From: Christian König <christian.koenig@xxxxxxx>
Starting with Stoney we support running UVD in VM mode as well.
v2: rebased, only enable on Polaris for now.
Signed-off-by: Christian König <christian.koenig at amd.com>
---
drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c | 82 +++++++++++++++++++++-
.../gpu/drm/amd/include/asic_reg/uvd/uvd_6_0_d.h | 2 +
2 files changed, 82 insertions(+), 2 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c b/drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c
index 69439ab..9ab7797 100644
--- a/drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/uvd_v6_0.c
@@ -34,6 +34,7 @@
#include "smu/smu_7_1_3_d.h"
#include "smu/smu_7_1_3_sh_mask.h"
#include "bif/bif_5_1_d.h"
+#include "gmc/gmc_8_1_d.h"
#include "vi.h"
static void uvd_v6_0_set_ring_funcs(struct amdgpu_device *adev);
@@ -674,6 +675,9 @@ static void uvd_v6_0_ring_emit_ib(struct amdgpu_ring *ring,
struct amdgpu_ib *ib,
unsigned vm_id, bool ctx_switch)
{
+ amdgpu_ring_write(ring, PACKET0(mmUVD_LMI_RBC_IB_VMID, 0));
+ amdgpu_ring_write(ring, vm_id);
+
amdgpu_ring_write(ring, PACKET0(mmUVD_LMI_RBC_IB_64BIT_BAR_LOW, 0));
amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
amdgpu_ring_write(ring, PACKET0(mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH, 0));
@@ -682,6 +686,57 @@ static void uvd_v6_0_ring_emit_ib(struct amdgpu_ring *ring,
amdgpu_ring_write(ring, ib->length_dw);
}
+static void uvd_v6_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
+ unsigned vm_id, uint64_t pd_addr)
+{
+ uint32_t reg;
+
+ if (vm_id < 8)
+ reg = mmVM_CONTEXT0_PAGE_TABLE_BASE_ADDR + vm_id;
+ else
+ reg = mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + vm_id - 8;
+
+ amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0));
+ amdgpu_ring_write(ring, reg << 2);
+ amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0));
+ amdgpu_ring_write(ring, pd_addr >> 12);
+ amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0));
+ amdgpu_ring_write(ring, 0x8);
+
+ amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0));
+ amdgpu_ring_write(ring, mmVM_INVALIDATE_REQUEST << 2);
+ amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0));
+ amdgpu_ring_write(ring, 1 << vm_id);
+ amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0));
+ amdgpu_ring_write(ring, 0x8);
+
+ amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0));
+ amdgpu_ring_write(ring, mmVM_INVALIDATE_REQUEST << 2);
+ amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0));
+ amdgpu_ring_write(ring, 0);
+ amdgpu_ring_write(ring, PACKET0(mmUVD_GP_SCRATCH8, 0));
+ amdgpu_ring_write(ring, 1 << vm_id); /* mask */
+ amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0));
+ amdgpu_ring_write(ring, 0xC);
+}
+
+static void uvd_v6_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
+{
+ uint32_t seq = ring->fence_drv.sync_seq;
+ uint64_t addr = ring->fence_drv.gpu_addr;
+
+ amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA0, 0));
+ amdgpu_ring_write(ring, lower_32_bits(addr));
+ amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_DATA1, 0));
+ amdgpu_ring_write(ring, upper_32_bits(addr));
+ amdgpu_ring_write(ring, PACKET0(mmUVD_GP_SCRATCH8, 0));
+ amdgpu_ring_write(ring, 0xffffffff); /* mask */
+ amdgpu_ring_write(ring, PACKET0(mmUVD_GP_SCRATCH9, 0));
+ amdgpu_ring_write(ring, seq);
+ amdgpu_ring_write(ring, PACKET0(mmUVD_GPCOM_VCPU_CMD, 0));
+ amdgpu_ring_write(ring, 0xE);
+}
+
static bool uvd_v6_0_is_idle(void *handle)
{
struct amdgpu_device *adev = (struct amdgpu_device *)handle;
@@ -918,7 +973,7 @@ const struct amd_ip_funcs uvd_v6_0_ip_funcs = {
.set_powergating_state = uvd_v6_0_set_powergating_state,
};
-static const struct amdgpu_ring_funcs uvd_v6_0_ring_funcs = {
+static const struct amdgpu_ring_funcs uvd_v6_0_ring_phys_funcs = {
.get_rptr = uvd_v6_0_ring_get_rptr,
.get_wptr = uvd_v6_0_ring_get_wptr,
.set_wptr = uvd_v6_0_ring_set_wptr,
@@ -933,9 +988,32 @@ static const struct amdgpu_ring_funcs uvd_v6_0_ring_funcs = {
.pad_ib = amdgpu_ring_generic_pad_ib,
};
+static const struct amdgpu_ring_funcs uvd_v6_0_ring_vm_funcs = {
+ .get_rptr = uvd_v6_0_ring_get_rptr,
+ .get_wptr = uvd_v6_0_ring_get_wptr,
+ .set_wptr = uvd_v6_0_ring_set_wptr,
+ .parse_cs = NULL,
+ .emit_ib = uvd_v6_0_ring_emit_ib,
+ .emit_fence = uvd_v6_0_ring_emit_fence,
+ .emit_vm_flush = uvd_v6_0_ring_emit_vm_flush,
+ .emit_pipeline_sync = uvd_v6_0_ring_emit_pipeline_sync,
+ .emit_hdp_flush = uvd_v6_0_ring_emit_hdp_flush,
+ .emit_hdp_invalidate = uvd_v6_0_ring_emit_hdp_invalidate,
+ .test_ring = uvd_v6_0_ring_test_ring,
+ .test_ib = amdgpu_uvd_ring_test_ib,
+ .insert_nop = amdgpu_ring_insert_nop,
+ .pad_ib = amdgpu_ring_generic_pad_ib,
+};
+
static void uvd_v6_0_set_ring_funcs(struct amdgpu_device *adev)
{
- adev->uvd.ring.funcs = &uvd_v6_0_ring_funcs;
+ if (adev->asic_type >= CHIP_STONEY) {
+ adev->uvd.ring.funcs = &uvd_v6_0_ring_vm_funcs;
+ DRM_INFO("UVD is enabled in VM mode\n");
+ } else {
+ adev->uvd.ring.funcs = &uvd_v6_0_ring_phys_funcs;
+ DRM_INFO("UVD is enabled in physical mode\n");
+ }
}
static const struct amdgpu_irq_src_funcs uvd_v6_0_irq_funcs = {
diff --git a/drivers/gpu/drm/amd/include/asic_reg/uvd/uvd_6_0_d.h b/drivers/gpu/drm/amd/include/asic_reg/uvd/uvd_6_0_d.h
index 6f6fb34..ec69869 100644
--- a/drivers/gpu/drm/amd/include/asic_reg/uvd/uvd_6_0_d.h
+++ b/drivers/gpu/drm/amd/include/asic_reg/uvd/uvd_6_0_d.h
@@ -111,6 +111,8 @@
#define mmUVD_MIF_RECON1_ADDR_CONFIG 0x39c5
#define ixUVD_MIF_SCLR_ADDR_CONFIG 0x4
#define mmUVD_JPEG_ADDR_CONFIG 0x3a1f
+#define mmUVD_GP_SCRATCH8 0x3c0a
+#define mmUVD_GP_SCRATCH9 0x3c0b
#define mmUVD_GP_SCRATCH4 0x3d38
#endif /* UVD_6_0_D_H */
--
2.5.0
