Qiuxu Zhuo | 98f2fc8 | 2019-01-30 11:15:18 -0800 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | /* |
| 3 | * EDAC driver for Intel(R) Xeon(R) Skylake processors |
| 4 | * Copyright (c) 2016, Intel Corporation. |
| 5 | */ |
| 6 | |
| 7 | #include <linux/kernel.h> |
| 8 | #include <linux/processor.h> |
| 9 | #include <asm/cpu_device_id.h> |
| 10 | #include <asm/intel-family.h> |
| 11 | #include <asm/mce.h> |
| 12 | |
| 13 | #include "edac_module.h" |
| 14 | #include "skx_common.h" |
| 15 | |
| 16 | #define EDAC_MOD_STR "skx_edac" |
| 17 | |
| 18 | /* |
| 19 | * Debug macros |
| 20 | */ |
| 21 | #define skx_printk(level, fmt, arg...) \ |
| 22 | edac_printk(level, "skx", fmt, ##arg) |
| 23 | |
| 24 | #define skx_mc_printk(mci, level, fmt, arg...) \ |
| 25 | edac_mc_chipset_printk(mci, level, "skx", fmt, ##arg) |
| 26 | |
| 27 | static struct list_head *skx_edac_list; |
| 28 | |
| 29 | static u64 skx_tolm, skx_tohm; |
| 30 | static int skx_num_sockets; |
| 31 | static unsigned int nvdimm_count; |
| 32 | |
| 33 | #define MASK26 0x3FFFFFF /* Mask for 2^26 */ |
| 34 | #define MASK29 0x1FFFFFFF /* Mask for 2^29 */ |
| 35 | |
| 36 | static struct skx_dev *get_skx_dev(struct pci_bus *bus, u8 idx) |
| 37 | { |
| 38 | struct skx_dev *d; |
| 39 | |
| 40 | list_for_each_entry(d, skx_edac_list, list) { |
| 41 | if (d->seg == pci_domain_nr(bus) && d->bus[idx] == bus->number) |
| 42 | return d; |
| 43 | } |
| 44 | |
| 45 | return NULL; |
| 46 | } |
| 47 | |
| 48 | enum munittype { |
Tony Luck | e80634a | 2019-08-15 14:53:28 -0700 | [diff] [blame] | 49 | CHAN0, CHAN1, CHAN2, SAD_ALL, UTIL_ALL, SAD, |
| 50 | ERRCHAN0, ERRCHAN1, ERRCHAN2, |
Qiuxu Zhuo | 98f2fc8 | 2019-01-30 11:15:18 -0800 | [diff] [blame] | 51 | }; |
| 52 | |
| 53 | struct munit { |
| 54 | u16 did; |
| 55 | u16 devfn[SKX_NUM_IMC]; |
| 56 | u8 busidx; |
| 57 | u8 per_socket; |
| 58 | enum munittype mtype; |
| 59 | }; |
| 60 | |
| 61 | /* |
| 62 | * List of PCI device ids that we need together with some device |
| 63 | * number and function numbers to tell which memory controller the |
| 64 | * device belongs to. |
| 65 | */ |
| 66 | static const struct munit skx_all_munits[] = { |
| 67 | { 0x2054, { }, 1, 1, SAD_ALL }, |
| 68 | { 0x2055, { }, 1, 1, UTIL_ALL }, |
| 69 | { 0x2040, { PCI_DEVFN(10, 0), PCI_DEVFN(12, 0) }, 2, 2, CHAN0 }, |
| 70 | { 0x2044, { PCI_DEVFN(10, 4), PCI_DEVFN(12, 4) }, 2, 2, CHAN1 }, |
| 71 | { 0x2048, { PCI_DEVFN(11, 0), PCI_DEVFN(13, 0) }, 2, 2, CHAN2 }, |
Tony Luck | e80634a | 2019-08-15 14:53:28 -0700 | [diff] [blame] | 72 | { 0x2043, { PCI_DEVFN(10, 3), PCI_DEVFN(12, 3) }, 2, 2, ERRCHAN0 }, |
| 73 | { 0x2047, { PCI_DEVFN(10, 7), PCI_DEVFN(12, 7) }, 2, 2, ERRCHAN1 }, |
| 74 | { 0x204b, { PCI_DEVFN(11, 3), PCI_DEVFN(13, 3) }, 2, 2, ERRCHAN2 }, |
Qiuxu Zhuo | 98f2fc8 | 2019-01-30 11:15:18 -0800 | [diff] [blame] | 75 | { 0x208e, { }, 1, 0, SAD }, |
| 76 | { } |
| 77 | }; |
| 78 | |
| 79 | static int get_all_munits(const struct munit *m) |
| 80 | { |
| 81 | struct pci_dev *pdev, *prev; |
| 82 | struct skx_dev *d; |
| 83 | u32 reg; |
| 84 | int i = 0, ndev = 0; |
| 85 | |
| 86 | prev = NULL; |
| 87 | for (;;) { |
| 88 | pdev = pci_get_device(PCI_VENDOR_ID_INTEL, m->did, prev); |
| 89 | if (!pdev) |
| 90 | break; |
| 91 | ndev++; |
| 92 | if (m->per_socket == SKX_NUM_IMC) { |
| 93 | for (i = 0; i < SKX_NUM_IMC; i++) |
| 94 | if (m->devfn[i] == pdev->devfn) |
| 95 | break; |
| 96 | if (i == SKX_NUM_IMC) |
| 97 | goto fail; |
| 98 | } |
| 99 | d = get_skx_dev(pdev->bus, m->busidx); |
| 100 | if (!d) |
| 101 | goto fail; |
| 102 | |
| 103 | /* Be sure that the device is enabled */ |
