Dave Jones wrote:
> On Tue, Jan 30, 2007 at 01:06:17AM +0100, Jes Sorensen wrote:
> > Then there is the issue of architectures, at least in my book KS should
> > focus on the ones that are really live and not in maintenance mode.
> > x86_64, x86_32, PPC, ia64, ARM seems to be the driving ones these days,
> > m68k, Sparc32, and others, somewhat less so .....
>
> Again, I don't recall us spending any time at all discussing m68k, or
> sparc, whilst the others you mention were well represented.
Well, others where represented, I was there looking after non-mmu m68k
for example (and other general non-mmu stuff). There just wasn't much
contentious stuff in that space that needed wider discussion.
Regards
Greg
------------------------------------------------------------------------
Greg Ungerer -- Chief Software Dude EMAIL: [email protected]
SnapGear -- a Secure Computing Company PHONE: +61 7 3435 2888
825 Stanley St, FAX: +61 7 3891 3630
Woolloongabba, QLD, 4102, Australia WEB: http://www.SnapGear.com
On Tue, Jan 30, 2007 at 01:30:56PM +1000, Greg Ungerer wrote:
>
> Dave Jones wrote:
> > On Tue, Jan 30, 2007 at 01:06:17AM +0100, Jes Sorensen wrote:
> > > Then there is the issue of architectures, at least in my book KS should
> > > focus on the ones that are really live and not in maintenance mode.
> > > x86_64, x86_32, PPC, ia64, ARM seems to be the driving ones these days,
> > > m68k, Sparc32, and others, somewhat less so .....
> >
> > Again, I don't recall us spending any time at all discussing m68k, or
> > sparc, whilst the others you mention were well represented.
>
> Well, others where represented, I was there looking after non-mmu m68k
> for example (and other general non-mmu stuff). There just wasn't much
> contentious stuff in that space that needed wider discussion.
Right, other than during the CPU architects panel, I don't remember
any non x86/ia64/ppc stuff being brought up at all.
Dave
--
http://www.codemonkey.org.uk
Dave Jones wrote:
> On Tue, Jan 30, 2007 at 01:30:56PM +1000, Greg Ungerer wrote:
> >
> > Dave Jones wrote:
> > > On Tue, Jan 30, 2007 at 01:06:17AM +0100, Jes Sorensen wrote:
> > > > Then there is the issue of architectures, at least in my book KS should
> > > > focus on the ones that are really live and not in maintenance mode.
> > > > x86_64, x86_32, PPC, ia64, ARM seems to be the driving ones these days,
> > > > m68k, Sparc32, and others, somewhat less so .....
> > >
> > > Again, I don't recall us spending any time at all discussing m68k, or
> > > sparc, whilst the others you mention were well represented.
> >
> > Well, others where represented, I was there looking after non-mmu m68k
> > for example (and other general non-mmu stuff). There just wasn't much
> > contentious stuff in that space that needed wider discussion.
>
> Right, other than during the CPU architects panel, I don't remember
> any non x86/ia64/ppc stuff being brought up at all.
Yep. IIRC the CPU architects panel was all x86/x86_64/ppc too wasn't it?
------------------------------------------------------------------------
Greg Ungerer -- Chief Software Dude EMAIL: [email protected]
SnapGear -- a Secure Computing Company PHONE: +61 7 3435 2888
825 Stanley St, FAX: +61 7 3891 3630
Woolloongabba, QLD, 4102, Australia WEB: http://www.SnapGear.com
Dave Jones wrote:
> On Tue, Jan 30, 2007 at 01:30:56PM +1000, Greg Ungerer wrote:
> >
> > Dave Jones wrote:
> > > On Tue, Jan 30, 2007 at 01:06:17AM +0100, Jes Sorensen wrote:
> > > > Then there is the issue of architectures, at least in my book KS should
> > > > focus on the ones that are really live and not in maintenance mode.
> > > > x86_64, x86_32, PPC, ia64, ARM seems to be the driving ones these days,
> > > > m68k, Sparc32, and others, somewhat less so .....
