[RFC PATCH] mtd: allow mmap for ram devices in memory address space.

[RFC PATCH] mtd: allow mmap for ram devices in memory address space.

[Philippe De Muyter]

Implement mmap for mtd-ram, by using mtd->_get_unmapped_area, as asked by
an ancient comment, to get the physical address (if any) in memory of the
mtd device.

Therefore, mapram_unmapped_area had to be changed to return the physical
address, not the virtual one, For the NOMMU case, that makes no difference,
but for the MMU case, this is needed by mapram_unmapped_area to implement mmap.

Signed-off-by: Philippe De Muyter <phdm@macqel.be>
Cc: David Howells <dhowells@redhat.com>
Cc: Markus Niebel <list-09@tqsc.de>
---
 drivers/mtd/chips/map_ram.c |    2 +-
 drivers/mtd/mtdchar.c       |   12 +++++++++++-
 2 files changed, 12 insertions(+), 2 deletions(-)

diff --git a/drivers/mtd/chips/map_ram.c b/drivers/mtd/chips/map_ram.c
index 991c2a1..e93d147 100644
--- a/drivers/mtd/chips/map_ram.c
+++ b/drivers/mtd/chips/map_ram.c
@@ -92,7 +92,7 @@ static unsigned long mapram_unmapped_area(struct mtd_info *mtd,
 					  unsigned long flags)
 {
 	struct map_info *map = mtd->priv;
-	return (unsigned long) map->virt + offset;
+	return (unsigned long) map->phys + offset;
 }
 
 static int mapram_read (struct mtd_info *mtd, loff_t from, size_t len, size_t *retlen, u_char *buf)
diff --git a/drivers/mtd/mtdchar.c b/drivers/mtd/mtdchar.c
index 2147e73..00eac7a 100644
--- a/drivers/mtd/mtdchar.c
+++ b/drivers/mtd/mtdchar.c
@@ -1112,13 +1112,23 @@ static int mtdchar_mmap(struct file *file, struct vm_area_struct *vma)
 #ifdef CONFIG_MMU
 	struct mtd_file_info *mfi = file->private_data;
 	struct mtd_info *mtd = mfi->mtd;
-	struct map_info *map = mtd->priv;
+	loff_t offset = vma->vm_pgoff << PAGE_SHIFT;
+	loff_t len = vma->vm_end - vma->vm_start;
 
+	if (offset > mtd->size - len)
+		return (unsigned long) -EINVAL;
+	if (mtd->_get_unmapped_area) {
+		phys_addr_t start;
+
+		start = mtd->_get_unmapped_area(mtd, len, offset, vma->vm_flags);
+		return vm_iomap_memory(vma, start, len);
+	}
         /* This is broken because it assumes the MTD device is map-based
 	   and that mtd->priv is a valid struct map_info.  It should be
 	   replaced with something that uses the mtd_get_unmapped_area()
 	   operation properly. */
 	if (0 /*mtd->type == MTD_RAM || mtd->type == MTD_ROM*/) {
+		struct map_info *map = mtd->priv;
 #ifdef pgprot_noncached
 		if (file->f_flags & O_DSYNC || map->phys >= __pa(high_memory))
 			vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
-- 
1.7.5.3

Re: [RFC PATCH] mtd: allow mmap for ram devices in memory address space.

[Brian Norris]

On Thu, Mar 06, 2014 at 08:32:26AM +0100, Philippe De Muyter wrote:
> Implement mmap for mtd-ram, by using mtd->_get_unmapped_area, as asked by
> an ancient comment, to get the physical address (if any) in memory of the
> mtd device.
> 
> Therefore, mapram_unmapped_area had to be changed to return the physical
> address, not the virtual one, For the NOMMU case, that makes no difference,
> but for the MMU case, this is needed by mapram_unmapped_area to implement mmap.
> 
> Signed-off-by: Philippe De Muyter <phdm@macqel.be>
> Cc: David Howells <dhowells@redhat.com>
> Cc: Markus Niebel <list-09@tqsc.de>
> ---
>  drivers/mtd/chips/map_ram.c |    2 +-
>  drivers/mtd/mtdchar.c       |   12 +++++++++++-
>  2 files changed, 12 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/mtd/chips/map_ram.c b/drivers/mtd/chips/map_ram.c
> index 991c2a1..e93d147 100644
> --- a/drivers/mtd/chips/map_ram.c
> +++ b/drivers/mtd/chips/map_ram.c
> @@ -92,7 +92,7 @@ static unsigned long mapram_unmapped_area(struct mtd_info *mtd,
>  					  unsigned long flags)
>  {
>  	struct map_info *map = mtd->priv;
> -	return (unsigned long) map->virt + offset;
> +	return (unsigned long) map->phys + offset;
>  }
>  
>  static int mapram_read (struct mtd_info *mtd, loff_t from, size_t len, size_t *retlen, u_char *buf)
> diff --git a/drivers/mtd/mtdchar.c b/drivers/mtd/mtdchar.c
> index 2147e73..00eac7a 100644
> --- a/drivers/mtd/mtdchar.c
> +++ b/drivers/mtd/mtdchar.c
> @@ -1112,13 +1112,23 @@ static int mtdchar_mmap(struct file *file, struct vm_area_struct *vma)
>  #ifdef CONFIG_MMU
>  	struct mtd_file_info *mfi = file->private_data;
>  	struct mtd_info *mtd = mfi->mtd;
> -	struct map_info *map = mtd->priv;
> +	loff_t offset = vma->vm_pgoff << PAGE_SHIFT;
> +	loff_t len = vma->vm_end - vma->vm_start;
>  
> +	if (offset > mtd->size - len)
> +		return (unsigned long) -EINVAL;

