[PATCH 1/2] powerpc/eeh: Rework pci_dev lookup in debugfs attributes

[PATCH 1/2] powerpc/eeh: Rework pci_dev lookup in debugfs attributes

[Oliver O'Halloran]

Pull the string -> pci_dev lookup stuff into a helper function. No functional change.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
---
 arch/powerpc/kernel/eeh.c | 71 ++++++++++++++++++++-------------------
 1 file changed, 37 insertions(+), 34 deletions(-)

diff --git a/arch/powerpc/kernel/eeh.c b/arch/powerpc/kernel/eeh.c
index 813713c9120c..f9182ff57804 100644
--- a/arch/powerpc/kernel/eeh.c
+++ b/arch/powerpc/kernel/eeh.c
@@ -1596,6 +1596,35 @@ static int proc_eeh_show(struct seq_file *m, void *v)
 }
 
 #ifdef CONFIG_DEBUG_FS
+
+
+static struct pci_dev *eeh_debug_lookup_pdev(struct file *filp,
+					     const char __user *user_buf,
+					     size_t count, loff_t *ppos)
+{
+	uint32_t domain, bus, dev, fn;
+	struct pci_dev *pdev;
+	char buf[20];
+	int ret;
+
+	memset(buf, 0, sizeof(buf));
+	ret = simple_write_to_buffer(buf, sizeof(buf)-1, ppos, user_buf, count);
+	if (!ret)
+		return ERR_PTR(-EFAULT);
+
+	ret = sscanf(buf, "%x:%x:%x.%x", &domain, &bus, &dev, &fn);
+	if (ret != 4) {
+		pr_err("%s: expected 4 args, got %d\n", __func__, ret);
+		return ERR_PTR(-EINVAL);
+	}
+
+	pdev = pci_get_domain_bus_and_slot(domain, bus, (dev << 3) | fn);
+	if (!pdev)
+		return ERR_PTR(-ENODEV);
+
+	return pdev;
+}
+
 static int eeh_enable_dbgfs_set(void *data, u64 val)
 {
 	if (val)
@@ -1688,26 +1717,13 @@ static ssize_t eeh_dev_check_write(struct file *filp,
 				const char __user *user_buf,
 				size_t count, loff_t *ppos)
 {
-	uint32_t domain, bus, dev, fn;
 	struct pci_dev *pdev;
 	struct eeh_dev *edev;
-	char buf[20];
 	int ret;
 
-	memset(buf, 0, sizeof(buf));
-	ret = simple_write_to_buffer(buf, sizeof(buf)-1, ppos, user_buf, count);
-	if (!ret)
-		return -EFAULT;
-
-	ret = sscanf(buf, "%x:%x:%x.%x", &domain, &bus, &dev, &fn);
-	if (ret != 4) {
-		pr_err("%s: expected 4 args, got %d\n", __func__, ret);
-		return -EINVAL;
-	}
-
-	pdev = pci_get_domain_bus_and_slot(domain, bus, (dev << 3) | fn);
-	if (!pdev)
-		return -ENODEV;
+	pdev = eeh_debug_lookup_pdev(filp, user_buf, count, ppos);
+	if (IS_ERR(pdev))
+		return PTR_ERR(pdev);
 
 	edev = pci_dev_to_eeh_dev(pdev);
 	if (!edev) {
@@ -1717,8 +1733,8 @@ static ssize_t eeh_dev_check_write(struct file *filp,
 	}
 
 	ret = eeh_dev_check_failure(edev);
-	pci_info(pdev, "eeh_dev_check_failure(%04x:%02x:%02x.%01x) = %d\n",
-			domain, bus, dev, fn, ret);
+	pci_info(pdev, "eeh_dev_check_failure(%s) = %d\n",
+			pci_name(pdev), ret);
 
 	pci_dev_put(pdev);
 
@@ -1829,25 +1845,12 @@ static ssize_t eeh_dev_break_write(struct file *filp,
 				const char __user *user_buf,
 				size_t count, loff_t *ppos)
 {
-	uint32_t domain, bus, dev, fn;
 	struct pci_dev *pdev;
-	char buf[20];
 	int ret;
 
