Re: [PATCH] laptop mode

Re: [PATCH] laptop mode

[Jeff Garzik]

Linux Kernel Mailing List wrote:
> ChangeSet 1.1150.1.52, 2003/10/08 10:49:45-03:00, axboe@suse.de
> 
> 	[PATCH] laptop mode
> 	
> 	Hi Marcelo,
> 	
> 	Lots of people have been using this patch with great success, and it's
> 	been in the SuSE kernel for some months now too. It is also in the -benh
> 	ppc tree
> 	
> 	Basically, it introduces a write back mode of dirty and journal data
> 	that is more suitable for laptops. At the block layer end, it schedules
> 	write out of dirty data after the disk has been idle for 5 seconds.
> 	
> 	Laptop mode can be switched on and off with /proc/sys/vm/laptop_mode.
> 	There is also a block_dump sysctl, which if enabled will dump who
> 	dirties and writes out data. This is very helpful in pinning down who is
> 	causing unnecessary writes to the disk.

Red Hat just dropped this patch since it was suspected of data 
corruption ;-(

	Jeff

Re: [PATCH] laptop mode

[Jens Axboe]

On Thu, Oct 09 2003, Jeff Garzik wrote:
> Linux Kernel Mailing List wrote:
> >ChangeSet 1.1150.1.52, 2003/10/08 10:49:45-03:00, axboe@suse.de
> >
> >	[PATCH] laptop mode
> >	
> >	Hi Marcelo,
> >	
> >	Lots of people have been using this patch with great success, and 
> >	it's
> >	been in the SuSE kernel for some months now too. It is also in the 
> >	-benh
> >	ppc tree
> >	
> >	Basically, it introduces a write back mode of dirty and journal data
> >	that is more suitable for laptops. At the block layer end, it 
> >	schedules
> >	write out of dirty data after the disk has been idle for 5 seconds.
> >	
> >	Laptop mode can be switched on and off with /proc/sys/vm/laptop_mode.
> >	There is also a block_dump sysctl, which if enabled will dump who
> >	dirties and writes out data. This is very helpful in pinning down 
> >	who is
> >	causing unnecessary writes to the disk.
> 
> Red Hat just dropped this patch since it was suspected of data 
> corruption ;-(

Eh? Care to explain a bit further? I'm not aware of any data corruption
issues there, and it's certainly simple enough to easily audit.

And how kind of Red Hat to not inform me of any suspicion in this
regard.

-- 
Jens Axboe

Re: [PATCH] laptop mode

[Marcelo Tosatti]

On Thu, 9 Oct 2003, Jeff Garzik wrote:

> Linux Kernel Mailing List wrote:
> > ChangeSet 1.1150.1.52, 2003/10/08 10:49:45-03:00, axboe@suse.de
> > 
> > 	[PATCH] laptop mode
> > 	
> > 	Hi Marcelo,
> > 	
> > 	Lots of people have been using this patch with great success, and it's
> > 	been in the SuSE kernel for some months now too. It is also in the -benh
> > 	ppc tree
> > 	
> > 	Basically, it introduces a write back mode of dirty and journal data
> > 	that is more suitable for laptops. At the block layer end, it schedules
> > 	write out of dirty data after the disk has been idle for 5 seconds.
> > 	
> > 	Laptop mode can be switched on and off with /proc/sys/vm/laptop_mode.
> > 	There is also a block_dump sysctl, which if enabled will dump who
> > 	dirties and writes out data. This is very helpful in pinning down who is
> > 	causing unnecessary writes to the disk.
> 
> Red Hat just dropped this patch since it was suspected of data 
> corruption ;-(

Uh, oh... Jens? 

