[PATCH 1/2] ACPI: TAD: introduce acpi_tad_rtc_available()

[PATCH 1/2] ACPI: TAD: introduce acpi_tad_rtc_available()

[Hamza Mahfooz]

Code outside of the TAD driver, such as the EFI subsystem, may need to
know whether or not the RTC can be driven via the ACPI TAD, for example
to avoid registering a redundant RTC.

For this purpose, add a new header, include/acpi/acpi_tad.h, containing
a define for the TAD's hardware ID, the TAD capability flags moved there
from the driver, and a helper, acpi_tad_rtc_available(), that returns
whether or not acpi-tad will register an RTC class device for a TAD.

acpi-tad does that for a TAD that implements the Get/Set Real Time
methods, as advertised by bit 2 of its _GCP return value, but only if
the TAD has been enumerated as a platform device, so make the helper
check both.

Signed-off-by: Hamza Mahfooz <hamzamahfooz@linux.microsoft.com>
---
 drivers/acpi/acpi_tad.c | 15 ++-------
 include/acpi/acpi_tad.h | 67 +++++++++++++++++++++++++++++++++++++++++
 2 files changed, 70 insertions(+), 12 deletions(-)
 create mode 100644 include/acpi/acpi_tad.h

diff --git a/drivers/acpi/acpi_tad.c b/drivers/acpi/acpi_tad.c
index 45459a4ca..5fb978523 100644
--- a/drivers/acpi/acpi_tad.c
+++ b/drivers/acpi/acpi_tad.c
@@ -33,21 +33,12 @@
 #include <linux/rtc.h>
 #include <linux/suspend.h>
 
+#include <acpi/acpi_tad.h>
+
 MODULE_DESCRIPTION("ACPI Time and Alarm (TAD) Device Driver");
 MODULE_LICENSE("GPL v2");
 MODULE_AUTHOR("Rafael J. Wysocki");
 
-/* ACPI TAD capability flags (ACPI 6.6, Section 9.17.2) */
-#define ACPI_TAD_AC_WAKE	BIT(0)
-#define ACPI_TAD_DC_WAKE	BIT(1)
-#define ACPI_TAD_RT		BIT(2)
-#define ACPI_TAD_RT_IN_MS	BIT(3)
-#define ACPI_TAD_S4_S5__GWS	BIT(4)
-#define ACPI_TAD_AC_S4_WAKE	BIT(5)
-#define ACPI_TAD_AC_S5_WAKE	BIT(6)
-#define ACPI_TAD_DC_S4_WAKE	BIT(7)
-#define ACPI_TAD_DC_S5_WAKE	BIT(8)
-
 /* ACPI TAD alarm timer selection */
 #define ACPI_TAD_AC_TIMER	(u32)0
 #define ACPI_TAD_DC_TIMER	(u32)1
@@ -895,7 +886,7 @@ static int acpi_tad_probe(struct platform_device *pdev)
 }
 
 static const struct acpi_device_id acpi_tad_ids[] = {
-	{"ACPI000E", 0},
+	{ACPI_TAD_HID, 0},
 	{}
 };
 
