diff mbox series

[1/6] arm64: Unconditionally call unflatten_device_tree()

Message ID 20240112200750.4062441-2-sboyd@kernel.org
State Superseded
Headers show
Series of: populate of_root node if bootloader doesn't | expand

Commit Message

Stephen Boyd Jan. 12, 2024, 8:07 p.m. UTC
Call this function unconditionally so that we can populate an empty DTB
on platforms that don't boot with a firmware provided or builtin DTB.
There's no harm in calling unflatten_device_tree() unconditionally. If
there isn't a valid initial_boot_params dtb then unflatten_device_tree()
returns early.

Cc: Rob Herring <robh+dt@kernel.org>
Cc: Frank Rowand <frowand.list@gmail.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will@kernel.org>
Cc: <linux-arm-kernel@lists.infradead.org>
Signed-off-by: Stephen Boyd <sboyd@kernel.org>
---
 arch/arm64/kernel/setup.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

Comments

Rob Herring (Arm) Jan. 15, 2024, 5:57 p.m. UTC | #1
On Fri, Jan 12, 2024 at 12:07:44PM -0800, Stephen Boyd wrote:
> Call this function unconditionally so that we can populate an empty DTB
> on platforms that don't boot with a firmware provided or builtin DTB.
> There's no harm in calling unflatten_device_tree() unconditionally. If
> there isn't a valid initial_boot_params dtb then unflatten_device_tree()
> returns early.

There's always a valid DTB because that's the boot params even for ACPI 
systems. This does also create a userspace visible change that 
/proc/device-tree will be populated. I don't see an issue with that.

There was worry when ACPI was added that systems would pass both DT and 
ACPI tables and that the kernel must only use ACPI. That was more to 
force ACPI adoption, but I'm not sure if that actually exists in any 
early system. I think we're past forcing adoption now.

> Cc: Rob Herring <robh+dt@kernel.org>
> Cc: Frank Rowand <frowand.list@gmail.com>
> Cc: Catalin Marinas <catalin.marinas@arm.com>
> Cc: Will Deacon <will@kernel.org>
> Cc: <linux-arm-kernel@lists.infradead.org>
> Signed-off-by: Stephen Boyd <sboyd@kernel.org>
> ---
>  arch/arm64/kernel/setup.c | 3 +--
>  1 file changed, 1 insertion(+), 2 deletions(-)
> 
> diff --git a/arch/arm64/kernel/setup.c b/arch/arm64/kernel/setup.c
> index 417a8a86b2db..ede3d59dabf0 100644
> --- a/arch/arm64/kernel/setup.c
> +++ b/arch/arm64/kernel/setup.c
> @@ -351,8 +351,7 @@ void __init __no_sanitize_address setup_arch(char **cmdline_p)
>  	/* Parse the ACPI tables for possible boot-time configuration */
>  	acpi_boot_table_init();
>  
> -	if (acpi_disabled)
> -		unflatten_device_tree();
> +	unflatten_device_tree();
>  
>  	bootmem_init();
>  
> -- 
> https://git.kernel.org/pub/scm/linux/kernel/git/clk/linux.git/
> https://git.kernel.org/pub/scm/linux/kernel/git/sboyd/spmi.git
>
Stephen Boyd Jan. 17, 2024, 1:27 a.m. UTC | #2
Quoting Mark Rutland (2024-01-16 03:51:14)
> Hi Stephen,
> 
> On Fri, Jan 12, 2024 at 12:07:44PM -0800, Stephen Boyd wrote:
> > Call this function unconditionally so that we can populate an empty DTB
> > on platforms that don't boot with a firmware provided or builtin DTB.
> > There's no harm in calling unflatten_device_tree() unconditionally.
> 
> For better or worse, that's not true: there are systems the provide both a DTB
> *and* ACPI tables, and we must not consume both at the same time as those can
> clash and cause all sorts of problems. In addition, we don't want people being
> "clever" and describing disparate portions of their system in ACPI and DT.
> 
> It is a very deliberate choice to not unflatten the DTB when ACPI is in use,
> and I don't think we want to reopen this can of worms.

