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 | 236 |
1 files changed, 107 insertions, 129 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c index ab6880958511..1081fa3d4b0f 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c @@ -738,9 +738,9 @@ static void gfx_v9_0_ring_emit_de_meta(struct amdgpu_ring *ring); static u64 gfx_v9_0_ring_get_rptr_compute(struct amdgpu_ring *ring); static int gfx_v9_0_query_ras_error_count(struct amdgpu_device *adev, void *ras_error_status); -static void gfx_v9_0_clear_ras_edc_counter(struct amdgpu_device *adev); static int gfx_v9_0_ras_error_inject(struct amdgpu_device *adev, void *inject_if); +static void gfx_v9_0_reset_ras_error_count(struct amdgpu_device *adev); static void gfx_v9_0_kiq_set_resources(struct amdgpu_ring *kiq_ring, uint64_t queue_mask) @@ -1847,6 +1847,10 @@ static int gfx_v9_0_rlc_init(struct amdgpu_device *adev) break; } + /* init spm vmid with 0xf */ + if (adev->gfx.rlc.funcs->update_spm_vmid) + adev->gfx.rlc.funcs->update_spm_vmid(adev, 0xf); + return 0; } @@ -1993,7 +1997,8 @@ static const struct amdgpu_gfx_funcs gfx_v9_0_gfx_funcs = { .read_wave_vgprs = &gfx_v9_0_read_wave_vgprs, .select_me_pipe_q = &gfx_v9_0_select_me_pipe_q, .ras_error_inject = &gfx_v9_0_ras_error_inject, - .query_ras_error_count = &gfx_v9_0_query_ras_error_count + .query_ras_error_count = &gfx_v9_0_query_ras_error_count, + .reset_ras_error_count = &gfx_v9_0_reset_ras_error_count, }; static const struct amdgpu_gfx_funcs gfx_v9_4_gfx_funcs = { @@ -2004,7 +2009,8 @@ static const struct amdgpu_gfx_funcs gfx_v9_4_gfx_funcs = { .read_wave_vgprs = &gfx_v9_0_read_wave_vgprs, .select_me_pipe_q = &gfx_v9_0_select_me_pipe_q, .ras_error_inject = &gfx_v9_4_ras_error_inject, - .query_ras_error_count = &gfx_v9_4_query_ras_error_count + .query_ras_error_count = &gfx_v9_4_query_ras_error_count, + .reset_ras_error_count = &gfx_v9_4_reset_ras_error_count, }; static int gfx_v9_0_gpu_early_init(struct amdgpu_device *adev) @@ -3310,6 +3316,22 @@ static void gfx_v9_0_kiq_setting(struct amdgpu_ring *ring) WREG32_SOC15_RLC(GC, 0, mmRLC_CP_SCHEDULERS, tmp); } +static void gfx_v9_0_mqd_set_priority(struct amdgpu_ring *ring, struct v9_mqd *mqd) +{ + struct amdgpu_device *adev = ring->adev; + + if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) { + if (amdgpu_gfx_is_high_priority_compute_queue(adev, ring->queue)) { + mqd->cp_hqd_pipe_priority = AMDGPU_GFX_PIPE_PRIO_HIGH; + ring->has_high_prio = true; + mqd->cp_hqd_queue_priority = + AMDGPU_GFX_QUEUE_PRIORITY_MAXIMUM; + } else { + ring->has_high_prio = false; + } + } +} + static int gfx_v9_0_mqd_init(struct amdgpu_ring *ring) { struct amdgpu_device *adev = ring->adev; @@ -3446,6 +3468,10 @@ static int gfx_v9_0_mqd_init(struct amdgpu_ring *ring) tmp = REG_SET_FIELD(tmp, CP_HQD_IB_CONTROL, MIN_IB_AVAIL_SIZE, 3); mqd->cp_hqd_ib_control = tmp; + /* set static priority for a queue/ring */ + gfx_v9_0_mqd_set_priority(ring, mqd); + mqd->cp_hqd_quantum = RREG32(mmCP_HQD_QUANTUM); + /* map_queues