arch/powerpc/platforms/cell/ras.c

   1 /*
   2  * Copyright 2006-2008, IBM Corporation.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public License
   6  * as published by the Free Software Foundation; either version
   7  * 2 of the License, or (at your option) any later version.
   8  */
   9
  10 #undef DEBUG
  11
  12 #include <linux/types.h>
  13 #include <linux/kernel.h>
  14 #include <linux/smp.h>
  15 #include <linux/reboot.h>
  16
  17 #include <asm/reg.h>
  18 #include <asm/io.h>
  19 #include <asm/prom.h>
  20 #include <asm/machdep.h>
  21 #include <asm/rtas.h>
  22 #include <asm/cell-regs.h>
  23
  24 #include "ras.h"
  25
  26
  27 static void dump_fir(int cpu)
  28 {
  29         struct cbe_pmd_regs __iomem *pregs = cbe_get_cpu_pmd_regs(cpu);
  30         struct cbe_iic_regs __iomem *iregs = cbe_get_cpu_iic_regs(cpu);
  31
  32         if (pregs == NULL)
  33                 return;
  34
  35         /* Todo: do some nicer parsing of bits and based on them go down
  36          * to other sub-units FIRs and not only IIC
  37          */
  38         printk(KERN_ERR "Global Checkstop FIR    : 0x%016lx\n",
  39                in_be64(&pregs->checkstop_fir));
  40         printk(KERN_ERR "Global Recoverable FIR  : 0x%016lx\n",
  41                in_be64(&pregs->checkstop_fir));
  42         printk(KERN_ERR "Global MachineCheck FIR : 0x%016lx\n",
  43                in_be64(&pregs->spec_att_mchk_fir));
  44
  45         if (iregs == NULL)
  46                 return;
  47         printk(KERN_ERR "IOC FIR                 : 0x%016lx\n",
  48                in_be64(&iregs->ioc_fir));
  49
  50 }
  51
  52 void cbe_system_error_exception(struct pt_regs *regs)
  53 {
  54         int cpu = smp_processor_id();
  55
  56         printk(KERN_ERR "System Error Interrupt on CPU %d !\n", cpu);
  57         dump_fir(cpu);
  58         dump_stack();
  59 }
  60
  61 void cbe_maintenance_exception(struct pt_regs *regs)
  62 {
  63         int cpu = smp_processor_id();
  64
  65         /*
  66          * Nothing implemented for the maintenance interrupt at this point
  67          */
  68
  69         printk(KERN_ERR "Unhandled Maintenance interrupt on CPU %d !\n", cpu);
  70         dump_stack();
  71 }
  72
  73 void cbe_thermal_exception(struct pt_regs *regs)
  74 {
  75         int cpu = smp_processor_id();
  76
  77         /*
  78          * Nothing implemented for the thermal interrupt at this point
  79          */
  80
  81         printk(KERN_ERR "Unhandled Thermal interrupt on CPU %d !\n", cpu);
  82         dump_stack();
  83 }
  84
  85 static int cbe_machine_check_handler(struct pt_regs *regs)
  86 {
  87         int cpu = smp_processor_id();
  88
  89         printk(KERN_ERR "Machine Check Interrupt on CPU %d !\n", cpu);
  90         dump_fir(cpu);
  91
  92         /* No recovery from this code now, lets continue */
  93         return 0;
  94 }
  95
  96 struct ptcal_area {
  97         struct list_head list;
  98         int nid;
  99         int order;
 100         struct page *pages;
 101 };
 102
 103 static LIST_HEAD(ptcal_list);
 104
 105 static int ptcal_start_tok, ptcal_stop_tok;
 106
 107 static int __init cbe_ptcal_enable_on_node(int nid, int order)
 108 {
 109         struct ptcal_area *area;
 110         int ret = -ENOMEM;
 111         unsigned long addr;
 112
 113 #ifdef CONFIG_CRASH_DUMP
 114         rtas_call(ptcal_stop_tok, 1, 1, NULL, nid);
 115 #endif
 116
 117         area = kmalloc(sizeof(*area), GFP_KERNEL);
 118         if (!area)
 119                 goto out_err;
 120
 121         area->nid = nid;
 122         area->order = order;
 123         area->pages = alloc_pages_node(area->nid, GFP_KERNEL, area->order);
 124
 125         if (!area->pages)
 126                 goto out_free_area;
 127
 128         addr = __pa(page_address(area->pages));
 129
 130         ret = -EIO;
 131         if (rtas_call(ptcal_start_tok, 3, 1, NULL, area->nid,
 132                                 (unsigned int)(addr >> 32),
 133                                 (unsigned int)(addr & 0xffffffff))) {
 134                 printk(KERN_ERR "%s: error enabling PTCAL on node %d!\n",
 135                                 __FUNCTION__, nid);
 136                 goto out_free_pages;
 137         }
 138
 139         list_add(&area->list, &ptcal_list);
 140
 141         return 0;
 142
 143 out_free_pages:
 144         __free_pages(area->pages, area->order);
 145 out_free_area:
 146         kfree(area);
 147 out_err:
 148         return ret;
 149 }
 150
