pe) {
480 pr_warn("%s: PHB#%x-PE#%x out of range [0, %x]\n",
481 __func__, phb->hose->global_number,
482 pe->addr, phb->ioda.total_pe);
Gavin Shan8c41a7f2013-06-20 13:21:09 +0800483 return EEH_STATE_NOT_SUPPORT;
484 }
485
Gavin Shanc979c702014-07-21 14:42:32 +1000486 if (pe->type & EEH_PE_PHB)
487 return ioda_eeh_get_phb_state(pe);
Gavin Shan78954702014-04-24 18:00:14 +1000488
Gavin Shanc979c702014-07-21 14:42:32 +1000489 return ioda_eeh_get_pe_state(pe);
Gavin Shan8c41a7f2013-06-20 13:21:09 +0800490}
491
Gavin Shan9d5cab02013-06-20 13:21:10 +0800492static s64 ioda_eeh_phb_poll(struct pnv_phb *phb)
493{
494 s64 rc = OPAL_HARDWARE;
495
496 while (1) {
497 rc = opal_pci_poll(phb->opal_id);
498 if (rc <= 0)
499 break;
500
Gavin Shan361f2a22014-04-24 18:00:25 +1000501 if (system_state < SYSTEM_RUNNING)
502 udelay(1000 * rc);
503 else
504 msleep(rc);
Gavin Shan9d5cab02013-06-20 13:21:10 +0800505 }
506
507 return rc;
508}
509
Gavin Shan361f2a22014-04-24 18:00:25 +1000510int ioda_eeh_phb_reset(struct pci_controller *hose, int option)
Gavin Shan9d5cab02013-06-20 13:21:10 +0800511{
512 struct pnv_phb *phb = hose->private_data;
513 s64 rc = OPAL_HARDWARE;
514
515 pr_debug("%s: Reset PHB#%x, option=%d\n",
516 __func__, hose->global_number, option);
517
518 /* Issue PHB complete reset request */
519 if (option == EEH_RESET_FUNDAMENTAL ||
520 option == EEH_RESET_HOT)
521 rc = opal_pci_reset(phb->opal_id,
Gavin Shand1a85ee2014-09-30 12:39:05 +1000522 OPAL_RESET_PHB_COMPLETE,
Gavin Shan9d5cab02013-06-20 13:21:10 +0800523 OPAL_ASSERT_RESET);
524 else if (option == EEH_RESET_DEACTIVATE)
525 rc = opal_pci_reset(phb->opal_id,
Gavin Shand1a85ee2014-09-30 12:39:05 +1000526 OPAL_RESET_PHB_COMPLETE,
Gavin Shan9d5cab02013-06-20 13:21:10 +0800527 OPAL_DEASSERT_RESET);
528 if (rc < 0)
529 goto out;
530
531 /*
532 * Poll state of the PHB until the request is done
Gavin Shan26833a52014-04-24 18:00:23 +1000533 * successfully. The PHB reset is usually PHB complete
534 * reset followed by hot reset on root bus. So we also
535 * need the PCI bus settlement delay.
Gavin Shan9d5cab02013-06-20 13:21:10 +0800536 */
537 rc = ioda_eeh_phb_poll(phb);
Gavin Shan361f2a22014-04-24 18:00:25 +1000538 if (option == EEH_RESET_DEACTIVATE) {
539 if (system_state < SYSTEM_RUNNING)
540 udelay(1000 * EEH_PE_RST_SETTLE_TIME);
541 else
542 msleep(EEH_PE_RST_SETTLE_TIME);
543 }
Gavin Shan9d5cab02013-06-20 13:21:10 +0800544out:
545 if (rc != OPAL_SUCCESS)
546 return -EIO;
547
548 return 0;
549}
550
551static int ioda_eeh_root_reset(struct pci_controller *hose, int option)
552{
553 struct pnv_phb *phb = hose->private_data;
554 s64 rc = OPAL_SUCCESS;
555
556 pr_debug("%s: Reset PHB#%x, option=%d\n",
557 __func__, hose->global_number, option);
558
559 /*
560 * During the reset deassert time, we needn't care
561 * the reset scope because the firmware does nothing
562 * for fundamental or hot reset during deassert phase.
