drivers/edac/edac_device.h

0001 /*
0002  * Defines, structures, APIs for edac_device
0003  *
0004  * (C) 2007 Linux Networx (http://lnxi.com)
0005  * This file may be distributed under the terms of the
0006  * GNU General Public License.
0007  *
0008  * Written by Thayne Harbaugh
0009  * Based on work by Dan Hollis <goemon at anime dot net> and others.
0010  *  http://www.anime.net/~goemon/linux-ecc/
0011  *
0012  * NMI handling support added by
0013  *     Dave Peterson <dsp@llnl.gov> <dave_peterson@pobox.com>
0014  *
0015  * Refactored for multi-source files:
0016  *  Doug Thompson <norsk5@xmission.com>
0017  *
0018  * Please look at Documentation/driver-api/edac.rst for more info about
0019  * EDAC core structs and functions.
0020  */
0021
0022 #ifndef _EDAC_DEVICE_H_
0023 #define _EDAC_DEVICE_H_
0024
0025 #include <linux/completion.h>
0026 #include <linux/device.h>
0027 #include <linux/edac.h>
0028 #include <linux/kobject.h>
0029 #include <linux/list.h>
0030 #include <linux/types.h>
0031 #include <linux/sysfs.h>
0032 #include <linux/workqueue.h>
0033
0034
0035 /*
0036  * The following are the structures to provide for a generic
0037  * or abstract 'edac_device'. This set of structures and the
0038  * code that implements the APIs for the same, provide for
0039  * registering EDAC type devices which are NOT standard memory.
0040  *
0041  * CPU caches (L1 and L2)
0042  * DMA engines
0043  * Core CPU switches
0044  * Fabric switch units
0045  * PCIe interface controllers
0046  * other EDAC/ECC type devices that can be monitored for
0047  * errors, etc.
0048  *
0049  * It allows for a 2 level set of hierarchy. For example:
0050  *
0051  * cache could be composed of L1, L2 and L3 levels of cache.
0052  * Each CPU core would have its own L1 cache, while sharing
0053  * L2 and maybe L3 caches.
0054  *
0055  * View them arranged, via the sysfs presentation:
0056  * /sys/devices/system/edac/..
0057  *
0058  *  mc/     <existing memory device directory>
0059  *  cpu/cpu0/.. <L1 and L2 block directory>
0060  *      /L1-cache/ce_count
0061  *           /ue_count
0062  *      /L2-cache/ce_count
0063  *           /ue_count
0064  *  cpu/cpu1/.. <L1 and L2 block directory>
0065  *      /L1-cache/ce_count
0066  *           /ue_count
0067  *      /L2-cache/ce_count
0068  *           /ue_count
0069  *  ...
0070  *
0071  *  the L1 and L2 directories would be "edac_device_block's"
0072  */
0073
0074 struct edac_device_counter {
0075     u32 ue_count;
0076     u32 ce_count;
0077 };
0078
0079 /* forward reference */
0080 struct edac_device_ctl_info;
0081 struct edac_device_block;
0082
0083 /* edac_dev_sysfs_attribute structure
0084  *  used for driver sysfs attributes in mem_ctl_info
0085  *  for extra controls and attributes:
0086  *      like high level error Injection controls
0087  */
0088 struct edac_dev_sysfs_attribute {
0089     struct attribute attr;
0090     ssize_t (*show)(struct edac_device_ctl_info *, char *);
0091     ssize_t (*store)(struct edac_device_ctl_info *, const char *, size_t);
0092 };
0093
0094 /* edac_dev_sysfs_block_attribute structure
0095  *
0096  *  used in leaf 'block' nodes for adding controls/attributes
0097  *
0098  *  each block in each instance of the containing control structure
0099  *  can have an array of the following. The show and store functions
0100  *  will be filled in with the show/store function in the
0101  *  low level driver.
