1// SPDX-License-Identifier: GPL-2.0-or-later 2/* 3 * APM X-Gene SoC EDAC (error detection and correction) 4 * 5 * Copyright (c) 2015, Applied Micro Circuits Corporation 6 * Author: Feng Kan <fkan@apm.com> 7 * Loc Ho <lho@apm.com> 8 */ 9 10#include <linux/ctype.h> 11#include <linux/edac.h> 12#include <linux/interrupt.h> 13#include <linux/mfd/syscon.h> 14#include <linux/module.h> 15#include <linux/of.h> 16#include <linux/of_address.h> 17#include <linux/regmap.h> 18 19#include "edac_module.h" 20 21#define EDAC_MOD_STR "xgene_edac" 22 23/* Global error configuration status registers (CSR) */ 24#define PCPHPERRINTSTS 0x0000 25#define PCPHPERRINTMSK 0x0004 26#define MCU_CTL_ERR_MASK BIT(12) 27#define IOB_PA_ERR_MASK BIT(11) 28#define IOB_BA_ERR_MASK BIT(10) 29#define IOB_XGIC_ERR_MASK BIT(9) 30#define IOB_RB_ERR_MASK BIT(8) 31#define L3C_UNCORR_ERR_MASK BIT(5) 32#define MCU_UNCORR_ERR_MASK BIT(4) 33#define PMD3_MERR_MASK BIT(3) 34#define PMD2_MERR_MASK BIT(2) 35#define PMD1_MERR_MASK BIT(1) 36#define PMD0_MERR_MASK BIT(0) 37#define PCPLPERRINTSTS 0x0008 38#define PCPLPERRINTMSK 0x000C 39#define CSW_SWITCH_TRACE_ERR_MASK BIT(2) 40#define L3C_CORR_ERR_MASK BIT(1) 41#define MCU_CORR_ERR_MASK BIT(0) 42#define MEMERRINTSTS 0x0010 43#define MEMERRINTMSK 0x0014 44 45struct xgene_edac { 46 struct device *dev; 47 struct regmap *csw_map; 48 struct regmap *mcba_map; 49 struct regmap *mcbb_map; 50 struct regmap *efuse_map; 51 struct regmap *rb_map; 52 void __iomem *pcp_csr; 53 spinlock_t lock; 54 struct dentry *dfs; 55 56 struct list_head mcus; 57 struct list_head pmds; 58 struct list_head l3s; 59 struct list_head socs; 60 61 struct mutex mc_lock; 62 int mc_active_mask; 63 int mc_registered_mask; 64}; 65 66static void xgene_edac_pcp_rd(struct xgene_edac *edac, u32 reg, u32 *val) 67{ 68 *val = readl(edac->pcp_csr + reg); 69} 70 71static void xgene_edac_pcp_clrbits(struct xgene_edac *edac, u32 reg, 72 u32 bits_mask) 73{ 74 u32 val; 75 76 spin_lock(&edac->lock); 77 val = readl(edac->pcp_csr + reg); 78 val &= ~bits_mask; 79 writel(val, edac->pcp_csr + reg); 80 spin_unlock(&edac->lock); 81} 82 83static void xgene_edac_pcp_setbits(struct xgene_edac *edac, u32 reg, 84 u32 bits_mask) 85{ 86 u32 val; 87 88 spin_lock(&edac->lock); 89 val = readl(edac->pcp_csr + reg); 90 val |= bits_mask; 91 writel(val, edac->pcp_csr + reg); 92 spin_unlock(&edac->lock); 93} 94 95/* Memory controller error CSR */ 96#define MCU_MAX_RANK 8 97#define MCU_RANK_STRIDE 0x40 98 99#define MCUGECR 0x0110 100#define MCU_GECR_DEMANDUCINTREN_MASK BIT(0) 101#define MCU_GECR_BACKUCINTREN_MASK BIT(1) 102#define MCU_GECR_CINTREN_MASK BIT(2) 103#define MUC_GECR_MCUADDRERREN_MASK BIT(9) 104#define MCUGESR 0x0114 105#define MCU_GESR_ADDRNOMATCH_ERR_MASK BIT(7) 106#define MCU_GESR_ADDRMULTIMATCH_ERR_MASK BIT(6) 107#define MCU_GESR_PHYP_ERR_MASK BIT(3) 108#define MCUESRR0 0x0314 109#define MCU_ESRR_MULTUCERR_MASK BIT(3) 110#define MCU_ESRR_BACKUCERR_MASK BIT(2) 111#define MCU_ESRR_DEMANDUCERR_MASK BIT(1) 112#define MCU_ESRR_CERR_MASK BIT(0) 113#define MCUESRRA0 0x0318 114#define MCUEBLRR0 0x031c 115#define MCU_EBLRR_ERRBANK_RD(src) (((src) & 0x00000007) >> 0) 116#define MCUERCRR0 0x0320 117#define MCU_ERCRR_ERRROW_RD(src) (((src) & 0xFFFF0000) >> 16) 118#define MCU_ERCRR_ERRCOL_RD(src) ((src) & 0x00000FFF) 119#define MCUSBECNT0 0x0324 120#define MCU_SBECNT_COUNT(src) ((src) & 0xFFFF) 121 122#define CSW_CSWCR 0x0000 123#define CSW_CSWCR_DUALMCB_MASK BIT(0) 124 125#define MCBADDRMR 0x0000 126#define MCBADDRMR_MCU_INTLV_MODE_MASK BIT(3) 127#define MCBADDRMR_DUALMCU_MODE_MASK BIT(2) 128#define MCBADDRMR_MCB_INTLV_MODE_MASK BIT(1) 129#define MCBADDRMR_ADDRESS_MODE_MASK BIT(0) 130 131struct xgene_edac_mc_ctx { 132 struct list_head next; 133 char *name; 134 struct mem_ctl_info *mci; 135 struct xgene_edac *edac; 136 void __iomem *mcu_csr; 137 u32 mcu_id; 138}; 139 140static ssize_t xgene_edac_mc_err_inject_write(struct file *file, 141 const char __user *data, 142 size_t count, loff_t *ppos) 143{ 144 struct mem_ctl_info *mci = file->private_data; 145 struct xgene_edac_mc_ctx *ctx = mci->pvt_info; 146 int i; 147 148 for (i = 0; i < MCU_MAX_RANK; i++) { 149 writel(MCU_ESRR_MULTUCERR_MASK | MCU_ESRR_BACKUCERR_MASK | 150 MCU_ESRR_DEMANDUCERR_MASK | MCU_ESRR_CERR_MASK, 151 ctx->mcu_csr + MCUESRRA0 + i * MCU_RANK_STRIDE); 152 } 153 return count; 154} 155 156static const struct file_operations xgene_edac_mc_debug_inject_fops = { 157 .open = simple_open, 158 .write = xgene_edac_mc_err_inject_write, 159 .llseek = generic_file_llseek, 160}; 161 162static void xgene_edac_mc_create_debugfs_node(struct mem_ctl_info *mci) 163{ 164 if (!IS_ENABLED(CONFIG_EDAC_DEBUG)) 165 return; 166 167 if (!mci->debugfs) 168 return; 169 170 edac_debugfs_create_file("inject_ctrl", S_IWUSR, mci->debugfs, mci, 171 &xgene_edac_mc_debug_inject_fops); 172} 173 174static void xgene_edac_mc_check(struct mem_ctl_info *mci) 175{ 176 struct xgene_edac_mc_ctx *ctx = mci->pvt_info; 177 unsigned int pcp_hp_stat; 178 unsigned int pcp_lp_stat; 179 u32 reg; 180 u32 rank; 181 u32 bank; 182 u32 count; 183 u32 col_row; 184 185 xgene_edac_pcp_rd(ctx->edac, PCPHPERRINTSTS, &pcp_hp_stat); 186 xgene_edac_pcp_rd(ctx->edac, PCPLPERRINTSTS, &pcp_lp_stat); 187 if (!((MCU_UNCORR_ERR_MASK & pcp_hp_stat) || 188 (MCU_CTL_ERR_MASK & pcp_hp_stat) || 189 (MCU_CORR_ERR_MASK & pcp_lp_stat))) 190 return; 191 192 for (rank = 0; rank < MCU_MAX_RANK; rank++) { 193 reg = readl(ctx->mcu_csr + MCUESRR0 + rank * MCU_RANK_STRIDE); 194 195 /* Detect uncorrectable memory error */ 196 if (reg & (MCU_ESRR_DEMANDUCERR_MASK | 197 MCU_ESRR_BACKUCERR_MASK)) { 198 /* Detected uncorrectable memory error */ 199 edac_mc_chipset_printk(mci, KERN_ERR, "X-Gene", 200 "MCU uncorrectable error at rank %d\n", rank); 201 202 edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci, 203 1, 0, 0, 0, 0, 0, -1, mci->ctl_name, ""); 204 } 205 206 /* Detect correctable memory error */ 207 if (reg & MCU_ESRR_CERR_MASK) { 208 bank = readl(ctx->mcu_csr + MCUEBLRR0 + 209 rank * MCU_RANK_STRIDE); 210 col_row = readl(ctx->mcu_csr + MCUERCRR0 + 211 rank * MCU_RANK_STRIDE); 212 count = readl(ctx->mcu_csr + MCUSBECNT0 + 213 rank * MCU_RANK_STRIDE); 214 edac_mc_chipset_printk(mci, KERN_WARNING, "X-Gene", 215 "MCU correctable error at rank %d bank %d column %d row %d count %d\n", 216 rank, MCU_EBLRR_ERRBANK_RD(bank), 217 MCU_ERCRR_ERRCOL_RD(col_row), 218 MCU_ERCRR_ERRROW_RD(col_row), 219 MCU_SBECNT_COUNT(count)); 220 221 edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci, 222 1, 0, 0, 0, 0, 0, -1, mci->ctl_name, ""); 223 } 224 225 /* Clear all error registers */ 226 writel(0x0, ctx->mcu_csr + MCUEBLRR0 + rank * MCU_RANK_STRIDE); 227 writel(0x0, ctx->mcu_csr + MCUERCRR0 + rank * MCU_RANK_STRIDE); 228 writel(0x0, ctx->mcu_csr + MCUSBECNT0 + 229 rank * MCU_RANK_STRIDE); 230 writel(reg, ctx->mcu_csr + MCUESRR0 + rank * MCU_RANK_STRIDE); 231 } 232 233 /* Detect memory controller error */ 234 reg = readl(ctx->mcu_csr + MCUGESR); 235 if (reg) { 236 if (reg & MCU_GESR_ADDRNOMATCH_ERR_MASK) 237 edac_mc_chipset_printk(mci, KERN_WARNING, "X-Gene", 238 "MCU address miss-match error\n"); 239 if (reg & MCU_GESR_ADDRMULTIMATCH_ERR_MASK) 240 edac_mc_chipset_printk(mci, KERN_WARNING, "X-Gene", 241 "MCU address multi-match error\n"); 242 243 writel(reg, ctx->mcu_csr + MCUGESR); 244 } 245} 246 247static void xgene_edac_mc_irq_ctl(struct mem_ctl_info *mci, bool enable) 248{ 249 struct xgene_edac_mc_ctx *ctx = mci->pvt_info; 250 unsigned int val; 251 252 if (edac_op_state != EDAC_OPSTATE_INT) 253 return; 254 255 mutex_lock(&ctx->edac->mc_lock); 256 257 /* 258 * As there is only single bit for enable error and interrupt mask, 259 * we must only enable top level interrupt after all MCUs are 260 * registered. Otherwise, if there is an error and the corresponding 261 * MCU has not registered, the interrupt will never get cleared. To 262 * determine all MCU have registered, we will keep track of active 263 * MCUs and registered MCUs. 264 */ 265 if (enable) { 266 /* Set registered MCU bit */ 267 ctx->edac->mc_registered_mask |= 1 << ctx->mcu_id; 268 269 /* Enable interrupt after all active MCU registered */ 270 if (ctx->edac->mc_registered_mask == 271 