[PATCH 1/3] drm/amd/amdgpu: Add wave reader to debugfs

[Nicolai Hähnle]

On 11.10.2016 21:18, Tom St Denis wrote:
> Currently supports CZ/VI.  Allows nearly atomic read
> of wave data from GPU.
>
> Signed-off-by: Tom St Denis <tom.stdenis at amd.com>
> ---
>  drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 74 ++++++++++++++++++++++++++++++
>  1 file changed, 74 insertions(+)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> index 89b353418195..b1ab6358fa0f 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> @@ -2914,6 +2914,72 @@ static ssize_t amdgpu_debugfs_sensor_read(struct file *f, char __user *buf,
>  	return !r ? 4 : r;
>  }
>
> +static uint32_t wave_read_ind(struct amdgpu_device *adev, uint32_t SQ_INDEX, uint32_t SQ_DATA, uint32_t simd, uint32_t wave, uint32_t address)
> +{
> +	WREG32(SQ_INDEX, (wave & 0xF) | ((simd & 0x3) << 4) | (address << 16) | (1 << 13));
> +	return RREG32(SQ_DATA);
> +}
> +
> +static ssize_t amdgpu_debugfs_wave_read(struct file *f, char __user *buf,
> +					size_t size, loff_t *pos)
> +{
> +	struct amdgpu_device *adev = f->f_inode->i_private;
> +	int r, x;
> +	ssize_t result=0;
> +	uint32_t offset, se, sh, cu, wave, simd, data[16];
> +	
> +	if (size & 3 || *pos & 3)
> +		return -EINVAL;
> +
> +	/* decode offset */
> +	offset = (*pos & 0x7F);
> +	se = ((*pos >> 7) & 0xFF);
> +	sh = ((*pos >> 15) & 0xFF);
> +	cu = ((*pos >> 23) & 0xFF);
> +	wave = ((*pos >> 31) & 0xFF);
> +	simd = ((*pos >> 37) & 0xFF);
> +	*pos &= 0x7F;
> +
> +	/* switch to the specific se/sh/cu */
> +	mutex_lock(&adev->grbm_idx_mutex);
> +	amdgpu_gfx_select_se_sh(adev, se, sh, cu);
> +
> +	x = 0;
> +	if (adev->family == AMDGPU_FAMILY_CZ || adev->family == AMDGPU_FAMILY_VI) {
> +		/* type 0 wave data */
> +		data[x++] = 0;
> +		data[x++] = wave_read_ind(adev, 0x2378, 0x2379, simd, wave, 0x12);
> +		data[x++] = wave_read_ind(adev, 0x2378, 0x2379, simd, wave, 0x18);
> +		data[x++] = wave_read_ind(adev, 0x2378, 0x2379, simd, wave, 0x19);
> +		data[x++] = wave_read_ind(adev, 0x2378, 0x2379, simd, wave, 0x27E);
> +		data[x++] = wave_read_ind(adev, 0x2378, 0x2379, simd, wave, 0x27F);
> +		data[x++] = wave_read_ind(adev, 0x2378, 0x2379, simd, wave, 0x14);
> +		data[x++] = wave_read_ind(adev, 0x2378, 0x2379, simd, wave, 0x1A);
> +		data[x++] = wave_read_ind(adev, 0x2378, 0x2379, simd, wave, 0x1B);

I know this is just debug code, but it's still annoying to have all 
these magic constants here, when there are perfectly good ixSQ_WAVE_* 
etc. defines in the asic_reg headers.

Nicolai

> +	} else {
> +		return -EINVAL;
> +	}
> +
> +	amdgpu_gfx_select_se_sh(adev, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF);
> +	mutex_unlock(&adev->grbm_idx_mutex);
> +
> +	while (size && (*pos < x * 4)) {
> +		uint32_t value;
> +
> +		value = data[*pos >> 2];
> +		r = put_user(value, (uint32_t *)buf);
> +		if (r)
> +			return r;
> +
> +		result += 4;
> +		buf += 4;
> +		*pos += 4;
> +		size -= 4;
> +	}
> +
> +	return result;
> +}
> +
>  static const struct file_operations amdgpu_debugfs_regs_fops = {
>  	.owner = THIS_MODULE,
>  	.read = amdgpu_debugfs_regs_read,
> @@ -2951,6 +3017,12 @@ static const struct file_operations amdgpu_debugfs_sensors_fops = {
>  	.llseek = default_llseek
>  };
>
> +static const struct file_operations amdgpu_debugfs_wave_fops = {
> +	.owner = THIS_MODULE,
> +	.read = amdgpu_debugfs_wave_read,
> +	.llseek = default_llseek
> +};
> +
>  static const struct file_operations *debugfs_regs[] = {
>  	&amdgpu_debugfs_regs_fops,
>  	&amdgpu_debugfs_regs_didt_fops,
> @@ -2958,6 +3030,7 @@ static const struct file_operations *debugfs_regs[] = {
>  	&amdgpu_debugfs_regs_smc_fops,
>  	&amdgpu_debugfs_gca_config_fops,
>  	&amdgpu_debugfs_sensors_fops,
> +	&amdgpu_debugfs_wave_fops,
>  };
>
>  static const char *debugfs_regs_names[] = {
> @@ -2967,6 +3040,7 @@ static const char *debugfs_regs_names[] = {
>  	"amdgpu_regs_smc",
>  	"amdgpu_gca_config",
>  	"amdgpu_sensors",
> +	"amdgpu_wave",
>  };
>
>  static int amdgpu_debugfs_regs_init(struct amdgpu_device *adev)
>