diff --git a/include/acpi/acpi_tad.h b/include/acpi/acpi_tad.h
new file mode 100644
index 000000000..4825e10e9
--- /dev/null
+++ b/include/acpi/acpi_tad.h
@@ -0,0 +1,67 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * ACPI Time and Alarm Device (TAD) definitions (ACPI 6.6, Section 9.17)
+ */
+#ifndef __ACPI_TAD_H
+#define __ACPI_TAD_H
+
+#include <linux/acpi.h>
+#include <linux/bits.h>
+#include <linux/platform_device.h>
+#include <linux/types.h>
+
+#define ACPI_TAD_HID		"ACPI000E"
+
+/* ACPI TAD capability flags (ACPI 6.6, Section 9.17.2) */
+#define ACPI_TAD_AC_WAKE	BIT(0)
+#define ACPI_TAD_DC_WAKE	BIT(1)
+#define ACPI_TAD_RT		BIT(2)
+#define ACPI_TAD_RT_IN_MS	BIT(3)
+#define ACPI_TAD_S4_S5__GWS	BIT(4)
+#define ACPI_TAD_AC_S4_WAKE	BIT(5)
+#define ACPI_TAD_AC_S5_WAKE	BIT(6)
+#define ACPI_TAD_DC_S4_WAKE	BIT(7)
+#define ACPI_TAD_DC_S5_WAKE	BIT(8)
+
+#if IS_ENABLED(CONFIG_ACPI_TAD) && IS_ENABLED(CONFIG_RTC_CLASS)
+/**
+ * acpi_tad_rtc_available - Check if the RTC can be driven via the ACPI TAD
+ *
+ * Check if there is a TAD that the acpi-tad driver will register an RTC class
+ * device for.
+ *
+ * Context: Process context, after the ACPI namespace has been scanned.
+ *
+ * Return: %true if acpi-tad will provide an RTC, %false otherwise.
+ */
+static inline bool acpi_tad_rtc_available(void)
+{
+	struct acpi_device *adev;
+	unsigned long long caps;
+	struct device *dev;
+	acpi_status status;
+
+	adev = acpi_dev_get_first_match_dev(ACPI_TAD_HID, NULL, -1);
+	if (!adev)
+		return false;
+
+	dev = bus_find_device_by_acpi_dev(&platform_bus_type, adev);
+	if (!dev) {
+		acpi_dev_put(adev);
+		return false;
+	}
+	put_device(dev);
+
+	status = acpi_evaluate_integer(adev->handle, "_GCP", NULL, &caps);
+	acpi_dev_put(adev);
+
+	return ACPI_SUCCESS(status) && (caps & ACPI_TAD_RT);
+}
+#else
+static inline bool acpi_tad_rtc_available(void)
+{
+	return false;
+}
+#endif /* CONFIG_ACPI_TAD && CONFIG_RTC_CLASS */
+
+#endif /* __ACPI_TAD_H */
-- 
2.56.0

[PATCH 2/2] efi: don't register rtc-efi if the ACPI TAD can drive the RTC

[Hamza Mahfooz]

EFI runtime services are best avoided where possible, and ACPI defines a
standard interface to the real time clock in the form of the Time and
Alarm Device (TAD), which the acpi-tad driver has been exposing as an
RTC class device since commit 8a1e7f4b1764 ("ACPI: TAD: Add RTC class
device interface").

So, don't register the rtc-efi platform device if acpi-tad is going to
register an RTC class device for the TAD, according to
acpi_tad_rtc_available().

Suggested-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Hamza Mahfooz <hamzamahfooz@linux.microsoft.com>
---
 drivers/firmware/efi/efi.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/drivers/firmware/efi/efi.c b/drivers/firmware/efi/efi.c
index 9bc22de0d..706064479 100644
--- a/drivers/firmware/efi/efi.c
+++ b/drivers/firmware/efi/efi.c
@@ -34,6 +34,8 @@
 #include <linux/security.h>
 #include <linux/notifier.h>
 
+#include <acpi/acpi_tad.h>
+
 #include <asm/early_ioremap.h>
 
 struct efi __read_mostly efi = {
@@ -459,7 +461,8 @@ static int __init efisubsys_init(void)
 	if (!efi_enabled(EFI_BOOT))
 		return 0;
 
-	if (efi_rt_services_supported(EFI_RT_SUPPORTED_TIME_SERVICES))
+	if (efi_rt_services_supported(EFI_RT_SUPPORTED_TIME_SERVICES) &&
+	    !acpi_tad_rtc_available())
 		platform_device_register_simple("rtc-efi", 0, NULL, 0);
 
 	/* We register the efi directory at /sys/firmware/efi */
-- 
2.56.0

Re: [PATCH 1/2] ACPI: TAD: introduce acpi_tad_rtc_available()

[Ard Biesheuvel]

