]>
Commit | Line | Data |
---|---|---|
806c35f5 AC |
1 | /* |
2 | * AMD 76x Memory Controller kernel module | |
3 | * (C) 2003 Linux Networx (http://lnxi.com) | |
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 | * $Id: edac_amd76x.c,v 1.4.2.5 2005/10/05 00:43:44 dsp_llnl Exp $ | |
12 | * | |
13 | */ | |
14 | ||
806c35f5 AC |
15 | #include <linux/module.h> |
16 | #include <linux/init.h> | |
806c35f5 AC |
17 | #include <linux/pci.h> |
18 | #include <linux/pci_ids.h> | |
c3c52bce | 19 | #include <linux/edac.h> |
78d88e8a | 20 | #include "edac_module.h" |
806c35f5 | 21 | |
929a40ec | 22 | #define EDAC_MOD_STR "amd76x_edac" |
37f04581 | 23 | |
537fba28 | 24 | #define amd76x_printk(level, fmt, arg...) \ |
e7ecd891 | 25 | edac_printk(level, "amd76x", fmt, ##arg) |
537fba28 DP |
26 | |
27 | #define amd76x_mc_printk(mci, level, fmt, arg...) \ | |
e7ecd891 | 28 | edac_mc_chipset_printk(mci, level, "amd76x", fmt, ##arg) |
537fba28 | 29 | |
806c35f5 | 30 | #define AMD76X_NR_CSROWS 8 |
806c35f5 AC |
31 | #define AMD76X_NR_DIMMS 4 |
32 | ||
806c35f5 | 33 | /* AMD 76x register addresses - device 0 function 0 - PCI bridge */ |
e7ecd891 | 34 | |
806c35f5 AC |
35 | #define AMD76X_ECC_MODE_STATUS 0x48 /* Mode and status of ECC (32b) |
36 | * | |
37 | * 31:16 reserved | |
38 | * 15:14 SERR enabled: x1=ue 1x=ce | |
39 | * 13 reserved | |
40 | * 12 diag: disabled, enabled | |
41 | * 11:10 mode: dis, EC, ECC, ECC+scrub | |
42 | * 9:8 status: x1=ue 1x=ce | |
43 | * 7:4 UE cs row | |
44 | * 3:0 CE cs row | |
45 | */ | |
e7ecd891 | 46 | |
806c35f5 AC |
47 | #define AMD76X_DRAM_MODE_STATUS 0x58 /* DRAM Mode and status (32b) |
48 | * | |
49 | * 31:26 clock disable 5 - 0 | |
50 | * 25 SDRAM init | |
51 | * 24 reserved | |
52 | * 23 mode register service | |
53 | * 22:21 suspend to RAM | |
54 | * 20 burst refresh enable | |
55 | * 19 refresh disable | |
56 | * 18 reserved | |
57 | * 17:16 cycles-per-refresh | |
58 | * 15:8 reserved | |
59 | * 7:0 x4 mode enable 7 - 0 | |
60 | */ | |
e7ecd891 | 61 | |
806c35f5 AC |
62 | #define AMD76X_MEM_BASE_ADDR 0xC0 /* Memory base address (8 x 32b) |
63 | * | |
64 | * 31:23 chip-select base | |
65 | * 22:16 reserved | |
66 | * 15:7 chip-select mask | |
67 | * 6:3 reserved | |
68 | * 2:1 address mode | |
69 | * 0 chip-select enable | |
70 | */ | |
71 | ||
806c35f5 AC |
72 | struct amd76x_error_info { |
73 | u32 ecc_mode_status; | |
74 | }; | |
75 | ||
806c35f5 AC |
76 | enum amd76x_chips { |
77 | AMD761 = 0, | |
78 | AMD762 | |
79 | }; | |
80 | ||
806c35f5 AC |
81 | struct amd76x_dev_info { |
82 | const char *ctl_name; | |
83 | }; | |
84 | ||
806c35f5 | 85 | static const struct amd76x_dev_info amd76x_devs[] = { |
e7ecd891 | 86 | [AMD761] = { |
052dfb45 | 87 | .ctl_name = "AMD761"}, |
e7ecd891 | 88 | [AMD762] = { |
