[PATCH 1/1] Char: moxa, fix and optimise empty timer

[PATCH 1/1] Char: moxa, fix and optimise empty timer

[Jiri Slaby]

moxa, fix and optimise empty timer

don't wait and delete empty timer in empty timer function. Also fire next
empty timer at rounded jiffies to save power.

Signed-off-by: Jiri Slaby <jirislaby@gmail.com>

---
commit da52793b6347e8b6b048526ce2422e29b20bb335
tree ad0bee78e45beef89dc740f81e0606d782296542
parent 3a69b463dcad1ff142f46e8fb74e7dc5a092eb60
author Jiri Slaby <jirislaby@gmail.com> Wed, 25 Jul 2007 23:42:49 +0200
committer Jiri Slaby <jirislaby@gmail.com> Wed, 25 Jul 2007 23:42:49 +0200

 drivers/char/moxa.c |    4 ++--
 1 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/char/moxa.c b/drivers/char/moxa.c
index ed76f0a..5000b3b 100644
--- a/drivers/char/moxa.c
+++ b/drivers/char/moxa.c
@@ -1040,14 +1040,14 @@ static void check_xmit_empty(unsigned long data)
 	struct moxa_port *ch;
 
 	ch = (struct moxa_port *) data;
-	del_timer_sync(&moxa_ports[ch->port].emptyTimer);
 	if (ch->tty && (ch->statusflags & EMPTYWAIT)) {
 		if (MoxaPortTxQueue(ch->port) == 0) {
 			ch->statusflags &= ~EMPTYWAIT;
 			tty_wakeup(ch->tty);
 			return;
 		}
-		mod_timer(&moxa_ports[ch->port].emptyTimer, jiffies + HZ);
+		mod_timer(&moxa_ports[ch->port].emptyTimer,
+				round_jiffies(jiffies + HZ));
 	} else
 		ch->statusflags &= ~EMPTYWAIT;
 }

Re: [PATCH 1/1] Char: moxa, fix and optimise empty timer

[Andrew Morton]

On Wed, 25 Jul 2007 23:43:29 +0200 (CEST) Jiri Slaby <jirislaby@gmail.com> wrote:

> moxa, fix and optimise empty timer
> 
> don't wait and delete empty timer in empty timer function. Also fire next
> empty timer at rounded jiffies to save power.
> 

What is actually being "fixed" here?

> 
> ---
> commit da52793b6347e8b6b048526ce2422e29b20bb335
> tree ad0bee78e45beef89dc740f81e0606d782296542
> parent 3a69b463dcad1ff142f46e8fb74e7dc5a092eb60
> author Jiri Slaby <jirislaby@gmail.com> Wed, 25 Jul 2007 23:42:49 +0200
> committer Jiri Slaby <jirislaby@gmail.com> Wed, 25 Jul 2007 23:42:49 +0200
> 
>  drivers/char/moxa.c |    4 ++--
>  1 files changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/char/moxa.c b/drivers/char/moxa.c
> index ed76f0a..5000b3b 100644
> --- a/drivers/char/moxa.c
> +++ b/drivers/char/moxa.c
> @@ -1040,14 +1040,14 @@ static void check_xmit_empty(unsigned long data)
>  	struct moxa_port *ch;
>  
>  	ch = (struct moxa_port *) data;
> -	del_timer_sync(&moxa_ports[ch->port].emptyTimer);
>  	if (ch->tty && (ch->statusflags & EMPTYWAIT)) {
>  		if (MoxaPortTxQueue(ch->port) == 0) {
>  			ch->statusflags &= ~EMPTYWAIT;
>  			tty_wakeup(ch->tty);
>  			return;
>  		}
> -		mod_timer(&moxa_ports[ch->port].emptyTimer, jiffies + HZ);
> +		mod_timer(&moxa_ports[ch->port].emptyTimer,
> +				round_jiffies(jiffies + HZ));
>  	} else
>  		ch->statusflags &= ~EMPTYWAIT;
>  }

A call to check_xmit_empty() used to guarantee that the timer was no longer
running (if the first `if' succeeds and the second does not), but that is
no longer the case.  You're sure this is correct?

Re: [PATCH 1/1] Char: moxa, fix and optimise empty timer

[Jiri Slaby]

Andrew Morton napsal(a):
> On Wed, 25 Jul 2007 23:43:29 +0200 (CEST) Jiri Slaby <jirislaby@gmail.com> wrote:
> 
>> moxa, fix and optimise empty timer
>>
>> don't wait and delete empty timer in empty timer function. Also fire next
>> empty timer at rounded jiffies to save power.
>>
> 
> What is actually being "fixed" here?

Lockup (see below).

>> ---
>> commit da52793b6347e8b6b048526ce2422e29b20bb335
>> tree ad0bee78e45beef89dc740f81e0606d782296542
>> parent 3a69b463dcad1ff142f46e8fb74e7dc5a092eb60
>> author Jiri Slaby <jirislaby@gmail.com> Wed, 25 Jul 2007 23:42:49 +0200
>> committer Jiri Slaby <jirislaby@gmail.com> Wed, 25 Jul 2007 23:42:49 +0200
>>
>>  drivers/char/moxa.c |    4 ++--
>>  1 files changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/char/moxa.c b/drivers/char/moxa.c
>> index ed76f0a..5000b3b 100644
>> --- a/drivers/char/moxa.c
>> +++ b/drivers/char/moxa.c
>> @@ -1040,14 +1040,14 @@ static void check_xmit_empty(unsigned long data)
>>  	struct moxa_port *ch;
>>  
>>  	ch = (struct moxa_port *) data;
>> -	del_timer_sync(&moxa_ports[ch->port].emptyTimer);

It's "wait until the code below me is finished". At least the _sync is buggy.

>>  	if (ch->tty && (ch->statusflags & EMPTYWAIT)) {
>>  		if (MoxaPortTxQueue(ch->port) == 0) {
>>  			ch->statusflags &= ~EMPTYWAIT;
>>  			tty_wakeup(ch->tty);
>>  			return;
>>  		}
>> -		mod_timer(&moxa_ports[ch->port].emptyTimer, jiffies + HZ);
>> +		mod_timer(&moxa_ports[ch->port].emptyTimer,
>> +				round_jiffies(jiffies + HZ));
>>  	} else
>>  		ch->statusflags &= ~EMPTYWAIT;
>>  }
> 
> A call to check_xmit_empty() used to guarantee that the timer was no longer
> running (if the first `if' succeeds and the second does not), but that is

Correct me if I'm wrong, but it guaranteed nothing. When the timer function is
called (note, that check_xmit_empty is timer.function), timer base lock is held
and there is no window to schedule the timer again when its function is running.

> no longer the case.  You're sure this is correct?

I still think so.

thanks,
-- 
Jiri Slaby (jirislaby@gmail.com)
Faculty of Informatics, Masaryk University