Re: The bug

[Kevin Price <kp@osnanet.de>]

I largely agree with Greg.

Am 13.12.23 um 16:33 schrieb Greg Wooledge:
> On Wed, Dec 13, 2023 at 04:13:44PM +0100, tomas@tuxteam.de wrote:
>> On Wed, Dec 13, 2023 at 10:10:37AM -0500, Greg Wooledge wrote:
>>> On Wed, Dec 13, 2023 at 09:56:46AM -0500, Stefan Monnier wrote:
>>>> If so, then IIUC the answer is a resounding "YES, it is safe!".

Safe not to fry your ext4 by Bug#1057843, yes, Stefan.
Safe in general, as originally asked by Rick? He might be lucky, or
maybe less so.

>>> Safety is subjective.  A great deal will depend on what kind of system
>>> is being upgraded.  If it's a remote server to which you have limited
>>> or no physical access, booting a kernel that may "just be unusable"
>>> (enough to prevent editing GRUB menus and rebooting) could be a disaster.

Absolutely, Greg.

>> ...but that one most probably won't be attached via a Broadcom to the 'net.

Tomas: Servers are most usually not connected through wifi alone. But
"the bug" (#1057967/#1057969) won't only disable the wifi adapter, but
would probably make the running computer largely unusable, even unable
to shut down. That's confirmed. In that case you still might have access
through LAN IOT possibly fix GRUB's configuration, if you find a way to
do that without sudo, and working around whatever problems you'll
encounter attempting that. But even then, that new GRUB configuration
will never come into effect until you forcibly reboot/power cycle the
computer. (which has always been a bad thing to do in the first place)

Under these circumstances, "the bug" can become a huge problem. Maybe
unlikely for many use cases, but then huge.

> My superficial understanding, after skimming through the bug report,
> is that problems could be triggered just by *loading* one of the
> affected wifi driver modules.

With less superficial understanding, I fully agree with Greg.

> This would happen for any machine that
> has one of the "right" kinds of wifi hardware, even if that hardware
> isn't actively being used.

Exactly. Mere presence of wifi adapters will cause debian to load their
respective wifi driver modules, that in turn will invoke cfg80211, maybe
or not triggering cfg80211's bug.

> (Not just Broadcom either; at least one
> person reported an issue with Realtek.)

IIUC, that was Olivier's rtl88x2bu non-free wifi driver too, causing the
bug, but not Alberto's r8169, which is for wired LAN.

> Perhaps I'm reading it incorrectly, but I still feel it's wise to wait
> a little while and see if any more problems pop up, if stability is
> important to you.

Yes. If you're up to gambling, throw in all the computers you're willing
to spare. In case of solely remote controlled servers: I wouldn't.
Risking to lose control of servers is too much of a bet for too little
of a win, IMHO.

> I also salute the courage of those who've tested
> these recent changes.  Thank you all.

Appreciation for my small part (in pointing the problem out in the first
place) accepted, but please send your muchos kudos to Salvatore
Bonaccorso <carnil@debian.org>, who deserves credits for solving it.
-- 
Kevin Price