Skip to contentSkip to conversation

high cpu load on omap3 using musb

PATCH Adam Wozniak 3 participants 4 messages Jan 29, 2014 – Jul 21, 2014

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

With a USB 2.0 webcam attached to the OTG port on an OMAP3 (applies to overo gumstix, beagleboard, probably others) we see a high CPU load in a kworker thread.

Between 2.6.33 and 2.6.34 musb_core.c changed.

IRQ handlers changed with the result that a worker in musb_core.c got scheduled far more frequently than needed.

I've included a patch below against 3.7, but i think it'll apply against mainline. [I apologize for any whitespace mangling. I've also attached the patch.]

I'd like more eyeballs to tell me if this is right. I'd also like to know who I need to talk to to get this pushed into mainline.

--Adam

# The MUSB IRQ schedules work on every interrupt. # This is unnecessary, and causes excessive CPU load. #
# Here we only schedule work if there is something for

# the worker to do.
drivers/usb/musb/musb_core.c +3 -1
Index: git/drivers/usb/musb/musb_core.c
===================================================================
@@ -925,7 +925,9 @@ b_host:
        }
 #endif
 
-       schedule_work(&musb->irq_work);
+       if (musb->xceiv->state != musb->xceiv_old_state) {
+               schedule_work(&musb->irq_work);
+       }
 
        return handled;
 }

# The MUSB IRQ schedules work on every interrupt. # This is unnecessary, and causes excessive CPU load. #
# Here we only schedule work if there is something for

# the worker to do.
drivers/usb/musb/musb_core.c +3 -1
Index: git/drivers/usb/musb/musb_core.c
===================================================================
@@ -925,7 +925,9 @@ b_host:
 	}
 #endif
 
-	schedule_work(&musb->irq_work);
+	if (musb->xceiv->state != musb->xceiv_old_state) {
+		schedule_work(&musb->irq_work);
+	}
 
 	return handled;
 }
#1 lore raw

Hi Adam,

Quoted 9 lines · On Wednesday 29 January 2014 08:44:57 Adam Wozniak wrote:

With a USB 2.0 webcam attached to the OTG port on an OMAP3 (applies to overo gumstix, beagleboard, probably others) we see a high CPU load in a kworker thread.

Between 2.6.33 and 2.6.34 musb_core.c changed.

IRQ handlers changed with the result that a worker in musb_core.c got scheduled far more frequently than needed.

Quoted 6 lines

I've included a patch below against 3.7, but i think it'll apply against mainline.
[I apologize for any whitespace mangling. I've also attached the patch.]

I'd like more eyeballs to tell me if this is right. I'd also like to know who I need to talk to to get this pushed into mainline.

Running the scripts/get_maintainer.pl script on your patch produces

Felipe Balbi <[email protected]> (maintainer:MUSB MULTIPOINT H...) Greg Kroah-Hartman <[email protected]> (supporter:USB SUBSYSTEM) [email protected] (open list:MUSB MULTIPOINT H...) [email protected] (open list)

Felipe Balbi (CC'ed) is the person who you should talk to.

While we're touching the subject of scripts, you should run the scripts/checkpatch.pl script and fix errors and warnings before submitting patches. Please see Documentation/SubmittingPatches.

Last (but not least) piece of advice, don't give up if you don't receive replies to your patches. People are busy and mails fall to cracks from time to time.

Felipe, apart from the coding style violation and the possibly missing locking, what's your opinion on this ? Does the patch make sense ?

Quoted 21 lines

# The MUSB IRQ schedules work on every interrupt. # This is unnecessary, and causes excessive CPU load. #
# Here we only schedule work if there is something for

# the worker to do.

Index: git/drivers/usb/musb/musb_core.c
===================================================================

drivers/usb/musb/musb_core.c +3 -1
@@ -925,7 +925,9 @@ b_host:
         }
  #endif
 
-       schedule_work(&musb->irq_work);
+       if (musb->xceiv->state != musb->xceiv_old_state) {
+               schedule_work(&musb->irq_work);
+       }
 
         return handled;
  }
Signature · 3 lines
Regards,

Laurent Pinchart
#2 lore raw
Quoted 32 lines · Laurent Pinchart

Hi Adam,

Quoted 9 lines · On Wednesday 29 January 2014 08:44:57 Adam Wozniak wrote:

With a USB 2.0 webcam attached to the OTG port on an OMAP3 (applies to overo gumstix, beagleboard, probably others) we see a high CPU load in a kworker thread.

Between 2.6.33 and 2.6.34 musb_core.c changed.

