[PATCH] staging: ion: make the pte default none PTE_RDONLY

[PATCH] staging: ion: make the pte default none PTE_RDONLY

[Chen Feng]

The page is already alloc at ion_alloc function,
ion_mmap map the alloced pages to user-space.

The default prot can be PTE_RDONLY. Take a look at
here:
set_pte_at()
arch/arm64/include/asm:
		if (pte_dirty(pte) && pte_write(pte))
			pte_val(pte) &= ~PTE_RDONLY;
		else
			pte_val(pte) |= PTE_RDONLY;

So with the dirty bit,it can improve the efficiency
and donnot need to handle memory fault when use access.

Signed-off-by: Chen Feng <puck.chen@hisilicon.com>
Signed-off-by: Wei Dong <weidong2@hisilicon.com>
Reviewed-by: Zhuangluan Su <suzhuangluan@hisilicon.com>
---
 drivers/staging/android/ion/ion.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/drivers/staging/android/ion/ion.c b/drivers/staging/android/ion/ion.c
index e237e9f..dba5942 100644
--- a/drivers/staging/android/ion/ion.c
+++ b/drivers/staging/android/ion/ion.c
@@ -1026,6 +1026,9 @@ static int ion_mmap(struct dma_buf *dmabuf, struct vm_area_struct *vma)
 	if (!(buffer->flags & ION_FLAG_CACHED))
 		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
 
+	/*Default writeable*/
+	vma->vm_page_prot = pte_mkdirty(vma->vm_page_prot);
+
 	mutex_lock(&buffer->lock);
 	/* now map it to userspace */
 	ret = buffer->heap->ops->map_user(buffer->heap, buffer, vma);
-- 
1.9.1

Re: [PATCH] staging: ion: make the pte default none PTE_RDONLY

[kbuild test robot]

[-- Attachment #1: Type: text/plain, Size: 1709 bytes --]

Hi Chen,

[auto build test ERROR on staging/staging-testing]
[also build test ERROR on v4.4 next-20160114]
[if your patch is applied to the wrong git tree, please drop us a note to help improving the system]

url:    https://github.com/0day-ci/linux/commits/Chen-Feng/staging-ion-make-the-pte-default-none-PTE_RDONLY/20160115-104450
config: x86_64-randconfig-x013-01140842 (attached as .config)
reproduce:
        # save the attached .config to linux build tree
        make ARCH=x86_64 

All errors (new ones prefixed by >>):

   drivers/staging/android/ion/ion.c: In function 'ion_mmap':
>> drivers/staging/android/ion/ion.c:1030:34: error: incompatible type for argument 1 of 'pte_mkdirty'
     vma->vm_page_prot = pte_mkdirty(vma->vm_page_prot);
                                     ^
   In file included from include/linux/mm.h:55:0,
                    from include/linux/memblock.h:18,
                    from drivers/staging/android/ion/ion.c:26:
   arch/x86/include/asm/pgtable.h:218:21: note: expected 'pte_t {aka struct <anonymous>}' but argument is of type 'pgprot_t {aka struct pgprot}'
    static inline pte_t pte_mkdirty(pte_t pte)
                        ^

vim +/pte_mkdirty +1030 drivers/staging/android/ion/ion.c

  1024		}
  1025	
  1026		if (!(buffer->flags & ION_FLAG_CACHED))
  1027			vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
  1028	
  1029		/*Default writeable*/
> 1030		vma->vm_page_prot = pte_mkdirty(vma->vm_page_prot);
  1031	
  1032		mutex_lock(&buffer->lock);
  1033		/* now map it to userspace */

---
0-DAY kernel test infrastructure                Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all                   Intel Corporation

[-- Attachment #2: .config.gz --]
[-- Type: application/octet-stream, Size: 28435 bytes --]

Re: [PATCH] staging: ion: make the pte default none PTE_RDONLY

[kbuild test robot]

[-- Attachment #1: Type: text/plain, Size: 2878 bytes --]

Hi Chen,

[auto build test ERROR on staging/staging-testing]
[also build test ERROR on v4.4 next-20160114]
[if your patch is applied to the wrong git tree, please drop us a note to help improving the system]

url:    https://github.com/0day-ci/linux/commits/Chen-Feng/staging-ion-make-the-pte-default-none-PTE_RDONLY/20160115-104450
config: ia64-allyesconfig (attached as .config)
reproduce:
        wget https://git.kernel.org/cgit/linux/kernel/git/wfg/lkp-tests.git/plain/sbin/make.cross -O ~/bin/make.cross
        chmod +x ~/bin/make.cross
        # save the attached .config to linux build tree
        make.cross ARCH=ia64 

