Re: [PATCH v8 3/4] driver core: shut down devices asynchronously

From: Christophe JAILLET
Date: Sun Sep 08 2024 - 10:44:53 EST


Le 22/08/2024 à 22:28, Stuart Hayes a écrit :
Add code to allow asynchronous shutdown of devices, ensuring that each
device is shut down before its parents & suppliers.

Only devices with drivers that have async_shutdown_enable enabled will be
shut down asynchronously.

This can dramatically reduce system shutdown/reboot time on systems that
have multiple devices that take many seconds to shut down (like certain
NVMe drives). On one system tested, the shutdown time went from 11 minutes
without this patch to 55 seconds with the patch.

Signed-off-by: Stuart Hayes <stuart.w.hayes@xxxxxxxxx>
Signed-off-by: David Jeffery <djeffery@xxxxxxxxxx>
---

...

+/**
+ * shutdown_one_device_async
+ * @data: the pointer to the struct device to be shutdown
+ * @cookie: not used
+ *
+ * Shuts down one device, after waiting for shutdown_after to complete.
+ * shutdown_after should be set to the cookie of the last child or consumer
+ * of this device to be shutdown (if any), or to the cookie of the previous
+ * device to be shut down for devices that don't enable asynchronous shutdown.
+ */
+static void shutdown_one_device_async(void *data, async_cookie_t cookie)
+{
+ struct device *dev = data;
+
+ async_synchronize_cookie_domain(dev->p->shutdown_after + 1, &sd_domain);
+
+ shutdown_one_device(dev);
+}
+
/**
* device_shutdown - call ->shutdown() on each device to shutdown.
*/
void device_shutdown(void)
{
struct device *dev, *parent;
+ async_cookie_t cookie = 0;
+ struct device_link *link;
+ int idx;
wait_for_device_probe();
device_block_probing();
@@ -4852,11 +4878,37 @@ void device_shutdown(void)
list_del_init(&dev->kobj.entry);
spin_unlock(&devices_kset->list_lock);
- shutdown_one_device(dev);
+
+ /*
+ * Set cookie for devices that will be shut down synchronously
+ */
+ if (!dev->driver || !dev->driver->async_shutdown_enable)
+ dev->p->shutdown_after = cookie;
+
+ get_device(dev);
+ get_device(parent);
+
+ cookie = async_schedule_domain(shutdown_one_device_async,
+ dev, &sd_domain);
+ /*
+ * Ensure parent & suppliers wait for this device to shut down
+ */
+ if (parent) {
+ parent->p->shutdown_after = cookie;
+ put_device(parent);

Would it make sense to have this put_device() out of the if block?

IIUC, the behavior would be exactly the same, but it is more intuitive to have a put_device(parent) called for each get_device(parent) call.

Another way to keep symmetry is to have:
if (parent)
get_device(parent);
above.

+ }
+
+ idx = device_links_read_lock();
+ list_for_each_entry_rcu(link, &dev->links.suppliers, c_node,
+ device_links_read_lock_held())
+ link->supplier->p->shutdown_after = cookie;
+ device_links_read_unlock(idx);
+ put_device(dev);
spin_lock(&devices_kset->list_lock);
}
spin_unlock(&devices_kset->list_lock);
+ async_synchronize_full_domain(&sd_domain);
}
/*
diff --git a/include/linux/device/driver.h b/include/linux/device/driver.h
index 1fc8b68786de..2b6127faaa25 100644
--- a/include/linux/device/driver.h
+++ b/include/linux/device/driver.h
@@ -56,6 +56,7 @@ enum probe_type {
* @mod_name: Used for built-in modules.
* @suppress_bind_attrs: Disables bind/unbind via sysfs.
* @probe_type: Type of the probe (synchronous or asynchronous) to use.
+ * @async_shutdown_enable: Enables devices to be shutdown asynchronously.
* @of_match_table: The open firmware table.
* @acpi_match_table: The ACPI match table.
* @probe: Called to query the existence of a specific device,
@@ -102,6 +103,7 @@ struct device_driver {
bool suppress_bind_attrs; /* disables bind/unbind via sysfs */
enum probe_type probe_type;
+ bool async_shutdown_enable;

Maybe keep these 2 bools together to potentially avoid hole?

Just my 2c.

CJ

const struct of_device_id *of_match_table;
const struct acpi_device_id *acpi_match_table;