Re: alloc_page leaks tracing

[Dan Carpenter <dan.carpenter@oracle.com>]

I've pushed some changes to the check_unwind.c script so now the results
say which function allocated it.

I still haven't added the alloc_pages() and related functions to the
function table.  I need to do that.  I'm also still thinking about the
correct solution to the return_implies_param_key_exact() problem.  I
guess on my system, I just hardcoded the return values.  There is no
harm in pushing that so let me do that.

There are bunch of common false positives like:
1) put_device() frees the page.  I've been sort of working towards
   making put_device() handled in the correct way where we figure out
   what function it calls and parse the container_of() correctly and
   then it's all handled transparently by the check_kernel.c.  But for
   now probably the right thing is to add a special case to
   check_unwind.c.
2) Arrays.  Again, probably add a special case in check_unwind.c
3) Direct calls to a function which frees pointers from container_of().
   Eventually this will work correctly.  Another option for this would
   be to do what check_locking does and just record whenever a function
   frees something and then in the caller try to guess which resource
   it was based on the name.

So there is quite a bit of work which can be done still.

regards,
dan carpenter