arch/powerpc/platforms/cell/ras.c

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