drivers/edac/al_mc_edac.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright 2019 Amazon.com, Inc. or its affiliates. All Rights Reserved.
   4  */
   5 #include <linux/bitfield.h>
   6 #include <linux/bitops.h>
   7 #include <linux/edac.h>
   8 #include <linux/of_irq.h>
   9 #include <linux/platform_device.h>
  10 #include <linux/spinlock.h>
  11 #include "edac_module.h"
  12
  13 /* Registers Offset */
  14 #define AL_MC_ECC_CFG           0x70
  15 #define AL_MC_ECC_CLEAR         0x7c
  16 #define AL_MC_ECC_ERR_COUNT     0x80
  17 #define AL_MC_ECC_CE_ADDR0      0x84
  18 #define AL_MC_ECC_CE_ADDR1      0x88
  19 #define AL_MC_ECC_UE_ADDR0      0xa4
  20 #define AL_MC_ECC_UE_ADDR1      0xa8
  21 #define AL_MC_ECC_CE_SYND0      0x8c
  22 #define AL_MC_ECC_CE_SYND1      0x90
  23 #define AL_MC_ECC_CE_SYND2      0x94
  24 #define AL_MC_ECC_UE_SYND0      0xac
  25 #define AL_MC_ECC_UE_SYND1      0xb0
  26 #define AL_MC_ECC_UE_SYND2      0xb4
  27
  28 /* Registers Fields */
  29 #define AL_MC_ECC_CFG_SCRUB_DISABLED    BIT(4)
  30
  31 #define AL_MC_ECC_CLEAR_UE_COUNT        BIT(3)
  32 #define AL_MC_ECC_CLEAR_CE_COUNT        BIT(2)
  33 #define AL_MC_ECC_CLEAR_UE_ERR          BIT(1)
  34 #define AL_MC_ECC_CLEAR_CE_ERR          BIT(0)
  35
  36 #define AL_MC_ECC_ERR_COUNT_UE          GENMASK(31, 16)
  37 #define AL_MC_ECC_ERR_COUNT_CE          GENMASK(15, 0)
  38
  39 #define AL_MC_ECC_CE_ADDR0_RANK         GENMASK(25, 24)
  40 #define AL_MC_ECC_CE_ADDR0_ROW          GENMASK(17, 0)
  41
  42 #define AL_MC_ECC_CE_ADDR1_BG           GENMASK(25, 24)
  43 #define AL_MC_ECC_CE_ADDR1_BANK         GENMASK(18, 16)
  44 #define AL_MC_ECC_CE_ADDR1_COLUMN       GENMASK(11, 0)
  45
  46 #define AL_MC_ECC_UE_ADDR0_RANK         GENMASK(25, 24)
  47 #define AL_MC_ECC_UE_ADDR0_ROW          GENMASK(17, 0)
  48
  49 #define AL_MC_ECC_UE_ADDR1_BG           GENMASK(25, 24)
  50 #define AL_MC_ECC_UE_ADDR1_BANK         GENMASK(18, 16)
  51 #define AL_MC_ECC_UE_ADDR1_COLUMN       GENMASK(11, 0)
  52
  53 #define DRV_NAME "al_mc_edac"
  54 #define AL_MC_EDAC_MSG_MAX 256
  55
  56 struct al_mc_edac {
  57         void __iomem *mmio_base;
  58         spinlock_t lock;
  59         int irq_ce;
  60         int irq_ue;
  61 };
  62
  63 static void prepare_msg(char *message, size_t buffer_size,
  64                         enum hw_event_mc_err_type type,
  65                         u8 rank, u32 row, u8 bg, u8 bank, u16 column,
  66                         u32 syn0, u32 syn1, u32 syn2)
  67 {
  68         snprintf(message, buffer_size,
  69                  "%s rank=0x%x row=0x%x bg=0x%x bank=0x%x col=0x%x syn0: 0x%x syn1: 0x%x syn2: 0x%x",
  70                  type == HW_EVENT_ERR_UNCORRECTED ? "UE" : "CE",
  71                  rank, row, bg, bank, column, syn0, syn1, syn2);
  72 }
  73
  74 static int handle_ce(struct mem_ctl_info *mci)
  75 {
  76         u32 eccerrcnt, ecccaddr0, ecccaddr1, ecccsyn0, ecccsyn1, ecccsyn2, row;
  77         struct al_mc_edac *al_mc = mci->pvt_info;
  78         char msg[AL_MC_EDAC_MSG_MAX];
  79         u16 ce_count, column;
  80         unsigned long flags;
  81         u8 rank, bg, bank;
  82
  83         eccerrcnt = readl_relaxed(al_mc->mmio_base + AL_MC_ECC_ERR_COUNT);
  84         ce_count = FIELD_GET(AL_MC_ECC_ERR_COUNT_CE, eccerrcnt);
  85         if (!ce_count)
  86                 return 0;
  87
