Commit | Line | Data |
---|---|---|
29310e5e GS |
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> | |
4cf17445 | 15 | #include <linux/debugfs.h> |
29310e5e GS |
16 | #include <linux/delay.h> |
17 | #include <linux/export.h> | |
18 | #include <linux/init.h> | |
19 | #include <linux/list.h> | |
20 | #include <linux/msi.h> | |
21 | #include <linux/of.h> | |
22 | #include <linux/pci.h> | |
23 | #include <linux/proc_fs.h> | |
24 | #include <linux/rbtree.h> | |
25 | #include <linux/sched.h> | |
26 | #include <linux/seq_file.h> | |
27 | #include <linux/spinlock.h> | |
28 | ||
29 | #include <asm/eeh.h> | |
30 | #include <asm/eeh_event.h> | |
31 | #include <asm/firmware.h> | |
32 | #include <asm/io.h> | |
33 | #include <asm/iommu.h> | |
34 | #include <asm/machdep.h> | |
35 | #include <asm/msi_bitmap.h> | |
36 | #include <asm/opal.h> | |
37 | #include <asm/ppc-pci.h> | |
38 | ||
39 | #include "powernv.h" | |
40 | #include "pci.h" | |
41 | ||
4cf17445 GS |
42 | static bool pnv_eeh_nb_init = false; |
43 | ||
29310e5e | 44 | /** |
01f3bfb7 | 45 | * pnv_eeh_init - EEH platform dependent initialization |
29310e5e GS |
46 | * |
47 | * EEH platform dependent initialization on powernv | |
48 | */ | |
01f3bfb7 | 49 | static int pnv_eeh_init(void) |
29310e5e | 50 | { |
dc561fb9 GS |
51 | struct pci_controller *hose; |
52 | struct pnv_phb *phb; | |
53 | ||
29310e5e GS |
54 | /* We require OPALv3 */ |
55 | if (!firmware_has_feature(FW_FEATURE_OPALv3)) { | |
0dae2743 GS |
56 | pr_warn("%s: OPALv3 is required !\n", |
57 | __func__); | |
29310e5e GS |
58 | return -EINVAL; |
59 | } | |
60 | ||
05b1721d GS |
61 | /* Set probe mode */ |
62 | eeh_add_flag(EEH_PROBE_MODE_DEV); | |
29310e5e | 63 | |
dc561fb9 GS |
64 | /* |
65 | * P7IOC blocks PCI config access to frozen PE, but PHB3 | |
66 | * doesn't do that. So we have to selectively enable I/O | |
67 | * prior to collecting error log. | |
68 | */ | |
69 | list_for_each_entry(hose, &hose_list, list_node) { | |
70 | phb = hose->private_data; | |
71 | ||
72 | if (phb->model == PNV_PHB_MODEL_P7IOC) | |
73 | eeh_add_flag(EEH_ENABLE_IO_FOR_LOG); | |
2aa5cf9e GS |
74 | |
75 | /* | |
76 | * PE#0 should be regarded as valid by EEH core | |
77 | * if it's not the reserved one. Currently, we | |
78 | * have the reserved PE#0 and PE#127 for PHB3 | |
79 | * and P7IOC separately. So we should regard | |
80 | * PE#0 as valid for P7IOC. | |
81 | */ | |
82 | if (phb->ioda.reserved_pe != 0) | |
83 | eeh_add_flag(EEH_VALID_PE_ZERO); | |
84 | ||
dc561fb9 GS |
85 | break; |
86 | } | |
87 | ||
29310e5e GS |
88 | return 0; |
89 | } | |
90 | ||
4cf17445 GS |
91 | static int pnv_eeh_event(struct notifier_block *nb, |
92 | unsigned long events, void *change) | |
93 | { | |
94 | uint64_t changed_evts = (uint64_t)change; | |
95 | ||
96 | /* | |
97 | * We simply send special EEH event if EEH has | |
98 | * been enabled, or clear pending events in | |
99 | * case that we enable EEH soon | |
100 | */ | |
101 | if (!(changed_evts & OPAL_EVENT_PCI_ERROR) || | |
102 | !(events & OPAL_EVENT_PCI_ERROR)) | |
103 | return 0; | |
104 | ||
105 | if (eeh_enabled()) | |
106 | eeh_send_failure_event(NULL); | |
107 | else | |
108 | opal_notifier_update_evt(OPAL_EVENT_PCI_ERROR, 0x0ul); | |
109 | ||
110 | return 0; | |
111 | } | |
112 | ||
113 | static struct notifier_block pnv_eeh_nb = { | |
114 | .notifier_call = pnv_eeh_event, | |
115 | .next = NULL, | |
116 | .priority = 0 | |
117 | }; | |
118 | ||
119 | #ifdef CONFIG_DEBUG_FS | |
120 | static ssize_t pnv_eeh_ei_write(struct file *filp, | |
121 | const char __user *user_buf, | |
122 | size_t count, loff_t *ppos) | |
123 | { | |
124 | struct pci_controller *hose = filp->private_data; | |
125 | struct eeh_dev *edev; | |
126 | struct eeh_pe *pe; | |
127 | int pe_no, type, func; | |
128 | unsigned long addr, mask; | |
129 | char buf[50]; | |
130 | int ret; | |
131 | ||
132 | if (!eeh_ops || !eeh_ops->err_inject) | |
133 | return -ENXIO; | |
134 | ||
