[v2,2/2] ARM: tegra: Use WFE for power-gating on Tegra30

Turned out that WFI doesn't work reliably on Tegra30 as a trigger for
the power-gating, it causes CPU hang under some circumstances like having
memory controller running of PLLP. The TRM doc states that WFI should be
used for the Big-Little "Cluster Switch", while WFE for the power-gating.
Hence let's use the WFE for CPU0 power-gating, like it is done for the
power-gating of a secondary cores. This fixes CPU hang after entering LP2
with memory running off PLLP.

Acked-by: Peter De Schrijver <pdeschrijver@nvidia.com>
Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
---
 arch/arm/mach-tegra/sleep-tegra30.S |  4 +++-
 drivers/soc/tegra/flowctrl.c        | 19 +++++++++++++++++--
 2 files changed, 20 insertions(+), 3 deletions(-)

The Ouya (tegra30) hard locks when the emc clock drops below 400mhz.
Discovered while testing the devfreq and emc drivers.
This patch resolves that issue.

Tested-by: Peter Geis <pgwipeout@gmail.com>

On Tue, Jul 30, 2019 at 2:10 PM Dmitry Osipenko <digetx@gmail.com> wrote:
>
> Turned out that WFI doesn't work reliably on Tegra30 as a trigger for
> the power-gating, it causes CPU hang under some circumstances like having
> memory controller running of PLLP. The TRM doc states that WFI should be
> used for the Big-Little "Cluster Switch", while WFE for the power-gating.
> Hence let's use the WFE for CPU0 power-gating, like it is done for the
> power-gating of a secondary cores. This fixes CPU hang after entering LP2
> with memory running off PLLP.
>
> Acked-by: Peter De Schrijver <pdeschrijver@nvidia.com>
> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
> ---
>  arch/arm/mach-tegra/sleep-tegra30.S |  4 +++-
>  drivers/soc/tegra/flowctrl.c        | 19 +++++++++++++++++--
>  2 files changed, 20 insertions(+), 3 deletions(-)
>
> diff --git a/arch/arm/mach-tegra/sleep-tegra30.S b/arch/arm/mach-tegra/sleep-tegra30.S
> index 6c28395d8c75..17f7a2a6a494 100644
> --- a/arch/arm/mach-tegra/sleep-tegra30.S
> +++ b/arch/arm/mach-tegra/sleep-tegra30.S
> @@ -683,10 +683,12 @@ tegra30_enter_sleep:
>         dsb
>         ldr     r0, [r6, r2] /* memory barrier */
>
> +       cmp     r10, #TEGRA30
>  halted:
>         isb
>         dsb
> -       wfi     /* CPU should be power gated here */
> +       wfine   /* CPU should be power gated here */
> +       wfeeq
>
>         /* !!!FIXME!!! Implement halt failure handler */
>         b       halted
> diff --git a/drivers/soc/tegra/flowctrl.c b/drivers/soc/tegra/flowctrl.c
> index b6bdeef33db1..eb96a3086d6d 100644
> --- a/drivers/soc/tegra/flowctrl.c
> +++ b/drivers/soc/tegra/flowctrl.c
> @@ -91,8 +91,23 @@ void flowctrl_cpu_suspend_enter(unsigned int cpuid)
>                 reg &= ~TEGRA30_FLOW_CTRL_CSR_WFE_BITMAP;
>                 /* clear wfi bitmap */
>                 reg &= ~TEGRA30_FLOW_CTRL_CSR_WFI_BITMAP;
> -               /* pwr gating on wfi */
> -               reg |= TEGRA30_FLOW_CTRL_CSR_WFI_CPU0 << cpuid;
> +
> +               if (tegra_get_chip_id() == TEGRA30) {
> +                       /*
> +                        * The wfi doesn't work well on Tegra30 because
> +                        * CPU hangs under some odd circumstances after
> +                        * power-gating (like memory running off PLLP),
> +                        * hence use wfe that is working perfectly fine.
> +                        * Note that Tegra30 TRM doc clearly stands that
> +                        * wfi should be used for the "Cluster Switching",
> +                        * while wfe for the power-gating, just like it
> +                        * is done on Tegra20.
> +                        */
> +                       reg |= TEGRA20_FLOW_CTRL_CSR_WFE_CPU0 << cpuid;
> +               } else {
> +                       /* pwr gating on wfi */
> +                       reg |= TEGRA30_FLOW_CTRL_CSR_WFI_CPU0 << cpuid;
> +               }
>                 break;
>         }
>         reg |= FLOW_CTRL_CSR_INTR_FLAG;                 /* clear intr flag */
> --
> 2.22.0
>