Skip to contentSkip to conversation

binder: Use kmap_local_page() in binder_alloc.c

PATCH Superseded Fabio M. De Francesco 3 participants 10 messages Apr 23, 2022 – Apr 25, 2022 Android drivers

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

cover binder: Use kmap_local_page() in binder_alloc.c

Use kmap_local_page() in binder_alloc.c because kmap() and kmap_atomic() are being deprecated and kmap_local_page is preferred where it is feasible.

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.

Fabio M. De Francesco (3):
binder: Use memset_page() in binder_alloc_clear_buf()
binder: Use kmap_local_page() in binder_alloc_copy_user_to_buffer()
binder: Use kmap_local_page() in binder_alloc_get_page()

 drivers/android/binder_alloc.c | 15 ++++++---------
 1 file changed, 6 insertions(+), 9 deletions(-)
Signature · 1 line
2.34.1
#1 lore raw

PATCH 1/3 binder: Use memset_page() in binder_alloc_clear_buf()

The use of kmap() is being deprecated in favor of kmap_local_page() where it is feasible. With kmap_local_page(), the mapping is per thread, CPU local and not globally visible.

binder_alloc_clear_buf() is a function where the use of kmap_local_page() in place of kmap() is correctly suited because the mapping is local to the thread.

Therefore, use kmap_local_page() / kunmap_local() but, instead of open coding these two functions and adding a memset() of the virtual address of the mapping, prefer memset_page().

---
 drivers/android/binder_alloc.c | 5 +----
 1 file changed, 1 insertion(+), 4 deletions(-)
drivers/android/binder_alloc.c +1 -4
index 2ac1008a5f39..0b3f2f569053 100644
@@ -1175,14 +1175,11 @@ static void binder_alloc_clear_buf(struct binder_alloc *alloc,
 		unsigned long size;
 		struct page *page;
 		pgoff_t pgoff;
-		void *kptr;
 
 		page = binder_alloc_get_page(alloc, buffer,
 					     buffer_offset, &pgoff);
 		size = min_t(size_t, bytes, PAGE_SIZE - pgoff);
-		kptr = kmap(page) + pgoff;
-		memset(kptr, 0, size);
-		kunmap(page);
+		memset_page(page, pgoff, 0, size);
 		bytes -= size;
 		buffer_offset += size;
 	}
Signature · 1 line
2.34.1
Acked-by on v1 1/3
#2 lore raw

Re: [PATCH 1/3] binder: Use memset_page() in binder_alloc_clear_buf()

Quoted 9 lines · Fabio M. De Francesco

The use of kmap() is being deprecated in favor of kmap_local_page() where it is feasible. With kmap_local_page(), the mapping is per thread, CPU local and not globally visible.

binder_alloc_clear_buf() is a function where the use of kmap_local_page() in place of kmap() is correctly suited because the mapping is local to the thread.

Quoted 5 lines

Therefore, use kmap_local_page() / kunmap_local() but, instead of open coding these two functions and adding a memset() of the virtual address of the mapping, prefer memset_page().

Quoted 27 lines
---
 drivers/android/binder_alloc.c | 5 +----
 1 file changed, 1 insertion(+), 4 deletions(-)
drivers/android/binder_alloc.c +1 -5
index 2ac1008a5f39..0b3f2f569053 100644
@@ -1175,14 +1175,11 @@ static void binder_alloc_clear_buf(struct binder_alloc *alloc,
                unsigned long size;
                struct page *page;
                pgoff_t pgoff;
-               void *kptr;
 
                page = binder_alloc_get_page(alloc, buffer,
                                             buffer_offset, &pgoff);
                size = min_t(size_t, bytes, PAGE_SIZE - pgoff);
-               kptr = kmap(page) + pgoff;
-               memset(kptr, 0, size);
-               kunmap(page);
+               memset_page(page, pgoff, 0, size);
                bytes -= size;
                buffer_offset += size;
        }
--

2.34.1

#3 lore raw

PATCH 2/3 binder: Use kmap_local_page() in binder_alloc_copy_user_to_buffer()

In reply to Fabio M. De Francesco

The use of kmap() is being deprecated in favor of kmap_local_page() where it is feasible. With kmap_local_page(), the mapping is per thread, CPU local and not globally visible.

binder_alloc_copy_user_to_buffer() is a function where the use of kmap_local_page() in place of kmap() is correctly suited because the mapping is local to the thread.