135 | /* Copy over argument buffer */ | |
136 | ret = simple_write_to_buffer(buf, sizeof(buf), ppos, user_buf, count); | |
137 | if (!ret) | |
138 | return -EFAULT; | |
139 | ||
140 | /* Retrieve parameters */ | |
141 | ret = sscanf(buf, "%x:%x:%x:%lx:%lx", | |
142 | &pe_no, &type, &func, &addr, &mask); | |
143 | if (ret != 5) | |
144 | return -EINVAL; | |
145 | ||
146 | /* Retrieve PE */ | |
147 | edev = kzalloc(sizeof(*edev), GFP_KERNEL); | |
148 | if (!edev) | |
149 | return -ENOMEM; | |
150 | edev->phb = hose; | |
151 | edev->pe_config_addr = pe_no; | |
152 | pe = eeh_pe_get(edev); | |
153 | kfree(edev); | |
154 | if (!pe) | |
155 | return -ENODEV; | |
156 | ||
157 | /* Do error injection */ | |
158 | ret = eeh_ops->err_inject(pe, type, func, addr, mask); | |
159 | return ret < 0 ? ret : count; | |
160 | } | |
161 | ||
162 | static const struct file_operations pnv_eeh_ei_fops = { | |
163 | .open = simple_open, | |
164 | .llseek = no_llseek, | |
165 | .write = pnv_eeh_ei_write, | |
166 | }; | |
167 | ||
168 | static int pnv_eeh_dbgfs_set(void *data, int offset, u64 val) | |
169 | { | |
170 | struct pci_controller *hose = data; | |
171 | struct pnv_phb *phb = hose->private_data; | |
172 | ||
173 | out_be64(phb->regs + offset, val); | |
174 | return 0; | |
175 | } | |
176 | ||
177 | static int pnv_eeh_dbgfs_get(void *data, int offset, u64 *val) | |
178 | { | |
179 | struct pci_controller *hose = data; | |
180 | struct pnv_phb *phb = hose->private_data; | |
181 | ||
182 | *val = in_be64(phb->regs + offset); | |
183 | return 0; | |
184 | } | |
185 | ||
186 | static int pnv_eeh_outb_dbgfs_set(void *data, u64 val) | |
187 | { | |
188 | return pnv_eeh_dbgfs_set(data, 0xD10, val); | |
189 | } | |
190 | ||
191 | static int pnv_eeh_outb_dbgfs_get(void *data, u64 *val) | |
192 | { | |
193 | return pnv_eeh_dbgfs_get(data, 0xD10, val); | |
194 | } | |
195 | ||
196 | static int pnv_eeh_inbA_dbgfs_set(void *data, u64 val) | |
197 | { | |
198 | return pnv_eeh_dbgfs_set(data, 0xD90, val); | |
199 | } | |
200 | ||
201 | static int pnv_eeh_inbA_dbgfs_get(void *data, u64 *val) | |
202 | { | |
203 | return pnv_eeh_dbgfs_get(data, 0xD90, val); | |
204 | } | |
205 | ||
206 | static int pnv_eeh_inbB_dbgfs_set(void *data, u64 val) | |
207 | { | |
208 | return pnv_eeh_dbgfs_set(data, 0xE10, val); | |
209 | } | |
210 | ||
211 | static int pnv_eeh_inbB_dbgfs_get(void *data, u64 *val) | |
212 | { | |
213 | return pnv_eeh_dbgfs_get(data, 0xE10, val); | |
214 | } | |
215 | ||
216 | DEFINE_SIMPLE_ATTRIBUTE(pnv_eeh_outb_dbgfs_ops, pnv_eeh_outb_dbgfs_get, | |
217 | pnv_eeh_outb_dbgfs_set, "0x%llx\n"); | |
218 | DEFINE_SIMPLE_ATTRIBUTE(pnv_eeh_inbA_dbgfs_ops, pnv_eeh_inbA_dbgfs_get, | |
219 | pnv_eeh_inbA_dbgfs_set, "0x%llx\n"); | |
220 | DEFINE_SIMPLE_ATTRIBUTE(pnv_eeh_inbB_dbgfs_ops, pnv_eeh_inbB_dbgfs_get, | |
221 | pnv_eeh_inbB_dbgfs_set, "0x%llx\n"); | |
222 | #endif /* CONFIG_DEBUG_FS */ | |
223 | ||
29310e5e | 224 | /** |
01f3bfb7 | 225 | * pnv_eeh_post_init - EEH platform dependent post initialization |
29310e5e GS |
226 | * |
227 | * EEH platform dependent post initialization on powernv. When | |
228 | * the function is called, the EEH PEs and devices should have | |
229 | * been built. If the I/O cache staff has been built, EEH is | |
230 | * ready to supply service. | |
231 | */ | |
01f3bfb7 | 232 | static int pnv_eeh_post_init(void) |
29310e5e GS |
233 | { |
234 | struct pci_controller *hose; | |
235 | struct pnv_phb *phb; | |
236 | int ret = 0; | |
237 | ||
4cf17445 GS |
238 | /* Register OPAL event notifier */ |
239 | if (!pnv_eeh_nb_init) { | |
240 | ret = opal_notifier_register(&pnv_eeh_nb); | |
241 | if (ret) { | |
242 | pr_warn("%s: Can't register OPAL event notifier (%d)\n", | |
243 | __func__, ret); | |
244 | return ret; | |
245 | } | |
246 | ||
247 | pnv_eeh_nb_init = true; | |
248 | } | |
249 | ||
29310e5e GS |
250 | list_for_each_entry(hose, &hose_list, list_node) { |
251 | phb = hose->private_data; | |
252 | ||
