diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c')
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c | 73 |
1 files changed, 66 insertions, 7 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c index 5f4805e4d04a..65db88bb6cbc 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c @@ -38,7 +38,6 @@ #include "gc/gc_9_0_sh_mask.h" #include "vega10_enum.h" -#include "hdp/hdp_4_0_offset.h" #include "soc15_common.h" #include "clearstate_gfx9.h" @@ -53,6 +52,7 @@ #include "asic_reg/pwr/pwr_10_0_offset.h" #include "asic_reg/pwr/pwr_10_0_sh_mask.h" +#include "asic_reg/gc/gc_9_0_default.h" #define GFX9_NUM_GFX_RINGS 1 #define GFX9_MEC_HPD_SIZE 4096 @@ -2228,8 +2228,7 @@ static int gfx_v9_0_compute_ring_init(struct amdgpu_device *adev, int ring_id, irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP + ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec) + ring->pipe; - hw_prio = amdgpu_gfx_is_high_priority_compute_queue(adev, ring->pipe, - ring->queue) ? + hw_prio = amdgpu_gfx_is_high_priority_compute_queue(adev, ring) ? AMDGPU_GFX_PIPE_PRIO_HIGH : AMDGPU_GFX_PIPE_PRIO_NORMAL; /* type-2 packets are deprecated on MEC, use type-3 instead */ return amdgpu_ring_init(adev, ring, 1024, @@ -3391,9 +3390,7 @@ static void gfx_v9_0_mqd_set_priority(struct amdgpu_ring *ring, struct v9_mqd *m struct amdgpu_device *adev = ring->adev; if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) { - if (amdgpu_gfx_is_high_priority_compute_queue(adev, - ring->pipe, - ring->queue)) { + if (amdgpu_gfx_is_high_priority_compute_queue(adev, ring)) { mqd->cp_hqd_pipe_priority = AMDGPU_GFX_PIPE_PRIO_HIGH; mqd->cp_hqd_queue_priority = AMDGPU_GFX_QUEUE_PRIORITY_MAXIMUM; @@ -6671,6 +6668,65 @@ static void gfx_v9_0_emit_mem_sync(struct amdgpu_ring *ring) amdgpu_ring_write(ring, 0x0000000A); /* POLL_INTERVAL */ } +static void gfx_v9_0_emit_wave_limit_cs(struct amdgpu_ring *ring, + uint32_t pipe, bool enable) +{ + struct amdgpu_device *adev = ring->adev; + uint32_t val; + uint32_t wcl_cs_reg; + + /* mmSPI_WCL_PIPE_PERCENT_CS[0-7]_DEFAULT values are same */ + val = enable ? 0x1 : mmSPI_WCL_PIPE_PERCENT_CS0_DEFAULT; + + switch (pipe) { + case 0: + wcl_cs_reg = SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_CS0); + break; + case 1: + wcl_cs_reg = SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_CS1); + break; + case 2: + wcl_cs_reg = SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_CS2); + break; + case 3: + wcl_cs_reg = SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_CS3); + break; + default: + DRM_DEBUG("invalid pipe %d\n", pipe); + return; + } + + amdgpu_ring_emit_wreg(ring, wcl_cs_reg, val); + +} +static void gfx_v9_0_emit_wave_limit(struct amdgpu_ring *ring, bool enable) +{ + struct amdgpu_device *adev = ring->adev; + uint32_t val; + int i; + + + /* mmSPI_WCL_PIPE_PERCENT_GFX is 7 bit multiplier register to limit + * number of gfx waves. Setting 5 bit will make sure gfx only gets + * around 25% of gpu resources. + */ + val = enable ? 0x1f : mmSPI_WCL_PIPE_PERCENT_GFX_DEFAULT; + amdgpu_ring_emit_wreg(ring, + SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_GFX), + val); + + /* Restrict waves for normal/low priority compute queues as well + * to get best QoS for high priority compute jobs. + * + * amdgpu controls only 1st ME(0-3 CS pipes). + */ + for (i = 0; i < adev->gfx.mec.num_pipe_per_mec; i++) { + if (i != ring->pipe) + gfx_v9_0_emit_wave_limit_cs(ring, i, enable); + + } +} + static const struct amd_ip_funcs gfx_v9_0_ip_funcs = { .name = "gfx_v9_0", .early_init = gfx_v9_0_early_init, @@ -6760,7 +6816,9 @@ static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_compute = { SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 + 2 + /* gfx_v9_0_ring_emit_vm_flush */ 8 + 8 + 8 + /* gfx_v9_0_ring_emit_fence x3 for user fence, vm fence */ - 7, /* gfx_v9_0_emit_mem_sync */ + 7 + /* gfx_v9_0_emit_mem_sync */ + 5 + /* gfx_v9_0_emit_wave_limit for updating mmSPI_WCL_PIPE_PERCENT_GFX register */ + 15, /* for updating 3 mmSPI_WCL_PIPE_PERCENT_CS registers */ .emit_ib_size = 7, /* gfx_v9_0_ring_emit_ib_compute */ .emit_ib = gfx_v9_0_ring_emit_ib_compute, .emit_fence = gfx_v9_0_ring_emit_fence, @@ -6776,6 +6834,7 @@ static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_compute = { .emit_reg_wait = gfx_v9_0_ring_emit_reg_wait, .emit_reg_write_reg_wait = gfx_v9_0_ring_emit_reg_write_reg_wait, .emit_mem_sync = gfx_v9_0_emit_mem_sync, + .emit_wave_limit = gfx_v9_0_emit_wave_limit, }; static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_kiq = { |