Merge branch 'amdkfd-v6' of git://people.freedesktop.org/~gabbayo/linux into drm-next

Merge AMDKFD it seems clean enough.
* 'amdkfd-v6' of git://people.freedesktop.org/~gabbayo/linux: (29 commits)
  amdkfd: Implement the Get Version IOCTL
  amdkfd: Implement the Get Process Aperture IOCTL
  amdkfd: Implement the Get Clock Counters IOCTL
  amdkfd: Implement the Set Memory Policy IOCTL
  amdkfd: Implement the create/destroy/update queue IOCTLs
  amdkfd: Add interrupt handling module
  amdkfd: Add device queue manager module
  amdkfd: Add process queue manager module
  amdkfd: Add packet manager module
  amdkfd: Add module parameter of scheduling policy
  amdkfd: Add kernel queue module
  amdkfd: Add mqd_manager module
  amdkfd: Add queue module
  amdkfd: Add binding/unbinding calls to amd_iommu driver
  amdkfd: Add basic modules to amdkfd
  amdkfd: Add topology module to amdkfd
  amdkfd: Add amdkfd skeleton driver
  amdkfd: Add IOCTL set definitions of amdkfd
  Update MAINTAINERS and CREDITS files with amdkfd info
  drm/radeon: Add radeon <--> amdkfd interface
  ...

10 files changed, 914 insertions, 95 deletions

diff --git a/drivers/gpu/drm/radeon/Makefile b/drivers/gpu/drm/radeon/Makefile
index d01b87991422..bad6caa0a727 100644
--- a/drivers/gpu/drm/radeon/Makefile
+++ b/drivers/gpu/drm/radeon/Makefile
@@ -104,6 +104,7 @@ radeon-y += \
 	radeon_vce.o \
 	vce_v1_0.o \
 	vce_v2_0.o \
+	radeon_kfd.o
 
 radeon-$(CONFIG_COMPAT) += radeon_ioc32.o
 radeon-$(CONFIG_VGA_SWITCHEROO) += radeon_atpx_handler.o
diff --git a/drivers/gpu/drm/radeon/cik.c b/drivers/gpu/drm/radeon/cik.c
index 57a359d34830..bce73b6203ac 100644
--- a/drivers/gpu/drm/radeon/cik.c
+++ b/drivers/gpu/drm/radeon/cik.c
@@ -32,6 +32,7 @@
 #include "cik_blit_shaders.h"
 #include "radeon_ucode.h"
 #include "clearstate_ci.h"
+#include "radeon_kfd.h"
 
 MODULE_FIRMWARE("radeon/BONAIRE_pfp.bin");
 MODULE_FIRMWARE("radeon/BONAIRE_me.bin");
@@ -1563,6 +1564,8 @@ static const u32 godavari_golden_registers[] =
 
 static void cik_init_golden_registers(struct radeon_device *rdev)
 {
+	/* Some of the registers might be dependent on GRBM_GFX_INDEX */
+	mutex_lock(&rdev->grbm_idx_mutex);
 	switch (rdev->family) {
 	case CHIP_BONAIRE:
 		radeon_program_register_sequence(rdev,
@@ -1637,6 +1640,7 @@ static void cik_init_golden_registers(struct radeon_device *rdev)
 	default:
 		break;
 	}
+	mutex_unlock(&rdev->grbm_idx_mutex);
 }
 
 /**
@@ -3428,6 +3432,7 @@ static void cik_setup_rb(struct radeon_device *rdev,
 	u32 disabled_rbs = 0;
 	u32 enabled_rbs = 0;
 
+	mutex_lock(&rdev->grbm_idx_mutex);
 	for (i = 0; i < se_num; i++) {
 		for (j = 0; j < sh_per_se; j++) {
 			cik_select_se_sh(rdev, i, j);
@@ -3439,6 +3444,7 @@ static void cik_setup_rb(struct radeon_device *rdev,
 		}
 	}
 	cik_select_se_sh(rdev, 0xffffffff, 0xffffffff);
+	mutex_unlock(&rdev->grbm_idx_mutex);
 
 	mask = 1;
 	for (i = 0; i < max_rb_num_per_se * se_num; i++) {
@@ -3449,6 +3455,7 @@ static void cik_setup_rb(struct radeon_device *rdev,
 
 	rdev->config.cik.backend_enable_mask = enabled_rbs;
 
+	mutex_lock(&rdev->grbm_idx_mutex);
 	for (i = 0; i < se_num; i++) {
 		cik_select_se_sh(rdev, i, 0xffffffff);
 		data = 0;
@@ -3476,6 +3483,7 @@ static void cik_setup_rb(struct radeon_device *rdev,
 		WREG32(PA_SC_RASTER_CONFIG, data);
 	}
 	cik_select_se_sh(rdev, 0xffffffff, 0xffffffff);
+	mutex_unlock(&rdev->grbm_idx_mutex);
 }
 
 /**
@@ -3693,6 +3701,12 @@ static void cik_gpu_init(struct radeon_device *rdev)
 	/* set HW defaults for 3D engine */
 	WREG32(CP_MEQ_THRESHOLDS, MEQ1_START(0x30) | MEQ2_START(0x60));
 
+	mutex_lock(&rdev->grbm_idx_mutex);
+	/*
+	 * making sure that the following register writes will be broadcasted
+	 * to all the shaders
+	 */
+	cik_select_se_sh(rdev, 0xffffffff, 0xffffffff);
 	WREG32(SX_DEBUG_1, 0x20);
 
 	WREG32(TA_CNTL_AUX, 0x00010000);
@@ -3748,6 +3762,7 @@ static void cik_gpu_init(struct radeon_device *rdev)
 
 	WREG32(PA_CL_ENHANCE, CLIP_VTX_REORDER_ENA | NUM_CLIP_SEQ(3));
 	WREG32(PA_SC_ENHANCE, ENABLE_PA_SC_OUT_OF_ORDER);
+	mutex_unlock(&rdev->grbm_idx_mutex);
 
 	udelay(50);
 }
@@ -4684,12 +4699,11 @@ static int cik_mec_init(struct radeon_device *rdev)
 	/*
 	 * KV:    2 MEC, 4 Pipes/MEC, 8 Queues/Pipe - 64 Queues total
 	 * CI/KB: 1 MEC, 4 Pipes/MEC, 8 Queues/Pipe - 32 Queues total
+	 * Nonetheless, we assign only 1 pipe because all other pipes will
+	 * be handled by KFD
 	 */
-	if (rdev->family == CHIP_KAVERI)
-		rdev->mec.num_mec = 2;
-	else
-		rdev->mec.num_mec = 1;
-	rdev->mec.num_pipe = 4;
+	rdev->mec.num_mec = 1;
+	rdev->mec.num_pipe = 1;
 	rdev->mec.num_queue = rdev->mec.num_mec * rdev->mec.num_pipe * 8;
 
 	if (rdev->mec.hpd_eop_obj == NULL) {
@@ -4831,28 +4845,24 @@ static int cik_cp_compute_resume(struct radeon_device *rdev)
 
 	/* init the pipes */
 	mutex_lock(&rdev->srbm_mutex);
-	for (i = 0; i < (rdev->mec.num_pipe * rdev->mec.num_mec); i++) {
-		int me = (i < 4) ? 1 : 2;
-		int pipe = (i < 4) ? i : (i - 4);
 
-		eop_gpu_addr = rdev->mec.hpd_eop_gpu_addr + (i * MEC_HPD_SIZE * 2);
+	eop_gpu_addr = rdev->mec.hpd_eop_gpu_addr;
 
-		cik_srbm_select(rdev, me, pipe, 0, 0);
+	cik_srbm_select(rdev, 0, 0, 0, 0);
 
-		/* write the EOP addr */
-		WREG32(CP_HPD_EOP_BASE_ADDR, eop_gpu_addr >> 8);
-		WREG32(CP_HPD_EOP_BASE_ADDR_HI, upper_32_bits(eop_gpu_addr) >> 8);
+	/* write the EOP addr */
+	WREG32(CP_HPD_EOP_BASE_ADDR, eop_gpu_addr >> 8);
+	WREG32(CP_HPD_EOP_BASE_ADDR_HI, upper_32_bits(eop_gpu_addr) >> 8);
 