0102  *
0103  *  The 'value' field will be the actual value field used for
0104  *  counting
0105  */
0106 struct edac_dev_sysfs_block_attribute {
0107     struct attribute attr;
0108     ssize_t (*show)(struct kobject *, struct attribute *, char *);
0109     ssize_t (*store)(struct kobject *, struct attribute *,
0110             const char *, size_t);
0111     struct edac_device_block *block;
0112
0113     unsigned int value;
0114 };
0115
0116 /* device block control structure */
0117 struct edac_device_block {
0118     struct edac_device_instance *instance;  /* Up Pointer */
0119     char name[EDAC_DEVICE_NAME_LEN + 1];
0120
0121     struct edac_device_counter counters;    /* basic UE and CE counters */
0122
0123     int nr_attribs;     /* how many attributes */
0124
0125     /* this block's attributes, could be NULL */
0126     struct edac_dev_sysfs_block_attribute *block_attributes;
0127
0128     /* edac sysfs device control */
0129     struct kobject kobj;
0130 };
0131
0132 /* device instance control structure */
0133 struct edac_device_instance {
0134     struct edac_device_ctl_info *ctl;   /* Up pointer */
0135     char name[EDAC_DEVICE_NAME_LEN + 4];
0136
0137     struct edac_device_counter counters;    /* instance counters */
0138
0139     u32 nr_blocks;      /* how many blocks */
0140     struct edac_device_block *blocks;   /* block array */
0141
0142     /* edac sysfs device control */
0143     struct kobject kobj;
0144 };
0145
0146
0147 /*
0148  * Abstract edac_device control info structure
0149  *
0150  */
0151 struct edac_device_ctl_info {
0152     /* for global list of edac_device_ctl_info structs */
0153     struct list_head link;
0154
0155     struct module *owner;   /* Module owner of this control struct */
0156
0157     int dev_idx;
0158
0159     /* Per instance controls for this edac_device */
0160     int log_ue;     /* boolean for logging UEs */
0161     int log_ce;     /* boolean for logging CEs */
0162     int panic_on_ue;    /* boolean for panic'ing on an UE */
0163     unsigned poll_msec; /* number of milliseconds to poll interval */
0164     unsigned long delay;    /* number of jiffies for poll_msec */
0165
0166     /* Additional top controller level attributes, but specified
0167      * by the low level driver.
0168      *
0169      * Set by the low level driver to provide attributes at the
0170      * controller level, same level as 'ue_count' and 'ce_count' above.
0171      * An array of structures, NULL terminated
0172      *
0173      * If attributes are desired, then set to array of attributes
0174      * If no attributes are desired, leave NULL
0175      */
0176     struct edac_dev_sysfs_attribute *sysfs_attributes;
0177
0178     /* pointer to main 'edac' subsys in sysfs */
0179     struct bus_type *edac_subsys;
0180
0181     /* the internal state of this controller instance */
0182     int op_state;
0183     /* work struct for this instance */
0184     struct delayed_work work;
0185
0186     /* pointer to edac polling checking routine:
0187      *      If NOT NULL: points to polling check routine
0188      *      If NULL: Then assumes INTERRUPT operation, where
0189      *              MC driver will receive events
0190      */
0191     void (*edac_check) (struct edac_device_ctl_info * edac_dev);
0192
0193     struct device *dev; /* pointer to device structure */
0194
0195     const char *mod_name;   /* module name */
0196     const char *ctl_name;   /* edac controller  name */
0197     const char *dev_name;   /* pci/platform/etc... name */
0198
0199     void *pvt_info;     /* pointer to 'private driver' info */
0200
0201     unsigned long start_time;   /* edac_device load start time (jiffies) */
0202
0203     struct completion removal_complete;
0204
0205     /* sysfs top name under 'edac' directory
0206      * and instance name:
0207      *      cpu/cpu0/...
0208      *      cpu/cpu1/...
0209      *      cpu/cpu2/...
0210      *      ...
0211      */
0212     char name[EDAC_DEVICE_NAME_LEN + 1];
0213
0214     /* Number of instances supported on this control structure
0215      * and the array of those instances
0216      */
0217     u32 nr_instances;
0218     struct edac_device_instance *instances;
0219     struct edac_device_block *blocks;
0220     struct edac_dev_sysfs_block_attribute *attribs;
0221
0222     /* Event counters for the this whole EDAC Device */
0223     struct edac_device_counter counters;
0224
0225     /* edac sysfs device control for the 'name'
0226      * device this structure controls
0227      */
0228     struct kobject kobj;
0229 };
0230
0231 /* To get from the instance's wq to the beginning of the ctl structure */
0232 #define to_edac_mem_ctl_work(w) \
0233         container_of(w, struct mem_ctl_info, work)
0234
0235 #define to_edac_device_ctl_work(w) \
0236         container_of(w,struct edac_device_ctl_info,work)
0237
0238 /*
0239  * The alloc() and free() functions for the 'edac_device' control info
0240  * structure. A MC driver will allocate one of these for each edac_device
0241  * it is going to control/register with the EDAC CORE.
