Commit d4b8386c authored by Sonny Jiang's avatar Sonny Jiang Committed by Alex Deucher
Browse files

drm/amdgpu/jpeg5: Add support for DPG mode



Add DPG support for JPEG 5.0

Signed-off-by: default avatarSonny Jiang <sonjiang@amd.com>
Acked-by: default avatarSaleemkhan Jamadar <saleemkhan.jamadar@amd.com>
Reviewed-by: default avatarDavid (Ming Qiang) Wu <David.Wu3@amd.com>
Signed-off-by: default avatarAlex Deucher <alexander.deucher@amd.com>
parent 291af3f5
Loading
Loading
Loading
Loading
+31 −0
Original line number Diff line number Diff line
@@ -60,6 +60,37 @@
		RREG32_SOC15(JPEG, inst_idx, mmUVD_DPG_LMA_DATA);				\
	})

#define WREG32_SOC24_JPEG_DPG_MODE(inst_idx, offset, value, indirect)		\
	do {									\
		WREG32_SOC15(JPEG, GET_INST(JPEG, inst_idx),			\
			     regUVD_DPG_LMA_DATA, value);			\
		WREG32_SOC15(JPEG, GET_INST(JPEG, inst_idx),			\
			     regUVD_DPG_LMA_MASK, 0xFFFFFFFF);			\
		WREG32_SOC15(							\
			JPEG, GET_INST(JPEG, inst_idx),				\
			regUVD_DPG_LMA_CTL,					\
			(UVD_DPG_LMA_CTL__READ_WRITE_MASK |			\
			 offset << UVD_DPG_LMA_CTL__READ_WRITE_ADDR__SHIFT |	\
			 indirect << UVD_DPG_LMA_CTL__SRAM_SEL__SHIFT));	\
	} while (0)

#define RREG32_SOC24_JPEG_DPG_MODE(inst_idx, offset, mask_en)			\
	do {									\
		WREG32_SOC15(JPEG, GET_INST(JPEG, inst_idx),			\
			regUVD_DPG_LMA_MASK, 0xFFFFFFFF);			\
		WREG32_SOC15(JPEG, GET_INST(JPEG, inst_idx),			\
			regUVD_DPG_LMA_CTL,					\
			(UVD_DPG_LMA_CTL__MASK_EN_MASK |			\
			offset << UVD_DPG_LMA_CTL__READ_WRITE_ADDR__SHIFT));	\
		RREG32_SOC15(JPEG, inst_idx, regUVD_DPG_LMA_DATA);		\
	} while (0)

#define ADD_SOC24_JPEG_TO_DPG_SRAM(inst_idx, offset, value, indirect)		\
	do {									\
		*adev->jpeg.inst[inst_idx].dpg_sram_curr_addr++ = offset;	\
		*adev->jpeg.inst[inst_idx].dpg_sram_curr_addr++ = value;	\
	} while (0)

struct amdgpu_jpeg_reg{
	unsigned jpeg_pitch[AMDGPU_MAX_JPEG_RINGS];
};
+142 −17
Original line number Diff line number Diff line
@@ -31,6 +31,7 @@
#include "vcn/vcn_5_0_0_offset.h"
#include "vcn/vcn_5_0_0_sh_mask.h"
#include "ivsrcid/vcn/irqsrcs_vcn_4_0.h"
#include "jpeg_v5_0_0.h"

static void jpeg_v5_0_0_set_dec_ring_funcs(struct amdgpu_device *adev);
static void jpeg_v5_0_0_set_irq_funcs(struct amdgpu_device *adev);
@@ -137,6 +138,10 @@ static int jpeg_v5_0_0_hw_init(void *handle)
	adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell,
			(adev->doorbell_index.vcn.vcn_ring0_1 << 1), 0);

	/* Skip ring test because pause DPG is not implemented. */
	if (adev->pg_flags & AMD_PG_SUPPORT_JPEG_DPG)
		return 0;

	r = amdgpu_ring_test_helper(ring);
	if (r)
		return r;
@@ -235,7 +240,7 @@ static void jpeg_v5_0_0_enable_clock_gating(struct amdgpu_device *adev)
	WREG32_SOC15(JPEG, 0, regJPEG_CGC_GATE, data);
}

static int jpeg_v5_0_0_disable_static_power_gating(struct amdgpu_device *adev)
static int jpeg_v5_0_0_disable_power_gating(struct amdgpu_device *adev)
{
	uint32_t data = 0;

@@ -248,14 +253,10 @@ static int jpeg_v5_0_0_disable_static_power_gating(struct amdgpu_device *adev)
	WREG32_P(SOC15_REG_OFFSET(JPEG, 0, regUVD_JPEG_POWER_STATUS), 0,
		~UVD_JPEG_POWER_STATUS__JPEG_POWER_STATUS_MASK);

