2014-06-17 17:14:56

by Mark Salter

Subject: [PATCH] arm64: export __cpu_{clear,copy}_user_page functions

The __cpu_clear_user_page() and __cpu_copy_user_page() functions
are not currently exported. This prevents modules from using
clear_user_page() and opy_user_page().

Signed-off-by: Mark Salter <[email protected]>
---
arch/arm64/mm/copypage.c | 2 ++
1 file changed, 2 insertions(+)

diff --git a/arch/arm64/mm/copypage.c b/arch/arm64/mm/copypage.c
index 9aecbac..13bbc3be 100644
--- a/arch/arm64/mm/copypage.c
+++ b/arch/arm64/mm/copypage.c
@@ -27,8 +27,10 @@ void __cpu_copy_user_page(void *kto, const void *kfrom, unsigned long vaddr)
copy_page(kto, kfrom);
__flush_dcache_area(kto, PAGE_SIZE);
}
+EXPORT_SYMBOL_GPL(__cpu_copy_user_page);

void __cpu_clear_user_page(void *kaddr, unsigned long vaddr)
{
clear_page(kaddr);
}
+EXPORT_SYMBOL_GPL(__cpu_clear_user_page);
--
1.9.0

2014-06-18 10:40:05

by Catalin Marinas

Subject: Re: [PATCH] arm64: export __cpu_{clear,copy}_user_page functions

On Tue, Jun 17, 2014 at 06:14:26PM +0100, Mark Salter wrote:
> The __cpu_clear_user_page() and __cpu_copy_user_page() functions
> are not currently exported. This prevents modules from using
> clear_user_page() and opy_user_page().
>
> Signed-off-by: Mark Salter <[email protected]>
> ---
> arch/arm64/mm/copypage.c | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/arch/arm64/mm/copypage.c b/arch/arm64/mm/copypage.c
> index 9aecbac..13bbc3be 100644
> --- a/arch/arm64/mm/copypage.c
> +++ b/arch/arm64/mm/copypage.c
> @@ -27,8 +27,10 @@ void __cpu_copy_user_page(void *kto, const void *kfrom, unsigned long vaddr)
> copy_page(kto, kfrom);
> __flush_dcache_area(kto, PAGE_SIZE);
> }
> +EXPORT_SYMBOL_GPL(__cpu_copy_user_page);
>
> void __cpu_clear_user_page(void *kaddr, unsigned long vaddr)
> {
> clear_page(kaddr);
> }
> +EXPORT_SYMBOL_GPL(__cpu_clear_user_page);

I wonder why we still have these functions. I don't think we need any
D-cache flushing for copy_user_page(), so we could simply remove this
code and define copy_user_page() as copy_page() (similarly for
clear_user_page). I'll post a patch separately removing the copy_page,
copy_user_page etc. functions (now that we also have optimised memcpy
and memset).

--
Catalin

2014-06-18 13:54:29

by Mark Salter

Subject: Re: [PATCH] arm64: export __cpu_{clear,copy}_user_page functions

On Wed, 2014-06-18 at 11:39 +0100, Catalin Marinas wrote:
> On Tue, Jun 17, 2014 at 06:14:26PM +0100, Mark Salter wrote:
> > The __cpu_clear_user_page() and __cpu_copy_user_page() functions
> > are not currently exported. This prevents modules from using
> > clear_user_page() and opy_user_page().
> >
> > Signed-off-by: Mark Salter <[email protected]>
> > ---
> > arch/arm64/mm/copypage.c | 2 ++
> > 1 file changed, 2 insertions(+)
> >
> > diff --git a/arch/arm64/mm/copypage.c b/arch/arm64/mm/copypage.c
> > index 9aecbac..13bbc3be 100644
> > --- a/arch/arm64/mm/copypage.c
> > +++ b/arch/arm64/mm/copypage.c
> > @@ -27,8 +27,10 @@ void __cpu_copy_user_page(void *kto, const void *kfrom, unsigned long vaddr)
> > copy_page(kto, kfrom);
> > __flush_dcache_area(kto, PAGE_SIZE);
> > }
> > +EXPORT_SYMBOL_GPL(__cpu_copy_user_page);
> >
> > void __cpu_clear_user_page(void *kaddr, unsigned long vaddr)
> > {
> > clear_page(kaddr);
> > }
> > +EXPORT_SYMBOL_GPL(__cpu_clear_user_page);
>
> I wonder why we still have these functions. I don't think we need any
> D-cache flushing for copy_user_page(), so we could simply remove this
> code and define copy_user_page() as copy_page() (similarly for
> clear_user_page). I'll post a patch separately removing the copy_page,
> copy_user_page etc. functions (now that we also have optimised memcpy
> and memset).
>

Even better. I much prefer removing code than adding it. Thanks.

2014-06-23 11:31:20

by Catalin Marinas

Subject: Re: [PATCH] arm64: export __cpu_{clear,copy}_user_page functions

On Wed, Jun 18, 2014 at 02:54:03PM +0100, Mark Salter wrote:
> On Wed, 2014-06-18 at 11:39 +0100, Catalin Marinas wrote:
> > On Tue, Jun 17, 2014 at 06:14:26PM +0100, Mark Salter wrote:
> > > The __cpu_clear_user_page() and __cpu_copy_user_page() functions
> > > are not currently exported. This prevents modules from using
> > > clear_user_page() and opy_user_page().
> > >
> > > Signed-off-by: Mark Salter <[email protected]>
> > > ---
> > > arch/arm64/mm/copypage.c | 2 ++
> > > 1 file changed, 2 insertions(+)
> > >
> > > diff --git a/arch/arm64/mm/copypage.c b/arch/arm64/mm/copypage.c
> > > index 9aecbac..13bbc3be 100644
> > > --- a/arch/arm64/mm/copypage.c
> > > +++ b/arch/arm64/mm/copypage.c
> > > @@ -27,8 +27,10 @@ void __cpu_copy_user_page(void *kto, const void *kfrom, unsigned long vaddr)
> > > copy_page(kto, kfrom);
> > > __flush_dcache_area(kto, PAGE_SIZE);
> > > }
> > > +EXPORT_SYMBOL_GPL(__cpu_copy_user_page);
> > >
> > > void __cpu_clear_user_page(void *kaddr, unsigned long vaddr)
> > > {
> > > clear_page(kaddr);
> > > }
> > > +EXPORT_SYMBOL_GPL(__cpu_clear_user_page);
> >
> > I wonder why we still have these functions. I don't think we need any
> > D-cache flushing for copy_user_page(), so we could simply remove this
> > code and define copy_user_page() as copy_page() (similarly for
> > clear_user_page). I'll post a patch separately removing the copy_page,
> > copy_user_page etc. functions (now that we also have optimised memcpy
> > and memset).
>
> Even better. I much prefer removing code than adding it. Thanks.

For the time being I'll merge your patch since using memset for
clear_page adds a small overhead with checking the alignment and the
fill value (not much but consistently slower in my mmap benchmarks).

--
Catalin