Skip to contentSkip to conversation

ACPI: Device objects for all namespace nodes and PCI root hotplug integration

PATCH In review Rafael J. Wysocki 4 participants 44 messages Nov 17, 2013 – Dec 29, 2013

Subsystems (6)
Documentation ACPI PCI Driver core +2 more subsystems
lore
Share & tools
lore
#0 lore raw

cover ACPI: Device objects for all namespace nodes and PCI root hotplug integration

Hi,

This is 3.14 material, but I'm posting it today to let you know it is coming. Of course, if you see any immediate problems with the following patches, please let me know.

Patch [1/10] is a place holder until we get an equivalent from ACPICA upstream, which should happen early after 3.13-rc1. The remaining patches are:

[2/10] ACPI / scan: Define non-empty removal handler to passed to
acpi_attach_data() for struct acpi_device objects.

[3/10] ACPI / scan: Create struct acpi_device objects for all device nodes in
the namespace regardless of the current result of _STA (the reasons are explained in the changelog).

[4/10] ACPI / hotplug: Do not fail bus check and device check notifications
if hotplug is not enabled for the given group of devices via sysfs (failing those notifications only skips device enumeration, but doesn't
prevent devices from being physically added to the system).

[5/10] ACPI / hotplug: Introduce common device hotplug function
acpi_device_hotplug() and reduce some code duplication.

[6/10] ACPI / hotplug: Make PCI host bridge hotplug use common code (instead
of defining its own notify handler).

[7/10] ACPI / hotplug: Move container-specific code out of the common code.

[8/10] ACPI / hotplug: Rework the common device hotplug code to cover surprise
removals of devices.

[9/10] ACPI / hotplug: Drop unfinished global hotplug notification handling
routines.

[10/10] ACPI: Clean up some revoltingly ugly stuff.

Thanks!

Signature · 2 lines
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.
#1 lore raw

PATCH 1/10 ACPICA: Delete all attached data objects on node deletion

From: Rafael J. Wysocki <[email protected]>

Since many data objects may be attached to a single namespace node, acpi_ns_delete_node() should take that into account and delete all of those objects along with the namespace node itself, calling deletion handlers for all of them in the process. Make that happen.

---
 drivers/acpi/acpica/nsalloc.c |    6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
drivers/acpi/acpica/nsalloc.c +5 -1
Index: linux-pm/drivers/acpi/acpica/nsalloc.c
===================================================================
@@ -106,6 +106,7 @@ struct acpi_namespace_node *acpi_ns_crea
 void acpi_ns_delete_node(struct acpi_namespace_node *node)
 {
 	union acpi_operand_object *obj_desc;
+	union acpi_operand_object *next_obj_desc;
 
 	ACPI_FUNCTION_NAME(ns_delete_node);
 
@@ -119,7 +120,8 @@ void acpi_ns_delete_node(struct acpi_nam
 	 * detached above, the only possible remaining object is a data object.
 	 */
 	obj_desc = node->object;
-	if (obj_desc && (obj_desc->common.type == ACPI_TYPE_LOCAL_DATA)) {
+	while (obj_desc && obj_desc->common.type == ACPI_TYPE_LOCAL_DATA) {
+		next_obj_desc = obj_desc->common.next_object;
 
 		/* Invoke the attached data deletion handler if present */
 
@@ -128,6 +130,8 @@ void acpi_ns_delete_node(struct acpi_nam
 		}
 
 		acpi_ut_remove_reference(obj_desc);
+
+		obj_desc = next_obj_desc;
 	}
 
 	/* Now we can delete the node */
#2 lore raw

PATCH 2/10 ACPI / scan: Define non-empty device removal handler

In reply to Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

If an ACPI namespace node is removed (usually, as a result of a table unload), and there is a data object attached to that node, acpi_ns_delete_node() executes the removal handler submitted to acpi_attach_data() for that object. That handler is currently empty for struct acpi_device objects, so it is necessary to detach those objects from the corresponding ACPI namespace nodes in advance every time a table unload may happen. That is cumbersome and inefficient and leads to some design constraints that turn out to be quite inconvenient (in particular, struct acpi_device objects cannot be registered for namespace nodes representing devices that are not reported as present or functional by _STA).

For this reason, introduce a non-empty removal handler for ACPI device objects that will unregister them when their ACPI namespace nodes go away.

This code modification alone should not change functionality except for the ordering of the ACPI hotplug workqueue which should not matter (without subsequent code changes).

---
 drivers/acpi/internal.h |    2 
 drivers/acpi/osl.c      |    6 ++
 drivers/acpi/scan.c     |  114 +++++++++++++++++++++++++++++++++++++++---------
 include/acpi/acpi_bus.h |    1 
 4 files changed, 102 insertions(+), 21 deletions(-)
drivers/acpi/scan.c +94 -20
Index: linux-pm/drivers/acpi/scan.c
===================================================================
@@ -14,6 +14,8 @@
 
 #include <acpi/acpi_drivers.h>
 
+#include <asm/pgtable.h>
+
 #include "internal.h"
 
 #define _COMPONENT		ACPI_BUS_COMPONENT
@@ -27,6 +29,8 @@ extern struct acpi_device *acpi_root;
 
 #define ACPI_IS_ROOT_DEVICE(device)    (!(device)->parent)
 
+#define INVALID_ACPI_HANDLE	((acpi_handle)empty_zero_page)
+
 /*
  * If set, devices will be hot-removed even if they cannot be put offline
  * gracefully (from the kernel's standpoint).
@@ -907,9 +911,91 @@ struct bus_type acpi_bus_type = {
 	.uevent		= acpi_device_uevent,
 };
 
-static void acpi_bus_data_handler(acpi_handle handle, void *context)
+static void acpi_device_del(struct acpi_device *device)
+{
+	mutex_lock(&acpi_device_lock);
+	if (device->parent)
+		list_del(&device->node);
+
+	list_del(&device->wakeup_list);
+	mutex_unlock(&acpi_device_lock);
+
+	acpi_power_add_remove_device(device, false);
+	acpi_device_remove_files(device);
+	if (device->remove)
+		device->remove(device);
+
+	device_del(&device->dev);
+}
+
+static LIST_HEAD(acpi_device_del_list);
+static DEFINE_MUTEX(acpi_device_del_lock);
+
+static void acpi_device_del_work_fn(struct work_struct *work_not_used)
+{
+	for (;;) {
+		struct acpi_device *adev;
+
+		mutex_lock(&acpi_device_del_lock);
+
+		if (list_empty(&acpi_device_del_list)) {
+			mutex_unlock(&acpi_device_del_lock);
+			break;
+		}
+		adev = list_first_entry(&acpi_device_del_list,
+					struct acpi_device, del_list);
+		list_del(&adev->del_list);
+
+		mutex_unlock(&acpi_device_del_lock);
+
+		acpi_device_del(adev);
+		/*
+		 * Drop references to all power resources that might have been
+		 * used by the device.
+		 */
+		acpi_power_transition(adev, ACPI_STATE_D3_COLD);
+		put_device(&adev->dev);
+	}
+}
+
+/**
+ * acpi_scan_drop_device - Drop an ACPI device object.
+ * @handle: Handle of an ACPI namespace node, not used.
+ * @context: Address of the ACPI device object to drop.
+ *
+ * This is invoked by acpi_ns_delete_node() during the removal of the ACPI
+ * namespace node the device object pointed to by @context is attached to.
+ *
+ * The unregistration is carried out asynchronously to avoid running
+ * acpi_device_del() under the ACPICA's namespace mutex and the list is used to
+ * ensure the correct ordering (the device objects must be unregistered in the
+ * same order in which the corresponding namespace nodes are deleted).
+ */
+static void acpi_scan_drop_device(acpi_handle handle, void *context)
 {
-	/* Intentionally empty. */
+	static DECLARE_WORK(work, acpi_device_del_work_fn);
+	struct acpi_device *adev = context;
+
+	mutex_lock(&acpi_device_del_lock);
+
+	/*
+	 * Use the ACPI hotplug workqueue which is ordered, so this work item
+	 * won't run after any hotplug work items submitted subsequently.  That
+	 * prevents attempts to register device objects identical to those being
+	 * deleted from happening concurrently (such attempts result from
+	 * hotplug events handled via the ACPI hotplug workqueue).  It also will
+	 * run after all of the work items submitted previosuly, which helps
+	 * those work items to ensure that they are not accessing stale device
+	 * objects.
+	 */
+	if (list_empty(&acpi_device_del_list))
+		acpi_queue_hotplug_work(&work);
+
+	list_add_tail(&adev->del_list, &acpi_device_del_list);
+	/* Make acpi_ns_validate_handle() return NULL for this handle. */
+	adev->handle = INVALID_ACPI_HANDLE;
+
+	mutex_unlock(&acpi_device_del_lock);
 }
 
 int acpi_bus_get_device(acpi_handle handle, struct acpi_device **device)
@@ -919,7 +1005,7 @@ int acpi_bus_get_device(acpi_handle hand
 	if (!device)
 		return -EINVAL;
 
-	status = acpi_get_data(handle, acpi_bus_data_handler, (void **)device);
+	status = acpi_get_data(handle, acpi_scan_drop_device, (void **)device);
 	if (ACPI_FAILURE(status) || !*device) {
 		ACPI_DEBUG_PRINT((ACPI_DB_INFO, "No context for object [%p]\n",
 				  handle));
@@ -939,7 +1025,7 @@ int acpi_device_add(struct acpi_device *
 	if (device->handle) {
 		acpi_status status;
 
-		status = acpi_attach_data(device->handle, acpi_bus_data_handler,
+		status = acpi_attach_data(device->handle, acpi_scan_drop_device,
 					  device);
 		if (ACPI_FAILURE(status)) {
 			acpi_handle_err(device->handle,
@@ -957,6 +1043,7 @@ int acpi_device_add(struct acpi_device *
 	INIT_LIST_HEAD(&device->node);
 	INIT_LIST_HEAD(&device->wakeup_list);
 	INIT_LIST_HEAD(&device->physical_node_list);
+	INIT_LIST_HEAD(&device->del_list);
 	mutex_init(&device->physical_node_lock);
 
 	new_bus_id = kzalloc(sizeof(struct acpi_device_bus_id), GFP_KERNEL);
@@ -1020,27 +1107,14 @@ int acpi_device_add(struct acpi_device *
 	mutex_unlock(&acpi_device_lock);
 
  err_detach:
-	acpi_detach_data(device->handle, acpi_bus_data_handler);
+	acpi_detach_data(device->handle, acpi_scan_drop_device);
 	return result;
 }
 
 static void acpi_device_unregister(struct acpi_device *device)
 {
-	mutex_lock(&acpi_device_lock);
-	if (device->parent)
-		list_del(&device->node);
-
-	list_del(&device->wakeup_list);
-	mutex_unlock(&acpi_device_lock);
-
-	acpi_detach_data(device->handle, acpi_bus_data_handler);
-
-	acpi_power_add_remove_device(device, false);
-	acpi_device_remove_files(device);
-	if (device->remove)
-		device->remove(device);
-
-	device_del(&device->dev);
+	acpi_detach_data(device->handle, acpi_scan_drop_device);
+	acpi_device_del(device);
 	/*
 	 * Transition the device to D3cold to drop the reference counts of all
 	 * power resources the device depends on and turn off the ones that have
include/acpi/acpi_bus.h +1 -0
Index: linux-pm/include/acpi/acpi_bus.h
===================================================================
@@ -298,6 +298,7 @@ struct acpi_device {
 	struct list_head children;
 	struct list_head node;
 	struct list_head wakeup_list;
+	struct list_head del_list;
 	struct acpi_device_status status;
 	struct acpi_device_flags flags;
 	struct acpi_device_pnp pnp;
drivers/acpi/osl.c +5 -1
Index: linux-pm/drivers/acpi/osl.c
===================================================================
@@ -1215,6 +1215,10 @@ acpi_status acpi_hotplug_execute(acpi_hp
 	return AE_OK;
 }
 
+bool acpi_queue_hotplug_work(struct work_struct *work)
+{
+	return queue_work(kacpi_hotplug_wq, work);
+}
 
 acpi_status
 acpi_os_create_semaphore(u32 max_units, u32 initial_units, acpi_handle * handle)
@@ -1794,7 +1798,7 @@ acpi_status __init acpi_os_initialize1(v
 {
 	kacpid_wq = alloc_workqueue("kacpid", 0, 1);
 	kacpi_notify_wq = alloc_workqueue("kacpi_notify", 0, 1);
-	kacpi_hotplug_wq = alloc_workqueue("kacpi_hotplug", 0, 1);
+	kacpi_hotplug_wq = alloc_ordered_workqueue("kacpi_hotplug", 0);
 	BUG_ON(!kacpid_wq);
 	BUG_ON(!kacpi_notify_wq);
 	BUG_ON(!kacpi_hotplug_wq);
drivers/acpi/internal.h +2 -0
Index: linux-pm/drivers/acpi/internal.h
===================================================================
@@ -73,6 +73,8 @@ void acpi_lpss_init(void);
 static inline void acpi_lpss_init(void) {}
 #endif
 
+bool acpi_queue_hotplug_work(struct work_struct *work);
+
 /* --------------------------------------------------------------------------
                      Device Node Initialization / Removal
    -------------------------------------------------------------------------- */
#3 lore raw

PATCH 3/10 ACPI / scan: Add acpi_device objects for all device nodes in the namespace

In reply to Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Modify the ACPI namespace scanning code to register a struct acpi_device object for every namespace node representing a device, processor and so on, even if the device represented by that namespace node is reported to be not present and not functional by _STA.

There are multiple reasons to do that. First of all, it avoids quite a lot of overhead when struct acpi_device objects are deleted every time acpi_bus_trim() is run and then added again by a subsequent acpi_bus_scan() for the same scope, although the namespace objects they correspond to stay in memory all the time (which always is the case on a vast majority of systems).

Second, it will allow user space to see that there are namespace nodes representing devices that are not present at the moment and may be added to the system. It will also allow user space to evaluate _SUN for those nodes to check what physical slots the "missing" devices may be put into and it will make sense to add a sysfs attribute for _STA evaluation after this change (that will be useful for thermal management on some systems).

Next, it will help to consolidate the ACPI hotplug handling among subsystems by making it possible to store hotplug-related information in struct acpi_device objects in a standard common way.

Finally, it will help to avoid a race condition related to the deletion of ACPI namespace nodes. Namely, namespace nodes may be deleted as a result of a table unload triggered by _EJ0 or _DCK. If a hotplug notification for one of those nodes is triggered right before the deletion and it executes a hotplug callback via acpi_hotplug_execute(), the ACPI handle passed to that callback may be stale when the callback actually runs. One way to work around that is to always pass struct acpi_device pointers to hotplug callbacks after doing a get_device() on the objects in question which eliminates the use-after-free possibility (the ACPI handles in those objects are invalidated by acpi_scan_drop_device(), so they will trigger ACPICA errors on attempts to use them).

---
 Documentation/acpi/namespace.txt   |   10 ---
 drivers/acpi/device_pm.c           |   22 ++++++-
 drivers/acpi/dock.c                |    9 +--
 drivers/acpi/internal.h            |    3 +
 drivers/acpi/pci_root.c            |    5 +
 drivers/acpi/scan.c                |  107 +++++++++++++++++--------------------
 drivers/pci/hotplug/acpiphp_glue.c |    2 
 drivers/xen/xen-acpi-cpuhotplug.c  |    8 +-
 drivers/xen/xen-acpi-memhotplug.c  |    7 +-
 include/acpi/acpi_bus.h            |    9 ++-
 10 files changed, 98 insertions(+), 84 deletions(-)
include/acpi/acpi_bus.h +8 -1
Index: linux-pm/include/acpi/acpi_bus.h
===================================================================
@@ -168,7 +168,9 @@ struct acpi_device_flags {
 	u32 ejectable:1;
 	u32 power_manageable:1;
 	u32 match_driver:1;
-	u32 reserved:27;
+	u32 initialized:1;
+	u32 visited:1;
+	u32 reserved:25;
 };
 
 /* File System */
@@ -385,6 +387,11 @@ int acpi_match_device_ids(struct acpi_de
 int acpi_create_dir(struct acpi_device *);
 void acpi_remove_dir(struct acpi_device *);
 
+static inline bool acpi_device_enumerated(struct acpi_device *adev)
+{
+	return adev && adev->flags.initialized && adev->flags.visited;
+}
+
 typedef void (*acpi_hp_callback)(void *data, u32 src);
 
 acpi_status acpi_hotplug_execute(acpi_hp_callback func, void *data, u32 src);
drivers/acpi/scan.c +51 -56

205 lines not on this page · raw message

drivers/acpi/device_pm.c +18 -4
Index: linux-pm/drivers/acpi/device_pm.c
===================================================================
@@ -256,6 +256,8 @@ int acpi_bus_init_power(struct acpi_devi
 		return -EINVAL;
 
 	device->power.state = ACPI_STATE_UNKNOWN;
+	if (!acpi_device_is_present(device))
+		return 0;
 
 	result = acpi_device_get_power(device, &state);
 	if (result)
@@ -302,15 +304,18 @@ int acpi_device_fix_up_power(struct acpi
 	return ret;
 }
 
-int acpi_bus_update_power(acpi_handle handle, int *state_p)
+int acpi_device_update_power(struct acpi_device *device, int *state_p)
 {
-	struct acpi_device *device;
 	int state;
 	int result;
 
-	result = acpi_bus_get_device(handle, &device);
-	if (result)
+	if (device->power.state == ACPI_STATE_UNKNOWN) {
+		result = acpi_bus_init_power(device);
+		if (!result && state_p)
+			*state_p = device->power.state;
+
 		return result;
+	}
 
 	result = acpi_device_get_power(device, &state);
 	if (result)
@@ -338,6 +343,15 @@ int acpi_bus_update_power(acpi_handle ha
 
 	return 0;
 }
+
+int acpi_bus_update_power(acpi_handle handle, int *state_p)
+{
+	struct acpi_device *device;
+	int result;
+
+	result = acpi_bus_get_device(handle, &device);
+	return result ? result : acpi_device_update_power(device, state_p);
+}
 EXPORT_SYMBOL_GPL(acpi_bus_update_power);
 
 bool acpi_bus_power_manageable(acpi_handle handle)
drivers/acpi/internal.h +3 -0
Index: linux-pm/drivers/acpi/internal.h
===================================================================
@@ -90,6 +90,7 @@ void acpi_free_pnp_ids(struct acpi_devic
 int acpi_bind_one(struct device *dev, acpi_handle handle);
 int acpi_unbind_one(struct device *dev);
 void acpi_bus_device_eject(void *data, u32 ost_src);
+bool acpi_device_is_present(struct acpi_device *adev);
 
 /* --------------------------------------------------------------------------
                                   Power Resource
@@ -107,6 +108,8 @@ int acpi_power_get_inferred_state(struct
 int acpi_power_on_resources(struct acpi_device *device, int state);
 int acpi_power_transition(struct acpi_device *device, int state);
 
+int acpi_device_update_power(struct acpi_device *device, int *state_p);
+
 int acpi_wakeup_device_init(void);
 void acpi_early_processor_set_pdc(void);
 
drivers/pci/hotplug/acpiphp_glue.c +1 -1
Index: linux-pm/drivers/pci/hotplug/acpiphp_glue.c
===================================================================
@@ -489,7 +489,7 @@ static void acpiphp_bus_add(acpi_handle
 
 	acpi_bus_scan(handle);
 	acpi_bus_get_device(handle, &adev);
-	if (adev)
+	if (acpi_device_enumerated(adev))
 		acpi_device_set_power(adev, ACPI_STATE_D0);
 }
 
drivers/acpi/dock.c +3 -6
Index: linux-pm/drivers/acpi/dock.c
===================================================================
@@ -323,14 +323,11 @@ static int dock_present(struct dock_stat
  */
 static void dock_create_acpi_device(acpi_handle handle)
 {
-	struct acpi_device *device;
+	struct acpi_device *device = NULL;
 	int ret;
 
-	if (acpi_bus_get_device(handle, &device)) {
-		/*
-		 * no device created for this object,
-		 * so we should create one.
-		 */
+	acpi_bus_get_device(handle, &device);
+	if (!acpi_device_enumerated(device)) {
 		ret = acpi_bus_scan(handle);
 		if (ret)
 			pr_debug("error adding bus, %x\n", -ret);
drivers/acpi/pci_root.c +3 -2
Index: linux-pm/drivers/acpi/pci_root.c
===================================================================
@@ -631,9 +631,10 @@ void __init acpi_pci_root_init(void)
 
 static void handle_root_bridge_insertion(acpi_handle handle)
 {
-	struct acpi_device *device;
+	struct acpi_device *device = NULL;
 
-	if (!acpi_bus_get_device(handle, &device)) {
+	acpi_bus_get_device(handle, &device);
+	if (acpi_device_enumerated(device)) {
 		dev_printk(KERN_DEBUG, &device->dev,
 			   "acpi device already exists; ignoring notify\n");
 		return;
drivers/xen/xen-acpi-cpuhotplug.c +5 -3
Index: linux-pm/drivers/xen/xen-acpi-cpuhotplug.c
===================================================================
@@ -269,7 +269,8 @@ static void acpi_processor_hotplug_notif
 		if (!is_processor_present(handle))
 			break;
 
-		if (!acpi_bus_get_device(handle, &device))
+		acpi_bus_get_device(handle, &device);
+		if (acpi_device_enumerated(device))
 			break;
 
 		result = acpi_bus_scan(handle);
@@ -277,8 +278,9 @@ static void acpi_processor_hotplug_notif
 			pr_err(PREFIX "Unable to add the device\n");
 			break;
 		}
-		result = acpi_bus_get_device(handle, &device);
-		if (result) {
+		device = NULL;
+		acpi_bus_get_device(handle, &device);
+		if (!acpi_device_enumerated(device)) {
 			pr_err(PREFIX "Missing device object\n");
 			break;
 		}
drivers/xen/xen-acpi-memhotplug.c +4 -3
Index: linux-pm/drivers/xen/xen-acpi-memhotplug.c
===================================================================
@@ -169,7 +169,7 @@ static int acpi_memory_get_device(acpi_h
 	acpi_scan_lock_acquire();
 
 	acpi_bus_get_device(handle, &device);
-	if (device)
+	if (acpi_device_enumerated(device))
 		goto end;
 
 	/*
@@ -182,8 +182,9 @@ static int acpi_memory_get_device(acpi_h
 		result = -EINVAL;
 		goto out;
 	}
-	result = acpi_bus_get_device(handle, &device);
-	if (result) {
+	device = NULL;
+	acpi_bus_get_device(handle, &device);
+	if (!acpi_device_enumerated(device)) {
 		pr_warn(PREFIX "Missing device object\n");
 		result = -EINVAL;
 		goto out;
Documentation/acpi/namespace.txt +2 -8
Index: linux-pm/Documentation/acpi/namespace.txt
===================================================================
@@ -235,10 +235,6 @@ Wysocki <[email protected]>.
       named object's type in the second column).  In that case the object's
       directory in sysfs will contain the 'path' attribute whose value is
       the full path to the node from the namespace root.
-      struct acpi_device objects are created for the ACPI namespace nodes
-      whose _STA control methods return PRESENT or FUNCTIONING.  The power
-      resource nodes or nodes without _STA are assumed to be both PRESENT
-      and FUNCTIONING.
    F:
       The struct acpi_device object is created for a fixed hardware
       feature (as indicated by the fixed feature flag's name in the second
@@ -340,7 +336,7 @@ Wysocki <[email protected]>.
      | +-------------+-------+----------------+
      |   |
      |   | +- - - - - - - +- - - - - - +- - - - - - - -+
-     |   +-| * PNP0C0D:00 | \_SB_.LID0 | acpi:PNP0C0D: |
+     |   +-| PNP0C0D:00 | \_SB_.LID0 | acpi:PNP0C0D: |
      |   | +- - - - - - - +- - - - - - +- - - - - - - -+
      |   |
      |   | +------------+------------+-----------------------+
@@ -390,6 +386,4 @@ Wysocki <[email protected]>.
             attribute (as described earlier in this document).
    NOTE: N/A indicates the device object does not have the 'path' or the
          'modalias' attribute.
-   NOTE: The PNP0C0D device listed above is highlighted (marked by "*")
-         to indicate it will be created only when its _STA methods return
-         PRESENT or FUNCTIONING.
+
#4 lore raw

PATCH 4/10 ACPI / hotplug: Do not fail bus and device checks for disabled hotplug

In reply to Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

If the scan handler for the given device has hotplug.enabled unset, it doesn't really make sense to fail bus check and device check notifications.

First, bus check may not have anything to do with the device it is signaled for, but it may concern another device on the bus below this one. For this reason, bus check notifications should not be failed if hotplug is disabled for the target device.

Second, device check notifications are signaled only after a device has already appeared (or disappeared), so failing it can only prevent scan handlers and drivers from attaching to that (already existing) device, which is not very useful.

Consequently, if device hotplug is disabled through the device's scan handler, fail eject request notifications only.

---
 drivers/acpi/scan.c |   40 +++++++---------------------------------
 1 file changed, 7 insertions(+), 33 deletions(-)
drivers/acpi/scan.c +7 -33
Index: linux-pm/drivers/acpi/scan.c
===================================================================
@@ -363,43 +363,13 @@ static void acpi_scan_bus_device_check(v
 	unlock_device_hotplug();
 }
 
-static void acpi_hotplug_unsupported(acpi_handle handle, u32 type)
-{
-	u32 ost_status;
-
-	switch (type) {
-	case ACPI_NOTIFY_BUS_CHECK:
-		acpi_handle_debug(handle,
-			"ACPI_NOTIFY_BUS_CHECK event: unsupported\n");
-		ost_status = ACPI_OST_SC_INSERT_NOT_SUPPORTED;
-		break;
-	case ACPI_NOTIFY_DEVICE_CHECK:
-		acpi_handle_debug(handle,
-			"ACPI_NOTIFY_DEVICE_CHECK event: unsupported\n");
-		ost_status = ACPI_OST_SC_INSERT_NOT_SUPPORTED;
-		break;
-	case ACPI_NOTIFY_EJECT_REQUEST:
-		acpi_handle_debug(handle,
-			"ACPI_NOTIFY_EJECT_REQUEST event: unsupported\n");
-		ost_status = ACPI_OST_SC_EJECT_NOT_SUPPORTED;
-		break;
-	default:
-		/* non-hotplug event; possibly handled by other handler */
-		return;
-	}
-
-	acpi_evaluate_hotplug_ost(handle, type, ost_status, NULL);
-}
-
 static void acpi_hotplug_notify_cb(acpi_handle handle, u32 type, void *data)
 {
+	u32 ost_code = ACPI_OST_SC_NON_SPECIFIC_FAILURE;
 	struct acpi_scan_handler *handler = data;
 	struct acpi_device *adev;
 	acpi_status status;
 
-	if (!handler->hotplug.enabled)
-		return acpi_hotplug_unsupported(handle, type);
-
 	switch (type) {
 	case ACPI_NOTIFY_BUS_CHECK:
 		acpi_handle_debug(handle, "ACPI_NOTIFY_BUS_CHECK event\n");
@@ -409,6 +379,11 @@ static void acpi_hotplug_notify_cb(acpi_
 		break;
 	case ACPI_NOTIFY_EJECT_REQUEST:
 		acpi_handle_debug(handle, "ACPI_NOTIFY_EJECT_REQUEST event\n");
+		if (!handler->hotplug.enabled) {
+			acpi_handle_err(handle, "Eject disabled\n");
+			ost_code = ACPI_OST_SC_EJECT_NOT_SUPPORTED;
+			goto err_out;
+		}
 		if (acpi_bus_get_device(handle, &adev))
 			goto err_out;
 
@@ -428,8 +403,7 @@ static void acpi_hotplug_notify_cb(acpi_
 		return;
 
  err_out:
-	acpi_evaluate_hotplug_ost(handle, type,
-				  ACPI_OST_SC_NON_SPECIFIC_FAILURE, NULL);
+	acpi_evaluate_hotplug_ost(handle, type, ost_code, NULL);
 }
 
 static ssize_t real_power_state_show(struct device *dev,
#5 lore raw

PATCH 5/10 ACPI / hotplug: Introduce common hotplug function acpi_device_hotplug()

In reply to Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Modify the common ACPI device hotplug code to always queue up the same function, acpi_device_hotplug(), using acpi_hotplug_execute() and make the PCI host bridge hotplug code use that function too for device hot removal.

This allows some code duplication to be reduced and a race condition where the relevant ACPI handle may become invalid between the notification handler and the function queued up by it via acpi_hotplug_execute() to be avoided.

---
 drivers/acpi/internal.h |    2 
 drivers/acpi/pci_root.c |    4 +
 drivers/acpi/scan.c     |  141 +++++++++++++++++++++---------------------------
 3 files changed, 68 insertions(+), 79 deletions(-)
drivers/acpi/internal.h +1 -1
Index: linux-pm/drivers/acpi/internal.h
===================================================================
@@ -89,7 +89,7 @@ void acpi_device_add_finalize(struct acp
 void acpi_free_pnp_ids(struct acpi_device_pnp *pnp);
 int acpi_bind_one(struct device *dev, acpi_handle handle);
 int acpi_unbind_one(struct device *dev);
-void acpi_bus_device_eject(void *data, u32 ost_src);
+void acpi_device_hotplug(void *data, u32 ost_src);
 bool acpi_device_is_present(struct acpi_device *adev);
 
 /* --------------------------------------------------------------------------
drivers/acpi/pci_root.c +3 -1
Index: linux-pm/drivers/acpi/pci_root.c
===================================================================
@@ -680,11 +680,13 @@ static void hotplug_event_root(void *dat
 		if (!root)
 			break;
 
+		acpi_evaluate_hotplug_ost(handle, ACPI_NOTIFY_EJECT_REQUEST,
+					  ACPI_OST_SC_EJECT_IN_PROGRESS, NULL);
 		get_device(&root->device->dev);
 
 		acpi_scan_lock_release();
 
-		acpi_bus_device_eject(root->device, ACPI_NOTIFY_EJECT_REQUEST);
+		acpi_device_hotplug(root->device, ACPI_NOTIFY_EJECT_REQUEST);
 		return;
 	default:
 		acpi_handle_warn(handle,
drivers/acpi/scan.c +64 -77

213 lines not on this page · raw message

#6 lore raw

PATCH 6/10 ACPI / hotplug: Make ACPI PCI root hotplug use common hotplug code

In reply to Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Rework the common ACPI device hotplug code so that it is suitable for PCI host bridge hotplug and switch the PCI host bridge scan handler to using the common hotplug code.

This allows quite a few lines of code that are not necessary any more to be dropped from the PCI host bridge scan handler and removes arbitrary differences in behavior between PCI host bridge hotplug and ACPI-based hotplug of other components, like CPUs and memory.

Also acpi_device_hotplug() can be static now.

