On Tue, Apr 01, 2025 at 12:28:33PM -0700, Nathan Chancellor wrote:
> Hi Charlie,
> 
> On Wed, Mar 19, 2025 at 11:35:20AM -0700, Charlie Jenkins wrote:
> > Implement the runtime constant infrastructure for riscv. Use this
> > infrastructure to generate constants to be used by the d_hash()
> > function.
> > 
> > This is the riscv variant of commit 94a2bc0f611c ("arm64: add 'runtime
> > constant' support") and commit e3c92e81711d ("runtime constants: add
> > x86 architecture support").
> > 
> > Signed-off-by: Charlie Jenkins <char...@rivosinc.com>
> > Reviewed-by: Alexandre Ghiti <alexgh...@rivosinc.com>
> ...
> > diff --git a/arch/riscv/include/asm/runtime-const.h 
> > b/arch/riscv/include/asm/runtime-const.h
> ...
> > +#define RISCV_RUNTIME_CONST_64_ZBA                         \
> > +   ".option push\n\t"                                      \
> > +   ".option arch,+zba\n\t"                                 \
> > +   "slli   %[__tmp],%[__tmp],32\n\t"                       \
> > +   "add.uw %[__ret],%[__ret],%[__tmp]\n\t"                 \
> > +   "nop\n\t"                                               \
> > +   "nop\n\t"                                               \
> > +   ".option pop\n\t"                                       \
> ...
> > +#if defined(CONFIG_RISCV_ISA_ZBA) && defined(CONFIG_RISCV_ISA_ZBKB)
> ...
> > +#elif defined(CONFIG_RISCV_ISA_ZBA)
> > +#define runtime_const_ptr(sym)                                             
> > \
> > +({                                                                 \
> > +   typeof(sym) __ret, __tmp;                                       \
> > +   asm_inline(RISCV_RUNTIME_CONST_64_PREAMBLE                      \
> > +           ALTERNATIVE(                                            \
> > +                   RISCV_RUNTIME_CONST_64_BASE,                    \
> > +                   RISCV_RUNTIME_CONST_64_ZBA,                     \
> > +                   0, RISCV_ISA_EXT_ZBA, 1                         \
> > +           )                                                       \
> > +           RISCV_RUNTIME_CONST_64_POSTAMBLE(sym)                   \
> > +           : [__ret] "=r" (__ret), [__tmp] "=r" (__tmp));          \
> > +   __ret;                                                          \
> > +})
> 
> This breaks the build for clang versions 16 and earlier because they do
> not support '.option arch' and it is used in CONFIG_RISCV_ISA_ZBA, which
> has no dependencies and it is default on.
> 
>   $ make -skj"$(nproc)" ARCH=riscv LLVM=1 mrproper defconfig fs/dcache.o
>   fs/dcache.c:117:9: warning: unknown option, expected 'push', 'pop', 'rvc', 
> 'norvc', 'relax' or 'norelax' [-Winline-asm]
>           return runtime_const_ptr(dentry_hashtable) +
>                  ^
>   arch/riscv/include/asm/runtime-const.h:103:4: note: expanded from macro 
> 'runtime_const_ptr'
>                           RISCV_RUNTIME_CONST_64_ZBA,                     \
>                           ^
>   arch/riscv/include/asm/runtime-const.h:57:17: note: expanded from macro 
> 'RISCV_RUNTIME_CONST_64_ZBA'
>           ".option push\n\t"                                      \
>                          ^
>   <inline asm>:32:10: note: instantiated into assembly here
>           .option arch,+zba
>                   ^
>   fs/dcache.c:117:9: error: instruction requires the following: 'Zba' 
> (Address Generation Instructions)
>           return runtime_const_ptr(dentry_hashtable) +
>                  ^
>   arch/riscv/include/asm/runtime-const.h:103:4: note: expanded from macro 
> 'runtime_const_ptr'
>                           RISCV_RUNTIME_CONST_64_ZBA,                     \
>                           ^
>   arch/riscv/include/asm/runtime-const.h:59:30: note: expanded from macro 
> 'RISCV_RUNTIME_CONST_64_ZBA'
>           "slli   %[__tmp],%[__tmp],32\n\t"                       \
>                                         ^
>   <inline asm>:34:2: note: instantiated into assembly here
>           add.uw a2,a2,a3
>           ^
>   ...
> 
>   $ rg 'OPTION_ARCH|ZBA' .config
>   364:CONFIG_RISCV_ISA_ZBA=y
> 
> Should it grow a dependency on AS_HAS_OPTION_ARCH or should there be a
> different fix?

This should have been fixed by Alex's patch [1]. Zba is in an awkward
state because BPF generates Zba code without the need for toolchain
support.

[1] https://lore.kernel.org/all/20250328115422.253670-1-alexgh...@rivosinc.com/

> 
> Cheers,
> Nathan

Reply via email to