-		/* set the VMID assigned */
-		WREG32(CP_HPD_EOP_VMID, 0);
+	/* set the VMID assigned */
+	WREG32(CP_HPD_EOP_VMID, 0);
+
+	/* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
+	tmp = RREG32(CP_HPD_EOP_CONTROL);
+	tmp &= ~EOP_SIZE_MASK;
+	tmp |= order_base_2(MEC_HPD_SIZE / 8);
+	WREG32(CP_HPD_EOP_CONTROL, tmp);
 
-		/* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
-		tmp = RREG32(CP_HPD_EOP_CONTROL);
-		tmp &= ~EOP_SIZE_MASK;
-		tmp |= order_base_2(MEC_HPD_SIZE / 8);
-		WREG32(CP_HPD_EOP_CONTROL, tmp);
-	}
-	cik_srbm_select(rdev, 0, 0, 0, 0);
 	mutex_unlock(&rdev->srbm_mutex);
 
 	/* init the queues.  Just two for now. */
@@ -5906,8 +5916,13 @@ int cik_ib_parse(struct radeon_device *rdev, struct radeon_ib *ib)
  */
 int cik_vm_init(struct radeon_device *rdev)
 {
-	/* number of VMs */
-	rdev->vm_manager.nvm = 16;
+	/*
+	 * number of VMs
+	 * VMID 0 is reserved for System
+	 * radeon graphics/compute will use VMIDs 1-7
+	 * amdkfd will use VMIDs 8-15
+	 */
+	rdev->vm_manager.nvm = RADEON_NUM_OF_VMIDS;
 	/* base offset of vram pages */
 	if (rdev->flags & RADEON_IS_IGP) {
 		u64 tmp = RREG32(MC_VM_FB_OFFSET);
@@ -6068,6 +6083,7 @@ static void cik_wait_for_rlc_serdes(struct radeon_device *rdev)
 	u32 i, j, k;
 	u32 mask;
 
+	mutex_lock(&rdev->grbm_idx_mutex);
 	for (i = 0; i < rdev->config.cik.max_shader_engines; i++) {
 		for (j = 0; j < rdev->config.cik.max_sh_per_se; j++) {
 			cik_select_se_sh(rdev, i, j);
@@ -6079,6 +6095,7 @@ static void cik_wait_for_rlc_serdes(struct radeon_device *rdev)
 		}
 	}
 	cik_select_se_sh(rdev, 0xffffffff, 0xffffffff);
+	mutex_unlock(&rdev->grbm_idx_mutex);
 
 	mask = SE_MASTER_BUSY_MASK | GC_MASTER_BUSY | TC0_MASTER_BUSY | TC1_MASTER_BUSY;
 	for (k = 0; k < rdev->usec_timeout; k++) {
@@ -6213,10 +6230,12 @@ static int cik_rlc_resume(struct radeon_device *rdev)
 	WREG32(RLC_LB_CNTR_INIT, 0);
 	WREG32(RLC_LB_CNTR_MAX, 0x00008000);
 
+	mutex_lock(&rdev->grbm_idx_mutex);
 	cik_select_se_sh(rdev, 0xffffffff, 0xffffffff);
 	WREG32(RLC_LB_INIT_CU_MASK, 0xffffffff);
 	WREG32(RLC_LB_PARAMS, 0x00600408);
 	WREG32(RLC_LB_CNTL, 0x80000004);
+	mutex_unlock(&rdev->grbm_idx_mutex);
 
 	WREG32(RLC_MC_CNTL, 0);
 	WREG32(RLC_UCODE_CNTL, 0);
@@ -6283,11 +6302,13 @@ static void cik_enable_cgcg(struct radeon_device *rdev, bool enable)
 
 		tmp = cik_halt_rlc(rdev);
 
+		mutex_lock(&rdev->grbm_idx_mutex);
 		cik_select_se_sh(rdev, 0xffffffff, 0xffffffff);
 		WREG32(RLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
 		WREG32(RLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
 		tmp2 = BPM_ADDR_MASK | CGCG_OVERRIDE_0 | CGLS_ENABLE;
 		WREG32(RLC_SERDES_WR_CTRL, tmp2);
+		mutex_unlock(&rdev->grbm_idx_mutex);
 
 		cik_update_rlc(rdev, tmp);
 
@@ -6329,11 +6350,13 @@ static void cik_enable_mgcg(struct radeon_device *rdev, bool enable)
 
 		tmp = cik_halt_rlc(rdev);
 
+		mutex_lock(&rdev->grbm_idx_mutex);
 		cik_select_se_sh(rdev, 0xffffffff, 0xffffffff);
 		WREG32(RLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
 		WREG32(RLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
 		data = BPM_ADDR_MASK | MGCG_OVERRIDE_0;
 		WREG32(RLC_SERDES_WR_CTRL, data);
+		mutex_unlock(&rdev->grbm_idx_mutex);
 
 		cik_update_rlc(rdev, tmp);
 
@@ -6377,11 +6400,13 @@ static void cik_enable_mgcg(struct radeon_device *rdev, bool enable)
 
 		tmp = cik_halt_rlc(rdev);
 
+		mutex_lock(&rdev->grbm_idx_mutex);
 		cik_select_se_sh(rdev, 0xffffffff, 0xffffffff);
 		WREG32(RLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
 		WREG32(RLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
 		data = BPM_ADDR_MASK | MGCG_OVERRIDE_1;
 		WREG32(RLC_SERDES_WR_CTRL, data);
+		mutex_unlock(&rdev->grbm_idx_mutex);
 
 		cik_update_rlc(rdev, tmp);
 	}
@@ -6810,10 +6835,12 @@ static u32 cik_get_cu_active_bitmap(struct radeon_device *rdev, u32 se, u32 sh)
 	u32 mask = 0, tmp, tmp1;
 	int i;
 
+	mutex_lock(&rdev->grbm_idx_mutex);
 	cik_select_se_sh(rdev, se, sh);
 	tmp = RREG32(CC_GC_SHADER_ARRAY_CONFIG);
 	tmp1 = RREG32(GC_USER_SHADER_ARRAY_CONFIG);
 	cik_select_se_sh(rdev, 0xffffffff, 0xffffffff);
+	mutex_unlock(&rdev->grbm_idx_mutex);
 
 	tmp &= 0xffff0000;
 
@@ -7297,8 +7324,7 @@ static int cik_irq_init(struct radeon_device *rdev)
 int cik_irq_set(struct radeon_device *rdev)
 {
 	u32 cp_int_cntl;
-	u32 cp_m1p0, cp_m1p1, cp_m1p2, cp_m1p3;
-	u32 cp_m2p0, cp_m2p1, cp_m2p2, cp_m2p3;
+	u32 cp_m1p0;
 	u32 crtc1 = 0, crtc2 = 0, crtc3 = 0, crtc4 = 0, crtc5 = 0, crtc6 = 0;
 	u32 hpd1, hpd2, hpd3, hpd4, hpd5, hpd6;
 	u32 grbm_int_cntl = 0;
@@ -7332,13 +7358,6 @@ int cik_irq_set(struct radeon_device *rdev)
 	dma_cntl1 = RREG32(SDMA0_CNTL + SDMA1_REGISTER_OFFSET) & ~TRAP_ENABLE;
 
