[0/3] x86: Don't support kernel-mode FPU with hardirqs disabled

This series returns to my earlier suggestion to make x86 not support
kernel-mode FPU when hardirqs are disabled, aligning it with arm64
(https://lore.kernel.org/r/20250220051325.340691-2-ebiggers@kernel.org).
To make this possible despite the use of the kernel-mode FPU functions
by __save_processor_state() (which I mentioned at
https://lore.kernel.org/r/20250228035924.GC5588@sol.localdomain), I've
changed __save_processor_state() to use a new function instead of
(mis)using the kernel-mode FPU functions.

This slightly reduces the overhead of kernel-mode FPU (since the result
is fewer checks), and it fixes the issue reported at
https://lore.kernel.org/r/20250516112217.GBaCcf6Yoc6LkIIryP@fat_crate.local/
where irq_fpu_usable() incorrectly returned false during CPU
initialization, causing a crash in the SHA-256 library code.

Eric Biggers (3):
  x86/fpu: Add fpu_save_state() for __save_processor_state()
  x86/pm: Use fpu_save_state() in __save_processor_state()
  x86/fpu: Don't support kernel-mode FPU when irqs_disabled()

 arch/x86/include/asm/fpu/api.h |  1 +
 arch/x86/kernel/fpu/core.c     | 92 ++++++++++++++++++++--------------
 arch/x86/power/cpu.c           | 18 +++----
 3 files changed, 62 insertions(+), 49 deletions(-)


base-commit: 8566fc3b96539e3235909d6bdda198e1282beaed

On Fri, May 16, 2025 at 04:18:55PM -0700, Eric Biggers wrote:
> This series returns to my earlier suggestion to make x86 not support
> kernel-mode FPU when hardirqs are disabled, aligning it with arm64
> (https://lore.kernel.org/r/20250220051325.340691-2-ebiggers@kernel.org).
> To make this possible despite the use of the kernel-mode FPU functions
> by __save_processor_state() (which I mentioned at
> https://lore.kernel.org/r/20250228035924.GC5588@sol.localdomain), I've
> changed __save_processor_state() to use a new function instead of
> (mis)using the kernel-mode FPU functions.
> 
> This slightly reduces the overhead of kernel-mode FPU (since the result
> is fewer checks), and it fixes the issue reported at
> https://lore.kernel.org/r/20250516112217.GBaCcf6Yoc6LkIIryP@fat_crate.local/
> where irq_fpu_usable() incorrectly returned false during CPU
> initialization, causing a crash in the SHA-256 library code.
> 
> Eric Biggers (3):
>   x86/fpu: Add fpu_save_state() for __save_processor_state()
>   x86/pm: Use fpu_save_state() in __save_processor_state()
>   x86/fpu: Don't support kernel-mode FPU when irqs_disabled()
> 
>  arch/x86/include/asm/fpu/api.h |  1 +
>  arch/x86/kernel/fpu/core.c     | 92 ++++++++++++++++++++--------------
>  arch/x86/power/cpu.c           | 18 +++----
>  3 files changed, 62 insertions(+), 49 deletions(-)

I realized I forgot about EFI again.  Ard had mentioned that earlier.

I think we'll need a !irq_disabled()-safe solution for efi_fpu_begin().  It
could be a different function from the regular kernel_fpu_begin(), though.

- Eric