| 104 | if (unlikely(pci_enable_device(pdev) < 0)) { |
| 105 | skx_printk(KERN_ERR, "Couldn't enable device %04x:%04x\n", |
| 106 | PCI_VENDOR_ID_INTEL, m->did); |
| 107 | goto fail; |
| 108 | } |
| 109 | |
| 110 | switch (m->mtype) { |
Tony Luck | e80634a | 2019-08-15 14:53:28 -0700 | [diff] [blame] | 111 | case CHAN0: |
| 112 | case CHAN1: |
| 113 | case CHAN2: |
Qiuxu Zhuo | 98f2fc8 | 2019-01-30 11:15:18 -0800 | [diff] [blame] | 114 | pci_dev_get(pdev); |
| 115 | d->imc[i].chan[m->mtype].cdev = pdev; |
| 116 | break; |
Tony Luck | e80634a | 2019-08-15 14:53:28 -0700 | [diff] [blame] | 117 | case ERRCHAN0: |
| 118 | case ERRCHAN1: |
| 119 | case ERRCHAN2: |
| 120 | pci_dev_get(pdev); |
| 121 | d->imc[i].chan[m->mtype - ERRCHAN0].edev = pdev; |
| 122 | break; |
Qiuxu Zhuo | 98f2fc8 | 2019-01-30 11:15:18 -0800 | [diff] [blame] | 123 | case SAD_ALL: |
| 124 | pci_dev_get(pdev); |
| 125 | d->sad_all = pdev; |
| 126 | break; |
| 127 | case UTIL_ALL: |
| 128 | pci_dev_get(pdev); |
| 129 | d->util_all = pdev; |
| 130 | break; |
| 131 | case SAD: |
| 132 | /* |
| 133 | * one of these devices per core, including cores |
| 134 | * that don't exist on this SKU. Ignore any that |
| 135 | * read a route table of zero, make sure all the |
| 136 | * non-zero values match. |
| 137 | */ |
| 138 | pci_read_config_dword(pdev, 0xB4, ®); |
| 139 | if (reg != 0) { |
| 140 | if (d->mcroute == 0) { |
| 141 | d->mcroute = reg; |
| 142 | } else if (d->mcroute != reg) { |
| 143 | skx_printk(KERN_ERR, "mcroute mismatch\n"); |
| 144 | goto fail; |
| 145 | } |
| 146 | } |
| 147 | ndev--; |
| 148 | break; |
| 149 | } |
| 150 | |
| 151 | prev = pdev; |
| 152 | } |
| 153 | |
| 154 | return ndev; |
| 155 | fail: |
| 156 | pci_dev_put(pdev); |
| 157 | return -ENODEV; |
| 158 | } |
| 159 | |
Qiuxu Zhuo | ee5340a | 2020-04-24 20:18:14 +0800 | [diff] [blame] | 160 | static struct res_config skx_cfg = { |
| 161 | .type = SKX, |
| 162 | .decs_did = 0x2016, |
| 163 | .busno_cfg_offset = 0xcc, |
| 164 | }; |
| 165 | |
Qiuxu Zhuo | 98f2fc8 | 2019-01-30 11:15:18 -0800 | [diff] [blame] | 166 | static const struct x86_cpu_id skx_cpuids[] = { |
Qiuxu Zhuo | ee5340a | 2020-04-24 20:18:14 +0800 | [diff] [blame] | 167 | X86_MATCH_INTEL_FAM6_MODEL(SKYLAKE_X, &skx_cfg), |
Qiuxu Zhuo | 98f2fc8 | 2019-01-30 11:15:18 -0800 | [diff] [blame] | 168 | { } |
| 169 | }; |
| 170 | MODULE_DEVICE_TABLE(x86cpu, skx_cpuids); |
| 171 | |
Qiuxu Zhuo | 1032095 | 2020-05-15 20:34:06 +0800 | [diff] [blame] | 172 | static bool skx_check_ecc(u32 mcmtr) |
Qiuxu Zhuo | 98f2fc8 | 2019-01-30 11:15:18 -0800 | [diff] [blame] | 173 | { |
Qiuxu Zhuo | 1032095 | 2020-05-15 20:34:06 +0800 | [diff] [blame] | 174 | return !!GET_BITFIELD(mcmtr, 2, 2); |
Qiuxu Zhuo | 98f2fc8 | 2019-01-30 11:15:18 -0800 | [diff] [blame] | 175 | } |
| 176 | |
| 177 | static int skx_get_dimm_config(struct mem_ctl_info *mci) |
| 178 | { |
| 179 | struct skx_pvt *pvt = mci->pvt_info; |
Qiuxu Zhuo | 1032095 | 2020-05-15 20:34:06 +0800 | [diff] [blame] | 180 | u32 mtr, mcmtr, amap, mcddrtcfg; |
Qiuxu Zhuo | 98f2fc8 | 2019-01-30 11:15:18 -0800 | [diff] [blame] | 181 | struct skx_imc *imc = pvt->imc; |
Qiuxu Zhuo | 98f2fc8 | 2019-01-30 11:15:18 -0800 | [diff] [blame] | 182 | struct dimm_info *dimm; |
| 183 | int i, j; |
| 184 | int ndimms; |
| 185 | |
Qiuxu Zhuo | 1032095 | 2020-05-15 20:34:06 +0800 | [diff] [blame] | 186 | /* Only the mcmtr on the first channel is effective */ |
| 187 | pci_read_config_dword(imc->chan[0].cdev, 0x87c, &mcmtr); |
| 188 | |
Qiuxu Zhuo | 98f2fc8 | 2019-01-30 11:15:18 -0800 | [diff] [blame] | 189 | for (i = 0; i < SKX_NUM_CHANNELS; i++) { |