---
 drivers/acpi/internal.h |    2 
 drivers/acpi/pci_root.c |  126 ++++--------------------------------------------
 drivers/acpi/scan.c     |   40 +++++++++------
 include/acpi/acpi_bus.h |    1 
 4 files changed, 38 insertions(+), 131 deletions(-)
include/acpi/acpi_bus.h +1 -0
Index: linux-pm/include/acpi/acpi_bus.h
===================================================================
@@ -101,6 +101,7 @@ struct acpi_hotplug_profile {
 	struct kobject kobj;
 	bool enabled:1;
 	enum acpi_hotplug_mode mode;
+	int (*scan_dependent)(struct acpi_device *adev);
 };
 
 static inline struct acpi_hotplug_profile *to_acpi_hotplug_profile(
drivers/acpi/scan.c +24 -16
Index: linux-pm/drivers/acpi/scan.c
===================================================================
@@ -283,17 +283,6 @@ static int acpi_scan_device_check(struct
 {
 	int error;
 
-	/*
-	 * This function is only called for device objects for which matching
-	 * scan handlers exist.  The only situation in which the scan handler is
-	 * not attached to this device object yet is when the device has just
-	 * appeared (either it wasn't present at all before or it was removed
-	 * and then added again).
-	 */
-	if (adev->handler) {
-		dev_warn(&adev->dev, "Already enumerated\n");
-		return -EBUSY;
-	}
 	error = acpi_bus_scan(adev->handle);
 	if (error) {
 		dev_warn(&adev->dev, "Namespace scan failure\n");
@@ -309,10 +298,11 @@ static int acpi_scan_device_check(struct
 	return 0;
 }
 
-void acpi_device_hotplug(void *data, u32 src)
+static void acpi_device_hotplug(void *data, u32 src)
 {
 	u32 ost_code = ACPI_OST_SC_NON_SPECIFIC_FAILURE;
 	struct acpi_device *adev = data;
+	struct acpi_scan_handler *handler;
 	int error;
 
 	lock_device_hotplug();
@@ -326,12 +316,32 @@ void acpi_device_hotplug(void *data, u32
 	if (adev->handle == INVALID_ACPI_HANDLE)
 		goto out;
 
+	handler = adev->handler;
+
 	switch (src) {
 	case ACPI_NOTIFY_BUS_CHECK:
-		error = acpi_bus_scan(adev->handle);
+		if (handler) {
+			error = handler->hotplug.scan_dependent ?
+					handler->hotplug.scan_dependent(adev) :
+					acpi_bus_scan(adev->handle);
+		} else {
+			error = acpi_scan_device_check(adev);
+		}
 		break;
 	case ACPI_NOTIFY_DEVICE_CHECK:
-		error = acpi_scan_device_check(adev);
+		/*
+		 * This code is only run for device objects for which matching
+		 * scan handlers exist.  The only situation in which the scan
+		 * handler is not attached to this device object yet is when the
+		 * device has just appeared (either it wasn't present at all
+		 * before or it was removed and then added again).
+		 */
+		if (adev->handler) {
+			dev_warn(&adev->dev, "Already enumerated\n");
+			error = -EBUSY;
+		} else {
+			error = acpi_scan_device_check(adev);
+		}
 		break;
 	case ACPI_NOTIFY_EJECT_REQUEST:
 	case ACPI_OST_EC_OSPM_EJECT:
@@ -2081,8 +2091,6 @@ int __init acpi_scan_init(void)
 
 	acpi_update_all_gpes();
 
-	acpi_pci_root_hp_init();
-
  out:
 	mutex_unlock(&acpi_scan_lock);
 	return result;
drivers/acpi/pci_root.c +13 -113

144 lines not on this page · raw message

drivers/acpi/internal.h +0 -2
Index: linux-pm/drivers/acpi/internal.h
===================================================================
@@ -28,7 +28,6 @@ int init_acpi_device_notify(void);
 int acpi_scan_init(void);
 void acpi_pci_root_init(void);
 void acpi_pci_link_init(void);
-void acpi_pci_root_hp_init(void);
 void acpi_processor_init(void);
 void acpi_platform_init(void);
 int acpi_sysfs_init(void);
@@ -89,7 +88,6 @@ void acpi_device_add_finalize(struct acp
 void acpi_free_pnp_ids(struct acpi_device_pnp *pnp);
 int acpi_bind_one(struct device *dev, acpi_handle handle);
 int acpi_unbind_one(struct device *dev);
-void acpi_device_hotplug(void *data, u32 ost_src);
 bool acpi_device_is_present(struct acpi_device *adev);
 
 /* --------------------------------------------------------------------------
#7 lore raw

PATCH 7/10 ACPI / hotplug: Move container-specific code out of the core

In reply to Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Move container-specific uevents from the core hotplug code to the container scan handler's .attach() and .detach() callbacks.

This way the core will not have to special-case containers and the uevents will be guaranteed to happen every time a container is either scanned or trimmed as appropriate.

---
 drivers/acpi/container.c |   11 ++++++++---
 drivers/acpi/scan.c      |    8 +-------
 include/acpi/acpi_bus.h  |    7 -------
 3 files changed, 9 insertions(+), 17 deletions(-)
drivers/acpi/container.c +8 -3
Index: linux-pm/drivers/acpi/container.c
===================================================================
@@ -44,19 +44,24 @@ static const struct acpi_device_id conta
 	{"", 0},
 };
 
-static int container_device_attach(struct acpi_device *device,
+static int container_device_attach(struct acpi_device *adev,
 				   const struct acpi_device_id *not_used)
 {
-	/* This is necessary for container hotplug to work. */
+	kobject_uevent(&adev->dev.kobj, KOBJ_ONLINE);
 	return 1;
 }
 
+static void container_device_detach(struct acpi_device *adev)
+{
+	kobject_uevent(&adev->dev.kobj, KOBJ_OFFLINE);
+}
+
 static struct acpi_scan_handler container_handler = {
 	.ids = container_device_ids,
 	.attach = container_device_attach,
+	.detach = container_device_detach,
 	.hotplug = {
 		.enabled = true,
-		.mode = AHM_CONTAINER,
 	},
 };
 
drivers/acpi/scan.c +1 -7
Index: linux-pm/drivers/acpi/scan.c
===================================================================
@@ -206,9 +206,6 @@ static int acpi_scan_hot_remove(struct a
 	acpi_status status;
 	unsigned long long sta;
 
-	if (device->handler && device->handler->hotplug.mode == AHM_CONTAINER)
-		kobject_uevent(&device->dev.kobj, KOBJ_OFFLINE);
-
 	/*
 	 * Carry out two passes here and ignore errors in the first pass,
 	 * because if the devices in question are memory blocks and
@@ -288,10 +285,7 @@ static int acpi_scan_device_check(struct
 		dev_warn(&adev->dev, "Namespace scan failure\n");
 		return error;
 	}
-	if (adev->handler) {
-		if (adev->handler->hotplug.mode == AHM_CONTAINER)
-			kobject_uevent(&adev->dev.kobj, KOBJ_ONLINE);
-	} else {
+	if (!adev->handler) {
 		dev_warn(&adev->dev, "Enumeration failure\n");
 		return -ENODEV;
 	}
include/acpi/acpi_bus.h +0 -7
Index: linux-pm/include/acpi/acpi_bus.h
===================================================================
@@ -91,16 +91,9 @@ struct acpi_device;
  * -----------------
  */
 
-enum acpi_hotplug_mode {
-	AHM_GENERIC = 0,
-	AHM_CONTAINER,
-	AHM_COUNT
-};
-
 struct acpi_hotplug_profile {
 	struct kobject kobj;
 	bool enabled:1;
-	enum acpi_hotplug_mode mode;
 	int (*scan_dependent)(struct acpi_device *adev);
 };
 
#8 lore raw

Re: [PATCH 7/10] ACPI / hotplug: Move container-specific code out of the core

Hi Rafael,

Replying to this mail may be wrong.

Do you remember following your patch?
http://lkml.org/lkml/2013/2/23/97

I want to add autoeject variable in acpi_hotplug_profile structure and set autoecjet of container device "false".

Currently, I have a problem on ejecting container device. Since linux-3.12, container device is removed by acpi_scan_hot_remove.

I think this has two problems.

1. easily fail
My container device has CPU device and Memory device, and maximum size of memory is 3Tbyte. In my environment, hot removing container device fails on offlining memory if memory is used by application.
I think if offlininig memory, we must retly to offline memory several times.
2. cannot work with userland's application
Hot removing CPU and memory on container device, we need take care of userland application. Before linux-3.12, container device just notifies KOBJ_OFFLINE to udev. So by using udev, if application binds to removed CPU or node, applications can change them before hot removing container device.
Currently, KOBJ_OFFLINE is notified to udev. But acpi_scan_hot_remove also runs simultaneously for hot removing container device. So when applications runs for corresponding to the deletion of the devices, the devices may have been deleted.

I don't know what devices are on hotpluggable conatainer device of other vendors. At least, my container device cannot be hot removed correctly. Then I want to add autoeject variable in acpi_hotplug_profile so that user can change the parameter to "true" or "false".

Thanks,
Yasuaki Ishimatsu

Quoted 101 lines · Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Move container-specific uevents from the core hotplug code to the container scan handler's .attach() and .detach() callbacks.

This way the core will not have to special-case containers and the uevents will be guaranteed to happen every time a container is either scanned or trimmed as appropriate.

---
  drivers/acpi/container.c |   11 ++++++++---
  drivers/acpi/scan.c      |    8 +-------
  include/acpi/acpi_bus.h  |    7 -------
  3 files changed, 9 insertions(+), 17 deletions(-)

Index: linux-pm/drivers/acpi/container.c
===================================================================

drivers/acpi/container.c +8 -3
@@ -44,19 +44,24 @@ static const struct acpi_device_id conta
  	{"", 0},
  };
 
-static int container_device_attach(struct acpi_device *device,
+static int container_device_attach(struct acpi_device *adev,
  				   const struct acpi_device_id *not_used)
  {
-	/* This is necessary for container hotplug to work. */
+	kobject_uevent(&adev->dev.kobj, KOBJ_ONLINE);
  	return 1;
  }
 
+static void container_device_detach(struct acpi_device *adev)
+{
+	kobject_uevent(&adev->dev.kobj, KOBJ_OFFLINE);
+}
+
  static struct acpi_scan_handler container_handler = {
  	.ids = container_device_ids,
  	.attach = container_device_attach,
+	.detach = container_device_detach,
  	.hotplug = {
  		.enabled = true,
-		.mode = AHM_CONTAINER,
  	},
  };
 

Index: linux-pm/drivers/acpi/scan.c
===================================================================

drivers/acpi/scan.c +1 -7
@@ -206,9 +206,6 @@ static int acpi_scan_hot_remove(struct a
  	acpi_status status;
  	unsigned long long sta;
 
-	if (device->handler && device->handler->hotplug.mode == AHM_CONTAINER)
-		kobject_uevent(&device->dev.kobj, KOBJ_OFFLINE);
-
  	/*
  	 * Carry out two passes here and ignore errors in the first pass,
  	 * because if the devices in question are memory blocks and
@@ -288,10 +285,7 @@ static int acpi_scan_device_check(struct
  		dev_warn(&adev->dev, "Namespace scan failure\n");
  		return error;
  	}
-	if (adev->handler) {
-		if (adev->handler->hotplug.mode == AHM_CONTAINER)
-			kobject_uevent(&adev->dev.kobj, KOBJ_ONLINE);
-	} else {
+	if (!adev->handler) {
  		dev_warn(&adev->dev, "Enumeration failure\n");
  		return -ENODEV;
  	}

Index: linux-pm/include/acpi/acpi_bus.h
===================================================================

include/acpi/acpi_bus.h +0 -8
@@ -91,16 +91,9 @@ struct acpi_device;
   * -----------------
   */
 
-enum acpi_hotplug_mode {
-	AHM_GENERIC = 0,
-	AHM_CONTAINER,
-	AHM_COUNT
-};
-
  struct acpi_hotplug_profile {
  	struct kobject kobj;
  	bool enabled:1;
-	enum acpi_hotplug_mode mode;
  	int (*scan_dependent)(struct acpi_device *adev);
  };
 
 
--

To unsubscribe from this list: send the line "unsubscribe linux-acpi" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#9 lore raw

Re: [PATCH 7/10] ACPI / hotplug: Move container-specific code out of the core

Quoted 1 line · Yasuaki Ishimatsu

Hi Rafael,

Hi,

Quoted 1 line

Replying to this mail may be wrong.

OK, so this particular patch doesn't break things any more?

Quoted 5 lines

Do you remember following your patch?
http://lkml.org/lkml/2013/2/23/97

I want to add autoeject variable in acpi_hotplug_profile structure and set autoecjet of container device "false".

Then after the series the $subject patch belongs to it will work almost the same way as /sys/firmware/acpi/container/enabled (hot add will still work after patch [4/10] if "enabled" is 0), but only for containers.

Quoted 11 lines

Currently, I have a problem on ejecting container device. Since linux-3.12, container device is removed by acpi_scan_hot_remove.

I think this has two problems.

1. easily fail
My container device has CPU device and Memory device, and maximum size of memory is 3Tbyte. In my environment, hot removing container device fails on offlining memory if memory is used by application.
I think if offlininig memory, we must retly to offline memory several times.

Yes, that's correct. But then you can try to offline the memory upfront and only remove the container after that has been successful.

Quoted 10 lines

2. cannot work with userland's application
Hot removing CPU and memory on container device, we need take care of userland application. Before linux-3.12, container device just notifies KOBJ_OFFLINE to udev. So by using udev, if application binds to removed CPU or node, applications can change them before hot removing container device.
Currently, KOBJ_OFFLINE is notified to udev. But acpi_scan_hot_remove also runs simultaneously for hot removing container device. So when applications runs for corresponding to the deletion of the devices, the devices may have been deleted.

So the expectation is that the container will refuse to offline, but instead it will emit KOBJ_OFFLINE so that user space can do some cleanup and offline it through the "eject" attribute, right?

Quoted 4 lines

I don't know what devices are on hotpluggable conatainer device of other vendors. At least, my container device cannot be hot removed correctly. Then I want to add autoeject variable in acpi_hotplug_profile so that user can change the parameter to "true" or "false".

I have a different idea.

Why don't we create a bus type for containers in analogy with CPUs and memory and make it support offline. Then, the container scan handler will create a "physical" container device under that bus type and the new bus type code will implement the logic you need (that is, it will have a sysfs flag that will cause the offline to fail emitting a uevent of some sort if set and will allow the offline to happen when unset). That "physical" container device will go away (again, via the container scan handler) during container removal.

The eject work flow can be:
(1) an eject event occurs,
(2) the container "physical" device fails offline in acpi_scan_hot_remove() emmitting, say, KOBJ_CHANGE for the "physical" device,
(3) user space notices the KOBJ_CHANGE and does the cleanup as needed,
(4) user space changes the "physical" container device flag controlling offline to 0,
(5) user space uses the sysfs "eject" attribute of the ACPI container object to finally eject the container,
(6) the offline in acpi_scan_hot_remove() is now successful, because the flag controlling it has been set to 0 in step (4),
(7) the "physical" container device goes away before executing _EJ0,
(8) the container is ejected.

Of course, if the flag controlling container offline is 0 to start with, step
(6) will now occur directly after (1), so whoever wants containers to be hot-removed automatically may just clear that flag for all of them on boot.

How does that sound?

Rafael

#10 lore raw

Re: [PATCH 7/10] ACPI / hotplug: Move container-specific code out of the core

Quoted 8 lines · Rafael J. Wysocki
Quoted 1 line · Yasuaki Ishimatsu

Hi Rafael,

Hi,

Quoted 1 line

Replying to this mail may be wrong.

OK, so this particular patch doesn't break things any more?

Yes.

Quoted 38 lines
Quoted 6 lines

Do you remember following your patch?
http://lkml.org/lkml/2013/2/23/97

I want to add autoeject variable in acpi_hotplug_profile structure and set autoecjet of container device "false".

Then after the series the $subject patch belongs to it will work almost the same way as /sys/firmware/acpi/container/enabled (hot add will still work after patch [4/10] if "enabled" is 0), but only for containers.

Quoted 11 lines

Currently, I have a problem on ejecting container device. Since linux-3.12, container device is removed by acpi_scan_hot_remove.

I think this has two problems.

1. easily fail
My container device has CPU device and Memory device, and maximum size of memory is 3Tbyte. In my environment, hot removing container device fails on offlining memory if memory is used by application.
I think if offlininig memory, we must retly to offline memory several times.

Yes, that's correct. But then you can try to offline the memory upfront and only remove the container after that has been successful.

Quoted 10 lines

2. cannot work with userland's application
Hot removing CPU and memory on container device, we need take care of userland application. Before linux-3.12, container device just notifies KOBJ_OFFLINE to udev. So by using udev, if application binds to removed CPU or node, applications can change them before hot removing container device.
Currently, KOBJ_OFFLINE is notified to udev. But acpi_scan_hot_remove also runs simultaneously for hot removing container device. So when applications runs for corresponding to the deletion of the devices, the devices may have been deleted.

Quoted 3 lines

So the expectation is that the container will refuse to offline, but instead it will emit KOBJ_OFFLINE so that user space can do some cleanup and offline it through the "eject" attribute, right?

Yes, that's right.

Quoted 31 lines
Quoted 5 lines

I don't know what devices are on hotpluggable conatainer device of other vendors. At least, my container device cannot be hot removed correctly. Then I want to add autoeject variable in acpi_hotplug_profile so that user can change the parameter to "true" or "false".

I have a different idea.

Why don't we create a bus type for containers in analogy with CPUs and memory and make it support offline. Then, the container scan handler will create a "physical" container device under that bus type and the new bus type code will implement the logic you need (that is, it will have a sysfs flag that will cause the offline to fail emitting a uevent of some sort if set and will allow the offline to happen when unset). That "physical" container device will go away (again, via the container scan handler) during container removal.

The eject work flow can be:
(1) an eject event occurs,
(2) the container "physical" device fails offline in acpi_scan_hot_remove() emmitting, say, KOBJ_CHANGE for the "physical" device,
(3) user space notices the KOBJ_CHANGE and does the cleanup as needed,
(4) user space changes the "physical" container device flag controlling offline to 0,
(5) user space uses the sysfs "eject" attribute of the ACPI container object to finally eject the container,
(6) the offline in acpi_scan_hot_remove() is now successful, because the flag controlling it has been set to 0 in step (4),
(7) the "physical" container device goes away before executing _EJ0,
(8) the container is ejected.

Quoted 5 lines

Of course, if the flag controlling container offline is 0 to start with, step
(6) will now occur directly after (1), so whoever wants containers to be hot-removed automatically may just clear that flag for all of them on boot.

How does that sound?

The above ideas are almost O.K. I want kernel to notify user space of KOBJ_OFFLINE. Even if user space catches "KOBJ_CHANGE", user doesn't know whether the notification is offline or not.

Thanks,
Yasuaki Ishimatsu

Quoted 7 lines

Rafael

--
To unsubscribe from this list: send the line "unsubscribe linux-acpi" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#11 lore raw

Re: [PATCH 7/10] ACPI / hotplug: Move container-specific code out of the core

Quoted 91 lines · Yasuaki Ishimatsu
Quoted 8 lines · Rafael J. Wysocki
Quoted 1 line · Yasuaki Ishimatsu

Hi Rafael,

Hi,

Quoted 1 line

Replying to this mail may be wrong.

OK, so this particular patch doesn't break things any more?

Yes.

Quoted 38 lines
Quoted 6 lines

Do you remember following your patch?
http://lkml.org/lkml/2013/2/23/97

I want to add autoeject variable in acpi_hotplug_profile structure and set autoecjet of container device "false".

Then after the series the $subject patch belongs to it will work almost the same way as /sys/firmware/acpi/container/enabled (hot add will still work after patch [4/10] if "enabled" is 0), but only for containers.

Quoted 11 lines

Currently, I have a problem on ejecting container device. Since linux-3.12, container device is removed by acpi_scan_hot_remove.

I think this has two problems.

1. easily fail
My container device has CPU device and Memory device, and maximum size of memory is 3Tbyte. In my environment, hot removing container device fails on offlining memory if memory is used by application.
I think if offlininig memory, we must retly to offline memory several times.

Yes, that's correct. But then you can try to offline the memory upfront and only remove the container after that has been successful.

Quoted 10 lines

2. cannot work with userland's application
Hot removing CPU and memory on container device, we need take care of userland application. Before linux-3.12, container device just notifies KOBJ_OFFLINE to udev. So by using udev, if application binds to removed CPU or node, applications can change them before hot removing container device.
Currently, KOBJ_OFFLINE is notified to udev. But acpi_scan_hot_remove also runs simultaneously for hot removing container device. So when applications runs for corresponding to the deletion of the devices, the devices may have been deleted.

Quoted 3 lines

So the expectation is that the container will refuse to offline, but instead it will emit KOBJ_OFFLINE so that user space can do some cleanup and offline it through the "eject" attribute, right?

Yes, that's right.

Quoted 35 lines
Quoted 5 lines

I don't know what devices are on hotpluggable conatainer device of other vendors. At least, my container device cannot be hot removed correctly. Then I want to add autoeject variable in acpi_hotplug_profile so that user can change the parameter to "true" or "false".

I have a different idea.

Why don't we create a bus type for containers in analogy with CPUs and memory and make it support offline. Then, the container scan handler will create a "physical" container device under that bus type and the new bus type code will implement the logic you need (that is, it will have a sysfs flag that will cause the offline to fail emitting a uevent of some sort if set and will allow the offline to happen when unset). That "physical" container device will go away (again, via the container scan handler) during container removal.

The eject work flow can be:
(1) an eject event occurs,
(2) the container "physical" device fails offline in acpi_scan_hot_remove() emmitting, say, KOBJ_CHANGE for the "physical" device,
(3) user space notices the KOBJ_CHANGE and does the cleanup as needed,
(4) user space changes the "physical" container device flag controlling offline to 0,
(5) user space uses the sysfs "eject" attribute of the ACPI container object to finally eject the container,
(6) the offline in acpi_scan_hot_remove() is now successful, because the flag controlling it has been set to 0 in step (4),
(7) the "physical" container device goes away before executing _EJ0,
(8) the container is ejected.

Of course, if the flag controlling container offline is 0 to start with, step
(6) will now occur directly after (1), so whoever wants containers to be hot-removed automatically may just clear that flag for all of them on boot.

Quoted 6 lines
Quoted 1 line

How does that sound?

The above ideas are almost O.K. I want kernel to notify user space of KOBJ_OFFLINE. Even if user space catches "KOBJ_CHANGE", user doesn't know whether the notification is offline or not.

It is easy to figure out, though. Since the KOBJ_CHANGE will be emitted for container devices only in that situation, user space can see that (1) it is from a container and (2) it is KOBJ_CHANGE, so it must mean "container offline has been attempted".

My concern with using KOBJ_OFFLINE for that is that device_offline() emits it too on success and it may be easily confused with the one emitted on failure for containers.

Thanks,
Rafael

#12 lore raw

Re: [PATCH 7/10] ACPI / hotplug: Move container-specific code out of the core

Quoted 100 lines · Rafael J. Wysocki
Quoted 91 lines · Yasuaki Ishimatsu
Quoted 8 lines · Rafael J. Wysocki
Quoted 1 line · Yasuaki Ishimatsu

Hi Rafael,

Hi,

Quoted 1 line

Replying to this mail may be wrong.

OK, so this particular patch doesn't break things any more?

Yes.

Quoted 38 lines
Quoted 6 lines

Do you remember following your patch?
http://lkml.org/lkml/2013/2/23/97

I want to add autoeject variable in acpi_hotplug_profile structure and set autoecjet of container device "false".

Then after the series the $subject patch belongs to it will work almost the same way as /sys/firmware/acpi/container/enabled (hot add will still work after patch [4/10] if "enabled" is 0), but only for containers.

Quoted 11 lines

Currently, I have a problem on ejecting container device. Since linux-3.12, container device is removed by acpi_scan_hot_remove.

I think this has two problems.

1. easily fail
My container device has CPU device and Memory device, and maximum size of memory is 3Tbyte. In my environment, hot removing container device fails on offlining memory if memory is used by application.
I think if offlininig memory, we must retly to offline memory several times.

Yes, that's correct. But then you can try to offline the memory upfront and only remove the container after that has been successful.

Quoted 10 lines

2. cannot work with userland's application
Hot removing CPU and memory on container device, we need take care of userland application. Before linux-3.12, container device just notifies KOBJ_OFFLINE to udev. So by using udev, if application binds to removed CPU or node, applications can change them before hot removing container device.
Currently, KOBJ_OFFLINE is notified to udev. But acpi_scan_hot_remove also runs simultaneously for hot removing container device. So when applications runs for corresponding to the deletion of the devices, the devices may have been deleted.

Quoted 3 lines

So the expectation is that the container will refuse to offline, but instead it will emit KOBJ_OFFLINE so that user space can do some cleanup and offline it through the "eject" attribute, right?

Yes, that's right.

Quoted 35 lines
Quoted 5 lines

I don't know what devices are on hotpluggable conatainer device of other vendors. At least, my container device cannot be hot removed correctly. Then I want to add autoeject variable in acpi_hotplug_profile so that user can change the parameter to "true" or "false".

I have a different idea.

Why don't we create a bus type for containers in analogy with CPUs and memory and make it support offline. Then, the container scan handler will create a "physical" container device under that bus type and the new bus type code will implement the logic you need (that is, it will have a sysfs flag that will cause the offline to fail emitting a uevent of some sort if set and will allow the offline to happen when unset). That "physical" container device will go away (again, via the container scan handler) during container removal.

The eject work flow can be:
(1) an eject event occurs,
(2) the container "physical" device fails offline in acpi_scan_hot_remove() emmitting, say, KOBJ_CHANGE for the "physical" device,
(3) user space notices the KOBJ_CHANGE and does the cleanup as needed,
(4) user space changes the "physical" container device flag controlling offline to 0,
(5) user space uses the sysfs "eject" attribute of the ACPI container object to finally eject the container,
(6) the offline in acpi_scan_hot_remove() is now successful, because the flag controlling it has been set to 0 in step (4),
(7) the "physical" container device goes away before executing _EJ0,
(8) the container is ejected.

Of course, if the flag controlling container offline is 0 to start with, step
(6) will now occur directly after (1), so whoever wants containers to be hot-removed automatically may just clear that flag for all of them on boot.

Quoted 6 lines
Quoted 1 line

How does that sound?

The above ideas are almost O.K. I want kernel to notify user space of KOBJ_OFFLINE. Even if user space catches "KOBJ_CHANGE", user doesn't know whether the notification is offline or not.

Quoted 8 lines

It is easy to figure out, though. Since the KOBJ_CHANGE will be emitted for container devices only in that situation, user space can see that (1) it is from a container and (2) it is KOBJ_CHANGE, so it must mean "container offline has been attempted".

My concern with using KOBJ_OFFLINE for that is that device_offline() emits it too on success and it may be easily confused with the one emitted on failure for containers.

I have no objection.

Thanks,
Yasuaki Ishimatsu

Quoted 4 lines

Thanks,
Rafael

#13 lore raw

Re: [PATCH 7/10] ACPI / hotplug: Move container-specific code out of the core

In reply to Rafael J. Wysocki

Hi Rafael,

Please share your more detailed idea. I started to implement the following idea. But the idea has one problem.

Quoted 5 lines
Quoted 5 lines
Quoted 5 lines

The eject work flow can be:
(1) an eject event occurs,
(2) the container "physical" device fails offline in acpi_scan_hot_remove() emmitting, say, KOBJ_CHANGE for the "physical" device,
(3) user space notices the KOBJ_CHANGE and does the cleanup as needed,

Quoted 8 lines
Quoted 8 lines
Quoted 8 lines

(4) user space changes the "physical" container device flag controlling offline to 0,
(5) user space uses the sysfs "eject" attribute of the ACPI container object to finally eject the container,
(6) the offline in acpi_scan_hot_remove() is now successful, because the flag controlling it has been set to 0 in step (4),
(7) the "physical" container device goes away before executing _EJ0,
(8) the container is ejected.

I want to emit KOBJ_CHANGE before offlining devices on container device at (2). But acpi_scan_hot_remove() offlines devices on container device at first. So when offline container device, devices on container has been offlined.

Thus the idea cannot fulfill my necessary feature.

Thanks,
Yasauaki Ishimatsu

Thanks,
Yasauaki Ishimatsu

Quoted 24 lines
Quoted 11 lines
Quoted 6 lines

Of course, if the flag controlling container offline is 0 to start with, step
(6) will now occur directly after (1), so whoever wants containers to be hot-removed automatically may just clear that flag for all of them on boot.

How does that sound?

The above ideas are almost O.K. I want kernel to notify user space of KOBJ_OFFLINE. Even if user space catches "KOBJ_CHANGE", user doesn't know whether the notification is offline or not.

It is easy to figure out, though. Since the KOBJ_CHANGE will be emitted for container devices only in that situation, user space can see that (1) it is from a container and (2) it is KOBJ_CHANGE, so it must mean "container offline has been attempted".

My concern with using KOBJ_OFFLINE for that is that device_offline() emits it too on success and it may be easily confused with the one emitted on failure for containers.

Thanks,
Rafael

#14 lore raw

Re: [PATCH 7/10] ACPI / hotplug: Move container-specific code out of the core

Quoted 1 line · Yasuaki Ishimatsu

Hi Rafael,

Hi,

Quoted 17 lines

Please share your more detailed idea. I started to implement the following idea. But the idea has one problem.

Quoted 5 lines
Quoted 5 lines
Quoted 5 lines

The eject work flow can be:
(1) an eject event occurs,
(2) the container "physical" device fails offline in acpi_scan_hot_remove() emmitting, say, KOBJ_CHANGE for the "physical" device,
(3) user space notices the KOBJ_CHANGE and does the cleanup as needed,

Quoted 8 lines
Quoted 8 lines
Quoted 8 lines

(4) user space changes the "physical" container device flag controlling offline to 0,
(5) user space uses the sysfs "eject" attribute of the ACPI container object to finally eject the container,
(6) the offline in acpi_scan_hot_remove() is now successful, because the flag controlling it has been set to 0 in step (4),
(7) the "physical" container device goes away before executing _EJ0,
(8) the container is ejected.

Quoted 5 lines

I want to emit KOBJ_CHANGE before offlining devices on container device at (2). But acpi_scan_hot_remove() offlines devices on container device at first. So when offline container device, devices on container has been offlined.

Thus the idea cannot fulfill my necessary feature.

Well, in that case we need to treat containers in a special way at the ACPI level. Which is a bit unfortunate so to speak.

To that end I'd try to add a new flag to struct acpi_hotplug_profile, say .verify_offline, such that if set, it would cause acpi_scan_hot_remove() to check if all of the "physical" companions of the top-level device are offline to start with, and if not, it would just emit KOBJ_CHANGE for the companions that are not offline and bail out.

