On Fri, Mar 09, 2018 at 07:37:04PM +1100, Balbir Singh wrote: > On Fri, Mar 9, 2018 at 7:12 PM, Ram Pai <linux...@us.ibm.com> wrote: > > Once an address range is associated with an allocated pkey, it cannot be > > reverted back to key-0. There is no valid reason for the above behavior. On > > the contrary applications need the ability to do so. > > > > The patch relaxes the restriction. > > I looked at the code and my observation was going to be that we need > to change mm_pkey_is_allocated. I still fail to understand what > happens if pkey 0 is reserved? What is the default key is it the first > available key? Assuming 0 is the default key may work and seems to > work, but I am sure its mostly by accident. It would be nice, if we > could have a notion of the default key. I don't like the special > meaning given to key 0 here. Remember on powerpc if 0 is reserved and > UAMOR/AMOR does not allow modification because it's reserved, setting > 0 will still fail
The linux pkey API, assumes pkey-0 is the default key. If no key is explicitly associated with a page, the default key gets associated. When a default key gets associated with a page, the permissions on the page are not dictated by the permissions of the default key, but by the permission of other bits in the pte; i.e _PAGE_RWX. On powerpc, and AFAICT on x86, neither the hardware nor the hypervisor reserves key-0. Hence the OS is free to use the key value, the way it chooses. On Linux we choose to associate key-0 the special status called default-key. However I see your point. If some cpu architecture takes away key-0 from Linux, than implementing the special status for key-0 on that architecture can become challenging, though not impossible. That architecture implementation can internally map key-0 value to some other available key, and associate that key to the page. And offcourse make sure that the hardware/MMU uses the pte's RWX bits to enforce permissions, for that key. -- Ram Pai