What's the cast doing? This function returns 'int'.

And anyway, you're duplicating the checks that are already in
mtd_get_unmapped_area(). See below.

> +	if (mtd->_get_unmapped_area) {
> +		phys_addr_t start;
> +
> +		start = mtd->_get_unmapped_area(mtd, len, offset, vma->vm_flags);

You should be using mtd_get_unmapped_area(), as the comment below says.
The underscore in front of the function pointer is to prevent people
from using them directly. You can check for -EOPNOTSUPP after calling
mtd_get_unmapped_area().

> +		return vm_iomap_memory(vma, start, len);
> +	}
>          /* This is broken because it assumes the MTD device is map-based
>  	   and that mtd->priv is a valid struct map_info.  It should be
>  	   replaced with something that uses the mtd_get_unmapped_area()
>  	   operation properly. */
>  	if (0 /*mtd->type == MTD_RAM || mtd->type == MTD_ROM*/) {
> +		struct map_info *map = mtd->priv;

You're adding dead code. It's behind an 'if (0)'. Could you perhaps take
a harder look at this code and fix up or kill any cruft?

>  #ifdef pgprot_noncached
>  		if (file->f_flags & O_DSYNC || map->phys >= __pa(high_memory))
>  			vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);

Brian

[PATCH v2] mtd: mtdchar: allow mmap for ram devices in memory address space.

[Philippe De Muyter]

use mtd_get_unmapped_area, as asked by an ancient comment, to get
the physical address (if any) in memory of the mtd device.

Therefore, mapram_unmapped_area had to be changed to return the physical
address, not the virtual one, For the NOMMU case, that makes no difference,
but for the MMU case, it is needed by mapram_unmapped_area to implement mmap.

Signed-off-by: Philippe De Muyter <phdm@macqel.be>
Cc: David Howells <dhowells@redhat.com>
Cc: Markus Niebel <Markus.Niebel@tqs.de>
Cc: Brian Norris <computersforpeace@gmail.com>
---
v2: cleanup following Brian's comments

 drivers/mtd/chips/map_ram.c |    2 +-
 drivers/mtd/mtdchar.c       |   22 ++++++++--------------
 2 files changed, 9 insertions(+), 15 deletions(-)

diff --git a/drivers/mtd/chips/map_ram.c b/drivers/mtd/chips/map_ram.c
index 991c2a1..e93d147 100644
--- a/drivers/mtd/chips/map_ram.c
+++ b/drivers/mtd/chips/map_ram.c
@@ -92,7 +92,7 @@ static unsigned long mapram_unmapped_area(struct mtd_info *mtd,
 					  unsigned long flags)
 {
 	struct map_info *map = mtd->priv;
-	return (unsigned long) map->virt + offset;
+	return (unsigned long) map->phys + offset;
 }
 