| 190 | ndimms = 0; |
| 191 | pci_read_config_dword(imc->chan[i].cdev, 0x8C, &amap); |
| 192 | pci_read_config_dword(imc->chan[i].cdev, 0x400, &mcddrtcfg); |
| 193 | for (j = 0; j < SKX_NUM_DIMMS; j++) { |
Robert Richter | bc9ad9e | 2019-11-06 09:33:02 +0000 | [diff] [blame] | 194 | dimm = edac_get_dimm(mci, i, j, 0); |
Qiuxu Zhuo | 98f2fc8 | 2019-01-30 11:15:18 -0800 | [diff] [blame] | 195 | pci_read_config_dword(imc->chan[i].cdev, |
| 196 | 0x80 + 4 * j, &mtr); |
| 197 | if (IS_DIMM_PRESENT(mtr)) { |
Qiuxu Zhuo | 1032095 | 2020-05-15 20:34:06 +0800 | [diff] [blame] | 198 | ndimms += skx_get_dimm_info(mtr, mcmtr, amap, dimm, imc, i, j); |
Qiuxu Zhuo | 98f2fc8 | 2019-01-30 11:15:18 -0800 | [diff] [blame] | 199 | } else if (IS_NVDIMM_PRESENT(mcddrtcfg, j)) { |
| 200 | ndimms += skx_get_nvdimm_info(dimm, imc, i, j, |
| 201 | EDAC_MOD_STR); |
| 202 | nvdimm_count++; |
| 203 | } |
| 204 | } |
Qiuxu Zhuo | 1032095 | 2020-05-15 20:34:06 +0800 | [diff] [blame] | 205 | if (ndimms && !skx_check_ecc(mcmtr)) { |
Qiuxu Zhuo | 98f2fc8 | 2019-01-30 11:15:18 -0800 | [diff] [blame] | 206 | skx_printk(KERN_ERR, "ECC is disabled on imc %d\n", imc->mc); |
| 207 | return -ENODEV; |
| 208 | } |
| 209 | } |
| 210 | |
| 211 | return 0; |
| 212 | } |
| 213 | |
| 214 | #define SKX_MAX_SAD 24 |
| 215 | |
| 216 | #define SKX_GET_SAD(d, i, reg) \ |
| 217 | pci_read_config_dword((d)->sad_all, 0x60 + 8 * (i), &(reg)) |
| 218 | #define SKX_GET_ILV(d, i, reg) \ |
| 219 | pci_read_config_dword((d)->sad_all, 0x64 + 8 * (i), &(reg)) |
| 220 | |
| 221 | #define SKX_SAD_MOD3MODE(sad) GET_BITFIELD((sad), 30, 31) |
| 222 | #define SKX_SAD_MOD3(sad) GET_BITFIELD((sad), 27, 27) |
| 223 | #define SKX_SAD_LIMIT(sad) (((u64)GET_BITFIELD((sad), 7, 26) << 26) | MASK26) |
| 224 | #define SKX_SAD_MOD3ASMOD2(sad) GET_BITFIELD((sad), 5, 6) |
| 225 | #define SKX_SAD_ATTR(sad) GET_BITFIELD((sad), 3, 4) |
| 226 | #define SKX_SAD_INTERLEAVE(sad) GET_BITFIELD((sad), 1, 2) |
| 227 | #define SKX_SAD_ENABLE(sad) GET_BITFIELD((sad), 0, 0) |
| 228 | |
| 229 | #define SKX_ILV_REMOTE(tgt) (((tgt) & 8) == 0) |
| 230 | #define SKX_ILV_TARGET(tgt) ((tgt) & 7) |
| 231 | |
Tony Luck | e80634a | 2019-08-15 14:53:28 -0700 | [diff] [blame] | 232 | static void skx_show_retry_rd_err_log(struct decoded_addr *res, |
| 233 | char *msg, int len) |
| 234 | { |
| 235 | u32 log0, log1, log2, log3, log4; |
| 236 | u32 corr0, corr1, corr2, corr3; |
| 237 | struct pci_dev *edev; |
| 238 | int n; |
| 239 | |
| 240 | edev = res->dev->imc[res->imc].chan[res->channel].edev; |
| 241 | |
| 242 | pci_read_config_dword(edev, 0x154, &log0); |
| 243 | pci_read_config_dword(edev, 0x148, &log1); |
| 244 | pci_read_config_dword(edev, 0x150, &log2); |
| 245 | pci_read_config_dword(edev, 0x15c, &log3); |
| 246 | pci_read_config_dword(edev, 0x114, &log4); |
| 247 | |
| 248 | n = snprintf(msg, len, " retry_rd_err_log[%.8x %.8x %.8x %.8x %.8x]", |
| 249 | log0, log1, log2, log3, log4); |
| 250 | |
| 251 | pci_read_config_dword(edev, 0x104, &corr0); |
| 252 | pci_read_config_dword(edev, 0x108, &corr1); |
| 253 | pci_read_config_dword(edev, 0x10c, &corr2); |
| 254 | pci_read_config_dword(edev, 0x110, &corr3); |
| 255 | |
| 256 | if (len - n > 0) |
| 257 | snprintf(msg + n, len - n, |
| 258 | " correrrcnt[%.4x %.4x %.4x %.4x %.4x %.4x %.4x %.4x]", |
| 259 | corr0 & 0xffff, corr0 >> 16, |
| 260 | corr1 & 0xffff, corr1 >> 16, |
| 261 | corr2 & 0xffff, corr2 >> 16, |
| 262 | corr3 & 0xffff, corr3 >> 16); |
| 263 | } |
| 264 | |
Qiuxu Zhuo | 98f2fc8 | 2019-01-30 11:15:18 -0800 | [diff] [blame] | 265 | static bool skx_sad_decode(struct decoded_addr *res) |
