drm/amdgpu: Initialize compute sdma and memory from kgd

v2: add missing MTYPE_NONCACHED enum
Signed-off-by: NBen Goz <ben.goz@amd.com>
Acked-by: NOded Gabbay <oded.gabbay@amd.com>
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/amd/amdgpu/cikd.h

@@ -552,4 +552,10 @@
 #define VCE_CMD_IB_AUTO		0x00000005
 #define VCE_CMD_SEMAPHORE	0x00000006
 
+/* valid for both DEFAULT_MTYPE and APE1_MTYPE */
+enum {
+	MTYPE_CACHED = 0,
+	MTYPE_NONCACHED = 3
+};
+
 #endif

drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c

@@ -2009,6 +2009,46 @@ static void gfx_v7_0_setup_rb(struct amdgpu_device *adev,
 	mutex_unlock(&adev->grbm_idx_mutex);
 }
 
+/**
+ * gmc_v7_0_init_compute_vmid - gart enable
+ *
+ * @rdev: amdgpu_device pointer
+ *
+ * Initialize compute vmid sh_mem registers
+ *
+ */
+#define DEFAULT_SH_MEM_BASES	(0x6000)
+#define FIRST_COMPUTE_VMID	(8)
+#define LAST_COMPUTE_VMID	(16)
+static void gmc_v7_0_init_compute_vmid(struct amdgpu_device *adev)
+{
+	int i;
+	uint32_t sh_mem_config;
+	uint32_t sh_mem_bases;
+
+	/*
+	 * Configure apertures:
+	 * LDS:         0x60000000'00000000 - 0x60000001'00000000 (4GB)
+	 * Scratch:     0x60000001'00000000 - 0x60000002'00000000 (4GB)
+	 * GPUVM:       0x60010000'00000000 - 0x60020000'00000000 (1TB)
+	*/
+	sh_mem_bases = DEFAULT_SH_MEM_BASES | (DEFAULT_SH_MEM_BASES << 16);
+	sh_mem_config = SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
+			SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT;
+	sh_mem_config |= MTYPE_NONCACHED << SH_MEM_CONFIG__DEFAULT_MTYPE__SHIFT;
+	mutex_lock(&adev->srbm_mutex);
+	for (i = FIRST_COMPUTE_VMID; i < LAST_COMPUTE_VMID; i++) {
+		cik_srbm_select(adev, 0, 0, 0, i);
+		/* CP and shaders */
+		WREG32(mmSH_MEM_CONFIG, sh_mem_config);
+		WREG32(mmSH_MEM_APE1_BASE, 1);
+		WREG32(mmSH_MEM_APE1_LIMIT, 0);
+		WREG32(mmSH_MEM_BASES, sh_mem_bases);
+	}
+	cik_srbm_select(adev, 0, 0, 0, 0);
+	mutex_unlock(&adev->srbm_mutex);
+}
+
 /**
  * gfx_v7_0_gpu_init - setup the 3D engine
  *
@@ -2230,6 +2270,8 @@ static void gfx_v7_0_gpu_init(struct amdgpu_device *adev)
 	cik_srbm_select(adev, 0, 0, 0, 0);
 	mutex_unlock(&adev->srbm_mutex);
 
+	gmc_v7_0_init_compute_vmid(adev);
+
 	WREG32(mmSX_DEBUG_1, 0x20);
 
 	WREG32(mmTA_CNTL_AUX, 0x00010000);

drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c

@@ -1894,6 +1894,51 @@ static void gfx_v8_0_setup_rb(struct amdgpu_device *adev,
 	mutex_unlock(&adev->grbm_idx_mutex);
 }
 
