>>> Bastian Blank <bastian@xxxxxxxxxxxx> 08.09.06 21:33 >>>
>On Fri, Sep 08, 2006 at 06:00:25PM +0100, Jan Beulich wrote:
>> If the vm86 code isn't running with an i/o bitmap permitting access (which
>> supposedly libint10 doesn't do), all i/o related faults should alread be
>> reflected back to the guest, shouldn't they? If otoh there was permission
>> granted from the kernel, then we shouldn't assume that code is prepared
>> to deal with gp faults from respective port accesses.
>Yes, but it does weird things, so I think there is something broken.
It would probably be good to first exactly understand what exactly is
>> However, the original patch would, if to be integrated, in my opinion need
>> quite a bit of additional work - it should honor non-zero segment bases
>> generally, and it should do proper limit checking for non-flat segments.
>Isn't this already done by copy_from_user?
How can it, if it isn't being passed a segment limit (and generally, shouldn't)?
Xen-devel mailing list