ohci : strange looking use of round_jiffies_relative. possible bug ?

ohci : strange looking use of round_jiffies_relative. possible bug ?

[Richard Kennedy]

I was looking at the uses of round_jiffies and noticed that the ohci
unlink_watchdog_func() calls 
mod_timer(&ohci->unlink_watchdog, round_jiffies_relative(HZ));

Which looks a bit strange as it will set the timer to 0 <= t < HZ.
Reading the comments I think that is really is expecting to setup timer
to (now + 1s)

mod_timer(&ohci->unlink_watchdog, round_jiffies(jiffies + HZ));

If this is not true then perhaps it needs a comment to explain what it
_is_ trying to do, because it's got me baffled ;)


Cheers
Richard

Re: ohci : strange looking use of round_jiffies_relative. possible bug ?

[Alan Stern]

On Wed, 26 Mar 2008, Richard Kennedy wrote:

> I was looking at the uses of round_jiffies and noticed that the ohci
> unlink_watchdog_func() calls 
> mod_timer(&ohci->unlink_watchdog, round_jiffies_relative(HZ));
> 
> Which looks a bit strange as it will set the timer to 0 <= t < HZ.
> Reading the comments I think that is really is expecting to setup timer
> to (now + 1s)
> 
> mod_timer(&ohci->unlink_watchdog, round_jiffies(jiffies + HZ));
> 
> If this is not true then perhaps it needs a comment to explain what it
> _is_ trying to do, because it's got me baffled ;)

I'd say it's a bug.  Or rather a pair of bugs, since that mod_timer() 
call appears in two places.

Alan Stern

Re: ohci : strange looking use of round_jiffies_relative. possible bug ?

[Richard Kennedy]

On Wed, 2008-03-26 at 11:37 -0400, Alan Stern wrote:
> On Wed, 26 Mar 2008, Richard Kennedy wrote:
> 
> > I was looking at the uses of round_jiffies and noticed that the ohci
> > unlink_watchdog_func() calls 
> > mod_timer(&ohci->unlink_watchdog, round_jiffies_relative(HZ));
> > 
> > Which looks a bit strange as it will set the timer to 0 <= t < HZ.
> > Reading the comments I think that is really is expecting to setup timer
> > to (now + 1s)
> > 
> > mod_timer(&ohci->unlink_watchdog, round_jiffies(jiffies + HZ));
> > 
> > If this is not true then perhaps it needs a comment to explain what it
> > _is_ trying to do, because it's got me baffled ;)
> 
> I'd say it's a bug.  Or rather a pair of bugs, since that mod_timer() 
> call appears in two places.
> 
> Alan Stern
> 
Thanks. In that case I'll post a patch.
Richard 

[patch] usb ohci : fix 2 timers to fire at jiffies + 1s

[Richard Kennedy]

    Code inspection discovered in 2 places timers were being
    incorrectly setup using round_jiffies_relative(HZ).
    The timer would then fire at time (0 <= T < HZ).
    
    Fix them to use round_jiffies(jiffies + HZ);
    
    Signed-off-by: Richard Kennedy <richard@rsk.demon.co.uk>
---

compiled on latest v2.6.25-rc7-11 git head but not otherwise tested.
Richard

diff --git a/drivers/usb/host/ohci-hcd.c b/drivers/usb/host/ohci-hcd.c
index dd4798e..33f1c1c 100644
--- a/drivers/usb/host/ohci-hcd.c
+++ b/drivers/usb/host/ohci-hcd.c
@@ -467,7 +467,7 @@ static void unlink_watchdog_func(unsigned long _ohci)
 out:
 	kfree(seen);
 	if (ohci->eds_scheduled)
-		mod_timer(&ohci->unlink_watchdog, round_jiffies_relative(HZ));
+		mod_timer(&ohci->unlink_watchdog, round_jiffies(jiffies + HZ));
 done:
 	spin_unlock_irqrestore(&ohci->lock, flags);
 }
diff --git a/drivers/usb/host/ohci-q.c b/drivers/usb/host/ohci-q.c
index 5181732..9c9f3b5 100644
--- a/drivers/usb/host/ohci-q.c
+++ b/drivers/usb/host/ohci-q.c
@@ -169,7 +169,7 @@ static int ed_schedule (struct ohci_hcd *ohci, struct ed *ed)
 	if (quirk_zfmicro(ohci)
 			&& (ed->type == PIPE_INTERRUPT)
 			&& !(ohci->eds_scheduled++))
-		mod_timer(&ohci->unlink_watchdog, round_jiffies_relative(HZ));
+		mod_timer(&ohci->unlink_watchdog, round_jiffies(jiffies + HZ));
 	wmb ();
 
 	/* we care about rm_list when setting CLE/BLE in case the HC was at

Re: ohci : strange looking use of round_jiffies_relative. possible bug ?

