Re: [PATCH v1] RISC-V: separate Zbb optimisations requiring and not requiring toolchain support

From: Andrew Jones
Date: Thu May 16 2024 - 03:59:56 EST

Next message: Francesco Dolcini: "Re: [PATCH] arm64: dts: imx8mp: Enable HDMI on i.MX8MP DHCOM PDK2 and PDK3"
Previous message: Hyunjun Ko: "Re: [PATCH v2 00/34] Qualcomm video encoder and decoder driver"
In reply to: Conor Dooley: "Re: [PATCH v1] RISC-V: separate Zbb optimisations requiring and not requiring toolchain support"
Next in thread: Conor Dooley: "Re: [PATCH v1] RISC-V: separate Zbb optimisations requiring and not requiring toolchain support"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

On Wed, May 15, 2024 at 04:27:40PM GMT, Conor Dooley wrote:
> From: Conor Dooley <conor.dooley@xxxxxxxxxxxxx>
>
> It seems a bit ridiculous to require toolchain support for BPF to
> assemble Zbb instructions, so introduce hidden a Kconfig option that

s/hidden a/a hidden/

> controls the use of any toolchain-requiring optimisations while support
> is available.
>
> Zbb support has always depended on alternatives, so while adjusting the
> config options guarding optimisations, remove any checks for
> whether or not alternatives are enabled.
>
> Signed-off-by: Conor Dooley <conor.dooley@xxxxxxxxxxxxx>
> ---
>
> This patch stems out of a conversation about Zba optimisations in BPF.
> I'm not super sold on the approach in all honesty, even though we
> recently had a conversation about respecting the Kconfig options - but
> at this point I'd be convinced to just add some wording to the Kconfig
> options mentioning that BPF optimisations are excluded.
> Having hidden options that mean someone can turn what on what they
> think are Zbb optimisations but not actually get any cos their toolchain
> doesn't support it seems crap to me. I don't wanna add another
> user-visible option for that situation cos I wanna try to minimise the
> extent of our extension-related Kconfig options, not blow them up like
> Augustus Gloop!
>
> Cheers,
> Conor.
>
> CC: xiao.w.wang@xxxxxxxxx
> CC: Andrew Jones <ajones@xxxxxxxxxxxxxxxx>
> CC: pulehui@xxxxxxxxxx
> CC: Charlie Jenkins <charlie@xxxxxxxxxxxx>
> CC: Paul Walmsley <paul.walmsley@xxxxxxxxxx>
> CC: Palmer Dabbelt <palmer@xxxxxxxxxxx>
> CC: Conor Dooley <conor.dooley@xxxxxxxxxxxxx>
> CC: linux-riscv@xxxxxxxxxxxxxxxxxxx
> CC: linux-kernel@xxxxxxxxxxxxxxx
> ---
> arch/riscv/Kconfig | 15 ++++++++++++---
> arch/riscv/include/asm/arch_hweight.h | 4 ++--
> arch/riscv/include/asm/bitops.h | 4 ++--
> arch/riscv/include/asm/checksum.h | 3 +--
> arch/riscv/lib/csum.c | 9 +++------
> arch/riscv/lib/strcmp.S | 4 ++--
> arch/riscv/lib/strlen.S | 4 ++--
> arch/riscv/lib/strncmp.S | 4 ++--
> 8 files changed, 26 insertions(+), 21 deletions(-)
>
> diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig
> index e927b52b420c..f216a52ed181 100644
> --- a/arch/riscv/Kconfig
> +++ b/arch/riscv/Kconfig
> @@ -605,14 +605,23 @@ config TOOLCHAIN_HAS_VECTOR_CRYPTO
> def_bool $(as-instr, .option arch$(comma) +v$(comma) +zvkb)
> depends on AS_HAS_OPTION_ARCH
>
> -config RISCV_ISA_ZBB
> - bool "Zbb extension support for bit manipulation instructions"
> +config RISCV_ISA_ZBB_ALT
> + def_bool RISCV_ISA_ZBB
> depends on TOOLCHAIN_HAS_ZBB
> depends on RISCV_ALTERNATIVE
> + help
> + This option controls whether or not we build optimisations that
> + depend on toolchain support. It's automatically enabled whenever the
> + toolchain in use supports assembling Zbb instructions and
> + RISCV_ISA_ZBB is set.
> +
> +config RISCV_ISA_ZBB
> + bool "Zbb extension support for bit manipulation instructions"
> default y
> help
> Add support for enabling optimisations in the kernel when the
> - Zbb extension is detected at boot.
> + Zbb extension is detected at boot. Some optimisations may
> + additionally depend on toolchain support for Zbb.
>
> The Zbb extension provides instructions to accelerate a number
> of bit-specific operations (count bit population, sign extending,
> diff --git a/arch/riscv/include/asm/arch_hweight.h b/arch/riscv/include/asm/arch_hweight.h
> index 85b2c443823e..a677f6b82228 100644
> --- a/arch/riscv/include/asm/arch_hweight.h
> +++ b/arch/riscv/include/asm/arch_hweight.h
> @@ -19,7 +19,7 @@
>
> static __always_inline unsigned int __arch_hweight32(unsigned int w)
> {
> -#ifdef CONFIG_RISCV_ISA_ZBB
> +#ifdef CONFIG_RISCV_ISA_ZBB_ALT