So the above algorithm would become:

(1) an eject event occurs,
(2) acpi_scan_hot_remove() checks the verify_offline flag in the target device's scan_handler structure,
(3) if set (it would always be set for containers), acpi_scan_hot_remove() checks the status of the target device's "physical" companions; if at least one of them is offline, KOBJ_CHANGE is emitted for that "physical" device, and acpi_scan_hot_remove() returns, [I guess we can just emit KOBJ_CHANGE for the first companion that is not offline at this point.]
(4) user space notices the KOBJ_CHANGE and does the cleanup as needed; in the process it carries out the offline operation for the container's "physical" companion (there's only one such companion for each container), [That operation for the container itself is trivial, but to succeed it requires all devices below the container to be taken offline in advance.]
(5) user space uses the sysfs "eject" attribute of the ACPI container object to finally eject the container,
(6) acpi_scan_hot_remove() is now successful, because the container's "physical" companion is now offline,
(7) the "physical" container device goes away before executing _EJ0,
(8) the container is ejected.

I think that should work for you.

Thanks,
Rafael

#15 lore raw

Re: [PATCH 7/10] ACPI / hotplug: Move container-specific code out of the core

Quoted 59 lines · Rafael J. Wysocki
Quoted 1 line · Yasuaki Ishimatsu

Hi Rafael,

Hi,

Quoted 17 lines

Please share your more detailed idea. I started to implement the following idea. But the idea has one problem.

Quoted 5 lines
Quoted 5 lines
Quoted 5 lines

The eject work flow can be:
(1) an eject event occurs,
(2) the container "physical" device fails offline in acpi_scan_hot_remove() emmitting, say, KOBJ_CHANGE for the "physical" device,
(3) user space notices the KOBJ_CHANGE and does the cleanup as needed,

Quoted 8 lines
Quoted 8 lines
Quoted 8 lines

(4) user space changes the "physical" container device flag controlling offline to 0,
(5) user space uses the sysfs "eject" attribute of the ACPI container object to finally eject the container,
(6) the offline in acpi_scan_hot_remove() is now successful, because the flag controlling it has been set to 0 in step (4),
(7) the "physical" container device goes away before executing _EJ0,
(8) the container is ejected.

Quoted 5 lines

I want to emit KOBJ_CHANGE before offlining devices on container device at (2). But acpi_scan_hot_remove() offlines devices on container device at first. So when offline container device, devices on container has been offlined.

Thus the idea cannot fulfill my necessary feature.

Well, in that case we need to treat containers in a special way at the ACPI level. Which is a bit unfortunate so to speak.

To that end I'd try to add a new flag to struct acpi_hotplug_profile, say .verify_offline, such that if set, it would cause acpi_scan_hot_remove() to check if all of the "physical" companions of the top-level device are offline to start with, and if not, it would just emit KOBJ_CHANGE for the companions that are not offline and bail out.

So the above algorithm would become:

(1) an eject event occurs,
(2) acpi_scan_hot_remove() checks the verify_offline flag in the target device's scan_handler structure,
(3) if set (it would always be set for containers), acpi_scan_hot_remove() checks the status of the target device's "physical" companions; if at least one of them is offline, KOBJ_CHANGE is emitted for that "physical" device, and acpi_scan_hot_remove() returns, [I guess we can just emit KOBJ_CHANGE for the first companion that is not offline at this point.]
(4) user space notices the KOBJ_CHANGE and does the cleanup as needed; in the process it carries out the offline operation for the container's "physical" companion (there's only one such companion for each container), [That operation for the container itself is trivial, but to succeed it requires all devices below the container to be taken offline in advance.]
(5) user space uses the sysfs "eject" attribute of the ACPI container object to finally eject the container,
(6) acpi_scan_hot_remove() is now successful, because the container's "physical" companion is now offline,
(7) the "physical" container device goes away before executing _EJ0,
(8) the container is ejected.

Quoted 1 line

I think that should work for you.

This idea seems to same as your previous work.
http://lkml.org/lkml/2013/2/23/97

How about add autoremove flag into acpi_hotplug_profile and check it as follow:

---
 drivers/acpi/scan.c | 5 +++++
 1 file changed, 5 insertions(+)
drivers/acpi/scan.c +5 -0
index 5383c81..c43d110 100644
@@ -409,6 +409,11 @@ static void acpi_hotplug_notify_cb(acpi_handle handle, u32 type, void *data)
 			ost_code = ACPI_OST_SC_EJECT_NOT_SUPPORTED;
 			goto err_out;
 		}
+		if (!handler->hotplug.autoremove) {
+			kobject_uevent(&device->dev.kobj, KOBJ_CHANGE);
+			ost_code = ACPI_OST_SC_EJECT_NON_SPECIFIC_FAILURE;
+			goto err_out;
+		}
 		acpi_evaluate_hotplug_ost(handle, ACPI_NOTIFY_EJECT_REQUEST,
 					  ACPI_OST_SC_EJECT_IN_PROGRESS, NULL);
 		break;

Adding the check into "acpi_hotplug_notify_cb()", user need not change the flag for removing container device by "sysfs eject".

Thanks,
Yasuaki Ishimatsu

Quoted 4 lines

Thanks,
Rafael

#16 lore raw

Re: [PATCH 7/10] ACPI / hotplug: Move container-specific code out of the core

Quoted 60 lines · Yasuaki Ishimatsu
Quoted 59 lines · Rafael J. Wysocki
Quoted 1 line · Yasuaki Ishimatsu

Hi Rafael,

Hi,

Quoted 17 lines

Please share your more detailed idea. I started to implement the following idea. But the idea has one problem.

Quoted 5 lines
Quoted 5 lines
Quoted 5 lines

The eject work flow can be:
(1) an eject event occurs,
(2) the container "physical" device fails offline in acpi_scan_hot_remove() emmitting, say, KOBJ_CHANGE for the "physical" device,
(3) user space notices the KOBJ_CHANGE and does the cleanup as needed,

Quoted 8 lines
Quoted 8 lines
Quoted 8 lines

(4) user space changes the "physical" container device flag controlling offline to 0,
(5) user space uses the sysfs "eject" attribute of the ACPI container object to finally eject the container,
(6) the offline in acpi_scan_hot_remove() is now successful, because the flag controlling it has been set to 0 in step (4),
(7) the "physical" container device goes away before executing _EJ0,
(8) the container is ejected.

Quoted 5 lines

I want to emit KOBJ_CHANGE before offlining devices on container device at (2). But acpi_scan_hot_remove() offlines devices on container device at first. So when offline container device, devices on container has been offlined.

Thus the idea cannot fulfill my necessary feature.

Well, in that case we need to treat containers in a special way at the ACPI level. Which is a bit unfortunate so to speak.

To that end I'd try to add a new flag to struct acpi_hotplug_profile, say .verify_offline, such that if set, it would cause acpi_scan_hot_remove() to check if all of the "physical" companions of the top-level device are offline to start with, and if not, it would just emit KOBJ_CHANGE for the companions that are not offline and bail out.

So the above algorithm would become:

(1) an eject event occurs,
(2) acpi_scan_hot_remove() checks the verify_offline flag in the target device's scan_handler structure,
(3) if set (it would always be set for containers), acpi_scan_hot_remove() checks the status of the target device's "physical" companions; if at least one of them is offline, KOBJ_CHANGE is emitted for that "physical" device, and acpi_scan_hot_remove() returns, [I guess we can just emit KOBJ_CHANGE for the first companion that is not offline at this point.]
(4) user space notices the KOBJ_CHANGE and does the cleanup as needed; in the process it carries out the offline operation for the container's "physical" companion (there's only one such companion for each container), [That operation for the container itself is trivial, but to succeed it requires all devices below the container to be taken offline in advance.]
(5) user space uses the sysfs "eject" attribute of the ACPI container object to finally eject the container,
(6) acpi_scan_hot_remove() is now successful, because the container's "physical" companion is now offline,
(7) the "physical" container device goes away before executing _EJ0,
(8) the container is ejected.

Quoted 4 lines
Quoted 1 line

I think that should work for you.

This idea seems to same as your previous work.
http://lkml.org/lkml/2013/2/23/97

No, it is not. That one didn't involve physical device representations.

Quoted 1 line

How about add autoremove flag into acpi_hotplug_profile and check it as follow:

This is very similar to "enable" except that it generates the uevent and "enable" doesn't. You might as well modify "enable" to trigger a uevent if eject is not enabled (note that with the latest patches in linux-next "enable" only applies to eject).

That said I don't think we should generate any uevents for struct acpi_device objects, because they are not devices.

Quoted 21 lines
---
  drivers/acpi/scan.c | 5 +++++
  1 file changed, 5 insertions(+)
drivers/acpi/scan.c +5 -0
index 5383c81..c43d110 100644
@@ -409,6 +409,11 @@ static void acpi_hotplug_notify_cb(acpi_handle handle, u32 type, void *data)
  			ost_code = ACPI_OST_SC_EJECT_NOT_SUPPORTED;
  			goto err_out;
  		}
+		if (!handler->hotplug.autoremove) {
+			kobject_uevent(&device->dev.kobj, KOBJ_CHANGE);
+			ost_code = ACPI_OST_SC_EJECT_NON_SPECIFIC_FAILURE;
+			goto err_out;
+		}
  		acpi_evaluate_hotplug_ost(handle, ACPI_NOTIFY_EJECT_REQUEST,
  					  ACPI_OST_SC_EJECT_IN_PROGRESS, NULL);
  		break;
Quoted 2 lines

Adding the check into "acpi_hotplug_notify_cb()", user need not change the flag for removing container device by "sysfs eject".

Which is utterly confusing. There is no reason whatsoever why the sysfs eject attribute should work differently from the event-triggered eject. Quite the opposite is the case: it should work in the same way in my opinion so that it is possible to test the eject code path using that attribute.

I'm traveling now, but when I get back home (next week), I'll try to implement the thing I was talking about above.

Thanks,
Rafael

#17 lore raw

Re: [PATCH 7/10] ACPI / hotplug: Move container-specific code out of the core

Quoted 107 lines · Rafael J. Wysocki
Quoted 60 lines · Yasuaki Ishimatsu
Quoted 59 lines · Rafael J. Wysocki
Quoted 1 line · Yasuaki Ishimatsu

Hi Rafael,

Hi,

Quoted 17 lines

Please share your more detailed idea. I started to implement the following idea. But the idea has one problem.

Quoted 5 lines
Quoted 5 lines
Quoted 5 lines

The eject work flow can be:
(1) an eject event occurs,
(2) the container "physical" device fails offline in acpi_scan_hot_remove() emmitting, say, KOBJ_CHANGE for the "physical" device,
(3) user space notices the KOBJ_CHANGE and does the cleanup as needed,

Quoted 8 lines
Quoted 8 lines
Quoted 8 lines

(4) user space changes the "physical" container device flag controlling offline to 0,
(5) user space uses the sysfs "eject" attribute of the ACPI container object to finally eject the container,
(6) the offline in acpi_scan_hot_remove() is now successful, because the flag controlling it has been set to 0 in step (4),
(7) the "physical" container device goes away before executing _EJ0,
(8) the container is ejected.

Quoted 5 lines

I want to emit KOBJ_CHANGE before offlining devices on container device at (2). But acpi_scan_hot_remove() offlines devices on container device at first. So when offline container device, devices on container has been offlined.

Thus the idea cannot fulfill my necessary feature.

Well, in that case we need to treat containers in a special way at the ACPI level. Which is a bit unfortunate so to speak.

To that end I'd try to add a new flag to struct acpi_hotplug_profile, say .verify_offline, such that if set, it would cause acpi_scan_hot_remove() to check if all of the "physical" companions of the top-level device are offline to start with, and if not, it would just emit KOBJ_CHANGE for the companions that are not offline and bail out.

So the above algorithm would become:

(1) an eject event occurs,
(2) acpi_scan_hot_remove() checks the verify_offline flag in the target device's scan_handler structure,
(3) if set (it would always be set for containers), acpi_scan_hot_remove() checks the status of the target device's "physical" companions; if at least one of them is offline, KOBJ_CHANGE is emitted for that "physical" device, and acpi_scan_hot_remove() returns, [I guess we can just emit KOBJ_CHANGE for the first companion that is not offline at this point.]
(4) user space notices the KOBJ_CHANGE and does the cleanup as needed; in the process it carries out the offline operation for the container's "physical" companion (there's only one such companion for each container), [That operation for the container itself is trivial, but to succeed it requires all devices below the container to be taken offline in advance.]
(5) user space uses the sysfs "eject" attribute of the ACPI container object to finally eject the container,
(6) acpi_scan_hot_remove() is now successful, because the container's "physical" companion is now offline,
(7) the "physical" container device goes away before executing _EJ0,
(8) the container is ejected.

Quoted 4 lines
Quoted 1 line

I think that should work for you.

This idea seems to same as your previous work.
http://lkml.org/lkml/2013/2/23/97

No, it is not. That one didn't involve physical device representations.

Quoted 1 line

How about add autoremove flag into acpi_hotplug_profile and check it as follow:

This is very similar to "enable" except that it generates the uevent and "enable" doesn't. You might as well modify "enable" to trigger a uevent if eject is not enabled (note that with the latest patches in linux-next "enable" only applies to eject).

That said I don't think we should generate any uevents for struct acpi_device objects, because they are not devices.

Quoted 21 lines
---
  drivers/acpi/scan.c | 5 +++++
  1 file changed, 5 insertions(+)
drivers/acpi/scan.c +5 -0
index 5383c81..c43d110 100644
@@ -409,6 +409,11 @@ static void acpi_hotplug_notify_cb(acpi_handle handle, u32 type, void *data)
  			ost_code = ACPI_OST_SC_EJECT_NOT_SUPPORTED;
  			goto err_out;
  		}
+		if (!handler->hotplug.autoremove) {
+			kobject_uevent(&device->dev.kobj, KOBJ_CHANGE);
+			ost_code = ACPI_OST_SC_EJECT_NON_SPECIFIC_FAILURE;
+			goto err_out;
+		}
  		acpi_evaluate_hotplug_ost(handle, ACPI_NOTIFY_EJECT_REQUEST,
  					  ACPI_OST_SC_EJECT_IN_PROGRESS, NULL);
  		break;
Quoted 2 lines

Adding the check into "acpi_hotplug_notify_cb()", user need not change the flag for removing container device by "sysfs eject".

Which is utterly confusing. There is no reason whatsoever why the sysfs eject attribute should work differently from the event-triggered eject. Quite the opposite is the case: it should work in the same way in my opinion so that it is possible to test the eject code path using that attribute.

Quoted 2 lines

I'm traveling now, but when I get back home (next week), I'll try to implement the thing I was talking about above.

It took some more time than I had expected, but I finally was able to get to that.

The following two patches implement the idea. This is the minimum (in my opinion) implementation and it may be extended in some ways.

Patch [1/2] introduces a new demand_offline flag for struct acpi_hotplug_profile that makes acpi_scan_hot_remove() check the offline status of the device object's companion physical devices to start with and return -EBUSY if at least one of them is not offline.

Patch [2/2] uses that flag to implement the container handling. The details are in the changelog, but that's how it is supposed to work.

During the initial namespace scan the container ACPI scan handler should create "physical" system container device under /sys/devices/system/container/ for each ACPI container object (the sysfs name of that device should be the same as the sysfs name of the corresponding container object and they should be linked to each other via the firmware_node and physical_node symbolic links, respectively). Those system container devices are initially online.

When a container eject event happens, acpi_scan_hot_remove() will notice that hotplug.demand_offline is set in the device object's scan handler and will check the online status of its "physical" companion device, which is online (that is the system container device the above paragraph is about). That will cause KOBJ_CHANGE to be emitted for the system container device and -EBUSY to be returned by acpi_scan_hot_remove().

Now, user space needs to offline the system container device through its online sysfs attribute (that should be present, because the bus type for containers provides the online and offline callbacks). However, the offline for system container devices will only succeed if the physical devices right below the container are all offline, so user space will have to offline those devices before attempting to offline the system container device itself. When finished, user space can trigger the container removal with the help of the eject sysfs attribute of the ACPI container object pointed to by the system container device's firmware_node link (this time the check in acpi_scan_hot_remove() will succeed, because the system container device in question is now offline).

The way it is implemented is a bit hackish (the driver_data pointer is slightly abused), but that's a special case and I wanted to avoid adding new fields to struct device just for handling it.

The patches haven't been tested yet. I'm going to do that later today, but first I need to take care of some other things, so that has to wait.

Thanks,
Rafael

#18 lore raw

PATCH 1/2 ACPI / hotplug: Add demand_offline hotplug profile flag

From: Rafael J. Wysocki <[email protected]>

Add a new ACPI hotplug profile flag, demand_offline, such that if set for the given ACPI device object's scan handler, it will cause acpi_scan_hot_remove() to check if that device object's physical companions are offline upfront and fail the hot removal if that is not the case.

That flag will be useful to overcome a problem with containers on some system where they can only be hot-removed after some cleanup operations carried out by user space, which needs to be notified of the container hot-removal before the kernel attempts to offline devices in the container. In those cases the current implementation of acpi_scan_hot_remove() is not sufficient, because it first tries to offline the devices in the container and only if that is suffcessful it tries to offline the container itself. As a result, the container hot-removal notification is not delivered to user space at the right time.

---
 drivers/acpi/scan.c     |   41 +++++++++++++++++++++++++++++++++++++----
 include/acpi/acpi_bus.h |    3 ++-
 2 files changed, 39 insertions(+), 5 deletions(-)
drivers/acpi/scan.c +37 -4
Index: linux-pm/drivers/acpi/scan.c
===================================================================
@@ -126,6 +126,24 @@ acpi_device_modalias_show(struct device
 }
 static DEVICE_ATTR(modalias, 0444, acpi_device_modalias_show, NULL);
 
+static bool acpi_scan_is_offline(struct acpi_device *adev)
+{
+	struct acpi_device_physical_node *pn;
+	bool offline = true;
+
+	mutex_lock(&adev->physical_node_lock);
+
+	list_for_each_entry(pn, &adev->physical_node_list, node)
+		if (!pn->dev->offline) {
+			kobject_uevent(&pn->dev->kobj, KOBJ_CHANGE);
+			offline = false;
+			break;
+		}
+
+	mutex_unlock(&adev->physical_node_lock);
+	return offline;
+}
+
 static acpi_status acpi_bus_offline(acpi_handle handle, u32 lvl, void *data,
 				    void **ret_p)
 {
@@ -196,12 +214,11 @@ static acpi_status acpi_bus_online(acpi_
 	return AE_OK;
 }
 
-static int acpi_scan_hot_remove(struct acpi_device *device)
+static int acpi_scan_try_to_offline(struct acpi_device *device)
 {
 	acpi_handle handle = device->handle;
-	struct device *errdev;
+	struct device *errdev = NULL;
 	acpi_status status;
-	unsigned long long sta;
 
 	/*
 	 * Carry out two passes here and ignore errors in the first pass,
@@ -212,7 +229,6 @@ static int acpi_scan_hot_remove(struct a
 	 *
 	 * If the first pass is successful, the second one isn't needed, though.
 	 */
-	errdev = NULL;
 	status = acpi_walk_namespace(ACPI_TYPE_ANY, handle, ACPI_UINT32_MAX,
 				     NULL, acpi_bus_offline, (void *)false,
 				     (void **)&errdev);
@@ -241,6 +257,23 @@ static int acpi_scan_hot_remove(struct a
 			return -EBUSY;
 		}
 	}
+	return 0;
+}
+
+static int acpi_scan_hot_remove(struct acpi_device *device)
+{
+	acpi_handle handle = device->handle;
+	unsigned long long sta;
+	acpi_status status;
+
+	if (device->handler->hotplug.demand_offline && !acpi_force_hot_remove) {
+		if (!acpi_scan_is_offline(device))
+			return -EBUSY;
+	} else {
+		int error = acpi_scan_try_to_offline(device);
+		if (error)
+			return error;
+	}
 
 	ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 		"Hot-removing device %s...\n", dev_name(&device->dev)));
include/acpi/acpi_bus.h +2 -1
Index: linux-pm/include/acpi/acpi_bus.h
===================================================================
@@ -91,8 +91,9 @@ struct acpi_device;
 
 struct acpi_hotplug_profile {
 	struct kobject kobj;
-	bool enabled:1;
 	int (*scan_dependent)(struct acpi_device *adev);
+	bool enabled:1;
+	bool demand_offline:1;
 };
 
 static inline struct acpi_hotplug_profile *to_acpi_hotplug_profile(
#19 lore raw

Re: [PATCH 1/2][Untested] ACPI / hotplug: Add demand_offline hotplug profile flag

Quoted 33 lines · Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Add a new ACPI hotplug profile flag, demand_offline, such that if set for the given ACPI device object's scan handler, it will cause acpi_scan_hot_remove() to check if that device object's physical companions are offline upfront and fail the hot removal if that is not the case.

That flag will be useful to overcome a problem with containers on some system where they can only be hot-removed after some cleanup operations carried out by user space, which needs to be notified of the container hot-removal before the kernel attempts to offline devices in the container. In those cases the current implementation of acpi_scan_hot_remove() is not sufficient, because it first tries to offline the devices in the container and only if that is suffcessful it tries to offline the container itself. As a result, the container hot-removal notification is not delivered to user space at the right time.

---
  drivers/acpi/scan.c     |   41 +++++++++++++++++++++++++++++++++++++----
  include/acpi/acpi_bus.h |    3 ++-
  2 files changed, 39 insertions(+), 5 deletions(-)

Index: linux-pm/drivers/acpi/scan.c
===================================================================

drivers/acpi/scan.c +0 -0
@@ -126,6 +126,24 @@ acpi_device_modalias_show(struct device
  }
  static DEVICE_ATTR(modalias, 0444, acpi_device_modalias_show, NULL);
 

drivers/acpi/scan.c acpi_device_modalias_show(struct device · line 138

Quoted 10 lines
drivers/acpi/scan.c +9 -0
+static bool acpi_scan_is_offline(struct acpi_device *adev)
+{
+	struct acpi_device_physical_node *pn;
+	bool offline = true;
+
+	mutex_lock(&adev->physical_node_lock);
+
+	list_for_each_entry(pn, &adev->physical_node_list, node)
 
+		if (!pn->dev->offline) {

Please check pn->dev->bus and pn->dev->bus->offline too as follow:

		if (pn->dev->bus && pn->dev->bus->offline &&
		    !pn->dev->offline) {

My container has CPU and Memory and PCI root bridge. PCI root bridge does not has offline function (pn->dev->bus->offline). So I cannot offline the device and pn->dev->offline of the device is always 0. Therefore, following operation always returns -EBUSY even if I offline CPUs and all memory sections on a container device.

echo 0 > /sys/bus/container/devices/ACPI0004:01/online

Thanks,
Yasuaki Ishimatsu

Quoted 75 lines
drivers/acpi/scan.c +9 -0
+			kobject_uevent(&pn->dev->kobj, KOBJ_CHANGE);
+			offline = false;
+			break;
+		}
+
+	mutex_unlock(&adev->physical_node_lock);
+	return offline;
+}
+
  static acpi_status acpi_bus_offline(acpi_handle handle, u32 lvl, void *data,
  				    void **ret_p)
  {
drivers/acpi/scan.c +19 -4
@@ -196,12 +214,11 @@ static acpi_status acpi_bus_online(acpi_
  	return AE_OK;
  }
 
-static int acpi_scan_hot_remove(struct acpi_device *device)
+static int acpi_scan_try_to_offline(struct acpi_device *device)
  {
  	acpi_handle handle = device->handle;
-	struct device *errdev;
+	struct device *errdev = NULL;
  	acpi_status status;
-	unsigned long long sta;
 
  	/*
  	 * Carry out two passes here and ignore errors in the first pass,
@@ -212,7 +229,6 @@ static int acpi_scan_hot_remove(struct a
  	 *
  	 * If the first pass is successful, the second one isn't needed, though.
  	 */
-	errdev = NULL;
  	status = acpi_walk_namespace(ACPI_TYPE_ANY, handle, ACPI_UINT32_MAX,
  				     NULL, acpi_bus_offline, (void *)false,
  				     (void **)&errdev);
@@ -241,6 +257,23 @@ static int acpi_scan_hot_remove(struct a
  			return -EBUSY;
  		}
  	}
+	return 0;
+}
+
+static int acpi_scan_hot_remove(struct acpi_device *device)
+{
+	acpi_handle handle = device->handle;
+	unsigned long long sta;
+	acpi_status status;
+
+	if (device->handler->hotplug.demand_offline && !acpi_force_hot_remove) {
+		if (!acpi_scan_is_offline(device))
+			return -EBUSY;
+	} else {
+		int error = acpi_scan_try_to_offline(device);
+		if (error)
+			return error;
+	}
 
  	ACPI_DEBUG_PRINT((ACPI_DB_INFO,
  		"Hot-removing device %s...\n", dev_name(&device->dev)));

Index: linux-pm/include/acpi/acpi_bus.h
===================================================================

include/acpi/acpi_bus.h +2 -1
@@ -91,8 +91,9 @@ struct acpi_device;
 
  struct acpi_hotplug_profile {
  	struct kobject kobj;
-	bool enabled:1;
  	int (*scan_dependent)(struct acpi_device *adev);
+	bool enabled:1;
+	bool demand_offline:1;
  };
 
  static inline struct acpi_hotplug_profile *to_acpi_hotplug_profile(
#20 lore raw

Re: [PATCH 1/2][Untested] ACPI / hotplug: Add demand_offline hotplug profile flag

Quoted 43 lines · Yasuaki Ishimatsu
Quoted 41 lines · Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Add a new ACPI hotplug profile flag, demand_offline, such that if set for the given ACPI device object's scan handler, it will cause acpi_scan_hot_remove() to check if that device object's physical companions are offline upfront and fail the hot removal if that is not the case.

That flag will be useful to overcome a problem with containers on some system where they can only be hot-removed after some cleanup operations carried out by user space, which needs to be notified of the container hot-removal before the kernel attempts to offline devices in the container. In those cases the current implementation of acpi_scan_hot_remove() is not sufficient, because it first tries to offline the devices in the container and only if that is suffcessful it tries to offline the container itself. As a result, the container hot-removal notification is not delivered to user space at the right time.

---
  drivers/acpi/scan.c     |   41 +++++++++++++++++++++++++++++++++++++----
  include/acpi/acpi_bus.h |    3 ++-
  2 files changed, 39 insertions(+), 5 deletions(-)

Index: linux-pm/drivers/acpi/scan.c
===================================================================

drivers/acpi/scan.c +8 -0
@@ -126,6 +126,24 @@ acpi_device_modalias_show(struct device
  }
  static DEVICE_ATTR(modalias, 0444, acpi_device_modalias_show, NULL);
 
+static bool acpi_scan_is_offline(struct acpi_device *adev)
+{
+    struct acpi_device_physical_node *pn;
+    bool offline = true;
+
+    mutex_lock(&adev->physical_node_lock);
+
+    list_for_each_entry(pn, &adev->physical_node_list, node)
Quoted 8 lines

drivers/acpi/scan.c acpi_device_modalias_show(struct device · line 137

Quoted 1 line
drivers/acpi/scan.c +1 -0
+        if (!pn->dev->offline) {

Please check pn->dev->bus and pn->dev->bus->offline too as follow:

         if (pn->dev->bus && pn->dev->bus->offline &&
             !pn->dev->offline) {

Adding above check, I could remove container device by using eject sysfs. But following messages were shown:

[ 1017.543000] ------------[ cut here ]------------
[ 1017.543000] WARNING: CPU: 0 PID: 6 at drivers/base/core.c:251 device_release+0x92/0xa0()
[ 1017.543000] Device 'ACPI0004:01' does not have a release() function, it is broken and must be fixed.
[ 1017.653000] Modules linked in: xt_CHECKSUM nf_conntrack_netbios_ns nf_conntrack_broadcast ipt_MASQUERADE ip6t_REJECT ipmi_devintf ipt_REJECT cfg80211 xt_conntrack rfkill ebtable_nat ebtable_broute bridge stp llc ebtable_filter ebtables ip6table_nat nf_conntrack_ipv6 nf_defrag_ipv6 nf_nat_ipv6 ip6table_mangle ip6table_security ip6table_raw ip6table_filter ip6_tables iptable_nat nf_conntrack_ipv4 nf_defrag_ipv4 nf_nat_ipv4 nf_nat nf_conntrack iptable_mangle iptable_security iptable_raw iptable_filter sg ip_tables vfat fat x86_pkg_temp_thermal coretemp iTCO_wdt iTCO_vendor_support kvm_intel kvm crct10dif_pclmul crc32_pclmul crc32c_intel ghash_clmulni_intel aesni_intel lrw gf128mul glue_helper ablk_helper cryptd dm_mirror dm_region_hash dm_log dm_mod microcode lpc_ich igb sb_edac e1000e pcspkr i2c_i801
[ 1017.653000]  edac_core mfd_core dca ptp pps_core shpchp ipmi_si ipmi_msghandler tpm_infineon nfsd auth_rpcgss nfs_acl lockd sunrpc xfs libcrc32c sd_mod mgag200 syscopyarea sysfillrect sysimgblt lpfc i2c_algo_bit drm_kms_helper ttm drm crc_t10dif crct10dif_common scsi_transport_fc megaraid_sas i2c_core scsi_tgt
[ 1017.653000] CPU: 0 PID: 6 Comm: kworker/u512:0 Tainted: G        W    3.13.0-rc5+ #5
[ 1017.653000] Hardware name:
[ 1017.653000] Workqueue: kacpi_hotplug acpi_hotplug_work_fn
[ 1017.653000]  0000000000000009 ffff880873a6dc68 ffffffff815d85ca ffff880873a6dcb0
[ 1017.653000]  ffff880873a6dca0 ffffffff8106594d ffff8a07d221c010 ffff8a07d221c000
[ 1017.653000]  ffff8808715472c0 ffff880871e91018 0000000000000103 ffff880873a6dd00
[ 1017.653000] Call Trace:
[ 1017.653000]  [<ffffffff815d85ca>] dump_stack+0x45/0x56
[ 1017.653000]  [<ffffffff8106594d>] warn_slowpath_common+0x7d/0xa0
[ 1017.653000]  [<ffffffff810659bc>] warn_slowpath_fmt+0x4c/0x50
[ 1017.653000]  [<ffffffff813ad892>] device_release+0x92/0xa0
[ 1017.653000]  [<ffffffff812b7197>] kobject_cleanup+0x77/0x1b0
[ 1017.653000]  [<ffffffff812b7045>] kobject_put+0x35/0x70
[ 1017.653000]  [<ffffffff813ae38c>] device_unregister+0x2c/0x60
[ 1017.653000]  [<ffffffff8134c87c>] container_device_detach+0x28/0x2a
[ 1017.653000]  [<ffffffff81323096>] acpi_bus_trim+0x56/0x89
[ 1017.653000]  [<ffffffff813246ae>] acpi_device_hotplug+0x168/0x383
[ 1017.653000]  [<ffffffff8131efba>] acpi_hotplug_work_fn+0x1c/0x27
[ 1017.653000]  [<ffffffff81080f1b>] process_one_work+0x17b/0x460
[ 1017.653000]  [<ffffffff81081ccb>] worker_thread+0x11b/0x400
[ 1017.653000]  [<ffffffff81081bb0>] ? rescuer_thread+0x3e0/0x3e0
[ 1017.653000]  [<ffffffff81088a12>] kthread+0xd2/0xf0
[ 1017.653000]  [<ffffffff81088940>] ? kthread_create_on_node+0x180/0x180
[ 1017.653000]  [<ffffffff815e82fc>] ret_from_fork+0x7c/0xb0
[ 1017.653000]  [<ffffffff81088940>] ? kthread_create_on_node+0x180/0x180
[ 1017.653000] ---[ end trace 41394323eb4b690a ]---

Thanks,
Yasuaki Ishimatsu

Quoted 87 lines

My container has CPU and Memory and PCI root bridge. PCI root bridge does not has offline function (pn->dev->bus->offline). So I cannot offline the device and pn->dev->offline of the device is always 0. Therefore, following operation always returns -EBUSY even if I offline CPUs and all memory sections on a container device.

echo 0 > /sys/bus/container/devices/ACPI0004:01/online

Thanks,
Yasuaki Ishimatsu

Quoted 75 lines
drivers/acpi/scan.c +28 -4
+            kobject_uevent(&pn->dev->kobj, KOBJ_CHANGE);
+            offline = false;
+            break;
+        }
+
+    mutex_unlock(&adev->physical_node_lock);
+    return offline;
+}
+
  static acpi_status acpi_bus_offline(acpi_handle handle, u32 lvl, void *data,
                      void **ret_p)
  {
@@ -196,12 +214,11 @@ static acpi_status acpi_bus_online(acpi_
      return AE_OK;
  }
 
-static int acpi_scan_hot_remove(struct acpi_device *device)
+static int acpi_scan_try_to_offline(struct acpi_device *device)
  {
      acpi_handle handle = device->handle;
-    struct device *errdev;
+    struct device *errdev = NULL;
      acpi_status status;
-    unsigned long long sta;
 
      /*
       * Carry out two passes here and ignore errors in the first pass,
@@ -212,7 +229,6 @@ static int acpi_scan_hot_remove(struct a
       *
       * If the first pass is successful, the second one isn't needed, though.
       */
-    errdev = NULL;
      status = acpi_walk_namespace(ACPI_TYPE_ANY, handle, ACPI_UINT32_MAX,
                       NULL, acpi_bus_offline, (void *)false,
                       (void **)&errdev);
@@ -241,6 +257,23 @@ static int acpi_scan_hot_remove(struct a
              return -EBUSY;
          }
      }
+    return 0;
+}
+
+static int acpi_scan_hot_remove(struct acpi_device *device)
+{
+    acpi_handle handle = device->handle;
+    unsigned long long sta;
+    acpi_status status;
+
+    if (device->handler->hotplug.demand_offline && !acpi_force_hot_remove) {
+        if (!acpi_scan_is_offline(device))
+            return -EBUSY;
+    } else {
+        int error = acpi_scan_try_to_offline(device);
+        if (error)
+            return error;
+    }
 
      ACPI_DEBUG_PRINT((ACPI_DB_INFO,
          "Hot-removing device %s...\n", dev_name(&device->dev)));

Index: linux-pm/include/acpi/acpi_bus.h
===================================================================

include/acpi/acpi_bus.h +2 -1
@@ -91,8 +91,9 @@ struct acpi_device;
 
  struct acpi_hotplug_profile {
      struct kobject kobj;
-    bool enabled:1;
      int (*scan_dependent)(struct acpi_device *adev);
+    bool enabled:1;
+    bool demand_offline:1;
  };
 
  static inline struct acpi_hotplug_profile *to_acpi_hotplug_profile(
#21 lore raw

Re: [PATCH 1/2][Untested] ACPI / hotplug: Add demand_offline hotplug profile flag

Quoted 47 lines · Yasuaki Ishimatsu
Quoted 45 lines · Yasuaki Ishimatsu
Quoted 43 lines · Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Add a new ACPI hotplug profile flag, demand_offline, such that if set for the given ACPI device object's scan handler, it will cause acpi_scan_hot_remove() to check if that device object's physical companions are offline upfront and fail the hot removal if that is not the case.