ctx->edac->mc_active_mask) { 272 /* Enable memory controller top level interrupt */ 273 xgene_edac_pcp_clrbits(ctx->edac, PCPHPERRINTMSK, 274 MCU_UNCORR_ERR_MASK | 275 MCU_CTL_ERR_MASK); 276 xgene_edac_pcp_clrbits(ctx->edac, PCPLPERRINTMSK, 277 MCU_CORR_ERR_MASK); 278 } 279 280 /* Enable MCU interrupt and error reporting */ 281 val = readl(ctx->mcu_csr + MCUGECR); 282 val |= MCU_GECR_DEMANDUCINTREN_MASK | 283 MCU_GECR_BACKUCINTREN_MASK | 284 MCU_GECR_CINTREN_MASK | 285 MUC_GECR_MCUADDRERREN_MASK; 286 writel(val, ctx->mcu_csr + MCUGECR); 287 } else { 288 /* Disable MCU interrupt */ 289 val = readl(ctx->mcu_csr + MCUGECR); 290 val &= ~(MCU_GECR_DEMANDUCINTREN_MASK | 291 MCU_GECR_BACKUCINTREN_MASK | 292 MCU_GECR_CINTREN_MASK | 293 MUC_GECR_MCUADDRERREN_MASK); 294 writel(val, ctx->mcu_csr + MCUGECR); 295 296 /* Disable memory controller top level interrupt */ 297 xgene_edac_pcp_setbits(ctx->edac, PCPHPERRINTMSK, 298 MCU_UNCORR_ERR_MASK | MCU_CTL_ERR_MASK); 299 xgene_edac_pcp_setbits(ctx->edac, PCPLPERRINTMSK, 300 MCU_CORR_ERR_MASK); 301 302 /* Clear registered MCU bit */ 303 ctx->edac->mc_registered_mask &= ~(1 << ctx->mcu_id); 304 } 305 306 mutex_unlock(&ctx->edac->mc_lock); 307} 308 309static int xgene_edac_mc_is_active(struct xgene_edac_mc_ctx *ctx, int mc_idx) 310{ 311 unsigned int reg; 312 u32 mcu_mask; 313 314 if (regmap_read(ctx->edac->csw_map, CSW_CSWCR, ®)) 315 return 0; 316 317 if (reg & CSW_CSWCR_DUALMCB_MASK) { 318 /* 319 * Dual MCB active - Determine if all 4 active or just MCU0 320 * and MCU2 active 321 */ 322 if (regmap_read(ctx->edac->mcbb_map, MCBADDRMR, ®)) 323 return 0; 324 mcu_mask = (reg & MCBADDRMR_DUALMCU_MODE_MASK) ? 0xF : 0x5; 325 } else { 326 /* 327 * Single MCB active - Determine if MCU0/MCU1 or just MCU0 328 * active 329 */ 330 if (regmap_read(ctx->edac->mcba_map, MCBADDRMR, ®)) 331 return 0; 332 mcu_mask = (reg & MCBADDRMR_DUALMCU_MODE_MASK) ? 0x3 : 0x1; 333 } 334 335 /* Save active MC mask if hasn't set already */ 336 if (!ctx->edac->mc_active_mask) 337 ctx->edac->mc_active_mask = mcu_mask; 338 339 return (mcu_mask & (1 << mc_idx)) ? 1 : 0; 340} 341 342static int xgene_edac_mc_add(struct xgene_edac *edac, struct device_node *np) 343{ 344 struct mem_ctl_info *mci; 345 struct edac_mc_layer layers[2]; 346 struct xgene_edac_mc_ctx tmp_ctx; 347 struct xgene_edac_mc_ctx *ctx; 348 struct resource res; 349 int rc; 350 351 memset(&tmp_ctx, 0, sizeof(tmp_ctx)); 352 tmp_ctx.edac = edac; 353 354 if (!devres_open_group(edac->dev, xgene_edac_mc_add, GFP_KERNEL)) 355 return -ENOMEM; 356 357 rc = of_address_to_resource(np, 0, &res); 358 if (rc < 0) { 359 dev_err(edac->dev, "no MCU resource address\n"); 360 goto err_group; 361 } 362 tmp_ctx.mcu_csr = devm_ioremap_resource(edac->dev, &res); 363 if (IS_ERR(tmp_ctx.mcu_csr)) { 364 dev_err(edac->dev, "unable to map MCU resource\n"); 365 rc = PTR_ERR(tmp_ctx.mcu_csr); 366 goto err_group; 367 } 368 369 /* Ignore non-active MCU */ 370 if (of_property_read_u32(np, "memory-controller", &tmp_ctx.mcu_id)) { 371 dev_err(edac->dev, "no memory-controller property\n"); 372 rc = -ENODEV; 373 goto err_group; 374 } 375 if (!xgene_edac_mc_is_active(&tmp_ctx, tmp_ctx.mcu_id)) { 376 rc = -ENODEV; 377 goto err_group; 378 } 379 380 layers[0].type = EDAC_MC_LAYER_CHIP_SELECT; 381 layers[0].size = 4; 382 layers[0].is_virt_csrow = true; 383 layers[1].type = EDAC_MC_LAYER_CHANNEL; 384 layers[1].size = 2; 385 layers[1].is_virt_csrow = false; 386 mci = edac_mc_alloc(tmp_ctx.mcu_id, ARRAY_SIZE(layers), layers, 387 sizeof(*ctx)); 388 if (!mci) { 389 rc = -ENOMEM; 390 goto err_group; 391 } 392 393 ctx = mci->pvt_info; 394 *ctx = tmp_ctx; /* Copy over resource value */ 395 ctx->name = "xgene_edac_mc_err"; 396 ctx->mci = mci; 397 mci->pdev = &mci->dev; 398 mci->ctl_name = ctx->name; 399 mci->dev_name = ctx->name; 400 401 mci->mtype_cap = MEM_FLAG_RDDR | MEM_FLAG_RDDR2 | MEM_FLAG_RDDR3 | 402 MEM_FLAG_DDR | MEM_FLAG_DDR2 | MEM_FLAG_DDR3; 403 mci->edac_ctl_cap = EDAC_FLAG_SECDED; 404 mci->edac_cap = EDAC_FLAG_SECDED; 405 mci->mod_name = EDAC_MOD_STR; 406 mci->ctl_page_to_phys = NULL; 407 mci->scrub_cap = SCRUB_FLAG_HW_SRC; 408 mci->scrub_mode = SCRUB_HW_SRC; 409 410 if (edac_op_state == EDAC_OPSTATE_POLL) 411 mci->edac_check = xgene_edac_mc_check; 412 413 if (edac_mc_add_mc(mci)) { 414 dev_err(edac->dev, "edac_mc_add_mc failed\n"); 415 rc = -EINVAL; 416 goto err_free; 417 } 418 419 xgene_edac_mc_create_debugfs_node(mci); 420 421 list_add(&ctx->next, &edac->mcus); 422 423 xgene_edac_mc_irq_ctl(mci, true); 424 425 devres_remove_group(edac->dev, xgene_edac_mc_add); 426 427 dev_info(edac->dev, "X-Gene EDAC MC registered\n"); 428 return 0; 429 430err_free: 431 edac_mc_free(mci); 432err_group: 433 devres_release_group(edac->dev, xgene_edac_mc_add); 434 return rc; 435} 436 437static int xgene_edac_mc_remove(struct xgene_edac_mc_ctx *mcu) 438{ 439 xgene_edac_mc_irq_ctl(mcu->mci, false); 440 edac_mc_del_mc(&mcu->mci->dev); 441 edac_mc_free(mcu->mci); 442 return 0; 443} 444 445/* CPU L1/L2 error CSR */ 446#define MAX_CPU_PER_PMD 2 447#define CPU_CSR_STRIDE 0x00100000 448#define CPU_L2C_PAGE 0x000D0000 449#define CPU_MEMERR_L2C_PAGE 0x000E0000 450#define CPU_MEMERR_CPU_PAGE 0x000F0000 451 452#define MEMERR_CPU_ICFECR_PAGE_OFFSET 0x0000 453#define MEMERR_CPU_ICFESR_PAGE_OFFSET 0x0004 454#define MEMERR_CPU_ICFESR_ERRWAY_RD(src) (((src) & 0xFF000000) >> 24) 455#define MEMERR_CPU_ICFESR_ERRINDEX_RD(src) (((src) & 0x003F0000) >> 16) 456#define MEMERR_CPU_ICFESR_ERRINFO_RD(src) (((src) & 0x0000FF00) >> 8) 457#define MEMERR_CPU_ICFESR_ERRTYPE_RD(src) (((src) & 0x00000070) >> 4) 458#define MEMERR_CPU_ICFESR_MULTCERR_MASK BIT(2) 459#define MEMERR_CPU_ICFESR_CERR_MASK BIT(0) 460#define MEMERR_CPU_LSUESR_PAGE_OFFSET 0x000c 461#define MEMERR_CPU_LSUESR_ERRWAY_RD(src) (((src) & 0xFF000000) >> 24) 462#define MEMERR_CPU_LSUESR_ERRINDEX_RD(src) (((src) & 0x003F0000) >> 16) 463#define MEMERR_CPU_LSUESR_ERRINFO_RD(src) (((src) & 0x0000FF00) >> 8) 464#define MEMERR_CPU_LSUESR_ERRTYPE_RD(src) (((src) & 0x00000070) >> 4) 465#define MEMERR_CPU_LSUESR_MULTCERR_MASK BIT(2) 466#define MEMERR_CPU_LSUESR_CERR_MASK BIT(0) 467#define MEMERR_CPU_LSUECR_PAGE_OFFSET 0x0008 468#define MEMERR_CPU_MMUECR_PAGE_OFFSET 0x0010 469#define MEMERR_CPU_MMUESR_PAGE_OFFSET 0x0014 470#define MEMERR_CPU_MMUESR_ERRWAY_RD(src) (((src) & 0xFF000000) >> 24) 471#define MEMERR_CPU_MMUESR_ERRINDEX_RD(src) (((src) & 0x007F0000) >> 16) 472#define MEMERR_CPU_MMUESR_ERRINFO_RD(src) (((src) & 0x0000FF00) >> 8) 473#define MEMERR_CPU_MMUESR_ERRREQSTR_LSU_MASK BIT(7) 474#define MEMERR_CPU_MMUESR_ERRTYPE_RD(src) (((src) & 0x00000070) >> 4) 475#define MEMERR_CPU_MMUESR_MULTCERR_MASK BIT(2) 476#define MEMERR_CPU_MMUESR_CERR_MASK BIT(0) 477#define MEMERR_CPU_ICFESRA_PAGE_OFFSET 0x0804 478#define MEMERR_CPU_LSUESRA_PAGE_OFFSET 0x080c 479#define MEMERR_CPU_MMUESRA_PAGE_OFFSET 0x0814 480 481#define MEMERR_L2C_L2ECR_PAGE_OFFSET 0x0000 482#define MEMERR_L2C_L2ESR_PAGE_OFFSET 0x0004 483#define MEMERR_L2C_L2ESR_ERRSYN_RD(src) (((src) & 0xFF000000) >> 24) 484#define MEMERR_L2C_L2ESR_ERRWAY_RD(src) (((src) & 0x00FC0000) >> 18) 485#define MEMERR_L2C_L2ESR_ERRCPU_RD(src) (((src) & 0x00020000) >> 17) 486#define MEMERR_L2C_L2ESR_ERRGROUP_RD(src) (((src) & 0x0000E000) >> 13) 487#define MEMERR_L2C_L2ESR_ERRACTION_RD(src) (((src) & 0x00001C00) >> 10) 488#define MEMERR_L2C_L2ESR_ERRTYPE_RD(src) (((src) & 0x00000300) >> 8) 489#define MEMERR_L2C_L2ESR_MULTUCERR_MASK BIT(3) 490#define MEMERR_L2C_L2ESR_MULTICERR_MASK BIT(2) 491#define MEMERR_L2C_L2ESR_UCERR_MASK BIT(1) 492#define MEMERR_L2C_L2ESR_ERR_MASK BIT(0) 493#define MEMERR_L2C_L2EALR_PAGE_OFFSET 0x0008 494#define CPUX_L2C_L2RTOCR_PAGE_OFFSET 0x0010 495#define MEMERR_L2C_L2EAHR_PAGE_OFFSET 0x000c 496#define CPUX_L2C_L2RTOSR_PAGE_OFFSET 0x0014 497#define MEMERR_L2C_L2RTOSR_MULTERR_MASK BIT(1) 498#define MEMERR_L2C_L2RTOSR_ERR_MASK BIT(0) 499#define CPUX_L2C_L2RTOALR_PAGE_OFFSET 0x0018 500#define CPUX_L2C_L2RTOAHR_PAGE_OFFSET 0x001c 501#define MEMERR_L2C_L2ESRA_PAGE_OFFSET 0x0804 502 503/* 504 * Processor Module Domain (PMD) context - Context for a pair of processors. 505 * Each PMD consists of 2 CPUs and a shared L2 cache. Each CPU consists of 506 * its own L1 cache. 