 	cp_m1p0 = RREG32(CP_ME1_PIPE0_INT_CNTL) & ~TIME_STAMP_INT_ENABLE;
-	cp_m1p1 = RREG32(CP_ME1_PIPE1_INT_CNTL) & ~TIME_STAMP_INT_ENABLE;
-	cp_m1p2 = RREG32(CP_ME1_PIPE2_INT_CNTL) & ~TIME_STAMP_INT_ENABLE;
-	cp_m1p3 = RREG32(CP_ME1_PIPE3_INT_CNTL) & ~TIME_STAMP_INT_ENABLE;
-	cp_m2p0 = RREG32(CP_ME2_PIPE0_INT_CNTL) & ~TIME_STAMP_INT_ENABLE;
-	cp_m2p1 = RREG32(CP_ME2_PIPE1_INT_CNTL) & ~TIME_STAMP_INT_ENABLE;
-	cp_m2p2 = RREG32(CP_ME2_PIPE2_INT_CNTL) & ~TIME_STAMP_INT_ENABLE;
-	cp_m2p3 = RREG32(CP_ME2_PIPE3_INT_CNTL) & ~TIME_STAMP_INT_ENABLE;
 
 	if (rdev->flags & RADEON_IS_IGP)
 		thermal_int = RREG32_SMC(CG_THERMAL_INT_CTRL) &
@@ -7360,33 +7379,6 @@ int cik_irq_set(struct radeon_device *rdev)
 			case 0:
 				cp_m1p0 |= TIME_STAMP_INT_ENABLE;
 				break;
-			case 1:
-				cp_m1p1 |= TIME_STAMP_INT_ENABLE;
-				break;
-			case 2:
-				cp_m1p2 |= TIME_STAMP_INT_ENABLE;
-				break;
-			case 3:
-				cp_m1p2 |= TIME_STAMP_INT_ENABLE;
-				break;
-			default:
-				DRM_DEBUG("si_irq_set: sw int cp1 invalid pipe %d\n", ring->pipe);
-				break;
-			}
-		} else if (ring->me == 2) {
-			switch (ring->pipe) {
-			case 0:
-				cp_m2p0 |= TIME_STAMP_INT_ENABLE;
-				break;
-			case 1:
-				cp_m2p1 |= TIME_STAMP_INT_ENABLE;
-				break;
-			case 2:
-				cp_m2p2 |= TIME_STAMP_INT_ENABLE;
-				break;
-			case 3:
-				cp_m2p2 |= TIME_STAMP_INT_ENABLE;
-				break;
 			default:
 				DRM_DEBUG("si_irq_set: sw int cp1 invalid pipe %d\n", ring->pipe);
 				break;
@@ -7403,33 +7395,6 @@ int cik_irq_set(struct radeon_device *rdev)
 			case 0:
 				cp_m1p0 |= TIME_STAMP_INT_ENABLE;
 				break;
-			case 1:
-				cp_m1p1 |= TIME_STAMP_INT_ENABLE;
-				break;
-			case 2:
-				cp_m1p2 |= TIME_STAMP_INT_ENABLE;
-				break;
-			case 3:
-				cp_m1p2 |= TIME_STAMP_INT_ENABLE;
-				break;
-			default:
-				DRM_DEBUG("si_irq_set: sw int cp2 invalid pipe %d\n", ring->pipe);
-				break;
-			}
-		} else if (ring->me == 2) {
-			switch (ring->pipe) {
-			case 0:
-				cp_m2p0 |= TIME_STAMP_INT_ENABLE;
-				break;
-			case 1:
-				cp_m2p1 |= TIME_STAMP_INT_ENABLE;
-				break;
-			case 2:
-				cp_m2p2 |= TIME_STAMP_INT_ENABLE;
-				break;
-			case 3:
-				cp_m2p2 |= TIME_STAMP_INT_ENABLE;
-				break;
 			default:
 				DRM_DEBUG("si_irq_set: sw int cp2 invalid pipe %d\n", ring->pipe);
 				break;
@@ -7518,13 +7483,6 @@ int cik_irq_set(struct radeon_device *rdev)
 	WREG32(SDMA0_CNTL + SDMA1_REGISTER_OFFSET, dma_cntl1);
 
 	WREG32(CP_ME1_PIPE0_INT_CNTL, cp_m1p0);
-	WREG32(CP_ME1_PIPE1_INT_CNTL, cp_m1p1);
-	WREG32(CP_ME1_PIPE2_INT_CNTL, cp_m1p2);
-	WREG32(CP_ME1_PIPE3_INT_CNTL, cp_m1p3);
-	WREG32(CP_ME2_PIPE0_INT_CNTL, cp_m2p0);
-	WREG32(CP_ME2_PIPE1_INT_CNTL, cp_m2p1);
-	WREG32(CP_ME2_PIPE2_INT_CNTL, cp_m2p2);
-	WREG32(CP_ME2_PIPE3_INT_CNTL, cp_m2p3);
 
 	WREG32(GRBM_INT_CNTL, grbm_int_cntl);
 
@@ -7841,6 +7799,10 @@ restart_ih:
 	while (rptr != wptr) {
 		/* wptr/rptr are in bytes! */
 		ring_index = rptr / 4;
+
+		radeon_kfd_interrupt(rdev,
+				(const void *) &rdev->ih.ring[ring_index]);
+
 		src_id =  le32_to_cpu(rdev->ih.ring[ring_index]) & 0xff;
 		src_data = le32_to_cpu(rdev->ih.ring[ring_index + 1]) & 0xfffffff;
 		ring_id = le32_to_cpu(rdev->ih.ring[ring_index + 2]) & 0xff;
@@ -8530,6 +8492,10 @@ static int cik_startup(struct radeon_device *rdev)
 	if (r)
 		return r;
 
+	r = radeon_kfd_resume(rdev);
+	if (r)
+		return r;
+
 	return 0;
 }
 
@@ -8578,6 +8544,7 @@ int cik_resume(struct radeon_device *rdev)
  */
 int cik_suspend(struct radeon_device *rdev)
 {
+	radeon_kfd_suspend(rdev);
 	radeon_pm_suspend(rdev);
 	dce6_audio_fini(rdev);
 	radeon_vm_manager_fini(rdev);
diff --git a/drivers/gpu/drm/radeon/cik_reg.h b/drivers/gpu/drm/radeon/cik_reg.h
index ca1bb6133580..79c45e8a536b 100644
--- a/drivers/gpu/drm/radeon/cik_reg.h
+++ b/drivers/gpu/drm/radeon/cik_reg.h
@@ -147,4 +147,140 @@
 
