Re: [PATCH RFC] ACPI: processor: idle: Do not propagate acpi_processor_ffh_lpi_probe() -ENODEV

[Breno Leitao]

Hello Sudeep,

On Tue, Apr 14, 2026 at 03:10:03PM +0100, Sudeep Holla wrote:
> On Tue, Apr 14, 2026 at 06:14:19AM -0700, Breno Leitao wrote:
> > Hello Sudeep,
> >
> > On Tue, Apr 14, 2026 at 01:25:53PM +0100, Sudeep Holla wrote:
> > > So while I understand that the kernel did not report an error previously, that
> > > does not mean the _LPI table is merely moot on this platform when it contains
> > > only a WFI state.
> >
> > Can you clarify whether datacenter ARM systems are expected to expose
> > deeper idle states beyond WFI in their _LPI tables?
> >
>
> Of course any system that prefers to save power when its idling must have
> these _LPI deeper idle states.
>
> > Backing up, I'm observing 72 pr_err() messages during boot on these
> > hosts and trying to determine whether this indicates a firmware issue or
> > if the kernel needs adjustment.
>
> I consider this a firmware issue, but not a fatal one. What matters more is
> the behavior after those errors are reported.

I understand. While I'm not a hardware or firmware vendor myself, I can
see how they might consider power management features _optional_ for certain
server configurations.

> If you force success, either through your change or through the PSCI approach
> suggested by lihuisong, then in practice you are only enabling a cpuidle
> driver with a single usable state: WFI. That is not inherently wrong, but it
> also does not provide much benefit.

Given that this isn't a critical error, would it make sense to downgrade
the pr_err() to pr_debug()? is it a reasonable compromise. I just want
to avoid these pr_err() all accross the board, affecting kernel health
metrics in large fleets.

My proposal:

commit c98007f9e10fe229672d29c3844c96705cecaed5
Author: Breno Leitao <leitao@xxxxxxxxxx>
Date:   Tue Apr 14 05:28:28 2026 -0700

    ACPI: processor: idle: Downgrade FFH LPI probe failure message to pr_debug()
    
    The "Invalid FFH LPI data" message is printed at pr_err() level for every
    CPU when acpi_processor_ffh_lpi_probe() fails. On platforms where the FFH
    probe legitimately returns an error (e.g., no deep idle states beyond
    WFI), this floods the kernel log with per-CPU error messages that are not
    actionable.
    
    Downgrade to pr_debug() since this is a diagnostic message, not a
    critical error.
    
    Signed-off-by: Breno Leitao <leitao@xxxxxxxxxx>

diff --git a/drivers/acpi/processor_idle.c b/drivers/acpi/processor_idle.c
index ee5facccbe10c..ab93a2c10a9ad 100644
--- a/drivers/acpi/processor_idle.c
+++ b/drivers/acpi/processor_idle.c
@@ -1259,7 +1259,7 @@ static int acpi_processor_get_power_info(struct acpi_processor *pr)
 	if (pr->flags.has_lpi) {
 		ret = acpi_processor_ffh_lpi_probe(pr->id);
 		if (ret)
-			pr_err("CPU%u: Invalid FFH LPI data\n", pr->id);
+			pr_debug("CPU%u: Invalid FFH LPI data\n", pr->id);
 	}
 
 	return ret;