Commit | Line | Data |
---|---|---|
da9bb1d2 AC |
1 | /* |
2 | * edac_mc kernel module | |
49c0dab7 | 3 | * (C) 2005, 2006 Linux Networx (http://lnxi.com) |
da9bb1d2 AC |
4 | * This file may be distributed under the terms of the |
5 | * GNU General Public License. | |
6 | * | |
7 | * Written by Thayne Harbaugh | |
8 | * Based on work by Dan Hollis <goemon at anime dot net> and others. | |
9 | * http://www.anime.net/~goemon/linux-ecc/ | |
10 | * | |
11 | * Modified by Dave Peterson and Doug Thompson | |
12 | * | |
13 | */ | |
14 | ||
da9bb1d2 AC |
15 | #include <linux/module.h> |
16 | #include <linux/proc_fs.h> | |
17 | #include <linux/kernel.h> | |
18 | #include <linux/types.h> | |
19 | #include <linux/smp.h> | |
20 | #include <linux/init.h> | |
21 | #include <linux/sysctl.h> | |
22 | #include <linux/highmem.h> | |
23 | #include <linux/timer.h> | |
24 | #include <linux/slab.h> | |
25 | #include <linux/jiffies.h> | |
26 | #include <linux/spinlock.h> | |
27 | #include <linux/list.h> | |
da9bb1d2 | 28 | #include <linux/ctype.h> |
c0d12172 | 29 | #include <linux/edac.h> |
53f2d028 | 30 | #include <linux/bitops.h> |
da9bb1d2 AC |
31 | #include <asm/uaccess.h> |
32 | #include <asm/page.h> | |
20bcb7a8 | 33 | #include "edac_core.h" |
7c9281d7 | 34 | #include "edac_module.h" |
53f2d028 MCC |
35 | #include <ras/ras_event.h> |
36 | ||
b01aec9b BP |
37 | #ifdef CONFIG_EDAC_ATOMIC_SCRUB |
38 | #include <asm/edac.h> | |
39 | #else | |
40 | #define edac_atomic_scrub(va, size) do { } while (0) | |
41 | #endif | |
42 | ||
da9bb1d2 | 43 | /* lock to memory controller's control array */ |
63b7df91 | 44 | static DEFINE_MUTEX(mem_ctls_mutex); |
ff6ac2a6 | 45 | static LIST_HEAD(mc_devices); |
da9bb1d2 | 46 | |
80cc7d87 MCC |
47 | /* |
48 | * Used to lock EDAC MC to just one module, avoiding two drivers e. g. | |
49 | * apei/ghes and i7core_edac to be used at the same time. | |
50 | */ | |
51 | static void const *edac_mc_owner; | |
52 | ||
88d84ac9 BP |
53 | static struct bus_type mc_bus[EDAC_MAX_MCS]; |
54 | ||
6e84d359 MCC |
55 | unsigned edac_dimm_info_location(struct dimm_info *dimm, char *buf, |
56 | unsigned len) | |
57 | { | |
58 | struct mem_ctl_info *mci = dimm->mci; | |
59 | int i, n, count = 0; | |
60 | char *p = buf; | |
61 | ||
62 | for (i = 0; i < mci->n_layers; i++) { | |
63 | n = snprintf(p, len, "%s %d ", | |
64 | edac_layer_name[mci->layers[i].type], | |
65 | dimm->location[i]); | |
66 | p += n; | |
67 | len -= n; | |
68 | count += n; | |
69 | if (!len) | |
70 | break; | |
71 | } | |
72 | ||
73 | return count; | |
74 | } | |
75 | ||
da9bb1d2 AC |
76 | #ifdef CONFIG_EDAC_DEBUG |
77 | ||
a4b4be3f | 78 | static void edac_mc_dump_channel(struct rank_info *chan) |
da9bb1d2 | 79 | { |
6e84d359 MCC |
80 | edac_dbg(4, " channel->chan_idx = %d\n", chan->chan_idx); |
81 | edac_dbg(4, " channel = %p\n", chan); | |
82 | edac_dbg(4, " channel->csrow = %p\n", chan->csrow); | |
83 | edac_dbg(4, " channel->dimm = %p\n", chan->dimm); | |
4275be63 MCC |
84 | } |
85 | ||
6e84d359 | 86 | static void edac_mc_dump_dimm(struct dimm_info *dimm, int number) |
4275be63 | 87 | { |
6e84d359 MCC |
88 | char location[80]; |
89 | ||
90 | edac_dimm_info_location(dimm, location, sizeof(location)); | |
91 | ||
92 | edac_dbg(4, "%s%i: %smapped as virtual row %d, chan %d\n", | |
9713faec | 93 | dimm->mci->csbased ? "rank" : "dimm", |
6e84d359 MCC |
94 | number, location, dimm->csrow, dimm->cschannel); |
95 | edac_dbg(4, " dimm = %p\n", dimm); | |
96 | edac_dbg(4, " dimm->label = '%s'\n", dimm->label); | |
97 | edac_dbg(4, " dimm->nr_pages = 0x%x\n", dimm->nr_pages); | |
98 | edac_dbg(4, " dimm->grain = %d\n", dimm->grain); | |
99 | edac_dbg(4, " dimm->nr_pages = 0x%x\n", dimm->nr_pages); | |
da9bb1d2 AC |
100 | } |
101 | ||
2da1c119 | 102 | static void edac_mc_dump_csrow(struct csrow_info *csrow) |
da9bb1d2 | 103 | { |
6e84d359 MCC |
104 | edac_dbg(4, "csrow->csrow_idx = %d\n", csrow->csrow_idx); |
105 | edac_dbg(4, " csrow = %p\n", csrow); | |
106 | edac_dbg(4, " csrow->first_page = 0x%lx\n", csrow->first_page); | |
107 | edac_dbg(4, " csrow->last_page = 0x%lx\n", csrow->last_page); | |
108 | edac_dbg(4, " csrow->page_mask = 0x%lx\n", csrow->page_mask); | |
109 | edac_dbg(4, " csrow->nr_channels = %d\n", csrow->nr_channels); | |
110 | edac_dbg(4, " csrow->channels = %p\n", csrow->channels); | |
111 | edac_dbg(4, " csrow->mci = %p\n", csrow->mci); | |
da9bb1d2 AC |
112 | } |
113 | ||
2da1c119 | 114 | static void edac_mc_dump_mci(struct mem_ctl_info *mci) |
da9bb1d2 | 115 | { |
956b9ba1 JP |
116 | edac_dbg(3, "\tmci = %p\n", mci); |
117 | edac_dbg(3, "\tmci->mtype_cap = %lx\n", mci->mtype_cap); | |
118 | edac_dbg(3, "\tmci->edac_ctl_cap = %lx\n", mci->edac_ctl_cap); | |
119 | edac_dbg(3, "\tmci->edac_cap = %lx\n", mci->edac_cap); | |
120 | edac_dbg(4, "\tmci->edac_check = %p\n", mci->edac_check); | |
121 | edac_dbg(3, "\tmci->nr_csrows = %d, csrows = %p\n", | |
122 | mci->nr_csrows, mci->csrows); | |
123 | edac_dbg(3, "\tmci->nr_dimms = %d, dimms = %p\n", | |
124 | mci->tot_dimms, mci->dimms); | |
125 | edac_dbg(3, "\tdev = %p\n", mci->pdev); | |
126 | edac_dbg(3, "\tmod_name:ctl_name = %s:%s\n", | |
127 | mci->mod_name, mci->ctl_name); | |
128 | edac_dbg(3, "\tpvt_info = %p\n\n", mci->pvt_info); | |
da9bb1d2 AC |
129 | } |
130 | ||
24f9a7fe BP |
131 | #endif /* CONFIG_EDAC_DEBUG */ |
132 | ||
f4ce6eca | 133 | const char * const edac_mem_types[] = { |
4cfc3a40 BP |
134 | [MEM_EMPTY] = "Empty csrow", |
135 | [MEM_RESERVED] = "Reserved csrow type", | |
136 | [MEM_UNKNOWN] = "Unknown csrow type", | |
137 | [MEM_FPM] = "Fast page mode RAM", | |
138 | [MEM_EDO] = "Extended data out RAM", | |
139 | [MEM_BEDO] = "Burst Extended data out RAM", | |
140 | [MEM_SDR] = "Single data rate SDRAM", | |
141 | [MEM_RDR] = "Registered single data rate SDRAM", | |
142 | [MEM_DDR] = "Double data rate SDRAM", | |
143 | [MEM_RDDR] = "Registered Double data rate SDRAM", | |
144 | [MEM_RMBS] = "Rambus DRAM", | |
145 | [MEM_DDR2] = "Unbuffered DDR2 RAM", | |
146 | [MEM_FB_DDR2] = "Fully buffered DDR2", | |
147 | [MEM_RDDR2] = "Registered DDR2 RAM", | |
148 | [MEM_XDR] = "Rambus XDR", | |
149 | [MEM_DDR3] = "Unbuffered DDR3 RAM", | |
150 | [MEM_RDDR3] = "Registered DDR3 RAM", | |
151 | [MEM_LRDDR3] = "Load-Reduced DDR3 RAM", | |
152 | [MEM_DDR4] = "Unbuffered DDR4 RAM", | |
153 | [MEM_RDDR4] = "Registered DDR4 RAM", | |
239642fe BP |
154 | }; |
