Re: [Tails-dev] Please review'n'merge bugfix/back-to-linux-3…

Supprimer ce message

Répondre à ce message
Auteur: winterfairy
Date:  
À: tails-dev
Sujet: Re: [Tails-dev] Please review'n'merge bugfix/back-to-linux-3.10 [Was: Fix sdmem on Intel graphic hardware, please review]
intrigeri wrote:
> I just run a bunch of tests on three different laptops.
>
> Below, "emergency" means that I've removed the boot USB stick once the
> greeter appeared, "applet" means that I've used the GNOME panel applet
> to shutdown the system, and N/M means N successes on M attempts.
>
> ThinkPenguin Royal:
>
> * 0.22~rc1: emergency = 0/3, applet = 0/3
> * bugfix/sdmem_on_intel_gpu: emergency = 0/3, applet = 0/3
> * bugfix/back-to-linux-3.10: emergency = 3/3, applet = 3/3
>
> ThinkPad X201:
>
> * bugfix/sdmem_on_intel_gpu: emergency = 1/3, applet = 1/3
> * bugfix/back-to-linux-3.10: emergency = 3/3, applet = 3/3
>
> ThinkPad X32:
>
> * bugfix/back-to-linux-3.10: emergency = 3/3, applet = 3/3
>
> So, I'm hereby requesting a review'n'merge of
> bugfix/back-to-linux-3.10 into testing and devel (there's an APT merge
> to do, too). bertagaz will take care of it. Merged into experimental,
> pushed to origin and lizard so the next nightly build from
> experimental will have this stuff.
>
> Once this branch is merged, I'll file a ticket to unblock the
> situation later, as we can't ship Linux 3.10 forever, and will edit
> the call for testing 0.22~rc1 so that we get feedback from
> non-Intel-graphics users and hopefully understand better who's
> affected exactly.


My tests, applet only (success/total):

 * before any fix:         0/3
 * sdmem_on_intel_gpu [1]: 2/3
 * back-to-linux-3.10 [2]: 3/3


Apparently my fix does not fix it completely, so reverting to an older
kernel version seems to be the most sane thing to do right now, yes.

I won't try to fix this problem anymore, so unless anyone else is up to
it, we have to hope a future linux kernel version works again.

I believe the underlying problem is that kexec doesn't play well with all
drivers/hardware, since they are not properly deinitialized or cannot be
properly deinitialized, leaving memory-mapped regions the newly loaded
kernel does not know about. The longstanding known issues page lists
similar problems for other computers, some with non-Intel graphic
hardware.

[1] tails-i386-experimental-0.23-20131202T0830Z-07817d7.iso
[2] tails-i386-experimental-0.23-20131202T1339Z-92b9fbb.iso