 #define CIK_LB_DESKTOP_HEIGHT                     0x6b0c
 
+#define CP_HQD_IQ_RPTR					0xC970u
+#define AQL_ENABLE					(1U << 0)
+
+#define IDLE					(1 << 2)
+
+struct cik_mqd {
+	uint32_t header;
+	uint32_t compute_dispatch_initiator;
+	uint32_t compute_dim_x;
+	uint32_t compute_dim_y;
+	uint32_t compute_dim_z;
+	uint32_t compute_start_x;
+	uint32_t compute_start_y;
+	uint32_t compute_start_z;
+	uint32_t compute_num_thread_x;
+	uint32_t compute_num_thread_y;
+	uint32_t compute_num_thread_z;
+	uint32_t compute_pipelinestat_enable;
+	uint32_t compute_perfcount_enable;
+	uint32_t compute_pgm_lo;
+	uint32_t compute_pgm_hi;
+	uint32_t compute_tba_lo;
+	uint32_t compute_tba_hi;
+	uint32_t compute_tma_lo;
+	uint32_t compute_tma_hi;
+	uint32_t compute_pgm_rsrc1;
+	uint32_t compute_pgm_rsrc2;
+	uint32_t compute_vmid;
+	uint32_t compute_resource_limits;
+	uint32_t compute_static_thread_mgmt_se0;
+	uint32_t compute_static_thread_mgmt_se1;
+	uint32_t compute_tmpring_size;
+	uint32_t compute_static_thread_mgmt_se2;
+	uint32_t compute_static_thread_mgmt_se3;
+	uint32_t compute_restart_x;
+	uint32_t compute_restart_y;
+	uint32_t compute_restart_z;
+	uint32_t compute_thread_trace_enable;
+	uint32_t compute_misc_reserved;
+	uint32_t compute_user_data_0;
+	uint32_t compute_user_data_1;
+	uint32_t compute_user_data_2;
+	uint32_t compute_user_data_3;
+	uint32_t compute_user_data_4;
+	uint32_t compute_user_data_5;
+	uint32_t compute_user_data_6;
+	uint32_t compute_user_data_7;
+	uint32_t compute_user_data_8;
+	uint32_t compute_user_data_9;
+	uint32_t compute_user_data_10;
+	uint32_t compute_user_data_11;
+	uint32_t compute_user_data_12;
+	uint32_t compute_user_data_13;
+	uint32_t compute_user_data_14;
+	uint32_t compute_user_data_15;
+	uint32_t cp_compute_csinvoc_count_lo;
+	uint32_t cp_compute_csinvoc_count_hi;
+	uint32_t cp_mqd_base_addr_lo;
+	uint32_t cp_mqd_base_addr_hi;
+	uint32_t cp_hqd_active;
+	uint32_t cp_hqd_vmid;
+	uint32_t cp_hqd_persistent_state;
+	uint32_t cp_hqd_pipe_priority;
+	uint32_t cp_hqd_queue_priority;
+	uint32_t cp_hqd_quantum;
+	uint32_t cp_hqd_pq_base_lo;
+	uint32_t cp_hqd_pq_base_hi;
+	uint32_t cp_hqd_pq_rptr;
+	uint32_t cp_hqd_pq_rptr_report_addr_lo;
+	uint32_t cp_hqd_pq_rptr_report_addr_hi;
+	uint32_t cp_hqd_pq_wptr_poll_addr_lo;
+	uint32_t cp_hqd_pq_wptr_poll_addr_hi;
+	uint32_t cp_hqd_pq_doorbell_control;
+	uint32_t cp_hqd_pq_wptr;
+	uint32_t cp_hqd_pq_control;
+	uint32_t cp_hqd_ib_base_addr_lo;
+	uint32_t cp_hqd_ib_base_addr_hi;
+	uint32_t cp_hqd_ib_rptr;
+	uint32_t cp_hqd_ib_control;
+	uint32_t cp_hqd_iq_timer;
+	uint32_t cp_hqd_iq_rptr;
+	uint32_t cp_hqd_dequeue_request;
+	uint32_t cp_hqd_dma_offload;
+	uint32_t cp_hqd_sema_cmd;
+	uint32_t cp_hqd_msg_type;
+	uint32_t cp_hqd_atomic0_preop_lo;
+	uint32_t cp_hqd_atomic0_preop_hi;
+	uint32_t cp_hqd_atomic1_preop_lo;
+	uint32_t cp_hqd_atomic1_preop_hi;
+	uint32_t cp_hqd_hq_status0;
+	uint32_t cp_hqd_hq_control0;
+	uint32_t cp_mqd_control;
+	uint32_t cp_mqd_query_time_lo;
+	uint32_t cp_mqd_query_time_hi;
+	uint32_t cp_mqd_connect_start_time_lo;
+	uint32_t cp_mqd_connect_start_time_hi;
+	uint32_t cp_mqd_connect_end_time_lo;
+	uint32_t cp_mqd_connect_end_time_hi;
+	uint32_t cp_mqd_connect_end_wf_count;
+	uint32_t cp_mqd_connect_end_pq_rptr;
+	uint32_t cp_mqd_connect_end_pq_wptr;
+	uint32_t cp_mqd_connect_end_ib_rptr;
+	uint32_t reserved_96;
+	uint32_t reserved_97;
+	uint32_t reserved_98;
+	uint32_t reserved_99;
+	uint32_t iqtimer_pkt_header;
+	uint32_t iqtimer_pkt_dw0;
+	uint32_t iqtimer_pkt_dw1;
+	uint32_t iqtimer_pkt_dw2;
+	uint32_t iqtimer_pkt_dw3;
+	uint32_t iqtimer_pkt_dw4;
+	uint32_t iqtimer_pkt_dw5;
+	uint32_t iqtimer_pkt_dw6;
+	uint32_t reserved_108;
+	uint32_t reserved_109;
+	uint32_t reserved_110;
+	uint32_t reserved_111;
+	uint32_t queue_doorbell_id0;
+	uint32_t queue_doorbell_id1;
+	uint32_t queue_doorbell_id2;
+	uint32_t queue_doorbell_id3;
+	uint32_t queue_doorbell_id4;
+	uint32_t queue_doorbell_id5;
+	uint32_t queue_doorbell_id6;
+	uint32_t queue_doorbell_id7;
+	uint32_t queue_doorbell_id8;
+	uint32_t queue_doorbell_id9;
+	uint32_t queue_doorbell_id10;
+	uint32_t queue_doorbell_id11;
+	uint32_t queue_doorbell_id12;
+	uint32_t queue_doorbell_id13;
+	uint32_t queue_doorbell_id14;
+	uint32_t queue_doorbell_id15;
+};
+
 #endif
diff --git a/drivers/gpu/drm/radeon/cikd.h b/drivers/gpu/drm/radeon/cikd.h
index 0c6e1b55d968..068cbb019326 100644
--- a/drivers/gpu/drm/radeon/cikd.h
+++ b/drivers/gpu/drm/radeon/cikd.h
@@ -30,6 +30,8 @@
 #define CIK_RB_BITMAP_WIDTH_PER_SH     2
 #define HAWAII_RB_BITMAP_WIDTH_PER_SH  4
 
+#define RADEON_NUM_OF_VMIDS	8
+
 /* DIDT IND registers */
 #define DIDT_SQ_CTRL0                                     0x0
 #       define DIDT_CTRL_EN                               (1 << 0)
@@ -1137,6 +1139,9 @@
 #define			SH_MEM_ALIGNMENT_MODE_UNALIGNED			3
 #define		DEFAULT_MTYPE(x)				((x) << 4)
 #define		APE1_MTYPE(x)					((x) << 7)
+/* valid for both DEFAULT_MTYPE and APE1_MTYPE */
+#define	MTYPE_CACHED					0
+#define	MTYPE_NONCACHED					3
 
 #define	SX_DEBUG_1					0x9060
 
@@ -1447,6 +1452,16 @@
 #define CP_HQD_ACTIVE                                     0xC91C
 #define CP_HQD_VMID                                       0xC920
 
+#define CP_HQD_PERSISTENT_STATE				0xC924u
+#define	DEFAULT_CP_HQD_PERSISTENT_STATE			(0x33U << 8)
+
+#define CP_HQD_PIPE_PRIORITY				0xC928u
+#define CP_HQD_QUEUE_PRIORITY				0xC92Cu
+#define CP_HQD_QUANTUM					0xC930u
+#define	QUANTUM_EN					1U
+#define	QUANTUM_SCALE_1MS				(1U << 4)
+#define	QUANTUM_DURATION(x)				((x) << 8)
+
 #define CP_HQD_PQ_BASE                                    0xC934
 #define CP_HQD_PQ_BASE_HI                                 0xC938
 #define CP_HQD_PQ_RPTR                                    0xC93C
@@ -1474,12 +1489,32 @@
 #define		PRIV_STATE      			(1 << 30)
 #define		KMD_QUEUE      				(1 << 31)
 
-#define CP_HQD_DEQUEUE_REQUEST                          0xC974
+#define CP_HQD_IB_BASE_ADDR				0xC95Cu
+#define CP_HQD_IB_BASE_ADDR_HI			0xC960u
+#define CP_HQD_IB_RPTR					0xC964u
+#define CP_HQD_IB_CONTROL				0xC968u
+#define	IB_ATC_EN					(1U << 23)
+#define	DEFAULT_MIN_IB_AVAIL_SIZE			(3U << 20)
+
+#define CP_HQD_DEQUEUE_REQUEST			0xC974
+#define	DEQUEUE_REQUEST_DRAIN				1
+#define DEQUEUE_REQUEST_RESET				2
 
 #define CP_MQD_CONTROL                                  0xC99C
 #define		MQD_VMID(x)				((x) << 0)
 #define		MQD_VMID_MASK      			(0xf << 0)
 