155 | EXPORT_SYMBOL_GPL(edac_mem_types); | |
156 | ||
93e4fe64 MCC |
157 | /** |
158 | * edac_align_ptr - Prepares the pointer offsets for a single-shot allocation | |
159 | * @p: pointer to a pointer with the memory offset to be used. At | |
160 | * return, this will be incremented to point to the next offset | |
161 | * @size: Size of the data structure to be reserved | |
162 | * @n_elems: Number of elements that should be reserved | |
da9bb1d2 AC |
163 | * |
164 | * If 'size' is a constant, the compiler will optimize this whole function | |
93e4fe64 MCC |
165 | * down to either a no-op or the addition of a constant to the value of '*p'. |
166 | * | |
167 | * The 'p' pointer is absolutely needed to keep the proper advancing | |
168 | * further in memory to the proper offsets when allocating the struct along | |
169 | * with its embedded structs, as edac_device_alloc_ctl_info() does it | |
170 | * above, for example. | |
171 | * | |
172 | * At return, the pointer 'p' will be incremented to be used on a next call | |
173 | * to this function. | |
da9bb1d2 | 174 | */ |
93e4fe64 | 175 | void *edac_align_ptr(void **p, unsigned size, int n_elems) |
da9bb1d2 AC |
176 | { |
177 | unsigned align, r; | |
93e4fe64 | 178 | void *ptr = *p; |
da9bb1d2 | 179 | |
93e4fe64 MCC |
180 | *p += size * n_elems; |
181 | ||
182 | /* | |
183 | * 'p' can possibly be an unaligned item X such that sizeof(X) is | |
184 | * 'size'. Adjust 'p' so that its alignment is at least as | |
185 | * stringent as what the compiler would provide for X and return | |
186 | * the aligned result. | |
187 | * Here we assume that the alignment of a "long long" is the most | |
da9bb1d2 AC |
188 | * stringent alignment that the compiler will ever provide by default. |
189 | * As far as I know, this is a reasonable assumption. | |
190 | */ | |
191 | if (size > sizeof(long)) | |
192 | align = sizeof(long long); | |
193 | else if (size > sizeof(int)) | |
194 | align = sizeof(long); | |
195 | else if (size > sizeof(short)) | |
196 | align = sizeof(int); | |
197 | else if (size > sizeof(char)) | |
198 | align = sizeof(short); | |
199 | else | |
079708b9 | 200 | return (char *)ptr; |
da9bb1d2 | 201 | |
8447c4d1 | 202 | r = (unsigned long)p % align; |
da9bb1d2 AC |
203 | |
204 | if (r == 0) | |
079708b9 | 205 | return (char *)ptr; |
da9bb1d2 | 206 | |
93e4fe64 MCC |
207 | *p += align - r; |
208 | ||
7391c6dc | 209 | return (void *)(((unsigned long)ptr) + align - r); |
da9bb1d2 AC |
210 | } |
211 | ||
faa2ad09 SR |
212 | static void _edac_mc_free(struct mem_ctl_info *mci) |
213 | { | |
214 | int i, chn, row; | |
215 | struct csrow_info *csr; | |
216 | const unsigned int tot_dimms = mci->tot_dimms; | |
217 | const unsigned int tot_channels = mci->num_cschannel; | |
218 | const unsigned int tot_csrows = mci->nr_csrows; | |
219 | ||
220 | if (mci->dimms) { | |
221 | for (i = 0; i < tot_dimms; i++) | |
222 | kfree(mci->dimms[i]); | |
223 | kfree(mci->dimms); | |
224 | } | |
225 | if (mci->csrows) { | |
226 | for (row = 0; row < tot_csrows; row++) { | |
227 | csr = mci->csrows[row]; | |
228 | if (csr) { | |
229 | if (csr->channels) { | |
230 | for (chn = 0; chn < tot_channels; chn++) | |
231 | kfree(csr->channels[chn]); | |
232 | kfree(csr->channels); | |
233 | } | |
234 | kfree(csr); | |
235 | } | |
236 | } | |
237 | kfree(mci->csrows); | |
238 | } | |
239 | kfree(mci); | |
240 | } | |
241 | ||
da9bb1d2 | 242 | /** |
4275be63 MCC |
243 | * edac_mc_alloc: Allocate and partially fill a struct mem_ctl_info structure |
244 | * @mc_num: Memory controller number | |
245 | * @n_layers: Number of MC hierarchy layers | |
246 | * layers: Describes each layer as seen by the Memory Controller | |
247 | * @size_pvt: size of private storage needed | |
248 | * | |
da9bb1d2 AC |
249 | * |
250 | * Everything is kmalloc'ed as one big chunk - more efficient. | |
251 | * Only can be used if all structures have the same lifetime - otherwise | |
252 | * you have to allocate and initialize your own structures. | |
253 | * | |
254 | * Use edac_mc_free() to free mc structures allocated by this function. | |
255 | * | |
4275be63 MCC |
256 | * NOTE: drivers handle multi-rank memories in different ways: in some |
257 | * drivers, one multi-rank memory stick is mapped as one entry, while, in | |
258 | * others, a single multi-rank memory stick would be mapped into several | |
259 | * entries. Currently, this function will allocate multiple struct dimm_info | |
260 | * on such scenarios, as grouping the multiple ranks require drivers change. | |
261 | * | |
da9bb1d2 | 262 | * Returns: |
ca0907b9 MCC |
263 | * On failure: NULL |
264 | * On success: struct mem_ctl_info pointer | |
da9bb1d2 | 265 | */ |
ca0907b9 MCC |
266 | struct mem_ctl_info *edac_mc_alloc(unsigned mc_num, |
267 | unsigned n_layers, | |
268 | struct edac_mc_layer *layers, | |
269 | unsigned sz_pvt) | |
da9bb1d2 AC |
270 | { |
271 | struct mem_ctl_info *mci; | |
4275be63 | 272 | struct edac_mc_layer *layer; |
de3910eb MCC |
273 | struct csrow_info *csr; |
274 | struct rank_info *chan; | |
a7d7d2e1 | 275 | struct dimm_info *dimm; |
4275be63 MCC |
276 | u32 *ce_per_layer[EDAC_MAX_LAYERS], *ue_per_layer[EDAC_MAX_LAYERS]; |
277 | unsigned pos[EDAC_MAX_LAYERS]; | |
4275be63 MCC |
278 | unsigned size, tot_dimms = 1, count = 1; |
279 | unsigned tot_csrows = 1, tot_channels = 1, tot_errcount = 0; | |
5926ff50 | 280 | void *pvt, *p, *ptr = NULL; |
de3910eb | 281 | int i, j, row, chn, n, len, off; |
4275be63 MCC |
282 | bool per_rank = false; |
283 | ||
284 | BUG_ON(n_layers > EDAC_MAX_LAYERS || n_layers == 0); | |
285 | /* | |
286 | * Calculate the total amount of dimms and csrows/cschannels while | |
287 | * in the old API emulation mode | |
288 | */ | |
289 | for (i = 0; i < n_layers; i++) { | |
290 | tot_dimms *= layers[i].size; | |
291 | if (layers[i].is_virt_csrow) | |
292 | tot_csrows *= layers[i].size; | |
293 | else | |
294 | tot_channels *= layers[i].size; | |
295 | ||
296 | if (layers[i].type == EDAC_MC_LAYER_CHIP_SELECT) | |
297 | per_rank = true; | |
298 | } | |
da9bb1d2 AC |
299 | |
300 | /* Figure out the offsets of the various items from the start of an mc | |
301 | * structure. We want the alignment of each item to be at least as | |
302 | * stringent as what the compiler would provide if we could simply | |
303 | * hardcode everything into a single struct. | |
304 | */ | |
