1 /* SPDX-License-Identifier: GPL-2.0-or-later */
2 /*
3 * Copyright (C) 2001 Dave Engebretsen & Todd Inglett IBM Corporation.
4 * Copyright 2001-2012 IBM Corporation.
5 */
6
7 #ifndef _POWERPC_EEH_H
8 #define _POWERPC_EEH_H
9 #ifdef __KERNEL__
10
11 #include <linux/init.h>
12 #include <linux/list.h>
13 #include <linux/string.h>
14 #include <linux/time.h>
15 #include <linux/atomic.h>
16
17 #include <uapi/asm/eeh.h>
18
19 struct pci_dev;
20 struct pci_bus;
21 struct pci_dn;
22
23 #ifdef CONFIG_EEH
24
25 /* EEH subsystem flags */
26 #define EEH_ENABLED 0x01 /* EEH enabled */
27 #define EEH_FORCE_DISABLED 0x02 /* EEH disabled */
28 #define EEH_PROBE_MODE_DEV 0x04 /* From PCI device */
29 #define EEH_PROBE_MODE_DEVTREE 0x08 /* From device tree */
30 #define EEH_ENABLE_IO_FOR_LOG 0x20 /* Enable IO for log */
31 #define EEH_EARLY_DUMP_LOG 0x40 /* Dump log immediately */
32
33 /*
34 * Delay for PE reset, all in ms
35 *
36 * PCI specification has reset hold time of 100 milliseconds.
37 * We have 250 milliseconds here. The PCI bus settlement time
38 * is specified as 1.5 seconds and we have 1.8 seconds.
39 */
40 #define EEH_PE_RST_HOLD_TIME 250
41 #define EEH_PE_RST_SETTLE_TIME 1800
42
43 /*
44 * The struct is used to trace PE related EEH functionality.
45 * In theory, there will have one instance of the struct to
46 * be created against particular PE. In nature, PEs correlate
47 * to each other. the struct has to reflect that hierarchy in
48 * order to easily pick up those affected PEs when one particular
49 * PE has EEH errors.
50 *
51 * Also, one particular PE might be composed of PCI device, PCI
52 * bus and its subordinate components. The struct also need ship
53 * the information. Further more, one particular PE is only meaingful
54 * in the corresponding PHB. Therefore, the root PEs should be created
55 * against existing PHBs in on-to-one fashion.
56 */
57 #define EEH_PE_INVALID (1 << 0) /* Invalid */
58 #define EEH_PE_PHB (1 << 1) /* PHB PE */
59 #define EEH_PE_DEVICE (1 << 2) /* Device PE */
60 #define EEH_PE_BUS (1 << 3) /* Bus PE */
61 #define EEH_PE_VF (1 << 4) /* VF PE */
62
63 #define EEH_PE_ISOLATED (1 << 0) /* Isolated PE */
64 #define EEH_PE_RECOVERING (1 << 1) /* Recovering PE */
65 #define EEH_PE_CFG_BLOCKED (1 << 2) /* Block config access */
66 #define EEH_PE_RESET (1 << 3) /* PE reset in progress */
67
68 #define EEH_PE_KEEP (1 << 8) /* Keep PE on hotplug */
69 #define EEH_PE_CFG_RESTRICTED (1 << 9) /* Block config on error */
70 #define EEH_PE_REMOVED (1 << 10) /* Removed permanently */
71 #define EEH_PE_PRI_BUS (1 << 11) /* Cached primary bus */
72
73 struct eeh_pe {
74 int type; /* PE type: PHB/Bus/Device */
75 int state; /* PE EEH dependent mode */
76 int addr; /* PE configuration address */
77 struct pci_controller *phb; /* Associated PHB */
78 struct pci_bus *bus; /* Top PCI bus for bus PE */
79 int check_count; /* Times of ignored error */
80 int freeze_count; /* Times of froze up */
81 time64_t tstamp; /* Time on first-time freeze */
82 int false_positives; /* Times of reported #ff's */
83 atomic_t pass_dev_cnt; /* Count of passed through devs */
84 struct eeh_pe *parent; /* Parent PE */
85 void *data; /* PE auxiliary data */
86 struct list_head child_list; /* List of PEs below this PE */
87 struct list_head child; /* Memb. child_list/eeh_phb_pe */
88 struct list_head edevs; /* List of eeh_dev in this PE */
89
90 #ifdef CONFIG_STACKTRACE
91 /*
92 * Saved stack trace. When we find a PE freeze in eeh_dev_check_failure
93 * the stack trace is saved here so we can print it in the recovery
94 * thread if it turns out to due to a real problem rather than
95 * a hot-remove.