packet doesn't need activate the queue, * so only kiq need set this field. */ @@ -3964,6 +3990,63 @@ static int gfx_v9_0_soft_reset(void *handle) return 0; } +static uint64_t gfx_v9_0_kiq_read_clock(struct amdgpu_device *adev) +{ + signed long r, cnt = 0; + unsigned long flags; + uint32_t seq; + struct amdgpu_kiq *kiq = &adev->gfx.kiq; + struct amdgpu_ring *ring = &kiq->ring; + + BUG_ON(!ring->funcs->emit_rreg); + + spin_lock_irqsave(&kiq->ring_lock, flags); + amdgpu_ring_alloc(ring, 32); + amdgpu_ring_write(ring, PACKET3(PACKET3_COPY_DATA, 4)); + amdgpu_ring_write(ring, 9 | /* src: register*/ + (5 << 8) | /* dst: memory */ + (1 << 16) | /* count sel */ + (1 << 20)); /* write confirm */ + amdgpu_ring_write(ring, 0); + amdgpu_ring_write(ring, 0); + amdgpu_ring_write(ring, lower_32_bits(adev->wb.gpu_addr + + kiq->reg_val_offs * 4)); + amdgpu_ring_write(ring, upper_32_bits(adev->wb.gpu_addr + + kiq->reg_val_offs * 4)); + amdgpu_fence_emit_polling(ring, &seq); + amdgpu_ring_commit(ring); + spin_unlock_irqrestore(&kiq->ring_lock, flags); + + r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT); + + /* don't wait anymore for gpu reset case because this way may + * block gpu_recover() routine forever, e.g. this virt_kiq_rreg + * is triggered in TTM and ttm_bo_lock_delayed_workqueue() will + * never return if we keep waiting in virt_kiq_rreg, which cause + * gpu_recover() hang there. + * + * also don't wait anymore for IRQ context + * */ + if (r < 1 && (adev->in_gpu_reset || in_interrupt())) + goto failed_kiq_read; + + might_sleep(); + while (r < 1 && cnt++ < MAX_KIQ_REG_TRY) { + msleep(MAX_KIQ_REG_BAILOUT_INTERVAL); + r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT); + } + + if (cnt > MAX_KIQ_REG_TRY) + goto failed_kiq_read; + + return (uint64_t)adev->wb.wb[kiq->reg_val_offs] | + (uint64_t)adev->wb.wb[kiq->reg_val_offs + 1 ] << 32ULL; + +failed_kiq_read: + pr_err("failed to read gpu clock\n"); + return ~0; +} + static uint64_t gfx_v9_0_get_gpu_clock_counter(struct amdgpu_device *adev) { uint64_t clock; @@ -3971,16 +4054,7 @@ static uint64_t gfx_v9_0_get_gpu_clock_counter(struct amdgpu_device *adev) amdgpu_gfx_off_ctrl(adev, false); mutex_lock(&adev->gfx.gpu_clock_mutex); if (adev->asic_type == CHIP_VEGA10 && amdgpu_sriov_runtime(adev)) { - uint32_t tmp, lsb, msb, i = 0; - do { - if (i != 0) - udelay(1); - tmp = RREG32_SOC15(GC, 0, mmRLC_REFCLOCK_TIMESTAMP_MSB); - lsb = RREG32_SOC15(GC, 0, mmRLC_REFCLOCK_TIMESTAMP_LSB); - msb = RREG32_SOC15(GC, 0, mmRLC_REFCLOCK_TIMESTAMP_MSB); - i++; - } while (unlikely(tmp != msb) && (i < adev->usec_timeout)); - clock = (uint64_t)lsb | ((uint64_t)msb << 32ULL); + clock = gfx_v9_0_kiq_read_clock(adev); } else { WREG32_SOC15(GC, 0, mmRLC_CAPTURE_GPU_CLOCK_COUNT, 1); clock = (uint64_t)RREG32_SOC15(GC, 0, mmRLC_GPU_CLOCK_COUNT_LSB) | @@ -4142,7 +4216,6 @@ static const struct soc15_reg_entry gfx_v9_0_edc_counter_regs[] = { { SOC15_REG_ENTRY(GC, 0, mmTCC_EDC_CNT2), 0, 1, 