507 */ 508struct xgene_edac_pmd_ctx { 509 struct list_head next; 510 struct device ddev; 511 char *name; 512 struct xgene_edac *edac; 513 struct edac_device_ctl_info *edac_dev; 514 void __iomem *pmd_csr; 515 u32 pmd; 516 int version; 517}; 518 519static void xgene_edac_pmd_l1_check(struct edac_device_ctl_info *edac_dev, 520 int cpu_idx) 521{ 522 struct xgene_edac_pmd_ctx *ctx = edac_dev->pvt_info; 523 void __iomem *pg_f; 524 u32 val; 525 526 pg_f = ctx->pmd_csr + cpu_idx * CPU_CSR_STRIDE + CPU_MEMERR_CPU_PAGE; 527 528 val = readl(pg_f + MEMERR_CPU_ICFESR_PAGE_OFFSET); 529 if (!val) 530 goto chk_lsu; 531 dev_err(edac_dev->dev, 532 "CPU%d L1 memory error ICF 0x%08X Way 0x%02X Index 0x%02X Info 0x%02X\n", 533 ctx->pmd * MAX_CPU_PER_PMD + cpu_idx, val, 534 MEMERR_CPU_ICFESR_ERRWAY_RD(val), 535 MEMERR_CPU_ICFESR_ERRINDEX_RD(val), 536 MEMERR_CPU_ICFESR_ERRINFO_RD(val)); 537 if (val & MEMERR_CPU_ICFESR_CERR_MASK) 538 dev_err(edac_dev->dev, "One or more correctable error\n"); 539 if (val & MEMERR_CPU_ICFESR_MULTCERR_MASK) 540 dev_err(edac_dev->dev, "Multiple correctable error\n"); 541 switch (MEMERR_CPU_ICFESR_ERRTYPE_RD(val)) { 542 case 1: 543 dev_err(edac_dev->dev, "L1 TLB multiple hit\n"); 544 break; 545 case 2: 546 dev_err(edac_dev->dev, "Way select multiple hit\n"); 547 break; 548 case 3: 549 dev_err(edac_dev->dev, "Physical tag parity error\n"); 550 break; 551 case 4: 552 case 5: 553 dev_err(edac_dev->dev, "L1 data parity error\n"); 554 break; 555 case 6: 556 dev_err(edac_dev->dev, "L1 pre-decode parity error\n"); 557 break; 558 } 559 560 /* Clear any HW errors */ 561 writel(val, pg_f + MEMERR_CPU_ICFESR_PAGE_OFFSET); 562 563 if (val & (MEMERR_CPU_ICFESR_CERR_MASK | 564 MEMERR_CPU_ICFESR_MULTCERR_MASK)) 565 edac_device_handle_ce(edac_dev, 0, 0, edac_dev->ctl_name); 566 567chk_lsu: 568 val = readl(pg_f + MEMERR_CPU_LSUESR_PAGE_OFFSET); 569 if (!val) 570 goto chk_mmu; 571 dev_err(edac_dev->dev, 572 "CPU%d memory error LSU 0x%08X Way 0x%02X Index 0x%02X Info 0x%02X\n", 573 ctx->pmd * MAX_CPU_PER_PMD + cpu_idx, val, 574 MEMERR_CPU_LSUESR_ERRWAY_RD(val), 575 MEMERR_CPU_LSUESR_ERRINDEX_RD(val), 576 MEMERR_CPU_LSUESR_ERRINFO_RD(val)); 577 if (val & MEMERR_CPU_LSUESR_CERR_MASK) 578 dev_err(edac_dev->dev, "One or more correctable error\n"); 579 if (val & MEMERR_CPU_LSUESR_MULTCERR_MASK) 580 dev_err(edac_dev->dev, "Multiple correctable error\n"); 581 switch (MEMERR_CPU_LSUESR_ERRTYPE_RD(val)) { 582 case 0: 583 dev_err(edac_dev->dev, "Load tag error\n"); 584 break; 585 case 1: 586 dev_err(edac_dev->dev, "Load data error\n"); 587 break; 588 case 2: 589 dev_err(edac_dev->dev, "WSL multihit error\n"); 590 break; 591 case 3: 592 dev_err(edac_dev->dev, "Store tag error\n"); 593 break; 594 case 4: 595 dev_err(edac_dev->dev, 596 "DTB multihit from load pipeline error\n"); 597 break; 598 case 5: 599 dev_err(edac_dev->dev, 600 "DTB multihit from store pipeline error\n"); 601 break; 602 } 603 604 /* Clear any HW errors */ 605 writel(val, pg_f + MEMERR_CPU_LSUESR_PAGE_OFFSET); 606 607 if (val & (MEMERR_CPU_LSUESR_CERR_MASK | 608 MEMERR_CPU_LSUESR_MULTCERR_MASK)) 609 edac_device_handle_ce(edac_dev, 0, 0, edac_dev->ctl_name); 610 611chk_mmu: 612 val = readl(pg_f + MEMERR_CPU_MMUESR_PAGE_OFFSET); 613 if (!val) 614 return; 615 dev_err(edac_dev->dev, 616 "CPU%d memory error MMU 0x%08X Way 0x%02X Index 0x%02X Info 0x%02X %s\n", 617 ctx->pmd * MAX_CPU_PER_PMD + cpu_idx, val, 618 MEMERR_CPU_MMUESR_ERRWAY_RD(val), 619 MEMERR_CPU_MMUESR_ERRINDEX_RD(val), 620 MEMERR_CPU_MMUESR_ERRINFO_RD(val), 621 val & MEMERR_CPU_MMUESR_ERRREQSTR_LSU_MASK ? "LSU" : "ICF"); 622 if (val & MEMERR_CPU_MMUESR_CERR_MASK) 623 dev_err(edac_dev->dev, "One or more correctable error\n"); 624 if (val & MEMERR_CPU_MMUESR_MULTCERR_MASK) 625 dev_err(edac_dev->dev, "Multiple correctable error\n"); 626 switch (MEMERR_CPU_MMUESR_ERRTYPE_RD(val)) { 627 case 0: 628 dev_err(edac_dev->dev, "Stage 1 UTB hit error\n"); 629 break; 630 case 1: 631 dev_err(edac_dev->dev, "Stage 1 UTB miss error\n"); 632 break; 633 case 2: 634 dev_err(edac_dev->dev, "Stage 1 UTB allocate error\n"); 635 break; 636 case 3: 637 dev_err(edac_dev->dev, "TMO operation single bank error\n"); 638 break; 639 case 4: 640 dev_err(edac_dev->dev, "Stage 2 UTB error\n"); 641 break; 642 case 5: 643 dev_err(edac_dev->dev, "Stage 2 UTB miss error\n"); 644 break; 645 case 6: 646 dev_err(edac_dev->dev, "Stage 2 UTB allocate error\n"); 647 break; 648 case 7: 649 dev_err(edac_dev->dev, "TMO operation multiple bank error\n"); 650 break; 651 } 652 653 /* Clear any HW errors */ 654 writel(val, pg_f + MEMERR_CPU_MMUESR_PAGE_OFFSET); 655 656 edac_device_handle_ce(edac_dev, 0, 0, edac_dev->ctl_name); 657} 658 659static void xgene_edac_pmd_l2_check(struct edac_device_ctl_info *edac_dev) 660{ 661 struct xgene_edac_pmd_ctx *ctx = edac_dev->pvt_info; 662 void __iomem *pg_d; 663 void __iomem *pg_e; 664 u32 val_hi; 665 u32 val_lo; 666 u32 val; 667 668 /* Check L2 */ 669 pg_e = ctx->pmd_csr + CPU_MEMERR_L2C_PAGE; 670 val = readl(pg_e + MEMERR_L2C_L2ESR_PAGE_OFFSET); 671 if (!val) 672 goto chk_l2c; 673 val_lo = readl(pg_e + MEMERR_L2C_L2EALR_PAGE_OFFSET); 674 val_hi = readl(pg_e + MEMERR_L2C_L2EAHR_PAGE_OFFSET); 675 dev_err(edac_dev->dev, 676 "PMD%d memory error L2C L2ESR 0x%08X @ 0x%08X.%08X\n", 677 ctx->pmd, val, val_hi, val_lo); 678 dev_err(edac_dev->dev, 679 "ErrSyndrome 0x%02X ErrWay 0x%02X ErrCpu %d ErrGroup 0x%02X ErrAction 0x%02X\n", 680 MEMERR_L2C_L2ESR_ERRSYN_RD(val), 681 MEMERR_L2C_L2ESR_ERRWAY_RD(val), 682 MEMERR_L2C_L2ESR_ERRCPU_RD(val), 683 MEMERR_L2C_L2ESR_ERRGROUP_RD(val), 684 MEMERR_L2C_L2ESR_ERRACTION_RD(val)); 685 686 if (val & MEMERR_L2C_L2ESR_ERR_MASK) 687 dev_err(edac_dev->dev, "One or more correctable error\n"); 688 if (val & MEMERR_L2C_L2ESR_MULTICERR_MASK) 689 dev_err(edac_dev->dev, "Multiple correctable error\n"); 690 if (val & MEMERR_L2C_L2ESR_UCERR_MASK) 691 dev_err(edac_dev->dev, "One or more uncorrectable error\n"); 692 if (val & MEMERR_L2C_L2ESR_MULTUCERR_MASK) 693 dev_err(edac_dev->dev, "Multiple uncorrectable error\n"); 694 695 switch (MEMERR_L2C_L2ESR_ERRTYPE_RD(val)) { 696 case 0: 697 dev_err(edac_dev->dev, "Outbound SDB parity error\n"); 698 break; 699 case 1: 700 dev_err(edac_dev->dev, "Inbound SDB parity error\n"); 701 break; 702 case 2: 703 dev_err(edac_dev->dev, "Tag ECC error\n"); 704 break; 705 case 3: 706 dev_err(edac_dev->dev, "Data ECC error\n"); 707 break; 708 } 709 710 /* Clear any HW errors */ 711 writel(val, pg_e + MEMERR_L2C_L2ESR_PAGE_OFFSET); 712 713 if (val & (MEMERR_L2C_L2ESR_ERR_MASK | 714 MEMERR_L2C_L2ESR_MULTICERR_MASK)) 715 edac_device_handle_ce(edac_dev, 0, 0, edac_dev->ctl_name); 716 if (val & (MEMERR_L2C_L2ESR_UCERR_MASK | 717 MEMERR_L2C_L2ESR_MULTUCERR_MASK)) 718 edac_device_handle_ue(edac_dev, 0, 0, edac_dev->ctl_name); 719 720chk_l2c: 721 /* Check if any memory request timed out on L2 cache */ 722 pg_d = ctx->pmd_csr + CPU_L2C_PAGE; 723 val = readl(pg_d + CPUX_L2C_L2RTOSR_PAGE_OFFSET); 724 if (val) { 725 val_lo = readl(pg_d + CPUX_L2C_L2RTOALR_PAGE_OFFSET); 726 val_hi = readl(pg_d + CPUX_L2C_L2RTOAHR_PAGE_OFFSET); 727 dev_err(edac_dev->dev, 728 "PMD%d L2C error L2C RTOSR 0x%08X @ 0x%08X.%08X\n", 729 ctx->pmd, val, val_hi, val_lo); 730 writel(val, pg_d + CPUX_L2C_L2RTOSR_PAGE_OFFSET); 731 } 732} 733 734static void xgene_edac_pmd_check(struct edac_device_ctl_info *edac_dev) 735{ 736 struct xgene_edac_pmd_ctx *ctx = edac_dev->pvt_info; 737 unsigned int pcp_hp_stat; 738 int i; 739 740 xgene_edac_pcp_rd(ctx->edac, PCPHPERRINTSTS, &pcp_hp_stat); 741 if (!