4cf17445 GS |
253 | /* |
254 | * If EEH is enabled, we're going to rely on that. | |
255 | * Otherwise, we restore to conventional mechanism | |
256 | * to clear frozen PE during PCI config access. | |
257 | */ | |
258 | if (eeh_enabled()) | |
259 | phb->flags |= PNV_PHB_FLAG_EEH; | |
260 | else | |
261 | phb->flags &= ~PNV_PHB_FLAG_EEH; | |
262 | ||
263 | /* Create debugfs entries */ | |
264 | #ifdef CONFIG_DEBUG_FS | |
265 | if (phb->has_dbgfs || !phb->dbgfs) | |
266 | continue; | |
267 | ||
268 | phb->has_dbgfs = 1; | |
269 | debugfs_create_file("err_injct", 0200, | |
270 | phb->dbgfs, hose, | |
271 | &pnv_eeh_ei_fops); | |
272 | ||
273 | debugfs_create_file("err_injct_outbound", 0600, | |
274 | phb->dbgfs, hose, | |
275 | &pnv_eeh_outb_dbgfs_ops); | |
276 | debugfs_create_file("err_injct_inboundA", 0600, | |
277 | phb->dbgfs, hose, | |
278 | &pnv_eeh_inbA_dbgfs_ops); | |
279 | debugfs_create_file("err_injct_inboundB", 0600, | |
280 | phb->dbgfs, hose, | |
281 | &pnv_eeh_inbB_dbgfs_ops); | |
282 | #endif /* CONFIG_DEBUG_FS */ | |
29310e5e GS |
283 | } |
284 | ||
4cf17445 | 285 | |
29310e5e GS |
286 | return ret; |
287 | } | |
288 | ||
289 | /** | |
01f3bfb7 | 290 | * pnv_eeh_dev_probe - Do probe on PCI device |
29310e5e GS |
291 | * @dev: PCI device |
292 | * @flag: unused | |
293 | * | |
294 | * When EEH module is installed during system boot, all PCI devices | |
295 | * are checked one by one to see if it supports EEH. The function | |
296 | * is introduced for the purpose. By default, EEH has been enabled | |
297 | * on all PCI devices. That's to say, we only need do necessary | |
298 | * initialization on the corresponding eeh device and create PE | |
299 | * accordingly. | |
300 | * | |
301 | * It's notable that's unsafe to retrieve the EEH device through | |
302 | * the corresponding PCI device. During the PCI device hotplug, which | |
303 | * was possiblly triggered by EEH core, the binding between EEH device | |
304 | * and the PCI device isn't built yet. | |
305 | */ | |
01f3bfb7 | 306 | static int pnv_eeh_dev_probe(struct pci_dev *dev, void *flag) |
29310e5e GS |
307 | { |
308 | struct pci_controller *hose = pci_bus_to_host(dev->bus); | |
309 | struct pnv_phb *phb = hose->private_data; | |
310 | struct device_node *dn = pci_device_to_OF_node(dev); | |
311 | struct eeh_dev *edev = of_node_to_eeh_dev(dn); | |
dadcd6d6 | 312 | int ret; |
29310e5e GS |
313 | |
314 | /* | |
315 | * When probing the root bridge, which doesn't have any | |
316 | * subordinate PCI devices. We don't have OF node for | |
317 | * the root bridge. So it's not reasonable to continue | |
318 | * the probing. | |
319 | */ | |
f5c57710 | 320 | if (!dn || !edev || edev->pe) |
29310e5e GS |
321 | return 0; |
322 | ||
323 | /* Skip for PCI-ISA bridge */ | |
324 | if ((dev->class >> 8) == PCI_CLASS_BRIDGE_ISA) | |
325 | return 0; | |
326 | ||
327 | /* Initialize eeh device */ | |
ab55d218 GS |
328 | edev->class_code = dev->class; |
329 | edev->mode &= 0xFFFFFF00; | |
4b83bd45 GS |
330 | if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) |
331 | edev->mode |= EEH_DEV_BRIDGE; | |
2a18dfc6 | 332 | edev->pcix_cap = pci_find_capability(dev, PCI_CAP_ID_PCIX); |
4b83bd45 GS |
333 | if (pci_is_pcie(dev)) { |
334 | edev->pcie_cap = pci_pcie_cap(dev); | |
335 | ||
336 | if (pci_pcie_type(dev) == PCI_EXP_TYPE_ROOT_PORT) | |
337 | edev->mode |= EEH_DEV_ROOT_PORT; | |
338 | else if (pci_pcie_type(dev) == PCI_EXP_TYPE_DOWNSTREAM) | |
339 | edev->mode |= EEH_DEV_DS_PORT; | |
2a18dfc6 GS |
340 | |
341 | edev->aer_cap = pci_find_ext_capability(dev, | |
342 | PCI_EXT_CAP_ID_ERR); | |
4b83bd45 GS |
343 | } |
344 | ||
29310e5e GS |
345 | edev->config_addr = ((dev->bus->number << 8) | dev->devfn); |
346 | edev->pe_config_addr = phb->bdfn_to_pe(phb, dev->bus, dev->devfn & 0xff); | |
347 | ||
348 | /* Create PE */ | |
dadcd6d6 MQ |
349 | ret = eeh_add_to_parent_pe(edev); |
350 | if (ret) { | |
351 | pr_warn("%s: Can't add PCI dev %s to parent PE (%d)\n", | |