+#define CP_HQD_SEMA_CMD					0xC97Cu
+#define CP_HQD_MSG_TYPE					0xC980u
+#define CP_HQD_ATOMIC0_PREOP_LO			0xC984u
+#define CP_HQD_ATOMIC0_PREOP_HI			0xC988u
+#define CP_HQD_ATOMIC1_PREOP_LO			0xC98Cu
+#define CP_HQD_ATOMIC1_PREOP_HI			0xC990u
+#define CP_HQD_HQ_SCHEDULER0			0xC994u
+#define CP_HQD_HQ_SCHEDULER1			0xC998u
+
+#define SH_STATIC_MEM_CONFIG			0x9604u
+
 #define DB_RENDER_CONTROL                               0x28000
 
 #define PA_SC_RASTER_CONFIG                             0x28350
@@ -2069,4 +2104,20 @@
 #define VCE_CMD_IB_AUTO		0x00000005
 #define VCE_CMD_SEMAPHORE	0x00000006
 
+#define ATC_VMID0_PASID_MAPPING					0x339Cu
+#define	ATC_VMID_PASID_MAPPING_UPDATE_STATUS	0x3398u
+#define	ATC_VMID_PASID_MAPPING_VALID				(1U << 31)
+
+#define ATC_VM_APERTURE0_CNTL					0x3310u
+#define	ATS_ACCESS_MODE_NEVER						0
+#define	ATS_ACCESS_MODE_ALWAYS						1
+
+#define ATC_VM_APERTURE0_CNTL2					0x3318u
+#define ATC_VM_APERTURE0_HIGH_ADDR				0x3308u
+#define ATC_VM_APERTURE0_LOW_ADDR				0x3300u
+#define ATC_VM_APERTURE1_CNTL					0x3314u
+#define ATC_VM_APERTURE1_CNTL2					0x331Cu
+#define ATC_VM_APERTURE1_HIGH_ADDR				0x330Cu
+#define ATC_VM_APERTURE1_LOW_ADDR				0x3304u
+
 #endif
diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h
index 908f349a6362..1f61ff089c9e 100644
--- a/drivers/gpu/drm/radeon/radeon.h
+++ b/drivers/gpu/drm/radeon/radeon.h
@@ -701,6 +701,10 @@ struct radeon_doorbell {
 
 int radeon_doorbell_get(struct radeon_device *rdev, u32 *page);
 void radeon_doorbell_free(struct radeon_device *rdev, u32 doorbell);
+void radeon_doorbell_get_kfd_info(struct radeon_device *rdev,
+				  phys_addr_t *aperture_base,
+				  size_t *aperture_size,
+				  size_t *start_offset);
 
 /*
  * IRQS.
@@ -2393,6 +2397,8 @@ struct radeon_device {
 	struct radeon_atcs		atcs;
 	/* srbm instance registers */
 	struct mutex			srbm_mutex;
+	/* GRBM index mutex. Protects concurrents access to GRBM index */
+	struct mutex			grbm_idx_mutex;
 	/* clock, powergating flags */
 	u32 cg_flags;
 	u32 pg_flags;
@@ -2405,6 +2411,10 @@ struct radeon_device {
 	u64 vram_pin_size;
 	u64 gart_pin_size;
 
+	/* amdkfd interface */
+	struct kfd_dev		*kfd;
+	struct radeon_sa_manager	kfd_bo;
+
 	struct mutex	mn_lock;
 	DECLARE_HASHTABLE(mn_hash, 7);
 };
diff --git a/drivers/gpu/drm/radeon/radeon_device.c b/drivers/gpu/drm/radeon/radeon_device.c
index ea2676954dde..ae87310fd96e 100644
--- a/drivers/gpu/drm/radeon/radeon_device.c
+++ b/drivers/gpu/drm/radeon/radeon_device.c
@@ -377,6 +377,37 @@ void radeon_doorbell_free(struct radeon_device *rdev, u32 doorbell)
 		__clear_bit(doorbell, rdev->doorbell.used);
 }
 
+/**
+ * radeon_doorbell_get_kfd_info - Report doorbell configuration required to
+ *                                setup KFD
+ *
+ * @rdev: radeon_device pointer
+ * @aperture_base: output returning doorbell aperture base physical address
+ * @aperture_size: output returning doorbell aperture size in bytes
+ * @start_offset: output returning # of doorbell bytes reserved for radeon.
+ *
+ * Radeon and the KFD share the doorbell aperture. Radeon sets it up,
+ * takes doorbells required for its own rings and reports the setup to KFD.
+ * Radeon reserved doorbells are at the start of the doorbell aperture.
+ */
+void radeon_doorbell_get_kfd_info(struct radeon_device *rdev,
+				  phys_addr_t *aperture_base,
+				  size_t *aperture_size,
+				  size_t *start_offset)
+{
+	/* The first num_doorbells are used by radeon.
+	 * KFD takes whatever's left in the aperture. */
+	if (rdev->doorbell.size > rdev->doorbell.num_doorbells * sizeof(u32)) {
+		*aperture_base = rdev->doorbell.base;
+		*aperture_size = rdev->doorbell.size;
+		*start_offset = rdev->doorbell.num_doorbells * sizeof(u32);
+	} else {
+		*aperture_base = 0;
+		*aperture_size = 0;
+		*start_offset = 0;
+	}
+}
+
 /*
  * radeon_wb_*()
  * Writeback is the the method by which the the GPU updates special pages
@@ -1272,6 +1303,7 @@ int radeon_device_init(struct radeon_device *rdev,
 	mutex_init(&rdev->pm.mutex);
 	mutex_init(&rdev->gpu_clock_mutex);
 	mutex_init(&rdev->srbm_mutex);
+	mutex_init(&rdev->grbm_idx_mutex);
 	init_rwsem(&rdev->pm.mclk_lock);
 	init_rwsem(&rdev->exclusive_lock);
 	init_waitqueue_head(&rdev->irq.vblank_queue);
diff --git a/drivers/gpu/drm/radeon/radeon_drv.c b/drivers/gpu/drm/radeon/radeon_drv.c
index dcffa30ee2db..4f50fb0e3d93 100644
--- a/drivers/gpu/drm/radeon/radeon_drv.c
+++ b/drivers/gpu/drm/radeon/radeon_drv.c
@@ -41,6 +41,8 @@
 #include <drm/drm_gem.h>
 
 #include "drm_crtc_helper.h"
+#include "radeon_kfd.h"
+
 /*
  * KMS wrapper.
  * - 2.0.0 - initial interface
@@ -654,12 +656,15 @@ static int __init radeon_init(void)
 #endif
 	}
 
+	radeon_kfd_init();
+
 	/* let modprobe override vga console setting */
 	return drm_pci_init(driver, pdriver);
 }
 
