1 /* 2 * Intel X38 Memory Controller kernel module 3 * Copyright (C) 2008 Cluster Computing, Inc. 4 * 5 * This file may be distributed under the terms of the 6 * GNU General Public License. 7 * 8 * This file is based on i3200_edac.c 9 * 10 */ 11 12 #include <linux/module.h> 13 #include <linux/init.h> 14 #include <linux/pci.h> 15 #include <linux/pci_ids.h> 16 #include <linux/edac.h> 17 #include "edac_core.h" 18 19 #define X38_REVISION "1.1" 20 21 #define EDAC_MOD_STR "x38_edac" 22 23 #define PCI_DEVICE_ID_INTEL_X38_HB 0x29e0 24 25 #define X38_RANKS 8 26 #define X38_RANKS_PER_CHANNEL 4 27 #define X38_CHANNELS 2 28 29 /* Intel X38 register addresses - device 0 function 0 - DRAM Controller */ 30 31 #define X38_MCHBAR_LOW 0x48 /* MCH Memory Mapped Register BAR */ 32 #define X38_MCHBAR_HIGH 0x4c 33 #define X38_MCHBAR_MASK 0xfffffc000ULL /* bits 35:14 */ 34 #define X38_MMR_WINDOW_SIZE 16384 35 36 #define X38_TOM 0xa0 /* Top of Memory (16b) 37 * 38 * 15:10 reserved 39 * 9:0 total populated physical memory 40 */ 41 #define X38_TOM_MASK 0x3ff /* bits 9:0 */ 42 #define X38_TOM_SHIFT 26 /* 64MiB grain */ 43 44 #define X38_ERRSTS 0xc8 /* Error Status Register (16b) 45 * 46 * 15 reserved 47 * 14 Isochronous TBWRR Run Behind FIFO Full 48 * (ITCV) 49 * 13 Isochronous TBWRR Run Behind FIFO Put 50 * (ITSTV) 51 * 12 reserved 52 * 11 MCH Thermal Sensor Event 53 * for SMI/SCI/SERR (GTSE) 54 * 10 reserved 55 * 9 LOCK to non-DRAM Memory Flag (LCKF) 56 * 8 reserved 57 * 7 DRAM Throttle Flag (DTF) 58 * 6:2 reserved 59 * 1 Multi-bit DRAM ECC Error Flag (DMERR) 60 * 0 Single-bit DRAM ECC Error Flag (DSERR) 61 */ 62 #define X38_ERRSTS_UE 0x0002 63 #define X38_ERRSTS_CE 0x0001 64 #define X38_ERRSTS_BITS (X38_ERRSTS_UE | X38_ERRSTS_CE) 65 66 67 /* Intel MMIO register space - device 0 function 0 - MMR space */ 68 69 #define X38_C0DRB 0x200 /* Channel 0 DRAM Rank Boundary (16b x 4) 70 * 71 * 15:10 reserved 72 * 9:0 Channel 0 DRAM Rank Boundary Address 73 */ 74 #define X38_C1DRB 0x600 /* Channel 1 DRAM Rank Boundary (16b x 4) */ 75 #define X38_DRB_MASK 0x3ff /* bits 9:0 */ 76 #define X38_DRB_SHIFT 26 /* 64MiB grain */ 77 78 #define X38_C0ECCERRLOG 0x280 /* Channel 0 ECC Error Log (64b) 79 * 80 * 63:48 Error Column Address (ERRCOL) 81 * 47:32 Error Row Address (ERRROW) 82 * 31:29 Error Bank Address (ERRBANK) 83 * 28:27 Error Rank Address (ERRRANK) 84 * 26:24 reserved 85 * 23:16 Error Syndrome (ERRSYND) 86 * 15: 2 reserved 87 * 1 Multiple Bit Error Status (MERRSTS) 88 * 0 Correctable Error Status (CERRSTS) 89 */ 90 #define X38_C1ECCERRLOG 0x680 /* Channel 1 ECC Error Log (64b) */ 91 #define X38_ECCERRLOG_CE 0x1 92 #define X38_ECCERRLOG_UE 0x2 93 #define X38_ECCERRLOG_RANK_BITS 0x18000000 94 #define