1 /* SPDX-License-Identifier: GPL-2.0-or-later */
3 * Copyright (C) 2001 Dave Engebretsen & Todd Inglett IBM Corporation.
4 * Copyright 2001-2012 IBM Corporation.
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>
17 #include <uapi/asm/eeh.h>
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_VALID_PE_ZERO 0x10 /* PE#0 is valid */
31 #define EEH_ENABLE_IO_FOR_LOG 0x20 /* Enable IO for log */
32 #define EEH_EARLY_DUMP_LOG 0x40 /* Dump log immediately */
35 * Delay for PE reset, all in ms
37 * PCI specification has reset hold time of 100 milliseconds.
38 * We have 250 milliseconds here. The PCI bus settlement time
39 * is specified as 1.5 seconds and we have 1.8 seconds.
41 #define EEH_PE_RST_HOLD_TIME 250
42 #define EEH_PE_RST_SETTLE_TIME 1800
45 * The struct is used to trace PE related EEH functionality.
46 * In theory, there will have one instance of the struct to
47 * be created against particular PE. In nature, PEs correlate
48 * to each other. the struct has to reflect that hierarchy in
49 * order to easily pick up those affected PEs when one particular
52 * Also, one particular PE might be composed of PCI device, PCI
53 * bus and its subordinate components. The struct also need ship
54 * the information. Further more, one particular PE is only meaingful
55 * in the corresponding PHB. Therefore, the root PEs should be created
56 * against existing PHBs in on-to-one fashion.
58 #define EEH_PE_INVALID (1 << 0) /* Invalid */
59 #define EEH_PE_PHB (1 << 1) /* PHB PE */
60 #define EEH_PE_DEVICE (1 << 2) /* Device PE */
61 #define EEH_PE_BUS (1 << 3) /* Bus PE */
62 #define EEH_PE_VF (1 << 4) /* VF PE */
64 #define EEH_PE_ISOLATED (1 << 0) /* Isolated PE */
65 #define EEH_PE_RECOVERING (1 << 1) /* Recovering PE */
66 #define EEH_PE_CFG_BLOCKED (1 << 2) /* Block config access */
67 #define EEH_PE_RESET (1 << 3) /* PE reset in progress */
69 #define EEH_PE_KEEP (1 << 8) /* Keep PE on hotplug */
70 #define EEH_PE_CFG_RESTRICTED (1 << 9) /* Block config on error */
71 #define EEH_PE_REMOVED (1 << 10) /* Removed permanently */
72 #define EEH_PE_PRI_BUS (1 << 11) /* Cached primary bus */
75 int type
; /* PE type: PHB/Bus/Device */
76 int state
; /* PE EEH dependent mode */
77 int config_addr
; /* Traditional PCI address */
78 int addr
; /* PE configuration address */
79 struct pci_controller
*phb
; /* Associated PHB */
80 struct pci_bus
*bus
; /* Top PCI bus for bus PE */
81 int check_count
; /* Times of ignored error */
82 int freeze_count
; /* Times of froze up */
83 time64_t tstamp
; /* Time on first-time freeze */
84 int false_positives
; /* Times of reported #ff's */
85 atomic_t pass_dev_cnt
; /* Count of passed through devs */
86 struct eeh_pe
*parent
; /* Parent PE */
87 void *data
; /* PE auxillary data */
88 struct list_head child_list
; /* List of PEs below this PE */
89 struct list_head child
; /* Memb. child_list/eeh_phb_pe */
90 struct list_head edevs
; /* List of eeh_dev in this PE */
92 #ifdef CONFIG_STACKTRACE
94 * Saved stack trace. When we find a PE freeze in eeh_dev_check_failure
95 * the stack trace is saved here so we can print it in the recovery
96 * thread if it turns out to due to a real problem rather than
99 * A max of 64 entries might be overkill, but it also might not be.