So the new hidden config is a shorthand for

#if defined(CONFIG_RISCV_ISA_ZBB) && \
defined(CONFIG_TOOLCHAIN_HAS_ZBB) && \
defined(CONFIG_RISCV_ALTERNATIVE)

which is reasonable to add, since that's a mouthful, but I'm not sure the
name, RISCV_ISA_ZBB_ALT, does a good job conveying all that.

If we instead just dropped the 'depends on TOOLCHAIN_HAS_ZBB' from
config RISCV_ISA_ZBB (keeping the 'depends on RISCV_ALTERNATIVE',
since nobody is really complaining about that), then we could change
this to

#if defined(CONFIG_RISCV_ISA_ZBB) && defined(CONFIG_TOOLCHAIN_HAS_ZBB)

which is verbose, but easy to understand.

> asm goto(ALTERNATIVE("j %l[legacy]", "nop", 0,
> RISCV_ISA_EXT_ZBB, 1)
> : : : : legacy);
> @@ -50,7 +50,7 @@ static inline unsigned int __arch_hweight8(unsigned int w)
> #if BITS_PER_LONG == 64
> static __always_inline unsigned long __arch_hweight64(__u64 w)
> {
> -# ifdef CONFIG_RISCV_ISA_ZBB
> +# ifdef CONFIG_RISCV_ISA_ZBB_ALT

nit: While we're here we could remove that space after the # (and the
corresponding one on the # endif)

> asm goto(ALTERNATIVE("j %l[legacy]", "nop", 0,
> RISCV_ISA_EXT_ZBB, 1)
> : : : : legacy);
> diff --git a/arch/riscv/include/asm/bitops.h b/arch/riscv/include/asm/bitops.h
> index 880606b0469a..3ed810a6123d 100644
> --- a/arch/riscv/include/asm/bitops.h
> +++ b/arch/riscv/include/asm/bitops.h
> @@ -15,7 +15,7 @@
> #include <asm/barrier.h>
> #include <asm/bitsperlong.h>
>
> -#if !defined(CONFIG_RISCV_ISA_ZBB) || defined(NO_ALTERNATIVE)
> +#if !defined(CONFIG_RISCV_ISA_ZBB_ALT) || defined(NO_ALTERNATIVE)

nit: It's sufficient to check !defined(CONFIG_RISCV_ISA_ZBB), so no need
for this change or its #endif comment change below.