93e4fe64 | 305 | mci = edac_align_ptr(&ptr, sizeof(*mci), 1); |
4275be63 | 306 | layer = edac_align_ptr(&ptr, sizeof(*layer), n_layers); |
4275be63 MCC |
307 | for (i = 0; i < n_layers; i++) { |
308 | count *= layers[i].size; | |
956b9ba1 | 309 | edac_dbg(4, "errcount layer %d size %d\n", i, count); |
4275be63 MCC |
310 | ce_per_layer[i] = edac_align_ptr(&ptr, sizeof(u32), count); |
311 | ue_per_layer[i] = edac_align_ptr(&ptr, sizeof(u32), count); | |
312 | tot_errcount += 2 * count; | |
313 | } | |
314 | ||
956b9ba1 | 315 | edac_dbg(4, "allocating %d error counters\n", tot_errcount); |
93e4fe64 | 316 | pvt = edac_align_ptr(&ptr, sz_pvt, 1); |
079708b9 | 317 | size = ((unsigned long)pvt) + sz_pvt; |
da9bb1d2 | 318 | |
956b9ba1 JP |
319 | edac_dbg(1, "allocating %u bytes for mci data (%d %s, %d csrows/channels)\n", |
320 | size, | |
321 | tot_dimms, | |
322 | per_rank ? "ranks" : "dimms", | |
323 | tot_csrows * tot_channels); | |
de3910eb | 324 | |
8096cfaf DT |
325 | mci = kzalloc(size, GFP_KERNEL); |
326 | if (mci == NULL) | |
da9bb1d2 AC |
327 | return NULL; |
328 | ||
329 | /* Adjust pointers so they point within the memory we just allocated | |
330 | * rather than an imaginary chunk of memory located at address 0. | |
331 | */ | |
4275be63 | 332 | layer = (struct edac_mc_layer *)(((char *)mci) + ((unsigned long)layer)); |
4275be63 MCC |
333 | for (i = 0; i < n_layers; i++) { |
334 | mci->ce_per_layer[i] = (u32 *)((char *)mci + ((unsigned long)ce_per_layer[i])); | |
335 | mci->ue_per_layer[i] = (u32 *)((char *)mci + ((unsigned long)ue_per_layer[i])); | |
336 | } | |
079708b9 | 337 | pvt = sz_pvt ? (((char *)mci) + ((unsigned long)pvt)) : NULL; |
da9bb1d2 | 338 | |
b8f6f975 | 339 | /* setup index and various internal pointers */ |
4275be63 | 340 | mci->mc_idx = mc_num; |
4275be63 | 341 | mci->tot_dimms = tot_dimms; |
da9bb1d2 | 342 | mci->pvt_info = pvt; |
4275be63 MCC |
343 | mci->n_layers = n_layers; |
344 | mci->layers = layer; | |
345 | memcpy(mci->layers, layers, sizeof(*layer) * n_layers); | |
346 | mci->nr_csrows = tot_csrows; | |
347 | mci->num_cschannel = tot_channels; | |
9713faec | 348 | mci->csbased = per_rank; |
da9bb1d2 | 349 | |
a7d7d2e1 | 350 | /* |
de3910eb | 351 | * Alocate and fill the csrow/channels structs |
a7d7d2e1 | 352 | */ |
d3d09e18 | 353 | mci->csrows = kcalloc(tot_csrows, sizeof(*mci->csrows), GFP_KERNEL); |
de3910eb MCC |
354 | if (!mci->csrows) |
355 | goto error; | |
4275be63 | 356 | for (row = 0; row < tot_csrows; row++) { |
de3910eb MCC |
357 | csr = kzalloc(sizeof(**mci->csrows), GFP_KERNEL); |
358 | if (!csr) | |
359 | goto error; | |
360 | mci->csrows[row] = csr; | |
4275be63 MCC |
361 | csr->csrow_idx = row; |
362 | csr->mci = mci; | |
363 | csr->nr_channels = tot_channels; | |
d3d09e18 | 364 | csr->channels = kcalloc(tot_channels, sizeof(*csr->channels), |
de3910eb MCC |
365 | GFP_KERNEL); |
366 | if (!csr->channels) | |
367 | goto error; | |
4275be63 MCC |
368 | |
369 | for (chn = 0; chn < tot_channels; chn++) { | |
de3910eb MCC |
370 | chan = kzalloc(sizeof(**csr->channels), GFP_KERNEL); |
371 | if (!chan) | |
372 | goto error; | |
373 | csr->channels[chn] = chan; | |
da9bb1d2 | 374 | chan->chan_idx = chn; |
4275be63 MCC |
375 | chan->csrow = csr; |
376 | } | |
377 | } | |
378 | ||
379 | /* | |
de3910eb | 380 | * Allocate and fill the dimm structs |
4275be63 | 381 | */ |
d3d09e18 | 382 | mci->dimms = kcalloc(tot_dimms, sizeof(*mci->dimms), GFP_KERNEL); |
de3910eb MCC |
383 | if (!mci->dimms) |
384 | goto error; | |
385 | ||
4275be63 MCC |
386 | memset(&pos, 0, sizeof(pos)); |
387 | row = 0; | |
388 | chn = 0; | |
4275be63 | 389 | for (i = 0; i < tot_dimms; i++) { |
de3910eb MCC |
390 | chan = mci->csrows[row]->channels[chn]; |
391 | off = EDAC_DIMM_OFF(layer, n_layers, pos[0], pos[1], pos[2]); | |
392 | if (off < 0 || off >= tot_dimms) { | |
393 | edac_mc_printk(mci, KERN_ERR, "EDAC core bug: EDAC_DIMM_OFF is trying to do an illegal data access\n"); | |
394 | goto error; | |
395 | } | |
4275be63 | 396 | |
de3910eb | 397 | dimm = kzalloc(sizeof(**mci->dimms), GFP_KERNEL); |
08a4a136 DC |
398 | if (!dimm) |
399 | goto error; | |
de3910eb | 400 | mci->dimms[off] = dimm; |
4275be63 | 401 | dimm->mci = mci; |
4275be63 | 402 | |
5926ff50 MCC |
403 | /* |
404 | * Copy DIMM location and initialize it. | |
405 | */ | |
406 | len = sizeof(dimm->label); | |
407 | p = dimm->label; | |
408 | n = snprintf(p, len, "mc#%u", mc_num); | |
409 | p += n; | |
410 | len -= n; | |
411 | for (j = 0; j < n_layers; j++) { | |
412 | n = snprintf(p, len, "%s#%u", | |
413 | edac_layer_name[layers[j].type], | |
414 | pos[j]); | |
415 | p += n; | |
416 | len -= n; | |
4275be63 MCC |
417 | dimm->location[j] = pos[j]; |
418 | ||
5926ff50 MCC |
419 | if (len <= 0) |
420 | break; | |
421 | } | |
422 | ||
4275be63 MCC |
423 | /* Link it to the csrows old API data */ |
424 | chan->dimm = dimm; | |
425 | dimm->csrow = row; | |
426 | dimm->cschannel = chn; | |
427 | ||
428 | /* Increment csrow location */ | |
24bef66e | 429 | if (layers[0].is_virt_csrow) { |
4275be63 | 430 | chn++; |
24bef66e MCC |
431 | if (chn == tot_channels) { |
432 | chn = 0; | |
433 | row++; | |
434 | } | |
435 | } else { | |
436 | row++; | |
437 | if (row == tot_csrows) { | |
438 | row = 0; | |
439 | chn++; | |
440 | } | |
4275be63 | 441 | } |
a7d7d2e1 | 442 | |
4275be63 MCC |
443 | /* Increment dimm location */ |
444 | for (j = n_layers - 1; j >= 0; j--) { | |
445 | pos[j]++; | |
446 | if (pos[j] < layers[j].size) | |
447 | break; | |
448 | pos[j] = 0; | |
da9bb1d2 AC |
449 | } |
450 | } | |
451 | ||
81d87cb1 | 452 | mci->op_state = OP_ALLOC; |
8096cfaf | 453 | |
da9bb1d2 | 454 | return mci; |
de3910eb MCC |
455 | |
456 | error: | |
faa2ad09 | 457 | _edac_mc_free(mci); |
de3910eb MCC |
458 | |
459 | return NULL; | |
4275be63 | 460 | } |
9110540f | 461 | EXPORT_SYMBOL_GPL(edac_mc_alloc); |
da9bb1d2 | 462 | |
da9bb1d2 | 463 | /** |
8096cfaf DT |
464 | * edac_mc_free |
465 | * 'Free' a previously allocated 'mci' structure | |
da9bb1d2 | 466 | * @mci: pointer to a struct mem_ctl_info structure |
da9bb1d2 AC |
467 | */ |
468 | void edac_mc_free(struct mem_ctl_info *mci) | |
469 | { | |
956b9ba1 | 470 | edac_dbg(1, "\n"); |
bbc560ae | 471 | |
faa2ad09 SR |
472 | /* If we're not yet registered with sysfs free only what was allocated |
473 | * in edac_mc_alloc(). | |
474 | */ | |
475 | if (!device_is_registered(&mci->dev)) { | |
