[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-Id: <20200515053500.215929-4-saravanak@google.com>
Date: Thu, 14 May 2020 22:34:59 -0700
From: Saravana Kannan <saravanak@...gle.com>
To: Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
"Rafael J. Wysocki" <rafael@...nel.org>,
Rob Herring <robh+dt@...nel.org>,
Frank Rowand <frowand.list@...il.com>,
Len Brown <lenb@...nel.org>
Cc: Saravana Kannan <saravanak@...gle.com>, kernel-team@...roid.com,
linux-kernel@...r.kernel.org, devicetree@...r.kernel.org,
linux-acpi@...r.kernel.org, Ji Luo <ji.luo@....com>
Subject: [PATCH v1 3/4] driver core: fw_devlink: Add support for batching
fwnode parsing
The amount of time spent parsing fwnodes of devices can become really
high if the devices are added in an non-ideal order. Worst case can be
O(N^2) when N devices are added. But this can be optimized to O(N) by
adding all the devices and then parsing all their fwnodes in one batch.
This commit adds fw_devlink_pause() and fw_devlink_resume() to allow
doing this.
Signed-off-by: Saravana Kannan <saravanak@...gle.com>
---
drivers/base/base.h | 1 +
drivers/base/core.c | 116 ++++++++++++++++++++++++++++++++++++++---
drivers/base/dd.c | 8 +++
include/linux/fwnode.h | 2 +
4 files changed, 120 insertions(+), 7 deletions(-)
diff --git a/drivers/base/base.h b/drivers/base/base.h
index 40fb069a8a7e..95c22c0f9036 100644
--- a/drivers/base/base.h
+++ b/drivers/base/base.h
@@ -153,6 +153,7 @@ extern char *make_class_name(const char *name, struct kobject *kobj);
extern int devres_release_all(struct device *dev);
extern void device_block_probing(void);
extern void device_unblock_probing(void);
+extern void driver_deferred_probe_force_trigger(void);
/* /sys/devices directory */
extern struct kset *devices_kset;
diff --git a/drivers/base/core.c b/drivers/base/core.c
index f585d92e09d0..84c569726d75 100644
--- a/drivers/base/core.c
+++ b/drivers/base/core.c
@@ -49,6 +49,9 @@ static LIST_HEAD(wait_for_suppliers);
static DEFINE_MUTEX(wfs_lock);
static LIST_HEAD(deferred_sync);
static unsigned int defer_sync_state_count = 1;
+static unsigned int defer_fw_devlink_count;
+static DEFINE_MUTEX(defer_fw_devlink_lock);
+static bool fw_devlink_is_permissive(void);
#ifdef CONFIG_SRCU
static DEFINE_MUTEX(device_links_lock);
@@ -527,7 +530,7 @@ static void device_link_add_missing_supplier_links(void)
int ret = fwnode_call_int_op(dev->fwnode, add_links, dev);
if (!ret)
list_del_init(&dev->links.needs_suppliers);
- else if (ret != -ENODEV)
+ else if (ret != -ENODEV || fw_devlink_is_permissive())
dev->links.need_for_probe = false;
}
mutex_unlock(&wfs_lock);
@@ -1177,17 +1180,116 @@ static void fw_devlink_link_device(struct device *dev)
{
int fw_ret;
- device_link_add_missing_supplier_links();
+ if (!fw_devlink_flags)
+ return;
+
+ mutex_lock(&defer_fw_devlink_lock);
+ if (!defer_fw_devlink_count)
+ device_link_add_missing_supplier_links();
+
+ /*
+ * The device's fwnode not having add_links() doesn't affect if other
+ * consumers can find this device as a supplier. So, this check is
+ * intentionally placed after device_link_add_missing_supplier_links().
+ */
+ if (!fwnode_has_op(dev->fwnode, add_links))
+ goto out;
- if (fw_devlink_flags && fwnode_has_op(dev->fwnode, add_links)) {
+ /*
+ * If fw_devlink is being deferred, assume all devices have mandatory
+ * suppliers they need to link to later. Then, when the fw_devlink is
+ * resumed, all these devices will get a chance to try and link to any
+ * suppliers they have.
