[PATCH] efi_loader: Don't carve out memory reservations too early

[Mark Kettenis]

> Date: Sat, 17 Feb 2024 12:02:56 +0100
> From: Heinrich Schuchardt <xypron.glpk at gmx.de>

Hi Heinrich,

> On 2/16/24 3:17 PM, Mark Kettenis wrote:
> >> Date: Fri, 16 Feb 2024 00:38:25 +0100
> >> From: Heinrich Schuchardt <xypron.glpk at gmx.de>
> >>
> >> Am 16. Februar 2024 00:25:34 MEZ schrieb Mark Kettenis <kettenis at openbsd.org>:
> >>> Moving the efi_carve_out_dt_rsv() call in commit 1be415b21b2d
> >>> ("efi_loader: create memory reservations in ACPI case")
> >>> broke boards that create additional memory reservations in
> >>> ft_board_setup() since it is now called before those additional
> >>> memory reservations are made.  This is the case for the rk3588
> >>> boards and breaks booting OpenBSD on those boards.
> >>>
> >>> Move the call back to its original location and add a call in
> >>> the code path used for ACPI.
> >>>
> >>> Fixes: 1be415b21b2d ("efi_loader: create memory reservations in ACPI case")
> >>> Signed-off-by: Mark Kettenis <kettenis at openbsd.org>
> >>> ---
> >>> lib/efi_loader/efi_helper.c | 11 +++++++----
> >>> 1 file changed, 7 insertions(+), 4 deletions(-)
> >>>
> >>> diff --git a/lib/efi_loader/efi_helper.c b/lib/efi_loader/efi_helper.c
> >>> index 5dd9cc876e..58761fae78 100644
> >>> --- a/lib/efi_loader/efi_helper.c
> >>> +++ b/lib/efi_loader/efi_helper.c
> >>> @@ -456,11 +456,11 @@ efi_status_t efi_install_fdt(void *fdt)
> >>> 		return EFI_LOAD_ERROR;
> >>> 	}
> >>>
> >>> -	/* Create memory reservations as indicated by the device tree */
> >>> -	efi_carve_out_dt_rsv(fdt);
> >>> -
> >>> -	if (CONFIG_IS_ENABLED(GENERATE_ACPI_TABLE))
> >>> +	if (CONFIG_IS_ENABLED(GENERATE_ACPI_TABLE)) {
> >>> +		/* Create memory reservations as indicated by the device tree */
> >>
> >> Imagine booting the rk3588 board with ACPI.
> >
> > I'd rather not, thank you ;)
> >
> >> Wouldn't we miss creating the ft_board_setup() reservations before
> >> efi_carve_out_dt_rsv(fdt)?
> >
> > Yes.  And arguably the these memory reservations should be made way
> > earlier, at the the time that efi_memory_init() runs.  I think we're
> > just lucky that efi_allocate_pages() doesn't hand us memory from these
> > areas in copy_fdt().
> >
> > Better ideas?
> >
> 
> image_setup_libfdt(&img, fdt, NULL) must be called before
> efi_carve_out_dt_rsv().
> 
> Could you, please, add this call in this path, too. Otherwise the patch
> looks correct.

Yes I can.  Just sent out v2.

Thanks,

Mark


> >
> >>> +		efi_carve_out_dt_rsv(fdt);
> >>> 		return EFI_SUCCESS;
> >>> +	}
> >>>
> >>> 	/* Prepare device tree for payload */
> >>> 	ret = copy_fdt(&fdt);
> >>> @@ -474,6 +474,9 @@ efi_status_t efi_install_fdt(void *fdt)
> >>> 		return EFI_LOAD_ERROR;
> >>> 	}
> >>>
> >>> +	/* Create memory reservations as indicated by the device tree */
> >>> +	efi_carve_out_dt_rsv(fdt);
> >>> +
> >>> 	efi_try_purge_kaslr_seed(fdt);
> >>>
> >>> 	if (CONFIG_IS_ENABLED(EFI_TCG2_PROTOCOL_MEASURE_DTB)) {
> >>
> >
> 
>