352 | __func__, pci_name(dev), ret); | |
353 | return ret; | |
354 | } | |
355 | ||
b6541db1 GS |
356 | /* |
357 | * If the PE contains any one of following adapters, the | |
358 | * PCI config space can't be accessed when dumping EEH log. | |
359 | * Otherwise, we will run into fenced PHB caused by shortage | |
360 | * of outbound credits in the adapter. The PCI config access | |
361 | * should be blocked until PE reset. MMIO access is dropped | |
362 | * by hardware certainly. In order to drop PCI config requests, | |
363 | * one more flag (EEH_PE_CFG_RESTRICTED) is introduced, which | |
364 | * will be checked in the backend for PE state retrival. If | |
365 | * the PE becomes frozen for the first time and the flag has | |
366 | * been set for the PE, we will set EEH_PE_CFG_BLOCKED for | |
367 | * that PE to block its config space. | |
368 | * | |
369 | * Broadcom Austin 4-ports NICs (14e4:1657) | |
179ea48b | 370 | * Broadcom Shiner 2-ports 10G NICs (14e4:168e) |
b6541db1 | 371 | */ |
179ea48b GS |
372 | if ((dev->vendor == PCI_VENDOR_ID_BROADCOM && dev->device == 0x1657) || |
373 | (dev->vendor == PCI_VENDOR_ID_BROADCOM && dev->device == 0x168e)) | |
b6541db1 GS |
374 | edev->pe->state |= EEH_PE_CFG_RESTRICTED; |
375 | ||
dadcd6d6 MQ |
376 | /* |
377 | * Cache the PE primary bus, which can't be fetched when | |
378 | * full hotplug is in progress. In that case, all child | |
379 | * PCI devices of the PE are expected to be removed prior | |
380 | * to PE reset. | |
381 | */ | |
382 | if (!edev->pe->bus) | |
383 | edev->pe->bus = dev->bus; | |
29310e5e GS |
384 | |
385 | /* | |
386 | * Enable EEH explicitly so that we will do EEH check | |
387 | * while accessing I/O stuff | |
29310e5e | 388 | */ |
05b1721d | 389 | eeh_add_flag(EEH_ENABLED); |
29310e5e GS |
390 | |
391 | /* Save memory bars */ | |
392 | eeh_save_bars(edev); | |
393 | ||
394 | return 0; | |
395 | } | |
396 | ||
397 | /** | |
01f3bfb7 | 398 | * pnv_eeh_set_option - Initialize EEH or MMIO/DMA reenable |
29310e5e GS |
399 | * @pe: EEH PE |
400 | * @option: operation to be issued | |
401 | * | |
402 | * The function is used to control the EEH functionality globally. | |
403 | * Currently, following options are support according to PAPR: | |
404 | * Enable EEH, Disable EEH, Enable MMIO and Enable DMA | |
405 | */ | |
01f3bfb7 | 406 | static int pnv_eeh_set_option(struct eeh_pe *pe, int option) |
29310e5e GS |
407 | { |
408 | struct pci_controller *hose = pe->phb; | |
409 | struct pnv_phb *phb = hose->private_data; | |
7e3e4f8d GS |
410 | bool freeze_pe = false; |
411 | int opt, ret = 0; | |
412 | s64 rc; | |
413 | ||
414 | /* Sanity check on option */ | |
415 | switch (option) { | |
416 | case EEH_OPT_DISABLE: | |
417 | return -EPERM; | |
418 | case EEH_OPT_ENABLE: | |
419 | return 0; | |
420 | case EEH_OPT_THAW_MMIO: | |
421 | opt = OPAL_EEH_ACTION_CLEAR_FREEZE_MMIO; | |
422 | break; | |
423 | case EEH_OPT_THAW_DMA: | |
424 | opt = OPAL_EEH_ACTION_CLEAR_FREEZE_DMA; | |
425 | break; | |
426 | case EEH_OPT_FREEZE_PE: | |
427 | freeze_pe = true; | |
428 | opt = OPAL_EEH_ACTION_SET_FREEZE_ALL; | |
429 | break; | |
430 | default: | |
431 | pr_warn("%s: Invalid option %d\n", __func__, option); | |
432 | return -EINVAL; | |
433 | } | |
29310e5e | 434 | |
7e3e4f8d GS |
435 | /* If PHB supports compound PE, to handle it */ |
436 | if (freeze_pe) { | |
437 | if (phb->freeze_pe) { | |
438 | phb->freeze_pe(phb, pe->addr); | |
439 | } else { | |
440 | rc = opal_pci_eeh_freeze_set(phb->opal_id, | |
441 | pe->addr, opt); | |
442 | if (rc != OPAL_SUCCESS) { | |
443 | pr_warn("%s: Failure %lld freezing " | |
444 | "PHB#%x-PE#%x\n", | |
445 | __func__, rc, | |
446 | phb->hose->global_number, pe->addr); | |
447 | ret = -EIO; | |
448 | } | |
449 | } | |
450 | } else { | |
451 | if (phb->unfreeze_pe) { | |
452 | ret = phb->unfreeze_pe(phb, pe->addr, opt); | |
453 | } else { | |
454 | rc = opal_pci_eeh_freeze_clear(phb->opal_id, | |
