Re: Is there any supported watchdog hardware straight out of the box?

2019-04-27 Thread Igor Podlesny
On Sat, 27 Apr 2019 at 17:55, Stuart Henderson  wrote:
[...]
> It is not true. They don't have *wide* support but there is some
> supported hw. If someone wants to change this, I suggest adding acpi
> watchdog support would give the best return for time spent.

Got it.

> Also I don't see what this has to do with watchdogs

Well, it's just due to:
> >> Recompiling would be needed.

> you would normally
> not want to wait until userland starts and LKMs are loaded before arming
> the watchdog.

Well, in theory it couldn't be more correct. In practice though hangs
aren't happening that
often so a module won't have time to load and start. At least if it
would be that bad
I won't rely on watchdogs at all.

Thank you nonetheless. :)

-- 
End of message. Next message?



Re: Is there any supported watchdog hardware straight out of the box?

2019-04-27 Thread Stuart Henderson
On 2019-04-27, Igor Podlesny  wrote:
> On Fri, 26 Apr 2019 at 22:58, Stuart Henderson  wrote:
>> On 2019-04-26, Igor Podlesny  wrote:
>> > Or would kernel's recompiling be needed anyways?
> [...]
>> Recompiling would be needed.
>>
>> If you want to try it, see faq 5 about fetching the source tree,
>> add "ichwdt* at pci?" to /sys/arch/amd64/conf/GENERIC. then see faq 5
>> about building a kernel.
>
> Thanks for confirmation and brief how-to even. I'd like to clarify the matter 
> in
> more general ways though.
>
> 1) Is it true that more or less fresh OpenBSD generic kernels come with
> no support of any watchdog hw?

It is not true. They don't have *wide* support but there is some
supported hw. If someone wants to change this, I suggest adding acpi
watchdog support would give the best return for time spent.

> 2) I heard that kernel modules were intentionally rid of in OpenBSD
> primarily due
> security concerns -- did it really happen for that reason? If so, and
> I assume that
> happened long ago, were there any developer's opinions to undo this? Actually
> even not taking crypto verification approach (modules signing) one
> could always have
> secure level increased high enough to cut down this vector of attacks
> completely.

LKM added a bunch of complexity to all kernels with only a small benefit
to a small subset of users, and there's a viable alternative (build your
own kernels rather than just the module). It's not like a kernel build
takes all that long.

> OTOH, it's well known that dynamic loading approach greatly expands
> functionality of
> OS and makes it more convenient to use.

They also gave an easy way for people to add crap to their kernels.
At least with static kernels we can identify from dmesg when somebody
reporting a problem is running something other than a standard kernel
build. With LKMs this is gone, the most we'll have is a printf, but
people reporting bugs have a tendency to remove things they don't want
to show or think are unimportant.

Also I don't see what this has to do with watchdogs, you would normally
not want to wait until userland starts and LKMs are loaded before arming
the watchdog.



Re: Is there any supported watchdog hardware straight out of the box?

2019-04-26 Thread Igor Podlesny
On Sat, 27 Apr 2019 at 12:12, Theo de Raadt  wrote:
> Igor Podlesny  wrote:
[...]
> > 1) Is it true that more or less fresh OpenBSD generic kernels come with
> > no support of any watchdog hw?
> No.

I see.

> > 2) I heard that kernel modules were intentionally rid of in OpenBSD
> > primarily due security concerns -- did it really happen for that reason?
> No.

I see.

> > If so, and I assume that happened long ago, were there any developer's
> > opinions to undo this?
> No.

Well, since it was rid of not due security -- that q-n doesn't matter.
But thanks!

[...]
> > completely.
>
> Huh?  crypto implies security is amateur logic.  If only it were

I didn't say so. Just asked w/o any implies.

> that simple, we'd add crypto to garbage and get secure garbage.

I looked for pertinent reply on the matter not garbage analogies. ;-P
Look, if you're bored with that q-ns why don't you just skip it? I
didn't ask personally you.

> > OTOH, it's well known that dynamic loading approach greatly expands
> > functionality of
> > OS and makes it more convenient to use.
>
> OK, then I am happy to inform you Linux has everything you want, except
> it also has a ton of other things you probably don't want ... alas.

I'm well informed about Linux, NetBSD and you Theo and about lots of
different things. Some of them surely don't really change with time. ;-P
Traditions11

-- 
End of message. Next message?



Re: Is there any supported watchdog hardware straight out of the box?