16}, { SOC15_REG_ENTRY(GC, 0, mmTCA_EDC_CNT), 0, 1, 2}, { SOC15_REG_ENTRY(GC, 0, mmSQC_EDC_CNT3), 0, 4, 6}, - { SOC15_REG_ENTRY(HDP, 0, mmHDP_EDC_CNT), 0, 1, 1}, }; static int gfx_v9_0_do_edc_gds_workarounds(struct amdgpu_device *adev) @@ -4343,18 +4416,6 @@ static int gfx_v9_0_do_edc_gpr_workarounds(struct amdgpu_device *adev) goto fail; } - switch (adev->asic_type) - { - case CHIP_VEGA20: - gfx_v9_0_clear_ras_edc_counter(adev); - break; - case CHIP_ARCTURUS: - gfx_v9_4_clear_ras_edc_counter(adev); - break; - default: - break; - } - fail: amdgpu_ib_free(adev, &ib, NULL); dma_fence_put(f); @@ -4402,6 +4463,10 @@ static int gfx_v9_0_ecc_late_init(void *handle) if (r) return r; + if (adev->gfx.funcs && + adev->gfx.funcs->reset_ras_error_count) + adev->gfx.funcs->reset_ras_error_count(adev); + r = amdgpu_gfx_ras_late_init(adev); if (r) return r; @@ -4706,6 +4771,18 @@ static int gfx_v9_0_update_gfx_clock_gating(struct amdgpu_device *adev, return 0; } +static void gfx_v9_0_update_spm_vmid(struct amdgpu_device *adev, unsigned vmid) +{ + u32 data; + + data = RREG32_SOC15(GC, 0, mmRLC_SPM_MC_CNTL); + + data &= ~RLC_SPM_MC_CNTL__RLC_SPM_VMID_MASK; + data |= (vmid & RLC_SPM_MC_CNTL__RLC_SPM_VMID_MASK) << RLC_SPM_MC_CNTL__RLC_SPM_VMID__SHIFT; + + WREG32_SOC15(GC, 0, mmRLC_SPM_MC_CNTL, data); +} + static const struct amdgpu_rlc_funcs gfx_v9_0_rlc_funcs = { .is_rlc_enabled = gfx_v9_0_is_rlc_enabled, .set_safe_mode = gfx_v9_0_set_safe_mode, @@ -4717,7 +4794,8 @@ static const struct amdgpu_rlc_funcs gfx_v9_0_rlc_funcs = { .resume = gfx_v9_0_rlc_resume, .stop = gfx_v9_0_rlc_stop, .reset = gfx_v9_0_rlc_reset, - .start = gfx_v9_0_rlc_start + .start = gfx_v9_0_rlc_start, + .update_spm_vmid = gfx_v9_0_update_spm_vmid }; static int gfx_v9_0_set_powergating_state(void *handle, @@ -4920,7 +4998,7 @@ static void gfx_v9_0_ring_emit_ib_gfx(struct amdgpu_ring *ring, if (amdgpu_sriov_vf(ring->adev) && (ib->flags & AMDGPU_IB_FLAG_PREEMPT)) { control |= INDIRECT_BUFFER_PRE_ENB(1); - if (!(ib->flags & AMDGPU_IB_FLAG_CE)) + if (!(ib->flags & AMDGPU_IB_FLAG_CE) && vmid) gfx_v9_0_ring_emit_de_meta(ring); } @@ -5045,105 +5123,6 @@ static u64 gfx_v9_0_ring_get_wptr_compute(struct amdgpu_ring *ring) return wptr; } -static void gfx_v9_0_ring_set_pipe_percent(struct amdgpu_ring *ring, - bool acquire) -{ - struct amdgpu_device *adev = ring->adev; - int pipe_num, tmp, reg; - int pipe_percent = acquire ? SPI_WCL_PIPE_PERCENT_GFX__VALUE_MASK : 0x1; - - pipe_num = ring->me * adev->gfx.mec.num_pipe_per_mec + ring->pipe; - - /* first me only has 2 entries, GFX and HP3D */ - if (ring->me > 0) - pipe_num -= 2; - - reg = SOC15_REG_OFFSET(GC, 0, mmSPI_WCL_PIPE_PERCENT_GFX) + pipe_num; - tmp = RREG32(reg); - tmp = REG_SET_FIELD(tmp, SPI_WCL_PIPE_PERCENT_GFX, VALUE, pipe_percent); - WREG32(reg, tmp); -} - -static void gfx_v9_0_pipe_reserve_resources(struct amdgpu_device *adev, - struct amdgpu_ring *ring, - bool acquire) -{ - int i, pipe; - bool reserve; - struct amdgpu_ring *iring; - - mutex_lock(&adev->gfx.pipe_reserve_mutex); - pipe = amdgpu_gfx_mec_queue_to_bit(adev, ring->me, ring->pipe, 0); - if (acquire) - set_bit(pipe, adev->gfx.pipe_reserve_bitmap); - else - clear_bit(pipe, adev->gfx.pipe_reserve_bitmap); - - if (!bitmap_weight(adev->gfx.pipe_reserve_bitmap, AMDGPU_MAX_COMPUTE_QUEUES)) { - /* Clear all reservations - everyone reacquires all resources */ - for (i = 0; i < adev->gfx.num_gfx_rings; ++i) - gfx_v9_0_ring_set_pipe_percent(&adev->gfx.gfx_ring[i], - true); - - for (i = 0; i < adev->gfx.num_compute_rings; ++i) - gfx_v9_0_ring_set_pipe_percent(&adev->gfx.compute_ring[i], - true); - } else { - /* Lower all pipes without a current reservation */ - for (i = 0; i < adev->gfx.num_gfx_rings; ++i) { - iring = &adev->gfx.gfx_ring[i]; - pipe = amdgpu_gfx_mec_queue_to_bit(adev, - iring->me, - iring->pipe, - 0); - reserve = test_bit(pipe, adev->gfx.pipe_reserve_bitmap); - gfx_v9_0_ring_set_pipe_percent(iring, reserve); - } - - for (i = 0; i < adev->gfx.num_compute_rings; ++i) { - iring = &adev->gfx.compute_ring[i]; - pipe = amdgpu_gfx_mec_queue_to_bit(adev, - iring->me, - iring->pipe, - 0); - reserve = test_bit(pipe, adev->gfx.pipe_reserve_bitmap); - gfx_v9_0_ring_set_pipe_percent(iring, reserve); - } - } - - mutex_unlock(&adev->gfx.pipe_reserve_mutex); -} - -static void gfx_v9_0_hqd_set_priority(struct amdgpu_device *adev, - struct amdgpu_ring *ring, - bool acquire) -{ - uint32_t pipe_priority = acquire ? 0x2 : 0x0; - uint32_t queue_priority = acquire ? 0xf : 0x0; - - mutex_lock(&adev->srbm_mutex); - soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0); - - WREG32_SOC15_RLC(GC, 0, mmCP_HQD_PIPE_PRIORITY, pipe_priority); - WREG32_SOC15_RLC(GC, 0, mmCP_HQD_QUEUE_PRIORITY, queue_priority); - - soc15_grbm_select(adev, 0, 0, 0, 0); - mutex_unlock(&adev->srbm_mutex); -} - -static void gfx_v9_0_ring_set_priority_compute(struct amdgpu_ring *ring, - enum drm_sched_priority priority) -{ - struct amdgpu_device *adev = ring->adev; - bool acquire = priority == DRM_SCHED_PRIORITY_HIGH_HW; - - if (ring->funcs->type != AMDGPU_RING_TYPE_COMPUTE) - return; - - gfx_v9_0_hqd_set_priority(adev, ring, acquire); - gfx_v9_0_pipe_reserve_resources(adev, ring, acquire); -} - static void gfx_v9_0_ring_set_wptr_compute(struct amdgpu_ring *ring) { struct amdgpu_device *adev = ring->adev; @@ -6323,7 +6302,7 @@ static int gfx_v9_0_ras_error_count(const struct soc15_reg_entry *reg, return 0; } -static void gfx_v9_0_clear_ras_edc_counter(struct amdgpu_device *adev) +static void gfx_v9_0_reset_ras_error_count(struct amdgpu_device *adev) { int i, j, k; @@ -6514,7 +6493,6 @@ static const struct amdgpu_ring_funcs gfx_v9_0_ring_funcs_compute = { .test_ib = gfx_v9_0_ring_test_ib, .insert_nop = amdgpu_ring_insert_nop, .pad_ib = amdgpu_ring_generic_pad_ib, - .set_priority = gfx_v9_0_ring_set_priority_compute, .emit_wreg = gfx_v9_0_ring_emit_wreg, .emit_reg_wait = gfx_v9_0_ring_emit_reg_wait, .emit_reg_write_reg_wait = gfx_v9_0_ring_emit_reg_write_reg_wait, |