arch_faults_on_old_pte() relies on the calling context being
non-preemptible. CONFIG_PREEMPT_RT turns the PTE lock into a sleepable
spinlock, which doesn't disable preemption once acquired, triggering the
warning in arch_faults_on_old_pte().
It does however disable migration, ensuring the task remains on the same
CPU during the entirety of the critical section, making the read of
cpu_has_hw_af() safe and stable.
Make arch_faults_on_old_pte() check cant_migrate() instead of preemptible().
Suggested-by: Sebastian Andrzej Siewior <[email protected]>
Signed-off-by: Valentin Schneider <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
---
v3 -> v4: Dropped migratable(), reuse cant_migrate() (Sebastian)
---
arch/arm64/include/asm/pgtable.h | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/arch/arm64/include/asm/pgtable.h b/arch/arm64/include/asm/pgtable.h
index c4ba047a82d2..3caf6346ea95 100644
--- a/arch/arm64/include/asm/pgtable.h
+++ b/arch/arm64/include/asm/pgtable.h
@@ -1001,7 +1001,8 @@ static inline void update_mmu_cache(struct vm_area_struct *vma,
*/
static inline bool arch_faults_on_old_pte(void)
{
- WARN_ON(preemptible());
+ /* The register read below requires a stable CPU to make any sense */
+ cant_migrate();
return !cpu_has_hw_af();
}
--
2.25.1
On 2022-01-27 19:24:37 [+0000], Valentin Schneider wrote:
> arch_faults_on_old_pte() relies on the calling context being
> non-preemptible. CONFIG_PREEMPT_RT turns the PTE lock into a sleepable
> spinlock, which doesn't disable preemption once acquired, triggering the
> warning in arch_faults_on_old_pte().
>
> It does however disable migration, ensuring the task remains on the same
> CPU during the entirety of the critical section, making the read of
> cpu_has_hw_af() safe and stable.
>
> Make arch_faults_on_old_pte() check cant_migrate() instead of preemptible().
>
> Suggested-by: Sebastian Andrzej Siewior <[email protected]>
> Signed-off-by: Valentin Schneider <[email protected]>
> Link: https://lore.kernel.org/r/[email protected]
Let me pick that up so I can drop the other two. I hope the ARM64 folks
follow my lead ;)
Acked-by: Sebastian Andrzej Siewior <[email protected]>
Sebastian
On 2022-02-04 18:55:17 [+0000], Catalin Marinas wrote:
>
> The patch looks alright but what's the plan with it? Does it go into the
> preempt-rt tree, mainline? It's not a fix to be queued now, so:
It would be nice if if it could go mainline, too.
> Acked-by: Catalin Marinas <[email protected]>
Sebastian
On Thu, Jan 27, 2022 at 07:24:37PM +0000, Valentin Schneider wrote:
> arch_faults_on_old_pte() relies on the calling context being
> non-preemptible. CONFIG_PREEMPT_RT turns the PTE lock into a sleepable
> spinlock, which doesn't disable preemption once acquired, triggering the
> warning in arch_faults_on_old_pte().
>
> It does however disable migration, ensuring the task remains on the same
> CPU during the entirety of the critical section, making the read of
> cpu_has_hw_af() safe and stable.
>
> Make arch_faults_on_old_pte() check cant_migrate() instead of preemptible().
>
> Suggested-by: Sebastian Andrzej Siewior <[email protected]>
> Signed-off-by: Valentin Schneider <[email protected]>
> Link: https://lore.kernel.org/r/[email protected]
> ---
> v3 -> v4: Dropped migratable(), reuse cant_migrate() (Sebastian)
> ---
> arch/arm64/include/asm/pgtable.h | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/arch/arm64/include/asm/pgtable.h b/arch/arm64/include/asm/pgtable.h
> index c4ba047a82d2..3caf6346ea95 100644
> --- a/arch/arm64/include/asm/pgtable.h
> +++ b/arch/arm64/include/asm/pgtable.h
> @@ -1001,7 +1001,8 @@ static inline void update_mmu_cache(struct vm_area_struct *vma,
> */
> static inline bool arch_faults_on_old_pte(void)
> {
> - WARN_ON(preemptible());
> + /* The register read below requires a stable CPU to make any sense */
> + cant_migrate();
FWIW: There's a patch in the multi-generational LRU series that drops this
WARN_ON() entirely. I'm not sure if/when that lot will land, but just wanted
to let you know.
Will