Re: [PATCH] laptop mode

[Jens Axboe]

On Thu, Oct 09 2003, Marcelo Tosatti wrote:
> 
> 
> On Thu, 9 Oct 2003, Jeff Garzik wrote:
> 
> > Linux Kernel Mailing List wrote:
> > > ChangeSet 1.1150.1.52, 2003/10/08 10:49:45-03:00, axboe@suse.de
> > > 
> > > 	[PATCH] laptop mode
> > > 	
> > > 	Hi Marcelo,
> > > 	
> > > 	Lots of people have been using this patch with great success, and it's
> > > 	been in the SuSE kernel for some months now too. It is also in the -benh
> > > 	ppc tree
> > > 	
> > > 	Basically, it introduces a write back mode of dirty and journal data
> > > 	that is more suitable for laptops. At the block layer end, it schedules
> > > 	write out of dirty data after the disk has been idle for 5 seconds.
> > > 	
> > > 	Laptop mode can be switched on and off with /proc/sys/vm/laptop_mode.
> > > 	There is also a block_dump sysctl, which if enabled will dump who
> > > 	dirties and writes out data. This is very helpful in pinning down who is
> > > 	causing unnecessary writes to the disk.
> > 
> > Red Hat just dropped this patch since it was suspected of data 
> > corruption ;-(
> 
> Uh, oh... Jens? 

See my previous mail. I don't see any problems with it, and I've
certainly not heard of (or experienced myself) problems with the patch.
I'm waiting for Jeff to expand on his mail, surely he/RH must know more
about this issue.

-- 
Jens Axboe

Re: [PATCH] laptop mode

[Dave Jones]

On Thu, Oct 09, 2003 at 04:05:47PM +0200, Jens Axboe wrote:

 > > Red Hat just dropped this patch since it was suspected of data 
 > > corruption ;-(
 > Eh? Care to explain a bit further? I'm not aware of any data corruption
 > issues there, and it's certainly simple enough to easily audit.

3-4 cases of random data corruption, all using Quantum Fireball drives,
all with different IDE chipsets.

 > And how kind of Red Hat to not inform me of any suspicion in this
 > regard.

I want to get facts right before crying wolf.
Right now laptopmode/aam is just a suspect. There are still 1-2 other
small patches against IDE which could be the reason.  We've dropped
laptopmode/aam for the time being to see if the folks seeing repeatable
corruption suddenly start behaving again.

		Dave

-- 
 Dave Jones     http://www.codemonkey.org.uk

Re: [PATCH] laptop mode

[Jeff Garzik]

Jens Axboe wrote:
> On Thu, Oct 09 2003, Marcelo Tosatti wrote:
> 
>>
>>On Thu, 9 Oct 2003, Jeff Garzik wrote:
>>
>>
>>>Linux Kernel Mailing List wrote:
>>>
>>>>ChangeSet 1.1150.1.52, 2003/10/08 10:49:45-03:00, axboe@suse.de
>>>>
>>>>	[PATCH] laptop mode
>>>>	
>>>>	Hi Marcelo,
>>>>	
>>>>	Lots of people have been using this patch with great success, and it's
>>>>	been in the SuSE kernel for some months now too. It is also in the -benh
>>>>	ppc tree
>>>>	
>>>>	Basically, it introduces a write back mode of dirty and journal data
>>>>	that is more suitable for laptops. At the block layer end, it schedules
>>>>	write out of dirty data after the disk has been idle for 5 seconds.
>>>>	
>>>>	Laptop mode can be switched on and off with /proc/sys/vm/laptop_mode.
>>>>	There is also a block_dump sysctl, which if enabled will dump who
>>>>	dirties and writes out data. This is very helpful in pinning down who is
>>>>	causing unnecessary writes to the disk.
>>>
>>>Red Hat just dropped this patch since it was suspected of data 
>>>corruption ;-(
>>
>>Uh, oh... Jens? 
> 
> 
> See my previous mail. I don't see any problems with it, and I've
> certainly not heard of (or experienced myself) problems with the patch.
> I'm waiting for Jeff to expand on his mail, surely he/RH must know more
> about this issue.


That's 100% of my knowledge.  Talk to arjan/davej/sct for more info...

