Hot kernel change

Hot kernel change

[Carlos Silva]

hi,

i would like to know if isn't it possible to implement a hot kernel
change, i mean, without reboot. i would do it myself if i had the knoledge
to do it but i'm starting with kernel-level programing now. i think it
would be possible if we make something like M$'s OS do when it hibernates,
copy all the memory, registers, etc to the disc and then put all back
again.

am i dreaming or this is possible? :)

Greetings,

Carlos "r3pek" Silva

Re: Hot kernel change

[Richard B. Johnson]

On Thu, 19 Feb 2004, Carlos Silva wrote:

> hi,
>
> i would like to know if isn't it possible to implement a hot kernel
> change, i mean, without reboot. i would do it myself if i had the knoledge
> to do it but i'm starting with kernel-level programing now. i think it
> would be possible if we make something like M$'s OS do when it hibernates,
> copy all the memory, registers, etc to the disc and then put all back
> again.
>
> am i dreaming or this is possible? :)
>
> Greetings,
>
> Carlos "r3pek" Silva

Sure it's possible. However, you can't change to a new kernel this
way because the addresses of the hardware devices like the PCI bus
devices may change with a new kernel. Since the displacements of
the kernel's internal workings will change with a new version, there
would need to be considerable work done in re-designing the kernel
so that it wouldn't matter.

The best you can do, right now, is reload the same kernel. It
will take about as much time as a reboot, so why bother? Oh, you
intend to keep the same processes running, do you? You expect to
be writing a letter in X-windows and hit the reset switch, magically
returning to the same state after the machine has a new kernel
installed? Well, well-written software is indistinguishable from
magic, but first you need to find out how to make time run back-
wards because, at the very least, the new kernel will be installed
in the future which means many things will have changed (like
network IP addresses, dynamic routes, remote file discriptors, etc.)
Sounds like a neat project for a College Student who wants to
learn to solve problems they haven't even dreamed of yet.

Cheers,
Dick Johnson
Penguin : Linux version 2.4.24 on an Intel Pentium III machine (797.90 BogoMips).
            Note 96.31% of all statistics are fiction.

Re: Hot kernel change

[Jim Richardson]

On Thu, Feb 19, 2004 at 09:05:25AM -0500, Richard B. Johnson wrote:
>On Thu, 19 Feb 2004, Carlos Silva wrote:
>
>> hi,
>>
>> i would like to know if isn't it possible to implement a hot kernel
>> change, i mean, without reboot. i would do it myself if i had the knoledge
>> to do it but i'm starting with kernel-level programing now. i think it
>> would be possible if we make something like M$'s OS do when it hibernates,
>> copy all the memory, registers, etc to the disc and then put all back
>> again.
>>
>> am i dreaming or this is possible? :)
>>
>> Greetings,
>>
>> Carlos "r3pek" Silva
>
>Sure it's possible. However, you can't change to a new kernel this
>way because the addresses of the hardware devices like the PCI bus
>devices may change with a new kernel. Since the displacements of
>the kernel's internal workings will change with a new version, there
>would need to be considerable work done in re-designing the kernel
>so that it wouldn't matter.
>
>The best you can do, right now, is reload the same kernel. It
>will take about as much time as a reboot, so why bother? Oh, you
>intend to keep the same processes running, do you? You expect to
>be writing a letter in X-windows and hit the reset switch, magically
>returning to the same state after the machine has a new kernel
>installed? Well, well-written software is indistinguishable from
>magic, but first you need to find out how to make time run back-
>wards because, at the very least, the new kernel will be installed
>in the future which means many things will have changed (like
>network IP addresses, dynamic routes, remote file discriptors, etc.)
>Sounds like a neat project for a College Student who wants to
>learn to solve problems they haven't even dreamed of yet.
>


What you could do, is use MkLinux, (is that still around?) It had the
ability to run simultaneous kernels, IIRC, then you might be able to
gradually push over new processess to the new kernel, and eventually,
kill the old one. It's been quite a while since I used MkLinux, on PPC
for what it matters, and I could be remembering wrong...


