Message-ID: <CALHNRZ8DJNAd=TrehuoydCad=iDKwUZ3xgEZC1f0+yLHWi3HAQ@mail.gmail.com>
Date: Tue, 26 Aug 2025 11:34:22 -0500
From: Aaron Kling <webgeek1234@...il.com>
To: Ard Biesheuvel <ardb@...nel.org>
Cc: Jan Kiszka <jan.kiszka@...mens.com>, linux-efi@...r.kernel.org, 
	linux-kernel@...r.kernel.org
Subject: Re: [PATCH] efistub: Lower default log level

On Tue, Aug 26, 2025 at 7:43 AM Ard Biesheuvel <ardb@...nel.org> wrote:
>
> On Tue, 26 Aug 2025 at 10:16, Ard Biesheuvel <ardb@...nel.org> wrote:
> >
> > On Tue, 26 Aug 2025 at 09:23, Jan Kiszka <jan.kiszka@...mens.com> wrote:
> > >
> > > On 26.08.25 00:09, Aaron Kling wrote:
> > > > On Mon, Aug 25, 2025 at 4:28 PM Ard Biesheuvel <ardb@...nel.org> wrote:
> > > >>
> > > >> On Tue, 26 Aug 2025 at 02:34, Aaron Kling <webgeek1234@...il.com> wrote:
> > > >>>
> > > >>> On Mon, Aug 25, 2025 at 5:44 AM Ard Biesheuvel <ardb@...nel.org> wrote:
> > > >>>>
> > > >>>> On Sun, 24 Aug 2025 at 16:47, Jan Kiszka <jan.kiszka@...mens.com> wrote:
> > > >>>>>
> > > >>>>> On 24.08.25 02:31, Ard Biesheuvel wrote:
> > > >>>>>> On Sat, 16 Aug 2025 at 16:52, Jan Kiszka <jan.kiszka@...mens.com> wrote:
> > > >>>>>>>
> > > >>>>>>> On 15.07.25 03:35, Ard Biesheuvel wrote:
> > > >>>>>>>> On Tue, 8 Jul 2025 at 17:31, Aaron Kling <webgeek1234@...il.com> wrote:
> > > >>>>>>>>>
> > > >>>>>>>>> On Tue, Jul 8, 2025 at 2:30 AM Aaron Kling via B4 Relay
> > > >>>>>>>>> <devnull+webgeek1234.gmail.com@...nel.org> wrote:
> > > >>>>>>>>>>
> > > >>>>>>>>>> From: Aaron Kling <webgeek1234@...il.com>
> > > >>>>>>>>>>
> > > >>>>>>>>>> Some uefi implementations will write the efistub logs to the display
> > > >>>>>>>>>> over a splash image. This is not desirable for debug and info logs, so
> > > >>>>>>>>>> lower the default efi log level to exclude them.
> > > >>>>>>>>>>
> > > >>>>>>>>>> Suggested-by: Ard Biesheuvel <ardb@...nel.org>
> > > >>>>>>>>>> Signed-off-by: Aaron Kling <webgeek1234@...il.com>
> > > >>>>>>>>>> ---
> > > >>>>>>>>>>  drivers/firmware/efi/libstub/printk.c | 4 ++--
> > > >>>>>>>>>>  1 file changed, 2 insertions(+), 2 deletions(-)
> > > >>>>>>>>>>
> > > >>>>>>>>>> diff --git a/drivers/firmware/efi/libstub/printk.c b/drivers/firmware/efi/libstub/printk.c
> > > >>>>>>>>>> index 3a67a2cea7bdf1aa215d48dbf9ece4ceec6e4c28..bc599212c05dd746a9c54abbbe61a4bf70f1a8c4 100644
> > > >>>>>>>>>> --- a/drivers/firmware/efi/libstub/printk.c
> > > >>>>>>>>>> +++ b/drivers/firmware/efi/libstub/printk.c
> > > >>>>>>>>>> @@ -5,13 +5,13 @@
> > > >>>>>>>>>>  #include <linux/ctype.h>
> > > >>>>>>>>>>  #include <linux/efi.h>
> > > >>>>>>>>>>  #include <linux/kernel.h>
> > > >>>>>>>>>> -#include <linux/printk.h> /* For CONSOLE_LOGLEVEL_* */
> > > >>>>>>>>>> +#include <linux/kern_levels.h>
> > > >>>>>>>>>>  #include <asm/efi.h>
> > > >>>>>>>>>>  #include <asm/setup.h>
> > > >>>>>>>>>>
> > > >>>>>>>>>>  #include "efistub.h"
> > > >>>>>>>>>>
> > > >>>>>>>>>> -int efi_loglevel = CONSOLE_LOGLEVEL_DEFAULT;
> > > >>>>>>>>>> +int efi_loglevel = LOGLEVEL_NOTICE;
> > > >>>>>>>>>>
> > > >>>>>>>>>>  /**
> > > >>>>>>>>>>   * efi_char16_puts() - Write a UCS-2 encoded string to the console
> > > >>>>>>>>>>
> > > >>>>>>>>>> ---
> > > >>>>>>>>>> base-commit: d7b8f8e20813f0179d8ef519541a3527e7661d3a
> > > >>>>>>>>>> change-id: 20250708-efi-default-loglevel-4da5a36cac87
> > > >>>>>>>>>>
> > > >>>>>>>>>> Best regards,
> > > >>>>>>>>>> --
> > > >>>>>>>>>> Aaron Kling <webgeek1234@...il.com>
> > > >>>>>>>>>
> > > >>>>>>>>> This patch was originally suggested a few months ago [0], but as far
> > > >>>>>>>>> as I can tell was never queued for merge. Since I'm also hitting a
> > > >>>>>>>>> case where this is relevant, I'm sending this in to bring attention
> > > >>>>>>>>> back to it.
> > > >>>>>>>>>
> > > >>>>>>>>
> > > >>>>>>>> I've queued this up now - thanks.
> > > >>>>>>>>
> > > >>>>>>>
> > > >>>>>>> And how can I get back the loglevel info? It seems I can only choose
> > > >>>>>>> between notice, silent and debug now. And the latter two only by also
> > > >>>>>>> touching the kernel's loglevel.
> > > >>>>>>>
> > > >>>>>>> I'm particularly missing [1] in my UART logs now which is helpful in
> > > >>>>>>> understanding this essential system state.
> > > >>>>>>>
> > > >>>>>>
> > > >>>>>> Hi Jan,
