diff mbox series

[v4,4/4] phytec: phycore_imx8mp: override env_get_location

Message ID 20220131165859.261923-5-tommaso.merciai@amarulasolutions.com
State Changes Requested
Delegated to: Stefano Babic
Headers show
Series imx8m: move env_get_location for imx8mn and imx8mp at board level | expand

Commit Message

Tommaso Merciai Jan. 31, 2022, 4:58 p.m. UTC
Override env_get_location function at board level, previously dropped
down from arch/arm/mach-imx/imx8m/soc.c

References:
 - commit 98af80d3c969e69a1b8ce98bb20e5ad844022da2

Signed-off-by: Tommaso Merciai <tommaso.merciai@amarulasolutions.com>
---
 board/phytec/phycore_imx8mp/phycore-imx8mp.c | 33 ++++++++++++++++++++
 1 file changed, 33 insertions(+)

Comments

Tommaso Merciai Jan. 31, 2022, 10:15 p.m. UTC | #1
On Mon, Jan 31, 2022 at 06:03:58PM +0100, Marek Vasut wrote:
> On 1/31/22 17:58, Tommaso Merciai wrote:
> > Override env_get_location function at board level, previously dropped
> > down from arch/arm/mach-imx/imx8m/soc.c
> > 
> > References:
> >   - commit 98af80d3c969e69a1b8ce98bb20e5ad844022da2
> > 
> > Signed-off-by: Tommaso Merciai <tommaso.merciai@amarulasolutions.com>
> > ---
> >   board/phytec/phycore_imx8mp/phycore-imx8mp.c | 33 ++++++++++++++++++++
> >   1 file changed, 33 insertions(+)
> > 
> > diff --git a/board/phytec/phycore_imx8mp/phycore-imx8mp.c b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
> > index a8f0821437..05926eefa3 100644
> > --- a/board/phytec/phycore_imx8mp/phycore-imx8mp.c
> > +++ b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
> > @@ -11,9 +11,42 @@
> >   #include <asm/mach-imx/boot_mode.h>
> >   #include <env.h>
> >   #include <miiphy.h>
> > +#include <env_internal.h>
> >   DECLARE_GLOBAL_DATA_PTR;
> > +enum env_location env_get_location(enum env_operation op, int prio)
> > +{
> 
> Why don't you just turn this into default __weak function and override it on
> board level when it is really needed to be overridden ?

Hi Marek,
env_get_location is already declared as __weak, check env/env.c. We
can't override it 2 times.

References:
 - https://developer.arm.com/documentation/dui0491/i/Compiler-specific-Features/--weak
Adam Ford Feb. 1, 2022, 12:20 a.m. UTC | #2
On Mon, Jan 31, 2022 at 4:16 PM Tommaso Merciai
<tommaso.merciai@amarulasolutions.com> wrote:
>
> On Mon, Jan 31, 2022 at 06:03:58PM +0100, Marek Vasut wrote:
> > On 1/31/22 17:58, Tommaso Merciai wrote:
> > > Override env_get_location function at board level, previously dropped
> > > down from arch/arm/mach-imx/imx8m/soc.c
> > >
> > > References:
> > >   - commit 98af80d3c969e69a1b8ce98bb20e5ad844022da2
> > >
> > > Signed-off-by: Tommaso Merciai <tommaso.merciai@amarulasolutions.com>
> > > ---
> > >   board/phytec/phycore_imx8mp/phycore-imx8mp.c | 33 ++++++++++++++++++++
> > >   1 file changed, 33 insertions(+)
> > >
> > > diff --git a/board/phytec/phycore_imx8mp/phycore-imx8mp.c b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
> > > index a8f0821437..05926eefa3 100644
> > > --- a/board/phytec/phycore_imx8mp/phycore-imx8mp.c
> > > +++ b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
> > > @@ -11,9 +11,42 @@
> > >   #include <asm/mach-imx/boot_mode.h>
> > >   #include <env.h>
> > >   #include <miiphy.h>
> > > +#include <env_internal.h>
> > >   DECLARE_GLOBAL_DATA_PTR;
> > > +enum env_location env_get_location(enum env_operation op, int prio)
> > > +{
> >
> > Why don't you just turn this into default __weak function and override it on
> > board level when it is really needed to be overridden ?
>
> Hi Marek,
> env_get_location is already declared as __weak, check env/env.c. We
> can't override it 2 times.

The original version (before it was added in 2707faf01f04
("imx8mn/imx8mp: override env_get_offset and env_get_location") is
located in env/env.c and for my board, this is the preferred method.
This replacement method actually is the opposite behavior from what I
want, which is to force the environment to a fixed location regardless
of the boot device.

I think Tommaso's method is better, because as it is, the users cannot
override it any more.

adam
>
> References:
>  - https://developer.arm.com/documentation/dui0491/i/Compiler-specific-Features/--weak
>
> --
> Tommaso Merciai
> Embedded Linux Developer
> tommaso.merciai@amarulasolutions.com
> __________________________________
>
> Amarula Solutions SRL
> Via Le Canevare 30, 31100 Treviso, Veneto, IT
> T. +39 042 243 5310
> info@amarulasolutions.com
> www.amarulasolutions.com
Marek Vasut Feb. 1, 2022, 3:16 a.m. UTC | #3
On 1/31/22 23:15, Tommaso Merciai wrote:
> On Mon, Jan 31, 2022 at 06:03:58PM +0100, Marek Vasut wrote:
>> On 1/31/22 17:58, Tommaso Merciai wrote:
>>> Override env_get_location function at board level, previously dropped
>>> down from arch/arm/mach-imx/imx8m/soc.c
>>>
>>> References:
>>>    - commit 98af80d3c969e69a1b8ce98bb20e5ad844022da2
>>>
>>> Signed-off-by: Tommaso Merciai <tommaso.merciai@amarulasolutions.com>
>>> ---
>>>    board/phytec/phycore_imx8mp/phycore-imx8mp.c | 33 ++++++++++++++++++++
>>>    1 file changed, 33 insertions(+)
>>>
>>> diff --git a/board/phytec/phycore_imx8mp/phycore-imx8mp.c b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
>>> index a8f0821437..05926eefa3 100644
>>> --- a/board/phytec/phycore_imx8mp/phycore-imx8mp.c
>>> +++ b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
>>> @@ -11,9 +11,42 @@
>>>    #include <asm/mach-imx/boot_mode.h>
>>>    #include <env.h>
>>>    #include <miiphy.h>
>>> +#include <env_internal.h>
>>>    DECLARE_GLOBAL_DATA_PTR;
>>> +enum env_location env_get_location(enum env_operation op, int prio)
>>> +{
>>
>> Why don't you just turn this into default __weak function and override it on
>> board level when it is really needed to be overridden ?
> 
> Hi Marek,
> env_get_location is already declared as __weak, check env/env.c. We
> can't override it 2 times.

Oh, it is this problem with missing ability to define multiple levels of 
symbol strength.

__weak enum env_location arch_env_get_location(enum env_operation op, 
int prio)
{
         if (prio >= ARRAY_SIZE(env_locations))
                 return ENVL_UNKNOWN;

         return env_locations[prio];
}

__weak enum env_location board_env_get_location(enum env_operation op, 
int prio)
{
	return arch_env_get_location(op, prio);
}

__weak enum env_location env_get_location(enum env_operation op, int prio)
{
	return board_env_get_location(op, prio);
}

By default, the compiler will optimize it all out. If you have 
arch-specific default (like imx does), implement 
arch_env_get_location(), if you have even board-specific default (like 
your board likely does), implement board_env_get_location(), if you need 
to override both, then override env_get_location() (unlikely).