> #include <asm-generic/bitops/__ffs.h>
> #include <asm-generic/bitops/__fls.h>
> #include <asm-generic/bitops/ffs.h>
> @@ -175,7 +175,7 @@ static __always_inline int variable_fls(unsigned int x)
> variable_fls(x_); \
> })
>
> -#endif /* !defined(CONFIG_RISCV_ISA_ZBB) || defined(NO_ALTERNATIVE) */
> +#endif /* !defined(CONFIG_RISCV_ISA_ZBB_ALT) || defined(NO_ALTERNATIVE) */
>
> #include <asm-generic/bitops/ffz.h>
> #include <asm-generic/bitops/fls64.h>
> diff --git a/arch/riscv/include/asm/checksum.h b/arch/riscv/include/asm/checksum.h
> index 88e6f1499e88..956224ea8199 100644
> --- a/arch/riscv/include/asm/checksum.h
> +++ b/arch/riscv/include/asm/checksum.h
> @@ -49,8 +49,7 @@ static inline __sum16 ip_fast_csum(const void *iph, unsigned int ihl)
> * ZBB only saves three instructions on 32-bit and five on 64-bit so not
> * worth checking if supported without Alternatives.
> */
> - if (IS_ENABLED(CONFIG_RISCV_ISA_ZBB) &&
> - IS_ENABLED(CONFIG_RISCV_ALTERNATIVE)) {
> + if (IS_ENABLED(CONFIG_RISCV_ISA_ZBB_ALT)) {

When we can use C like this, rather than #if, we can create a helper to
reduce the verbosity of what I proposed above, e.g.

static inline bool csum_zbb(void)
{
return IS_ENABLED(CONFIG_RISCV_ISA_ZBB) &&
IS_ENABLED(CONFIG_TOOLCHAIN_HAS_ZBB);
}

> unsigned long fold_temp;
>
> asm goto(ALTERNATIVE("j %l[no_zbb]", "nop", 0,
> diff --git a/arch/riscv/lib/csum.c b/arch/riscv/lib/csum.c
> index 7fb12c59e571..7a97394c252b 100644
> --- a/arch/riscv/lib/csum.c
> +++ b/arch/riscv/lib/csum.c
> @@ -44,8 +44,7 @@ __sum16 csum_ipv6_magic(const struct in6_addr *saddr,
> * Zbb support saves 4 instructions, so not worth checking without
> * alternatives if supported
> */
> - if (IS_ENABLED(CONFIG_RISCV_ISA_ZBB) &&
> - IS_ENABLED(CONFIG_RISCV_ALTERNATIVE)) {
> + if (IS_ENABLED(CONFIG_RISCV_ISA_ZBB_ALT)) {
> unsigned long fold_temp;
>
> /*
> @@ -161,8 +160,7 @@ do_csum_with_alignment(const unsigned char *buff, int len)
> * Zbb support saves 6 instructions, so not worth checking without
> * alternatives if supported
> */
> - if (IS_ENABLED(CONFIG_RISCV_ISA_ZBB) &&
> - IS_ENABLED(CONFIG_RISCV_ALTERNATIVE)) {
> + if (IS_ENABLED(CONFIG_RISCV_ISA_ZBB_ALT)) {
> unsigned long fold_temp;
>
> /*
> @@ -248,8 +246,7 @@ do_csum_no_alignment(const unsigned char *buff, int len)
> * Zbb support saves 6 instructions, so not worth checking without
> * alternatives if supported
> */
> - if (IS_ENABLED(CONFIG_RISCV_ISA_ZBB) &&
> - IS_ENABLED(CONFIG_RISCV_ALTERNATIVE)) {
> + if (IS_ENABLED(CONFIG_RISCV_ISA_ZBB_ALT)) {
> unsigned long fold_temp;
>
> /*
> diff --git a/arch/riscv/lib/strcmp.S b/arch/riscv/lib/strcmp.S
> index 687b2bea5c43..a4dd2ac306f1 100644
> --- a/arch/riscv/lib/strcmp.S
> +++ b/arch/riscv/lib/strcmp.S
> @@ -8,7 +8,7 @@
> /* int strcmp(const char *cs, const char *ct) */
> SYM_FUNC_START(strcmp)
>
> - ALTERNATIVE("nop", "j strcmp_zbb", 0, RISCV_ISA_EXT_ZBB, CONFIG_RISCV_ISA_ZBB)
> + ALTERNATIVE("nop", "j strcmp_zbb", 0, RISCV_ISA_EXT_ZBB, CONFIG_RISCV_ISA_ZBB_ALT)