| 266 | { |
| 267 | struct skx_dev *d = list_first_entry(skx_edac_list, typeof(*d), list); |
| 268 | u64 addr = res->addr; |
| 269 | int i, idx, tgt, lchan, shift; |
| 270 | u32 sad, ilv; |
| 271 | u64 limit, prev_limit; |
| 272 | int remote = 0; |
| 273 | |
| 274 | /* Simple sanity check for I/O space or out of range */ |
| 275 | if (addr >= skx_tohm || (addr >= skx_tolm && addr < BIT_ULL(32))) { |
| 276 | edac_dbg(0, "Address 0x%llx out of range\n", addr); |
| 277 | return false; |
| 278 | } |
| 279 | |
| 280 | restart: |
| 281 | prev_limit = 0; |
| 282 | for (i = 0; i < SKX_MAX_SAD; i++) { |
| 283 | SKX_GET_SAD(d, i, sad); |
| 284 | limit = SKX_SAD_LIMIT(sad); |
| 285 | if (SKX_SAD_ENABLE(sad)) { |
| 286 | if (addr >= prev_limit && addr <= limit) |
| 287 | goto sad_found; |
| 288 | } |
| 289 | prev_limit = limit + 1; |
| 290 | } |
| 291 | edac_dbg(0, "No SAD entry for 0x%llx\n", addr); |
| 292 | return false; |
| 293 | |
| 294 | sad_found: |
| 295 | SKX_GET_ILV(d, i, ilv); |
| 296 | |
| 297 | switch (SKX_SAD_INTERLEAVE(sad)) { |
| 298 | case 0: |
| 299 | idx = GET_BITFIELD(addr, 6, 8); |
| 300 | break; |
| 301 | case 1: |
| 302 | idx = GET_BITFIELD(addr, 8, 10); |
| 303 | break; |
| 304 | case 2: |
| 305 | idx = GET_BITFIELD(addr, 12, 14); |
| 306 | break; |
| 307 | case 3: |
| 308 | idx = GET_BITFIELD(addr, 30, 32); |
| 309 | break; |
| 310 | } |
| 311 | |
| 312 | tgt = GET_BITFIELD(ilv, 4 * idx, 4 * idx + 3); |
| 313 | |
| 314 | /* If point to another node, find it and start over */ |
| 315 | if (SKX_ILV_REMOTE(tgt)) { |
| 316 | if (remote) { |
| 317 | edac_dbg(0, "Double remote!\n"); |
| 318 | return false; |
| 319 | } |
| 320 | remote = 1; |
| 321 | list_for_each_entry(d, skx_edac_list, list) { |
| 322 | if (d->imc[0].src_id == SKX_ILV_TARGET(tgt)) |
| 323 | goto restart; |
| 324 | } |
| 325 | edac_dbg(0, "Can't find node %d\n", SKX_ILV_TARGET(tgt)); |
| 326 | return false; |
| 327 | } |
| 328 | |
| 329 | if (SKX_SAD_MOD3(sad) == 0) { |
| 330 | lchan = SKX_ILV_TARGET(tgt); |
| 331 | } else { |
| 332 | switch (SKX_SAD_MOD3MODE(sad)) { |
| 333 | case 0: |
| 334 | shift = 6; |
| 335 | break; |
| 336 | case 1: |
| 337 | shift = 8; |
| 338 | break; |
| 339 | case 2: |
| 340 | shift = 12; |
| 341 | break; |
| 342 | default: |
| 343 | edac_dbg(0, "illegal mod3mode\n"); |
| 344 | return false; |
| 345 | } |
| 346 | switch (SKX_SAD_MOD3ASMOD2(sad)) { |
| 347 | case 0: |
| 348 | lchan = (addr >> shift) % 3; |
| 349 | break; |
| 350 | case 1: |
| 351 | lchan = (addr >> shift) % 2; |
| 352 | break; |
| 353 | case 2: |
| 354 | lchan = (addr >> shift) % 2; |
| 355 | lchan = (lchan << 1) | !lchan; |
| 356 | break; |
| 357 | case 3: |
| 358 | lchan = ((addr >> shift) % 2) << 1; |
| 359 | break; |
| 360 | } |
| 361 | lchan = (lchan << 1) | (SKX_ILV_TARGET(tgt) & 1); |
| 362 | } |
| 363 | |
| 364 | res->dev = d; |
| 365 | res->socket = d->imc[0].src_id; |
| 366 | res->imc = GET_BITFIELD(d->mcroute, lchan * 3, lchan * 3 + 2); |
| 367 | res->channel = GET_BITFIELD(d->mcroute, lchan * 2 + 18, lchan * 2 + 19); |
| 368 | |
| 369 | edac_dbg(2, "0x%llx: socket=%d imc=%d channel=%d\n", |
| 370 | res->addr, res->socket, res->imc, res->channel); |
| 371 | return true; |
| 372 | } |
| 373 | |
| 374 | #define SKX_MAX_TAD 8 |
| 375 | |
| 376 | #define SKX_GET_TADBASE(d, mc, i, reg) \ |
| 377 | pci_read_config_dword((d)->imc[mc].chan[0].cdev, 0x850 + 4 * (i), &(reg)) |
| 378 | #define SKX_GET_TADWAYNESS(d, mc, i, reg) \ |
| 379 | pci_read_config_dword((d)->imc[mc].chan[0].cdev, 0x880 + 4 * (i), &(reg)) |