This is also inline with all the other arch_*() and board_*() functions 
we have, and you won't have much duplication either.
Marek Vasut Feb. 1, 2022, 3:18 a.m. UTC | #4
On 2/1/22 01:20, Adam Ford wrote:
> On Mon, Jan 31, 2022 at 4:16 PM Tommaso Merciai
> <tommaso.merciai@amarulasolutions.com> wrote:
>>
>> On Mon, Jan 31, 2022 at 06:03:58PM +0100, Marek Vasut wrote:
>>> On 1/31/22 17:58, Tommaso Merciai wrote:
>>>> Override env_get_location function at board level, previously dropped
>>>> down from arch/arm/mach-imx/imx8m/soc.c
>>>>
>>>> References:
>>>>    - commit 98af80d3c969e69a1b8ce98bb20e5ad844022da2
>>>>
>>>> Signed-off-by: Tommaso Merciai <tommaso.merciai@amarulasolutions.com>
>>>> ---
>>>>    board/phytec/phycore_imx8mp/phycore-imx8mp.c | 33 ++++++++++++++++++++
>>>>    1 file changed, 33 insertions(+)
>>>>
>>>> diff --git a/board/phytec/phycore_imx8mp/phycore-imx8mp.c b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
>>>> index a8f0821437..05926eefa3 100644
>>>> --- a/board/phytec/phycore_imx8mp/phycore-imx8mp.c
>>>> +++ b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
>>>> @@ -11,9 +11,42 @@
>>>>    #include <asm/mach-imx/boot_mode.h>
>>>>    #include <env.h>
>>>>    #include <miiphy.h>
>>>> +#include <env_internal.h>
>>>>    DECLARE_GLOBAL_DATA_PTR;
>>>> +enum env_location env_get_location(enum env_operation op, int prio)
>>>> +{
>>>
>>> Why don't you just turn this into default __weak function and override it on
>>> board level when it is really needed to be overridden ?
>>
>> Hi Marek,
>> env_get_location is already declared as __weak, check env/env.c. We
>> can't override it 2 times.
> 
> The original version (before it was added in 2707faf01f04
> ("imx8mn/imx8mp: override env_get_offset and env_get_location") is
> located in env/env.c and for my board, this is the preferred method.
> This replacement method actually is the opposite behavior from what I
> want, which is to force the environment to a fixed location regardless
> of the boot device.
> 
> I think Tommaso's method is better, because as it is, the users cannot
> override it any more.

Doesn't 1/4 patch break env on pretty much every single imx8m board ?

See my other reply regarding arch and board level __weak symbols, that's 
likely the way to go here.
Tommaso Merciai Feb. 1, 2022, 9:09 a.m. UTC | #5
On Tue, Feb 01, 2022 at 04:16:52AM +0100, Marek Vasut wrote:
> On 1/31/22 23:15, Tommaso Merciai wrote:
> > On Mon, Jan 31, 2022 at 06:03:58PM +0100, Marek Vasut wrote:
> > > On 1/31/22 17:58, Tommaso Merciai wrote:
> > > > Override env_get_location function at board level, previously dropped
> > > > down from arch/arm/mach-imx/imx8m/soc.c
> > > > 
> > > > References:
> > > >    - commit 98af80d3c969e69a1b8ce98bb20e5ad844022da2
> > > > 
> > > > Signed-off-by: Tommaso Merciai <tommaso.merciai@amarulasolutions.com>
> > > > ---
> > > >    board/phytec/phycore_imx8mp/phycore-imx8mp.c | 33 ++++++++++++++++++++
> > > >    1 file changed, 33 insertions(+)
> > > > 
> > > > diff --git a/board/phytec/phycore_imx8mp/phycore-imx8mp.c b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
> > > > index a8f0821437..05926eefa3 100644
> > > > --- a/board/phytec/phycore_imx8mp/phycore-imx8mp.c
> > > > +++ b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
> > > > @@ -11,9 +11,42 @@
> > > >    #include <asm/mach-imx/boot_mode.h>
> > > >    #include <env.h>
> > > >    #include <miiphy.h>
> > > > +#include <env_internal.h>
> > > >    DECLARE_GLOBAL_DATA_PTR;
> > > > +enum env_location env_get_location(enum env_operation op, int prio)
> > > > +{
> > > 
> > > Why don't you just turn this into default __weak function and override it on
> > > board level when it is really needed to be overridden ?
> > 
> > Hi Marek,
> > env_get_location is already declared as __weak, check env/env.c. We
> > can't override it 2 times.
> 
> Oh, it is this problem with missing ability to define multiple levels of
> symbol strength.
> 
> __weak enum env_location arch_env_get_location(enum env_operation op, int
> prio)
> {
>         if (prio >= ARRAY_SIZE(env_locations))
>                 return ENVL_UNKNOWN;
> 
>         return env_locations[prio];
> }
> 
> __weak enum env_location board_env_get_location(enum env_operation op, int
> prio)
> {
> 	return arch_env_get_location(op, prio);
> }
> 
> __weak enum env_location env_get_location(enum env_operation op, int prio)
> {
> 	return board_env_get_location(op, prio);
> }
> 
> By default, the compiler will optimize it all out. If you have arch-specific
> default (like imx does), implement arch_env_get_location(), if you have even
> board-specific default (like your board likely does), implement
> board_env_get_location(), if you need to override both, then override
> env_get_location() (unlikely).
> 
> This is also inline with all the other arch_*() and board_*() functions we
> have, and you won't have much duplication either.

Hi Marek,
Thanks for the tips, then if I understand correctly, your idea is: use:

arch_env_get_location in (soc.c)

In this way imx8m users can override this function at board level using:

board_env_get_location

right?

Thanks,
Tommaso
Adam Ford Feb. 1, 2022, 11:16 a.m. UTC | #6
On Mon, Jan 31, 2022 at 9:18 PM Marek Vasut <marex@denx.de> wrote:
>
> On 2/1/22 01:20, Adam Ford wrote:
> > On Mon, Jan 31, 2022 at 4:16 PM Tommaso Merciai
> > <tommaso.merciai@amarulasolutions.com> wrote:
> >>
> >> On Mon, Jan 31, 2022 at 06:03:58PM +0100, Marek Vasut wrote:
> >>> On 1/31/22 17:58, Tommaso Merciai wrote:
> >>>> Override env_get_location function at board level, previously dropped
> >>>> down from arch/arm/mach-imx/imx8m/soc.c
> >>>>
> >>>> References:
> >>>>    - commit 98af80d3c969e69a1b8ce98bb20e5ad844022da2
> >>>>
> >>>> Signed-off-by: Tommaso Merciai <tommaso.merciai@amarulasolutions.com>
> >>>> ---
> >>>>    board/phytec/phycore_imx8mp/phycore-imx8mp.c | 33 ++++++++++++++++++++
> >>>>    1 file changed, 33 insertions(+)
> >>>>
> >>>> diff --git a/board/phytec/phycore_imx8mp/phycore-imx8mp.c b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
> >>>> index a8f0821437..05926eefa3 100644
> >>>> --- a/board/phytec/phycore_imx8mp/phycore-imx8mp.c
> >>>> +++ b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
> >>>> @@ -11,9 +11,42 @@
> >>>>    #include <asm/mach-imx/boot_mode.h>
> >>>>    #include <env.h>
> >>>>    #include <miiphy.h>
> >>>> +#include <env_internal.h>
> >>>>    DECLARE_GLOBAL_DATA_PTR;
> >>>> +enum env_location env_get_location(enum env_operation op, int prio)
> >>>> +{
> >>>
> >>> Why don't you just turn this into default __weak function and override it on
> >>> board level when it is really needed to be overridden ?
> >>
> >> Hi Marek,
> >> env_get_location is already declared as __weak, check env/env.c. We
> >> can't override it 2 times.
> >
> > The original version (before it was added in 2707faf01f04
> > ("imx8mn/imx8mp: override env_get_offset and env_get_location") is
> > located in env/env.c and for my board, this is the preferred method.
> > This replacement method actually is the opposite behavior from what I
> > want, which is to force the environment to a fixed location regardless
> > of the boot device.
> >
> > I think Tommaso's method is better, because as it is, the users cannot
> > override it any more.
>
> Doesn't 1/4 patch break env on pretty much every single imx8m board ?

For me, patch 1/4 fixed the issue of not being able to define a fixed
environment location.  It now sits where I put it.  It also allows me
to boot over USB without having to define ENV_IS_NOWHERE.