96 *
97 * A max of 64 entries might be overkill, but it also might not be.
98 */
99 unsigned long stack_trace[64];
100 int trace_entries;
101 #endif /* CONFIG_STACKTRACE */
102 };
103
104 #define eeh_pe_for_each_dev(pe, edev, tmp) \
105 list_for_each_entry_safe(edev, tmp, &pe->edevs, entry)
106
107 #define eeh_for_each_pe(root, pe) \
108 for (pe = root; pe; pe = eeh_pe_next(pe, root))
109
eeh_pe_passed(struct eeh_pe * pe)110 static inline bool eeh_pe_passed(struct eeh_pe *pe)
111 {
112 return pe ? !!atomic_read(&pe->pass_dev_cnt) : false;
113 }
114
115 /*
116 * The struct is used to trace EEH state for the associated
117 * PCI device node or PCI device. In future, it might
118 * represent PE as well so that the EEH device to form
119 * another tree except the currently existing tree of PCI
120 * buses and PCI devices
121 */
122 #define EEH_DEV_BRIDGE (1 << 0) /* PCI bridge */
123 #define EEH_DEV_ROOT_PORT (1 << 1) /* PCIe root port */
124 #define EEH_DEV_DS_PORT (1 << 2) /* Downstream port */
125 #define EEH_DEV_IRQ_DISABLED (1 << 3) /* Interrupt disabled */
126 #define EEH_DEV_DISCONNECTED (1 << 4) /* Removing from PE */
127
128 #define EEH_DEV_NO_HANDLER (1 << 8) /* No error handler */
129 #define EEH_DEV_SYSFS (1 << 9) /* Sysfs created */
130 #define EEH_DEV_REMOVED (1 << 10) /* Removed permanently */
131
132 struct eeh_dev {
133 int mode; /* EEH mode */
134 int bdfn; /* bdfn of device (for cfg ops) */
135 struct pci_controller *controller;
136 int pe_config_addr; /* PE config address */
137 u32 config_space[16]; /* Saved PCI config space */
138 int pcix_cap; /* Saved PCIx capability */
139 int pcie_cap; /* Saved PCIe capability */
140 int aer_cap; /* Saved AER capability */
141 int af_cap; /* Saved AF capability */
142 struct eeh_pe *pe; /* Associated PE */
143 struct list_head entry; /* Membership in eeh_pe.edevs */
144 struct list_head rmv_entry; /* Membership in rmv_list */
145 struct pci_dn *pdn; /* Associated PCI device node */
146 struct pci_dev *pdev; /* Associated PCI device */
147 bool in_error; /* Error flag for edev */
148
149 /* VF specific properties */
150 struct pci_dev *physfn; /* Associated SRIOV PF */
151 int vf_index; /* Index of this VF */
152 };
153
154 /* "fmt" must be a simple literal string */
155 #define EEH_EDEV_PRINT(level, edev, fmt, ...) \
156 pr_##level("PCI %04x:%02x:%02x.%x#%04x: EEH: " fmt, \
157 (edev)->controller->global_number, PCI_BUSNO((edev)->bdfn), \
158 PCI_SLOT((edev)->bdfn), PCI_FUNC((edev)->bdfn), \
159 ((edev)->pe ? (edev)->pe_config_addr : 0xffff), ##__VA_ARGS__)
160 #define eeh_edev_dbg(edev, fmt, ...) EEH_EDEV_PRINT(debug, (edev), fmt, ##__VA_ARGS__)
161 #define eeh_edev_info(edev, fmt, ...) EEH_EDEV_PRINT(info, (edev), fmt, ##__VA_ARGS__)
