Re: mmap() and ioctl()

[Matthew Dharm]

On Mon, Apr 04, 2005 at 03:02:26PM -0400, Richard B. Johnson wrote:
> On Mon, 4 Apr 2005, Matthew Dharm wrote:
> 
> >This probably is a silly question, but....
> >
> >Is is possible to open a file, mmap() it into memory, then pass the address
> >of that map via an ioctl() call to the kernel, which will copy_from_user()
> >that data?
> >
> 
> Yes. A user-mode pointer, passed via ioctl() is valid in the kernel
> in the context of the user, i.e., during read() write() ioctl().
> 
> However, it is not valid if it is accessed by some other process or
> an interrupt. In other words, you can't store it somewhere and
> access it later in some other context.

Right, I've got that part.  The plan has been to mmap(), ioctl(), and
inside the ioctl do a copy_from_user() into a kernel-context buffer.

> >Yeah, that's an odd concept, I know... I could always malloc() some
> >memory, read the file in, and then ioctl() it.  But, if I could get away
> >with a direct mmap(), that would be much better for me.
> 
> Since you need to copy anyway, you could mmap() your kernel
> data (impliment mmap in your driver). Then you mmap both
> "files" the same way and copy to/from in user-mode.

That's an interesting concept, and one I'm not familiar with.  Any useful
pointers (beyond UTSL)?  I'll admit to being much more familiar with SCSI
and USB internals than I am with something like device-layer interfacing.

It sounds like you're saying that my driver can implement an mmap() method
(similar to the ioctl method), and then I can just mmap the source file and
the driver /dev node and do a memcpy() between them. 

That's an interesting idea, but potentially not what I need.  The data
needs to go with some command information and a buffer to stuff the results
in.  This is basically a co-processor device I'm talking to.  The basic
data path here is from a file, through the driver, to a custom piece of
hardware (and back again).

Tho, anything that allows me to move the data from the disk up to a place
where I can pci_map_single() it faster is a Good Thing(tm).

Matt

-- 
Matthew Dharm                              Home: mdharm-usb@xxxxxxxxxxxxxxxxxx 
Maintainer, Linux USB Mass Storage Driver

Sir, for the hundreth time, we do NOT carry 600-round boxes of belt-fed 
suction darts!
					-- Salesperson to Greg
User Friendly, 12/30/1997

Attachment: pgp00000.pgp
Description: PGP signature