563 */
564 if (option == EEH_RESET_FUNDAMENTAL)
565 rc = opal_pci_reset(phb->opal_id,
Gavin Shand1a85ee2014-09-30 12:39:05 +1000566 OPAL_RESET_PCI_FUNDAMENTAL,
Gavin Shan9d5cab02013-06-20 13:21:10 +0800567 OPAL_ASSERT_RESET);
568 else if (option == EEH_RESET_HOT)
569 rc = opal_pci_reset(phb->opal_id,
Gavin Shand1a85ee2014-09-30 12:39:05 +1000570 OPAL_RESET_PCI_HOT,
Gavin Shan9d5cab02013-06-20 13:21:10 +0800571 OPAL_ASSERT_RESET);
572 else if (option == EEH_RESET_DEACTIVATE)
573 rc = opal_pci_reset(phb->opal_id,
Gavin Shand1a85ee2014-09-30 12:39:05 +1000574 OPAL_RESET_PCI_HOT,
Gavin Shan9d5cab02013-06-20 13:21:10 +0800575 OPAL_DEASSERT_RESET);
576 if (rc < 0)
577 goto out;
578
579 /* Poll state of the PHB until the request is done */
580 rc = ioda_eeh_phb_poll(phb);
Gavin Shan26833a52014-04-24 18:00:23 +1000581 if (option == EEH_RESET_DEACTIVATE)
582 msleep(EEH_PE_RST_SETTLE_TIME);
Gavin Shan9d5cab02013-06-20 13:21:10 +0800583out:
584 if (rc != OPAL_SUCCESS)
585 return -EIO;
586
587 return 0;
588}
589
Gavin Shan1d9a5442014-04-24 18:00:13 +1000590static int ioda_eeh_bridge_reset(struct pci_dev *dev, int option)
Gavin Shan9d5cab02013-06-20 13:21:10 +0800591
Gavin Shan1d9a5442014-04-24 18:00:13 +1000592{
593 struct device_node *dn = pci_device_to_OF_node(dev);
Gavin Shand92a2082014-04-24 18:00:24 +1000594 struct eeh_dev *edev = of_node_to_eeh_dev(dn);
595 int aer = edev ? edev->aer_cap : 0;
Benjamin Herrenschmidt965b5602014-05-20 10:20:49 +1000596 u32 ctrl;
Gavin Shan1d9a5442014-04-24 18:00:13 +1000597
598 pr_debug("%s: Reset PCI bus %04x:%02x with option %d\n",
599 __func__, pci_domain_nr(dev->bus),
600 dev->bus->number, option);
Gavin Shan9d5cab02013-06-20 13:21:10 +0800601
602 switch (option) {
603 case EEH_RESET_FUNDAMENTAL:
604 case EEH_RESET_HOT:
Benjamin Herrenschmidt965b5602014-05-20 10:20:49 +1000605 /* Don't report linkDown event */
Gavin Shand92a2082014-04-24 18:00:24 +1000606 if (aer) {
Gavin Shanb2b5efc2014-04-24 18:00:27 +1000607 eeh_ops->read_config(dn, aer + PCI_ERR_UNCOR_MASK,
Benjamin Herrenschmidt965b5602014-05-20 10:20:49 +1000608 4, &ctrl);
609 ctrl |= PCI_ERR_UNC_SURPDN;
Gavin Shanb2b5efc2014-04-24 18:00:27 +1000610 eeh_ops->write_config(dn, aer + PCI_ERR_UNCOR_MASK,
Benjamin Herrenschmidt965b5602014-05-20 10:20:49 +1000611 4, ctrl);
612 }
Gavin Shanb2b5efc2014-04-24 18:00:27 +1000613
Benjamin Herrenschmidt965b5602014-05-20 10:20:49 +1000614 eeh_ops->read_config(dn, PCI_BRIDGE_CONTROL, 2, &ctrl);
615 ctrl |= PCI_BRIDGE_CTL_BUS_RESET;
616 eeh_ops->write_config(dn, PCI_BRIDGE_CONTROL, 2, ctrl);
Gavin Shan26833a52014-04-24 18:00:23 +1000617 msleep(EEH_PE_RST_HOLD_TIME);
Gavin Shand92a2082014-04-24 18:00:24 +1000618
Gavin Shan9d5cab02013-06-20 13:21:10 +0800619 break;
620 case EEH_RESET_DEACTIVATE:
Benjamin Herrenschmidt965b5602014-05-20 10:20:49 +1000621 eeh_ops->read_config(dn, PCI_BRIDGE_CONTROL, 2, &ctrl);
622 ctrl &= ~PCI_BRIDGE_CTL_BUS_RESET;
623 eeh_ops->write_config(dn, PCI_BRIDGE_CONTROL, 2, ctrl);
Gavin Shan26833a52014-04-24 18:00:23 +1000624 msleep(EEH_PE_RST_SETTLE_TIME);
Gavin Shand92a2082014-04-24 18:00:24 +1000625
Benjamin Herrenschmidt965b5602014-05-20 10:20:49 +1000626 /* Continue reporting linkDown event */
Gavin Shand92a2082014-04-24 18:00:24 +1000627 if (aer) {
628 eeh_ops->read_config(dn, aer + PCI_ERR_UNCOR_MASK,
Benjamin Herrenschmidt965b5602014-05-20 10:20:49 +1000629 4, &ctrl);
630 ctrl &= ~PCI_ERR_UNC_SURPDN;
Gavin Shand92a2082014-04-24 18:00:24 +1000631 eeh_ops->write_config(dn, aer + PCI_ERR_UNCOR_MASK,
Benjamin Herrenschmidt965b5602014-05-20 10:20:49 +1000632 4, ctrl);
Gavin Shand92a2082014-04-24 18:00:24 +1000633 }
634
Gavin Shan9d5cab02013-06-20 13:21:10 +0800635 break;
636 }
637
638 return 0;
639}
640
Gavin Shand92a2082014-04-24 18:00:24 +1000641void pnv_pci_reset_secondary_bus(struct pci_dev *dev)
642{
643 struct pci_controller *hose;
644
645 if (pci_is_root_bus(dev->bus)) {
646 hose = pci_bus_to_host(dev->bus);
647 ioda_eeh_root_reset(hose, EEH_RESET_HOT);
648 ioda_eeh_root_reset(hose, EEH_RESET_DEACTIVATE);
649 } else {
650 ioda_eeh_bridge_reset(dev, EEH_RESET_HOT);
651 ioda_eeh_bridge_reset(dev, EEH_RESET_DEACTIVATE);
652 }
653}
654
Gavin Shan9d5cab02013-06-20 13:21:10 +0800655/**
656 * ioda_eeh_reset - Reset the indicated PE
657 * @pe: EEH PE
658 * @option: reset option
659 *
660 * Do reset on the indicated PE. For PCI bus sensitive PE,
661 * we need to reset the parent p2p bridge. The PHB has to
662 * be reinitialized if the p2p bridge is root bridge. For
663 * PCI device sensitive PE, we will try to reset the device
664 * through FLR. For now, we don't have OPAL APIs to do HARD
665 * reset yet, so all reset would be SOFT (HOT) reset.