476 | _edac_mc_free(mci); | |
477 | return; | |
478 | } | |
479 | ||
de3910eb | 480 | /* the mci instance is freed here, when the sysfs object is dropped */ |
7a623c03 | 481 | edac_unregister_sysfs(mci); |
da9bb1d2 | 482 | } |
9110540f | 483 | EXPORT_SYMBOL_GPL(edac_mc_free); |
da9bb1d2 | 484 | |
bce19683 | 485 | |
939747bd | 486 | /** |
bce19683 DT |
487 | * find_mci_by_dev |
488 | * | |
489 | * scan list of controllers looking for the one that manages | |
490 | * the 'dev' device | |
939747bd | 491 | * @dev: pointer to a struct device related with the MCI |
bce19683 | 492 | */ |
939747bd | 493 | struct mem_ctl_info *find_mci_by_dev(struct device *dev) |
da9bb1d2 AC |
494 | { |
495 | struct mem_ctl_info *mci; | |
496 | struct list_head *item; | |
497 | ||
956b9ba1 | 498 | edac_dbg(3, "\n"); |
da9bb1d2 AC |
499 | |
500 | list_for_each(item, &mc_devices) { | |
501 | mci = list_entry(item, struct mem_ctl_info, link); | |
502 | ||
fd687502 | 503 | if (mci->pdev == dev) |
da9bb1d2 AC |
504 | return mci; |
505 | } | |
506 | ||
507 | return NULL; | |
508 | } | |
939747bd | 509 | EXPORT_SYMBOL_GPL(find_mci_by_dev); |
da9bb1d2 | 510 | |
81d87cb1 DJ |
511 | /* |
512 | * handler for EDAC to check if NMI type handler has asserted interrupt | |
513 | */ | |
514 | static int edac_mc_assert_error_check_and_clear(void) | |
515 | { | |
66ee2f94 | 516 | int old_state; |
81d87cb1 | 517 | |
079708b9 | 518 | if (edac_op_state == EDAC_OPSTATE_POLL) |
81d87cb1 DJ |
519 | return 1; |
520 | ||
66ee2f94 DJ |
521 | old_state = edac_err_assert; |
522 | edac_err_assert = 0; | |
81d87cb1 | 523 | |
66ee2f94 | 524 | return old_state; |
81d87cb1 DJ |
525 | } |
526 | ||
527 | /* | |
528 | * edac_mc_workq_function | |
529 | * performs the operation scheduled by a workq request | |
530 | */ | |
81d87cb1 DJ |
531 | static void edac_mc_workq_function(struct work_struct *work_req) |
532 | { | |
fbeb4384 | 533 | struct delayed_work *d_work = to_delayed_work(work_req); |
81d87cb1 | 534 | struct mem_ctl_info *mci = to_edac_mem_ctl_work(d_work); |
81d87cb1 DJ |
535 | |
536 | mutex_lock(&mem_ctls_mutex); | |
537 | ||
bf52fa4a DT |
538 | /* if this control struct has movd to offline state, we are done */ |
539 | if (mci->op_state == OP_OFFLINE) { | |
540 | mutex_unlock(&mem_ctls_mutex); | |
541 | return; | |
542 | } | |
543 | ||
81d87cb1 DJ |
544 | /* Only poll controllers that are running polled and have a check */ |
545 | if (edac_mc_assert_error_check_and_clear() && (mci->edac_check != NULL)) | |
546 | mci->edac_check(mci); | |
547 | ||
81d87cb1 DJ |
548 | mutex_unlock(&mem_ctls_mutex); |
549 | ||
550 | /* Reschedule */ | |
4de78c68 | 551 | queue_delayed_work(edac_workqueue, &mci->work, |
052dfb45 | 552 | msecs_to_jiffies(edac_mc_get_poll_msec())); |
81d87cb1 DJ |
553 | } |
554 | ||
555 | /* | |
556 | * edac_mc_workq_setup | |
557 | * initialize a workq item for this mci | |
558 | * passing in the new delay period in msec | |
bf52fa4a DT |
559 | * |
560 | * locking model: | |
561 | * | |
562 | * called with the mem_ctls_mutex held | |
81d87cb1 | 563 | */ |
cb6ef42e BP |
564 | static void edac_mc_workq_setup(struct mem_ctl_info *mci, unsigned msec, |
565 | bool init) | |
81d87cb1 | 566 | { |
956b9ba1 | 567 | edac_dbg(0, "\n"); |
81d87cb1 | 568 | |
bf52fa4a DT |
569 | /* if this instance is not in the POLL state, then simply return */ |
570 | if (mci->op_state != OP_RUNNING_POLL) | |
571 | return; | |
572 | ||
cb6ef42e BP |
573 | if (init) |
574 | INIT_DELAYED_WORK(&mci->work, edac_mc_workq_function); | |
575 | ||
41f63c53 | 576 | mod_delayed_work(edac_workqueue, &mci->work, msecs_to_jiffies(msec)); |
81d87cb1 DJ |
577 | } |
578 | ||
579 | /* | |
580 | * edac_mc_workq_teardown | |
581 | * stop the workq processing on this mci | |
bf52fa4a DT |
582 | * |
583 | * locking model: | |
584 | * | |
585 | * called WITHOUT lock held | |
81d87cb1 | 586 | */ |
bf52fa4a | 587 | static void edac_mc_workq_teardown(struct mem_ctl_info *mci) |
81d87cb1 DJ |
588 | { |
589 | int status; | |
590 | ||
00740c58 BP |
591 | if (mci->op_state != OP_RUNNING_POLL) |
592 | return; | |
593 | ||
bce19683 DT |
594 | status = cancel_delayed_work(&mci->work); |
595 | if (status == 0) { | |
956b9ba1 | 596 | edac_dbg(0, "not canceled, flush the queue\n"); |
bf52fa4a | 597 | |
bce19683 DT |
598 | /* workq instance might be running, wait for it */ |
599 | flush_workqueue(edac_workqueue); | |
81d87cb1 DJ |
600 | } |
601 | } | |
602 | ||
603 | /* | |
bce19683 DT |
604 | * edac_mc_reset_delay_period(unsigned long value) |
605 | * | |
606 | * user space has updated our poll period value, need to | |
607 | * reset our workq delays | |
81d87cb1 | 608 | */ |
9da21b15 | 609 | void edac_mc_reset_delay_period(unsigned long value) |
81d87cb1 | 610 | { |
bce19683 DT |
611 | struct mem_ctl_info *mci; |
612 | struct list_head *item; | |
613 | ||
614 | mutex_lock(&mem_ctls_mutex); | |
615 | ||
bce19683 DT |
616 | list_for_each(item, &mc_devices) { |
617 | mci = list_entry(item, struct mem_ctl_info, link); | |
618 | ||
cb6ef42e | 619 | edac_mc_workq_setup(mci, value, false); |
bce19683 | 620 | } |
81d87cb1 DJ |
621 | |
622 | mutex_unlock(&mem_ctls_mutex); | |
623 | } | |
624 | ||
bce19683 DT |
625 | |
626 | ||
2d7bbb91 DT |
627 | /* Return 0 on success, 1 on failure. |
628 | * Before calling this function, caller must | |
629 | * assign a unique value to mci->mc_idx. | |
bf52fa4a DT |
630 | * |
631 | * locking model: | |
632 | * | |
633 | * called with the mem_ctls_mutex lock held | |
2d7bbb91 | 634 | */ |
079708b9 | 635 | static int add_mc_to_global_list(struct mem_ctl_info *mci) |
da9bb1d2 AC |
636 | { |
637 | struct list_head *item, *insert_before; | |
638 | struct mem_ctl_info *p; | |
da9bb1d2 | 639 | |
2d7bbb91 | 640 | insert_before = &mc_devices; |
da9bb1d2 | 641 | |
fd687502 | 642 | p = find_mci_by_dev(mci->pdev); |
bf52fa4a | 643 | if (unlikely(p != NULL)) |
2d7bbb91 | 644 | goto fail0; |
da9bb1d2 | 645 | |
2d7bbb91 DT |
646 | list_for_each(item, &mc_devices) { |
647 | p = list_entry(item, struct mem_ctl_info, link); | |
da9bb1d2 | 648 | |
2d7bbb91 DT |
649 | if (p->mc_idx >= mci->mc_idx) { |
650 | if (unlikely(p->mc_idx == mci->mc_idx)) | |
651 | goto fail1; | |
da9bb1d2 | 652 | |
2d7bbb91 DT |
653 | insert_before = item; |
654 | break; | |
da9bb1d2 | 655 | } |
da9bb1d2 AC |
656 | } |
657 | ||
658 | list_add_tail_rcu(&mci->link, insert_before); | |