>
> See my other reply regarding arch and board level __weak symbols, that's
> likely the way to go here.
Marek Vasut Feb. 1, 2022, 11:21 a.m. UTC | #7
On 2/1/22 12:16, Adam Ford wrote:
> On Mon, Jan 31, 2022 at 9:18 PM Marek Vasut <marex@denx.de> wrote:
>>
>> On 2/1/22 01:20, Adam Ford wrote:
>>> On Mon, Jan 31, 2022 at 4:16 PM Tommaso Merciai
>>> <tommaso.merciai@amarulasolutions.com> wrote:
>>>>
>>>> On Mon, Jan 31, 2022 at 06:03:58PM +0100, Marek Vasut wrote:
>>>>> On 1/31/22 17:58, Tommaso Merciai wrote:
>>>>>> Override env_get_location function at board level, previously dropped
>>>>>> down from arch/arm/mach-imx/imx8m/soc.c
>>>>>>
>>>>>> References:
>>>>>>     - commit 98af80d3c969e69a1b8ce98bb20e5ad844022da2
>>>>>>
>>>>>> Signed-off-by: Tommaso Merciai <tommaso.merciai@amarulasolutions.com>
>>>>>> ---
>>>>>>     board/phytec/phycore_imx8mp/phycore-imx8mp.c | 33 ++++++++++++++++++++
>>>>>>     1 file changed, 33 insertions(+)
>>>>>>
>>>>>> diff --git a/board/phytec/phycore_imx8mp/phycore-imx8mp.c b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
>>>>>> index a8f0821437..05926eefa3 100644
>>>>>> --- a/board/phytec/phycore_imx8mp/phycore-imx8mp.c
>>>>>> +++ b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
>>>>>> @@ -11,9 +11,42 @@
>>>>>>     #include <asm/mach-imx/boot_mode.h>
>>>>>>     #include <env.h>
>>>>>>     #include <miiphy.h>
>>>>>> +#include <env_internal.h>
>>>>>>     DECLARE_GLOBAL_DATA_PTR;
>>>>>> +enum env_location env_get_location(enum env_operation op, int prio)
>>>>>> +{
>>>>>
>>>>> Why don't you just turn this into default __weak function and override it on
>>>>> board level when it is really needed to be overridden ?
>>>>
>>>> Hi Marek,
>>>> env_get_location is already declared as __weak, check env/env.c. We
>>>> can't override it 2 times.
>>>
>>> The original version (before it was added in 2707faf01f04
>>> ("imx8mn/imx8mp: override env_get_offset and env_get_location") is
>>> located in env/env.c and for my board, this is the preferred method.
>>> This replacement method actually is the opposite behavior from what I
>>> want, which is to force the environment to a fixed location regardless
>>> of the boot device.
>>>
>>> I think Tommaso's method is better, because as it is, the users cannot
>>> override it any more.
>>
>> Doesn't 1/4 patch break env on pretty much every single imx8m board ?
> 
> For me, patch 1/4 fixed the issue of not being able to define a fixed
> environment location.  It now sits where I put it.  It also allows me
> to boot over USB without having to define ENV_IS_NOWHERE.

Sure, it also likely breaks every other MX8M board which does not define 
that env_get_location() on board level, i.e. every MX8M board.
Adam Ford Feb. 1, 2022, 11:22 a.m. UTC | #8
On Tue, Feb 1, 2022 at 3:09 AM Tommaso Merciai
<tommaso.merciai@amarulasolutions.com> wrote:
>
> On Tue, Feb 01, 2022 at 04:16:52AM +0100, Marek Vasut wrote:
> > On 1/31/22 23:15, Tommaso Merciai wrote:
> > > On Mon, Jan 31, 2022 at 06:03:58PM +0100, Marek Vasut wrote:
> > > > On 1/31/22 17:58, Tommaso Merciai wrote:
> > > > > Override env_get_location function at board level, previously dropped
> > > > > down from arch/arm/mach-imx/imx8m/soc.c
> > > > >
> > > > > References:
> > > > >    - commit 98af80d3c969e69a1b8ce98bb20e5ad844022da2
> > > > >
> > > > > Signed-off-by: Tommaso Merciai <tommaso.merciai@amarulasolutions.com>
> > > > > ---
> > > > >    board/phytec/phycore_imx8mp/phycore-imx8mp.c | 33 ++++++++++++++++++++
> > > > >    1 file changed, 33 insertions(+)
> > > > >
> > > > > diff --git a/board/phytec/phycore_imx8mp/phycore-imx8mp.c b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
> > > > > index a8f0821437..05926eefa3 100644
> > > > > --- a/board/phytec/phycore_imx8mp/phycore-imx8mp.c
> > > > > +++ b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
> > > > > @@ -11,9 +11,42 @@
> > > > >    #include <asm/mach-imx/boot_mode.h>
> > > > >    #include <env.h>
> > > > >    #include <miiphy.h>
> > > > > +#include <env_internal.h>
> > > > >    DECLARE_GLOBAL_DATA_PTR;
> > > > > +enum env_location env_get_location(enum env_operation op, int prio)
> > > > > +{
> > > >
> > > > Why don't you just turn this into default __weak function and override it on
> > > > board level when it is really needed to be overridden ?
> > >
> > > Hi Marek,
> > > env_get_location is already declared as __weak, check env/env.c. We
> > > can't override it 2 times.
> >
> > Oh, it is this problem with missing ability to define multiple levels of
> > symbol strength.
> >
> > __weak enum env_location arch_env_get_location(enum env_operation op, int
> > prio)
> > {
> >         if (prio >= ARRAY_SIZE(env_locations))
> >                 return ENVL_UNKNOWN;
> >
> >         return env_locations[prio];
> > }
> >
> > __weak enum env_location board_env_get_location(enum env_operation op, int
> > prio)
> > {
> >       return arch_env_get_location(op, prio);
> > }
> >
> > __weak enum env_location env_get_location(enum env_operation op, int prio)
> > {
> >       return board_env_get_location(op, prio);
> > }
> >
> > By default, the compiler will optimize it all out. If you have arch-specific
> > default (like imx does), implement arch_env_get_location(), if you have even
> > board-specific default (like your board likely does), implement
> > board_env_get_location(), if you need to override both, then override
> > env_get_location() (unlikely).
> >
> > This is also inline with all the other arch_*() and board_*() functions we
> > have, and you won't have much duplication either.
>
> Hi Marek,
> Thanks for the tips, then if I understand correctly, your idea is: use:
>
> arch_env_get_location in (soc.c)
>
> In this way imx8m users can override this function at board level using:
>
> board_env_get_location
>
> right?

What about those of us who want to use the default option found in
env.c?  It seems like we're creating more abstraction to address the
abstraction we don't all want.  From my interpretation, the whole
point of creating the default in env.c was to let people define the
location of their environment, and this function in soc.c undid that.
If people want it for their boards, put this function in their boards,
otherwise, just use the default, or write your own.

adam
>
> Thanks,
> Tommaso
>
> --
> Tommaso Merciai
> Embedded Linux Developer
> tommaso.merciai@amarulasolutions.com
> __________________________________
>
> Amarula Solutions SRL
> Via Le Canevare 30, 31100 Treviso, Veneto, IT
> T. +39 042 243 5310
> info@amarulasolutions.com
> www.amarulasolutions.com
Marek Vasut Feb. 1, 2022, 11:22 a.m. UTC | #9
On 2/1/22 10:09, Tommaso Merciai wrote:
> On Tue, Feb 01, 2022 at 04:16:52AM +0100, Marek Vasut wrote:
>> On 1/31/22 23:15, Tommaso Merciai wrote:
>>> On Mon, Jan 31, 2022 at 06:03:58PM +0100, Marek Vasut wrote:
>>>> On 1/31/22 17:58, Tommaso Merciai wrote:
>>>>> Override env_get_location function at board level, previously dropped
>>>>> down from arch/arm/mach-imx/imx8m/soc.c
>>>>>
>>>>> References:
>>>>>     - commit 98af80d3c969e69a1b8ce98bb20e5ad844022da2
>>>>>
>>>>> Signed-off-by: Tommaso Merciai <tommaso.merciai@amarulasolutions.com>
>>>>> ---
>>>>>     board/phytec/phycore_imx8mp/phycore-imx8mp.c | 33 ++++++++++++++++++++
>>>>>     1 file changed, 33 insertions(+)
>>>>>
>>>>> diff --git a/board/phytec/phycore_imx8mp/phycore-imx8mp.c b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
>>>>> index a8f0821437..05926eefa3 100644
>>>>> --- a/board/phytec/phycore_imx8mp/phycore-imx8mp.c
>>>>> +++ b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
>>>>> @@ -11,9 +11,42 @@
>>>>>     #include <asm/mach-imx/boot_mode.h>
>>>>>     #include <env.h>
>>>>>     #include <miiphy.h>
>>>>> +#include <env_internal.h>
>>>>>     DECLARE_GLOBAL_DATA_PTR;
>>>>> +enum env_location env_get_location(enum env_operation op, int prio)
>>>>> +{
>>>>
>>>> Why don't you just turn this into default __weak function and override it on
>>>> board level when it is really needed to be overridden ?
>>>
>>> Hi Marek,
>>> env_get_location is already declared as __weak, check env/env.c. We
>>> can't override it 2 times.
>>
>> Oh, it is this problem with missing ability to define multiple levels of
>> symbol strength.
>>
>> __weak enum env_location arch_env_get_location(enum env_operation op, int
>> prio)
>> {
>>          if (prio >= ARRAY_SIZE(env_locations))
>>                  return ENVL_UNKNOWN;
>>
>>          return env_locations[prio];
>> }
>>
>> __weak enum env_location board_env_get_location(enum env_operation op, int
>> prio)
>> {
>> 	return arch_env_get_location(op, prio);
>> }
>>
>> __weak enum env_location env_get_location(enum env_operation op, int prio)
>> {
>> 	return board_env_get_location(op, prio);
>> }
>>
>> By default, the compiler will optimize it all out. If you have arch-specific
>> default (like imx does), implement arch_env_get_location(), if you have even
>> board-specific default (like your board likely does), implement
>> board_env_get_location(), if you need to override both, then override
>> env_get_location() (unlikely).
>>
>> This is also inline with all the other arch_*() and board_*() functions we
>> have, and you won't have much duplication either.
> 
> Hi Marek,
> Thanks for the tips, then if I understand correctly, your idea is: use:
> 
> arch_env_get_location in (soc.c)
> 
> In this way imx8m users can override this function at board level using:
> 
> board_env_get_location
> 
> right?

