drm/amdgpu: add vcn enc ring type and functions
authorLeo Liu <leo.liu@amd.com>
Tue, 21 Feb 2017 15:36:15 +0000 (10:36 -0500)
committerAlex Deucher <alexander.deucher@amd.com>
Wed, 24 May 2017 21:41:41 +0000 (17:41 -0400)
Add the ring function callbacks for the encode rings.

Signed-off-by: Leo Liu <leo.liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h
drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h
drivers/gpu/drm/amd/amdgpu/vcn_v1_0.c

index 9bf3100f273e44a459f85fb4b88b19a56f3d88a2..a9223a8de8c290a8022c35046dbf338f3a397947 100644 (file)
@@ -48,7 +48,8 @@ enum amdgpu_ring_type {
        AMDGPU_RING_TYPE_VCE,
        AMDGPU_RING_TYPE_KIQ,
        AMDGPU_RING_TYPE_UVD_ENC,
-       AMDGPU_RING_TYPE_VCN_DEC
+       AMDGPU_RING_TYPE_VCN_DEC,
+       AMDGPU_RING_TYPE_VCN_ENC
 };
 
 struct amdgpu_device;
index 444fed503f4414b1fe046b36efa664db05547672..d50ba06578542e5a7df60dbc963a75edd1d1b672 100644 (file)
 #define VCN_DEC_CMD_PACKET_START       0x0000000a
 #define VCN_DEC_CMD_PACKET_END         0x0000000b
 
+#define VCN_ENC_CMD_NO_OP              0x00000000
+#define VCN_ENC_CMD_END                0x00000001
+#define VCN_ENC_CMD_IB                 0x00000002
+#define VCN_ENC_CMD_FENCE              0x00000003
+#define VCN_ENC_CMD_TRAP               0x00000004
+#define VCN_ENC_CMD_REG_WRITE          0x0000000b
+#define VCN_ENC_CMD_REG_WAIT           0x0000000c
+
 struct amdgpu_vcn {
        struct amdgpu_bo        *vcpu_bo;
        void                    *cpu_addr;
index b8f4e7713921f6fbf9c0ee483b069d493e4fc7bf..e15a81ffa2cae4f4618a424d2ce7c132755e18dd 100644 (file)
@@ -38,6 +38,7 @@
 static int vcn_v1_0_start(struct amdgpu_device *adev);
 static int vcn_v1_0_stop(struct amdgpu_device *adev);
 static void vcn_v1_0_set_dec_ring_funcs(struct amdgpu_device *adev);
+static void vcn_v1_0_set_enc_ring_funcs(struct amdgpu_device *adev);
 static void vcn_v1_0_set_irq_funcs(struct amdgpu_device *adev);
 
 /**
@@ -54,6 +55,7 @@ static int vcn_v1_0_early_init(void *handle)
        adev->vcn.num_enc_rings = 2;
 
        vcn_v1_0_set_dec_ring_funcs(adev);
+       vcn_v1_0_set_enc_ring_funcs(adev);
        vcn_v1_0_set_irq_funcs(adev);
 
        return 0;
@@ -688,6 +690,141 @@ static void vcn_v1_0_dec_ring_emit_vm_flush(struct amdgpu_ring *ring,
        vcn_v1_0_dec_vm_reg_wait(ring, data0, data1, mask);
 }
 
