diff mbox series

[-fixes] riscv: Fix BUILTIN_DTB for sifive and microchip soc

Message ID 20210604120639.1447869-1-alex@ghiti.fr (mailing list archive)
State New
Headers show
Series [-fixes] riscv: Fix BUILTIN_DTB for sifive and microchip soc | expand

Commit Message

Alex Ghiti June 4, 2021, 12:06 p.m. UTC
Fix BUILTIN_DTB config which resulted in a dtb that was actually not built
into the Linux image: in the same manner as Canaan soc does, create an object
file from the dtb file that will get linked into the Linux image.

Signed-off-by: Alexandre Ghiti <alex@ghiti.fr>
---
 arch/riscv/boot/dts/microchip/Makefile | 1 +
 arch/riscv/boot/dts/sifive/Makefile    | 1 +
 2 files changed, 2 insertions(+)

Comments

Arnd Bergmann June 4, 2021, 1:08 p.m. UTC | #1
On Fri, Jun 4, 2021 at 2:06 PM Alexandre Ghiti <alex@ghiti.fr> wrote:
>
> Fix BUILTIN_DTB config which resulted in a dtb that was actually not built
> into the Linux image: in the same manner as Canaan soc does, create an object
> file from the dtb file that will get linked into the Linux image.
>
> Signed-off-by: Alexandre Ghiti <alex@ghiti.fr>

Along the same lines as the comment that Jisheng Zhang made on the fixed
address, building a dtb into the kernel itself fundamentally breaks generic
kernel images.

I can understand using it on K210, which is extremely limited and wouldn't
run a generic kernel anyway, but for normal platforms like microchip and
sifive, it would be better to disallow CONFIG_BUILTIN_DTB in Kconfig
and require a non-broken boot loader.

      Arnd
Palmer Dabbelt June 4, 2021, 3:51 p.m. UTC | #2
On Fri, 04 Jun 2021 06:08:05 PDT (-0700), Arnd Bergmann wrote:
> On Fri, Jun 4, 2021 at 2:06 PM Alexandre Ghiti <alex@ghiti.fr> wrote:
>>
>> Fix BUILTIN_DTB config which resulted in a dtb that was actually not built
>> into the Linux image: in the same manner as Canaan soc does, create an object
>> file from the dtb file that will get linked into the Linux image.
>>
>> Signed-off-by: Alexandre Ghiti <alex@ghiti.fr>
>
> Along the same lines as the comment that Jisheng Zhang made on the fixed
> address, building a dtb into the kernel itself fundamentally breaks generic
> kernel images.
>
> I can understand using it on K210, which is extremely limited and wouldn't
> run a generic kernel anyway, but for normal platforms like microchip and
> sifive, it would be better to disallow CONFIG_BUILTIN_DTB in Kconfig
> and require a non-broken boot loader.

When we first added BUILTIN_DTB we actually had a compatibility 
mechanism in there.  There isn't enough in the ISA to handle board 
compatibility, but we were hoping to get something to deal with that.  
It didn't pan out so we dropped the compatibility mechanism, which is 
how we ended up here.

Maybe the right thing to do here is to add some sort of "be compatible 
with the platform spec" Kconfig, which we could then use to disallow all 
these features that result in non-portable kernels?
Vitaly Wool June 4, 2021, 5:45 p.m. UTC | #3
Hey Arnd,

On Fri, Jun 4, 2021 at 3:18 PM Arnd Bergmann <arnd@arndb.de> wrote:
>
> On Fri, Jun 4, 2021 at 2:06 PM Alexandre Ghiti <alex@ghiti.fr> wrote:
> >
> > Fix BUILTIN_DTB config which resulted in a dtb that was actually not built
> > into the Linux image: in the same manner as Canaan soc does, create an object
> > file from the dtb file that will get linked into the Linux image.
> >
> > Signed-off-by: Alexandre Ghiti <alex@ghiti.fr>
>
> Along the same lines as the comment that Jisheng Zhang made on the fixed
> address, building a dtb into the kernel itself fundamentally breaks generic
> kernel images.
>
> I can understand using it on K210, which is extremely limited and wouldn't
> run a generic kernel anyway, but for normal platforms like microchip and
> sifive, it would be better to disallow CONFIG_BUILTIN_DTB in Kconfig
> and require a non-broken boot loader.

can't quite agree here. If we take XIP, it does make sense to have
BUILTIN_DTB, since 1) this will not be a generic kernel anyway 2) we
may want to skip the bootloader altogether or at least make it as thin
as possible and 3) copying device tree binaries from bootloader to RAM
as opposed to having it handy compiled in the kernel will be just a
waste of RAM.

