Re: [PATCH v2 2/2] ACPI / Sleep: Check low power idle constraints for debug only

On Fri, 2017-08-11 at 14:43 +0000, Mario.Limonciello@dell.com wrote:

quoted

-----Original Message-----
From: Srinivas Pandruvada [mailto:srinivas.pandruvada@linux.intel.c
om]
Sent: Thursday, August 10, 2017 5:54 PM
To: Limonciello, Mario <redacted>; rjw@rjwysocki.
net;
lenb@kernel.org
Cc: linux-pm@vger.kernel.org; linux-kernel@vger.kernel.org; linux-
acpi@vger.kernel.org; lukas@wunner.de
Subject: Re: [PATCH v2 2/2] ACPI / Sleep: Check low power idle
constraints for
debug only

On Thu, 2017-08-10 at 22:07 +0000, Mario.Limonciello@dell.com
wrote:

quoted

quoted

[...]

quoted

quoted

+
+		ret = acpi_device_get_power(adev, &state);
+		if (!ret)
+			pr_debug("LPI: %s required min power
state
%d, current
power state %d, real power state %d\n",
+				 lpi_constraints_table[i].name
,
+				 lpi_constraints_table[i].min_
dsta
te,
+				 adev->power.state, state);

Isn't this superfluous to be showing the state returned from
acpi_device_get_power and
also probing directly at the state? You can't just rely on the
information you got
back from apci_device_get_power?

They can be different as one is real power state and the other is
what
was set.
For example on Dell 9365 it shows

[ 1924.393653] LPI: \_SB.PCI0.XHC required min power state 3,
current
power state 3, real power state 255

Isn't 255 ACPI_STATE_UNKNOWN?  That makes it seem like it
is a logic problem in acpi_device_get_power (or somewhere down the
chain)
doesn't it?

There is no _PSC for XHC device. So it will return unknown. This is an
optional object, so I think that dumping the status is fine, but
matching with output of acpi_device_get_power() as it relies on _PSC is
not correct for the constraint.

Thanks,
Srinivas

quoted

quoted

quoted

+
+		if (adev->flags.power_manageable && adev-

quoted

power.state <

+					lpi_constraints_table[
i].m
in_dstate)
+			pr_info("LPI: Constraint [%s] not
matched\n",
+				 lpi_constraints_table[i].name
);

Similarly here, can't you just compare against &state instead?

The problem then the check will fail for XHCI on Dell 9365. So not
using "state".

Thanks,
Srinivas

quoted

quoted

+	}
+}
+
 static void acpi_sleep_run_lps0_dsm(unsigned int func)
 {
 	union acpi_object *out_obj;

@@ -729,6 +886,9 @@ static int lps0_device_attach(struct

acpi_device *adev,
 				  "_DSM function 0 evaluation
failed\n");
 	}
 	ACPI_FREE(out_obj);
+
+	lpi_device_get_constraints();
+
 	return 0;
 }

@@ -773,6 +933,8 @@ static void acpi_freeze_wake(void)

 	 */
 	if (acpi_sci_irq_valid() &&
 	    !irqd_is_wakeup_armed(irq_get_irq_data(acpi_sci_ir
q)))
{
+		if (pm_debug_messages_enabled())
+			lpi_check_constraints();
 		pm_system_cancel_wakeup();
 		s2idle_wakeup = true;
 	}
--
2.7.5