IRQ handlers changed with the result that a worker in musb_core.c got scheduled far more frequently than needed.

Quoted 6 lines

I've included a patch below against 3.7, but i think it'll apply against mainline.
[I apologize for any whitespace mangling. I've also attached the patch.]

I'd like more eyeballs to tell me if this is right. I'd also like to know who I need to talk to to get this pushed into mainline.

Running the scripts/get_maintainer.pl script on your patch produces

Felipe Balbi <[email protected]> (maintainer:MUSB MULTIPOINT H...) Greg Kroah-Hartman <[email protected]> (supporter:USB SUBSYSTEM) [email protected] (open list:MUSB MULTIPOINT H...) [email protected] (open list)

Felipe Balbi (CC'ed) is the person who you should talk to.

While we're touching the subject of scripts, you should run the scripts/checkpatch.pl script and fix errors and warnings before submitting patches. Please see Documentation/SubmittingPatches.

Quoted 6 lines

Last (but not least) piece of advice, don't give up if you don't receive replies to your patches. People are busy and mails fall to cracks from time to time.

Felipe, apart from the coding style violation and the possibly missing locking, what's your opinion on this ? Does the patch make sense ?

It's a duplication of the check which is already in musb_irq_work():

1742 static void musb_irq_work(struct work_struct *data) 1743 {
1744 struct musb *musb = container_of(data, struct musb, irq_work);
1745
1746 if (musb->xceiv->state != musb->xceiv_old_state) {
1747 musb->xceiv_old_state = musb->xceiv->state;
1748 sysfs_notify(&musb->controller->kobj, NULL, "mode");
1749 }
1750 }

That does look better, but I'd need the check inside musb_irq_work() to be removed and commit log would have to improve a bit.

ps: there's no missing locking, musb_stage0_irq() is called within musb_interrupt() which is called within a locked IRQ handler.

Signature · 1 line
balbi
#3 lore raw

Hi Felipe and Adam,

Quoted 51 lines · On Monday 21 July 2014 10:40:52 Felipe Balbi wrote:
Quoted 31 lines · Laurent Pinchart
Quoted 9 lines · On Wednesday 29 January 2014 08:44:57 Adam Wozniak wrote:

With a USB 2.0 webcam attached to the OTG port on an OMAP3 (applies to overo gumstix, beagleboard, probably others) we see a high CPU load in a kworker thread.

Between 2.6.33 and 2.6.34 musb_core.c changed.

IRQ handlers changed with the result that a worker in musb_core.c got scheduled far more frequently than needed.

Quoted 7 lines

I've included a patch below against 3.7, but i think it'll apply against mainline.
[I apologize for any whitespace mangling. I've also attached the patch.]

I'd like more eyeballs to tell me if this is right. I'd also like to know who I need to talk to to get this pushed into mainline.

Running the scripts/get_maintainer.pl script on your patch produces

Felipe Balbi <[email protected]> (maintainer:MUSB MULTIPOINT H...) Greg Kroah-Hartman <[email protected]> (supporter:USB SUBSYSTEM) [email protected] (open list:MUSB MULTIPOINT H...) [email protected] (open list)

Felipe Balbi (CC'ed) is the person who you should talk to.

While we're touching the subject of scripts, you should run the scripts/checkpatch.pl script and fix errors and warnings before submitting patches. Please see Documentation/SubmittingPatches.

Quoted 6 lines

Last (but not least) piece of advice, don't give up if you don't receive replies to your patches. People are busy and mails fall to cracks from time to time.

Felipe, apart from the coding style violation and the possibly missing locking, what's your opinion on this ? Does the patch make sense ?

It's a duplication of the check which is already in musb_irq_work():

1742 static void musb_irq_work(struct work_struct *data) 1743 {
1744 struct musb *musb = container_of(data, struct musb, irq_work);
1745
1746 if (musb->xceiv->state != musb->xceiv_old_state) {
1747 musb->xceiv_old_state = musb->xceiv->state;
1748 sysfs_notify(&musb->controller->kobj, NULL, "mode");
1749 }
1750 }

Quoted 2 lines

That does look better, but I'd need the check inside musb_irq_work() to be removed and commit log would have to improve a bit.

OK. Adam, could you please modify the patch accordingly and resubmit it ?

Quoted 2 lines

ps: there's no missing locking, musb_stage0_irq() is called within musb_interrupt() which is called within a locked IRQ handler.

I hadn't checked that, thank you for the confirmation.

Signature · 3 lines
Regards,

Laurent Pinchart