((PMD0_MERR_MASK << ctx->pmd) & pcp_hp_stat)) 742 return; 743 744 /* Check CPU L1 error */ 745 for (i = 0; i < MAX_CPU_PER_PMD; i++) 746 xgene_edac_pmd_l1_check(edac_dev, i); 747 748 /* Check CPU L2 error */ 749 xgene_edac_pmd_l2_check(edac_dev); 750} 751 752static void xgene_edac_pmd_cpu_hw_cfg(struct edac_device_ctl_info *edac_dev, 753 int cpu) 754{ 755 struct xgene_edac_pmd_ctx *ctx = edac_dev->pvt_info; 756 void __iomem *pg_f = ctx->pmd_csr + cpu * CPU_CSR_STRIDE + 757 CPU_MEMERR_CPU_PAGE; 758 759 /* 760 * Enable CPU memory error: 761 * MEMERR_CPU_ICFESRA, MEMERR_CPU_LSUESRA, and MEMERR_CPU_MMUESRA 762 */ 763 writel(0x00000301, pg_f + MEMERR_CPU_ICFECR_PAGE_OFFSET); 764 writel(0x00000301, pg_f + MEMERR_CPU_LSUECR_PAGE_OFFSET); 765 writel(0x00000101, pg_f + MEMERR_CPU_MMUECR_PAGE_OFFSET); 766} 767 768static void xgene_edac_pmd_hw_cfg(struct edac_device_ctl_info *edac_dev) 769{ 770 struct xgene_edac_pmd_ctx *ctx = edac_dev->pvt_info; 771 void __iomem *pg_d = ctx->pmd_csr + CPU_L2C_PAGE; 772 void __iomem *pg_e = ctx->pmd_csr + CPU_MEMERR_L2C_PAGE; 773 774 /* Enable PMD memory error - MEMERR_L2C_L2ECR and L2C_L2RTOCR */ 775 writel(0x00000703, pg_e + MEMERR_L2C_L2ECR_PAGE_OFFSET); 776 /* Configure L2C HW request time out feature if supported */ 777 if (ctx->version > 1) 778 writel(0x00000119, pg_d + CPUX_L2C_L2RTOCR_PAGE_OFFSET); 779} 780 781static void xgene_edac_pmd_hw_ctl(struct edac_device_ctl_info *edac_dev, 782 bool enable) 783{ 784 struct xgene_edac_pmd_ctx *ctx = edac_dev->pvt_info; 785 int i; 786 787 /* Enable PMD error interrupt */ 788 if (edac_dev->op_state == OP_RUNNING_INTERRUPT) { 789 if (enable) 790 xgene_edac_pcp_clrbits(ctx->edac, PCPHPERRINTMSK, 791 PMD0_MERR_MASK << ctx->pmd); 792 else 793 xgene_edac_pcp_setbits(ctx->edac, PCPHPERRINTMSK, 794 PMD0_MERR_MASK << ctx->pmd); 795 } 796 797 if (enable) { 798 xgene_edac_pmd_hw_cfg(edac_dev); 799 800 /* Two CPUs per a PMD */ 801 for (i = 0; i < MAX_CPU_PER_PMD; i++) 802 xgene_edac_pmd_cpu_hw_cfg(edac_dev, i); 803 } 804} 805 806static ssize_t xgene_edac_pmd_l1_inject_ctrl_write(struct file *file, 807 const char __user *data, 808 size_t count, loff_t *ppos) 809{ 810 struct edac_device_ctl_info *edac_dev = file->private_data; 811 struct xgene_edac_pmd_ctx *ctx = edac_dev->pvt_info; 812 void __iomem *cpux_pg_f; 813 int i; 814 815 for (i = 0; i < MAX_CPU_PER_PMD; i++) { 816 cpux_pg_f = ctx->pmd_csr + i * CPU_CSR_STRIDE + 817 CPU_MEMERR_CPU_PAGE; 818 819 writel(MEMERR_CPU_ICFESR_MULTCERR_MASK | 820 MEMERR_CPU_ICFESR_CERR_MASK, 821 cpux_pg_f + MEMERR_CPU_ICFESRA_PAGE_OFFSET); 822 writel(MEMERR_CPU_LSUESR_MULTCERR_MASK | 823 MEMERR_CPU_LSUESR_CERR_MASK, 824 cpux_pg_f + MEMERR_CPU_LSUESRA_PAGE_OFFSET); 825 writel(MEMERR_CPU_MMUESR_MULTCERR_MASK | 826 MEMERR_CPU_MMUESR_CERR_MASK, 827 cpux_pg_f + MEMERR_CPU_MMUESRA_PAGE_OFFSET); 828 } 829 return count; 830} 831 832static ssize_t xgene_edac_pmd_l2_inject_ctrl_write(struct file *file, 833 const char __user *data, 834 size_t count, loff_t *ppos) 835{ 836 struct edac_device_ctl_info *edac_dev = file->private_data; 837 struct xgene_edac_pmd_ctx *ctx = edac_dev->pvt_info; 838 void __iomem *pg_e = ctx->pmd_csr + CPU_MEMERR_L2C_PAGE; 839 840 writel(MEMERR_L2C_L2ESR_MULTUCERR_MASK | 841 MEMERR_L2C_L2ESR_MULTICERR_MASK | 842 MEMERR_L2C_L2ESR_UCERR_MASK | 843 MEMERR_L2C_L2ESR_ERR_MASK, 844 pg_e + MEMERR_L2C_L2ESRA_PAGE_OFFSET); 845 return count; 846} 847 848static const struct file_operations xgene_edac_pmd_debug_inject_fops[] = { 849 { 850 .open = simple_open, 851 .write = xgene_edac_pmd_l1_inject_ctrl_write, 852 .llseek = generic_file_llseek, }, 853 { 854 .open = simple_open, 855 .write = xgene_edac_pmd_l2_inject_ctrl_write, 856 .llseek = generic_file_llseek, }, 857 { } 858}; 859 860static void 861xgene_edac_pmd_create_debugfs_nodes(struct edac_device_ctl_info *edac_dev) 862{ 863 struct xgene_edac_pmd_ctx *ctx = edac_dev->pvt_info; 864 struct dentry *dbgfs_dir; 865 char name[10]; 866 867 if (!IS_ENABLED(CONFIG_EDAC_DEBUG) || !ctx->edac->dfs) 868 return; 869 870 snprintf(name, sizeof(name), "PMD%d", ctx->pmd); 871 dbgfs_dir = edac_debugfs_create_dir_at(name, ctx->edac->dfs); 872 if (!dbgfs_dir) 873 return; 874 875 edac_debugfs_create_file("l1_inject_ctrl", S_IWUSR, dbgfs_dir, edac_dev, 876 &xgene_edac_pmd_debug_inject_fops[0]); 877 edac_debugfs_create_file("l2_inject_ctrl", S_IWUSR, dbgfs_dir, edac_dev, 878 &xgene_edac_pmd_debug_inject_fops[1]); 879} 880 881static int xgene_edac_pmd_available(u32 efuse, int pmd) 882{ 883 return (efuse & (1 << pmd)) ? 0 : 1; 884} 885 886static int xgene_edac_pmd_add(struct xgene_edac *edac, struct device_node *np, 887 int version) 888{ 889 struct edac_device_ctl_info *edac_dev; 890 struct xgene_edac_pmd_ctx *ctx; 891 struct resource res; 892 char edac_name[10]; 893 u32 pmd; 894 int rc; 895 u32 val; 896 897 if (!devres_open_group(edac->dev, xgene_edac_pmd_add, GFP_KERNEL)) 898 return -ENOMEM; 899 900 /* Determine if this PMD is disabled */ 901 if (of_property_read_u32(np, "pmd-controller", &pmd)) { 902 dev_err(edac->dev, "no pmd-controller property\n"); 903 rc = -ENODEV; 904 goto err_group; 905 } 906 rc = regmap_read(edac->efuse_map, 0, &val); 907 if (rc) 908 goto err_group; 909 if (!xgene_edac_pmd_available(val, pmd)) { 910 rc = -ENODEV; 911 goto err_group; 912 } 913 914 snprintf(edac_name, sizeof(edac_name), "l2c%d", pmd); 915 edac_dev = edac_device_alloc_ctl_info(sizeof(*ctx), 916 edac_name, 1, "l2c", 1, 2, NULL, 917 0, edac_device_alloc_index()); 918 if (!edac_dev) { 919 rc = -ENOMEM; 920 goto err_group; 921 } 922 923 ctx = edac_dev->pvt_info; 924 ctx->name = "xgene_pmd_err"; 925 ctx->pmd = pmd; 926 ctx->edac = edac; 927 ctx->edac_dev = edac_dev; 928 ctx->ddev = *edac->dev; 929 ctx->version = version; 930 edac_dev->dev = &ctx->ddev; 931 edac_dev->ctl_name = ctx->name; 932 edac_dev->dev_name = ctx->name; 933 edac_dev->mod_name = EDAC_MOD_STR; 934 935 rc = of_address_to_resource(np, 0, &res); 936 if (rc < 0) { 937 dev_err(edac->dev, "no PMD resource address\n"); 938 goto err_free; 939 } 940 ctx->pmd_csr = devm_ioremap_resource(edac->dev, &res); 941 if (IS_ERR(ctx->pmd_csr)) { 942 dev_err(edac->dev, 943 "devm_ioremap_resource failed for PMD resource address\n"); 944 rc = PTR_ERR(ctx->pmd_csr); 945 goto err_free; 946 } 947 948 if (edac_op_state == EDAC_OPSTATE_POLL) 949 edac_dev->edac_check = xgene_edac_pmd_check; 950 951 xgene_edac_pmd_create_debugfs_nodes(edac_dev); 952 953 rc = edac_device_add_device(edac_dev); 954 if (rc > 0) { 955 dev_err(edac->dev, "edac_device_add_device failed\n"); 956 rc = -ENOMEM; 957 goto err_free; 958 } 959 960 if (edac_op_state == EDAC_OPSTATE_INT) 961 edac_dev->op_state = OP_RUNNING_INTERRUPT; 962 963 list_add(&ctx->next, &edac->pmds); 964 965 xgene_edac_pmd_hw_ctl(edac_dev, 1); 966 967 devres_remove_group(edac->dev, xgene_edac_pmd_add); 968 969 dev_info(edac->dev, "X-Gene EDAC PMD%d registered\n", ctx->pmd); 970 return 0; 971 972err_free: 973 edac_device_free_ctl_info(edac_dev); 974err_group: 975 devres_release_group(edac->dev, xgene_edac_pmd_add); 976 return rc; 977} 978 979static int xgene_edac_pmd_remove(struct xgene_edac_pmd_ctx *pmd) 980{ 981 struct edac_device_ctl_info *edac_dev = pmd->edac_dev; 982 983 xgene_edac_pmd_hw_ctl(edac_dev, 0); 984 edac_device_del_device(edac_dev->dev); 985 edac_device_free_ctl_info(edac_dev); 986 return 0; 987} 988 989/* L3 Error device */ 990#define L3C_ESR (0x0A * 4) 991#define L3C_ESR_DATATAG_MASK BIT(9) 992#define L3C_ESR_MULTIHIT_MASK BIT(8) 993#define L3C_ESR_UCEVICT_MASK BIT(6) 994#define L3C_ESR_MULTIUCERR_MASK BIT(5) 995#define L3C_ESR_MULTICERR_MASK BIT(4) 996#define L3C_ESR_UCERR_MASK BIT(3) 997#define L3C_ESR_CERR_MASK BIT(2) 998#define L3C_ESR_UCERRINTR_MASK BIT(1) 999#define L3C_ESR_CERRINTR_MASK BIT(0) 1000#define L3C_ECR (0x0B * 4) 1001#define L3C_ECR_UCINTREN BIT(3) 1002#define L3C_ECR_CINTREN BIT(2) 1003#define L3C_UCERREN BIT(1) 1004#define L3C_CERREN BIT(0) 1005#define L3C_ELR (0x0C * 4) 1006#define L3C_ELR_ERRSYN(src) ((src & 0xFF800000) >> 23) 1007#define L3C_ELR_ERRWAY(src) ((src & 0x007E0000) >> 17) 1008#define L3C_ELR_AGENTID(src) ((src & 0x0001E000) >> 13) 1009#define L3C_ELR_ERRGRP(src) ((src & 0x00000F00) >> 8) 1010#define L3C_ELR_OPTYPE(src) ((src & 0x000000F0) >> 4) 1011#define L3C_ELR_PADDRHIGH(src) (src & 0x0000000F) 1012#define L3C_AELR (0x0D * 4) 1013#define L3C_BELR (0x0E * 4) 1014#define L3C_BELR_BANK(src) (src & 0x0000000F) 1015 1016struct xgene_edac_dev_ctx { 1017 struct list_head next; 1018 struct device ddev; 1019 char *name; 1020 struct xgene_edac *edac; 1021 struct edac_device_ctl_info *edac_dev; 1022 int edac_idx; 1023 void __iomem *dev_csr; 1024 int version; 1025}; 1026 1027/* 1028 * Version 1 of the L3 controller has broken single bit correctable logic for 1029 * certain error syndromes. Log them as uncorrectable in that case. 1030 */ 1031static bool xgene_edac_l3_promote_to_uc_err(u32 l3cesr, u32 l3celr) 1032{ 1033 if (l3cesr & L3C_ESR_DATATAG_MASK) { 1034 switch (L3C_ELR_ERRSYN(l3celr)) { 1035 case 0x13C: 1036 case 0x0B4: 1037 case 0x007: 1038 case 0x00D: 1039 case 0x00E: 1040 case 0x019: 1041 case 0x01A: 1042 case 0x01C: 1043 case 0x04E: 1044 case 0x041: 1045 return true; 1046 } 1047 } else if (L3C_ELR_ERRWAY(l3celr) == 9) 1048 return true; 1049 1050 return false; 1051} 1052 1053static void xgene_edac_l3_check(struct edac_device_ctl_info *edac_dev) 1054{ 1055 struct xgene_edac_dev_ctx *ctx = edac_dev->pvt_info; 1056 u32 l3cesr; 1057 u32 l3celr; 1058 u32 l3caelr; 1059 u32 l3cbelr; 1060 1061 l3cesr = readl(ctx->dev_csr + L3C_ESR); 1062 if (!