  88         ecccaddr0 = readl_relaxed(al_mc->mmio_base + AL_MC_ECC_CE_ADDR0);
  89         ecccaddr1 = readl_relaxed(al_mc->mmio_base + AL_MC_ECC_CE_ADDR1);
  90         ecccsyn0 = readl_relaxed(al_mc->mmio_base + AL_MC_ECC_CE_SYND0);
  91         ecccsyn1 = readl_relaxed(al_mc->mmio_base + AL_MC_ECC_CE_SYND1);
  92         ecccsyn2 = readl_relaxed(al_mc->mmio_base + AL_MC_ECC_CE_SYND2);
  93
  94         writel_relaxed(AL_MC_ECC_CLEAR_CE_COUNT | AL_MC_ECC_CLEAR_CE_ERR,
  95                        al_mc->mmio_base + AL_MC_ECC_CLEAR);
  96
  97         dev_dbg(mci->pdev, "eccuaddr0=0x%08x eccuaddr1=0x%08x\n",
  98                 ecccaddr0, ecccaddr1);
  99
 100         rank = FIELD_GET(AL_MC_ECC_CE_ADDR0_RANK, ecccaddr0);
 101         row = FIELD_GET(AL_MC_ECC_CE_ADDR0_ROW, ecccaddr0);
 102
 103         bg = FIELD_GET(AL_MC_ECC_CE_ADDR1_BG, ecccaddr1);
 104         bank = FIELD_GET(AL_MC_ECC_CE_ADDR1_BANK, ecccaddr1);
 105         column = FIELD_GET(AL_MC_ECC_CE_ADDR1_COLUMN, ecccaddr1);
 106
 107         prepare_msg(msg, sizeof(msg), HW_EVENT_ERR_CORRECTED,
 108                     rank, row, bg, bank, column,
 109                     ecccsyn0, ecccsyn1, ecccsyn2);
 110
 111         spin_lock_irqsave(&al_mc->lock, flags);
 112         edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci,
 113                              ce_count, 0, 0, 0, 0, 0, -1, mci->ctl_name, msg);
 114         spin_unlock_irqrestore(&al_mc->lock, flags);
 115
 116         return ce_count;
 117 }
 118
 119 static int handle_ue(struct mem_ctl_info *mci)
 120 {
 121         u32 eccerrcnt, eccuaddr0, eccuaddr1, eccusyn0, eccusyn1, eccusyn2, row;
 122         struct al_mc_edac *al_mc = mci->pvt_info;
 123         char msg[AL_MC_EDAC_MSG_MAX];
 124         u16 ue_count, column;
 125         unsigned long flags;
 126         u8 rank, bg, bank;
 127
 128         eccerrcnt = readl_relaxed(al_mc->mmio_base + AL_MC_ECC_ERR_COUNT);
 129         ue_count = FIELD_GET(AL_MC_ECC_ERR_COUNT_UE, eccerrcnt);
 130         if (!ue_count)
 131                 return 0;
 132
 133         eccuaddr0 = readl_relaxed(al_mc->mmio_base + AL_MC_ECC_UE_ADDR0);
 134         eccuaddr1 = readl_relaxed(al_mc->mmio_base + AL_MC_ECC_UE_ADDR1);
 135         eccusyn0 = readl_relaxed(al_mc->mmio_base + AL_MC_ECC_UE_SYND0);
 136         eccusyn1 = readl_relaxed(al_mc->mmio_base + AL_MC_ECC_UE_SYND1);
 137         eccusyn2 = readl_relaxed(al_mc->mmio_base + AL_MC_ECC_UE_SYND2);
 138
 139         writel_relaxed(AL_MC_ECC_CLEAR_UE_COUNT | AL_MC_ECC_CLEAR_UE_ERR,
 140                        al_mc->mmio_base + AL_MC_ECC_CLEAR);
 141
 142         dev_dbg(mci->pdev, "eccuaddr0=0x%08x eccuaddr1=0x%08x\n",
 143                 eccuaddr0, eccuaddr1);
 144
 145         rank = FIELD_GET(AL_MC_ECC_UE_ADDR0_RANK, eccuaddr0);
 146         row = FIELD_GET(AL_MC_ECC_UE_ADDR0_ROW, eccuaddr0);
 147
 148         bg = FIELD_GET(AL_MC_ECC_UE_ADDR1_BG, eccuaddr1);
 149         bank = FIELD_GET(AL_MC_ECC_UE_ADDR1_BANK, eccuaddr1);
 150         column = FIELD_GET(AL_MC_ECC_UE_ADDR1_COLUMN, eccuaddr1);
 151
 152         prepare_msg(msg, sizeof(msg), HW_EVENT_ERR_UNCORRECTED,
 153                     rank, row, bg, bank, column,
 154                     eccusyn0, eccusyn1, eccusyn2);
 155
 156         spin_lock_irqsave(&al_mc->lock, flags);
 157         edac_mc_handle_error(HW_EVENT_ERR_UNCORRECTED, mci,
 158                              ue_count, 0, 0, 0, 0, 0, -1, mci->ctl_name, msg);