Hmm ok. I missed this part. Can we knock out the initial_boot_params in
this case so that we don't unflatten a DTB when ACPI is in use?
Rob Herring (Arm) Jan. 17, 2024, 5:54 p.m. UTC | #3
On Tue, Jan 16, 2024 at 05:27:18PM -0800, Stephen Boyd wrote:
> Quoting Mark Rutland (2024-01-16 03:51:14)
> > Hi Stephen,
> > 
> > On Fri, Jan 12, 2024 at 12:07:44PM -0800, Stephen Boyd wrote:
> > > Call this function unconditionally so that we can populate an empty DTB
> > > on platforms that don't boot with a firmware provided or builtin DTB.
> > > There's no harm in calling unflatten_device_tree() unconditionally.
> > 
> > For better or worse, that's not true: there are systems the provide both a DTB
> > *and* ACPI tables, and we must not consume both at the same time as those can
> > clash and cause all sorts of problems. In addition, we don't want people being
> > "clever" and describing disparate portions of their system in ACPI and DT.
> > 
> > It is a very deliberate choice to not unflatten the DTB when ACPI is in use,
> > and I don't think we want to reopen this can of worms.
> 
> Hmm ok. I missed this part. Can we knock out the initial_boot_params in
> this case so that we don't unflatten a DTB when ACPI is in use?

You mean so we don't unflatten the boot DTB, but instead unflatten the 
empty one, right? That sounds fine.

Another thing to check is kexec because it will still need the original 
DTB I think. Though if you are doing ACPI boot and kexec'ing, kexec may 
write out everything needed by the next kernel and the empty DTB would 
work just fine. Of course those users booting with ACPI and then 
kexec'ing to DT boot will be broken. Perhaps that's a feature...

Rob
Stephen Boyd Jan. 17, 2024, 11 p.m. UTC | #4
Quoting Rob Herring (2024-01-17 09:54:48)
> On Tue, Jan 16, 2024 at 05:27:18PM -0800, Stephen Boyd wrote:
> > Quoting Mark Rutland (2024-01-16 03:51:14)
> > > Hi Stephen,
> > > 
> > > On Fri, Jan 12, 2024 at 12:07:44PM -0800, Stephen Boyd wrote:
> > > > Call this function unconditionally so that we can populate an empty DTB
> > > > on platforms that don't boot with a firmware provided or builtin DTB.
> > > > There's no harm in calling unflatten_device_tree() unconditionally.
> > > 
> > > For better or worse, that's not true: there are systems the provide both a DTB
> > > *and* ACPI tables, and we must not consume both at the same time as those can
> > > clash and cause all sorts of problems. In addition, we don't want people being
> > > "clever" and describing disparate portions of their system in ACPI and DT.
> > > 
> > > It is a very deliberate choice to not unflatten the DTB when ACPI is in use,
> > > and I don't think we want to reopen this can of worms.
> > 
> > Hmm ok. I missed this part. Can we knock out the initial_boot_params in
> > this case so that we don't unflatten a DTB when ACPI is in use?
> 
> You mean so we don't unflatten the boot DTB, but instead unflatten the 
> empty one, right? That sounds fine.

Yes. Note, I don't have any ACPI arm64 system on hand to test anything
with :-(

> 
> Another thing to check is kexec because it will still need the original 
> DTB I think. Though if you are doing ACPI boot and kexec'ing, kexec may 
> write out everything needed by the next kernel and the empty DTB would 
> work just fine.

Yeah, it looks like dt_is_stub() will keep doing its thing there. The
empty DTB will have nothing in it and so kexec with ACPI and the empty
DTB will continue to use ACPI, and then the empty DTB will be added in
again.

> Of course those users booting with ACPI and then 
> kexec'ing to DT boot will be broken. Perhaps that's a feature...

I don't know how this part works. If you kexec to DT boot won't you run
through startup again and initial_boot_params will have a non-empty DTB
in it? I'd think this would keep working.
Mark Rutland Jan. 18, 2024, 3:23 p.m. UTC | #5
On Tue, Jan 16, 2024 at 03:13:42PM +0100, Geert Uytterhoeven wrote:
> Hi Mark,
> 
> On Tue, Jan 16, 2024 at 12:51 PM Mark Rutland <mark.rutland@arm.com> wrote:
> > On Fri, Jan 12, 2024 at 12:07:44PM -0800, Stephen Boyd wrote:
> > > Call this function unconditionally so that we can populate an empty DTB
> > > on platforms that don't boot with a firmware provided or builtin DTB.
> > > There's no harm in calling unflatten_device_tree() unconditionally.
> >
> > For better or worse, that's not true: there are systems the provide both a DTB
> > *and* ACPI tables, and we must not consume both at the same time as those can
> > clash and cause all sorts of problems. In addition, we don't want people being
> > "clever" and describing disparate portions of their system in ACPI and DT.
> 
> We'd get to the latter anyway, when plugging in a USB device where the
> circuitry on/behind the USB device is described in DT.