(l3cesr & (L3C_ESR_UCERR_MASK | L3C_ESR_CERR_MASK))) 1063 return; 1064 1065 if (l3cesr & L3C_ESR_UCERR_MASK) 1066 dev_err(edac_dev->dev, "L3C uncorrectable error\n"); 1067 if (l3cesr & L3C_ESR_CERR_MASK) 1068 dev_warn(edac_dev->dev, "L3C correctable error\n"); 1069 1070 l3celr = readl(ctx->dev_csr + L3C_ELR); 1071 l3caelr = readl(ctx->dev_csr + L3C_AELR); 1072 l3cbelr = readl(ctx->dev_csr + L3C_BELR); 1073 if (l3cesr & L3C_ESR_MULTIHIT_MASK) 1074 dev_err(edac_dev->dev, "L3C multiple hit error\n"); 1075 if (l3cesr & L3C_ESR_UCEVICT_MASK) 1076 dev_err(edac_dev->dev, 1077 "L3C dropped eviction of line with error\n"); 1078 if (l3cesr & L3C_ESR_MULTIUCERR_MASK) 1079 dev_err(edac_dev->dev, "L3C multiple uncorrectable error\n"); 1080 if (l3cesr & L3C_ESR_DATATAG_MASK) 1081 dev_err(edac_dev->dev, 1082 "L3C data error syndrome 0x%X group 0x%X\n", 1083 L3C_ELR_ERRSYN(l3celr), L3C_ELR_ERRGRP(l3celr)); 1084 else 1085 dev_err(edac_dev->dev, 1086 "L3C tag error syndrome 0x%X Way of Tag 0x%X Agent ID 0x%X Operation type 0x%X\n", 1087 L3C_ELR_ERRSYN(l3celr), L3C_ELR_ERRWAY(l3celr), 1088 L3C_ELR_AGENTID(l3celr), L3C_ELR_OPTYPE(l3celr)); 1089 /* 1090 * NOTE: Address [41:38] in L3C_ELR_PADDRHIGH(l3celr). 1091 * Address [37:6] in l3caelr. Lower 6 bits are zero. 1092 */ 1093 dev_err(edac_dev->dev, "L3C error address 0x%08X.%08X bank %d\n", 1094 L3C_ELR_PADDRHIGH(l3celr) << 6 | (l3caelr >> 26), 1095 (l3caelr & 0x3FFFFFFF) << 6, L3C_BELR_BANK(l3cbelr)); 1096 dev_err(edac_dev->dev, 1097 "L3C error status register value 0x%X\n", l3cesr); 1098 1099 /* Clear L3C error interrupt */ 1100 writel(0, ctx->dev_csr + L3C_ESR); 1101 1102 if (ctx->version <= 1 && 1103 xgene_edac_l3_promote_to_uc_err(l3cesr, l3celr)) { 1104 edac_device_handle_ue(edac_dev, 0, 0, edac_dev->ctl_name); 1105 return; 1106 } 1107 if (l3cesr & L3C_ESR_CERR_MASK) 1108 edac_device_handle_ce(edac_dev, 0, 0, edac_dev->ctl_name); 1109 if (l3cesr & L3C_ESR_UCERR_MASK) 1110 edac_device_handle_ue(edac_dev, 0, 0, edac_dev->ctl_name); 1111} 1112 1113static void xgene_edac_l3_hw_init(struct edac_device_ctl_info *edac_dev, 1114 bool enable) 1115{ 1116 struct xgene_edac_dev_ctx *ctx = edac_dev->pvt_info; 1117 u32 val; 1118 1119 val = readl(ctx->dev_csr + L3C_ECR); 1120 val |= L3C_UCERREN | L3C_CERREN; 1121 /* On disable, we just disable interrupt but keep error enabled */ 1122 if (edac_dev->op_state == OP_RUNNING_INTERRUPT) { 1123 if (enable) 1124 val |= L3C_ECR_UCINTREN | L3C_ECR_CINTREN; 1125 else 1126 val &= ~(L3C_ECR_UCINTREN | L3C_ECR_CINTREN); 1127 } 1128 writel(val, ctx->dev_csr + L3C_ECR); 1129 1130 if (edac_dev->op_state == OP_RUNNING_INTERRUPT) { 1131 /* Enable/disable L3 error top level interrupt */ 1132 if (enable) { 1133 xgene_edac_pcp_clrbits(ctx->edac, PCPHPERRINTMSK, 1134 L3C_UNCORR_ERR_MASK); 1135 xgene_edac_pcp_clrbits(ctx->edac, PCPLPERRINTMSK, 1136 L3C_CORR_ERR_MASK); 1137 } else { 1138 xgene_edac_pcp_setbits(ctx->edac, PCPHPERRINTMSK, 1139 L3C_UNCORR_ERR_MASK); 1140 xgene_edac_pcp_setbits(ctx->edac, PCPLPERRINTMSK, 1141 L3C_CORR_ERR_MASK); 1142 } 1143 } 1144} 1145 1146static ssize_t xgene_edac_l3_inject_ctrl_write(struct file *file, 1147 const char __user *data, 1148 size_t count, loff_t *ppos) 1149{ 1150 struct edac_device_ctl_info *edac_dev = file->private_data; 1151 struct xgene_edac_dev_ctx *ctx = edac_dev->pvt_info; 1152 1153 /* Generate all errors */ 1154 writel(0xFFFFFFFF, ctx->dev_csr + L3C_ESR); 1155 return count; 1156} 1157 1158static const struct file_operations xgene_edac_l3_debug_inject_fops = { 1159 .open = simple_open, 1160 .write = xgene_edac_l3_inject_ctrl_write, 1161 .llseek = generic_file_llseek 1162}; 1163 1164static void 1165xgene_edac_l3_create_debugfs_nodes(struct edac_device_ctl_info *edac_dev) 1166{ 1167 struct xgene_edac_dev_ctx *ctx = edac_dev->pvt_info; 1168 struct dentry *dbgfs_dir; 1169 char name[10]; 1170 1171 if (!IS_ENABLED(CONFIG_EDAC_DEBUG) || !ctx->edac->dfs) 1172 return; 1173 1174 snprintf(name, sizeof(name), "l3c%d", ctx->edac_idx); 1175 dbgfs_dir = edac_debugfs_create_dir_at(name, ctx->edac->dfs); 1176 if (!dbgfs_dir) 1177 return; 1178 1179 debugfs_create_file("l3_inject_ctrl", S_IWUSR, dbgfs_dir, edac_dev, 1180 &xgene_edac_l3_debug_inject_fops); 1181} 1182 1183static int xgene_edac_l3_add(struct xgene_edac *edac, struct device_node *np, 1184 int version) 1185{ 1186 struct edac_device_ctl_info *edac_dev; 1187 struct xgene_edac_dev_ctx *ctx; 1188 struct resource res; 1189 void __iomem *dev_csr; 1190 int edac_idx; 1191 int rc = 0; 1192 1193 if (!devres_open_group(edac->dev, xgene_edac_l3_add, GFP_KERNEL)) 1194 return -ENOMEM; 1195 1196 rc = of_address_to_resource(np, 0, &res); 1197 if (rc < 0) { 1198 dev_err(edac->dev, "no L3 resource address\n"); 1199 goto err_release_group; 1200 } 1201 dev_csr = devm_ioremap_resource(edac->dev, &res); 1202 if (IS_ERR(dev_csr)) { 1203 dev_err(edac->dev, 1204 "devm_ioremap_resource failed for L3 resource address\n"); 1205 rc = PTR_ERR(dev_csr); 1206 goto err_release_group; 1207 } 1208 1209 edac_idx = edac_device_alloc_index(); 1210 edac_dev = edac_device_alloc_ctl_info(sizeof(*ctx), 1211 "l3c", 1, "l3c", 1, 0, NULL, 0, 1212 edac_idx); 1213 if (!edac_dev) { 1214 rc = -ENOMEM; 1215 goto err_release_group; 1216 } 1217 1218 ctx = edac_dev->pvt_info; 1219 ctx->dev_csr = dev_csr; 1220 ctx->name = "xgene_l3_err"; 1221 ctx->edac_idx = edac_idx; 1222 ctx->edac = edac; 1223 ctx->edac_dev = edac_dev; 1224 ctx->ddev = *edac->dev; 1225 ctx->version = version; 1226 edac_dev->dev = &ctx->ddev; 1227 edac_dev->ctl_name = ctx->name; 1228 edac_dev->dev_name = ctx->name; 1229 edac_dev->mod_name = EDAC_MOD_STR; 1230 1231 if (edac_op_state == EDAC_OPSTATE_POLL) 1232 edac_dev->edac_check = xgene_edac_l3_check; 1233 1234 xgene_edac_l3_create_debugfs_nodes(edac_dev); 1235 1236 rc = edac_device_add_device(edac_dev); 1237 if (rc > 0) { 1238 dev_err(edac->dev, "failed edac_device_add_device()\n"); 1239 rc = -ENOMEM; 1240 goto err_ctl_free; 1241 } 1242 1243 if (edac_op_state == EDAC_OPSTATE_INT) 1244 edac_dev->op_state = OP_RUNNING_INTERRUPT; 1245 1246 list_add(&ctx->next, &edac->l3s); 1247 1248 xgene_edac_l3_hw_init(edac_dev, 1); 1249 1250 devres_remove_group(edac->dev, xgene_edac_l3_add); 1251 1252 dev_info(edac->dev, "X-Gene EDAC L3 registered\n"); 1253 return 0; 1254 1255err_ctl_free: 1256 edac_device_free_ctl_info(edac_dev); 1257err_release_group: 1258 devres_release_group(edac->dev, xgene_edac_l3_add); 1259 return rc; 1260} 1261 1262static int xgene_edac_l3_remove(struct xgene_edac_dev_ctx *l3) 1263{ 1264 struct edac_device_ctl_info *edac_dev = l3->edac_dev; 1265 1266 xgene_edac_l3_hw_init(edac_dev, 0); 1267 edac_device_del_device(l3->edac->dev); 1268 edac_device_free_ctl_info(edac_dev); 1269 return 0; 1270} 1271 1272/* SoC error device */ 1273#define IOBAXIS0TRANSERRINTSTS 0x0000 1274#define IOBAXIS0_M_ILLEGAL_ACCESS_MASK BIT(1) 1275#define IOBAXIS0_ILLEGAL_ACCESS_MASK BIT(0) 1276#define IOBAXIS0TRANSERRINTMSK 0x0004 1277#define IOBAXIS0TRANSERRREQINFOL 0x0008 1278#define IOBAXIS0TRANSERRREQINFOH 0x000c 1279#define REQTYPE_RD(src) (((src) & BIT(0))) 1280#define ERRADDRH_RD(src) (((src) & 0xffc00000) >> 22) 1281#define IOBAXIS1TRANSERRINTSTS 0x0010 1282#define IOBAXIS1TRANSERRINTMSK 0x0014 1283#define IOBAXIS1TRANSERRREQINFOL 0x0018 1284#define IOBAXIS1TRANSERRREQINFOH 0x001c 1285#define IOBPATRANSERRINTSTS 0x0020 1286#define IOBPA_M_REQIDRAM_CORRUPT_MASK BIT(7) 1287#define IOBPA_REQIDRAM_CORRUPT_MASK BIT(6) 1288#define IOBPA_M_TRANS_CORRUPT_MASK BIT(5) 1289#define IOBPA_TRANS_CORRUPT_MASK BIT(4) 1290#define