 static int mapram_read (struct mtd_info *mtd, loff_t from, size_t len, size_t *retlen, u_char *buf)
diff --git a/drivers/mtd/mtdchar.c b/drivers/mtd/mtdchar.c
index 2147e73..a1c7096 100644
--- a/drivers/mtd/mtdchar.c
+++ b/drivers/mtd/mtdchar.c
@@ -1112,20 +1112,14 @@ static int mtdchar_mmap(struct file *file, struct vm_area_struct *vma)
 #ifdef CONFIG_MMU
 	struct mtd_file_info *mfi = file->private_data;
 	struct mtd_info *mtd = mfi->mtd;
-	struct map_info *map = mtd->priv;
-
-        /* This is broken because it assumes the MTD device is map-based
-	   and that mtd->priv is a valid struct map_info.  It should be
-	   replaced with something that uses the mtd_get_unmapped_area()
-	   operation properly. */
-	if (0 /*mtd->type == MTD_RAM || mtd->type == MTD_ROM*/) {
-#ifdef pgprot_noncached
-		if (file->f_flags & O_DSYNC || map->phys >= __pa(high_memory))
-			vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
-#endif
-		return vm_iomap_memory(vma, map->phys, map->size);
-	}
-	return -ENODEV;
+	loff_t offset = vma->vm_pgoff << PAGE_SHIFT;
+	loff_t len = vma->vm_end - vma->vm_start;
+	phys_addr_t start;
+
+	if (offset > mtd->size - len)
+		return -EINVAL;
+	start = mtd_get_unmapped_area(mtd, len, offset, vma->vm_flags);
+	return start == -EOPNOTSUPP ? -ENODEV : vm_iomap_memory(vma, start, len);
 #else
 	return vma->vm_flags & VM_SHARED ? 0 : -EACCES;
 #endif
-- 
1.7.5.3

[PATCH v3] mtd: mtdchar: allow mmap for ram devices in memory address space.

[Philippe De Muyter]

use mtd_get_unmapped_area, as asked by an ancient comment, to get
the physical address (if any) in memory of the mtd device.

Therefore, mapram_unmapped_area had to be changed to return the physical
address, not the virtual one, For the NOMMU case, that makes no difference,
but for the MMU case, it is needed by mapram_unmapped_area to implement mmap.

Signed-off-by: Philippe De Muyter <phdm@macqel.be>
Cc: David Howells <dhowells@redhat.com>
Cc: Markus Niebel <Markus.Niebel@tqs.de>
Cc: Brian Norris <computersforpeace@gmail.com>
---
v3: avoid duplicating test done by mtd_get_unmapped_area

 drivers/mtd/chips/map_ram.c |    2 +-
 drivers/mtd/mtdchar.c       |   24 ++++++++++--------------
 2 files changed, 11 insertions(+), 15 deletions(-)

diff --git a/drivers/mtd/chips/map_ram.c b/drivers/mtd/chips/map_ram.c
index 991c2a1..e93d147 100644
--- a/drivers/mtd/chips/map_ram.c
+++ b/drivers/mtd/chips/map_ram.c
@@ -92,7 +92,7 @@ static unsigned long mapram_unmapped_area(struct mtd_info *mtd,
 					  unsigned long flags)
 {
 	struct map_info *map = mtd->priv;
-	return (unsigned long) map->virt + offset;
+	return (unsigned long) map->phys + offset;
 }
 
 static int mapram_read (struct mtd_info *mtd, loff_t from, size_t len, size_t *retlen, u_char *buf)
diff --git a/drivers/mtd/mtdchar.c b/drivers/mtd/mtdchar.c
index 2147e73..5956246 100644
--- a/drivers/mtd/mtdchar.c
+++ b/drivers/mtd/mtdchar.c
@@ -1112,20 +1112,16 @@ static int mtdchar_mmap(struct file *file, struct vm_area_struct *vma)
 #ifdef CONFIG_MMU
 	struct mtd_file_info *mfi = file->private_data;
 	struct mtd_info *mtd = mfi->mtd;
-	struct map_info *map = mtd->priv;
-
-        /* This is broken because it assumes the MTD device is map-based
-	   and that mtd->priv is a valid struct map_info.  It should be
-	   replaced with something that uses the mtd_get_unmapped_area()
-	   operation properly. */
-	if (0 /*mtd->type == MTD_RAM || mtd->type == MTD_ROM*/) {
-#ifdef pgprot_noncached
-		if (file->f_flags & O_DSYNC || map->phys >= __pa(high_memory))
-			vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
-#endif
-		return vm_iomap_memory(vma, map->phys, map->size);
-	}
-	return -ENODEV;
+	loff_t offset = vma->vm_pgoff << PAGE_SHIFT;
+	loff_t len = vma->vm_end - vma->vm_start;
+	phys_addr_t start;
+
+	start = mtd_get_unmapped_area(mtd, len, offset, vma->vm_flags);
+	if (start == -EOPNOTSUPP)
+		return -ENODEV;
+	if (start == -EINVAL)
+		return -EINVAL;
+	return vm_iomap_memory(vma, start, len);
 #else
 	return vma->vm_flags & VM_SHARED ? 0 : -EACCES;
 #endif
-- 
1.7.5.3

Re: [PATCH v3] mtd: mtdchar: allow mmap for ram devices in memory address space.