Yes, I think that is the most flexible way with least duplication.
Adam Ford Feb. 1, 2022, 11:23 a.m. UTC | #10
On Tue, Feb 1, 2022 at 5:22 AM Marek Vasut <marex@denx.de> wrote:
>
> On 2/1/22 12:16, Adam Ford wrote:
> > On Mon, Jan 31, 2022 at 9:18 PM Marek Vasut <marex@denx.de> wrote:
> >>
> >> On 2/1/22 01:20, Adam Ford wrote:
> >>> On Mon, Jan 31, 2022 at 4:16 PM Tommaso Merciai
> >>> <tommaso.merciai@amarulasolutions.com> wrote:
> >>>>
> >>>> On Mon, Jan 31, 2022 at 06:03:58PM +0100, Marek Vasut wrote:
> >>>>> On 1/31/22 17:58, Tommaso Merciai wrote:
> >>>>>> Override env_get_location function at board level, previously dropped
> >>>>>> down from arch/arm/mach-imx/imx8m/soc.c
> >>>>>>
> >>>>>> References:
> >>>>>>     - commit 98af80d3c969e69a1b8ce98bb20e5ad844022da2
> >>>>>>
> >>>>>> Signed-off-by: Tommaso Merciai <tommaso.merciai@amarulasolutions.com>
> >>>>>> ---
> >>>>>>     board/phytec/phycore_imx8mp/phycore-imx8mp.c | 33 ++++++++++++++++++++
> >>>>>>     1 file changed, 33 insertions(+)
> >>>>>>
> >>>>>> diff --git a/board/phytec/phycore_imx8mp/phycore-imx8mp.c b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
> >>>>>> index a8f0821437..05926eefa3 100644
> >>>>>> --- a/board/phytec/phycore_imx8mp/phycore-imx8mp.c
> >>>>>> +++ b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
> >>>>>> @@ -11,9 +11,42 @@
> >>>>>>     #include <asm/mach-imx/boot_mode.h>
> >>>>>>     #include <env.h>
> >>>>>>     #include <miiphy.h>
> >>>>>> +#include <env_internal.h>
> >>>>>>     DECLARE_GLOBAL_DATA_PTR;
> >>>>>> +enum env_location env_get_location(enum env_operation op, int prio)
> >>>>>> +{
> >>>>>
> >>>>> Why don't you just turn this into default __weak function and override it on
> >>>>> board level when it is really needed to be overridden ?
> >>>>
> >>>> Hi Marek,
> >>>> env_get_location is already declared as __weak, check env/env.c. We
> >>>> can't override it 2 times.
> >>>
> >>> The original version (before it was added in 2707faf01f04
> >>> ("imx8mn/imx8mp: override env_get_offset and env_get_location") is
> >>> located in env/env.c and for my board, this is the preferred method.
> >>> This replacement method actually is the opposite behavior from what I
> >>> want, which is to force the environment to a fixed location regardless
> >>> of the boot device.
> >>>
> >>> I think Tommaso's method is better, because as it is, the users cannot
> >>> override it any more.
> >>
> >> Doesn't 1/4 patch break env on pretty much every single imx8m board ?
> >
> > For me, patch 1/4 fixed the issue of not being able to define a fixed
> > environment location.  It now sits where I put it.  It also allows me
> > to boot over USB without having to define ENV_IS_NOWHERE.
>
> Sure, it also likely breaks every other MX8M board which does not define
> that env_get_location() on board level, i.e. every MX8M board.

env_get_location is defined in env/env.c
Marek Vasut Feb. 1, 2022, 11:43 a.m. UTC | #11
On 2/1/22 12:23, Adam Ford wrote:
> On Tue, Feb 1, 2022 at 5:22 AM Marek Vasut <marex@denx.de> wrote:
>>
>> On 2/1/22 12:16, Adam Ford wrote:
>>> On Mon, Jan 31, 2022 at 9:18 PM Marek Vasut <marex@denx.de> wrote:
>>>>
>>>> On 2/1/22 01:20, Adam Ford wrote:
>>>>> On Mon, Jan 31, 2022 at 4:16 PM Tommaso Merciai
>>>>> <tommaso.merciai@amarulasolutions.com> wrote:
>>>>>>
>>>>>> On Mon, Jan 31, 2022 at 06:03:58PM +0100, Marek Vasut wrote:
>>>>>>> On 1/31/22 17:58, Tommaso Merciai wrote:
>>>>>>>> Override env_get_location function at board level, previously dropped
>>>>>>>> down from arch/arm/mach-imx/imx8m/soc.c
>>>>>>>>
>>>>>>>> References:
>>>>>>>>      - commit 98af80d3c969e69a1b8ce98bb20e5ad844022da2
>>>>>>>>
>>>>>>>> Signed-off-by: Tommaso Merciai <tommaso.merciai@amarulasolutions.com>
>>>>>>>> ---
>>>>>>>>      board/phytec/phycore_imx8mp/phycore-imx8mp.c | 33 ++++++++++++++++++++
>>>>>>>>      1 file changed, 33 insertions(+)
>>>>>>>>
>>>>>>>> diff --git a/board/phytec/phycore_imx8mp/phycore-imx8mp.c b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
>>>>>>>> index a8f0821437..05926eefa3 100644
>>>>>>>> --- a/board/phytec/phycore_imx8mp/phycore-imx8mp.c
>>>>>>>> +++ b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
>>>>>>>> @@ -11,9 +11,42 @@
>>>>>>>>      #include <asm/mach-imx/boot_mode.h>
>>>>>>>>      #include <env.h>
>>>>>>>>      #include <miiphy.h>
>>>>>>>> +#include <env_internal.h>
>>>>>>>>      DECLARE_GLOBAL_DATA_PTR;
>>>>>>>> +enum env_location env_get_location(enum env_operation op, int prio)
>>>>>>>> +{
>>>>>>>
>>>>>>> Why don't you just turn this into default __weak function and override it on
>>>>>>> board level when it is really needed to be overridden ?
>>>>>>
>>>>>> Hi Marek,
>>>>>> env_get_location is already declared as __weak, check env/env.c. We
>>>>>> can't override it 2 times.
>>>>>
>>>>> The original version (before it was added in 2707faf01f04
>>>>> ("imx8mn/imx8mp: override env_get_offset and env_get_location") is
>>>>> located in env/env.c and for my board, this is the preferred method.
>>>>> This replacement method actually is the opposite behavior from what I
>>>>> want, which is to force the environment to a fixed location regardless
>>>>> of the boot device.
>>>>>
>>>>> I think Tommaso's method is better, because as it is, the users cannot
>>>>> override it any more.
>>>>
>>>> Doesn't 1/4 patch break env on pretty much every single imx8m board ?
>>>
>>> For me, patch 1/4 fixed the issue of not being able to define a fixed
>>> environment location.  It now sits where I put it.  It also allows me
>>> to boot over USB without having to define ENV_IS_NOWHERE.
>>
>> Sure, it also likely breaks every other MX8M board which does not define
>> that env_get_location() on board level, i.e. every MX8M board.
> 
> env_get_location is defined in env/env.c

