Re: cdevpriv and mmap(2)

On Sun, 2008-09-14 at 20:48 +0300, Kostik Belousov wrote:
> When implementing cdevpriv, I completeley missed the support for
> d_mmap() driver method. This method is called by the kernel (at least)
> twice: one time to validate the mmaped region and protection mode (see
> vm/device_pager.c:dev_pager_alloc()), and second time to obtain the
> physical address when serving page fault (see dev_pager_getpages()).
> 
> Support for cdevpriv for the first call(s) is trivial, and is
> implemented by the patch below. Second calls are much harder and
> essentially require attaching cdevpriv bookkeeping data to the struct
> vm_map_entry. In fact, I am not sure whether this support for the second
> time calls is needed at all in real usage.
> 
> I Cc:ed people that pointed me to the issue, please evaluate the
> patch against your needs. I think I will commit it shortly after your
> feedback. On the other hand, I would think about implementing full
> support for d_mmap only if actually needed.

The needs of drm in this case are very minor, but after talking with
rwatson the other night, I was going to work around this issue.  But it
would be great if this works, so I'll try it as soon as I get caught up
on mail...

robert.

> Thanks !
> 
> diff --git a/sys/vm/vm_mmap.c b/sys/vm/vm_mmap.c
> index 7e6b04f..c3f08b0 100644
> --- a/sys/vm/vm_mmap.c
> +++ b/sys/vm/vm_mmap.c
> _at__at_ -391,8 +391,10 _at__at_ map:
>  		goto done;
>  	}
>  
> +	td->td_fpop = fp;
>  	error = vm_mmap(&vms->vm_map, &addr, size, prot, maxprot,
>  	    flags, handle_type, handle, pos);
> +	td->td_fpop = NULL;
>  #ifdef HWPMC_HOOKS
>  	/* inform hwpmc(4) if an executable is being mapped */
>  	if (error == 0 && handle_type == OBJT_VNODE &&