455 | pe->addr, opt); | |
456 | if (rc != OPAL_SUCCESS) { | |
457 | pr_warn("%s: Failure %lld enable %d " | |
458 | "for PHB#%x-PE#%x\n", | |
459 | __func__, rc, option, | |
460 | phb->hose->global_number, pe->addr); | |
461 | ret = -EIO; | |
462 | } | |
463 | } | |
464 | } | |
29310e5e GS |
465 | |
466 | return ret; | |
467 | } | |
468 | ||
469 | /** | |
01f3bfb7 | 470 | * pnv_eeh_get_pe_addr - Retrieve PE address |
29310e5e GS |
471 | * @pe: EEH PE |
472 | * | |
473 | * Retrieve the PE address according to the given tranditional | |
474 | * PCI BDF (Bus/Device/Function) address. | |
475 | */ | |
01f3bfb7 | 476 | static int pnv_eeh_get_pe_addr(struct eeh_pe *pe) |
29310e5e GS |
477 | { |
478 | return pe->addr; | |
479 | } | |
480 | ||
40ae5f69 GS |
481 | static void pnv_eeh_get_phb_diag(struct eeh_pe *pe) |
482 | { | |
483 | struct pnv_phb *phb = pe->phb->private_data; | |
484 | s64 rc; | |
485 | ||
486 | rc = opal_pci_get_phb_diag_data2(phb->opal_id, pe->data, | |
487 | PNV_PCI_DIAG_BUF_SIZE); | |
488 | if (rc != OPAL_SUCCESS) | |
489 | pr_warn("%s: Failure %lld getting PHB#%x diag-data\n", | |
490 | __func__, rc, pe->phb->global_number); | |
491 | } | |
492 | ||
493 | static int pnv_eeh_get_phb_state(struct eeh_pe *pe) | |
494 | { | |
495 | struct pnv_phb *phb = pe->phb->private_data; | |
496 | u8 fstate; | |
497 | __be16 pcierr; | |
498 | s64 rc; | |
499 | int result = 0; | |
500 | ||
501 | rc = opal_pci_eeh_freeze_status(phb->opal_id, | |
502 | pe->addr, | |
503 | &fstate, | |
504 | &pcierr, | |
505 | NULL); | |
506 | if (rc != OPAL_SUCCESS) { | |
507 | pr_warn("%s: Failure %lld getting PHB#%x state\n", | |
508 | __func__, rc, phb->hose->global_number); | |
509 | return EEH_STATE_NOT_SUPPORT; | |
510 | } | |
511 | ||
512 | /* | |
513 | * Check PHB state. If the PHB is frozen for the | |
514 | * first time, to dump the PHB diag-data. | |
515 | */ | |
516 | if (be16_to_cpu(pcierr) != OPAL_EEH_PHB_ERROR) { | |
517 | result = (EEH_STATE_MMIO_ACTIVE | | |
518 | EEH_STATE_DMA_ACTIVE | | |
519 | EEH_STATE_MMIO_ENABLED | | |
520 | EEH_STATE_DMA_ENABLED); | |
521 | } else if (!(pe->state & EEH_PE_ISOLATED)) { | |
522 | eeh_pe_state_mark(pe, EEH_PE_ISOLATED); | |
523 | pnv_eeh_get_phb_diag(pe); | |
524 | ||
525 | if (eeh_has_flag(EEH_EARLY_DUMP_LOG)) | |
526 | pnv_pci_dump_phb_diag_data(pe->phb, pe->data); | |
527 | } | |
528 | ||
529 | return result; | |
530 | } | |
531 | ||
532 | static int pnv_eeh_get_pe_state(struct eeh_pe *pe) | |
533 | { | |
534 | struct pnv_phb *phb = pe->phb->private_data; | |
535 | u8 fstate; | |
536 | __be16 pcierr; | |
537 | s64 rc; | |
538 | int result; | |
539 | ||
540 | /* | |
541 | * We don't clobber hardware frozen state until PE | |
542 | * reset is completed. In order to keep EEH core | |
543 | * moving forward, we have to return operational | |
544 | * state during PE reset. | |
545 | */ | |
546 | if (pe->state & EEH_PE_RESET) { | |
547 | result = (EEH_STATE_MMIO_ACTIVE | | |
548 | EEH_STATE_DMA_ACTIVE | | |
549 | EEH_STATE_MMIO_ENABLED | | |
550 | EEH_STATE_DMA_ENABLED); | |
551 | return result; | |
552 | } | |
553 | ||
554 | /* | |
555 | * Fetch PE state from hardware. If the PHB | |
556 | * supports compound PE, let it handle that. | |
557 | */ | |
558 | if (phb->get_pe_state) { | |
559 | fstate = phb->get_pe_state(phb, pe->addr); | |
560 | } else { | |
561 | rc = opal_pci_eeh_freeze_status(phb->opal_id, | |
562 | pe->addr, | |
563 | &fstate, | |
564 | &pcierr, | |
565 | NULL); | |
566 | if (rc != OPAL_SUCCESS) { | |
567 | pr_warn("%s: Failure %lld getting PHB#%x-PE%x state\n", | |
568 | __func__, rc, phb->hose->global_number, | |
569 | pe->addr); | |
570 | return EEH_STATE_NOT_SUPPORT; | |
571 | } | |
572 | } | |
573 | ||
574 | /* Figure out state */ | |
575 | switch (fstate) { | |
576 | case OPAL_EEH_STOPPED_NOT_FROZEN: | |
577 | result = (EEH_STATE_MMIO_ACTIVE | | |
578 | EEH_STATE_DMA_ACTIVE | | |
579 | EEH_STATE_MMIO_ENABLED | | |