Best regards,
   Vitaly

>       Arnd
>
> _______________________________________________
> linux-riscv mailing list
> linux-riscv@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-riscv
Arnd Bergmann June 4, 2021, 6:33 p.m. UTC | #4
On Fri, Jun 4, 2021 at 7:45 PM Vitaly Wool <vitaly.wool@konsulko.com> wrote:
> On Fri, Jun 4, 2021 at 3:18 PM Arnd Bergmann <arnd@arndb.de> wrote:
> > On Fri, Jun 4, 2021 at 2:06 PM Alexandre Ghiti <alex@ghiti.fr> wrote:
> > >
> > > Fix BUILTIN_DTB config which resulted in a dtb that was actually not built
> > > into the Linux image: in the same manner as Canaan soc does, create an object
> > > file from the dtb file that will get linked into the Linux image.
> > >
> > > Signed-off-by: Alexandre Ghiti <alex@ghiti.fr>
> >
> > Along the same lines as the comment that Jisheng Zhang made on the fixed
> > address, building a dtb into the kernel itself fundamentally breaks generic
> > kernel images.
> >
> > I can understand using it on K210, which is extremely limited and wouldn't
> > run a generic kernel anyway, but for normal platforms like microchip and
> > sifive, it would be better to disallow CONFIG_BUILTIN_DTB in Kconfig
> > and require a non-broken boot loader.
>
> can't quite agree here. If we take XIP, it does make sense to have
> BUILTIN_DTB, since 1) this will not be a generic kernel anyway 2) we
> may want to skip the bootloader altogether or at least make it as thin
> as possible and 3) copying device tree binaries from bootloader to RAM
> as opposed to having it handy compiled in the kernel will be just a
> waste of RAM.

Indeed, it does make sense in combination with XIP. Maybe there could
be a Kconfig option that depends on CONFIG_EXPERT and that can
be used to guard non-generic options like this?

       Arnd
Arnd Bergmann June 4, 2021, 6:34 p.m. UTC | #5
On Fri, Jun 4, 2021 at 5:51 PM Palmer Dabbelt <palmer@dabbelt.com> wrote:
> On Fri, 04 Jun 2021 06:08:05 PDT (-0700), Arnd Bergmann wrote:
> > On Fri, Jun 4, 2021 at 2:06 PM Alexandre Ghiti <alex@ghiti.fr> wrote:
> >>
> >> Fix BUILTIN_DTB config which resulted in a dtb that was actually not built
> >> into the Linux image: in the same manner as Canaan soc does, create an object
> >> file from the dtb file that will get linked into the Linux image.
> >>
> >> Signed-off-by: Alexandre Ghiti <alex@ghiti.fr>
> >
> > Along the same lines as the comment that Jisheng Zhang made on the fixed
> > address, building a dtb into the kernel itself fundamentally breaks generic
> > kernel images.
> >
> > I can understand using it on K210, which is extremely limited and wouldn't
> > run a generic kernel anyway, but for normal platforms like microchip and
> > sifive, it would be better to disallow CONFIG_BUILTIN_DTB in Kconfig
> > and require a non-broken boot loader.
>
> When we first added BUILTIN_DTB we actually had a compatibility
> mechanism in there.  There isn't enough in the ISA to handle board
> compatibility, but we were hoping to get something to deal with that.
> It didn't pan out so we dropped the compatibility mechanism, which is
> how we ended up here.
>
> Maybe the right thing to do here is to add some sort of "be compatible
> with the platform spec" Kconfig, which we could then use to disallow all
> these features that result in non-portable kernels?