	/* keep the JPEG in static PG mode */
	WREG32_P(SOC15_REG_OFFSET(JPEG, 0, regUVD_JPEG_POWER_STATUS), 0,
		~UVD_JPEG_POWER_STATUS__JPEG_PG_MODE_MASK);

	return 0;
}

static int jpeg_v5_0_0_enable_static_power_gating(struct amdgpu_device *adev)
static int jpeg_v5_0_0_enable_power_gating(struct amdgpu_device *adev)
{
	/* enable anti hang mechanism */
	WREG32_P(SOC15_REG_OFFSET(JPEG, 0, regUVD_JPEG_POWER_STATUS),
@@ -273,6 +274,121 @@ static int jpeg_v5_0_0_enable_static_power_gating(struct amdgpu_device *adev)
	return 0;
}

static void jpeg_engine_5_0_0_dpg_clock_gating_mode(struct amdgpu_device *adev,
	       int inst_idx, uint8_t indirect)
{
	uint32_t data = 0;

	// JPEG disable CGC
	if (adev->cg_flags & AMD_CG_SUPPORT_JPEG_MGCG)
		data = 1 << JPEG_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
	else
		data = 0 << JPEG_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;

	data |= 1 << JPEG_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
	data |= 4 << JPEG_CGC_CTRL__CLK_OFF_DELAY__SHIFT;

	if (indirect) {
		ADD_SOC24_JPEG_TO_DPG_SRAM(inst_idx, vcnipJPEG_CGC_CTRL, data, indirect);

		// Turn on All JPEG clocks
		data = 0;
		ADD_SOC24_JPEG_TO_DPG_SRAM(inst_idx, vcnipJPEG_CGC_GATE, data, indirect);
	} else {
		WREG32_SOC24_JPEG_DPG_MODE(inst_idx, vcnipJPEG_CGC_CTRL, data, indirect);

		// Turn on All JPEG clocks
		data = 0;
		WREG32_SOC24_JPEG_DPG_MODE(inst_idx, vcnipJPEG_CGC_GATE, data, indirect);
	}
}

/**
 * jpeg_v5_0_0_start_dpg_mode - Jpeg start with dpg mode
 *
 * @adev: amdgpu_device pointer
 * @inst_idx: instance number index
 * @indirect: indirectly write sram
 *
 * Start JPEG block with dpg mode
 */
static int jpeg_v5_0_0_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
{
	struct amdgpu_ring *ring = adev->jpeg.inst[inst_idx].ring_dec;
	uint32_t reg_data = 0;

	jpeg_v5_0_0_enable_power_gating(adev);

	// enable dynamic power gating mode
	reg_data = RREG32_SOC15(JPEG, inst_idx, regUVD_JPEG_POWER_STATUS);
	reg_data |= UVD_JPEG_POWER_STATUS__JPEG_PG_MODE_MASK;
	WREG32_SOC15(JPEG, inst_idx, regUVD_JPEG_POWER_STATUS, reg_data);

	if (indirect)
		adev->jpeg.inst[inst_idx].dpg_sram_curr_addr =
			(uint32_t *)adev->jpeg.inst[inst_idx].dpg_sram_cpu_addr;

	jpeg_engine_5_0_0_dpg_clock_gating_mode(adev, inst_idx, indirect);

	/* MJPEG global tiling registers */
	if (indirect)
		ADD_SOC24_JPEG_TO_DPG_SRAM(inst_idx, vcnipJPEG_DEC_GFX10_ADDR_CONFIG,
			adev->gfx.config.gb_addr_config, indirect);
	else
		WREG32_SOC24_JPEG_DPG_MODE(inst_idx, vcnipJPEG_DEC_GFX10_ADDR_CONFIG,
			adev->gfx.config.gb_addr_config, 1);

	/* enable System Interrupt for JRBC */
	if (indirect)
		ADD_SOC24_JPEG_TO_DPG_SRAM(inst_idx, vcnipJPEG_SYS_INT_EN,
			JPEG_SYS_INT_EN__DJRBC0_MASK, indirect);
	else
		WREG32_SOC24_JPEG_DPG_MODE(inst_idx, vcnipJPEG_SYS_INT_EN,
			JPEG_SYS_INT_EN__DJRBC0_MASK, 1);

	if (indirect) {
		/* add nop to workaround PSP size check */
		ADD_SOC24_JPEG_TO_DPG_SRAM(inst_idx, vcnipUVD_NO_OP, 0, indirect);

		amdgpu_jpeg_psp_update_sram(adev, inst_idx, 0);
	}

	WREG32_SOC15(VCN, 0, regVCN_JPEG_DB_CTRL,
		ring->doorbell_index << VCN_JPEG_DB_CTRL__OFFSET__SHIFT |
		VCN_JPEG_DB_CTRL__EN_MASK);

