* [PATCH] highmem: Don't disable preemption on RT in kmap_atomic()
@ 2021-08-10 9:11 Sebastian Andrzej Siewior
2021-08-10 10:33 ` Vlastimil Babka
0 siblings, 1 reply; 2+ messages in thread
From: Sebastian Andrzej Siewior @ 2021-08-10 9:11 UTC (permalink / raw)
To: linux-kernel, linux-mm; +Cc: Thomas Gleixner, Peter Zijlstra, Andrew Morton
kmap_atomic() disables preemption and pagefaults for historical
reasons. The conversion to kmap_local(), which only disables
migration, cannot be done wholesale because quite some call sites need
to be updated to accommodate with the changed semantics.
On PREEMPT_RT enabled kernels the kmap_atomic() semantics are
problematic due to the implicit disabling of preemption which makes it
impossible to acquire 'sleeping' spinlocks within the kmap atomic
sections.
PREEMPT_RT replaces the preempt_disable() with a migrate_disable() for
more than a decade. It could be argued that this is a justification to
do this unconditionally, but PREEMPT_RT covers only a limited number of
architectures and it disables some functionality which limits the
coverage further.
Limit the replacement to PREEMPT_RT for now.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
---
include/linux/highmem-internal.h | 27 ++++++++++++++++++++++-----
1 file changed, 22 insertions(+), 5 deletions(-)
diff --git a/include/linux/highmem-internal.h b/include/linux/highmem-internal.h
index 7902c7d8b55f9..4aa1031d3e4c3 100644
--- a/include/linux/highmem-internal.h
+++ b/include/linux/highmem-internal.h
@@ -90,7 +90,11 @@ static inline void __kunmap_local(void *vaddr)
static inline void *kmap_atomic_prot(struct page *page, pgprot_t prot)
{
- preempt_disable();
+ if (IS_ENABLED(CONFIG_PREEMPT_RT))
+ migrate_disable();
+ else
+ preempt_disable();
+
pagefault_disable();
return __kmap_local_page_prot(page, prot);
}
@@ -102,7 +106,11 @@ static inline void *kmap_atomic(struct page *page)
static inline void *kmap_atomic_pfn(unsigned long pfn)
{
- preempt_disable();
+ if (IS_ENABLED(CONFIG_PREEMPT_RT))
+ migrate_disable();
+ else
+ preempt_disable();
+
pagefault_disable();
return __kmap_local_pfn_prot(pfn, kmap_prot);
}
@@ -111,7 +119,10 @@ static inline void __kunmap_atomic(void *addr)
{
kunmap_local_indexed(addr);
pagefault_enable();
- preempt_enable();
+ if (IS_ENABLED(CONFIG_PREEMPT_RT))
+ migrate_enable();
+ else
+ preempt_enable();
}
unsigned int __nr_free_highpages(void);
@@ -179,7 +190,10 @@ static inline void __kunmap_local(void *addr)
static inline void *kmap_atomic(struct page *page)
{
- preempt_disable();
+ if (IS_ENABLED(CONFIG_PREEMPT_RT))
+ migrate_disable();
+ else
+ preempt_disable();
pagefault_disable();
return page_address(page);
}
@@ -200,7 +214,10 @@ static inline void __kunmap_atomic(void *addr)
kunmap_flush_on_unmap(addr);
#endif
pagefault_enable();
- preempt_enable();
+ if (IS_ENABLED(CONFIG_PREEMPT_RT))
+ migrate_enable();
+ else
+ preempt_enable();
}
static inline unsigned int nr_free_highpages(void) { return 0; }
--
2.32.0
^ permalink raw reply related [flat|nested] 2+ messages in thread
* Re: [PATCH] highmem: Don't disable preemption on RT in kmap_atomic()
2021-08-10 9:11 [PATCH] highmem: Don't disable preemption on RT in kmap_atomic() Sebastian Andrzej Siewior
@ 2021-08-10 10:33 ` Vlastimil Babka
0 siblings, 0 replies; 2+ messages in thread
From: Vlastimil Babka @ 2021-08-10 10:33 UTC (permalink / raw)
To: Sebastian Andrzej Siewior, linux-kernel, linux-mm
Cc: Thomas Gleixner, Peter Zijlstra, Andrew Morton
On 8/10/21 11:11 AM, Sebastian Andrzej Siewior wrote:
> kmap_atomic() disables preemption and pagefaults for historical
> reasons. The conversion to kmap_local(), which only disables
> migration, cannot be done wholesale because quite some call sites need
> to be updated to accommodate with the changed semantics.
