Commit 7e9084b3 authored by Oza Pawandeep's avatar Oza Pawandeep Committed by Bjorn Helgaas

PCI/AER: Handle ERR_FATAL with removal and re-enumeration of devices

PCIe ERR_FATAL errors mean the Link is unreliable.  Components on the Link
may need to be reset to return to reliable operation (PCIe r4.0, sec
6.2.2).  We previously handled these errors much differently depending on
whether the platform supports Downstream Port Containment (DPC) (PCIe r4.0,
sec 6.2.10) or not.

The AER driver has historically logged the error details, called
driver-supplied pci_error_handlers callbacks, and reset the Link.  This
reset downstream devices, but did not remove them from the PCI subsystem,
re-enumerate them, or call their driver .remove() or .probe() methods.

DPC is different because the hardware automatically disables the Link when
it detects ERR_FATAL, which resets downstream devices.  There's no
opportunity for pci_error_handlers callbacks before resetting the Link.
The DPC driver removes affected devices (which calls their driver .remove()
methods), brings the Link back up, and re-enumerates (which calls driver
.probe() methods).

Align AER ERR_FATAL handling with DPC by resetting the Link in software,
skipping the driver pci_error_handlers callbacks, removing the devices from
the PCI subsystem, and re-enumerating.  The idea is that drivers and
devices should see the same behavior for ERR_FATAL events, regardless of
whether they're handled by AER or DPC.

Here are the basic ERR_FATAL recovery steps, showing the previous AER
behavior, the AER behavior after this patch, and the DPC behavior:

                          AER        AER      DPC
                          previous   new      behavior
                          --------   ---      --------
  Log error               yes        yes      yes (minimal)
  drv.error_detected()    yes        no       no
  Reset Link              yes        yes      yes
  drv.mmio_enabled()      yes        no       no
  drv.slot_reset()        yes        no       no
  drv.resume()            yes        no       no
  Remove PCI devices      no         yes      yes
    (calls drv.remove())
  Re-enumerate            no         yes      yes
    (calls drv.probe())