Hello Hamza,

On Thu, 22 Oct 2026, at 23:13, Hamza Mahfooz wrote:
> Code outside of the TAD driver, such as the EFI subsystem, may need to
> know whether or not the RTC can be driven via the ACPI TAD, for example
> to avoid registering a redundant RTC.
>
> For this purpose, add a new header, include/acpi/acpi_tad.h, containing
> a define for the TAD's hardware ID, the TAD capability flags moved there
> from the driver, and a helper, acpi_tad_rtc_available(), that returns
> whether or not acpi-tad will register an RTC class device for a TAD.
>
> acpi-tad does that for a TAD that implements the Get/Set Real Time
> methods, as advertised by bit 2 of its _GCP return value, but only if
> the TAD has been enumerated as a platform device, so make the helper
> check both.
>
> Signed-off-by: Hamza Mahfooz <hamzamahfooz@linux.microsoft.com>
> ---
>  drivers/acpi/acpi_tad.c | 15 ++-------
>  include/acpi/acpi_tad.h | 67 +++++++++++++++++++++++++++++++++++++++++
>  2 files changed, 70 insertions(+), 12 deletions(-)
>  create mode 100644 include/acpi/acpi_tad.h
>
> diff --git a/drivers/acpi/acpi_tad.c b/drivers/acpi/acpi_tad.c
> index 45459a4ca..5fb978523 100644
> --- a/drivers/acpi/acpi_tad.c
> +++ b/drivers/acpi/acpi_tad.c
> @@ -33,21 +33,12 @@
>  #include <linux/rtc.h>
>  #include <linux/suspend.h>
> 
> +#include <acpi/acpi_tad.h>
> +
>  MODULE_DESCRIPTION("ACPI Time and Alarm (TAD) Device Driver");
>  MODULE_LICENSE("GPL v2");
>  MODULE_AUTHOR("Rafael J. Wysocki");
> 
> -/* ACPI TAD capability flags (ACPI 6.6, Section 9.17.2) */
> -#define ACPI_TAD_AC_WAKE	BIT(0)
> -#define ACPI_TAD_DC_WAKE	BIT(1)
> -#define ACPI_TAD_RT		BIT(2)
> -#define ACPI_TAD_RT_IN_MS	BIT(3)
> -#define ACPI_TAD_S4_S5__GWS	BIT(4)
> -#define ACPI_TAD_AC_S4_WAKE	BIT(5)
> -#define ACPI_TAD_AC_S5_WAKE	BIT(6)
> -#define ACPI_TAD_DC_S4_WAKE	BIT(7)
> -#define ACPI_TAD_DC_S5_WAKE	BIT(8)
> -
>  /* ACPI TAD alarm timer selection */
>  #define ACPI_TAD_AC_TIMER	(u32)0
>  #define ACPI_TAD_DC_TIMER	(u32)1
> @@ -895,7 +886,7 @@ static int acpi_tad_probe(struct platform_device *pdev)
>  }
> 
>  static const struct acpi_device_id acpi_tad_ids[] = {
> -	{"ACPI000E", 0},
> +	{ACPI_TAD_HID, 0},
>  	{}
>  };
> 
> diff --git a/include/acpi/acpi_tad.h b/include/acpi/acpi_tad.h
> new file mode 100644
> index 000000000..4825e10e9
> --- /dev/null
> +++ b/include/acpi/acpi_tad.h
> @@ -0,0 +1,67 @@
> +/* SPDX-License-Identifier: GPL-2.0 */
> +/*
> + * ACPI Time and Alarm Device (TAD) definitions (ACPI 6.6, Section 9.17)
> + */
> +#ifndef __ACPI_TAD_H
> +#define __ACPI_TAD_H
> +
> +#include <linux/acpi.h>
> +#include <linux/bits.h>
> +#include <linux/platform_device.h>
> +#include <linux/types.h>
> +
> +#define ACPI_TAD_HID		"ACPI000E"
> +
> +/* ACPI TAD capability flags (ACPI 6.6, Section 9.17.2) */
> +#define ACPI_TAD_AC_WAKE	BIT(0)
> +#define ACPI_TAD_DC_WAKE	BIT(1)
> +#define ACPI_TAD_RT		BIT(2)
> +#define ACPI_TAD_RT_IN_MS	BIT(3)
> +#define ACPI_TAD_S4_S5__GWS	BIT(4)
> +#define ACPI_TAD_AC_S4_WAKE	BIT(5)
> +#define ACPI_TAD_AC_S5_WAKE	BIT(6)
> +#define ACPI_TAD_DC_S4_WAKE	BIT(7)
> +#define ACPI_TAD_DC_S5_WAKE	BIT(8)
> +
> +#if IS_ENABLED(CONFIG_ACPI_TAD) && IS_ENABLED(CONFIG_RTC_CLASS)
> +/**
> + * acpi_tad_rtc_available - Check if the RTC can be driven via the ACPI TAD
> + *
> + * Check if there is a TAD that the acpi-tad driver will register an RTC class
> + * device for.
> + *
> + * Context: Process context, after the ACPI namespace has been scanned.
> + *
> + * Return: %true if acpi-tad will provide an RTC, %false otherwise.
> + */
> +static inline bool acpi_tad_rtc_available(void)
> +{
> +	struct acpi_device *adev;
> +	unsigned long long caps;
> +	struct device *dev;
> +	acpi_status status;
> +