I don't understand what you mean there; where is the DT description of the USB
device coming from if the DTB hasn't been unflattened?

Mark.
Geert Uytterhoeven Jan. 18, 2024, 4:23 p.m. UTC | #6
Hi Mark,

On Thu, Jan 18, 2024 at 4:27 PM Mark Rutland <mark.rutland@arm.com> wrote:
> On Tue, Jan 16, 2024 at 05:27:18PM -0800, Stephen Boyd wrote:
> > Quoting Mark Rutland (2024-01-16 03:51:14)
> > > On Fri, Jan 12, 2024 at 12:07:44PM -0800, Stephen Boyd wrote:
> > > > Call this function unconditionally so that we can populate an empty DTB
> > > > on platforms that don't boot with a firmware provided or builtin DTB.
> > > > There's no harm in calling unflatten_device_tree() unconditionally.
> > >
> > > For better or worse, that's not true: there are systems the provide both a DTB
> > > *and* ACPI tables, and we must not consume both at the same time as those can
> > > clash and cause all sorts of problems. In addition, we don't want people being
> > > "clever" and describing disparate portions of their system in ACPI and DT.
> > >
> > > It is a very deliberate choice to not unflatten the DTB when ACPI is in use,
> > > and I don't think we want to reopen this can of worms.
> >
> > Hmm ok. I missed this part. Can we knock out the initial_boot_params in
> > this case so that we don't unflatten a DTB when ACPI is in use?
>
> Why is that better than just not calling unflatten_device_tree(), as we do
> today?
>
> The cover letter says this is all so that we can run DT tests for the clk
> framework; why can't that just depend on the system being booted with DT rather
> than ACPI? We have other tests which are architecture and/or configuration
> dependent...

There is definitely a merit in running (platform-independent) DT tests
on any platform, whether the platform actually uses DT to boot or not.

Gr{oetje,eeting}s,

                        Geert
Rob Herring (Arm) Jan. 19, 2024, 11:10 p.m. UTC | #7
On Thu, Jan 18, 2024 at 03:26:43PM +0000, Mark Rutland wrote:
> On Tue, Jan 16, 2024 at 05:27:18PM -0800, Stephen Boyd wrote:
> > Quoting Mark Rutland (2024-01-16 03:51:14)
> > > Hi Stephen,
> > > 
> > > On Fri, Jan 12, 2024 at 12:07:44PM -0800, Stephen Boyd wrote:
> > > > Call this function unconditionally so that we can populate an empty DTB
> > > > on platforms that don't boot with a firmware provided or builtin DTB.
> > > > There's no harm in calling unflatten_device_tree() unconditionally.
> > > 
> > > For better or worse, that's not true: there are systems the provide both a DTB
> > > *and* ACPI tables, and we must not consume both at the same time as those can
> > > clash and cause all sorts of problems. In addition, we don't want people being
> > > "clever" and describing disparate portions of their system in ACPI and DT.
> > > 
> > > It is a very deliberate choice to not unflatten the DTB when ACPI is in use,
> > > and I don't think we want to reopen this can of worms.
> > 
> > Hmm ok. I missed this part. Can we knock out the initial_boot_params in
> > this case so that we don't unflatten a DTB when ACPI is in use?
> 
> Why is that better than just not calling unflatten_device_tree(), as we do
> today?
> 
> The cover letter says this is all so that we can run DT tests for the clk
> framework; why can't that just depend on the system being booted with DT rather
> than ACPI? 

Because then the tests can never run on x86 and some people still use 
those systems. It's no different than why do we compile !x86 drivers on 
x86. It is convenient.

> We have other tests which are architecture and/or configuration
> dependent...

There's another usecase of non-discoverable devices behind discoverable 
devices. See my LPC session slides for more details. For this we will 
need some base DT to apply overlays to on DT AND ACPI systems. This is 
what Geert was getting at. Yes, it could be done with some other code 
path, but the DT unittest has done that hack for years and this series 
is getting rid of it.

Rob
diff mbox series

Patch

diff --git a/arch/arm64/kernel/setup.c b/arch/arm64/kernel/setup.c
index 417a8a86b2db..ede3d59dabf0 100644
--- a/arch/arm64/kernel/setup.c
+++ b/arch/arm64/kernel/setup.c
@@ -351,8 +351,7 @@  void __init __no_sanitize_address setup_arch(char **cmdline_p)
 	/* Parse the ACPI tables for possible boot-time configuration */
 	acpi_boot_table_init();
 
-	if (acpi_disabled)
-		unflatten_device_tree();
+	unflatten_device_tree();
 
 	bootmem_init();