052dfb45 | 89 | .ctl_name = "AMD762"}, |
806c35f5 AC |
90 | }; |
91 | ||
456a2f95 DJ |
92 | static struct edac_pci_ctl_info *amd76x_pci; |
93 | ||
806c35f5 AC |
94 | /** |
95 | * amd76x_get_error_info - fetch error information | |
96 | * @mci: Memory controller | |
97 | * @info: Info to fill in | |
98 | * | |
99 | * Fetch and store the AMD76x ECC status. Clear pending status | |
100 | * on the chip so that further errors will be reported | |
101 | */ | |
e7ecd891 | 102 | static void amd76x_get_error_info(struct mem_ctl_info *mci, |
052dfb45 | 103 | struct amd76x_error_info *info) |
806c35f5 | 104 | { |
37f04581 DT |
105 | struct pci_dev *pdev; |
106 | ||
fd687502 | 107 | pdev = to_pci_dev(mci->pdev); |
37f04581 | 108 | pci_read_config_dword(pdev, AMD76X_ECC_MODE_STATUS, |
052dfb45 | 109 | &info->ecc_mode_status); |
806c35f5 AC |
110 | |
111 | if (info->ecc_mode_status & BIT(8)) | |
37f04581 | 112 | pci_write_bits32(pdev, AMD76X_ECC_MODE_STATUS, |
67cb2b61 | 113 | (u32) BIT(8), (u32) BIT(8)); |
806c35f5 AC |
114 | |
115 | if (info->ecc_mode_status & BIT(9)) | |
37f04581 | 116 | pci_write_bits32(pdev, AMD76X_ECC_MODE_STATUS, |
67cb2b61 | 117 | (u32) BIT(9), (u32) BIT(9)); |
806c35f5 AC |
118 | } |
119 | ||
806c35f5 AC |
120 | /** |
121 | * amd76x_process_error_info - Error check | |
122 | * @mci: Memory controller | |
123 | * @info: Previously fetched information from chip | |
124 | * @handle_errors: 1 if we should do recovery | |
125 | * | |
126 | * Process the chip state and decide if an error has occurred. | |
127 | * A return of 1 indicates an error. Also if handle_errors is true | |
128 | * then attempt to handle and clean up after the error | |
129 | */ | |
e7ecd891 | 130 | static int amd76x_process_error_info(struct mem_ctl_info *mci, |
052dfb45 DT |
131 | struct amd76x_error_info *info, |
132 | int handle_errors) | |
806c35f5 AC |
133 | { |
134 | int error_found; | |
135 | u32 row; | |
136 | ||
137 | error_found = 0; | |
138 | ||
139 | /* | |
67cb2b61 | 140 | * Check for an uncorrectable error |
806c35f5 AC |
141 | */ |
142 | if (info->ecc_mode_status & BIT(8)) { | |
143 | error_found = 1; | |
144 | ||
145 | if (handle_errors) { | |
146 | row = (info->ecc_mode_status >> 4) & 0xf; | |
9eb07a7f | 147 | edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci, 1, |
de3910eb | 148 | mci->csrows[row]->first_page, 0, 0, |
d8c34af4 | 149 | row, 0, -1, |
03f7eae8 | 150 | mci->ctl_name, ""); |
806c35f5 AC |
151 | } |
152 | } | |
153 | ||
154 | /* | |
67cb2b61 | 155 | * Check for a correctable error |
806c35f5 AC |
156 | */ |
157 | if (info->ecc_mode_status & BIT(9)) { | |
158 | error_found = 1; | |
159 | ||
160 | if (handle_errors) { | |
161 | row = info->ecc_mode_status & 0xf; | |
9eb07a7f | 162 | edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, 1, |
