Merge branch 'for-4.5/for-jens' of git://git.kernel.org/pub/scm/linux/kernel/git...
[cascardo/linux.git] / drivers / gpu / drm / amd / amdgpu / vce_v3_0.c
index 370c6c9..e99af81 100644 (file)
@@ -103,6 +103,108 @@ static void vce_v3_0_ring_set_wptr(struct amdgpu_ring *ring)
                WREG32(mmVCE_RB_WPTR2, ring->wptr);
 }
 
+static void vce_v3_0_override_vce_clock_gating(struct amdgpu_device *adev, bool override)
+{
+       u32 tmp, data;
+
+       tmp = data = RREG32(mmVCE_RB_ARB_CTRL);
+       if (override)
+               data |= VCE_RB_ARB_CTRL__VCE_CGTT_OVERRIDE_MASK;
+       else
+               data &= ~VCE_RB_ARB_CTRL__VCE_CGTT_OVERRIDE_MASK;
+
+       if (tmp != data)
+               WREG32(mmVCE_RB_ARB_CTRL, data);
+}
+
+static void vce_v3_0_set_vce_sw_clock_gating(struct amdgpu_device *adev,
+                                            bool gated)
+{
+       u32 tmp, data;
+       /* Set Override to disable Clock Gating */
+       vce_v3_0_override_vce_clock_gating(adev, true);
+
+       if (!gated) {
+               /* Force CLOCK ON for VCE_CLOCK_GATING_B,
+                * {*_FORCE_ON, *_FORCE_OFF} = {1, 0}
+                * VREG can be FORCE ON or set to Dynamic, but can't be OFF
+                */
+               tmp = data = RREG32(mmVCE_CLOCK_GATING_B);
+               data |= 0x1ff;
+               data &= ~0xef0000;
+               if (tmp != data)
+                       WREG32(mmVCE_CLOCK_GATING_B, data);
+
+               /* Force CLOCK ON for VCE_UENC_CLOCK_GATING,
+                * {*_FORCE_ON, *_FORCE_OFF} = {1, 0}
+                */
+               tmp = data = RREG32(mmVCE_UENC_CLOCK_GATING);
+               data |= 0x3ff000;
+               data &= ~0xffc00000;
+               if (tmp != data)
+                       WREG32(mmVCE_UENC_CLOCK_GATING, data);
+
+               /* set VCE_UENC_CLOCK_GATING_2 */
+               tmp = data = RREG32(mmVCE_UENC_CLOCK_GATING_2);
+               data |= 0x2;
+               data &= ~0x2;
+               if (tmp != data)
+                       WREG32(mmVCE_UENC_CLOCK_GATING_2, data);
+
+               /* Force CLOCK ON for VCE_UENC_REG_CLOCK_GATING */
+               tmp = data = RREG32(mmVCE_UENC_REG_CLOCK_GATING);
+               data |= 0x37f;
+               if (tmp != data)
+                       WREG32(mmVCE_UENC_REG_CLOCK_GATING, data);
+
+               /* Force VCE_UENC_DMA_DCLK_CTRL Clock ON */
+               tmp = data = RREG32(mmVCE_UENC_DMA_DCLK_CTRL);
+               data |= VCE_UENC_DMA_DCLK_CTRL__WRDMCLK_FORCEON_MASK |
+                               VCE_UENC_DMA_DCLK_CTRL__RDDMCLK_FORCEON_MASK |
+                               VCE_UENC_DMA_DCLK_CTRL__REGCLK_FORCEON_MASK  |
+                               0x8;
+               if (tmp != data)
+                       WREG32(mmVCE_UENC_DMA_DCLK_CTRL, data);
+       } else {
+               /* Force CLOCK OFF for VCE_CLOCK_GATING_B,
+                * {*, *_FORCE_OFF} = {*, 1}
+                * set VREG to Dynamic, as it can't be OFF
+                */
+               tmp = data = RREG32(mmVCE_CLOCK_GATING_B);
+               data &= ~0x80010;
+               data |= 0xe70008;
+               if (tmp != data)
+                       WREG32(mmVCE_CLOCK_GATING_B, data);
+               /* Force CLOCK OFF for VCE_UENC_CLOCK_GATING,
+                * Force ClOCK OFF takes precedent over Force CLOCK ON setting.
+                * {*_FORCE_ON, *_FORCE_OFF} = {*, 1}
+                */
+               tmp = data = RREG32(mmVCE_UENC_CLOCK_GATING);
+               data |= 0xffc00000;
+               if (tmp != data)
+                       WREG32(mmVCE_UENC_CLOCK_GATING, data);
+               /* Set VCE_UENC_CLOCK_GATING_2 */
+               tmp = data = RREG32(mmVCE_UENC_CLOCK_GATING_2);
+               data |= 0x10000;
+               if (tmp != data)
+                       WREG32(mmVCE_UENC_CLOCK_GATING_2, data);
+               /* Set VCE_UENC_REG_CLOCK_GATING to dynamic */
+               tmp = data = RREG32(mmVCE_UENC_REG_CLOCK_GATING);
+               data &= ~0xffc00000;
+               if (tmp != data)
+                       WREG32(mmVCE_UENC_REG_CLOCK_GATING, data);
+               /* Set VCE_UENC_DMA_DCLK_CTRL CG always in dynamic mode */
+               tmp = data = RREG32(mmVCE_UENC_DMA_DCLK_CTRL);
+               data &= ~(VCE_UENC_DMA_DCLK_CTRL__WRDMCLK_FORCEON_MASK |
+                               VCE_UENC_DMA_DCLK_CTRL__RDDMCLK_FORCEON_MASK |
+                               VCE_UENC_DMA_DCLK_CTRL__REGCLK_FORCEON_MASK  |
+                               0x8);
+               if (tmp != data)
+                       WREG32(mmVCE_UENC_DMA_DCLK_CTRL, data);
+       }
+       vce_v3_0_override_vce_clock_gating(adev, false);
+}
+
 /**
  * vce_v3_0_start - start VCE block
  *
@@ -121,7 +223,7 @@ static int vce_v3_0_start(struct amdgpu_device *adev)
                if (adev->vce.harvest_config & (1 << idx))
                        continue;
 