666 */
667static int ioda_eeh_reset(struct eeh_pe *pe, int option)
668{
669 struct pci_controller *hose = pe->phb;
Gavin Shan5b2e1982014-02-12 15:24:54 +0800670 struct pci_bus *bus;
Gavin Shan9d5cab02013-06-20 13:21:10 +0800671 int ret;
672
673 /*
Gavin Shanfd5cee72014-04-24 18:00:22 +1000674 * For PHB reset, we always have complete reset. For those PEs whose
675 * primary bus derived from root complex (root bus) or root port
676 * (usually bus#1), we apply hot or fundamental reset on the root port.
677 * For other PEs, we always have hot reset on the PE primary bus.
Gavin Shan78954702014-04-24 18:00:14 +1000678 *
679 * Here, we have different design to pHyp, which always clear the
680 * frozen state during PE reset. However, the good idea here from
681 * benh is to keep frozen state before we get PE reset done completely
682 * (until BAR restore). With the frozen state, HW drops illegal IO
683 * or MMIO access, which can incur recrusive frozen PE during PE
684 * reset. The side effect is that EEH core has to clear the frozen
685 * state explicitly after BAR restore.
Gavin Shan9d5cab02013-06-20 13:21:10 +0800686 */
687 if (pe->type & EEH_PE_PHB) {
688 ret = ioda_eeh_phb_reset(hose, option);
689 } else {
Gavin Shand9df1b52014-09-30 12:38:58 +1000690 struct pnv_phb *phb;
691 s64 rc;
692
693 /*
694 * The frozen PE might be caused by PAPR error injection
695 * registers, which are expected to be cleared after hitting
696 * frozen PE as stated in the hardware spec. Unfortunately,
697 * that's not true on P7IOC. So we have to clear it manually
698 * to avoid recursive EEH errors during recovery.
699 */
700 phb = hose->private_data;
701 if (phb->model == PNV_PHB_MODEL_P7IOC &&
702 (option == EEH_RESET_HOT ||
703 option == EEH_RESET_FUNDAMENTAL)) {
704 rc = opal_pci_reset(phb->opal_id,
Gavin Shand1a85ee2014-09-30 12:39:05 +1000705 OPAL_RESET_PHB_ERROR,
Gavin Shand9df1b52014-09-30 12:38:58 +1000706 OPAL_ASSERT_RESET);
707 if (rc != OPAL_SUCCESS) {
708 pr_warn("%s: Failure %lld clearing "
709 "error injection registers\n",
710 __func__, rc);
711 return -EIO;
712 }
713 }
714
Gavin Shan5b2e1982014-02-12 15:24:54 +0800715 bus = eeh_pe_bus_get(pe);
Gavin Shanfd5cee72014-04-24 18:00:22 +1000716 if (pci_is_root_bus(bus) ||
717 pci_is_root_bus(bus->parent))
Gavin Shan9d5cab02013-06-20 13:21:10 +0800718 ret = ioda_eeh_root_reset(hose, option);
719 else
Gavin Shan1d9a5442014-04-24 18:00:13 +1000720 ret = ioda_eeh_bridge_reset(bus->self, option);
Gavin Shan9d5cab02013-06-20 13:21:10 +0800721 }
722
723 return ret;
724}
725
Gavin Shanbf90dfe2013-06-20 13:21:11 +0800726/**
Gavin Shanbb593c02014-07-17 14:41:43 +1000727 * ioda_eeh_get_log - Retrieve error log
728 * @pe: frozen PE
729 * @severity: permanent or temporary error
730 * @drv_log: device driver log
731 * @len: length of device driver log
732 *
733 * Retrieve error log, which contains log from device driver
734 * and firmware.
