Patch for initial CapsLock

Dear Alan,

Here's a USB problem which can only be resolved in generic code. Vojtech is the owner of the code, but did not reply. Do you think you can keep this patch in -ac?

Thanks in advance,
-- Pete

----- Forwarded message from Pete Zaitcev <[email protected]> -----

Date: Tue, 24 Dec 2002 13:41:03 -0500
From: Pete Zaitcev <[email protected]>
To: [email protected]
Cc: [email protected], [email protected]
Subject: Patch for initial CapsLock
User-Agent: Mutt/1.2.5.1i

Hi, Vojtech:

In 2.4.21-pre2, if Caps Lock is on and a USB keyboard is connected, the LED will not be lit. It mostly affects KVM users like those of IBM BladeCenter, because they "connect" keyboards all day long. Would you be so kind to review the attached patch?

-- Pete

diff -ur -X dontdiff linux-2.4.21-pre2/drivers/char/keyboard.c linux-2.4.21-pre2-usb/drivers/char/keyboard.c

drivers/char/keyboard.c +16 -3

@@ -64,6 +64,7 @@
 void (*kbd_ledfunc)(unsigned int led);
 EXPORT_SYMBOL(handle_scancode);
 EXPORT_SYMBOL(kbd_ledfunc);
+EXPORT_SYMBOL(kbd_refresh_leds);
 
 extern void ctrl_alt_del(void);
 

@@ -899,9 +900,9 @@
  * Aside from timing (which isn't really that important for
  * keyboard interrupts as they happen often), using the software
  * interrupt routines for this thing allows us to easily mask
- * this when we don't want any of the above to happen. Not yet
- * used, but this allows for easy and efficient race-condition
- * prevention later on.
+ * this when we don't want any of the above to happen.
+ * This allows for easy and efficient race-condition prevention
+ * for kbd_ledfunc => input_event(dev, EV_LED, ...) => ...
  */
 static void kbd_bh(unsigned long dummy)
 {

@@ -917,6 +918,18 @@
 EXPORT_SYMBOL(keyboard_tasklet);
 DECLARE_TASKLET_DISABLED(keyboard_tasklet, kbd_bh, 0);
 
+/*
+ * This allows a newly plugged keyboard to pick the LED state.
+ * We do it in this seemindly backwards fashion to ensure proper locking.
+ * Built-in keyboard does refresh on its own.
+ */
+void kbd_refresh_leds(void)
+{
+	tasklet_disable(&keyboard_tasklet);
+	if (ledstate != 0xff && kbd_ledfunc != NULL) kbd_ledfunc(ledstate);
+	tasklet_enable(&keyboard_tasklet);
+}
+
 typedef void (pm_kbd_func) (void);
 
 pm_callback pm_kbd_request_override = NULL;

diff -ur -X dontdiff linux-2.4.21-pre2/drivers/input/keybdev.c linux-2.4.21-pre2-usb/drivers/input/keybdev.c

drivers/input/keybdev.c +2 -0

@@ -201,6 +201,7 @@
 	input_open_device(handle);
 
 //	printk(KERN_INFO "keybdev.c: Adding keyboard: input%d\n", dev->number);
+	kbd_refresh_leds();
 
 	return handle;
 }

@@ -222,6 +223,7 @@
 {
 	input_register_handler(&keybdev_handler);
 	kbd_ledfunc = keybdev_ledfunc;
+	kbd_refresh_leds();
 
 	if (jp_kbd_109) {
 		x86_keycodes[0xb5] = 0x73;	/* backslash, underscore */

diff -ur -X dontdiff linux-2.4.21-pre2/include/linux/kbd_kern.h linux-2.4.21-pre2-usb/include/linux/kbd_kern.h

include/linux/kbd_kern.h +1 -1

@@ -72,6 +72,7 @@
 extern int do_poke_blanked_console;
 
 extern void (*kbd_ledfunc)(unsigned int led);
+extern void kbd_refresh_leds(void);
 
 extern void set_console(int nr);
 extern void schedule_console_callback(void);
 
----- End forwarded message -----

Date: Mon, 6 Jan 2003 22:52:18 +0100
From: Vojtech Pavlik <[email protected]>

Sorry for the delay - yes, it's OK.

Quoted 4 lines

Date: Mon, 6 Jan 2003 22:52:18 +0100
From: Vojtech Pavlik <[email protected]>

Sorry for the delay - yes, it's OK.

Do you want me to do the same for 2.5? It should be a little simpler there, and no new exports.

Quoted 2 lines

Do you want me to do the same for 2.5? It should be a little simpler there, and no new exports.

Yes, please.

Quoted 4 lines

Quoted 2 lines

Do you want me to do the same for 2.5? It should be a little simpler there, and no new exports.

Yes, please.

Done, works for me.

I toyed with the idea to trigger keyboard tasklet instead of locking it out and calling methods, but I did not see how to force updates for unchanged ledstate. Changing ledstate to 0xff just looked so very wrong...

-- Pete

drivers/char/keyboard.c +20 -3

@@ -894,9 +894,9 @@
  * Aside from timing (which isn't really that important for
  * keyboard interrupts as they happen often), using the software
  * interrupt routines for this thing allows us to easily mask
- * this when we don't want any of the above to happen. Not yet
- * used, but this allows for easy and efficient race-condition
- * prevention later on.
+ * this when we don't want any of the above to happen.
+ * This allows for easy and efficient race-condition prevention
+ * for kbd_refresh_leds => input_event(dev, EV_LED, ...) => ...
  */
 
 static void kbd_bh(unsigned long dummy)

@@ -918,6 +918,22 @@
 
 DECLARE_TASKLET_DISABLED(keyboard_tasklet, kbd_bh, 0);
 
+/*
+ * This allows a newly plugged keyboard to pick the LED state.
+ */
+void kbd_refresh_leds(struct input_handle *handle)
+{
+	unsigned char leds = ledstate;
+
+	tasklet_disable(&keyboard_tasklet);
+	if (leds != 0xff) {
+		input_event(handle->dev, EV_LED, LED_SCROLLL, !!(leds & 0x01));
+		input_event(handle->dev, EV_LED, LED_NUML,    !!(leds & 0x02));
+		input_event(handle->dev, EV_LED, LED_CAPSL,   !!(leds & 0x04));
+	}
+	tasklet_enable(&keyboard_tasklet);
+}
+
 #if defined(CONFIG_X86) || defined(CONFIG_IA64) || defined(CONFIG_ALPHA) || defined(CONFIG_MIPS) || defined(CONFIG_PPC) || defined(CONFIG_SPARC32) || defined(CONFIG_SPARC64)
 
 static unsigned short x86_keycodes[256] =

@@ -1159,6 +1175,7 @@
 	handle->name = kbd_name;
 
 	input_open_device(handle);
+	kbd_refresh_leds(handle);
 
 	return handle;
 }