162 #define eeh_edev_warn(edev, fmt, ...) EEH_EDEV_PRINT(warn, (edev), fmt, ##__VA_ARGS__)
163 #define eeh_edev_err(edev, fmt, ...) EEH_EDEV_PRINT(err, (edev), fmt, ##__VA_ARGS__)
164
eeh_dev_to_pdn(struct eeh_dev * edev)165 static inline struct pci_dn *eeh_dev_to_pdn(struct eeh_dev *edev)
166 {
167 return edev ? edev->pdn : NULL;
168 }
169
eeh_dev_to_pci_dev(struct eeh_dev * edev)170 static inline struct pci_dev *eeh_dev_to_pci_dev(struct eeh_dev *edev)
171 {
172 return edev ? edev->pdev : NULL;
173 }
174
eeh_dev_to_pe(struct eeh_dev * edev)175 static inline struct eeh_pe *eeh_dev_to_pe(struct eeh_dev* edev)
176 {
177 return edev ? edev->pe : NULL;
178 }
179
180 /* Return values from eeh_ops::next_error */
181 enum {
182 EEH_NEXT_ERR_NONE = 0,
183 EEH_NEXT_ERR_INF,
184 EEH_NEXT_ERR_FROZEN_PE,
185 EEH_NEXT_ERR_FENCED_PHB,
186 EEH_NEXT_ERR_DEAD_PHB,
187 EEH_NEXT_ERR_DEAD_IOC
188 };
189
190 /*
191 * The struct is used to trace the registered EEH operation
192 * callback functions. Actually, those operation callback
193 * functions are heavily platform dependent. That means the
194 * platform should register its own EEH operation callback
195 * functions before any EEH further operations.
196 */
197 #define EEH_OPT_DISABLE 0 /* EEH disable */
198 #define EEH_OPT_ENABLE 1 /* EEH enable */
199 #define EEH_OPT_THAW_MMIO 2 /* MMIO enable */
200 #define EEH_OPT_THAW_DMA 3 /* DMA enable */
201 #define EEH_OPT_FREEZE_PE 4 /* Freeze PE */
202 #define EEH_STATE_UNAVAILABLE (1 << 0) /* State unavailable */
203 #define EEH_STATE_NOT_SUPPORT (1 << 1) /* EEH not supported */
204 #define EEH_STATE_RESET_ACTIVE (1 << 2) /* Active reset */
205 #define EEH_STATE_MMIO_ACTIVE (1 << 3) /* Active MMIO */
206 #define EEH_STATE_DMA_ACTIVE (1 << 4) /* Active DMA */
207 #define EEH_STATE_MMIO_ENABLED (1 << 5) /* MMIO enabled */
208 #define EEH_STATE_DMA_ENABLED (1 << 6) /* DMA enabled */
209 #define EEH_RESET_DEACTIVATE 0 /* Deactivate the PE reset */
210 #define EEH_RESET_HOT 1 /* Hot reset */
211 #define EEH_RESET_FUNDAMENTAL 3 /* Fundamental reset */
212 #define EEH_LOG_TEMP 1 /* EEH temporary error log */
213 #define EEH_LOG_PERM 2 /* EEH permanent error log */
214
215 struct eeh_ops {
216 char *name;
217 struct eeh_dev *(*probe)(struct pci_dev *pdev);
218 int (*set_option)(struct eeh_pe *pe, int option);
219 int (*get_state)(struct eeh_pe *pe, int *delay);
220 int (*reset)(struct eeh_pe *pe, int option);
221 int (*get_log)(struct eeh_pe *pe, int severity, char *drv_log, unsigned long len);
222 int (*configure_bridge)(struct eeh_pe *pe);
223 int (*err_inject)(struct eeh_pe *pe, int type, int func,
224 unsigned long addr, unsigned long mask);