735 */
Anton Blancharde51df2c2014-08-20 08:55:18 +1000736static int ioda_eeh_get_log(struct eeh_pe *pe, int severity,
737 char *drv_log, unsigned long len)
Gavin Shanbb593c02014-07-17 14:41:43 +1000738{
Gavin Shana450e8f2014-11-22 21:58:09 +1100739 if (!eeh_has_flag(EEH_EARLY_DUMP_LOG))
740 pnv_pci_dump_phb_diag_data(pe->phb, pe->data);
Gavin Shanbb593c02014-07-17 14:41:43 +1000741
742 return 0;
743}
744
745/**
Gavin Shanbf90dfe2013-06-20 13:21:11 +0800746 * ioda_eeh_configure_bridge - Configure the PCI bridges for the indicated PE
747 * @pe: EEH PE
748 *
749 * For particular PE, it might have included PCI bridges. In order
750 * to make the PE work properly, those PCI bridges should be configured
751 * correctly. However, we need do nothing on P7IOC since the reset
752 * function will do everything that should be covered by the function.
753 */
754static int ioda_eeh_configure_bridge(struct eeh_pe *pe)
755{
756 return 0;
757}
758
Gavin Shan131c1232014-09-30 12:38:56 +1000759static int ioda_eeh_err_inject(struct eeh_pe *pe, int type, int func,
760 unsigned long addr, unsigned long mask)
761{
762 struct pci_controller *hose = pe->phb;
763 struct pnv_phb *phb = hose->private_data;
764 s64 ret;
765
766 /* Sanity check on error type */
767 if (type != OPAL_ERR_INJECT_TYPE_IOA_BUS_ERR &&
768 type != OPAL_ERR_INJECT_TYPE_IOA_BUS_ERR64) {
769 pr_warn("%s: Invalid error type %d\n",
770 __func__, type);
771 return -ERANGE;
772 }
773
774 if (func < OPAL_ERR_INJECT_FUNC_IOA_LD_MEM_ADDR ||
775 func > OPAL_ERR_INJECT_FUNC_IOA_DMA_WR_TARGET) {
776 pr_warn("%s: Invalid error function %d\n",
777 __func__, func);
778 return -ERANGE;
779 }
780
781 /* Firmware supports error injection ? */
782 if (!opal_check_token(OPAL_PCI_ERR_INJECT)) {
783 pr_warn("%s: Firmware doesn't support error injection\n",
784 __func__);
785 return -ENXIO;
786 }
787
788 /* Do error injection */
789 ret = opal_pci_err_inject(phb->opal_id, pe->addr,
790 type, func, addr, mask);
791 if (ret != OPAL_SUCCESS) {
792 pr_warn("%s: Failure %lld injecting error "
793 "%d-%d to PHB#%x-PE#%x\n",
794 __func__, ret, type, func,
795 hose->global_number, pe->addr);
796 return -EIO;
797 }
798
799 return 0;
800}
801
Gavin Shan70f942d2013-06-20 13:21:12 +0800802static void ioda_eeh_hub_diag_common(struct OpalIoP7IOCErrorData *data)
803{
804 /* GEM */
Gavin Shanf18440f2014-07-17 14:41:42 +1000805 if (data->gemXfir || data->gemRfir ||
806 data->gemRirqfir || data->gemMask || data->gemRwof)
807 pr_info(" GEM: %016llx %016llx %016llx %016llx %016llx\n",
808 be64_to_cpu(data->gemXfir),
809 be64_to_cpu(data->gemRfir),
810 be64_to_cpu(data->gemRirqfir),
811 be64_to_cpu(data->gemMask),
812 be64_to_cpu(data->gemRwof));
Gavin Shan70f942d2013-06-20 13:21:12 +0800813
814 /* LEM */
Gavin Shanf18440f2014-07-17 14:41:42 +1000815 if (data->lemFir || data->lemErrMask ||
816 data->lemAction0 || data->lemAction1 || data->lemWof)
817 pr_info(" LEM: %016llx %016llx %016llx %016llx %016llx\n",
818 be64_to_cpu(data->lemFir),
819 be64_to_cpu(data->lemErrMask),
820 be64_to_cpu(data->lemAction0),
821 be64_to_cpu(data->lemAction1),
822 be64_to_cpu(data->lemWof));
Gavin Shan70f942d2013-06-20 13:21:12 +0800823}
824
825static void ioda_eeh_hub_diag(struct pci_controller *hose)
826{
827 struct pnv_phb *phb = hose->private_data;
Brian W Hartca1de5d2013-12-20 13:06:01 -0600828 struct OpalIoP7IOCErrorData *data = &phb->diag.hub_diag;
Gavin Shan70f942d2013-06-20 13:21:12 +0800829 long rc;
830
Brian W Hartca1de5d2013-12-20 13:06:01 -0600831 rc = opal_pci_get_hub_diag_data(phb->hub_id, data, sizeof(*data));
Gavin Shan70f942d2013-06-20 13:21:12 +0800832 if (rc != OPAL_SUCCESS) {
Gavin Shan0dae2742014-07-17 14:41:41 +1000833 pr_warn("%s: Failed to get HUB#%llx diag-data (%ld)\n",
