[RFC PATCH 0/3] Remove unused EFI runtime APIs

Ard Biesheuvel posted 3 patches 2 months, 3 weeks ago
arch/x86/platform/efi/efi_64.c          |  22 ----
drivers/firmware/efi/runtime-wrappers.c |  68 ------------
drivers/firmware/efi/test/efi_test.c    | 108 +-------------------
drivers/rtc/rtc-efi.c                   |  76 +-------------
drivers/xen/efi.c                       |  56 ----------
include/linux/efi.h                     |   6 --
6 files changed, 4 insertions(+), 332 deletions(-)
[RFC PATCH 0/3] Remove unused EFI runtime APIs
Posted by Ard Biesheuvel 2 months, 3 weeks ago
From: Ard Biesheuvel <ardb@kernel.org>

Using EFI runtime services to program the RTC to wake up the system is
supported in theory, but rarely works in practice. Fortunately, this
functionality is rarely [if ever] used to begin with so we can just drop
it. (Note that the EFI rtc driver is not used by x86, which programs the
CMOS rtc directly)

The same applies to GetNextHighMonoCount(), which, if implemented,
usually relies on SetVariable() under the hood *, which is often not
supported at runtime by non-x86 platforms. But it has no known users
either so let's drop support for it as well.

This means we need to drop the slightly pointless tests for it too.

* EDK2 based EFI implementations usually have a MTC variable carrying
  the monotonic counter variable, which is therefore not truly
  monotonic, given that SetVariable() will happily overwrite it. 

Cc: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
Cc: Feng Tang <feng.tang@linux.alibaba.com>
Cc: Alexandre Belloni <alexandre.belloni@bootlin.com>
Cc: Juergen Gross <jgross@suse.com>
Cc: Stefano Stabellini <sstabellini@kernel.org>
Cc: Oleksandr Tyshchenko <oleksandr_tyshchenko@epam.com> 
Cc: Sunil V L <sunilvl@ventanamicro.com>
Cc: Bibo Mao <maobibo@loongson.cn>
Cc: linux-rtc@vger.kernel.org
Cc: linux-efi@vger.kernel.org
Cc: xen-devel@lists.xenproject.org
Cc: x86@kernel.org
Cc: linux-riscv@lists.infradead.org
Cc: loongarch@lists.linux.dev

Ard Biesheuvel (3):
  efi-rtc: Remove wakeup functionality
  efi/test: Don't bother pseudo-testing unused EFI services
  efi: Remove support for pointless, unused EFI services

 arch/x86/platform/efi/efi_64.c          |  22 ----
 drivers/firmware/efi/runtime-wrappers.c |  68 ------------
 drivers/firmware/efi/test/efi_test.c    | 108 +-------------------
 drivers/rtc/rtc-efi.c                   |  76 +-------------
 drivers/xen/efi.c                       |  56 ----------
 include/linux/efi.h                     |   6 --
 6 files changed, 4 insertions(+), 332 deletions(-)

-- 
2.50.0.727.gbf7dc18ff4-goog
Re: [RFC PATCH 0/3] Remove unused EFI runtime APIs
Posted by Heinrich Schuchardt 2 months, 3 weeks ago
On 7/14/25 08:08, Ard Biesheuvel wrote:
> From: Ard Biesheuvel <ardb@kernel.org>
> 
> Using EFI runtime services to program the RTC to wake up the system is
> supported in theory, but rarely works in practice. Fortunately, this
> functionality is rarely [if ever] used to begin with so we can just drop
> it. (Note that the EFI rtc driver is not used by x86, which programs the
> CMOS rtc directly)

The main problem I see with firmware offering access to the RTC via UEFI 
services is that two different drivers, the firmware one and the Linux 
one might be trying to access the same busses or registers which might 
lead to unexpected results.

Recently there was a discussion in the RISC-V technical group for the 
server platform specification where the same issue was discussed 
concerning SetTime().

As a UEFI firmware should not care which operating system is booted, it 
should be up to the OS to disable EFI access to the RTC if it has native 
access.

Could we disable all the EFI services for the RTC in Linux dynamically 
when an RTC driver is successfully probed?

