drivers/edac/cell_edac.c

   1 /*
   2  * Cell MIC driver for ECC counting
   3  *
   4  * Copyright 2007 Benjamin Herrenschmidt, IBM Corp.
   5  *                <benh@kernel.crashing.org>
   6  *
   7  * This file may be distributed under the terms of the
   8  * GNU General Public License.
   9  */
  10 #undef DEBUG
  11
  12 #include <linux/module.h>
  13 #include <linux/init.h>
  14 #include <linux/platform_device.h>
  15 #include <linux/stop_machine.h>
  16 #include <linux/io.h>
  17 #include <asm/machdep.h>
  18 #include <asm/cell-regs.h>
  19
  20 #include "edac_core.h"
  21
  22 struct cell_edac_priv
  23 {
  24         struct cbe_mic_tm_regs __iomem  *regs;
  25         int                             node;
  26         int                             chanmask;
  27 #ifdef DEBUG
  28         u64                             prev_fir;
  29 #endif
  30 };
  31
  32 static void cell_edac_count_ce(struct mem_ctl_info *mci, int chan, u64 ar)
  33 {
  34         struct cell_edac_priv           *priv = mci->pvt_info;
  35         struct csrow_info               *csrow = &mci->csrows[0];
  36         unsigned long                   address, pfn, offset, syndrome;
  37
  38         dev_dbg(mci->dev, "ECC CE err on node %d, channel %d, ar = 0x%016lx\n",
  39                 priv->node, chan, ar);
  40
  41         /* Address decoding is likely a bit bogus, to dbl check */
  42         address = (ar & 0xffffffffe0000000ul) >> 29;
  43         if (priv->chanmask == 0x3)
  44                 address = (address << 1) | chan;
  45         pfn = address >> PAGE_SHIFT;
  46         offset = address & ~PAGE_MASK;
  47         syndrome = (ar & 0x000000001fe00000ul) >> 21;
  48
  49         /* TODO: Decoding of the error addresss */
  50         edac_mc_handle_ce(mci, csrow->first_page + pfn, offset,
  51                           syndrome, 0, chan, "");
  52 }
  53
  54 static void cell_edac_count_ue(struct mem_ctl_info *mci, int chan, u64 ar)
  55 {
  56         struct cell_edac_priv           *priv = mci->pvt_info;
  57         struct csrow_info               *csrow = &mci->csrows[0];
  58         unsigned long                   address, pfn, offset;
  59
  60         dev_dbg(mci->dev, "ECC UE err on node %d, channel %d, ar = 0x%016lx\n",
  61                 priv->node, chan, ar);
  62
  63         /* Address decoding is likely a bit bogus, to dbl check */
  64         address = (ar & 0xffffffffe0000000ul) >> 29;
  65         if (priv->chanmask == 0x3)
  66                 address = (address << 1) | chan;
  67         pfn = address >> PAGE_SHIFT;
  68         offset = address & ~PAGE_MASK;
  69
  70         /* TODO: Decoding of the error addresss */
  71         edac_mc_handle_ue(mci, csrow->first_page + pfn, offset, 0, "");
  72 }
  73
  74 static void cell_edac_check(struct mem_ctl_info *mci)
  75 {
  76         struct cell_edac_priv           *priv = mci->pvt_info;
  77         u64                             fir, addreg, clear = 0;
  78
  79         fir = in_be64(&priv->regs->mic_fir);
  80 #ifdef DEBUG
  81         if (fir != priv->prev_fir) {
  82                 dev_dbg(mci->dev, "fir change : 0x%016lx\n", fir);
  83                 priv->prev_fir = fir;
  84         }
  85 #endif
  86         if ((priv->chanmask & 0x1) && (fir & CBE_MIC_FIR_ECC_SINGLE_0_ERR)) {
  87                 addreg = in_be64(&priv->regs->mic_df_ecc_address_0);
  88                 clear |= CBE_MIC_FIR_ECC_SINGLE_0_RESET;
  89                 cell_edac_count_ce(mci, 0, addreg);
  90         }