> > >
> > > Again, I don't recall us spending any time at all discussing m68k, or
> > > sparc, whilst the others you mention were well represented.
> >
> > Well, others where represented, I was there looking after non-mmu m68k
> > for example (and other general non-mmu stuff). There just wasn't much
> > contentious stuff in that space that needed wider discussion.
>
> Right, other than during the CPU architects panel, I don't remember
> any non x86/ia64/ppc stuff being brought up at all.
Yep. IIRC the CPU architects panel was all x86/x86_64/ppc too wasn't it?
------------------------------------------------------------------------
Greg Ungerer -- Chief Software Dude EMAIL: [email protected]
SnapGear -- a Secure Computing Company PHONE: +61 7 3435 2888
825 Stanley St, FAX: +61 7 3891 3630
Woolloongabba, QLD, 4102, Australia WEB: http://www.SnapGear.com
On Tue, Jan 30, 2007 at 02:01:07PM +1000, Greg Ungerer wrote:
> Dave Jones wrote:
> >Right, other than during the CPU architects panel, I don't remember
> >any non x86/ia64/ppc stuff being brought up at all.
>
> Yep. IIRC the CPU architects panel was all x86/x86_64/ppc too wasn't it?
>
Similarly, it would be nice if we could avoid marketing oriented CPU
roadmap presentations this year, before any other vendors start
getting ideas..
On Tue, Jan 30, 2007 at 02:01:07PM +1000, Greg Ungerer wrote:
> > > > Again, I don't recall us spending any time at all discussing m68k, or
> > > > sparc, whilst the others you mention were well represented.
> > >
> > > Well, others where represented, I was there looking after non-mmu m68k
> > > for example (and other general non-mmu stuff). There just wasn't much
> > > contentious stuff in that space that needed wider discussion.
> >
> > Right, other than during the CPU architects panel, I don't remember
> > any non x86/ia64/ppc stuff being brought up at all.
>
> Yep. IIRC the CPU architects panel was all x86/x86_64/ppc too wasn't it?
I thought there was coldfire mentioned too, or maybe my memory is
playing tricks on me. Maybe I'm misremembering the ppc bit.
Dave
--
http://www.codemonkey.org.uk
Dave Jones wrote:
> On Tue, Jan 30, 2007 at 02:01:07PM +1000, Greg Ungerer wrote:
>
> > > > > Again, I don't recall us spending any time at all discussing m68k, or
> > > > > sparc, whilst the others you mention were well represented.
> > > >
> > > > Well, others where represented, I was there looking after non-mmu m68k
> > > > for example (and other general non-mmu stuff). There just wasn't much
> > > > contentious stuff in that space that needed wider discussion.
> > >
> > > Right, other than during the CPU architects panel, I don't remember
> > > any non x86/ia64/ppc stuff being brought up at all.
> >
> > Yep. IIRC the CPU architects panel was all x86/x86_64/ppc too wasn't it?
>
> I thought there was coldfire mentioned too, or maybe my memory is
> playing tricks on me. Maybe I'm misremembering the ppc bit.
Your right, the person from Freescale did mention it.
------------------------------------------------------------------------
Greg Ungerer -- Chief Software Dude EMAIL: [email protected]
SnapGear -- a Secure Computing Company PHONE: +61 7 3435 2888
825 Stanley St, FAX: +61 7 3891 3630
Woolloongabba, QLD, 4102, Australia WEB: http://www.SnapGear.com
On 1/29/07 8:10 PM, "Dave Jones" <[email protected]> wrote:
>
>>>>> Again, I don't recall us spending any time at all discussing m68k, or
>>>>> sparc, whilst the others you mention were well represented.
>>>>
>>>> Well, others where represented, I was there looking after non-mmu m68k
>>>> for example (and other general non-mmu stuff). There just wasn't much
>>>> contentious stuff in that space that needed wider discussion.
>>>
>>> Right, other than during the CPU architects panel, I don't remember
>>> any non x86/ia64/ppc stuff being brought up at all.