c0d12172 | 659 | atomic_inc(&edac_handlers); |
da9bb1d2 | 660 | return 0; |
2d7bbb91 | 661 | |
052dfb45 | 662 | fail0: |
2d7bbb91 | 663 | edac_printk(KERN_WARNING, EDAC_MC, |
fd687502 | 664 | "%s (%s) %s %s already assigned %d\n", dev_name(p->pdev), |
17aa7e03 | 665 | edac_dev_name(mci), p->mod_name, p->ctl_name, p->mc_idx); |
2d7bbb91 DT |
666 | return 1; |
667 | ||
052dfb45 | 668 | fail1: |
2d7bbb91 | 669 | edac_printk(KERN_WARNING, EDAC_MC, |
052dfb45 DT |
670 | "bug in low-level driver: attempt to assign\n" |
671 | " duplicate mc_idx %d in %s()\n", p->mc_idx, __func__); | |
2d7bbb91 | 672 | return 1; |
da9bb1d2 AC |
673 | } |
674 | ||
80cc7d87 | 675 | static int del_mc_from_global_list(struct mem_ctl_info *mci) |
a1d03fcc | 676 | { |
80cc7d87 | 677 | int handlers = atomic_dec_return(&edac_handlers); |
a1d03fcc | 678 | list_del_rcu(&mci->link); |
e2e77098 LJ |
679 | |
680 | /* these are for safe removal of devices from global list while | |
681 | * NMI handlers may be traversing list | |
682 | */ | |
683 | synchronize_rcu(); | |
684 | INIT_LIST_HEAD(&mci->link); | |
80cc7d87 MCC |
685 | |
686 | return handlers; | |
a1d03fcc DP |
687 | } |
688 | ||
5da0831c DT |
689 | /** |
690 | * edac_mc_find: Search for a mem_ctl_info structure whose index is 'idx'. | |
691 | * | |
692 | * If found, return a pointer to the structure. | |
693 | * Else return NULL. | |
694 | * | |
695 | * Caller must hold mem_ctls_mutex. | |
696 | */ | |
079708b9 | 697 | struct mem_ctl_info *edac_mc_find(int idx) |
5da0831c DT |
698 | { |
699 | struct list_head *item; | |
700 | struct mem_ctl_info *mci; | |
701 | ||
702 | list_for_each(item, &mc_devices) { | |
703 | mci = list_entry(item, struct mem_ctl_info, link); | |
704 | ||
705 | if (mci->mc_idx >= idx) { | |
706 | if (mci->mc_idx == idx) | |
707 | return mci; | |
708 | ||
709 | break; | |
710 | } | |
711 | } | |
712 | ||
713 | return NULL; | |
714 | } | |
715 | EXPORT_SYMBOL(edac_mc_find); | |
716 | ||
da9bb1d2 | 717 | /** |
4e8d230d TI |
718 | * edac_mc_add_mc_with_groups: Insert the 'mci' structure into the mci |
719 | * global list and create sysfs entries associated with mci structure | |
da9bb1d2 | 720 | * @mci: pointer to the mci structure to be added to the list |
4e8d230d | 721 | * @groups: optional attribute groups for the driver-specific sysfs entries |
da9bb1d2 AC |
722 | * |
723 | * Return: | |
724 | * 0 Success | |
725 | * !0 Failure | |
726 | */ | |
727 | ||
728 | /* FIXME - should a warning be printed if no error detection? correction? */ | |
4e8d230d TI |
729 | int edac_mc_add_mc_with_groups(struct mem_ctl_info *mci, |
730 | const struct attribute_group **groups) | |
da9bb1d2 | 731 | { |
80cc7d87 | 732 | int ret = -EINVAL; |
956b9ba1 | 733 | edac_dbg(0, "\n"); |
b8f6f975 | 734 | |
88d84ac9 BP |
735 | if (mci->mc_idx >= EDAC_MAX_MCS) { |
736 | pr_warn_once("Too many memory controllers: %d\n", mci->mc_idx); | |
737 | return -ENODEV; | |
738 | } | |
739 | ||
da9bb1d2 AC |
740 | #ifdef CONFIG_EDAC_DEBUG |
741 | if (edac_debug_level >= 3) | |
742 | edac_mc_dump_mci(mci); | |
e7ecd891 | 743 | |
da9bb1d2 AC |
744 | if (edac_debug_level >= 4) { |
745 | int i; | |
746 | ||
747 | for (i = 0; i < mci->nr_csrows; i++) { | |
6e84d359 MCC |
748 | struct csrow_info *csrow = mci->csrows[i]; |
749 | u32 nr_pages = 0; | |
da9bb1d2 | 750 | int j; |
e7ecd891 | 751 | |
6e84d359 MCC |
752 | for (j = 0; j < csrow->nr_channels; j++) |
753 | nr_pages += csrow->channels[j]->dimm->nr_pages; | |
754 | if (!nr_pages) | |
755 | continue; | |
756 | edac_mc_dump_csrow(csrow); | |
757 | for (j = 0; j < csrow->nr_channels; j++) | |
758 | if (csrow->channels[j]->dimm->nr_pages) | |
759 | edac_mc_dump_channel(csrow->channels[j]); | |
da9bb1d2 | 760 | } |
4275be63 | 761 | for (i = 0; i < mci->tot_dimms; i++) |
6e84d359 MCC |
762 | if (mci->dimms[i]->nr_pages) |
763 | edac_mc_dump_dimm(mci->dimms[i], i); | |
da9bb1d2 AC |
764 | } |
765 | #endif | |
63b7df91 | 766 | mutex_lock(&mem_ctls_mutex); |
da9bb1d2 | 767 | |
80cc7d87 MCC |
768 | if (edac_mc_owner && edac_mc_owner != mci->mod_name) { |
769 | ret = -EPERM; | |
770 | goto fail0; | |
771 | } | |
772 | ||
da9bb1d2 | 773 | if (add_mc_to_global_list(mci)) |
028a7b6d | 774 | goto fail0; |
da9bb1d2 AC |
775 | |
776 | /* set load time so that error rate can be tracked */ | |
777 | mci->start_time = jiffies; | |
778 | ||
88d84ac9 BP |
779 | mci->bus = &mc_bus[mci->mc_idx]; |
780 | ||
4e8d230d | 781 | if (edac_create_sysfs_mci_device(mci, groups)) { |
9794f33d | 782 | edac_mc_printk(mci, KERN_WARNING, |
052dfb45 | 783 | "failed to create sysfs device\n"); |
9794f33d | 784 | goto fail1; |
785 | } | |
da9bb1d2 | 786 | |
81d87cb1 DJ |
787 | /* If there IS a check routine, then we are running POLLED */ |
788 | if (mci->edac_check != NULL) { | |
789 | /* This instance is NOW RUNNING */ | |
790 | mci->op_state = OP_RUNNING_POLL; | |
791 | ||
cb6ef42e | 792 | edac_mc_workq_setup(mci, edac_mc_get_poll_msec(), true); |
81d87cb1 DJ |
793 | } else { |
794 | mci->op_state = OP_RUNNING_INTERRUPT; | |
795 | } | |
796 | ||
da9bb1d2 | 797 | /* Report action taken */ |
7270a608 RR |
798 | edac_mc_printk(mci, KERN_INFO, |
799 | "Giving out device to module %s controller %s: DEV %s (%s)\n", | |
800 | mci->mod_name, mci->ctl_name, mci->dev_name, | |
801 | edac_op_state_to_string(mci->op_state)); | |
da9bb1d2 | 802 | |
80cc7d87 MCC |
803 | edac_mc_owner = mci->mod_name; |
804 | ||
63b7df91 | 805 | mutex_unlock(&mem_ctls_mutex); |
028a7b6d | 806 | return 0; |
da9bb1d2 | 807 | |
052dfb45 | 808 | fail1: |
028a7b6d DP |
809 | del_mc_from_global_list(mci); |
810 | ||
052dfb45 | 811 | fail0: |
63b7df91 | 812 | mutex_unlock(&mem_ctls_mutex); |
80cc7d87 | 813 | return ret; |
da9bb1d2 | 814 | } |
4e8d230d | 815 | EXPORT_SYMBOL_GPL(edac_mc_add_mc_with_groups); |
da9bb1d2 | 816 | |
da9bb1d2 | 817 | /** |
472678eb DP |
818 | * edac_mc_del_mc: Remove sysfs entries for specified mci structure and |
819 | * remove mci structure from global list | |
37f04581 | 820 | * @pdev: Pointer to 'struct device' representing mci structure to remove. |
da9bb1d2 | 821 | * |
18dbc337 | 822 | * Return pointer to removed mci structure, or NULL if device not found. |
da9bb1d2 | 823 | */ |
079708b9 | 824 | struct mem_ctl_info *edac_mc_del_mc(struct device *dev) |
da9bb1d2 | 825 | { |
18dbc337 | 826 | struct mem_ctl_info *mci; |
da9bb1d2 | 827 | |
956b9ba1 | 828 | edac_dbg(0, "\n"); |
bf52fa4a | 829 | |