That flag will be useful to overcome a problem with containers on some system where they can only be hot-removed after some cleanup operations carried out by user space, which needs to be notified of the container hot-removal before the kernel attempts to offline devices in the container. In those cases the current implementation of acpi_scan_hot_remove() is not sufficient, because it first tries to offline the devices in the container and only if that is suffcessful it tries to offline the container itself. As a result, the container hot-removal notification is not delivered to user space at the right time.

---
  drivers/acpi/scan.c     |   41 +++++++++++++++++++++++++++++++++++++----
  include/acpi/acpi_bus.h |    3 ++-
  2 files changed, 39 insertions(+), 5 deletions(-)

Index: linux-pm/drivers/acpi/scan.c
===================================================================

drivers/acpi/scan.c +9 -0
@@ -126,6 +126,24 @@ acpi_device_modalias_show(struct device
  }
  static DEVICE_ATTR(modalias, 0444, acpi_device_modalias_show, NULL);
 
+static bool acpi_scan_is_offline(struct acpi_device *adev)
+{
+    struct acpi_device_physical_node *pn;
+    bool offline = true;
+
+    mutex_lock(&adev->physical_node_lock);
+
+    list_for_each_entry(pn, &adev->physical_node_list, node)
 
+        if (!pn->dev->offline) {
Quoted 8 lines
Quoted 5 lines

Please check pn->dev->bus and pn->dev->bus->offline too as follow:

         if (pn->dev->bus && pn->dev->bus->offline &&
             !pn->dev->offline) {

Adding above check, I could remove container device by using eject sysfs. But following messages were shown:

Well, it looks like I have overlooked that error during my testing.

Quoted 26 lines
[ 1017.543000] ------------[ cut here ]------------
[ 1017.543000] WARNING: CPU: 0 PID: 6 at drivers/base/core.c:251 device_release+0x92/0xa0()
[ 1017.543000] Device 'ACPI0004:01' does not have a release() function, it is broken and must be fixed.
[ 1017.653000] Modules linked in: xt_CHECKSUM nf_conntrack_netbios_ns nf_conntrack_broadcast ipt_MASQUERADE ip6t_REJECT ipmi_devintf ipt_REJECT cfg80211 xt_conntrack rfkill ebtable_nat ebtable_broute bridge stp llc ebtable_filter ebtables ip6table_nat nf_conntrack_ipv6 nf_defrag_ipv6 nf_nat_ipv6 
ip6table_mangle ip6table_security ip6table_raw ip6table_filter ip6_tables iptable_nat nf_conntrack_ipv4 nf_defrag_ipv4 nf_nat_ipv4 nf_nat nf_conntrack iptable_mangle iptable_security iptable_raw iptable_filter sg ip_tables vfat fat x86_pkg_temp_thermal coretemp iTCO_wdt iTCO_vendor_support kvm_intel 
kvm crct10dif_pclmul crc32_pclmul crc32c_intel ghash_clmulni_intel aesni_intel lrw gf128mul glue_helper ablk_helper cryptd dm_mirror dm_region_hash dm_log dm_mod microcode lpc_ich igb sb_edac e1000e pcspkr i2c_i801
[ 1017.653000]  edac_core mfd_core dca ptp pps_core shpchp ipmi_si ipmi_msghandler tpm_infineon nfsd auth_rpcgss nfs_acl lockd sunrpc xfs libcrc32c sd_mod mgag200 syscopyarea sysfillrect sysimgblt lpfc i2c_algo_bit drm_kms_helper ttm drm crc_t10dif crct10dif_common scsi_transport_fc megaraid_sas 
i2c_core scsi_tgt
[ 1017.653000] CPU: 0 PID: 6 Comm: kworker/u512:0 Tainted: G        W    3.13.0-rc5+ #5
[ 1017.653000] Hardware name:
[ 1017.653000] Workqueue: kacpi_hotplug acpi_hotplug_work_fn
[ 1017.653000]  0000000000000009 ffff880873a6dc68 ffffffff815d85ca ffff880873a6dcb0
[ 1017.653000]  ffff880873a6dca0 ffffffff8106594d ffff8a07d221c010 ffff8a07d221c000
[ 1017.653000]  ffff8808715472c0 ffff880871e91018 0000000000000103 ffff880873a6dd00
[ 1017.653000] Call Trace:
[ 1017.653000]  [<ffffffff815d85ca>] dump_stack+0x45/0x56
[ 1017.653000]  [<ffffffff8106594d>] warn_slowpath_common+0x7d/0xa0
[ 1017.653000]  [<ffffffff810659bc>] warn_slowpath_fmt+0x4c/0x50
[ 1017.653000]  [<ffffffff813ad892>] device_release+0x92/0xa0
[ 1017.653000]  [<ffffffff812b7197>] kobject_cleanup+0x77/0x1b0
[ 1017.653000]  [<ffffffff812b7045>] kobject_put+0x35/0x70
[ 1017.653000]  [<ffffffff813ae38c>] device_unregister+0x2c/0x60
[ 1017.653000]  [<ffffffff8134c87c>] container_device_detach+0x28/0x2a
[ 1017.653000]  [<ffffffff81323096>] acpi_bus_trim+0x56/0x89
[ 1017.653000]  [<ffffffff813246ae>] acpi_device_hotplug+0x168/0x383
[ 1017.653000]  [<ffffffff8131efba>] acpi_hotplug_work_fn+0x1c/0x27
Quoted 8 lines
[ 1017.653000]  [<ffffffff81080f1b>] process_one_work+0x17b/0x460
[ 1017.653000]  [<ffffffff81081ccb>] worker_thread+0x11b/0x400
[ 1017.653000]  [<ffffffff81081bb0>] ? rescuer_thread+0x3e0/0x3e0
[ 1017.653000]  [<ffffffff81088a12>] kthread+0xd2/0xf0
[ 1017.653000]  [<ffffffff81088940>] ? kthread_create_on_node+0x180/0x180
[ 1017.653000]  [<ffffffff815e82fc>] ret_from_fork+0x7c/0xb0
[ 1017.653000]  [<ffffffff81088940>] ? kthread_create_on_node+0x180/0x180
[ 1017.653000] ---[ end trace 41394323eb4b690a ]---

Below is an updated version of patch [2/2] that should fix this problem (and the other one with the PCI host bridge not supporting offline too).

Thanks,
Rafael

---
From: Rafael J. Wysocki <[email protected]>
Subject: ACPI / hotplug / driver core: Handle containers in a special way

ACPI container devices require special hotplug handling, at least on some systems, since generally user space needs to carry out system-specific cleanup before it makes sense to offline devices in the container. However, the current ACPI hotplug code for containers first attempts to offline devices in the container and only then it notifies user space of the container offline.

Moreover, after commit 202317a573b2 (ACPI / scan: Add acpi_device objects for all device nodes in the namespace), ACPI device objects representing containers are present as long as the ACPI namespace nodes corresponding to them are present, which may be forever, even if the container devices are physically detached from the system (the return values of the corresponding _STA methods change in those cases, but generally the namespace nodes themselves are still there). Thus it is useful to introduce entities representing containers that will go away during container hot-unplug.

The goal of this change is to address both the above issues.

The idea is to create a "companion" container system device for each of the ACPI container device objects during the initial namespace scan or on a hotplug event making the container present. That system device will be unregistered on container removal. A new bus type for container devices is added for this purpose, because device offline and online operations need to be defined for them. The online operation is a trivial function that is always successful and the offline uses a callback pointed to by the container device's offline member.

For ACPI containers that callback simply walks the list of ACPI device objects right below the container object (its children) and checks if all of their physical companion devices are offline. If that's not the case, it returns -EBUSY and the container system devivce cannot be put offline. Consequently, to put the container system device offline, it is necessary to put all of the physical devices depending on its ACPI companion object offline beforehand.

Container system devices created for ACPI container objects are initially online. They are created by the container ACPI scan handler whose hotplug.demand_offline flag is set. That causes acpi_scan_hot_remove() to check if the companion container system device is offline before attempting to remove an ACPI container or any devices below it. If the check fails, a KOBJ_CHANGE uevent is emitted for the container system device in question and user space is expected to offline all devices below the container and the container itself in response to it. Then, user space can finalize the removal of the container with the help of its ACPI device object's eject attribute in sysfs.

---
 drivers/acpi/container.c  |   48 ++++++++++++++++++++++++++++++++++++++++++----
 drivers/acpi/internal.h   |    1 
 drivers/acpi/scan.c       |   10 +++++----
 drivers/base/Makefile     |    2 -
 drivers/base/base.h       |    1 
 drivers/base/container.c  |   44 ++++++++++++++++++++++++++++++++++++++++++
 drivers/base/init.c       |    1 
 include/linux/container.h |   25 +++++++++++++++++++++++
 8 files changed, 123 insertions(+), 9 deletions(-)
include/linux/container.h new file +25 -0
Index: linux-pm/include/linux/container.h
===================================================================
@@ -0,0 +1,25 @@
+/*
+ * Definitions for container bus type.
+ *
+ * Copyright (C) 2013, Intel Corporation
+ * Author: Rafael J. Wysocki <[email protected]>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include <linux/device.h>
+
+/* drivers/base/power/container.c */
+extern struct bus_type container_subsys;
+
+struct container_dev {
+	struct device dev;
+	int (*offline)(struct container_dev *cdev);
+};
+
+static inline struct container_dev *to_container_dev(struct device *dev)
+{
+	return container_of(dev, struct container_dev, dev);
+}
drivers/base/container.c new file +44 -0

44 lines not on this page · raw message

drivers/base/base.h +1 -0
Index: linux-pm/drivers/base/base.h
===================================================================
@@ -100,6 +100,7 @@ static inline int hypervisor_init(void)
 #endif
 extern int platform_bus_init(void);
 extern void cpu_dev_init(void);
+extern void container_dev_init(void);
 
 struct kobject *virtual_device_parent(struct device *dev);
 
drivers/base/init.c +1 -0
Index: linux-pm/drivers/base/init.c
===================================================================
@@ -33,4 +33,5 @@ void __init driver_init(void)
 	platform_bus_init();
 	cpu_dev_init();
 	memory_dev_init();
+	container_dev_init();
 }
drivers/base/Makefile +1 -1
Index: linux-pm/drivers/base/Makefile
===================================================================
@@ -4,7 +4,7 @@ obj-y			:= core.o bus.o dd.o syscore.o \
 			   driver.o class.o platform.o \
 			   cpu.o firmware.o init.o map.o devres.o \
 			   attribute_container.o transport_class.o \
-			   topology.o
+			   topology.o container.o
 obj-$(CONFIG_DEVTMPFS)	+= devtmpfs.o
 obj-$(CONFIG_DMA_CMA) += dma-contiguous.o
 obj-y			+= power/
drivers/acpi/internal.h +1 -0
Index: linux-pm/drivers/acpi/internal.h
===================================================================
@@ -73,6 +73,7 @@ static inline void acpi_lpss_init(void)
 #endif
 
 bool acpi_queue_hotplug_work(struct work_struct *work);
+bool acpi_scan_is_offline(struct acpi_device *adev, bool uevent);
 
 /* --------------------------------------------------------------------------
                      Device Node Initialization / Removal
drivers/acpi/scan.c +6 -4
Index: linux-pm/drivers/acpi/scan.c
===================================================================
@@ -126,7 +126,7 @@ acpi_device_modalias_show(struct device
 }
 static DEVICE_ATTR(modalias, 0444, acpi_device_modalias_show, NULL);
 
-static bool acpi_scan_is_offline(struct acpi_device *adev)
+bool acpi_scan_is_offline(struct acpi_device *adev, bool uevent)
 {
 	struct acpi_device_physical_node *pn;
 	bool offline = true;
@@ -134,8 +134,10 @@ static bool acpi_scan_is_offline(struct
 	mutex_lock(&adev->physical_node_lock);
 
 	list_for_each_entry(pn, &adev->physical_node_list, node)
-		if (!pn->dev->offline) {
-			kobject_uevent(&pn->dev->kobj, KOBJ_CHANGE);
+		if (device_supports_offline(pn->dev) && !pn->dev->offline) {
+			if (uevent)
+				kobject_uevent(&pn->dev->kobj, KOBJ_CHANGE);
+
 			offline = false;
 			break;
 		}
@@ -267,7 +269,7 @@ static int acpi_scan_hot_remove(struct a
 	acpi_status status;
 
 	if (device->handler->hotplug.demand_offline && !acpi_force_hot_remove) {
-		if (!acpi_scan_is_offline(device))
+		if (!acpi_scan_is_offline(device, true))
 			return -EBUSY;
 	} else {
 		int error = acpi_scan_try_to_offline(device);
drivers/acpi/container.c +44 -4

74 lines not on this page · raw message

#22 lore raw

Re: [PATCH 1/2][Untested] ACPI / hotplug: Add demand_offline hotplug profile flag

Quoted 94 lines · Rafael J. Wysocki
Quoted 47 lines · Yasuaki Ishimatsu
Quoted 45 lines · Yasuaki Ishimatsu
Quoted 43 lines · Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Add a new ACPI hotplug profile flag, demand_offline, such that if set for the given ACPI device object's scan handler, it will cause acpi_scan_hot_remove() to check if that device object's physical companions are offline upfront and fail the hot removal if that is not the case.

That flag will be useful to overcome a problem with containers on some system where they can only be hot-removed after some cleanup operations carried out by user space, which needs to be notified of the container hot-removal before the kernel attempts to offline devices in the container. In those cases the current implementation of acpi_scan_hot_remove() is not sufficient, because it first tries to offline the devices in the container and only if that is suffcessful it tries to offline the container itself. As a result, the container hot-removal notification is not delivered to user space at the right time.

---
   drivers/acpi/scan.c     |   41 +++++++++++++++++++++++++++++++++++++----
   include/acpi/acpi_bus.h |    3 ++-
   2 files changed, 39 insertions(+), 5 deletions(-)

Index: linux-pm/drivers/acpi/scan.c
===================================================================

drivers/acpi/scan.c +9 -0
@@ -126,6 +126,24 @@ acpi_device_modalias_show(struct device
   }
   static DEVICE_ATTR(modalias, 0444, acpi_device_modalias_show, NULL);
 
+static bool acpi_scan_is_offline(struct acpi_device *adev)
+{
+    struct acpi_device_physical_node *pn;
+    bool offline = true;
+
+    mutex_lock(&adev->physical_node_lock);
+
+    list_for_each_entry(pn, &adev->physical_node_list, node)
 
+        if (!pn->dev->offline) {
Quoted 8 lines
Quoted 5 lines

Please check pn->dev->bus and pn->dev->bus->offline too as follow:

          if (pn->dev->bus && pn->dev->bus->offline &&
              !pn->dev->offline) {

Adding above check, I could remove container device by using eject sysfs. But following messages were shown:

Well, it looks like I have overlooked that error during my testing.

Quoted 26 lines
[ 1017.543000] ------------[ cut here ]------------
[ 1017.543000] WARNING: CPU: 0 PID: 6 at drivers/base/core.c:251 device_release+0x92/0xa0()
[ 1017.543000] Device 'ACPI0004:01' does not have a release() function, it is broken and must be fixed.
[ 1017.653000] Modules linked in: xt_CHECKSUM nf_conntrack_netbios_ns nf_conntrack_broadcast ipt_MASQUERADE ip6t_REJECT ipmi_devintf ipt_REJECT cfg80211 xt_conntrack rfkill ebtable_nat ebtable_broute bridge stp llc ebtable_filter ebtables ip6table_nat nf_conntrack_ipv6 nf_defrag_ipv6 nf_nat_ipv6
ip6table_mangle ip6table_security ip6table_raw ip6table_filter ip6_tables iptable_nat nf_conntrack_ipv4 nf_defrag_ipv4 nf_nat_ipv4 nf_nat nf_conntrack iptable_mangle iptable_security iptable_raw iptable_filter sg ip_tables vfat fat x86_pkg_temp_thermal coretemp iTCO_wdt iTCO_vendor_support kvm_intel
kvm crct10dif_pclmul crc32_pclmul crc32c_intel ghash_clmulni_intel aesni_intel lrw gf128mul glue_helper ablk_helper cryptd dm_mirror dm_region_hash dm_log dm_mod microcode lpc_ich igb sb_edac e1000e pcspkr i2c_i801
[ 1017.653000]  edac_core mfd_core dca ptp pps_core shpchp ipmi_si ipmi_msghandler tpm_infineon nfsd auth_rpcgss nfs_acl lockd sunrpc xfs libcrc32c sd_mod mgag200 syscopyarea sysfillrect sysimgblt lpfc i2c_algo_bit drm_kms_helper ttm drm crc_t10dif crct10dif_common scsi_transport_fc megaraid_sas
i2c_core scsi_tgt
[ 1017.653000] CPU: 0 PID: 6 Comm: kworker/u512:0 Tainted: G        W    3.13.0-rc5+ #5
[ 1017.653000] Hardware name:
[ 1017.653000] Workqueue: kacpi_hotplug acpi_hotplug_work_fn
[ 1017.653000]  0000000000000009 ffff880873a6dc68 ffffffff815d85ca ffff880873a6dcb0
[ 1017.653000]  ffff880873a6dca0 ffffffff8106594d ffff8a07d221c010 ffff8a07d221c000
[ 1017.653000]  ffff8808715472c0 ffff880871e91018 0000000000000103 ffff880873a6dd00
[ 1017.653000] Call Trace:
[ 1017.653000]  [<ffffffff815d85ca>] dump_stack+0x45/0x56
[ 1017.653000]  [<ffffffff8106594d>] warn_slowpath_common+0x7d/0xa0
[ 1017.653000]  [<ffffffff810659bc>] warn_slowpath_fmt+0x4c/0x50
[ 1017.653000]  [<ffffffff813ad892>] device_release+0x92/0xa0
[ 1017.653000]  [<ffffffff812b7197>] kobject_cleanup+0x77/0x1b0
[ 1017.653000]  [<ffffffff812b7045>] kobject_put+0x35/0x70
[ 1017.653000]  [<ffffffff813ae38c>] device_unregister+0x2c/0x60
[ 1017.653000]  [<ffffffff8134c87c>] container_device_detach+0x28/0x2a
[ 1017.653000]  [<ffffffff81323096>] acpi_bus_trim+0x56/0x89
[ 1017.653000]  [<ffffffff813246ae>] acpi_device_hotplug+0x168/0x383
[ 1017.653000]  [<ffffffff8131efba>] acpi_hotplug_work_fn+0x1c/0x27
Quoted 8 lines
[ 1017.653000]  [<ffffffff81080f1b>] process_one_work+0x17b/0x460
[ 1017.653000]  [<ffffffff81081ccb>] worker_thread+0x11b/0x400
[ 1017.653000]  [<ffffffff81081bb0>] ? rescuer_thread+0x3e0/0x3e0
[ 1017.653000]  [<ffffffff81088a12>] kthread+0xd2/0xf0
[ 1017.653000]  [<ffffffff81088940>] ? kthread_create_on_node+0x180/0x180
[ 1017.653000]  [<ffffffff815e82fc>] ret_from_fork+0x7c/0xb0
[ 1017.653000]  [<ffffffff81088940>] ? kthread_create_on_node+0x180/0x180
[ 1017.653000] ---[ end trace 41394323eb4b690a ]---
Quoted 2 lines

Below is an updated version of patch [2/2] that should fix this problem (and the other one with the PCI host bridge not supporting offline too).

By updated patch, I can offline the container device and the above messages disappeared.

BTW, when I hot remove PCI root bridge, following messages were shown:

...

[  116.758000] ------------[ cut here ]------------
[  116.758000] WARNING: CPU: 0 PID: 6 at fs/sysfs/group.c:214 sysfs_remove_group+0xc6/0xd0()
[  116.758000] sysfs group ffffffff819ac5c0 not found for kobject '0001:ff:10.2'
[  116.758000] Modules linked in: xt_CHECKSUM nf_conntrack_netbios_ns nf_conntrack_broadcast ipt_MASQUERADE ip6t_REJECT ipmi_devintf ipt_REJECT xt_conntrack cfg80211 rfkill ebtable_nat ebtable_broute bridge stp llc ebtable_filter ebtables ip6table_nat nf_conntrack_ipv6 nf_defrag_ipv6 nf_nat_ipv6 ip6table_mangle ip6table_security ip6table_raw ip6table_filter ip6_tables iptable_nat nf_conntrack_ipv4 nf_defrag_ipv4 nf_nat_ipv4 nf_nat nf_conntrack iptable_mangle iptable_security iptable_raw iptable_filter ip_tables sg vfat fat x86_pkg_temp_thermal coretemp kvm_intel kvm crct10dif_pclmul crc32_pclmul crc32c_intel ghash_clmulni_intel aesni_intel lrw gf128mul glue_helper ablk_helper cryptd dm_mirror dm_region_hash e1000e dm_log dm_mod iTCO_wdt iTCO_vendor_support microcode sb_edac igb pcspkr edac_core i2c_i801
[  116.758000]  ptp lpc_ich pps_core dca mfd_core shpchp ipmi_si tpm_infineon ipmi_msghandler nfsd auth_rpcgss nfs_acl lockd sunrpc xfs libcrc32c sd_mod mgag200 syscopyarea sysfillrect sysimgblt i2c_algo_bit drm_kms_helper lpfc ttm drm crc_t10dif crct10dif_common scsi_transport_fc megaraid_sas i2c_core scsi_tgt
[  116.758000] CPU: 0 PID: 6 Comm: kworker/u512:0 Tainted: G        W    3.13.0-rc5+ #11
[  116.758000] Hardware name:
[  116.758000] Workqueue: kacpi_hotplug acpi_hotplug_work_fn
[  116.758000]  0000000000000009 ffff8808738d3bd8 ffffffff815d84ea ffff8808738d3c20
[  116.758000]  ffff8808738d3c10 ffffffff8106594d 0000000000000000 ffffffff819ac5c0
[  116.758000]  ffff880871b9d0a8 ffff8a07d1895000 0000000000000103 ffff8808738d3c70
[  116.758000] Call Trace:
[  116.758000]  [<ffffffff815d84ea>] dump_stack+0x45/0x56
[  116.758000]  [<ffffffff8106594d>] warn_slowpath_common+0x7d/0xa0
[  116.758000]  [<ffffffff810659bc>] warn_slowpath_fmt+0x4c/0x50
[  116.758000]  [<ffffffff8122b52e>] ? sysfs_get_dirent_ns+0x4e/0x70
[  116.758000]  [<ffffffff8122c806>] sysfs_remove_group+0xc6/0xd0
[  116.758000]  [<ffffffff813b83f3>] dpm_sysfs_remove+0x43/0x50
[  116.758000]  [<ffffffff813ae105>] device_del+0x45/0x1c0
[  116.758000]  [<ffffffff812e51f6>] pci_remove_bus_device+0x66/0xd0
[  116.758000]  [<ffffffff812e5363>] pci_remove_root_bus+0x73/0x80
[  116.758000]  [<ffffffff813276ab>] acpi_pci_root_remove+0x42/0x4f
[  116.758000]  [<ffffffff81323070>] acpi_bus_trim+0x56/0x89
[  116.758000]  [<ffffffff81323052>] acpi_bus_trim+0x38/0x89
[  116.758000]  [<ffffffff813245df>] acpi_device_hotplug+0x137/0x33b
[  116.758000]  [<ffffffff8131efba>] acpi_hotplug_work_fn+0x1c/0x27
[  116.758000]  [<ffffffff81080f1b>] process_one_work+0x17b/0x460
[  116.758000]  [<ffffffff81081ccb>] worker_thread+0x11b/0x400
[  116.758000]  [<ffffffff81081bb0>] ? rescuer_thread+0x3e0/0x3e0
[  116.758000]  [<ffffffff81088a12>] kthread+0xd2/0xf0
[  116.758000]  [<ffffffff81088940>] ? kthread_create_on_node+0x180/0x180
[  116.758000]  [<ffffffff815e823c>] ret_from_fork+0x7c/0xb0
[  116.758000]  [<ffffffff81088940>] ? kthread_create_on_node+0x180/0x180
[  116.758000] ---[ end trace b403db9d0ec9fc9e ]---

...

It is a know issue? The message are shown when we use latest bleeding-edge.

Thanks,
Yasuaki Ishimatsu

Quoted 312 lines

Thanks,
Rafael

---
From: Rafael J. Wysocki <[email protected]>
Subject: ACPI / hotplug / driver core: Handle containers in a special way

ACPI container devices require special hotplug handling, at least on some systems, since generally user space needs to carry out system-specific cleanup before it makes sense to offline devices in the container. However, the current ACPI hotplug code for containers first attempts to offline devices in the container and only then it notifies user space of the container offline.

Moreover, after commit 202317a573b2 (ACPI / scan: Add acpi_device objects for all device nodes in the namespace), ACPI device objects representing containers are present as long as the ACPI namespace nodes corresponding to them are present, which may be forever, even if the container devices are physically detached from the system (the return values of the corresponding _STA methods change in those cases, but generally the namespace nodes themselves are still there). Thus it is useful to introduce entities representing containers that will go away during container hot-unplug.

The goal of this change is to address both the above issues.

The idea is to create a "companion" container system device for each of the ACPI container device objects during the initial namespace scan or on a hotplug event making the container present. That system device will be unregistered on container removal. A new bus type for container devices is added for this purpose, because device offline and online operations need to be defined for them. The online operation is a trivial function that is always successful and the offline uses a callback pointed to by the container device's offline member.

For ACPI containers that callback simply walks the list of ACPI device objects right below the container object (its children) and checks if all of their physical companion devices are offline. If that's not the case, it returns -EBUSY and the container system devivce cannot be put offline. Consequently, to put the container system device offline, it is necessary to put all of the physical devices depending on its ACPI companion object offline beforehand.

Container system devices created for ACPI container objects are initially online. They are created by the container ACPI scan handler whose hotplug.demand_offline flag is set. That causes acpi_scan_hot_remove() to check if the companion container system device is offline before attempting to remove an ACPI container or any devices below it. If the check fails, a KOBJ_CHANGE uevent is emitted for the container system device in question and user space is expected to offline all devices below the container and the container itself in response to it. Then, user space can finalize the removal of the container with the help of its ACPI device object's eject attribute in sysfs.