>>
>> Yep. IIRC the CPU architects panel was all x86/x86_64/ppc too wasn't it?
>
> I thought there was coldfire mentioned too, or maybe my memory is
> playing tricks on me. Maybe I'm misremembering the ppc bit.
I believe we had AMD, Freescale and Intel last year, no PPC.
/D
On 1/29/07 8:22 PM, "Greg Ungerer" <[email protected]> wrote:
>>> Yep. IIRC the CPU architects panel was all x86/x86_64/ppc too wasn't it?
>>
>> I thought there was coldfire mentioned too, or maybe my memory is
>> playing tricks on me. Maybe I'm misremembering the ppc bit.
>
> Your right, the person from Freescale did mention it.
Oops yeah. Hit send to early. The presenter from Freescale spoke about PPC
as well.
/D
On Tue, Jan 30, 2007 at 01:08:26PM +0900, Paul Mundt wrote:
> On Tue, Jan 30, 2007 at 02:01:07PM +1000, Greg Ungerer wrote:
> > Dave Jones wrote:
> > >Right, other than during the CPU architects panel, I don't remember
> > >any non x86/ia64/ppc stuff being brought up at all.
> >
> > Yep. IIRC the CPU architects panel was all x86/x86_64/ppc too wasn't it?
> >
> Similarly, it would be nice if we could avoid marketing oriented CPU
> roadmap presentations this year, before any other vendors start
> getting ideas..
It might be worth putting together a list of do's and don'ts for the
CPU architects if we have a panel again this year (and its usually
a fairly popular session, so I'd be surprised if it got dropped).
something along the lines of
Do:
- Detail what new features next-gen cpu out in Q1'08 will have
that we may need to care about
- Ask for input on what features _we_ would like in CPUs in Q1 2010
- Tell us how we're taking advantage of things before that other OS,
it makes us happy ;)
Don't:
- Waffle about process shrink roadmaps.
Whilst it's good to hear, it doesn't affect the code we write.
- Moves to new substrates are fascinating to CPU manufacturers,
not so much for kernel engineers.
- If something can't be discussed other than under NDA, don't
bother bringing it up. Those interested will likely find
out about it through their employers anyway.
This came up last year with a number of "we can't tell you"
reponses, which made one presentation almost worthless.
or perhaps it was particularly "we can't tell hch" :-)
Any others?
Dave
--
http://www.codemonkey.org.uk
Greg Ungerer wrote:
> Dave Jones wrote:
>> Again, I don't recall us spending any time at all discussing m68k, or
>> sparc, whilst the others you mention were well represented.
>
> Well, others where represented, I was there looking after non-mmu m68k
> for example (and other general non-mmu stuff). There just wasn't much
> contentious stuff in that space that needed wider discussion.
Which is good, provided that non-mmu work is actively driving some of
the decisions in mainline. Given that I don't follow the non-mmu work
at all, I don't have a feel for whether that is the case, but I could
imagine that it would have some impact that needs to be taken into
account at times. My worry is for spending time and slots on things /
people dealing with classic architectures which are no longer being
manufactured and are only being maintained in catch-up mode.
Cheers,
Jes
On Tuesday 30 January 2007 04:41, Dave Jones wrote:
> Right, other than during the CPU architects panel, I don't remember
> any non x86/ia64/ppc stuff being brought up at all.
No IA64 stuff that I can remember. And there was a presentation on PPC.
But that was planned to be differently with more focus on embedded,
unfortunately the comittee didn't manage to find more embedded CPU
people in time.
My personal preference would be to go for a chipset panel this year
instead. Chipsets seem to impact kernels much more than CPUs.
-Andi
On Tue, Jan 30, 2007 at 06:11:18AM +0100, Andi Kleen wrote:
> On Tuesday 30 January 2007 04:41, Dave Jones wrote:
>
> > Right, other than during the CPU architects panel, I don't remember
> > any non x86/ia64/ppc stuff being brought up at all.
