On Mon, 2024-10-21 at 19:27 -0300, Gustavo Sousa wrote:
> Some display MMIO transactions for offsets in the range that requires
> the DMC wakelock happen in atomic context (this has been confirmed
> during tests on PTL). That means that we need to use a non-sleeping
> variant of MMIO waiting function.
>
> Implement __intel_de_wait_for_register_atomic_nowl() and use it when
> waiting for acknowledgment of acquire/release.
>
> Signed-off-by: Gustavo Sousa <gustavo.sousa@xxxxxxxxx>
> ---
> drivers/gpu/drm/i915/display/intel_de.h | 11 +++++++++++
> drivers/gpu/drm/i915/display/intel_dmc_wl.c | 20 ++++++++++++--------
> 2 files changed, 23 insertions(+), 8 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/display/intel_de.h b/drivers/gpu/drm/i915/display/intel_de.h
> index e017cd4a8168..4116783a62dd 100644
> --- a/drivers/gpu/drm/i915/display/intel_de.h
> +++ b/drivers/gpu/drm/i915/display/intel_de.h
> @@ -121,6 +121,17 @@ ____intel_de_wait_for_register_nowl(struct intel_display *display,
> }
> #define __intel_de_wait_for_register_nowl(p,...) ____intel_de_wait_for_register_nowl(__to_intel_display(p), __VA_ARGS__)
>
> +static inline int
> +____intel_de_wait_for_register_atomic_nowl(struct intel_display *display,
> + i915_reg_t reg,
> + u32 mask, u32 value,
> + unsigned int fast_timeout_us)
> +{
> + return __intel_wait_for_register(__to_uncore(display), reg, mask,
> + value, fast_timeout_us, 0, NULL);
> +}
> +#define __intel_de_wait_for_register_atomic_nowl(p,...) ____intel_de_wait_for_register_atomic_nowl(__to_intel_display(p), __VA_ARGS__)
> +
> static inline int
> __intel_de_wait(struct intel_display *display, i915_reg_t reg,
> u32 mask, u32 value, unsigned int timeout)
> diff --git a/drivers/gpu/drm/i915/display/intel_dmc_wl.c b/drivers/gpu/drm/i915/display/intel_dmc_wl.c
> index 5634ff07269d..8056a3c8666c 100644
> --- a/drivers/gpu/drm/i915/display/intel_dmc_wl.c
> +++ b/drivers/gpu/drm/i915/display/intel_dmc_wl.c
> @@ -39,7 +39,7 @@
> * potential future use.
> */
>
> -#define DMC_WAKELOCK_CTL_TIMEOUT 5
> +#define DMC_WAKELOCK_CTL_TIMEOUT_US 5000
Maybe this deserves a small explanation in the commit message saying
that you need to use the microsecond variant (fast_timeout_us) because
that makes it atomic further down the call chain?
--
Cheers,
Luca.
