Skip to contentSkip to conversation

CDROM: replace jiffies busyloop with msleep

PATCH In review Thomas Gleixner 6 participants 9 messages May 22, 2007 – May 23, 2007

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 CDROM: replace jiffies busyloop with msleep

From: Ingo Molnar <[email protected]>

The SJCD driver uses a jiffies busy loop. Replace it with msleep.

---
 drivers/cdrom/sjcd.c |    6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
drivers/cdrom/sjcd.c +3 -3
Index: linux-2.6/drivers/cdrom/sjcd.c
===================================================================
@@ -69,6 +69,7 @@
 #include <linux/string.h>
 #include <linux/major.h>
 #include <linux/init.h>
+#include <linux/delay.h>
 
 #include <asm/system.h>
 #include <asm/io.h>
@@ -1709,12 +1710,11 @@ static int __init sjcd_init(void)
 	printk(KERN_INFO "SJCD: Resetting: ");
 	sjcd_send_cmd(SCMD_RESET);
 	for (i = 1000; i > 0 && !sjcd_status_valid; --i) {
-		unsigned long timer;
-
 		/*
 		 * Wait 10ms approx.
 		 */
-		for (timer = jiffies; time_before_eq(jiffies, timer););
+		msleep(10);
+
 		if ((i % 100) == 0)
 			printk(".");
 		(void) sjcd_check_status();
#1 lore raw
Quoted 25 lines · Thomas Gleixner

From: Ingo Molnar <[email protected]>

The SJCD driver uses a jiffies busy loop. Replace it with msleep.

---
 drivers/cdrom/sjcd.c |    6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

Index: linux-2.6/drivers/cdrom/sjcd.c
===================================================================

drivers/cdrom/sjcd.c +1 -0
@@ -69,6 +69,7 @@
 #include <linux/string.h>
 #include <linux/major.h>
 #include <linux/init.h>
+#include <linux/delay.h>
 
 #include <asm/system.h>
 #include <asm/io.h>
@@ -1709,12 +1710,11 @@ static int __init sjcd_init(void)
 	printk(KERN_INFO "SJCD: Resetting: ");

drivers/cdrom/sjcd.c static int __init sjcd_init(void) · line 1717

Quoted 10 lines
drivers/cdrom/sjcd.c +2 -3
 	sjcd_send_cmd(SCMD_RESET);
 	for (i = 1000; i > 0 && !sjcd_status_valid; --i) {
-		unsigned long timer;
-
 		/*
 		 * Wait 10ms approx.
 		 */
-		for (timer = jiffies; time_before_eq(jiffies, timer););
+		msleep(10);
+

I always worry when I see code like this, wondering if there is some arcane reason that I cannot fathom, that is being removed. You gotta wonder how long it has been since this driver was used, by anybody.

Oh well, I cannot find fault with the patch, paranoia worries aside.

ACK.

Jeff

#2 lore raw
Quoted 17 lines · Jeff Garzik

drivers/cdrom/sjcd.c static int __init sjcd_init(void) · line 1717

Quoted 12 lines
drivers/cdrom/sjcd.c +2 -3
@@ -1709,12 +1710,11 @@ static int __init sjcd_init(void)
 	printk(KERN_INFO "SJCD: Resetting: ");
 	sjcd_send_cmd(SCMD_RESET);
 	for (i = 1000; i > 0 && !sjcd_status_valid; --i) {
-		unsigned long timer;
-
 		/*
 		 * Wait 10ms approx.
 		 */
-		for (timer = jiffies; time_before_eq(jiffies, timer););
+		msleep(10);
+

I always worry when I see code like this, wondering if there is some arcane reason that I cannot fathom, that is being removed. You gotta wonder how long it has been since this driver was used, by anybody.

Quoted 1 line

Oh well, I cannot find fault with the patch, paranoia worries aside.

well, while i dont have that hardware, i found this by booting an allyesconfig bzImage which runs the code above, which locks up without this change. And then it booted fine with this change. So arcane issues aside, it does visibly improve things ;-)

Ingo

#3 lore raw
Quoted 20 lines · Ingo Molnar
Quoted 17 lines · Jeff Garzik

drivers/cdrom/sjcd.c static int __init sjcd_init(void) · line 1717

Quoted 12 lines
drivers/cdrom/sjcd.c +2 -3
@@ -1709,12 +1710,11 @@ static int __init sjcd_init(void)
 	printk(KERN_INFO "SJCD: Resetting: ");
 	sjcd_send_cmd(SCMD_RESET);
 	for (i = 1000; i > 0 && !sjcd_status_valid; --i) {
-		unsigned long timer;
-
 		/*
 		 * Wait 10ms approx.
 		 */
-		for (timer = jiffies; time_before_eq(jiffies, timer););
+		msleep(10);
+

I always worry when I see code like this, wondering if there is some arcane reason that I cannot fathom, that is being removed. You gotta wonder how long it has been since this driver was used, by anybody.

