Re: [PATCH 6/6] efi/earlycon: Remap entire framebuffer after page initialization

From: Ard Biesheuvel
Date: Wed Dec 11 2019 - 06:26:23 EST


On Tue, 10 Dec 2019 at 21:05, Arvind Sankar <nivedita@xxxxxxxxxxxx> wrote:
>
> On Mon, Dec 09, 2019 at 07:24:13PM +0000, Ard Biesheuvel wrote:
> > On Mon, 9 Dec 2019 at 20:12, Arvind Sankar <nivedita@xxxxxxxxxxxx> wrote:
> > >
> > > On Fri, Dec 06, 2019 at 04:55:42PM +0000, Ard Biesheuvel wrote:
> > > > From: Andy Shevchenko <andriy.shevchenko@xxxxxxxxxxxxxxx>
> > > >
> > > > When commit 69c1f396f25b
> > > >
> > > > "efi/x86: Convert x86 EFI earlyprintk into generic earlycon implementation"
> > > >
> > > > moved the x86 specific EFI earlyprintk implementation to a shared location,
> > > > it also tweaked the behaviour. In particular, it dropped a trick with full
> > > > framebuffer remapping after page initialization, leading to two regressions:
> > > > 1) very slow scrolling after page initialization,
> > > > 2) kernel hang when the 'keep_bootcon' command line argument is passed.
> > > >
> > > > Putting the tweak back fixes #2 and mitigates #1, i.e., it limits the slow
> > > > behavior to the early boot stages, presumably due to eliminating heavy
> > > > map()/unmap() operations per each pixel line on the screen.
> > > >
> > >
> > > Could the efi earlycon have an interaction with PCI resource allocation,
> > > similar to what commit dcf8f5ce3165 ("drivers/fbdev/efifb: Allow BAR to
> > > be moved instead of claiming it") fixed for efifb?
> >
> > Yes. If the BAR gets moved, things will break. This is mostly an issue
> > for the keep_bootcon case, but that is documented as being a debug
> > feature specifically for addressing console initialization related
> > issues. Earlycon itself is also a debug feature, so if you hit the BAR
> > reallocation issue, you're simply out of luck. Note that this happens
> > rarely in practice, only on non-x86 systems where the firmware and the
> > kernel have very different policies regarding BAR allocation, and on
> > DT based systems, you can force the OS to honour the existing
> > allocation by using linux,pci-probe-only
>
> Thanks. Another q -- I tried out the earlycon=efifb, and it seems like
> it gets disabled (without keep_bootcon) as soon as dummycon takes over,
> which is well before the real console.
>
> DUMMY_CONSOLE is defined as
> depends on VGA_CONSOLE!=y || SGI_NEWPORT_CONSOLE!=y
> default y
>
> so it seems like it will pretty much always be enabled, as it doesn't
> seem likely that VGA_CONSOLE=y and SGI_NEWPORT_CONSOLE=y would ever be
> true simultaneously.
>
> Am I missing something or is this the way it's supposed to work? So
> keep_bootcon seems almost necessary with the EFI boot console? Would a
> patch to not disable boot console when dummycon is initialized, but wait
> for a real console, be useful?
>

Well spotted!

I have traced this down to [0] which combined various arch specific
definitions into one, and obviously chose the wrong boolean operator
for combining the conditions.

Patches welcome.


[0] https://git.kernel.org/pub/scm/linux/kernel/git/tglx/history.git/commit/drivers/video/console/Kconfig?id=31d2a7d36d6989c714b792ec00358ada24c039e7