-               if(idx == 0)
+               if (idx == 0)
                        WREG32_P(mmGRBM_GFX_INDEX, 0,
                                ~GRBM_GFX_INDEX__VCE_INSTANCE_MASK);
                else
@@ -174,6 +276,10 @@ static int vce_v3_0_start(struct amdgpu_device *adev)
                /* clear BUSY flag */
                WREG32_P(mmVCE_STATUS, 0, ~1);
 
+               /* Set Clock-Gating off */
+               if (adev->cg_flags & AMDGPU_CG_SUPPORT_VCE_MGCG)
+                       vce_v3_0_set_vce_sw_clock_gating(adev, false);
+
                if (r) {
                        DRM_ERROR("VCE not responding, giving up!!!\n");
                        mutex_unlock(&adev->grbm_idx_mutex);
@@ -208,14 +314,11 @@ static int vce_v3_0_start(struct amdgpu_device *adev)
 static unsigned vce_v3_0_get_harvest_config(struct amdgpu_device *adev)
 {
        u32 tmp;
-       unsigned ret;
 
        /* Fiji, Stoney are single pipe */
        if ((adev->asic_type == CHIP_FIJI) ||
-           (adev->asic_type == CHIP_STONEY)){
-               ret = AMDGPU_VCE_HARVEST_VCE1;
-               return ret;
-       }
+           (adev->asic_type == CHIP_STONEY))
+               return AMDGPU_VCE_HARVEST_VCE1;
 