That's what RH's field testing showed us, ignore it if you wish... :)

	Jeff

Re: [PATCH] laptop mode

[Jens Axboe]

On Thu, Oct 09 2003, Jeff Garzik wrote:
> Jens Axboe wrote:
> >On Thu, Oct 09 2003, Marcelo Tosatti wrote:
> >
> >>
> >>On Thu, 9 Oct 2003, Jeff Garzik wrote:
> >>
> >>
> >>>Linux Kernel Mailing List wrote:
> >>>
> >>>>ChangeSet 1.1150.1.52, 2003/10/08 10:49:45-03:00, axboe@suse.de
> >>>>
> >>>>	[PATCH] laptop mode
> >>>>	
> >>>>	Hi Marcelo,
> >>>>	
> >>>>	Lots of people have been using this patch with great success, and 
> >>>>	it's
> >>>>	been in the SuSE kernel for some months now too. It is also in the 
> >>>>	-benh
> >>>>	ppc tree
> >>>>	
> >>>>	Basically, it introduces a write back mode of dirty and journal data
> >>>>	that is more suitable for laptops. At the block layer end, it 
> >>>>	schedules
> >>>>	write out of dirty data after the disk has been idle for 5 seconds.
> >>>>	
> >>>>	Laptop mode can be switched on and off with /proc/sys/vm/laptop_mode.
> >>>>	There is also a block_dump sysctl, which if enabled will dump who
> >>>>	dirties and writes out data. This is very helpful in pinning down 
> >>>>	who is
> >>>>	causing unnecessary writes to the disk.
> >>>
> >>>Red Hat just dropped this patch since it was suspected of data 
> >>>corruption ;-(
> >>
> >>Uh, oh... Jens? 
> >
> >
> >See my previous mail. I don't see any problems with it, and I've
> >certainly not heard of (or experienced myself) problems with the patch.
> >I'm waiting for Jeff to expand on his mail, surely he/RH must know more
> >about this issue.
> 
> 
> That's 100% of my knowledge.  Talk to arjan/davej/sct for more info...

CC'ed

> That's what RH's field testing showed us, ignore it if you wish... :)

As far as I can tell, laptop-mode cannot cause any corruption on it own
that cannot otherwise happen without laptop mode. So if you see
corruption, it is most likely exposing some other bug. Or the field
users run out of battery and 10 minutes of work goes down the drain,
that's a calcuated risk though.

-- 
Jens Axboe

Re: [PATCH] laptop mode

[Jens Axboe]

On Thu, Oct 09 2003, Dave Jones wrote:
> On Thu, Oct 09, 2003 at 04:05:47PM +0200, Jens Axboe wrote:
> 
>  > > Red Hat just dropped this patch since it was suspected of data 
>  > > corruption ;-(
>  > Eh? Care to explain a bit further? I'm not aware of any data corruption
>  > issues there, and it's certainly simple enough to easily audit.
> 
> 3-4 cases of random data corruption, all using Quantum Fireball drives,
> all with different IDE chipsets.
> 
>  > And how kind of Red Hat to not inform me of any suspicion in this
>  > regard.
> 
> I want to get facts right before crying wolf.

Fair enough.

> Right now laptopmode/aam is just a suspect. There are still 1-2 other
> small patches against IDE which could be the reason.  We've dropped
> laptopmode/aam for the time being to see if the folks seeing repeatable
> corruption suddenly start behaving again.

aam patch is far more risky, it's a far more likely suspect. That patch
never reall did go out of beta. Dropping laptop-mode and aam at the same
time is bad engineering practice :).

laptop-mode cannot cause corruption that cannot show otherwise.

-- 
Jens Axboe

Re: [PATCH] laptop mode

[Dave Jones]

On Thu, Oct 09, 2003 at 04:11:43PM +0200, Jens Axboe wrote:

 > > > Red Hat just dropped this patch since it was suspected of data 
 > > > corruption ;-(
 > > 
 > > Uh, oh... Jens? 
 > 
 > See my previous mail. I don't see any problems with it, and I've
 > certainly not heard of (or experienced myself) problems with the patch.
 > I'm waiting for Jeff to expand on his mail, surely he/RH must know more
 > about this issue.


Laptop mode is likely fine, the problem we've seen is very likely caused
by the AAM counterpart.

		Dave

-- 
 Dave Jones     http://www.codemonkey.org.uk

Re: [PATCH] laptop mode

[Pavel Machek]

Hi!

