2022-04-23 15:50:08

by Fabio M. De Francesco

[permalink] [raw]
Subject: [PATCH 3/3] binder: Use kmap_local_page() in binder_alloc_get_page()

The use of kmap_atomic() is being deprecated in favor of kmap_local_page()
where it is feasible. Each call of kmap_atomic() in the kernel creates
a non-preemptible section and disable pagefaults. This could be a source
of unwanted latency, so it should be only used if it is absolutely
required, otherwise kmap_local_page() should be preferred.

With kmap_local_page(), the mapping is per thread, CPU local and not
globally visible. Furthermore, the mapping can be acquired from any context
(including interrupts).

Therefore, use kmap_local_page() / kunmap_local() in place of
kmap_atomic() / kunmap_atomic().

Signed-off-by: Fabio M. De Francesco <[email protected]>
---
drivers/android/binder_alloc.c | 6 +++---
1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/drivers/android/binder_alloc.c b/drivers/android/binder_alloc.c
index 0875c463c002..058595cc7ff0 100644
--- a/drivers/android/binder_alloc.c
+++ b/drivers/android/binder_alloc.c
@@ -1250,17 +1250,17 @@ static int binder_alloc_do_buffer_copy(struct binder_alloc *alloc,
page = binder_alloc_get_page(alloc, buffer,
buffer_offset, &pgoff);
size = min_t(size_t, bytes, PAGE_SIZE - pgoff);
- base_ptr = kmap_atomic(page);
+ base_ptr = kmap_local_page(page);
tmpptr = base_ptr + pgoff;
if (to_buffer)
memcpy(tmpptr, ptr, size);
else
memcpy(ptr, tmpptr, size);
/*
- * kunmap_atomic() takes care of flushing the cache
+ * kunmap_local() takes care of flushing the cache
* if this device has VIVT cache arch
*/
- kunmap_atomic(base_ptr);
+ kunmap_local(base_ptr);
bytes -= size;
pgoff = 0;
ptr = ptr + size;
--
2.34.1


2022-04-23 18:14:09

by Christophe JAILLET

[permalink] [raw]
Subject: Re: [PATCH 3/3] binder: Use kmap_local_page() in binder_alloc_get_page()

Hi,

Le 23/04/2022 à 12:24, Fabio M. De Francesco a écrit :
> The use of kmap_atomic() is being deprecated in favor of kmap_local_page()
> where it is feasible. Each call of kmap_atomic() in the kernel creates
> a non-preemptible section and disable pagefaults. This could be a source
> of unwanted latency, so it should be only used if it is absolutely
> required, otherwise kmap_local_page() should be preferred.
>
> With kmap_local_page(), the mapping is per thread, CPU local and not
> globally visible. Furthermore, the mapping can be acquired from any context
> (including interrupts).
>
> Therefore, use kmap_local_page() / kunmap_local() in place of
> kmap_atomic() / kunmap_atomic().
>
> Signed-off-by: Fabio M. De Francesco <[email protected]>
> ---
> drivers/android/binder_alloc.c | 6 +++---
> 1 file changed, 3 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/android/binder_alloc.c b/drivers/android/binder_alloc.c
> index 0875c463c002..058595cc7ff0 100644
> --- a/drivers/android/binder_alloc.c
> +++ b/drivers/android/binder_alloc.c
> @@ -1250,17 +1250,17 @@ static int binder_alloc_do_buffer_copy(struct binder_alloc *alloc,
> page = binder_alloc_get_page(alloc, buffer,
> buffer_offset, &pgoff);
> size = min_t(size_t, bytes, PAGE_SIZE - pgoff);
> - base_ptr = kmap_atomic(page);
> + base_ptr = kmap_local_page(page);
> tmpptr = base_ptr + pgoff;
> if (to_buffer)
> memcpy(tmpptr, ptr, size);
> else
> memcpy(ptr, tmpptr, size);

in the same spirit as patch 1/3, memcpy_to_page() and memcpy_from_page()
looks good candidate to avoid code duplication.

Not checked in details, but looks mostly the same.

Just my 2c.

