[PATCH v2] efi_loader: Get rid of kaslr-seed

[Ilias Apalodimas]

Hi Mark, 

On Sun, Jan 02, 2022 at 10:27:50PM +0100, Mark Kettenis wrote:
> > Date: Sun, 02 Jan 2022 22:06:11 +0100
> > From: Heinrich Schuchardt <xypron.glpk at gmx.de>
> > 
> > Am 2. Januar 2022 21:50:35 MEZ schrieb Ilias Apalodimas <ilias.apalodimas at linaro.org>:
> > >Hi Heinrich,
> > >
> > >> > > > > 
> > >
> > >[...]
> > >
> > >> > > > > diff --git a/cmd/bootefi.c b/cmd/bootefi.c
> > >> > > > > index d77d3b6e943d..57f13ce701ec 100644
> > >> > > > > --- a/cmd/bootefi.c
> > >> > > > > +++ b/cmd/bootefi.c
> > >> > > > > @@ -310,6 +310,8 @@ efi_status_t efi_install_fdt(void *fdt)
> > >> > > > >   	/* Create memory reservations as indicated by the device tree */
> > >> > > > >   	efi_carve_out_dt_rsv(fdt);
> > >> > > > > 
> > >> > > > > +	efi_try_purge_kaslr_seed(fdt);
> > >> 
> > >> This function should only be invoked for CONFIG_EFI_TCG2_PROTOCOL=y.
> > >
> > >Why?  As we discussed the kernel ignores the kaslr-seed for the
> > >physical randomization.  The only reason we would like to keep it is 
> > >for the randomization of the virtual address.  But if the EFI
> > >RNG protocol is installed the EFI stub is already doing the right thing. 
> > >So I really think purging it if EFI RNG is installed is the best option
> > >here (regardless of TPM measurements)
> > >
> > 
> > The only reason to delete kaslr-seed is that it conflicts with
> > measured boot. If an OS prefers the RNG protocol over kaslr-seed is
> > the decision of the OS and nothing U-Boot has to care about.
> 
> But it is also up to the OS to decide whether it cares about measured
> boot or not.
>  
> > You will have to delete kaslr-seed no matter if you have a RNG
> > protocol or not if and only if you want to use measured boot.
> 
> So you shouldn't just unconditionally delete kaslr-seed if the
> CONFIG_EFI_TCG2_PROTOCOL has been enabled, but only if it is actually
> used...  But is that even possible?

I can't think of any (sane) way to figure this out.

> 
> So maybe you should just specify the certain properties (such as
> kaslr-seed) should be skipped when calculating the hash of the device
> tree.
> 

We could, but I'd like to avoid that unless we don't have a better
alternative.

Thanks
/Ilias