IOBPA_M_WDATA_CORRUPT_MASK BIT(3) 1291#define IOBPA_WDATA_CORRUPT_MASK BIT(2) 1292#define IOBPA_M_RDATA_CORRUPT_MASK BIT(1) 1293#define IOBPA_RDATA_CORRUPT_MASK BIT(0) 1294#define IOBBATRANSERRINTSTS 0x0030 1295#define M_ILLEGAL_ACCESS_MASK BIT(15) 1296#define ILLEGAL_ACCESS_MASK BIT(14) 1297#define M_WIDRAM_CORRUPT_MASK BIT(13) 1298#define WIDRAM_CORRUPT_MASK BIT(12) 1299#define M_RIDRAM_CORRUPT_MASK BIT(11) 1300#define RIDRAM_CORRUPT_MASK BIT(10) 1301#define M_TRANS_CORRUPT_MASK BIT(9) 1302#define TRANS_CORRUPT_MASK BIT(8) 1303#define M_WDATA_CORRUPT_MASK BIT(7) 1304#define WDATA_CORRUPT_MASK BIT(6) 1305#define M_RBM_POISONED_REQ_MASK BIT(5) 1306#define RBM_POISONED_REQ_MASK BIT(4) 1307#define M_XGIC_POISONED_REQ_MASK BIT(3) 1308#define XGIC_POISONED_REQ_MASK BIT(2) 1309#define M_WRERR_RESP_MASK BIT(1) 1310#define WRERR_RESP_MASK BIT(0) 1311#define IOBBATRANSERRREQINFOL 0x0038 1312#define IOBBATRANSERRREQINFOH 0x003c 1313#define REQTYPE_F2_RD(src) ((src) & BIT(0)) 1314#define ERRADDRH_F2_RD(src) (((src) & 0xffc00000) >> 22) 1315#define IOBBATRANSERRCSWREQID 0x0040 1316#define XGICTRANSERRINTSTS 0x0050 1317#define M_WR_ACCESS_ERR_MASK BIT(3) 1318#define WR_ACCESS_ERR_MASK BIT(2) 1319#define M_RD_ACCESS_ERR_MASK BIT(1) 1320#define RD_ACCESS_ERR_MASK BIT(0) 1321#define XGICTRANSERRINTMSK 0x0054 1322#define XGICTRANSERRREQINFO 0x0058 1323#define REQTYPE_MASK BIT(26) 1324#define ERRADDR_RD(src) ((src) & 0x03ffffff) 1325#define GLBL_ERR_STS 0x0800 1326#define MDED_ERR_MASK BIT(3) 1327#define DED_ERR_MASK BIT(2) 1328#define MSEC_ERR_MASK BIT(1) 1329#define SEC_ERR_MASK BIT(0) 1330#define GLBL_SEC_ERRL 0x0810 1331#define GLBL_SEC_ERRH 0x0818 1332#define GLBL_MSEC_ERRL 0x0820 1333#define GLBL_MSEC_ERRH 0x0828 1334#define GLBL_DED_ERRL 0x0830 1335#define GLBL_DED_ERRLMASK 0x0834 1336#define GLBL_DED_ERRH 0x0838 1337#define GLBL_DED_ERRHMASK 0x083c 1338#define GLBL_MDED_ERRL 0x0840 1339#define GLBL_MDED_ERRLMASK 0x0844 1340#define GLBL_MDED_ERRH 0x0848 1341#define GLBL_MDED_ERRHMASK 0x084c 1342 1343/* IO Bus Registers */ 1344#define RBCSR 0x0000 1345#define STICKYERR_MASK BIT(0) 1346#define RBEIR 0x0008 1347#define AGENT_OFFLINE_ERR_MASK BIT(30) 1348#define UNIMPL_RBPAGE_ERR_MASK BIT(29) 1349#define WORD_ALIGNED_ERR_MASK BIT(28) 1350#define PAGE_ACCESS_ERR_MASK BIT(27) 1351#define WRITE_ACCESS_MASK BIT(26) 1352 1353static const char * const soc_mem_err_v1[] = { 1354 "10GbE0", 1355 "10GbE1", 1356 "Security", 1357 "SATA45", 1358 "SATA23/ETH23", 1359 "SATA01/ETH01", 1360 "USB1", 1361 "USB0", 1362 "QML", 1363 "QM0", 1364 "QM1 (XGbE01)", 1365 "PCIE4", 1366 "PCIE3", 1367 "PCIE2", 1368 "PCIE1", 1369 "PCIE0", 1370 "CTX Manager", 1371 "OCM", 1372 "1GbE", 1373 "CLE", 1374 "AHBC", 1375 "PktDMA", 1376 "GFC", 1377 "MSLIM", 1378 "10GbE2", 1379 "10GbE3", 1380 "QM2 (XGbE23)", 1381 "IOB", 1382 "unknown", 1383 "unknown", 1384 "unknown", 1385 "unknown", 1386}; 1387 1388static void xgene_edac_iob_gic_report(struct edac_device_ctl_info *edac_dev) 1389{ 1390 struct xgene_edac_dev_ctx *ctx = edac_dev->pvt_info; 1391 u32 err_addr_lo; 1392 u32 err_addr_hi; 1393 u32 reg; 1394 u32 info; 1395 1396 /* GIC transaction error interrupt */ 1397 reg = readl(ctx->dev_csr + XGICTRANSERRINTSTS); 1398 if (!reg) 1399 goto chk_iob_err; 1400 dev_err(edac_dev->dev, "XGIC transaction error\n"); 1401 if (reg & RD_ACCESS_ERR_MASK) 1402 dev_err(edac_dev->dev, "XGIC read size error\n"); 1403 if (reg & M_RD_ACCESS_ERR_MASK) 1404 dev_err(edac_dev->dev, "Multiple XGIC read size error\n"); 1405 if (reg & WR_ACCESS_ERR_MASK) 1406 dev_err(edac_dev->dev, "XGIC write size error\n"); 1407 if (reg & M_WR_ACCESS_ERR_MASK) 1408 dev_err(edac_dev->dev, "Multiple XGIC write size error\n"); 1409 info = readl(ctx->dev_csr + XGICTRANSERRREQINFO); 1410 dev_err(edac_dev->dev, "XGIC %s access @ 0x%08X (0x%08X)\n", 1411 info & REQTYPE_MASK ? "read" : "write", ERRADDR_RD(info), 1412 info); 1413 writel(reg, ctx->dev_csr + XGICTRANSERRINTSTS); 1414 1415chk_iob_err: 1416 /* IOB memory error */ 1417 reg = readl(ctx->dev_csr + GLBL_ERR_STS); 1418 if (!reg) 1419 return; 1420 if (reg & SEC_ERR_MASK) { 1421 err_addr_lo = readl(ctx->dev_csr + GLBL_SEC_ERRL); 1422 err_addr_hi = readl(ctx->dev_csr + GLBL_SEC_ERRH); 1423 dev_err(edac_dev->dev, 1424 "IOB single-bit correctable memory at 0x%08X.%08X error\n", 1425 err_addr_lo, err_addr_hi); 1426 writel(err_addr_lo, ctx->dev_csr + GLBL_SEC_ERRL); 1427 writel(err_addr_hi, ctx->dev_csr + GLBL_SEC_ERRH); 1428 } 1429 if (reg & MSEC_ERR_MASK) { 1430 err_addr_lo = readl(ctx->dev_csr + GLBL_MSEC_ERRL); 1431 err_addr_hi = readl(ctx->dev_csr + GLBL_MSEC_ERRH); 1432 dev_err(edac_dev->dev, 1433 "IOB multiple single-bit correctable memory at 0x%08X.%08X error\n", 1434 err_addr_lo, err_addr_hi); 1435 writel(err_addr_lo, ctx->dev_csr + GLBL_MSEC_ERRL); 1436 writel(err_addr_hi, ctx->dev_csr + GLBL_MSEC_ERRH); 1437 } 1438 if (reg & (SEC_ERR_MASK | MSEC_ERR_MASK)) 1439 edac_device_handle_ce(edac_dev, 0, 0, edac_dev->ctl_name); 1440 1441 if (reg & DED_ERR_MASK) { 1442 err_addr_lo = readl(ctx->dev_csr + GLBL_DED_ERRL); 1443 err_addr_hi = readl(ctx->dev_csr + GLBL_DED_ERRH); 1444 dev_err(edac_dev->dev, 1445 "IOB double-bit uncorrectable memory at 0x%08X.%08X error\n", 1446 err_addr_lo, err_addr_hi); 1447 writel(err_addr_lo, ctx->dev_csr + GLBL_DED_ERRL); 1448 writel(err_addr_hi, ctx->dev_csr + GLBL_DED_ERRH); 1449 } 1450 if (reg & MDED_ERR_MASK) { 1451 err_addr_lo = readl(ctx->dev_csr + GLBL_MDED_ERRL); 1452 err_addr_hi = readl(ctx->dev_csr + GLBL_MDED_ERRH); 1453 dev_err(edac_dev->dev, 1454 "Multiple IOB double-bit uncorrectable memory at 0x%08X.%08X error\n", 1455 err_addr_lo, err_addr_hi); 1456 writel(err_addr_lo, ctx->dev_csr + GLBL_MDED_ERRL); 1457 writel(err_addr_hi, ctx->dev_csr + GLBL_MDED_ERRH); 1458 } 1459 if (reg & (DED_ERR_MASK | MDED_ERR_MASK)) 1460 edac_device_handle_ue(edac_dev, 0, 0, edac_dev->ctl_name); 1461} 1462 1463static void xgene_edac_rb_report(struct edac_device_ctl_info *edac_dev) 1464{ 1465 struct xgene_edac_dev_ctx *ctx = edac_dev->pvt_info; 1466 u32 err_addr_lo; 1467 u32 err_addr_hi; 1468 u32 reg; 1469 1470 /* If the register bus resource isn't available, just skip it */ 1471 if (!ctx->edac->rb_map) 1472 goto rb_skip; 1473 1474 /* 1475 * Check RB access errors 1476 * 1. Out of range 1477 * 2. Un-implemented page 1478 * 3. Un-aligned access 1479 * 4. Offline slave IP 1480 */ 1481 if (regmap_read(ctx->edac->rb_map, RBCSR, ®)) 1482 return; 1483 if (reg & STICKYERR_MASK) { 1484 bool write; 1485 1486 dev_err(edac_dev->dev, "IOB bus access error(s)\n"); 1487 if (regmap_read(ctx->edac->rb_map, RBEIR, ®)) 1488 return; 1489 write = reg & WRITE_ACCESS_MASK ? 1 : 0; 1490 if (reg & AGENT_OFFLINE_ERR_MASK) 1491 dev_err(edac_dev->dev, 1492 "IOB bus %s access to offline agent error\n", 1493 write ? "write" : "read"); 1494 if (reg & UNIMPL_RBPAGE_ERR_MASK) 1495 dev_err(edac_dev->dev, 1496 "IOB bus %s access to unimplemented page error\n", 1497 write ? "write" : "read"); 1498 if (reg & WORD_ALIGNED_ERR_MASK) 1499 dev_err(edac_dev->dev, 1500 "IOB bus %s word aligned access error\n", 1501 write ? "write" : "read"); 1502 if (reg & PAGE_ACCESS_ERR_MASK) 1503 dev_err(edac_dev->dev, 1504 "IOB bus %s to page out of range access error\n", 1505 write ? "write" : "read"); 1506 if (regmap_write(ctx->edac->rb_map, RBEIR, 0)) 1507 return; 1508 if (regmap_write(ctx->edac->rb_map, RBCSR, 0)) 1509 return; 1510 } 1511rb_skip: 1512 1513 /* IOB Bridge agent transaction error interrupt */ 1514 reg = readl(ctx->dev_csr + IOBBATRANSERRINTSTS); 1515 if (!reg) 1516 return; 1517 1518 dev_err(edac_dev->dev, "IOB bridge agent (BA) transaction error\n"); 1519 if (reg & WRERR_RESP_MASK) 1520 dev_err(edac_dev->dev, "IOB BA write response error\n"); 1521 if (reg & M_WRERR_RESP_MASK) 1522 dev_err(edac_dev->dev, 1523 "Multiple IOB BA write response error\n"); 1524 if (reg & XGIC_POISONED_REQ_MASK) 1525 dev_err(edac_dev->dev, "IOB BA XGIC poisoned write error\n"); 1526 if (reg & M_XGIC_POISONED_REQ_MASK) 1527 dev_err(edac_dev->dev, 1528 "Multiple IOB BA XGIC poisoned write error\n"); 1529 if (reg & RBM_POISONED_REQ_MASK) 1530 dev_err(edac_dev->dev, "IOB BA RBM poisoned write error\n"); 1531 if (reg & M_RBM_POISONED_REQ_MASK) 1532 dev_err(edac_dev->dev, 