X38_ECCERRLOG_SYNDROME_BITS 0xff0000 95 96 #define X38_CAPID0 0xe0 /* see P.94 of spec for details */ 97 98 static int x38_channel_num; 99 100 static int how_many_channel(struct pci_dev *pdev) 101 { 102 unsigned char capid0_8b; /* 8th byte of CAPID0 */ 103 104 pci_read_config_byte(pdev, X38_CAPID0 + 8, &capid0_8b); 105 if (capid0_8b & 0x20) { /* check DCD: Dual Channel Disable */ 106 edac_dbg(0, "In single channel mode\n"); 107 x38_channel_num = 1; 108 } else { 109 edac_dbg(0, "In dual channel mode\n"); 110 x38_channel_num = 2; 111 } 112 113 return x38_channel_num; 114 } 115 116 static unsigned long eccerrlog_syndrome(u64 log) 117 { 118 return (log & X38_ECCERRLOG_SYNDROME_BITS) >> 16; 119 } 120 121 static int eccerrlog_row(int channel, u64 log) 122 { 123 return ((log & X38_ECCERRLOG_RANK_BITS) >> 27) | 124 (channel * X38_RANKS_PER_CHANNEL); 125 } 126 127 enum x38_chips { 128 X38 = 0, 129 }; 130 131 struct x38_dev_info { 132 const char *ctl_name; 133 }; 134 135 struct x38_error_info { 136 u16 errsts; 137 u16 errsts2; 138 u64 eccerrlog[X38_CHANNELS]; 139 }; 140 141 static const struct x38_dev_info x38_devs[] = { 142 [X38] = { 143 .ctl_name = "x38"}, 144 }; 145 146 static struct pci_dev *mci_pdev; 147 static int x38_registered = 1; 148 149 150 static void x38_clear_error_info(struct mem_ctl_info *mci) 151 { 152 struct pci_dev *pdev; 153 154 pdev = to_pci_dev(mci->pdev); 155 156 /* 157 * Clear any error bits. 158 * (Yes, we really clear bits by writing 1 to them.) 159 */ 160 pci_write_bits16(pdev, X38_ERRSTS, X38_ERRSTS_BITS, 161 X38_ERRSTS_BITS); 162 } 163 164 static u64 x38_readq(const void __iomem *addr) 165 { 166 return readl(addr) | (((u64)readl(addr + 4)) << 32); 167 } 168 169 static void x38_get_and_clear_error_info(struct mem_ctl_info *mci, 170 struct x38_error_info *info) 171 { 172 struct pci_dev *pdev; 173 void __iomem *window = mci->pvt_info; 174 175 pdev = to_pci_dev(mci->pdev); 176 177 /* 178 * This is a mess because there is no atomic way to read all the 179 * registers at once and the registers can transition from CE being 180 * overwritten by UE. 181 */ 182 pci_read_config_word(pdev, X38_ERRSTS, &info->errsts); 183 if (!(info->errsts & X38_ERRSTS_BITS)) 184 return; 185 186 info->eccerrlog[0] = x38_readq(window + X38_C0ECCERRLOG); 187 if (x38_channel_num == 2) 188 info->eccerrlog[1] = x38_readq(window + X38_C1ECCERRLOG); 189 190 pci_read_config_word(pdev, X38_ERRSTS, &info->errsts2); 191 192 /* 193 * If the error is the same for both reads then the first set 194 * of reads is valid. If there is a change then there is a CE 195 * with no info and the second set of reads is valid and 196 * should be UE info. 197 */ 198 if ((info->errsts ^ info->errsts2) & X38_ERRSTS_BITS) { 199 info->eccerrlog[0] = x38_readq(window + X38_C0ECCERRLOG); 200 if (x38_channel_num == 2) 201 info->eccerrlog[1] = 202 x38_readq(window + X38_C1ECCERRLOG); 203 } 204 205 x38_clear_error_info(mci); 206 } 207 208 static void x38_process_error_info(struct mem_ctl_info *mci, 209 struct x38_error_info *info) 210 { 211 int channel; 212 u64 log; 213 214 if (!