Skip to contentSkip to conversation

driver core: remove polling for driver_probe_done(v2)

PATCH In review [email protected] 3 participants 7 messages Jan 30, 2009

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 driver core: remove polling for driver_probe_done(v2)

From: Ming Lei <[email protected]>

This patch renames driver_probe_done to driver_wait_probe_done, and make it wait on condition of probing done to remove polling for it in fs initialization. Also add driver_wait_probe_done_and_dev_appear(name) to allow callers to wait until all probing is done AND until the device @name exists.

Removing polling in fs initialization may lead to a faster boot.

---
 drivers/base/dd.c      |   26 +++++++++++++++++++-------
 include/linux/device.h |    3 ++-
 init/do_mounts.c       |    9 ++++-----
 init/do_mounts_md.c    |    5 +++--
 4 files changed, 28 insertions(+), 15 deletions(-)
drivers/base/dd.c +19 -7
index 9b721d3..ab8c30f 100644
@@ -21,6 +21,7 @@
 #include <linux/module.h>
 #include <linux/kthread.h>
 #include <linux/wait.h>
+#include <linux/mount.h>
 
 #include "base.h"
 #include "power/power.h"
@@ -153,21 +154,32 @@ done:
 }
 
 /**
- * driver_probe_done
- * Determine if the probe sequence is finished or not.
+ * driver_probe_wait_done
+ * wait until the probe sequence is finished.
  *
- * Should somehow figure out how to use a semaphore, not an atomic variable...
  */
-int driver_probe_done(void)
+void driver_wait_probe_done(void)
 {
 	pr_debug("%s: probe_count = %d\n", __func__,
 		 atomic_read(&probe_count));
-	if (atomic_read(&probe_count))
-		return -EBUSY;
-	return 0;
+
+	wait_event(probe_waitqueue, atomic_read(&probe_count) == 0);
 }
 
 /**
+ * driver_wait_probe_done_and_dev_appear
+ * wait until the probe sequence is finished _and_ the dev @name appears.
+ *
+ */
+void driver_wait_probe_done_and_dev_appear(char *name)
+{
+	pr_debug("%s: probe_count = %d\n", __func__,
+		 atomic_read(&probe_count));
+
+	wait_event(probe_waitqueue, (atomic_read(&probe_count) == 0) &&
+		  name_to_dev_t(name));
+}
+/**
  * driver_probe_device - attempt to bind device & driver together
  * @drv: driver to bind a device to
  * @dev: device to try to bind to the driver
include/linux/device.h +2 -1
index 45e5b19..7e2e138 100644
@@ -146,7 +146,8 @@ extern struct device_driver *get_driver(struct device_driver *drv);
 extern void put_driver(struct device_driver *drv);
 extern struct device_driver *driver_find(const char *name,
 					 struct bus_type *bus);
-extern int driver_probe_done(void);
+extern void driver_wait_probe_done(void);
+extern void driver_wait_probe_done_and_dev_appear(char *name);
 
 /* sysfs interface for exporting driver attributes */
 
init/do_mounts.c +4 -5
index 708105e..1bf34fc 100644
@@ -371,8 +371,8 @@ void __init prepare_namespace(void)
 	}
 
 	/* wait for the known devices to complete their probing */
-	while (driver_probe_done() != 0)
-		msleep(100);
+	driver_wait_probe_done();
+
 	async_synchronize_full();
 
 	md_run_setup();
@@ -396,9 +396,8 @@ void __init prepare_namespace(void)
 	if ((ROOT_DEV == 0) && root_wait) {
 		printk(KERN_INFO "Waiting for root device %s...\n",
 			saved_root_name);
-		while (driver_probe_done() != 0 ||
-			(ROOT_DEV = name_to_dev_t(saved_root_name)) == 0)
-			msleep(100);
+		driver_wait_probe_done_and_dev_appear(saved_root_name);
+		ROOT_DEV = name_to_dev_t(saved_root_name);
 	}
 
 	is_floppy = MAJOR(ROOT_DEV) == FLOPPY_MAJOR;
init/do_mounts_md.c +3 -2
index ff95e31..bcea03b 100644
@@ -281,8 +281,9 @@ static void __init autodetect_raid(void)
 	 */
 	printk(KERN_INFO "md: Waiting for all devices to be available before autodetect\n");
 	printk(KERN_INFO "md: If you don't use raid, use raid=noautodetect\n");
-	while (driver_probe_done() < 0)
-		msleep(100);
+
+	driver_wait_probe_done();
+
 	fd = sys_open("/dev/md0", 0, 0);
 	if (fd >= 0) {
 		sys_ioctl(fd, RAID_AUTORUN, raid_autopart);
Signature · 1 line
1.6.0
#1 lore raw
Quoted 9 lines · On Fri, 30 Jan 2009 10:31:42 +0800 [email protected] wrote:

From: Ming Lei <[email protected]>

This patch renames driver_probe_done to driver_wait_probe_done, and make it wait on condition of probing done to remove polling for it in fs initialization. Also add driver_wait_probe_done_and_dev_appear(name) to allow callers to wait until all probing is done AND until the device @name exists.

Quoted 4 lines

/**
+ * driver_wait_probe_done_and_dev_appear
+ * wait until the probe sequence is finished _and_ the dev @name appears.