 159         spin_unlock_irqrestore(&al_mc->lock, flags);
 160
 161         return ue_count;
 162 }
 163
 164 static void al_mc_edac_check(struct mem_ctl_info *mci)
 165 {
 166         struct al_mc_edac *al_mc = mci->pvt_info;
 167
 168         if (al_mc->irq_ue <= 0)
 169                 handle_ue(mci);
 170
 171         if (al_mc->irq_ce <= 0)
 172                 handle_ce(mci);
 173 }
 174
 175 static irqreturn_t al_mc_edac_irq_handler_ue(int irq, void *info)
 176 {
 177         struct platform_device *pdev = info;
 178         struct mem_ctl_info *mci = platform_get_drvdata(pdev);
 179
 180         if (handle_ue(mci))
 181                 return IRQ_HANDLED;
 182         return IRQ_NONE;
 183 }
 184
 185 static irqreturn_t al_mc_edac_irq_handler_ce(int irq, void *info)
 186 {
 187         struct platform_device *pdev = info;
 188         struct mem_ctl_info *mci = platform_get_drvdata(pdev);
 189
 190         if (handle_ce(mci))
 191                 return IRQ_HANDLED;
 192         return IRQ_NONE;
 193 }
 194
 195 static enum scrub_type get_scrub_mode(void __iomem *mmio_base)
 196 {
 197         u32 ecccfg0;
 198
 199         ecccfg0 = readl(mmio_base + AL_MC_ECC_CFG);
 200
 201         if (FIELD_GET(AL_MC_ECC_CFG_SCRUB_DISABLED, ecccfg0))
 202                 return SCRUB_NONE;
 203         else
 204                 return SCRUB_HW_SRC;
 205 }
 206
 207 static void devm_al_mc_edac_free(void *data)
 208 {
 209         edac_mc_free(data);
 210 }
 211
 212 static void devm_al_mc_edac_del(void *data)
 213 {
 214         edac_mc_del_mc(data);
 215 }
 216
 217 static int al_mc_edac_probe(struct platform_device *pdev)
 218 {
 219         struct edac_mc_layer layers[1];
 220         struct mem_ctl_info *mci;
 221         struct al_mc_edac *al_mc;
 222         void __iomem *mmio_base;
 223         struct dimm_info *dimm;
 224         int ret;
 225
 226         mmio_base = devm_platform_ioremap_resource(pdev, 0);
 227         if (IS_ERR(mmio_base)) {
 228                 dev_err(&pdev->dev, "failed to ioremap memory (%ld)\n",
 229                         PTR_ERR(mmio_base));
 230                 return PTR_ERR(mmio_base);
 231         }
 232
 233         layers[0].type = EDAC_MC_LAYER_CHIP_SELECT;
 234         layers[0].size = 1;
 235         layers[0].is_virt_csrow = false;
 236         mci = edac_mc_alloc(0, ARRAY_SIZE(layers), layers,
 237                             sizeof(struct al_mc_edac));
 238         if (!mci)
 239                 return -ENOMEM;
 240
 241         ret = devm_add_action(&pdev->dev, devm_al_mc_edac_free, mci);
 242         if (ret) {
 243                 edac_mc_free(mci);
 244                 return ret;
 245         }
 246
 247         platform_set_drvdata(pdev, mci);
 248         al_mc = mci->pvt_info;
 249
 250         al_mc->mmio_base = mmio_base;
 251
 252         al_mc->irq_ue = of_irq_get_byname(pdev->dev.of_node, "ue");
 253         if (al_mc->irq_ue <= 0)
 254                 dev_dbg(&pdev->dev,
 255                         "no IRQ defined for UE - falling back to polling\n");
 256
 257         al_mc->irq_ce = of_irq_get_byname(pdev->dev.of_node, "ce");
 258         if (al_mc->irq_ce <= 0)
 259                 dev_dbg(&pdev->dev,
 260                         "no IRQ defined for CE - falling back to polling\n");
 261
 262         /*
 263          * In case both interrupts (ue/ce) are to be found, use interrupt mode.