All error/warnings (new ones prefixed by >>):

   In file included from arch/ia64/include/asm/ptrace.h:45:0,
                    from arch/ia64/include/asm/processor.h:19,
                    from include/linux/mutex.h:19,
                    from include/linux/kernfs.h:13,
                    from include/linux/sysfs.h:15,
                    from include/linux/kobject.h:21,
                    from include/linux/device.h:17,
                    from drivers/staging/android/ion/ion.c:18:
   drivers/staging/android/ion/ion.c: In function 'ion_mmap':
>> arch/ia64/include/asm/page.h:183:25: error: 'pgprot_t' has no member named 'pte'
    # define pte_val(x) ((x).pte)
                            ^
   arch/ia64/include/asm/page.h:191:31: note: in definition of macro '__pte'
    # define __pte(x) ((pte_t) { (x) } )
                                  ^
>> arch/ia64/include/asm/pgtable.h:311:33: note: in expansion of macro 'pte_val'
    #define pte_mkdirty(pte) (__pte(pte_val(pte) | _PAGE_D))
                                    ^
>> drivers/staging/android/ion/ion.c:1030:22: note: in expansion of macro 'pte_mkdirty'
     vma->vm_page_prot = pte_mkdirty(vma->vm_page_prot);
                         ^

vim +/pte_mkdirty +1030 drivers/staging/android/ion/ion.c

  1014			return -EINVAL;
  1015		}
  1016	
  1017		if (ion_buffer_fault_user_mappings(buffer)) {
  1018			vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND |
  1019								VM_DONTDUMP;
  1020			vma->vm_private_data = buffer;
  1021			vma->vm_ops = &ion_vma_ops;
  1022			ion_vm_open(vma);
  1023			return 0;
  1024		}
  1025	
  1026		if (!(buffer->flags & ION_FLAG_CACHED))
  1027			vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
  1028	
  1029		/*Default writeable*/
> 1030		vma->vm_page_prot = pte_mkdirty(vma->vm_page_prot);
  1031	
  1032		mutex_lock(&buffer->lock);
  1033		/* now map it to userspace */
  1034		ret = buffer->heap->ops->map_user(buffer->heap, buffer, vma);
  1035		mutex_unlock(&buffer->lock);
  1036	
  1037		if (ret)
  1038			pr_err("%s: failure mapping buffer to userspace\n",

---
0-DAY kernel test infrastructure                Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all                   Intel Corporation

[-- Attachment #2: .config.gz --]
[-- Type: application/octet-stream, Size: 44528 bytes --]

[PATCH] staging: ion: make the pte default none PTE_RDONLY

[Laura Abbott]

(adding linux-arm and a few people)

On 01/14/2016 06:42 PM, Chen Feng wrote:
> The page is already alloc at ion_alloc function,
> ion_mmap map the alloced pages to user-space.
>
> The default prot can be PTE_RDONLY. Take a look at
> here:
> set_pte_at()
> arch/arm64/include/asm:
> 		if (pte_dirty(pte) && pte_write(pte))
> 			pte_val(pte) &= ~PTE_RDONLY;
> 		else
> 			pte_val(pte) |= PTE_RDONLY;
>
> So with the dirty bit,it can improve the efficiency
> and donnot need to handle memory fault when use access.
>
> Signed-off-by: Chen Feng <puck.chen@hisilicon.com>
> Signed-off-by: Wei Dong <weidong2@hisilicon.com>
> Reviewed-by: Zhuangluan Su <suzhuangluan@hisilicon.com>
> ---
>   drivers/staging/android/ion/ion.c | 3 +++
>   1 file changed, 3 insertions(+)
>
> diff --git a/drivers/staging/android/ion/ion.c b/drivers/staging/android/ion/ion.c
> index e237e9f..dba5942 100644
> --- a/drivers/staging/android/ion/ion.c
> +++ b/drivers/staging/android/ion/ion.c
> @@ -1026,6 +1026,9 @@ static int ion_mmap(struct dma_buf *dmabuf, struct vm_area_struct *vma)
>   	if (!(buffer->flags & ION_FLAG_CACHED))
>   		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
>
> +	/*Default writeable*/
> +	vma->vm_page_prot = pte_mkdirty(vma->vm_page_prot);
> +
>   	mutex_lock(&buffer->lock);
>   	/* now map it to userspace */
>   	ret = buffer->heap->ops->map_user(buffer->heap, buffer, vma);
>

The extra fault is unfortunate but I'm skeptical about just setting
pte_mkdirty.

