[PATCH] raspberrypi: reserve first two pages in efi memory map

[PATCH] raspberrypi: reserve first two pages in efi memory map

[kevans at FreeBSD.org]

From: Kyle Evans <kevans@FreeBSD.org>

The psci stub provided by Raspberry Pi is around 5k in size, thus residing
in the first two pages of memory. Without this reservation, the next stage
or OS assume they're free to use the second page and may get catastrophic
results from clobbering it.

Signed-off-by: Kyle Evans <kevans@FreeBSD.org>
Cc: Matthias Brugger <mbrugger@suse.com>
---
 board/raspberrypi/rpi/rpi.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/board/raspberrypi/rpi/rpi.c b/board/raspberrypi/rpi/rpi.c
index e367ba3092..d3f9bd16ac 100644
--- a/board/raspberrypi/rpi/rpi.c
+++ b/board/raspberrypi/rpi/rpi.c
@@ -488,8 +488,8 @@ int ft_board_setup(void *blob, bd_t *bd)
 	lcd_dt_simplefb_add_node(blob);
 
 #ifdef CONFIG_EFI_LOADER
-	/* Reserve the spin table */
-	efi_add_memory_map(0, 1, EFI_RESERVED_MEMORY_TYPE, 0);
+	/* Reserve the first two pages for spin table/psci stub. */
+	efi_add_memory_map(0, 2, EFI_RESERVED_MEMORY_TYPE, 0);
 #endif
 
 	return 0;
-- 
2.25.1

[PATCH] raspberrypi: reserve first two pages in efi memory map

[Matthias Brugger]

On 26/02/2020 15:03, kevans at FreeBSD.org wrote:
> From: Kyle Evans <kevans@FreeBSD.org>
> 
> The psci stub provided by Raspberry Pi is around 5k in size, thus residing
> in the first two pages of memory. Without this reservation, the next stage
> or OS assume they're free to use the second page and may get catastrophic
> results from clobbering it.
> 
> Signed-off-by: Kyle Evans <kevans@FreeBSD.org>
> Cc: Matthias Brugger <mbrugger@suse.com>
> ---
>  board/raspberrypi/rpi/rpi.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/board/raspberrypi/rpi/rpi.c b/board/raspberrypi/rpi/rpi.c
> index e367ba3092..d3f9bd16ac 100644
> --- a/board/raspberrypi/rpi/rpi.c
> +++ b/board/raspberrypi/rpi/rpi.c
> @@ -488,8 +488,8 @@ int ft_board_setup(void *blob, bd_t *bd)
>  	lcd_dt_simplefb_add_node(blob);
>  
>  #ifdef CONFIG_EFI_LOADER
> -	/* Reserve the spin table */
> -	efi_add_memory_map(0, 1, EFI_RESERVED_MEMORY_TYPE, 0);
> +	/* Reserve the first two pages for spin table/psci stub. */
> +	efi_add_memory_map(0, 2, EFI_RESERVED_MEMORY_TYPE, 0);

Is this that changed recently or was the memory reservation wrong from the
beginning?

Reagrds,
Matthias

>  #endif
>  
>  	return 0;
> 

[PATCH] raspberrypi: reserve first two pages in efi memory map

[Kyle Evans]

On Wed, Feb 26, 2020 at 12:06 PM Matthias Brugger <mbrugger@suse.com> wrote:
>
>
>
> On 26/02/2020 15:03, kevans at FreeBSD.org wrote:
> > From: Kyle Evans <kevans@FreeBSD.org>
> >
> > The psci stub provided by Raspberry Pi is around 5k in size, thus residing
> > in the first two pages of memory. Without this reservation, the next stage
> > or OS assume they're free to use the second page and may get catastrophic
> > results from clobbering it.
> >
> > Signed-off-by: Kyle Evans <kevans@FreeBSD.org>
> > Cc: Matthias Brugger <mbrugger@suse.com>
> > ---
> >  board/raspberrypi/rpi/rpi.c | 4 ++--
> >  1 file changed, 2 insertions(+), 2 deletions(-)
> >
> > diff --git a/board/raspberrypi/rpi/rpi.c b/board/raspberrypi/rpi/rpi.c
> > index e367ba3092..d3f9bd16ac 100644
> > --- a/board/raspberrypi/rpi/rpi.c
> > +++ b/board/raspberrypi/rpi/rpi.c
> > @@ -488,8 +488,8 @@ int ft_board_setup(void *blob, bd_t *bd)
> >       lcd_dt_simplefb_add_node(blob);
> >
> >  #ifdef CONFIG_EFI_LOADER
> > -     /* Reserve the spin table */
> > -     efi_add_memory_map(0, 1, EFI_RESERVED_MEMORY_TYPE, 0);
> > +     /* Reserve the first two pages for spin table/psci stub. */
> > +     efi_add_memory_map(0, 2, EFI_RESERVED_MEMORY_TYPE, 0);
>
> Is this that changed recently or was the memory reservation wrong from the
> beginning?
>

