Re: [PATCH 2/3] soc: samsung: exynos-pmu: add support for PMU_ALIVE non atomic registers

[William McVicker <willmcvicker@google.com>]

On 06/26/2024, Peter Griffin wrote:
> Not all registers in PMU_ALIVE block support atomic set/clear operations.
> GS101_SYSIP_DAT0 and GS101_SYSTEM_CONFIGURATION registers are two regs
> where attempting atomic access fails.
> 
> As documentation on exactly which registers support atomic operations is
> not forthcoming. We default to atomic access, unless the register is
> explicitly added to the tensor_is_atomic() function. Update the comment
> to reflect this as well.
> 
> Signed-off-by: Peter Griffin <peter.griffin@linaro.org>

Reviewed-by: Will McVicker <willmcvicker@google.com>
Tested-by: Will McVicker <willmcvicker@google.com>

I verified reboot and power off on my Pixel 6 Pro.

> ---
>  drivers/soc/samsung/exynos-pmu.c            | 16 ++++++++++++++--
>  include/linux/soc/samsung/exynos-regs-pmu.h |  4 ++++
>  2 files changed, 18 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/soc/samsung/exynos-pmu.c b/drivers/soc/samsung/exynos-pmu.c
> index 624324f4001c..5556acc7c092 100644
> --- a/drivers/soc/samsung/exynos-pmu.c
> +++ b/drivers/soc/samsung/exynos-pmu.c
> @@ -129,14 +129,26 @@ static int tensor_set_bits_atomic(void *ctx, unsigned int offset, u32 val,
>  	return ret;
>  }
>  
> +static bool tensor_is_atomic(unsigned int reg)
> +{
> +	switch (reg) {
> +	case GS101_SYSIP_DAT0:
> +	case GS101_SYSTEM_CONFIGURATION:
> +		return false;
> +	default:
> +		return true;
> +	}
> +}
> +
>  static int tensor_sec_update_bits(void *ctx, unsigned int reg,
>  				  unsigned int mask, unsigned int val)
>  {
>  	/*
>  	 * Use atomic operations for PMU_ALIVE registers (offset 0~0x3FFF)
> -	 * as the target registers can be accessed by multiple masters.
> +	 * as the target registers can be accessed by multiple masters. Some
> +	 * SFRs don't support this however as reported by tensor_is_atomic()
>  	 */
> -	if (reg > PMUALIVE_MASK)
> +	if (reg > PMUALIVE_MASK || !tensor_is_atomic(reg))

nit: Should we just move all the logic, e.g. `reg > PMUALIVE_MASK` into `tensor_is_atomic()`?

[...]

Thanks,
Will