Catalin/Will, do you have any thoughts? Right now it seems like any
range mapped with remap_pfn_range will have this extra fault
behavior. Is marking the range dirty the best solution?

Thanks,
Laura

Re: [PATCH] staging: ion: make the pte default none PTE_RDONLY

[Laura Abbott]

(adding linux-arm and a few people)

On 01/14/2016 06:42 PM, Chen Feng wrote:
> The page is already alloc at ion_alloc function,
> ion_mmap map the alloced pages to user-space.
>
> The default prot can be PTE_RDONLY. Take a look at
> here:
> set_pte_at()
> arch/arm64/include/asm:
> 		if (pte_dirty(pte) && pte_write(pte))
> 			pte_val(pte) &= ~PTE_RDONLY;
> 		else
> 			pte_val(pte) |= PTE_RDONLY;
>
> So with the dirty bit,it can improve the efficiency
> and donnot need to handle memory fault when use access.
>
> Signed-off-by: Chen Feng <puck.chen@hisilicon.com>
> Signed-off-by: Wei Dong <weidong2@hisilicon.com>
> Reviewed-by: Zhuangluan Su <suzhuangluan@hisilicon.com>
> ---
>   drivers/staging/android/ion/ion.c | 3 +++
>   1 file changed, 3 insertions(+)
>
> diff --git a/drivers/staging/android/ion/ion.c b/drivers/staging/android/ion/ion.c
> index e237e9f..dba5942 100644
> --- a/drivers/staging/android/ion/ion.c
> +++ b/drivers/staging/android/ion/ion.c
> @@ -1026,6 +1026,9 @@ static int ion_mmap(struct dma_buf *dmabuf, struct vm_area_struct *vma)
>   	if (!(buffer->flags & ION_FLAG_CACHED))
>   		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
>
> +	/*Default writeable*/
> +	vma->vm_page_prot = pte_mkdirty(vma->vm_page_prot);
> +
>   	mutex_lock(&buffer->lock);
>   	/* now map it to userspace */
>   	ret = buffer->heap->ops->map_user(buffer->heap, buffer, vma);
>

The extra fault is unfortunate but I'm skeptical about just setting
pte_mkdirty.

Catalin/Will, do you have any thoughts? Right now it seems like any
range mapped with remap_pfn_range will have this extra fault
behavior. Is marking the range dirty the best solution?

Thanks,
Laura

[PATCH] staging: ion: make the pte default none PTE_RDONLY

[Russell King - ARM Linux]

On Fri, Jan 15, 2016 at 03:03:42PM -0800, Laura Abbott wrote:
> (adding linux-arm and a few people)
> 
> On 01/14/2016 06:42 PM, Chen Feng wrote:
> >The page is already alloc at ion_alloc function,
> >ion_mmap map the alloced pages to user-space.
> >
> >The default prot can be PTE_RDONLY. Take a look at
> >here:
> >set_pte_at()
> >arch/arm64/include/asm:
> >		if (pte_dirty(pte) && pte_write(pte))
> >			pte_val(pte) &= ~PTE_RDONLY;
> >		else
> >			pte_val(pte) |= PTE_RDONLY;
> >
> >So with the dirty bit,it can improve the efficiency
> >and donnot need to handle memory fault when use access.
> >
> >Signed-off-by: Chen Feng <puck.chen@hisilicon.com>
> >Signed-off-by: Wei Dong <weidong2@hisilicon.com>
> >Reviewed-by: Zhuangluan Su <suzhuangluan@hisilicon.com>
> >---
> >  drivers/staging/android/ion/ion.c | 3 +++
> >  1 file changed, 3 insertions(+)
> >
> >diff --git a/drivers/staging/android/ion/ion.c b/drivers/staging/android/ion/ion.c
> >index e237e9f..dba5942 100644
> >--- a/drivers/staging/android/ion/ion.c
> >+++ b/drivers/staging/android/ion/ion.c
> >@@ -1026,6 +1026,9 @@ static int ion_mmap(struct dma_buf *dmabuf, struct vm_area_struct *vma)
> >  	if (!(buffer->flags & ION_FLAG_CACHED))
> >  		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
> >
> >+	/*Default writeable*/
> >+	vma->vm_page_prot = pte_mkdirty(vma->vm_page_prot);
> >+
> >  	mutex_lock(&buffer->lock);
> >  	/* now map it to userspace */
> >  	ret = buffer->heap->ops->map_user(buffer->heap, buffer, vma);
> >
> 
> The extra fault is unfortunate but I'm skeptical about just setting
> pte_mkdirty.
> 
> Catalin/Will, do you have any thoughts? Right now it seems like any
> range mapped with remap_pfn_range will have this extra fault
> behavior. Is marking the range dirty the best solution?

