[PATCH] init: no need to wait device probe

[PATCH] init: no need to wait device probe

[ning.a.zhang]

From: Zhang Ning <ning.a.zhang@intel.com>

there are 2 reasons for no need to wait device probe

reason 1:
mount root device is very late in kernel initial stage.
all initcalls are finished. that means most of probe functions
are returned.

and deferred probe are also finished by late_initcall.
only async probe driver are possible in  probing.

no block devices, device-mapper or nfs are use async probe.
so no need to wait device probe.

reason 2:
let's check dd.c, probe_count is increased and decreased only
in really_probe, and when really_probe returns, probe_count
always be 0.

when someone really wants to wait device-B probe.
but code looks like:

thread-1:			thread-2:
probe device-A;                 wait_for_device_probe();
msleep(30);
probe device-B

when device-A probe finishes, thread-2 will be wakeup,
but device-B is not probed.

wait_for_device_probe can't make sure the device you want is probed.

based on above 2 reasons, no need to wait for device probe.

Signed-off-by: Zhang Ning <ning.a.zhang@intel.com>
---
 init/do_mounts.c    | 9 ---------
 init/do_mounts_md.c | 2 --
 2 files changed, 11 deletions(-)

diff --git a/init/do_mounts.c b/init/do_mounts.c
index 7cf4f6dafd5f..a9fb2ad44964 100644
--- a/init/do_mounts.c
+++ b/init/do_mounts.c
@@ -555,15 +555,6 @@ void __init prepare_namespace(void)
 		ssleep(root_delay);
 	}
 
-	/*
-	 * wait for the known devices to complete their probing
-	 *
-	 * Note: this is a potential source of long boot delays.
-	 * For example, it is not atypical to wait 5 seconds here
-	 * for the touchpad of a laptop to initialize.
-	 */
-	wait_for_device_probe();
-
 	md_run_setup();
 
 	if (saved_root_name[0]) {
diff --git a/init/do_mounts_md.c b/init/do_mounts_md.c
index 3f733c760a8c..4aab3492e71d 100644
--- a/init/do_mounts_md.c
+++ b/init/do_mounts_md.c
@@ -292,8 +292,6 @@ 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");
 
-	wait_for_device_probe();
-
 	fd = sys_open("/dev/md0", 0, 0);
 	if (fd >= 0) {
 		sys_ioctl(fd, RAID_AUTORUN, raid_autopart);
-- 
2.14.2

Re: [PATCH] init: no need to wait device probe

[Randy Dunlap]

Hi,

On 03/15/2018 12:20 AM, ning.a.zhang@intel.com wrote:
> From: Zhang Ning <ning.a.zhang@intel.com>

meta comment (i.e., not about the merits of the patch itself):

You'll need to send the patch to someone if you want it to be merged.
Maintainers don't mine mailing lists for patches to apply.