        /* Tonga and CZ are dual or single pipe */
        if (adev->flags & AMD_IS_APU)
@@ -229,19 +332,14 @@ static unsigned vce_v3_0_get_harvest_config(struct amdgpu_device *adev)
 
        switch (tmp) {
        case 1:
-               ret = AMDGPU_VCE_HARVEST_VCE0;
-               break;
+               return AMDGPU_VCE_HARVEST_VCE0;
        case 2:
-               ret = AMDGPU_VCE_HARVEST_VCE1;
-               break;
+               return AMDGPU_VCE_HARVEST_VCE1;
        case 3:
-               ret = AMDGPU_VCE_HARVEST_VCE0 | AMDGPU_VCE_HARVEST_VCE1;
-               break;
+               return AMDGPU_VCE_HARVEST_VCE0 | AMDGPU_VCE_HARVEST_VCE1;
        default:
-               ret = 0;
+               return 0;
        }
-
-       return ret;
 }
 
 static int vce_v3_0_early_init(void *handle)
@@ -316,28 +414,22 @@ static int vce_v3_0_sw_fini(void *handle)
 
 static int vce_v3_0_hw_init(void *handle)
 {
-       struct amdgpu_ring *ring;
-       int r;
+       int r, i;
        struct amdgpu_device *adev = (struct amdgpu_device *)handle;
 
        r = vce_v3_0_start(adev);
        if (r)
                return r;
 
-       ring = &adev->vce.ring[0];
-       ring->ready = true;
-       r = amdgpu_ring_test_ring(ring);
-       if (r) {
-               ring->ready = false;
-               return r;
-       }
+       adev->vce.ring[0].ready = false;
+       adev->vce.ring[1].ready = false;
 
-       ring = &adev->vce.ring[1];
-       ring->ready = true;
-       r = amdgpu_ring_test_ring(ring);
-       if (r) {
-               ring->ready = false;
-               return r;
+       for (i = 0; i < 2; i++) {
+               r = amdgpu_ring_test_ring(&adev->vce.ring[i]);
+               if (r)
+                       return r;
+               else
+                       adev->vce.ring[i].ready = true;
        }
 
        DRM_INFO("VCE initialized successfully.\n");
@@ -437,17 +529,9 @@ static bool vce_v3_0_is_idle(void *handle)
 {
        struct amdgpu_device *adev = (struct amdgpu_device *)handle;
        u32 mask = 0;
-       int idx;
-
-       for (idx = 0; idx < 2; ++idx) {
-               if (adev->vce.harvest_config & (1 << idx))
-                       continue;
 
-               if (idx == 0)
-                       mask |= SRBM_STATUS2__VCE0_BUSY_MASK;
-               else
-                       mask |= SRBM_STATUS2__VCE1_BUSY_MASK;
-       }
+       mask |= (adev->vce.harvest_config & AMDGPU_VCE_HARVEST_VCE0) ? 0 : SRBM_STATUS2__VCE0_BUSY_MASK;
+       mask |= (adev->vce.harvest_config & AMDGPU_VCE_HARVEST_VCE1) ? 0 : SRBM_STATUS2__VCE1_BUSY_MASK;
 
        return !(RREG32(mmSRBM_STATUS2) & mask);
 }
@@ -456,23 +540,11 @@ static int vce_v3_0_wait_for_idle(void *handle)
 {
        unsigned i;
        struct amdgpu_device *adev = (struct amdgpu_device *)handle;
-       u32 mask = 0;
-       int idx;
-
-       for (idx = 0; idx < 2; ++idx) {
-               if (adev->vce.harvest_config & (1 << idx))
-                       continue;
-
-               if (idx == 0)
-                       mask |= SRBM_STATUS2__VCE0_BUSY_MASK;
-               else
-                       mask |= SRBM_STATUS2__VCE1_BUSY_MASK;
-       }
 
-       for (i = 0; i < adev->usec_timeout; i++) {
-               if (!(RREG32(mmSRBM_STATUS2) & mask))
+       for (i = 0; i < adev->usec_timeout; i++)
+               if (vce_v3_0_is_idle(handle))
                        return 0;
-       }
+
        return -ETIMEDOUT;
 }
 