de3910eb | 163 | mci->csrows[row]->first_page, 0, 0, |
d8c34af4 | 164 | row, 0, -1, |
03f7eae8 | 165 | mci->ctl_name, ""); |
806c35f5 AC |
166 | } |
167 | } | |
e7ecd891 | 168 | |
806c35f5 AC |
169 | return error_found; |
170 | } | |
171 | ||
172 | /** | |
173 | * amd76x_check - Poll the controller | |
174 | * @mci: Memory controller | |
175 | * | |
176 | * Called by the poll handlers this function reads the status | |
177 | * from the controller and checks for errors. | |
178 | */ | |
806c35f5 AC |
179 | static void amd76x_check(struct mem_ctl_info *mci) |
180 | { | |
181 | struct amd76x_error_info info; | |
956b9ba1 | 182 | edac_dbg(3, "\n"); |
806c35f5 AC |
183 | amd76x_get_error_info(mci, &info); |
184 | amd76x_process_error_info(mci, &info, 1); | |
185 | } | |
186 | ||
13189525 | 187 | static void amd76x_init_csrows(struct mem_ctl_info *mci, struct pci_dev *pdev, |
052dfb45 | 188 | enum edac_type edac_mode) |
13189525 DT |
189 | { |
190 | struct csrow_info *csrow; | |
084a4fcc | 191 | struct dimm_info *dimm; |
13189525 DT |
192 | u32 mba, mba_base, mba_mask, dms; |
193 | int index; | |
194 | ||
195 | for (index = 0; index < mci->nr_csrows; index++) { | |
de3910eb MCC |
196 | csrow = mci->csrows[index]; |
197 | dimm = csrow->channels[0]->dimm; | |
13189525 DT |
198 | |
199 | /* find the DRAM Chip Select Base address and mask */ | |
200 | pci_read_config_dword(pdev, | |
052dfb45 | 201 | AMD76X_MEM_BASE_ADDR + (index * 4), &mba); |
13189525 DT |
202 | |
203 | if (!(mba & BIT(0))) | |
204 | continue; | |
205 | ||
206 | mba_base = mba & 0xff800000UL; | |
207 | mba_mask = ((mba & 0xff80) << 16) | 0x7fffffUL; | |
208 | pci_read_config_dword(pdev, AMD76X_DRAM_MODE_STATUS, &dms); | |
209 | csrow->first_page = mba_base >> PAGE_SHIFT; | |
a895bf8b MCC |
210 | dimm->nr_pages = (mba_mask + 1) >> PAGE_SHIFT; |
211 | csrow->last_page = csrow->first_page + dimm->nr_pages - 1; | |
13189525 | 212 | csrow->page_mask = mba_mask >> PAGE_SHIFT; |
a895bf8b | 213 | dimm->grain = dimm->nr_pages << PAGE_SHIFT; |
084a4fcc MCC |
214 | dimm->mtype = MEM_RDDR; |
215 | dimm->dtype = ((dms >> index) & 0x1) ? DEV_X4 : DEV_UNKNOWN; | |
216 | dimm->edac_mode = edac_mode; | |
13189525 DT |
217 | } |
218 | } | |
219 | ||
806c35f5 AC |
220 | /** |
221 | * amd76x_probe1 - Perform set up for detected device | |
222 | * @pdev; PCI device detected | |
223 | * @dev_idx: Device type index | |
224 | * | |
225 | * We have found an AMD76x and now need to set up the memory | |
226 | * controller status reporting. We configure and set up the | |
227 | * memory controller reporting and claim the device. | |
228 | */ | |
806c35f5 AC |
229 | static int amd76x_probe1(struct pci_dev *pdev, int dev_idx) |
230 | { | |