> > Right now laptopmode/aam is just a suspect. There are still 1-2 other
> > small patches against IDE which could be the reason.  We've dropped
> > laptopmode/aam for the time being to see if the folks seeing repeatable
> > corruption suddenly start behaving again.
> 
> aam patch is far more risky, it's a far more likely suspect. That patch
> never reall did go out of beta. Dropping laptop-mode and aam at the same
> time is bad engineering practice :).
> 
> laptop-mode cannot cause corruption that cannot show otherwise.

Well, if fireballs have data-corrupting that happens just after spinup
(for example), I can see how laptop-mode affects that. W/o
laptop-mode, disk is spinning all the time so sleep bugs can not
manifest themselves. You add laptop-mode, and...
								Pavel
-- 
When do you have a heart between your knees?
[Johanka's followup: and *two* hearts?]

Re: [PATCH] laptop mode

[Jens Axboe]

On Sat, Oct 11 2003, Pavel Machek wrote:
> Hi!
> 
> > > Right now laptopmode/aam is just a suspect. There are still 1-2 other
> > > small patches against IDE which could be the reason.  We've dropped
> > > laptopmode/aam for the time being to see if the folks seeing repeatable
> > > corruption suddenly start behaving again.
> > 
> > aam patch is far more risky, it's a far more likely suspect. That patch
> > never reall did go out of beta. Dropping laptop-mode and aam at the same
> > time is bad engineering practice :).
> > 
> > laptop-mode cannot cause corruption that cannot show otherwise.
> 
> Well, if fireballs have data-corrupting that happens just after spinup
> (for example), I can see how laptop-mode affects that. W/o
> laptop-mode, disk is spinning all the time so sleep bugs can not
> manifest themselves. You add laptop-mode, and...

Not very likely, imho. People have been using spin down with hdparm for
years (in Linux and elsewhere), while acoustic management is a bit more
esoteric.

-- 
Jens Axboe

Re: [PATCH] laptop mode

[Barry K. Nathan]

On Sat, Oct 11, 2003 at 03:59:43PM +0200, Jens Axboe wrote:
> Not very likely, imho. People have been using spin down with hdparm for
> years (in Linux and elsewhere), while acoustic management is a bit more
> esoteric.

I'm having trouble finding this on Google now, but I've heard rumors
over the years of old Fireball drives corrupting data if they receive
write commands too soon after spinning up (i.e., the drive doesn't
bother waiting to spin up fully first). Maybe I'm not remembering the
details correctly, but it was something about the drive trying to act on
commands before it was fully spun up and malfunctioning as a result.

-Barry K. Nathan <barryn@pobox.com>

Re: [PATCH] laptop mode

[Jeff Garzik]

Barry K. Nathan wrote:
> On Sat, Oct 11, 2003 at 03:59:43PM +0200, Jens Axboe wrote:
> 
>>Not very likely, imho. People have been using spin down with hdparm for
>>years (in Linux and elsewhere), while acoustic management is a bit more
>>esoteric.
> 
> 
> I'm having trouble finding this on Google now, but I've heard rumors
> over the years of old Fireball drives corrupting data if they receive
> write commands too soon after spinning up (i.e., the drive doesn't
> bother waiting to spin up fully first). Maybe I'm not remembering the
> details correctly, but it was something about the drive trying to act on
> commands before it was fully spun up and malfunctioning as a result.


Well, there exists devices which comply with the delayed spin-up part of 
ATA/ATAPI specification...   maybe our code doesn't cover that.  I don't 
put much stock in rumors, but they are occasionally clues... :)

	Jeff

Re: [PATCH] laptop mode

[Jan Rychter]

[-- Attachment #1: Type: text/plain, Size: 2008 bytes --]

