Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
From:   "Guilherme G. Piccoli" <gpiccoli@igalia.com>
To:     ardb@kernel.org, linux-efi@vger.kernel.org
Cc:     linux-kernel@vger.kernel.org, kernel-dev@igalia.com,
        kernel@gpiccoli.net, anton@enomsg.org, ccross@android.com,
        keescook@chromium.org, matt@codeblueprint.co.uk,
        matthew.garrett@nebula.com, tony.luck@intel.com,
        "Guilherme G. Piccoli" <gpiccoli@igalia.com>
Subject: [PATCH 0/2] UEFI panic notification mechanism
Date:   Fri, 20 May 2022 16:50:26 -0300
Message-Id: <20220520195028.1347426-1-gpiccoli@igalia.com>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Precedence: bulk

Hi folks, this patch set is about a mechanism to notify the UEFI
firmware about a panic event in the kernel, so the firmware is
enabled to take some action. The specific use case for us is to
show an alternative splash screen if a panic happened.
The base for the idea is to explore the panic notifiers mechanism,
that allows a callback to execute in the panic path.

Patch 1 is kind of a "clean-up" in a way; just taking a helper out
of efibc and adding it on generic efi.h header, so we have common
code used by 3 modules (efibc, efi-pstore and efi-panic).

Patch 2 is the efi-panic module itself; it is *strongly* based on
efibc, and for that I'd like to hereby thank to the authors.
The efibc code is very clear!


Some design decisions to be discussed:

(1) The variable name and value - I called it PanicWarn, and the
data it holds is just a byte, set by default to 0xFF (though users
can change that via module parameter). I have no attachment to
these, we can improve naming and the size of the data for example,
suggestions are appreciated!


(2) The 3 modules (efibc, efi-pstore and efi-panic) share a lot of
ideas or code; both efi-{pstore,panic} deal with panic events, and
both efi{bc,-panic} rely on notifiers and share code.
Should we unify some of them or anything like that? It seemed to me
the proper approach would be a simple and small standalone module,
but I'm open for suggestions.


(3) I've noticed a behavior that also happens in efi-pstore, I'm not
sure if that's something to care about. In the efi-panic module, after
a fresh boot, we check if PanicWarn is there an if so, we print a
message and _delete_ that variable from the NVRAM. But...the variable
is still present in sysfs, in the list created by efivars. Same happens
with efi-pstore, if we delete the dumps generated from /sys/fs/pstore.

In my case, I've used the __efivar_entry_delete() variant, so it doesn't
delete from any list, only from the firmware area. Should this be handled?
Or we just don't care with the empty variable reference in the sysfs tree?


I appreciate feedbacks and suggestions, thanks in advance for the attention!
Cheers,


Guilherme


Guilherme G. Piccoli (2):
  efi: Add a generic helper to convert strings to unicode
  efi-panic: Introduce the UEFI panic notification module

 drivers/firmware/efi/Kconfig      | 10 ++++
 drivers/firmware/efi/Makefile     |  1 +
 drivers/firmware/efi/efi-panic.c  | 97 +++++++++++++++++++++++++++++++
 drivers/firmware/efi/efi-pstore.c |  5 +-
 drivers/firmware/efi/efibc.c      | 16 ++---
 include/linux/efi.h               | 16 +++++
 6 files changed, 130 insertions(+), 15 deletions(-)
 create mode 100644 drivers/firmware/efi/efi-panic.c

-- 
2.36.0