225 int (*read_config)(struct eeh_dev *edev, int where, int size, u32 *val);
226 int (*write_config)(struct eeh_dev *edev, int where, int size, u32 val);
227 int (*next_error)(struct eeh_pe **pe);
228 int (*restore_config)(struct eeh_dev *edev);
229 int (*notify_resume)(struct eeh_dev *edev);
230 };
231
232 extern int eeh_subsystem_flags;
233 extern u32 eeh_max_freezes;
234 extern bool eeh_debugfs_no_recover;
235 extern struct eeh_ops *eeh_ops;
236 extern raw_spinlock_t confirm_error_lock;
237
eeh_add_flag(int flag)238 static inline void eeh_add_flag(int flag)
239 {
240 eeh_subsystem_flags |= flag;
241 }
242
eeh_clear_flag(int flag)243 static inline void eeh_clear_flag(int flag)
244 {
245 eeh_subsystem_flags &= ~flag;
246 }
247
eeh_has_flag(int flag)248 static inline bool eeh_has_flag(int flag)
249 {
250 return !!(eeh_subsystem_flags & flag);
251 }
252
eeh_enabled(void)253 static inline bool eeh_enabled(void)
254 {
255 return eeh_has_flag(EEH_ENABLED) && !eeh_has_flag(EEH_FORCE_DISABLED);
256 }
257
eeh_serialize_lock(unsigned long * flags)258 static inline void eeh_serialize_lock(unsigned long *flags)
259 {
260 raw_spin_lock_irqsave(&confirm_error_lock, *flags);
261 }
262
eeh_serialize_unlock(unsigned long flags)263 static inline void eeh_serialize_unlock(unsigned long flags)
264 {
265 raw_spin_unlock_irqrestore(&confirm_error_lock, flags);
266 }
267
eeh_state_active(int state)268 static inline bool eeh_state_active(int state)
269 {
270 return (state & (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE))
271 == (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE);
272 }
273
274 typedef void (*eeh_edev_traverse_func)(struct eeh_dev *edev, void *flag);
275 typedef void *(*eeh_pe_traverse_func)(struct eeh_pe *pe, void *flag);
276 void eeh_set_pe_aux_size(int size);
277 int eeh_phb_pe_create(struct pci_controller *phb);
278 int eeh_wait_state(struct eeh_pe *pe, int max_wait);
279 struct eeh_pe *eeh_phb_pe_get(struct pci_controller *phb);
280 struct eeh_pe *eeh_pe_next(struct eeh_pe *pe, struct eeh_pe *root);
281 struct eeh_pe *eeh_pe_get(struct pci_controller *phb, int pe_no);
282 int eeh_pe_tree_insert(struct eeh_dev *edev, struct eeh_pe *new_pe_parent);
283 int eeh_pe_tree_remove(struct eeh_dev *edev);
284 void eeh_pe_update_time_stamp(struct eeh_pe *pe);
285 void *eeh_pe_traverse(struct eeh_pe *root,
286 eeh_pe_traverse_func fn, void *flag);
287 void eeh_pe_dev_traverse(struct eeh_pe *root,
288 eeh_edev_traverse_func fn, void *flag);
289 void eeh_pe_restore_bars(struct eeh_pe *pe);
290 const char *eeh_pe_loc_get(struct eeh_pe *pe);
291 struct pci_bus *eeh_pe_bus_get(struct eeh_pe *pe);
292 const char *eeh_pe_loc_get_bus(struct pci_bus *bus);
293 struct pci_bus *eeh_pe_bus_get_nolock(struct eeh_pe *pe);
