Skip to contentSkip to conversation

Current List of Kernel Summit suggested topics from the discuss list

DISCUSSION James Bottomley 10 participants 12 messages Jun 27, 2008 – Jul 3, 2008

lore
Share & tools
lore
#0 lore raw

Hi All,

This is the synopsis of the currently suggested topics so far. That's not to say this is the list we will be doing, just to say that if there's something you think we should be discussing and it's not on the list, now would be a good time to add it (don't trust the programme committee magically to add it at the last minute ...)

James

---

Meta Topics:

1. Do KS as an unconference - Matthew Wilcox
2. Do a Mini Hackfest for 50% of the time - Greg KH
3. Expand to three days and do Tech/Process/Tech on each day - Dave Miller

Agenda Topics:

1. Asynchronous Operations - Ulrich Drepper
* How do we make programs more parallel (driven by multicore)
* Can we use tasklets for this
2. Fixing the Kernel Janitor's Project - James Bottomley
* KJ Isn't a good intro to kernel hacking, need better ones like bug finding and fixing.
* Need to find more ways for non coders to make useful contributions
* Need to think more about what 'useful contributions' are
* What about the kernel tester's project [Adrian Bunk]
3. Moving firmware Blobs out of the Kernel - David Woodhouse
* remove all firmware and repackage in ways that would either be pulled in at runtime or could be linked into the kernel (distro choice)
* Need to be careful about licensing compatibility
* Where should we actually keep the extracted firmware
4. Barriers - Neil Brown
* Need to define what semantics filesystems actually want
* Need to relate this to what devices and subsystems can actually provide.
5. Tracking Regressions - Rafael Wysocki
* Describe experiences with the current running of the regression lists
* How could we make the current list and practice more useful
6. Discuss new Suspend/Resume Framework - Rafael Wysocki
* Discuss semantics of what drivers should be expecting and what best practices are.
* Need better documentation on this
* can we make subsystem libraries of helpers for their driver suspend/resume to ease the pain?
7. Hack/Fix session for Laptop Suspend and NOHZ/idle - Thomas Gleixner
* Perhaps this should be done at Plumbers instead
8. Content Accessed Filesystems - Lars Noschinski
* Filesystem that would store the same block only once
* Would be useful for git [Pavel Machek]
9. When should Drivers be Merged - James Bottomley
* Options seem to be ASAP, Wait for a bit for it to mature, when the vendor has fixed the obvious bugs
* Have a fast submission track for drivers for which the hardware documentation is available [Tony Luck]
* Run driver submissions early through staging trees while they get fixed
* How should these be run? Centrally or in subsystem trees
* Should staging trees be part of linux-next

#1 lore raw
Quoted 10 lines · James Bottomley

Hi All,

This is the synopsis of the currently suggested topics so far. That's not to say this is the list we will be doing, just to say that if there's something you think we should be discussing and it's not on the list, now would be a good time to add it (don't trust the programme committee magically to add it at the last minute ...)

James

I'd like to propose a topic about kernel quality; there's been a lot of lkml traffic about it (in ups and downs). As part of this topic I could present trends, data and conclusions from the kerneloops.org project. I'm sure Andrew has a set of data and views from his part of the world, and the regressions topic could fall under this too. If we, unlike last year, ask various subsystem maintainers to prepare data or at least something more solid than "eh I think we're fine" ahead of the conference, we could have a more thought out set of inputs from that direction as well.

#2 lore raw
Quoted 12 lines · Arjan van de Ven
Quoted 10 lines · James Bottomley

Hi All,

This is the synopsis of the currently suggested topics so far. That's not to say this is the list we will be doing, just to say that if there's something you think we should be discussing and it's not on the list, now would be a good time to add it (don't trust the programme committee magically to add it at the last minute ...)

James

Quoted 7 lines

I'd like to propose a topic about kernel quality; there's been a lot of lkml traffic about it (in ups and downs). As part of this topic I could present trends, data and conclusions from the kerneloops.org project. I'm sure Andrew has a set of data and views from his part of the world, and the regressions topic could fall under this too. If we, unlike last year, ask various subsystem maintainers to prepare data or at least something more solid than "eh I think we're fine" ahead of the conference, we could have a more thought out set of inputs from that direction as well.

Really, then you want this topic:

Quoted 5 lines

 5. Tracking Regressions - Rafael Wysocki
* Describe experiences with the current running of the regression lists
* How could we make the current list and practice more useful

broadened to include all quality issues?

James

#3 lore raw
Quoted 24 lines · James Bottomley
Quoted 12 lines · Arjan van de Ven
Quoted 10 lines · James Bottomley

Hi All,

This is the synopsis of the currently suggested topics so far. That's not to say this is the list we will be doing, just to say that if there's something you think we should be discussing and it's not on the list, now would be a good time to add it (don't trust the programme committee magically to add it at the last minute ...)

James

Quoted 7 lines

I'd like to propose a topic about kernel quality; there's been a lot of lkml traffic about it (in ups and downs). As part of this topic I could present trends, data and conclusions from the kerneloops.org project. I'm sure Andrew has a set of data and views from his part of the world, and the regressions topic could fall under this too. If we, unlike last year, ask various subsystem maintainers to prepare data or at least something more solid than "eh I think we're fine" ahead of the conference, we could have a more thought out set of inputs from that direction as well.

Really, then you want this topic:

Quoted 7 lines
Quoted 5 lines

 5. Tracking Regressions - Rafael Wysocki
* Describe experiences with the current running of the regression lists
* How could we make the current list and practice more useful

broadened to include all quality issues?

That would be fine by me.

