Skip to contentSkip to conversation

- Replaces yield() with

PATCH In review Gustavo Franco 4 participants 7 messages Apr 14, 2004 – Apr 25, 2004 Kernel janitors

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 - Replaces yield() with

Hi,

I'm new here, and this is my first patch, following the issue described by Matthew Wilcox: "Calling yield() Considered Harmful" on TODO list.

There is no doubt that isn't necessary replace all yield() calls, but i'm planning to check each case on the tree.Comments?

Hope that helps,
Gustavo Franco
p.s: This e-mail on the "from" is only for ML traffic, my main e-mail is stratus at acm.dot org.

cdrom/cdu31a.c 2004-04-13 22:44:17.000000000 -0300 renamed from cdrom/cdu31a.c.orig 2004-04-13 22:43:55.000000000 -0300 +1 -1
@@ -386,7 +386,7 @@ static inline void sony_sleep(void)
        unsigned long flags;
                                                                                         
 
        if (cdu31a_irq <= 0) {
-               yield();
+               schedule_timeout(1);
        } else {                /* Interrupt driven */
                                                                                         
                save_flags(flags);

_______________________________________________
Kernel-janitors mailing list
[email protected]
http://lists.osdl.org/mailman/listinfo/kernel-janitors

#1 lore raw

On Tue, 13 Apr 2004 22:59:10 -0300 Gustavo Franco wrote:

| Hi,
| 
| I'm new here, and this is my first patch, following the issue described by
| Matthew Wilcox: "Calling yield() Considered Harmful" on TODO list.
| 
| There is no doubt that isn't necessary replace all yield() calls, but i'm
| planning to check each case on the tree.Comments?
| 
| Hope that helps,
| Gustavo Franco
| p.s: This e-mail on the "from" is only for ML traffic, my main e-mail is
| stratus at acm.dot org.
| 
| 
| --- drivers/cdrom/cdu31a.c.orig 2004-04-13 22:43:55.000000000 -0300
| +++ drivers/cdrom/cdu31a.c      2004-04-13 22:44:17.000000000 -0300
| @@ -386,7 +386,7 @@ static inline void sony_sleep(void)
|         unsigned long flags;
|                                                                                          
| 
|         if (cdu31a_irq <= 0) {
| -               yield();
| +               schedule_timeout(1);
|         } else {                /* Interrupt driven */
| 
|                 save_flags(flags);

Makes sense, except that the comment above the function bothers me:

/*
* Wait a little while (used for polling the drive). If in initialization,
* setting a timeout doesn't work, so just loop for a while.
*/

I expect that cdu31a_irq <= 0 means "in initializatgion," so is using schedule_timeout() safe here? anyone checked this?

--
~Randy

_______________________________________________
Kernel-janitors mailing list
[email protected]
http://lists.osdl.org/mailman/listinfo/kernel-janitors

#2 lore raw
In reply to Gustavo Franco
Quoted 31 lines · Randy.Dunlap

On Tue, 13 Apr 2004 22:59:10 -0300 Gustavo Franco wrote:

| Hi,
| 
| I'm new here, and this is my first patch, following the issue described by
| Matthew Wilcox: "Calling yield() Considered Harmful" on TODO list.
| 
| There is no doubt that isn't necessary replace all yield() calls, but i'm
| planning to check each case on the tree.Comments?
| 
| Hope that helps,
| Gustavo Franco
| p.s: This e-mail on the "from" is only for ML traffic, my main e-mail is
| stratus at acm.dot org.
| 
| 
| --- drivers/cdrom/cdu31a.c.orig 2004-04-13 22:43:55.000000000 -0300
| +++ drivers/cdrom/cdu31a.c      2004-04-13 22:44:17.000000000 -0300
| @@ -386,7 +386,7 @@ static inline void sony_sleep(void)
|         unsigned long flags;
|                                                                                          
| 
|         if (cdu31a_irq <= 0) {
| -               yield();
| +               schedule_timeout(1);
|         } else {                /* Interrupt driven */
| 
|                 save_flags(flags);

Makes sense, except that the comment above the function bothers me:

Quoted 10 lines

/*
* Wait a little while (used for polling the drive). If in initialization,
* setting a timeout doesn't work, so just loop for a while.
*/

I expect that cdu31a_irq <= 0 means "in initializatgion," so is using schedule_timeout() safe here? anyone checked this?