+ */
+ if (!defer_fw_devlink_count) {
fw_ret = fwnode_call_int_op(dev->fwnode, add_links, dev);
- if (fw_ret == -ENODEV && !fw_devlink_is_permissive())
- device_link_wait_for_mandatory_supplier(dev);
- else if (fw_ret)
- device_link_wait_for_optional_supplier(dev);
+ if (fw_ret == -ENODEV && fw_devlink_is_permissive())
+ fw_ret = -EAGAIN;
+ } else {
+ fw_ret = -ENODEV;
}
+
+ if (fw_ret == -ENODEV)
+ device_link_wait_for_mandatory_supplier(dev);
+ else if (fw_ret)
+ device_link_wait_for_optional_supplier(dev);
+
+out:
+ mutex_unlock(&defer_fw_devlink_lock);
}
+/**
+ * fw_devlink_pause - Pause parsing of fwnode to create device links
+ *
+ * Calling this function defers any fwnode parsing to create device links until
+ * fw_devlink_resume() is called. Both these functions are ref counted and the
+ * caller needs to match the calls.
+ *
+ * While fw_devlink is paused:
+ * - Any device that is added won't have its fwnode parsed to create device
+ * links.
+ * - The probe of the device will also be deferred during this period.
+ * - Any devices that were already added, but waiting for suppliers won't be
+ * able to link to newly added devices.
+ *
+ * Once fw_devlink_resume():
+ * - All the fwnodes that was not parsed will be parsed.
+ * - All the devices that were deferred probing will be reattempted if they
+ * aren't waiting for any more suppliers.
+ *
+ * This pair of functions, is mainly meant to optimize the parsing of fwnodes
+ * when a lot of devices that need to link to each other are added in a short
+ * interval of time. For example, adding all the top level devices in a system.
+ *
+ * For example, if N devices are added and:
+ * - All the consumers are added before their suppliers
+ * - All the suppliers of the N devices are part of the N devices
+ *
+ * Then:
+ *
+ * - With the use of fw_devlink_pause() and fw_devlink_resume(), each device
+ * will only need one parsing of its fwnode because it is guaranteed to find
+ * all the supplier devices already registered and ready to link to. It won't
+ * have to do another pass later to find one or more suppliers it couldn't
+ * find in the first parse of the fwnode. So, we'll only need O(N) fwnode
+ * parses.
+ *
+ * - Without the use of fw_devlink_pause() and fw_devlink_resume(), we would
+ * end up doing O(N^2) parses of fwnodes because every device that's added is
+ * guaranteed to trigger a parse of the fwnode of every device added before
+ * it. This O(N^2) parse is made worse by the fact that when a fwnode of a
+ * device is parsed, all it descendant devices might need to have their
+ * fwnodes parsed too (even if the devices themselves aren't added).
+ */
+void fw_devlink_pause(void)
+{
+ mutex_lock(&defer_fw_devlink_lock);
+ defer_fw_devlink_count++;
+ mutex_unlock(&defer_fw_devlink_lock);
+}
+
+/** fw_devlink_resume - Resume parsing of fwnode to create device links
+ *
+ * This function is used in conjunction with fw_devlink_pause() and is ref
+ * counted. See documentation for fw_devlink_pause() for more details.
+ */
+void fw_devlink_resume(void)
+{
+ mutex_lock(&defer_fw_devlink_lock);
+ if (!defer_fw_devlink_count) {
+ WARN(true, "Unmatched fw_devlink pause/resume!");
+ goto out;
+ }
+
+ defer_fw_devlink_count--;
+ if (defer_fw_devlink_count)
+ goto out;
+
+ device_link_add_missing_supplier_links();
+ driver_deferred_probe_force_trigger();
+out:
+ mutex_unlock(&defer_fw_devlink_lock);
+}
/* Device links support end. */
int (*platform_notify)(struct device *dev) = NULL;
diff --git a/drivers/base/dd.c b/drivers/base/dd.c
index 48ca81cb8ebc..63991d97adcc 100644
--- a/drivers/base/dd.c
+++ b/drivers/base/dd.c
@@ -160,6 +160,14 @@ static bool driver_deferred_probe_enable = false;
* again.
*/
static void driver_deferred_probe_trigger(void)
+{
+ if (!driver_deferred_probe_enable)
+ return;
+
+ driver_deferred_probe_force_trigger();
+}
+
+void driver_deferred_probe_force_trigger(void)
{
if (!driver_deferred_probe_enable)
return;
diff --git a/include/linux/fwnode.h b/include/linux/fwnode.h
index e0abafbb17f8..9506f8ec0974 100644
--- a/include/linux/fwnode.h
+++ b/include/linux/fwnode.h
@@ -171,5 +171,7 @@ struct fwnode_operations {
#define get_dev_from_fwnode(fwnode) get_device((fwnode)->dev)
extern u32 fw_devlink_get_flags(void);
+void fw_devlink_pause(void);
+void fw_devlink_resume(void);
#endif
--
2.26.2.761.g0e0b3e54be-goog
Powered by blists - more mailing lists