834 __func__, phb->hub_id, rc);
Gavin Shan70f942d2013-06-20 13:21:12 +0800835 return;
836 }
837
838 switch (data->type) {
839 case OPAL_P7IOC_DIAG_TYPE_RGC:
840 pr_info("P7IOC diag-data for RGC\n\n");
841 ioda_eeh_hub_diag_common(data);
Gavin Shanf18440f2014-07-17 14:41:42 +1000842 if (data->rgc.rgcStatus || data->rgc.rgcLdcp)
843 pr_info(" RGC: %016llx %016llx\n",
844 be64_to_cpu(data->rgc.rgcStatus),
845 be64_to_cpu(data->rgc.rgcLdcp));
Gavin Shan70f942d2013-06-20 13:21:12 +0800846 break;
847 case OPAL_P7IOC_DIAG_TYPE_BI:
848 pr_info("P7IOC diag-data for BI %s\n\n",
849 data->bi.biDownbound ? "Downbound" : "Upbound");
850 ioda_eeh_hub_diag_common(data);
Gavin Shanf18440f2014-07-17 14:41:42 +1000851 if (data->bi.biLdcp0 || data->bi.biLdcp1 ||
852 data->bi.biLdcp2 || data->bi.biFenceStatus)
853 pr_info(" BI: %016llx %016llx %016llx %016llx\n",
854 be64_to_cpu(data->bi.biLdcp0),
855 be64_to_cpu(data->bi.biLdcp1),
856 be64_to_cpu(data->bi.biLdcp2),
857 be64_to_cpu(data->bi.biFenceStatus));
Gavin Shan70f942d2013-06-20 13:21:12 +0800858 break;
859 case OPAL_P7IOC_DIAG_TYPE_CI:
Gavin Shanf18440f2014-07-17 14:41:42 +1000860 pr_info("P7IOC diag-data for CI Port %d\n\n",
Gavin Shan70f942d2013-06-20 13:21:12 +0800861 data->ci.ciPort);
862 ioda_eeh_hub_diag_common(data);
Gavin Shanf18440f2014-07-17 14:41:42 +1000863 if (data->ci.ciPortStatus || data->ci.ciPortLdcp)
864 pr_info(" CI: %016llx %016llx\n",
865 be64_to_cpu(data->ci.ciPortStatus),
866 be64_to_cpu(data->ci.ciPortLdcp));
Gavin Shan70f942d2013-06-20 13:21:12 +0800867 break;
868 case OPAL_P7IOC_DIAG_TYPE_MISC:
869 pr_info("P7IOC diag-data for MISC\n\n");
870 ioda_eeh_hub_diag_common(data);
871 break;
872 case OPAL_P7IOC_DIAG_TYPE_I2C:
873 pr_info("P7IOC diag-data for I2C\n\n");
874 ioda_eeh_hub_diag_common(data);
875 break;
876 default:
Gavin Shan0dae2742014-07-17 14:41:41 +1000877 pr_warn("%s: Invalid type of HUB#%llx diag-data (%d)\n",
878 __func__, phb->hub_id, data->type);
Gavin Shan70f942d2013-06-20 13:21:12 +0800879 }
880}
881
Gavin Shan70f942d2013-06-20 13:21:12 +0800882static int ioda_eeh_get_pe(struct pci_controller *hose,
883 u16 pe_no, struct eeh_pe **pe)
884{
Gavin Shan58287902014-07-21 14:42:34 +1000885 struct pnv_phb *phb = hose->private_data;
886 struct pnv_ioda_pe *pnv_pe;
887 struct eeh_pe *dev_pe;
888 struct eeh_dev edev;
Gavin Shan70f942d2013-06-20 13:21:12 +0800889
Gavin Shan58287902014-07-21 14:42:34 +1000890 /*
891 * If PHB supports compound PE, to fetch
892 * the master PE because slave PE is invisible
893 * to EEH core.
894 */
Gavin Shan372fb802014-09-30 12:39:09 +1000895 pnv_pe = &phb->ioda.pe_array[pe_no];
896 if (pnv_pe->flags & PNV_IODA_PE_SLAVE) {
897 pnv_pe = pnv_pe->master;
898 WARN_ON(!pnv_pe ||
899 !(pnv_pe->flags & PNV_IODA_PE_MASTER));
900 pe_no = pnv_pe->pe_number;
Gavin Shan58287902014-07-21 14:42:34 +1000901 }
Gavin Shan70f942d2013-06-20 13:21:12 +0800902
903 /* Find the PE according to PE# */
Gavin Shan58287902014-07-21 14:42:34 +1000904 memset(&edev, 0, sizeof(struct eeh_dev));
905 edev.phb = hose;
906 edev.pe_config_addr = pe_no;
907 dev_pe = eeh_pe_get(&edev);
908 if (!dev_pe)
909 return -EEXIST;
Gavin Shan70f942d2013-06-20 13:21:12 +0800910
Gavin Shan372fb802014-09-30 12:39:09 +1000911 /* Freeze the (compound) PE */
Gavin Shan70f942d2013-06-20 13:21:12 +0800912 *pe = dev_pe;
Gavin Shan372fb802014-09-30 12:39:09 +1000913 if (!(dev_pe->state & EEH_PE_ISOLATED))
Gavin Shan58287902014-07-21 14:42:34 +1000914 phb->freeze_pe(phb, pe_no);
915
Gavin Shan372fb802014-09-30 12:39:09 +1000916 /*
917 * At this point, we're sure the (compound) PE should
918 * have been frozen. However, we still need poke until
919 * hitting the frozen PE on top level.