Here we have a problem with the compound config expression, but we could
rework these functions to be like this

#if defined(CONFIG_RISCV_ISA_ZBB) && defined(CONFIG_TOOLCHAIN_HAS_ZBB)
ALTERNATIVE("j strcmp_nozbb", "nop", 0, RISCV_ISA_EXT_ZBB, 1)

...zbb impl...
ret
#endif
strcmp_nozbb:
...no zbb impl...

>
> /*
> * Returns
> @@ -43,7 +43,7 @@ SYM_FUNC_START(strcmp)
> * The code was published as part of the bitmanip manual
> * in Appendix A.
> */
> -#ifdef CONFIG_RISCV_ISA_ZBB
> +#ifdef CONFIG_RISCV_ISA_ZBB_ALT
> strcmp_zbb:
>
> .option push
> diff --git a/arch/riscv/lib/strlen.S b/arch/riscv/lib/strlen.S
> index 8ae3064e45ff..3ab1310a7b83 100644
> --- a/arch/riscv/lib/strlen.S
> +++ b/arch/riscv/lib/strlen.S
> @@ -8,7 +8,7 @@
> /* int strlen(const char *s) */
> SYM_FUNC_START(strlen)
>
> - ALTERNATIVE("nop", "j strlen_zbb", 0, RISCV_ISA_EXT_ZBB, CONFIG_RISCV_ISA_ZBB)
> + ALTERNATIVE("nop", "j strlen_zbb", 0, RISCV_ISA_EXT_ZBB, CONFIG_RISCV_ISA_ZBB_ALT)
>
> /*
> * Returns
> @@ -33,7 +33,7 @@ SYM_FUNC_START(strlen)
> /*
> * Variant of strlen using the ZBB extension if available
> */
> -#ifdef CONFIG_RISCV_ISA_ZBB
> +#ifdef CONFIG_RISCV_ISA_ZBB_ALT
> strlen_zbb:
>
> #ifdef CONFIG_CPU_BIG_ENDIAN
> diff --git a/arch/riscv/lib/strncmp.S b/arch/riscv/lib/strncmp.S
> index aba5b3148621..aeed830804d7 100644
> --- a/arch/riscv/lib/strncmp.S
> +++ b/arch/riscv/lib/strncmp.S
> @@ -8,7 +8,7 @@
> /* int strncmp(const char *cs, const char *ct, size_t count) */
> SYM_FUNC_START(strncmp)
>
> - ALTERNATIVE("nop", "j strncmp_zbb", 0, RISCV_ISA_EXT_ZBB, CONFIG_RISCV_ISA_ZBB)
> + ALTERNATIVE("nop", "j strncmp_zbb", 0, RISCV_ISA_EXT_ZBB, CONFIG_RISCV_ISA_ZBB_ALT)
>
> /*
> * Returns
> @@ -46,7 +46,7 @@ SYM_FUNC_START(strncmp)
> /*
> * Variant of strncmp using the ZBB extension if available
> */
> -#ifdef CONFIG_RISCV_ISA_ZBB
> +#ifdef CONFIG_RISCV_ISA_ZBB_ALT
> strncmp_zbb:
>
> .option push
> --
> 2.43.0
>

Thanks,
drew

Next message: Francesco Dolcini: "Re: [PATCH] arm64: dts: imx8mp: Enable HDMI on i.MX8MP DHCOM PDK2 and PDK3"
Previous message: Hyunjun Ko: "Re: [PATCH v2 00/34] Qualcomm video encoder and decoder driver"
In reply to: Conor Dooley: "Re: [PATCH v1] RISC-V: separate Zbb optimisations requiring and not requiring toolchain support"
Next in thread: Conor Dooley: "Re: [PATCH v1] RISC-V: separate Zbb optimisations requiring and not requiring toolchain support"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]