POSIX.Memory_Mapping.Map_Memory

POSIX.Memory_Mapping.Map_Memory

[Adrian Hoe]

Hi,

I am working on video_device with Video_4_Linux based on the work of
Anders Gidenstam. According to latest V4L documentation, capturing with
Read has been deprecated and the only method now is by the use of mmap.

I open a device and read the device with the following procedures, Open
and Map_Image as below. In Procedure Map_Image, I make a call to
Map_Memory (POSIX.Memory_Mapping) and raise an exception:

POSIX.POSIX_ERROR : PERMISSION_DENIED

-----------------------------------------------------------------------
...
type Intensity is new Natural range 0..255;
for Intensity'Size use 8;
type Bitmap is array (Positive range <>) of Intensity;
type Bitmap_Access is access Bitmap;
...
private

type Image_Type is new Ada.Finalization.Controlled with
record
Class  : Image_Class;
Width  : Natural := 0;
Height : Natural := 0;
BM     : Bitmap_Access := null;
end record;



------------------------------------------------------------------------
procedure Open (Device    :    out Video_Device;
File_Name : in     String) is
Win : Video_Window;
begin
if Is_Character_Special_File (To_POSIX_String (File_Name)) then
Device.Fd := Open (To_POSIX_String (File_Name),
Read_Only);
Get_Capture_Window (Device, Win);
Device.Width  := Win.Width;
Device.Height := Win.Height;
else
raise IO_Error;
end if;
end Open;

-- Close video device.
procedure Close (Device : in Video_Device) is
begin
POSIX.IO.Close (Device.Fd);
Put_Line ("Device closed");
end Close;

-- Map image
procedure Map_Image (Device  : in     Video_Device;
Image   : in out Image_Type) is

package Convert is new System.Address_To_Access_Conversions
(Bitmap_Access);
use Convert;

Buf_Size : Natural;

begin
if Get_Width (Image) /= Device.Width or
Get_Height (Image) /= Device.Height then

case Get_Class (Image) is
when Grey =>
Buf_Size := Device.Width * Device.Height;

when Color =>
Buf_Size := Device.Width * Device.Height * 3;
end case;

declare
Map      : System.Address;
begin

-- Permission Denied exception was raised here at the line below!!!
Map := Map_Memory (System.Storage_Elements.Storage_Offset
(Buf_Size),
POSIX.Memory_Mapping.Allow_Read +
POSIX.Memory_Mapping.Allow_Write,
POSIX.Memory_Mapping.Map_Shared,
Device.Fd,
0);
Set_Bitmap (Image,
Device.Width,
Device.Height,
Bitmap_Access (Convert.To_Pointer (Map).all));
end;
end if;
end Map_Image;

--------------------------------------------------------------------------

What could I have missed? Any pointer please?
Best regards,
-- 
Adrian Hoe

Re: POSIX.Memory_Mapping.Map_Memory

[Mark Lorenzen]

"Adrian Hoe" <byhoe@greenlime.com> writes:

> Hi,
> 
> I am working on video_device with Video_4_Linux based on the work of
> Anders Gidenstam. According to latest V4L documentation, capturing with
> Read has been deprecated and the only method now is by the use of mmap.
> 
> I open a device and read the device with the following procedures, Open
> and Map_Image as below. In Procedure Map_Image, I make a call to
> Map_Memory (POSIX.Memory_Mapping) and raise an exception:
> 
> POSIX.POSIX_ERROR : PERMISSION_DENIED
> 
> -----------------------------------------------------------------------

[cut]

> Device.Fd := Open (To_POSIX_String (File_Name),
> Read_Only);

You open the device in read-only.

[cut]

> 
> -- Permission Denied exception was raised here at the line below!!!
> Map := Map_Memory (System.Storage_Elements.Storage_Offset
> (Buf_Size),
> POSIX.Memory_Mapping.Allow_Read +
> POSIX.Memory_Mapping.Allow_Write,
> POSIX.Memory_Mapping.Map_Shared,
> Device.Fd,
> 0);

But you try to map it in read/write mode.