Except not the NXP SoC-specific variant of it which is pulled in by most 
of the MX8M boards, and which is removed in 1/4, thus 1/4 changes the 
behavior of many boards.
Marek Vasut Feb. 1, 2022, 11:45 a.m. UTC | #12
On 2/1/22 12:22, Adam Ford wrote:
> On Tue, Feb 1, 2022 at 3:09 AM Tommaso Merciai
> <tommaso.merciai@amarulasolutions.com> wrote:
>>
>> On Tue, Feb 01, 2022 at 04:16:52AM +0100, Marek Vasut wrote:
>>> On 1/31/22 23:15, Tommaso Merciai wrote:
>>>> On Mon, Jan 31, 2022 at 06:03:58PM +0100, Marek Vasut wrote:
>>>>> On 1/31/22 17:58, Tommaso Merciai wrote:
>>>>>> Override env_get_location function at board level, previously dropped
>>>>>> down from arch/arm/mach-imx/imx8m/soc.c
>>>>>>
>>>>>> References:
>>>>>>     - commit 98af80d3c969e69a1b8ce98bb20e5ad844022da2
>>>>>>
>>>>>> Signed-off-by: Tommaso Merciai <tommaso.merciai@amarulasolutions.com>
>>>>>> ---
>>>>>>     board/phytec/phycore_imx8mp/phycore-imx8mp.c | 33 ++++++++++++++++++++
>>>>>>     1 file changed, 33 insertions(+)
>>>>>>
>>>>>> diff --git a/board/phytec/phycore_imx8mp/phycore-imx8mp.c b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
>>>>>> index a8f0821437..05926eefa3 100644
>>>>>> --- a/board/phytec/phycore_imx8mp/phycore-imx8mp.c
>>>>>> +++ b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
>>>>>> @@ -11,9 +11,42 @@
>>>>>>     #include <asm/mach-imx/boot_mode.h>
>>>>>>     #include <env.h>
>>>>>>     #include <miiphy.h>
>>>>>> +#include <env_internal.h>
>>>>>>     DECLARE_GLOBAL_DATA_PTR;
>>>>>> +enum env_location env_get_location(enum env_operation op, int prio)
>>>>>> +{
>>>>>
>>>>> Why don't you just turn this into default __weak function and override it on
>>>>> board level when it is really needed to be overridden ?
>>>>
>>>> Hi Marek,
>>>> env_get_location is already declared as __weak, check env/env.c. We
>>>> can't override it 2 times.
>>>
>>> Oh, it is this problem with missing ability to define multiple levels of
>>> symbol strength.
>>>
>>> __weak enum env_location arch_env_get_location(enum env_operation op, int
>>> prio)
>>> {
>>>          if (prio >= ARRAY_SIZE(env_locations))
>>>                  return ENVL_UNKNOWN;
>>>
>>>          return env_locations[prio];
>>> }
>>>
>>> __weak enum env_location board_env_get_location(enum env_operation op, int
>>> prio)
>>> {
>>>        return arch_env_get_location(op, prio);
>>> }
>>>
>>> __weak enum env_location env_get_location(enum env_operation op, int prio)
>>> {
>>>        return board_env_get_location(op, prio);
>>> }
>>>
>>> By default, the compiler will optimize it all out. If you have arch-specific
>>> default (like imx does), implement arch_env_get_location(), if you have even
>>> board-specific default (like your board likely does), implement
>>> board_env_get_location(), if you need to override both, then override
>>> env_get_location() (unlikely).
>>>
>>> This is also inline with all the other arch_*() and board_*() functions we
>>> have, and you won't have much duplication either.
>>
>> Hi Marek,
>> Thanks for the tips, then if I understand correctly, your idea is: use:
>>
>> arch_env_get_location in (soc.c)
>>
>> In this way imx8m users can override this function at board level using:
>>
>> board_env_get_location
>>
>> right?
> 
> What about those of us who want to use the default option found in
> env.c?  It seems like we're creating more abstraction to address the
> abstraction we don't all want.

On MX8M platforms which have their own arch-specific default, you 
override the env settings on your board level with the same stuff that 
is in env/env.c already. Most of the MX8M users will likely stick to the 
arch-specific MX8M default as they do so far.

> From my interpretation, the whole
> point of creating the default in env.c was to let people define the
> location of their environment, and this function in soc.c undid that.
> If people want it for their boards, put this function in their boards,
> otherwise, just use the default, or write your own.

[...]
Adam Ford Feb. 1, 2022, 11:51 a.m. UTC | #13
On Tue, Feb 1, 2022 at 5:46 AM Marek Vasut <marex@denx.de> wrote:
>
> On 2/1/22 12:22, Adam Ford wrote:
> > On Tue, Feb 1, 2022 at 3:09 AM Tommaso Merciai
> > <tommaso.merciai@amarulasolutions.com> wrote:
> >>
> >> On Tue, Feb 01, 2022 at 04:16:52AM +0100, Marek Vasut wrote:
> >>> On 1/31/22 23:15, Tommaso Merciai wrote:
> >>>> On Mon, Jan 31, 2022 at 06:03:58PM +0100, Marek Vasut wrote:
> >>>>> On 1/31/22 17:58, Tommaso Merciai wrote:
> >>>>>> Override env_get_location function at board level, previously dropped
> >>>>>> down from arch/arm/mach-imx/imx8m/soc.c
> >>>>>>
> >>>>>> References:
> >>>>>>     - commit 98af80d3c969e69a1b8ce98bb20e5ad844022da2
> >>>>>>
> >>>>>> Signed-off-by: Tommaso Merciai <tommaso.merciai@amarulasolutions.com>
> >>>>>> ---
> >>>>>>     board/phytec/phycore_imx8mp/phycore-imx8mp.c | 33 ++++++++++++++++++++
> >>>>>>     1 file changed, 33 insertions(+)
> >>>>>>
> >>>>>> diff --git a/board/phytec/phycore_imx8mp/phycore-imx8mp.c b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
> >>>>>> index a8f0821437..05926eefa3 100644
> >>>>>> --- a/board/phytec/phycore_imx8mp/phycore-imx8mp.c
> >>>>>> +++ b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
> >>>>>> @@ -11,9 +11,42 @@
> >>>>>>     #include <asm/mach-imx/boot_mode.h>
> >>>>>>     #include <env.h>
> >>>>>>     #include <miiphy.h>
> >>>>>> +#include <env_internal.h>
> >>>>>>     DECLARE_GLOBAL_DATA_PTR;
> >>>>>> +enum env_location env_get_location(enum env_operation op, int prio)
> >>>>>> +{
> >>>>>
> >>>>> Why don't you just turn this into default __weak function and override it on
> >>>>> board level when it is really needed to be overridden ?
> >>>>
> >>>> Hi Marek,
> >>>> env_get_location is already declared as __weak, check env/env.c. We
> >>>> can't override it 2 times.
> >>>
> >>> Oh, it is this problem with missing ability to define multiple levels of
> >>> symbol strength.
> >>>
> >>> __weak enum env_location arch_env_get_location(enum env_operation op, int
> >>> prio)
> >>> {
> >>>          if (prio >= ARRAY_SIZE(env_locations))
> >>>                  return ENVL_UNKNOWN;
> >>>
> >>>          return env_locations[prio];
> >>> }
> >>>
> >>> __weak enum env_location board_env_get_location(enum env_operation op, int
> >>> prio)
> >>> {
> >>>        return arch_env_get_location(op, prio);
> >>> }
> >>>
> >>> __weak enum env_location env_get_location(enum env_operation op, int prio)
> >>> {
> >>>        return board_env_get_location(op, prio);
> >>> }
> >>>
> >>> By default, the compiler will optimize it all out. If you have arch-specific
> >>> default (like imx does), implement arch_env_get_location(), if you have even
> >>> board-specific default (like your board likely does), implement
> >>> board_env_get_location(), if you need to override both, then override
> >>> env_get_location() (unlikely).
> >>>
> >>> This is also inline with all the other arch_*() and board_*() functions we
> >>> have, and you won't have much duplication either.
> >>
> >> Hi Marek,
> >> Thanks for the tips, then if I understand correctly, your idea is: use:
> >>
> >> arch_env_get_location in (soc.c)
> >>
> >> In this way imx8m users can override this function at board level using:
> >>
> >> board_env_get_location
> >>
> >> right?
> >
> > What about those of us who want to use the default option found in
> > env.c?  It seems like we're creating more abstraction to address the
> > abstraction we don't all want.
>
> On MX8M platforms which have their own arch-specific default, you
> override the env settings on your board level with the same stuff that
> is in env/env.c already. Most of the MX8M users will likely stick to the
> arch-specific MX8M default as they do so far.

This function is encapsulated in an ifdef which is only true for 8MN
and 8MP.  It's not defined for 8MM or 8MQ, so those boards already
default back to the __weak function in env.c

