2014-06-24 18:23:25

by Alan Stern

[permalink] [raw]
Subject: CONFIG_UEVENT_HELPER default

Michael and Greg:

The help text for CONFIG_UEVENT_HELPER says (among other things):

This should not be used today, because usual systems create
many events at bootup or device discovery in a very short time
frame.

If it shouldn't be used, why does it default to 'y'?

Alan Stern


2014-06-24 18:33:23

by Greg KH

[permalink] [raw]
Subject: Re: CONFIG_UEVENT_HELPER default

On Tue, Jun 24, 2014 at 11:30:05AM -0700, Michael Marineau wrote:
> On Jun 24, 2014 11:23 AM, "Alan Stern" <[email protected]> wrote:
> >
> > Michael and Greg:
> >
> > The help text for CONFIG_UEVENT_HELPER says (among other things):
> >
> > ? ? ? ? ? This should not be used today, because usual systems create
> > ? ? ? ? ? many events at bootup or device discovery in a very short time
> > ? ? ? ? ? frame.
> >
> > If it shouldn't be used, why does it default to 'y'?
> >
> > Alan Stern
> >
>
> To introduce the option but not change the default behavior. (yet?) I don't
> really have an opinion one way or the other, I just defaulted to being
> conservative.

Yes, being conservative is good as turning this off with older systems
(like the pathological Fedora 3 system that some kernel developers still
use for testing), would result in a non-booting box. So if you know
that your system is "new enough", it's safe to turn off, but if you have
a doubt, leave it on to be safe.

thanks,

greg k-h

2014-06-24 19:55:32

by Andy Lutomirski

[permalink] [raw]
Subject: Re: CONFIG_UEVENT_HELPER default

On 06/24/2014 11:33 AM, Greg KH wrote:
> On Tue, Jun 24, 2014 at 11:30:05AM -0700, Michael Marineau wrote:
>> On Jun 24, 2014 11:23 AM, "Alan Stern" <[email protected]> wrote:
>>>
>>> Michael and Greg:
>>>
>>> The help text for CONFIG_UEVENT_HELPER says (among other things):
>>>
>>> This should not be used today, because usual systems create
>>> many events at bootup or device discovery in a very short time
>>> frame.
>>>
>>> If it shouldn't be used, why does it default to 'y'?
>>>
>>> Alan Stern
>>>
>>
>> To introduce the option but not change the default behavior. (yet?) I don't
>> really have an opinion one way or the other, I just defaulted to being
>> conservative.
>
> Yes, being conservative is good as turning this off with older systems
> (like the pathological Fedora 3 system that some kernel developers still
> use for testing), would result in a non-booting box. So if you know
> that your system is "new enough", it's safe to turn off, but if you have
> a doubt, leave it on to be safe.

As far as I know, there's no real requirement that a defconfig kernel be
able to boot old userspace. We want an oldconfig kernel to be able to
boot old userspace, but changing the default won't affect that.

For example, a defconfig kernel won't boot opensuse 9.

--Andy

2014-06-24 22:35:15

by Ken Moffat

[permalink] [raw]
Subject: Re: CONFIG_UEVENT_HELPER default

On Tue, Jun 24, 2014 at 12:55:26PM -0700, Andy Lutomirski wrote:
> On 06/24/2014 11:33 AM, Greg KH wrote:
> > On Tue, Jun 24, 2014 at 11:30:05AM -0700, Michael Marineau wrote:
> >> On Jun 24, 2014 11:23 AM, "Alan Stern" <[email protected]> wrote:
> >>>
> >>> Michael and Greg:
> >>>
> >>> The help text for CONFIG_UEVENT_HELPER says (among other things):
> >>>
> >>> This should not be used today, because usual systems create
> >>> many events at bootup or device discovery in a very short time
> >>> frame.
> >>>
> >>> If it shouldn't be used, why does it default to 'y'?
> >>>
> >>> Alan Stern
> >>>
> >>
> >> To introduce the option but not change the default behavior. (yet?) I don't
> >> really have an opinion one way or the other, I just defaulted to being
> >> conservative.
> >
> > Yes, being conservative is good as turning this off with older systems
> > (like the pathological Fedora 3 system that some kernel developers still
> > use for testing), would result in a non-booting box. So if you know
> > that your system is "new enough", it's safe to turn off, but if you have
> > a doubt, leave it on to be safe.
>
> As far as I know, there's no real requirement that a defconfig kernel be
> able to boot old userspace. We want an oldconfig kernel to be able to
> boot old userspace, but changing the default won't affect that.
>
> For example, a defconfig kernel won't boot opensuse 9.
>
> --Andy

I noticed this help message yesterday, and decided that I
almost-certainly did NOT want it (that system is about 6 weeks old,
with then-current releases of everything). But I was not able to
complete the kernel build because of other problems (possibly
related to gcc-4.9.1) and I have other things to do at the moment.
All of which means that I can not, for the moment, review what will
happen if I let this option get enabled.

Two things about this default concern me:

(i.) I got the option with 'make oldconfig' and, after reading the
help, made a decision. But, in the absence of other problems, and
if the help text is correct, it looks as if a kernel built after
accepting the default 'Y' here with recent userspace might grind to
a halt after "successfully" booting? That sounds slightly better
than "fails to boot", but only slightly. Maybe the problem needs
a lot of modules, or is it something like "it will hang for a minute,
then boot" ?

(ii.) I understand that people continue to use ancient userspace,
and for that the 'Y' option is needed. Using ancient userspace is
a worthwhile thing for _somebody_ to try. But where is the line
between "you need to enable this" and "enabling this might be a
really bad idea" ? Maybe a specific version of udev ?

ĸen
--
Nanny Ogg usually went to bed early. After all, she was an old lady.
Sometimes she went to bed as early as 6 a.m.

2014-07-03 21:24:15

by Ken Moffat

[permalink] [raw]
Subject: Re: CONFIG_UEVENT_HELPER default

On Tue, Jun 24, 2014 at 11:35:09PM +0100, Ken Moffat wrote:
> On Tue, Jun 24, 2014 at 12:55:26PM -0700, Andy Lutomirski wrote:
> > On 06/24/2014 11:33 AM, Greg KH wrote:
> > > On Tue, Jun 24, 2014 at 11:30:05AM -0700, Michael Marineau wrote:
> > >> On Jun 24, 2014 11:23 AM, "Alan Stern" <[email protected]> wrote:
> > >>>
[ snipping most of this ]
> (i.) I got the option with 'make oldconfig' and, after reading the
> help, made a decision. But, in the absence of other problems, and
> if the help text is correct, it looks as if a kernel built after
> accepting the default 'Y' here with recent userspace might grind to
> a halt after "successfully" booting? That sounds slightly better
> than "fails to boot", but only slightly. Maybe the problem needs
> a lot of modules, or is it something like "it will hang for a minute,
> then boot" ?
>
> (ii.) I understand that people continue to use ancient userspace,
> and for that the 'Y' option is needed. Using ancient userspace is
> a worthwhile thing for _somebody_ to try. But where is the line
> between "you need to enable this" and "enabling this might be a
> really bad idea" ? Maybe a specific version of udev ?
>
Now that I have managed to boot -rc3, with the default 'Y' on a
recent system (linuxfromscratch from May), it appears to work fine.

So, the text implies that bad things might happen, but so far I have
not seen them. I'll stop caring.

ĸen
--
Nanny Ogg usually went to bed early. After all, she was an old lady.
Sometimes she went to bed as early as 6 a.m.