Maybe thats the case of the exception. Do you have to be root in order
to open the device at all?

[cut]

> Adrian Hoe

- Mark Lorenzen

Re: POSIX.Memory_Mapping.Map_Memory

[Adrian Hoe]

Mark Lorenzen wrote:
> "Adrian Hoe" <byhoe@greenlime.com> writes:
>
> > Hi,
> >
> > I am working on video_device with Video_4_Linux based on the work
of
> > Anders Gidenstam. According to latest V4L documentation, capturing
with
> > Read has been deprecated and the only method now is by the use of
mmap.
>

>>>> snip <<<<


> > Device.Fd := Open (To_POSIX_String (File_Name),
> > Read_Only);
>
> You open the device in read-only.
>
> [cut]
>
> >
> > -- Permission Denied exception was raised here at the line below!!!
> > Map := Map_Memory (System.Storage_Elements.Storage_Offset
> > (Buf_Size),
> > POSIX.Memory_Mapping.Allow_Read +
> > POSIX.Memory_Mapping.Allow_Write,
> > POSIX.Memory_Mapping.Map_Shared,
> > Device.Fd,
> > 0);
>
> But you try to map it in read/write mode.

I open with Read_Write option and raised POSIX.POSIX_ERROR :
INVALID_ARGUMENT at the same line.

> Maybe thats the case of the exception. Do you have to be root in
order
> to open the device at all?


No, I don't have to be root to open the device. Some applications are
able to open it and the camera and video card work.



--
Adrian Hoe

Re: POSIX.Memory_Mapping.Map_Memory

[Mark Lorenzen]

"Adrian Hoe" <byhoe@greenlime.com> writes:

> Mark Lorenzen wrote:
> > "Adrian Hoe" <byhoe@greenlime.com> writes:
> >
> > > Hi,
> > >
> > > I am working on video_device with Video_4_Linux based on the work
> of
> > > Anders Gidenstam. According to latest V4L documentation, capturing
> with
> > > Read has been deprecated and the only method now is by the use of
> mmap.
> >
> 
> >>>> snip <<<<
> 
> 
> > > Device.Fd := Open (To_POSIX_String (File_Name),
> > > Read_Only);
> >
> > You open the device in read-only.
> >
> > [cut]
> >
> > >
> > > -- Permission Denied exception was raised here at the line below!!!
> > > Map := Map_Memory (System.Storage_Elements.Storage_Offset
> > > (Buf_Size),
> > > POSIX.Memory_Mapping.Allow_Read +
> > > POSIX.Memory_Mapping.Allow_Write,
> > > POSIX.Memory_Mapping.Map_Shared,
> > > Device.Fd,
> > > 0);
> >
> > But you try to map it in read/write mode.
> 
> I open with Read_Write option and raised POSIX.POSIX_ERROR :
> INVALID_ARGUMENT at the same line.

In the way you are using Map_Memeory, it is probably because the value
of Offset (i.e. 0) is not page aligned.

On many implementations, you can only map a multiple of the page size
and the memory has to be page aligned.

> 
> > Maybe thats the case of the exception. Do you have to be root in
> order
> > to open the device at all?
> 
> 
> No, I don't have to be root to open the device. Some applications are
> able to open it and the camera and video card work.
> 
> 
> 
> --
> Adrian Hoe

- Mark Lorenzen

Re: POSIX.Memory_Mapping.Map_Memory

[Adrian Hoe]

Mark Lorenzen wrote:
>
> In the way you are using Map_Memeory, it is probably because the
value
> of Offset (i.e. 0) is not page aligned.
>
> On many implementations, you can only map a multiple of the page size
> and the memory has to be page aligned.
>



I wrote a short C program to test mmap. It works and seems like no
problem doing mmap. The C code is as below:

---------------------------------------------------------------------
#include <stdio.h>
#include <fcntl.h>
#include "linux/ioctl.h"
#include "sys/ioctl.h"
#include "sys/mman.h"

int main ()
{
int fd;
int bufsize;
int result;
char *buf;

fd = open ("/dev/video0", O_RDWR);

if (fd < 0) {
printf ("Error open device.\n");
exit (1);
}

bufsize = 320 * 240 * 3;
buf=mmap(0, bufsize, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);

if (!buf) {
printf ("mmap error.\n");
exit (1);
}
else {
printf ("mmap successful.\n");
}

result = munmap (buf, bufsize);
printf ("unmap successful.\n");
close (fd);
printf ("device closed.\n");

return (0);
}
---------------------------------------------------------------------

Is this gnat/florist error or just as what you mentioned about page
alignment?

