Re: [CRASH][BISECTED] 6.4.1 crash in boot

[Kees Cook]

On Mon, Jul 03, 2023 at 09:03:38AM +0200, Mirsad Goran Todorovac wrote:
> On 3.7.2023. 7:41, Kees Cook wrote:
> > On Mon, Jul 03, 2023 at 07:18:57AM +0200, Mirsad Goran Todorovac wrote:
> > > I apologise for confusion. In fact, I have cloned the Torvalds tree after
> > > 6.4.1 was released, but I actually cloned the Torvalds tree, not the 6.4.1
> > > from the stable branch as the Subject line might have misled.
> > 
> > Thanks, no worries! I got myself confused too. :)
> > 
> > The config you sent looks like I'd expect now too. Questions for you, if
> > you have time to diagnose further:
> > 
> > - Are you able to catch the very beginning of the crash, where the Oops
> >    starts?
> 
> It scrolls up very quickly. Couldn't catch that with the camera.
> 
> > - Does pstore work for you to catch the crash?
> 
> Haven't tried that yet. I will have to do some homework.

Try adding this to the .config:

# Enable PSTORE support
CONFIG_PSTORE=y
CONFIG_PSTORE_DEFAULT_KMSG_BYTES=10240
CONFIG_PSTORE_COMPRESS=y
CONFIG_PSTORE_DEFLATE_COMPRESS=y
# Enable UEFI pstore backend
CONFIG_EFI_VARS_PSTORE=y
# CONFIG_EFI_VARS_PSTORE_DEFAULT_DISABLE is not set
# Enable ACPI ERST pstore backend
CONFIG_ACPI=y
CONFIG_ACPI_APEI=y

A go write-up about using it is here:
https://blogs.oracle.com/linux/post/pstore-linux-kernel-persistent-storage-file-system
and covers the systemd-pstore details too. Note that in the config I
suggested, I've enabled the efi backend by default.

> > - Can you try booting with this patch applied?
> >    https://lore.kernel.org/lkml/20230629190900.never.787-kees@xxxxxxxxxx/
> 
> Sure, but after 4 PM UTC+02 I suppose.

Cool. xhci-hub is in your backtrace, and the above patch was made for
something very similar (though, again, I don't see why you're getting a
_crash_, it should _warn_ and continue normally). And, actually, also
include this patch:
https://lore.kernel.org/lkml/20230614181307.gonna.256-kees@xxxxxxxxxx/

> > I'll try to see if I can figure out anything more from the images you
> > posted.

Yeah, the xhci-hub bit is the only clue I can see here. It's also in the
IRQ handler, which reminds me of this bug that we still don't have a
root-cause for the _crash_ during the warning here:
https://lore.kernel.org/oe-lkp/202306131354.A499DE60@keescook/
but I the new patch I linked to above fixes the source of the warning.

> I really couldn't figure out myself what went wrong with this one?

Having the crash scroll off the page is pretty frustrating. I wonder if
the kernel crash handler could changed to repeat the RIP at the end of
the crash...

-Kees

-- 
Kees Cook