580 | EEH_STATE_DMA_ENABLED); | |
581 | break; | |
582 | case OPAL_EEH_STOPPED_MMIO_FREEZE: | |
583 | result = (EEH_STATE_DMA_ACTIVE | | |
584 | EEH_STATE_DMA_ENABLED); | |
585 | break; | |
586 | case OPAL_EEH_STOPPED_DMA_FREEZE: | |
587 | result = (EEH_STATE_MMIO_ACTIVE | | |
588 | EEH_STATE_MMIO_ENABLED); | |
589 | break; | |
590 | case OPAL_EEH_STOPPED_MMIO_DMA_FREEZE: | |
591 | result = 0; | |
592 | break; | |
593 | case OPAL_EEH_STOPPED_RESET: | |
594 | result = EEH_STATE_RESET_ACTIVE; | |
595 | break; | |
596 | case OPAL_EEH_STOPPED_TEMP_UNAVAIL: | |
597 | result = EEH_STATE_UNAVAILABLE; | |
598 | break; | |
599 | case OPAL_EEH_STOPPED_PERM_UNAVAIL: | |
600 | result = EEH_STATE_NOT_SUPPORT; | |
601 | break; | |
602 | default: | |
603 | result = EEH_STATE_NOT_SUPPORT; | |
604 | pr_warn("%s: Invalid PHB#%x-PE#%x state %x\n", | |
605 | __func__, phb->hose->global_number, | |
606 | pe->addr, fstate); | |
607 | } | |
608 | ||
609 | /* | |
610 | * If PHB supports compound PE, to freeze all | |
611 | * slave PEs for consistency. | |
612 | * | |
613 | * If the PE is switching to frozen state for the | |
614 | * first time, to dump the PHB diag-data. | |
615 | */ | |
616 | if (!(result & EEH_STATE_NOT_SUPPORT) && | |
617 | !(result & EEH_STATE_UNAVAILABLE) && | |
618 | !(result & EEH_STATE_MMIO_ACTIVE) && | |
619 | !(result & EEH_STATE_DMA_ACTIVE) && | |
620 | !(pe->state & EEH_PE_ISOLATED)) { | |
621 | if (phb->freeze_pe) | |
622 | phb->freeze_pe(phb, pe->addr); | |
623 | ||
624 | eeh_pe_state_mark(pe, EEH_PE_ISOLATED); | |
625 | pnv_eeh_get_phb_diag(pe); | |
626 | ||
627 | if (eeh_has_flag(EEH_EARLY_DUMP_LOG)) | |
628 | pnv_pci_dump_phb_diag_data(pe->phb, pe->data); | |
629 | } | |
630 | ||
631 | return result; | |
632 | } | |
633 | ||
29310e5e | 634 | /** |
01f3bfb7 | 635 | * pnv_eeh_get_state - Retrieve PE state |
29310e5e GS |
636 | * @pe: EEH PE |
637 | * @delay: delay while PE state is temporarily unavailable | |
638 | * | |
639 | * Retrieve the state of the specified PE. For IODA-compitable | |
640 | * platform, it should be retrieved from IODA table. Therefore, | |
641 | * we prefer passing down to hardware implementation to handle | |
642 | * it. | |
643 | */ | |
01f3bfb7 | 644 | static int pnv_eeh_get_state(struct eeh_pe *pe, int *delay) |
29310e5e | 645 | { |
40ae5f69 | 646 | int ret; |
29310e5e | 647 | |
40ae5f69 GS |
648 | if (pe->type & EEH_PE_PHB) |
649 | ret = pnv_eeh_get_phb_state(pe); | |
650 | else | |
651 | ret = pnv_eeh_get_pe_state(pe); | |
29310e5e | 652 | |
40ae5f69 GS |
653 | if (!delay) |
654 | return ret; | |
655 | ||
656 | /* | |
657 | * If the PE state is temporarily unavailable, | |
658 | * to inform the EEH core delay for default | |
659 | * period (1 second) | |
660 | */ | |
661 | *delay = 0; | |
662 | if (ret & EEH_STATE_UNAVAILABLE) | |
663 | *delay = 1000; | |
29310e5e GS |
664 | |
665 | return ret; | |
666 | } | |
667 | ||
668 | /** | |
01f3bfb7 | 669 | * pnv_eeh_reset - Reset the specified PE |
29310e5e GS |
670 | * @pe: EEH PE |
671 | * @option: reset option | |
672 | * | |
673 | * Reset the specified PE | |
674 | */ | |
01f3bfb7 | 675 | static int pnv_eeh_reset(struct eeh_pe *pe, int option) |
29310e5e GS |
676 | { |
677 | struct pci_controller *hose = pe->phb; | |
678 | struct pnv_phb *phb = hose->private_data; | |
679 | int ret = -EEXIST; | |
680 | ||
681 | if (phb->eeh_ops && phb->eeh_ops->reset) | |
682 | ret = phb->eeh_ops->reset(pe, option); | |
683 | ||
684 | return ret; | |
685 | } | |
686 | ||
687 | /** | |
01f3bfb7 | 688 | * pnv_eeh_wait_state - Wait for PE state |
29310e5e GS |
689 | * @pe: EEH PE |
690 | * @max_wait: maximal period in microsecond | |
691 | * | |
692 | * Wait for the state of associated PE. It might take some time | |
693 | * to retrieve the PE's state. | |
694 | */ | |
01f3bfb7 | 695 | static int pnv_eeh_wait_state(struct eeh_pe *pe, int max_wait) |
29310e5e GS |
696 | { |
697 | int ret; | |
698 | int mwait; | |
699 | ||
700 | while (1) { | |
