Skip to contentSkip to conversation

rtc: pm8xxx: do not fail probe when UEFI offset variable is missing

PATCH In review Greg Ociepka 3 participants 3 messages Aug 21 – Aug 24

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 rtc: pm8xxx: do not fail probe when UEFI offset variable is missing

On machines using "qcom,uefi-rtc-info" the Unix epoch offset for the read-only PMIC RTC counter lives in the RTCInfo UEFI variable. The driver reads it at probe time and treats every failure as fatal, including EFI_NOT_FOUND.

That turns a merely-uninitialized clock into a permanently absent one on firmware that never creates the variable. The ASUS Zenbook A16 (UX3607OA, Snapdragon X2 Elite Extreme "Glymur", InsydeH2O UEFI) is such a machine: qseecom and uefisecapp come up fine, other variables in the same Qualcomm vendor GUID (882f8c2b-9646-435f-8de5-f208ff80c1bd) exist and are readable, but among the 110 variables exposed through efivarfs there is no RTCInfo -- and Windows on the same machine does not create one either. Probe then fails:

rtc-pm8xxx c426000.spmi:pmic@0:rtc@6100: probe with driver rtc-pm8xxx
failed with error -2

This is a chicken-and-egg failure: pm8xxx_rtc_write_uefi_offset() would create the variable (the set path uses EFI_VARIABLE_NON_VOLATILE attributes and efivar_set_variable() creates missing variables), but it can only run from the RTC set_time path -- and the RTC device never registers because probe failed. The variable can never come into existence.

Treat a missing variable like an unset clock instead: warn, keep the zero offset, and register the RTC. Reads expose the raw counter until the first clock set (typically the NTP-triggered RTC synchronization) computes the offset and creates the variable; from then on the machine keeps time across reboots. Userspace already copes with an implausible RTC value at boot -- systemd only steps the clock forward from its persistent timestamp.

Other read errors still fail the probe as before.

Verified on the Zenbook A16 across consecutive boots: on the first boot the driver warns, registers rtc0 and sets the system clock from the raw counter (1970-01-04 here -- the counter had been running for three days); the first NTP-triggered clock set creates RTCInfo; every following boot restores correct wall time from the RTC before any network is up.

Fixes bba38b874886 ("rtc: pm8xxx: add support for uefi offset")
Assisted-by Claude:fable-5
---
 drivers/rtc/rtc-pm8xxx.c | 16 +++++++++++++++-
 1 file changed, 15 insertions(+), 1 deletion(-)
drivers/rtc/rtc-pm8xxx.c +15 -1
index e624f84..4627f4e 100644
@@ -589,7 +589,21 @@ static int pm8xxx_rtc_probe_offset(struct pm8xxx_rtc *rtc_dd)
 		rtc_dd->use_uefi = false;
 	}
 
-	return pm8xxx_rtc_read_uefi_offset(rtc_dd);
+	rc = pm8xxx_rtc_read_uefi_offset(rtc_dd);
+	if (rc == -ENOENT) {
+		/*
+		 * The variable does not exist until something stores an
+		 * offset: pm8xxx_rtc_write_uefi_offset() creates it on the
+		 * first clock set. Keep probing with a zero offset instead
+		 * of failing -- otherwise the RTC never registers and the
+		 * variable can never come into existence.
+		 */
+		dev_warn(rtc_dd->dev,
+			 "UEFI offset variable not found, will be created on first clock set\n");
+		return 0;
+	}
+
+	return rc;
 }
 
 static int pm8xxx_rtc_probe(struct platform_device *pdev)
#1 lore raw
Sashiko has reviewed this patch and found no issues. It looks great!
Quoted 14 lines

On machines using "qcom,uefi-rtc-info" the Unix epoch offset for the read-only PMIC RTC counter lives in the RTCInfo UEFI variable. The driver reads it at probe time and treats every failure as fatal, including EFI_NOT_FOUND.

That turns a merely-uninitialized clock into a permanently absent one on firmware that never creates the variable. The ASUS Zenbook A16 (UX3607OA, Snapdragon X2 Elite Extreme "Glymur", InsydeH2O UEFI) is such a machine: qseecom and uefisecapp come up fine, other variables in the same Qualcomm vendor GUID (882f8c2b-9646-435f-8de5-f208ff80c1bd) exist and are readable, but among the 110 variables exposed through efivarfs there is no RTCInfo -- and Windows on the same machine does not create one either. Probe then fails:

Quoted 7 lines

rtc-pm8xxx c426000.spmi:pmic@0:rtc@6100: probe with driver rtc-pm8xxx
[ ... ]
following boot restores correct wall time from the RTC before any network is up.