adam

>
> > From my interpretation, the whole
> > point of creating the default in env.c was to let people define the
> > location of their environment, and this function in soc.c undid that.
> > If people want it for their boards, put this function in their boards,
> > otherwise, just use the default, or write your own.
>
> [...]
Marek Vasut Feb. 1, 2022, 11:52 a.m. UTC | #14
On 2/1/22 12:51, Adam Ford wrote:
> On Tue, Feb 1, 2022 at 5:46 AM Marek Vasut <marex@denx.de> wrote:
>>
>> On 2/1/22 12:22, Adam Ford wrote:
>>> On Tue, Feb 1, 2022 at 3:09 AM Tommaso Merciai
>>> <tommaso.merciai@amarulasolutions.com> wrote:
>>>>
>>>> On Tue, Feb 01, 2022 at 04:16:52AM +0100, Marek Vasut wrote:
>>>>> On 1/31/22 23:15, Tommaso Merciai wrote:
>>>>>> On Mon, Jan 31, 2022 at 06:03:58PM +0100, Marek Vasut wrote:
>>>>>>> On 1/31/22 17:58, Tommaso Merciai wrote:
>>>>>>>> Override env_get_location function at board level, previously dropped
>>>>>>>> down from arch/arm/mach-imx/imx8m/soc.c
>>>>>>>>
>>>>>>>> References:
>>>>>>>>      - commit 98af80d3c969e69a1b8ce98bb20e5ad844022da2
>>>>>>>>
>>>>>>>> Signed-off-by: Tommaso Merciai <tommaso.merciai@amarulasolutions.com>
>>>>>>>> ---
>>>>>>>>      board/phytec/phycore_imx8mp/phycore-imx8mp.c | 33 ++++++++++++++++++++
>>>>>>>>      1 file changed, 33 insertions(+)
>>>>>>>>
>>>>>>>> diff --git a/board/phytec/phycore_imx8mp/phycore-imx8mp.c b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
>>>>>>>> index a8f0821437..05926eefa3 100644
>>>>>>>> --- a/board/phytec/phycore_imx8mp/phycore-imx8mp.c
>>>>>>>> +++ b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
>>>>>>>> @@ -11,9 +11,42 @@
>>>>>>>>      #include <asm/mach-imx/boot_mode.h>
>>>>>>>>      #include <env.h>
>>>>>>>>      #include <miiphy.h>
>>>>>>>> +#include <env_internal.h>
>>>>>>>>      DECLARE_GLOBAL_DATA_PTR;
>>>>>>>> +enum env_location env_get_location(enum env_operation op, int prio)
>>>>>>>> +{
>>>>>>>
>>>>>>> Why don't you just turn this into default __weak function and override it on
>>>>>>> board level when it is really needed to be overridden ?
>>>>>>
>>>>>> Hi Marek,
>>>>>> env_get_location is already declared as __weak, check env/env.c. We
>>>>>> can't override it 2 times.
>>>>>
>>>>> Oh, it is this problem with missing ability to define multiple levels of
>>>>> symbol strength.
>>>>>
>>>>> __weak enum env_location arch_env_get_location(enum env_operation op, int
>>>>> prio)
>>>>> {
>>>>>           if (prio >= ARRAY_SIZE(env_locations))
>>>>>                   return ENVL_UNKNOWN;
>>>>>
>>>>>           return env_locations[prio];
>>>>> }
>>>>>
>>>>> __weak enum env_location board_env_get_location(enum env_operation op, int
>>>>> prio)
>>>>> {
>>>>>         return arch_env_get_location(op, prio);
>>>>> }
>>>>>
>>>>> __weak enum env_location env_get_location(enum env_operation op, int prio)
>>>>> {
>>>>>         return board_env_get_location(op, prio);
>>>>> }
>>>>>
>>>>> By default, the compiler will optimize it all out. If you have arch-specific
>>>>> default (like imx does), implement arch_env_get_location(), if you have even
>>>>> board-specific default (like your board likely does), implement
>>>>> board_env_get_location(), if you need to override both, then override
>>>>> env_get_location() (unlikely).
>>>>>
>>>>> This is also inline with all the other arch_*() and board_*() functions we
>>>>> have, and you won't have much duplication either.
>>>>
>>>> Hi Marek,
>>>> Thanks for the tips, then if I understand correctly, your idea is: use:
>>>>
>>>> arch_env_get_location in (soc.c)
>>>>
>>>> In this way imx8m users can override this function at board level using:
>>>>
>>>> board_env_get_location
>>>>
>>>> right?
>>>
>>> What about those of us who want to use the default option found in
>>> env.c?  It seems like we're creating more abstraction to address the
>>> abstraction we don't all want.
>>
>> On MX8M platforms which have their own arch-specific default, you
>> override the env settings on your board level with the same stuff that
>> is in env/env.c already. Most of the MX8M users will likely stick to the
>> arch-specific MX8M default as they do so far.
> 
> This function is encapsulated in an ifdef which is only true for 8MN
> and 8MP.  It's not defined for 8MM or 8MQ, so those boards already
> default back to the __weak function in env.c

So, it is also inconsistent, very nice.

Anyway, if we have arch and board level overrides, then we cover all the 
requirements on all MX8Ms, without breaking existing boards.
Tommaso Merciai Feb. 1, 2022, 11:56 a.m. UTC | #15
On Tue, Feb 01, 2022 at 05:22:06AM -0600, Adam Ford wrote:
> On Tue, Feb 1, 2022 at 3:09 AM Tommaso Merciai
> <tommaso.merciai@amarulasolutions.com> wrote:
> >
> > On Tue, Feb 01, 2022 at 04:16:52AM +0100, Marek Vasut wrote:
> > > On 1/31/22 23:15, Tommaso Merciai wrote:
> > > > On Mon, Jan 31, 2022 at 06:03:58PM +0100, Marek Vasut wrote:
> > > > > On 1/31/22 17:58, Tommaso Merciai wrote:
> > > > > > Override env_get_location function at board level, previously dropped
> > > > > > down from arch/arm/mach-imx/imx8m/soc.c
> > > > > >
> > > > > > References:
> > > > > >    - commit 98af80d3c969e69a1b8ce98bb20e5ad844022da2
> > > > > >
> > > > > > Signed-off-by: Tommaso Merciai <tommaso.merciai@amarulasolutions.com>
> > > > > > ---
> > > > > >    board/phytec/phycore_imx8mp/phycore-imx8mp.c | 33 ++++++++++++++++++++
> > > > > >    1 file changed, 33 insertions(+)
> > > > > >
> > > > > > diff --git a/board/phytec/phycore_imx8mp/phycore-imx8mp.c b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
> > > > > > index a8f0821437..05926eefa3 100644
> > > > > > --- a/board/phytec/phycore_imx8mp/phycore-imx8mp.c
> > > > > > +++ b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
> > > > > > @@ -11,9 +11,42 @@
> > > > > >    #include <asm/mach-imx/boot_mode.h>
> > > > > >    #include <env.h>
> > > > > >    #include <miiphy.h>
> > > > > > +#include <env_internal.h>
> > > > > >    DECLARE_GLOBAL_DATA_PTR;
> > > > > > +enum env_location env_get_location(enum env_operation op, int prio)
> > > > > > +{
> > > > >
> > > > > Why don't you just turn this into default __weak function and override it on
> > > > > board level when it is really needed to be overridden ?
> > > >
> > > > Hi Marek,
> > > > env_get_location is already declared as __weak, check env/env.c. We
> > > > can't override it 2 times.
> > >
> > > Oh, it is this problem with missing ability to define multiple levels of
> > > symbol strength.
> > >
> > > __weak enum env_location arch_env_get_location(enum env_operation op, int
> > > prio)
> > > {
> > >         if (prio >= ARRAY_SIZE(env_locations))
> > >                 return ENVL_UNKNOWN;
> > >
> > >         return env_locations[prio];
> > > }
> > >
> > > __weak enum env_location board_env_get_location(enum env_operation op, int
> > > prio)
> > > {
> > >       return arch_env_get_location(op, prio);
> > > }
> > >
> > > __weak enum env_location env_get_location(enum env_operation op, int prio)
> > > {
> > >       return board_env_get_location(op, prio);
> > > }
> > >
> > > By default, the compiler will optimize it all out. If you have arch-specific
> > > default (like imx does), implement arch_env_get_location(), if you have even
> > > board-specific default (like your board likely does), implement
> > > board_env_get_location(), if you need to override both, then override
> > > env_get_location() (unlikely).
> > >
> > > This is also inline with all the other arch_*() and board_*() functions we
> > > have, and you won't have much duplication either.
> >
> > Hi Marek,
> > Thanks for the tips, then if I understand correctly, your idea is: use:
> >
> > arch_env_get_location in (soc.c)
> >
> > In this way imx8m users can override this function at board level using:
> >
> > board_env_get_location
> >
> > right?
> 
> What about those of us who want to use the default option found in
> env.c?

Hi Adam,
You are right. Mmm in this way you have to duplicate the code of env.c
into your board.c. This doesn't look very functional.
I think remove it from soc.c is the right way.

> It seems like we're creating more abstraction to address the
> abstraction we don't all want.  From my interpretation, the whole
> point of creating the default in env.c was to let people define the
> location of their environment, and this function in soc.c undid that.
> If people want it for their boards, put this function in their boards,
> otherwise, just use the default, or write your own.