294
295 void eeh_show_enabled(void);
296 int __init eeh_init(struct eeh_ops *ops);
297 int eeh_check_failure(const volatile void __iomem *token);
298 int eeh_dev_check_failure(struct eeh_dev *edev);
299 void eeh_addr_cache_init(void);
300 void eeh_probe_device(struct pci_dev *pdev);
301 void eeh_remove_device(struct pci_dev *);
302 int eeh_unfreeze_pe(struct eeh_pe *pe);
303 int eeh_pe_reset_and_recover(struct eeh_pe *pe);
304 int eeh_dev_open(struct pci_dev *pdev);
305 void eeh_dev_release(struct pci_dev *pdev);
306 struct eeh_pe *eeh_iommu_group_to_pe(struct iommu_group *group);
307 int eeh_pe_set_option(struct eeh_pe *pe, int option);
308 int eeh_pe_get_state(struct eeh_pe *pe);
309 int eeh_pe_reset(struct eeh_pe *pe, int option, bool include_passed);
310 int eeh_pe_configure(struct eeh_pe *pe);
311 int eeh_pe_inject_err(struct eeh_pe *pe, int type, int func,
312 unsigned long addr, unsigned long mask);
313 int eeh_pe_inject_mmio_error(struct pci_dev *pdev);
314
315 /**
316 * EEH_POSSIBLE_ERROR() -- test for possible MMIO failure.
317 *
318 * If this macro yields TRUE, the caller relays to eeh_check_failure()
319 * which does further tests out of line.
320 */
321 #define EEH_POSSIBLE_ERROR(val, type) ((val) == (type)~0 && eeh_enabled())
322
323 /*
324 * Reads from a device which has been isolated by EEH will return
325 * all 1s. This macro gives an all-1s value of the given size (in
326 * bytes: 1, 2, or 4) for comparing with the result of a read.
327 */
328 #define EEH_IO_ERROR_VALUE(size) (~0U >> ((4 - (size)) * 8))
329
330 #else /* !CONFIG_EEH */
331
eeh_enabled(void)332 static inline bool eeh_enabled(void)
333 {
334 return false;
335 }
336
eeh_show_enabled(void)337 static inline void eeh_show_enabled(void) { }
338
eeh_check_failure(const volatile void __iomem * token)339 static inline int eeh_check_failure(const volatile void __iomem *token)
340 {
341 return 0;
342 }
343
344 #define eeh_dev_check_failure(x) (0)
345
eeh_addr_cache_init(void)346 static inline void eeh_addr_cache_init(void) { }
347
eeh_probe_device(struct pci_dev * dev)348 static inline void eeh_probe_device(struct pci_dev *dev) { }
349
eeh_remove_device(struct pci_dev * dev)350 static inline void eeh_remove_device(struct pci_dev *dev) { }
351
352 #define EEH_POSSIBLE_ERROR(val, type) (0)
353 #define EEH_IO_ERROR_VALUE(size) (-1UL)
eeh_phb_pe_create(struct pci_controller * phb)354 static inline int eeh_phb_pe_create(struct pci_controller *phb) { return 0; }
355 #endif /* CONFIG_EEH */
356
357 #if defined(CONFIG_PPC_PSERIES) && defined(CONFIG_EEH)
358 void pseries_eeh_init_edev_recursive(struct pci_dn *pdn);
359 #endif
360
361 #ifdef CONFIG_PPC64
362 /*
363 * MMIO read/write operations with EEH support.