101 unsigned long stack_trace
[64];
103 #endif /* CONFIG_STACKTRACE */
106 #define eeh_pe_for_each_dev(pe, edev, tmp) \
107 list_for_each_entry_safe(edev, tmp, &pe->edevs, entry)
109 #define eeh_for_each_pe(root, pe) \
110 for (pe = root; pe; pe = eeh_pe_next(pe, root))
112 static inline bool eeh_pe_passed(struct eeh_pe
*pe
)
114 return pe
? !!atomic_read(&pe
->pass_dev_cnt
) : false;
118 * The struct is used to trace EEH state for the associated
119 * PCI device node or PCI device. In future, it might
120 * represent PE as well so that the EEH device to form
121 * another tree except the currently existing tree of PCI
122 * buses and PCI devices
124 #define EEH_DEV_BRIDGE (1 << 0) /* PCI bridge */
125 #define EEH_DEV_ROOT_PORT (1 << 1) /* PCIe root port */
126 #define EEH_DEV_DS_PORT (1 << 2) /* Downstream port */
127 #define EEH_DEV_IRQ_DISABLED (1 << 3) /* Interrupt disabled */
128 #define EEH_DEV_DISCONNECTED (1 << 4) /* Removing from PE */
130 #define EEH_DEV_NO_HANDLER (1 << 8) /* No error handler */
131 #define EEH_DEV_SYSFS (1 << 9) /* Sysfs created */
132 #define EEH_DEV_REMOVED (1 << 10) /* Removed permanently */
135 int mode
; /* EEH mode */
136 int class_code
; /* Class code of the device */
137 int bdfn
; /* bdfn of device (for cfg ops) */
138 struct pci_controller
*controller
;
139 int pe_config_addr
; /* PE config address */
140 u32 config_space
[16]; /* Saved PCI config space */
141 int pcix_cap
; /* Saved PCIx capability */
142 int pcie_cap
; /* Saved PCIe capability */
143 int aer_cap
; /* Saved AER capability */
144 int af_cap
; /* Saved AF capability */
145 struct eeh_pe
*pe
; /* Associated PE */
146 struct list_head entry
; /* Membership in eeh_pe.edevs */
147 struct list_head rmv_entry
; /* Membership in rmv_list */
148 struct pci_dn
*pdn
; /* Associated PCI device node */
149 struct pci_dev
*pdev
; /* Associated PCI device */
150 bool in_error
; /* Error flag for edev */
151 struct pci_dev
*physfn
; /* Associated SRIOV PF */
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__)
165 static inline struct pci_dn
*eeh_dev_to_pdn(struct eeh_dev
*edev
)
167 return edev
? edev
->pdn
: NULL
;
170 static inline struct pci_dev
*eeh_dev_to_pci_dev(struct eeh_dev
*edev
)
172 return edev
? edev
->pdev
: NULL
;
175 static inline struct eeh_pe
*eeh_dev_to_pe(struct eeh_dev
* edev
)
177 return edev
? edev
->pe
: NULL
;
180 /* Return values from eeh_ops::next_error */
182 EEH_NEXT_ERR_NONE
= 0,
184 EEH_NEXT_ERR_FROZEN_PE
,
185 EEH_NEXT_ERR_FENCED_PHB
,
186 EEH_NEXT_ERR_DEAD_PHB
,
187 EEH_NEXT_ERR_DEAD_IOC
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.
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 */
218 void* (*probe
)(struct pci_dn
*pdn
, void *data
);
219 int (*set_option
)(struct eeh_pe
*pe
, int option
);
220 int (*get_pe_addr
)(struct eeh_pe
*pe
);
221 int (*get_state
)(struct eeh_pe
*pe
, int *delay
);
222 int (*reset
)(struct eeh_pe
*pe
, int option
);
223 int (*get_log
)(struct eeh_pe
*pe
, int severity
, char *drv_log
, unsigned long len
);
224 int (*configure_bridge
)(struct eeh_pe
*pe
);
225 int (*err_inject
)(struct eeh_pe
*pe
, int type
, int func
,
226 unsigned long addr
, unsigned long mask
);
227 int (*read_config
)(struct pci_dn
*pdn
, int where
, int size
, u32
*val
);
228 int (*write_config
)(struct pci_dn
*pdn
, int where
, int size
, u32 val
);
229 int (*next_error
)(struct eeh_pe
**pe
);
230 int (*restore_config
)(struct pci_dn
*pdn
);
231 int (*notify_resume
)(struct pci_dn
*pdn
);
234 extern int eeh_subsystem_flags
;
235 extern u32 eeh_max_freezes
;
236 extern bool eeh_debugfs_no_recover
;
237 extern struct eeh_ops
*eeh_ops
;
238 extern raw_spinlock_t confirm_error_lock
;