Yes, I should have read your email before I replied with the same
suggestion to Vitaly ;-)

       Arnd
Alex Ghiti June 5, 2021, 6:33 a.m. UTC | #6
Hi Arnd,

Le 4/06/2021 à 15:08, Arnd Bergmann a écrit :
> On Fri, Jun 4, 2021 at 2:06 PM Alexandre Ghiti <alex@ghiti.fr> wrote:
>>
>> Fix BUILTIN_DTB config which resulted in a dtb that was actually not built
>> into the Linux image: in the same manner as Canaan soc does, create an object
>> file from the dtb file that will get linked into the Linux image.
>>
>> Signed-off-by: Alexandre Ghiti <alex@ghiti.fr>
> 
> Along the same lines as the comment that Jisheng Zhang made on the fixed
> address, building a dtb into the kernel itself fundamentally breaks generic
> kernel images.
> 
> I can understand using it on K210, which is extremely limited and wouldn't
> run a generic kernel anyway, but for normal platforms like microchip and
> sifive, it would be better to disallow CONFIG_BUILTIN_DTB in Kconfig
> and require a non-broken boot loader.

I kind of disagree because if I want to build a custom kernel for those 
platforms with a builtin dtb for some reasons (debug, development..Etc), 
I think I should be able to do so.

> 
>        Arnd
> 
> _______________________________________________
> linux-riscv mailing list
> linux-riscv@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-riscv
>
Arnd Bergmann June 5, 2021, 11 a.m. UTC | #7
On Sat, Jun 5, 2021 at 8:37 AM Alex Ghiti <alex@ghiti.fr> wrote:
> Le 4/06/2021 à 15:08, Arnd Bergmann a écrit :
> > On Fri, Jun 4, 2021 at 2:06 PM Alexandre Ghiti <alex@ghiti.fr> wrote:
> >>
> >> Fix BUILTIN_DTB config which resulted in a dtb that was actually not built
> >> into the Linux image: in the same manner as Canaan soc does, create an object
> >> file from the dtb file that will get linked into the Linux image.
> >>
> >> Signed-off-by: Alexandre Ghiti <alex@ghiti.fr>
> >
> > Along the same lines as the comment that Jisheng Zhang made on the fixed
> > address, building a dtb into the kernel itself fundamentally breaks generic
> > kernel images.
> >
> > I can understand using it on K210, which is extremely limited and wouldn't
> > run a generic kernel anyway, but for normal platforms like microchip and
> > sifive, it would be better to disallow CONFIG_BUILTIN_DTB in Kconfig
> > and require a non-broken boot loader.
>
> I kind of disagree because if I want to build a custom kernel for those
> platforms with a builtin dtb for some reasons (debug, development..Etc),
> I think I should be able to do so.

How is the builtin dtb better than appended dtb, or passing the dtb to the
boot loader in that case?

         Arnd
Alex Ghiti June 6, 2021, 7:40 a.m. UTC | #8
Le 5/06/2021 à 13:00, Arnd Bergmann a écrit :
> On Sat, Jun 5, 2021 at 8:37 AM Alex Ghiti <alex@ghiti.fr> wrote:
>> Le 4/06/2021 à 15:08, Arnd Bergmann a écrit :
>>> On Fri, Jun 4, 2021 at 2:06 PM Alexandre Ghiti <alex@ghiti.fr> wrote:
>>>>
>>>> Fix BUILTIN_DTB config which resulted in a dtb that was actually not built
>>>> into the Linux image: in the same manner as Canaan soc does, create an object
>>>> file from the dtb file that will get linked into the Linux image.
>>>>
>>>> Signed-off-by: Alexandre Ghiti <alex@ghiti.fr>
>>>
>>> Along the same lines as the comment that Jisheng Zhang made on the fixed
>>> address, building a dtb into the kernel itself fundamentally breaks generic
>>> kernel images.
>>>
>>> I can understand using it on K210, which is extremely limited and wouldn't
>>> run a generic kernel anyway, but for normal platforms like microchip and
>>> sifive, it would be better to disallow CONFIG_BUILTIN_DTB in Kconfig
>>> and require a non-broken boot loader.
>>
>> I kind of disagree because if I want to build a custom kernel for those
>> platforms with a builtin dtb for some reasons (debug, development..Etc),
>> I think I should be able to do so.
> 
> How is the builtin dtb better than appended dtb, or passing the dtb to the
> boot loader in that case?