This needs a acpi_disabled check I think

> +	adev = acpi_dev_get_first_match_dev(ACPI_TAD_HID, NULL, -1);
> +	if (!adev)
> +		return false;
> +
> +	dev = bus_find_device_by_acpi_dev(&platform_bus_type, adev);
> +	if (!dev) {
> +		acpi_dev_put(adev);
> +		return false;
> +	}
> +	put_device(dev);
> +
> +	status = acpi_evaluate_integer(adev->handle, "_GCP", NULL, &caps);
> +	acpi_dev_put(adev);
> +
> +	return ACPI_SUCCESS(status) && (caps & ACPI_TAD_RT);
> +}
> +#else
> +static inline bool acpi_tad_rtc_available(void)
> +{
> +	return false;
> +}
> +#endif /* CONFIG_ACPI_TAD && CONFIG_RTC_CLASS */
> +
> +#endif /* __ACPI_TAD_H */

I get that the mutual exclusion is a bit tricky to model here, but
I'm not crazy about the general shape of this patch.

Can we make this look more like 'acpi_has_watchdog()' today, i.e.,
move this into the acpi-tad driver and export it, and call it from
the rtc-efi driver? That will make rtc-efi.ko depend on acpi-tad.ko
when both are built as modules, which is what we actually want to
happen in this case because we might get stuck without an RTC
otherwise if the module is denylisted etc.

E.g., add this to linux/acpi.h

#if IS_REACHABLE(CONFIG_ACPI_TAD)
bool acpi_has_rtc(void);
#else
static inline bool acpi_has_rtc(void) { return false; }
#endif

and decline the rtc-efi driver probe if it returns true.

We'll need to add a rule to RTC_EFI to ensure that both
drivers are available, i.e.,

depends on !ACPI || ACPI_TAD

Even nicer would be if we could unbind acpi-tad from ACPI000E:00
and bind rtc-efi instead at runtime, but that would require acpi_has_rtc()
to take into account whether anything is bound to the driver, so I'm
not sure if that is feasible.

