eeh.c

linux-2.6.15.6

	/* Log the error with the rtas logger */
	spin_lock_irqsave(&slot_errbuf_lock, flags);
	memset(slot_errbuf, 0, eeh_error_buf_size);

	rc = rtas_call(ibm_slot_error_detail,
	               8, 1, NULL, pdn->eeh_config_addr,
	               BUID_HI(pdn->phb->buid),
	               BUID_LO(pdn->phb->buid), NULL, 0,
	               virt_to_phys(slot_errbuf),
	               eeh_error_buf_size,
	               severity);

	if (rc == 0)
		log_error(slot_errbuf, ERR_TYPE_RTAS_LOG, 0);
	spin_unlock_irqrestore(&slot_errbuf_lock, flags);
}

/**
 * read_slot_reset_state - Read the reset state of a device node's slot
 * @dn: device node to read
 * @rets: array to return results in
 */
static int read_slot_reset_state(struct pci_dn *pdn, int rets[])
{
	int token, outputs;

	if (ibm_read_slot_reset_state2 != RTAS_UNKNOWN_SERVICE) {
		token = ibm_read_slot_reset_state2;
		outputs = 4;
	} else {
		token = ibm_read_slot_reset_state;
		rets[2] = 0; /* fake PE Unavailable info */
		outputs = 3;
	}

	return rtas_call(token, 3, outputs, rets, pdn->eeh_config_addr,
			 BUID_HI(pdn->phb->buid), BUID_LO(pdn->phb->buid));
}

/**
 * eeh_token_to_phys - convert EEH address token to phys address
 * @token i/o token, should be address in the form 0xA....
 */
static inline unsigned long eeh_token_to_phys(unsigned long token)
{
	pte_t *ptep;
	unsigned long pa;

	ptep = find_linux_pte(init_mm.pgd, token);
	if (!ptep)
		return token;
	pa = pte_pfn(*ptep) << PAGE_SHIFT;

	return pa | (token & (PAGE_SIZE-1));
}

/** 
 * Return the "partitionable endpoint" (pe) under which this device lies
 */
static struct device_node * find_device_pe(struct device_node *dn)
{
	while ((dn->parent) && PCI_DN(dn->parent) &&
	      (PCI_DN(dn->parent)->eeh_mode & EEH_MODE_SUPPORTED)) {
		dn = dn->parent;
	}
	return dn;
}

/** Mark all devices that are peers of this device as failed.
 *  Mark the device driver too, so that it can see the failure
 *  immediately; this is critical, since some drivers poll
 *  status registers in interrupts ... If a driver is polling,
 *  and the slot is frozen, then the driver can deadlock in
 *  an interrupt context, which is bad.
 */

static void __eeh_mark_slot (struct device_node *dn, int mode_flag)
{
	while (dn) {
		if (PCI_DN(dn)) {
			PCI_DN(dn)->eeh_mode |= mode_flag;

			if (dn->child)
				__eeh_mark_slot (dn->child, mode_flag);
		}
		dn = dn->sibling;
	}
}

void eeh_mark_slot (struct device_node *dn, int mode_flag)
{
	dn = find_device_pe (dn);
	PCI_DN(dn)->eeh_mode |= mode_flag;
	__eeh_mark_slot (dn->child, mode_flag);
}

static void __eeh_clear_slot (struct device_node *dn, int mode_flag)
{
	while (dn) {
		if (PCI_DN(dn)) {
			PCI_DN(dn)->eeh_mode &= ~mode_flag;
			PCI_DN(dn)->eeh_check_count = 0;
			if (dn->child)
				__eeh_clear_slot (dn->child, mode_flag);
		}
		dn = dn->sibling;
	}
}

void eeh_clear_slot (struct device_node *dn, int mode_flag)
{
	unsigned long flags;
	spin_lock_irqsave(&confirm_error_lock, flags);
	dn = find_device_pe (dn);
	PCI_DN(dn)->eeh_mode &= ~mode_flag;
	PCI_DN(dn)->eeh_check_count = 0;
	__eeh_clear_slot (dn->child, mode_flag);
	spin_unlock_irqrestore(&confirm_error_lock, flags);
}

/**
 * eeh_dn_check_failure - check if all 1's data is due to EEH slot freeze
 * @dn device node
 * @dev pci device, if known
 *
 * Check for an EEH failure for the given device node.  Call this
 * routine if the result of a read was all 0xff's and you want to
 * find out if this is due to an EEH slot freeze.  This routine
 * will query firmware for the EEH status.
 *
 * Returns 0 if there has not been an EEH error; otherwise returns
 * a non-zero value and queues up a slot isolation event notification.
 *
 * It is safe to call this routine in an interrupt context.
 */
int eeh_dn_check_failure(struct device_node *dn, struct pci_dev *dev)
{
	int ret;
	int rets[3];
	unsigned long flags;
	struct pci_dn *pdn;
	int rc = 0;