Good question. Read this bit from comments on the source.

[...]
* This interface is (unfortunately) a polled interface. This is
* because most Sony interfaces are set up with DMA and interrupts
* disables. Some (like mine) do not even have the capability to
* handle interrupts or DMA. For this reason you will see a lot of
* the following:
*
* retry_count = jiffies+ SONY_JIFFIES_TIMEOUT;
* while (time_before(jiffies, retry_count) && (! <some condition to
wait for))
* {
* while (handle_sony_cd_attention())
* ;
*
* sony_sleep();

 *   }
 *   if (the condition not met)
 *   {

* return an error;
* }
*
* This ugly hack waits for something to happen, sleeping a little
* between every try. it also handles attentions, which are
* asynchronous events from the drive informing the driver that a disk
* has been inserted, removed, etc.
[...]

If it's related with initialization why it isn't being called from cdu31a_setup() ?
sony_sleep() is called from other functions as described above on the comments.

Hope that helps,
Gustavo Franco

_______________________________________________
Kernel-janitors mailing list
[email protected]
http://lists.osdl.org/mailman/listinfo/kernel-janitors

#3 lore raw
In reply to Gustavo Franco

On Thu, 15 Apr 2004 23:13:30 -0300 Gustavo Franco <[email protected]> wrote:

| Randy.Dunlap wrote:
| 
| >On Tue, 13 Apr 2004 22:59:10 -0300 Gustavo Franco wrote:
| >
| >| Hi,
| >| 
| >| I'm new here, and this is my first patch, following the issue described by
| >| Matthew Wilcox: "Calling yield() Considered Harmful" on TODO list.
| >| 
| >| There is no doubt that isn't necessary replace all yield() calls, but i'm
| >| planning to check each case on the tree.Comments?
| >| 
| >| Hope that helps,
| >| Gustavo Franco
| >| p.s: This e-mail on the "from" is only for ML traffic, my main e-mail is
| >| stratus at acm.dot org.
| >| 
| >| 
| >| --- drivers/cdrom/cdu31a.c.orig 2004-04-13 22:43:55.000000000 -0300
| >| +++ drivers/cdrom/cdu31a.c      2004-04-13 22:44:17.000000000 -0300
| >| @@ -386,7 +386,7 @@ static inline void sony_sleep(void)
| >|         unsigned long flags;
| >|                                                                                          
| >| 
| >|         if (cdu31a_irq <= 0) {
| >| -               yield();
| >| +               schedule_timeout(1);
| >|         } else {                /* Interrupt driven */
| >| 
| >|                 save_flags(flags);
| >
| >Makes sense, except that the comment above the function bothers me:
| >
| >/*
| > * Wait a little while (used for polling the drive).  If in initialization,
| > * setting a timeout doesn't work, so just loop for a while.
| > */
| >
| >I expect that cdu31a_irq <= 0 means "in initializatgion," so is
| >using schedule_timeout() safe here?  anyone checked this?
| >
| >  
| >
| 
| Good question. Read this bit from comments on the source.
| 
| [...]
|  * This interface is (unfortunately) a polled interface.  This is
|  * because most Sony interfaces are set up with DMA and interrupts
|  * disables.  Some (like mine) do not even have the capability to
|  * handle interrupts or DMA.  For this reason you will see a lot of
|  * the following:
|  *
|  *   retry_count = jiffies+ SONY_JIFFIES_TIMEOUT;
|  *   while (time_before(jiffies, retry_count) && (! <some condition to 
| wait for))
|  *   {
|  *      while (handle_sony_cd_attention())
|  *         ;
|  *
|  *      sony_sleep();
|  *   }
|  *   if (the condition not met)
|  *   {
|  *      return an error;
|  *   }
|  *
|  * This ugly hack waits for something to happen, sleeping a little
|  * between every try.  it also handles attentions, which are
|  * asynchronous events from the drive informing the driver that a disk
|  * has been inserted, removed, etc.
| [...]
| 
| If it's related with initialization why it isn't being called from 
| cdu31a_setup() ?
| sony_sleep() is called from other functions as described above on the 
| comments.

