Le 12/09/2024 à 19:12, Ard Biesheuvel a écrit :
On Thu, 12 Sept 2024 at 19:11, Marion & Christophe JAILLET
<christophe.jaillet@xxxxxxxxxx> wrote:
(trying to merge t and cc fields from Ard's and Andy's messages)
Le 12/09/2024 à 12:46, Ard Biesheuvel a écrit :
From: Ard Biesheuvel <ardb@xxxxxxxxxx>
ACPI boot does not provide clocks and regulators, but instead, provides
the PCLK rate directly, and enables the clock in firmware. So deal
gracefully with this.
Fixes: 55750148e559 ("i2c: synquacer: Fix an error handling path in synquacer_i2c_probe()")
Hi,
If that matters, I'm not sure that the Fixes tag is correct.
IIUC, either it is a new functionally that is added (now it works with
ACPI...), or if considered as a fix, then I think that it is linked to
commit 0d676a6c4390 ("i2c: add support for Socionext SynQuacer I2C
controller").
I don't think that 55750148e559 introduced a regression. The issue seems
to be there since the beginning. Agreed?
No.
The original code used IS_ERR_OR_NULL() to explicitly permit the case
where no clock exists at all.
Got it, this is not related to the removed _OR_NULL, but to the change
of logic I've introduce.
if (!IS_ERR)
do_something_and_continue;
if_NOENT_was_returned_we_still_get_there();
which became:
if (IS_ERR)
return;
we_can_get_there_with_NOENT_anymore();
My bad!
CJ
This has worked fine with ACPI boot for many years before this fix was applied.
If yes, then it may be needed to backport it in older kernels too.
No, it used to work. The fix is what broke ACPI boot.