---
  drivers/acpi/container.c  |   48 ++++++++++++++++++++++++++++++++++++++++++----
  drivers/acpi/internal.h   |    1
  drivers/acpi/scan.c       |   10 +++++----
  drivers/base/Makefile     |    2 -
  drivers/base/base.h       |    1
  drivers/base/container.c  |   44 ++++++++++++++++++++++++++++++++++++++++++
  drivers/base/init.c       |    1
  include/linux/container.h |   25 +++++++++++++++++++++++
  8 files changed, 123 insertions(+), 9 deletions(-)

Index: linux-pm/include/linux/container.h
===================================================================

include/linux/container.h new file +25 -0
@@ -0,0 +1,25 @@
+/*
+ * Definitions for container bus type.
+ *
+ * Copyright (C) 2013, Intel Corporation
+ * Author: Rafael J. Wysocki <[email protected]>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include <linux/device.h>
+
+/* drivers/base/power/container.c */
+extern struct bus_type container_subsys;
+
+struct container_dev {
+	struct device dev;
+	int (*offline)(struct container_dev *cdev);
+};
+
+static inline struct container_dev *to_container_dev(struct device *dev)
+{
+	return container_of(dev, struct container_dev, dev);
+}

Index: linux-pm/drivers/base/container.c
===================================================================

drivers/base/container.c new file +44 -0
@@ -0,0 +1,44 @@
+/*
+ * System bus type for containers.
+ *
+ * Copyright (C) 2013, Intel Corporation
+ * Author: Rafael J. Wysocki <[email protected]>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include <linux/container.h>
+
+#include "base.h"
+
+#define CONTAINER_BUS_NAME	"container"
+
+static int trivial_online(struct device *dev)
+{
+	return 0;
+}
+
+static int container_offline(struct device *dev)
+{
+	struct container_dev *cdev = to_container_dev(dev);
+
+	return cdev->offline ? cdev->offline(cdev) : 0;
+}
+
+struct bus_type container_subsys = {
+	.name = CONTAINER_BUS_NAME,
+	.dev_name = CONTAINER_BUS_NAME,
+	.online = trivial_online,
+	.offline = container_offline,
+};
+
+void __init container_dev_init(void)
+{
+	int ret;
+
+	ret = subsys_system_register(&container_subsys, NULL);
+	if (ret)
+		pr_err("%s() failed: %d\n", __func__, ret);
+}

Index: linux-pm/drivers/base/base.h
===================================================================

drivers/base/base.h +1 -0
@@ -100,6 +100,7 @@ static inline int hypervisor_init(void)
  #endif
  extern int platform_bus_init(void);
  extern void cpu_dev_init(void);
+extern void container_dev_init(void);
 
  struct kobject *virtual_device_parent(struct device *dev);
 

Index: linux-pm/drivers/base/init.c
===================================================================

drivers/base/init.c +1 -0
@@ -33,4 +33,5 @@ void __init driver_init(void)
  	platform_bus_init();
  	cpu_dev_init();
  	memory_dev_init();
+	container_dev_init();
  }

Index: linux-pm/drivers/base/Makefile
===================================================================

drivers/base/Makefile +1 -1
@@ -4,7 +4,7 @@ obj-y			:= core.o bus.o dd.o syscore.o \
  			   driver.o class.o platform.o \
  			   cpu.o firmware.o init.o map.o devres.o \
  			   attribute_container.o transport_class.o \
-			   topology.o
+			   topology.o container.o
  obj-$(CONFIG_DEVTMPFS)	+= devtmpfs.o
  obj-$(CONFIG_DMA_CMA) += dma-contiguous.o
  obj-y			+= power/

Index: linux-pm/drivers/acpi/internal.h
===================================================================

drivers/acpi/internal.h +1 -0
@@ -73,6 +73,7 @@ static inline void acpi_lpss_init(void)
  #endif
 
  bool acpi_queue_hotplug_work(struct work_struct *work);
+bool acpi_scan_is_offline(struct acpi_device *adev, bool uevent);
 
  /* --------------------------------------------------------------------------
                       Device Node Initialization / Removal

Index: linux-pm/drivers/acpi/scan.c
===================================================================

drivers/acpi/scan.c +6 -4
@@ -126,7 +126,7 @@ acpi_device_modalias_show(struct device
  }
  static DEVICE_ATTR(modalias, 0444, acpi_device_modalias_show, NULL);
 
-static bool acpi_scan_is_offline(struct acpi_device *adev)
+bool acpi_scan_is_offline(struct acpi_device *adev, bool uevent)
  {
  	struct acpi_device_physical_node *pn;
  	bool offline = true;
@@ -134,8 +134,10 @@ static bool acpi_scan_is_offline(struct
  	mutex_lock(&adev->physical_node_lock);
 
  	list_for_each_entry(pn, &adev->physical_node_list, node)
-		if (!pn->dev->offline) {
-			kobject_uevent(&pn->dev->kobj, KOBJ_CHANGE);
+		if (device_supports_offline(pn->dev) && !pn->dev->offline) {
+			if (uevent)
+				kobject_uevent(&pn->dev->kobj, KOBJ_CHANGE);
+
  			offline = false;
  			break;
  		}
@@ -267,7 +269,7 @@ static int acpi_scan_hot_remove(struct a
  	acpi_status status;
 
  	if (device->handler->hotplug.demand_offline && !acpi_force_hot_remove) {
-		if (!acpi_scan_is_offline(device))
+		if (!acpi_scan_is_offline(device, true))
  			return -EBUSY;
  	} else {
  		int error = acpi_scan_try_to_offline(device);

Index: linux-pm/drivers/acpi/container.c
===================================================================

drivers/acpi/container.c +44 -4
@@ -27,8 +27,7 @@
   * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   */
  #include <linux/acpi.h>
-
-#include "internal.h"
+#include <linux/container.h>
 
  #include "internal.h"
 
@@ -44,16 +43,56 @@ static const struct acpi_device_id conta
  	{"", 0},
  };
 
+static int acpi_container_offline(struct container_dev *cdev)
+{
+	struct acpi_device *adev = ACPI_COMPANION(&cdev->dev);
+	struct acpi_device *child;
+
+	/* Check all of the dependent devices' physical companions. */
+	list_for_each_entry(child, &adev->children, node)
+		if (!acpi_scan_is_offline(child, false))
+			return -EBUSY;
+
+	return 0;
+}
+
+static void acpi_container_release(struct device *dev)
+{
+	kfree(to_container_dev(dev));
+}
+
  static int container_device_attach(struct acpi_device *adev,
  				   const struct acpi_device_id *not_used)
  {
-	kobject_uevent(&adev->dev.kobj, KOBJ_ONLINE);
+	struct container_dev *cdev;
+	struct device *dev;
+	int ret;
+
+	cdev = kzalloc(sizeof(*cdev), GFP_KERNEL);
+	if (!cdev)
+		return -ENOMEM;
+
+	cdev->offline = acpi_container_offline;
+	dev = &cdev->dev;
+	dev->bus = &container_subsys;
+	dev_set_name(dev, "%s", dev_name(&adev->dev));
+	ACPI_COMPANION_SET(dev, adev);
+	dev->release = acpi_container_release;
+	ret = device_register(dev);
+	if (ret)
+		return ret;
+
+	adev->driver_data = dev;
  	return 1;
  }
 
  static void container_device_detach(struct acpi_device *adev)
  {
-	kobject_uevent(&adev->dev.kobj, KOBJ_OFFLINE);
+	struct device *dev = acpi_driver_data(adev);
+
+	adev->driver_data = NULL;
+	if (dev)
+		device_unregister(dev);
  }
 
  static struct acpi_scan_handler container_handler = {
@@ -62,6 +101,7 @@ static struct acpi_scan_handler containe
  	.detach = container_device_detach,
  	.hotplug = {
  		.enabled = true,
+		.demand_offline = true,
  	},
  };
 
#23 lore raw

Re: [PATCH 1/2][Untested] ACPI / hotplug: Add demand_offline hotplug profile flag

Quoted 141 lines · Yasuaki Ishimatsu
Quoted 94 lines · Rafael J. Wysocki
Quoted 47 lines · Yasuaki Ishimatsu
Quoted 45 lines · Yasuaki Ishimatsu
Quoted 43 lines · Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Add a new ACPI hotplug profile flag, demand_offline, such that if set for the given ACPI device object's scan handler, it will cause acpi_scan_hot_remove() to check if that device object's physical companions are offline upfront and fail the hot removal if that is not the case.

That flag will be useful to overcome a problem with containers on some system where they can only be hot-removed after some cleanup operations carried out by user space, which needs to be notified of the container hot-removal before the kernel attempts to offline devices in the container. In those cases the current implementation of acpi_scan_hot_remove() is not sufficient, because it first tries to offline the devices in the container and only if that is suffcessful it tries to offline the container itself. As a result, the container hot-removal notification is not delivered to user space at the right time.

---
   drivers/acpi/scan.c     |   41 +++++++++++++++++++++++++++++++++++++----
   include/acpi/acpi_bus.h |    3 ++-
   2 files changed, 39 insertions(+), 5 deletions(-)

Index: linux-pm/drivers/acpi/scan.c
===================================================================

drivers/acpi/scan.c +9 -0
@@ -126,6 +126,24 @@ acpi_device_modalias_show(struct device
   }
   static DEVICE_ATTR(modalias, 0444, acpi_device_modalias_show, NULL);
 
+static bool acpi_scan_is_offline(struct acpi_device *adev)
+{
+    struct acpi_device_physical_node *pn;
+    bool offline = true;
+
+    mutex_lock(&adev->physical_node_lock);
+
+    list_for_each_entry(pn, &adev->physical_node_list, node)
 
+        if (!pn->dev->offline) {
Quoted 8 lines
Quoted 5 lines

Please check pn->dev->bus and pn->dev->bus->offline too as follow:

          if (pn->dev->bus && pn->dev->bus->offline &&
              !pn->dev->offline) {

Adding above check, I could remove container device by using eject sysfs. But following messages were shown:

Well, it looks like I have overlooked that error during my testing.

Quoted 26 lines
[ 1017.543000] ------------[ cut here ]------------
[ 1017.543000] WARNING: CPU: 0 PID: 6 at drivers/base/core.c:251 device_release+0x92/0xa0()
[ 1017.543000] Device 'ACPI0004:01' does not have a release() function, it is broken and must be fixed.
[ 1017.653000] Modules linked in: xt_CHECKSUM nf_conntrack_netbios_ns nf_conntrack_broadcast ipt_MASQUERADE ip6t_REJECT ipmi_devintf ipt_REJECT cfg80211 xt_conntrack rfkill ebtable_nat ebtable_broute bridge stp llc ebtable_filter ebtables ip6table_nat nf_conntrack_ipv6 nf_defrag_ipv6 nf_nat_ipv6
ip6table_mangle ip6table_security ip6table_raw ip6table_filter ip6_tables iptable_nat nf_conntrack_ipv4 nf_defrag_ipv4 nf_nat_ipv4 nf_nat nf_conntrack iptable_mangle iptable_security iptable_raw iptable_filter sg ip_tables vfat fat x86_pkg_temp_thermal coretemp iTCO_wdt iTCO_vendor_support kvm_intel
kvm crct10dif_pclmul crc32_pclmul crc32c_intel ghash_clmulni_intel aesni_intel lrw gf128mul glue_helper ablk_helper cryptd dm_mirror dm_region_hash dm_log dm_mod microcode lpc_ich igb sb_edac e1000e pcspkr i2c_i801
[ 1017.653000]  edac_core mfd_core dca ptp pps_core shpchp ipmi_si ipmi_msghandler tpm_infineon nfsd auth_rpcgss nfs_acl lockd sunrpc xfs libcrc32c sd_mod mgag200 syscopyarea sysfillrect sysimgblt lpfc i2c_algo_bit drm_kms_helper ttm drm crc_t10dif crct10dif_common scsi_transport_fc megaraid_sas
i2c_core scsi_tgt
[ 1017.653000] CPU: 0 PID: 6 Comm: kworker/u512:0 Tainted: G        W    3.13.0-rc5+ #5
[ 1017.653000] Hardware name:
[ 1017.653000] Workqueue: kacpi_hotplug acpi_hotplug_work_fn
[ 1017.653000]  0000000000000009 ffff880873a6dc68 ffffffff815d85ca ffff880873a6dcb0
[ 1017.653000]  ffff880873a6dca0 ffffffff8106594d ffff8a07d221c010 ffff8a07d221c000
[ 1017.653000]  ffff8808715472c0 ffff880871e91018 0000000000000103 ffff880873a6dd00
[ 1017.653000] Call Trace:
[ 1017.653000]  [<ffffffff815d85ca>] dump_stack+0x45/0x56
[ 1017.653000]  [<ffffffff8106594d>] warn_slowpath_common+0x7d/0xa0
[ 1017.653000]  [<ffffffff810659bc>] warn_slowpath_fmt+0x4c/0x50
[ 1017.653000]  [<ffffffff813ad892>] device_release+0x92/0xa0
[ 1017.653000]  [<ffffffff812b7197>] kobject_cleanup+0x77/0x1b0
[ 1017.653000]  [<ffffffff812b7045>] kobject_put+0x35/0x70
[ 1017.653000]  [<ffffffff813ae38c>] device_unregister+0x2c/0x60
[ 1017.653000]  [<ffffffff8134c87c>] container_device_detach+0x28/0x2a
[ 1017.653000]  [<ffffffff81323096>] acpi_bus_trim+0x56/0x89
[ 1017.653000]  [<ffffffff813246ae>] acpi_device_hotplug+0x168/0x383
[ 1017.653000]  [<ffffffff8131efba>] acpi_hotplug_work_fn+0x1c/0x27
Quoted 8 lines
[ 1017.653000]  [<ffffffff81080f1b>] process_one_work+0x17b/0x460
[ 1017.653000]  [<ffffffff81081ccb>] worker_thread+0x11b/0x400
[ 1017.653000]  [<ffffffff81081bb0>] ? rescuer_thread+0x3e0/0x3e0
[ 1017.653000]  [<ffffffff81088a12>] kthread+0xd2/0xf0
[ 1017.653000]  [<ffffffff81088940>] ? kthread_create_on_node+0x180/0x180
[ 1017.653000]  [<ffffffff815e82fc>] ret_from_fork+0x7c/0xb0
[ 1017.653000]  [<ffffffff81088940>] ? kthread_create_on_node+0x180/0x180
[ 1017.653000] ---[ end trace 41394323eb4b690a ]---
Quoted 2 lines

Below is an updated version of patch [2/2] that should fix this problem (and the other one with the PCI host bridge not supporting offline too).

By updated patch, I can offline the container device and the above messages disappeared.

BTW, when I hot remove PCI root bridge, following messages were shown:

...

[  116.758000] ------------[ cut here ]------------
[  116.758000] WARNING: CPU: 0 PID: 6 at fs/sysfs/group.c:214 sysfs_remove_group+0xc6/0xd0()
[  116.758000] sysfs group ffffffff819ac5c0 not found for kobject '0001:ff:10.2'
[  116.758000] Modules linked in: xt_CHECKSUM nf_conntrack_netbios_ns nf_conntrack_broadcast ipt_MASQUERADE ip6t_REJECT ipmi_devintf ipt_REJECT xt_conntrack cfg80211 rfkill ebtable_nat ebtable_broute bridge stp llc ebtable_filter ebtables ip6table_nat nf_conntrack_ipv6 nf_defrag_ipv6 nf_nat_ipv6
ip6table_mangle ip6table_security ip6table_raw ip6table_filter ip6_tables iptable_nat nf_conntrack_ipv4 nf_defrag_ipv4 nf_nat_ipv4 nf_nat nf_conntrack iptable_mangle iptable_security iptable_raw iptable_filter ip_tables sg vfat fat x86_pkg_temp_thermal coretemp kvm_intel kvm crct10dif_pclmul
crc32_pclmul crc32c_intel ghash_clmulni_intel aesni_intel lrw gf128mul glue_helper ablk_helper cryptd dm_mirror dm_region_hash e1000e dm_log dm_mod iTCO_wdt iTCO_vendor_support microcode sb_edac igb pcspkr edac_core i2c_i801
[  116.758000]  ptp lpc_ich pps_core dca mfd_core shpchp ipmi_si tpm_infineon ipmi_msghandler nfsd auth_rpcgss nfs_acl lockd sunrpc xfs libcrc32c sd_mod mgag200 syscopyarea sysfillrect sysimgblt i2c_algo_bit drm_kms_helper lpfc ttm drm crc_t10dif crct10dif_common scsi_transport_fc megaraid_sas
i2c_core scsi_tgt
[  116.758000] CPU: 0 PID: 6 Comm: kworker/u512:0 Tainted: G        W    3.13.0-rc5+ #11
[  116.758000] Hardware name:
[  116.758000] Workqueue: kacpi_hotplug acpi_hotplug_work_fn
[  116.758000]  0000000000000009 ffff8808738d3bd8 ffffffff815d84ea ffff8808738d3c20
[  116.758000]  ffff8808738d3c10 ffffffff8106594d 0000000000000000 ffffffff819ac5c0
[  116.758000]  ffff880871b9d0a8 ffff8a07d1895000 0000000000000103 ffff8808738d3c70
[  116.758000] Call Trace:
[  116.758000]  [<ffffffff815d84ea>] dump_stack+0x45/0x56
[  116.758000]  [<ffffffff8106594d>] warn_slowpath_common+0x7d/0xa0
[  116.758000]  [<ffffffff810659bc>] warn_slowpath_fmt+0x4c/0x50
[  116.758000]  [<ffffffff8122b52e>] ? sysfs_get_dirent_ns+0x4e/0x70
[  116.758000]  [<ffffffff8122c806>] sysfs_remove_group+0xc6/0xd0
[  116.758000]  [<ffffffff813b83f3>] dpm_sysfs_remove+0x43/0x50
[  116.758000]  [<ffffffff813ae105>] device_del+0x45/0x1c0
[  116.758000]  [<ffffffff812e51f6>] pci_remove_bus_device+0x66/0xd0
[  116.758000]  [<ffffffff812e5363>] pci_remove_root_bus+0x73/0x80
[  116.758000]  [<ffffffff813276ab>] acpi_pci_root_remove+0x42/0x4f
[  116.758000]  [<ffffffff81323070>] acpi_bus_trim+0x56/0x89
[  116.758000]  [<ffffffff81323052>] acpi_bus_trim+0x38/0x89
[  116.758000]  [<ffffffff813245df>] acpi_device_hotplug+0x137/0x33b
[  116.758000]  [<ffffffff8131efba>] acpi_hotplug_work_fn+0x1c/0x27
[  116.758000]  [<ffffffff81080f1b>] process_one_work+0x17b/0x460
[  116.758000]  [<ffffffff81081ccb>] worker_thread+0x11b/0x400
[  116.758000]  [<ffffffff81081bb0>] ? rescuer_thread+0x3e0/0x3e0
[  116.758000]  [<ffffffff81088a12>] kthread+0xd2/0xf0
[  116.758000]  [<ffffffff81088940>] ? kthread_create_on_node+0x180/0x180
[  116.758000]  [<ffffffff815e823c>] ret_from_fork+0x7c/0xb0
[  116.758000]  [<ffffffff81088940>] ? kthread_create_on_node+0x180/0x180
[  116.758000] ---[ end trace b403db9d0ec9fc9e ]---
Quoted 6 lines

...

It is a know issue? The message are shown when we use latest bleeding-edge.

Thanks,
Yasuaki Ishimatsu

I'll have new year vacation starting tomorrow. Thus I'll review and test the updated patch from 6 Jan.

Thanks,
Yasuaki Ishimatsu

Quoted 317 lines
Quoted 316 lines

Thanks,
Rafael

---
From: Rafael J. Wysocki <[email protected]>
Subject: ACPI / hotplug / driver core: Handle containers in a special way

ACPI container devices require special hotplug handling, at least on some systems, since generally user space needs to carry out system-specific cleanup before it makes sense to offline devices in the container. However, the current ACPI hotplug code for containers first attempts to offline devices in the container and only then it notifies user space of the container offline.

Moreover, after commit 202317a573b2 (ACPI / scan: Add acpi_device objects for all device nodes in the namespace), ACPI device objects representing containers are present as long as the ACPI namespace nodes corresponding to them are present, which may be forever, even if the container devices are physically detached from the system (the return values of the corresponding _STA methods change in those cases, but generally the namespace nodes themselves are still there). Thus it is useful to introduce entities representing containers that will go away during container hot-unplug.

The goal of this change is to address both the above issues.

The idea is to create a "companion" container system device for each of the ACPI container device objects during the initial namespace scan or on a hotplug event making the container present. That system device will be unregistered on container removal. A new bus type for container devices is added for this purpose, because device offline and online operations need to be defined for them. The online operation is a trivial function that is always successful and the offline uses a callback pointed to by the container device's offline member.

For ACPI containers that callback simply walks the list of ACPI device objects right below the container object (its children) and checks if all of their physical companion devices are offline. If that's not the case, it returns -EBUSY and the container system devivce cannot be put offline. Consequently, to put the container system device offline, it is necessary to put all of the physical devices depending on its ACPI companion object offline beforehand.

Container system devices created for ACPI container objects are initially online. They are created by the container ACPI scan handler whose hotplug.demand_offline flag is set. That causes acpi_scan_hot_remove() to check if the companion container system device is offline before attempting to remove an ACPI container or any devices below it. If the check fails, a KOBJ_CHANGE uevent is emitted for the container system device in question and user space is expected to offline all devices below the container and the container itself in response to it. Then, user space can finalize the removal of the container with the help of its ACPI device object's eject attribute in sysfs.

---
  drivers/acpi/container.c  |   48 ++++++++++++++++++++++++++++++++++++++++++----
  drivers/acpi/internal.h   |    1
  drivers/acpi/scan.c       |   10 +++++----
  drivers/base/Makefile     |    2 -
  drivers/base/base.h       |    1
  drivers/base/container.c  |   44 ++++++++++++++++++++++++++++++++++++++++++
  drivers/base/init.c       |    1
  include/linux/container.h |   25 +++++++++++++++++++++++
  8 files changed, 123 insertions(+), 9 deletions(-)

Index: linux-pm/include/linux/container.h
===================================================================

include/linux/container.h new file +25 -0
@@ -0,0 +1,25 @@
+/*
+ * Definitions for container bus type.
+ *
+ * Copyright (C) 2013, Intel Corporation
+ * Author: Rafael J. Wysocki <[email protected]>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include <linux/device.h>
+
+/* drivers/base/power/container.c */
+extern struct bus_type container_subsys;
+
+struct container_dev {
+    struct device dev;
+    int (*offline)(struct container_dev *cdev);
+};
+
+static inline struct container_dev *to_container_dev(struct device *dev)
+{
+    return container_of(dev, struct container_dev, dev);
+}

Index: linux-pm/drivers/base/container.c
===================================================================

drivers/base/container.c new file +44 -0
@@ -0,0 +1,44 @@
+/*
+ * System bus type for containers.
+ *
+ * Copyright (C) 2013, Intel Corporation
+ * Author: Rafael J. Wysocki <[email protected]>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include <linux/container.h>
+
+#include "base.h"
+
+#define CONTAINER_BUS_NAME    "container"
+
+static int trivial_online(struct device *dev)
+{
+    return 0;
+}
+
+static int container_offline(struct device *dev)
+{
+    struct container_dev *cdev = to_container_dev(dev);
+
+    return cdev->offline ? cdev->offline(cdev) : 0;
+}
+
+struct bus_type container_subsys = {
+    .name = CONTAINER_BUS_NAME,
+    .dev_name = CONTAINER_BUS_NAME,
+    .online = trivial_online,
+    .offline = container_offline,
+};
+
+void __init container_dev_init(void)
+{
+    int ret;
+
+    ret = subsys_system_register(&container_subsys, NULL);
+    if (ret)
+        pr_err("%s() failed: %d\n", __func__, ret);
+}

Index: linux-pm/drivers/base/base.h
===================================================================

drivers/base/base.h +1 -0
@@ -100,6 +100,7 @@ static inline int hypervisor_init(void)
  #endif
  extern int platform_bus_init(void);
  extern void cpu_dev_init(void);
+extern void container_dev_init(void);
 
  struct kobject *virtual_device_parent(struct device *dev);
 

Index: linux-pm/drivers/base/init.c
===================================================================

drivers/base/init.c +1 -0
@@ -33,4 +33,5 @@ void __init driver_init(void)
      platform_bus_init();
      cpu_dev_init();
      memory_dev_init();
+    container_dev_init();
  }

Index: linux-pm/drivers/base/Makefile
===================================================================

drivers/base/Makefile +1 -1
@@ -4,7 +4,7 @@ obj-y            := core.o bus.o dd.o syscore.o \
                 driver.o class.o platform.o \
                 cpu.o firmware.o init.o map.o devres.o \
                 attribute_container.o transport_class.o \
-               topology.o
+               topology.o container.o
  obj-$(CONFIG_DEVTMPFS)    += devtmpfs.o
  obj-$(CONFIG_DMA_CMA) += dma-contiguous.o
  obj-y            += power/

Index: linux-pm/drivers/acpi/internal.h
===================================================================

drivers/acpi/internal.h +1 -0
@@ -73,6 +73,7 @@ static inline void acpi_lpss_init(void)
  #endif
 
  bool acpi_queue_hotplug_work(struct work_struct *work);
+bool acpi_scan_is_offline(struct acpi_device *adev, bool uevent);
 
  /* --------------------------------------------------------------------------
                       Device Node Initialization / Removal

Index: linux-pm/drivers/acpi/scan.c
===================================================================

drivers/acpi/scan.c +6 -4
@@ -126,7 +126,7 @@ acpi_device_modalias_show(struct device
  }
  static DEVICE_ATTR(modalias, 0444, acpi_device_modalias_show, NULL);
 
-static bool acpi_scan_is_offline(struct acpi_device *adev)
+bool acpi_scan_is_offline(struct acpi_device *adev, bool uevent)
  {
      struct acpi_device_physical_node *pn;
      bool offline = true;
@@ -134,8 +134,10 @@ static bool acpi_scan_is_offline(struct
      mutex_lock(&adev->physical_node_lock);
 
      list_for_each_entry(pn, &adev->physical_node_list, node)
-        if (!pn->dev->offline) {
-            kobject_uevent(&pn->dev->kobj, KOBJ_CHANGE);
+        if (device_supports_offline(pn->dev) && !pn->dev->offline) {
+            if (uevent)
+                kobject_uevent(&pn->dev->kobj, KOBJ_CHANGE);
+
              offline = false;
              break;
          }
@@ -267,7 +269,7 @@ static int acpi_scan_hot_remove(struct a
      acpi_status status;
 
      if (device->handler->hotplug.demand_offline && !acpi_force_hot_remove) {
-        if (!acpi_scan_is_offline(device))
+        if (!acpi_scan_is_offline(device, true))
              return -EBUSY;
      } else {
          int error = acpi_scan_try_to_offline(device);

Index: linux-pm/drivers/acpi/container.c
===================================================================

drivers/acpi/container.c +44 -4
@@ -27,8 +27,7 @@
   * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   */
  #include <linux/acpi.h>
-
-#include "internal.h"
+#include <linux/container.h>
 
  #include "internal.h"
 
@@ -44,16 +43,56 @@ static const struct acpi_device_id conta
      {"", 0},
  };
 
+static int acpi_container_offline(struct container_dev *cdev)
+{
+    struct acpi_device *adev = ACPI_COMPANION(&cdev->dev);
+    struct acpi_device *child;
+
+    /* Check all of the dependent devices' physical companions. */
+    list_for_each_entry(child, &adev->children, node)
+        if (!acpi_scan_is_offline(child, false))
+            return -EBUSY;
+
+    return 0;
+}
+
+static void acpi_container_release(struct device *dev)
+{
+    kfree(to_container_dev(dev));
+}
+
  static int container_device_attach(struct acpi_device *adev,
                     const struct acpi_device_id *not_used)
  {
-    kobject_uevent(&adev->dev.kobj, KOBJ_ONLINE);
+    struct container_dev *cdev;
+    struct device *dev;
+    int ret;
+
+    cdev = kzalloc(sizeof(*cdev), GFP_KERNEL);
+    if (!cdev)
+        return -ENOMEM;
+
+    cdev->offline = acpi_container_offline;
+    dev = &cdev->dev;
+    dev->bus = &container_subsys;
+    dev_set_name(dev, "%s", dev_name(&adev->dev));
+    ACPI_COMPANION_SET(dev, adev);
+    dev->release = acpi_container_release;
+    ret = device_register(dev);
+    if (ret)
+        return ret;
+
+    adev->driver_data = dev;
      return 1;
  }
 
  static void container_device_detach(struct acpi_device *adev)
  {
-    kobject_uevent(&adev->dev.kobj, KOBJ_OFFLINE);
+    struct device *dev = acpi_driver_data(adev);
+
+    adev->driver_data = NULL;
+    if (dev)
+        device_unregister(dev);
  }
 
  static struct acpi_scan_handler container_handler = {
@@ -62,6 +101,7 @@ static struct acpi_scan_handler containe
      .detach = container_device_detach,
      .hotplug = {
          .enabled = true,
+        .demand_offline = true,
      },
  };
 
#24 lore raw

Re: [PATCH 1/2][Untested] ACPI / hotplug: Add demand_offline hotplug profile flag

Quoted 144 lines · Yasuaki Ishimatsu
Quoted 143 lines · Yasuaki Ishimatsu
Quoted 94 lines · Rafael J. Wysocki
Quoted 47 lines · Yasuaki Ishimatsu
Quoted 45 lines · Yasuaki Ishimatsu
Quoted 43 lines · Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Add a new ACPI hotplug profile flag, demand_offline, such that if set for the given ACPI device object's scan handler, it will cause acpi_scan_hot_remove() to check if that device object's physical companions are offline upfront and fail the hot removal if that is not the case.

That flag will be useful to overcome a problem with containers on some system where they can only be hot-removed after some cleanup operations carried out by user space, which needs to be notified of the container hot-removal before the kernel attempts to offline devices in the container. In those cases the current implementation of acpi_scan_hot_remove() is not sufficient, because it first tries to offline the devices in the container and only if that is suffcessful it tries to offline the container itself. As a result, the container hot-removal notification is not delivered to user space at the right time.

---
   drivers/acpi/scan.c     |   41 +++++++++++++++++++++++++++++++++++++----
   include/acpi/acpi_bus.h |    3 ++-
   2 files changed, 39 insertions(+), 5 deletions(-)

Index: linux-pm/drivers/acpi/scan.c
===================================================================

drivers/acpi/scan.c +9 -0
@@ -126,6 +126,24 @@ acpi_device_modalias_show(struct device
   }
   static DEVICE_ATTR(modalias, 0444, acpi_device_modalias_show, NULL);
 
+static bool acpi_scan_is_offline(struct acpi_device *adev)
+{
+    struct acpi_device_physical_node *pn;
+    bool offline = true;
+
+    mutex_lock(&adev->physical_node_lock);
+
+    list_for_each_entry(pn, &adev->physical_node_list, node)
 
+        if (!pn->dev->offline) {
Quoted 8 lines
Quoted 5 lines

Please check pn->dev->bus and pn->dev->bus->offline too as follow:

          if (pn->dev->bus && pn->dev->bus->offline &&
              !pn->dev->offline) {

Adding above check, I could remove container device by using eject sysfs. But following messages were shown:

Well, it looks like I have overlooked that error during my testing.

