Re: [PATCH] drm/panthor: Add 64-bit and poll register accessors

[Boris Brezillon <boris.brezillon@collabora.com>]

On Thu, 10 Apr 2025 17:35:46 +0100
Karunika Choo <karunika.choo@arm.com> wrote:

> This patch adds 64-bit register accessors to simplify register access in
> Panthor. It also adds 32-bit and 64-bit variants for read_poll_timeout.
> 
> This patch also updates Panthor to use the new 64-bit accessors and poll
> functions.
> 
> Signed-off-by: Karunika Choo <karunika.choo@arm.com>
> ---
>  drivers/gpu/drm/panthor/panthor_device.h |  71 ++++++++++++
>  drivers/gpu/drm/panthor/panthor_fw.c     |   9 +-
>  drivers/gpu/drm/panthor/panthor_gpu.c    | 142 ++++++-----------------
>  drivers/gpu/drm/panthor/panthor_mmu.c    |  34 ++----
>  drivers/gpu/drm/panthor/panthor_regs.h   |   6 -
>  5 files changed, 124 insertions(+), 138 deletions(-)
> 
> diff --git a/drivers/gpu/drm/panthor/panthor_device.h b/drivers/gpu/drm/panthor/panthor_device.h
> index da6574021664..5028e25f5e0d 100644
> --- a/drivers/gpu/drm/panthor/panthor_device.h
> +++ b/drivers/gpu/drm/panthor/panthor_device.h
> @@ -428,4 +428,75 @@ static int panthor_request_ ## __name ## _irq(struct panthor_device *ptdev,			\
>  
>  extern struct workqueue_struct *panthor_cleanup_wq;
>  
> +static inline void gpu_write(struct panthor_device *ptdev, u32 reg, u32 data)
> +{
> +	writel(data, ptdev->iomem + reg);
> +}
> +
> +static inline u32 gpu_read(struct panthor_device *ptdev, u32 reg)
> +{
> +	return readl(ptdev->iomem + reg);
> +}
> +
> +static inline u32 gpu_read_relaxed(struct panthor_device *ptdev, u32 reg)
> +{
> +	return readl_relaxed(ptdev->iomem + reg);
> +}
> +
> +static inline void gpu_write64(struct panthor_device *ptdev, u32 reg, u64 data)
> +{
> +	gpu_write(ptdev, reg, lower_32_bits(data));
> +	gpu_write(ptdev, reg + 4, upper_32_bits(data));
> +}
> +
> +static inline u64 gpu_read64(struct panthor_device *ptdev, u32 reg)
> +{
> +	return (gpu_read(ptdev, reg) | ((u64)gpu_read(ptdev, reg + 4) << 32));
> +}
> +
> +static inline u64 gpu_read64_relaxed(struct panthor_device *ptdev, u32 reg)
> +{
> +	return (gpu_read_relaxed(ptdev, reg) |
> +		((u64)gpu_read_relaxed(ptdev, reg + 4) << 32));
> +}
> +
> +static inline u64 gpu_read64_counter(struct panthor_device *ptdev, u32 reg)
> +{
> +	u32 lo, hi1, hi2;
> +	do {
> +		hi1 = gpu_read(ptdev, reg + 4);
> +		lo = gpu_read(ptdev, reg);
> +		hi2 = gpu_read(ptdev, reg + 4);
> +	} while (hi1 != hi2);
> +	return lo | ((u64)hi2 << 32);
> +}
> +
> +#define gpu_read_poll_timeout(dev, reg, val, cond, delay_us, timeout_us)    \
> +	read_poll_timeout(gpu_read, val, cond, delay_us, timeout_us, false, \
> +			  dev, reg)

nit: can use use tabs to pad till the '\' at the end of the line so we
can have a consistent formatting across these definitions?