> > > >>>>>>
> > > >>>>>> Is efi=debug too noisy for you?
> > > >>>>>
> > > >>>>> Yes, also because it affects the kernel even more. I'm looking for
> > > >>>>> "efi=info".
> > > >>>>>
> > > >>>>> I don't get the reason behind this change anymore as well. If you have a
> > > >>>>> splash screen shown, weren't you booting with "quiet" before already,
> > > >>>>> thus also without any stub messages?
> > > >>>>>
> > > >>>>
> > > >>>> Yeah, good point. IIRC that came up in the discussion but I can't
> > > >>>> remember the motivation so it can't have been very convincing.
> > > >>>>
> > > >>>> So should we just revert this change?
> > > >>>
> > > >>> I'd prefer not to have to set quiet to get a clean splash screen. That
> > > >>> doesn't seem like an unreasonable expectation, getting default
> > > >>> non-debug logs and not having stuff written on top of the splash
> > > >>> image.
> > > >>
> > > >> Perhaps you could remind us why this only applies to the efistub
> > > >> output, and having the output of the kernel itself corrupting the
> > > >> splash screen is fine?
> > > >
> > > > I'm not greatly knowledgeable about the efi standard and what's
> > > > happening under the hood, so I will just speak to what I saw in my use
> > > > case. I'm working on Nvidia Tegra devices, newer generations of which
> > > > use EDK2 as the last stage bootloader. The target os is Android, which
> > > > has a pretty strictly controlled defconfig. Prior to this change, the
> > > > kernel efistub logs were getting passed to the efi impl, which was
> > > > then printing those lines to the display. The kernel logs were not
> > > > being printed to the screen, as none of the console drivers were
> > > > enabled to do so. So after this change, regardless of the kernel log
> > > > level, the boot splash will remain untouched until the kernel display
> > > > driver takes over the display and the os renders to it. Because no
> > > > efistub log lines are being printed.
> > > >
> > >
> > > That makes sense now, and surely don't mind having some build-time or
> > > runtime configuration switch that allow to tune the system into such
> > > settings. It's just not so nice to take away the freedom of full-scale
> > > loglevel control from the efistub.
> > >
> >
> > Yeah, that would be my fault, I guess. I suggested simplifying this to
> > the current approach.
> >
> > Would it be sufficient to make the EFI stub loglevel a separate
> > compile time Kconfig option? I'd prefer that over adding more runtime
> > logic.
>
> Would the below work for you Aaron?
>
> diff --git a/drivers/firmware/efi/Kconfig b/drivers/firmware/efi/Kconfig
> index eb1bff6968a5..f7552f36ab51 100644
> --- a/drivers/firmware/efi/Kconfig
> +++ b/drivers/firmware/efi/Kconfig
> @@ -72,6 +72,11 @@ config EFI_RUNTIME_WRAPPERS
>  config EFI_GENERIC_STUB
>         bool
>
> +config EFI_STUB_LOGLEVEL
> +       int "Loglevel for the EFI stub console"
> +       range 1 15
> +       default CONSOLE_LOGLEVEL_DEFAULT
> +
>  config EFI_ZBOOT
>         bool "Enable the generic EFI decompressor"
>         depends on EFI_GENERIC_STUB && !ARM
> diff --git a/drivers/firmware/efi/libstub/printk.c
> b/drivers/firmware/efi/libstub/printk.c
> index bc599212c05d..782d1330c1cc 100644
> --- a/drivers/firmware/efi/libstub/printk.c
> +++ b/drivers/firmware/efi/libstub/printk.c
> @@ -5,13 +5,12 @@
>  #include <linux/ctype.h>
>  #include <linux/efi.h>
>  #include <linux/kernel.h>
> -#include <linux/kern_levels.h>
>  #include <asm/efi.h>
>  #include <asm/setup.h>
>
>  #include "efistub.h"
>
> -int efi_loglevel = LOGLEVEL_NOTICE;
> +int efi_loglevel = CONFIG_EFI_STUB_LOGLEVEL;
>
>  /**
>   * efi_char16_puts() - Write a UCS-2 encoded string to the console

The only issue for me is that the Android defconfig is set by Google.
Specific devices build kernel modules in a device specific manner, but
the core kernel is fixed. And since efistub is =y, that means vendors
and third party projects like me cannot change this as a build time
config. This works for me only if I can convince Google to set the log
level lower in their defconfig. If this is what you wish to do, I can
suggest such a change to aosp once they fork the next lts.

Aaron

Powered by blists - more mailing lists