13189525 | 231 | static const enum edac_type ems_modes[] = { |
806c35f5 AC |
232 | EDAC_NONE, |
233 | EDAC_EC, | |
234 | EDAC_SECDED, | |
235 | EDAC_SECDED | |
236 | }; | |
d8c34af4 MCC |
237 | struct mem_ctl_info *mci; |
238 | struct edac_mc_layer layers[2]; | |
806c35f5 AC |
239 | u32 ems; |
240 | u32 ems_mode; | |
749ede57 | 241 | struct amd76x_error_info discard; |
806c35f5 | 242 | |
956b9ba1 | 243 | edac_dbg(0, "\n"); |
806c35f5 AC |
244 | pci_read_config_dword(pdev, AMD76X_ECC_MODE_STATUS, &ems); |
245 | ems_mode = (ems >> 10) & 0x3; | |
806c35f5 | 246 | |
d8c34af4 MCC |
247 | layers[0].type = EDAC_MC_LAYER_CHIP_SELECT; |
248 | layers[0].size = AMD76X_NR_CSROWS; | |
249 | layers[0].is_virt_csrow = true; | |
250 | layers[1].type = EDAC_MC_LAYER_CHANNEL; | |
251 | layers[1].size = 1; | |
252 | layers[1].is_virt_csrow = false; | |
ca0907b9 | 253 | mci = edac_mc_alloc(0, ARRAY_SIZE(layers), layers, 0); |
d8c34af4 MCC |
254 | |
255 | if (mci == NULL) | |
13189525 | 256 | return -ENOMEM; |
806c35f5 | 257 | |
956b9ba1 | 258 | edac_dbg(0, "mci = %p\n", mci); |
fd687502 | 259 | mci->pdev = &pdev->dev; |
806c35f5 | 260 | mci->mtype_cap = MEM_FLAG_RDDR; |
806c35f5 AC |
261 | mci->edac_ctl_cap = EDAC_FLAG_NONE | EDAC_FLAG_EC | EDAC_FLAG_SECDED; |
262 | mci->edac_cap = ems_mode ? | |
052dfb45 | 263 | (EDAC_FLAG_EC | EDAC_FLAG_SECDED) : EDAC_FLAG_NONE; |
680cbbbb | 264 | mci->mod_name = EDAC_MOD_STR; |
806c35f5 | 265 | mci->ctl_name = amd76x_devs[dev_idx].ctl_name; |
c4192705 | 266 | mci->dev_name = pci_name(pdev); |
806c35f5 AC |
267 | mci->edac_check = amd76x_check; |
268 | mci->ctl_page_to_phys = NULL; | |
269 | ||
13189525 | 270 | amd76x_init_csrows(mci, pdev, ems_modes[ems_mode]); |
67cb2b61 | 271 | amd76x_get_error_info(mci, &discard); /* clear counters */ |
806c35f5 | 272 | |
2d7bbb91 DT |
273 | /* Here we assume that we will never see multiple instances of this |
274 | * type of memory controller. The ID is therefore hardcoded to 0. | |
275 | */ | |
b8f6f975 | 276 | if (edac_mc_add_mc(mci)) { |
956b9ba1 | 277 | edac_dbg(3, "failed edac_mc_add_mc()\n"); |
806c35f5 AC |
278 | goto fail; |
279 | } | |
280 | ||
456a2f95 DJ |
281 | /* allocating generic PCI control info */ |
282 | amd76x_pci = edac_pci_create_generic_ctl(&pdev->dev, EDAC_MOD_STR); | |
283 | if (!amd76x_pci) { | |
284 | printk(KERN_WARNING | |
285 | "%s(): Unable to create PCI control\n", | |
286 | __func__); | |
287 | printk(KERN_WARNING | |
288 | "%s(): PCI error report via EDAC not setup\n", | |
289 | __func__); | |
290 | } | |
291 | ||
806c35f5 | 292 | /* get this far and it's successful */ |
956b9ba1 | 293 | edac_dbg(3, "success\n"); |
806c35f5 AC |
294 | return 0; |
295 | ||
052dfb45 | 296 | fail: |
13189525 DT |
297 | edac_mc_free(mci); |
298 | return -ENODEV; | |
806c35f5 AC |
299 | } |
