Re: [PATCH v1] RISC-V: Only default to spinwait on SBI-0.1 and M-mode

[Palmer Dabbelt]

On Thu, 21 Apr 2022 10:03:55 PDT (-0700), Palmer Dabbelt wrote:
> From: Palmer Dabbelt <palmer@xxxxxxxxxxxx>
>
> The spinwait boot method has been superseeded by the SBI HSM extension
> for some time now, but it still enabled by default.  This causes some
> issues on large hart count systems, which will hang if a physical hart
> exists that is larger than NR_CPUS.
>
> Users on modern SBI implemenation don't need spinwait, and while it's
> probably possible to deal with some of the spinwait issues let's just
> restrict the default to systems that are likely to actually use it.
>
> Signed-off-by: Palmer Dabbelt <palmer@xxxxxxxxxxxx>
> ---
> This seems to be the source of many of my new hangs when trying to test
> the NR_CPUS=512 support.  It's not really related, just fallout from
> testing different setups.
> ---
>  arch/riscv/Kconfig | 8 +++++++-
>  1 file changed, 7 insertions(+), 1 deletion(-)
>
> diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig
> index 00fd9c548f26..dd5e975abe37 100644
> --- a/arch/riscv/Kconfig
> +++ b/arch/riscv/Kconfig
> @@ -359,7 +359,7 @@ config RISCV_SBI_V01
>  config RISCV_BOOT_SPINWAIT
>  	bool "Spinwait booting method"
>  	depends on SMP
> -	default y
> +	default y if RISCV_SBI_V01 || RISCV_M_MODE
>  	help
>  	  This enables support for booting Linux via spinwait method. In the
>  	  spinwait method, all cores randomly jump to Linux. One of the cores
> @@ -370,6 +370,12 @@ config RISCV_BOOT_SPINWAIT
>  	  rely on ordered booting via SBI HSM extension which gets chosen
>  	  dynamically at runtime if the firmware supports it.
>
> +	  Since spinwait is incompatible with sparse hart IDs, it requires
> +	  NR_CPUS be large enough to contain the physical hart ID of the first
> +	  hart to enter Linux.
> +
> +	  If unsure what to do here, say N.
> +
>  config KEXEC
>  	bool "Kexec system call"
>  	select KEXEC_CORE

This is on for-next, with some typos fixed.