[Brian Norris]

In case this gets resurrected...

On Fri, Mar 07, 2014 at 02:54:39PM +0100, Philippe De Muyter wrote:
> use mtd_get_unmapped_area, as asked by an ancient comment, to get
> the physical address (if any) in memory of the mtd device.
> 
> Therefore, mapram_unmapped_area had to be changed to return the physical
> address, not the virtual one, For the NOMMU case, that makes no difference,
> but for the MMU case, it is needed by mapram_unmapped_area to implement mmap.
> 
> Signed-off-by: Philippe De Muyter <phdm@macqel.be>
> Cc: David Howells <dhowells@redhat.com>
> Cc: Markus Niebel <Markus.Niebel@tqs.de>
> Cc: Brian Norris <computersforpeace@gmail.com>
> ---
> v3: avoid duplicating test done by mtd_get_unmapped_area
> 
>  drivers/mtd/chips/map_ram.c |    2 +-
>  drivers/mtd/mtdchar.c       |   24 ++++++++++--------------
>  2 files changed, 11 insertions(+), 15 deletions(-)
> 
> diff --git a/drivers/mtd/chips/map_ram.c b/drivers/mtd/chips/map_ram.c
> index 991c2a1..e93d147 100644
> --- a/drivers/mtd/chips/map_ram.c
> +++ b/drivers/mtd/chips/map_ram.c
> @@ -92,7 +92,7 @@ static unsigned long mapram_unmapped_area(struct mtd_info *mtd,
>  					  unsigned long flags)
>  {
>  	struct map_info *map = mtd->priv;
> -	return (unsigned long) map->virt + offset;
> +	return (unsigned long) map->phys + offset;
>  }
>  
>  static int mapram_read (struct mtd_info *mtd, loff_t from, size_t len, size_t *retlen, u_char *buf)
> diff --git a/drivers/mtd/mtdchar.c b/drivers/mtd/mtdchar.c
> index 2147e73..5956246 100644
> --- a/drivers/mtd/mtdchar.c
> +++ b/drivers/mtd/mtdchar.c
> @@ -1112,20 +1112,16 @@ static int mtdchar_mmap(struct file *file, struct vm_area_struct *vma)
>  #ifdef CONFIG_MMU
>  	struct mtd_file_info *mfi = file->private_data;
>  	struct mtd_info *mtd = mfi->mtd;
> -	struct map_info *map = mtd->priv;
> -
> -        /* This is broken because it assumes the MTD device is map-based
> -	   and that mtd->priv is a valid struct map_info.  It should be
> -	   replaced with something that uses the mtd_get_unmapped_area()
> -	   operation properly. */
> -	if (0 /*mtd->type == MTD_RAM || mtd->type == MTD_ROM*/) {
> -#ifdef pgprot_noncached
> -		if (file->f_flags & O_DSYNC || map->phys >= __pa(high_memory))
> -			vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
> -#endif
> -		return vm_iomap_memory(vma, map->phys, map->size);
> -	}
> -	return -ENODEV;
> +	loff_t offset = vma->vm_pgoff << PAGE_SHIFT;
> +	loff_t len = vma->vm_end - vma->vm_start;
> +	phys_addr_t start;
> +
> +	start = mtd_get_unmapped_area(mtd, len, offset, vma->vm_flags);

Hmm, you're storing an 'unsigned long' in 'phys_addr_t', then treating
it as unsigned for the error checking. Is that guaranteed to work right?
Especially if sizeof(phys_addr_t) != sizeof(unsigned long)? e.g., if
phys_addr_t is 64 bits (with PAE?), but unsigned long is 32-bits, we
might see:

	start = mtd_get_unmapped_area(); // -EOPNOTSUPP

But -EOPNOTSUPP in a 32-bit unsigned long is only:

	0xffffffa1

which extends to:

	0x00000000ffffffa1

and doesn't match favorably with '-EOPNOTSUPP' when you compare it
later.

I think it makes more sense for 'start' to either match the return type
of mtd_get_unmapped_area() (i.e., unsigned long), or else just be a
signed type, so we don't have the unsigned/signed comparison issues
below.

Regards,
Brian

> +	if (start == -EOPNOTSUPP)
> +		return -ENODEV;
> +	if (start == -EINVAL)
> +		return -EINVAL;
> +	return vm_iomap_memory(vma, start, len);
>  #else
>  	return vma->vm_flags & VM_SHARED ? 0 : -EACCES;
>  #endif
> -- 
> 1.7.5.3
>