63b7df91 | 830 | mutex_lock(&mem_ctls_mutex); |
18dbc337 | 831 | |
bf52fa4a DT |
832 | /* find the requested mci struct in the global list */ |
833 | mci = find_mci_by_dev(dev); | |
834 | if (mci == NULL) { | |
63b7df91 | 835 | mutex_unlock(&mem_ctls_mutex); |
18dbc337 DP |
836 | return NULL; |
837 | } | |
838 | ||
80cc7d87 MCC |
839 | if (!del_mc_from_global_list(mci)) |
840 | edac_mc_owner = NULL; | |
63b7df91 | 841 | mutex_unlock(&mem_ctls_mutex); |
bf52fa4a | 842 | |
bb31b312 | 843 | /* flush workq processes */ |
bf52fa4a | 844 | edac_mc_workq_teardown(mci); |
bb31b312 BP |
845 | |
846 | /* marking MCI offline */ | |
847 | mci->op_state = OP_OFFLINE; | |
848 | ||
849 | /* remove from sysfs */ | |
bf52fa4a DT |
850 | edac_remove_sysfs_mci_device(mci); |
851 | ||
537fba28 | 852 | edac_printk(KERN_INFO, EDAC_MC, |
052dfb45 | 853 | "Removed device %d for %s %s: DEV %s\n", mci->mc_idx, |
17aa7e03 | 854 | mci->mod_name, mci->ctl_name, edac_dev_name(mci)); |
bf52fa4a | 855 | |
18dbc337 | 856 | return mci; |
da9bb1d2 | 857 | } |
9110540f | 858 | EXPORT_SYMBOL_GPL(edac_mc_del_mc); |
da9bb1d2 | 859 | |
2da1c119 AB |
860 | static void edac_mc_scrub_block(unsigned long page, unsigned long offset, |
861 | u32 size) | |
da9bb1d2 AC |
862 | { |
863 | struct page *pg; | |
864 | void *virt_addr; | |
865 | unsigned long flags = 0; | |
866 | ||
956b9ba1 | 867 | edac_dbg(3, "\n"); |
da9bb1d2 AC |
868 | |
869 | /* ECC error page was not in our memory. Ignore it. */ | |
079708b9 | 870 | if (!pfn_valid(page)) |
da9bb1d2 AC |
871 | return; |
872 | ||
873 | /* Find the actual page structure then map it and fix */ | |
874 | pg = pfn_to_page(page); | |
875 | ||
876 | if (PageHighMem(pg)) | |
877 | local_irq_save(flags); | |
878 | ||
4e5df7ca | 879 | virt_addr = kmap_atomic(pg); |
da9bb1d2 AC |
880 | |
881 | /* Perform architecture specific atomic scrub operation */ | |
b01aec9b | 882 | edac_atomic_scrub(virt_addr + offset, size); |
da9bb1d2 AC |
883 | |
884 | /* Unmap and complete */ | |
4e5df7ca | 885 | kunmap_atomic(virt_addr); |
da9bb1d2 AC |
886 | |
887 | if (PageHighMem(pg)) | |
888 | local_irq_restore(flags); | |
889 | } | |
890 | ||
da9bb1d2 | 891 | /* FIXME - should return -1 */ |
e7ecd891 | 892 | int edac_mc_find_csrow_by_page(struct mem_ctl_info *mci, unsigned long page) |
da9bb1d2 | 893 | { |
de3910eb | 894 | struct csrow_info **csrows = mci->csrows; |
a895bf8b | 895 | int row, i, j, n; |
da9bb1d2 | 896 | |
956b9ba1 | 897 | edac_dbg(1, "MC%d: 0x%lx\n", mci->mc_idx, page); |
da9bb1d2 AC |
898 | row = -1; |
899 | ||
900 | for (i = 0; i < mci->nr_csrows; i++) { | |
de3910eb | 901 | struct csrow_info *csrow = csrows[i]; |
a895bf8b MCC |
902 | n = 0; |
903 | for (j = 0; j < csrow->nr_channels; j++) { | |
de3910eb | 904 | struct dimm_info *dimm = csrow->channels[j]->dimm; |
a895bf8b MCC |
905 | n += dimm->nr_pages; |
906 | } | |
907 | if (n == 0) | |
da9bb1d2 AC |
908 | continue; |
909 | ||
956b9ba1 JP |
910 | edac_dbg(3, "MC%d: first(0x%lx) page(0x%lx) last(0x%lx) mask(0x%lx)\n", |
911 | mci->mc_idx, | |
912 | csrow->first_page, page, csrow->last_page, | |
913 | csrow->page_mask); | |
da9bb1d2 AC |
914 | |
915 | if ((page >= csrow->first_page) && | |
916 | (page <= csrow->last_page) && | |
917 | ((page & csrow->page_mask) == | |
918 | (csrow->first_page & csrow->page_mask))) { | |
919 | row = i; | |
920 | break; | |
921 | } | |
922 | } | |
923 | ||
924 | if (row == -1) | |
537fba28 | 925 | edac_mc_printk(mci, KERN_ERR, |
052dfb45 DT |
926 | "could not look up page error address %lx\n", |
927 | (unsigned long)page); | |
da9bb1d2 AC |
928 | |
929 | return row; | |
930 | } | |
9110540f | 931 | EXPORT_SYMBOL_GPL(edac_mc_find_csrow_by_page); |
da9bb1d2 | 932 | |
4275be63 MCC |
933 | const char *edac_layer_name[] = { |
934 | [EDAC_MC_LAYER_BRANCH] = "branch", | |
935 | [EDAC_MC_LAYER_CHANNEL] = "channel", | |
936 | [EDAC_MC_LAYER_SLOT] = "slot", | |
937 | [EDAC_MC_LAYER_CHIP_SELECT] = "csrow", | |
c66b5a79 | 938 | [EDAC_MC_LAYER_ALL_MEM] = "memory", |
4275be63 MCC |
939 | }; |
940 | EXPORT_SYMBOL_GPL(edac_layer_name); | |
941 | ||
942 | static void edac_inc_ce_error(struct mem_ctl_info *mci, | |
9eb07a7f MCC |
943 | bool enable_per_layer_report, |
944 | const int pos[EDAC_MAX_LAYERS], | |
945 | const u16 count) | |
da9bb1d2 | 946 | { |
4275be63 | 947 | int i, index = 0; |
da9bb1d2 | 948 | |
9eb07a7f | 949 | mci->ce_mc += count; |
da9bb1d2 | 950 | |
4275be63 | 951 | if (!enable_per_layer_report) { |
9eb07a7f | 952 | mci->ce_noinfo_count += count; |
da9bb1d2 AC |
953 | return; |
954 | } | |
e7ecd891 | 955 | |
4275be63 MCC |
956 | for (i = 0; i < mci->n_layers; i++) { |
957 | if (pos[i] < 0) | |
958 | break; | |
959 | index += pos[i]; | |
9eb07a7f | 960 | mci->ce_per_layer[i][index] += count; |
4275be63 MCC |
961 | |
962 | if (i < mci->n_layers - 1) | |
963 | index *= mci->layers[i + 1].size; | |
964 | } | |
965 | } | |
966 | ||
967 | static void edac_inc_ue_error(struct mem_ctl_info *mci, | |
968 | bool enable_per_layer_report, | |
9eb07a7f MCC |
969 | const int pos[EDAC_MAX_LAYERS], |
970 | const u16 count) | |
4275be63 MCC |
971 | { |
972 | int i, index = 0; | |
973 | ||
9eb07a7f | 974 | mci->ue_mc += count; |
4275be63 MCC |
975 | |
976 | if (!enable_per_layer_report) { | |
9eb07a7f | 977 | mci->ce_noinfo_count += count; |
da9bb1d2 AC |
978 | return; |
979 | } | |
980 | ||
4275be63 MCC |
981 | for (i = 0; i < mci->n_layers; i++) { |
982 | if (pos[i] < 0) | |
983 | break; | |
984 | index += pos[i]; | |
9eb07a7f | 985 | mci->ue_per_layer[i][index] += count; |
a7d7d2e1 | 986 | |
4275be63 MCC |
987 | if (i < mci->n_layers - 1) |
988 | index *= mci->layers[i + 1].size; | |
989 | } | |
990 | } | |
da9bb1d2 | 991 | |
4275be63 | 992 | static void edac_ce_error(struct mem_ctl_info *mci, |
9eb07a7f | 993 | const u16 error_count, |
4275be63 MCC |
994 | const int pos[EDAC_MAX_LAYERS], |
995 | const char *msg, | |
996 | const char *location, | |
997 | const char *label, | |
998 | const char *detail, | |
999 | const char *other_detail, | |
1000 | const bool enable_per_layer_report, | |
1001 | const unsigned long page_frame_number, | |
1002 | const unsigned long offset_in_page, | |
53f2d028 | 1003 | long grain) |
4275be63 MCC |
1004 | { |
1005 | unsigned long remapped_page; | |
f430d570 BP |
1006 | char *msg_aux = ""; |
1007 | ||
1008 | if (*msg) | |
1009 | msg_aux = " "; | |
4275be63 MCC |
1010 | |
1011 | if (edac_mc_get_log_ce()) { | |