 151 static int __init cbe_ptcal_enable(void)
 152 {
 153         const u32 *size;
 154         struct device_node *np;
 155         int order, found_mic = 0;
 156
 157         np = of_find_node_by_path("/rtas");
 158         if (!np)
 159                 return -ENODEV;
 160
 161         size = of_get_property(np, "ibm,cbe-ptcal-size", NULL);
 162         if (!size)
 163                 return -ENODEV;
 164
 165         pr_debug("%s: enabling PTCAL, size = 0x%x\n", __FUNCTION__, *size);
 166         order = get_order(*size);
 167         of_node_put(np);
 168
 169         /* support for malta device trees, with be@/mic@ nodes */
 170         for_each_node_by_type(np, "mic-tm") {
 171                 cbe_ptcal_enable_on_node(of_node_to_nid(np), order);
 172                 found_mic = 1;
 173         }
 174
 175         if (found_mic)
 176                 return 0;
 177
 178         /* support for older device tree - use cpu nodes */
 179         for_each_node_by_type(np, "cpu") {
 180                 const u32 *nid = of_get_property(np, "node-id", NULL);
 181                 if (!nid) {
 182                         printk(KERN_ERR "%s: node %s is missing node-id?\n",
 183                                         __FUNCTION__, np->full_name);
 184                         continue;
 185                 }
 186                 cbe_ptcal_enable_on_node(*nid, order);
 187                 found_mic = 1;
 188         }
 189
 190         return found_mic ? 0 : -ENODEV;
 191 }
 192
 193 static int cbe_ptcal_disable(void)
 194 {
 195         struct ptcal_area *area, *tmp;
 196         int ret = 0;
 197
 198         pr_debug("%s: disabling PTCAL\n", __FUNCTION__);
 199
 200         list_for_each_entry_safe(area, tmp, &ptcal_list, list) {
 201                 /* disable ptcal on this node */
 202                 if (rtas_call(ptcal_stop_tok, 1, 1, NULL, area->nid)) {
 203                         printk(KERN_ERR "%s: error disabling PTCAL "
 204                                         "on node %d!\n", __FUNCTION__,
 205                                         area->nid);
 206                         ret = -EIO;
 207                         continue;
 208                 }
 209
 210                 /* ensure we can access the PTCAL area */
 211                 memset(page_address(area->pages), 0,
 212                                 1 << (area->order + PAGE_SHIFT));
 213
 214                 /* clean up */
 215                 list_del(&area->list);
 216                 __free_pages(area->pages, area->order);
 217                 kfree(area);
 218         }
 219
 220         return ret;
 221 }
 222
 223 static int cbe_ptcal_notify_reboot(struct notifier_block *nb,
 224                 unsigned long code, void *data)
 225 {
 226         return cbe_ptcal_disable();
 227 }
 228
 229 static struct notifier_block cbe_ptcal_reboot_notifier = {
 230         .notifier_call = cbe_ptcal_notify_reboot
 231 };
 232
 233 int __init cbe_ptcal_init(void)
 234 {
 235         int ret;
 236         ptcal_start_tok = rtas_token("ibm,cbe-start-ptcal");
 237         ptcal_stop_tok = rtas_token("ibm,cbe-stop-ptcal");
 238
 239         if (ptcal_start_tok == RTAS_UNKNOWN_SERVICE
 240                         || ptcal_stop_tok == RTAS_UNKNOWN_SERVICE)
 241                 return -ENODEV;
 242
 243         ret = register_reboot_notifier(&cbe_ptcal_reboot_notifier);
 244         if (ret) {
 245                 printk(KERN_ERR "Can't disable PTCAL, so not enabling\n");
 246                 return ret;
 247         }
 248
 249         return cbe_ptcal_enable();
 250 }
 251
 252 arch_initcall(cbe_ptcal_init);
 253
 254 void __init cbe_ras_init(void)
 255 {
 256         unsigned long hid0;
 257
 258         /*
 259          * Enable System Error & thermal interrupts and wakeup conditions
 260          */
 261
 262         hid0 = mfspr(SPRN_HID0);
 263         hid0 |= HID0_CBE_THERM_INT_EN | HID0_CBE_THERM_WAKEUP |
 264                 HID0_CBE_SYSERR_INT_EN | HID0_CBE_SYSERR_WAKEUP;
 265         mtspr(SPRN_HID0, hid0);
 266         mb();
 267
 268         /*
 269          * Install machine check handler. Leave setting of precise mode to
 270          * what the firmware did for now
 271          */
 272         ppc_md.machine_check_exception = cbe_machine_check_handler;
 273         mb();
 274
 275         /*
 276          * For now, we assume that IOC_FIR is already set to forward some
 277          * error conditions to the System Error handler. If that is not true
 278          * then it will have to be fixed up here.
 279          */
 280 }