N.B. With DPC, the Link reset happens before the driver .remove() calls,
while with AER, the reset happens *after* the .remove() calls.  The goal is
to eventually do the reset before .remove() for AER as well.
Signed-off-by: default avatarOza Pawandeep <>
[bhelgaas: changelog, squash doc patch into this, remove unused
Signed-off-by: default avatarBjorn Helgaas <>
Reviewed-by: Keith Busch's avatarKeith Busch <>
parent 9f5a70f1
......@@ -110,7 +110,7 @@ The actual steps taken by a platform to recover from a PCI error
event will be platform-dependent, but will follow the general
sequence described below.
STEP 0: Error Event
A PCI bus error is detected by the PCI hardware. On powerpc, the slot
is isolated, in that all I/O is blocked: all reads return 0xffffffff,
......@@ -228,13 +228,7 @@ proceeds to either STEP3 (Link Reset) or to STEP 5 (Resume Operations).
If any driver returned PCI_ERS_RESULT_NEED_RESET, then the platform
proceeds to STEP 4 (Slot Reset)
STEP 3: Link Reset
The platform resets the link. This is a PCI-Express specific step
and is done whenever a fatal error has been detected that can be
"solved" by resetting the link.
STEP 4: Slot Reset
STEP 3: Slot Reset
In response to a return value of PCI_ERS_RESULT_NEED_RESET, the
......@@ -320,7 +314,7 @@ Failure).
>>> However, it probably should.
STEP 5: Resume Operations
STEP 4: Resume Operations
The platform will call the resume() callback on all affected device
drivers if all drivers on the segment have returned
......@@ -332,7 +326,7 @@ a result code.
At this point, if a new error happens, the platform will restart
a new error recovery sequence.
STEP 6: Permanent Failure
STEP 5: Permanent Failure
A "permanent failure" has occurred, and the platform cannot recover
the device. The platform will call error_detected() with a
......@@ -355,6 +349,27 @@ errors. See the discussion in powerpc/eeh-pci-error-recovery.txt
for additional detail on real-life experience of the causes of
software errors.
STEP 0: Error Event: ERR_FATAL
PCI bus error is detected by the PCI hardware. On powerpc, the slot is
isolated, in that all I/O is blocked: all reads return 0xffffffff, all
writes are ignored.
STEP 1: Remove devices
Platform removes the devices depending on the error agent, it could be
this port for all subordinates or upstream component (likely downstream
STEP 2: Reset link
The platform resets the link. This is a PCI-Express specific step and is
done whenever a fatal error has been detected that can be "solved" by
resetting the link.
STEP 3: Re-enumerate the devices
Initiates the re-enumeration.
Conclusion; General Remarks
......@@ -353,10 +353,7 @@ static void aer_error_resume(struct pci_dev *dev)
pos = dev->aer_cap;
pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, &status);
pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_SEVER, &mask);
if (dev->error_state == pci_channel_io_normal)
status &= ~mask; /* Clear corresponding nonfatal bits */
status &= mask; /* Clear corresponding fatal bits */
status &= ~mask; /* Clear corresponding nonfatal bits */
pci_write_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, status);
......@@ -20,6 +20,7 @@
#include <linux/slab.h>
#include <linux/kfifo.h>
#include "aerdrv.h"
#include "../../pci.h"
......@@ -475,35 +476,81 @@ static pci_ers_result_t reset_link(struct pci_dev *dev)
* do_recovery - handle nonfatal/fatal error recovery process
* do_fatal_recovery - handle fatal error recovery process
* @dev: pointer to a pci_dev data structure of agent detecting an error
* @severity: error severity type
* Invoked when an error is nonfatal/fatal. Once being invoked, broadcast
* Invoked when an error is fatal. Once being invoked, removes the devices
* beneath this AER agent, followed by reset link e.g. secondary bus reset
* followed by re-enumeration of devices.
static void do_fatal_recovery(struct pci_dev *dev)
struct pci_dev *udev;
struct pci_bus *parent;
struct pci_dev *pdev, *temp;
pci_ers_result_t result;
if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE)
udev = dev;
udev = dev->bus->self;
parent = udev->subordinate;
list_for_each_entry_safe_reverse(pdev, temp, &parent->devices,
bus_list) {
pci_dev_set_disconnected(pdev, NULL);
if (pci_has_subordinate(pdev))
pci_dev_set_disconnected, NULL);
result = reset_link(udev);
if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
* If the error is reported by a bridge, we think this error
* is related to the downstream link of the bridge, so we
* do error recovery on all subordinates of the bridge instead
* of the bridge and clear the error status of the bridge.
if (pcie_wait_for_link(udev, true))
} else {
pci_uevent_ers(dev, PCI_ERS_RESULT_DISCONNECT);
pci_info(dev, "AER: Device recovery from fatal error failed\n");
* do_nonfatal_recovery - handle nonfatal error recovery process
* @dev: pointer to a pci_dev data structure of agent detecting an error
* Invoked when an error is nonfatal. Once being invoked, broadcast
* error detected message to all downstream drivers within a hierarchy in
* question and return the returned code.
static void do_recovery(struct pci_dev *dev, int severity)
static void do_nonfatal_recovery(struct pci_dev *dev)
pci_ers_result_t status, result = PCI_ERS_RESULT_RECOVERED;
pci_ers_result_t status;
enum pci_channel_state state;
if (severity == AER_FATAL)
state = pci_channel_io_frozen;
state = pci_channel_io_normal;
state = pci_channel_io_normal;
status = broadcast_error_message(dev,
if (severity == AER_FATAL) {
result = reset_link(dev);
goto failed;
status = broadcast_error_message(dev,
......@@ -562,8 +609,10 @@ static void handle_error_source(struct pcie_device *aerdev,
if (pos)
pci_write_config_dword(dev, pos + PCI_ERR_COR_STATUS,
} else
do_recovery(dev, info->severity);
} else if (info->severity == AER_NONFATAL)
else if (info->severity == AER_FATAL)
......@@ -627,8 +676,10 @@ static void aer_recover_work_func(struct work_struct *work)
cper_print_aer(pdev, entry.severity, entry.regs);
if (entry.severity != AER_CORRECTABLE)
do_recovery(pdev, entry.severity);
if (entry.severity == AER_NONFATAL)
else if (entry.severity == AER_FATAL)
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment