[RFC PATCH 2/4] riscv: Cleanup ERRATA_THEAD_PBMT for rv32 svpbmt compile

[Guo Ren]

On Fri, Jul 8, 2022 at 4:51 PM Heiko Stübner <heiko at sntech.de> wrote:
>
> Hi Guo,
>
> Am Dienstag, 5. Juli 2022, 12:05:21 CEST schrieb guoren at kernel.org:
> > From: Guo Ren <guoren at linux.alibaba.com>
> >
> > Make compile cleaner and don't reference the THEAD_PBMT data struct when
> > CONFIG_ERRATA_THEAD_PBMT=y. Next, we could cleanly make svpbmt to
> > support rv32.
> >
> > Signed-off-by: Guo Ren <guoren at linux.alibaba.com>
> > Signed-off-by: Guo Ren <guoren at kernel.org>
> > ---
> >  arch/riscv/include/asm/errata_list.h | 11 ++++++++++-
> >  1 file changed, 10 insertions(+), 1 deletion(-)
> >
> > diff --git a/arch/riscv/include/asm/errata_list.h b/arch/riscv/include/asm/errata_list.h
> > index 416ead0f9a65..47175d91773d 100644
> > --- a/arch/riscv/include/asm/errata_list.h
> > +++ b/arch/riscv/include/asm/errata_list.h
> > @@ -47,6 +47,8 @@ asm(ALTERNATIVE("sfence.vma %0", "sfence.vma", SIFIVE_VENDOR_ID,    \
> >   * in the default case.
> >   */
> >  #define ALT_SVPBMT_SHIFT 61
> > +
> > +#ifdef CONFIG_ERRATA_THEAD_PBMT
>
> I don't really think that is necessary and actually makes the code
> more complex than needed.
>
> Each alternative-entry already has the dependency on
> CONFIG_* ... i.e. CONFIG_RISCV_ISA_SVPBMT and CONFIG_ERRATA_THEAD_PBMT
>
> When you look at alternative-macros.h you'll see this translating to the
> enable argument in ALT_NEW_CONTENT.
>
> So only when that is active is the alternative section added to the build.
> I.e. that translates to:
>
>         .if IS_ENABLED(CONFIG_ERRATA_THEAD_PBMT)
>         .pushsection .alternative, "a"
>         ...

Above can't affect the below:
                : "I"(prot##_SVPBMT >> ALT_SVPBMT_SHIFT),               \
                  "I"(prot##_THEAD >> ALT_THEAD_PBMT_SHIFT),            \
                  "I"(ALT_SVPBMT_SHIFT),                                \
                  "I"(ALT_THEAD_PBMT_SHIFT))

So CONFIG_ERRATA_THEAD_PBMT is not clean as you think, compiler still
process the  "I"(prot##_THEAD >> ALT_THEAD_PBMT_SHIFT)" &
"I"(ALT_THEAD_PBMT_SHIFT).

>
> So when CONFIG_ERRATA_THEAD_PBMT is disabled the whole alternative
> part never gets added already, so there shouldn't be any need to make the
> source more complicated.
>
>
> Heiko
>
> >  #define ALT_THEAD_PBMT_SHIFT 59
> >  #define ALT_SVPBMT(_val, prot)                                               \
> >  asm(ALTERNATIVE_2("li %0, 0\t\nnop",                                 \
> > @@ -60,7 +62,6 @@ asm(ALTERNATIVE_2("li %0, 0\t\nnop",                                        \
> >                 "I"(ALT_SVPBMT_SHIFT),                                \
> >                 "I"(ALT_THEAD_PBMT_SHIFT))
> >
> > -#ifdef CONFIG_ERRATA_THEAD_PBMT
> >  /*
> >   * IO/NOCACHE memory types are handled together with svpbmt,
> >   * so on T-Head chips, check if no other memory type is set,
> > @@ -90,6 +91,14 @@ asm volatile(ALTERNATIVE(                                          \
> >         "I"(ALT_THEAD_PBMT_SHIFT)                                     \
> >       : "t3")
> >  #else
> > +#define ALT_SVPBMT(_val, prot)                                               \
> > +asm(ALTERNATIVE("li %0, 0\t\nnop",                                   \
> > +              "li %0, %1\t\nslli %0,%0,%2", 0,                       \
> > +                     CPUFEATURE_SVPBMT, CONFIG_RISCV_ISA_SVPBMT)     \
> > +             : "=r"(_val)                                            \
> > +             : "I"(prot##_SVPBMT >> ALT_SVPBMT_SHIFT),               \
> > +               "I"(ALT_SVPBMT_SHIFT))
> > +
> >  #define ALT_THEAD_PMA(_val)
> >  #endif
> >
> >
>
>
>
>


-- 
Best Regards
 Guo Ren