01f3bfb7 | 701 | ret = pnv_eeh_get_state(pe, &mwait); |
29310e5e GS |
702 | |
703 | /* | |
704 | * If the PE's state is temporarily unavailable, | |
705 | * we have to wait for the specified time. Otherwise, | |
706 | * the PE's state will be returned immediately. | |
707 | */ | |
708 | if (ret != EEH_STATE_UNAVAILABLE) | |
709 | return ret; | |
710 | ||
711 | max_wait -= mwait; | |
712 | if (max_wait <= 0) { | |
0dae2743 GS |
713 | pr_warn("%s: Timeout getting PE#%x's state (%d)\n", |
714 | __func__, pe->addr, max_wait); | |
29310e5e GS |
715 | return EEH_STATE_NOT_SUPPORT; |
716 | } | |
717 | ||
718 | msleep(mwait); | |
719 | } | |
720 | ||
721 | return EEH_STATE_NOT_SUPPORT; | |
722 | } | |
723 | ||
724 | /** | |
01f3bfb7 | 725 | * pnv_eeh_get_log - Retrieve error log |
29310e5e GS |
726 | * @pe: EEH PE |
727 | * @severity: temporary or permanent error log | |
728 | * @drv_log: driver log to be combined with retrieved error log | |
729 | * @len: length of driver log | |
730 | * | |
731 | * Retrieve the temporary or permanent error from the PE. | |
732 | */ | |
01f3bfb7 GS |
733 | static int pnv_eeh_get_log(struct eeh_pe *pe, int severity, |
734 | char *drv_log, unsigned long len) | |
29310e5e | 735 | { |
95edcdea GS |
736 | if (!eeh_has_flag(EEH_EARLY_DUMP_LOG)) |
737 | pnv_pci_dump_phb_diag_data(pe->phb, pe->data); | |
29310e5e | 738 | |
95edcdea | 739 | return 0; |
29310e5e GS |
740 | } |
741 | ||
742 | /** | |
01f3bfb7 | 743 | * pnv_eeh_configure_bridge - Configure PCI bridges in the indicated PE |
29310e5e GS |
744 | * @pe: EEH PE |
745 | * | |
746 | * The function will be called to reconfigure the bridges included | |
747 | * in the specified PE so that the mulfunctional PE would be recovered | |
748 | * again. | |
749 | */ | |
01f3bfb7 | 750 | static int pnv_eeh_configure_bridge(struct eeh_pe *pe) |
29310e5e | 751 | { |
bbe170ed | 752 | return 0; |
29310e5e GS |
753 | } |
754 | ||
131c123a | 755 | /** |
01f3bfb7 | 756 | * pnv_pe_err_inject - Inject specified error to the indicated PE |
131c123a GS |
757 | * @pe: the indicated PE |
758 | * @type: error type | |
759 | * @func: specific error type | |
760 | * @addr: address | |
761 | * @mask: address mask | |
762 | * | |
763 | * The routine is called to inject specified error, which is | |
764 | * determined by @type and @func, to the indicated PE for | |
765 | * testing purpose. | |
766 | */ | |
01f3bfb7 GS |
767 | static int pnv_eeh_err_inject(struct eeh_pe *pe, int type, int func, |
768 | unsigned long addr, unsigned long mask) | |
131c123a GS |
769 | { |
770 | struct pci_controller *hose = pe->phb; | |
771 | struct pnv_phb *phb = hose->private_data; | |
fa646c3c GS |
772 | s64 rc; |
773 | ||
774 | /* Sanity check on error type */ | |
775 | if (type != OPAL_ERR_INJECT_TYPE_IOA_BUS_ERR && | |
776 | type != OPAL_ERR_INJECT_TYPE_IOA_BUS_ERR64) { | |
777 | pr_warn("%s: Invalid error type %d\n", | |
778 | __func__, type); | |
779 | return -ERANGE; | |
780 | } | |
131c123a | 781 | |
fa646c3c GS |
782 | if (func < OPAL_ERR_INJECT_FUNC_IOA_LD_MEM_ADDR || |
783 | func > OPAL_ERR_INJECT_FUNC_IOA_DMA_WR_TARGET) { | |
784 | pr_warn("%s: Invalid error function %d\n", | |
785 | __func__, func); | |
786 | return -ERANGE; | |
787 | } | |
131c123a | 788 | |
fa646c3c GS |
789 | /* Firmware supports error injection ? */ |
790 | if (!opal_check_token(OPAL_PCI_ERR_INJECT)) { | |
791 | pr_warn("%s: Firmware doesn't support error injection\n", | |
792 | __func__); | |
793 | return -ENXIO; | |
794 | } | |
795 | ||
796 | /* Do error injection */ | |
797 | rc = opal_pci_err_inject(phb->opal_id, pe->addr, | |
798 | type, func, addr, mask); | |
799 | if (rc != OPAL_SUCCESS) { | |
800 | pr_warn("%s: Failure %lld injecting error " | |
801 | "%d-%d to PHB#%x-PE#%x\n", | |
802 | __func__, rc, type, func, | |
803 | hose->global_number, pe->addr); | |
804 | return -EIO; | |
805 | } | |
806 | ||
807 | return 0; | |
131c123a GS |
808 | } |
809 | ||