| 380 | #define SKX_GET_TADCHNILVOFFSET(d, mc, ch, i, reg) \ |
| 381 | pci_read_config_dword((d)->imc[mc].chan[ch].cdev, 0x90 + 4 * (i), &(reg)) |
| 382 | |
| 383 | #define SKX_TAD_BASE(b) ((u64)GET_BITFIELD((b), 12, 31) << 26) |
| 384 | #define SKX_TAD_SKT_GRAN(b) GET_BITFIELD((b), 4, 5) |
| 385 | #define SKX_TAD_CHN_GRAN(b) GET_BITFIELD((b), 6, 7) |
| 386 | #define SKX_TAD_LIMIT(b) (((u64)GET_BITFIELD((b), 12, 31) << 26) | MASK26) |
| 387 | #define SKX_TAD_OFFSET(b) ((u64)GET_BITFIELD((b), 4, 23) << 26) |
| 388 | #define SKX_TAD_SKTWAYS(b) (1 << GET_BITFIELD((b), 10, 11)) |
| 389 | #define SKX_TAD_CHNWAYS(b) (GET_BITFIELD((b), 8, 9) + 1) |
| 390 | |
| 391 | /* which bit used for both socket and channel interleave */ |
| 392 | static int skx_granularity[] = { 6, 8, 12, 30 }; |
| 393 | |
| 394 | static u64 skx_do_interleave(u64 addr, int shift, int ways, u64 lowbits) |
| 395 | { |
| 396 | addr >>= shift; |
| 397 | addr /= ways; |
| 398 | addr <<= shift; |
| 399 | |
| 400 | return addr | (lowbits & ((1ull << shift) - 1)); |
| 401 | } |
| 402 | |
| 403 | static bool skx_tad_decode(struct decoded_addr *res) |
| 404 | { |
| 405 | int i; |
| 406 | u32 base, wayness, chnilvoffset; |
| 407 | int skt_interleave_bit, chn_interleave_bit; |
| 408 | u64 channel_addr; |
| 409 | |
| 410 | for (i = 0; i < SKX_MAX_TAD; i++) { |
| 411 | SKX_GET_TADBASE(res->dev, res->imc, i, base); |
| 412 | SKX_GET_TADWAYNESS(res->dev, res->imc, i, wayness); |
| 413 | if (SKX_TAD_BASE(base) <= res->addr && res->addr <= SKX_TAD_LIMIT(wayness)) |
| 414 | goto tad_found; |
| 415 | } |
| 416 | edac_dbg(0, "No TAD entry for 0x%llx\n", res->addr); |
| 417 | return false; |
| 418 | |
| 419 | tad_found: |
| 420 | res->sktways = SKX_TAD_SKTWAYS(wayness); |
| 421 | res->chanways = SKX_TAD_CHNWAYS(wayness); |
| 422 | skt_interleave_bit = skx_granularity[SKX_TAD_SKT_GRAN(base)]; |
| 423 | chn_interleave_bit = skx_granularity[SKX_TAD_CHN_GRAN(base)]; |
| 424 | |
| 425 | SKX_GET_TADCHNILVOFFSET(res->dev, res->imc, res->channel, i, chnilvoffset); |
| 426 | channel_addr = res->addr - SKX_TAD_OFFSET(chnilvoffset); |
| 427 | |
| 428 | if (res->chanways == 3 && skt_interleave_bit > chn_interleave_bit) { |
| 429 | /* Must handle channel first, then socket */ |
| 430 | channel_addr = skx_do_interleave(channel_addr, chn_interleave_bit, |
| 431 | res->chanways, channel_addr); |
| 432 | channel_addr = skx_do_interleave(channel_addr, skt_interleave_bit, |
| 433 | res->sktways, channel_addr); |
| 434 | } else { |
| 435 | /* Handle socket then channel. Preserve low bits from original address */ |
| 436 | channel_addr = skx_do_interleave(channel_addr, skt_interleave_bit, |
| 437 | res->sktways, res->addr); |
| 438 | channel_addr = skx_do_interleave(channel_addr, chn_interleave_bit, |
| 439 | res->chanways, res->addr); |
| 440 | } |
| 441 | |
| 442 | res->chan_addr = channel_addr; |
| 443 | |
| 444 | edac_dbg(2, "0x%llx: chan_addr=0x%llx sktways=%d chanways=%d\n", |
| 445 | res->addr, res->chan_addr, res->sktways, res->chanways); |
| 446 | return true; |
| 447 | } |
| 448 | |
| 449 | #define SKX_MAX_RIR 4 |
| 450 | |
| 451 | #define SKX_GET_RIRWAYNESS(d, mc, ch, i, reg) \ |
| 452 | pci_read_config_dword((d)->imc[mc].chan[ch].cdev, \ |
| 453 | 0x108 + 4 * (i), &(reg)) |
| 454 | #define SKX_GET_RIRILV(d, mc, ch, idx, i, reg) \ |
| 455 | pci_read_config_dword((d)->imc[mc].chan[ch].cdev, \ |
| 456 | 0x120 + 16 * (idx) + 4 * (i), &(reg)) |
| 457 | |
| 458 | #define SKX_RIR_VALID(b) GET_BITFIELD((b), 31, 31) |
| 459 | #define SKX_RIR_LIMIT(b) (((u64)GET_BITFIELD((b), 1, 11) << 29) | MASK29) |
| 460 | #define SKX_RIR_WAYS(b) (1 << GET_BITFIELD((b), 28, 29)) |
