Re: [PATCH] [8/7] oprofile - dcookies need to use u32

[John Levon]

On Tue, Oct 15, 2002 at 07:00:19PM -0700, David S. Miller wrote:

>    +               return (u32)dentry;
>    
>    Um, isn't this supposed to uniquely identify the dentry?
>    On a platform with 64-bit pointers there's now the theoretical
>    possibility of different dentries getting the same cookie ...
> 
> That's true.
> 
> We dealt with this (trying to use a kernel pointer as a cache held by
> userspace) in tcp_diag by making the actual object opaque.  It was
> actually two u32's, and that way it worked independant of kernel
> vs. user word size.

I'm not sure that's an option :

o userspace needs to know the size of the cookie in the event buffer
o userspace would like to use the cookie as a hash value to avoid
  repeated lookups

Perhaps the best solution would be to use a separate u32 ID value,
allocated linearly. I could just refuse to allocate new dcookies in
theoretical case of overflow.

The other possibility is a dcookiefs (cat
/dev/oprofile/dcookie/34343234) but that's a lot of extra
code/complexity ...

regards
john
-- 
"It's a cardboard universe ... and if you lean too hard against it, you fall
 through." 
	- Philip K. Dick