Re: DRM-based Oops viewer

[Ahmed S. Darwish]

On Mon, Mar 11, 2019 at 02:49:41PM +0100, Daniel Vetter wrote:
> On Mon, Mar 11, 2019 at 11:04:19AM +0200, Jani Nikula wrote:
> > On Sun, 10 Mar 2019, "Ahmed S. Darwish" <darwish.07@xxxxxxxxx> wrote:
> > > Hello DRM/UEFI maintainers,
> > >
> > > Several years ago, I wrote a set of patches to dump the kernel
> > > log to disk upon panic -- through BIOS INT 0x13 services. [1]
> > >
> > > The overwhelming response was that it's unsafe to do this in a
> > > generic manner. Linus proposed a video-based viewer instead: [2]
> > >
> > >     If you want to do the BIOS services thing, do it for video: copy the
> > >     oops to low RAM, return to real mode, re-run the graphics card POST
> > >     routines to initialize text-mode, and use the BIOS to print out the
> > >     oops.  That is WAY less scary than writing to disk.
> > >
> > > Of course it's 2019 now though, and it's quite known that
> > > Intel is officially obsoleting the PC/AT BIOS by 2020.. [3]
> > >
> > > Researching whether this can be done from UEFI, it was also clear
> > > that UEFI "Runtime Services" do not provide any re-initialization
> > > routines. [4]
> > >
> > > The maximum possible that UEFI can provide is a GOP-provided
> > > framebuffer that's ready to use by the OS -- even after the UEFI
> > > boot phase is marked as done through ExitBootServices(). [5]
> > >
> > > Of course, once native drivers like i915 or radeon take over,
> > > such a framebuffer is toast... [6]
> > >
> > > Thus a possible remaining option, is to display the oops through
> > > "minimal" DRM drivers provided for each HW variant... Since
> > > these special drivers will run only and fully under a panic()
> > > context though, several constraints exist:
> > >
> > >   - The code should be fully synchronous (irqs are disabled)
> > >   - It should not allocate any dynamic memory
> > >   - It should make minimal assumptions about HW state
> > >   - It should not chain into any other kernel subsystem
> > >   - It has ample freedom to use delay-based loops and the
> > >     like, the kernel is already dead.
> > >
> > > How feasible is it to have such a special "DRM viewoops"
> > > framework + its minimal drivers in the kernel?
> >
> > Please first better define what you want to achieve.
> >
> > Do you want to store the dmesg or oops (like your original series
> > suggests) or do you want to display the oops? Do you want the facility
> > to be functioning at all times, or only when specifically requested in
> > advance by the user? If you want to display the oops, do you want it to
> > also work when the display is disabled at the time of the oops? What if
> > the display is at attached to a port on a dock?
> >
> > There's at least kdump, ramoops, and netconsole that can be used to
> > achieve some of what you want. How do they fall short for you?
>
> Assuming the use-case is to get an oops to display on a kms driver, we do
> have a fairly comprehensive plan of what that's should look like:
>
> https://dri.freedesktop.org/docs/drm/gpu/todo.html#make-panic-handling-work
>
> This takes into account all the failed previous attempts at trying to get
> an oops to display. It's conceptually a match with your viewoops framework
> I think.

Thanks a lot Daniel for the reference! Yup, this is a conceptual
match indeed!

It's great to know that at the maintainer level there is some
agreement on, awareness of, and plans for, the general topic...

I'll jump to Noralf Trønnes's just-posted patches then and see how
to move from there :)

all the best,

--darwi
http://darwish.chasingpointers.com