[PATCH 1/3] input: mt: Simplify event bypass logic

[PATCH 1/3] input: mt: Simplify event bypass logic

[Henrik Rydberg]

The input bypass mechanism is only used by the MT events. Since all
MT events use bypassing by construction, the current list is a mere
duplication of input.h. This patch replaces the bypass list by a
simple range test.

Signed-off-by: Henrik Rydberg <rydberg@euromail.se>
---
 drivers/input/input.c |   31 +------------------------------
 include/linux/input.h |    3 +++
 2 files changed, 4 insertions(+), 30 deletions(-)

diff --git a/drivers/input/input.c b/drivers/input/input.c
index 9c79bd5..187dd8b 100644
--- a/drivers/input/input.c
+++ b/drivers/input/input.c
@@ -33,25 +33,6 @@ MODULE_LICENSE("GPL");
 
 #define INPUT_DEVICES	256
 
-/*
- * EV_ABS events which should not be cached are listed here.
- */
-static unsigned int input_abs_bypass_init_data[] __initdata = {
-	ABS_MT_TOUCH_MAJOR,
-	ABS_MT_TOUCH_MINOR,
-	ABS_MT_WIDTH_MAJOR,
-	ABS_MT_WIDTH_MINOR,
-	ABS_MT_ORIENTATION,
-	ABS_MT_POSITION_X,
-	ABS_MT_POSITION_Y,
-	ABS_MT_TOOL_TYPE,
-	ABS_MT_BLOB_ID,
-	ABS_MT_TRACKING_ID,
-	ABS_MT_PRESSURE,
-	0
-};
-static unsigned long input_abs_bypass[BITS_TO_LONGS(ABS_CNT)];
-
 static LIST_HEAD(input_dev_list);
 static LIST_HEAD(input_handler_list);
 
@@ -235,7 +216,7 @@ static void input_handle_event(struct input_dev *dev,
 	case EV_ABS:
 		if (is_event_supported(code, dev->absbit, ABS_MAX)) {
 
-			if (test_bit(code, input_abs_bypass)) {
+			if (code >= ABS_MT_MIN && code <= ABS_MT_MAX) {
 				disposition = INPUT_PASS_TO_HANDLERS;
 				break;
 			}
@@ -1926,20 +1907,10 @@ static const struct file_operations input_fops = {
 	.open = input_open_file,
 };
 
-static void __init input_init_abs_bypass(void)
-{
-	const unsigned int *p;
-
-	for (p = input_abs_bypass_init_data; *p; p++)
-		input_abs_bypass[BIT_WORD(*p)] |= BIT_MASK(*p);
-}
-
 static int __init input_init(void)
 {
 	int err;
 
-	input_init_abs_bypass();
-
 	err = class_register(&input_class);
 	if (err) {
 		printk(KERN_ERR "input: unable to register input_dev class\n");
diff --git a/include/linux/input.h b/include/linux/input.h
index 7ed2251..d4ad53f 100644
--- a/include/linux/input.h
+++ b/include/linux/input.h
@@ -706,6 +706,9 @@ struct input_absinfo {
 #define ABS_MT_TRACKING_ID	0x39	/* Unique ID of initiated contact */
 #define ABS_MT_PRESSURE		0x3a	/* Pressure on contact area */
 
+#define ABS_MT_MIN		0x30
+#define ABS_MT_MAX		0x3a
+
 #define ABS_MAX			0x3f
 #define ABS_CNT			(ABS_MAX+1)
 
-- 
1.6.3.3

Re: [PATCH 1/3] input: mt: Simplify event bypass logic

[Dmitry Torokhov]

On Sat, May 01, 2010 at 03:49:39PM +0200, Henrik Rydberg wrote:
> The input bypass mechanism is only used by the MT events. Since all
> MT events use bypassing by construction, the current list is a mere
> duplication of input.h. This patch replaces the bypass list by a
> simple range test.
> 

Henrik,

I disagree with the patch. I can not guarantee that we will be able to
maintain continuous range of ABS_MT_* events and thus current scheme is
more flexible.

Thanks.

-- 
Dmitry

Re: [PATCH 1/3] input: mt: Simplify event bypass logic

[Henrik Rydberg]

Dmitry Torokhov wrote:
> On Sat, May 01, 2010 at 03:49:39PM +0200, Henrik Rydberg wrote:
>> The input bypass mechanism is only used by the MT events. Since all
>> MT events use bypassing by construction, the current list is a mere
>> duplication of input.h. This patch replaces the bypass list by a
>> simple range test.
>>
> 
> Henrik,
> 
> I disagree with the patch. I can not guarantee that we will be able to
> maintain continuous range of ABS_MT_* events and thus current scheme is
> more flexible.
> 
> Thanks.
> 

Yes, I was hesitant myself. I would rather keep all [ABS_MAX + 1 - ABS_MT_MIN]
in the slot. Introduces a memory slack but is more flexible.

Henrik

Re: [PATCH 1/3] input: mt: Simplify event bypass logic

[Henrik Rydberg]

Dmitry Torokhov wrote:
> On Sat, May 01, 2010 at 03:49:39PM +0200, Henrik Rydberg wrote:
>> The input bypass mechanism is only used by the MT events. Since all
>> MT events use bypassing by construction, the current list is a mere
>> duplication of input.h. This patch replaces the bypass list by a
>> simple range test.
>>
> 
> Henrik,
> 
> I disagree with the patch. I can not guarantee that we will be able to
> maintain continuous range of ABS_MT_* events and thus current scheme is
> more flexible.

Coming back to this, the current scheme has problems, too.

With the introduction of slots, all events, not only EV_ABS events, are
implicitly partitioned into global events and per-slot events. In the current
approach, the partitioning is controlled via the hidden list in input.c, and
hinted using the prefix MT. However, expecting the usage of slots to grow
outside of a linear range of EV_ABS events, the user will inevitably need to
known the semantics of particular events in order to parse correctly. And once
the user starts to deal with that information, it can just as well be dynamic,
and controlled by the device driver.

My proposition is to add a bit field in input_dev which controls whether a
particular event is used in the slot context or the global context. This
information should be user-readable via EVIO. The driver should be able to alter
these settings, and the default should match the current behavior.

Thoughts?

Cheers,
Henrik