-- 
Jim Richardson     http://www.eskimo.com/~warlock
"Microsoft Wheel.  Now with 8 sides for a smoother ride."
	-- The Ghost in the Machine, in COLA

Re: Hot kernel change

[Carlos Silva]

> On Thu, Feb 19, 2004 at 09:05:25AM -0500, Richard B. Johnson wrote:
>>On Thu, 19 Feb 2004, Carlos Silva wrote:
>>
>>> hi,
>>>
>>> i would like to know if isn't it possible to implement a hot kernel
>>> change, i mean, without reboot. i would do it myself if i had the
>>> knoledge
>>> to do it but i'm starting with kernel-level programing now. i think it
>>> would be possible if we make something like M$'s OS do when it
>>> hibernates,
>>> copy all the memory, registers, etc to the disc and then put all back
>>> again.
>>>
>>> am i dreaming or this is possible? :)
>>>
>>> Greetings,
>>>
>>> Carlos "r3pek" Silva
>>
>>Sure it's possible. However, you can't change to a new kernel this
>>way because the addresses of the hardware devices like the PCI bus
>>devices may change with a new kernel. Since the displacements of
>>the kernel's internal workings will change with a new version, there
>>would need to be considerable work done in re-designing the kernel
>>so that it wouldn't matter.
>>
>>The best you can do, right now, is reload the same kernel. It
>>will take about as much time as a reboot, so why bother? Oh, you
>>intend to keep the same processes running, do you? You expect to
>>be writing a letter in X-windows and hit the reset switch, magically
>>returning to the same state after the machine has a new kernel
>>installed? Well, well-written software is indistinguishable from
>>magic, but first you need to find out how to make time run back-
>>wards because, at the very least, the new kernel will be installed
>>in the future which means many things will have changed (like
>>network IP addresses, dynamic routes, remote file discriptors, etc.)
>>Sounds like a neat project for a College Student who wants to
>>learn to solve problems they haven't even dreamed of yet.
>>
>
>
> What you could do, is use MkLinux, (is that still around?) It had the
> ability to run simultaneous kernels, IIRC, then you might be able to
> gradually push over new processess to the new kernel, and eventually,
> kill the old one. It's been quite a while since I used MkLinux, on PPC
> for what it matters, and I could be remembering wrong...
>

well, that's what i had in mind... if this could be possible under x86
would be great. i say x86 'cause for what a read, MkLinux looks like a
MacLinux Distribution, correct me if i am wrong.
like i said in the first place, i don't program for the kernel (yet, i
intend to), so i don't know what are the big/small changes that have to be
made for somthing like this to work. but i would really like to see this
working :D

Re: Hot kernel change

[Randy.Dunlap]

On Thu, 19 Feb 2004 13:33:43 -0000 (WET) "Carlos Silva" <r3pek@r3pek.homelinux.org> wrote:

| hi,
| 
| i would like to know if isn't it possible to implement a hot kernel
| change, i mean, without reboot. i would do it myself if i had the knoledge
| to do it but i'm starting with kernel-level programing now. i think it
| would be possible if we make something like M$'s OS do when it hibernates,
| copy all the memory, registers, etc to the disc and then put all back
| again.
| 
| am i dreaming or this is possible? :)

The kexec patch is basically "linux reboots linux".
It bypasses the firmware/BIOS to do the reboot.

Patches for 2.6.0 and 2.6.1 are here (I haven't updated for
2.6.2 or 2.6.3 yet):
  http://developer.osdl.org/rddunlap/kexec/

Patches for some 2.5.x kernels are here:
  http://www.xmission.com/~ebiederm/files/kexec/

kexec does reduce reboot time quite a bit on some machines, but
there is still a noticeable pause.

