<div dir="auto">I doubt "linux/pci: " matches the powerpc convention and I know it doesn't match the drivers/pci convention.<div dir="auto"><br></div><div dir="auto">I'd suggest matching one or the other. In drivers/pci I would be using "PCI/AER: ".</div></div><div class="gmail_extra"><br><div class="gmail_quote">On Jan 5, 2018 10:46 AM, "Bryant G. Ly" <<a href="mailto:bryantly@linux.vnet.ibm.com">bryantly@linux.vnet.ibm.com</a>> wrote:<br type="attribution"><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Devices can go offline when erors reported. This<br>
patch adds a change to the kernel object and lets udev<br>
know of error. When device resumes, a change is also set<br>
reporting device as online. Therefore, EEH and AER events<br>
are better propagated to user space for PCI devices in<br>
all arches.<br>
<br>
Signed-off-by: Bryant G. Ly <<a href="mailto:bryantly@linux.vnet.ibm.com">bryantly@linux.vnet.ibm.com</a>><br>
Signed-off-by: Juan J. Alvarez <<a href="mailto:jjalvare@linux.vnet.ibm.com">jjalvare@linux.vnet.ibm.com</a>><br>
Acked-by: Bjorn Helgaas <<a href="mailto:bhelgaas@google.com">bhelgaas@google.com</a>><br>
---<br>
arch/powerpc/kernel/eeh_<wbr>driver.c | 6 ++++++<br>
drivers/pci/pcie/aer/aerdrv_<wbr>core.c | 3 +++<br>
include/linux/pci.h | 36 ++++++++++++++++++++++++++++++<wbr>++++++<br>
3 files changed, 45 insertions(+)<br>
<br>
diff --git a/arch/powerpc/kernel/eeh_<wbr>driver.c b/arch/powerpc/kernel/eeh_<wbr>driver.c<br>
index 3c0fa99c5533..beea2182d754 100644<br>
--- a/arch/powerpc/kernel/eeh_<wbr>driver.c<br>
+++ b/arch/powerpc/kernel/eeh_<wbr>driver.c<br>
@@ -228,6 +228,7 @@ static void *eeh_report_error(void *data, void *userdata)<br>
<br>
edev->in_error = true;<br>
eeh_pcid_put(dev);<br>
+ pci_uevent_ers(dev, PCI_ERS_RESULT_NONE);<br>
return NULL;<br>
}<br>
<br>
@@ -381,6 +382,10 @@ static void *eeh_report_resume(void *data, void *userdata)<br>
driver->err_handler->resume(<wbr>dev);<br>
<br>
eeh_pcid_put(dev);<br>
+ pci_uevent_ers(dev, PCI_ERS_RESULT_RECOVERED);<br>
+#ifdef CONFIG_PCI_IOV<br>
+ eeh_ops->notify_resume(eeh_<wbr>dev_to_pdn(edev));<br>
+#endif<br>
return NULL;<br>
}<br>
<br>
@@ -416,6 +421,7 @@ static void *eeh_report_failure(void *data, void *userdata)<br>
driver->err_handler->error_<wbr>detected(dev, pci_channel_io_perm_failure);<br>
<br>
eeh_pcid_put(dev);<br>
+ pci_uevent_ers(dev, PCI_ERS_RESULT_DISCONNECT);<br>
return NULL;<br>
}<br>
<br>
diff --git a/drivers/pci/pcie/aer/aerdrv_<wbr>core.c b/drivers/pci/pcie/aer/aerdrv_<wbr>core.c<br>
index 744805232155..8d7448063fd1 100644<br>
--- a/drivers/pci/pcie/aer/aerdrv_<wbr>core.c<br>
+++ b/drivers/pci/pcie/aer/aerdrv_<wbr>core.c<br>
@@ -278,6 +278,7 @@ static int report_error_detected(struct pci_dev *dev, void *data)<br>
} else {<br>
err_handler = dev->driver->err_handler;<br>
vote = err_handler->error_detected(<wbr>dev, result_data->state);<br>
+ pci_uevent_ers(dev, PCI_ERS_RESULT_NONE);<br>
}<br>
<br>
result_data->result = merge_result(result_data-><wbr>result, vote);<br>
@@ -341,6 +342,7 @@ static int report_resume(struct pci_dev *dev, void *data)<br>
<br>
err_handler = dev->driver->err_handler;<br>
err_handler->resume(dev);<br>
+ pci_uevent_ers(dev, PCI_ERS_RESULT_RECOVERED);<br>
out:<br>
device_unlock(&dev->dev);<br>
return 0;<br>
@@ -541,6 +543,7 @@ static void do_recovery(struct pci_dev *dev, int severity)<br>
return;<br>
<br>
failed:<br>
+ pci_uevent_ers(dev, PCI_ERS_RESULT_DISCONNECT);<br>
/* TODO: Should kernel panic here? */<br>
dev_info(&dev->dev, "AER: Device recovery failed\n");<br>
}<br>
diff --git a/include/linux/pci.h b/include/linux/pci.h<br>
index e3e94467687a..405630441b74 100644<br>
--- a/include/linux/pci.h<br>
+++ b/include/linux/pci.h<br>
@@ -2277,6 +2277,42 @@ static inline bool pci_is_thunderbolt_attached(<wbr>struct pci_dev *pdev)<br>
return false;<br>
}<br>
<br>
+/**<br>
+ * pci_uevent_ers - emit a uevent during recovery path of pci device<br>
+ * @pdev: pci device to check<br>
+ * @err_type: type of error event<br>
+ *<br>
+ */<br>
+static inline void pci_uevent_ers(struct pci_dev *pdev,<br>
+ enum pci_ers_result err_type)<br>
+{<br>
+ int idx = 0;<br>
+ char *envp[3];<br>
+<br>
+ switch (err_type) {<br>
+ case PCI_ERS_RESULT_NONE:<br>
+ case PCI_ERS_RESULT_CAN_RECOVER:<br>
+ envp[idx++] = "ERROR_EVENT=BEGIN_RECOVERY";<br>
+ envp[idx++] = "DEVICE_ONLINE=0";<br>
+ break;<br>
+ case PCI_ERS_RESULT_RECOVERED:<br>
+ envp[idx++] = "ERROR_EVENT=SUCCESSFUL_<wbr>RECOVERY";<br>
+ envp[idx++] = "DEVICE_ONLINE=1";<br>
+ break;<br>
+ case PCI_ERS_RESULT_DISCONNECT:<br>
+ envp[idx++] = "ERROR_EVENT=FAILED_RECOVERY";<br>
+ envp[idx++] = "DEVICE_ONLINE=0";<br>
+ break;<br>
+ default:<br>
+ break;<br>
+ }<br>
+<br>
+ if (idx > 0) {<br>
+ envp[idx++] = NULL;<br>
+ kobject_uevent_env(&pdev->dev.<wbr>kobj, KOBJ_CHANGE, envp);<br>
+ }<br>
+}<br>
+<br>
/* provide the legacy pci_dma_* API */<br>
#include <linux/pci-dma-compat.h><br>
<br>
--<br>
2.14.3 (Apple Git-98)<br>
<br>
</blockquote></div></div>