(info->errsts & X38_ERRSTS_BITS)) 215 return; 216 217 if ((info->errsts ^ info->errsts2) & X38_ERRSTS_BITS) { 218 edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci, 1, 0, 0, 0, 219 -1, -1, -1, 220 "UE overwrote CE", ""); 221 info->errsts = info->errsts2; 222 } 223 224 for (channel = 0; channel < x38_channel_num; channel++) { 225 log = info->eccerrlog[channel]; 226 if (log & X38_ECCERRLOG_UE) { 227 edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci, 1, 228 0, 0, 0, 229 eccerrlog_row(channel, log), 230 -1, -1, 231 "x38 UE", ""); 232 } else if (log & X38_ECCERRLOG_CE) { 233 edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, 1, 234 0, 0, eccerrlog_syndrome(log), 235 eccerrlog_row(channel, log), 236 -1, -1, 237 "x38 CE", ""); 238 } 239 } 240 } 241 242 static void x38_check(struct mem_ctl_info *mci) 243 { 244 struct x38_error_info info; 245 246 edac_dbg(1, "MC%d\n", mci->mc_idx); 247 x38_get_and_clear_error_info(mci, &info); 248 x38_process_error_info(mci, &info); 249 } 250 251 252 void __iomem *x38_map_mchbar(struct pci_dev *pdev) 253 { 254 union { 255 u64 mchbar; 256 struct { 257 u32 mchbar_low; 258 u32 mchbar_high; 259 }; 260 } u; 261 void __iomem *window; 262 263 pci_read_config_dword(pdev, X38_MCHBAR_LOW, &u.mchbar_low); 264 pci_write_config_dword(pdev, X38_MCHBAR_LOW, u.mchbar_low | 0x1); 265 pci_read_config_dword(pdev, X38_MCHBAR_HIGH, &u.mchbar_high); 266 u.mchbar &= X38_MCHBAR_MASK; 267 268 if (u.mchbar != (resource_size_t)u.mchbar) { 269 printk(KERN_ERR 270 "x38: mmio space beyond accessible range (0x%llx)\n", 271 (unsigned long long)u.mchbar); 272 return NULL; 273 } 274 275 window = ioremap_nocache(u.mchbar, X38_MMR_WINDOW_SIZE); 276 if (!window) 277 printk(KERN_ERR "x38: cannot map mmio space at 0x%llx\n", 278 (unsigned long long)u.mchbar); 279 280 return window; 281 } 282 283 284 static void x38_get_drbs(void __iomem *window, 285 u16 drbs[X38_CHANNELS][X38_RANKS_PER_CHANNEL]) 286 { 287 int i; 288 289 for (i = 0; i < X38_RANKS_PER_CHANNEL; i++) { 290 drbs[0][i] = readw(window + X38_C0DRB + 2*i) & X38_DRB_MASK; 291 drbs[1][i] = readw(window + X38_C1DRB + 2*i) & X38_DRB_MASK; 292 } 293 } 294 295 static bool x38_is_stacked(struct pci_dev *pdev, 296 u16 drbs[X38_CHANNELS][X38_RANKS_PER_CHANNEL]) 297 { 298 u16 tom; 299 300 pci_read_config_word(pdev, X38_TOM, &tom); 301 tom &= X38_TOM_MASK; 