Quoted 6 lines
Quoted 1 line

Oh well, I cannot find fault with the patch, paranoia worries aside.

well, while i dont have that hardware, i found this by booting an allyesconfig bzImage which runs the code above, which locks up without this change. And then it booted fine with this change. So arcane issues aside, it does visibly improve things ;-)

But if you looked at the driver, you'd see 2 other identical busy loops in the very same function :-)

Signature · 1 line
Jens Axboe
#4 lore raw
In reply to Thomas Gleixner
Quoted 6 lines · Thomas Gleixner

From: Ingo Molnar <[email protected]>

The SJCD driver uses a jiffies busy loop. Replace it with msleep.

Okay, that's just waiting for a reset to complete, which seems perfectly fine with a sleeping loop like that, but...

I re-started working on a rewrite of the mitsumi legacy cdrom driver that Pekka Enberg and I have been doing this afternoon again and I ran into not being able to use sleeping loops there an hour ago!

The trouble there is that unless you poll the bloody thing like mad too much of the Q subchannels passes below you and you need a huge number of retries to get anything out of it. I noticed when I started adding audio bits that the driver took full seconds to complete some audio requests while the old driver was snappy in that regard. When I replaced our sleeping loop with a busy-wait same as the original the snappyness returned and moreover, reading the TOC from the CD went from something close to a minute to approximately a second. Thought that minute was just because I was dealing with an old junk 1-speed drive...

Now, as said, this looks to be fine since it's just waiting for a reset to complete, but unless you have the hardware to actually test, be careful in there.

Or in fact, maybe even decide there's not much point. The current plan is to do mitsumi, sony and panasonic and then throw the rest away (that trio is special in so far that controllers for them are still available on a lot of old ISA soundcards). Or do you actually have the hardware?

Rene.

#5 lore raw
Quoted 10 lines · Rene Herman

The trouble there is that unless you poll the bloody thing like mad too much of the Q subchannels passes below you and you need a huge number of retries to get anything out of it. I noticed when I started adding audio bits that the driver took full seconds to complete some audio requests while the old driver was snappy in that regard. When I replaced our sleeping loop with a busy-wait same as the original the snappyness returned and moreover, reading the TOC from the CD went from something close to a minute to approximately a second. Thought that minute was just because I was dealing with an old junk 1-speed drive...

ouch ...

then i guess the right approach is to not use jiffies but to use a calculated number of udelay(10) calls or so.

Ingo

#6 lore raw
Quoted 5 lines · Ingo Molnar
Quoted 1 line · Rene Herman

The trouble there is that unless you poll the bloody thing like mad too much of the Q subchannels passes below you and you need a huge number of retries to get anything out of it. I noticed when I started adding audio bits that the driver took full seconds to complete some audio requests while the old driver was snappy in that regard. When I replaced our sleeping loop with a busy-wait same as the original the snappyness returned and moreover, reading the TOC from the CD went from something close to a minute to approximately a second. Thought that minute was just because I was dealing with an old junk 1-speed drive...

ouch ...

Indeed...

Quoted 1 line

then i guess the right approach is to not use jiffies but to use a calculated number of udelay(10) calls or so.

I'm not sure that will do -- the disc is still spinning below us while we're delaying/sleeping. Alan just now suggested checking need_resched which sounds like okay advice. I'll be trying what still works...

(given that noone uses these drives for anything serious anymore I'm in fact also not overly worried about being a busy-looping pig and just saying "this hardware sucks, why are you using it?").

Rene.

#7 lore raw
In reply to Rene Herman
Quoted 5 lines

driver was snappy in that regard. When I replaced our sleeping loop with a busy-wait same as the original the snappyness returned and moreover, reading the TOC from the CD went from something close to a minute to approximately a second. Thought that minute was just because I was dealing with an old junk 1-speed drive...

It happens. If you want to be fairer you can check need_resched in the loop so that you still share the CPU fairly with other users.

#8 lore raw
Quoted 4 lines · Alan Cox
Quoted 1 line

driver was snappy in that regard. When I replaced our sleeping loop with a busy-wait same as the original the snappyness returned and moreover, reading the TOC from the CD went from something close to a minute to approximately a second. Thought that minute was just because I was dealing with an old junk 1-speed drive...

It happens. If you want to be fairer you can check need_resched in the loop so that you still share the CPU fairly with other users.

Thanks for the tip -- I'll do this.

Rene.