240 static inline void eeh_add_flag(int flag
)
242 eeh_subsystem_flags
|= flag
;
245 static inline void eeh_clear_flag(int flag
)
247 eeh_subsystem_flags
&= ~flag
;
250 static inline bool eeh_has_flag(int flag
)
252 return !!(eeh_subsystem_flags
& flag
);
255 static inline bool eeh_enabled(void)
257 return eeh_has_flag(EEH_ENABLED
) && !eeh_has_flag(EEH_FORCE_DISABLED
);
260 static inline void eeh_serialize_lock(unsigned long *flags
)
262 raw_spin_lock_irqsave(&confirm_error_lock
, *flags
);
265 static inline void eeh_serialize_unlock(unsigned long flags
)
267 raw_spin_unlock_irqrestore(&confirm_error_lock
, flags
);
270 static inline bool eeh_state_active(int state
)
272 return (state
& (EEH_STATE_MMIO_ACTIVE
| EEH_STATE_DMA_ACTIVE
))
273 == (EEH_STATE_MMIO_ACTIVE
| EEH_STATE_DMA_ACTIVE
);
276 typedef void (*eeh_edev_traverse_func
)(struct eeh_dev
*edev
, void *flag
);
277 typedef void *(*eeh_pe_traverse_func
)(struct eeh_pe
*pe
, void *flag
);
278 void eeh_set_pe_aux_size(int size
);
279 int eeh_phb_pe_create(struct pci_controller
*phb
);
280 int eeh_wait_state(struct eeh_pe
*pe
, int max_wait
);
281 struct eeh_pe
*eeh_phb_pe_get(struct pci_controller
*phb
);
282 struct eeh_pe
*eeh_pe_next(struct eeh_pe
*pe
, struct eeh_pe
*root
);
283 struct eeh_pe
*eeh_pe_get(struct pci_controller
*phb
,
284 int pe_no
, int config_addr
);
285 int eeh_add_to_parent_pe(struct eeh_dev
*edev
);
286 int eeh_rmv_from_parent_pe(struct eeh_dev
*edev
);
287 void eeh_pe_update_time_stamp(struct eeh_pe
*pe
);
288 void *eeh_pe_traverse(struct eeh_pe
*root
,
289 eeh_pe_traverse_func fn
, void *flag
);
290 void eeh_pe_dev_traverse(struct eeh_pe
*root
,
291 eeh_edev_traverse_func fn
, void *flag
);
292 void eeh_pe_restore_bars(struct eeh_pe
*pe
);
293 const char *eeh_pe_loc_get(struct eeh_pe
*pe
);
294 struct pci_bus
*eeh_pe_bus_get(struct eeh_pe
*pe
);
296 struct eeh_dev
*eeh_dev_init(struct pci_dn
*pdn
);
297 void eeh_dev_phb_init_dynamic(struct pci_controller
*phb
);
298 void eeh_show_enabled(void);
299 int __init
eeh_ops_register(struct eeh_ops
*ops
);
300 int __exit
eeh_ops_unregister(const char *name
);
301 int eeh_check_failure(const volatile void __iomem
*token
);
302 int eeh_dev_check_failure(struct eeh_dev
*edev
);
303 void eeh_addr_cache_init(void);
304 void eeh_add_device_early(struct pci_dn
*);
305 void eeh_add_device_tree_early(struct pci_dn
*);
306 void eeh_add_device_late(struct pci_dev
*);
307 void eeh_add_device_tree_late(struct pci_bus
*);
308 void eeh_add_sysfs_files(struct pci_bus
*);
309 void eeh_remove_device(struct pci_dev
*);
310 int eeh_unfreeze_pe(struct eeh_pe
*pe
);
311 int eeh_pe_reset_and_recover(struct eeh_pe
*pe
);
312 int eeh_dev_open(struct pci_dev
*pdev
);
313 void eeh_dev_release(struct pci_dev
*pdev
);
314 struct eeh_pe
*eeh_iommu_group_to_pe(struct iommu_group
*group
);
315 int eeh_pe_set_option(struct eeh_pe
*pe
, int option
);
316 int eeh_pe_get_state(struct eeh_pe
*pe
);
317 int eeh_pe_reset(struct eeh_pe
*pe
, int option
, bool include_passed
);
318 int eeh_pe_configure(struct eeh_pe
*pe
);
319 int eeh_pe_inject_err(struct eeh_pe
*pe
, int type
, int func
,
320 unsigned long addr
, unsigned long mask
);
321 int eeh_restore_vf_config(struct pci_dn
*pdn
);
324 * EEH_POSSIBLE_ERROR() -- test for possible MMIO failure.
326 * If this macro yields TRUE, the caller relays to eeh_check_failure()
327 * which does further tests out of line.
329 #define EEH_POSSIBLE_ERROR(val, type) ((val) == (type)~0 && eeh_enabled())
332 * Reads from a device which has been isolated by EEH will return
333 * all 1s. This macro gives an all-1s value of the given size (in
334 * bytes: 1, 2, or 4) for comparing with the result of a read.