Best regards

Heinrich

> 
> The same applies to GetNextHighMonoCount(), which, if implemented,
> usually relies on SetVariable() under the hood *, which is often not
> supported at runtime by non-x86 platforms. But it has no known users
> either so let's drop support for it as well.
> 
> This means we need to drop the slightly pointless tests for it too.
> 
> * EDK2 based EFI implementations usually have a MTC variable carrying
>    the monotonic counter variable, which is therefore not truly
>    monotonic, given that SetVariable() will happily overwrite it.
> 
> Cc: Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
> Cc: Feng Tang <feng.tang@linux.alibaba.com>
> Cc: Alexandre Belloni <alexandre.belloni@bootlin.com>
> Cc: Juergen Gross <jgross@suse.com>
> Cc: Stefano Stabellini <sstabellini@kernel.org>
> Cc: Oleksandr Tyshchenko <oleksandr_tyshchenko@epam.com>
> Cc: Sunil V L <sunilvl@ventanamicro.com>
> Cc: Bibo Mao <maobibo@loongson.cn>
> Cc: linux-rtc@vger.kernel.org
> Cc: linux-efi@vger.kernel.org
> Cc: xen-devel@lists.xenproject.org
> Cc: x86@kernel.org
> Cc: linux-riscv@lists.infradead.org
> Cc: loongarch@lists.linux.dev
> 
> Ard Biesheuvel (3):
>    efi-rtc: Remove wakeup functionality
>    efi/test: Don't bother pseudo-testing unused EFI services
>    efi: Remove support for pointless, unused EFI services
> 
>   arch/x86/platform/efi/efi_64.c          |  22 ----
>   drivers/firmware/efi/runtime-wrappers.c |  68 ------------
>   drivers/firmware/efi/test/efi_test.c    | 108 +-------------------
>   drivers/rtc/rtc-efi.c                   |  76 +-------------
>   drivers/xen/efi.c                       |  56 ----------
>   include/linux/efi.h                     |   6 --
>   6 files changed, 4 insertions(+), 332 deletions(-)
>
Re: [RFC PATCH 0/3] Remove unused EFI runtime APIs
Posted by Ard Biesheuvel 2 months, 3 weeks ago
On Mon, 14 Jul 2025 at 18:11, Heinrich Schuchardt
<heinrich.schuchardt@canonical.com> wrote:
>
> On 7/14/25 08:08, Ard Biesheuvel wrote:
> > From: Ard Biesheuvel <ardb@kernel.org>
> >
> > Using EFI runtime services to program the RTC to wake up the system is
> > supported in theory, but rarely works in practice. Fortunately, this
> > functionality is rarely [if ever] used to begin with so we can just drop
> > it. (Note that the EFI rtc driver is not used by x86, which programs the
> > CMOS rtc directly)
>
> The main problem I see with firmware offering access to the RTC via UEFI
> services is that two different drivers, the firmware one and the Linux
> one might be trying to access the same busses or registers which might
> lead to unexpected results.
>
> Recently there was a discussion in the RISC-V technical group for the
> server platform specification where the same issue was discussed
> concerning SetTime().
>
> As a UEFI firmware should not care which operating system is booted, it
> should be up to the OS to disable EFI access to the RTC if it has native
> access.
>
> Could we disable all the EFI services for the RTC in Linux dynamically
> when an RTC driver is successfully probed?
>

I don't think this would be the right way to do it.

It also depends on whether ACPI or DT is being used to describe the
platform to the OS.

ACPI does not support describing the RTC device, so it should provide
the EFI services.

DT can describe the RTC device directly, so I think it is acceptable
for such firmware to mark all RTC routines unsupported in the RT_PROP
table, and just expose the RTC device directly.