1533 "Multiple IOB BA RBM poisoned write error\n"); 1534 if (reg & WDATA_CORRUPT_MASK) 1535 dev_err(edac_dev->dev, "IOB BA write error\n"); 1536 if (reg & M_WDATA_CORRUPT_MASK) 1537 dev_err(edac_dev->dev, "Multiple IOB BA write error\n"); 1538 if (reg & TRANS_CORRUPT_MASK) 1539 dev_err(edac_dev->dev, "IOB BA transaction error\n"); 1540 if (reg & M_TRANS_CORRUPT_MASK) 1541 dev_err(edac_dev->dev, "Multiple IOB BA transaction error\n"); 1542 if (reg & RIDRAM_CORRUPT_MASK) 1543 dev_err(edac_dev->dev, 1544 "IOB BA RDIDRAM read transaction ID error\n"); 1545 if (reg & M_RIDRAM_CORRUPT_MASK) 1546 dev_err(edac_dev->dev, 1547 "Multiple IOB BA RDIDRAM read transaction ID error\n"); 1548 if (reg & WIDRAM_CORRUPT_MASK) 1549 dev_err(edac_dev->dev, 1550 "IOB BA RDIDRAM write transaction ID error\n"); 1551 if (reg & M_WIDRAM_CORRUPT_MASK) 1552 dev_err(edac_dev->dev, 1553 "Multiple IOB BA RDIDRAM write transaction ID error\n"); 1554 if (reg & ILLEGAL_ACCESS_MASK) 1555 dev_err(edac_dev->dev, 1556 "IOB BA XGIC/RB illegal access error\n"); 1557 if (reg & M_ILLEGAL_ACCESS_MASK) 1558 dev_err(edac_dev->dev, 1559 "Multiple IOB BA XGIC/RB illegal access error\n"); 1560 1561 err_addr_lo = readl(ctx->dev_csr + IOBBATRANSERRREQINFOL); 1562 err_addr_hi = readl(ctx->dev_csr + IOBBATRANSERRREQINFOH); 1563 dev_err(edac_dev->dev, "IOB BA %s access at 0x%02X.%08X (0x%08X)\n", 1564 REQTYPE_F2_RD(err_addr_hi) ? "read" : "write", 1565 ERRADDRH_F2_RD(err_addr_hi), err_addr_lo, err_addr_hi); 1566 if (reg & WRERR_RESP_MASK) 1567 dev_err(edac_dev->dev, "IOB BA requestor ID 0x%08X\n", 1568 readl(ctx->dev_csr + IOBBATRANSERRCSWREQID)); 1569 writel(reg, ctx->dev_csr + IOBBATRANSERRINTSTS); 1570} 1571 1572static void xgene_edac_pa_report(struct edac_device_ctl_info *edac_dev) 1573{ 1574 struct xgene_edac_dev_ctx *ctx = edac_dev->pvt_info; 1575 u32 err_addr_lo; 1576 u32 err_addr_hi; 1577 u32 reg; 1578 1579 /* IOB Processing agent transaction error interrupt */ 1580 reg = readl(ctx->dev_csr + IOBPATRANSERRINTSTS); 1581 if (!reg) 1582 goto chk_iob_axi0; 1583 dev_err(edac_dev->dev, "IOB processing agent (PA) transaction error\n"); 1584 if (reg & IOBPA_RDATA_CORRUPT_MASK) 1585 dev_err(edac_dev->dev, "IOB PA read data RAM error\n"); 1586 if (reg & IOBPA_M_RDATA_CORRUPT_MASK) 1587 dev_err(edac_dev->dev, 1588 "Multiple IOB PA read data RAM error\n"); 1589 if (reg & IOBPA_WDATA_CORRUPT_MASK) 1590 dev_err(edac_dev->dev, "IOB PA write data RAM error\n"); 1591 if (reg & IOBPA_M_WDATA_CORRUPT_MASK) 1592 dev_err(edac_dev->dev, 1593 "Multiple IOB PA write data RAM error\n"); 1594 if (reg & IOBPA_TRANS_CORRUPT_MASK) 1595 dev_err(edac_dev->dev, "IOB PA transaction error\n"); 1596 if (reg & IOBPA_M_TRANS_CORRUPT_MASK) 1597 dev_err(edac_dev->dev, "Multiple IOB PA transaction error\n"); 1598 if (reg & IOBPA_REQIDRAM_CORRUPT_MASK) 1599 dev_err(edac_dev->dev, "IOB PA transaction ID RAM error\n"); 1600 if (reg & IOBPA_M_REQIDRAM_CORRUPT_MASK) 1601 dev_err(edac_dev->dev, 1602 "Multiple IOB PA transaction ID RAM error\n"); 1603 writel(reg, ctx->dev_csr + IOBPATRANSERRINTSTS); 1604 1605chk_iob_axi0: 1606 /* IOB AXI0 Error */ 1607 reg = readl(ctx->dev_csr + IOBAXIS0TRANSERRINTSTS); 1608 if (!reg) 1609 goto chk_iob_axi1; 1610 err_addr_lo = readl(ctx->dev_csr + IOBAXIS0TRANSERRREQINFOL); 1611 err_addr_hi = readl(ctx->dev_csr + IOBAXIS0TRANSERRREQINFOH); 1612 dev_err(edac_dev->dev, 1613 "%sAXI slave 0 illegal %s access @ 0x%02X.%08X (0x%08X)\n", 1614 reg & IOBAXIS0_M_ILLEGAL_ACCESS_MASK ? "Multiple " : "", 1615 REQTYPE_RD(err_addr_hi) ? "read" : "write", 1616 ERRADDRH_RD(err_addr_hi), err_addr_lo, err_addr_hi); 1617 writel(reg, ctx->dev_csr + IOBAXIS0TRANSERRINTSTS); 1618 1619chk_iob_axi1: 1620 /* IOB AXI1 Error */ 1621 reg = readl(ctx->dev_csr + IOBAXIS1TRANSERRINTSTS); 1622 if (!reg) 1623 return; 1624 err_addr_lo = readl(ctx->dev_csr + IOBAXIS1TRANSERRREQINFOL); 1625 err_addr_hi = readl(ctx->dev_csr + IOBAXIS1TRANSERRREQINFOH); 1626 dev_err(edac_dev->dev, 1627 "%sAXI slave 1 illegal %s access @ 0x%02X.%08X (0x%08X)\n", 1628 reg & IOBAXIS0_M_ILLEGAL_ACCESS_MASK ? "Multiple " : "", 1629 REQTYPE_RD(err_addr_hi) ? "read" : "write", 1630 ERRADDRH_RD(err_addr_hi), err_addr_lo, err_addr_hi); 1631 writel(reg, ctx->dev_csr + IOBAXIS1TRANSERRINTSTS); 1632} 1633 1634static void xgene_edac_soc_check(struct edac_device_ctl_info *edac_dev) 1635{ 1636 struct xgene_edac_dev_ctx *ctx = edac_dev->pvt_info; 1637 const char * const *soc_mem_err = NULL; 1638 u32 pcp_hp_stat; 1639 u32 pcp_lp_stat; 1640 u32 reg; 1641 int i; 1642 1643 xgene_edac_pcp_rd(ctx->edac, PCPHPERRINTSTS, &pcp_hp_stat); 1644 xgene_edac_pcp_rd(ctx->edac, PCPLPERRINTSTS, &pcp_lp_stat); 1645 xgene_edac_pcp_rd(ctx->edac, MEMERRINTSTS, ®); 1646 if (!((pcp_hp_stat & (IOB_PA_ERR_MASK | IOB_BA_ERR_MASK | 1647 IOB_XGIC_ERR_MASK | IOB_RB_ERR_MASK)) || 1648 (pcp_lp_stat & CSW_SWITCH_TRACE_ERR_MASK) || reg)) 1649 return; 1650 1651 if (pcp_hp_stat & IOB_XGIC_ERR_MASK) 1652 xgene_edac_iob_gic_report(edac_dev); 1653 1654 if (pcp_hp_stat & (IOB_RB_ERR_MASK | IOB_BA_ERR_MASK)) 1655 xgene_edac_rb_report(edac_dev); 1656 1657 if (pcp_hp_stat & IOB_PA_ERR_MASK) 1658 xgene_edac_pa_report(edac_dev); 1659 1660 if (pcp_lp_stat & CSW_SWITCH_TRACE_ERR_MASK) { 1661 dev_info(edac_dev->dev, 1662 "CSW switch trace correctable memory parity error\n"); 1663 edac_device_handle_ce(edac_dev, 0, 0, edac_dev->ctl_name); 1664 } 1665 1666 if (!reg) 1667 return; 1668 if (ctx->version == 1) 1669 soc_mem_err = soc_mem_err_v1; 1670 if (!soc_mem_err) { 1671 dev_err(edac_dev->dev, "SoC memory parity error 0x%08X\n", 1672 reg); 1673 edac_device_handle_ue(edac_dev, 0, 0, edac_dev->ctl_name); 1674 return; 1675 } 1676 for (i = 0; i < 31; i++) { 1677 if (reg & (1 << i)) { 1678 dev_err(edac_dev->dev, "%s memory parity error\n", 1679 soc_mem_err[i]); 1680 edac_device_handle_ue(edac_dev, 0, 0, 1681 edac_dev->ctl_name); 1682 } 1683 } 1684} 1685 1686static void xgene_edac_soc_hw_init(struct edac_device_ctl_info *edac_dev, 1687 bool enable) 1688{ 1689 struct xgene_edac_dev_ctx *ctx = edac_dev->pvt_info; 1690 1691 /* Enable SoC IP error interrupt */ 1692 if (edac_dev->op_state == OP_RUNNING_INTERRUPT) { 1693 if (enable) { 1694 xgene_edac_pcp_clrbits(ctx->edac, PCPHPERRINTMSK, 1695 IOB_PA_ERR_MASK | 1696 IOB_BA_ERR_MASK | 1697 IOB_XGIC_ERR_MASK | 1698 IOB_RB_ERR_MASK); 1699 xgene_edac_pcp_clrbits(ctx->edac, PCPLPERRINTMSK, 1700 CSW_SWITCH_TRACE_ERR_MASK); 1701 } else { 1702 xgene_edac_pcp_setbits(ctx->edac, PCPHPERRINTMSK, 1703 IOB_PA_ERR_MASK | 1704 IOB_BA_ERR_MASK | 1705 IOB_XGIC_ERR_MASK | 1706 IOB_RB_ERR_MASK); 1707 xgene_edac_pcp_setbits(ctx->edac, PCPLPERRINTMSK, 1708 CSW_SWITCH_TRACE_ERR_MASK); 1709 } 1710 1711 writel(enable ? 0x0 : 0xFFFFFFFF, 1712 ctx->dev_csr + IOBAXIS0TRANSERRINTMSK); 1713 writel(enable ? 0x0 : 0xFFFFFFFF, 1714 ctx->dev_csr + IOBAXIS1TRANSERRINTMSK); 1715 writel(enable ? 0x0 : 0xFFFFFFFF, 1716 ctx->dev_csr + XGICTRANSERRINTMSK); 1717 1718 xgene_edac_pcp_setbits(ctx->edac, MEMERRINTMSK, 1719 enable ? 