  91         if ((priv->chanmask & 0x2) && (fir & CBE_MIC_FIR_ECC_SINGLE_1_ERR)) {
  92                 addreg = in_be64(&priv->regs->mic_df_ecc_address_1);
  93                 clear |= CBE_MIC_FIR_ECC_SINGLE_1_RESET;
  94                 cell_edac_count_ce(mci, 1, addreg);
  95         }
  96         if ((priv->chanmask & 0x1) && (fir & CBE_MIC_FIR_ECC_MULTI_0_ERR)) {
  97                 addreg = in_be64(&priv->regs->mic_df_ecc_address_0);
  98                 clear |= CBE_MIC_FIR_ECC_MULTI_0_RESET;
  99                 cell_edac_count_ue(mci, 0, addreg);
 100         }
 101         if ((priv->chanmask & 0x2) && (fir & CBE_MIC_FIR_ECC_MULTI_1_ERR)) {
 102                 addreg = in_be64(&priv->regs->mic_df_ecc_address_1);
 103                 clear |= CBE_MIC_FIR_ECC_MULTI_1_RESET;
 104                 cell_edac_count_ue(mci, 1, addreg);
 105         }
 106
 107         /* The procedure for clearing FIR bits is a bit ... weird */
 108         if (clear) {
 109                 fir &= ~(CBE_MIC_FIR_ECC_ERR_MASK | CBE_MIC_FIR_ECC_SET_MASK);
 110                 fir |= CBE_MIC_FIR_ECC_RESET_MASK;
 111                 fir &= ~clear;
 112                 out_be64(&priv->regs->mic_fir, fir);
 113                 (void)in_be64(&priv->regs->mic_fir);
 114
 115                 mb();   /* sync up */
 116 #ifdef DEBUG
 117                 fir = in_be64(&priv->regs->mic_fir);
 118                 dev_dbg(mci->dev, "fir clear  : 0x%016lx\n", fir);
 119 #endif
 120         }
 121 }
 122
 123 static void __devinit cell_edac_init_csrows(struct mem_ctl_info *mci)
 124 {
 125         struct csrow_info               *csrow = &mci->csrows[0];
 126         struct cell_edac_priv           *priv = mci->pvt_info;
 127         struct device_node              *np;
 128
 129         for (np = NULL;
 130              (np = of_find_node_by_name(np, "memory")) != NULL;) {
 131                 struct resource r;
 132
 133                 /* We "know" that the Cell firmware only creates one entry
 134                  * in the "memory" nodes. If that changes, this code will
 135                  * need to be adapted.
 136                  */
 137                 if (of_address_to_resource(np, 0, &r))
 138                         continue;
 139                 if (of_node_to_nid(np) != priv->node)
 140                         continue;
 141                 csrow->first_page = r.start >> PAGE_SHIFT;
 142                 csrow->nr_pages = (r.end - r.start + 1) >> PAGE_SHIFT;
 143                 csrow->last_page = csrow->first_page + csrow->nr_pages - 1;
 144                 csrow->mtype = MEM_XDR;
 145                 csrow->edac_mode = EDAC_FLAG_EC | EDAC_FLAG_SECDED;
 146                 dev_dbg(mci->dev,
 147                         "Initialized on node %d, chanmask=0x%x,"
 148                         " first_page=0x%lx, nr_pages=0x%x\n",
 149                         priv->node, priv->chanmask,
 150                         csrow->first_page, csrow->nr_pages);
 151                 break;
 152         }
 153 }
 154
 155 static int __devinit cell_edac_probe(struct platform_device *pdev)
 156 {
 157         struct cbe_mic_tm_regs __iomem  *regs;
 158         struct mem_ctl_info             *mci;
 159         struct cell_edac_priv           *priv;
 160         u64                             reg;
 161         int                             rc, chanmask;
 162
 163         regs = cbe_get_cpu_mic_tm_regs(cbe_node_to_cpu(pdev->id));
 164         if (regs == NULL)
 165                 return -ENODEV;
 166
 167         /* Get channel population */
 168         reg = in_be64(&regs->mic_mnt_cfg);