-	memset(buf, 0, sizeof(buf));
-	ret = simple_write_to_buffer(buf, sizeof(buf)-1, ppos, user_buf, count);
-	if (!ret)
-		return -EFAULT;
-
-	ret = sscanf(buf, "%x:%x:%x.%x", &domain, &bus, &dev, &fn);
-	if (ret != 4) {
-		pr_err("%s: expected 4 args, got %d\n", __func__, ret);
-		return -EINVAL;
-	}
-
-	pdev = pci_get_domain_bus_and_slot(domain, bus, (dev << 3) | fn);
-	if (!pdev)
-		return -ENODEV;
+	pdev = eeh_debug_lookup_pdev(filp, user_buf, count, ppos);
+	if (IS_ERR(pdev))
+		return PTR_ERR(pdev);
 
 	ret = eeh_debugfs_break_device(pdev);
 	pci_dev_put(pdev);
-- 
2.26.2

[PATCH 2/2] powerpc/eeh: Add a debugfs interface to check if a driver supports recovery

[Oliver O'Halloran]

If a PCI device's current driver implements the error handling callbacks
EEH can use them to recover the device after an error occurs. For devices
without the error handling callbacks we recover them by removing the device
and re-scanning it so the PCI core puts the device back into a known good
state.

Currently there's no way for userspace to determine if the driver supports
recovery or not which makes it difficult to write automated tests for EEH.
This patch addressing that by adding a debugfs interface for querying if
a specific device can be recovered or not.

Signed-off-by: Oliver O'Halloran <oohall@gmail.com>
---
 arch/powerpc/kernel/eeh.c | 50 +++++++++++++++++++++++++++++++++++++++
 1 file changed, 50 insertions(+)

diff --git a/arch/powerpc/kernel/eeh.c b/arch/powerpc/kernel/eeh.c
index f9182ff57804..cd60bc1c8701 100644
--- a/arch/powerpc/kernel/eeh.c
+++ b/arch/powerpc/kernel/eeh.c
@@ -1868,6 +1868,53 @@ static const struct file_operations eeh_dev_break_fops = {
 	.read   = eeh_debugfs_dev_usage,
 };
 
+static ssize_t eeh_dev_can_recover(struct file *filp,
+				   const char __user *user_buf,
+				   size_t count, loff_t *ppos)
+{
+	struct pci_driver *drv;
+	struct pci_dev *pdev;
+	size_t ret;
+
+	pdev = eeh_debug_lookup_pdev(filp, user_buf, count, ppos);
+	if (IS_ERR(pdev))
+		return PTR_ERR(pdev);
+
+	/*
+	 * In order for error recovery to work the driver needs to implement
+	 * .error_detected(), so it can quiesce IO to the device, and
+	 * .slot_reset() so it can re-initialise the device after a reset.
+	 *
+	 * Ideally they'd implement .resume() too, but some drivers which
+	 * we need to support (notably IPR) don't so I guess we can tolerate
+	 * that.
+	 *
+	 * .mmio_enabled() is mostly there as a work-around for devices which
+	 * take forever to re-init after a hot reset. Implementing that is
+	 * strictly optional.
+	 */
+	drv = pci_dev_driver(pdev);
+	if (drv &&
+	    drv->err_handler &&
+	    drv->err_handler->error_detected &&
+	    drv->err_handler->slot_reset) {
+		ret = count;
+	} else {
+		ret = -EOPNOTSUPP;
+	}
+
+	pci_dev_put(pdev);
+
+	return ret;
+}
+
+static const struct file_operations eeh_dev_can_recover_fops = {
+	.open	= simple_open,
+	.llseek	= no_llseek,
+	.write	= eeh_dev_can_recover,
+	.read   = eeh_debugfs_dev_usage,
+};
+
 #endif
 
 static int __init eeh_init_proc(void)
@@ -1892,6 +1939,9 @@ static int __init eeh_init_proc(void)
 		debugfs_create_file_unsafe("eeh_force_recover", 0600,
 				powerpc_debugfs_root, NULL,
 				&eeh_force_recover_fops);
+		debugfs_create_file_unsafe("eeh_dev_can_recover", 0600,
+				powerpc_debugfs_root, NULL,
+				&eeh_dev_can_recover_fops);
 		eeh_cache_debugfs_init();
 #endif
 	}
-- 
2.26.2

Re: [PATCH 1/2] powerpc/eeh: Rework pci_dev lookup in debugfs attributes

[Michael Ellerman]

On Tue, 3 Nov 2020 16:15:11 +1100, Oliver O'Halloran wrote:
> Pull the string -> pci_dev lookup stuff into a helper function. No functional change.

Applied to powerpc/next.

[1/2] powerpc/eeh: Rework pci_dev lookup in debugfs attributes
      https://git.kernel.org/powerpc/c/b5e904b83067bbbd7dc83ea3734c119f8796d79f
[2/2] powerpc/eeh: Add a debugfs interface to check if a driver supports recovery
      https://git.kernel.org/powerpc/c/9e857416833d9701a406ecd6f03a695405ada5e6

cheers