Re: [PATCH 1/2] ACPI: TAD: introduce acpi_tad_rtc_available()

[Hamza Mahfooz]

Hi Ard,

On Fri, Oct 23, 2026 at 03:50:20PM +0200, Ard Biesheuvel wrote:
> Hello Hamza,
> 
> On Thu, 22 Oct 2026, at 23:13, Hamza Mahfooz wrote:
> > Code outside of the TAD driver, such as the EFI subsystem, may need to
> > know whether or not the RTC can be driven via the ACPI TAD, for example
> > to avoid registering a redundant RTC.
> >
> > For this purpose, add a new header, include/acpi/acpi_tad.h, containing
> > a define for the TAD's hardware ID, the TAD capability flags moved there
> > from the driver, and a helper, acpi_tad_rtc_available(), that returns
> > whether or not acpi-tad will register an RTC class device for a TAD.
> >
> > acpi-tad does that for a TAD that implements the Get/Set Real Time
> > methods, as advertised by bit 2 of its _GCP return value, but only if
> > the TAD has been enumerated as a platform device, so make the helper
> > check both.
> >
> > Signed-off-by: Hamza Mahfooz <hamzamahfooz@linux.microsoft.com>
> > ---
> >  drivers/acpi/acpi_tad.c | 15 ++-------
> >  include/acpi/acpi_tad.h | 67 +++++++++++++++++++++++++++++++++++++++++
> >  2 files changed, 70 insertions(+), 12 deletions(-)
> >  create mode 100644 include/acpi/acpi_tad.h
> >
> > diff --git a/drivers/acpi/acpi_tad.c b/drivers/acpi/acpi_tad.c
> > index 45459a4ca..5fb978523 100644
> > --- a/drivers/acpi/acpi_tad.c
> > +++ b/drivers/acpi/acpi_tad.c
> > @@ -33,21 +33,12 @@
> >  #include <linux/rtc.h>
> >  #include <linux/suspend.h>
> > 
> > +#include <acpi/acpi_tad.h>
> > +
> >  MODULE_DESCRIPTION("ACPI Time and Alarm (TAD) Device Driver");
> >  MODULE_LICENSE("GPL v2");
> >  MODULE_AUTHOR("Rafael J. Wysocki");
> > 
> > -/* ACPI TAD capability flags (ACPI 6.6, Section 9.17.2) */
> > -#define ACPI_TAD_AC_WAKE	BIT(0)
> > -#define ACPI_TAD_DC_WAKE	BIT(1)
> > -#define ACPI_TAD_RT		BIT(2)
> > -#define ACPI_TAD_RT_IN_MS	BIT(3)
> > -#define ACPI_TAD_S4_S5__GWS	BIT(4)
> > -#define ACPI_TAD_AC_S4_WAKE	BIT(5)
> > -#define ACPI_TAD_AC_S5_WAKE	BIT(6)
> > -#define ACPI_TAD_DC_S4_WAKE	BIT(7)
> > -#define ACPI_TAD_DC_S5_WAKE	BIT(8)
> > -
> >  /* ACPI TAD alarm timer selection */
> >  #define ACPI_TAD_AC_TIMER	(u32)0
> >  #define ACPI_TAD_DC_TIMER	(u32)1
> > @@ -895,7 +886,7 @@ static int acpi_tad_probe(struct platform_device *pdev)
> >  }
> > 
> >  static const struct acpi_device_id acpi_tad_ids[] = {
> > -	{"ACPI000E", 0},
> > +	{ACPI_TAD_HID, 0},
> >  	{}
> >  };
> > 
> > diff --git a/include/acpi/acpi_tad.h b/include/acpi/acpi_tad.h
> > new file mode 100644
> > index 000000000..4825e10e9
> > --- /dev/null
> > +++ b/include/acpi/acpi_tad.h
> > @@ -0,0 +1,67 @@
> > +/* SPDX-License-Identifier: GPL-2.0 */
> > +/*
> > + * ACPI Time and Alarm Device (TAD) definitions (ACPI 6.6, Section 9.17)
> > + */
> > +#ifndef __ACPI_TAD_H
> > +#define __ACPI_TAD_H
> > +
> > +#include <linux/acpi.h>
> > +#include <linux/bits.h>
> > +#include <linux/platform_device.h>
> > +#include <linux/types.h>
> > +
> > +#define ACPI_TAD_HID		"ACPI000E"
> > +
> > +/* ACPI TAD capability flags (ACPI 6.6, Section 9.17.2) */
