diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c')
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c | 130 |
1 files changed, 112 insertions, 18 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c index 2f3f09dfb1fd2..f57cc72c43cf4 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c @@ -515,6 +515,9 @@ int amdgpu_gfx_disable_kcq(struct amdgpu_device *adev, int xcc_id) if (!kiq->pmf || !kiq->pmf->kiq_unmap_queues) return -EINVAL; + if (!kiq_ring->sched.ready || adev->job_hang || amdgpu_in_reset(adev)) + return 0; + spin_lock(&kiq->ring_lock); if (amdgpu_ring_alloc(kiq_ring, kiq->pmf->unmap_queues_size * adev->gfx.num_compute_rings)) { @@ -528,20 +531,15 @@ int amdgpu_gfx_disable_kcq(struct amdgpu_device *adev, int xcc_id) &adev->gfx.compute_ring[j], RESET_QUEUES, 0, 0); } - - /** - * This is workaround: only skip kiq_ring test - * during ras recovery in suspend stage for gfx9.4.3 + /* Submit unmap queue packet */ + amdgpu_ring_commit(kiq_ring); + /* + * Ring test will do a basic scratch register change check. Just run + * this to ensure that unmap queues that is submitted before got + * processed successfully before returning. */ - if ((amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 3) || - amdgpu_ip_version(adev, GC_HWIP, 0) == IP_VERSION(9, 4, 4)) && - amdgpu_ras_in_recovery(adev)) { - spin_unlock(&kiq->ring_lock); - return 0; - } + r = amdgpu_ring_test_helper(kiq_ring); - if (kiq_ring->sched.ready && !adev->job_hang) - r = amdgpu_ring_test_helper(kiq_ring); spin_unlock(&kiq->ring_lock); return r; @@ -569,8 +567,11 @@ int amdgpu_gfx_disable_kgq(struct amdgpu_device *adev, int xcc_id) if (!kiq->pmf || !kiq->pmf->kiq_unmap_queues) return -EINVAL; - spin_lock(&kiq->ring_lock); + if (!adev->gfx.kiq[0].ring.sched.ready || adev->job_hang) + return 0; + if (amdgpu_gfx_is_master_xcc(adev, xcc_id)) { + spin_lock(&kiq->ring_lock); if (amdgpu_ring_alloc(kiq_ring, kiq->pmf->unmap_queues_size * adev->gfx.num_gfx_rings)) { spin_unlock(&kiq->ring_lock); @@ -583,11 +584,17 @@ int amdgpu_gfx_disable_kgq(struct amdgpu_device *adev, int xcc_id) &adev->gfx.gfx_ring[j], PREEMPT_QUEUES, 0, 0); } - } + /* Submit unmap queue packet */ + amdgpu_ring_commit(kiq_ring); - if (adev->gfx.kiq[0].ring.sched.ready && !adev->job_hang) + /* + * Ring test will do a basic scratch register change check. + * Just run this to ensure that unmap queues that is submitted + * before got processed successfully before returning. + */ r = amdgpu_ring_test_helper(kiq_ring); - spin_unlock(&kiq->ring_lock); + spin_unlock(&kiq->ring_lock); + } return r; } @@ -692,7 +699,13 @@ int amdgpu_gfx_enable_kcq(struct amdgpu_device *adev, int xcc_id) kiq->pmf->kiq_map_queues(kiq_ring, &adev->gfx.compute_ring[j]); } - + /* Submit map queue packet */ + amdgpu_ring_commit(kiq_ring); + /* + * Ring test will do a basic scratch register change check. Just run + * this to ensure that map queues that is submitted before got + * processed successfully before returning. + */ r = amdgpu_ring_test_helper(kiq_ring); spin_unlock(&kiq->ring_lock); if (r) @@ -743,7 +756,13 @@ int amdgpu_gfx_enable_kgq(struct amdgpu_device *adev, int xcc_id) &adev->gfx.gfx_ring[j]); } } - + /* Submit map queue packet */ + amdgpu_ring_commit(kiq_ring); + /* + * Ring test will do a basic scratch register change check. Just run + * this to ensure that map queues that is submitted before got + * processed successfully before returning. + */ r = amdgpu_ring_test_helper(kiq_ring); spin_unlock(&kiq->ring_lock); if (r) @@ -885,6 +904,9 @@ int