CJ

> /*
> - * kunmap_atomic() takes care of flushing the cache
> + * kunmap_local() takes care of flushing the cache
> * if this device has VIVT cache arch
> */
> - kunmap_atomic(base_ptr);
> + kunmap_local(base_ptr);
> bytes -= size;
> pgoff = 0;
> ptr = ptr + size;


2022-04-25 02:00:40

by Fabio M. De Francesco

[permalink] [raw]
Subject: Re: [PATCH 3/3] binder: Use kmap_local_page() in binder_alloc_get_page()

On sabato 23 aprile 2022 18:02:48 CEST Christophe JAILLET wrote:
> Hi,
>
> Le 23/04/2022 à 12:24, Fabio M. De Francesco a écrit :
> > The use of kmap_atomic() is being deprecated in favor of
kmap_local_page()
> > where it is feasible. Each call of kmap_atomic() in the kernel creates
> > a non-preemptible section and disable pagefaults. This could be a
source
> > of unwanted latency, so it should be only used if it is absolutely
> > required, otherwise kmap_local_page() should be preferred.
> >
> > With kmap_local_page(), the mapping is per thread, CPU local and not
> > globally visible. Furthermore, the mapping can be acquired from any
context
> > (including interrupts).
> >
> > Therefore, use kmap_local_page() / kunmap_local() in place of
> > kmap_atomic() / kunmap_atomic().
> >
> > Signed-off-by: Fabio M. De Francesco <[email protected]>
> > ---
> > drivers/android/binder_alloc.c | 6 +++---
> > 1 file changed, 3 insertions(+), 3 deletions(-)
> >
> > diff --git a/drivers/android/binder_alloc.c b/drivers/android/
binder_alloc.c
> > index 0875c463c002..058595cc7ff0 100644
> > --- a/drivers/android/binder_alloc.c
> > +++ b/drivers/android/binder_alloc.c
> > @@ -1250,17 +1250,17 @@ static int binder_alloc_do_buffer_copy(struct
binder_alloc *alloc,
> > page = binder_alloc_get_page(alloc, buffer,
> > buffer_offset,
&pgoff);
> > size = min_t(size_t, bytes, PAGE_SIZE - pgoff);
> > - base_ptr = kmap_atomic(page);
> > + base_ptr = kmap_local_page(page);
> > tmpptr = base_ptr + pgoff;
> > if (to_buffer)
> > memcpy(tmpptr, ptr, size);
> > else
> > memcpy(ptr, tmpptr, size);
>
> in the same spirit as patch 1/3, memcpy_to_page() and memcpy_from_page()
> looks good candidate to avoid code duplication.

Hello Christophe, Todd,

I had thought to use memcpy_to_page() and memcpy_from_page() (exactly as I
did in other conversions I have been working on during the latest couple of
weeks).

However, I decided to avoid to use them for I should also have deleted the
comment which is before "kunmap_local(base_ptr);".

I don't know how much Maintainers think it is necessary to make readers
notice that "kunmap_local() takes care of flushing the cache []" (exactly
as kunmap_atomic() does). Actually I'd delete that comment that looks
redundant and unnecessary to me, but I cannot know if Todd wants it to
remain there.

@Todd: Can you please say what you think about this topic? Should I leave
the patch as-is or I should use memcpy_{to,from}_page() and delete that
comment?

I won't send any v2 unless I have your confirmation.

Thanks,

Fabio

>
> Not checked in details, but looks mostly the same.
>
> Just my 2c.
>
> CJ
>
> > /*
> > - * kunmap_atomic() takes care of flushing the cache
> > + * kunmap_local() takes care of flushing the cache
> > * if this device has VIVT cache arch
> > */
> > - kunmap_atomic(base_ptr);
> > + kunmap_local(base_ptr);
> > bytes -= size;
> > pgoff = 0;
> > ptr = ptr + size;
>
>




2022-04-25 03:46:38

by Todd Kjos

[permalink] [raw]
Subject: Re: [PATCH 3/3] binder: Use kmap_local_page() in binder_alloc_get_page()

On Sat, Apr 23, 2022 at 3:24 AM Fabio M. De Francesco
<[email protected]> wrote:
>
> The use of kmap_atomic() is being deprecated in favor of kmap_local_page()
> where it is feasible. Each call of kmap_atomic() in the kernel creates
> a non-preemptible section and disable pagefaults. This could be a source
> of unwanted latency, so it should be only used if it is absolutely
> required, otherwise kmap_local_page() should be preferred.
>
> With kmap_local_page(), the mapping is per thread, CPU local and not
> globally visible. Furthermore, the mapping can be acquired from any context
> (including interrupts).
>
> Therefore, use kmap_local_page() / kunmap_local() in place of
> kmap_atomic() / kunmap_atomic().
>
> Signed-off-by: Fabio M. De Francesco <[email protected]>

