[PATCH] fix SMP support in cdu535 cdrom driver

[PATCH] fix SMP support in cdu535 cdrom driver

[Felipe W Damasio]

[-- Attachment #1: Type: text/plain, Size: 821 bytes --]

	Hi Jens,

	Patch against 2.6.0-test6.

	Even though cdu535 is polled, not interrupt driven, Matthew pointed 
out[1] that is a good idea to fix SMP support on that bit, so that 
people who wants to fix cdu31a don't get that wrong.

	Move prepare_to_wait before enable_interrupts, so that if we enter 
the interrupt handler (after enable_interrupts) on a different CPU 
that's executing sony_sleep's poll version, "wait" is already on the 
irq_wait list, which I think fixes the race that Matthew pointed out, 
doesn't it?

	CC'ing kernel-janitors since it also replaces yield for 
schedule_timeout[2], as Matthew suggested.

	Please consider applying.

	Thanks,

Felipe

[1] http://marc.theaimsgroup.com/?l=linux-kernel&m=106322465303649&w=2

[2] 
http://lists.osdl.org/pipermail/kernel-janitors/2003-October/000173.html

[-- Attachment #2: sonycd535-yield.patch --]
[-- Type: text/plain, Size: 639 bytes --]

--- linux-2.6.0-test6/drivers/cdrom/sonycd535.c.orig	2003-10-08 14:42:43.000000000 -0300
+++ linux-2.6.0-test6/drivers/cdrom/sonycd535.c	2003-10-08 16:00:09.000000000 -0300
@@ -342,13 +342,14 @@
 sony_sleep(void)
 {
 	if (sony535_irq_used <= 0) {	/* poll */
-		yield();
+		set_current_state(TASK_INTERRUPTIBLE);
+		schedule_timeout(1);
 	} else {	/* Interrupt driven */
 		DEFINE_WAIT(wait);
-		
+
 		spin_lock_irq(&sonycd535_lock);
-		enable_interrupts();
 		prepare_to_wait(&cdu535_irq_wait, &wait, TASK_INTERRUPTIBLE);
+		enable_interrupts();
 		spin_unlock_irq(&sonycd535_lock);
 		schedule();
 		finish_wait(&cdu535_irq_wait, &wait);

[PATCH] fix SMP support in cdu535 cdrom driver

[Felipe W Damasio]

[-- Attachment #1: Type: text/plain, Size: 818 bytes --]

	Hi Jens,

	Patch against 2.6.0-test6.

	Even though cdu535 is polled, not interrupt driven, Matthew pointed 
out[1] that is a good idea to fix SMP support on that bit, so that 
people who wants to fix cdu31a don't get that wrong.

	Move prepare_to_wait before enable_interrupts, so that if we enter 
the interrupt handler (after enable_interrupts) on a different CPU 
that's executing sony_sleep's poll version, "wait" is already on the 
irq_wait list, which I think fixes the race that Matthew pointed out, 
doesn't it?

	CC'ing kernel-janitors since it also replaces yield for
schedule_timeout[2], as Matthew suggested.

	Please consider applying.

	Thanks,

Felipe

[1]http://marc.theaimsgroup.com/?l=linux-kernel&m=106322465303649&w=2

[2]http://lists.osdl.org/pipermail/kernel-janitors/2003-October/000173.html


[-- Attachment #2: sonycd535-yield.patch --]
[-- Type: text/plain, Size: 640 bytes --]

--- linux-2.6.0-test6/drivers/cdrom/sonycd535.c.orig	2003-10-08 14:42:43.000000000 -0300
+++ linux-2.6.0-test6/drivers/cdrom/sonycd535.c	2003-10-08 16:00:09.000000000 -0300
@@ -342,13 +342,14 @@
 sony_sleep(void)
 {
 	if (sony535_irq_used <= 0) {	/* poll */
-		yield();
+		set_current_state(TASK_INTERRUPTIBLE);
+		schedule_timeout(1);
 	} else {	/* Interrupt driven */
 		DEFINE_WAIT(wait);
-		
+
 		spin_lock_irq(&sonycd535_lock);
-		enable_interrupts();
 		prepare_to_wait(&cdu535_irq_wait, &wait, TASK_INTERRUPTIBLE);
+		enable_interrupts();
 		spin_unlock_irq(&sonycd535_lock);
 		schedule();
 		finish_wait(&cdu535_irq_wait, &wait);