s390/pci: implement minimal PCI error recovery

When the platform detects an error on a PCI function or a service action
has been performed it is put in the error state and an error event
notification is provided to the OS.

Currently we treat all error event notifications the same and simply set
pdev->error_state = pci_channel_io_perm_failure requiring user
intervention such as use of the recover attribute to get the device
usable again. Despite requiring a manual step this also has the
disadvantage that the device is completely torn down and recreated
resulting in higher level devices such as a block or network device
being recreated. In case of a block device this also means that it may
need to be removed and added to a software raid even if that could
otherwise survive with a temporary degradation.

This is of course not ideal more so since an error notification with PEC
0x3A indicates that the platform already performed error recovery
successfully or that the error state was caused by a service action that
is now finished.

At least in this case we can assume that the error state can be reset
and the function made usable again. So as not to have the disadvantage
of a full tear down and recreation we need to coordinate this recovery
with the driver. Thankfully there is already a well defined recovery
flow for this described in Documentation/PCI/pci-error-recovery.rst.

The implementation of this is somewhat straight forward and simplified
by the fact that our recovery flow is defined per PCI function. As
a reset we use the newly introduced zpci_hot_reset_device() which also
takes the PCI function out of the error state.

Reviewed-by: Pierre Morel <pmorel@linux.ibm.com>
Acked-by: Matthew Rosato <mjrosato@linux.ibm.com>
Signed-off-by: Niklas Schnelle <schnelle@linux.ibm.com>
Signed-off-by: Vasily Gorbik <gor@linux.ibm.com>

1 files changed, 53 insertions, 0 deletions

diff --git a/arch/s390/pci/pci.c b/arch/s390/pci/pci.c
index e95b08e5cc5b..2f9b78fa82a5 100644
--- a/arch/s390/pci/pci.c
+++ b/arch/s390/pci/pci.c
@@ -990,6 +990,59 @@ int zpci_report_error(struct pci_dev *pdev,
 }
 EXPORT_SYMBOL(zpci_report_error);
 
+/**
+ * zpci_clear_error_state() - Clears the zPCI error state of the device
+ * @zdev: The zdev for which the zPCI error state should be reset
+ *
+ * Clear the zPCI error state of the device. If clearing the zPCI error state
+ * fails the device is left in the error state. In this case it may make sense
+ * to call zpci_io_perm_failure() on the associated pdev if it exists.
+ *
+ * Returns: 0 on success, -EIO otherwise
+ */
+int zpci_clear_error_state(struct zpci_dev *zdev)
+{
+	u64 req = ZPCI_CREATE_REQ(zdev->fh, 0, ZPCI_MOD_FC_RESET_ERROR);
+	struct zpci_fib fib = {0};
+	u8 status;
+	int cc;
+
+	cc = zpci_mod_fc(req, &fib, &status);
+	if (cc) {
+		zpci_dbg(3, "ces fid:%x, cc:%d, status:%x\n", zdev->fid, cc, status);
+		return -EIO;
+	}
+
+	return 0;
+}
+
+/**
+ * zpci_reset_load_store_blocked() - Re-enables L/S from error state
+ * @zdev: The zdev for which to unblock load/store access
+ *
+ * Re-enables load/store access for a PCI function in the error state while
+ * keeping DMA blocked. In this state drivers can poke MMIO space to determine
+ * if error recovery is possible while catching any rogue DMA access from the
+ * device.
+ *
+ * Returns: 0 on success, -EIO otherwise
+ */
+int zpci_reset_load_store_blocked(struct zpci_dev *zdev)
+{
+	u64 req = ZPCI_CREATE_REQ(zdev->fh, 0, ZPCI_MOD_FC_RESET_BLOCK);
+	struct zpci_fib fib = {0};
+	u8 status;
+	int cc;
+
+	cc = zpci_mod_fc(req, &fib, &status);
+	if (cc) {
+		zpci_dbg(3, "rls fid:%x, cc:%d, status:%x\n", zdev->fid, cc, status);
+		return -EIO;
+	}
+
+	return 0;
+}
+
 static int zpci_mem_init(void)
 {
 	BUILD_BUG_ON(!is_power_of_2(__alignof__(struct zpci_fmb)) ||