> there are 2 reasons for no need to wait device probe
> 
> reason 1:
> mount root device is very late in kernel initial stage.
> all initcalls are finished. that means most of probe functions
> are returned.
> 
> and deferred probe are also finished by late_initcall.
> only async probe driver are possible in  probing.
> 
> no block devices, device-mapper or nfs are use async probe.
> so no need to wait device probe.
> 
> reason 2:
> let's check dd.c, probe_count is increased and decreased only
> in really_probe, and when really_probe returns, probe_count
> always be 0.
> 
> when someone really wants to wait device-B probe.
> but code looks like:
> 
> thread-1:			thread-2:
> probe device-A;                 wait_for_device_probe();
> msleep(30);
> probe device-B
> 
> when device-A probe finishes, thread-2 will be wakeup,
> but device-B is not probed.
> 
> wait_for_device_probe can't make sure the device you want is probed.
> 
> based on above 2 reasons, no need to wait for device probe.
> 
> Signed-off-by: Zhang Ning <ning.a.zhang@intel.com>
> ---
>  init/do_mounts.c    | 9 ---------
>  init/do_mounts_md.c | 2 --
>  2 files changed, 11 deletions(-)
> 
> diff --git a/init/do_mounts.c b/init/do_mounts.c
> index 7cf4f6dafd5f..a9fb2ad44964 100644
> --- a/init/do_mounts.c
> +++ b/init/do_mounts.c
> @@ -555,15 +555,6 @@ void __init prepare_namespace(void)
>  		ssleep(root_delay);
>  	}
>  
> -	/*
> -	 * wait for the known devices to complete their probing
> -	 *
> -	 * Note: this is a potential source of long boot delays.
> -	 * For example, it is not atypical to wait 5 seconds here
> -	 * for the touchpad of a laptop to initialize.
> -	 */
> -	wait_for_device_probe();
> -
>  	md_run_setup();
>  
>  	if (saved_root_name[0]) {
> diff --git a/init/do_mounts_md.c b/init/do_mounts_md.c
> index 3f733c760a8c..4aab3492e71d 100644
> --- a/init/do_mounts_md.c
> +++ b/init/do_mounts_md.c
> @@ -292,8 +292,6 @@ 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");
>  
> -	wait_for_device_probe();
> -
>  	fd = sys_open("/dev/md0", 0, 0);
>  	if (fd >= 0) {
>  		sys_ioctl(fd, RAID_AUTORUN, raid_autopart);
> 


-- 
~Randy

Re: [PATCH] init: no need to wait device probe

[Zhang, Ning A]

在 2018-03-17六的 12:17 -0700，Randy Dunlap写道：
> Maintainers don't mine mailing lists for patches to apply.

thank you very much, I will resend my patch.

BR.
Ning.

[PATCH] init: no need to wait device probe

[ning.a.zhang]

From: Zhang Ning <ning.a.zhang@intel.com>

there are 2 reasons for no need to wait device probe

reason 1:
mount root device is very late in kernel initial stage.
all initcalls are finished. that means most of probe functions
are returned.

and deferred probe are also finished by late_initcall.
only async probe driver are possible in  probing.

no block devices, device-mapper or nfs are use async probe.
so no need to wait device probe.

reason 2:
let's check dd.c, probe_count is increased and decreased only
in really_probe, and when really_probe returns, probe_count
always be 0.

when someone really wants to wait device-B probe.
but code looks like:

thread-1:			thread-2:
probe device-A;                 wait_for_device_probe();
msleep(30);
probe device-B

when device-A probe finishes, thread-2 will be wakeup,
but device-B is not probed.

wait_for_device_probe can't make sure the device you want is probed.

based on above 2 reasons, no need to wait for device probe.

Signed-off-by: Zhang Ning <ning.a.zhang@intel.com>
---
 init/do_mounts.c    | 9 ---------
 init/do_mounts_md.c | 2 --
 2 files changed, 11 deletions(-)

diff --git a/init/do_mounts.c b/init/do_mounts.c
index 7cf4f6dafd5f..a9fb2ad44964 100644
--- a/init/do_mounts.c
+++ b/init/do_mounts.c
@@ -555,15 +555,6 @@ void __init prepare_namespace(void)
 		ssleep(root_delay);
 	}
 
-	/*
-	 * wait for the known devices to complete their probing
-	 *
-	 * Note: this is a potential source of long boot delays.
-	 * For example, it is not atypical to wait 5 seconds here
-	 * for the touchpad of a laptop to initialize.
-	 */
-	wait_for_device_probe();
-
 	md_run_setup();
 
 	if (saved_root_name[0]) {
diff --git a/init/do_mounts_md.c b/init/do_mounts_md.c
index 3f733c760a8c..4aab3492e71d 100644
--- a/init/do_mounts_md.c
+++ b/init/do_mounts_md.c
@@ -292,8 +292,6 @@ 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");
 
-	wait_for_device_probe();
-
 	fd = sys_open("/dev/md0", 0, 0);
 	if (fd >= 0) {
 		sys_ioctl(fd, RAID_AUTORUN, raid_autopart);
-- 
2.14.2