Re: [PATCH] riscv: Disallow to build XIP_KERNEL with SOC_SIFIVE

[Palmer Dabbelt <palmer@dabbelt.com>]

On Fri, 30 Apr 2021 01:09:14 PDT (-0700), alex@ghiti.fr wrote:
> Le 4/29/21 à 8:13 AM, Alex Ghiti a écrit :
>> Le 4/29/21 à 5:11 AM, Vitaly Wool a écrit :
>>> On Thu, Apr 29, 2021 at 10:47 AM Alexandre Ghiti <alex@ghiti.fr> wrote:
>>>>
>>>> RISCV_ERRATA_ALTERNATIVE patches text at runtime which is not
>>>> possible when
>>>> the kernel is executed from the flash in XIP mode, and as the SIFIVE
>>>> errata must be fixed somehow, disallow to build a XIP kernel that
>>>> supports SIFIVE socs.
>>>
>>> Could you please hold off this patch for a bit? I will try to come up
>>> with an alternative solution. It should be possible to define a
>>> special section within the RW area and place the functions that need
>>> such patching there.
>>> Not that I like that much but at least we'll keep the ability to use
>>> XIP on SiFive.
>>
>> Ok, I'm wondering why I did not think of that...I'll give it a try just
>> to punish myself.
>>
>> Thanks Vitaly,
>>
>> Alex
>>
>
> I tried what you proposed and it works well, *callers* must be placed
> into this writable section in RAM and that's it, that may be more
> complicated if at some point the patched functions are generic but I
> think we can use an intermediate riscv function to patch instead or
> something else.
>
> The modifications I did only consist in putting alternative section in
> RAM and place the exception vector table in this section.
>
> If you can do the proper patch, I'll let you do it, otherwise I'll do
> that later as I have other things to do before.
>
> So Palmer you can drop this patch.

Great.  I was going to push back and just say "we should swap the 
alternatives over to the errata implemnetation for XIP, as the current 
ones are just a performance hit when unnecessary and XIP users are 
likely to build for a single SOC anyway".  That would mean we'd need to 
sort something out WRT errata that rely on design-specific 
functionality, and everything I came up with there was a headache.

It's way better if we can just full support the alternatives stuff, so 
I'll just wait for that.

>
> Thanks again,
>
>>>
>>> Best regards,
>>>     Vitaly
>>>
>>>> Signed-off-by: Alexandre Ghiti <alex@ghiti.fr>
>>>> ---
>>>>   arch/riscv/Kconfig.erratas | 2 +-
>>>>   arch/riscv/Kconfig.socs    | 1 +
>>>>   2 files changed, 2 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/arch/riscv/Kconfig.erratas b/arch/riscv/Kconfig.erratas
>>>> index d5d03ae8d685..9537dbd67357 100644
>>>> --- a/arch/riscv/Kconfig.erratas
>>>> +++ b/arch/riscv/Kconfig.erratas
>>>> @@ -2,7 +2,7 @@ menu "CPU errata selection"
>>>>
>>>>   config RISCV_ERRATA_ALTERNATIVE
>>>>          bool "RISC-V alternative scheme"
>>>> -       default y
>>>> +       default y if !XIP_KERNEL
>>>>          help
>>>>            This Kconfig allows the kernel to automatically patch the
>>>>            errata required by the execution platform at run time. The
>>>> diff --git a/arch/riscv/Kconfig.socs b/arch/riscv/Kconfig.socs
>>>> index 00c2b205654c..9cb38bc9d7cd 100644
>>>> --- a/arch/riscv/Kconfig.socs
>>>> +++ b/arch/riscv/Kconfig.socs
>>>> @@ -9,6 +9,7 @@ config SOC_MICROCHIP_POLARFIRE
>>>>
>>>>   config SOC_SIFIVE
>>>>          bool "SiFive SoCs"
>>>> +       depends on !XIP_KERNEL
>>>>          select SERIAL_SIFIVE if TTY
>>>>          select SERIAL_SIFIVE_CONSOLE if TTY
>>>>          select CLK_SIFIVE
>>>> --
>>>> 2.20.1
>>>>
>>>>
>>>> _______________________________________________
>>>> linux-riscv mailing list
>>>> linux-riscv@lists.infradead.org
>>>> http://lists.infradead.org/mailman/listinfo/linux-riscv
>>>
>>> _______________________________________________
>>> linux-riscv mailing list
>>> linux-riscv@lists.infradead.org
>>> http://lists.infradead.org/mailman/listinfo/linux-riscv
>>>
>>
>> _______________________________________________
>> linux-riscv mailing list
>> linux-riscv@lists.infradead.org
>> http://lists.infradead.org/mailman/listinfo/linux-riscv