amdgpu_gfx_ras_late_init(struct amdgpu_device *adev, struct ras_common_if *r if (r) return r; + if (amdgpu_sriov_vf(adev)) + return r; + if (adev->gfx.cp_ecc_error_irq.funcs) { r = amdgpu_irq_get(adev, &adev->gfx.cp_ecc_error_irq, 0); if (r) @@ -1576,9 +1598,11 @@ static ssize_t amdgpu_gfx_set_enforce_isolation(struct device *dev, if (adev->enforce_isolation[i] && !partition_values[i]) { /* Going from enabled to disabled */ amdgpu_vmid_free_reserved(adev, AMDGPU_GFXHUB(i)); + amdgpu_mes_set_enforce_isolation(adev, i, false); } else if (!adev->enforce_isolation[i] && partition_values[i]) { /* Going from disabled to enabled */ amdgpu_vmid_alloc_reserved(adev, AMDGPU_GFXHUB(i)); + amdgpu_mes_set_enforce_isolation(adev, i, true); } adev->enforce_isolation[i] = partition_values[i]; } @@ -1588,6 +1612,32 @@ static ssize_t amdgpu_gfx_set_enforce_isolation(struct device *dev, return count; } +static ssize_t amdgpu_gfx_get_gfx_reset_mask(struct device *dev, + struct device_attribute *attr, + char *buf) +{ + struct drm_device *ddev = dev_get_drvdata(dev); + struct amdgpu_device *adev = drm_to_adev(ddev); + + if (!adev) + return -ENODEV; + + return amdgpu_show_reset_mask(buf, adev->gfx.gfx_supported_reset); +} + +static ssize_t amdgpu_gfx_get_compute_reset_mask(struct device *dev, + struct device_attribute *attr, + char *buf) +{ + struct drm_device *ddev = dev_get_drvdata(dev); + struct amdgpu_device *adev = drm_to_adev(ddev); + + if (!adev) + return -ENODEV; + + return amdgpu_show_reset_mask(buf, adev->gfx.compute_supported_reset); +} + static DEVICE_ATTR(run_cleaner_shader, 0200, NULL, amdgpu_gfx_set_run_cleaner_shader); @@ -1601,6 +1651,11 @@ static DEVICE_ATTR(current_compute_partition, 0644, static DEVICE_ATTR(available_compute_partition, 0444, amdgpu_gfx_get_available_compute_partition, NULL); +static DEVICE_ATTR(gfx_reset_mask, 0444, + amdgpu_gfx_get_gfx_reset_mask, NULL); + +static DEVICE_ATTR(compute_reset_mask, 0444, + amdgpu_gfx_get_compute_reset_mask, NULL); static int amdgpu_gfx_sysfs_xcp_init(struct amdgpu_device *adev) { @@ -1666,6 +1721,40 @@ static void amdgpu_gfx_sysfs_isolation_shader_fini(struct amdgpu_device *adev) device_remove_file(adev->dev, &dev_attr_run_cleaner_shader); } +static int amdgpu_gfx_sysfs_reset_mask_init(struct amdgpu_device *adev) +{ + int r = 0; + + if (!amdgpu_gpu_recovery) + return r; + + if (adev->gfx.num_gfx_rings) { + r = device_create_file(adev->dev, &dev_attr_gfx_reset_mask); + if (r) + return r; + } + + if (adev->gfx.num_compute_rings) { + r = device_create_file(adev->dev, &dev_attr_compute_reset_mask); + if (r) + return r; + } + + return r; +} + +static void amdgpu_gfx_sysfs_reset_mask_fini(struct amdgpu_device *adev) +{ + if (!amdgpu_gpu_recovery) + return; + + if (adev->gfx.num_gfx_rings) + device_remove_file(adev->dev, &dev_attr_gfx_reset_mask); + + if (adev->gfx.num_compute_rings) + device_remove_file(adev->dev, &dev_attr_compute_reset_mask); +} + int amdgpu_gfx_sysfs_init(struct amdgpu_device *adev) { int r; @@ -1680,6 +1769,10 @@ int amdgpu_gfx_sysfs_init(struct amdgpu_device *adev) if (r) dev_err(adev->dev, "failed to create isolation sysfs files"); + r = amdgpu_gfx_sysfs_reset_mask_init(adev); + if (r) + dev_err(adev->dev, "failed to create reset mask sysfs files"); + return r; } @@ -1687,6 +1780,7 @@ void amdgpu_gfx_sysfs_fini(struct amdgpu_device *adev) { amdgpu_gfx_sysfs_xcp_fini(adev); amdgpu_gfx_sysfs_isolation_shader_fini(adev); + amdgpu_gfx_sysfs_reset_mask_fini(adev); } int amdgpu_gfx_cleaner_shader_sw_init(struct amdgpu_device *adev, |