Fixes bba38b874886 ("rtc: pm8xxx: add support for uefi offset")

Sashiko has reviewed this patch and found no issues. It looks great!

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1
#2 lore raw
In reply to Greg Ociepka

[ +CC: Konrad and Bjorn ]

Quoted 4 lines · Greg Ociepka

On machines using "qcom,uefi-rtc-info" the Unix epoch offset for the read-only PMIC RTC counter lives in the RTCInfo UEFI variable. The driver reads it at probe time and treats every failure as fatal, including EFI_NOT_FOUND.

This is intentional as "qcom,uefi-rtc-info" should not be set unless the offset *is* stored (by the firmware) in said UEFI variable as documented in the binding.

Quoted 8 lines

That turns a merely-uninitialized clock into a permanently absent one on firmware that never creates the variable. The ASUS Zenbook A16 (UX3607OA, Snapdragon X2 Elite Extreme "Glymur", InsydeH2O UEFI) is such a machine: qseecom and uefisecapp come up fine, other variables in the same Qualcomm vendor GUID (882f8c2b-9646-435f-8de5-f208ff80c1bd) exist and are readable, but among the 110 variables exposed through efivarfs there is no RTCInfo -- and Windows on the same machine does not create one either. Probe then fails:

Then the devicetree added by e8fbbca94db7 ("arm64: dts: qcom: glymur:
Add Asus Zenbook A16 (UX3607OA)") for 7.3 is incorrect and you should drop the corresponding DT property.

Quoted 9 lines

rtc-pm8xxx c426000.spmi:pmic@0:rtc@6100: probe with driver rtc-pm8xxx
failed with error -2

This is a chicken-and-egg failure: pm8xxx_rtc_write_uefi_offset() would create the variable (the set path uses EFI_VARIABLE_NON_VOLATILE attributes and efivar_set_variable() creates missing variables), but it can only run from the RTC set_time path -- and the RTC device never registers because probe failed. The variable can never come into existence.

No, it's not. If your firmware doesn't use the EFI variable for the offset then you need to figure out how the RTC is supposed to work on these platforms (e.g. if you can set the time using some other measure and should not be using an external offset all).

Quoted 7 lines

Treat a missing variable like an unset clock instead: warn, keep the zero offset, and register the RTC. Reads expose the raw counter until the first clock set (typically the NTP-triggered RTC synchronization) computes the offset and creates the variable; from then on the machine keeps time across reboots. Userspace already copes with an implausible RTC value at boot -- systemd only steps the clock forward from its persistent timestamp.

No, this is a hack. And it, for example, prevents you from having the RTC provide the same time in BIOS, Windows and Linux.

Quoted 10 lines

Other read errors still fail the probe as before.

Verified on the Zenbook A16 across consecutive boots: on the first boot the driver warns, registers rtc0 and sets the system clock from the raw counter (1970-01-04 here -- the counter had been running for three days); the first NTP-triggered clock set creates RTCInfo; every following boot restores correct wall time from the RTC before any network is up.

Fixes bba38b874886 ("rtc: pm8xxx: add support for uefi offset")

The current driver is not buggy, it's working as intended so there should be no Fixes tag.

Quoted 23 lines
Assisted-by Claude:fable-5
---
 drivers/rtc/rtc-pm8xxx.c | 16 +++++++++++++++-
 1 file changed, 15 insertions(+), 1 deletion(-)
drivers/rtc/rtc-pm8xxx.c +8 -1
index e624f84..4627f4e 100644
@@ -589,7 +589,21 @@ static int pm8xxx_rtc_probe_offset(struct pm8xxx_rtc *rtc_dd)
 		rtc_dd->use_uefi = false;
 	}
 
-	return pm8xxx_rtc_read_uefi_offset(rtc_dd);
+	rc = pm8xxx_rtc_read_uefi_offset(rtc_dd);
+	if (rc == -ENOENT) {
+		/*
+		 * The variable does not exist until something stores an
+		 * offset: pm8xxx_rtc_write_uefi_offset() creates it on the
+		 * first clock set. Keep probing with a zero offset instead
+		 * of failing -- otherwise the RTC never registers and the
+		 * variable can never come into existence.

drivers/rtc/rtc-pm8xxx.c static int pm8xxx_rtc_probe_offset(struct pm8xxx_rtc *rtc_dd) · line 609

Quoted 10 lines
drivers/rtc/rtc-pm8xxx.c +7 -0
+		 */
+		dev_warn(rtc_dd->dev,
+			 "UEFI offset variable not found, will be created on first clock set\n");
+		return 0;
+	}
+
+	return rc;
 }
 
 static int pm8xxx_rtc_probe(struct platform_device *pdev)

Johan