Ah never said it was better, just it was available so there is no reason 
we could not allow it :)

> 
>           Arnd
> 
> _______________________________________________
> linux-riscv mailing list
> linux-riscv@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-riscv
>
Palmer Dabbelt June 12, 2021, 4:10 a.m. UTC | #9
On Sun, 06 Jun 2021 00:40:34 PDT (-0700), alex@ghiti.fr wrote:
> Le 5/06/2021 à 13:00, Arnd Bergmann a écrit :
>> On Sat, Jun 5, 2021 at 8:37 AM Alex Ghiti <alex@ghiti.fr> wrote:
>>> Le 4/06/2021 à 15:08, Arnd Bergmann a écrit :
>>>> On Fri, Jun 4, 2021 at 2:06 PM Alexandre Ghiti <alex@ghiti.fr> wrote:
>>>>>
>>>>> Fix BUILTIN_DTB config which resulted in a dtb that was actually not built
>>>>> into the Linux image: in the same manner as Canaan soc does, create an object
>>>>> file from the dtb file that will get linked into the Linux image.
>>>>>
>>>>> Signed-off-by: Alexandre Ghiti <alex@ghiti.fr>
>>>>
>>>> Along the same lines as the comment that Jisheng Zhang made on the fixed
>>>> address, building a dtb into the kernel itself fundamentally breaks generic
>>>> kernel images.
>>>>
>>>> I can understand using it on K210, which is extremely limited and wouldn't
>>>> run a generic kernel anyway, but for normal platforms like microchip and
>>>> sifive, it would be better to disallow CONFIG_BUILTIN_DTB in Kconfig
>>>> and require a non-broken boot loader.
>>>
>>> I kind of disagree because if I want to build a custom kernel for those
>>> platforms with a builtin dtb for some reasons (debug, development..Etc),
>>> I think I should be able to do so.
>>
>> How is the builtin dtb better than appended dtb, or passing the dtb to the
>> boot loader in that case?
>
> Ah never said it was better, just it was available so there is no reason
> we could not allow it :)

I agree: I'm not really a fan of BUILTIN_DTB (and I tried pretty hard 
not to have it in the first place), but whatever we have shouldn't be 
broken.

This is on fixes.
diff mbox series

Patch

diff --git a/arch/riscv/boot/dts/microchip/Makefile b/arch/riscv/boot/dts/microchip/Makefile
index 622b12771fd3..855c1502d912 100644
--- a/arch/riscv/boot/dts/microchip/Makefile
+++ b/arch/riscv/boot/dts/microchip/Makefile
@@ -1,2 +1,3 @@ 
 # SPDX-License-Identifier: GPL-2.0
 dtb-$(CONFIG_SOC_MICROCHIP_POLARFIRE) += microchip-mpfs-icicle-kit.dtb
+obj-$(CONFIG_BUILTIN_DTB) += $(addsuffix .o, $(dtb-y))
diff --git a/arch/riscv/boot/dts/sifive/Makefile b/arch/riscv/boot/dts/sifive/Makefile
index 74c47fe9fc22..d90e4eb0ade8 100644
--- a/arch/riscv/boot/dts/sifive/Makefile
+++ b/arch/riscv/boot/dts/sifive/Makefile
@@ -1,3 +1,4 @@ 
 # SPDX-License-Identifier: GPL-2.0
 dtb-$(CONFIG_SOC_SIFIVE) += hifive-unleashed-a00.dtb \
 			    hifive-unmatched-a00.dtb
+obj-$(CONFIG_BUILTIN_DTB) += $(addsuffix .o, $(dtb-y))