Quoted 26 lines
[ 1017.543000] ------------[ cut here ]------------
[ 1017.543000] WARNING: CPU: 0 PID: 6 at drivers/base/core.c:251 device_release+0x92/0xa0()
[ 1017.543000] Device 'ACPI0004:01' does not have a release() function, it is broken and must be fixed.
[ 1017.653000] Modules linked in: xt_CHECKSUM nf_conntrack_netbios_ns nf_conntrack_broadcast ipt_MASQUERADE ip6t_REJECT ipmi_devintf ipt_REJECT cfg80211 xt_conntrack rfkill ebtable_nat ebtable_broute bridge stp llc ebtable_filter ebtables ip6table_nat nf_conntrack_ipv6 nf_defrag_ipv6 nf_nat_ipv6
ip6table_mangle ip6table_security ip6table_raw ip6table_filter ip6_tables iptable_nat nf_conntrack_ipv4 nf_defrag_ipv4 nf_nat_ipv4 nf_nat nf_conntrack iptable_mangle iptable_security iptable_raw iptable_filter sg ip_tables vfat fat x86_pkg_temp_thermal coretemp iTCO_wdt iTCO_vendor_support kvm_intel
kvm crct10dif_pclmul crc32_pclmul crc32c_intel ghash_clmulni_intel aesni_intel lrw gf128mul glue_helper ablk_helper cryptd dm_mirror dm_region_hash dm_log dm_mod microcode lpc_ich igb sb_edac e1000e pcspkr i2c_i801
[ 1017.653000]  edac_core mfd_core dca ptp pps_core shpchp ipmi_si ipmi_msghandler tpm_infineon nfsd auth_rpcgss nfs_acl lockd sunrpc xfs libcrc32c sd_mod mgag200 syscopyarea sysfillrect sysimgblt lpfc i2c_algo_bit drm_kms_helper ttm drm crc_t10dif crct10dif_common scsi_transport_fc megaraid_sas
i2c_core scsi_tgt
[ 1017.653000] CPU: 0 PID: 6 Comm: kworker/u512:0 Tainted: G        W    3.13.0-rc5+ #5
[ 1017.653000] Hardware name:
[ 1017.653000] Workqueue: kacpi_hotplug acpi_hotplug_work_fn
[ 1017.653000]  0000000000000009 ffff880873a6dc68 ffffffff815d85ca ffff880873a6dcb0
[ 1017.653000]  ffff880873a6dca0 ffffffff8106594d ffff8a07d221c010 ffff8a07d221c000
[ 1017.653000]  ffff8808715472c0 ffff880871e91018 0000000000000103 ffff880873a6dd00
[ 1017.653000] Call Trace:
[ 1017.653000]  [<ffffffff815d85ca>] dump_stack+0x45/0x56
[ 1017.653000]  [<ffffffff8106594d>] warn_slowpath_common+0x7d/0xa0
[ 1017.653000]  [<ffffffff810659bc>] warn_slowpath_fmt+0x4c/0x50
[ 1017.653000]  [<ffffffff813ad892>] device_release+0x92/0xa0
[ 1017.653000]  [<ffffffff812b7197>] kobject_cleanup+0x77/0x1b0
[ 1017.653000]  [<ffffffff812b7045>] kobject_put+0x35/0x70
[ 1017.653000]  [<ffffffff813ae38c>] device_unregister+0x2c/0x60
[ 1017.653000]  [<ffffffff8134c87c>] container_device_detach+0x28/0x2a
[ 1017.653000]  [<ffffffff81323096>] acpi_bus_trim+0x56/0x89
[ 1017.653000]  [<ffffffff813246ae>] acpi_device_hotplug+0x168/0x383
[ 1017.653000]  [<ffffffff8131efba>] acpi_hotplug_work_fn+0x1c/0x27
Quoted 8 lines
[ 1017.653000]  [<ffffffff81080f1b>] process_one_work+0x17b/0x460
[ 1017.653000]  [<ffffffff81081ccb>] worker_thread+0x11b/0x400
[ 1017.653000]  [<ffffffff81081bb0>] ? rescuer_thread+0x3e0/0x3e0
[ 1017.653000]  [<ffffffff81088a12>] kthread+0xd2/0xf0
[ 1017.653000]  [<ffffffff81088940>] ? kthread_create_on_node+0x180/0x180
[ 1017.653000]  [<ffffffff815e82fc>] ret_from_fork+0x7c/0xb0
[ 1017.653000]  [<ffffffff81088940>] ? kthread_create_on_node+0x180/0x180
[ 1017.653000] ---[ end trace 41394323eb4b690a ]---
Quoted 2 lines

Below is an updated version of patch [2/2] that should fix this problem (and the other one with the PCI host bridge not supporting offline too).

By updated patch, I can offline the container device and the above messages disappeared.

BTW, when I hot remove PCI root bridge, following messages were shown:

...

[  116.758000] ------------[ cut here ]------------
[  116.758000] WARNING: CPU: 0 PID: 6 at fs/sysfs/group.c:214 sysfs_remove_group+0xc6/0xd0()
[  116.758000] sysfs group ffffffff819ac5c0 not found for kobject '0001:ff:10.2'
[  116.758000] Modules linked in: xt_CHECKSUM nf_conntrack_netbios_ns nf_conntrack_broadcast ipt_MASQUERADE ip6t_REJECT ipmi_devintf ipt_REJECT xt_conntrack cfg80211 rfkill ebtable_nat ebtable_broute bridge stp llc ebtable_filter ebtables ip6table_nat nf_conntrack_ipv6 nf_defrag_ipv6 nf_nat_ipv6
ip6table_mangle ip6table_security ip6table_raw ip6table_filter ip6_tables iptable_nat nf_conntrack_ipv4 nf_defrag_ipv4 nf_nat_ipv4 nf_nat nf_conntrack iptable_mangle iptable_security iptable_raw iptable_filter ip_tables sg vfat fat x86_pkg_temp_thermal coretemp kvm_intel kvm crct10dif_pclmul
crc32_pclmul crc32c_intel ghash_clmulni_intel aesni_intel lrw gf128mul glue_helper ablk_helper cryptd dm_mirror dm_region_hash e1000e dm_log dm_mod iTCO_wdt iTCO_vendor_support microcode sb_edac igb pcspkr edac_core i2c_i801
[  116.758000]  ptp lpc_ich pps_core dca mfd_core shpchp ipmi_si tpm_infineon ipmi_msghandler nfsd auth_rpcgss nfs_acl lockd sunrpc xfs libcrc32c sd_mod mgag200 syscopyarea sysfillrect sysimgblt i2c_algo_bit drm_kms_helper lpfc ttm drm crc_t10dif crct10dif_common scsi_transport_fc megaraid_sas
i2c_core scsi_tgt
[  116.758000] CPU: 0 PID: 6 Comm: kworker/u512:0 Tainted: G        W    3.13.0-rc5+ #11
[  116.758000] Hardware name:
[  116.758000] Workqueue: kacpi_hotplug acpi_hotplug_work_fn
[  116.758000]  0000000000000009 ffff8808738d3bd8 ffffffff815d84ea ffff8808738d3c20
[  116.758000]  ffff8808738d3c10 ffffffff8106594d 0000000000000000 ffffffff819ac5c0
[  116.758000]  ffff880871b9d0a8 ffff8a07d1895000 0000000000000103 ffff8808738d3c70
[  116.758000] Call Trace:
[  116.758000]  [<ffffffff815d84ea>] dump_stack+0x45/0x56
[  116.758000]  [<ffffffff8106594d>] warn_slowpath_common+0x7d/0xa0
[  116.758000]  [<ffffffff810659bc>] warn_slowpath_fmt+0x4c/0x50
[  116.758000]  [<ffffffff8122b52e>] ? sysfs_get_dirent_ns+0x4e/0x70
[  116.758000]  [<ffffffff8122c806>] sysfs_remove_group+0xc6/0xd0
[  116.758000]  [<ffffffff813b83f3>] dpm_sysfs_remove+0x43/0x50
[  116.758000]  [<ffffffff813ae105>] device_del+0x45/0x1c0
[  116.758000]  [<ffffffff812e51f6>] pci_remove_bus_device+0x66/0xd0
[  116.758000]  [<ffffffff812e5363>] pci_remove_root_bus+0x73/0x80
[  116.758000]  [<ffffffff813276ab>] acpi_pci_root_remove+0x42/0x4f
[  116.758000]  [<ffffffff81323070>] acpi_bus_trim+0x56/0x89
[  116.758000]  [<ffffffff81323052>] acpi_bus_trim+0x38/0x89
[  116.758000]  [<ffffffff813245df>] acpi_device_hotplug+0x137/0x33b
[  116.758000]  [<ffffffff8131efba>] acpi_hotplug_work_fn+0x1c/0x27
[  116.758000]  [<ffffffff81080f1b>] process_one_work+0x17b/0x460
[  116.758000]  [<ffffffff81081ccb>] worker_thread+0x11b/0x400
[  116.758000]  [<ffffffff81081bb0>] ? rescuer_thread+0x3e0/0x3e0
[  116.758000]  [<ffffffff81088a12>] kthread+0xd2/0xf0
[  116.758000]  [<ffffffff81088940>] ? kthread_create_on_node+0x180/0x180
[  116.758000]  [<ffffffff815e823c>] ret_from_fork+0x7c/0xb0
[  116.758000]  [<ffffffff81088940>] ? kthread_create_on_node+0x180/0x180
[  116.758000] ---[ end trace b403db9d0ec9fc9e ]---

...

Quoted 8 lines
Quoted 4 lines

It is a know issue? The message are shown when we use latest bleeding-edge.

Thanks,
Yasuaki Ishimatsu

I'll have new year vacation starting tomorrow. Thus I'll review and test the updated patch from 6 Jan.

Thanks a lot for the work already done and have a great vacation!

Rafael

#25 lore raw

Re: [PATCH 1/2][Untested] ACPI / hotplug: Add demand_offline hotplug profile flag

In reply to Yasuaki Ishimatsu
Quoted 95 lines · Yasuaki Ishimatsu
Quoted 94 lines · Rafael J. Wysocki
Quoted 47 lines · Yasuaki Ishimatsu
Quoted 45 lines · Yasuaki Ishimatsu
Quoted 43 lines · Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Add a new ACPI hotplug profile flag, demand_offline, such that if set for the given ACPI device object's scan handler, it will cause acpi_scan_hot_remove() to check if that device object's physical companions are offline upfront and fail the hot removal if that is not the case.

That flag will be useful to overcome a problem with containers on some system where they can only be hot-removed after some cleanup operations carried out by user space, which needs to be notified of the container hot-removal before the kernel attempts to offline devices in the container. In those cases the current implementation of acpi_scan_hot_remove() is not sufficient, because it first tries to offline the devices in the container and only if that is suffcessful it tries to offline the container itself. As a result, the container hot-removal notification is not delivered to user space at the right time.

---
   drivers/acpi/scan.c     |   41 +++++++++++++++++++++++++++++++++++++----
   include/acpi/acpi_bus.h |    3 ++-
   2 files changed, 39 insertions(+), 5 deletions(-)

Index: linux-pm/drivers/acpi/scan.c
===================================================================

drivers/acpi/scan.c +9 -0
@@ -126,6 +126,24 @@ acpi_device_modalias_show(struct device
   }
   static DEVICE_ATTR(modalias, 0444, acpi_device_modalias_show, NULL);
 
+static bool acpi_scan_is_offline(struct acpi_device *adev)
+{
+    struct acpi_device_physical_node *pn;
+    bool offline = true;
+
+    mutex_lock(&adev->physical_node_lock);
+
+    list_for_each_entry(pn, &adev->physical_node_list, node)
 
+        if (!pn->dev->offline) {
Quoted 8 lines
Quoted 5 lines

Please check pn->dev->bus and pn->dev->bus->offline too as follow:

          if (pn->dev->bus && pn->dev->bus->offline &&
              !pn->dev->offline) {

Adding above check, I could remove container device by using eject sysfs. But following messages were shown:

Well, it looks like I have overlooked that error during my testing.

Quoted 34 lines
[ 1017.543000] ------------[ cut here ]------------
[ 1017.543000] WARNING: CPU: 0 PID: 6 at drivers/base/core.c:251 device_release+0x92/0xa0()
[ 1017.543000] Device 'ACPI0004:01' does not have a release() function, it is broken and must be fixed.
[ 1017.653000] Modules linked in: xt_CHECKSUM nf_conntrack_netbios_ns nf_conntrack_broadcast ipt_MASQUERADE ip6t_REJECT ipmi_devintf ipt_REJECT cfg80211 xt_conntrack rfkill ebtable_nat ebtable_broute bridge stp llc ebtable_filter ebtables ip6table_nat nf_conntrack_ipv6 nf_defrag_ipv6 nf_nat_ipv6
ip6table_mangle ip6table_security ip6table_raw ip6table_filter ip6_tables iptable_nat nf_conntrack_ipv4 nf_defrag_ipv4 nf_nat_ipv4 nf_nat nf_conntrack iptable_mangle iptable_security iptable_raw iptable_filter sg ip_tables vfat fat x86_pkg_temp_thermal coretemp iTCO_wdt iTCO_vendor_support kvm_intel
kvm crct10dif_pclmul crc32_pclmul crc32c_intel ghash_clmulni_intel aesni_intel lrw gf128mul glue_helper ablk_helper cryptd dm_mirror dm_region_hash dm_log dm_mod microcode lpc_ich igb sb_edac e1000e pcspkr i2c_i801
[ 1017.653000]  edac_core mfd_core dca ptp pps_core shpchp ipmi_si ipmi_msghandler tpm_infineon nfsd auth_rpcgss nfs_acl lockd sunrpc xfs libcrc32c sd_mod mgag200 syscopyarea sysfillrect sysimgblt lpfc i2c_algo_bit drm_kms_helper ttm drm crc_t10dif crct10dif_common scsi_transport_fc megaraid_sas
i2c_core scsi_tgt
[ 1017.653000] CPU: 0 PID: 6 Comm: kworker/u512:0 Tainted: G        W    3.13.0-rc5+ #5
[ 1017.653000] Hardware name:
[ 1017.653000] Workqueue: kacpi_hotplug acpi_hotplug_work_fn
[ 1017.653000]  0000000000000009 ffff880873a6dc68 ffffffff815d85ca ffff880873a6dcb0
[ 1017.653000]  ffff880873a6dca0 ffffffff8106594d ffff8a07d221c010 ffff8a07d221c000
[ 1017.653000]  ffff8808715472c0 ffff880871e91018 0000000000000103 ffff880873a6dd00
[ 1017.653000] Call Trace:
[ 1017.653000]  [<ffffffff815d85ca>] dump_stack+0x45/0x56
[ 1017.653000]  [<ffffffff8106594d>] warn_slowpath_common+0x7d/0xa0
[ 1017.653000]  [<ffffffff810659bc>] warn_slowpath_fmt+0x4c/0x50
[ 1017.653000]  [<ffffffff813ad892>] device_release+0x92/0xa0
[ 1017.653000]  [<ffffffff812b7197>] kobject_cleanup+0x77/0x1b0
[ 1017.653000]  [<ffffffff812b7045>] kobject_put+0x35/0x70
[ 1017.653000]  [<ffffffff813ae38c>] device_unregister+0x2c/0x60
[ 1017.653000]  [<ffffffff8134c87c>] container_device_detach+0x28/0x2a
[ 1017.653000]  [<ffffffff81323096>] acpi_bus_trim+0x56/0x89
[ 1017.653000]  [<ffffffff813246ae>] acpi_device_hotplug+0x168/0x383
[ 1017.653000]  [<ffffffff8131efba>] acpi_hotplug_work_fn+0x1c/0x27
[ 1017.653000]  [<ffffffff81080f1b>] process_one_work+0x17b/0x460
[ 1017.653000]  [<ffffffff81081ccb>] worker_thread+0x11b/0x400
[ 1017.653000]  [<ffffffff81081bb0>] ? rescuer_thread+0x3e0/0x3e0
[ 1017.653000]  [<ffffffff81088a12>] kthread+0xd2/0xf0
[ 1017.653000]  [<ffffffff81088940>] ? kthread_create_on_node+0x180/0x180
[ 1017.653000]  [<ffffffff815e82fc>] ret_from_fork+0x7c/0xb0
[ 1017.653000]  [<ffffffff81088940>] ? kthread_create_on_node+0x180/0x180
[ 1017.653000] ---[ end trace 41394323eb4b690a ]---
Quoted 5 lines
Quoted 2 lines

Below is an updated version of patch [2/2] that should fix this problem (and the other one with the PCI host bridge not supporting offline too).

By updated patch, I can offline the container device and the above messages disappeared.

Awesome, thanks for testing!

Quoted 40 lines

BTW, when I hot remove PCI root bridge, following messages were shown:

...

[  116.758000] ------------[ cut here ]------------
[  116.758000] WARNING: CPU: 0 PID: 6 at fs/sysfs/group.c:214 sysfs_remove_group+0xc6/0xd0()
[  116.758000] sysfs group ffffffff819ac5c0 not found for kobject '0001:ff:10.2'
[  116.758000] Modules linked in: xt_CHECKSUM nf_conntrack_netbios_ns nf_conntrack_broadcast ipt_MASQUERADE ip6t_REJECT ipmi_devintf ipt_REJECT xt_conntrack cfg80211 rfkill ebtable_nat ebtable_broute bridge stp llc ebtable_filter ebtables ip6table_nat nf_conntrack_ipv6 nf_defrag_ipv6 nf_nat_ipv6 
ip6table_mangle ip6table_security ip6table_raw ip6table_filter ip6_tables iptable_nat nf_conntrack_ipv4 nf_defrag_ipv4 nf_nat_ipv4 nf_nat nf_conntrack iptable_mangle iptable_security iptable_raw iptable_filter ip_tables sg vfat fat x86_pkg_temp_thermal coretemp kvm_intel kvm crct10dif_pclmul 
crc32_pclmul crc32c_intel ghash_clmulni_intel aesni_intel lrw gf128mul glue_helper ablk_helper cryptd dm_mirror dm_region_hash e1000e dm_log dm_mod iTCO_wdt iTCO_vendor_support microcode sb_edac igb pcspkr edac_core i2c_i801
[  116.758000]  ptp lpc_ich pps_core dca mfd_core shpchp ipmi_si tpm_infineon ipmi_msghandler nfsd auth_rpcgss nfs_acl lockd sunrpc xfs libcrc32c sd_mod mgag200 syscopyarea sysfillrect sysimgblt i2c_algo_bit drm_kms_helper lpfc ttm drm crc_t10dif crct10dif_common scsi_transport_fc megaraid_sas 
i2c_core scsi_tgt
[  116.758000] CPU: 0 PID: 6 Comm: kworker/u512:0 Tainted: G        W    3.13.0-rc5+ #11
[  116.758000] Hardware name:
[  116.758000] Workqueue: kacpi_hotplug acpi_hotplug_work_fn
[  116.758000]  0000000000000009 ffff8808738d3bd8 ffffffff815d84ea ffff8808738d3c20
[  116.758000]  ffff8808738d3c10 ffffffff8106594d 0000000000000000 ffffffff819ac5c0
[  116.758000]  ffff880871b9d0a8 ffff8a07d1895000 0000000000000103 ffff8808738d3c70
[  116.758000] Call Trace:
[  116.758000]  [<ffffffff815d84ea>] dump_stack+0x45/0x56
[  116.758000]  [<ffffffff8106594d>] warn_slowpath_common+0x7d/0xa0
[  116.758000]  [<ffffffff810659bc>] warn_slowpath_fmt+0x4c/0x50
[  116.758000]  [<ffffffff8122b52e>] ? sysfs_get_dirent_ns+0x4e/0x70
[  116.758000]  [<ffffffff8122c806>] sysfs_remove_group+0xc6/0xd0
[  116.758000]  [<ffffffff813b83f3>] dpm_sysfs_remove+0x43/0x50
[  116.758000]  [<ffffffff813ae105>] device_del+0x45/0x1c0
[  116.758000]  [<ffffffff812e51f6>] pci_remove_bus_device+0x66/0xd0
[  116.758000]  [<ffffffff812e5363>] pci_remove_root_bus+0x73/0x80
[  116.758000]  [<ffffffff813276ab>] acpi_pci_root_remove+0x42/0x4f
[  116.758000]  [<ffffffff81323070>] acpi_bus_trim+0x56/0x89
[  116.758000]  [<ffffffff81323052>] acpi_bus_trim+0x38/0x89
[  116.758000]  [<ffffffff813245df>] acpi_device_hotplug+0x137/0x33b
[  116.758000]  [<ffffffff8131efba>] acpi_hotplug_work_fn+0x1c/0x27
[  116.758000]  [<ffffffff81080f1b>] process_one_work+0x17b/0x460
[  116.758000]  [<ffffffff81081ccb>] worker_thread+0x11b/0x400
[  116.758000]  [<ffffffff81081bb0>] ? rescuer_thread+0x3e0/0x3e0
[  116.758000]  [<ffffffff81088a12>] kthread+0xd2/0xf0
[  116.758000]  [<ffffffff81088940>] ? kthread_create_on_node+0x180/0x180
[  116.758000]  [<ffffffff815e823c>] ret_from_fork+0x7c/0xb0
[  116.758000]  [<ffffffff81088940>] ? kthread_create_on_node+0x180/0x180
[  116.758000] ---[ end trace b403db9d0ec9fc9e ]---
Quoted 3 lines

...

It is a know issue? The message are shown when we use latest bleeding-edge.

Well, this particular one isn't, but it belongs to the group of similar issues uncovered by recent sysfs changes.

It basically means that the removal ordering is most likely wrong somewhere.

I'll have a look into that, but it is not related to the patches in this thread.

Thanks,
Rafael

#26 lore raw

cover ACPI / hotplug / driver core: Special handling for container devices

Hi Greg,

The following series of 2 patches (patch [2/2] in particular) make changes needed to handle hot-removal of system container devices (represented by ACPI container and module device objects) on Fujitsu systems. Those devices represent things like CPU packages, so we need to be able to take care of them cleanly for things like in-the-field CPU socked replacement to work.

The problem being addressed here is that on the systems in question the removal of container devices has to be carried out with the help of user space that needs to be notified of the container removal before the kernel attempts to offline any devices below the container (e.g. in the package represented by the container device object in the ACPI tables). However, our current code works the other way around and the entire thing is messed up.

This patchset adds the bare bones of what's needed to address that issue and it should be possible to build on top of the code added by it in the future if need be.

Patch [1/2] introduces a new demand_offline flag for struct acpi_hotplug_profile that makes acpi_scan_hot_remove() check the offline status of the device object's companion physical devices to start with and return -EBUSY if at least one of them is not offline.

Patch [2/2] uses that flag to implement the container handling. The details are in the changelog, but that's how it works.

During the initial namespace scan the container ACPI scan handler creates "physical" system container device under /sys/devices/system/container/ for each ACPI container object whose status is "present" at that time (the sysfs name of that device is the same as the sysfs name of the corresponding container object and they are linked to each other via the firmware_node and physical_node symbolic links, respectively). Those system container devices are initially online.

The container ACPI scan handler has the demand_offline flag set in its hotplug profile, so when a container eject event happens, acpi_scan_hot_remove() will notice that the flag is set in the device object's scan handler and will check the online status of its "physical" companion device, which is online (that is the system container device the above paragraph is about). That will cause KOBJ_CHANGE to be emitted for the system container device and -EBUSY to be returned by acpi_scan_hot_remove(). User space is expected to respond to that KOBJ_CHANGE by doing what's necessary to remove the container.

To that end, it needs to offline the system container device through its online sysfs attribute (which is present, because the bus type for containers provides the online and offline callbacks). However, the offline for system container devices will only succeed if the physical devices right below the container (e.g. in the package represented by it) are all offline, so user space will have to offline those devices before attempting to offline the system container device itself. When finished, user space can trigger the container removal with the help of the eject sysfs attribute of the ACPI container object pointed to by the system container device's firmware_node link (this time the check in acpi_scan_hot_remove() will succeed, because the system container device in question is now offline).

Please let me know if you have any objections. If not, I'd like to queue up these patches for 3.14.

Thanks,
Rafael

#27 lore raw

PATCH 1/2 ACPI / hotplug: Add demand_offline hotplug profile flag

From: Rafael J. Wysocki <[email protected]>

Add a new ACPI hotplug profile flag, demand_offline, such that if set for the given ACPI device object's scan handler, it will cause acpi_scan_hot_remove() to check if that device object's physical companions are offline upfront and fail the hot removal if that is not the case.

That flag will be useful to overcome a problem with containers on some system where they can only be hot-removed after some cleanup operations carried out by user space, which needs to be notified of the container hot-removal before the kernel attempts to offline devices in the container. In those cases the current implementation of acpi_scan_hot_remove() is not sufficient, because it first tries to offline the devices in the container and only if that is suffcessful it tries to offline the container itself. As a result, the container hot-removal notification is not delivered to user space at the right time.

---
 drivers/acpi/scan.c     |   41 +++++++++++++++++++++++++++++++++++++----
 include/acpi/acpi_bus.h |    3 ++-
 2 files changed, 39 insertions(+), 5 deletions(-)
drivers/acpi/scan.c +37 -4
Index: linux-pm/drivers/acpi/scan.c
===================================================================
@@ -126,6 +126,24 @@ acpi_device_modalias_show(struct device
 }
 static DEVICE_ATTR(modalias, 0444, acpi_device_modalias_show, NULL);
 
+static bool acpi_scan_is_offline(struct acpi_device *adev)
+{
+	struct acpi_device_physical_node *pn;
+	bool offline = true;
+
+	mutex_lock(&adev->physical_node_lock);
+
+	list_for_each_entry(pn, &adev->physical_node_list, node)
+		if (device_supports_offline(pn->dev) && !pn->dev->offline) {
+			kobject_uevent(&pn->dev->kobj, KOBJ_CHANGE);
+			offline = false;
+			break;
+		}
+
+	mutex_unlock(&adev->physical_node_lock);
+	return offline;
+}
+
 static acpi_status acpi_bus_offline(acpi_handle handle, u32 lvl, void *data,
 				    void **ret_p)
 {
@@ -196,12 +214,11 @@ static acpi_status acpi_bus_online(acpi_
 	return AE_OK;
 }
 
-static int acpi_scan_hot_remove(struct acpi_device *device)
+static int acpi_scan_try_to_offline(struct acpi_device *device)
 {
 	acpi_handle handle = device->handle;
-	struct device *errdev;
+	struct device *errdev = NULL;
 	acpi_status status;
-	unsigned long long sta;
 
 	/*
 	 * Carry out two passes here and ignore errors in the first pass,
@@ -212,7 +229,6 @@ static int acpi_scan_hot_remove(struct a
 	 *
 	 * If the first pass is successful, the second one isn't needed, though.
 	 */
-	errdev = NULL;
 	status = acpi_walk_namespace(ACPI_TYPE_ANY, handle, ACPI_UINT32_MAX,
 				     NULL, acpi_bus_offline, (void *)false,
 				     (void **)&errdev);
@@ -241,6 +257,23 @@ static int acpi_scan_hot_remove(struct a
 			return -EBUSY;
 		}
 	}
+	return 0;
+}
+
+static int acpi_scan_hot_remove(struct acpi_device *device)
+{
+	acpi_handle handle = device->handle;
+	unsigned long long sta;
+	acpi_status status;
+
+	if (device->handler->hotplug.demand_offline && !acpi_force_hot_remove) {
+		if (!acpi_scan_is_offline(device))
+			return -EBUSY;
+	} else {
+		int error = acpi_scan_try_to_offline(device);
+		if (error)
+			return error;
+	}
 
 	ACPI_DEBUG_PRINT((ACPI_DB_INFO,
 		"Hot-removing device %s...\n", dev_name(&device->dev)));
include/acpi/acpi_bus.h +2 -1
Index: linux-pm/include/acpi/acpi_bus.h
===================================================================
@@ -91,8 +91,9 @@ struct acpi_device;
 
 struct acpi_hotplug_profile {
 	struct kobject kobj;
-	bool enabled:1;
 	int (*scan_dependent)(struct acpi_device *adev);
+	bool enabled:1;
+	bool demand_offline:1;
 };
 
 static inline struct acpi_hotplug_profile *to_acpi_hotplug_profile(
#28 lore raw

PATCH 2/2 ACPI / hotplug / driver core: Handle containers in a special way

In reply to Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

ACPI container devices require special hotplug handling, at least on some systems, since generally user space needs to carry out system-specific cleanup before it makes sense to offline devices in the container. However, the current ACPI hotplug code for containers first attempts to offline devices in the container and only then it notifies user space of the container offline.

Moreover, after commit 202317a573b2 (ACPI / scan: Add acpi_device objects for all device nodes in the namespace), ACPI device objects representing containers are present as long as the ACPI namespace nodes corresponding to them are present, which may be forever, even if the container devices are physically detached from the system (the return values of the corresponding _STA methods change in those cases, but generally the namespace nodes themselves are still there). Thus it is useful to introduce entities representing containers that will go away during container hot-unplug.

The goal of this change is to address both the above issues.

The idea is to create a "companion" container system device for each of the ACPI container device objects during the initial namespace scan or on a hotplug event making the container present. That system device will be unregistered on container removal. A new bus type for container devices is added for this purpose, because device offline and online operations need to be defined for them. The online operation is a trivial function that is always successful and the offline uses a callback pointed to by the container device's offline member.

For ACPI containers that callback simply walks the list of ACPI device objects right below the container object (its children) and checks if all of their physical companion devices are offline. If that's not the case, it returns -EBUSY and the container system device cannot be put offline. Consequently, to put the container system device offline, it is necessary to put all of the physical devices depending on its ACPI companion object offline beforehand.

Container system devices created for ACPI container objects are initially online. They are created by the container ACPI scan handler whose hotplug.demand_offline flag is set. That causes acpi_scan_hot_remove() to check if the companion container system device is offline before attempting to remove an ACPI container or any devices below it. If the check fails, a KOBJ_CHANGE uevent is emitted for the container system device in question and user space is expected to offline all devices below the container and the container itself in response to it. Then, user space can finalize the removal of the container with the help of its ACPI device object's eject attribute in sysfs.