@@ -480,17 +552,10 @@ static int vce_v3_0_soft_reset(void *handle)
 {
        struct amdgpu_device *adev = (struct amdgpu_device *)handle;
        u32 mask = 0;
-       int idx;
 
-       for (idx = 0; idx < 2; ++idx) {
-               if (adev->vce.harvest_config & (1 << idx))
-                       continue;
+       mask |= (adev->vce.harvest_config & AMDGPU_VCE_HARVEST_VCE0) ? 0 : SRBM_SOFT_RESET__SOFT_RESET_VCE0_MASK;
+       mask |= (adev->vce.harvest_config & AMDGPU_VCE_HARVEST_VCE1) ? 0 : SRBM_SOFT_RESET__SOFT_RESET_VCE1_MASK;
 
-               if (idx == 0)
-                       mask |= SRBM_SOFT_RESET__SOFT_RESET_VCE0_MASK;
-               else
-                       mask |= SRBM_SOFT_RESET__SOFT_RESET_VCE1_MASK;
-       }
        WREG32_P(mmSRBM_SOFT_RESET, mask,
                 ~(SRBM_SOFT_RESET__SOFT_RESET_VCE0_MASK |
                   SRBM_SOFT_RESET__SOFT_RESET_VCE1_MASK));
@@ -592,10 +657,8 @@ static int vce_v3_0_process_interrupt(struct amdgpu_device *adev,
 
        switch (entry->src_data) {
        case 0:
-               amdgpu_fence_process(&adev->vce.ring[0]);
-               break;
        case 1:
-               amdgpu_fence_process(&adev->vce.ring[1]);
+               amdgpu_fence_process(&adev->vce.ring[entry->src_data]);
                break;
        default:
                DRM_ERROR("Unhandled interrupt: %d %d\n",
@@ -609,6 +672,47 @@ static int vce_v3_0_process_interrupt(struct amdgpu_device *adev,
 static int vce_v3_0_set_clockgating_state(void *handle,
                                          enum amd_clockgating_state state)
 {
+       struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+       bool enable = (state == AMD_CG_STATE_GATE) ? true : false;
+       int i;
+
+       if (!(adev->cg_flags & AMDGPU_CG_SUPPORT_VCE_MGCG))
+               return 0;
+
+       mutex_lock(&adev->grbm_idx_mutex);
+       for (i = 0; i < 2; i++) {
+               /* Program VCE Instance 0 or 1 if not harvested */
+               if (adev->vce.harvest_config & (1 << i))
+                       continue;
+
+               if (i == 0)
+                       WREG32_P(mmGRBM_GFX_INDEX, 0,
+                                       ~GRBM_GFX_INDEX__VCE_INSTANCE_MASK);
+               else
+                       WREG32_P(mmGRBM_GFX_INDEX,
+                                       GRBM_GFX_INDEX__VCE_INSTANCE_MASK,
+                                       ~GRBM_GFX_INDEX__VCE_INSTANCE_MASK);
+
+               if (enable) {
+                       /* initialize VCE_CLOCK_GATING_A: Clock ON/OFF delay */
+                       uint32_t data = RREG32(mmVCE_CLOCK_GATING_A);
+                       data &= ~(0xf | 0xff0);
+                       data |= ((0x0 << 0) | (0x04 << 4));
+                       WREG32(mmVCE_CLOCK_GATING_A, data);
+
+                       /* initialize VCE_UENC_CLOCK_GATING: Clock ON/OFF delay */
+                       data = RREG32(mmVCE_UENC_CLOCK_GATING);
+                       data &= ~(0xf | 0xff0);
+                       data |= ((0x0 << 0) | (0x04 << 4));
+                       WREG32(mmVCE_UENC_CLOCK_GATING, data);
+               }
+
+               vce_v3_0_set_vce_sw_clock_gating(adev, enable);
+       }
+
+       WREG32_P(mmGRBM_GFX_INDEX, 0, ~GRBM_GFX_INDEX__VCE_INSTANCE_MASK);
+       mutex_unlock(&adev->grbm_idx_mutex);
+
        return 0;
 }