0x0 : 0xFFFFFFFF); 1720 } 1721} 1722 1723static int xgene_edac_soc_add(struct xgene_edac *edac, struct device_node *np, 1724 int version) 1725{ 1726 struct edac_device_ctl_info *edac_dev; 1727 struct xgene_edac_dev_ctx *ctx; 1728 void __iomem *dev_csr; 1729 struct resource res; 1730 int edac_idx; 1731 int rc; 1732 1733 if (!devres_open_group(edac->dev, xgene_edac_soc_add, GFP_KERNEL)) 1734 return -ENOMEM; 1735 1736 rc = of_address_to_resource(np, 0, &res); 1737 if (rc < 0) { 1738 dev_err(edac->dev, "no SoC resource address\n"); 1739 goto err_release_group; 1740 } 1741 dev_csr = devm_ioremap_resource(edac->dev, &res); 1742 if (IS_ERR(dev_csr)) { 1743 dev_err(edac->dev, 1744 "devm_ioremap_resource failed for soc resource address\n"); 1745 rc = PTR_ERR(dev_csr); 1746 goto err_release_group; 1747 } 1748 1749 edac_idx = edac_device_alloc_index(); 1750 edac_dev = edac_device_alloc_ctl_info(sizeof(*ctx), 1751 "SOC", 1, "SOC", 1, 2, NULL, 0, 1752 edac_idx); 1753 if (!edac_dev) { 1754 rc = -ENOMEM; 1755 goto err_release_group; 1756 } 1757 1758 ctx = edac_dev->pvt_info; 1759 ctx->dev_csr = dev_csr; 1760 ctx->name = "xgene_soc_err"; 1761 ctx->edac_idx = edac_idx; 1762 ctx->edac = edac; 1763 ctx->edac_dev = edac_dev; 1764 ctx->ddev = *edac->dev; 1765 ctx->version = version; 1766 edac_dev->dev = &ctx->ddev; 1767 edac_dev->ctl_name = ctx->name; 1768 edac_dev->dev_name = ctx->name; 1769 edac_dev->mod_name = EDAC_MOD_STR; 1770 1771 if (edac_op_state == EDAC_OPSTATE_POLL) 1772 edac_dev->edac_check = xgene_edac_soc_check; 1773 1774 rc = edac_device_add_device(edac_dev); 1775 if (rc > 0) { 1776 dev_err(edac->dev, "failed edac_device_add_device()\n"); 1777 rc = -ENOMEM; 1778 goto err_ctl_free; 1779 } 1780 1781 if (edac_op_state == EDAC_OPSTATE_INT) 1782 edac_dev->op_state = OP_RUNNING_INTERRUPT; 1783 1784 list_add(&ctx->next, &edac->socs); 1785 1786 xgene_edac_soc_hw_init(edac_dev, 1); 1787 1788 devres_remove_group(edac->dev, xgene_edac_soc_add); 1789 1790 dev_info(edac->dev, "X-Gene EDAC SoC registered\n"); 1791 1792 return 0; 1793 1794err_ctl_free: 1795 edac_device_free_ctl_info(edac_dev); 1796err_release_group: 1797 devres_release_group(edac->dev, xgene_edac_soc_add); 1798 return rc; 1799} 1800 1801static int xgene_edac_soc_remove(struct xgene_edac_dev_ctx *soc) 1802{ 1803 struct edac_device_ctl_info *edac_dev = soc->edac_dev; 1804 1805 xgene_edac_soc_hw_init(edac_dev, 0); 1806 edac_device_del_device(soc->edac->dev); 1807 edac_device_free_ctl_info(edac_dev); 1808 return 0; 1809} 1810 1811static irqreturn_t xgene_edac_isr(int irq, void *dev_id) 1812{ 1813 struct xgene_edac *ctx = dev_id; 1814 struct xgene_edac_pmd_ctx *pmd; 1815 struct xgene_edac_dev_ctx *node; 1816 unsigned int pcp_hp_stat; 1817 unsigned int pcp_lp_stat; 1818 1819 xgene_edac_pcp_rd(ctx, PCPHPERRINTSTS, &pcp_hp_stat); 1820 xgene_edac_pcp_rd(ctx, PCPLPERRINTSTS, &pcp_lp_stat); 1821 if ((MCU_UNCORR_ERR_MASK & pcp_hp_stat) || 1822 (MCU_CTL_ERR_MASK & pcp_hp_stat) || 1823 (MCU_CORR_ERR_MASK & pcp_lp_stat)) { 1824 struct xgene_edac_mc_ctx *mcu; 1825 1826 list_for_each_entry(mcu, &ctx->mcus, next) 1827 xgene_edac_mc_check(mcu->mci); 1828 } 1829 1830 list_for_each_entry(pmd, &ctx->pmds, next) { 1831 if ((PMD0_MERR_MASK << pmd->pmd) & pcp_hp_stat) 1832 xgene_edac_pmd_check(pmd->edac_dev); 1833 } 1834 1835 list_for_each_entry(node, &ctx->l3s, next) 1836 xgene_edac_l3_check(node->edac_dev); 1837 1838 list_for_each_entry(node, &ctx->socs, next) 1839 xgene_edac_soc_check(node->edac_dev); 1840 1841 return IRQ_HANDLED; 1842} 1843 1844static int xgene_edac_probe(struct platform_device *pdev) 1845{ 1846 struct xgene_edac *edac; 1847 struct device_node *child; 1848 struct resource *res; 1849 int rc; 1850 1851 edac = devm_kzalloc(&pdev->dev, sizeof(*edac), GFP_KERNEL); 1852 if (!edac) 1853 return -ENOMEM; 1854 1855 edac->dev = &pdev->dev; 1856 platform_set_drvdata(pdev, edac); 1857 INIT_LIST_HEAD(&edac->mcus); 1858 INIT_LIST_HEAD(&edac->pmds); 1859 INIT_LIST_HEAD(&edac->l3s); 1860 INIT_LIST_HEAD(&edac->socs); 1861 spin_lock_init(&edac->lock); 1862 mutex_init(&edac->mc_lock); 1863 1864 edac->csw_map = syscon_regmap_lookup_by_phandle(pdev->dev.of_node, 1865 "regmap-csw"); 1866 if (IS_ERR(edac->csw_map)) { 1867 dev_err(edac->dev, "unable to get syscon regmap csw\n"); 1868 rc = PTR_ERR(edac->csw_map); 1869 goto out_err; 1870 } 1871 1872 edac->mcba_map = syscon_regmap_lookup_by_phandle(pdev->dev.of_node, 1873 "regmap-mcba"); 1874 if (IS_ERR(edac->mcba_map)) { 1875 dev_err(edac->dev, "unable to get syscon regmap mcba\n"); 1876 rc = PTR_ERR(edac->mcba_map); 1877 goto out_err; 1878 } 1879 1880 edac->mcbb_map = syscon_regmap_lookup_by_phandle(pdev->dev.of_node, 1881 "regmap-mcbb"); 1882 if (IS_ERR(edac->mcbb_map)) { 1883 dev_err(edac->dev, "unable to get syscon regmap mcbb\n"); 1884 rc = PTR_ERR(edac->mcbb_map); 1885 goto out_err; 1886 } 1887 edac->efuse_map = syscon_regmap_lookup_by_phandle(pdev->dev.of_node, 1888 "regmap-efuse"); 1889 if (IS_ERR(edac->efuse_map)) { 1890 dev_err(edac->dev, "unable to get syscon regmap efuse\n"); 1891 rc = PTR_ERR(edac->efuse_map); 1892 goto out_err; 1893 } 1894 1895 /* 1896 * NOTE: The register bus resource is optional for compatibility 1897 * reason. 1898 */ 1899 edac->rb_map = syscon_regmap_lookup_by_phandle(pdev->dev.of_node, 1900 "regmap-rb"); 1901 if (IS_ERR(edac->rb_map)) { 1902 dev_warn(edac->dev, "missing syscon regmap rb\n"); 1903 edac->rb_map = NULL; 1904 } 1905 1906 res = platform_get_resource(pdev, IORESOURCE_MEM, 0); 1907 edac->pcp_csr = devm_ioremap_resource(&pdev->dev, res); 1908 if (IS_ERR(edac->pcp_csr)) { 1909 dev_err(&pdev->dev, "no PCP resource address\n"); 1910 rc = PTR_ERR(edac->pcp_csr); 1911 goto out_err; 1912 } 1913 1914 if (edac_op_state == EDAC_OPSTATE_INT) { 1915 int irq; 1916 int i; 1917 1918 for (i = 0; i < 3; i++) { 1919 irq = platform_get_irq_optional(pdev, i); 1920 if (irq < 0) { 1921 dev_err(&pdev->dev, "No IRQ resource\n"); 1922 rc = irq; 1923 goto out_err; 1924 } 1925 rc = devm_request_irq(&pdev->dev, irq, 1926 xgene_edac_isr, IRQF_SHARED, 1927 dev_name(&pdev->dev), edac); 1928 if (rc) { 1929 dev_err(&pdev->dev, 1930 "Could not request IRQ %d\n", irq); 1931 goto out_err; 1932 } 1933 } 1934 } 1935 1936 edac->dfs = edac_debugfs_create_dir(pdev->dev.kobj.name); 1937 1938 for_each_child_of_node(pdev->dev.of_node, child) { 1939 if (!of_device_is_available(child)) 1940 continue; 1941 if (of_device_is_compatible(child, "apm,xgene-edac-mc")) 1942 xgene_edac_mc_add(edac, child); 1943 if (of_device_is_compatible(child, "apm,xgene-edac-pmd")) 1944 xgene_edac_pmd_add(edac, child, 1); 1945 if (of_device_is_compatible(child, "apm,xgene-edac-pmd-v2")) 1946 xgene_edac_pmd_add(edac, child, 2); 1947 if (of_device_is_compatible(child, "apm,xgene-edac-l3")) 1948 xgene_edac_l3_add(edac, child, 1); 1949 if (of_device_is_compatible(child, "apm,xgene-edac-l3-v2")) 1950 xgene_edac_l3_add(edac, child, 2); 1951 if (of_device_is_compatible(child, "apm,xgene-edac-soc")) 1952 xgene_edac_soc_add(edac, child, 0); 1953 if (of_device_is_compatible(child, "apm,xgene-edac-soc-v1")) 1954 xgene_edac_soc_add(edac, child, 1); 1955 } 1956 1957 return 0; 1958 1959out_err: 1960 return rc; 1961} 1962 1963static void xgene_edac_remove(struct platform_device *pdev) 1964{ 1965 struct xgene_edac *edac = dev_get_drvdata(&pdev->dev); 1966 struct xgene_edac_mc_ctx *mcu; 1967 struct xgene_edac_mc_ctx *temp_mcu; 1968 struct xgene_edac_pmd_ctx *pmd; 1969 struct xgene_edac_pmd_ctx *temp_pmd; 1970 struct xgene_edac_dev_ctx *node; 1971 struct xgene_edac_dev_ctx *temp_node; 1972 1973 list_for_each_entry_safe(mcu, temp_mcu, &edac->mcus, next) 1974 xgene_edac_mc_remove(mcu); 1975 1976 list_for_each_entry_safe(pmd, temp_pmd, &edac->pmds, next) 1977 xgene_edac_pmd_remove(pmd); 1978 1979 list_for_each_entry_safe(node, temp_node, &edac->l3s, next) 1980 xgene_edac_l3_remove(node); 1981 1982 list_for_each_entry_safe(node, temp_node, &edac->socs, next) 1983 xgene_edac_soc_remove(node); 1984} 1985 1986static const struct of_device_id xgene_edac_of_match[] = { 1987 { .compatible = "apm,xgene-edac" }, 1988 {}, 1989}; 1990MODULE_DEVICE_TABLE(of, xgene_edac_of_match); 1991 1992static struct platform_driver xgene_edac_driver = { 1993 .probe = xgene_edac_probe, 1994 .remove_new = xgene_edac_remove, 1995 .driver = { 1996 .name = "xgene-edac", 1997 .of_match_table = xgene_edac_of_match, 1998 }, 1999}; 2000 2001static int __init xgene_edac_init(void) 2002{ 2003 int rc; 2004 2005 if (ghes_get_devices()) 2006 return -EBUSY; 2007 2008 /* Make sure error reporting method is sane */ 2009 switch (edac_op_state) { 2010 case EDAC_OPSTATE_POLL: 2011 case EDAC_OPSTATE_INT: 2012 break; 2013 default: 2014 edac_op_state = EDAC_OPSTATE_INT; 2015 break; 2016 } 2017 2018 rc = platform_driver_register(&xgene_edac_driver); 2019 if (rc) { 2020 edac_printk(KERN_ERR, EDAC_MOD_STR, 2021 "EDAC fails to register\n"); 2022 goto reg_failed; 2023 } 2024 2025 return 0; 2026 2027reg_failed: 2028 return rc; 2029} 2030module_init(xgene_edac_init); 2031 2032static void __exit xgene_edac_exit(void) 2033{ 2034 platform_driver_unregister(&xgene_edac_driver); 2035} 2036module_exit(xgene_edac_exit); 2037 2038MODULE_LICENSE("GPL"); 2039MODULE_AUTHOR("Feng Kan <fkan@apm.com>"); 2040MODULE_DESCRIPTION("APM X-Gene EDAC driver"); 2041module_param(edac_op_state, int, 0444); 2042MODULE_PARM_DESC(edac_op_state, 2043 "EDAC error reporting state: 0=Poll, 2=Interrupt"); 2044