> > +#define ACPI_TAD_AC_WAKE	BIT(0)
> > +#define ACPI_TAD_DC_WAKE	BIT(1)
> > +#define ACPI_TAD_RT		BIT(2)
> > +#define ACPI_TAD_RT_IN_MS	BIT(3)
> > +#define ACPI_TAD_S4_S5__GWS	BIT(4)
> > +#define ACPI_TAD_AC_S4_WAKE	BIT(5)
> > +#define ACPI_TAD_AC_S5_WAKE	BIT(6)
> > +#define ACPI_TAD_DC_S4_WAKE	BIT(7)
> > +#define ACPI_TAD_DC_S5_WAKE	BIT(8)
> > +
> > +#if IS_ENABLED(CONFIG_ACPI_TAD) && IS_ENABLED(CONFIG_RTC_CLASS)
> > +/**
> > + * acpi_tad_rtc_available - Check if the RTC can be driven via the ACPI TAD
> > + *
> > + * Check if there is a TAD that the acpi-tad driver will register an RTC class
> > + * device for.
> > + *
> > + * Context: Process context, after the ACPI namespace has been scanned.
> > + *
> > + * Return: %true if acpi-tad will provide an RTC, %false otherwise.
> > + */
> > +static inline bool acpi_tad_rtc_available(void)
> > +{
> > +	struct acpi_device *adev;
> > +	unsigned long long caps;
> > +	struct device *dev;
> > +	acpi_status status;
> > +
> 
> This needs a acpi_disabled check I think
> 
> > +	adev = acpi_dev_get_first_match_dev(ACPI_TAD_HID, NULL, -1);
> > +	if (!adev)
> > +		return false;
> > +
> > +	dev = bus_find_device_by_acpi_dev(&platform_bus_type, adev);
> > +	if (!dev) {
> > +		acpi_dev_put(adev);
> > +		return false;
> > +	}
> > +	put_device(dev);
> > +
> > +	status = acpi_evaluate_integer(adev->handle, "_GCP", NULL, &caps);
> > +	acpi_dev_put(adev);
> > +
> > +	return ACPI_SUCCESS(status) && (caps & ACPI_TAD_RT);
> > +}
> > +#else
> > +static inline bool acpi_tad_rtc_available(void)
> > +{
> > +	return false;
> > +}
> > +#endif /* CONFIG_ACPI_TAD && CONFIG_RTC_CLASS */
> > +
> > +#endif /* __ACPI_TAD_H */
> 
> I get that the mutual exclusion is a bit tricky to model here, but
> I'm not crazy about the general shape of this patch.
> 
> Can we make this look more like 'acpi_has_watchdog()' today, i.e.,
> move this into the acpi-tad driver and export it, and call it from
> the rtc-efi driver? That will make rtc-efi.ko depend on acpi-tad.ko
> when both are built as modules, which is what we actually want to
> happen in this case because we might get stuck without an RTC
> otherwise if the module is denylisted etc.
> 
> E.g., add this to linux/acpi.h
> 
> #if IS_REACHABLE(CONFIG_ACPI_TAD)
> bool acpi_has_rtc(void);
> #else
> static inline bool acpi_has_rtc(void) { return false; }
> #endif
> 
> and decline the rtc-efi driver probe if it returns true.
> 
> We'll need to add a rule to RTC_EFI to ensure that both
> drivers are available, i.e.,
> 
> depends on !ACPI || ACPI_TAD
> 
> Even nicer would be if we could unbind acpi-tad from ACPI000E:00
> and bind rtc-efi instead at runtime, but that would require acpi_has_rtc()
> to take into account whether anything is bound to the driver, so I'm
> not sure if that is feasible.

I did look into this suggestion but I couldn't think of a clean way to
implement it. Also, it appears that we would have to switch rtc-efi from
module_platform_driver_probe() to module_platform_driver() (of which I
don't know if there is an appetite for). So, I've left it be for now.