At 2011-07-08 02:37:46，"Hans J. Koch" <hjk@hansjkoch.de> wrote:

On Thu, Jul 07, 2011 at 07:15:25AM -0700, Greg KH wrote:

...

On Thu, Jul 07, 2011 at 03:32:54PM +0800, bill wrote:

...

Hi, Greg

I've been reading the UIO code, one place puzzled me a lot. May I ask one question about it?

------------------------------------------------------------------------------- drivers/uio/uio.c static int uio_find_mem_index(struct vm_area_struct *vma) { int mi; struct uio_device *idev = vma->vm_private_data;

for (mi = 0; mi < MAX_UIO_MAPS; mi++) { if (idev->info->mem[mi].size == 0) return -1; if (vma->vm_pgoff == mi) return mi; } return -1; }

I don't get it why use vma->vm_pgoff as the index for struct uio_mem mem[MAX_UIO_MAPS], Suppose there are twohardware memory space, if user intends to map the 2nd mem, then the last parameter of mmap should be 1,

mapaddr = mmap(NULL, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 1);

Is my understanding correct?

No, it isn't. The last parameter of mmap is divided by PAGE_SIZE before it reaches the UIO core. Therefore, it has to be set to n * getpagesize(). This is explained in the docs under Documentation/DocBook/. Use "make htmldocs" to generate HTML documentation. A faster way is to do

dblatex Documentation/DocBook/uio-howto.tmpl

This generates a PDF with the UIO documentation, provided you have a working LaTeX system, and have dblatex installed.

Thanks, Hans

Thanks for your explanation. So vma->vm_pgoff is not used as its original meanings, that's the point puzzled me. Now it's clear:)

...

I really don't know, sorry.

But Hans should know, Hans?