300 | ||
301 | /* returns count (>= 0), or negative on error */ | |
9b3c6e85 GKH |
302 | static int amd76x_init_one(struct pci_dev *pdev, |
303 | const struct pci_device_id *ent) | |
806c35f5 | 304 | { |
956b9ba1 | 305 | edac_dbg(0, "\n"); |
806c35f5 | 306 | |
ee6583f6 | 307 | /* don't need to call pci_enable_device() */ |
806c35f5 AC |
308 | return amd76x_probe1(pdev, ent->driver_data); |
309 | } | |
310 | ||
806c35f5 AC |
311 | /** |
312 | * amd76x_remove_one - driver shutdown | |
313 | * @pdev: PCI device being handed back | |
314 | * | |
315 | * Called when the driver is unloaded. Find the matching mci | |
316 | * structure for the device then delete the mci and free the | |
317 | * resources. | |
318 | */ | |
9b3c6e85 | 319 | static void amd76x_remove_one(struct pci_dev *pdev) |
806c35f5 AC |
320 | { |
321 | struct mem_ctl_info *mci; | |
322 | ||
956b9ba1 | 323 | edac_dbg(0, "\n"); |
806c35f5 | 324 | |
456a2f95 DJ |
325 | if (amd76x_pci) |
326 | edac_pci_release_generic_ctl(amd76x_pci); | |
327 | ||
37f04581 | 328 | if ((mci = edac_mc_del_mc(&pdev->dev)) == NULL) |
806c35f5 | 329 | return; |
18dbc337 | 330 | |
806c35f5 AC |
331 | edac_mc_free(mci); |
332 | } | |
333 | ||
ba935f40 | 334 | static const struct pci_device_id amd76x_pci_tbl[] = { |
e7ecd891 | 335 | { |
67cb2b61 DT |
336 | PCI_VEND_DEV(AMD, FE_GATE_700C), PCI_ANY_ID, PCI_ANY_ID, 0, 0, |
337 | AMD762}, | |
e7ecd891 | 338 | { |
67cb2b61 DT |
339 | PCI_VEND_DEV(AMD, FE_GATE_700E), PCI_ANY_ID, PCI_ANY_ID, 0, 0, |
340 | AMD761}, | |
e7ecd891 | 341 | { |
67cb2b61 DT |
342 | 0, |
343 | } /* 0 terminated list. */ | |
806c35f5 AC |
344 | }; |
345 | ||
346 | MODULE_DEVICE_TABLE(pci, amd76x_pci_tbl); | |
347 | ||
806c35f5 | 348 | static struct pci_driver amd76x_driver = { |
680cbbbb | 349 | .name = EDAC_MOD_STR, |
806c35f5 | 350 | .probe = amd76x_init_one, |
9b3c6e85 | 351 | .remove = amd76x_remove_one, |
806c35f5 AC |
352 | .id_table = amd76x_pci_tbl, |
353 | }; | |
354 | ||
da9bb1d2 | 355 | static int __init amd76x_init(void) |
806c35f5 | 356 | { |
c3c52bce HM |
357 | /* Ensure that the OPSTATE is set correctly for POLL or NMI */ |
358 | opstate_init(); | |
359 | ||
806c35f5 AC |
360 | return pci_register_driver(&amd76x_driver); |
361 | } | |
362 | ||
363 | static void __exit amd76x_exit(void) | |
364 | { | |
365 | pci_unregister_driver(&amd76x_driver); | |
366 | } | |
367 | ||
368 | module_init(amd76x_init); | |
369 | module_exit(amd76x_exit); | |
370 | ||
371 | MODULE_LICENSE("GPL"); | |
372 | MODULE_AUTHOR("Linux Networx (http://lnxi.com) Thayne Harbaugh"); | |
373 | MODULE_DESCRIPTION("MC support for AMD 76x memory controllers"); | |
c3c52bce HM |
374 | |
375 | module_param(edac_op_state, int, 0444); | |
376 | MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI"); |