920 */
921 dev_pe = dev_pe->parent;
922 while (dev_pe && !(dev_pe->type & EEH_PE_PHB)) {
923 int ret;
924 int active_flags = (EEH_STATE_MMIO_ACTIVE |
925 EEH_STATE_DMA_ACTIVE);
926
927 ret = eeh_ops->get_state(dev_pe, NULL);
928 if (ret <= 0 || (ret & active_flags) == active_flags) {
929 dev_pe = dev_pe->parent;
930 continue;
931 }
932
933 /* Frozen parent PE */
934 *pe = dev_pe;
935 if (!(dev_pe->state & EEH_PE_ISOLATED))
936 phb->freeze_pe(phb, dev_pe->addr);
937
938 /* Next one */
939 dev_pe = dev_pe->parent;
940 }
941
Gavin Shan70f942d2013-06-20 13:21:12 +0800942 return 0;
943}
944
945/**
946 * ioda_eeh_next_error - Retrieve next error for EEH core to handle
947 * @pe: The affected PE
948 *
949 * The function is expected to be called by EEH core while it gets
950 * special EEH event (without binding PE). The function calls to
951 * OPAL APIs for next error to handle. The informational error is
952 * handled internally by platform. However, the dead IOC, dead PHB,
953 * fenced PHB and frozen PE should be handled by EEH core eventually.
954 */
955static int ioda_eeh_next_error(struct eeh_pe **pe)
956{
Gavin Shan7e4e7862014-01-15 13:16:11 +0800957 struct pci_controller *hose;
Gavin Shan70f942d2013-06-20 13:21:12 +0800958 struct pnv_phb *phb;
Gavin Shan1ad7a722014-05-05 09:29:03 +1000959 struct eeh_pe *phb_pe, *parent_pe;
Guo Chaoddf03222014-06-09 16:58:51 +0800960 __be64 frozen_pe_no;
961 __be16 err_type, severity;
Gavin Shan1ad7a722014-05-05 09:29:03 +1000962 int active_flags = (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE);
Gavin Shan70f942d2013-06-20 13:21:12 +0800963 long rc;
Gavin Shan1ad7a722014-05-05 09:29:03 +1000964 int state, ret = EEH_NEXT_ERR_NONE;
Gavin Shan70f942d2013-06-20 13:21:12 +0800965
Gavin Shan7cb9d932013-06-20 18:13:24 +0800966 /*
967 * While running here, it's safe to purge the event queue.
968 * And we should keep the cached OPAL notifier event sychronized
969 * between the kernel and firmware.
970 */
Gavin Shan5c7a35e2014-06-04 17:31:52 +1000971 eeh_remove_event(NULL, false);
Gavin Shan7cb9d932013-06-20 18:13:24 +0800972 opal_notifier_update_evt(OPAL_EVENT_PCI_ERROR, 0x0ul);
Gavin Shan70f942d2013-06-20 13:21:12 +0800973
Gavin Shan7e4e7862014-01-15 13:16:11 +0800974 list_for_each_entry(hose, &hose_list, list_node) {
Gavin Shan70f942d2013-06-20 13:21:12 +0800975 /*
976 * If the subordinate PCI buses of the PHB has been
Gavin Shan467f79a2014-04-24 18:00:08 +1000977 * removed or is exactly under error recovery, we
978 * needn't take care of it any more.
Gavin Shan70f942d2013-06-20 13:21:12 +0800979 */
980 phb = hose->private_data;
Gavin Shan467f79a2014-04-24 18:00:08 +1000981 phb_pe = eeh_phb_pe_get(hose);
982 if (!phb_pe || (phb_pe->state & EEH_PE_ISOLATED))
Gavin Shan70f942d2013-06-20 13:21:12 +0800983 continue;
984
985 rc = opal_pci_next_error(phb->opal_id,
986 &frozen_pe_no, &err_type, &severity);
987
988 /* If OPAL API returns error, we needn't proceed */
989 if (rc != OPAL_SUCCESS) {
Mike Qiu20212702013-08-12 02:15:36 -0400990 pr_devel("%s: Invalid return value on "
991 "PHB#%x (0x%lx) from opal_pci_next_error",
992 __func__, hose->global_number, rc);
Gavin Shan70f942d2013-06-20 13:21:12 +0800993 continue;
994 }
995
996 /* If the PHB doesn't have error, stop processing */
Guo Chaoddf03222014-06-09 16:58:51 +0800997 if (be16_to_cpu(err_type) == OPAL_EEH_NO_ERROR ||
998 be16_to_cpu(severity) == OPAL_EEH_SEV_NO_ERROR) {
Mike Qiu20212702013-08-12 02:15:36 -0400999 pr_devel("%s: No error found on PHB#%x\n",
1000 __func__, hose->global_number);
Gavin Shan70f942d2013-06-20 13:21:12 +08001001 continue;
1002 }
1003
1004 /*
1005 * Processing the error. We're expecting the error with
1006 * highest priority reported upon multiple errors on the
1007 * specific PHB.