[Mike Nuss]

On Wed, Mar 26, 2008 at 11:46 AM, Richard Kennedy
<richard@rsk.demon.co.uk> wrote:
>
>  On Wed, 2008-03-26 at 11:37 -0400, Alan Stern wrote:
>  > On Wed, 26 Mar 2008, Richard Kennedy wrote:
>  >
>  > > I was looking at the uses of round_jiffies and noticed that the ohci
>  > > unlink_watchdog_func() calls
>  > > mod_timer(&ohci->unlink_watchdog, round_jiffies_relative(HZ));
>  > >
>  > > Which looks a bit strange as it will set the timer to 0 <= t < HZ.
>  > > Reading the comments I think that is really is expecting to setup timer
>  > > to (now + 1s)
>  > >
>  > > mod_timer(&ohci->unlink_watchdog, round_jiffies(jiffies + HZ));
>  > >
>  > > If this is not true then perhaps it needs a comment to explain what it
>  > > _is_ trying to do, because it's got me baffled ;)

Yes, my intent there was a roughly 1 second delay. You're correct, it's wrong.

>  > I'd say it's a bug.  Or rather a pair of bugs, since that mod_timer()
>  > call appears in two places.

Yup.

Mike

Re: [patch] usb ohci : fix 2 timers to fire at jiffies + 1s

[David Brownell]

On Wednesday 26 March 2008, Richard Kennedy wrote:
>     Code inspection discovered in 2 places timers were being
>     incorrectly setup using round_jiffies_relative(HZ).
>     The timer would then fire at time (0 <= T < HZ).
>     
>     Fix them to use round_jiffies(jiffies + HZ);
>     
>     Signed-off-by: Richard Kennedy <richard@rsk.demon.co.uk>

Acked-by: David Brownell <dbrownell@users.sourceforge.net>

> ---
> 
> compiled on latest v2.6.25-rc7-11 git head but not otherwise tested.
> Richard
> 
> diff --git a/drivers/usb/host/ohci-hcd.c b/drivers/usb/host/ohci-hcd.c
> index dd4798e..33f1c1c 100644
> --- a/drivers/usb/host/ohci-hcd.c
> +++ b/drivers/usb/host/ohci-hcd.c
> @@ -467,7 +467,7 @@ static void unlink_watchdog_func(unsigned long _ohci)
>  out:
>  	kfree(seen);
>  	if (ohci->eds_scheduled)
> -		mod_timer(&ohci->unlink_watchdog, round_jiffies_relative(HZ));
> +		mod_timer(&ohci->unlink_watchdog, round_jiffies(jiffies + HZ));
>  done:
>  	spin_unlock_irqrestore(&ohci->lock, flags);
>  }
> diff --git a/drivers/usb/host/ohci-q.c b/drivers/usb/host/ohci-q.c
> index 5181732..9c9f3b5 100644
> --- a/drivers/usb/host/ohci-q.c
> +++ b/drivers/usb/host/ohci-q.c
> @@ -169,7 +169,7 @@ static int ed_schedule (struct ohci_hcd *ohci, struct ed *ed)
>  	if (quirk_zfmicro(ohci)
>  			&& (ed->type == PIPE_INTERRUPT)
>  			&& !(ohci->eds_scheduled++))
> -		mod_timer(&ohci->unlink_watchdog, round_jiffies_relative(HZ));
> +		mod_timer(&ohci->unlink_watchdog, round_jiffies(jiffies + HZ));
>  	wmb ();
>  
>  	/* we care about rm_list when setting CLE/BLE in case the HC was at
> 
> 

Re: ohci : strange looking use of round_jiffies_relative. possible bug ?

[David Brownell]

On Wednesday 26 March 2008, Mike Nuss wrote:
> On Wed, Mar 26, 2008 at 11:46 AM, Richard Kennedy
> <richard@rsk.demon.co.uk> wrote:
> 
> >  > I'd say it's a bug.  Or rather a pair of bugs, since that mod_timer()
> >  > call appears in two places.
> 
> Yup.

Although ... one specific to the Compaq ZF Micro chipset,
so it's a bug with rather limited impact.  Do Not Panic.