queue usb hotplug events

PATCH In review Olaf Hering 2 participants 9 messages Jun 7, 2003 – Jun 9, 2003

lore

Share & tools

lore

Olaf Hering

author

2003-06-07 11:43 UTC #0 lore raw

PATCH 1/1 queue usb hotplug events

Hi,

there is a hardcoded sleep 3 in the usb.agent. But this is wrong, because the kernel runs all hotplug events at once for a new hub. The result is that every event still runs in parallel, just 3 seconds later.

This patch implements an userland fifo.

touch an unique file
wait one second to let things settle
build a list with lockfiles, oldest last
grab the pid of the oldest agent
if this pid match the running agent, proceed
if all lockfiles are gone, proceed as well, provides a way to workaround a hanging event
otherwise wait a second and try again

Index: etc/hotplug/usb.agent
=================================--- etc/hotplug/usb.agent (revision 13)
+++ etc/hotplug/usb.agent (working copy)
@@ -367,7 +367,15 @@

     # with slow HID devices and "usbmodules".
     # starting with 2.5, all hcds must queue control traffic.
     if [ "$DEVPATH" = "" ]; then
-	sleep 3
+	touch "${REMOVER}.queue.$$"
+    	until [ `sleep 1` ] ; do
+		lockfiles=`ls -t /var/run/usb/*.queue.* 2>/dev/null`
+		pid=${lockfiles##*.}
+		debug_mesg $pid $lockfiles
+		test "$pid" = "$$" && { debug_mesg pid $$ proceeds ; break; }
+		test "$lockfiles" = "" && { debug_mesg all lockfiles gone ; break; }
+		debug_mesg wait 1 second for pid $pid
+	done
     fi

usb_convert_vars
@@ -426,6 +434,10 @@
/sbin/devlabel restart
fi

+ if [ -e "${REMOVER}.queue.$$" ]; then
+ debug_mesg pid $$ done
+ rm -f "${REMOVER}.queue.$$"
+ fi

;;

remove)

Signature · 11 lines

USB is for mice, FireWire is for men!


-------------------------------------------------------
This SF.net email is sponsored by:  Etnus, makers of TotalView, The best
thread debugger on the planet. Designed with thread debugging features
you've never dreamed of, try TotalView 6 free at www.etnus.com.
_______________________________________________
Linux-hotplug-devel mailing list  http://linux-hotplug.sourceforge.net
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel

David Brownell

2003-06-07 19:22 UTC #1 lore raw

Quoted 5 lines · Olaf Hering

Hi,
there is a hardcoded sleep 3 in the usb.agent. But this is wrong, because the kernel runs all hotplug events at once for a new hub. The result is that every event still runs in parallel, just 3 seconds later.

That "sleep" is a workaround for "uhci" and "usb-uhci" bugs:
they don't queue control transfers. No complete user mode workaround is possible; it doesn't affect just hotplug, or even just usbfs-based applications.

The real fix is just to get rid of it on 2.5, where "uhci-hcd" has been fixed, rather than to keep making all USB hotplugging be annoyingly slow.

That's why I won't merge this patch.

- Dave

-------------------------------------------------------
This SF.net email is sponsored by: Etnus, makers of TotalView, The best thread debugger on the planet. Designed with thread debugging features you've never dreamed of, try TotalView 6 free at www.etnus.com.
_______________________________________________
Linux-hotplug-devel mailing list http://linux-hotplug.sourceforge.net
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel

Olaf Hering

author

2003-06-07 19:25 UTC #2 lore raw

In reply to Olaf Hering

Quoted 11 lines · David Brownell

Quoted 5 lines · Olaf Hering
Hi,
there is a hardcoded sleep 3 in the usb.agent. But this is wrong, because the kernel runs all hotplug events at once for a new hub. The result is that every event still runs in parallel, just 3 seconds later.
That "sleep" is a workaround for "uhci" and "usb-uhci" bugs:
they don't queue control transfers. No complete user mode workaround is possible; it doesn't affect just hotplug, or even just usbfs-based applications.

my patch fixes the timeouts. Have you even tried it?

Signature · 11 lines

USB is for mice, FireWire is for men!


-------------------------------------------------------
This SF.net email is sponsored by:  Etnus, makers of TotalView, The best
thread debugger on the planet. Designed with thread debugging features
you've never dreamed of, try TotalView 6 free at www.etnus.com.
_______________________________________________
Linux-hotplug-devel mailing list  http://linux-hotplug.sourceforge.net
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel

David Brownell

2003-06-07 19:34 UTC #3 lore raw

In reply to Olaf Hering

Quoted 7 lines · Olaf Hering

Quoted 4 lines
That "sleep" is a workaround for "uhci" and "usb-uhci" bugs:
they don't queue control transfers. No complete user mode workaround is possible; it doesn't affect just hotplug, or even just usbfs-based applications.
my patch fixes the timeouts. Have you even tried it?

Actually it's already fixed on 2.5, the check for $DEVPATH will fail on earlier kernels.