I am using gcc/gnat 3.3.3 and florist 3.15p on SuSE Linux 9.1 Pro on
AMD Athlon. I have no problem with both gcc/gnat and florist. My other
programs using both for RS-232 interface have no problems at all.
Now, I'm in despair with curiosity! :-)
--
Adrian Hoe

Re: POSIX.Memory_Mapping.Map_Memory

[Adrian Hoe]

Adrian Hoe wrote:
> Mark Lorenzen wrote:
> >
> > In the way you are using Map_Memeory, it is probably because the
> value
> > of Offset (i.e. 0) is not page aligned.
> >
> > On many implementations, you can only map a multiple of the page
size
> > and the memory has to be page aligned.
> >
>
>
>
> I wrote a short C program to test mmap. It works and seems like no
> problem doing mmap. The C code is as below:
>
> ---------------------------------------------------------------------
> #include <stdio.h>
> #include <fcntl.h>
> #include "linux/ioctl.h"
> #include "sys/ioctl.h"
> #include "sys/mman.h"
>
> int main ()
> {
> int fd;
> int bufsize;
> int result;
> char *buf;
>
> fd = open ("/dev/video0", O_RDWR);
>
> if (fd < 0) {
> printf ("Error open device.\n");
> exit (1);
> }
>
> bufsize = 320 * 240 * 3;
> buf=mmap(0, bufsize, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
>
> if (!buf) {
> printf ("mmap error.\n");
> exit (1);
> }
> else {
> printf ("mmap successful.\n");
> }
>
> result = munmap (buf, bufsize);
> printf ("unmap successful.\n");
> close (fd);
> printf ("device closed.\n");
>
> return (0);
> }
> ---------------------------------------------------------------------
>
> Is this gnat/florist error or just as what you mentioned about page
> alignment?
>
> I am using gcc/gnat 3.3.3 and florist 3.15p on SuSE Linux 9.1 Pro on
> AMD Athlon. I have no problem with both gcc/gnat and florist. My
other
> programs using both for RS-232 interface have no problems at all.
> Now, I'm in despair with curiosity! :-)
> --
> Adrian Hoe


I noticed the identation of my code is gone. I hope this is the error
of the Beta Google Groups. Sorry for the mess.
--
Adrian Hoe

Re: POSIX.Memory_Mapping.Map_Memory

[Mark Lorenzen]

"Adrian Hoe" <byhoe@greenlime.com> writes:

> Mark Lorenzen wrote:
> >
> > In the way you are using Map_Memeory, it is probably because the
> value
> > of Offset (i.e. 0) is not page aligned.
> >
> > On many implementations, you can only map a multiple of the page size
> > and the memory has to be page aligned.
> >
> 
> 
> 
> I wrote a short C program to test mmap. It works and seems like no
> problem doing mmap. The C code is as below:
> 
> ---------------------------------------------------------------------
> #include <stdio.h>
> #include <fcntl.h>
> #include "linux/ioctl.h"
> #include "sys/ioctl.h"
> #include "sys/mman.h"
> 
> int main ()
> {
> int fd;
> int bufsize;
> int result;
> char *buf;
> 
> fd = open ("/dev/video0", O_RDWR);
> 
> if (fd < 0) {
> printf ("Error open device.\n");
> exit (1);
> }
> 
> bufsize = 320 * 240 * 3;
> buf=mmap(0, bufsize, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
> 
> if (!buf) {

Note that mmap wil NEVER return 0, so the check should be "if (buf ==
MAP_FAILED)".