+/**
+ * vcn_v1_0_enc_ring_get_rptr - get enc read pointer
+ *
+ * @ring: amdgpu_ring pointer
+ *
+ * Returns the current hardware enc read pointer
+ */
+static uint64_t vcn_v1_0_enc_ring_get_rptr(struct amdgpu_ring *ring)
+{
+       struct amdgpu_device *adev = ring->adev;
+
+       if (ring == &adev->vcn.ring_enc[0])
+               return RREG32(SOC15_REG_OFFSET(UVD, 0, mmUVD_RB_RPTR));
+       else
+               return RREG32(SOC15_REG_OFFSET(UVD, 0, mmUVD_RB_RPTR2));
+}
+
+ /**
+ * vcn_v1_0_enc_ring_get_wptr - get enc write pointer
+ *
+ * @ring: amdgpu_ring pointer
+ *
+ * Returns the current hardware enc write pointer
+ */
+static uint64_t vcn_v1_0_enc_ring_get_wptr(struct amdgpu_ring *ring)
+{
+       struct amdgpu_device *adev = ring->adev;
+
+       if (ring == &adev->vcn.ring_enc[0])
+               return RREG32(SOC15_REG_OFFSET(UVD, 0, mmUVD_RB_WPTR));
+       else
+               return RREG32(SOC15_REG_OFFSET(UVD, 0, mmUVD_RB_WPTR2));
+}
+
+ /**
+ * vcn_v1_0_enc_ring_set_wptr - set enc write pointer
+ *
+ * @ring: amdgpu_ring pointer
+ *
+ * Commits the enc write pointer to the hardware
+ */
+static void vcn_v1_0_enc_ring_set_wptr(struct amdgpu_ring *ring)
+{
+       struct amdgpu_device *adev = ring->adev;
+
+       if (ring == &adev->vcn.ring_enc[0])
+               WREG32(SOC15_REG_OFFSET(UVD, 0, mmUVD_RB_WPTR),
+                       lower_32_bits(ring->wptr));
+       else
+               WREG32(SOC15_REG_OFFSET(UVD, 0, mmUVD_RB_WPTR2),
+                       lower_32_bits(ring->wptr));
+}
+
+/**
+ * vcn_v1_0_enc_ring_emit_fence - emit an enc fence & trap command
+ *
+ * @ring: amdgpu_ring pointer
+ * @fence: fence to emit
+ *
+ * Write enc a fence and a trap command to the ring.
+ */
+static void vcn_v1_0_enc_ring_emit_fence(struct amdgpu_ring *ring, u64 addr,
+                       u64 seq, unsigned flags)
+{
+       WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
+
+       amdgpu_ring_write(ring, VCN_ENC_CMD_FENCE);
+       amdgpu_ring_write(ring, addr);
+       amdgpu_ring_write(ring, upper_32_bits(addr));
+       amdgpu_ring_write(ring, seq);
+       amdgpu_ring_write(ring, VCN_ENC_CMD_TRAP);
+}
+
+static void vcn_v1_0_enc_ring_insert_end(struct amdgpu_ring *ring)
+{
+       amdgpu_ring_write(ring, VCN_ENC_CMD_END);
+}
+
+/**
+ * vcn_v1_0_enc_ring_emit_ib - enc execute indirect buffer
+ *
+ * @ring: amdgpu_ring pointer
+ * @ib: indirect buffer to execute
+ *
+ * Write enc ring commands to execute the indirect buffer
+ */
+static void vcn_v1_0_enc_ring_emit_ib(struct amdgpu_ring *ring,
+               struct amdgpu_ib *ib, unsigned int vm_id, bool ctx_switch)
+{
+       amdgpu_ring_write(ring, VCN_ENC_CMD_IB);
+       amdgpu_ring_write(ring, vm_id);
+       amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
+       amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
+       amdgpu_ring_write(ring, ib->length_dw);
+}
+
+static void vcn_v1_0_enc_ring_emit_vm_flush(struct amdgpu_ring *ring,
+                        unsigned int vm_id, uint64_t pd_addr)
+{
+       struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->funcs->vmhub];
+       uint32_t req = ring->adev->gart.gart_funcs->get_invalidate_req(vm_id);
+       unsigned eng = ring->vm_inv_eng;
+
+       pd_addr = pd_addr | 0x1; /* valid bit */
+       /* now only use physical base address of PDE and valid */
+       BUG_ON(pd_addr & 0xFFFF00000000003EULL);
+
+       amdgpu_ring_write(ring, VCN_ENC_CMD_REG_WRITE);
+       amdgpu_ring_write(ring,
+                         (hub->ctx0_ptb_addr_hi32 + vm_id * 2) << 2);
+       amdgpu_ring_write(ring, upper_32_bits(pd_addr));
+
+       amdgpu_ring_write(ring, VCN_ENC_CMD_REG_WRITE);
+       amdgpu_ring_write(ring,
+                         (hub->ctx0_ptb_addr_lo32 + vm_id * 2) << 2);
+       amdgpu_ring_write(ring, lower_32_bits(pd_addr));
+
+       amdgpu_ring_write(ring, VCN_ENC_CMD_REG_WAIT);
+       amdgpu_ring_write(ring,
+                         (hub->ctx0_ptb_addr_lo32 + vm_id * 2) << 2);
+       amdgpu_ring_write(ring, 0xffffffff);
+       amdgpu_ring_write(ring, lower_32_bits(pd_addr));
+
+       /* flush TLB */
+       amdgpu_ring_write(ring, VCN_ENC_CMD_REG_WRITE);
+       amdgpu_ring_write(ring, (hub->vm_inv_eng0_req + eng) << 2);
+       amdgpu_ring_write(ring, req);
+
+       /* wait for flush */
+       amdgpu_ring_write(ring, VCN_ENC_CMD_REG_WAIT);
+       amdgpu_ring_write(ring, (hub->vm_inv_eng0_ack + eng) << 2);
+       amdgpu_ring_write(ring, 1 << vm_id);
+       amdgpu_ring_write(ring, 1 << vm_id);
+}
+
 static int vcn_v1_0_set_interrupt_state(struct amdgpu_device *adev,
                                        struct amdgpu_irq_src *source,
                                        unsigned type,
@@ -755,12 +892,45 @@ static const struct amdgpu_ring_funcs vcn_v1_0_dec_ring_vm_funcs = {
        .end_use = amdgpu_vcn_ring_end_use,
 };
 
