Changes in vm_operations_struct 2.2.x => 2.4.x

Changes in vm_operations_struct 2.2.x => 2.4.x

[Martin Maletinsky]

Hi,

Playing around with kernel memory allocated by a module and mmap()ed by a user space process, I noticed that in the 2.4.x kernel many of the 2.2.x vm_operation_struct
fields have gone. In particular there is no unmap operation any more. Is there any way for a driver to get notified when the user space process mapping memory exported by
the driver unmaps parts of that memory (as far as I know, the close operation is invoked when the entire memory is unmapped)? Does anyone know the reason why the number of
operations in vm_operation_struct has been reduced?

regards
Martin

--
Supercomputing System AG          email: maletinsky@scs.ch
Martin Maletinsky                 phone: +41 (0)1 445 16 05
Technoparkstrasse 1               fax:   +41 (0)1 445 16 10
CH-8005 Zurich


--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/

Re: Changes in vm_operations_struct 2.2.x => 2.4.x

[Ingo Oeser]

On Fri, Aug 03, 2001 at 10:01:22AM +0200, Martin Maletinsky wrote:
> Does anyone know the reason why the number of operations in
> vm_operation_struct has been reduced?

Al Viro reduced it, because nobody used them for several years.
Nobody complained after removing them, also.

Maybe you can explain more, what you try to do in your module and
people can help you.

Regards

Ingo Oeser
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/

Re: Changes in vm_operations_struct 2.2.x => 2.4.x

[Martin Maletinsky]

Hi Ingo,

Thank's for your reply. The details are as follows:
My module allocates a block of memory, and exports that block to user space processes, by registering as a character device and implementing a mmap file operation, so that
user space processes can map that memory block into their virtual address space by calling mmap().

A block of memory may only be mapped through mmap() by one process, subsequent mmap() calls to map that memory block should fail, until the block is no longer mapped into
any processes memory (note, that following fork() calls the block may be mapped into the virtual memory space of several processes simultaneously, but this is OK).

It would have simplified my job, if I could have assumed that a process has either the entire block mapped into it's virtual address space, or that it has none of the block
mapped at all (i.e. there are never fragment's of the block mapped into a process' virtual address space). To ensure that at the time of mapping, I would have checked the
length of the processes memory region (vma->vm_end -vma->vm_start) in the module's mmap() file operation. To prevent partial unmapping of the block, I would have registered
an unmap() vm_area operation with the vm_area_struct in the module's mmap() file operation. The unmap() vm_area operation would have  checked the addr and len parameter, to
detect partial unmappings of the memory block by a user process. In case of a partial unmapping of the memory block, unmap() would have simply sent a SIGKILL to the
process.

Regards
Martin

Ingo Oeser wrote:

> On Fri, Aug 03, 2001 at 10:01:22AM +0200, Martin Maletinsky wrote:
> > Does anyone know the reason why the number of operations in
> > vm_operation_struct has been reduced?
>
> Al Viro reduced it, because nobody used them for several years.
> Nobody complained after removing them, also.
>
> Maybe you can explain more, what you try to do in your module and
> people can help you.
>
> Regards
>
> Ingo Oeser
> -
> Kernelnewbies: Help each other learn about the Linux kernel.
> Archive:       http://mail.nl.linux.org/kernelnewbies/
> IRC Channel:   irc.openprojects.net / #kernelnewbies
> Web Page:      http://www.kernelnewbies.org/

--
Supercomputing System AG          email: maletinsky@scs.ch
Martin Maletinsky                 phone: +41 (0)1 445 16 05
Technoparkstrasse 1               fax:   +41 (0)1 445 16 10
CH-8005 Zurich


--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/

Re: Changes in vm_operations_struct 2.2.x => 2.4.x

[Thomas Hofer]

Martin Maletinsky wrote (Donnerstag,  9. August 2001 09:02):
> My module allocates a block of memory, and exports that block to user
> space processes, by registering as a character device and
> implementing a mmap file operation, so that user space processes can
> map that memory block into their virtual address space by calling
> mmap().

Would it be possible to do this with shared memory (shmget/shmat)? 
What's the advantage of making a device and mmaping it? Sounds 
more complicated. (I did only user-space programs as yet, so forgive me 
my naivity)

Thomas.
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/