Skip to contentSkip to conversation

USB input ati_remote autorepeat problem

DISCUSSION Marko Macek 4 participants 6 messages Jun 27, 2006 – Jul 2, 2006

lore
Share & tools
lore
#0 lore raw

Hello!

I have problems with autorepeat in ati_remote (drivers/usb/input) driver in "recent" kernels: all keys start repeating immediately without some delay.

This makes some things, like changing the channel prev/next or toggling fullscreen, etc... impossible/hard.

The problem seems to be related to FILTER_TIME and HZ=250 (which I forgot to change).

FILTER_TIME is defined to HZ / 20, and since 250 is not divisible by 20, the time will be too short to ignore enough events.

Defining FILTER_TIME to HZ / 20 + 1 seems to fix things, but I'm not sure if there are any bad side effects.

Mark

#1 lore raw

I have also experience this problem with the ati_remote driver, but didn't know the fix. If I am correct, the HZ always used to default to 1000HZ, but not it is configurable in the kernel as 1000HZ, 250HZ, or 100HZ. By my calculations, the FILTER_TIME will be at most 20% longer than before, but as little as 2% bigger.

       HZ       Old        New        % Difference
      100       5          6          20%
      250       12.5       13.5       8%
     1000       50         51         2%

Those are my calculations based on your previously stated definitions. The FILTER_TIME const is only used 1 time in ati_remote.c, so I doubt there would be a problem redefining it. Would a redefinition to 50 be more appropriate (so keep the repeat delay the same across all platforms)?

Tonight I will recompile and test the driver with the 2 new definitions and report on my findings.

John

Quoted 18 lines · Marko Macek

Hello!

I have problems with autorepeat in ati_remote (drivers/usb/input) driver in "recent" kernels: all keys start repeating immediately without some delay.

This makes some things, like changing the channel prev/next or toggling fullscreen, etc... impossible/hard.

The problem seems to be related to FILTER_TIME and HZ=250 (which I forgot to change).

FILTER_TIME is defined to HZ / 20, and since 250 is not divisible by 20, the time will be too short to ignore enough events.

Defining FILTER_TIME to HZ / 20 + 1 seems to fix things, but I'm not sure if there are any bad side effects.

Mark
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

#2 lore raw
In reply to Marko Macek
Quoted 9 lines · Marko Macek

Hello!

I have problems with autorepeat in ati_remote (drivers/usb/input) driver in "recent" kernels: all keys start repeating immediately without some delay.

This makes some things, like changing the channel prev/next or toggling fullscreen, etc... impossible/hard.

Quoted 8 lines

The problem seems to be related to FILTER_TIME and HZ=250 (which I forgot to change).

FILTER_TIME is defined to HZ / 20, and since 250 is not divisible by 20, the time will be too short to ignore enough events.

Defining FILTER_TIME to HZ / 20 + 1 seems to fix things, but I'm not sure if there are any bad side effects.

Can you try just defining it to msecs_to_jiffies(50)? That should handle the various HZ cases just fine.

Thanks,
Nish

#3 lore raw
Quoted 13 lines · Nish Aravamudan
Quoted 12 lines · Marko Macek

Hello!

I have problems with autorepeat in ati_remote (drivers/usb/input) driver in "recent" kernels: all keys start repeating immediately without some delay.

This makes some things, like changing the channel prev/next or toggling fullscreen, etc... impossible/hard.

The problem seems to be related to FILTER_TIME and HZ=250 (which I forgot to change).

Quoted 8 lines
Quoted 5 lines

FILTER_TIME is defined to HZ / 20, and since 250 is not divisible by 20, the time will be too short to ignore enough events.

Defining FILTER_TIME to HZ / 20 + 1 seems to fix things, but I'm not sure if there are any bad side effects.

Can you try just defining it to msecs_to_jiffies(50)? That should handle the various HZ cases just fine.

Indeed, that would be thr right solution. Even better would be to

	#define	FILTER_TIME	50	/* 50 msec */

and later use

msecs_to_jiffies(FILTER_TIME)

in the code.

Signature · 2 lines
Vojtech Pavlik
Director SuSE Labs
#4 lore raw
Quoted 28 lines · Vojtech Pavlik
Quoted 14 lines · Nish Aravamudan
Quoted 12 lines · Marko Macek

Hello!

I have problems with autorepeat in ati_remote (drivers/usb/input) driver in "recent" kernels: all keys start repeating immediately without some delay.

This makes some things, like changing the channel prev/next or toggling fullscreen, etc... impossible/hard.

The problem seems to be related to FILTER_TIME and HZ=250 (which I forgot to change).

Quoted 9 lines
Quoted 6 lines

FILTER_TIME is defined to HZ / 20, and since 250 is not divisible by 20, the time will be too short to ignore enough events.

Defining FILTER_TIME to HZ / 20 + 1 seems to fix things, but I'm not sure if there are any bad side effects.

Can you try just defining it to msecs_to_jiffies(50)? That should handle the various HZ cases just fine.

Indeed, that would be thr right solution. Even better would be to

Quoted 7 lines
	#define	FILTER_TIME	50	/* 50 msec */

and later use

msecs_to_jiffies(FILTER_TIME)

in the code.