+static const struct amdgpu_ring_funcs vcn_v1_0_enc_ring_vm_funcs = {
+       .type = AMDGPU_RING_TYPE_VCN_ENC,
+       .align_mask = 0x3f,
+       .nop = VCN_ENC_CMD_NO_OP,
+       .support_64bit_ptrs = false,
+       .get_rptr = vcn_v1_0_enc_ring_get_rptr,
+       .get_wptr = vcn_v1_0_enc_ring_get_wptr,
+       .set_wptr = vcn_v1_0_enc_ring_set_wptr,
+       .emit_frame_size =
+               17 * AMDGPU_MAX_VMHUBS + /* vcn_v1_0_enc_ring_emit_vm_flush */
+               5 + 5 + /* vcn_v1_0_enc_ring_emit_fence x2 vm fence */
+               1, /* vcn_v1_0_enc_ring_insert_end */
+       .emit_ib_size = 5, /* vcn_v1_0_enc_ring_emit_ib */
+       .emit_ib = vcn_v1_0_enc_ring_emit_ib,
+       .emit_fence = vcn_v1_0_enc_ring_emit_fence,
+       .emit_vm_flush = vcn_v1_0_enc_ring_emit_vm_flush,
+       .insert_nop = amdgpu_ring_insert_nop,
+       .insert_end = vcn_v1_0_enc_ring_insert_end,
+       .pad_ib = amdgpu_ring_generic_pad_ib,
+       .begin_use = amdgpu_vcn_ring_begin_use,
+       .end_use = amdgpu_vcn_ring_end_use,
+};
+
 static void vcn_v1_0_set_dec_ring_funcs(struct amdgpu_device *adev)
 {
        adev->vcn.ring_dec.funcs = &vcn_v1_0_dec_ring_vm_funcs;
        DRM_INFO("VCN decode is enabled in VM mode\n");
 }
 
+static void vcn_v1_0_set_enc_ring_funcs(struct amdgpu_device *adev)
+{
+       int i;
+
+       for (i = 0; i < adev->vcn.num_enc_rings; ++i)
+               adev->vcn.ring_enc[i].funcs = &vcn_v1_0_enc_ring_vm_funcs;
+
+       DRM_INFO("VCN encode is enabled in VM mode\n");
+}
+
 static const struct amdgpu_irq_src_funcs vcn_v1_0_irq_funcs = {
        .set = vcn_v1_0_set_interrupt_state,
        .process = vcn_v1_0_process_interrupt,