336 #define EEH_IO_ERROR_VALUE(size) (~0U >> ((4 - (size)) * 8))
338 #else /* !CONFIG_EEH */
340 static inline bool eeh_enabled(void)
345 static inline void eeh_show_enabled(void) { }
347 static inline void *eeh_dev_init(struct pci_dn
*pdn
, void *data
)
352 static inline void eeh_dev_phb_init_dynamic(struct pci_controller
*phb
) { }
354 static inline int eeh_check_failure(const volatile void __iomem
*token
)
359 #define eeh_dev_check_failure(x) (0)
361 static inline void eeh_addr_cache_init(void) { }
363 static inline void eeh_add_device_early(struct pci_dn
*pdn
) { }
365 static inline void eeh_add_device_tree_early(struct pci_dn
*pdn
) { }
367 static inline void eeh_add_device_late(struct pci_dev
*dev
) { }
369 static inline void eeh_add_device_tree_late(struct pci_bus
*bus
) { }
371 static inline void eeh_add_sysfs_files(struct pci_bus
*bus
) { }
373 static inline void eeh_remove_device(struct pci_dev
*dev
) { }
375 #define EEH_POSSIBLE_ERROR(val, type) (0)
376 #define EEH_IO_ERROR_VALUE(size) (-1UL)
377 #endif /* CONFIG_EEH */
381 * MMIO read/write operations with EEH support.
383 static inline u8
eeh_readb(const volatile void __iomem
*addr
)
386 if (EEH_POSSIBLE_ERROR(val
, u8
))
387 eeh_check_failure(addr
);
391 static inline u16
eeh_readw(const volatile void __iomem
*addr
)
393 u16 val
= in_le16(addr
);
394 if (EEH_POSSIBLE_ERROR(val
, u16
))
395 eeh_check_failure(addr
);
399 static inline u32
eeh_readl(const volatile void __iomem
*addr
)
401 u32 val
= in_le32(addr
);
402 if (EEH_POSSIBLE_ERROR(val
, u32
))
403 eeh_check_failure(addr
);
407 static inline u64
eeh_readq(const volatile void __iomem
*addr
)
409 u64 val
= in_le64(addr
);
410 if (EEH_POSSIBLE_ERROR(val
, u64
))
411 eeh_check_failure(addr
);
415 static inline u16
eeh_readw_be(const volatile void __iomem
*addr
)
417 u16 val
= in_be16(addr
);
418 if (EEH_POSSIBLE_ERROR(val
, u16
))
419 eeh_check_failure(addr
);
423 static inline u32
eeh_readl_be(const volatile void __iomem
*addr
)
425 u32 val
= in_be32(addr
);
426 if (EEH_POSSIBLE_ERROR(val
, u32
))
427 eeh_check_failure(addr
);
431 static inline u64
eeh_readq_be(const volatile void __iomem
*addr
)
433 u64 val
= in_be64(addr
);
434 if (EEH_POSSIBLE_ERROR(val
, u64
))
435 eeh_check_failure(addr
);
439 static inline void eeh_memcpy_fromio(void *dest
, const
440 volatile void __iomem
*src
,
443 _memcpy_fromio(dest
, src
, n
);
445 /* Look for ffff's here at dest[n]. Assume that at least 4 bytes
446 * were copied. Check all four bytes.
448 if (n
>= 4 && EEH_POSSIBLE_ERROR(*((u32
*)(dest
+ n
- 4)), u32
))
449 eeh_check_failure(src
);
452 /* in-string eeh macros */
453 static inline void eeh_readsb(const volatile void __iomem
*addr
, void * buf
,
456 _insb(addr
, buf
, ns
);
457 if (EEH_POSSIBLE_ERROR((*(((u8
*)buf
)+ns
-1)), u8
))
458 eeh_check_failure(addr
);
461 static inline void eeh_readsw(const volatile void __iomem
*addr
, void * buf
,
464 _insw(addr
, buf
, ns
);
465 if (EEH_POSSIBLE_ERROR((*(((u16
*)buf
)+ns
-1)), u16
))
466 eeh_check_failure(addr
);
469 static inline void eeh_readsl(const volatile void __iomem
*addr
, void * buf
,
472 _insl(addr
, buf
, nl
);
473 if (EEH_POSSIBLE_ERROR((*(((u32
*)buf
)+nl
-1)), u32
))
474 eeh_check_failure(addr
);
478 void eeh_cache_debugfs_init(void);
480 #endif /* CONFIG_PPC64 */
481 #endif /* __KERNEL__ */
482 #endif /* _POWERPC_EEH_H */