There is still a problem (reproducible in HZ=100, at least), because msec_to_jiffies

#if HZ <= MSEC_PER_SEC && !(MSEC_PER_SEC % HZ)
       return (m + (MSEC_PER_SEC / HZ) - 1) / (MSEC_PER_SEC / HZ);

Calculates 5 ticks for 50ms, which might seem to be correct, but it really isn't, since 5 ticks can happen in as little as 40 (+eps) ms.

I wonder if this usage of msec_to_jiffies is correct (seems wrong to me).

A working (but not clean) patch might look like this:

--- linux-2.6.17.orig/drivers/usb/input/ati_remote.c 2006-06-29 21:18:15.000000000 +0200

+++ linux-2.6.17/drivers/usb/input/ati_remote.c 2006-07-02 20:10:17.000000000 +0200

@@ -155,9 +155,8 @@

* events. The hardware generates 5 events for the first keypress

* and we have to take this into account for an accurate repeat

* behaviour.

- * (HZ / 20) == 50 ms and works well for me.

*/

-#define FILTER_TIME (HZ / 20)

+#define FILTER_TIME 51 /* msec */

struct ati_remote {

    struct input_dev *idev;

@@ -470,7 +469,7 @@

        /* Filter duplicate events which happen "too close" together. */

        if ((ati_remote->old_data[0] == data[1]) &&

            (ati_remote->old_data[1] == data[2]) &&

- time_before(jiffies, ati_remote->old_jiffies + FILTER_TIME)) {

+ time_before(jiffies, ati_remote->old_jiffies + msecs_to_jiffies(FILTER_TIME))) {

            ati_remote->repeat_count++;

        } else {

            ati_remote->repeat_count = 0;

Some googling reveals that an old patch used HZ >> 4 (HZ / 16) instead of HZ / 20;

Perhaps using msec_to_jiffies(50) + 1 would be the correct fix?

Mark

#5 lore raw
Quoted 42 lines · Marko Macek
Quoted 28 lines · Vojtech Pavlik
Quoted 14 lines · Nish Aravamudan
Quoted 12 lines · Marko Macek

Hello!

I have problems with autorepeat in ati_remote (drivers/usb/input) driver in "recent" kernels: all keys start repeating immediately without some delay.

This makes some things, like changing the channel prev/next or toggling fullscreen, etc... impossible/hard.

The problem seems to be related to FILTER_TIME and HZ=250 (which I forgot to change).

Quoted 9 lines
Quoted 6 lines

FILTER_TIME is defined to HZ / 20, and since 250 is not divisible by 20, the time will be too short to ignore enough events.

Defining FILTER_TIME to HZ / 20 + 1 seems to fix things, but I'm not sure if there are any bad side effects.

Can you try just defining it to msecs_to_jiffies(50)? That should handle the various HZ cases just fine.

Indeed, that would be thr right solution. Even better would be to

Quoted 7 lines
	#define	FILTER_TIME	50	/* 50 msec */

and later use

msecs_to_jiffies(FILTER_TIME)

in the code.

There is still a problem (reproducible in HZ=100, at least), because msec_to_jiffies

#if HZ <= MSEC_PER_SEC && !(MSEC_PER_SEC % HZ)
       return (m + (MSEC_PER_SEC / HZ) - 1) / (MSEC_PER_SEC / HZ);
Quoted 6 lines

Calculates 5 ticks for 50ms, which might seem to be correct, but it really isn't, since 5 ticks can happen in as little as 40 (+eps) ms.

I wonder if this usage of msec_to_jiffies is correct (seems wrong to me).

A working (but not clean) patch might look like this:

The patch looks OK to me. You could as well use 60 msec to be on the safe side, and closer to the "/ 16" version.

Quoted 57 lines

--- linux-2.6.17.orig/drivers/usb/input/ati_remote.c 2006-06-29
21:18:15.000000000 +0200

+++ linux-2.6.17/drivers/usb/input/ati_remote.c 2006-07-02
20:10:17.000000000 +0200

(quoted diff) +0 -1
@@ -155,9 +155,8 @@
 
 * events. The hardware generates 5 events for the first keypress
 
 * and we have to take this into account for an accurate repeat
 
 * behaviour.
 
- * (HZ / 20) == 50 ms and works well for me.
 

*/

(quoted diff) +1 -1
-#define FILTER_TIME (HZ / 20)
 
+#define FILTER_TIME 51 /* msec */

struct ati_remote {

    struct input_dev *idev;
(quoted diff) +1 -1
@@ -470,7 +469,7 @@
 
        /* Filter duplicate events which happen "too close" together. */
 
        if ((ati_remote->old_data[0] == data[1]) &&
 
            (ati_remote->old_data[1] == data[2]) &&
 
-            time_before(jiffies, ati_remote->old_jiffies + FILTER_TIME)) {
 
+            time_before(jiffies, ati_remote->old_jiffies + 

msecs_to_jiffies(FILTER_TIME))) {

            ati_remote->repeat_count++;

        } else {

            ati_remote->repeat_count = 0;

Some googling reveals that an old patch used HZ >> 4 (HZ / 16) instead of HZ / 20;

Perhaps using msec_to_jiffies(50) + 1 would be the correct fix?

Mark

Signature · 2 lines
Vojtech Pavlik
Director SuSE Labs