[PATCH] ARM64: kernel: implement ACPI parking protocol

[Lorenzo Pieralisi]

On Wed, Aug 26, 2015 at 08:13:22PM +0100, Mark Salter wrote:
> On Wed, 2015-08-26 at 17:07 +0100, Lorenzo Pieralisi wrote:
> > Mark,
> > 
> > On Thu, Jul 16, 2015 at 06:40:49PM +0100, Mark Salter wrote:
> > 
> > [...]
> > 
> > > > > The kernel will only add cached memory regions to linear mapping 
> > > > > and
> > > > > presumably, the FW will mark the mailboxes as uncached. Otherwise, 
> > > > > it
> > > > > is a FW bug. But I suppose we could run into problems with kernels
> > > > > using 64K pagesize since firmware assumes 4k.
> > > > 
> > > > Nope, ioremap takes care of that, everything should be fine.
> > > 
> > > The mailbox is 4K. If it is next to a cached UEFI region, the kernel 
> > > may
> > > have to overlap the mailbox with a cached 64K mapping in order to 
> > > include
> > > the adjoining UEFI region in the linear map. Then the ioremap would 
> > > fail
> > > because the mailbox is included in the linear mapping.
> > 
> > You have to acknowledge that what you describe is a bit of a corner
> > case (and a silly FW set-up), are you aware of any existing FW set-up
> > where we actually hit the corner case above ?
> 
> Oh sure. It could definitely be a corner case. I'm not aware of any
> existing firmware supporting the protocol implemented by your patch.

So basically there is no existing firmware implementing the ARM64 ACPI
parking protocol.

> So time will tell how well firmware vendors which do implement it
> avoid the corner case problem. I am aware of some existing firmwares
> which put the mailbox area in RAM which is also in the kernel linear
> mapping as cached memory. But those wouldn't work with your patch for
> more reasons than that.

They will have to fix them, that's the reason why I am pushing this
patch upstream, and not a version that is massaged to make their
non-compliant firmware compliant.

> > I think it is fine to leave code as-is, at least the mailbox
> > mappings, I will check to see I can improve the MADT parsing,
> > somehow.
> 
> I'm not strenuously opposed to leaving it as-is. But we need to
> acknowledge that if the 64K pagesize issue isn't covered by the
> protocol spec, we're relying on firmware vendors to be diligent
> not to break 64K pagesize kernels.

Should we try to amend the protocol spec ? I will raise the point,
I do not see how we can prevent this issue otherwise, other than
relying on vendors to be aware of the issue and prevent it.

Thanks,
Lorenzo