	__get_cpu_var(total_mmio_ffs)++;

	if (!eeh_subsystem_enabled)
		return 0;

	if (!dn) {
		__get_cpu_var(no_dn)++;
		return 0;
	}
	pdn = PCI_DN(dn);

	/* Access to IO BARs might get this far and still not want checking. */
	if (!(pdn->eeh_mode & EEH_MODE_SUPPORTED) ||
	    pdn->eeh_mode & EEH_MODE_NOCHECK) {
		__get_cpu_var(ignored_check)++;
#ifdef DEBUG
		printk ("EEH:ignored check (%x) for %s %s\n", 
		        pdn->eeh_mode, pci_name (dev), dn->full_name);
#endif
		return 0;
	}

	if (!pdn->eeh_config_addr) {
		__get_cpu_var(no_cfg_addr)++;
		return 0;
	}

	/* If we already have a pending isolation event for this
	 * slot, we know it's bad already, we don't need to check.
	 * Do this checking under a lock; as multiple PCI devices
	 * in one slot might report errors simultaneously, and we
	 * only want one error recovery routine running.
	 */
	spin_lock_irqsave(&confirm_error_lock, flags);
	rc = 1;
	if (pdn->eeh_mode & EEH_MODE_ISOLATED) {
		pdn->eeh_check_count ++;
		if (pdn->eeh_check_count >= EEH_MAX_FAILS) {
			printk (KERN_ERR "EEH: Device driver ignored %d bad reads, panicing\n",
			        pdn->eeh_check_count);
			dump_stack();
			
			/* re-read the slot reset state */
			if (read_slot_reset_state(pdn, rets) != 0)
				rets[0] = -1;	/* reset state unknown */

			/* If we are here, then we hit an infinite loop. Stop. */
			panic("EEH: MMIO halt (%d) on device:%s\n", rets[0], pci_name(dev));
		}
		goto dn_unlock;
	}

	/*
	 * Now test for an EEH failure.  This is VERY expensive.
	 * Note that the eeh_config_addr may be a parent device
	 * in the case of a device behind a bridge, or it may be
	 * function zero of a multi-function device.
	 * In any case they must share a common PHB.
	 */
	ret = read_slot_reset_state(pdn, rets);

	/* If the call to firmware failed, punt */
	if (ret != 0) {
		printk(KERN_WARNING "EEH: read_slot_reset_state() failed; rc=%d dn=%s\n",
		       ret, dn->full_name);
		__get_cpu_var(false_positives)++;
		rc = 0;
		goto dn_unlock;
	}

	/* If EEH is not supported on this device, punt. */
	if (rets[1] != 1) {
		printk(KERN_WARNING "EEH: event on unsupported device, rc=%d dn=%s\n",
		       ret, dn->full_name);
		__get_cpu_var(false_positives)++;
		rc = 0;
		goto dn_unlock;
	}

	/* If not the kind of error we know about, punt. */
	if (rets[0] != 2 && rets[0] != 4 && rets[0] != 5) {
		__get_cpu_var(false_positives)++;
		rc = 0;
		goto dn_unlock;
	}

	/* Note that config-io to empty slots may fail;
	 * we recognize empty because they don't have children. */
	if ((rets[0] == 5) && (dn->child == NULL)) {
		__get_cpu_var(false_positives)++;
		rc = 0;
		goto dn_unlock;
	}

	__get_cpu_var(slot_resets)++;
 
	/* Avoid repeated reports of this failure, including problems
	 * with other functions on this device, and functions under
	 * bridges. */
	eeh_mark_slot (dn, EEH_MODE_ISOLATED);
	spin_unlock_irqrestore(&confirm_error_lock, flags);

	eeh_send_failure_event (dn, dev, rets[0], rets[2]);
	
	/* Most EEH events are due to device driver bugs.  Having
	 * a stack trace will help the device-driver authors figure
	 * out what happened.  So print that out. */
	if (rets[0] != 5) dump_stack();
	return 1;

dn_unlock:
	spin_unlock_irqrestore(&confirm_error_lock, flags);
	return rc;
}

EXPORT_SYMBOL_GPL(eeh_dn_check_failure);