1012 | if (other_detail && *other_detail) | |
1013 | edac_mc_printk(mci, KERN_WARNING, | |
f430d570 BP |
1014 | "%d CE %s%son %s (%s %s - %s)\n", |
1015 | error_count, msg, msg_aux, label, | |
1016 | location, detail, other_detail); | |
4275be63 MCC |
1017 | else |
1018 | edac_mc_printk(mci, KERN_WARNING, | |
f430d570 BP |
1019 | "%d CE %s%son %s (%s %s)\n", |
1020 | error_count, msg, msg_aux, label, | |
1021 | location, detail); | |
4275be63 | 1022 | } |
9eb07a7f | 1023 | edac_inc_ce_error(mci, enable_per_layer_report, pos, error_count); |
da9bb1d2 | 1024 | |
aa2064d7 | 1025 | if (mci->scrub_mode == SCRUB_SW_SRC) { |
da9bb1d2 | 1026 | /* |
4275be63 MCC |
1027 | * Some memory controllers (called MCs below) can remap |
1028 | * memory so that it is still available at a different | |
1029 | * address when PCI devices map into memory. | |
1030 | * MC's that can't do this, lose the memory where PCI | |
1031 | * devices are mapped. This mapping is MC-dependent | |
1032 | * and so we call back into the MC driver for it to | |
1033 | * map the MC page to a physical (CPU) page which can | |
1034 | * then be mapped to a virtual page - which can then | |
1035 | * be scrubbed. | |
1036 | */ | |
da9bb1d2 | 1037 | remapped_page = mci->ctl_page_to_phys ? |
052dfb45 DT |
1038 | mci->ctl_page_to_phys(mci, page_frame_number) : |
1039 | page_frame_number; | |
da9bb1d2 | 1040 | |
4275be63 MCC |
1041 | edac_mc_scrub_block(remapped_page, |
1042 | offset_in_page, grain); | |
da9bb1d2 AC |
1043 | } |
1044 | } | |
1045 | ||
4275be63 | 1046 | static void edac_ue_error(struct mem_ctl_info *mci, |
9eb07a7f | 1047 | const u16 error_count, |
4275be63 MCC |
1048 | const int pos[EDAC_MAX_LAYERS], |
1049 | const char *msg, | |
1050 | const char *location, | |
1051 | const char *label, | |
1052 | const char *detail, | |
1053 | const char *other_detail, | |
1054 | const bool enable_per_layer_report) | |
da9bb1d2 | 1055 | { |
f430d570 BP |
1056 | char *msg_aux = ""; |
1057 | ||
1058 | if (*msg) | |
1059 | msg_aux = " "; | |
1060 | ||
4275be63 MCC |
1061 | if (edac_mc_get_log_ue()) { |
1062 | if (other_detail && *other_detail) | |
1063 | edac_mc_printk(mci, KERN_WARNING, | |
f430d570 BP |
1064 | "%d UE %s%son %s (%s %s - %s)\n", |
1065 | error_count, msg, msg_aux, label, | |
1066 | location, detail, other_detail); | |
4275be63 MCC |
1067 | else |
1068 | edac_mc_printk(mci, KERN_WARNING, | |
f430d570 BP |
1069 | "%d UE %s%son %s (%s %s)\n", |
1070 | error_count, msg, msg_aux, label, | |
1071 | location, detail); | |
4275be63 | 1072 | } |
e7ecd891 | 1073 | |
4275be63 MCC |
1074 | if (edac_mc_get_panic_on_ue()) { |
1075 | if (other_detail && *other_detail) | |
f430d570 BP |
1076 | panic("UE %s%son %s (%s%s - %s)\n", |
1077 | msg, msg_aux, label, location, detail, other_detail); | |
4275be63 | 1078 | else |
f430d570 BP |
1079 | panic("UE %s%son %s (%s%s)\n", |
1080 | msg, msg_aux, label, location, detail); | |
4275be63 MCC |
1081 | } |
1082 | ||
9eb07a7f | 1083 | edac_inc_ue_error(mci, enable_per_layer_report, pos, error_count); |
da9bb1d2 AC |
1084 | } |
1085 | ||
e7e24830 MCC |
1086 | /** |
1087 | * edac_raw_mc_handle_error - reports a memory event to userspace without doing | |
1088 | * anything to discover the error location | |
1089 | * | |
1090 | * @type: severity of the error (CE/UE/Fatal) | |
1091 | * @mci: a struct mem_ctl_info pointer | |
1092 | * @e: error description | |
1093 | * | |
1094 | * This raw function is used internally by edac_mc_handle_error(). It should | |
1095 | * only be called directly when the hardware error come directly from BIOS, | |
1096 | * like in the case of APEI GHES driver. | |
1097 | */ | |
1098 | void edac_raw_mc_handle_error(const enum hw_event_mc_err_type type, | |
1099 | struct mem_ctl_info *mci, | |
1100 | struct edac_raw_error_desc *e) | |
1101 | { | |
1102 | char detail[80]; | |
1103 | int pos[EDAC_MAX_LAYERS] = { e->top_layer, e->mid_layer, e->low_layer }; | |
1104 | ||
1105 | /* Memory type dependent details about the error */ | |
1106 | if (type == HW_EVENT_ERR_CORRECTED) { | |
1107 | snprintf(detail, sizeof(detail), | |
1108 | "page:0x%lx offset:0x%lx grain:%ld syndrome:0x%lx", | |
1109 | e->page_frame_number, e->offset_in_page, | |
1110 | e->grain, e->syndrome); | |
1111 | edac_ce_error(mci, e->error_count, pos, e->msg, e->location, e->label, | |
1112 | detail, e->other_detail, e->enable_per_layer_report, | |
1113 | e->page_frame_number, e->offset_in_page, e->grain); | |
1114 | } else { | |
1115 | snprintf(detail, sizeof(detail), | |
1116 | "page:0x%lx offset:0x%lx grain:%ld", | |
1117 | e->page_frame_number, e->offset_in_page, e->grain); | |
1118 | ||
1119 | edac_ue_error(mci, e->error_count, pos, e->msg, e->location, e->label, | |
1120 | detail, e->other_detail, e->enable_per_layer_report); | |
1121 | } | |
1122 | ||
1123 | ||
1124 | } | |
1125 | EXPORT_SYMBOL_GPL(edac_raw_mc_handle_error); | |
53f2d028 MCC |
1126 | |
1127 | /** | |
1128 | * edac_mc_handle_error - reports a memory event to userspace | |
1129 | * | |
1130 | * @type: severity of the error (CE/UE/Fatal) | |
1131 | * @mci: a struct mem_ctl_info pointer | |
9eb07a7f | 1132 | * @error_count: Number of errors of the same type |
53f2d028 MCC |
1133 | * @page_frame_number: mem page where the error occurred |
1134 | * @offset_in_page: offset of the error inside the page | |
1135 | * @syndrome: ECC syndrome | |
1136 | * @top_layer: Memory layer[0] position | |
1137 | * @mid_layer: Memory layer[1] position | |
1138 | * @low_layer: Memory layer[2] position | |
1139 | * @msg: Message meaningful to the end users that | |
1140 | * explains the event | |
1141 | * @other_detail: Technical details about the event that | |
1142 | * may help hardware manufacturers and | |
1143 | * EDAC developers to analyse the event | |
53f2d028 | 1144 | */ |
4275be63 MCC |
1145 | void edac_mc_handle_error(const enum hw_event_mc_err_type type, |
1146 | struct mem_ctl_info *mci, | |
9eb07a7f | 1147 | const u16 error_count, |
4275be63 MCC |
1148 | const unsigned long page_frame_number, |
1149 | const unsigned long offset_in_page, | |
1150 | const unsigned long syndrome, | |
53f2d028 MCC |
1151 | const int top_layer, |
1152 | const int mid_layer, | |
1153 | const int low_layer, | |
4275be63 | 1154 | const char *msg, |
03f7eae8 | 1155 | const char *other_detail) |
da9bb1d2 | 1156 | { |
4275be63 MCC |
1157 | char *p; |