	WREG32_SOC15(JPEG, inst_idx, regUVD_LMI_JRBC_RB_VMID, 0);
	WREG32_SOC15(JPEG, inst_idx, regUVD_JRBC_RB_CNTL, (0x00000001L | 0x00000002L));
	WREG32_SOC15(JPEG, inst_idx, regUVD_LMI_JRBC_RB_64BIT_BAR_LOW,
		lower_32_bits(ring->gpu_addr));
	WREG32_SOC15(JPEG, inst_idx, regUVD_LMI_JRBC_RB_64BIT_BAR_HIGH,
		upper_32_bits(ring->gpu_addr));
	WREG32_SOC15(JPEG, inst_idx, regUVD_JRBC_RB_RPTR, 0);
	WREG32_SOC15(JPEG, inst_idx, regUVD_JRBC_RB_WPTR, 0);
	WREG32_SOC15(JPEG, inst_idx, regUVD_JRBC_RB_CNTL, 0x00000002L);
	WREG32_SOC15(JPEG, inst_idx, regUVD_JRBC_RB_SIZE, ring->ring_size / 4);
	ring->wptr = RREG32_SOC15(JPEG, inst_idx, regUVD_JRBC_RB_WPTR);

	return 0;
}

/**
 * jpeg_v5_0_0_stop_dpg_mode - Jpeg stop with dpg mode
 *
 * @adev: amdgpu_device pointer
 * @inst_idx: instance number index
 *
 * Stop JPEG block with dpg mode
 */
static void jpeg_v5_0_0_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx)
{
	uint32_t reg_data = 0;

	reg_data = RREG32_SOC15(JPEG, inst_idx, regUVD_JPEG_POWER_STATUS);
	reg_data &= ~UVD_JPEG_POWER_STATUS__JPEG_PG_MODE_MASK;
	WREG32_SOC15(JPEG, inst_idx, regUVD_JPEG_POWER_STATUS, reg_data);
}

/**
 * jpeg_v5_0_0_start - start JPEG block
 *
@@ -288,8 +404,13 @@ static int jpeg_v5_0_0_start(struct amdgpu_device *adev)
	if (adev->pm.dpm_enabled)
		amdgpu_dpm_enable_jpeg(adev, true);

	if (adev->pg_flags & AMD_PG_SUPPORT_JPEG_DPG) {
		r = jpeg_v5_0_0_start_dpg_mode(adev, 0, adev->jpeg.indirect_sram);
		return r;
	}

	/* disable power gating */
	r = jpeg_v5_0_0_disable_static_power_gating(adev);
	r = jpeg_v5_0_0_disable_power_gating(adev);
	if (r)
		return r;

@@ -300,7 +421,6 @@ static int jpeg_v5_0_0_start(struct amdgpu_device *adev)
	WREG32_SOC15(JPEG, 0, regJPEG_DEC_GFX10_ADDR_CONFIG,
		adev->gfx.config.gb_addr_config);


	/* enable JMI channel */
	WREG32_P(SOC15_REG_OFFSET(JPEG, 0, regUVD_JMI_CNTL), 0,
		~UVD_JMI_CNTL__SOFT_RESET_MASK);
@@ -340,6 +460,10 @@ static int jpeg_v5_0_0_stop(struct amdgpu_device *adev)
{
	int r;

	if (adev->pg_flags & AMD_PG_SUPPORT_JPEG_DPG) {
		jpeg_v5_0_0_stop_dpg_mode(adev, 0);
	} else {

		/* reset JMI */
		WREG32_P(SOC15_REG_OFFSET(JPEG, 0, regUVD_JMI_CNTL),
			UVD_JMI_CNTL__SOFT_RESET_MASK,
@@ -348,9 +472,10 @@ static int jpeg_v5_0_0_stop(struct amdgpu_device *adev)
		jpeg_v5_0_0_enable_clock_gating(adev);

		/* enable power gating */
	r = jpeg_v5_0_0_enable_static_power_gating(adev);
		r = jpeg_v5_0_0_enable_power_gating(adev);
		if (r)
			return r;
	}

	if (adev->pm.dpm_enabled)
		amdgpu_dpm_enable_jpeg(adev, false);
+6 −0
Original line number Diff line number Diff line
@@ -24,6 +24,12 @@
#ifndef __JPEG_V5_0_0_H__
#define __JPEG_V5_0_0_H__

#define vcnipJPEG_CGC_GATE                                 0x4160
#define vcnipJPEG_CGC_CTRL                                 0x4161
#define vcnipJPEG_SYS_INT_EN                               0x4141
#define vcnipUVD_NO_OP                                     0x0029
#define vcnipJPEG_DEC_GFX10_ADDR_CONFIG                    0x404A

extern const struct amdgpu_ip_block_version jpeg_v5_0_0_ip_block;

#endif /* __JPEG_V5_0_0_H__ */
+1 −0
Original line number Diff line number Diff line
@@ -428,6 +428,7 @@ static int soc24_common_early_init(void *handle)

		adev->pg_flags = AMD_PG_SUPPORT_VCN |
			AMD_PG_SUPPORT_JPEG |
			AMD_PG_SUPPORT_JPEG_DPG |
			AMD_PG_SUPPORT_VCN_DPG;
		adev->external_rev_id = adev->rev_id + 0x50;
		break;