302 303 return drbs[X38_CHANNELS - 1][X38_RANKS_PER_CHANNEL - 1] == tom; 304 } 305 306 static unsigned long drb_to_nr_pages( 307 u16 drbs[X38_CHANNELS][X38_RANKS_PER_CHANNEL], 308 bool stacked, int channel, int rank) 309 { 310 int n; 311 312 n = drbs[channel][rank]; 313 if (rank > 0) 314 n -= drbs[channel][rank - 1]; 315 if (stacked && (channel == 1) && drbs[channel][rank] == 316 drbs[channel][X38_RANKS_PER_CHANNEL - 1]) { 317 n -= drbs[0][X38_RANKS_PER_CHANNEL - 1]; 318 } 319 320 n <<= (X38_DRB_SHIFT - PAGE_SHIFT); 321 return n; 322 } 323 324 static int x38_probe1(struct pci_dev *pdev, int dev_idx) 325 { 326 int rc; 327 int i, j; 328 struct mem_ctl_info *mci = NULL; 329 struct edac_mc_layer layers[2]; 330 u16 drbs[X38_CHANNELS][X38_RANKS_PER_CHANNEL]; 331 bool stacked; 332 void __iomem *window; 333 334 edac_dbg(0, "MC:\n"); 335 336 window = x38_map_mchbar(pdev); 337 if (!window) 338 return -ENODEV; 339 340 x38_get_drbs(window, drbs); 341 342 how_many_channel(pdev); 343 344 /* FIXME: unconventional pvt_info usage */ 345 layers[0].type = EDAC_MC_LAYER_CHIP_SELECT; 346 layers[0].size = X38_RANKS; 347 layers[0].is_virt_csrow = true; 348 layers[1].type = EDAC_MC_LAYER_CHANNEL; 349 layers[1].size = x38_channel_num; 350 layers[1].is_virt_csrow = false; 351 mci = edac_mc_alloc(0, ARRAY_SIZE(layers), layers, 0); 352 if (!mci) 353 return -ENOMEM; 354 355 edac_dbg(3, "MC: init mci\n"); 356 357 mci->pdev = &pdev->dev; 358 mci->mtype_cap = MEM_FLAG_DDR2; 359 360 mci->edac_ctl_cap = EDAC_FLAG_SECDED; 361 mci->edac_cap = EDAC_FLAG_SECDED; 362 363 mci->mod_name = EDAC_MOD_STR; 364 mci->mod_ver = X38_REVISION; 365 mci->ctl_name = x38_devs[dev_idx].ctl_name; 366 mci->dev_name = pci_name(pdev); 367 mci->edac_check = x38_check; 368 mci->ctl_page_to_phys = NULL; 369 mci->pvt_info = window; 370 371 stacked = x38_is_stacked(pdev, drbs); 372 373 /* 374 * The dram rank boundary (DRB) reg values are boundary addresses 375 * for each DRAM rank with a granularity of 64MB. DRB regs are 376 * cumulative; the last one will contain the total memory 377 * contained in all ranks. 378 */ 379 for (i = 0; i < mci->nr_csrows; i++) { 380 unsigned long nr_pages; 381 struct csrow_info *csrow = mci->csrows[i]; 382 383 nr_pages = drb_to_nr_pages(drbs, stacked, 384 i / X38_RANKS_PER_CHANNEL, 385 i % X38_RANKS_PER_CHANNEL); 386 387 if (nr_pages == 0) 388 continue; 389 390 for (j = 0; j < x38_channel_num; j++) { 391 struct dimm_info *dimm = csrow->channels[j]->dimm; 392 393 dimm->nr_pages = nr_pages / x38_channel_num; 394 dimm->grain = nr_pages << PAGE_SHIFT; 395 dimm->mtype = MEM_DDR2; 396 dimm->dtype = DEV_UNKNOWN; 397 dimm->edac_mode = EDAC_UNKNOWN; 398 } 399 } 400 401 x38_clear_error_info(mci); 402 403 