1008 */
Mike Qiu20212702013-08-12 02:15:36 -04001009 pr_devel("%s: Error (%d, %d, %llu) on PHB#%x\n",
Guo Chaoddf03222014-06-09 16:58:51 +08001010 __func__, be16_to_cpu(err_type), be16_to_cpu(severity),
1011 be64_to_cpu(frozen_pe_no), hose->global_number);
1012 switch (be16_to_cpu(err_type)) {
Gavin Shan70f942d2013-06-20 13:21:12 +08001013 case OPAL_EEH_IOC_ERROR:
Guo Chaoddf03222014-06-09 16:58:51 +08001014 if (be16_to_cpu(severity) == OPAL_EEH_SEV_IOC_DEAD) {
Gavin Shan56ca4fd2013-06-27 13:46:46 +08001015 pr_err("EEH: dead IOC detected\n");
Gavin Shan7e4e7862014-01-15 13:16:11 +08001016 ret = EEH_NEXT_ERR_DEAD_IOC;
Guo Chaoddf03222014-06-09 16:58:51 +08001017 } else if (be16_to_cpu(severity) == OPAL_EEH_SEV_INF) {
Gavin Shan56ca4fd2013-06-27 13:46:46 +08001018 pr_info("EEH: IOC informative error "
1019 "detected\n");
Gavin Shan70f942d2013-06-20 13:21:12 +08001020 ioda_eeh_hub_diag(hose);
Gavin Shan7e4e7862014-01-15 13:16:11 +08001021 ret = EEH_NEXT_ERR_NONE;
Gavin Shan56ca4fd2013-06-27 13:46:46 +08001022 }
Gavin Shan70f942d2013-06-20 13:21:12 +08001023
1024 break;
1025 case OPAL_EEH_PHB_ERROR:
Guo Chaoddf03222014-06-09 16:58:51 +08001026 if (be16_to_cpu(severity) == OPAL_EEH_SEV_PHB_DEAD) {
Gavin Shan467f79a2014-04-24 18:00:08 +10001027 *pe = phb_pe;
Gavin Shan357b2f32014-06-11 18:26:44 +10001028 pr_err("EEH: dead PHB#%x detected, "
1029 "location: %s\n",
1030 hose->global_number,
1031 eeh_pe_loc_get(phb_pe));
Gavin Shan7e4e7862014-01-15 13:16:11 +08001032 ret = EEH_NEXT_ERR_DEAD_PHB;
Guo Chaoddf03222014-06-09 16:58:51 +08001033 } else if (be16_to_cpu(severity) ==
1034 OPAL_EEH_SEV_PHB_FENCED) {
Gavin Shan467f79a2014-04-24 18:00:08 +10001035 *pe = phb_pe;
Gavin Shan357b2f32014-06-11 18:26:44 +10001036 pr_err("EEH: Fenced PHB#%x detected, "
1037 "location: %s\n",
1038 hose->global_number,
1039 eeh_pe_loc_get(phb_pe));
Gavin Shan7e4e7862014-01-15 13:16:11 +08001040 ret = EEH_NEXT_ERR_FENCED_PHB;
Guo Chaoddf03222014-06-09 16:58:51 +08001041 } else if (be16_to_cpu(severity) == OPAL_EEH_SEV_INF) {
Gavin Shan56ca4fd2013-06-27 13:46:46 +08001042 pr_info("EEH: PHB#%x informative error "
Gavin Shan357b2f32014-06-11 18:26:44 +10001043 "detected, location: %s\n",
1044 hose->global_number,
1045 eeh_pe_loc_get(phb_pe));
Gavin Shanbb593c02014-07-17 14:41:43 +10001046 ioda_eeh_phb_diag(phb_pe);
1047 pnv_pci_dump_phb_diag_data(hose, phb_pe->data);
Gavin Shan7e4e7862014-01-15 13:16:11 +08001048 ret = EEH_NEXT_ERR_NONE;
Gavin Shan56ca4fd2013-06-27 13:46:46 +08001049 }
Gavin Shan70f942d2013-06-20 13:21:12 +08001050
1051 break;
1052 case OPAL_EEH_PE_ERROR:
Gavin Shancb5b2422014-01-15 13:16:13 +08001053 /*
Gavin Shan71b540a2014-05-05 09:29:04 +10001054 * If we can't find the corresponding PE, we
1055 * just try to unfreeze.
