eeh.c

/*
 * eeh.c
 * Copyright (C) 2001 Dave Engebretsen & Todd Inglett IBM Corporation
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
 */

#include <linux/init.h>
#include <linux/list.h>
#include <linux/notifier.h>
#include <linux/pci.h>
#include <linux/proc_fs.h>
#include <linux/rbtree.h>
#include <linux/seq_file.h>
#include <linux/spinlock.h>
#include <asm/atomic.h>
#include <asm/eeh.h>
#include <asm/io.h>
#include <asm/machdep.h>
#include <asm/rtas.h>
#include <asm/atomic.h>
#include <asm/systemcfg.h>
#include <asm/ppc-pci.h>

#undef DEBUG

/** Overview:
 *  EEH, or "Extended Error Handling" is a PCI bridge technology for
 *  dealing with PCI bus errors that can't be dealt with within the
 *  usual PCI framework, except by check-stopping the CPU.  Systems
 *  that are designed for high-availability/reliability cannot afford
 *  to crash due to a "mere" PCI error, thus the need for EEH.
 *  An EEH-capable bridge operates by converting a detected error
 *  into a "slot freeze", taking the PCI adapter off-line, making
 *  the slot behave, from the OS'es point of view, as if the slot
 *  were "empty": all reads return 0xff's and all writes are silently
 *  ignored.  EEH slot isolation events can be triggered by parity
 *  errors on the address or data busses (e.g. during posted writes),
 *  which in turn might be caused by low voltage on the bus, dust,
 *  vibration, humidity, radioactivity or plain-old failed hardware.
 *
 *  Note, however, that one of the leading causes of EEH slot
 *  freeze events are buggy device drivers, buggy device microcode,
 *  or buggy device hardware.  This is because any attempt by the
 *  device to bus-master data to a memory address that is not
 *  assigned to the device will trigger a slot freeze.   (The idea
 *  is to prevent devices-gone-wild from corrupting system memory).
 *  Buggy hardware/drivers will have a miserable time co-existing
 *  with EEH.
 *
 *  Ideally, a PCI device driver, when suspecting that an isolation
 *  event has occured (e.g. by reading 0xff's), will then ask EEH
 *  whether this is the case, and then take appropriate steps to
 *  reset the PCI slot, the PCI device, and then resume operations.
 *  However, until that day,  the checking is done here, with the
 *  eeh_check_failure() routine embedded in the MMIO macros.  If
 *  the slot is found to be isolated, an "EEH Event" is synthesized
 *  and sent out for processing.
 */

/* EEH event workqueue setup. */
static DEFINE_SPINLOCK(eeh_eventlist_lock);
LIST_HEAD(eeh_eventlist);
static void eeh_event_handler(void *);
DECLARE_WORK(eeh_event_wq, eeh_event_handler, NULL);

static struct notifier_block *eeh_notifier_chain;

/*
 * If a device driver keeps reading an MMIO register in an interrupt
 * handler after a slot isolation event has occurred, we assume it
 * is broken and panic.  This sets the threshold for how many read
 * attempts we allow before panicking.
 */
#define EEH_MAX_FAILS	1000
static atomic_t eeh_fail_count;

/* RTAS tokens */
static int ibm_set_eeh_option;
static int ibm_set_slot_reset;
static int ibm_read_slot_reset_state;
static int ibm_read_slot_reset_state2;
static int ibm_slot_error_detail;

static int eeh_subsystem_enabled;

/* Lock to avoid races due to multiple reports of an error */
static DEFINE_SPINLOCK(confirm_error_lock);

/* Buffer for reporting slot-error-detail rtas calls */
static unsigned char slot_errbuf[RTAS_ERROR_LOG_MAX];
static DEFINE_SPINLOCK(slot_errbuf_lock);
static int eeh_error_buf_size;

/* System monitoring statistics */
static DEFINE_PER_CPU(unsigned long, no_device);
static DEFINE_PER_CPU(unsigned long, no_dn);
static DEFINE_PER_CPU(unsigned long, no_cfg_addr);
static DEFINE_PER_CPU(unsigned long, ignored_check);
static DEFINE_PER_CPU(unsigned long, total_mmio_ffs);
static DEFINE_PER_CPU(unsigned long, false_positives);
static DEFINE_PER_CPU(unsigned long, ignored_failures);
static DEFINE_PER_CPU(unsigned long, slot_resets);

/**
 * The pci address cache subsystem.  This subsystem places
 * PCI device address resources into a red-black tree, sorted
 * according to the address range, so that given only an i/o
 * address, the corresponding PCI device can be **quickly**
 * found. It is safe to perform an address lookup in an interrupt
 * context; this ability is an important feature.
 *
 * Currently, the only customer of this code is the EEH subsystem;
 * thus, this code has been somewhat tailored to suit EEH better.
 * In particular, the cache does *not* hold the addresses of devices
 * for which EEH is not enabled.
 *
 * (Implementation Note: The RB tree seems to be better/faster
 * than any hash algo I could think of for this problem, even
 * with the penalty of slow pointer chases for d-cache misses).
 */
struct pci_io_addr_range
{
	struct rb_node rb_node;
	unsigned long addr_lo;
	unsigned long addr_hi;
	struct pci_dev *pcidev;
	unsigned int flags;
};

static struct pci_io_addr_cache
{
	struct rb_root rb_root;
	spinlock_t piar_lock;
} pci_io_addr_cache_root;

static inline struct pci_dev *__pci_get_device_by_addr(unsigned long addr)
{
	struct rb_node *n = pci_io_addr_cache_root.rb_root.rb_node;

	while (n) {
		struct pci_io_addr_range *piar;
		piar = rb_entry(n, struct pci_io_addr_range, rb_node);

		if (addr < piar->addr_lo) {
			n = n->rb_left;
		} else {
			if (addr > piar->addr_hi) {
				n = n->rb_right;
			} else {
				pci_dev_get(piar->pcidev);
				return piar->pcidev;
			}
		}
	}

	return NULL;
}

/**
 * pci_get_device_by_addr - Get device, given only address
 * @addr: mmio (PIO) phys address or i/o port number
 *
 * Given an mmio phys address, or a port number, find a pci device
 * that implements this address.  Be sure to pci_dev_put the device
 * when finished.  I/O port numbers are assumed to be offset
 * from zero (that is, they do *not* have pci_io_addr added in).
 * It is safe to call this function within an interrupt.
 */
static struct pci_dev *pci_get_device_by_addr(unsigned long addr)
{
	struct pci_dev *dev;
	unsigned long flags;

	spin_lock_irqsave(&pci_io_addr_cache_root.piar_lock, flags);
	dev = __pci_get_device_by_addr(addr);
	spin_unlock_irqrestore(&pci_io_addr_cache_root.piar_lock, flags);
	return dev;
}

#ifdef DEBUG
/*
 * Handy-dandy debug print routine, does nothing more
 * than print out the contents of our addr cache.
 */
static void pci_addr_cache_print(struct pci_io_addr_cache *cache)
{
	struct rb_node *n;