 static void __exit radeon_exit(void)
 {
+	radeon_kfd_fini();
 	drm_pci_exit(driver, pdriver);
 	radeon_unregister_atpx_handler();
 }
diff --git a/drivers/gpu/drm/radeon/radeon_kfd.c b/drivers/gpu/drm/radeon/radeon_kfd.c
new file mode 100644
index 000000000000..065d02068ec3
--- /dev/null
+++ b/drivers/gpu/drm/radeon/radeon_kfd.c
@@ -0,0 +1,563 @@
+/*
+ * Copyright 2014 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+#include <linux/module.h>
+#include <linux/fdtable.h>
+#include <linux/uaccess.h>
+#include <drm/drmP.h>
+#include "radeon.h"
+#include "cikd.h"
+#include "cik_reg.h"
+#include "radeon_kfd.h"
+
+#define CIK_PIPE_PER_MEC	(4)
+
+struct kgd_mem {
+	struct radeon_sa_bo *sa_bo;
+	uint64_t gpu_addr;
+	void *ptr;
+};
+
+static int init_sa_manager(struct kgd_dev *kgd, unsigned int size);
+static void fini_sa_manager(struct kgd_dev *kgd);
+
+static int allocate_mem(struct kgd_dev *kgd, size_t size, size_t alignment,
+		enum kgd_memory_pool pool, struct kgd_mem **mem);
+
+static void free_mem(struct kgd_dev *kgd, struct kgd_mem *mem);
+
+static uint64_t get_vmem_size(struct kgd_dev *kgd);
+static uint64_t get_gpu_clock_counter(struct kgd_dev *kgd);
+
+static uint32_t get_max_engine_clock_in_mhz(struct kgd_dev *kgd);
+
+/*
+ * Register access functions
+ */
+
+static void kgd_program_sh_mem_settings(struct kgd_dev *kgd, uint32_t vmid,
+		uint32_t sh_mem_config,	uint32_t sh_mem_ape1_base,
+		uint32_t sh_mem_ape1_limit, uint32_t sh_mem_bases);
+
+static int kgd_set_pasid_vmid_mapping(struct kgd_dev *kgd, unsigned int pasid,
+					unsigned int vmid);
+
+static int kgd_init_memory(struct kgd_dev *kgd);
+
+static int kgd_init_pipeline(struct kgd_dev *kgd, uint32_t pipe_id,
+				uint32_t hpd_size, uint64_t hpd_gpu_addr);
+
+static int kgd_hqd_load(struct kgd_dev *kgd, void *mqd, uint32_t pipe_id,
+			uint32_t queue_id, uint32_t __user *wptr);
+
+static bool kgd_hqd_is_occupies(struct kgd_dev *kgd, uint64_t queue_address,
+				uint32_t pipe_id, uint32_t queue_id);
+
+static int kgd_hqd_destroy(struct kgd_dev *kgd, uint32_t reset_type,
+				unsigned int timeout, uint32_t pipe_id,
+				uint32_t queue_id);
+
+static const struct kfd2kgd_calls kfd2kgd = {
+	.init_sa_manager = init_sa_manager,
+	.fini_sa_manager = fini_sa_manager,
+	.allocate_mem = allocate_mem,
+	.free_mem = free_mem,
+	.get_vmem_size = get_vmem_size,
+	.get_gpu_clock_counter = get_gpu_clock_counter,
+	.get_max_engine_clock_in_mhz = get_max_engine_clock_in_mhz,
+	.program_sh_mem_settings = kgd_program_sh_mem_settings,
+	.set_pasid_vmid_mapping = kgd_set_pasid_vmid_mapping,
+	.init_memory = kgd_init_memory,
+	.init_pipeline = kgd_init_pipeline,
+	.hqd_load = kgd_hqd_load,
+	.hqd_is_occupies = kgd_hqd_is_occupies,
+	.hqd_destroy = kgd_hqd_destroy,
+};
+
+static const struct kgd2kfd_calls *kgd2kfd;
+
+bool radeon_kfd_init(void)
+{
+	bool (*kgd2kfd_init_p)(unsigned, const struct kfd2kgd_calls*,
+				const struct kgd2kfd_calls**);
+
+	kgd2kfd_init_p = symbol_request(kgd2kfd_init);
+
+	if (kgd2kfd_init_p == NULL)
+		return false;
+
+	if (!kgd2kfd_init_p(KFD_INTERFACE_VERSION, &kfd2kgd, &kgd2kfd)) {
+		symbol_put(kgd2kfd_init);
+		kgd2kfd = NULL;
+
+		return false;
+	}
+
+	return true;
+}
+
+void radeon_kfd_fini(void)
+{
+	if (kgd2kfd) {
+		kgd2kfd->exit();
+		symbol_put(kgd2kfd_init);
+	}
+}
+
+void radeon_kfd_device_probe(struct radeon_device *rdev)
+{
+	if (kgd2kfd)
+		rdev->kfd = kgd2kfd->probe((struct kgd_dev *)rdev, rdev->pdev);
+}
+
+void radeon_kfd_device_init(struct radeon_device *rdev)
+{
+	if (rdev->kfd) {
+		struct kgd2kfd_shared_resources gpu_resources = {
+			.compute_vmid_bitmap = 0xFF00,
+
+			.first_compute_pipe = 1,
+			.compute_pipe_count = 8 - 1,
+		};
+
+		radeon_doorbell_get_kfd_info(rdev,
+				&gpu_resources.doorbell_physical_address,
+				&gpu_resources.doorbell_aperture_size,
+				&gpu_resources.doorbell_start_offset);
+
+		kgd2kfd->device_init(rdev->kfd, &gpu_resources);
+	}
+}
+
+void radeon_kfd_device_fini(struct radeon_device *rdev)
+{
+	if (rdev->kfd) {
+		kgd2kfd->device_exit(rdev->kfd);
+		rdev->kfd = NULL;
+	}
+}
+
+void radeon_kfd_interrupt(struct radeon_device *rdev, const void *ih_ring_entry)
+{
+	if (rdev->kfd)
+		kgd2kfd->interrupt(rdev->kfd, ih_ring_entry);
+}
+
+void radeon_kfd_suspend(struct radeon_device *rdev)
+{
+	if (rdev->kfd)
+		kgd2kfd->suspend(rdev->kfd);
+}
+
+int radeon_kfd_resume(struct radeon_device *rdev)
+{
+	int r = 0;
+
+	if (rdev->kfd)
+		r = kgd2kfd->resume(rdev->kfd);
+
+	return r;
+}
+
+static u32 pool_to_domain(enum kgd_memory_pool p)
+{
+	switch (p) {
+	case KGD_POOL_FRAMEBUFFER: return RADEON_GEM_DOMAIN_VRAM;
+	default: return RADEON_GEM_DOMAIN_GTT;
+	}
+}
+
+static int init_sa_manager(struct kgd_dev *kgd, unsigned int size)
+{
+	struct radeon_device *rdev = (struct radeon_device *)kgd;
+	int r;
+
+	BUG_ON(kgd == NULL);
+
+	r = radeon_sa_bo_manager_init(rdev, &rdev->kfd_bo,
+				      size,
+				      RADEON_GPU_PAGE_SIZE,
+				      RADEON_GEM_DOMAIN_GTT,
+				      RADEON_GEM_GTT_WC);
+
+	if (r)
+		return r;
+
+	r = radeon_sa_bo_manager_start(rdev, &rdev->kfd_bo);
+	if (r)
+		radeon_sa_bo_manager_fini(rdev, &rdev->kfd_bo);
+
+	return r;
+}
+
+static void fini_sa_manager(struct kgd_dev *kgd)
+{
+	struct radeon_device *rdev = (struct radeon_device *)kgd;
+
+	BUG_ON(kgd == NULL);
+
+	radeon_sa_bo_manager_suspend(rdev, &rdev->kfd_bo);
+	radeon_sa_bo_manager_fini(rdev, &rdev->kfd_bo);
+}
+
+static int allocate_mem(struct kgd_dev *kgd, size_t size, size_t alignment,
+		enum kgd_memory_pool pool, struct kgd_mem **mem)
+{
+	struct radeon_device *rdev = (struct radeon_device *)kgd;
+	u32 domain;
+	int r;
+