Thanks,
Rafael

#4 lore raw
In reply to Arjan van de Ven
Quoted 11 lines · Arjan van de Ven
Quoted 9 lines · James Bottomley

Hi All,

This is the synopsis of the currently suggested topics so far. That's not to say this is the list we will be doing, just to say that if there's something you think we should be discussing and it's not on the list, now would be a good time to add it (don't trust the programme committee magically to add it at the last minute ...)

James

Quoted 8 lines

I'd like to propose a topic about kernel quality; there's been a lot of lkml traffic about it (in ups and downs). As part of this topic I could present trends, data and conclusions from the kerneloops.org project. I'm sure Andrew has a set of data and views from his part of the world, and the regressions topic could fall under this too. If we, unlike last year, ask various subsystem maintainers to prepare data or at least something more solid than "eh I think we're fine" ahead of the conference, we could have a more thought out set of inputs from that direction as well.

I think this could be an interesting topic, as long as we have data like what you collect at kerneloops.org. Unfortunately, data for other stuff is harder to come by.

Another question this raises is how we define "quality". Is it simply having a low bug count? Or just having few bugs that affect large numbers of people? Given that we'll always have bugs though, we could also measure quality in terms of how effectively we handle bugs, e.g. how quickly fixes are proposed and integrated, how well we work with distros to incorporate fixes for their high priority issues, etc.

If we measure quality using bug metrics, things are pretty hard. We have kerneloops.org which tracks one class if issues, but there are so few bugs filed at kernel.org, relative to other projects that use bug tracking heavily, that for PCI at least I could only paint an anecdotal picture of things (though we do have one fairly clear area of weakness that I've seen based on my triaging). Contrast this to the Intel graphics projects at freedesktop.org where we have a fairly large bug history and can say pretty convincingly that things are improving a lot, and fairly rapidly.

I guess I'm saying this *could* be an interesting topic, but it could also be the same discussion we've had for years, which never even reaches a conclusion about whether we're improving or not.

Jesse

#5 lore raw
Quoted 1 line · Jesse Barnes

If we measure quality using bug metrics, things are pretty hard.

One problem with any metric is that the metric becomes a driving factor in itself. Consider the whole whitespace issue, for example.

As far as the whitespace issue is concerned, we may want to consider just setting a flag day -- probably immediately before an -rc1 release -- and just run "cleanfile" on the whole tree. In-flux patches are still appliable with "patch -l", and by now we should have enough tools to avoid adding new whitespace problems, so we can just kill this issue once and for all.

-hpa

#6 lore raw
Quoted 5 lines · H. Peter Anvin
Quoted 1 line · Jesse Barnes

If we measure quality using bug metrics, things are pretty hard.

One problem with any metric is that the metric becomes a driving factor in itself. Consider the whole whitespace issue, for example.

Sure, but I don't think that's a reason to avoid metrics altogether. We've already seen that purely qualitative discussions don't really get us anywhere. For any discussion about kernel quality I think we have to:
a) define our goals (no oopses? fast bug fix turnaround? whatever)
b) define a way to measure progress against those goals
c) periodically re-evaluate both

I think all of these are fairly difficult tasks, and any goal or metric we create will have problems, but does that mean we should just ignore quality? Or limit ourselves to our current situation where everyone has a different idea of what it means and whether we're achieving it?

Jesse

#7 lore raw
In reply to James Bottomley
Quoted 3 lines · James Bottomley

8. Content Accessed Filesystems - Lars Noschinski
* Filesystem that would store the same block only once
* Would be useful for git [Pavel Machek]

This is an intersting research topic, but nothing that needs to be discussed at KS. It's a pure filesystem implementation issue, and not a kernel-wide dsicussion item.

#8 lore raw
In reply to James Bottomley
Quoted 4 lines · James Bottomley

4. Barriers - Neil Brown
* Need to define what semantics filesystems actually want
* Need to relate this to what devices and subsystems can actually provide.

Again? Barriers must be the most persistent topic at the kernel summit, yet nothing has happened since my initial implementation back in 2001 or so! :-)

This being a fairly wide and technical topic, I think it's actually better suited for mailing list discussion for starters.

Signature · 1 line
Jens Axboe
#9 lore raw
In reply to James Bottomley
Quoted 6 lines · James Bottomley

Agenda Topics:

1. Asynchronous Operations - Ulrich Drepper
* How do we make programs more parallel (driven by multicore)
* Can we use tasklets for this

I suppose you mean "threadlets" or "syslets" here, not "tasklets", which are a kernel internal concept, right?

	Arnd <><
#10 lore raw
In reply to James Bottomley
Quoted 7 lines · James Bottomley

Hi All,

This is the synopsis of the currently suggested topics so far. That's not to say this is the list we will be doing, just to say that if there's something you think we should be discussing and it's not on the list, now would be a good time to add it (don't trust the programme committee magically to add it at the last minute ...)

Another proposal:

Working well with other subsystems - Jean Delvare

There are a number of times where different subsystems in the kernel need to interact with each other in ways that are not always easy or obvious. This talk will be about some examples of how this has been done well, as well as work to figure out how to handle this better in the future.

thanks,

greg k-h

#11 lore raw
Quoted 10 lines · On Thu, 3 Jul 2008 11:27:07 -0700 Greg KH <[email protected]> wrote:

Another proposal:

Working well with other subsystems - Jean Delvare

There are a number of times where different subsystems in the kernel need to interact with each other in ways that are not always easy or obvious. This talk will be about some examples of how this has been done well, as well as work to figure out how to handle this better in the future.

Hear! Hear! :-)

This could be expanded into a general discussion of how to handle API change in our distributed environment.

Signature · 3 lines
Cheers,
Stephen Rothwell                    [email protected]
http://www.canb.auug.org.au/~sfr/