Olaf Hering

author

2003-06-07 19:34 UTC #4 lore raw

In reply to Olaf Hering

Quoted 8 lines · David Brownell

Quoted 6 lines · Olaf Hering
Quoted 4 lines
That "sleep" is a workaround for "uhci" and "usb-uhci" bugs:
they don't queue control transfers. No complete user mode workaround is possible; it doesn't affect just hotplug, or even just usbfs-based applications.

Quoted 4 lines

Quoted 1 line
my patch fixes the timeouts. Have you even tried it?
Actually it's already fixed on 2.5, the check for $DEVPATH will fail on earlier kernels.

Ok, DEVPATH does not exist in 2.4 kernels.
sleep 3 is executed everytime

Now with my patch, only sleep 1 is executed. Where is that slower?

Gruss Olaf

Signature · 11 lines

USB is for mice, FireWire is for men!


-------------------------------------------------------
This SF.net email is sponsored by:  Etnus, makers of TotalView, The best
thread debugger on the planet. Designed with thread debugging features
you've never dreamed of, try TotalView 6 free at www.etnus.com.
_______________________________________________
Linux-hotplug-devel mailing list  http://linux-hotplug.sourceforge.net
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel

Olaf Hering

author

2003-06-08 07:23 UTC #5 lore raw

In reply to Olaf Hering

Quoted 8 lines · David Brownell

Quoted 5 lines · Olaf Hering
Hi,
there is a hardcoded sleep 3 in the usb.agent. But this is wrong, because the kernel runs all hotplug events at once for a new hub. The result is that every event still runs in parallel, just 3 seconds later.
That "sleep" is a workaround for "uhci" and "usb-uhci" bugs:

you can change that to sleep 1 or sleep 42 or sleep $largenum, they will still run in parallel. This one solves it.

        touch "${REMOVER}.queue.$$"
        sleep 3 & MAX_DELAY=$!
        while [ -d /proc/$MAX_DELAY ] ; do
                lockfiles=`ls -t /var/run/usb/*.queue.* 2>/dev/null`
                pid=${lockfiles##*.}
                test "$pid" = "$$" && { debug_mesg pid $$ proceeds ; break; }
                test "$lockfiles" = "" && { debug_mesg all lockfiles gone ; break; }
                debug_mesg wait 1 second for pid $pid
                sleep 0.5
        done

Now you have zero delay for the first event.

Signature · 11 lines

USB is for mice, FireWire is for men!


-------------------------------------------------------
This SF.net email is sponsored by:  Etnus, makers of TotalView, The best
thread debugger on the planet. Designed with thread debugging features
you've never dreamed of, try TotalView 6 free at www.etnus.com.
_______________________________________________
Linux-hotplug-devel mailing list  http://linux-hotplug.sourceforge.net
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel

David Brownell

2003-06-08 19:10 UTC #6 lore raw

In reply to Olaf Hering

Quoted 9 lines · Olaf Hering

Quoted 5 lines
Quoted 3 lines
there is a hardcoded sleep 3 in the usb.agent. But this is wrong, because the kernel runs all hotplug events at once for a new hub. The result is that every event still runs in parallel, just 3 seconds later.
That "sleep" is a workaround for "uhci" and "usb-uhci" bugs:
you can change that to sleep 1 or sleep 42 or sleep $largenum, they will still run in parallel. This one solves it.

Running parallel isn't a bug, doesn't need to be changed. Typically the agents just finish faster that way.

Olaf Hering

author

2003-06-09 11:34 UTC #7 lore raw

In reply to Olaf Hering

Quoted 1 line · Olaf Hering

        touch "${REMOVER}.queue.$$"

It should be '> "${REMOVER}.queue.$$"' to avoid yet another process.

Gruss Olaf

Signature · 11 lines

USB is for mice, FireWire is for men!


-------------------------------------------------------
This SF.net email is sponsored by:  Etnus, makers of TotalView, The best
thread debugger on the planet. Designed with thread debugging features
you've never dreamed of, try TotalView 6 free at www.etnus.com.
_______________________________________________
Linux-hotplug-devel mailing list  http://linux-hotplug.sourceforge.net
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel

Olaf Hering

author

2003-06-09 15:25 UTC #8 lore raw

In reply to Olaf Hering

Quoted 5 lines · David Brownell

Quoted 4 lines · Olaf Hering
Quoted 4 lines
Quoted 3 lines
there is a hardcoded sleep 3 in the usb.agent. But this is wrong, because the kernel runs all hotplug events at once for a new hub. The result is that every event still runs in parallel, just 3 seconds later.

Quoted 8 lines

Quoted 5 lines
Quoted 1 line
That "sleep" is a workaround for "uhci" and "usb-uhci" bugs:
you can change that to sleep 1 or sleep 42 or sleep $largenum, they will still run in parallel. This one solves it.
Running parallel isn't a bug, doesn't need to be changed. Typically the agents just finish faster that way.

You could drop the sleep 3 at all if you argue that way.

Signature · 11 lines

USB is for mice, FireWire is for men!


-------------------------------------------------------
This SF.net email is sponsored by:  Etnus, makers of TotalView, The best
thread debugger on the planet. Designed with thread debugging features
you've never dreamed of, try TotalView 6 free at www.etnus.com.
_______________________________________________
Linux-hotplug-devel mailing list  http://linux-hotplug.sourceforge.net
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel