[openib-general] Re: [PATCH][RFC][0/4] InfiniBand userspace verbs implementation

[Roland Dreier]

    Andrew> umm, how about we

    Andrew> - force the special pages into a separate vma

    Andrew> - run get_user_pages() against it all

    Andrew> - use RLIMIT_MEMLOCK accounting to check whether the user
    Andrew> is allowed to do this thing

    Andrew> - undo the RMLIMIT_MEMLOCK accounting in ->release

    Andrew> This will all interact with user-initiated mlock/munlock
    Andrew> in messy ways. Maybe a new kernel-internal vma->vm_flag
    Andrew> which works like VM_LOCKED but is unaffected by
    Andrew> mlock/munlock activity is needed.

    Andrew> A bit of generalisation in do_mlock() should suit?

Yes, it seems that modifying do_mlock() to something like

	int do_mlock(unsigned long start, size_t len,
		     unsigned int set, unsigned int clear)

and then exporting a function along the lines of

	int do_mem_pin(unsigned long start, size_t len, int on)

that sets/clears (VM_LOCKED_KERNEL | VM_DONTCOPY) according to the on
flag.

Seem reasonable?  If so I can code this up.

 - R.