+/**
+ * gmc_v8_0_init_compute_vmid - gart enable
+ *
+ * @rdev: amdgpu_device pointer
+ *
+ * Initialize compute vmid sh_mem registers
+ *
+ */
+#define DEFAULT_SH_MEM_BASES	(0x6000)
+#define FIRST_COMPUTE_VMID	(8)
+#define LAST_COMPUTE_VMID	(16)
+static void gmc_v8_0_init_compute_vmid(struct amdgpu_device *adev)
+{
+	int i;
+	uint32_t sh_mem_config;
+	uint32_t sh_mem_bases;
+
+	/*
+	 * Configure apertures:
+	 * LDS:         0x60000000'00000000 - 0x60000001'00000000 (4GB)
+	 * Scratch:     0x60000001'00000000 - 0x60000002'00000000 (4GB)
+	 * GPUVM:       0x60010000'00000000 - 0x60020000'00000000 (1TB)
+	 */
+	sh_mem_bases = DEFAULT_SH_MEM_BASES | (DEFAULT_SH_MEM_BASES << 16);
+
+	sh_mem_config = SH_MEM_ADDRESS_MODE_HSA64 <<
+			SH_MEM_CONFIG__ADDRESS_MODE__SHIFT |
+			SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
+			SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT |
+			MTYPE_CC << SH_MEM_CONFIG__DEFAULT_MTYPE__SHIFT |
+			SH_MEM_CONFIG__PRIVATE_ATC_MASK;
+
+	mutex_lock(&adev->srbm_mutex);
+	for (i = FIRST_COMPUTE_VMID; i < LAST_COMPUTE_VMID; i++) {
+		vi_srbm_select(adev, 0, 0, 0, i);
+		/* CP and shaders */
+		WREG32(mmSH_MEM_CONFIG, sh_mem_config);
+		WREG32(mmSH_MEM_APE1_BASE, 1);
+		WREG32(mmSH_MEM_APE1_LIMIT, 0);
+		WREG32(mmSH_MEM_BASES, sh_mem_bases);
+	}
+	vi_srbm_select(adev, 0, 0, 0, 0);
+	mutex_unlock(&adev->srbm_mutex);
+}
+
 static void gfx_v8_0_gpu_init(struct amdgpu_device *adev)
 {
 	u32 gb_addr_config;
@@ -2113,6 +2158,8 @@ static void gfx_v8_0_gpu_init(struct amdgpu_device *adev)
 	vi_srbm_select(adev, 0, 0, 0, 0);
 	mutex_unlock(&adev->srbm_mutex);
 
+	gmc_v8_0_init_compute_vmid(adev);
+
 	mutex_lock(&adev->grbm_idx_mutex);
 	/*
 	 * making sure that the following register writes will be broadcasted

drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c

@@ -438,6 +438,31 @@ static void sdma_v3_0_rlc_stop(struct amdgpu_device *adev)
 	/* XXX todo */
 }
 
+/**
+ * sdma_v3_0_ctx_switch_enable - stop the async dma engines context switch
+ *
+ * @adev: amdgpu_device pointer
+ * @enable: enable/disable the DMA MEs context switch.
+ *
+ * Halt or unhalt the async dma engines context switch (VI).
+ */
+static void sdma_v3_0_ctx_switch_enable(struct amdgpu_device *adev, bool enable)
+{
+	u32 f32_cntl;
+	int i;
+
+	for (i = 0; i < SDMA_MAX_INSTANCE; i++) {
+		f32_cntl = RREG32(mmSDMA0_CNTL + sdma_offsets[i]);
+		if (enable)
+			f32_cntl = REG_SET_FIELD(f32_cntl, SDMA0_CNTL,
+					AUTO_CTXSW_ENABLE, 1);
+		else
+			f32_cntl = REG_SET_FIELD(f32_cntl, SDMA0_CNTL,
+					AUTO_CTXSW_ENABLE, 0);
+		WREG32(mmSDMA0_CNTL + sdma_offsets[i], f32_cntl);
+	}
+}
+
 /**
  * sdma_v3_0_enable - stop the async dma engines
  *
@@ -648,6 +673,8 @@ static int sdma_v3_0_start(struct amdgpu_device *adev)
 
 	/* unhalt the MEs */
 	sdma_v3_0_enable(adev, true);
+	/* enable sdma ring preemption */
+	sdma_v3_0_ctx_switch_enable(adev, true);
 
 	/* start the gfx rings and rlc compute queues */
 	r = sdma_v3_0_gfx_resume(adev);
@@ -1079,6 +1106,7 @@ static int sdma_v3_0_hw_fini(void *handle)
 {
 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 
+	sdma_v3_0_ctx_switch_enable(adev, false);
 	sdma_v3_0_enable(adev, false);
 
 	return 0;