Yes, I looked at it again and I think that it's safe. I'll put it into the next KJ patchset.

--
~Randy

_______________________________________________
Kernel-janitors mailing list
[email protected]
http://lists.osdl.org/mailman/listinfo/kernel-janitors

#4 lore raw
In reply to Gustavo Franco

...

Quoted 9 lines
| >| --- drivers/cdrom/cdu31a.c.orig 2004-04-13 22:43:55.000000000 -0300
| >| +++ drivers/cdrom/cdu31a.c      2004-04-13 22:44:17.000000000 -0300
| >| @@ -386,7 +386,7 @@ static inline void sony_sleep(void)
| >|         unsigned long flags;
| >|                                                                                          
| >| 
| >|         if (cdu31a_irq <= 0) {
| >| -               yield();
| >| +               schedule_timeout(1);

set_current_state() missing.

Domen

_______________________________________________
Kernel-janitors mailing list
[email protected]
http://lists.osdl.org/mailman/listinfo/kernel-janitors

#5 lore raw
In reply to Gustavo Franco
Quoted 13 lines · Domen Puncer

...

Quoted 9 lines
| >| --- drivers/cdrom/cdu31a.c.orig 2004-04-13 22:43:55.000000000 -0300
| >| +++ drivers/cdrom/cdu31a.c      2004-04-13 22:44:17.000000000 -0300
| >| @@ -386,7 +386,7 @@ static inline void sony_sleep(void)
| >|         unsigned long flags;
| >|                                                                                          | >| | >|         if (cdu31a_irq <= 0) {
| >| -               yield();
| >| +               schedule_timeout(1);
Quoted 7 lines

set_current_state() missing.

Domen

Domen,

Are you talking about something like:
<http://www.iglu.org.il/lxr/ident?i=set_current_state>
set_current_state(TASK_UNINTERRUPTIBLE);
schedule_timeout(1);

Is is really necessary?

Thanks,
Gustavo Franco -- <[email protected]>

_______________________________________________
Kernel-janitors mailing list
[email protected]
http://lists.osdl.org/mailman/listinfo/kernel-janitors

#6 lore raw
In reply to Gustavo Franco
Quoted 20 lines · Gustavo Franco
Quoted 15 lines · Domen Puncer

...

Quoted 11 lines
| >| --- drivers/cdrom/cdu31a.c.orig 2004-04-13 22:43:55.000000000 -0300
| >| +++ drivers/cdrom/cdu31a.c      2004-04-13 22:44:17.000000000 -0300
| >| @@ -386,7 +386,7 @@ static inline void sony_sleep(void)
| >|         unsigned long flags;
| >|                                                                      
| >| 
| >|         if (cdu31a_irq <= 0) {
| >| -               yield();
| >| +               schedule_timeout(1);
Quoted 3 lines

set_current_state() missing.

Quoted 6 lines

Are you talking about something like:
<http://www.iglu.org.il/lxr/ident?i=set_current_state>
set_current_state(TASK_UNINTERRUPTIBLE);
schedule_timeout(1);

Is is really necessary?

Yes, comment above schedule_timeout:
* Make the current task sleep until @timeout jiffies have
* elapsed. The routine will return immediately unless
* the current task state has been set (see set_current_state()).

...the "immediately" part

Quoted 3 lines

Thanks,
Gustavo Franco -- <[email protected]>

_______________________________________________
Kernel-janitors mailing list
[email protected]
http://lists.osdl.org/mailman/listinfo/kernel-janitors