364 */
eeh_readb(const volatile void __iomem * addr)365 static inline u8 eeh_readb(const volatile void __iomem *addr)
366 {
367 u8 val = in_8(addr);
368 if (EEH_POSSIBLE_ERROR(val, u8))
369 eeh_check_failure(addr);
370 return val;
371 }
372
eeh_readw(const volatile void __iomem * addr)373 static inline u16 eeh_readw(const volatile void __iomem *addr)
374 {
375 u16 val = in_le16(addr);
376 if (EEH_POSSIBLE_ERROR(val, u16))
377 eeh_check_failure(addr);
378 return val;
379 }
380
eeh_readl(const volatile void __iomem * addr)381 static inline u32 eeh_readl(const volatile void __iomem *addr)
382 {
383 u32 val = in_le32(addr);
384 if (EEH_POSSIBLE_ERROR(val, u32))
385 eeh_check_failure(addr);
386 return val;
387 }
388
eeh_readq(const volatile void __iomem * addr)389 static inline u64 eeh_readq(const volatile void __iomem *addr)
390 {
391 u64 val = in_le64(addr);
392 if (EEH_POSSIBLE_ERROR(val, u64))
393 eeh_check_failure(addr);
394 return val;
395 }
396
eeh_readw_be(const volatile void __iomem * addr)397 static inline u16 eeh_readw_be(const volatile void __iomem *addr)
398 {
399 u16 val = in_be16(addr);
400 if (EEH_POSSIBLE_ERROR(val, u16))
401 eeh_check_failure(addr);
402 return val;
403 }
404
eeh_readl_be(const volatile void __iomem * addr)405 static inline u32 eeh_readl_be(const volatile void __iomem *addr)
406 {
407 u32 val = in_be32(addr);
408 if (EEH_POSSIBLE_ERROR(val, u32))
409 eeh_check_failure(addr);
410 return val;
411 }
412
eeh_readq_be(const volatile void __iomem * addr)413 static inline u64 eeh_readq_be(const volatile void __iomem *addr)
414 {
415 u64 val = in_be64(addr);
416 if (EEH_POSSIBLE_ERROR(val, u64))
417 eeh_check_failure(addr);
418 return val;
419 }
420
eeh_memcpy_fromio(void * dest,const volatile void __iomem * src,unsigned long n)421 static inline void eeh_memcpy_fromio(void *dest, const
422 volatile void __iomem *src,
423 unsigned long n)
424 {
425 _memcpy_fromio(dest, src, n);
426
427 /* Look for ffff's here at dest[n]. Assume that at least 4 bytes
428 * were copied. Check all four bytes.
429 */
430 if (n >= 4 && EEH_POSSIBLE_ERROR(*((u32 *)(dest + n - 4)), u32))
431 eeh_check_failure(src);
432 }
433
434 /* in-string eeh macros */
eeh_readsb(const volatile void __iomem * addr,void * buf,int ns)435 static inline void eeh_readsb(const volatile void __iomem *addr, void * buf,
436 int ns)
437 {
438 _insb(addr, buf, ns);
439 if (EEH_POSSIBLE_ERROR((*(((u8*)buf)+ns-1)), u8))
440 eeh_check_failure(addr);
441 }
442
eeh_readsw(const volatile void __iomem * addr,void * buf,int ns)443 static inline void eeh_readsw(const volatile void __iomem *addr, void * buf,
444 int ns)
445 {
446 _insw(addr, buf, ns);
447 if (EEH_POSSIBLE_ERROR((*(((u16*)buf)+ns-1)), u16))
448 eeh_check_failure(addr);
449 }
450
eeh_readsl(const volatile void __iomem * addr,void * buf,int nl)451 static inline void eeh_readsl(const volatile void __iomem *addr, void * buf,
452 int nl)
453 {
454 _insl(addr, buf, nl);
455 if (EEH_POSSIBLE_ERROR((*(((u32*)buf)+nl-1)), u32))
456 eeh_check_failure(addr);
457 }
458
459
460 void __init eeh_cache_debugfs_init(void);
461
462 #endif /* CONFIG_PPC64 */
463 #endif /* __KERNEL__ */
464 #endif /* _POWERPC_EEH_H */
465