Crashes of tainted kernels

[Chris Murphy]

On Tue, Mar 17, 2015 at 2:41 PM, Rick Stevens <ricks at alldigital.com> wrote:
> On 03/17/2015 01:26 PM, Chris Murphy wrote:

>> a. It actually has to communicate with a server, so whose hosting this
>> "other" bugzilla?
>
>
> I think the point here is that a user could configure ADDITIONAL
> bugzilla sites.

Do you have examples of such additional bugzilla sites?


> If the kernel aborts and abrt determines it's a tainted
> kernel, then it could pop up this list of additional sites and let the
> user report the crash to a site on that list. Yes, this is an extension
> to abrt, but I think a reasonable (and positive) one.

If you know the site already, why not just file the bug directly on
that bugzilla? Why does it need to be automated?


>
>> b. Then what happens? Who looks at these reports?
>
>
> It rather depends on which site you're reporting to. I think bug
> reports to companies such as nVidia or AMD or HP or Intel or any one of
> a number of others would be looked at. I do development work on nVidia
> CUDA stuff and, while they don't have a bugzilla, they are responsive
> to bug reports using their other mechanisms.
>
>> Why bother with this infrastructure if no one is going to look at the
>> reports or do anything about them?
>
>
> You're assuming they won't without any reason to think that. The fact
> they set up a bugzilla in the first place (and the user has login
> credentials for it) is fairly indicative that they do want to fix
> problems.
>
> To assume that only the kernel developers or Red Hat (or Suse or Ubuntu
> or whomever) are the ONLY people that care to fix bugs is rather myopic.

That is true, I was only considering bugs going into the RHBZ. But I
don't really see much value in the automated reporting for a kernel
oops because the only thing that tends to get reported by abrt with
Fedora kernels is dmesg. It's not like you really gain anything by
having abrt file it. The OS installer often has something like a dozen
files to upload so in that case abrt is nice. But I just don't see the
value here unless there were some mechanism to help the user file the
bug with the proper BZ. And that's kinda difficult without assuming
the out of tree module is actually to blame or incidental to the oops.

The fact of the matter is, you have to understand the call trace and
the kernel and probably the module to a fair degree to know what party
to send the bug to.

OR, like Matt said several times, test with a stock kernel. Then you know.


-- 
Chris Murphy