Skip to content

Commit

Permalink
powerpc/eeh: Handle multiple EEH errors
Browse files Browse the repository at this point in the history 
For one PCI error relevant OPAL event, we possibly have multiple
EEH errors for that. For example, multiple frozen PEs detected on
different PHBs. Unfortunately, we didn't cover the case. The patch
enumarates the return value from eeh_ops::next_error() and change
eeh_handle_special_event() and eeh_ops::next_error() to handle all
existing EEH errors.

As Ben pointed out, we needn't list_for_each_entry_safe() since we
are not deleting any PHB from the hose_list and the EEH serialized
lock should be held while purging EEH events. The patch covers those
suggestions as well.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
  • Loading branch information
@shangw @ozbenh
shangw authored and ozbenh committed Jan 15, 2014
1 parent fac515d commit 7e4e786
Show file tree
Hide file tree
Showing 3 changed files with 112 additions and 87 deletions.
10 changes: 10 additions & 0 deletions arch/powerpc/include/asm/eeh.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -118,6 +118,16 @@ static inline struct pci_dev *eeh_dev_to_pci_dev(struct eeh_dev *edev)
return edev ? edev->pdev : NULL;
}

/* Return values from eeh_ops::next_error */
enum {
EEH_NEXT_ERR_NONE = 0,
EEH_NEXT_ERR_INF,
EEH_NEXT_ERR_FROZEN_PE,
EEH_NEXT_ERR_FENCED_PHB,
EEH_NEXT_ERR_DEAD_PHB,
EEH_NEXT_ERR_DEAD_IOC
};

/*
* The struct is used to trace the registered EEH operation
* callback functions. Actually, those operation callback
Expand Down
150 changes: 78 additions & 72 deletions arch/powerpc/kernel/eeh_driver.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -630,84 +630,90 @@ static void eeh_handle_special_event(void)
{
struct eeh_pe *pe, *phb_pe;
struct pci_bus *bus;
struct pci_controller *hose, *tmp;
struct pci_controller *hose;
unsigned long flags;
int rc = 0;
int rc;

/*
* The return value from next_error() has been classified as follows.
* It might be good to enumerate them. However, next_error() is only
* supported by PowerNV platform for now. So it would be fine to use
* integer directly:
*
* 4 - Dead IOC 3 - Dead PHB
* 2 - Fenced PHB 1 - Frozen PE
* 0 - No error found
*
*/
rc = eeh_ops->next_error(&pe);
if (rc <= 0)
return;

switch (rc) {
case 4:
/* Mark all PHBs in dead state */
eeh_serialize_lock(&flags);
list_for_each_entry_safe(hose, tmp,
&hose_list, list_node) {
phb_pe = eeh_phb_pe_get(hose);
if (!phb_pe) continue;

eeh_pe_state_mark(phb_pe,
EEH_PE_ISOLATED | EEH_PE_PHB_DEAD);
do {
rc = eeh_ops->next_error(&pe);

switch (rc) {
case EEH_NEXT_ERR_DEAD_IOC:
/* Mark all PHBs in dead state */
eeh_serialize_lock(&flags);

/* Purge all events */
eeh_remove_event(NULL);

list_for_each_entry(hose, &hose_list, list_node) {
phb_pe = eeh_phb_pe_get(hose);
if (!phb_pe) continue;

eeh_pe_state_mark(phb_pe,
EEH_PE_ISOLATED | EEH_PE_PHB_DEAD);
}

eeh_serialize_unlock(flags);

break;
case EEH_NEXT_ERR_FROZEN_PE:
case EEH_NEXT_ERR_FENCED_PHB:
case EEH_NEXT_ERR_DEAD_PHB:
/* Mark the PE in fenced state */
eeh_serialize_lock(&flags);

/* Purge all events of the PHB */
eeh_remove_event(pe);

if (rc == EEH_NEXT_ERR_DEAD_PHB)
eeh_pe_state_mark(pe,
EEH_PE_ISOLATED | EEH_PE_PHB_DEAD);
else
eeh_pe_state_mark(pe,
EEH_PE_ISOLATED | EEH_PE_RECOVERING);

eeh_serialize_unlock(flags);

break;
case EEH_NEXT_ERR_NONE:
return;
default:
pr_warn("%s: Invalid value %d from next_error()\n",
__func__, rc);
return;
}
eeh_serialize_unlock(flags);

/* Purge all events */
eeh_remove_event(NULL);
break;
case 3:
case 2:
case 1:
/* Mark the PE in fenced state */
eeh_serialize_lock(&flags);
if (rc == 3)
eeh_pe_state_mark(pe,
EEH_PE_ISOLATED | EEH_PE_PHB_DEAD);
else
eeh_pe_state_mark(pe,
EEH_PE_ISOLATED | EEH_PE_RECOVERING);
eeh_serialize_unlock(flags);

/* Purge all events of the PHB */
eeh_remove_event(pe);
break;
default:
pr_err("%s: Invalid value %d from next_error()\n",
__func__, rc);
return;
}