rc = -ENODEV; 404 if (edac_mc_add_mc(mci)) { 405 edac_dbg(3, "MC: failed edac_mc_add_mc()\n"); 406 goto fail; 407 } 408 409 /* get this far and it's successful */ 410 edac_dbg(3, "MC: success\n"); 411 return 0; 412 413 fail: 414 iounmap(window); 415 if (mci) 416 edac_mc_free(mci); 417 418 return rc; 419 } 420 421 static int __devinit x38_init_one(struct pci_dev *pdev, 422 const struct pci_device_id *ent) 423 { 424 int rc; 425 426 edac_dbg(0, "MC:\n"); 427 428 if (pci_enable_device(pdev) < 0) 429 return -EIO; 430 431 rc = x38_probe1(pdev, ent->driver_data); 432 if (!mci_pdev) 433 mci_pdev = pci_dev_get(pdev); 434 435 return rc; 436 } 437 438 static void __devexit x38_remove_one(struct pci_dev *pdev) 439 { 440 struct mem_ctl_info *mci; 441 442 edac_dbg(0, "\n"); 443 444 mci = edac_mc_del_mc(&pdev->dev); 445 if (!mci) 446 return; 447 448 iounmap(mci->pvt_info); 449 450 edac_mc_free(mci); 451 } 452 453 static DEFINE_PCI_DEVICE_TABLE(x38_pci_tbl) = { 454 { 455 PCI_VEND_DEV(INTEL, X38_HB), PCI_ANY_ID, PCI_ANY_ID, 0, 0, 456 X38}, 457 { 458 0, 459 } /* 0 terminated list. */ 460 }; 461 462 MODULE_DEVICE_TABLE(pci, x38_pci_tbl); 463 464 static struct pci_driver x38_driver = { 465 .name = EDAC_MOD_STR, 466 .probe = x38_init_one, 467 .remove = __devexit_p(x38_remove_one), 468 .id_table = x38_pci_tbl, 469 }; 470 471 static int __init x38_init(void) 472 { 473 int pci_rc; 474 475 edac_dbg(3, "MC:\n"); 476 477 /* Ensure that the OPSTATE is set correctly for POLL or NMI */ 478 opstate_init(); 479 480 pci_rc = pci_register_driver(&x38_driver); 481 if (pci_rc < 0) 482 goto fail0; 483 484 if (!mci_pdev) { 485 x38_registered = 0; 486 mci_pdev = pci_get_device(PCI_VENDOR_ID_INTEL, 487 PCI_DEVICE_ID_INTEL_X38_HB, NULL); 488 if (!mci_pdev) { 489 edac_dbg(0, "x38 pci_get_device fail\n"); 490 pci_rc = -ENODEV; 491 goto fail1; 492 } 493 494 pci_rc = x38_init_one(mci_pdev, x38_pci_tbl); 495 if (pci_rc < 0) { 496 edac_dbg(0, "x38 init fail\n"); 497 pci_rc = -ENODEV; 498 goto fail1; 499 } 500 } 501 502 return 0; 503 504 fail1: 505 pci_unregister_driver(&x38_driver); 506 507 fail0: 508 if (mci_pdev) 509 pci_dev_put(mci_pdev); 510 511 return pci_rc; 512 } 513 514 static void __exit x38_exit(void) 515 { 516 edac_dbg(3, "MC:\n"); 517 518 pci_unregister_driver(&x38_driver); 519 if (!x38_registered) { 520 x38_remove_one(mci_pdev); 521 pci_dev_put(mci_pdev); 522 } 523 } 524 525 module_init(x38_init); 526 module_exit(x38_exit); 527 528 MODULE_LICENSE("GPL"); 529 MODULE_AUTHOR("Cluster Computing, Inc. Hitoshi Mitake"); 530 MODULE_DESCRIPTION("MC support for Intel X38 memory hub controllers"); 531 532 module_param(edac_op_state, int, 0444); 533 MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI"); 534