>
> No IA64 stuff that I can remember. And there was a presentation on PPC.
>
> But that was planned to be differently with more focus on embedded,
> unfortunately the comittee didn't manage to find more embedded CPU
> people in time.
given we barely had enough time for freescale, perhaps that was
for the best.
> My personal preference would be to go for a chipset panel this year
> instead. Chipsets seem to impact kernels much more than CPUs.
That could be interesting. I wonder if its worth doing both ?
Depends if enough people are bored with CPU panels I guess :)
Dave
--
http://www.codemonkey.org.uk
On Tue, Jan 30, 2007 at 06:11:18AM +0100, Andi Kleen wrote:
> On Tuesday 30 January 2007 04:41, Dave Jones wrote:
> > Right, other than during the CPU architects panel, I don't remember
> > any non x86/ia64/ppc stuff being brought up at all.
>
> No IA64 stuff that I can remember. And there was a presentation on PPC.
>
> But that was planned to be differently with more focus on embedded,
> unfortunately the comittee didn't manage to find more embedded CPU
> people in time.
>
It might be interesting to have a more condensed CPU panel, something
like the OLS lightning talks. Gather a larger number of vendors, and give
each a small window to bring up the most relevant issues for them in the
future, while also allowing for some feedback and Q&A.
It would be nice to get more input from CPU architects, but only if it's
possible to keep it entirely technical and moving along without anyone
having to hurry their window of time due to someone else overstepping
theirs. If they only have a small window to present their concerns, I
think we'll see a lot of the fluff (as itemized by davej) go away.
As soon as a vendor starts rambling on about value-added IP blocks, we've
already lost..
On Tuesday 30 January 2007 05:24, Dirk Hohndel wrote:
>
> > I thought there was coldfire mentioned too, or maybe my memory is
> > playing tricks on me. ?Maybe I'm misremembering the ppc bit.
>
> I believe we had AMD, Freescale and Intel last year, no PPC.
Freescale's bigger CPUs are all PPC, although they also have stuff
like coldfire, which is not.
Arnd <><
On Mon, Jan 29, 2007 at 11:34:21PM -0500, Dave Jones wrote:
> It might be worth putting together a list of do's and don'ts for the
> CPU architects if we have a panel again this year (and its usually
> a fairly popular session, so I'd be surprised if it got dropped).
> something along the lines of
Count my vote for dropping the cpu panels session. It's been far
too marketing oriented, and all of the companies have far more interesting
meetings of their own where thos caring about a particular architecture
(and that includes much more than just the cpu!) can have usefull discussions.
> Don't:
> - Waffle about process shrink roadmaps.
Buy a graphics company, continue blocking 2D support and expect anyone to
even care about your hardware ... ?
On Tue, 2007-01-30 at 10:30 +0000, Christoph Hellwig wrote:
> On Mon, Jan 29, 2007 at 11:34:21PM -0500, Dave Jones wrote:
> > It might be worth putting together a list of do's and don'ts for the
> > CPU architects if we have a panel again this year (and its usually
> > a fairly popular session, so I'd be surprised if it got dropped).
> > something along the lines of
>
> Count my vote for dropping the cpu panels session. It's been far
> too marketing oriented, and all of the companies have far more interesting
> meetings of their own where thos caring about a particular architecture
> (and that includes much more than just the cpu!) can have usefull discussions.
Well, OK, but the next question is that is some form of panel of
outsiders still a useful feature?
Previous panels we've done have been:
* Device Drivers - Inputs from vendors trying to get code into the
kernel. I had feedback that this was reasonably useful; the
problem is that it tends to be composed of vendors already
making a big effort on the open source process and not the ones
(like graphics) who aren't.
* Customer Panel - inputs from various users deploying linux in
their enterprises. This did tend to degenerate quickly to a
list of requirements.
The one everyone seems to want is chipsets, so is this the one we want
to shoot for this year?