0242  */
0243 extern struct edac_device_ctl_info *edac_device_alloc_ctl_info(
0244         unsigned sizeof_private,
0245         char *edac_device_name, unsigned nr_instances,
0246         char *edac_block_name, unsigned nr_blocks,
0247         unsigned offset_value,
0248         struct edac_dev_sysfs_block_attribute *block_attributes,
0249         unsigned nr_attribs,
0250         int device_index);
0251
0252 /* The offset value can be:
0253  *  -1 indicating no offset value
0254  *  0 for zero-based block numbers
0255  *  1 for 1-based block number
0256  *  other for other-based block number
0257  */
0258 #define BLOCK_OFFSET_VALUE_OFF  ((unsigned) -1)
0259
0260 extern void edac_device_free_ctl_info(struct edac_device_ctl_info *ctl_info);
0261
0262 /**
0263  * edac_device_add_device - Insert the 'edac_dev' structure into the
0264  *   edac_device global list and create sysfs entries associated with
0265  *   edac_device structure.
0266  *
0267  * @edac_dev: pointer to edac_device structure to be added to the list
0268  *  'edac_device' structure.
0269  *
0270  * Returns:
0271  *  0 on Success, or an error code on failure
0272  */
0273 extern int edac_device_add_device(struct edac_device_ctl_info *edac_dev);
0274
0275 /**
0276  * edac_device_del_device - Remove sysfs entries for specified edac_device
0277  *  structure and then remove edac_device structure from global list
0278  *
0279  * @dev:
0280  *  Pointer to struct &device representing the edac device
0281  *  structure to remove.
0282  *
0283  * Returns:
0284  *  Pointer to removed edac_device structure,
0285  *  or %NULL if device not found.
0286  */
0287 extern struct edac_device_ctl_info *edac_device_del_device(struct device *dev);
0288
0289 /**
0290  * edac_device_handle_ce_count - Log correctable errors.
0291  *
0292  * @edac_dev: pointer to struct &edac_device_ctl_info
0293  * @inst_nr: number of the instance where the CE error happened
0294  * @count: Number of errors to log.
0295  * @block_nr: number of the block where the CE error happened
0296  * @msg: message to be printed
0297  */
0298 void edac_device_handle_ce_count(struct edac_device_ctl_info *edac_dev,
0299                  unsigned int count, int inst_nr, int block_nr,
0300                  const char *msg);
0301
0302 /**
0303  * edac_device_handle_ue_count - Log uncorrectable errors.
0304  *
0305  * @edac_dev: pointer to struct &edac_device_ctl_info
0306  * @inst_nr: number of the instance where the CE error happened
0307  * @count: Number of errors to log.
0308  * @block_nr: number of the block where the CE error happened
0309  * @msg: message to be printed
0310  */
0311 void edac_device_handle_ue_count(struct edac_device_ctl_info *edac_dev,
0312                  unsigned int count, int inst_nr, int block_nr,
0313                  const char *msg);
0314
0315 /**
0316  * edac_device_handle_ce(): Log a single correctable error
0317  *
0318  * @edac_dev: pointer to struct &edac_device_ctl_info
0319  * @inst_nr: number of the instance where the CE error happened
0320  * @block_nr: number of the block where the CE error happened
0321  * @msg: message to be printed
0322  */
0323 static inline void
0324 edac_device_handle_ce(struct edac_device_ctl_info *edac_dev, int inst_nr,
0325               int block_nr, const char *msg)
0326 {
0327     edac_device_handle_ce_count(edac_dev, 1, inst_nr, block_nr, msg);
0328 }
0329
0330 /**
0331  * edac_device_handle_ue(): Log a single uncorrectable error
0332  *
0333  * @edac_dev: pointer to struct &edac_device_ctl_info
0334  * @inst_nr: number of the instance where the UE error happened
0335  * @block_nr: number of the block where the UE error happened
0336  * @msg: message to be printed
0337  */
0338 static inline void
0339 edac_device_handle_ue(struct edac_device_ctl_info *edac_dev, int inst_nr,
0340               int block_nr, const char *msg)
0341 {
0342     edac_device_handle_ue_count(edac_dev, 1, inst_nr, block_nr, msg);
0343 }
0344
0345 /**
0346  * edac_device_alloc_index: Allocate a unique device index number
0347  *
0348  * Returns:
0349  *  allocated index number
0350  */
0351 extern int edac_device_alloc_index(void);
0352 extern const char *edac_layer_name[];
0353
0354 /* Free the actual struct */
0355 static inline void __edac_device_free_ctl_info(struct edac_device_ctl_info *ci)
0356 {
0357     if (ci) {
0358         kfree(ci->pvt_info);
0359         kfree(ci->attribs);
0360         kfree(ci->blocks);
0361         kfree(ci->instances);
0362         kfree(ci);
0363     }
0364 }
0365 #endif