>>>>> "Jens" == Jens Axboe <axboe@suse.de> writes:
 Jens> On Thu, Oct 09 2003, Marcelo Tosatti wrote:
 >> On Thu, 9 Oct 2003, Jeff Garzik wrote:
 > Linux Kernel Mailing List wrote:
 > > ChangeSet 1.1150.1.52, 2003/10/08 10:49:45-03:00, axboe@suse.de
 > > 	[PATCH] laptop mode
 > >
 > > 	Hi Marcelo,
 > >
 > > 	Lots of people have been using this patch with great success,
 > > 	and it's been in the SuSE kernel for some months now too. It is
 > > 	also in the -benh ppc tree
 > >
 > > 	Basically, it introduces a write back mode of dirty and journal
 > > 	data that is more suitable for laptops. At the block layer end,
 > > 	it schedules write out of dirty data after the disk has been
 > > 	idle for 5 seconds.
 > >
 > > 	Laptop mode can be switched on and off with
 > > 	/proc/sys/vm/laptop_mode.  There is also a block_dump sysctl,
 > > 	which if enabled will dump who dirties and writes out data. This
 > > 	is very helpful in pinning down who is causing unnecessary
 > > 	writes to the disk.
 >
 > Red Hat just dropped this patch since it was suspected of data
 > corruption ;-(
 >>
 >> Uh, oh... Jens?

 Jens> See my previous mail. I don't see any problems with it, and I've
 Jens> certainly not heard of (or experienced myself) problems with the
 Jens> patch.  I'm waiting for Jeff to expand on his mail, surely he/RH
 Jens> must know more about this issue.

FWIW, I've been using laptop-mode for a long time now, with no problems
whatsoever. It's a great addition and really helps laptop users.

On a related note, a very desirable feature would be to be able to trace
which processes do I/O. Unfortunately, Linux distributions aren't very
well adapted to laptops and many applications are way too
"trigger-happy" when it comes to hard drive accesses. I guess most
developers don't use laptops and just assume that they can write or
read data at will. It is very difficult to trace the culprits when your
HD light goes on every couple of seconds.

--J.

[-- Attachment #2: Type: application/pgp-signature, Size: 188 bytes --]

Re: [PATCH] laptop mode

[Jens Axboe]

On Fri, Oct 17 2003, Jan Rychter wrote:
> >>>>> "Jens" == Jens Axboe <axboe@suse.de> writes:
>  Jens> On Thu, Oct 09 2003, Marcelo Tosatti wrote:
>  >> On Thu, 9 Oct 2003, Jeff Garzik wrote:
>  > Linux Kernel Mailing List wrote:
>  > > ChangeSet 1.1150.1.52, 2003/10/08 10:49:45-03:00, axboe@suse.de
>  > > 	[PATCH] laptop mode
>  > >
>  > > 	Hi Marcelo,
>  > >
>  > > 	Lots of people have been using this patch with great success,
>  > > 	and it's been in the SuSE kernel for some months now too. It is
>  > > 	also in the -benh ppc tree
>  > >
>  > > 	Basically, it introduces a write back mode of dirty and journal
>  > > 	data that is more suitable for laptops. At the block layer end,
>  > > 	it schedules write out of dirty data after the disk has been
>  > > 	idle for 5 seconds.
>  > >
>  > > 	Laptop mode can be switched on and off with
>  > > 	/proc/sys/vm/laptop_mode.  There is also a block_dump sysctl,
>  > > 	which if enabled will dump who dirties and writes out data. This
>  > > 	is very helpful in pinning down who is causing unnecessary
>  > > 	writes to the disk.
>  >
>  > Red Hat just dropped this patch since it was suspected of data
>  > corruption ;-(
>  >>
>  >> Uh, oh... Jens?
> 
>  Jens> See my previous mail. I don't see any problems with it, and I've
>  Jens> certainly not heard of (or experienced myself) problems with the
>  Jens> patch.  I'm waiting for Jeff to expand on his mail, surely he/RH
>  Jens> must know more about this issue.
> 
> FWIW, I've been using laptop-mode for a long time now, with no problems
> whatsoever. It's a great addition and really helps laptop users.

Glad to hear that.

> On a related note, a very desirable feature would be to be able to trace
> which processes do I/O. Unfortunately, Linux distributions aren't very
> well adapted to laptops and many applications are way too
> "trigger-happy" when it comes to hard drive accesses. I guess most
> developers don't use laptops and just assume that they can write or
> read data at will. It is very difficult to trace the culprits when your
> HD light goes on every couple of seconds.

You do know that the laptop-mode patch included a feature to help with
just that very problem, block_dump? If you do

echo "1" > /proc/sys/vm/block_dump

(shut down klogd first), it will tell you which process caused the hd
activity, either directly or by dirtying a page of data.

-- 
Jens Axboe