> printf ("mmap error.\n");
> exit (1);
> }
> else {
> printf ("mmap successful.\n");
> }
> 
> result = munmap (buf, bufsize);
> printf ("unmap successful.\n");
> close (fd);
> printf ("device closed.\n");
> 
> return (0);
> }
> ---------------------------------------------------------------------
> 
> Is this gnat/florist error or just as what you mentioned about page
> alignment?

I corrected the above mentioned problem and successfully opened and
mapped a file. Strange.

> 
> I am using gcc/gnat 3.3.3 and florist 3.15p on SuSE Linux 9.1 Pro on
> AMD Athlon. I have no problem with both gcc/gnat and florist. My other
> programs using both for RS-232 interface have no problems at all.
> Now, I'm in despair with curiosity! :-)

Please try to run the program using strace and see how mmap is
actually called.

> --
> Adrian Hoe

- Mark Lorenzen

Re: POSIX.Memory_Mapping.Map_Memory

[Adrian Hoe]

Mark Lorenzen wrote:
>
> I corrected the above mentioned problem and successfully opened and
> mapped a file. Strange.

Yes. I made a mistake. It should be:

if (buf == MAP_FAILED)

and it gave me error as well. And the strace says:

open("/dev/video0", O_RDWR)             = 3
mmap2(NULL, 230400, PROT_READ|PROT_WRITE, MAP_SHARED, 3, 0) = -1 EINVAL
(Invalid argument)
fstat64(1, {st_mode=S_IFCHR|0620, st_rdev=makedev(136, 1), ...}) = 0
m

mmap gives EINVAL in my C and Ada programs.
--
Adrian Hoe

Re: POSIX.Memory_Mapping.Map_Memory

[Adrian Hoe]

Mark Lorenzen wrote:

>
> Note that mmap wil NEVER return 0, so the check should be "if (buf ==
> MAP_FAILED)".
>
>
> I corrected the above mentioned problem and successfully opened and
> mapped a file. Strange.
>

Did you managed to open and map /dev/video0 after you change the line
to if (buf == MAP_FAILED)?

Or could it be a problem with distro? Which Linux distro are you using?
--
Adrian Hoe

Re: POSIX.Memory_Mapping.Map_Memory

[Mark Lorenzen]

"Adrian Hoe" <byhoe@greenlime.com> writes:

> Mark Lorenzen wrote:
> >
> > I corrected the above mentioned problem and successfully opened and
> > mapped a file. Strange.
> 
> Yes. I made a mistake. It should be:
> 
> if (buf == MAP_FAILED)
> 
> and it gave me error as well. And the strace says:
> 
> open("/dev/video0", O_RDWR)             = 3
> mmap2(NULL, 230400, PROT_READ|PROT_WRITE, MAP_SHARED, 3, 0) = -1 EINVAL

I was just wondering: Is it allowed to map a device file as shared
memory? Try to use MAP_PRIVATE instead of MAP_SHARED.

> (Invalid argument)
> fstat64(1, {st_mode=S_IFCHR|0620, st_rdev=makedev(136, 1), ...}) = 0
> m
> 
> mmap gives EINVAL in my C and Ada programs.
> --
> Adrian Hoe

- Mark Lorenzen

Re: POSIX.Memory_Mapping.Map_Memory

[Mark Lorenzen]

"Adrian Hoe" <byhoe@greenlime.com> writes:

> Mark Lorenzen wrote:
> 
> >
> > Note that mmap wil NEVER return 0, so the check should be "if (buf ==
> > MAP_FAILED)".
> >
> >
> > I corrected the above mentioned problem and successfully opened and
> > mapped a file. Strange.
> >
> 
> Did you managed to open and map /dev/video0 after you change the line
> to if (buf == MAP_FAILED)?

I do not have that device special file. I opened and mapped a regular
file that i have read/write access to.

