kernel/arch/powerpc/platforms/powernv/opal-hmi.c

   1 /*
   2  * OPAL hypervisor Maintenance interrupt handling support in PowreNV.
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License as published by
   6  * the Free Software Foundation; either version 2 of the License, or
   7  * (at your option) any later version.
   8  *
   9  * This program is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; If not, see <http://www.gnu.org/licenses/>.
  16  *
  17  * Copyright 2014 IBM Corporation
  18  * Author: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
  19  */
  20
  21 #undef DEBUG
  22
  23 #include <linux/kernel.h>
  24 #include <linux/init.h>
  25 #include <linux/of.h>
  26 #include <linux/mm.h>
  27 #include <linux/slab.h>
  28
  29 #include <asm/opal.h>
  30 #include <asm/cputable.h>
  31 #include <asm/machdep.h>
  32
  33 static int opal_hmi_handler_nb_init;
  34 struct OpalHmiEvtNode {
  35         struct list_head list;
  36         struct OpalHMIEvent hmi_evt;
  37 };
  38 static LIST_HEAD(opal_hmi_evt_list);
  39 static DEFINE_SPINLOCK(opal_hmi_evt_lock);
  40
  41 static void print_hmi_event_info(struct OpalHMIEvent *hmi_evt)
  42 {
  43         const char *level, *sevstr, *error_info;
  44         static const char *hmi_error_types[] = {
  45                 "Malfunction Alert",
  46                 "Processor Recovery done",
  47                 "Processor recovery occurred again",
  48                 "Processor recovery occurred for masked error",
  49                 "Timer facility experienced an error",
  50                 "TFMR SPR is corrupted",
  51                 "UPS (Uniterrupted Power System) Overflow indication",
  52                 "An XSCOM operation failure",
  53                 "An XSCOM operation completed",
  54                 "SCOM has set a reserved FIR bit to cause recovery",
  55                 "Debug trigger has set a reserved FIR bit to cause recovery",
  56                 "A hypervisor resource error occurred"
  57         };
  58
  59         /* Print things out */
  60         if (hmi_evt->version < OpalHMIEvt_V1) {
  61                 pr_err("HMI Interrupt, Unknown event version %d !\n",
  62                         hmi_evt->version);
  63                 return;
  64         }
  65         switch (hmi_evt->severity) {
  66         case OpalHMI_SEV_NO_ERROR:
  67                 level = KERN_INFO;
  68                 sevstr = "Harmless";
  69                 break;
  70         case OpalHMI_SEV_WARNING:
  71                 level = KERN_WARNING;
  72                 sevstr = "";
  73                 break;
  74         case OpalHMI_SEV_ERROR_SYNC:
  75                 level = KERN_ERR;
  76                 sevstr = "Severe";
  77                 break;
  78         case OpalHMI_SEV_FATAL:
  79         default:
  80                 level = KERN_ERR;
  81                 sevstr = "Fatal";
  82                 break;
  83         }
  84
  85         printk("%s%s Hypervisor Maintenance interrupt [%s]\n",
  86                 level, sevstr,
  87                 hmi_evt->disposition == OpalHMI_DISPOSITION_RECOVERED ?
  88                 "Recovered" : "Not recovered");
  89         error_info = hmi_evt->type < ARRAY_SIZE(hmi_error_types) ?
