Re: [PATCH] x86/EFI: additional checks in efi_bgrt_init()

[Josh Triplett <josh-iaAMLnmF4UmaiuxdJuQwMA@public.gmane.org>]

On Tue, Nov 06, 2012 at 01:37:10PM +0000, Jan Beulich wrote:
> >>> On 06.11.12 at 13:55, Josh Triplett <josh-iaAMLnmF4UmaiuxdJuQwMA@public.gmane.org> wrote:
> > On Tue, Nov 06, 2012 at 08:57:06AM +0000, Jan Beulich wrote:
> >> The question here is how to properly invalidate that data then:
> >> So far I was assuming that clearing the valid bit would be the way
> >> to go, as the specification says nothing on e.g. the image address
> >> being zero having any specific meaning. I need to do that in Xen,
> >> at least for the time being, as I'm not inclined to postpone
> >> indefinitely the use of boot services memory for normal memory
> >> allocations (which we would have to do if we wanted Dom0 to
> >> be able to access the image pointed to here).
> > 
> > This doesn't postpone the use of boot services memory indefinitely; the
> > BGRT driver copies the image out, and the kernel then frees boot
> > services memory.  That does introduce a delay in the use of boot
> > services memory, but not an indefinite one.
> 
> If and when Dom0 tells the hypervisor that it's done using that
> data is unknown from the hypervisor perspective, so for Xen
> this _is_ indefinitely.

Ah, I see.

> >> I was quite bad a design decision to allow (and even suggest)
> >> the image to live in boot services memory - one can't generally
> >> expect the ACPI parser to become available in an OS before
> >> setting up memory allocation. To Linux this already has turned
> >> out to be a problem, on Xen dealing with this would turn out
> >> even more cumbersome.
> > 
> > What problems has this caused?
> 
> None so far - I'm trying to prevent any from occurring.
> 
> Or if you're referring to my use of the term "problems" above - I
> was referring to the fact that for this very reason the treatment
> of boot services memory had to be changed in Linux.

Oh, I definitely agree about the poor design decision of putting the
image in boot services memory.  I had specifically wondered what
problems had occurred in Linux and Xen due to needing the ACPI parser
before freeing boot services memory.

- Josh Triplett