---
 drivers/acpi/container.c  |   48 ++++++++++++++++++++++++++++++++++++++++++----
 drivers/acpi/internal.h   |    1 
 drivers/acpi/scan.c       |    8 ++++---
 drivers/base/Makefile     |    2 -
 drivers/base/base.h       |    1 
 drivers/base/container.c  |   44 ++++++++++++++++++++++++++++++++++++++++++
 drivers/base/init.c       |    1 
 include/linux/container.h |   25 +++++++++++++++++++++++
 8 files changed, 122 insertions(+), 8 deletions(-)
include/linux/container.h new file +25 -0
Index: linux-pm/include/linux/container.h
===================================================================
@@ -0,0 +1,25 @@
+/*
+ * Definitions for container bus type.
+ *
+ * Copyright (C) 2013, Intel Corporation
+ * Author: Rafael J. Wysocki <[email protected]>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include <linux/device.h>
+
+/* drivers/base/power/container.c */
+extern struct bus_type container_subsys;
+
+struct container_dev {
+	struct device dev;
+	int (*offline)(struct container_dev *cdev);
+};
+
+static inline struct container_dev *to_container_dev(struct device *dev)
+{
+	return container_of(dev, struct container_dev, dev);
+}
drivers/base/container.c new file +44 -0

44 lines not on this page · raw message

drivers/base/base.h +1 -0
Index: linux-pm/drivers/base/base.h
===================================================================
@@ -100,6 +100,7 @@ static inline int hypervisor_init(void)
 #endif
 extern int platform_bus_init(void);
 extern void cpu_dev_init(void);
+extern void container_dev_init(void);
 
 struct kobject *virtual_device_parent(struct device *dev);
 
drivers/base/init.c +1 -0
Index: linux-pm/drivers/base/init.c
===================================================================
@@ -33,4 +33,5 @@ void __init driver_init(void)
 	platform_bus_init();
 	cpu_dev_init();
 	memory_dev_init();
+	container_dev_init();
 }
drivers/base/Makefile +1 -1
Index: linux-pm/drivers/base/Makefile
===================================================================
@@ -4,7 +4,7 @@ obj-y			:= core.o bus.o dd.o syscore.o \
 			   driver.o class.o platform.o \
 			   cpu.o firmware.o init.o map.o devres.o \
 			   attribute_container.o transport_class.o \
-			   topology.o
+			   topology.o container.o
 obj-$(CONFIG_DEVTMPFS)	+= devtmpfs.o
 obj-$(CONFIG_DMA_CMA) += dma-contiguous.o
 obj-y			+= power/
drivers/acpi/internal.h +1 -0
Index: linux-pm/drivers/acpi/internal.h
===================================================================
@@ -73,6 +73,7 @@ static inline void acpi_lpss_init(void)
 #endif
 
 bool acpi_queue_hotplug_work(struct work_struct *work);
+bool acpi_scan_is_offline(struct acpi_device *adev, bool uevent);
 
 /* --------------------------------------------------------------------------
                      Device Node Initialization / Removal
drivers/acpi/scan.c +5 -3
Index: linux-pm/drivers/acpi/scan.c
===================================================================
@@ -126,7 +126,7 @@ acpi_device_modalias_show(struct device
 }
 static DEVICE_ATTR(modalias, 0444, acpi_device_modalias_show, NULL);
 
-static bool acpi_scan_is_offline(struct acpi_device *adev)
+bool acpi_scan_is_offline(struct acpi_device *adev, bool uevent)
 {
 	struct acpi_device_physical_node *pn;
 	bool offline = true;
@@ -135,7 +135,9 @@ static bool acpi_scan_is_offline(struct
 
 	list_for_each_entry(pn, &adev->physical_node_list, node)
 		if (device_supports_offline(pn->dev) && !pn->dev->offline) {
-			kobject_uevent(&pn->dev->kobj, KOBJ_CHANGE);
+			if (uevent)
+				kobject_uevent(&pn->dev->kobj, KOBJ_CHANGE);
+
 			offline = false;
 			break;
 		}
@@ -267,7 +269,7 @@ static int acpi_scan_hot_remove(struct a
 	acpi_status status;
 
 	if (device->handler->hotplug.demand_offline && !acpi_force_hot_remove) {
-		if (!acpi_scan_is_offline(device))
+		if (!acpi_scan_is_offline(device, true))
 			return -EBUSY;
 	} else {
 		int error = acpi_scan_try_to_offline(device);
drivers/acpi/container.c +44 -4

74 lines not on this page · raw message

maintainer Acked-by on v1 2/2
#29 lore raw

Re: [PATCH 2/2] ACPI / hotplug / driver core: Handle containers in a special way

Quoted 51 lines · Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

ACPI container devices require special hotplug handling, at least on some systems, since generally user space needs to carry out system-specific cleanup before it makes sense to offline devices in the container. However, the current ACPI hotplug code for containers first attempts to offline devices in the container and only then it notifies user space of the container offline.

Moreover, after commit 202317a573b2 (ACPI / scan: Add acpi_device objects for all device nodes in the namespace), ACPI device objects representing containers are present as long as the ACPI namespace nodes corresponding to them are present, which may be forever, even if the container devices are physically detached from the system (the return values of the corresponding _STA methods change in those cases, but generally the namespace nodes themselves are still there). Thus it is useful to introduce entities representing containers that will go away during container hot-unplug.

The goal of this change is to address both the above issues.

The idea is to create a "companion" container system device for each of the ACPI container device objects during the initial namespace scan or on a hotplug event making the container present. That system device will be unregistered on container removal. A new bus type for container devices is added for this purpose, because device offline and online operations need to be defined for them. The online operation is a trivial function that is always successful and the offline uses a callback pointed to by the container device's offline member.

For ACPI containers that callback simply walks the list of ACPI device objects right below the container object (its children) and checks if all of their physical companion devices are offline. If that's not the case, it returns -EBUSY and the container system device cannot be put offline. Consequently, to put the container system device offline, it is necessary to put all of the physical devices depending on its ACPI companion object offline beforehand.

Container system devices created for ACPI container objects are initially online. They are created by the container ACPI scan handler whose hotplug.demand_offline flag is set. That causes acpi_scan_hot_remove() to check if the companion container system device is offline before attempting to remove an ACPI container or any devices below it. If the check fails, a KOBJ_CHANGE uevent is emitted for the container system device in question and user space is expected to offline all devices below the container and the container itself in response to it. Then, user space can finalize the removal of the container with the help of its ACPI device object's eject attribute in sysfs.

Quoted 2 lines
#30 lore raw

Re: [PATCH 0/2] ACPI / hotplug / driver core: Special handling for container devices

In reply to Rafael J. Wysocki
Quoted 56 lines · Rafael J. Wysocki

Hi Greg,

The following series of 2 patches (patch [2/2] in particular) make changes needed to handle hot-removal of system container devices (represented by ACPI container and module device objects) on Fujitsu systems. Those devices represent things like CPU packages, so we need to be able to take care of them cleanly for things like in-the-field CPU socked replacement to work.

The problem being addressed here is that on the systems in question the removal of container devices has to be carried out with the help of user space that needs to be notified of the container removal before the kernel attempts to offline any devices below the container (e.g. in the package represented by the container device object in the ACPI tables). However, our current code works the other way around and the entire thing is messed up.

This patchset adds the bare bones of what's needed to address that issue and it should be possible to build on top of the code added by it in the future if need be.

Patch [1/2] introduces a new demand_offline flag for struct acpi_hotplug_profile that makes acpi_scan_hot_remove() check the offline status of the device object's companion physical devices to start with and return -EBUSY if at least one of them is not offline.

Patch [2/2] uses that flag to implement the container handling. The details are in the changelog, but that's how it works.

During the initial namespace scan the container ACPI scan handler creates "physical" system container device under /sys/devices/system/container/ for each ACPI container object whose status is "present" at that time (the sysfs name of that device is the same as the sysfs name of the corresponding container object and they are linked to each other via the firmware_node and physical_node symbolic links, respectively). Those system container devices are initially online.

The container ACPI scan handler has the demand_offline flag set in its hotplug profile, so when a container eject event happens, acpi_scan_hot_remove() will notice that the flag is set in the device object's scan handler and will check the online status of its "physical" companion device, which is online (that is the system container device the above paragraph is about). That will cause KOBJ_CHANGE to be emitted for the system container device and -EBUSY to be returned by acpi_scan_hot_remove(). User space is expected to respond to that KOBJ_CHANGE by doing what's necessary to remove the container.

To that end, it needs to offline the system container device through its online sysfs attribute (which is present, because the bus type for containers provides the online and offline callbacks). However, the offline for system container devices will only succeed if the physical devices right below the container (e.g. in the package represented by it) are all offline, so user space will have to offline those devices before attempting to offline the system container device itself. When finished, user space can trigger the container removal with the help of the eject sysfs attribute of the ACPI container object pointed to by the system container device's firmware_node link (this time the check in acpi_scan_hot_remove() will succeed, because the system container device in question is now offline).

Quoted 2 lines

Please let me know if you have any objections. If not, I'd like to queue up these patches for 3.14.

No objection from me, I've acked the second patch, feel free to take both of them in your tree.

thanks,

greg k-h

#31 lore raw

Re: [PATCH 0/2] ACPI / hotplug / driver core: Special handling for container devices

Quoted 57 lines · Greg Kroah-Hartman
Quoted 56 lines · Rafael J. Wysocki

Hi Greg,

The following series of 2 patches (patch [2/2] in particular) make changes needed to handle hot-removal of system container devices (represented by ACPI container and module device objects) on Fujitsu systems. Those devices represent things like CPU packages, so we need to be able to take care of them cleanly for things like in-the-field CPU socked replacement to work.

The problem being addressed here is that on the systems in question the removal of container devices has to be carried out with the help of user space that needs to be notified of the container removal before the kernel attempts to offline any devices below the container (e.g. in the package represented by the container device object in the ACPI tables). However, our current code works the other way around and the entire thing is messed up.

This patchset adds the bare bones of what's needed to address that issue and it should be possible to build on top of the code added by it in the future if need be.

Patch [1/2] introduces a new demand_offline flag for struct acpi_hotplug_profile that makes acpi_scan_hot_remove() check the offline status of the device object's companion physical devices to start with and return -EBUSY if at least one of them is not offline.

Patch [2/2] uses that flag to implement the container handling. The details are in the changelog, but that's how it works.

During the initial namespace scan the container ACPI scan handler creates "physical" system container device under /sys/devices/system/container/ for each ACPI container object whose status is "present" at that time (the sysfs name of that device is the same as the sysfs name of the corresponding container object and they are linked to each other via the firmware_node and physical_node symbolic links, respectively). Those system container devices are initially online.

The container ACPI scan handler has the demand_offline flag set in its hotplug profile, so when a container eject event happens, acpi_scan_hot_remove() will notice that the flag is set in the device object's scan handler and will check the online status of its "physical" companion device, which is online (that is the system container device the above paragraph is about). That will cause KOBJ_CHANGE to be emitted for the system container device and -EBUSY to be returned by acpi_scan_hot_remove(). User space is expected to respond to that KOBJ_CHANGE by doing what's necessary to remove the container.

To that end, it needs to offline the system container device through its online sysfs attribute (which is present, because the bus type for containers provides the online and offline callbacks). However, the offline for system container devices will only succeed if the physical devices right below the container (e.g. in the package represented by it) are all offline, so user space will have to offline those devices before attempting to offline the system container device itself. When finished, user space can trigger the container removal with the help of the eject sysfs attribute of the ACPI container object pointed to by the system container device's firmware_node link (this time the check in acpi_scan_hot_remove() will succeed, because the system container device in question is now offline).

Quoted 5 lines
Quoted 2 lines

Please let me know if you have any objections. If not, I'd like to queue up these patches for 3.14.

No objection from me, I've acked the second patch, feel free to take both of them in your tree.

I will, thanks a lot!

Rafael

#32 lore raw

Re: [PATCH 1/2][Untested] ACPI / hotplug: Add demand_offline hotplug profile flag

In reply to Yasuaki Ishimatsu
Quoted 50 lines · Yasuaki Ishimatsu
Quoted 33 lines · Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Add a new ACPI hotplug profile flag, demand_offline, such that if set for the given ACPI device object's scan handler, it will cause acpi_scan_hot_remove() to check if that device object's physical companions are offline upfront and fail the hot removal if that is not the case.

That flag will be useful to overcome a problem with containers on some system where they can only be hot-removed after some cleanup operations carried out by user space, which needs to be notified of the container hot-removal before the kernel attempts to offline devices in the container. In those cases the current implementation of acpi_scan_hot_remove() is not sufficient, because it first tries to offline the devices in the container and only if that is suffcessful it tries to offline the container itself. As a result, the container hot-removal notification is not delivered to user space at the right time.

---
  drivers/acpi/scan.c     |   41 +++++++++++++++++++++++++++++++++++++----
  include/acpi/acpi_bus.h |    3 ++-
  2 files changed, 39 insertions(+), 5 deletions(-)

Index: linux-pm/drivers/acpi/scan.c
===================================================================

drivers/acpi/scan.c +0 -0
@@ -126,6 +126,24 @@ acpi_device_modalias_show(struct device
  }
  static DEVICE_ATTR(modalias, 0444, acpi_device_modalias_show, NULL);
 

drivers/acpi/scan.c acpi_device_modalias_show(struct device · line 138

Quoted 10 lines
drivers/acpi/scan.c +9 -0
+static bool acpi_scan_is_offline(struct acpi_device *adev)
+{
+	struct acpi_device_physical_node *pn;
+	bool offline = true;
+
+	mutex_lock(&adev->physical_node_lock);
+
+	list_for_each_entry(pn, &adev->physical_node_list, node)
 
+		if (!pn->dev->offline) {

Please check pn->dev->bus and pn->dev->bus->offline too as follow:

		if (pn->dev->bus && pn->dev->bus->offline &&
		    !pn->dev->offline) {
Quoted 7 lines

My container has CPU and Memory and PCI root bridge. PCI root bridge does not has offline function (pn->dev->bus->offline). So I cannot offline the device and pn->dev->offline of the device is always 0. Therefore, following operation always returns -EBUSY even if I offline CPUs and all memory sections on a container device.

echo 0 > /sys/bus/container/devices/ACPI0004:01/online

I see. We simply can do

	if (device_supports_offline(pn->dev) && !pn->dev->offline) {

there I think.

Thanks,
Rafael

#33 lore raw

PATCH 2/2 ACPI / hotplug / driver core: Handle containers in a special way

In reply to Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

ACPI container devices require special hotplug handling, at least on some systems, since generally user space needs to carry out system-specific cleanup before it makes sense to offline devices in the container. However, the current ACPI hotplug code for containers first attempts to offline devices in the container and only then it notifies user space of the container offline.

Moreover, after commit 202317a573b2 (ACPI / scan: Add acpi_device objects for all device nodes in the namespace), ACPI device objects representing containers are present as long as the ACPI namespace nodes corresponding to them are present, which may be forever, even if the container devices are physically detached from the system (the return values of the corresponding _STA methods change in those cases, but generally the namespace nodes themselves are still there). Thus it is useful to introduce entities representing containers that will go away during container hot-unplug.

The goal of this change is to address both the above issues.

The idea is to create a "companion" container system device for each of the ACPI container device objects during the initial namespace scan or on a hotplug event making the container present. That system device will be unregistered on container removal. A new bus type for container devices is added for this purpose, because device offline and online operations need to be defined for them. The online operation is a trivial function that is always successful and the offline uses a callback provided by the container device creator through driver data.

For ACPI containers that callback simply walks the list of ACPI device objects right below the container object (its children) and checks if all of their physical companion devices are offline. If that's not the case, it returns -EBUSY and the container system devivce cannot be put offline. Consequently, to put the container system device offline, it is necessary to put all of the physical devices depending on its ACPI companion object offline beforehand.

Container system devices created for ACPI container objects are initially online. They are created by the container ACPI scan handler whose hotplug.demand_offline flag is set. That causes acpi_scan_hot_remove() to check if the companion container system device is offline before attempting to remove an ACPI container or any devices below it. If the check fails, a KOBJ_CHANGE uevent is emitted for the container system device in question and user space is expected to offline all devices below the container and the container itself in response to it. Then, user space can finalize the removal of the container with the help of its ACPI device object's eject attribute in sysfs.

---
 drivers/acpi/container.c |   43 ++++++++++++++++++++++++++++++++++++++----
 drivers/acpi/internal.h  |    1 
 drivers/acpi/scan.c      |    8 ++++---
 drivers/base/Makefile    |    2 -
 drivers/base/base.h      |    1 
 drivers/base/container.c |   48 +++++++++++++++++++++++++++++++++++++++++++++++
 drivers/base/init.c      |    1 
 include/linux/device.h   |    7 ++++++
 8 files changed, 103 insertions(+), 8 deletions(-)
drivers/base/container.c new file +48 -0
Index: linux-pm/drivers/base/container.c
===================================================================
@@ -0,0 +1,48 @@
+/*
+ * System bus type for containers.
+ *
+ * Copyright (C) 2013, Intel Corporation
+ * Author: Rafael J. Wysocki <[email protected]>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include <linux/device.h>
+
+#include "base.h"
+
+#define CONTAINER_BUS_NAME	"container"
+
+static int trivial_online(struct device *dev)
+{
+	return 0;
+}
+
+static int container_offline(struct device *dev)
+{
+	struct container_ops *ops = dev_get_drvdata(dev);
+	int ret = 0;
+
+	if (ops && ops->offline)
+		ret = ops->offline(dev);
+
+	return ret;
+}
+
+struct bus_type container_subsys = {
+	.name = CONTAINER_BUS_NAME,
+	.dev_name = CONTAINER_BUS_NAME,
+	.online = trivial_online,
+	.offline = container_offline,
+};
+
+void __init container_dev_init(void)
+{
+	int ret;
+
+	ret = subsys_system_register(&container_subsys, NULL);
+	if (ret)
+		pr_err("%s() failed: %d\n", __func__, ret);
+}
drivers/base/base.h +1 -0
Index: linux-pm/drivers/base/base.h
===================================================================
@@ -100,6 +100,7 @@ static inline int hypervisor_init(void)
 #endif
 extern int platform_bus_init(void);
 extern void cpu_dev_init(void);
+extern void container_dev_init(void);
 
 struct kobject *virtual_device_parent(struct device *dev);
 
drivers/base/init.c +1 -0
Index: linux-pm/drivers/base/init.c
===================================================================
@@ -33,4 +33,5 @@ void __init driver_init(void)
 	platform_bus_init();
 	cpu_dev_init();
 	memory_dev_init();
+	container_dev_init();
 }
include/linux/device.h +7 -0
Index: linux-pm/include/linux/device.h
===================================================================
@@ -1004,6 +1004,13 @@ static inline int devtmpfs_mount(const c
 /* drivers/base/power/shutdown.c */
 extern void device_shutdown(void);
 
+struct container_ops {
+	int (*offline)(struct device *dev);
+};
+
+/* drivers/base/power/container.c */
+extern struct bus_type container_subsys;
+
 /* debugging and troubleshooting/diagnostic helpers. */
 extern const char *dev_driver_string(const struct device *dev);
 
drivers/acpi/container.c +39 -4

69 lines not on this page · raw message

drivers/base/Makefile +1 -1
Index: linux-pm/drivers/base/Makefile
===================================================================
@@ -4,7 +4,7 @@ obj-y			:= core.o bus.o dd.o syscore.o \
 			   driver.o class.o platform.o \
 			   cpu.o firmware.o init.o map.o devres.o \
 			   attribute_container.o transport_class.o \
-			   topology.o
+			   topology.o container.o
 obj-$(CONFIG_DEVTMPFS)	+= devtmpfs.o
 obj-$(CONFIG_DMA_CMA) += dma-contiguous.o
 obj-y			+= power/
drivers/acpi/internal.h +1 -0
Index: linux-pm/drivers/acpi/internal.h
===================================================================
@@ -73,6 +73,7 @@ static inline void acpi_lpss_init(void)
 #endif
 
 bool acpi_queue_hotplug_work(struct work_struct *work);
+bool acpi_scan_is_offline(struct acpi_device *adev, bool uevent);
 
 /* --------------------------------------------------------------------------
                      Device Node Initialization / Removal
drivers/acpi/scan.c +5 -3
Index: linux-pm/drivers/acpi/scan.c
===================================================================
@@ -126,7 +126,7 @@ acpi_device_modalias_show(struct device
 }
 static DEVICE_ATTR(modalias, 0444, acpi_device_modalias_show, NULL);
 
-static bool acpi_scan_is_offline(struct acpi_device *adev)
+bool acpi_scan_is_offline(struct acpi_device *adev, bool uevent)
 {
 	struct acpi_device_physical_node *pn;
 	bool offline = true;
@@ -135,7 +135,9 @@ static bool acpi_scan_is_offline(struct
 
 	list_for_each_entry(pn, &adev->physical_node_list, node)
 		if (!pn->dev->offline) {
-			kobject_uevent(&pn->dev->kobj, KOBJ_CHANGE);
+			if (uevent)
+				kobject_uevent(&pn->dev->kobj, KOBJ_CHANGE);
+
 			offline = false;
 			break;
 		}
@@ -267,7 +269,7 @@ static int acpi_scan_hot_remove(struct a
 	acpi_status status;
 
 	if (device->handler->hotplug.demand_offline && !acpi_force_hot_remove) {
-		if (!acpi_scan_is_offline(device))
+		if (!acpi_scan_is_offline(device, true))
 			return -EBUSY;
 	} else {
 		int error = acpi_scan_try_to_offline(device);
#34 lore raw

PATCH 2/2 ACPI / hotplug / driver core: Handle containers in a special way

From: Rafael J. Wysocki <[email protected]>
Subject: ACPI / hotplug / driver core: Handle containers in a special way

ACPI container devices require special hotplug handling, at least on some systems, since generally user space needs to carry out system-specific cleanup before it makes sense to offline devices in the container. However, the current ACPI hotplug code for containers first attempts to offline devices in the container and only then it notifies user space of the container offline.

Moreover, after commit 202317a573b2 (ACPI / scan: Add acpi_device objects for all device nodes in the namespace), ACPI device objects representing containers are present as long as the ACPI namespace nodes corresponding to them are present, which may be forever, even if the container devices are physically detached from the system (the return values of the corresponding _STA methods change in those cases, but generally the namespace nodes themselves are still there). Thus it is useful to introduce entities representing containers that will go away during container hot-unplug.

The goal of this change is to address both the above issues.

The idea is to create a "companion" container system device for each of the ACPI container device objects during the initial namespace scan or on a hotplug event making the container present. That system device will be unregistered on container removal. A new bus type for container devices is added for this purpose, because device offline and online operations need to be defined for them. The online operation is a trivial function that is always successful and the offline uses a callback pointed to by the container device's offline member.

For ACPI containers that callback simply walks the list of ACPI device objects right below the container object (its children) and checks if all of their physical companion devices are offline. If that's not the case, it returns -EBUSY and the container system devivce cannot be put offline. Consequently, to put the container system device offline, it is necessary to put all of the physical devices depending on its ACPI companion object offline beforehand.

Container system devices created for ACPI container objects are initially online. They are created by the container ACPI scan handler whose hotplug.demand_offline flag is set. That causes acpi_scan_hot_remove() to check if the companion container system device is offline before attempting to remove an ACPI container or any devices below it. If the check fails, a KOBJ_CHANGE uevent is emitted for the container system device in question and user space is expected to offline all devices below the container and the container itself in response to it. Then, user space can finalize the removal of the container with the help of its ACPI device object's eject attribute in sysfs.

---

Hi,

I've realized that I don't need to abuse driver_data here if I define a struct container_dev based on struct device and containing a pointer to the offline routine, so here goes an update with that modification.

I checked that my venerable Toshiba test box boots the kernel with [1/2] and this patch applied and that the container bus type is present then as expected, so it is not totally untested any more. I still need to test the creation of container devices, though.

Thanks,
Rafael

---
 drivers/acpi/container.c  |   42 ++++++++++++++++++++++++++++++++++++++----
 drivers/acpi/internal.h   |    1 +
 drivers/acpi/scan.c       |    8 +++++---
 drivers/base/Makefile     |    2 +-
 drivers/base/base.h       |    1 +
 drivers/base/container.c  |   44 ++++++++++++++++++++++++++++++++++++++++++++
 drivers/base/init.c       |    1 +
 include/linux/container.h |   25 +++++++++++++++++++++++++
 8 files changed, 116 insertions(+), 8 deletions(-)
include/linux/container.h new file +25 -0
Index: linux-pm/include/linux/container.h
===================================================================
@@ -0,0 +1,25 @@
+/*
+ * Definitions for container bus type.
+ *
+ * Copyright (C) 2013, Intel Corporation
+ * Author: Rafael J. Wysocki <[email protected]>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include <linux/device.h>
+
+/* drivers/base/power/container.c */
+extern struct bus_type container_subsys;
+
+struct container_dev {
+	struct device dev;
+	int (*offline)(struct container_dev *cdev);
+};
+
+static inline struct container_dev *to_container_dev(struct device *dev)
+{
+	return container_of(dev, struct container_dev, dev);
+}
drivers/base/container.c new file +44 -0

44 lines not on this page · raw message

drivers/base/base.h +1 -0
Index: linux-pm/drivers/base/base.h
===================================================================
@@ -100,6 +100,7 @@ static inline int hypervisor_init(void)
 #endif
 extern int platform_bus_init(void);
 extern void cpu_dev_init(void);
+extern void container_dev_init(void);
 
 struct kobject *virtual_device_parent(struct device *dev);
 
drivers/base/init.c +1 -0
Index: linux-pm/drivers/base/init.c
===================================================================
@@ -33,4 +33,5 @@ void __init driver_init(void)
 	platform_bus_init();
 	cpu_dev_init();
 	memory_dev_init();
+	container_dev_init();
 }
drivers/base/Makefile +1 -1
Index: linux-pm/drivers/base/Makefile
===================================================================
@@ -4,7 +4,7 @@ obj-y			:= core.o bus.o dd.o syscore.o \
 			   driver.o class.o platform.o \
 			   cpu.o firmware.o init.o map.o devres.o \
 			   attribute_container.o transport_class.o \
-			   topology.o
+			   topology.o container.o
 obj-$(CONFIG_DEVTMPFS)	+= devtmpfs.o
 obj-$(CONFIG_DMA_CMA) += dma-contiguous.o
 obj-y			+= power/
drivers/acpi/internal.h +1 -0
Index: linux-pm/drivers/acpi/internal.h
===================================================================
@@ -73,6 +73,7 @@ static inline void acpi_lpss_init(void)
 #endif
 
 bool acpi_queue_hotplug_work(struct work_struct *work);
+bool acpi_scan_is_offline(struct acpi_device *adev, bool uevent);
 
 /* --------------------------------------------------------------------------
                      Device Node Initialization / Removal
drivers/acpi/scan.c +5 -3
Index: linux-pm/drivers/acpi/scan.c
===================================================================
@@ -126,7 +126,7 @@ acpi_device_modalias_show(struct device
 }
 static DEVICE_ATTR(modalias, 0444, acpi_device_modalias_show, NULL);
 
-static bool acpi_scan_is_offline(struct acpi_device *adev)
+bool acpi_scan_is_offline(struct acpi_device *adev, bool uevent)
 {
 	struct acpi_device_physical_node *pn;
 	bool offline = true;
@@ -135,7 +135,9 @@ static bool acpi_scan_is_offline(struct
 
 	list_for_each_entry(pn, &adev->physical_node_list, node)
 		if (!pn->dev->offline) {
-			kobject_uevent(&pn->dev->kobj, KOBJ_CHANGE);
+			if (uevent)
+				kobject_uevent(&pn->dev->kobj, KOBJ_CHANGE);
+
 			offline = false;
 			break;
 		}
@@ -267,7 +269,7 @@ static int acpi_scan_hot_remove(struct a
 	acpi_status status;
 
 	if (device->handler->hotplug.demand_offline && !acpi_force_hot_remove) {
-		if (!acpi_scan_is_offline(device))
+		if (!acpi_scan_is_offline(device, true))
 			return -EBUSY;
 	} else {
 		int error = acpi_scan_try_to_offline(device);
drivers/acpi/container.c +38 -4

68 lines not on this page · raw message

#35 lore raw

Re: [PATCH 7/10] ACPI / hotplug: Move container-specific code out of the core

In reply to Rafael J. Wysocki
Quoted 150 lines · Rafael J. Wysocki
Quoted 107 lines · Rafael J. Wysocki
Quoted 60 lines · Yasuaki Ishimatsu
Quoted 59 lines · Rafael J. Wysocki
Quoted 1 line · Yasuaki Ishimatsu

Hi Rafael,

Hi,

Quoted 17 lines

Please share your more detailed idea. I started to implement the following idea. But the idea has one problem.

Quoted 5 lines
Quoted 5 lines
Quoted 5 lines

The eject work flow can be:
(1) an eject event occurs,
(2) the container "physical" device fails offline in acpi_scan_hot_remove() emmitting, say, KOBJ_CHANGE for the "physical" device,
(3) user space notices the KOBJ_CHANGE and does the cleanup as needed,

Quoted 8 lines
Quoted 8 lines
Quoted 8 lines

(4) user space changes the "physical" container device flag controlling offline to 0,
(5) user space uses the sysfs "eject" attribute of the ACPI container object to finally eject the container,
(6) the offline in acpi_scan_hot_remove() is now successful, because the flag controlling it has been set to 0 in step (4),
(7) the "physical" container device goes away before executing _EJ0,
(8) the container is ejected.

Quoted 5 lines

I want to emit KOBJ_CHANGE before offlining devices on container device at (2). But acpi_scan_hot_remove() offlines devices on container device at first. So when offline container device, devices on container has been offlined.

Thus the idea cannot fulfill my necessary feature.

Well, in that case we need to treat containers in a special way at the ACPI level. Which is a bit unfortunate so to speak.

To that end I'd try to add a new flag to struct acpi_hotplug_profile, say .verify_offline, such that if set, it would cause acpi_scan_hot_remove() to check if all of the "physical" companions of the top-level device are offline to start with, and if not, it would just emit KOBJ_CHANGE for the companions that are not offline and bail out.

So the above algorithm would become:

(1) an eject event occurs,
(2) acpi_scan_hot_remove() checks the verify_offline flag in the target device's scan_handler structure,
(3) if set (it would always be set for containers), acpi_scan_hot_remove() checks the status of the target device's "physical" companions; if at least one of them is offline, KOBJ_CHANGE is emitted for that "physical" device, and acpi_scan_hot_remove() returns, [I guess we can just emit KOBJ_CHANGE for the first companion that is not offline at this point.]
(4) user space notices the KOBJ_CHANGE and does the cleanup as needed; in the process it carries out the offline operation for the container's "physical" companion (there's only one such companion for each container), [That operation for the container itself is trivial, but to succeed it requires all devices below the container to be taken offline in advance.]
(5) user space uses the sysfs "eject" attribute of the ACPI container object to finally eject the container,
(6) acpi_scan_hot_remove() is now successful, because the container's "physical" companion is now offline,
(7) the "physical" container device goes away before executing _EJ0,
(8) the container is ejected.

Quoted 4 lines
Quoted 1 line

I think that should work for you.

This idea seems to same as your previous work.
http://lkml.org/lkml/2013/2/23/97

No, it is not. That one didn't involve physical device representations.

Quoted 1 line

How about add autoremove flag into acpi_hotplug_profile and check it as follow:

This is very similar to "enable" except that it generates the uevent and "enable" doesn't. You might as well modify "enable" to trigger a uevent if eject is not enabled (note that with the latest patches in linux-next "enable" only applies to eject).

That said I don't think we should generate any uevents for struct acpi_device objects, because they are not devices.