  90                         hmi_error_types[hmi_evt->type]
  91                         : "Unknown";
  92         printk("%s Error detail: %s\n", level, error_info);
  93         printk("%s      HMER: %016llx\n", level, be64_to_cpu(hmi_evt->hmer));
  94         if ((hmi_evt->type == OpalHMI_ERROR_TFAC) ||
  95                 (hmi_evt->type == OpalHMI_ERROR_TFMR_PARITY))
  96                 printk("%s      TFMR: %016llx\n", level,
  97                                                 be64_to_cpu(hmi_evt->tfmr));
  98 }
  99
 100 static void hmi_event_handler(struct work_struct *work)
 101 {
 102         unsigned long flags;
 103         struct OpalHMIEvent *hmi_evt;
 104         struct OpalHmiEvtNode *msg_node;
 105         uint8_t disposition;
 106
 107         spin_lock_irqsave(&opal_hmi_evt_lock, flags);
 108         while (!list_empty(&opal_hmi_evt_list)) {
 109                 msg_node = list_entry(opal_hmi_evt_list.next,
 110                                            struct OpalHmiEvtNode, list);
 111                 list_del(&msg_node->list);
 112                 spin_unlock_irqrestore(&opal_hmi_evt_lock, flags);
 113
 114                 hmi_evt = (struct OpalHMIEvent *) &msg_node->hmi_evt;
 115                 print_hmi_event_info(hmi_evt);
 116                 disposition = hmi_evt->disposition;
 117                 kfree(msg_node);
 118
 119                 /*
 120                  * Check if HMI event has been recovered or not. If not
 121                  * then we can't continue, invoke panic.
 122                  */
 123                 if (disposition != OpalHMI_DISPOSITION_RECOVERED)
 124                         panic("Unrecoverable HMI exception");
 125
 126                 spin_lock_irqsave(&opal_hmi_evt_lock, flags);
 127         }
 128         spin_unlock_irqrestore(&opal_hmi_evt_lock, flags);
 129 }
 130
 131 static DECLARE_WORK(hmi_event_work, hmi_event_handler);
 132 /*
 133  * opal_handle_hmi_event - notifier handler that queues up HMI events
 134  * to be preocessed later.
 135  */
 136 static int opal_handle_hmi_event(struct notifier_block *nb,
 137                           unsigned long msg_type, void *msg)
 138 {
 139         unsigned long flags;
 140         struct OpalHMIEvent *hmi_evt;
 141         struct opal_msg *hmi_msg = msg;
 142         struct OpalHmiEvtNode *msg_node;
 143
 144         /* Sanity Checks */
 145         if (msg_type != OPAL_MSG_HMI_EVT)
 146                 return 0;
 147
 148         /* HMI event info starts from param[0] */
 149         hmi_evt = (struct OpalHMIEvent *)&hmi_msg->params[0];
 150
 151         /* Delay the logging of HMI events to workqueue. */
 152         msg_node = kzalloc(sizeof(*msg_node), GFP_ATOMIC);
 153         if (!msg_node) {
 154                 pr_err("HMI: out of memory, Opal message event not handled\n");
 155                 return -ENOMEM;
 156         }
 157         memcpy(&msg_node->hmi_evt, hmi_evt, sizeof(struct OpalHMIEvent));
 158
 159         spin_lock_irqsave(&opal_hmi_evt_lock, flags);
 160         list_add(&msg_node->list, &opal_hmi_evt_list);
 161         spin_unlock_irqrestore(&opal_hmi_evt_lock, flags);
 162
 163         schedule_work(&hmi_event_work);
 164         return 0;
 165 }
 166
 167 static struct notifier_block opal_hmi_handler_nb = {
 168         .notifier_call  = opal_handle_hmi_event,
 169         .next           = NULL,
 170         .priority       = 0,
 171 };
 172
 173 static int __init opal_hmi_handler_init(void)
 174 {
 175         int ret;
 176
 177         if (!opal_hmi_handler_nb_init) {
 178                 ret = opal_message_notifier_register(
 179                                 OPAL_MSG_HMI_EVT, &opal_hmi_handler_nb);
 180                 if (ret) {
 181                         pr_err("%s: Can't register OPAL event notifier (%d)\n",
 182                                __func__, ret);
 183                         return ret;
 184                 }
 185                 opal_hmi_handler_nb_init = 1;
 186         }
 187         return 0;
 188 }
 189 machine_subsys_initcall(powernv, opal_hmi_handler_init);