Therefore, use kmap_local_page() / kunmap_local().

---
 drivers/android/binder_alloc.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/android/binder_alloc.c +2 -2
index 0b3f2f569053..0875c463c002 100644
@@ -1217,9 +1217,9 @@ binder_alloc_copy_user_to_buffer(struct binder_alloc *alloc,
 		page = binder_alloc_get_page(alloc, buffer,
 					     buffer_offset, &pgoff);
 		size = min_t(size_t, bytes, PAGE_SIZE - pgoff);
-		kptr = kmap(page) + pgoff;
+		kptr = kmap_local_page(page) + pgoff;
 		ret = copy_from_user(kptr, from, size);
-		kunmap(page);
+		kunmap_local(kptr);
 		if (ret)
 			return bytes - size + ret;
 		bytes -= size;
Signature · 1 line
2.34.1
Acked-by on v1 2/3
#4 lore raw

Re: [PATCH 2/3] binder: Use kmap_local_page() in binder_alloc_copy_user_to_buffer()

Quoted 5 lines · Fabio M. De Francesco

The use of kmap() is being deprecated in favor of kmap_local_page() where it is feasible. With kmap_local_page(), the mapping is per thread, CPU local and not globally visible.

Quoted 7 lines

binder_alloc_copy_user_to_buffer() is a function where the use of kmap_local_page() in place of kmap() is correctly suited because the mapping is local to the thread.

Therefore, use kmap_local_page() / kunmap_local().

Quoted 23 lines
---
 drivers/android/binder_alloc.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/android/binder_alloc.c +2 -3
index 0b3f2f569053..0875c463c002 100644
@@ -1217,9 +1217,9 @@ binder_alloc_copy_user_to_buffer(struct binder_alloc *alloc,
                page = binder_alloc_get_page(alloc, buffer,
                                             buffer_offset, &pgoff);
                size = min_t(size_t, bytes, PAGE_SIZE - pgoff);
-               kptr = kmap(page) + pgoff;
+               kptr = kmap_local_page(page) + pgoff;
                ret = copy_from_user(kptr, from, size);
-               kunmap(page);
+               kunmap_local(kptr);
                if (ret)
                        return bytes - size + ret;
                bytes -= size;
--

2.34.1

#5 lore raw

PATCH 3/3 binder: Use kmap_local_page() in binder_alloc_get_page()

In reply to Fabio M. De Francesco

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().

---
 drivers/android/binder_alloc.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
drivers/android/binder_alloc.c +3 -3
index 0875c463c002..058595cc7ff0 100644
@@ -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;
Signature · 1 line
2.34.1
Acked-by on v1 3/3
#6 lore raw

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

Quoted 7 lines · Fabio M. De Francesco

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.

Quoted 8 lines

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().

Quoted 32 lines
---
 drivers/android/binder_alloc.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
drivers/android/binder_alloc.c +3 -4
index 0875c463c002..058595cc7ff0 100644
@@ -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

#7 lore raw

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

In reply to Fabio M. De Francesco

Hi,

Le 23/04/2022 à 12:24, Fabio M. De Francesco a écrit :

Quoted 23 lines

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().