01f3bfb7 | 810 | static inline bool pnv_eeh_cfg_blocked(struct device_node *dn) |
d2cfbcd7 GS |
811 | { |
812 | struct eeh_dev *edev = of_node_to_eeh_dev(dn); | |
813 | ||
814 | if (!edev || !edev->pe) | |
815 | return false; | |
816 | ||
817 | if (edev->pe->state & EEH_PE_CFG_BLOCKED) | |
818 | return true; | |
819 | ||
820 | return false; | |
821 | } | |
822 | ||
01f3bfb7 GS |
823 | static int pnv_eeh_read_config(struct device_node *dn, |
824 | int where, int size, u32 *val) | |
d2cfbcd7 | 825 | { |
01f3bfb7 | 826 | if (pnv_eeh_cfg_blocked(dn)) { |
d2cfbcd7 GS |
827 | *val = 0xFFFFFFFF; |
828 | return PCIBIOS_SET_FAILED; | |
829 | } | |
830 | ||
831 | return pnv_pci_cfg_read(dn, where, size, val); | |
832 | } | |
833 | ||
01f3bfb7 GS |
834 | static int pnv_eeh_write_config(struct device_node *dn, |
835 | int where, int size, u32 val) | |
d2cfbcd7 | 836 | { |
01f3bfb7 | 837 | if (pnv_eeh_cfg_blocked(dn)) |
d2cfbcd7 GS |
838 | return PCIBIOS_SET_FAILED; |
839 | ||
840 | return pnv_pci_cfg_write(dn, where, size, val); | |
841 | } | |
842 | ||
29310e5e | 843 | /** |
01f3bfb7 | 844 | * pnv_eeh_next_error - Retrieve next EEH error to handle |
29310e5e GS |
845 | * @pe: Affected PE |
846 | * | |
847 | * Using OPAL API, to retrieve next EEH error for EEH core to handle | |
848 | */ | |
01f3bfb7 | 849 | static int pnv_eeh_next_error(struct eeh_pe **pe) |
29310e5e GS |
850 | { |
851 | struct pci_controller *hose; | |
852 | struct pnv_phb *phb = NULL; | |
853 | ||
854 | list_for_each_entry(hose, &hose_list, list_node) { | |
855 | phb = hose->private_data; | |
856 | break; | |
857 | } | |
858 | ||
859 | if (phb && phb->eeh_ops->next_error) | |
860 | return phb->eeh_ops->next_error(pe); | |
861 | ||
862 | return -EEXIST; | |
863 | } | |
864 | ||
01f3bfb7 | 865 | static int pnv_eeh_restore_config(struct device_node *dn) |
9be3becc GS |
866 | { |
867 | struct eeh_dev *edev = of_node_to_eeh_dev(dn); | |
868 | struct pnv_phb *phb; | |
869 | s64 ret; | |
870 | ||
871 | if (!edev) | |
872 | return -EEXIST; | |
873 | ||
874 | phb = edev->phb->private_data; | |
875 | ret = opal_pci_reinit(phb->opal_id, | |
876 | OPAL_REINIT_PCI_DEV, edev->config_addr); | |
877 | if (ret) { | |
878 | pr_warn("%s: Can't reinit PCI dev 0x%x (%lld)\n", | |
879 | __func__, edev->config_addr, ret); | |
880 | return -EIO; | |
881 | } | |
882 | ||
883 | return 0; | |
884 | } | |
885 | ||
01f3bfb7 | 886 | static struct eeh_ops pnv_eeh_ops = { |
29310e5e | 887 | .name = "powernv", |
01f3bfb7 GS |
888 | .init = pnv_eeh_init, |
889 | .post_init = pnv_eeh_post_init, | |
29310e5e | 890 | .of_probe = NULL, |
01f3bfb7 GS |
891 | .dev_probe = pnv_eeh_dev_probe, |
892 | .set_option = pnv_eeh_set_option, | |
893 | .get_pe_addr = pnv_eeh_get_pe_addr, | |
894 | .get_state = pnv_eeh_get_state, | |
895 | .reset = pnv_eeh_reset, | |
896 | .wait_state = pnv_eeh_wait_state, | |
897 | .get_log = pnv_eeh_get_log, | |
898 | .configure_bridge = pnv_eeh_configure_bridge, | |
899 | .err_inject = pnv_eeh_err_inject, | |
900 | .read_config = pnv_eeh_read_config, | |
901 | .write_config = pnv_eeh_write_config, | |
902 | .next_error = pnv_eeh_next_error, | |
903 | .restore_config = pnv_eeh_restore_config | |
29310e5e GS |
904 | }; |
905 | ||
906 | /** | |
907 | * eeh_powernv_init - Register platform dependent EEH operations | |
908 | * | |
909 | * EEH initialization on powernv platform. This function should be | |
910 | * called before any EEH related functions. | |
911 | */ | |
912 | static int __init eeh_powernv_init(void) | |
913 | { | |
914 | int ret = -EINVAL; | |
915 | ||
bb593c00 | 916 | eeh_set_pe_aux_size(PNV_PCI_DIAG_BUF_SIZE); |
01f3bfb7 | 917 | ret = eeh_ops_register(&pnv_eeh_ops); |
29310e5e GS |
918 | if (!ret) |
919 | pr_info("EEH: PowerNV platform initialized\n"); | |
920 | else | |
921 | pr_info("EEH: Failed to initialize PowerNV platform (%d)\n", ret); | |
922 | ||
923 | return ret; | |
924 | } | |
b14726c5 | 925 | machine_early_initcall(powernv, eeh_powernv_init); |