 169         dev_dbg(&pdev->dev, "MIC_MNT_CFG = 0x%016lx\n", reg);
 170         chanmask = 0;
 171         if (reg & CBE_MIC_MNT_CFG_CHAN_0_POP)
 172                 chanmask |= 0x1;
 173         if (reg & CBE_MIC_MNT_CFG_CHAN_1_POP)
 174                 chanmask |= 0x2;
 175         if (chanmask == 0) {
 176                 dev_warn(&pdev->dev,
 177                          "Yuck ! No channel populated ? Aborting !\n");
 178                 return -ENODEV;
 179         }
 180         dev_dbg(&pdev->dev, "Initial FIR = 0x%016lx\n",
 181                 in_be64(&regs->mic_fir));
 182
 183         /* Allocate & init EDAC MC data structure */
 184         mci = edac_mc_alloc(sizeof(struct cell_edac_priv), 1,
 185                             chanmask == 3 ? 2 : 1, pdev->id);
 186         if (mci == NULL)
 187                 return -ENOMEM;
 188         priv = mci->pvt_info;
 189         priv->regs = regs;
 190         priv->node = pdev->id;
 191         priv->chanmask = chanmask;
 192         mci->dev = &pdev->dev;
 193         mci->mtype_cap = MEM_FLAG_XDR;
 194         mci->edac_ctl_cap = EDAC_FLAG_NONE | EDAC_FLAG_EC | EDAC_FLAG_SECDED;
 195         mci->edac_cap = EDAC_FLAG_EC | EDAC_FLAG_SECDED;
 196         mci->mod_name = "cell_edac";
 197         mci->ctl_name = "MIC";
 198         mci->dev_name = pdev->dev.bus_id;
 199         mci->edac_check = cell_edac_check;
 200         cell_edac_init_csrows(mci);
 201
 202         /* Register with EDAC core */
 203         rc = edac_mc_add_mc(mci);
 204         if (rc) {
 205                 dev_err(&pdev->dev, "failed to register with EDAC core\n");
 206                 edac_mc_free(mci);
 207                 return rc;
 208         }
 209
 210         return 0;
 211 }
 212
 213 static int __devexit cell_edac_remove(struct platform_device *pdev)
 214 {
 215         struct mem_ctl_info *mci = edac_mc_del_mc(&pdev->dev);
 216         if (mci)
 217                 edac_mc_free(mci);
 218         return 0;
 219 }
 220
 221 static struct platform_driver cell_edac_driver = {
 222         .driver         = {
 223                 .name   = "cbe-mic",
 224                 .owner  = THIS_MODULE,
 225         },
 226         .probe          = cell_edac_probe,
 227         .remove         = cell_edac_remove,
 228 };
 229
 230 static int __init cell_edac_init(void)
 231 {
 232         /* Sanity check registers data structure */
 233         BUILD_BUG_ON(offsetof(struct cbe_mic_tm_regs,
 234                               mic_df_ecc_address_0) != 0xf8);
 235         BUILD_BUG_ON(offsetof(struct cbe_mic_tm_regs,
 236                               mic_df_ecc_address_1) != 0x1b8);
 237         BUILD_BUG_ON(offsetof(struct cbe_mic_tm_regs,
 238                               mic_df_config) != 0x218);
 239         BUILD_BUG_ON(offsetof(struct cbe_mic_tm_regs,
 240                               mic_fir) != 0x230);
 241         BUILD_BUG_ON(offsetof(struct cbe_mic_tm_regs,
 242                               mic_mnt_cfg) != 0x210);
 243         BUILD_BUG_ON(offsetof(struct cbe_mic_tm_regs,
 244                               mic_exc) != 0x208);
 245
 246         return platform_driver_register(&cell_edac_driver);
 247 }
 248
 249 static void __exit cell_edac_exit(void)
 250 {
 251         platform_driver_unregister(&cell_edac_driver);
 252 }
 253
 254 module_init(cell_edac_init);
 255 module_exit(cell_edac_exit);
 256
 257 MODULE_LICENSE("GPL");
 258 MODULE_AUTHOR("Benjamin Herrenschmidt <benh@kernel.crashing.org>");
 259 MODULE_DESCRIPTION("ECC counting for Cell MIC");