---
  drivers/android/binder_alloc.c | 6 +++---
  1 file changed, 3 insertions(+), 3 deletions(-)
drivers/android/binder_alloc.c +0 -0
index 0875c463c002..058595cc7ff0 100644

drivers/android/binder_alloc.c static int binder_alloc_do_buffer_copy(struct binder_alloc *alloc, · line 1258

Quoted 10 lines
drivers/android/binder_alloc.c +1 -1
@@ -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

Quoted 10 lines
drivers/android/binder_alloc.c +2 -2
  		/*
-		 * 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;
#8 lore raw

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

Quoted 4 lines · On sabato 23 aprile 2022 18:02:48 CEST Christophe JAILLET wrote:

Hi,

Quoted 1 line · Fabio M. De Francesco

The use of kmap_atomic() is being deprecated in favor of

kmap_local_page()

Quoted 2 lines
Quoted 2 lines

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

Quoted 5 lines
Quoted 5 lines

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

Quoted 11 lines

drivers/android/

Quoted 11 lines

(including interrupts).

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

---
  drivers/android/binder_alloc.c | 6 +++---
  1 file changed, 3 insertions(+), 3 deletions(-)
drivers/android/ renamed from drivers/android/binder_alloc.c +0 -0

binder_alloc.c

Quoted 4 lines

drivers/android/binder_alloc.c static int binder_alloc_do_buffer_copy(struct

Quoted 4 lines

index 0875c463c002..058595cc7ff0 100644

drivers/android/binder_alloc.c +0 -0
@@ -1250,17 +1250,17 @@ static int binder_alloc_do_buffer_copy(struct 

binder_alloc *alloc,

Quoted 2 lines

drivers/android/binder_alloc.c static int binder_alloc_do_buffer_copy(struct · line 1251

Quoted 2 lines
drivers/android/binder_alloc.c +0 -0
  		page = binder_alloc_get_page(alloc, buffer,
  					     buffer_offset, 

&pgoff);

Quoted 11 lines

drivers/android/binder_alloc.c static int binder_alloc_do_buffer_copy(struct binder_alloc *alloc, · line 1258

Quoted 8 lines
drivers/android/binder_alloc.c +1 -1
  		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

Quoted 19 lines

Not checked in details, but looks mostly the same.

Just my 2c.

CJ

Quoted 10 lines
drivers/android/binder_alloc.c +2 -2
  		/*
-		 * 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;
#9 lore raw

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

Quoted 63 lines · Fabio M. De Francesco
Quoted 4 lines · On sabato 23 aprile 2022 18:02:48 CEST Christophe JAILLET wrote:

Hi,

Quoted 1 line · Fabio M. De Francesco

The use of kmap_atomic() is being deprecated in favor of

kmap_local_page()

Quoted 2 lines
Quoted 2 lines

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

Quoted 5 lines
Quoted 5 lines

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

Quoted 11 lines

drivers/android/

Quoted 11 lines

(including interrupts).

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

---
  drivers/android/binder_alloc.c | 6 +++---
  1 file changed, 3 insertions(+), 3 deletions(-)
drivers/android/ renamed from drivers/android/binder_alloc.c +0 -0

binder_alloc.c

Quoted 4 lines

drivers/android/binder_alloc.c static int binder_alloc_do_buffer_copy(struct

Quoted 4 lines

index 0875c463c002..058595cc7ff0 100644

drivers/android/binder_alloc.c +0 -0
@@ -1250,17 +1250,17 @@ static int binder_alloc_do_buffer_copy(struct

binder_alloc *alloc,

Quoted 2 lines

drivers/android/binder_alloc.c static int binder_alloc_do_buffer_copy(struct · line 1251

Quoted 2 lines
drivers/android/binder_alloc.c +0 -0
            page = binder_alloc_get_page(alloc, buffer,
                                         buffer_offset,

&pgoff);

Quoted 11 lines

drivers/android/binder_alloc.c static int binder_alloc_do_buffer_copy(struct binder_alloc *alloc, · line 1258

Quoted 8 lines
drivers/android/binder_alloc.c +1 -1
            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.

Quoted 3 lines

@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.

Quoted 30 lines

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

Thanks,

Fabio

Quoted 19 lines

Not checked in details, but looks mostly the same.

Just my 2c.

CJ

Quoted 10 lines
drivers/android/binder_alloc.c +2 -2
            /*
-            * 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;