James
On Tue, 30 Jan 2007 10:48:45 -0600 James Bottomley wrote:
> On Tue, 2007-01-30 at 10:30 +0000, Christoph Hellwig wrote:
> > On Mon, Jan 29, 2007 at 11:34:21PM -0500, Dave Jones wrote:
> > > It might be worth putting together a list of do's and don'ts for the
> > > CPU architects if we have a panel again this year (and its usually
> > > a fairly popular session, so I'd be surprised if it got dropped).
> > > something along the lines of
> >
> > Count my vote for dropping the cpu panels session. It's been far
> > too marketing oriented, and all of the companies have far more interesting
> > meetings of their own where thos caring about a particular architecture
> > (and that includes much more than just the cpu!) can have usefull discussions.
>
> Well, OK, but the next question is that is some form of panel of
> outsiders still a useful feature?
>
> Previous panels we've done have been:
>
> * Device Drivers - Inputs from vendors trying to get code into the
> kernel. I had feedback that this was reasonably useful; the
> problem is that it tends to be composed of vendors already
> making a big effort on the open source process and not the ones
> (like graphics) who aren't.
> * Customer Panel - inputs from various users deploying linux in
> their enterprises. This did tend to degenerate quickly to a
> list of requirements.
>
> The one everyone seems to want is chipsets, so is this the one we want
> to shoot for this year?
As usual, "it depends" on the content. Can we provide them with
sufficient instructions/guidance so that the listeners get the content
that is desired instead of just some pseudo-marketing or requirements
list? Any of those panels (Customer or CPU) could have been good or bad.
---
~Randy
On Tue, 2007-01-30 at 08:53 -0800, Randy Dunlap wrote:
> As usual, "it depends" on the content. Can we provide them with
> sufficient instructions/guidance so that the listeners get the content
> that is desired instead of just some pseudo-marketing or requirements
> list? Any of those panels (Customer or CPU) could have been good or
> bad.
This is a really nasty problem. By and large, only organisations who
are active participants in the Linux community are happy sending their
technical architects ungaurded to a developer summit (like we get for
the CPU panel). The objective is always to get technical (not
marketing) people who haven't been frightened into silence by their
legal department and, if people want chipsets, that's what we'll try to
do ... it just takes a lot of persuasion, so the earlier we start ...
James
Jes Sorensen wrote:
> Greg Ungerer wrote:
>> Dave Jones wrote:
>>> Again, I don't recall us spending any time at all discussing m68k, or
>>> sparc, whilst the others you mention were well represented.
>>
>> Well, others where represented, I was there looking after non-mmu m68k
>> for example (and other general non-mmu stuff). There just wasn't much
>> contentious stuff in that space that needed wider discussion.
>
> Which is good, provided that non-mmu work is actively driving some of
> the decisions in mainline. Given that I don't follow the non-mmu work
> at all, I don't have a feel for whether that is the case, but I could
> imagine that it would have some impact that needs to be taken into
> account at times.
Exactly.
> My worry is for spending time and slots on things /
> people dealing with classic architectures which are no longer being
> manufactured and are only being maintained in catch-up mode.
Agreed, but I haven't seen that in the past. The CPU panel at least
has always covered current (and future) hardware.
Regards
Greg
On Tue, Jan 30, 2007 at 10:48:45AM -0600, James Bottomley wrote:
> Well, OK, but the next question is that is some form of panel of
> outsiders still a useful feature?
>
> Previous panels we've done have been:
>
> * Device Drivers - Inputs from vendors trying to get code into the
> kernel. I had feedback that this was reasonably useful; the
> problem is that it tends to be composed of vendors already
> making a big effort on the open source process and not the ones
> (like graphics) who aren't.
> * Customer Panel - inputs from various users deploying linux in
> their enterprises. This did tend to degenerate quickly to a
> list of requirements.
>
> The one everyone seems to want is chipsets, so is this the one we want
> to shoot for this year?
chipsets is probably more interesting than cpus, yes. Most useful would
be other open source projects and their requirements/wishes from the kernel,
but we're already discussing that elsewhere in this maze of threads..