Gavin Shancb5b2422014-01-15 13:16:13 +08001056 */
Guo Chaoddf03222014-06-09 16:58:51 +08001057 if (ioda_eeh_get_pe(hose,
Gavin Shan71b540a2014-05-05 09:29:04 +10001058 be64_to_cpu(frozen_pe_no), pe)) {
1059 /* Try best to clear it */
1060 pr_info("EEH: Clear non-existing PHB#%x-PE#%llx\n",
1061 hose->global_number, frozen_pe_no);
Gavin Shan357b2f32014-06-11 18:26:44 +10001062 pr_info("EEH: PHB location: %s\n",
1063 eeh_pe_loc_get(phb_pe));
Gavin Shan71b540a2014-05-05 09:29:04 +10001064 opal_pci_eeh_freeze_clear(phb->opal_id, frozen_pe_no,
1065 OPAL_EEH_ACTION_CLEAR_FREEZE_ALL);
1066 ret = EEH_NEXT_ERR_NONE;
Gavin Shan05ec4242014-06-10 11:41:55 +10001067 } else if ((*pe)->state & EEH_PE_ISOLATED ||
1068 eeh_pe_passed(*pe)) {
Gavin Shan63796552014-04-24 18:00:20 +10001069 ret = EEH_NEXT_ERR_NONE;
Gavin Shancb5b2422014-01-15 13:16:13 +08001070 } else {
1071 pr_err("EEH: Frozen PE#%x on PHB#%x detected\n",
1072 (*pe)->addr, (*pe)->phb->global_number);
Gavin Shan357b2f32014-06-11 18:26:44 +10001073 pr_err("EEH: PE location: %s, PHB location: %s\n",
1074 eeh_pe_loc_get(*pe), eeh_pe_loc_get(phb_pe));
Gavin Shancb5b2422014-01-15 13:16:13 +08001075 ret = EEH_NEXT_ERR_FROZEN_PE;
1076 }
Gavin Shan70f942d2013-06-20 13:21:12 +08001077
Gavin Shan7e4e7862014-01-15 13:16:11 +08001078 break;
1079 default:
1080 pr_warn("%s: Unexpected error type %d\n",
Guo Chaoddf03222014-06-09 16:58:51 +08001081 __func__, be16_to_cpu(err_type));
Gavin Shan70f942d2013-06-20 13:21:12 +08001082 }
Gavin Shan7e4e7862014-01-15 13:16:11 +08001083
1084 /*
Gavin Shan94716602014-02-25 15:28:37 +08001085 * EEH core will try recover from fenced PHB or
1086 * frozen PE. In the time for frozen PE, EEH core
1087 * enable IO path for that before collecting logs,
1088 * but it ruins the site. So we have to dump the
1089 * log in advance here.
1090 */
1091 if ((ret == EEH_NEXT_ERR_FROZEN_PE ||
1092 ret == EEH_NEXT_ERR_FENCED_PHB) &&
1093 !((*pe)->state & EEH_PE_ISOLATED)) {
1094 eeh_pe_state_mark(*pe, EEH_PE_ISOLATED);
Gavin Shanbb593c02014-07-17 14:41:43 +10001095 ioda_eeh_phb_diag(*pe);
Gavin Shana450e8f2014-11-22 21:58:09 +11001096
1097 if (eeh_has_flag(EEH_EARLY_DUMP_LOG))
1098 pnv_pci_dump_phb_diag_data((*pe)->phb,
1099 (*pe)->data);
Gavin Shan94716602014-02-25 15:28:37 +08001100 }
1101
1102 /*
Gavin Shan1ad7a722014-05-05 09:29:03 +10001103 * We probably have the frozen parent PE out there and
1104 * we need have to handle frozen parent PE firstly.
1105 */
1106 if (ret == EEH_NEXT_ERR_FROZEN_PE) {
1107 parent_pe = (*pe)->parent;
1108 while (parent_pe) {
1109 /* Hit the ceiling ? */
1110 if (parent_pe->type & EEH_PE_PHB)
1111 break;
1112
1113 /* Frozen parent PE ? */
1114 state = ioda_eeh_get_state(parent_pe);
1115 if (state > 0 &&
1116 (state & active_flags) != active_flags)
1117 *pe = parent_pe;
1118
1119 /* Next parent level */
1120 parent_pe = parent_pe->parent;
1121 }
1122
1123 /* We possibly migrate to another PE */
1124 eeh_pe_state_mark(*pe, EEH_PE_ISOLATED);
1125 }
1126
1127 /*
Gavin Shan7e4e7862014-01-15 13:16:11 +08001128 * If we have no errors on the specific PHB or only
1129 * informative error there, we continue poking it.
1130 * Otherwise, we need actions to be taken by upper
1131 * layer.
1132 */
1133 if (ret > EEH_NEXT_ERR_INF)
1134 break;
Gavin Shan70f942d2013-06-20 13:21:12 +08001135 }
1136
Gavin Shan70f942d2013-06-20 13:21:12 +08001137 return ret;
1138}
1139
Gavin Shan8747f362013-06-20 13:21:06 +08001140struct pnv_eeh_ops ioda_eeh_ops = {
Gavin Shan73370c62013-06-20 13:21:07 +08001141 .post_init = ioda_eeh_post_init,
Gavin Shaneb005982013-06-20 13:21:08 +08001142 .set_option = ioda_eeh_set_option,
Gavin Shan8c41a7f2013-06-20 13:21:09 +08001143 .get_state = ioda_eeh_get_state,
Gavin Shan9d5cab02013-06-20 13:21:10 +08001144 .reset = ioda_eeh_reset,
Gavin Shanbb593c02014-07-17 14:41:43 +10001145 .get_log = ioda_eeh_get_log,
Gavin Shanbf90dfe2013-06-20 13:21:11 +08001146 .configure_bridge = ioda_eeh_configure_bridge,
Gavin Shan131c1232014-09-30 12:38:56 +10001147 .err_inject = ioda_eeh_err_inject,
Gavin Shan70f942d2013-06-20 13:21:12 +08001148 .next_error = ioda_eeh_next_error
Gavin Shan8747f362013-06-20 13:21:06 +08001149};