What happens if the mapping requested was read only - at the very
least, I don't think this should be done unconditionally.

-- 
RMK's Patch system: http://www.arm.linux.org.uk/developer/patches/
FTTC broadband for 0.8mile line: currently at 9.6Mbps down 400kbps up
according to speedtest.net.

Re: [PATCH] staging: ion: make the pte default none PTE_RDONLY

[Russell King - ARM Linux]

On Fri, Jan 15, 2016 at 03:03:42PM -0800, Laura Abbott wrote:
> (adding linux-arm and a few people)
> 
> On 01/14/2016 06:42 PM, Chen Feng wrote:
> >The page is already alloc at ion_alloc function,
> >ion_mmap map the alloced pages to user-space.
> >
> >The default prot can be PTE_RDONLY. Take a look at
> >here:
> >set_pte_at()
> >arch/arm64/include/asm:
> >		if (pte_dirty(pte) && pte_write(pte))
> >			pte_val(pte) &= ~PTE_RDONLY;
> >		else
> >			pte_val(pte) |= PTE_RDONLY;
> >
> >So with the dirty bit,it can improve the efficiency
> >and donnot need to handle memory fault when use access.
> >
> >Signed-off-by: Chen Feng <puck.chen@hisilicon.com>
> >Signed-off-by: Wei Dong <weidong2@hisilicon.com>
> >Reviewed-by: Zhuangluan Su <suzhuangluan@hisilicon.com>
> >---
> >  drivers/staging/android/ion/ion.c | 3 +++
> >  1 file changed, 3 insertions(+)
> >
> >diff --git a/drivers/staging/android/ion/ion.c b/drivers/staging/android/ion/ion.c
> >index e237e9f..dba5942 100644
> >--- a/drivers/staging/android/ion/ion.c
> >+++ b/drivers/staging/android/ion/ion.c
> >@@ -1026,6 +1026,9 @@ static int ion_mmap(struct dma_buf *dmabuf, struct vm_area_struct *vma)
> >  	if (!(buffer->flags & ION_FLAG_CACHED))
> >  		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
> >
> >+	/*Default writeable*/
> >+	vma->vm_page_prot = pte_mkdirty(vma->vm_page_prot);
> >+
> >  	mutex_lock(&buffer->lock);
> >  	/* now map it to userspace */
> >  	ret = buffer->heap->ops->map_user(buffer->heap, buffer, vma);
> >
> 
> The extra fault is unfortunate but I'm skeptical about just setting
> pte_mkdirty.
> 
> Catalin/Will, do you have any thoughts? Right now it seems like any
> range mapped with remap_pfn_range will have this extra fault
> behavior. Is marking the range dirty the best solution?

What happens if the mapping requested was read only - at the very
least, I don't think this should be done unconditionally.

-- 
RMK's Patch system: http://www.arm.linux.org.uk/developer/patches/
FTTC broadband for 0.8mile line: currently at 9.6Mbps down 400kbps up
according to speedtest.net.

[PATCH] staging: ion: make the pte default none PTE_RDONLY

[chenfeng]