--
~Randy

Re: Hot kernel change

[Carlos Silva]

> On Thu, 19 Feb 2004 13:33:43 -0000 (WET) "Carlos Silva"
> <r3pek@r3pek.homelinux.org> wrote:
>
> | hi,
> |
> | i would like to know if isn't it possible to implement a hot kernel
> | change, i mean, without reboot. i would do it myself if i had the
> knoledge
> | to do it but i'm starting with kernel-level programing now. i think it
> | would be possible if we make something like M$'s OS do when it
> hibernates,
> | copy all the memory, registers, etc to the disc and then put all back
> | again.
> |
> | am i dreaming or this is possible? :)
>
> The kexec patch is basically "linux reboots linux".
> It bypasses the firmware/BIOS to do the reboot.
>
> Patches for 2.6.0 and 2.6.1 are here (I haven't updated for
> 2.6.2 or 2.6.3 yet):
>   http://developer.osdl.org/rddunlap/kexec/
>
> Patches for some 2.5.x kernels are here:
>   http://www.xmission.com/~ebiederm/files/kexec/
>
> kexec does reduce reboot time quite a bit on some machines, but
> there is still a noticeable pause.
>

well, what can i say? this is nice, really nice :) but still reboots the
machine... :D
but it's a thing that i will test when a patch for the 2.6.3 kernel gets out.

Re: Hot kernel change

[Dominik Kubla]

On Thursday 19 February 2004 22:32, Carlos Silva wrote:

> well, what can i say? this is nice, really nice :) but still reboots the
> machine... :D
> but it's a thing that i will test when a patch for the 2.6.3 kernel gets
> out. -

The only way to avoid a system reboot on a kernel upgrade that i know of is a 
VMS cluster... But I guess that bit of information doesn't help you much. ;-)

Regards,
  Dominik
-- 
	   THE DAILY PLANET

	SUPERMAN SAVES DESSERT!
	Plans to "Eat it later"

Re: Hot kernel change

[Jim Richardson]

On Thu, Feb 19, 2004 at 08:25:43PM -0000, Carlos Silva wrote:

<snip>


>well, that's what i had in mind... if this could be possible under x86
>would be great. i say x86 'cause for what a read, MkLinux looks like a
>MacLinux Distribution, correct me if i am wrong.
>like i said in the first place, i don't program for the kernel (yet, i
>intend to), so i don't know what are the big/small changes that have to be
>made for somthing like this to work. but i would really like to see this
>working :D

MkLinux was available for x86, but I have no idea if it is still in
development. To be clear, it doesn't allow you to simply replace a
kernel, but to add a second one, and possibly, to start transferring
over tasks to it. 

You can do much the same thing with user mode linux also. Again, not a
kernel replacement in that sense, but something similar, sort of. 


-- 
Jim Richardson     http://www.eskimo.com/~warlock
RFC 882 put the dot in .com.

Re: Hot kernel change

[Nigel Rantor]

Richard B. Johnson wrote:
> Sounds like a neat project for a College Student who wants to
> learn to solve problems they haven't even dreamed of yet.

But, boy, you'd learn a *lot* trying...

   N

*lurk*

Re: Hot kernel change

[Yann Dirson]

Jim Richardson wrote:
>MkLinux was available for x86, but I have no idea if it is still in
>development. To be clear, it doesn't allow you to simply replace a
>kernel, but to add a second one, and possibly, to start transferring
>over tasks to it. 

Aside from mklinux, there is the L4Linux option, based on a more modern
microkernel than Mach, and supporting x86.

See http://www.l4ka.org/projects/l4linux/

-- 
Yann Dirson    <ydirson@altern.org> |    Why make M$-Bill richer & richer ?
Debian-related: <dirson@debian.org> |   Support Debian GNU/Linux:
Pro:    <yann.dirson@fr.alcove.com> |  Freedom, Power, Stability, Gratuity
     http://ydirson.free.fr/        | Check <http://www.debian.org/>

Re: Hot kernel change

[Steven Cole]

On Fri, 2004-04-02 at 09:16, Yann Dirson wrote:
> Jim Richardson wrote:
> >MkLinux was available for x86, but I have no idea if it is still in
> >development. To be clear, it doesn't allow you to simply replace a
> >kernel, but to add a second one, and possibly, to start transferring
> >over tasks to it. 
> 
> Aside from mklinux, there is the L4Linux option, based on a more modern
> microkernel than Mach, and supporting x86.
> 
> See http://www.l4ka.org/projects/l4linux/

For folks interested in developing/testing various fast boot or possible
future kernel hot swap projects, there is a mailing list here:

http://lists.osdl.org/mailman/listinfo/fastboot

Steven