| 461 | #define SKX_RIR_CHAN_RANK(b) GET_BITFIELD((b), 16, 19) |
| 462 | #define SKX_RIR_OFFSET(b) ((u64)(GET_BITFIELD((b), 2, 15) << 26)) |
| 463 | |
| 464 | static bool skx_rir_decode(struct decoded_addr *res) |
| 465 | { |
| 466 | int i, idx, chan_rank; |
| 467 | int shift; |
| 468 | u32 rirway, rirlv; |
| 469 | u64 rank_addr, prev_limit = 0, limit; |
| 470 | |
| 471 | if (res->dev->imc[res->imc].chan[res->channel].dimms[0].close_pg) |
| 472 | shift = 6; |
| 473 | else |
| 474 | shift = 13; |
| 475 | |
| 476 | for (i = 0; i < SKX_MAX_RIR; i++) { |
| 477 | SKX_GET_RIRWAYNESS(res->dev, res->imc, res->channel, i, rirway); |
| 478 | limit = SKX_RIR_LIMIT(rirway); |
| 479 | if (SKX_RIR_VALID(rirway)) { |
| 480 | if (prev_limit <= res->chan_addr && |
| 481 | res->chan_addr <= limit) |
| 482 | goto rir_found; |
| 483 | } |
| 484 | prev_limit = limit; |
| 485 | } |
| 486 | edac_dbg(0, "No RIR entry for 0x%llx\n", res->addr); |
| 487 | return false; |
| 488 | |
| 489 | rir_found: |
| 490 | rank_addr = res->chan_addr >> shift; |
| 491 | rank_addr /= SKX_RIR_WAYS(rirway); |
| 492 | rank_addr <<= shift; |
| 493 | rank_addr |= res->chan_addr & GENMASK_ULL(shift - 1, 0); |
| 494 | |
| 495 | res->rank_address = rank_addr; |
| 496 | idx = (res->chan_addr >> shift) % SKX_RIR_WAYS(rirway); |
| 497 | |
| 498 | SKX_GET_RIRILV(res->dev, res->imc, res->channel, idx, i, rirlv); |
| 499 | res->rank_address = rank_addr - SKX_RIR_OFFSET(rirlv); |
| 500 | chan_rank = SKX_RIR_CHAN_RANK(rirlv); |
| 501 | res->channel_rank = chan_rank; |
| 502 | res->dimm = chan_rank / 4; |
| 503 | res->rank = chan_rank % 4; |
| 504 | |
| 505 | edac_dbg(2, "0x%llx: dimm=%d rank=%d chan_rank=%d rank_addr=0x%llx\n", |
| 506 | res->addr, res->dimm, res->rank, |
| 507 | res->channel_rank, res->rank_address); |
| 508 | return true; |
| 509 | } |
| 510 | |
| 511 | static u8 skx_close_row[] = { |
| 512 | 15, 16, 17, 18, 20, 21, 22, 28, 10, 11, 12, 13, 29, 30, 31, 32, 33 |
| 513 | }; |
| 514 | |
| 515 | static u8 skx_close_column[] = { |
| 516 | 3, 4, 5, 14, 19, 23, 24, 25, 26, 27 |
| 517 | }; |
| 518 | |
| 519 | static u8 skx_open_row[] = { |
| 520 | 14, 15, 16, 20, 28, 21, 22, 23, 24, 25, 26, 27, 29, 30, 31, 32, 33 |
| 521 | }; |
| 522 | |
| 523 | static u8 skx_open_column[] = { |
| 524 | 3, 4, 5, 6, 7, 8, 9, 10, 11, 12 |
| 525 | }; |
| 526 | |
| 527 | static u8 skx_open_fine_column[] = { |
| 528 | 3, 4, 5, 7, 8, 9, 10, 11, 12, 13 |
| 529 | }; |
| 530 | |
| 531 | static int skx_bits(u64 addr, int nbits, u8 *bits) |
| 532 | { |
| 533 | int i, res = 0; |
| 534 | |
| 535 | for (i = 0; i < nbits; i++) |
| 536 | res |= ((addr >> bits[i]) & 1) << i; |
| 537 | return res; |
| 538 | } |
| 539 | |
| 540 | static int skx_bank_bits(u64 addr, int b0, int b1, int do_xor, int x0, int x1) |
| 541 | { |
| 542 | int ret = GET_BITFIELD(addr, b0, b0) | (GET_BITFIELD(addr, b1, b1) << 1); |
| 543 | |
| 544 | if (do_xor) |
| 545 | ret ^= GET_BITFIELD(addr, x0, x0) | (GET_BITFIELD(addr, x1, x1) << 1); |
| 546 | |
| 547 | return ret; |
| 548 | } |
| 549 | |
| 550 | static bool skx_mad_decode(struct decoded_addr *r) |
| 551 | { |
| 552 | struct skx_dimm *dimm = &r->dev->imc[r->imc].chan[r->channel].dimms[r->dimm]; |
| 553 | int bg0 = dimm->fine_grain_bank ? 6 : 13; |
| 554 | |
| 555 | if (dimm->close_pg) { |
| 556 | r->row = skx_bits(r->rank_address, dimm->rowbits, skx_close_row); |
| 557 | r->column = skx_bits(r->rank_address, dimm->colbits, skx_close_column); |
| 558 | r->column |= 0x400; /* C10 is autoprecharge, always set */ |
| 559 | r->bank_address = skx_bank_bits(r->rank_address, 8, 9, dimm->bank_xor_enable, 22, 28); |