 264          * In case none of the interrupt are foud, use polling mode.
 265          * In case only one interrupt is found, use interrupt mode for it but
 266          * keep polling mode enable for the other.
 267          */
 268         if (al_mc->irq_ue <= 0 || al_mc->irq_ce <= 0) {
 269                 edac_op_state = EDAC_OPSTATE_POLL;
 270                 mci->edac_check = al_mc_edac_check;
 271         } else {
 272                 edac_op_state = EDAC_OPSTATE_INT;
 273         }
 274
 275         spin_lock_init(&al_mc->lock);
 276
 277         mci->mtype_cap = MEM_FLAG_DDR3 | MEM_FLAG_DDR4;
 278         mci->edac_ctl_cap = EDAC_FLAG_NONE | EDAC_FLAG_SECDED;
 279         mci->edac_cap = EDAC_FLAG_SECDED;
 280         mci->mod_name = DRV_NAME;
 281         mci->ctl_name = "al_mc";
 282         mci->pdev = &pdev->dev;
 283         mci->scrub_mode = get_scrub_mode(mmio_base);
 284
 285         dimm = *mci->dimms;
 286         dimm->grain = 1;
 287
 288         ret = edac_mc_add_mc(mci);
 289         if (ret < 0) {
 290                 dev_err(&pdev->dev,
 291                         "fail to add memory controller device (%d)\n",
 292                         ret);
 293                 return ret;
 294         }
 295
 296         ret = devm_add_action(&pdev->dev, devm_al_mc_edac_del, &pdev->dev);
 297         if (ret) {
 298                 edac_mc_del_mc(&pdev->dev);
 299                 return ret;
 300         }
 301
 302         if (al_mc->irq_ue > 0) {
 303                 ret = devm_request_irq(&pdev->dev,
 304                                        al_mc->irq_ue,
 305                                        al_mc_edac_irq_handler_ue,
 306                                        IRQF_SHARED,
 307                                        pdev->name,
 308                                        pdev);
 309                 if (ret != 0) {
 310                         dev_err(&pdev->dev,
 311                                 "failed to request UE IRQ %d (%d)\n",
 312                                 al_mc->irq_ue, ret);
 313                         return ret;
 314                 }
 315         }
 316
 317         if (al_mc->irq_ce > 0) {
 318                 ret = devm_request_irq(&pdev->dev,
 319                                        al_mc->irq_ce,
 320                                        al_mc_edac_irq_handler_ce,
 321                                        IRQF_SHARED,
 322                                        pdev->name,
 323                                        pdev);
 324                 if (ret != 0) {
 325                         dev_err(&pdev->dev,
 326                                 "failed to request CE IRQ %d (%d)\n",
 327                                 al_mc->irq_ce, ret);
 328                         return ret;
 329                 }
 330         }
 331
 332         return 0;
 333 }
 334
 335 static const struct of_device_id al_mc_edac_of_match[] = {
 336         { .compatible = "amazon,al-mc-edac", },
 337         {},
 338 };
 339
 340 MODULE_DEVICE_TABLE(of, al_mc_edac_of_match);
 341
 342 static struct platform_driver al_mc_edac_driver = {
 343         .probe = al_mc_edac_probe,
 344         .driver = {
 345                 .name = DRV_NAME,
 346                 .of_match_table = al_mc_edac_of_match,
 347         },
 348 };
 349
 350 module_platform_driver(al_mc_edac_driver);
 351
 352 MODULE_LICENSE("GPL v2");
 353 MODULE_AUTHOR("Talel Shenhar");
 354 MODULE_DESCRIPTION("Amazon's Annapurna Lab's Memory Controller EDAC Driver");