Ack.

tommaso
> 
> adam
> >
> > Thanks,
> > Tommaso
> >
> > --
> > Tommaso Merciai
> > Embedded Linux Developer
> > tommaso.merciai@amarulasolutions.com
> > __________________________________
> >
> > Amarula Solutions SRL
> > Via Le Canevare 30, 31100 Treviso, Veneto, IT
> > T. +39 042 243 5310
> > info@amarulasolutions.com
> > www.amarulasolutions.com
Marek Vasut Feb. 1, 2022, 11:57 a.m. UTC | #16
On 2/1/22 12:56, Tommaso Merciai wrote:
> On Tue, Feb 01, 2022 at 05:22:06AM -0600, Adam Ford wrote:
>> On Tue, Feb 1, 2022 at 3:09 AM Tommaso Merciai
>> <tommaso.merciai@amarulasolutions.com> wrote:
>>>
>>> On Tue, Feb 01, 2022 at 04:16:52AM +0100, Marek Vasut wrote:
>>>> On 1/31/22 23:15, Tommaso Merciai wrote:
>>>>> On Mon, Jan 31, 2022 at 06:03:58PM +0100, Marek Vasut wrote:
>>>>>> On 1/31/22 17:58, Tommaso Merciai wrote:
>>>>>>> Override env_get_location function at board level, previously dropped
>>>>>>> down from arch/arm/mach-imx/imx8m/soc.c
>>>>>>>
>>>>>>> References:
>>>>>>>     - commit 98af80d3c969e69a1b8ce98bb20e5ad844022da2
>>>>>>>
>>>>>>> Signed-off-by: Tommaso Merciai <tommaso.merciai@amarulasolutions.com>
>>>>>>> ---
>>>>>>>     board/phytec/phycore_imx8mp/phycore-imx8mp.c | 33 ++++++++++++++++++++
>>>>>>>     1 file changed, 33 insertions(+)
>>>>>>>
>>>>>>> diff --git a/board/phytec/phycore_imx8mp/phycore-imx8mp.c b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
>>>>>>> index a8f0821437..05926eefa3 100644
>>>>>>> --- a/board/phytec/phycore_imx8mp/phycore-imx8mp.c
>>>>>>> +++ b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
>>>>>>> @@ -11,9 +11,42 @@
>>>>>>>     #include <asm/mach-imx/boot_mode.h>
>>>>>>>     #include <env.h>
>>>>>>>     #include <miiphy.h>
>>>>>>> +#include <env_internal.h>
>>>>>>>     DECLARE_GLOBAL_DATA_PTR;
>>>>>>> +enum env_location env_get_location(enum env_operation op, int prio)
>>>>>>> +{
>>>>>>
>>>>>> Why don't you just turn this into default __weak function and override it on
>>>>>> board level when it is really needed to be overridden ?
>>>>>
>>>>> Hi Marek,
>>>>> env_get_location is already declared as __weak, check env/env.c. We
>>>>> can't override it 2 times.
>>>>
>>>> Oh, it is this problem with missing ability to define multiple levels of
>>>> symbol strength.
>>>>
>>>> __weak enum env_location arch_env_get_location(enum env_operation op, int
>>>> prio)
>>>> {
>>>>          if (prio >= ARRAY_SIZE(env_locations))
>>>>                  return ENVL_UNKNOWN;
>>>>
>>>>          return env_locations[prio];
>>>> }
>>>>
>>>> __weak enum env_location board_env_get_location(enum env_operation op, int
>>>> prio)
>>>> {
>>>>        return arch_env_get_location(op, prio);
>>>> }
>>>>
>>>> __weak enum env_location env_get_location(enum env_operation op, int prio)
>>>> {
>>>>        return board_env_get_location(op, prio);
>>>> }
>>>>
>>>> By default, the compiler will optimize it all out. If you have arch-specific
>>>> default (like imx does), implement arch_env_get_location(), if you have even
>>>> board-specific default (like your board likely does), implement
>>>> board_env_get_location(), if you need to override both, then override
>>>> env_get_location() (unlikely).
>>>>
>>>> This is also inline with all the other arch_*() and board_*() functions we
>>>> have, and you won't have much duplication either.
>>>
>>> Hi Marek,
>>> Thanks for the tips, then if I understand correctly, your idea is: use:
>>>
>>> arch_env_get_location in (soc.c)
>>>
>>> In this way imx8m users can override this function at board level using:
>>>
>>> board_env_get_location
>>>
>>> right?
>>
>> What about those of us who want to use the default option found in
>> env.c?
> 
> Hi Adam,
> You are right. Mmm in this way you have to duplicate the code of env.c
> into your board.c. This doesn't look very functional.
> I think remove it from soc.c is the right way.

How do you propose to fix all the boards which depend on the current 
arch-specific behavior, duplicate that code in multiple copies into 
board code ? That will be a lot of duplication too.
Adam Ford Feb. 1, 2022, 12:26 p.m. UTC | #17
On Tue, Feb 1, 2022 at 5:58 AM Marek Vasut <marex@denx.de> wrote:
>
> On 2/1/22 12:56, Tommaso Merciai wrote:
> > On Tue, Feb 01, 2022 at 05:22:06AM -0600, Adam Ford wrote:
> >> On Tue, Feb 1, 2022 at 3:09 AM Tommaso Merciai
> >> <tommaso.merciai@amarulasolutions.com> wrote:
> >>>
> >>> On Tue, Feb 01, 2022 at 04:16:52AM +0100, Marek Vasut wrote:
> >>>> On 1/31/22 23:15, Tommaso Merciai wrote:
> >>>>> On Mon, Jan 31, 2022 at 06:03:58PM +0100, Marek Vasut wrote:
> >>>>>> On 1/31/22 17:58, Tommaso Merciai wrote:
> >>>>>>> Override env_get_location function at board level, previously dropped
> >>>>>>> down from arch/arm/mach-imx/imx8m/soc.c
> >>>>>>>
> >>>>>>> References:
> >>>>>>>     - commit 98af80d3c969e69a1b8ce98bb20e5ad844022da2
> >>>>>>>
> >>>>>>> Signed-off-by: Tommaso Merciai <tommaso.merciai@amarulasolutions.com>
> >>>>>>> ---
> >>>>>>>     board/phytec/phycore_imx8mp/phycore-imx8mp.c | 33 ++++++++++++++++++++
> >>>>>>>     1 file changed, 33 insertions(+)
> >>>>>>>
> >>>>>>> diff --git a/board/phytec/phycore_imx8mp/phycore-imx8mp.c b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
> >>>>>>> index a8f0821437..05926eefa3 100644
> >>>>>>> --- a/board/phytec/phycore_imx8mp/phycore-imx8mp.c
> >>>>>>> +++ b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
> >>>>>>> @@ -11,9 +11,42 @@
> >>>>>>>     #include <asm/mach-imx/boot_mode.h>
> >>>>>>>     #include <env.h>
> >>>>>>>     #include <miiphy.h>
> >>>>>>> +#include <env_internal.h>
> >>>>>>>     DECLARE_GLOBAL_DATA_PTR;
> >>>>>>> +enum env_location env_get_location(enum env_operation op, int prio)
> >>>>>>> +{
> >>>>>>
> >>>>>> Why don't you just turn this into default __weak function and override it on
> >>>>>> board level when it is really needed to be overridden ?
> >>>>>
> >>>>> Hi Marek,
> >>>>> env_get_location is already declared as __weak, check env/env.c. We
> >>>>> can't override it 2 times.
> >>>>
> >>>> Oh, it is this problem with missing ability to define multiple levels of
> >>>> symbol strength.
> >>>>
> >>>> __weak enum env_location arch_env_get_location(enum env_operation op, int
> >>>> prio)
> >>>> {
> >>>>          if (prio >= ARRAY_SIZE(env_locations))
> >>>>                  return ENVL_UNKNOWN;
> >>>>
> >>>>          return env_locations[prio];
> >>>> }
> >>>>
> >>>> __weak enum env_location board_env_get_location(enum env_operation op, int
> >>>> prio)
> >>>> {
> >>>>        return arch_env_get_location(op, prio);
> >>>> }
> >>>>
> >>>> __weak enum env_location env_get_location(enum env_operation op, int prio)
> >>>> {
> >>>>        return board_env_get_location(op, prio);
> >>>> }
> >>>>
> >>>> By default, the compiler will optimize it all out. If you have arch-specific
> >>>> default (like imx does), implement arch_env_get_location(), if you have even
> >>>> board-specific default (like your board likely does), implement
> >>>> board_env_get_location(), if you need to override both, then override
> >>>> env_get_location() (unlikely).
> >>>>
> >>>> This is also inline with all the other arch_*() and board_*() functions we
> >>>> have, and you won't have much duplication either.
> >>>
> >>> Hi Marek,
> >>> Thanks for the tips, then if I understand correctly, your idea is: use:
> >>>
> >>> arch_env_get_location in (soc.c)
> >>>
> >>> In this way imx8m users can override this function at board level using:
> >>>
> >>> board_env_get_location
> >>>
> >>> right?
> >>
> >> What about those of us who want to use the default option found in
> >> env.c?
> >
> > Hi Adam,
> > You are right. Mmm in this way you have to duplicate the code of env.c
> > into your board.c. This doesn't look very functional.
> > I think remove it from soc.c is the right way.
>
> How do you propose to fix all the boards which depend on the current
> arch-specific behavior, duplicate that code in multiple copies into
> board code ? That will be a lot of duplication too.