| 560 | r->bank_group = skx_bank_bits(r->rank_address, 6, 7, dimm->bank_xor_enable, 20, 21); |
| 561 | } else { |
| 562 | r->row = skx_bits(r->rank_address, dimm->rowbits, skx_open_row); |
| 563 | if (dimm->fine_grain_bank) |
| 564 | r->column = skx_bits(r->rank_address, dimm->colbits, skx_open_fine_column); |
| 565 | else |
| 566 | r->column = skx_bits(r->rank_address, dimm->colbits, skx_open_column); |
| 567 | r->bank_address = skx_bank_bits(r->rank_address, 18, 19, dimm->bank_xor_enable, 22, 23); |
| 568 | r->bank_group = skx_bank_bits(r->rank_address, bg0, 17, dimm->bank_xor_enable, 20, 21); |
| 569 | } |
| 570 | r->row &= (1u << dimm->rowbits) - 1; |
| 571 | |
| 572 | edac_dbg(2, "0x%llx: row=0x%x col=0x%x bank_addr=%d bank_group=%d\n", |
| 573 | r->addr, r->row, r->column, r->bank_address, |
| 574 | r->bank_group); |
| 575 | return true; |
| 576 | } |
| 577 | |
| 578 | static bool skx_decode(struct decoded_addr *res) |
| 579 | { |
| 580 | return skx_sad_decode(res) && skx_tad_decode(res) && |
| 581 | skx_rir_decode(res) && skx_mad_decode(res); |
| 582 | } |
| 583 | |
| 584 | static struct notifier_block skx_mce_dec = { |
| 585 | .notifier_call = skx_mce_check_error, |
| 586 | .priority = MCE_PRIO_EDAC, |
| 587 | }; |
| 588 | |
Qiuxu Zhuo | fe78351 | 2019-03-21 15:13:39 -0700 | [diff] [blame] | 589 | #ifdef CONFIG_EDAC_DEBUG |
| 590 | /* |
| 591 | * Debug feature. |
| 592 | * Exercise the address decode logic by writing an address to |
| 593 | * /sys/kernel/debug/edac/skx_test/addr. |
| 594 | */ |
| 595 | static struct dentry *skx_test; |
| 596 | |
| 597 | static int debugfs_u64_set(void *data, u64 val) |
| 598 | { |
| 599 | struct mce m; |
| 600 | |
| 601 | pr_warn_once("Fake error to 0x%llx injected via debugfs\n", val); |
| 602 | |
| 603 | memset(&m, 0, sizeof(m)); |
| 604 | /* ADDRV + MemRd + Unknown channel */ |
| 605 | m.status = MCI_STATUS_ADDRV + 0x90; |
| 606 | /* One corrected error */ |
| 607 | m.status |= BIT_ULL(MCI_STATUS_CEC_SHIFT); |
| 608 | m.addr = val; |
| 609 | skx_mce_check_error(NULL, 0, &m); |
| 610 | |
| 611 | return 0; |
| 612 | } |
| 613 | DEFINE_SIMPLE_ATTRIBUTE(fops_u64_wo, NULL, debugfs_u64_set, "%llu\n"); |
| 614 | |
| 615 | static void setup_skx_debug(void) |
| 616 | { |
| 617 | skx_test = edac_debugfs_create_dir("skx_test"); |
| 618 | if (!skx_test) |
| 619 | return; |
| 620 | |
| 621 | if (!edac_debugfs_create_file("addr", 0200, skx_test, |
| 622 | NULL, &fops_u64_wo)) { |
| 623 | debugfs_remove(skx_test); |
| 624 | skx_test = NULL; |
| 625 | } |
| 626 | } |
| 627 | |
| 628 | static void teardown_skx_debug(void) |
| 629 | { |
| 630 | debugfs_remove_recursive(skx_test); |
| 631 | } |
| 632 | #else |
| 633 | static inline void setup_skx_debug(void) {} |
| 634 | static inline void teardown_skx_debug(void) {} |
| 635 | #endif /*CONFIG_EDAC_DEBUG*/ |
| 636 | |
Qiuxu Zhuo | 98f2fc8 | 2019-01-30 11:15:18 -0800 | [diff] [blame] | 637 | /* |
| 638 | * skx_init: |
| 639 | * make sure we are running on the correct cpu model |
| 640 | * search for all the devices we need |
| 641 | * check which DIMMs are present. |
| 642 | */ |
| 643 | static int __init skx_init(void) |
| 644 | { |
| 645 | const struct x86_cpu_id *id; |
Qiuxu Zhuo | ee5340a | 2020-04-24 20:18:14 +0800 | [diff] [blame] | 646 | struct res_config *cfg; |
Qiuxu Zhuo | 98f2fc8 | 2019-01-30 11:15:18 -0800 | [diff] [blame] | 647 | const struct munit *m; |
| 648 | const char *owner; |
| 649 | int rc = 0, i, off[3] = {0xd0, 0xd4, 0xd8}; |
| 650 | u8 mc = 0, src_id, node_id; |
| 651 | struct skx_dev *d; |
| 652 | |
| 653 | edac_dbg(2, "\n"); |
