1219820Sjeff/* 2219820Sjeff * Copyright (c) 2005 Cisco Systems. All rights reserved. 3219820Sjeff * 4219820Sjeff * This software is available to you under a choice of one of two 5219820Sjeff * licenses. You may choose to be licensed under the terms of the GNU 6219820Sjeff * General Public License (GPL) Version 2, available from the file 7219820Sjeff * COPYING in the main directory of this source tree, or the 8219820Sjeff * OpenIB.org BSD license below: 9219820Sjeff * 10219820Sjeff * Redistribution and use in source and binary forms, with or 11219820Sjeff * without modification, are permitted provided that the following 12219820Sjeff * conditions are met: 13219820Sjeff * 14219820Sjeff * - Redistributions of source code must retain the above 15219820Sjeff * copyright notice, this list of conditions and the following 16219820Sjeff * disclaimer. 17219820Sjeff * 18219820Sjeff * - Redistributions in binary form must reproduce the above 19219820Sjeff * copyright notice, this list of conditions and the following 20219820Sjeff * disclaimer in the documentation and/or other materials 21219820Sjeff * provided with the distribution. 22219820Sjeff * 23219820Sjeff * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 24219820Sjeff * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 25219820Sjeff * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 26219820Sjeff * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 27219820Sjeff * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 28219820Sjeff * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 29219820Sjeff * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 30219820Sjeff * SOFTWARE. 31219820Sjeff */ 32219820Sjeff 33219820Sjeff#include <linux/jiffies.h> 34219820Sjeff#include <linux/timer.h> 35219820Sjeff#include <linux/workqueue.h> 36219820Sjeff 37219820Sjeff#include "mthca_dev.h" 38219820Sjeff 39219820Sjeffenum { 40219820Sjeff MTHCA_CATAS_TYPE_INTERNAL = 0, 41219820Sjeff MTHCA_CATAS_TYPE_UPLINK = 3, 42219820Sjeff MTHCA_CATAS_TYPE_DDR = 4, 43219820Sjeff MTHCA_CATAS_TYPE_PARITY = 5, 44219820Sjeff}; 45219820Sjeff 46219820Sjeff#define MTHCA_CATAS_POLL_INTERVAL (5 * HZ) 47219820Sjeff 48219820Sjeffstatic DEFINE_SPINLOCK(catas_lock); 49219820Sjeff 50219820Sjeffstatic LIST_HEAD(catas_list); 51219820Sjeffstatic struct workqueue_struct *catas_wq; 52219820Sjeffstatic struct work_struct catas_work; 53219820Sjeff 54219820Sjeffstatic int catas_reset_disable; 55219820Sjeffmodule_param_named(catas_reset_disable, catas_reset_disable, int, 0644); 56219820SjeffMODULE_PARM_DESC(catas_reset_disable, "disable reset on catastrophic event if nonzero"); 57219820Sjeff 58219820Sjeffstatic void catas_reset(struct work_struct *work) 59219820Sjeff{ 60219820Sjeff struct mthca_dev *dev, *tmpdev; 61219820Sjeff LIST_HEAD(tlist); 62219820Sjeff int ret; 63219820Sjeff 64219820Sjeff mutex_lock(&mthca_device_mutex); 65219820Sjeff 66219820Sjeff spin_lock_irq(&catas_lock); 67219820Sjeff list_splice_init(&catas_list, &tlist); 68219820Sjeff spin_unlock_irq(&catas_lock); 69219820Sjeff 70219820Sjeff list_for_each_entry_safe(dev, tmpdev, &tlist, catas_err.list) { 71219820Sjeff struct pci_dev *pdev = dev->pdev; 72219820Sjeff ret = __mthca_restart_one(dev->pdev); 73219820Sjeff /* 'dev' now is not valid */ 74219820Sjeff if (ret) 75219820Sjeff printk(KERN_ERR "mthca %s: Reset failed (%d)\n", 76219820Sjeff pci_name(pdev), ret); 77219820Sjeff else { 78219820Sjeff struct mthca_dev *d = pci_get_drvdata(pdev); 79219820Sjeff mthca_dbg(d, "Reset succeeded\n"); 80219820Sjeff } 81219820Sjeff } 82219820Sjeff 83219820Sjeff mutex_unlock(&mthca_device_mutex); 84219820Sjeff} 85219820Sjeff 86219820Sjeffstatic void handle_catas(struct mthca_dev *dev) 87219820Sjeff{ 88219820Sjeff struct ib_event event; 89219820Sjeff unsigned long flags; 90219820Sjeff const char *type; 91219820Sjeff int i; 92219820Sjeff 93219820Sjeff