[v3,26/38] arm64: Avoid cpus_have_const_cap() for ARM64_MTE

Commit Message

Mark Rutland Oct. 10, 2023, 10:31 a.m. UTC

In system_supports_mte() we use cpus_have_const_cap() to check for
ARM64_MTE, but this is not necessary and cpus_have_final_boot_cap()
would be preferable.

For historical reasons, cpus_have_const_cap() is more complicated than
it needs to be. Before cpucaps are finalized, it will perform a bitmap
test of the system_cpucaps bitmap, and once cpucaps are finalized it
will use an alternative branch. This used to be necessary to handle some
race conditions in the window between cpucap detection and the
subsequent patching of alternatives and static branches, where different
branches could be out-of-sync with one another (or w.r.t. alternative
sequences). Now that we use alternative branches instead of static
branches, these are all patched atomically w.r.t. one another, and there
are only a handful of cases that need special care in the window between
cpucap detection and alternative patching.

Due to the above, it would be nice to remove cpus_have_const_cap(), and
migrate callers over to alternative_has_cap_*(), cpus_have_final_cap(),
or cpus_have_cap() depending on when their requirements. This will
remove redundant instructions and improve code generation, and will make
it easier to determine how each callsite will behave before, during, and
after alternative patching.

The ARM64_MTE cpucap is a boot cpu feature which is detected and patched
early on the boot CPU under smp_prepare_boot_cpu(). In the window
between detecting the ARM64_MTE cpucap and patching alternatives,
nothing depends on the ARM64_MTE cpucap:

* The kasan_hw_tags_enabled() helper depends upon the kasan_flag_enabled
  static key, which is initialized later in kasan_init_hw_tags() after
  alternatives have been applied.

* No KVM code is called during this window, and KVM is not initialized
  until after system cpucaps have been detected and patched. KVM code
  can safely use cpus_have_final_cap() or alternative_has_cap_*().

* We don't context-switch prior to patching boot alternatives, and thus
  mte_thread_switch() is not reachable during this window. Thus, we can
  safely use cpus_have_final_boot_cap() or alternative_has_cap_*() in
  the context-switch code.

* IRQ and FIQ are masked during this window, and we can only take SError
  and Debug exceptions. SError exceptions are fatal at this point in
  time, and we do not expect to take Debug exceptions, thus:

  - It's fine to lave TCO set for exceptions taken during this window,
    and mte_disable_tco_entry() doesn't need to do anything.

  - We don't need to detect and report asynchronous tag cehck faults
    during this window, and neither mte_check_tfsr_entry() nor
    mte_check_tfsr_exit() need to do anything.

  Since we want to report any SErrors taken during thiw window, these
  cannot safely use cpus_have_final_boot_cap() or cpus_have_final_cap(),
  but these can safely use alternative_has_cap_*().

* The __set_pte_at() function is not used during this window. It is
  possible for this to be used on kernel mappings prior to boot cpucaps
  being finalized, so this cannot safely use cpus_have_final_boot_cap()
  or cpus_have_final_cap(), but this can safely use
  alternative_has_cap_*().

* No userspace translation tables have been created yet, and swap has
  not been initialized yet. Thus swapping is not possible and none of
  the following are called:

  - arch_thp_swp_supported()
  - arch_prepare_to_swap()
  - arch_swap_invalidate_page()
  - arch_swap_invalidate_area()
  - arch_swap_restore()

  These can safely use system_has_final_cap() or
  alternative_has_cap_*().

* The elfcore functions are only reachable after userspace is brought
  up, which happens after system cpucaps have been detected and patched.
  Thus the elfcore code can safely use cpus_have_final_cap() or
  alternative_has_cap_*().

* Hibernation is only possible after userspace is brought up, which
  happens after system cpucaps have been detected and patched. Thus the
  hibernate code can safely use cpus_have_final_cap() or
  alternative_has_cap_*().

* The set_tagged_addr_ctrl() function is only reachable after userspace
  is brought up, which happens after system cpucaps have been detected
  and patched. Thus this can safely use cpus_have_final_cap() or
  alternative_has_cap_*().

* The copy_user_highpage() and copy_highpage() functions are not used
  during this window, and can safely use alternative_has_cap_*().

This patch replaces the use of cpus_have_const_cap() with
alternative_has_cap_unlikely(), which avoid generating code to test the
system_cpucaps bitmap and should be better for all subsequent calls at
runtime.

Signed-off-by: Mark Rutland <[e-mail]>
Cc: Catalin Marinas <[e-mail]>
Cc: Peter Collingbourne <[e-mail]>
Cc: Suzuki K Poulose <[e-mail]>
Cc: Will Deacon <[e-mail]>
---
 arch/arm64/include/asm/cpufeature.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Patch

diff --git a/arch/arm64/include/asm/cpufeature.h b/arch/arm64/include/asm/cpufeature.h
index bceb007f9ea3a..386b339d5367d 100644
--- a/arch/arm64/include/asm/cpufeature.h
+++ b/arch/arm64/include/asm/cpufeature.h
@@ -826,7 +826,7 @@  static __always_inline bool system_uses_irq_prio_masking(void)
 
 static inline bool system_supports_mte(void)
 {
-	return cpus_have_const_cap(ARM64_MTE);
+	return alternative_has_cap_unlikely(ARM64_MTE);
 }
 
 static inline bool system_has_prio_mask_debugging(void)