| 654 | |
| 655 | owner = edac_get_owner(); |
| 656 | if (owner && strncmp(owner, EDAC_MOD_STR, sizeof(EDAC_MOD_STR))) |
| 657 | return -EBUSY; |
| 658 | |
| 659 | id = x86_match_cpu(skx_cpuids); |
| 660 | if (!id) |
| 661 | return -ENODEV; |
| 662 | |
Qiuxu Zhuo | ee5340a | 2020-04-24 20:18:14 +0800 | [diff] [blame] | 663 | cfg = (struct res_config *)id->driver_data; |
| 664 | |
Qiuxu Zhuo | 98f2fc8 | 2019-01-30 11:15:18 -0800 | [diff] [blame] | 665 | rc = skx_get_hi_lo(0x2034, off, &skx_tolm, &skx_tohm); |
| 666 | if (rc) |
| 667 | return rc; |
| 668 | |
Qiuxu Zhuo | ee5340a | 2020-04-24 20:18:14 +0800 | [diff] [blame] | 669 | rc = skx_get_all_bus_mappings(cfg, &skx_edac_list); |
Qiuxu Zhuo | 98f2fc8 | 2019-01-30 11:15:18 -0800 | [diff] [blame] | 670 | if (rc < 0) |
| 671 | goto fail; |
| 672 | if (rc == 0) { |
| 673 | edac_dbg(2, "No memory controllers found\n"); |
| 674 | return -ENODEV; |
| 675 | } |
| 676 | skx_num_sockets = rc; |
| 677 | |
| 678 | for (m = skx_all_munits; m->did; m++) { |
| 679 | rc = get_all_munits(m); |
| 680 | if (rc < 0) |
| 681 | goto fail; |
| 682 | if (rc != m->per_socket * skx_num_sockets) { |
| 683 | edac_dbg(2, "Expected %d, got %d of 0x%x\n", |
| 684 | m->per_socket * skx_num_sockets, rc, m->did); |
| 685 | rc = -ENODEV; |
| 686 | goto fail; |
| 687 | } |
| 688 | } |
| 689 | |
| 690 | list_for_each_entry(d, skx_edac_list, list) { |
Qiuxu Zhuo | 1dc78f1 | 2019-06-26 14:16:55 +0800 | [diff] [blame] | 691 | rc = skx_get_src_id(d, 0xf0, &src_id); |
Qiuxu Zhuo | 98f2fc8 | 2019-01-30 11:15:18 -0800 | [diff] [blame] | 692 | if (rc < 0) |
| 693 | goto fail; |
| 694 | rc = skx_get_node_id(d, &node_id); |
| 695 | if (rc < 0) |
| 696 | goto fail; |
| 697 | edac_dbg(2, "src_id=%d node_id=%d\n", src_id, node_id); |
| 698 | for (i = 0; i < SKX_NUM_IMC; i++) { |
| 699 | d->imc[i].mc = mc++; |
| 700 | d->imc[i].lmc = i; |
| 701 | d->imc[i].src_id = src_id; |
| 702 | d->imc[i].node_id = node_id; |
| 703 | rc = skx_register_mci(&d->imc[i], d->imc[i].chan[0].cdev, |
| 704 | "Skylake Socket", EDAC_MOD_STR, |
| 705 | skx_get_dimm_config); |
| 706 | if (rc < 0) |
| 707 | goto fail; |
| 708 | } |
| 709 | } |
| 710 | |
Tony Luck | e80634a | 2019-08-15 14:53:28 -0700 | [diff] [blame] | 711 | skx_set_decode(skx_decode, skx_show_retry_rd_err_log); |
Qiuxu Zhuo | 98f2fc8 | 2019-01-30 11:15:18 -0800 | [diff] [blame] | 712 | |
| 713 | if (nvdimm_count && skx_adxl_get() == -ENODEV) |
| 714 | skx_printk(KERN_NOTICE, "Only decoding DDR4 address!\n"); |
| 715 | |
| 716 | /* Ensure that the OPSTATE is set correctly for POLL or NMI */ |
| 717 | opstate_init(); |
| 718 | |
Qiuxu Zhuo | fe78351 | 2019-03-21 15:13:39 -0700 | [diff] [blame] | 719 | setup_skx_debug(); |
Qiuxu Zhuo | 98f2fc8 | 2019-01-30 11:15:18 -0800 | [diff] [blame] | 720 | |
| 721 | mce_register_decode_chain(&skx_mce_dec); |
| 722 | |
| 723 | return 0; |
| 724 | fail: |
| 725 | skx_remove(); |
| 726 | return rc; |
| 727 | } |
| 728 | |
| 729 | static void __exit skx_exit(void) |
| 730 | { |
| 731 | edac_dbg(2, "\n"); |
| 732 | mce_unregister_decode_chain(&skx_mce_dec); |
| 733 | teardown_skx_debug(); |
| 734 | if (nvdimm_count) |
| 735 | skx_adxl_put(); |
| 736 | skx_remove(); |
| 737 | } |
| 738 | |
| 739 | module_init(skx_init); |
| 740 | module_exit(skx_exit); |
| 741 | |
| 742 | module_param(edac_op_state, int, 0444); |
| 743 | MODULE_PARM_DESC(edac_op_state, "EDAC Error Reporting state: 0=Poll,1=NMI"); |
| 744 | |
| 745 | MODULE_LICENSE("GPL v2"); |
| 746 | MODULE_AUTHOR("Tony Luck"); |
| 747 | MODULE_DESCRIPTION("MC Driver for Intel Skylake server processors"); |