The OS shouldn't have to reason about these things: it is up to the
platform to describe itself unambiguously.
Re: [RFC PATCH 0/3] Remove unused EFI runtime APIs
Posted by Sunil V L 2 months, 3 weeks ago
On Tue, Jul 15, 2025 at 01:29:15PM +1000, Ard Biesheuvel wrote:
> On Mon, 14 Jul 2025 at 18:11, Heinrich Schuchardt
> <heinrich.schuchardt@canonical.com> wrote:
> >
> > On 7/14/25 08:08, Ard Biesheuvel wrote:
> > > From: Ard Biesheuvel <ardb@kernel.org>
> > >
> > > Using EFI runtime services to program the RTC to wake up the system is
> > > supported in theory, but rarely works in practice. Fortunately, this
> > > functionality is rarely [if ever] used to begin with so we can just drop
> > > it. (Note that the EFI rtc driver is not used by x86, which programs the
> > > CMOS rtc directly)
> >
> > The main problem I see with firmware offering access to the RTC via UEFI
> > services is that two different drivers, the firmware one and the Linux
> > one might be trying to access the same busses or registers which might
> > lead to unexpected results.
> >
> > Recently there was a discussion in the RISC-V technical group for the
> > server platform specification where the same issue was discussed
> > concerning SetTime().
> >
> > As a UEFI firmware should not care which operating system is booted, it
> > should be up to the OS to disable EFI access to the RTC if it has native
> > access.
> >
> > Could we disable all the EFI services for the RTC in Linux dynamically
> > when an RTC driver is successfully probed?
> >
> 
> I don't think this would be the right way to do it.
> 
> It also depends on whether ACPI or DT is being used to describe the
> platform to the OS.
> 
> ACPI does not support describing the RTC device, so it should provide
> the EFI services.
> 
Hi Ard,
IIUC, TAD is defined for this purpose, right? 
https://uefi.org/specs/ACPI/6.6/09_ACPI_Defined_Devices_and_Device_Specific_Objects.html#time-and-alarm-device

> DT can describe the RTC device directly, so I think it is acceptable
> for such firmware to mark all RTC routines unsupported in the RT_PROP
> table, and just expose the RTC device directly.
> 
> The OS shouldn't have to reason about these things: it is up to the
> platform to describe itself unambiguously.

I agree. But I think even with ACPI, EFI GetTime/SetTime can return
unsupported if there is a TAD exposed with proper _GRT/_SRT and _GCP.

Thanks,
Sunil
Re: [RFC PATCH 0/3] Remove unused EFI runtime APIs
Posted by Ard Biesheuvel 2 months, 3 weeks ago
On Wed, 16 Jul 2025 at 00:58, Sunil V L <sunilvl@ventanamicro.com> wrote:
>
> On Tue, Jul 15, 2025 at 01:29:15PM +1000, Ard Biesheuvel wrote:
> > On Mon, 14 Jul 2025 at 18:11, Heinrich Schuchardt
> > <heinrich.schuchardt@canonical.com> wrote:
> > >
> > > On 7/14/25 08:08, Ard Biesheuvel wrote:
> > > > From: Ard Biesheuvel <ardb@kernel.org>
> > > >
> > > > Using EFI runtime services to program the RTC to wake up the system is
> > > > supported in theory, but rarely works in practice. Fortunately, this
> > > > functionality is rarely [if ever] used to begin with so we can just drop
> > > > it. (Note that the EFI rtc driver is not used by x86, which programs the
> > > > CMOS rtc directly)
> > >
> > > The main problem I see with firmware offering access to the RTC via UEFI
> > > services is that two different drivers, the firmware one and the Linux
> > > one might be trying to access the same busses or registers which might
> > > lead to unexpected results.
> > >
> > > Recently there was a discussion in the RISC-V technical group for the
> > > server platform specification where the same issue was discussed
> > > concerning SetTime().
> > >
> > > As a UEFI firmware should not care which operating system is booted, it
> > > should be up to the OS to disable EFI access to the RTC if it has native
> > > access.
> > >
> > > Could we disable all the EFI services for the RTC in Linux dynamically
> > > when an RTC driver is successfully probed?
> > >
> >
> > I don't think this would be the right way to do it.
> >
> > It also depends on whether ACPI or DT is being used to describe the
> > platform to the OS.
> >
> > ACPI does not support describing the RTC device, so it should provide
> > the EFI services.
> >
> Hi Ard,
> IIUC, TAD is defined for this purpose, right?
> https://uefi.org/specs/ACPI/6.6/09_ACPI_Defined_Devices_and_Device_Specific_Objects.html#time-and-alarm-device
>
> > DT can describe the RTC device directly, so I think it is acceptable
> > for such firmware to mark all RTC routines unsupported in the RT_PROP
> > table, and just expose the RTC device directly.
> >
> > The OS shouldn't have to reason about these things: it is up to the
> > platform to describe itself unambiguously.
>
> I agree. But I think even with ACPI, EFI GetTime/SetTime can return
> unsupported if there is a TAD exposed with proper _GRT/_SRT and _GCP.
>

Thanks for the pointer. This device did not exist yet when I last
looked at this stuff.

So it seems like TAD is a suitable way for exposing a RTC to the OS
without the need for a hardware specific driver. However, the existing
Linux driver does not appear to support get/set time, and is not
hooked up to the RTC subsystem [yet].