+	BUG_ON(kgd == NULL);
+
+	domain = pool_to_domain(pool);
+	if (domain != RADEON_GEM_DOMAIN_GTT) {
+		dev_err(rdev->dev,
+			"Only allowed to allocate gart memory for kfd\n");
+		return -EINVAL;
+	}
+
+	*mem = kmalloc(sizeof(struct kgd_mem), GFP_KERNEL);
+	if ((*mem) == NULL)
+		return -ENOMEM;
+
+	r = radeon_sa_bo_new(rdev, &rdev->kfd_bo, &(*mem)->sa_bo, size,
+				alignment);
+	if (r) {
+		dev_err(rdev->dev, "failed to get memory for kfd (%d)\n", r);
+		return r;
+	}
+
+	(*mem)->ptr = radeon_sa_bo_cpu_addr((*mem)->sa_bo);
+	(*mem)->gpu_addr = radeon_sa_bo_gpu_addr((*mem)->sa_bo);
+
+	return 0;
+}
+
+static void free_mem(struct kgd_dev *kgd, struct kgd_mem *mem)
+{
+	struct radeon_device *rdev = (struct radeon_device *)kgd;
+
+	BUG_ON(kgd == NULL);
+
+	radeon_sa_bo_free(rdev, &mem->sa_bo, NULL);
+	kfree(mem);
+}
+
+static uint64_t get_vmem_size(struct kgd_dev *kgd)
+{
+	struct radeon_device *rdev = (struct radeon_device *)kgd;
+
+	BUG_ON(kgd == NULL);
+
+	return rdev->mc.real_vram_size;
+}
+
+static uint64_t get_gpu_clock_counter(struct kgd_dev *kgd)
+{
+	struct radeon_device *rdev = (struct radeon_device *)kgd;
+
+	return rdev->asic->get_gpu_clock_counter(rdev);
+}
+
+static uint32_t get_max_engine_clock_in_mhz(struct kgd_dev *kgd)
+{
+	struct radeon_device *rdev = (struct radeon_device *)kgd;
+
+	/* The sclk is in quantas of 10kHz */
+	return rdev->pm.dpm.dyn_state.max_clock_voltage_on_ac.sclk / 100;
+}
+
+static inline struct radeon_device *get_radeon_device(struct kgd_dev *kgd)
+{
+	return (struct radeon_device *)kgd;
+}
+
+static void write_register(struct kgd_dev *kgd, uint32_t offset, uint32_t value)
+{
+	struct radeon_device *rdev = get_radeon_device(kgd);
+
+	writel(value, (void __iomem *)(rdev->rmmio + offset));
+}
+
+static uint32_t read_register(struct kgd_dev *kgd, uint32_t offset)
+{
+	struct radeon_device *rdev = get_radeon_device(kgd);
+
+	return readl((void __iomem *)(rdev->rmmio + offset));
+}
+
+static void lock_srbm(struct kgd_dev *kgd, uint32_t mec, uint32_t pipe,
+			uint32_t queue, uint32_t vmid)
+{
+	struct radeon_device *rdev = get_radeon_device(kgd);
+	uint32_t value = PIPEID(pipe) | MEID(mec) | VMID(vmid) | QUEUEID(queue);
+
+	mutex_lock(&rdev->srbm_mutex);
+	write_register(kgd, SRBM_GFX_CNTL, value);
+}
+
+static void unlock_srbm(struct kgd_dev *kgd)
+{
+	struct radeon_device *rdev = get_radeon_device(kgd);
+
+	write_register(kgd, SRBM_GFX_CNTL, 0);
+	mutex_unlock(&rdev->srbm_mutex);
+}
+
+static void acquire_queue(struct kgd_dev *kgd, uint32_t pipe_id,
+				uint32_t queue_id)
+{
+	uint32_t mec = (++pipe_id / CIK_PIPE_PER_MEC) + 1;
+	uint32_t pipe = (pipe_id % CIK_PIPE_PER_MEC);
+
+	lock_srbm(kgd, mec, pipe, queue_id, 0);
+}
+
+static void release_queue(struct kgd_dev *kgd)
+{
+	unlock_srbm(kgd);
+}
+
+static void kgd_program_sh_mem_settings(struct kgd_dev *kgd, uint32_t vmid,
+					uint32_t sh_mem_config,
+					uint32_t sh_mem_ape1_base,
+					uint32_t sh_mem_ape1_limit,
+					uint32_t sh_mem_bases)
+{
+	lock_srbm(kgd, 0, 0, 0, vmid);
+
+	write_register(kgd, SH_MEM_CONFIG, sh_mem_config);
+	write_register(kgd, SH_MEM_APE1_BASE, sh_mem_ape1_base);
+	write_register(kgd, SH_MEM_APE1_LIMIT, sh_mem_ape1_limit);
+	write_register(kgd, SH_MEM_BASES, sh_mem_bases);
+
+	unlock_srbm(kgd);
+}
+
+static int kgd_set_pasid_vmid_mapping(struct kgd_dev *kgd, unsigned int pasid,
+					unsigned int vmid)
+{
+	/*
+	 * We have to assume that there is no outstanding mapping.
+	 * The ATC_VMID_PASID_MAPPING_UPDATE_STATUS bit could be 0
+	 * because a mapping is in progress or because a mapping finished and
+	 * the SW cleared it.
+	 * So the protocol is to always wait & clear.
+	 */
+	uint32_t pasid_mapping = (pasid == 0) ? 0 :
+				(uint32_t)pasid | ATC_VMID_PASID_MAPPING_VALID;
+
+	write_register(kgd, ATC_VMID0_PASID_MAPPING + vmid*sizeof(uint32_t),
+			pasid_mapping);
+
+	while (!(read_register(kgd, ATC_VMID_PASID_MAPPING_UPDATE_STATUS) &
+								(1U << vmid)))
+		cpu_relax();
+	write_register(kgd, ATC_VMID_PASID_MAPPING_UPDATE_STATUS, 1U << vmid);
+
+	return 0;
+}
+
+static int kgd_init_memory(struct kgd_dev *kgd)
+{
+	/*
+	 * Configure apertures:
+	 * LDS:         0x60000000'00000000 - 0x60000001'00000000 (4GB)
+	 * Scratch:     0x60000001'00000000 - 0x60000002'00000000 (4GB)
+	 * GPUVM:       0x60010000'00000000 - 0x60020000'00000000 (1TB)
+	 */
+	int i;
+	uint32_t sh_mem_bases = PRIVATE_BASE(0x6000) | SHARED_BASE(0x6000);
+
+	for (i = 8; i < 16; i++) {
+		uint32_t sh_mem_config;
+
+		lock_srbm(kgd, 0, 0, 0, i);
+
+		sh_mem_config = ALIGNMENT_MODE(SH_MEM_ALIGNMENT_MODE_UNALIGNED);
+		sh_mem_config |= DEFAULT_MTYPE(MTYPE_NONCACHED);
+
+		write_register(kgd, SH_MEM_CONFIG, sh_mem_config);
+
+		write_register(kgd, SH_MEM_BASES, sh_mem_bases);
+
+		/* Scratch aperture is not supported for now. */
+		write_register(kgd, SH_STATIC_MEM_CONFIG, 0);
+
+		/* APE1 disabled for now. */
+		write_register(kgd, SH_MEM_APE1_BASE, 1);
+		write_register(kgd, SH_MEM_APE1_LIMIT, 0);
+
+		unlock_srbm(kgd);
+	}
+
+	return 0;
+}
+
+static int kgd_init_pipeline(struct kgd_dev *kgd, uint32_t pipe_id,
+				uint32_t hpd_size, uint64_t hpd_gpu_addr)
+{
+	uint32_t mec = (++pipe_id / CIK_PIPE_PER_MEC) + 1;
+	uint32_t pipe = (pipe_id % CIK_PIPE_PER_MEC);
+
+	lock_srbm(kgd, mec, pipe, 0, 0);
+	write_register(kgd, CP_HPD_EOP_BASE_ADDR,
+			lower_32_bits(hpd_gpu_addr >> 8));
+	write_register(kgd, CP_HPD_EOP_BASE_ADDR_HI,
+			upper_32_bits(hpd_gpu_addr >> 8));
+	write_register(kgd, CP_HPD_EOP_VMID, 0);
+	write_register(kgd, CP_HPD_EOP_CONTROL, hpd_size);
+	unlock_srbm(kgd);
+
+	return 0;
+}
+
+static inline struct cik_mqd *get_mqd(void *mqd)
+{
