Skip to contentSkip to conversation

release() locking

DISCUSSION Udo A. Steinberg 3 participants 10 messages Dec 7, 2001 Char and misc drivers

lore
Share & tools
lore
#0 lore raw

Hi Andrew,

According to Linus' 2.5.1-pre changelog, the release locking changes introduced in -pre5 are your work. Those changes, however, seem to break the keyboard driver:

keyboard: Timeout - AT keyboard not present?(f4)

Other people (i.e. Mike Galbraith) have been experiencing the same.

Do you have an updated patch which fixes those issues? -pre6 still contains the same stuff as -pre5 and if it's broken then Linus should probably back it out.

Regards,
Udo.

#1 lore raw
Quoted 10 lines · Udo A. Steinberg

Hi Andrew,

According to Linus' 2.5.1-pre changelog, the release locking changes introduced in -pre5 are your work. Those changes, however, seem to break the keyboard driver:

keyboard: Timeout - AT keyboard not present?(f4)

Other people (i.e. Mike Galbraith) have been experiencing the same.

wasntmeididntdoit

Quoted 3 lines

Do you have an updated patch which fixes those issues? -pre6 still contains the same stuff as -pre5 and if it's broken then Linus should probably back it out.

My patch simply fixed a few things like holding spinlocks across sleeping functions, forgetting to release locks when returning from functions, etc.

Yes, others have suggested that the whole lot should be reverted, for several reasons. However it looks like that won't happen, so we need to debug the present code. But it works for me.

I can review the code, see if anything stands out. But it'll probably require someone who can reproduce it to be able to fix it.

-

#2 lore raw
In reply to Udo A. Steinberg
Quoted 6 lines · Andrew Morton

Yes, others have suggested that the whole lot should be reverted, for several reasons. However it looks like that won't happen, so we need to debug the present code. But it works for me.

I can review the code, see if anything stands out. But it'll probably require someone who can reproduce it to be able to fix it.

That would be good. I can reliably reproduce the problem, so if you want me to try out some patches, just send them here.

Regards,
Udo.

#3 lore raw
In reply to Udo A. Steinberg
Quoted 8 lines · Andrew Morton
Quoted 6 lines · Udo A. Steinberg

Hi Andrew,

According to Linus' 2.5.1-pre changelog, the release locking changes introduced in -pre5 are your work. Those changes, however, seem to break the keyboard driver:

Quoted 10 lines
Quoted 4 lines

keyboard: Timeout - AT keyboard not present?(f4)

Other people (i.e. Mike Galbraith) have been experiencing the same.

wasntmeididntdoit

Quoted 4 lines

Do you have an updated patch which fixes those issues? -pre6 still contains the same stuff as -pre5 and if it's broken then Linus should probably back it out.

I'm responsible for the release locking changes. But, I don't think that the problems are a result of those changes. There have been some other patches that might have caused the problem. Take a look at this thread:

http://marc.theaimsgroup.com/?l=linux-kernel&m=100745930928683&w=2

Jens Axboe posted a patch. I asked him:

Quoted 1 line

So, what was the actual problem?

bio_alloc() not waiting on the reserved pool for free entries, even though __GFP_WAIT was set. No need for __GFP_IO in that case too.

Udo, did you apply the patch that Jens sent?

#4 lore raw
In reply to Udo A. Steinberg
Quoted 7 lines · David C. Hansen

I'm responsible for the release locking changes. But, I don't think that the problems are a result of those changes. There have been some other patches that might have caused the problem. Take a look at this thread:

Jens Axboe posted a patch. I asked him:
[...]

The patch that Jens posted was the fix for the oops I posted. The oops was related to the bio stuff and not to the release() locking. I only mentioned the keyboard stuff because I initially didn't know if there was any connection between that and the oops.

Quoted 1 line

Udo, did you apply the patch that Jens sent?

Yes - and it cured the oops (the one in the thread you refer to), however the keyboard problems are still there.

I guess there's something wrong with the changes you made, and it only shows with the modifications that Andrew made - and since he says he only fixed some bits of the code, the broken bits must have been there before.

Regards,
Udo.

#5 lore raw
In reply to Udo A. Steinberg
Quoted 5 lines · Udo A. Steinberg

I guess there's something wrong with the changes you made, and it only shows with the modifications that Andrew made - and since he says he only fixed some bits of the code, the broken bits must have been there before.

Maybe so. Can you identify the exact kernel version at which the problem started?

#6 lore raw
In reply to Udo A. Steinberg
Quoted 10 lines · Andrew Morton
Quoted 5 lines · Udo A. Steinberg

I guess there's something wrong with the changes you made, and it only shows with the modifications that Andrew made - and since he says he only fixed some bits of the code, the broken bits must have been there before.

Maybe so. Can you identify the exact kernel version at which the problem started?

The release() functions patch went into pre3. It looks like Jens' bio changes went into pre4.

#7 lore raw
In reply to Udo A. Steinberg
Quoted 5 lines · David C. Hansen
Quoted 2 lines · Andrew Morton

Maybe so. Can you identify the exact kernel version at which the problem started?

Yes. I tried the entire 2.5.1-pre series:

-pre1 and -pre2 are fine.
-pre3 doesn't compile out of the box and with 3 trivial compile fixes to
pc_keyb.c shows the problem.

So anything including and after -pre3 is broken wrt. locking.

Quoted 2 lines

The release() functions patch went into pre3. It looks like Jens' bio changes went into pre4.

Like I said before - it has nothing to do with Jens' bio stuff. Also the fixes done by Andrew have nothing to do with it.
Moreover, if I back out the changes to pc_keyb.c, the problem goes away.

Regards,
Udo.

#8 lore raw
In reply to Udo A. Steinberg
Quoted 11 lines · Udo A. Steinberg
Quoted 5 lines · David C. Hansen
Quoted 2 lines · Andrew Morton

Maybe so. Can you identify the exact kernel version at which the problem started?

Yes. I tried the entire 2.5.1-pre series:

Review

Quoted 3 lines
(quoted diff) +0 -2
-pre1 and -pre2 are fine.
-pre3 doesn't compile out of the box and with 3 trivial compile fixes to
      pc_keyb.c shows the problem.

Hum. send_data() requires that local interrupts be enabled.

Does this fix it?

drivers/char/pc_keyb.c +1 -1
@@ -1090,6 +1090,7 @@ static int open_aux(struct inode * inode
 		spin_unlock_irqrestore(&aux_count_lock, flags);
 		return -EBUSY;
 	}
+	spin_unlock_irqrestore(&aux_count_lock, flags);
 	kbd_write_command_w(KBD_CCMD_MOUSE_ENABLE);	/* Enable the
 							   auxiliary port on
 							   controller. */
@@ -1099,7 +1100,6 @@ static int open_aux(struct inode * inode
 	mdelay(2);			/* Ensure we follow the kbc access delay rules.. */
 
 	send_data(KBD_CMD_ENABLE);	/* try to workaround toshiba4030cdt problem */
-	spin_unlock_irqrestore(&aux_count_lock, flags);
 	return 0;
 }
#9 lore raw
In reply to Udo A. Steinberg
Quoted 4 lines · Andrew Morton

Hum. send_data() requires that local interrupts be enabled.

Does this fix it?

[Patch snipped]

Yes, that fixes it. Thanks! Please submit it to Linus for inclusion in pre7.

Regards,
Udo.