1158 | int row = -1, chan = -1; | |
53f2d028 | 1159 | int pos[EDAC_MAX_LAYERS] = { top_layer, mid_layer, low_layer }; |
c7ef7645 | 1160 | int i, n_labels = 0; |
53f2d028 | 1161 | u8 grain_bits; |
c7ef7645 | 1162 | struct edac_raw_error_desc *e = &mci->error_desc; |
da9bb1d2 | 1163 | |
956b9ba1 | 1164 | edac_dbg(3, "MC%d\n", mci->mc_idx); |
da9bb1d2 | 1165 | |
c7ef7645 MCC |
1166 | /* Fills the error report buffer */ |
1167 | memset(e, 0, sizeof (*e)); | |
1168 | e->error_count = error_count; | |
1169 | e->top_layer = top_layer; | |
1170 | e->mid_layer = mid_layer; | |
1171 | e->low_layer = low_layer; | |
1172 | e->page_frame_number = page_frame_number; | |
1173 | e->offset_in_page = offset_in_page; | |
1174 | e->syndrome = syndrome; | |
1175 | e->msg = msg; | |
1176 | e->other_detail = other_detail; | |
1177 | ||
4275be63 MCC |
1178 | /* |
1179 | * Check if the event report is consistent and if the memory | |
1180 | * location is known. If it is known, enable_per_layer_report will be | |
1181 | * true, the DIMM(s) label info will be filled and the per-layer | |
1182 | * error counters will be incremented. | |
1183 | */ | |
1184 | for (i = 0; i < mci->n_layers; i++) { | |
1185 | if (pos[i] >= (int)mci->layers[i].size) { | |
4275be63 MCC |
1186 | |
1187 | edac_mc_printk(mci, KERN_ERR, | |
1188 | "INTERNAL ERROR: %s value is out of range (%d >= %d)\n", | |
1189 | edac_layer_name[mci->layers[i].type], | |
1190 | pos[i], mci->layers[i].size); | |
1191 | /* | |
1192 | * Instead of just returning it, let's use what's | |
1193 | * known about the error. The increment routines and | |
1194 | * the DIMM filter logic will do the right thing by | |
1195 | * pointing the likely damaged DIMMs. | |
1196 | */ | |
1197 | pos[i] = -1; | |
1198 | } | |
1199 | if (pos[i] >= 0) | |
c7ef7645 | 1200 | e->enable_per_layer_report = true; |
da9bb1d2 AC |
1201 | } |
1202 | ||
4275be63 MCC |
1203 | /* |
1204 | * Get the dimm label/grain that applies to the match criteria. | |
1205 | * As the error algorithm may not be able to point to just one memory | |
1206 | * stick, the logic here will get all possible labels that could | |
1207 | * pottentially be affected by the error. | |
1208 | * On FB-DIMM memory controllers, for uncorrected errors, it is common | |
1209 | * to have only the MC channel and the MC dimm (also called "branch") | |
1210 | * but the channel is not known, as the memory is arranged in pairs, | |
1211 | * where each memory belongs to a separate channel within the same | |
1212 | * branch. | |
1213 | */ | |
c7ef7645 | 1214 | p = e->label; |
4275be63 | 1215 | *p = '\0'; |
4da1b7bf | 1216 | |
4275be63 | 1217 | for (i = 0; i < mci->tot_dimms; i++) { |
de3910eb | 1218 | struct dimm_info *dimm = mci->dimms[i]; |
da9bb1d2 | 1219 | |
53f2d028 | 1220 | if (top_layer >= 0 && top_layer != dimm->location[0]) |
4275be63 | 1221 | continue; |
53f2d028 | 1222 | if (mid_layer >= 0 && mid_layer != dimm->location[1]) |
4275be63 | 1223 | continue; |
53f2d028 | 1224 | if (low_layer >= 0 && low_layer != dimm->location[2]) |
4275be63 | 1225 | continue; |
da9bb1d2 | 1226 | |
4275be63 | 1227 | /* get the max grain, over the error match range */ |
c7ef7645 MCC |
1228 | if (dimm->grain > e->grain) |
1229 | e->grain = dimm->grain; | |
9794f33d | 1230 | |
4275be63 MCC |
1231 | /* |
1232 | * If the error is memory-controller wide, there's no need to | |
1233 | * seek for the affected DIMMs because the whole | |
1234 | * channel/memory controller/... may be affected. | |
1235 | * Also, don't show errors for empty DIMM slots. | |
1236 | */ | |
c7ef7645 MCC |
1237 | if (e->enable_per_layer_report && dimm->nr_pages) { |
1238 | if (n_labels >= EDAC_MAX_LABELS) { | |
1239 | e->enable_per_layer_report = false; | |
1240 | break; | |
1241 | } | |
1242 | n_labels++; | |
1243 | if (p != e->label) { | |
4275be63 MCC |
1244 | strcpy(p, OTHER_LABEL); |
1245 | p += strlen(OTHER_LABEL); | |
1246 | } | |
1247 | strcpy(p, dimm->label); | |
1248 | p += strlen(p); | |
1249 | *p = '\0'; | |
1250 | ||
1251 | /* | |
1252 | * get csrow/channel of the DIMM, in order to allow | |
1253 | * incrementing the compat API counters | |
1254 | */ | |
956b9ba1 | 1255 | edac_dbg(4, "%s csrows map: (%d,%d)\n", |
9713faec | 1256 | mci->csbased ? "rank" : "dimm", |
956b9ba1 | 1257 | dimm->csrow, dimm->cschannel); |
4275be63 MCC |
1258 | if (row == -1) |
1259 | row = dimm->csrow; | |
1260 | else if (row >= 0 && row != dimm->csrow) | |
1261 | row = -2; | |
1262 | ||
1263 | if (chan == -1) | |
1264 | chan = dimm->cschannel; | |
1265 | else if (chan >= 0 && chan != dimm->cschannel) | |
1266 | chan = -2; | |
1267 | } | |
9794f33d | 1268 | } |
1269 | ||
c7ef7645 MCC |
1270 | if (!e->enable_per_layer_report) { |
1271 | strcpy(e->label, "any memory"); | |
4275be63 | 1272 | } else { |
956b9ba1 | 1273 | edac_dbg(4, "csrow/channel to increment: (%d,%d)\n", row, chan); |
c7ef7645 MCC |
1274 | if (p == e->label) |
1275 | strcpy(e->label, "unknown memory"); | |
4275be63 MCC |
1276 | if (type == HW_EVENT_ERR_CORRECTED) { |
1277 | if (row >= 0) { | |
9eb07a7f | 1278 | mci->csrows[row]->ce_count += error_count; |
4275be63 | 1279 | if (chan >= 0) |
9eb07a7f | 1280 | mci->csrows[row]->channels[chan]->ce_count += error_count; |
4275be63 MCC |
1281 | } |
1282 | } else | |
1283 | if (row >= 0) | |
9eb07a7f | 1284 | mci->csrows[row]->ue_count += error_count; |
9794f33d | 1285 | } |
1286 | ||
4275be63 | 1287 | /* Fill the RAM location data */ |
c7ef7645 | 1288 | p = e->location; |
4da1b7bf | 1289 | |
4275be63 MCC |
1290 | for (i = 0; i < mci->n_layers; i++) { |
1291 | if (pos[i] < 0) | |
1292 | continue; | |
9794f33d | 1293 | |
4275be63 MCC |
1294 | p += sprintf(p, "%s:%d ", |
1295 | edac_layer_name[mci->layers[i].type], | |
1296 | pos[i]); | |
9794f33d | 1297 | } |
c7ef7645 | 1298 | if (p > e->location) |
53f2d028 MCC |
1299 | *(p - 1) = '\0'; |
1300 | ||
1301 | /* Report the error via the trace interface */ | |
c7ef7645 MCC |
1302 | grain_bits = fls_long(e->grain) + 1; |
1303 | trace_mc_event(type, e->msg, e->label, e->error_count, | |
1304 | mci->mc_idx, e->top_layer, e->mid_layer, e->low_layer, | |
1305 | PAGES_TO_MiB(e->page_frame_number) | e->offset_in_page, | |
e7e24830 | 1306 | grain_bits, e->syndrome, e->other_detail); |
a7d7d2e1 | 1307 | |
e7e24830 | 1308 | edac_raw_mc_handle_error(type, mci, e); |
9794f33d | 1309 | } |
4275be63 | 1310 | EXPORT_SYMBOL_GPL(edac_mc_handle_error); |