>
> On PREEMPT_RT enabled kernels the kmap_atomic() semantics are
> problematic due to the implicit disabling of preemption which makes it
> impossible to acquire 'sleeping' spinlocks within the kmap atomic
> sections.
>
> PREEMPT_RT replaces the preempt_disable() with a migrate_disable() for
> more than a decade. It could be argued that this is a justification to
> do this unconditionally, but PREEMPT_RT covers only a limited number of
> architectures and it disables some functionality which limits the
> coverage further.
>
> Limit the replacement to PREEMPT_RT for now.
>
> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Note I use the same pattern in my SLUB series, but for performance reasons
(migrate_disable() is an unconditional function call etc). But I can guess what
would be the answer if I suggested this pattern to get a common shared wrapper,
so I won't :P
> ---
> include/linux/highmem-internal.h | 27 ++++++++++++++++++++++-----
> 1 file changed, 22 insertions(+), 5 deletions(-)
>
> diff --git a/include/linux/highmem-internal.h b/include/linux/highmem-internal.h
> index 7902c7d8b55f9..4aa1031d3e4c3 100644
> --- a/include/linux/highmem-internal.h
> +++ b/include/linux/highmem-internal.h
> @@ -90,7 +90,11 @@ static inline void __kunmap_local(void *vaddr)
>
> static inline void *kmap_atomic_prot(struct page *page, pgprot_t prot)
> {
> - preempt_disable();
> + if (IS_ENABLED(CONFIG_PREEMPT_RT))
> + migrate_disable();
> + else
> + preempt_disable();
> +
> pagefault_disable();
> return __kmap_local_page_prot(page, prot);
> }
> @@ -102,7 +106,11 @@ static inline void *kmap_atomic(struct page *page)
>
> static inline void *kmap_atomic_pfn(unsigned long pfn)
> {
> - preempt_disable();
> + if (IS_ENABLED(CONFIG_PREEMPT_RT))
> + migrate_disable();
> + else
> + preempt_disable();
> +
> pagefault_disable();
> return __kmap_local_pfn_prot(pfn, kmap_prot);
> }
> @@ -111,7 +119,10 @@ static inline void __kunmap_atomic(void *addr)
> {
> kunmap_local_indexed(addr);
> pagefault_enable();
> - preempt_enable();
> + if (IS_ENABLED(CONFIG_PREEMPT_RT))
> + migrate_enable();
> + else
> + preempt_enable();
> }
>
> unsigned int __nr_free_highpages(void);
> @@ -179,7 +190,10 @@ static inline void __kunmap_local(void *addr)
>
> static inline void *kmap_atomic(struct page *page)
> {
> - preempt_disable();
> + if (IS_ENABLED(CONFIG_PREEMPT_RT))
> + migrate_disable();
> + else
> + preempt_disable();
> pagefault_disable();
> return page_address(page);
> }
> @@ -200,7 +214,10 @@ static inline void __kunmap_atomic(void *addr)
> kunmap_flush_on_unmap(addr);
> #endif
> pagefault_enable();
> - preempt_enable();
> + if (IS_ENABLED(CONFIG_PREEMPT_RT))
> + migrate_enable();
> + else
> + preempt_enable();
> }
>
> static inline unsigned int nr_free_highpages(void) { return 0; }
>
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2021-08-10 10:33 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-08-10 9:11 [PATCH] highmem: Don't disable preemption on RT in kmap_atomic() Sebastian Andrzej Siewior
2021-08-10 10:33 ` Vlastimil Babka
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.