event.device = &dev->ib_dev; 94219820Sjeff event.event = IB_EVENT_DEVICE_FATAL; 95219820Sjeff event.element.port_num = 0; 96219820Sjeff dev->active = 0; 97219820Sjeff 98219820Sjeff ib_dispatch_event(&event); 99219820Sjeff 100219820Sjeff switch (swab32(readl(dev->catas_err.map)) >> 24) { 101219820Sjeff case MTHCA_CATAS_TYPE_INTERNAL: 102219820Sjeff type = "internal error"; 103219820Sjeff break; 104219820Sjeff case MTHCA_CATAS_TYPE_UPLINK: 105219820Sjeff type = "uplink bus error"; 106219820Sjeff break; 107219820Sjeff case MTHCA_CATAS_TYPE_DDR: 108219820Sjeff type = "DDR data error"; 109219820Sjeff break; 110219820Sjeff case MTHCA_CATAS_TYPE_PARITY: 111219820Sjeff type = "internal parity error"; 112219820Sjeff break; 113219820Sjeff default: 114219820Sjeff type = "unknown error"; 115219820Sjeff break; 116219820Sjeff } 117219820Sjeff 118219820Sjeff mthca_err(dev, "Catastrophic error detected: %s\n", type); 119219820Sjeff for (i = 0; i < dev->catas_err.size; ++i) 120219820Sjeff mthca_err(dev, " buf[%02x]: %08x\n", 121219820Sjeff i, swab32(readl(dev->catas_err.map + i))); 122219820Sjeff 123219820Sjeff if (catas_reset_disable) 124219820Sjeff return; 125219820Sjeff 126219820Sjeff spin_lock_irqsave(&catas_lock, flags); 127219820Sjeff list_add(&dev->catas_err.list, &catas_list); 128219820Sjeff queue_work(catas_wq, &catas_work); 129219820Sjeff spin_unlock_irqrestore(&catas_lock, flags); 130219820Sjeff} 131219820Sjeff 132219820Sjeffstatic void poll_catas(unsigned long dev_ptr) 133219820Sjeff{ 134219820Sjeff struct mthca_dev *dev = (struct mthca_dev *) dev_ptr; 135219820Sjeff int i; 136219820Sjeff 137219820Sjeff for (i = 0; i < dev->catas_err.size; ++i) 138219820Sjeff if (readl(dev->catas_err.map + i)) { 139219820Sjeff handle_catas(dev); 140219820Sjeff return; 141219820Sjeff } 142219820Sjeff 143219820Sjeff mod_timer(&dev->catas_err.timer, 144219820Sjeff round_jiffies(jiffies + MTHCA_CATAS_POLL_INTERVAL)); 145219820Sjeff} 146219820Sjeff 147219820Sjeffvoid mthca_start_catas_poll(struct mthca_dev *dev) 148219820Sjeff{ 149219820Sjeff unsigned long addr; 150219820Sjeff 151219820Sjeff init_timer(&dev->catas_err.timer); 152219820Sjeff dev->catas_err.map = NULL; 153219820Sjeff 154219820Sjeff addr = pci_resource_start(dev->pdev, 0) + 155219820Sjeff ((pci_resource_len(dev->pdev, 0) - 1) & 156219820Sjeff dev->catas_err.addr); 157219820Sjeff 158219820Sjeff dev->catas_err.map = ioremap(addr, dev->catas_err.size * 4); 159219820Sjeff if (!dev->catas_err.map) { 160219820Sjeff mthca_warn(dev, "couldn't map catastrophic error region " 161219820Sjeff "at 0x%lx/0x%x\n", addr, dev->catas_err.size * 4); 162219820Sjeff return; 163219820Sjeff } 164219820Sjeff 165219820Sjeff dev->catas_err.timer.data = (unsigned long) dev; 166219820Sjeff dev->catas_err.timer.function = poll_catas; 167219820Sjeff dev->catas_err.timer.expires = jiffies + MTHCA_CATAS_POLL_INTERVAL; 168219820Sjeff INIT_LIST_HEAD(&dev->catas_err.list); 169219820Sjeff add_timer(&dev->catas_err.timer); 170219820Sjeff} 171219820Sjeff 172219820Sjeffvoid mthca_stop_catas_poll(struct mthca_dev *dev) 173219820Sjeff{ 174219820Sjeff del_timer_sync(&dev->catas_err.timer); 175219820Sjeff 176219820Sjeff if (dev->catas_err.map) 177219820Sjeff iounmap(dev->catas_err.map); 178219820Sjeff 179219820Sjeff spin_lock_irq(&catas_lock); 180219820Sjeff list_del(&dev->catas_err.list); 181219820Sjeff spin_unlock_irq(&catas_lock); 182219820Sjeff} 183219820Sjeff 184219820Sjeffint __init mthca_catas_init(void) 185219820Sjeff{ 186219820Sjeff INIT_WORK(&catas_work, catas_reset); 187219820Sjeff 188219820Sjeff catas_wq = create_singlethread_workqueue("mthcacatas"); 189219820Sjeff if (!catas_wq) 190219820Sjeff return -ENOMEM; 191219820Sjeff 192219820Sjeff return 0; 193219820Sjeff} 194219820Sjeff 195219820Sjeffvoid mthca_catas_cleanup(void) 196219820Sjeff{ 197219820Sjeff destroy_workqueue(catas_wq); 198219820Sjeff} 199