On 2016/1/16 7:23, Russell King - ARM Linux wrote:
> On Fri, Jan 15, 2016 at 03:03:42PM -0800, Laura Abbott wrote:
>> (adding linux-arm and a few people)
>>
>> On 01/14/2016 06:42 PM, Chen Feng wrote:
>>> The page is already alloc at ion_alloc function,
>>> ion_mmap map the alloced pages to user-space.
>>>
>>> The default prot can be PTE_RDONLY. Take a look at
>>> here:
>>> set_pte_at()
>>> arch/arm64/include/asm:
>>> 		if (pte_dirty(pte) && pte_write(pte))
>>> 			pte_val(pte) &= ~PTE_RDONLY;
>>> 		else
>>> 			pte_val(pte) |= PTE_RDONLY;
>>>
>>> So with the dirty bit,it can improve the efficiency
>>> and donnot need to handle memory fault when use access.
>>>
>>> Signed-off-by: Chen Feng <puck.chen@hisilicon.com>
>>> Signed-off-by: Wei Dong <weidong2@hisilicon.com>
>>> Reviewed-by: Zhuangluan Su <suzhuangluan@hisilicon.com>
>>> ---
>>>  drivers/staging/android/ion/ion.c | 3 +++
>>>  1 file changed, 3 insertions(+)
>>>
>>> diff --git a/drivers/staging/android/ion/ion.c b/drivers/staging/android/ion/ion.c
>>> index e237e9f..dba5942 100644
>>> --- a/drivers/staging/android/ion/ion.c
>>> +++ b/drivers/staging/android/ion/ion.c
>>> @@ -1026,6 +1026,9 @@ static int ion_mmap(struct dma_buf *dmabuf, struct vm_area_struct *vma)
>>>  	if (!(buffer->flags & ION_FLAG_CACHED))
>>>  		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
>>>
>>> +	/*Default writeable*/
>>> +	vma->vm_page_prot = pte_mkdirty(vma->vm_page_prot);
>>> +
>>>  	mutex_lock(&buffer->lock);
>>>  	/* now map it to userspace */
>>>  	ret = buffer->heap->ops->map_user(buffer->heap, buffer, vma);
>>>
>>
>> The extra fault is unfortunate but I'm skeptical about just setting
>> pte_mkdirty.
>>
>> Catalin/Will, do you have any thoughts? Right now it seems like any
>> range mapped with remap_pfn_range will have this extra fault
>> behavior. Is marking the range dirty the best solution?
Laura Abbott,

I agree with you, it seems all the remap_pfn_range have this fault behavior.


> 
> What happens if the mapping requested was read only - at the very
> least, I don't think this should be done unconditionally.
> 
Russell,
I am not sure about doing this unconditionally, but it can waste memory&time
while handling page fault with ion alloced page.

And the page can be used directly.

Re: [PATCH] staging: ion: make the pte default none PTE_RDONLY

[chenfeng]

On 2016/1/16 7:23, Russell King - ARM Linux wrote:
> On Fri, Jan 15, 2016 at 03:03:42PM -0800, Laura Abbott wrote:
>> (adding linux-arm and a few people)
>>
>> On 01/14/2016 06:42 PM, Chen Feng wrote:
>>> The page is already alloc at ion_alloc function,
>>> ion_mmap map the alloced pages to user-space.
>>>
>>> The default prot can be PTE_RDONLY. Take a look at
>>> here:
>>> set_pte_at()
>>> arch/arm64/include/asm:
>>> 		if (pte_dirty(pte) && pte_write(pte))
>>> 			pte_val(pte) &= ~PTE_RDONLY;
>>> 		else
>>> 			pte_val(pte) |= PTE_RDONLY;
>>>
>>> So with the dirty bit,it can improve the efficiency
>>> and donnot need to handle memory fault when use access.
>>>
>>> Signed-off-by: Chen Feng <puck.chen@hisilicon.com>
>>> Signed-off-by: Wei Dong <weidong2@hisilicon.com>
>>> Reviewed-by: Zhuangluan Su <suzhuangluan@hisilicon.com>
>>> ---
>>>  drivers/staging/android/ion/ion.c | 3 +++
>>>  1 file changed, 3 insertions(+)
>>>
>>> diff --git a/drivers/staging/android/ion/ion.c b/drivers/staging/android/ion/ion.c
>>> index e237e9f..dba5942 100644
>>> --- a/drivers/staging/android/ion/ion.c
>>> +++ b/drivers/staging/android/ion/ion.c
>>> @@ -1026,6 +1026,9 @@ static int ion_mmap(struct dma_buf *dmabuf, struct vm_area_struct *vma)
>>>  	if (!(buffer->flags & ION_FLAG_CACHED))
>>>  		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
>>>
>>> +	/*Default writeable*/
>>> +	vma->vm_page_prot = pte_mkdirty(vma->vm_page_prot);
>>> +
>>>  	mutex_lock(&buffer->lock);
>>>  	/* now map it to userspace */
>>>  	ret = buffer->heap->ops->map_user(buffer->heap, buffer, vma);
>>>
>>
>> The extra fault is unfortunate but I'm skeptical about just setting
>> pte_mkdirty.
>>
>> Catalin/Will, do you have any thoughts? Right now it seems like any
>> range mapped with remap_pfn_range will have this extra fault
>> behavior. Is marking the range dirty the best solution?
Laura Abbott,

I agree with you, it seems all the remap_pfn_range have this fault behavior.


> 
> What happens if the mapping requested was read only - at the very
> least, I don't think this should be done unconditionally.
> 
Russell,
I am not sure about doing this unconditionally, but it can waste memory&time
while handling page fault with ion alloced page.

And the page can be used directly.