Re: [PATCH] printk: git rid of [sched_delayed] message for printk_deferred

[Steven Rostedt]

On Sat, 20 Sep 2014 07:12:24 +0200
Jan Kara <jack@xxxxxxx> wrote:

> On Thu 18-09-14 19:34:14, Peter Zijlstra wrote:
> > On Wed, Sep 17, 2014 at 08:31:35PM -0400, Steven Rostedt wrote:
> > > I totally didn't get what you wrote.
> > 
> > :-)
> > 
> > > We don't want to know if it got delayed, then the patch to remove that
> > > print seems correct.
> > 
> > Why would you not want to know that? Also was that the actual argument?
> > Lemme go check the earlier emails -- I cannot find that argument in the
> > first few emails.
>   Well, so what gets delayed is printing from kernel buffer to console.
> So this is the same as when you do printk() when console lock is taken by
> someone else. So it seems a bit strange to prepend [delayed] in some cases
> and not in others.
> 
> Another question is what the [delayed] prefix would be useful for? If the
> message eventually gets printed to console I don't see why you would care
> it was printed few ms after it entered the kernel buffer (after all the
> time stamp before the line will be the time when it entered the kernel
> buffer). And if the kernel crashes in such a way that the message doesn't
> get printed, then bad luck but prefix in the kernel log buffer isn't going
> to make that any better :)
> 
> This all feels like bikeshedding, I don't deeply care what gets done but I
> wanted to point out I don't really see a use for [delayed]...
> 

I pretty much agree with this assessment. I don't really care if
there's a "[delayed]" message or not. I now agree that it isn't really
that useful. Now what I do care about is that there's a bug with the
current code, and the non bikeshed argument is how to fix this bug.

The bug is that there's users of printk_deferred() that use KERN_WARN
in the format. This ends up showing "[delayed]<whacky-characters>
message". The fix needs to remove those whacky characters.

There's three ways to fix this bug.

1) change printk() to check for whacky characters before adding
"[delayed]" and either move them before it or remove them all together.

2) change all users of printk_deferred() to not add the KERN_WARN.

3) just get rid of adding the "[delayed]" message and make printk()
itself a bit cleaner.

I'm leaning towards #3 as I don't see the usefulness of that
"[delayed]" message if there's other cases that can also delay printk.
Remember, the code has been changed since the delayed message was added
to make sure that the printk_deferred() message gets into the printk
message in sequence of other printk's happening. The original way (when
the "delayed" message was added) used its own buffer and would write to
the log buffer at a later time, where the printk_deferred() could
actually come out of sequence with other printk()s, and the "[delayed]"
message would be useful for that case. But it's not useful anymore.

#1 of above will just makes printk more complex, and being such a
critical function which is already complex enough, I would like to not
do so.

#2 can fix the issue for now, until someone else adds a
printk_deferred() with another KERN_WARN. Worse yet, someone may add
one of the other KERN_* log levels and it will be ignored.


Getting beyond the bikeshedding, there's a real bug that should be
fixed. We just need to figure out what's the best way to do so.

-- Steve
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/