+	return (struct cik_mqd *)mqd;
+}
+
+static int kgd_hqd_load(struct kgd_dev *kgd, void *mqd, uint32_t pipe_id,
+			uint32_t queue_id, uint32_t __user *wptr)
+{
+	uint32_t wptr_shadow, is_wptr_shadow_valid;
+	struct cik_mqd *m;
+
+	m = get_mqd(mqd);
+
+	is_wptr_shadow_valid = !get_user(wptr_shadow, wptr);
+
+	acquire_queue(kgd, pipe_id, queue_id);
+	write_register(kgd, CP_MQD_BASE_ADDR, m->cp_mqd_base_addr_lo);
+	write_register(kgd, CP_MQD_BASE_ADDR_HI, m->cp_mqd_base_addr_hi);
+	write_register(kgd, CP_MQD_CONTROL, m->cp_mqd_control);
+
+	write_register(kgd, CP_HQD_PQ_BASE, m->cp_hqd_pq_base_lo);
+	write_register(kgd, CP_HQD_PQ_BASE_HI, m->cp_hqd_pq_base_hi);
+	write_register(kgd, CP_HQD_PQ_CONTROL, m->cp_hqd_pq_control);
+
+	write_register(kgd, CP_HQD_IB_CONTROL, m->cp_hqd_ib_control);
+	write_register(kgd, CP_HQD_IB_BASE_ADDR, m->cp_hqd_ib_base_addr_lo);
+	write_register(kgd, CP_HQD_IB_BASE_ADDR_HI, m->cp_hqd_ib_base_addr_hi);
+
+	write_register(kgd, CP_HQD_IB_RPTR, m->cp_hqd_ib_rptr);
+
+	write_register(kgd, CP_HQD_PERSISTENT_STATE,
+			m->cp_hqd_persistent_state);
+	write_register(kgd, CP_HQD_SEMA_CMD, m->cp_hqd_sema_cmd);
+	write_register(kgd, CP_HQD_MSG_TYPE, m->cp_hqd_msg_type);
+
+	write_register(kgd, CP_HQD_ATOMIC0_PREOP_LO,
+			m->cp_hqd_atomic0_preop_lo);
+
+	write_register(kgd, CP_HQD_ATOMIC0_PREOP_HI,
+			m->cp_hqd_atomic0_preop_hi);
+
+	write_register(kgd, CP_HQD_ATOMIC1_PREOP_LO,
+			m->cp_hqd_atomic1_preop_lo);
+
+	write_register(kgd, CP_HQD_ATOMIC1_PREOP_HI,
+			m->cp_hqd_atomic1_preop_hi);
+
+	write_register(kgd, CP_HQD_PQ_RPTR_REPORT_ADDR,
+			m->cp_hqd_pq_rptr_report_addr_lo);
+
+	write_register(kgd, CP_HQD_PQ_RPTR_REPORT_ADDR_HI,
+			m->cp_hqd_pq_rptr_report_addr_hi);
+
+	write_register(kgd, CP_HQD_PQ_RPTR, m->cp_hqd_pq_rptr);
+
+	write_register(kgd, CP_HQD_PQ_WPTR_POLL_ADDR,
+			m->cp_hqd_pq_wptr_poll_addr_lo);
+
+	write_register(kgd, CP_HQD_PQ_WPTR_POLL_ADDR_HI,
+			m->cp_hqd_pq_wptr_poll_addr_hi);
+
+	write_register(kgd, CP_HQD_PQ_DOORBELL_CONTROL,
+			m->cp_hqd_pq_doorbell_control);
+
+	write_register(kgd, CP_HQD_VMID, m->cp_hqd_vmid);
+
+	write_register(kgd, CP_HQD_QUANTUM, m->cp_hqd_quantum);
+
+	write_register(kgd, CP_HQD_PIPE_PRIORITY, m->cp_hqd_pipe_priority);
+	write_register(kgd, CP_HQD_QUEUE_PRIORITY, m->cp_hqd_queue_priority);
+
+	write_register(kgd, CP_HQD_IQ_RPTR, m->cp_hqd_iq_rptr);
+
+	if (is_wptr_shadow_valid)
+		write_register(kgd, CP_HQD_PQ_WPTR, wptr_shadow);
+
+	write_register(kgd, CP_HQD_ACTIVE, m->cp_hqd_active);
+	release_queue(kgd);
+
+	return 0;
+}
+
+static bool kgd_hqd_is_occupies(struct kgd_dev *kgd, uint64_t queue_address,
+				uint32_t pipe_id, uint32_t queue_id)
+{
+	uint32_t act;
+	bool retval = false;
+	uint32_t low, high;
+
+	acquire_queue(kgd, pipe_id, queue_id);
+	act = read_register(kgd, CP_HQD_ACTIVE);
+	if (act) {
+		low = lower_32_bits(queue_address >> 8);
+		high = upper_32_bits(queue_address >> 8);
+
+		if (low == read_register(kgd, CP_HQD_PQ_BASE) &&
+				high == read_register(kgd, CP_HQD_PQ_BASE_HI))
+			retval = true;
+	}
+	release_queue(kgd);
+	return retval;
+}
+
+static int kgd_hqd_destroy(struct kgd_dev *kgd, uint32_t reset_type,
+				unsigned int timeout, uint32_t pipe_id,
+				uint32_t queue_id)
+{
+	uint32_t temp;
+
+	acquire_queue(kgd, pipe_id, queue_id);
+	write_register(kgd, CP_HQD_PQ_DOORBELL_CONTROL, 0);
+
+	write_register(kgd, CP_HQD_DEQUEUE_REQUEST, reset_type);
+
+	while (true) {
+		temp = read_register(kgd, CP_HQD_ACTIVE);
+		if (temp & 0x1)
+			break;
+		if (timeout == 0) {
+			pr_err("kfd: cp queue preemption time out (%dms)\n",
+				temp);
+			return -ETIME;
+		}
+		msleep(20);
+		timeout -= 20;
+	}
+
+	release_queue(kgd);
+	return 0;
+}
diff --git a/drivers/gpu/drm/radeon/radeon_kfd.h b/drivers/gpu/drm/radeon/radeon_kfd.h
new file mode 100644
index 000000000000..f90e161ca507
--- /dev/null
+++ b/drivers/gpu/drm/radeon/radeon_kfd.h
@@ -0,0 +1,47 @@
+/*
+ * Copyright 2014 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+/*
+ * radeon_kfd.h defines the private interface between the
+ * AMD kernel graphics drivers and the AMD KFD.
+ */
+
+#ifndef RADEON_KFD_H_INCLUDED
+#define RADEON_KFD_H_INCLUDED
+
+#include <linux/types.h>
+#include "../amd/include/kgd_kfd_interface.h"
+
+struct radeon_device;
+
+bool radeon_kfd_init(void);
+void radeon_kfd_fini(void);
+
+void radeon_kfd_suspend(struct radeon_device *rdev);
+int radeon_kfd_resume(struct radeon_device *rdev);
+void radeon_kfd_interrupt(struct radeon_device *rdev,
+			const void *ih_ring_entry);
+void radeon_kfd_device_probe(struct radeon_device *rdev);
+void radeon_kfd_device_init(struct radeon_device *rdev);
+void radeon_kfd_device_fini(struct radeon_device *rdev);
+
+#endif /* RADEON_KFD_H_INCLUDED */
diff --git a/drivers/gpu/drm/radeon/radeon_kms.c b/drivers/gpu/drm/radeon/radeon_kms.c
index 8309b11e674d..6eb561d33eba 100644
--- a/drivers/gpu/drm/radeon/radeon_kms.c
+++ b/drivers/gpu/drm/radeon/radeon_kms.c
@@ -34,6 +34,8 @@
 #include <linux/slab.h>
 #include <linux/pm_runtime.h>
 
+#include "radeon_kfd.h"
+
 #if defined(CONFIG_VGA_SWITCHEROO)
 bool radeon_has_atpx(void);
 #else
@@ -63,6 +65,8 @@ int radeon_driver_unload_kms(struct drm_device *dev)
 
 	pm_runtime_get_sync(dev->dev);
 
+	radeon_kfd_device_fini(rdev);
+
 	radeon_acpi_fini(rdev);
 	
 	radeon_modeset_fini(rdev);
@@ -142,6 +146,9 @@ int radeon_driver_load_kms(struct drm_device *dev, unsigned long flags)
 				"Error during ACPI methods call\n");
 	}
 
+	radeon_kfd_device_probe(rdev);
+	radeon_kfd_device_init(rdev);
+
 	if (radeon_is_px(dev)) {
 		pm_runtime_use_autosuspend(dev->dev);
 		pm_runtime_set_autosuspend_delay(dev->dev, 5000);