Hi,

sorry to be the bad guy, but I think you turned the "either one" into a "both".... so this isn't going to work ;(

unless I'm just too tired and miss something subtle, but I am pretty sure DeMorgan does not work this way :)

Signature · 3 lines
Arjan van de Ven 	Intel Open Source Technology Centre
For development, discussion and tips for power savings, 
visit http://www.lesswatts.org
author
#2 lore raw

2009/1/30 Arjan van de Ven <[email protected]>:

Quoted 17 lines
Quoted 9 lines · On Fri, 30 Jan 2009 10:31:42 +0800 [email protected] wrote:

From: Ming Lei <[email protected]>

This patch renames driver_probe_done to driver_wait_probe_done, and make it wait on condition of probing done to remove polling for it in fs initialization. Also add driver_wait_probe_done_and_dev_appear(name) to allow callers to wait until all probing is done AND until the device @name exists.

Quoted 4 lines

/**
+ * driver_wait_probe_done_and_dev_appear
+ * wait until the probe sequence is finished _and_ the dev @name appears.

Quoted 4 lines

Hi,

sorry to be the bad guy, but I think you turned the "either one" into a "both".... so this isn't going to work ;(

IMHO, it will msleep if either one is ture, so we turn this into return( not sleep) if both are false.
( eg, A||B->C <=> ~C->~A&&~B ) :-)

           while (driver_probe_done() != 0 ||
                  (ROOT_DEV = name_to_dev_t(saved_root_name)) == 0)
                  msleep(100);
 here:
      driver_probe_done() == 0  &&  name_to_dev_t(saved_root_name)

===>

driver_wait_probe_done_and_dev_appear(name)

{
         if (driver_probe_done() == 0 &&
                  (ROOT_DEV = name_to_dev_t(saved_root_name)))
                  return;
         else
                  schedule();
}

Thanks!

Quoted 10 lines

unless I'm just too tired and miss something subtle, but I am pretty sure DeMorgan does not work this way :)

--
Arjan van de Ven Intel Open Source Technology Centre
For development, discussion and tips for power savings, visit http://www.lesswatts.org

Signature · 1 line
Lei Ming
#3 lore raw
In reply to [email protected]

drivers/base/dd.c done: · line 180

Quoted 12 lines · On Fri, 30 Jan 2009 10:31:42 +0800, [email protected] wrote:

/**
+ * driver_wait_probe_done_and_dev_appear
+ * wait until the probe sequence is finished _and_ the dev @name appears.
+ *
+ */

drivers/base/dd.c +7 -0
+void driver_wait_probe_done_and_dev_appear(char *name)
+{
+	pr_debug("%s: probe_count = %d\n", __func__,
+		 atomic_read(&probe_count));
+
+	wait_event(probe_waitqueue, (atomic_read(&probe_count) == 0) &&
+		  name_to_dev_t(name));

probe_waitqueue is only woken at the end of really_probe(), not when whatever needs to be done to make the device available is done...

Quoted 5 lines
drivers/base/dd.c +2 -0
+}
+/**
  * driver_probe_device - attempt to bind device & driver together
  * @drv: driver to bind a device to
  * @dev: device to try to bind to the driver
author
#4 lore raw

2009/1/30 Cornelia Huck <[email protected]>:

Quoted 4 lines

On Fri, 30 Jan 2009 10:31:42 +0800,
[email protected] wrote:

Quoted 15 lines

drivers/base/dd.c done: · line 180

Quoted 12 lines

/**
+ * driver_wait_probe_done_and_dev_appear
+ * wait until the probe sequence is finished _and_ the dev @name appears.
+ *
+ */

drivers/base/dd.c +7 -0
+void driver_wait_probe_done_and_dev_appear(char *name)
+{
+     pr_debug("%s: probe_count = %d\n", __func__,
+              atomic_read(&probe_count));
+
+     wait_event(probe_waitqueue, (atomic_read(&probe_count) == 0) &&
+               name_to_dev_t(name));

probe_waitqueue is only woken at the end of really_probe(), not when whatever needs to be done to make the device available is done...

Yes.

If probing done does not mean the device is available, this patch should be ignored or fixed further.

But polling is really not good, is there a approach to avoid the polling for appearence of the root device?

Thanks!

Quoted 7 lines
Quoted 6 lines
drivers/base/dd.c +2 -0
+}
+/**
  * driver_probe_device - attempt to bind device & driver together
  * @drv: driver to bind a device to
  * @dev: device to try to bind to the driver
Signature · 1 line
Lei Ming
#5 lore raw
Quoted 5 lines · Ming Lei

If probing done does not mean the device is available, this patch should be ignored or fixed further.

But polling is really not good, is there a approach to avoid the polling for appearence of the root device?

I think the parts of your patch that replace polling for driver_probe_done() with waiting on the waitqueue are fine. The problematic part is the rootwait special case, where polling seems unavoidable afaics.

author
#6 lore raw

2009/1/30 Cornelia Huck <[email protected]>:

Quoted 6 lines
Quoted 3 lines · Ming Lei

If probing done does not mean the device is available, this patch should be ignored or fixed further.

Quoted 7 lines
Quoted 2 lines

But polling is really not good, is there a approach to avoid the polling for appearence of the root device?

I think the parts of your patch that replace polling for driver_probe_done() with waiting on the waitqueue are fine. The problematic part is the rootwait special case, where polling seems unavoidable afaics.

Yes.

Maybe we should keep the polling for root device. And I will resend a patch ,which only replace the driver_probe_done().

Thanks.

Quoted 1 line
Signature · 1 line
Lei Ming