/**
 * eeh_check_failure - check if all 1's data is due to EEH slot freeze
 * @token i/o token, should be address in the form 0xA....
 * @val value, should be all 1's (XXX why do we need this arg??)
 *
 * Check for an EEH failure at the given token address.  Call this
 * routine if the result of a read was all 0xff's and you want to
 * find out if this is due to an EEH slot freeze event.  This routine
 * will query firmware for the EEH status.
 *
 * Note this routine is safe to call in an interrupt context.
 */
unsigned long eeh_check_failure(const volatile void __iomem *token, unsigned long val)
{
	unsigned long addr;
	struct pci_dev *dev;
	struct device_node *dn;

	/* Finding the phys addr + pci device; this is pretty quick. */
	addr = eeh_token_to_phys((unsigned long __force) token);
	dev = pci_get_device_by_addr(addr);
	if (!dev) {
		__get_cpu_var(no_device)++;
		return val;
	}

	dn = pci_device_to_OF_node(dev);
	eeh_dn_check_failure (dn, dev);

	pci_dev_put(dev);
	return val;
}

EXPORT_SYMBOL(eeh_check_failure);

/* ------------------------------------------------------------- */
/* The code below deals with error recovery */

/** Return negative value if a permanent error, else return
 * a number of milliseconds to wait until the PCI slot is
 * ready to be used.
 */
static int
eeh_slot_availability(struct pci_dn *pdn)
{
	int rc;
	int rets[3];

	rc = read_slot_reset_state(pdn, rets);

	if (rc) return rc;

	if (rets[1] == 0) return -1;  /* EEH is not supported */
	if (rets[0] == 0)  return 0;  /* Oll Korrect */
	if (rets[0] == 5) {
		if (rets[2] == 0) return -1; /* permanently unavailable */
		return rets[2]; /* number of millisecs to wait */
	}
	return -1;
}

/** rtas_pci_slot_reset raises/lowers the pci #RST line
 *  state: 1/0 to raise/lower the #RST
 *
 * Clear the EEH-frozen condition on a slot.  This routine
 * asserts the PCI #RST line if the 'state' argument is '1',
 * and drops the #RST line if 'state is '0'.  This routine is
 * safe to call in an interrupt context.
 *
 */

static void
rtas_pci_slot_reset(struct pci_dn *pdn, int state)
{
	int rc;

	BUG_ON (pdn==NULL); 

	if (!pdn->phb) {
		printk (KERN_WARNING "EEH: in slot reset, device node %s has no phb\n",
		        pdn->node->full_name);
		return;
	}

	rc = rtas_call(ibm_set_slot_reset,4,1, NULL,
	               pdn->eeh_config_addr,
	               BUID_HI(pdn->phb->buid),
	               BUID_LO(pdn->phb->buid),
	               state);
	if (rc) {
		printk (KERN_WARNING "EEH: Unable to reset the failed slot, (%d) #RST=%d dn=%s\n", 
		        rc, state, pdn->node->full_name);
		return;
	}
}

/** rtas_set_slot_reset -- assert the pci #RST line for 1/4 second
 *  dn -- device node to be reset.
 */

void
rtas_set_slot_reset(struct pci_dn *pdn)
{
	int i, rc;

	rtas_pci_slot_reset (pdn, 1);

	/* The PCI bus requires that the reset be held high for at least
	 * a 100 milliseconds. We wait a bit longer 'just in case'.  */

#define PCI_BUS_RST_HOLD_TIME_MSEC 250
	msleep (PCI_BUS_RST_HOLD_TIME_MSEC);
	
	/* We might get hit with another EEH freeze as soon as the 
	 * pci slot reset line is dropped. Make sure we don't miss
	 * these, and clear the flag now. */
	eeh_clear_slot (pdn->node, EEH_MODE_ISOLATED);

	rtas_pci_slot_reset (pdn, 0);

	/* After a PCI slot has been reset, the PCI Express spec requires
	 * a 1.5 second idle time for the bus to stabilize, before starting
	 * up traffic. */
#define PCI_BUS_SETTLE_TIME_MSEC 1800
	msleep (PCI_BUS_SETTLE_TIME_MSEC);

	/* Now double check with the firmware to make sure the device is
	 * ready to be used; if not, wait for recovery. */
	for (i=0; i<10; i++) {
		rc = eeh_slot_availability (pdn);
		if (rc <= 0) break;

		msleep (rc+100);
	}
}

/* ------------------------------------------------------- */
/** Save and restore of PCI BARs
 *
 * Although firmware will set up BARs during boot, it doesn't
 * set up device BAR's after a device reset, although it will,
 * if requested, set up bridge configuration. Thus, we need to
 * configure the PCI devices ourselves.  
 */