Quoted 21 lines
---
  drivers/acpi/scan.c | 5 +++++
  1 file changed, 5 insertions(+)
drivers/acpi/scan.c +5 -0
index 5383c81..c43d110 100644
@@ -409,6 +409,11 @@ static void acpi_hotplug_notify_cb(acpi_handle handle, u32 type, void *data)
  			ost_code = ACPI_OST_SC_EJECT_NOT_SUPPORTED;
  			goto err_out;
  		}
+		if (!handler->hotplug.autoremove) {
+			kobject_uevent(&device->dev.kobj, KOBJ_CHANGE);
+			ost_code = ACPI_OST_SC_EJECT_NON_SPECIFIC_FAILURE;
+			goto err_out;
+		}
  		acpi_evaluate_hotplug_ost(handle, ACPI_NOTIFY_EJECT_REQUEST,
  					  ACPI_OST_SC_EJECT_IN_PROGRESS, NULL);
  		break;
Quoted 2 lines

Adding the check into "acpi_hotplug_notify_cb()", user need not change the flag for removing container device by "sysfs eject".

Which is utterly confusing. There is no reason whatsoever why the sysfs eject attribute should work differently from the event-triggered eject. Quite the opposite is the case: it should work in the same way in my opinion so that it is possible to test the eject code path using that attribute.

Quoted 2 lines

I'm traveling now, but when I get back home (next week), I'll try to implement the thing I was talking about above.

It took some more time than I had expected, but I finally was able to get to that.

The following two patches implement the idea. This is the minimum (in my opinion) implementation and it may be extended in some ways.

Patch [1/2] introduces a new demand_offline flag for struct acpi_hotplug_profile that makes acpi_scan_hot_remove() check the offline status of the device object's companion physical devices to start with and return -EBUSY if at least one of them is not offline.

Patch [2/2] uses that flag to implement the container handling. The details are in the changelog, but that's how it is supposed to work.

During the initial namespace scan the container ACPI scan handler should create "physical" system container device under /sys/devices/system/container/ for each ACPI container object (the sysfs name of that device should be the same as the sysfs name of the corresponding container object and they should be linked to each other via the firmware_node and physical_node symbolic links, respectively). Those system container devices are initially online.

When a container eject event happens, acpi_scan_hot_remove() will notice that hotplug.demand_offline is set in the device object's scan handler and will check the online status of its "physical" companion device, which is online (that is the system container device the above paragraph is about). That will cause KOBJ_CHANGE to be emitted for the system container device and -EBUSY to be returned by acpi_scan_hot_remove().

Now, user space needs to offline the system container device through its online sysfs attribute (that should be present, because the bus type for containers provides the online and offline callbacks). However, the offline for system container devices will only succeed if the physical devices right below the container are all offline, so user space will have to offline those devices before attempting to offline the system container device itself. When finished, user space can trigger the container removal with the help of the eject sysfs attribute of the ACPI container object pointed to by the system container device's firmware_node link (this time the check in acpi_scan_hot_remove() will succeed, because the system container device in question is now offline).

Quoted 6 lines

The way it is implemented is a bit hackish (the driver_data pointer is slightly abused), but that's a special case and I wanted to avoid adding new fields to struct device just for handling it.

The patches haven't been tested yet. I'm going to do that later today, but first I need to take care of some other things, so that has to wait.

The series of the two patches:

[1/2] ACPI / hotplug: Add demand_offline hotplug profile flag

	https://patchwork.kernel.org/patch/3396711/

[2/2] ACPI / hotplug / driver core: Handle containers in a special way

	https://patchwork.kernel.org/patch/3399081/

has been tested now and seems to work as expected, at least for a container that has no children (that's one I could simulate easily in a meaningful way).

For this reason, if there are no objections, I'll resend them as an official submission during the next couple of days.

Thanks!

Signature · 2 lines
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.
#36 lore raw

Re: [PATCH 7/10] ACPI / hotplug: Move container-specific code out of the core

HI Rafael,

Quoted 151 lines · Rafael J. Wysocki
Quoted 150 lines · Rafael J. Wysocki
Quoted 107 lines · Rafael J. Wysocki
Quoted 60 lines · Yasuaki Ishimatsu
Quoted 59 lines · Rafael J. Wysocki
Quoted 1 line · Yasuaki Ishimatsu

Hi Rafael,

Hi,

Quoted 17 lines

Please share your more detailed idea. I started to implement the following idea. But the idea has one problem.

Quoted 5 lines
Quoted 5 lines
Quoted 5 lines

The eject work flow can be:
(1) an eject event occurs,
(2) the container "physical" device fails offline in acpi_scan_hot_remove() emmitting, say, KOBJ_CHANGE for the "physical" device,
(3) user space notices the KOBJ_CHANGE and does the cleanup as needed,

Quoted 8 lines
Quoted 8 lines
Quoted 8 lines

(4) user space changes the "physical" container device flag controlling offline to 0,
(5) user space uses the sysfs "eject" attribute of the ACPI container object to finally eject the container,
(6) the offline in acpi_scan_hot_remove() is now successful, because the flag controlling it has been set to 0 in step (4),
(7) the "physical" container device goes away before executing _EJ0,
(8) the container is ejected.

Quoted 5 lines

I want to emit KOBJ_CHANGE before offlining devices on container device at (2). But acpi_scan_hot_remove() offlines devices on container device at first. So when offline container device, devices on container has been offlined.

Thus the idea cannot fulfill my necessary feature.

Well, in that case we need to treat containers in a special way at the ACPI level. Which is a bit unfortunate so to speak.

To that end I'd try to add a new flag to struct acpi_hotplug_profile, say .verify_offline, such that if set, it would cause acpi_scan_hot_remove() to check if all of the "physical" companions of the top-level device are offline to start with, and if not, it would just emit KOBJ_CHANGE for the companions that are not offline and bail out.

So the above algorithm would become:

(1) an eject event occurs,
(2) acpi_scan_hot_remove() checks the verify_offline flag in the target device's scan_handler structure,
(3) if set (it would always be set for containers), acpi_scan_hot_remove() checks the status of the target device's "physical" companions; if at least one of them is offline, KOBJ_CHANGE is emitted for that "physical" device, and acpi_scan_hot_remove() returns, [I guess we can just emit KOBJ_CHANGE for the first companion that is not offline at this point.]
(4) user space notices the KOBJ_CHANGE and does the cleanup as needed; in the process it carries out the offline operation for the container's "physical" companion (there's only one such companion for each container), [That operation for the container itself is trivial, but to succeed it requires all devices below the container to be taken offline in advance.]
(5) user space uses the sysfs "eject" attribute of the ACPI container object to finally eject the container,
(6) acpi_scan_hot_remove() is now successful, because the container's "physical" companion is now offline,
(7) the "physical" container device goes away before executing _EJ0,
(8) the container is ejected.

Quoted 4 lines
Quoted 1 line

I think that should work for you.

This idea seems to same as your previous work.
http://lkml.org/lkml/2013/2/23/97

No, it is not. That one didn't involve physical device representations.

Quoted 1 line

How about add autoremove flag into acpi_hotplug_profile and check it as follow:

This is very similar to "enable" except that it generates the uevent and "enable" doesn't. You might as well modify "enable" to trigger a uevent if eject is not enabled (note that with the latest patches in linux-next "enable" only applies to eject).

That said I don't think we should generate any uevents for struct acpi_device objects, because they are not devices.

Quoted 21 lines
---
   drivers/acpi/scan.c | 5 +++++
   1 file changed, 5 insertions(+)
drivers/acpi/scan.c +5 -0
index 5383c81..c43d110 100644
@@ -409,6 +409,11 @@ static void acpi_hotplug_notify_cb(acpi_handle handle, u32 type, void *data)
   			ost_code = ACPI_OST_SC_EJECT_NOT_SUPPORTED;
   			goto err_out;
   		}
+		if (!handler->hotplug.autoremove) {
+			kobject_uevent(&device->dev.kobj, KOBJ_CHANGE);
+			ost_code = ACPI_OST_SC_EJECT_NON_SPECIFIC_FAILURE;
+			goto err_out;
+		}
   		acpi_evaluate_hotplug_ost(handle, ACPI_NOTIFY_EJECT_REQUEST,
   					  ACPI_OST_SC_EJECT_IN_PROGRESS, NULL);
   		break;
Quoted 2 lines

Adding the check into "acpi_hotplug_notify_cb()", user need not change the flag for removing container device by "sysfs eject".

Which is utterly confusing. There is no reason whatsoever why the sysfs eject attribute should work differently from the event-triggered eject. Quite the opposite is the case: it should work in the same way in my opinion so that it is possible to test the eject code path using that attribute.

Quoted 2 lines

I'm traveling now, but when I get back home (next week), I'll try to implement the thing I was talking about above.

It took some more time than I had expected, but I finally was able to get to that.

The following two patches implement the idea. This is the minimum (in my opinion) implementation and it may be extended in some ways.

Patch [1/2] introduces a new demand_offline flag for struct acpi_hotplug_profile that makes acpi_scan_hot_remove() check the offline status of the device object's companion physical devices to start with and return -EBUSY if at least one of them is not offline.

Patch [2/2] uses that flag to implement the container handling. The details are in the changelog, but that's how it is supposed to work.

During the initial namespace scan the container ACPI scan handler should create "physical" system container device under /sys/devices/system/container/ for each ACPI container object (the sysfs name of that device should be the same as the sysfs name of the corresponding container object and they should be linked to each other via the firmware_node and physical_node symbolic links, respectively). Those system container devices are initially online.

When a container eject event happens, acpi_scan_hot_remove() will notice that hotplug.demand_offline is set in the device object's scan handler and will check the online status of its "physical" companion device, which is online (that is the system container device the above paragraph is about). That will cause KOBJ_CHANGE to be emitted for the system container device and -EBUSY to be returned by acpi_scan_hot_remove().

Now, user space needs to offline the system container device through its online sysfs attribute (that should be present, because the bus type for containers provides the online and offline callbacks). However, the offline for system container devices will only succeed if the physical devices right below the container are all offline, so user space will have to offline those devices before attempting to offline the system container device itself. When finished, user space can trigger the container removal with the help of the eject sysfs attribute of the ACPI container object pointed to by the system container device's firmware_node link (this time the check in acpi_scan_hot_remove() will succeed, because the system container device in question is now offline).

Quoted 7 lines
Quoted 6 lines

The way it is implemented is a bit hackish (the driver_data pointer is slightly abused), but that's a special case and I wanted to avoid adding new fields to struct device just for handling it.

The patches haven't been tested yet. I'm going to do that later today, but first I need to take care of some other things, so that has to wait.

Thank you for implementing your idea.

Quoted 5 lines

The series of the two patches:

[1/2] ACPI / hotplug: Add demand_offline hotplug profile flag

	https://patchwork.kernel.org/patch/3396711/
Quoted 8 lines

[2/2] ACPI / hotplug / driver core: Handle containers in a special way

	https://patchwork.kernel.org/patch/3399081/

has been tested now and seems to work as expected, at least for a container that has no children (that's one I could simulate easily in a meaningful way).

For this reason, if there are no objections, I'll resend them as an official submission during the next couple of days.

I'm testing these patches now. If I have a comment, I send it to these threads.

Thanks,
Yasuaki Ishimatsu

Quoted 3 lines

Thanks!

#37 lore raw

Re: [PATCH 7/10] ACPI / hotplug: Move container-specific code out of the core

Quoted 158 lines · Yasuaki Ishimatsu

HI Rafael,

Quoted 155 lines · Rafael J. Wysocki
Quoted 154 lines · Rafael J. Wysocki
Quoted 107 lines · Rafael J. Wysocki
Quoted 60 lines · Yasuaki Ishimatsu
Quoted 59 lines · Rafael J. Wysocki
Quoted 1 line · Yasuaki Ishimatsu

Hi Rafael,

Hi,

Quoted 17 lines

Please share your more detailed idea. I started to implement the following idea. But the idea has one problem.

Quoted 5 lines
Quoted 5 lines
Quoted 5 lines

The eject work flow can be:
(1) an eject event occurs,
(2) the container "physical" device fails offline in acpi_scan_hot_remove() emmitting, say, KOBJ_CHANGE for the "physical" device,
(3) user space notices the KOBJ_CHANGE and does the cleanup as needed,

Quoted 8 lines
Quoted 8 lines
Quoted 8 lines

(4) user space changes the "physical" container device flag controlling offline to 0,
(5) user space uses the sysfs "eject" attribute of the ACPI container object to finally eject the container,
(6) the offline in acpi_scan_hot_remove() is now successful, because the flag controlling it has been set to 0 in step (4),
(7) the "physical" container device goes away before executing _EJ0,
(8) the container is ejected.

Quoted 5 lines

I want to emit KOBJ_CHANGE before offlining devices on container device at (2). But acpi_scan_hot_remove() offlines devices on container device at first. So when offline container device, devices on container has been offlined.

Thus the idea cannot fulfill my necessary feature.

Well, in that case we need to treat containers in a special way at the ACPI level. Which is a bit unfortunate so to speak.

To that end I'd try to add a new flag to struct acpi_hotplug_profile, say .verify_offline, such that if set, it would cause acpi_scan_hot_remove() to check if all of the "physical" companions of the top-level device are offline to start with, and if not, it would just emit KOBJ_CHANGE for the companions that are not offline and bail out.

So the above algorithm would become:

(1) an eject event occurs,
(2) acpi_scan_hot_remove() checks the verify_offline flag in the target device's scan_handler structure,
(3) if set (it would always be set for containers), acpi_scan_hot_remove() checks the status of the target device's "physical" companions; if at least one of them is offline, KOBJ_CHANGE is emitted for that "physical" device, and acpi_scan_hot_remove() returns, [I guess we can just emit KOBJ_CHANGE for the first companion that is not offline at this point.]
(4) user space notices the KOBJ_CHANGE and does the cleanup as needed; in the process it carries out the offline operation for the container's "physical" companion (there's only one such companion for each container), [That operation for the container itself is trivial, but to succeed it requires all devices below the container to be taken offline in advance.]
(5) user space uses the sysfs "eject" attribute of the ACPI container object to finally eject the container,
(6) acpi_scan_hot_remove() is now successful, because the container's "physical" companion is now offline,
(7) the "physical" container device goes away before executing _EJ0,
(8) the container is ejected.

Quoted 4 lines
Quoted 1 line

I think that should work for you.

This idea seems to same as your previous work.
http://lkml.org/lkml/2013/2/23/97

No, it is not. That one didn't involve physical device representations.

Quoted 1 line

How about add autoremove flag into acpi_hotplug_profile and check it as follow:

This is very similar to "enable" except that it generates the uevent and "enable" doesn't. You might as well modify "enable" to trigger a uevent if eject is not enabled (note that with the latest patches in linux-next "enable" only applies to eject).

That said I don't think we should generate any uevents for struct acpi_device objects, because they are not devices.

Quoted 21 lines
---
   drivers/acpi/scan.c | 5 +++++
   1 file changed, 5 insertions(+)
drivers/acpi/scan.c +5 -0
index 5383c81..c43d110 100644
@@ -409,6 +409,11 @@ static void acpi_hotplug_notify_cb(acpi_handle handle, u32 type, void *data)
   			ost_code = ACPI_OST_SC_EJECT_NOT_SUPPORTED;
   			goto err_out;
   		}
+		if (!handler->hotplug.autoremove) {
+			kobject_uevent(&device->dev.kobj, KOBJ_CHANGE);
+			ost_code = ACPI_OST_SC_EJECT_NON_SPECIFIC_FAILURE;
+			goto err_out;
+		}
   		acpi_evaluate_hotplug_ost(handle, ACPI_NOTIFY_EJECT_REQUEST,
   					  ACPI_OST_SC_EJECT_IN_PROGRESS, NULL);
   		break;
Quoted 2 lines

Adding the check into "acpi_hotplug_notify_cb()", user need not change the flag for removing container device by "sysfs eject".

Which is utterly confusing. There is no reason whatsoever why the sysfs eject attribute should work differently from the event-triggered eject. Quite the opposite is the case: it should work in the same way in my opinion so that it is possible to test the eject code path using that attribute.

Quoted 2 lines

I'm traveling now, but when I get back home (next week), I'll try to implement the thing I was talking about above.

It took some more time than I had expected, but I finally was able to get to that.

The following two patches implement the idea. This is the minimum (in my opinion) implementation and it may be extended in some ways.

Patch [1/2] introduces a new demand_offline flag for struct acpi_hotplug_profile that makes acpi_scan_hot_remove() check the offline status of the device object's companion physical devices to start with and return -EBUSY if at least one of them is not offline.

Patch [2/2] uses that flag to implement the container handling. The details are in the changelog, but that's how it is supposed to work.

During the initial namespace scan the container ACPI scan handler should create "physical" system container device under /sys/devices/system/container/ for each ACPI container object (the sysfs name of that device should be the same as the sysfs name of the corresponding container object and they should be linked to each other via the firmware_node and physical_node symbolic links, respectively). Those system container devices are initially online.

When a container eject event happens, acpi_scan_hot_remove() will notice that hotplug.demand_offline is set in the device object's scan handler and will check the online status of its "physical" companion device, which is online (that is the system container device the above paragraph is about). That will cause KOBJ_CHANGE to be emitted for the system container device and -EBUSY to be returned by acpi_scan_hot_remove().

Now, user space needs to offline the system container device through its online sysfs attribute (that should be present, because the bus type for containers provides the online and offline callbacks). However, the offline for system container devices will only succeed if the physical devices right below the container are all offline, so user space will have to offline those devices before attempting to offline the system container device itself. When finished, user space can trigger the container removal with the help of the eject sysfs attribute of the ACPI container object pointed to by the system container device's firmware_node link (this time the check in acpi_scan_hot_remove() will succeed, because the system container device in question is now offline).

The way it is implemented is a bit hackish (the driver_data pointer is slightly abused), but that's a special case and I wanted to avoid adding new fields to struct device just for handling it.

Quoted 5 lines
Quoted 3 lines
Quoted 2 lines

The patches haven't been tested yet. I'm going to do that later today, but first I need to take care of some other things, so that has to wait.

Thank you for implementing your idea.

You're very welcome! :-)

Quoted 8 lines
Quoted 8 lines

The series of the two patches:

[1/2] ACPI / hotplug: Add demand_offline hotplug profile flag

	https://patchwork.kernel.org/patch/3396711/

[2/2] ACPI / hotplug / driver core: Handle containers in a special way

	https://patchwork.kernel.org/patch/3399081/
Quoted 8 lines
Quoted 5 lines

has been tested now and seems to work as expected, at least for a container that has no children (that's one I could simulate easily in a meaningful way).

For this reason, if there are no objections, I'll resend them as an official submission during the next couple of days.

I'm testing these patches now. If I have a comment, I send it to these threads.

Thanks a lot!

Rafael

#38 lore raw

PATCH 8/10 ACPI / hotplug: Rework generic code to handle suprise removals

In reply to Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

The generic ACPI hotplug code used for several types of device doesn't handle surprise removals, mostly because those devices currently cannot be removed by surprise in the majority of systems. However, surprise removals should be handled by that code as well as surprise additions of devices, so make it do that.

---
 drivers/acpi/scan.c |   85 ++++++++++++++++++++++++++++++++++++----------------
 1 file changed, 59 insertions(+), 26 deletions(-)
drivers/acpi/scan.c +59 -26
Index: linux-pm/drivers/acpi/scan.c
===================================================================
@@ -276,18 +276,72 @@ static int acpi_scan_hot_remove(struct a
 	return 0;
 }
 
+static int acpi_scan_device_not_present(struct acpi_device *adev)
+{
+	if (!acpi_device_enumerated(adev)) {
+		dev_warn(&adev->dev, "Still not present\n");
+		return -EALREADY;
+	}
+	acpi_bus_trim(adev);
+	return 0;
+}
+
 static int acpi_scan_device_check(struct acpi_device *adev)
 {
 	int error;
 
+	acpi_bus_get_status(adev);
+	if (adev->status.present || adev->status.functional) {
+		/*
+		 * This function is only called for device objects for which
+		 * matching scan handlers exist.  The only situation in which
+		 * the scan handler is not attached to this device object yet
+		 * is when the device has just appeared (either it wasn't
+		 * present at all before or it was removed and then added
+		 * again).
+		 */
+		if (adev->handler) {
+			dev_warn(&adev->dev, "Already enumerated\n");
+			return -EALREADY;
+		}
+		error = acpi_bus_scan(adev->handle);
+		if (error) {
+			dev_warn(&adev->dev, "Namespace scan failure\n");
+			return error;
+		}
+		if (!adev->handler) {
+			dev_warn(&adev->dev, "Enumeration failure\n");
+			error = -ENODEV;
+		}
+	} else {
+		error = acpi_scan_device_not_present(adev);
+	}
+	return error;
+}
+
+static int acpi_scan_bus_check(struct acpi_device *adev)
+{
+	struct acpi_scan_handler *handler = adev->handler;
+	struct acpi_device *child;
+	int error;
+
+	acpi_bus_get_status(adev);
+	if (!(adev->status.present || adev->status.functional)) {
+		acpi_scan_device_not_present(adev);
+		return 0;
+	}
+	if (handler && handler->hotplug.scan_dependent)
+		return handler->hotplug.scan_dependent(adev);
+
 	error = acpi_bus_scan(adev->handle);
 	if (error) {
 		dev_warn(&adev->dev, "Namespace scan failure\n");
 		return error;
 	}
-	if (!adev->handler) {
-		dev_warn(&adev->dev, "Enumeration failure\n");
-		return -ENODEV;
+	list_for_each_entry(child, &adev->children, node) {
+		error = acpi_scan_bus_check(child);
+		if (error)
+			return error;
 	}
 	return 0;
 }
@@ -296,7 +350,6 @@ static void acpi_device_hotplug(void *da
 {
 	u32 ost_code = ACPI_OST_SC_NON_SPECIFIC_FAILURE;
 	struct acpi_device *adev = data;
-	struct acpi_scan_handler *handler;
 	int error;
 
 	lock_device_hotplug();
@@ -310,32 +363,12 @@ static void acpi_device_hotplug(void *da
 	if (adev->handle == INVALID_ACPI_HANDLE)
 		goto out;
 
-	handler = adev->handler;
-
 	switch (src) {
 	case ACPI_NOTIFY_BUS_CHECK:
-		if (handler) {
-			error = handler->hotplug.scan_dependent ?
-					handler->hotplug.scan_dependent(adev) :
-					acpi_bus_scan(adev->handle);
-		} else {
-			error = acpi_scan_device_check(adev);
-		}
+		error = acpi_scan_bus_check(adev);
 		break;
 	case ACPI_NOTIFY_DEVICE_CHECK:
-		/*
-		 * This code is only run for device objects for which matching
-		 * scan handlers exist.  The only situation in which the scan
-		 * handler is not attached to this device object yet is when the
-		 * device has just appeared (either it wasn't present at all
-		 * before or it was removed and then added again).
-		 */
-		if (adev->handler) {
-			dev_warn(&adev->dev, "Already enumerated\n");
-			error = -EBUSY;
-		} else {
-			error = acpi_scan_device_check(adev);
-		}
+		error = acpi_scan_device_check(adev);
 		break;
 	case ACPI_NOTIFY_EJECT_REQUEST:
 	case ACPI_OST_EC_OSPM_EJECT:
#39 lore raw

PATCH 9/10 ACPI / hotplug: Drop unfinished global notification handling routines

In reply to Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

There are two global hotplug notification handling routines in bus.c, acpi_bus_check_device() and acpi_bus_check_scope(), that have never been finished and don't do anything useful, so drop them.

---
 drivers/acpi/bus.c |   64 +----------------------------------------------------
 1 file changed, 2 insertions(+), 62 deletions(-)
drivers/acpi/bus.c +2 -62
Index: linux-pm/drivers/acpi/bus.c
===================================================================
@@ -329,58 +329,6 @@ static void acpi_bus_osc_support(void)
                              Notification Handling
    -------------------------------------------------------------------------- */
 
-static void acpi_bus_check_device(acpi_handle handle)
-{
-	struct acpi_device *device;
-	acpi_status status;
-	struct acpi_device_status old_status;
-
-	if (acpi_bus_get_device(handle, &device))
-		return;
-	if (!device)
-		return;
-
-	old_status = device->status;
-
-	/*
-	 * Make sure this device's parent is present before we go about
-	 * messing with the device.
-	 */
-	if (device->parent && !device->parent->status.present) {
-		device->status = device->parent->status;
-		return;
-	}
-
-	status = acpi_bus_get_status(device);
-	if (ACPI_FAILURE(status))
-		return;
-
-	if (STRUCT_TO_INT(old_status) == STRUCT_TO_INT(device->status))
-		return;
-
-	/*
-	 * Device Insertion/Removal
-	 */
-	if ((device->status.present) && !(old_status.present)) {
-		ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Device insertion detected\n"));
-		/* TBD: Handle device insertion */
-	} else if (!(device->status.present) && (old_status.present)) {
-		ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Device removal detected\n"));
-		/* TBD: Handle device removal */
-	}
-}
-
-static void acpi_bus_check_scope(acpi_handle handle)
-{
-	/* Status Change? */
-	acpi_bus_check_device(handle);
-
-	/*
-	 * TBD: Enumerate child devices within this device's scope and
-	 *       run acpi_bus_check_device()'s on them.
-	 */
-}
-
 /**
  * acpi_bus_notify
  * ---------------
@@ -397,19 +345,11 @@ static void acpi_bus_notify(acpi_handle
 	switch (type) {
 
 	case ACPI_NOTIFY_BUS_CHECK:
-		acpi_bus_check_scope(handle);
-		/*
-		 * TBD: We'll need to outsource certain events to non-ACPI
-		 *      drivers via the device manager (device.c).
-		 */
+		/* TBD */
 		break;
 
 	case ACPI_NOTIFY_DEVICE_CHECK:
-		acpi_bus_check_device(handle);
-		/*
-		 * TBD: We'll need to outsource certain events to non-ACPI
-		 *      drivers via the device manager (device.c).
-		 */
+		/* TBD */
 		break;
 
 	case ACPI_NOTIFY_DEVICE_WAKE:
#40 lore raw

PATCH 10/10 ACPI: Introduce acpi_set_device_status()

In reply to Rafael J. Wysocki

From: Rafael J. Wysocki <[email protected]>

Introduce a static inline function for setting the status field of struct acpi_device on the basis of a supplied u32 number, acpi_set_device_status(), and use it instead of the horrible horrible STRUCT_TO_INT() macro wherever applicable. Having done that, drop STRUCT_TO_INT() (and pretend that it has never existed).

---
 drivers/acpi/bus.c      |   11 +++--------
 drivers/acpi/scan.c     |    5 ++---
 include/acpi/acpi_bus.h |    5 +++++
 3 files changed, 10 insertions(+), 11 deletions(-)
drivers/acpi/bus.c +3 -8
Index: linux-pm/drivers/acpi/bus.c
===================================================================
@@ -52,9 +52,6 @@ struct acpi_device *acpi_root;
 struct proc_dir_entry *acpi_root_dir;
 EXPORT_SYMBOL(acpi_root_dir);
 
-#define STRUCT_TO_INT(s)	(*((int*)&s))
-
-
 #ifdef CONFIG_X86
 static int set_copy_dsdt(const struct dmi_system_id *id)
 {
@@ -115,18 +112,16 @@ int acpi_bus_get_status(struct acpi_devi
 	if (ACPI_FAILURE(status))
 		return -ENODEV;
 
-	STRUCT_TO_INT(device->status) = (int) sta;
+	acpi_set_device_status(device, sta);
 
 	if (device->status.functional && !device->status.present) {
 		ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Device [%s] status [%08x]: "
 		       "functional but not present;\n",
-			device->pnp.bus_id,
-			(u32) STRUCT_TO_INT(device->status)));
+			device->pnp.bus_id, (u32)sta));
 	}
 
 	ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Device [%s] status [%08x]\n",
-			  device->pnp.bus_id,
-			  (u32) STRUCT_TO_INT(device->status)));
+			  device->pnp.bus_id, (u32)sta));
 	return 0;
 }
 EXPORT_SYMBOL(acpi_bus_get_status);
drivers/acpi/scan.c +2 -3
Index: linux-pm/drivers/acpi/scan.c
===================================================================
@@ -20,7 +20,6 @@
 
 #define _COMPONENT		ACPI_BUS_COMPONENT
 ACPI_MODULE_NAME("scan");
-#define STRUCT_TO_INT(s)	(*((int*)&s))
 extern struct acpi_device *acpi_root;
 
 #define ACPI_BUS_CLASS			"system_bus"
@@ -1683,7 +1682,7 @@ void acpi_init_device_object(struct acpi
 	device->device_type = type;
 	device->handle = handle;
 	device->parent = acpi_bus_get_parent(handle);
-	STRUCT_TO_INT(device->status) = sta;
+	acpi_set_device_status(device, sta);
 	acpi_device_get_busid(device);
 	acpi_set_pnp_ids(handle, &device->pnp, type);
 	acpi_bus_get_flags(device);
@@ -1927,7 +1926,7 @@ static acpi_status acpi_bus_device_attac
 	if (acpi_bus_get_device(handle, &device))
 		return AE_CTRL_DEPTH;
 
-	STRUCT_TO_INT(device->status) = sta;
+	acpi_set_device_status(device, sta);
 	/* Skip devices that are not present. */
 	if (!acpi_device_is_present(device))
 		goto err;
include/acpi/acpi_bus.h +5 -0
Index: linux-pm/include/acpi/acpi_bus.h
===================================================================
@@ -320,6 +320,11 @@ static inline void *acpi_driver_data(str
 #define to_acpi_device(d)	container_of(d, struct acpi_device, dev)
 #define to_acpi_driver(d)	container_of(d, struct acpi_driver, drv)
 
+static inline void acpi_set_device_status(struct acpi_device *adev, u32 sta)
+{
+	*((u32 *)&adev->status) = sta;
+}
+
 /* acpi_device.dev.bus == &acpi_bus_type */
 extern struct bus_type acpi_bus_type;
 
#41 lore raw
In reply to Rafael J. Wysocki
Quoted 3 lines · Rafael J. Wysocki

This is 3.14 material, but I'm posting it today to let you know it is coming. Of course, if you see any immediate problems with the following patches, please let me know.

For what it's worth, I have tested this series on three different systems that have Thunderbolt and it still works fine after the series has been applied :)

#42 lore raw
Quoted 8 lines · Mika Westerberg
Quoted 3 lines · Rafael J. Wysocki

This is 3.14 material, but I'm posting it today to let you know it is coming. Of course, if you see any immediate problems with the following patches, please let me know.

For what it's worth, I have tested this series on three different systems that have Thunderbolt and it still works fine after the series has been applied :)

Awesome, thanks!

I hope you won't mind if I add your Tested-by to the series. :-)

Signature · 2 lines
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.
Tested-by
#43 lore raw
Quoted 6 lines · Rafael J. Wysocki
Quoted 5 lines · Mika Westerberg
Quoted 3 lines · Rafael J. Wysocki

This is 3.14 material, but I'm posting it today to let you know it is coming. Of course, if you see any immediate problems with the following patches, please let me know.

Quoted 7 lines
Quoted 3 lines

For what it's worth, I have tested this series on three different systems that have Thunderbolt and it still works fine after the series has been applied :)

Awesome, thanks!

I hope you won't mind if I add your Tested-by to the series. :-)

Not at all,