/*
* For fenced PHB and frozen PE, it's handled as normal
* event. We have to remove the affected PHBs for dead
* PHB and IOC
*/
if (rc == 2 || rc == 1)
eeh_handle_normal_event(pe);
else {
list_for_each_entry_safe(hose, tmp,
&hose_list, list_node) {
phb_pe = eeh_phb_pe_get(hose);
if (!phb_pe || !(phb_pe->state & EEH_PE_PHB_DEAD))
continue;

bus = eeh_pe_bus_get(phb_pe);
/* Notify all devices that they're about to go down. */
eeh_pe_dev_traverse(pe, eeh_report_failure, NULL);
pcibios_remove_pci_devices(bus);
/*
* For fenced PHB and frozen PE, it's handled as normal
* event. We have to remove the affected PHBs for dead
* PHB and IOC
*/
if (rc == EEH_NEXT_ERR_FROZEN_PE ||
rc == EEH_NEXT_ERR_FENCED_PHB) {
eeh_handle_normal_event(pe);
} else {
list_for_each_entry(hose, &hose_list, list_node) {
phb_pe = eeh_phb_pe_get(hose);
if (!phb_pe ||
!(phb_pe->state & EEH_PE_PHB_DEAD))
continue;

/* Notify all devices to be down */
bus = eeh_pe_bus_get(phb_pe);
eeh_pe_dev_traverse(pe,
eeh_report_failure, NULL);
pcibios_remove_pci_devices(bus);
}
}
}

/*
* If we have detected dead IOC, we needn't proceed
* any more since all PHBs would have been removed
*/
if (rc == EEH_NEXT_ERR_DEAD_IOC)
break;
} while (rc != EEH_NEXT_ERR_NONE);
}

/**
Expand Down
39 changes: 24 additions & 15 deletions arch/powerpc/platforms/powernv/eeh-ioda.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -735,12 +735,12 @@ static int ioda_eeh_get_pe(struct pci_controller *hose,
*/
static int ioda_eeh_next_error(struct eeh_pe **pe)
{
struct pci_controller *hose, *tmp;
struct pci_controller *hose;
struct pnv_phb *phb;
u64 frozen_pe_no;
u16 err_type, severity;
long rc;
int ret = 1;
int ret = EEH_NEXT_ERR_NONE;

/*
* While running here, it's safe to purge the event queue.
Expand All @@ -750,7 +750,7 @@ static int ioda_eeh_next_error(struct eeh_pe **pe)
eeh_remove_event(NULL);
opal_notifier_update_evt(OPAL_EVENT_PCI_ERROR, 0x0ul);

list_for_each_entry_safe(hose, tmp, &hose_list, list_node) {
list_for_each_entry(hose, &hose_list, list_node) {
/*
* If the subordinate PCI buses of the PHB has been
* removed, we needn't take care of it any more.
Expand Down Expand Up @@ -789,19 +789,19 @@ static int ioda_eeh_next_error(struct eeh_pe **pe)
switch (err_type) {
case OPAL_EEH_IOC_ERROR:
if (severity == OPAL_EEH_SEV_IOC_DEAD) {
list_for_each_entry_safe(hose, tmp,
&hose_list, list_node) {
list_for_each_entry(hose, &hose_list,
list_node) {
phb = hose->private_data;
phb->eeh_state |= PNV_EEH_STATE_REMOVED;
}

pr_err("EEH: dead IOC detected\n");
ret = 4;
goto out;
ret = EEH_NEXT_ERR_DEAD_IOC;
} else if (severity == OPAL_EEH_SEV_INF) {
pr_info("EEH: IOC informative error "
"detected\n");
ioda_eeh_hub_diag(hose);
ret = EEH_NEXT_ERR_NONE;
}

break;
Expand All @@ -813,21 +813,20 @@ static int ioda_eeh_next_error(struct eeh_pe **pe)
pr_err("EEH: dead PHB#%x detected\n",
hose->global_number);
phb->eeh_state |= PNV_EEH_STATE_REMOVED;
ret = 3;
goto out;
ret = EEH_NEXT_ERR_DEAD_PHB;
} else if (severity == OPAL_EEH_SEV_PHB_FENCED) {
if (ioda_eeh_get_phb_pe(hose, pe))
break;

pr_err("EEH: fenced PHB#%x detected\n",
hose->global_number);
ret = 2;
goto out;
ret = EEH_NEXT_ERR_FENCED_PHB;
} else if (severity == OPAL_EEH_SEV_INF) {
pr_info("EEH: PHB#%x informative error "
"detected\n",
hose->global_number);
ioda_eeh_phb_diag(hose);
ret = EEH_NEXT_ERR_NONE;
}

break;
Expand All @@ -837,13 +836,23 @@ static int ioda_eeh_next_error(struct eeh_pe **pe)

pr_err("EEH: Frozen PE#%x on PHB#%x detected\n",
(*pe)->addr, (*pe)->phb->global_number);
ret = 1;
goto out;
ret = EEH_NEXT_ERR_FROZEN_PE;
break;
default:
pr_warn("%s: Unexpected error type %d\n",
__func__, err_type);
}

/*
* If we have no errors on the specific PHB or only
* informative error there, we continue poking it.
* Otherwise, we need actions to be taken by upper
* layer.
*/
if (ret > EEH_NEXT_ERR_INF)
break;
}

ret = 0;
out:
return ret;
}

Expand Down

0 comments on commit 7e4e786

Please sign in to comment.