As far as I can tell it just wasn't quite accounted for in the initial
reservation -- Linux presumably uses spin-table as described in the
DTS, where-as FreeBSD opts for PSCI instead since it's available. I
don't have a good intuition for who else may be using PSCI for RPi; we
may be in the minority.

FWIW, /memreserve/ in the upstream (raspberrypi/firmware) DTB also do
not account for PSCI stubs -- I've filed an issue[1] there, as well.

Thanks,

Kyle Evans

[1] https://github.com/raspberrypi/firmware/issues/1340

[PATCH] raspberrypi: reserve first two pages in efi memory map

[Kyle Evans]

On Wed, Feb 26, 2020 at 12:26 PM Kyle Evans <kevans@freebsd.org> wrote:
>
> On Wed, Feb 26, 2020 at 12:06 PM Matthias Brugger <mbrugger@suse.com> wrote:
> >
> >
> >
> > On 26/02/2020 15:03, kevans at FreeBSD.org wrote:
> > > From: Kyle Evans <kevans@FreeBSD.org>
> > >
> > > The psci stub provided by Raspberry Pi is around 5k in size, thus residing
> > > in the first two pages of memory. Without this reservation, the next stage
> > > or OS assume they're free to use the second page and may get catastrophic
> > > results from clobbering it.
> > >
> > > Signed-off-by: Kyle Evans <kevans@FreeBSD.org>
> > > Cc: Matthias Brugger <mbrugger@suse.com>
> > > ---
> > >  board/raspberrypi/rpi/rpi.c | 4 ++--
> > >  1 file changed, 2 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/board/raspberrypi/rpi/rpi.c b/board/raspberrypi/rpi/rpi.c
> > > index e367ba3092..d3f9bd16ac 100644
> > > --- a/board/raspberrypi/rpi/rpi.c
> > > +++ b/board/raspberrypi/rpi/rpi.c
> > > @@ -488,8 +488,8 @@ int ft_board_setup(void *blob, bd_t *bd)
> > >       lcd_dt_simplefb_add_node(blob);
> > >
> > >  #ifdef CONFIG_EFI_LOADER
> > > -     /* Reserve the spin table */
> > > -     efi_add_memory_map(0, 1, EFI_RESERVED_MEMORY_TYPE, 0);
> > > +     /* Reserve the first two pages for spin table/psci stub. */
> > > +     efi_add_memory_map(0, 2, EFI_RESERVED_MEMORY_TYPE, 0);
> >
> > Is this that changed recently or was the memory reservation wrong from the
> > beginning?
> >
>
> As far as I can tell it just wasn't quite accounted for in the initial
> reservation -- Linux presumably uses spin-table as described in the
> DTS, where-as FreeBSD opts for PSCI instead since it's available. I
> don't have a good intuition for who else may be using PSCI for RPi; we
> may be in the minority.
>
> FWIW, /memreserve/ in the upstream (raspberrypi/firmware) DTB also do
> not account for PSCI stubs -- I've filed an issue[1] there, as well.
>
> Thanks,
>
> Kyle Evans
>
> [1] https://github.com/raspberrypi/firmware/issues/1340

Upon further examination, it looks like I've managed to confuse myself
and we're using our own stub derived from the upstream spintable
stuff... let's disregard this patch, I'll look into rolling up a new
one as both upstream implementation and ours should pass the required
reservation size through to U-Boot via x1.