> 
> Or could it be a problem with distro? Which Linux distro are you using?
> --
> Adrian Hoe

- Mark Lorenzen

Re: POSIX.Memory_Mapping.Map_Memory

[Mark Lorenzen]

Mark Lorenzen <mark.lorenzen@ofir.dk> writes:

> "Adrian Hoe" <byhoe@greenlime.com> writes:
> 
> > Mark Lorenzen wrote:
> > >
> > > I corrected the above mentioned problem and successfully opened and
> > > mapped a file. Strange.
> > 
> > Yes. I made a mistake. It should be:
> > 
> > if (buf == MAP_FAILED)
> > 
> > and it gave me error as well. And the strace says:
> > 
> > open("/dev/video0", O_RDWR)             = 3
> > mmap2(NULL, 230400, PROT_READ|PROT_WRITE, MAP_SHARED, 3, 0) = -1 EINVAL
> 
> I was just wondering: Is it allowed to map a device file as shared
> memory? Try to use MAP_PRIVATE instead of MAP_SHARED.

Furthermore: I do not think that it is possible to map character
special files, since it only allows sequential access. You can
probably only map block special files.

> 
> > (Invalid argument)
> > fstat64(1, {st_mode=S_IFCHR|0620, st_rdev=makedev(136, 1), ...}) = 0
> > m
> > 
> > mmap gives EINVAL in my C and Ada programs.
> > --
> > Adrian Hoe
> 
> - Mark Lorenzen

Re: POSIX.Memory_Mapping.Map_Memory

[Florian Weimer]

* Mark Lorenzen:

>> I was just wondering: Is it allowed to map a device file as shared
>> memory? Try to use MAP_PRIVATE instead of MAP_SHARED.
>
> Furthermore: I do not think that it is possible to map character
> special files, since it only allows sequential access. You can
> probably only map block special files.

It's possible to map /dev/zero.  In fact, there is little difference
between character and block devices these days.

Re: POSIX.Memory_Mapping.Map_Memory

[Adrian Hoe]

Florian Weimer wrote:
> * Mark Lorenzen:
>
> >> I was just wondering: Is it allowed to map a device file as shared
> >> memory? Try to use MAP_PRIVATE instead of MAP_SHARED.
> >
> > Furthermore: I do not think that it is possible to map character
> > special files, since it only allows sequential access. You can
> > probably only map block special files.
>
> It's possible to map /dev/zero.  In fact, there is little difference
> between character and block devices these days.



Just for the purpose of recording for people who may need this in the
future:

I've successfully mapped the memory with
POSIX.Memory_Mapping.Map_Memory.

The VIDIOCGMBUF ioctl has to be issued before Map_Memory. This ioctl
reports the size of buffer to Memory_Map and the offset within the
buffer for each frame.

My application is far from completion but I have resolved this
Memory_Map issue. I believe the wheel has just started rolling again.
:-)

Regards,

--
Adrian Hoe

Re: POSIX.Memory_Mapping.Map_Memory

[Mark Lorenzen]

"Adrian Hoe" <byhoe@greenlime.com> writes:

> Florian Weimer wrote:
> > * Mark Lorenzen:
> >
> > >> I was just wondering: Is it allowed to map a device file as shared
> > >> memory? Try to use MAP_PRIVATE instead of MAP_SHARED.
> > >
> > > Furthermore: I do not think that it is possible to map character
> > > special files, since it only allows sequential access. You can
> > > probably only map block special files.
> >
> > It's possible to map /dev/zero.  In fact, there is little difference
> > between character and block devices these days.
> 
> 
> 
> Just for the purpose of recording for people who may need this in the
> future:
> 
> I've successfully mapped the memory with
> POSIX.Memory_Mapping.Map_Memory.
> 
> The VIDIOCGMBUF ioctl has to be issued before Map_Memory. This ioctl
> reports the size of buffer to Memory_Map and the offset within the
> buffer for each frame.
> 
> My application is far from completion but I have resolved this
> Memory_Map issue. I believe the wheel has just started rolling again.
> :-)
> 
> Regards,
> 
> --
> Adrian Hoe

Thanks for the info. So we can fortunately conclude that the problem
is neither Ada nor Florist related, but is simply a special issue with
this type of device.

- Mark Lorenzen