2019-04-26 Thread Theo de Raadt
Igor Podlesny  wrote:

> On Fri, 26 Apr 2019 at 22:58, Stuart Henderson  wrote:
> > On 2019-04-26, Igor Podlesny  wrote:
> > > Or would kernel's recompiling be needed anyways?
> [...]
> > Recompiling would be needed.
> >
> > If you want to try it, see faq 5 about fetching the source tree,
> > add "ichwdt* at pci?" to /sys/arch/amd64/conf/GENERIC. then see faq 5
> > about building a kernel.
> 
> Thanks for confirmation and brief how-to even. I'd like to clarify the matter 
> in
> more general ways though.
> 
> 1) Is it true that more or less fresh OpenBSD generic kernels come with
> no support of any watchdog hw?

No.
> 
> 2) I heard that kernel modules were intentionally rid of in OpenBSD
> primarily due
> security concerns -- did it really happen for that reason?

No.

> If so, and
> I assume that
> happened long ago, were there any developer's opinions to undo this?

No.

> Actually
> even not taking crypto verification approach (modules signing) one
> could always have
> secure level increased high enough to cut down this vector of attacks
> completely.

Huh?  crypto implies security is amateur logic.  If only it were
that simple, we'd add crypto to garbage and get secure garbage.

> OTOH, it's well known that dynamic loading approach greatly expands
> functionality of
> OS and makes it more convenient to use.

OK, then I am happy to inform you Linux has everything you want, except
it also has a ton of other things you probably don't want ... alas.



Re: Is there any supported watchdog hardware straight out of the box?

2019-04-26 Thread Igor Podlesny
On Fri, 26 Apr 2019 at 22:58, Stuart Henderson  wrote:
> On 2019-04-26, Igor Podlesny  wrote:
> > Or would kernel's recompiling be needed anyways?
[...]
> Recompiling would be needed.
>
> If you want to try it, see faq 5 about fetching the source tree,
> add "ichwdt* at pci?" to /sys/arch/amd64/conf/GENERIC. then see faq 5
> about building a kernel.

Thanks for confirmation and brief how-to even. I'd like to clarify the matter in
more general ways though.

1) Is it true that more or less fresh OpenBSD generic kernels come with
no support of any watchdog hw?

2) I heard that kernel modules were intentionally rid of in OpenBSD
primarily due
security concerns -- did it really happen for that reason? If so, and
I assume that
happened long ago, were there any developer's opinions to undo this? Actually
even not taking crypto verification approach (modules signing) one
could always have
secure level increased high enough to cut down this vector of attacks
completely.
OTOH, it's well known that dynamic loading approach greatly expands
functionality of
OS and makes it more convenient to use.

-- 
End of message. Next message?



Re: Is there any supported watchdog hardware straight out of the box?

2019-04-26 Thread Stuart Henderson
On 2019-04-26, Igor Podlesny  wrote:
> Or would kernel's recompiling be needed anyways?
>
> Moreover, I'm actually interested in intersection of watchdogs
> provided by KVM and
> supported by OpenBSD (as KVM's guest). At least as to KVM's it's gonna
> be a short list:
>
> 1) i6300esb (PCI)
> 2) ib700 (ISA)
>
> Attempt with 1st item shows its driver isn't compiled. Moreover, I've
> found this in mail list archives:
>
> revision 1.411
> date: 2005/05/02 17:26:00;  author: grange;  state: Exp;  lines: +2 -1;
> Add ichwdt(4): Intel 6300ESB ICH watchdog timer driver. Disabled for
> now due to lack of testing. If you have a machine that uses this
> device please contact me.
>
> Is there no solution of this task in OpenBSD 6.x?
>

Recompiling would be needed.

If you want to try it, see faq 5 about fetching the source tree,
add "ichwdt* at pci?" to /sys/arch/amd64/conf/GENERIC. then see faq 5
about building a kernel.




Is there any supported watchdog hardware straight out of the box?

2019-04-26 Thread Igor Podlesny
Or would kernel's recompiling be needed anyways?

Moreover, I'm actually interested in intersection of watchdogs
provided by KVM and
supported by OpenBSD (as KVM's guest). At least as to KVM's it's gonna
be a short list:

1) i6300esb (PCI)
2) ib700 (ISA)

Attempt with 1st item shows its driver isn't compiled. Moreover, I've
found this in mail list archives:

revision 1.411
date: 2005/05/02 17:26:00;  author: grange;  state: Exp;  lines: +2 -1;
Add ichwdt(4): Intel 6300ESB ICH watchdog timer driver. Disabled for
now due to lack of testing. If you have a machine that uses this
device please contact me.

Is there no solution of this task in OpenBSD 6.x?

-- 
End of message. Next message?