Acked-by: Todd Kjos <[email protected]>

> ---
> drivers/android/binder_alloc.c | 6 +++---
> 1 file changed, 3 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/android/binder_alloc.c b/drivers/android/binder_alloc.c
> index 0875c463c002..058595cc7ff0 100644
> --- a/drivers/android/binder_alloc.c
> +++ b/drivers/android/binder_alloc.c
> @@ -1250,17 +1250,17 @@ static int binder_alloc_do_buffer_copy(struct binder_alloc *alloc,
> page = binder_alloc_get_page(alloc, buffer,
> buffer_offset, &pgoff);
> size = min_t(size_t, bytes, PAGE_SIZE - pgoff);
> - base_ptr = kmap_atomic(page);
> + base_ptr = kmap_local_page(page);
> tmpptr = base_ptr + pgoff;
> if (to_buffer)
> memcpy(tmpptr, ptr, size);
> else
> memcpy(ptr, tmpptr, size);
> /*
> - * kunmap_atomic() takes care of flushing the cache
> + * kunmap_local() takes care of flushing the cache
> * if this device has VIVT cache arch
> */
> - kunmap_atomic(base_ptr);
> + kunmap_local(base_ptr);
> bytes -= size;
> pgoff = 0;
> ptr = ptr + size;
> --
> 2.34.1
>

2022-04-25 17:25:18

by Todd Kjos

[permalink] [raw]
Subject: Re: [PATCH 3/3] binder: Use kmap_local_page() in binder_alloc_get_page()

On Sun, Apr 24, 2022 at 2:40 AM Fabio M. De Francesco
<[email protected]> wrote:
>
> On sabato 23 aprile 2022 18:02:48 CEST Christophe JAILLET wrote:
> > Hi,
> >
> > Le 23/04/2022 à 12:24, Fabio M. De Francesco a écrit :
> > > The use of kmap_atomic() is being deprecated in favor of
> kmap_local_page()
> > > where it is feasible. Each call of kmap_atomic() in the kernel creates
> > > a non-preemptible section and disable pagefaults. This could be a
> source
> > > of unwanted latency, so it should be only used if it is absolutely
> > > required, otherwise kmap_local_page() should be preferred.
> > >
> > > With kmap_local_page(), the mapping is per thread, CPU local and not
> > > globally visible. Furthermore, the mapping can be acquired from any
> context
> > > (including interrupts).
> > >
> > > Therefore, use kmap_local_page() / kunmap_local() in place of
> > > kmap_atomic() / kunmap_atomic().
> > >
> > > Signed-off-by: Fabio M. De Francesco <[email protected]>
> > > ---
> > > drivers/android/binder_alloc.c | 6 +++---
> > > 1 file changed, 3 insertions(+), 3 deletions(-)
> > >
> > > diff --git a/drivers/android/binder_alloc.c b/drivers/android/
> binder_alloc.c
> > > index 0875c463c002..058595cc7ff0 100644
> > > --- a/drivers/android/binder_alloc.c
> > > +++ b/drivers/android/binder_alloc.c
> > > @@ -1250,17 +1250,17 @@ static int binder_alloc_do_buffer_copy(struct
> binder_alloc *alloc,
> > > page = binder_alloc_get_page(alloc, buffer,
> > > buffer_offset,
> &pgoff);
> > > size = min_t(size_t, bytes, PAGE_SIZE - pgoff);
> > > - base_ptr = kmap_atomic(page);
> > > + base_ptr = kmap_local_page(page);
> > > tmpptr = base_ptr + pgoff;
> > > if (to_buffer)
> > > memcpy(tmpptr, ptr, size);
> > > else
> > > memcpy(ptr, tmpptr, size);
> >
> > in the same spirit as patch 1/3, memcpy_to_page() and memcpy_from_page()
> > looks good candidate to avoid code duplication.
>
> Hello Christophe, Todd,
>
> I had thought to use memcpy_to_page() and memcpy_from_page() (exactly as I
> did in other conversions I have been working on during the latest couple of
> weeks).
>
> However, I decided to avoid to use them for I should also have deleted the
> comment which is before "kunmap_local(base_ptr);".
>
> I don't know how much Maintainers think it is necessary to make readers
> notice that "kunmap_local() takes care of flushing the cache []" (exactly
> as kunmap_atomic() does). Actually I'd delete that comment that looks
> redundant and unnecessary to me, but I cannot know if Todd wants it to
> remain there.
>
> @Todd: Can you please say what you think about this topic? Should I leave
> the patch as-is or I should use memcpy_{to,from}_page() and delete that
> comment?

I'm fine with using memcpy_{to,from}_page() and removing the comment.

>
> I won't send any v2 unless I have your confirmation.
>
> Thanks,
>
> Fabio
>
> >
> > Not checked in details, but looks mostly the same.
> >
> > Just my 2c.
> >
> > CJ
> >
> > > /*
> > > - * kunmap_atomic() takes care of flushing the cache
> > > + * kunmap_local() takes care of flushing the cache
> > > * if this device has VIVT cache arch
> > > */
> > > - kunmap_atomic(base_ptr);
> > > + kunmap_local(base_ptr);
> > > bytes -= size;
> > > pgoff = 0;
> > > ptr = ptr + size;
> >
> >
>
>
>
>