I would think we should ask the users of 8MP and 8MN to test theirs
without this patch to see if it impacts them.  The original patch was
done by NXP, so I assume they want it for their EVK's.  This code
explicitly makes it impossible to set the environment location to a
device that was not a boot device.  From what I can tell, we're
talking about a small number of boards since it's just 8MN and 8MP.

What about a compromise:

Rename env_get_location() in the soc.c to imx_env_get_location().

From the 8MN and 8MP boards (for people who want it or need it), write
a board function called env_get_location() which calls
imx_env_get_location().  This way the code remains, and people can
choose whether they want this behavior, or the behavior or the default
behavior used by 8MM and 8MQ.   It's less duplicated code, and it
doesn't break the default behavior for those who want that.

adam
Marek Vasut Feb. 1, 2022, 12:38 p.m. UTC | #18
On 2/1/22 13:26, Adam Ford wrote:
> On Tue, Feb 1, 2022 at 5:58 AM Marek Vasut <marex@denx.de> wrote:
>>
>> On 2/1/22 12:56, Tommaso Merciai wrote:
>>> On Tue, Feb 01, 2022 at 05:22:06AM -0600, Adam Ford wrote:
>>>> On Tue, Feb 1, 2022 at 3:09 AM Tommaso Merciai
>>>> <tommaso.merciai@amarulasolutions.com> wrote:
>>>>>
>>>>> On Tue, Feb 01, 2022 at 04:16:52AM +0100, Marek Vasut wrote:
>>>>>> On 1/31/22 23:15, Tommaso Merciai wrote:
>>>>>>> On Mon, Jan 31, 2022 at 06:03:58PM +0100, Marek Vasut wrote:
>>>>>>>> On 1/31/22 17:58, Tommaso Merciai wrote:
>>>>>>>>> Override env_get_location function at board level, previously dropped
>>>>>>>>> down from arch/arm/mach-imx/imx8m/soc.c
>>>>>>>>>
>>>>>>>>> References:
>>>>>>>>>      - commit 98af80d3c969e69a1b8ce98bb20e5ad844022da2
>>>>>>>>>
>>>>>>>>> Signed-off-by: Tommaso Merciai <tommaso.merciai@amarulasolutions.com>
>>>>>>>>> ---
>>>>>>>>>      board/phytec/phycore_imx8mp/phycore-imx8mp.c | 33 ++++++++++++++++++++
>>>>>>>>>      1 file changed, 33 insertions(+)
>>>>>>>>>
>>>>>>>>> diff --git a/board/phytec/phycore_imx8mp/phycore-imx8mp.c b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
>>>>>>>>> index a8f0821437..05926eefa3 100644
>>>>>>>>> --- a/board/phytec/phycore_imx8mp/phycore-imx8mp.c
>>>>>>>>> +++ b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
>>>>>>>>> @@ -11,9 +11,42 @@
>>>>>>>>>      #include <asm/mach-imx/boot_mode.h>
>>>>>>>>>      #include <env.h>
>>>>>>>>>      #include <miiphy.h>
>>>>>>>>> +#include <env_internal.h>
>>>>>>>>>      DECLARE_GLOBAL_DATA_PTR;
>>>>>>>>> +enum env_location env_get_location(enum env_operation op, int prio)
>>>>>>>>> +{
>>>>>>>>
>>>>>>>> Why don't you just turn this into default __weak function and override it on
>>>>>>>> board level when it is really needed to be overridden ?
>>>>>>>
>>>>>>> Hi Marek,
>>>>>>> env_get_location is already declared as __weak, check env/env.c. We
>>>>>>> can't override it 2 times.
>>>>>>
>>>>>> Oh, it is this problem with missing ability to define multiple levels of
>>>>>> symbol strength.
>>>>>>
>>>>>> __weak enum env_location arch_env_get_location(enum env_operation op, int
>>>>>> prio)
>>>>>> {
>>>>>>           if (prio >= ARRAY_SIZE(env_locations))
>>>>>>                   return ENVL_UNKNOWN;
>>>>>>
>>>>>>           return env_locations[prio];
>>>>>> }
>>>>>>
>>>>>> __weak enum env_location board_env_get_location(enum env_operation op, int
>>>>>> prio)
>>>>>> {
>>>>>>         return arch_env_get_location(op, prio);
>>>>>> }
>>>>>>
>>>>>> __weak enum env_location env_get_location(enum env_operation op, int prio)
>>>>>> {
>>>>>>         return board_env_get_location(op, prio);
>>>>>> }
>>>>>>
>>>>>> By default, the compiler will optimize it all out. If you have arch-specific
>>>>>> default (like imx does), implement arch_env_get_location(), if you have even
>>>>>> board-specific default (like your board likely does), implement
>>>>>> board_env_get_location(), if you need to override both, then override
>>>>>> env_get_location() (unlikely).
>>>>>>
>>>>>> This is also inline with all the other arch_*() and board_*() functions we
>>>>>> have, and you won't have much duplication either.
>>>>>
>>>>> Hi Marek,
>>>>> Thanks for the tips, then if I understand correctly, your idea is: use:
>>>>>
>>>>> arch_env_get_location in (soc.c)
>>>>>
>>>>> In this way imx8m users can override this function at board level using:
>>>>>
>>>>> board_env_get_location
>>>>>
>>>>> right?
>>>>
>>>> What about those of us who want to use the default option found in
>>>> env.c?
>>>
>>> Hi Adam,
>>> You are right. Mmm in this way you have to duplicate the code of env.c
>>> into your board.c. This doesn't look very functional.
>>> I think remove it from soc.c is the right way.
>>
>> How do you propose to fix all the boards which depend on the current
>> arch-specific behavior, duplicate that code in multiple copies into
>> board code ? That will be a lot of duplication too.
> 
> I would think we should ask the users of 8MP and 8MN to test theirs
> without this patch to see if it impacts them.  The original patch was
> done by NXP, so I assume they want it for their EVK's.  This code
> explicitly makes it impossible to set the environment location to a
> device that was not a boot device.  From what I can tell, we're
> talking about a small number of boards since it's just 8MN and 8MP.
> 
> What about a compromise:
> 
> Rename env_get_location() in the soc.c to imx_env_get_location().
> 
>>From the 8MN and 8MP boards (for people who want it or need it), write
> a board function called env_get_location() which calls
> imx_env_get_location().  This way the code remains, and people can
> choose whether they want this behavior, or the behavior or the default
> behavior used by 8MM and 8MQ.   It's less duplicated code, and it
> doesn't break the default behavior for those who want that.

We are still talking about 40 lines of code being copied to 12 different 
places, or, 12 boards which need to be fixed up -- instead of -- one 
board which needs to be configured correctly to pick env from wherever 
it should pick it from instead of the NXP default.
diff mbox series

Patch

diff --git a/board/phytec/phycore_imx8mp/phycore-imx8mp.c b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
index a8f0821437..05926eefa3 100644
--- a/board/phytec/phycore_imx8mp/phycore-imx8mp.c
+++ b/board/phytec/phycore_imx8mp/phycore-imx8mp.c
@@ -11,9 +11,42 @@ 
 #include <asm/mach-imx/boot_mode.h>
 #include <env.h>
 #include <miiphy.h>
+#include <env_internal.h>
 
 DECLARE_GLOBAL_DATA_PTR;
 
+enum env_location env_get_location(enum env_operation op, int prio)
+{
+	enum boot_device dev = get_boot_device();
+	enum env_location env_loc = ENVL_UNKNOWN;
+
+	if (prio)
+		return env_loc;
+
+	if (IS_ENABLED(CONFIG_ENV_IS_IN_SPI_FLASH) && dev == QSPI_BOOT) {
+		env_loc = ENVL_SPI_FLASH;
+	} else if (IS_ENABLED(CONFIG_ENV_IS_IN_NAND) && dev == NAND_BOOT) {
+		env_loc = ENVL_NAND;
+	} else if (IS_ENABLED(CONFIG_ENV_IS_IN_MMC)) {
+		switch (dev) {
+		case SD1_BOOT:
+		case SD2_BOOT:
+		case SD3_BOOT:
+		case MMC1_BOOT:
+		case MMC2_BOOT:
+		case MMC3_BOOT:
+			env_loc = ENVL_MMC;
+			break;
+		default:
+			break;
+		}
+	} else if (IS_ENABLED(CONFIG_ENV_IS_NOWHERE)) {
+		env_loc = ENVL_MMC;
+	}
+
+	return env_loc;
+}
+
 static int setup_fec(void)
 {
 	struct iomuxc_gpr_base_regs *gpr =