> +
> +#define gpu_read_poll_timeout_atomic(dev, reg, val, cond, delay_us,         \
> +				     timeout_us)                            \
> +	read_poll_timeout_atomic(gpu_read, val, cond, delay_us, timeout_us, \
> +				 false, dev, reg)
> +
> +#define gpu_read64_poll_timeout(dev, reg, val, cond, delay_us, timeout_us)    \
> +	read_poll_timeout(gpu_read64, val, cond, delay_us, timeout_us, false, \
> +			  dev, reg)
> +
> +#define gpu_read64_poll_timeout_atomic(dev, reg, val, cond, delay_us,         \
> +				       timeout_us)                            \
> +	read_poll_timeout_atomic(gpu_read64, val, cond, delay_us, timeout_us, \
> +				 false, dev, reg)
> +
> +#define gpu_read_relaxed_poll_timeout_atomic(dev, reg, val, cond, delay_us, \
> +					     timeout_us)                    \
> +	read_poll_timeout_atomic(gpu_read_relaxed, val, cond, delay_us,     \
> +				 timeout_us, false, dev, reg)
> +
> +#define gpu_read64_relaxed_poll_timeout(dev, reg, val, cond, delay_us,         \
> +					timeout_us)                            \
> +	read_poll_timeout(gpu_read64_relaxed, val, cond, delay_us, timeout_us, \
> +			  false, dev, reg)
> +
>  #endif
> diff --git a/drivers/gpu/drm/panthor/panthor_fw.c b/drivers/gpu/drm/panthor/panthor_fw.c
> index 0f52766a3120..ecfbe0456f89 100644
> --- a/drivers/gpu/drm/panthor/panthor_fw.c
> +++ b/drivers/gpu/drm/panthor/panthor_fw.c
> @@ -1059,8 +1059,8 @@ static void panthor_fw_stop(struct panthor_device *ptdev)
>  	u32 status;
>  
>  	gpu_write(ptdev, MCU_CONTROL, MCU_CONTROL_DISABLE);
> -	if (readl_poll_timeout(ptdev->iomem + MCU_STATUS, status,
> -			       status == MCU_STATUS_DISABLED, 10, 100000))
> +	if (gpu_read_poll_timeout(ptdev, MCU_STATUS, status,
> +				  status == MCU_STATUS_DISABLED, 10, 100000))
>  		drm_err(&ptdev->base, "Failed to stop MCU");
>  }
>  
> @@ -1085,8 +1085,9 @@ void panthor_fw_pre_reset(struct panthor_device *ptdev, bool on_hang)
>  
>  		panthor_fw_update_reqs(glb_iface, req, GLB_HALT, GLB_HALT);
>  		gpu_write(ptdev, CSF_DOORBELL(CSF_GLB_DOORBELL_ID), 1);
> -		if (!readl_poll_timeout(ptdev->iomem + MCU_STATUS, status,
> -					status == MCU_STATUS_HALT, 10, 100000)) {
> +		if (!gpu_read_poll_timeout(ptdev, MCU_STATUS, status,
> +					   status == MCU_STATUS_HALT, 10,
> +					   100000)) {
>  			ptdev->reset.fast = true;
>  		} else {
>  			drm_warn(&ptdev->base, "Failed to cleanly suspend MCU");
> diff --git a/drivers/gpu/drm/panthor/panthor_gpu.c b/drivers/gpu/drm/panthor/panthor_gpu.c
> index 671049020afa..fd09f0928019 100644
> --- a/drivers/gpu/drm/panthor/panthor_gpu.c
> +++ b/drivers/gpu/drm/panthor/panthor_gpu.c
> @@ -108,14 +108,9 @@ static void panthor_gpu_init_info(struct panthor_device *ptdev)
>  
>  	ptdev->gpu_info.as_present = gpu_read(ptdev, GPU_AS_PRESENT);
>  
> -	ptdev->gpu_info.shader_present = gpu_read(ptdev, GPU_SHADER_PRESENT_LO);
> -	ptdev->gpu_info.shader_present |= (u64)gpu_read(ptdev, GPU_SHADER_PRESENT_HI) << 32;
> -
> -	ptdev->gpu_info.tiler_present = gpu_read(ptdev, GPU_TILER_PRESENT_LO);
> -	ptdev->gpu_info.tiler_present |= (u64)gpu_read(ptdev, GPU_TILER_PRESENT_HI) << 32;
> -
> -	ptdev->gpu_info.l2_present = gpu_read(ptdev, GPU_L2_PRESENT_LO);
> -	ptdev->gpu_info.l2_present |= (u64)gpu_read(ptdev, GPU_L2_PRESENT_HI) << 32;
> +	ptdev->gpu_info.shader_present = gpu_read64(ptdev, GPU_SHADER_PRESENT_LO);
> +	ptdev->gpu_info.tiler_present = gpu_read64(ptdev, GPU_TILER_PRESENT_LO);
> +	ptdev->gpu_info.l2_present = gpu_read64(ptdev, GPU_L2_PRESENT_LO);

Now that we have proper 64-bit accessors, I think I would drop the
_LO/_HI definitions and just go a single def per register that replaces
the _LO one.