diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdkfd')
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_device.c | 748 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 8 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_doorbell.c | 8 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c | 2 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c | 18 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_iommu.c | 2 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c | 6 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_migrate.c | 4 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c | 2 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_vi.c | 4 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 3 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_process.c | 18 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c | 9 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 233 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_topology.c | 18 |
15 files changed, 329 insertions, 754 deletions
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c index e1294fba0c26..facc28f58c1f 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c @@ -53,680 +53,242 @@ extern const struct kfd2kgd_calls aldebaran_kfd2kgd; extern const struct kfd2kgd_calls gfx_v10_kfd2kgd; extern const struct kfd2kgd_calls gfx_v10_3_kfd2kgd; -#ifdef KFD_SUPPORT_IOMMU_V2 -static const struct kfd_device_info kaveri_device_info = { - .asic_name = "kaveri", - .gfx_target_version = 70000, - .max_pasid_bits = 16, - /* max num of queues for KV.TODO should be a dynamic value */ - .max_no_of_hqd = 24, - .doorbell_size = 4, - .ih_ring_entry_size = 4 * sizeof(uint32_t), - .event_interrupt_class = &event_interrupt_class_cik, - .num_of_watch_points = 4, - .mqd_size_aligned = MQD_SIZE_ALIGNED, - .supports_cwsr = false, - .needs_iommu_device = true, - .needs_pci_atomics = false, - .num_sdma_queues_per_engine = 2, -}; - -static const struct kfd_device_info carrizo_device_info = { - .asic_name = "carrizo", - .gfx_target_version = 80001, - .max_pasid_bits = 16, - /* max num of queues for CZ.TODO should be a dynamic value */ - .max_no_of_hqd = 24, - .doorbell_size = 4, - .ih_ring_entry_size = 4 * sizeof(uint32_t), - .event_interrupt_class = &event_interrupt_class_cik, - .num_of_watch_points = 4, - .mqd_size_aligned = MQD_SIZE_ALIGNED, - .supports_cwsr = true, - .needs_iommu_device = true, - .needs_pci_atomics = false, - .num_sdma_queues_per_engine = 2, -}; - -static const struct kfd_device_info raven_device_info = { - .asic_name = "raven", - .gfx_target_version = 90002, - .max_pasid_bits = 16, - .max_no_of_hqd = 24, - .doorbell_size = 8, - .ih_ring_entry_size = 8 * sizeof(uint32_t), - .event_interrupt_class = &event_interrupt_class_v9, - .num_of_watch_points = 4, - .mqd_size_aligned = MQD_SIZE_ALIGNED, - .supports_cwsr = true, - .needs_iommu_device = true, - .needs_pci_atomics = true, - .num_sdma_queues_per_engine = 2, -}; -#endif - -#ifdef CONFIG_DRM_AMDGPU_CIK -static const struct kfd_device_info hawaii_device_info = { - .asic_name = "hawaii", - .gfx_target_version = 70001, - .max_pasid_bits = 16, - /* max num of queues for KV.TODO should be a dynamic value */ - .max_no_of_hqd = 24, - .doorbell_size = 4, - .ih_ring_entry_size = 4 * sizeof(uint32_t), - .event_interrupt_class = &event_interrupt_class_cik, - .num_of_watch_points = 4, - .mqd_size_aligned = MQD_SIZE_ALIGNED, - .supports_cwsr = false, - .needs_iommu_device = false, - .needs_pci_atomics = false, - .num_sdma_queues_per_engine = 2, -}; -#endif - -static const struct kfd_device_info tonga_device_info = { - .asic_name = "tonga", - .gfx_target_version = 80002, - .max_pasid_bits = 16, - .max_no_of_hqd = 24, - .doorbell_size = 4, - .ih_ring_entry_size = 4 * sizeof(uint32_t), - .event_interrupt_class = &event_interrupt_class_cik, - .num_of_watch_points = 4, - .mqd_size_aligned = MQD_SIZE_ALIGNED, - .supports_cwsr = false, - .needs_iommu_device = false, - .needs_pci_atomics = true, - .num_sdma_queues_per_engine = 2, -}; - -static const struct kfd_device_info fiji_device_info = { - .asic_name = "fiji", - .gfx_target_version = 80003, - .max_pasid_bits = 16, - .max_no_of_hqd = 24, - .doorbell_size = 4, - .ih_ring_entry_size = 4 * sizeof(uint32_t), - .event_interrupt_class = &event_interrupt_class_cik, - .num_of_watch_points = 4, - .mqd_size_aligned = MQD_SIZE_ALIGNED, - .supports_cwsr = true, - .needs_iommu_device = false, - .needs_pci_atomics = true, - .num_sdma_queues_per_engine = 2, -}; - -static const struct kfd_device_info fiji_vf_device_info = { - .asic_name = "fiji", - .gfx_target_version = 80003, - .max_pasid_bits = 16, - .max_no_of_hqd = 24, - .doorbell_size = 4, - .ih_ring_entry_size = 4 * sizeof(uint32_t), - .event_interrupt_class = &event_interrupt_class_cik, - .num_of_watch_points = 4, - .mqd_size_aligned = MQD_SIZE_ALIGNED, - .supports_cwsr = true, - .needs_iommu_device = false, - .needs_pci_atomics = false, - .num_sdma_queues_per_engine = 2, -}; - - -static const struct kfd_device_info polaris10_device_info = { - .asic_name = "polaris10", - .gfx_target_version = 80003, - .max_pasid_bits = 16, - .max_no_of_hqd = 24, - .doorbell_size = 4, - .ih_ring_entry_size = 4 * sizeof(uint32_t), - .event_interrupt_class = &event_interrupt_class_cik, - .num_of_watch_points = 4, - .mqd_size_aligned = MQD_SIZE_ALIGNED, - .supports_cwsr = true, - .needs_iommu_device = false, - .needs_pci_atomics = true, - .num_sdma_queues_per_engine = 2, -}; - -static const struct kfd_device_info polaris10_vf_device_info = { - .asic_name = "polaris10", - .gfx_target_version = 80003, - .max_pasid_bits = 16, - .max_no_of_hqd = 24, - .doorbell_size = 4, - .ih_ring_entry_size = 4 * sizeof(uint32_t), - .event_interrupt_class = &event_interrupt_class_cik, - .num_of_watch_points = 4, - .mqd_size_aligned = MQD_SIZE_ALIGNED, - .supports_cwsr = true, - .needs_iommu_device = false, - .needs_pci_atomics = false, - .num_sdma_queues_per_engine = 2, -}; - -static const struct kfd_device_info polaris11_device_info = { - .asic_name = "polaris11", - .gfx_target_version = 80003, - .max_pasid_bits = 16, - .max_no_of_hqd = 24, - .doorbell_size = 4, - .ih_ring_entry_size = 4 * sizeof(uint32_t), - .event_interrupt_class = &event_interrupt_class_cik, - .num_of_watch_points = 4, - .mqd_size_aligned = MQD_SIZE_ALIGNED, - .supports_cwsr = true, - .needs_iommu_device = false, - .needs_pci_atomics = true, - .num_sdma_queues_per_engine = 2, -}; - -static const struct kfd_device_info polaris12_device_info = { - .asic_name = "polaris12", - .gfx_target_version = 80003, - .max_pasid_bits = 16, - .max_no_of_hqd = 24, - .doorbell_size = 4, - .ih_ring_entry_size = 4 * sizeof(uint32_t), - .event_interrupt_class = &event_interrupt_class_cik, - .num_of_watch_points = 4, - .mqd_size_aligned = MQD_SIZE_ALIGNED, - .supports_cwsr = true, - .needs_iommu_device = false, - .needs_pci_atomics = true, - .num_sdma_queues_per_engine = 2, -}; - -static const struct kfd_device_info vegam_device_info = { - .asic_name = "vegam", - .gfx_target_version = 80003, - .max_pasid_bits = 16, - .max_no_of_hqd = 24, - .doorbell_size = 4, - .ih_ring_entry_size = 4 * sizeof(uint32_t), - .event_interrupt_class = &event_interrupt_class_cik, - .num_of_watch_points = 4, - .mqd_size_aligned = MQD_SIZE_ALIGNED, - .supports_cwsr = true, - .needs_iommu_device = false, - .needs_pci_atomics = true, - .num_sdma_queues_per_engine = 2, -}; - -static const struct kfd_device_info vega10_device_info = { - .asic_name = "vega10", - .gfx_target_version = 90000, - .max_pasid_bits = 16, - .max_no_of_hqd = 24, - .doorbell_size = 8, - .ih_ring_entry_size = 8 * sizeof(uint32_t), - .event_interrupt_class = &event_interrupt_class_v9, - .num_of_watch_points = 4, - .mqd_size_aligned = MQD_SIZE_ALIGNED, - .supports_cwsr = true, - .needs_iommu_device = false, - .needs_pci_atomics = false, - .num_sdma_queues_per_engine = 2, -}; - -static const struct kfd_device_info vega10_vf_device_info = { - .asic_name = "vega10", - .gfx_target_version = 90000, - .max_pasid_bits = 16, - .max_no_of_hqd = 24, - .doorbell_size = 8, - .ih_ring_entry_size = 8 * sizeof(uint32_t), - .event_interrupt_class = &event_interrupt_class_v9, - .num_of_watch_points = 4, - .mqd_size_aligned = MQD_SIZE_ALIGNED, - .supports_cwsr = true, - .needs_iommu_device = false, - .needs_pci_atomics = false, - .num_sdma_queues_per_engine = 2, -}; - -static const struct kfd_device_info vega12_device_info = { - .asic_name = "vega12", - .gfx_target_version = 90004, - .max_pasid_bits = 16, - .max_no_of_hqd = 24, - .doorbell_size = 8, - .ih_ring_entry_size = 8 * sizeof(uint32_t), - .event_interrupt_class = &event_interrupt_class_v9, - .num_of_watch_points = 4, - .mqd_size_aligned = MQD_SIZE_ALIGNED, - .supports_cwsr = true, - .needs_iommu_device = false, - .needs_pci_atomics = false, - .num_sdma_queues_per_engine = 2, -}; - -static const struct kfd_device_info vega20_device_info = { - .asic_name = "vega20", - .gfx_target_version = 90006, - .max_pasid_bits = 16, - .max_no_of_hqd = 24, - .doorbell_size = 8, - .ih_ring_entry_size = 8 * sizeof(uint32_t), - .event_interrupt_class = &event_interrupt_class_v9, - .num_of_watch_points = 4, - .mqd_size_aligned = MQD_SIZE_ALIGNED, - .supports_cwsr = true, - .needs_iommu_device = false, - .needs_pci_atomics = false, - .num_sdma_queues_per_engine = 8, -}; - -static const struct kfd_device_info arcturus_device_info = { - .asic_name = "arcturus", - .gfx_target_version = 90008, - .max_pasid_bits = 16, - .max_no_of_hqd = 24, - .doorbell_size = 8, - .ih_ring_entry_size = 8 * sizeof(uint32_t), - .event_interrupt_class = &event_interrupt_class_v9, - .num_of_watch_points = 4, - .mqd_size_aligned = MQD_SIZE_ALIGNED, - .supports_cwsr = true, - .needs_iommu_device = false, - .needs_pci_atomics = false, - .num_sdma_queues_per_engine = 8, -}; - -static const struct kfd_device_info aldebaran_device_info = { - .asic_name = "aldebaran", - .gfx_target_version = 90010, - .max_pasid_bits = 16, - .max_no_of_hqd = 24, - .doorbell_size = 8, - .ih_ring_entry_size = 8 * sizeof(uint32_t), - .event_interrupt_class = &event_interrupt_class_v9, - .num_of_watch_points = 4, - .mqd_size_aligned = MQD_SIZE_ALIGNED, - .supports_cwsr = true, - .needs_iommu_device = false, - .needs_pci_atomics = false, - .num_sdma_queues_per_engine = 8, -}; - -static const struct kfd_device_info renoir_device_info = { - .asic_name = "renoir", - .gfx_target_version = 90012, - .max_pasid_bits = 16, - .max_no_of_hqd = 24, - .doorbell_size = 8, - .ih_ring_entry_size = 8 * sizeof(uint32_t), - .event_interrupt_class = &event_interrupt_class_v9, - .num_of_watch_points = 4, - .mqd_size_aligned = MQD_SIZE_ALIGNED, - .supports_cwsr = true, - .needs_iommu_device = false, - .needs_pci_atomics = false, - .num_sdma_queues_per_engine = 2, -}; - -static const struct kfd_device_info navi10_device_info = { - .asic_name = "navi10", - .gfx_target_version = 100100, - .max_pasid_bits = 16, - .max_no_of_hqd = 24, - .doorbell_size = 8, - .ih_ring_entry_size = 8 * sizeof(uint32_t), - .event_interrupt_class = &event_interrupt_class_v9, - .num_of_watch_points = 4, - .mqd_size_aligned = MQD_SIZE_ALIGNED, - .needs_iommu_device = false, - .supports_cwsr = true, - .needs_pci_atomics = true, - .no_atomic_fw_version = 145, - .num_sdma_queues_per_engine = 8, -}; - -static const struct kfd_device_info navi12_device_info = { - .asic_name = "navi12", - .gfx_target_version = 100101, - .max_pasid_bits = 16, - .max_no_of_hqd = 24, - .doorbell_size = 8, - .ih_ring_entry_size = 8 * sizeof(uint32_t), - .event_interrupt_class = &event_interrupt_class_v9, - .num_of_watch_points = 4, - .mqd_size_aligned = MQD_SIZE_ALIGNED, - .needs_iommu_device = false, - .supports_cwsr = true, - .needs_pci_atomics = true, - .no_atomic_fw_version = 145, - .num_sdma_queues_per_engine = 8, -}; - -static const struct kfd_device_info navi14_device_info = { - .asic_name = "navi14", - .gfx_target_version = 100102, - .max_pasid_bits = 16, - .max_no_of_hqd = 24, - .doorbell_size = 8, - .ih_ring_entry_size = 8 * sizeof(uint32_t), - .event_interrupt_class = &event_interrupt_class_v9, - .num_of_watch_points = 4, - .mqd_size_aligned = MQD_SIZE_ALIGNED, - .needs_iommu_device = false, - .supports_cwsr = true, - .needs_pci_atomics = true, - .no_atomic_fw_version = 145, - .num_sdma_queues_per_engine = 8, -}; - -static const struct kfd_device_info sienna_cichlid_device_info = { - .asic_name = "sienna_cichlid", - .gfx_target_version = 100300, - .max_pasid_bits = 16, - .max_no_of_hqd = 24, - .doorbell_size = 8, - .ih_ring_entry_size = 8 * sizeof(uint32_t), - .event_interrupt_class = &event_interrupt_class_v9, - .num_of_watch_points = 4, - .mqd_size_aligned = MQD_SIZE_ALIGNED, - .needs_iommu_device = false, - .supports_cwsr = true, - .needs_pci_atomics = true, - .no_atomic_fw_version = 92, - .num_sdma_queues_per_engine = 8, -}; - -static const struct kfd_device_info navy_flounder_device_info = { - .asic_name = "navy_flounder", - .gfx_target_version = 100301, - .max_pasid_bits = 16, - .max_no_of_hqd = 24, - .doorbell_size = 8, - .ih_ring_entry_size = 8 * sizeof(uint32_t), - .event_interrupt_class = &event_interrupt_class_v9, - .num_of_watch_points = 4, - .mqd_size_aligned = MQD_SIZE_ALIGNED, - .needs_iommu_device = false, - .supports_cwsr = true, - .needs_pci_atomics = true, - .no_atomic_fw_version = 92, - .num_sdma_queues_per_engine = 8, -}; - -static const struct kfd_device_info vangogh_device_info = { - .asic_name = "vangogh", - .gfx_target_version = 100303, - .max_pasid_bits = 16, - .max_no_of_hqd = 24, - .doorbell_size = 8, - .ih_ring_entry_size = 8 * sizeof(uint32_t), - .event_interrupt_class = &event_interrupt_class_v9, - .num_of_watch_points = 4, - .mqd_size_aligned = MQD_SIZE_ALIGNED, - .needs_iommu_device = false, - .supports_cwsr = true, - .needs_pci_atomics = true, - .no_atomic_fw_version = 92, - .num_sdma_queues_per_engine = 2, -}; - -static const struct kfd_device_info dimgrey_cavefish_device_info = { - .asic_name = "dimgrey_cavefish", - .gfx_target_version = 100302, - .max_pasid_bits = 16, - .max_no_of_hqd = 24, - .doorbell_size = 8, - .ih_ring_entry_size = 8 * sizeof(uint32_t), - .event_interrupt_class = &event_interrupt_class_v9, - .num_of_watch_points = 4, - .mqd_size_aligned = MQD_SIZE_ALIGNED, - .needs_iommu_device = false, - .supports_cwsr = true, - .needs_pci_atomics = true, - .no_atomic_fw_version = 92, - .num_sdma_queues_per_engine = 8, -}; - -static const struct kfd_device_info beige_goby_device_info = { - .asic_name = "beige_goby", - .gfx_target_version = 100304, - .max_pasid_bits = 16, - .max_no_of_hqd = 24, - .doorbell_size = 8, - .ih_ring_entry_size = 8 * sizeof(uint32_t), - .event_interrupt_class = &event_interrupt_class_v9, - .num_of_watch_points = 4, - .mqd_size_aligned = MQD_SIZE_ALIGNED, - .needs_iommu_device = false, - .supports_cwsr = true, - .needs_pci_atomics = true, - .no_atomic_fw_version = 92, - .num_sdma_queues_per_engine = 8, -}; - -static const struct kfd_device_info yellow_carp_device_info = { - .asic_name = "yellow_carp", - .gfx_target_version = 100305, - .max_pasid_bits = 16, - .max_no_of_hqd = 24, - .doorbell_size = 8, - .ih_ring_entry_size = 8 * sizeof(uint32_t), - .event_interrupt_class = &event_interrupt_class_v9, - .num_of_watch_points = 4, - .mqd_size_aligned = MQD_SIZE_ALIGNED, - .needs_iommu_device = false, - .supports_cwsr = true, - .needs_pci_atomics = true, - .no_atomic_fw_version = 92, - .num_sdma_queues_per_engine = 2, -}; - -static const struct kfd_device_info cyan_skillfish_device_info = { - .asic_name = "cyan_skillfish", - .gfx_target_version = 100103, - .max_pasid_bits = 16, - .max_no_of_hqd = 24, - .doorbell_size = 8, - .ih_ring_entry_size = 8 * sizeof(uint32_t), - .event_interrupt_class = &event_interrupt_class_v9, - .num_of_watch_points = 4, - .mqd_size_aligned = MQD_SIZE_ALIGNED, - .needs_iommu_device = false, - .supports_cwsr = true, - .needs_pci_atomics = true, - .num_sdma_queues_per_engine = 8, -}; - static int kfd_gtt_sa_init(struct kfd_dev *kfd, unsigned int buf_size, unsigned int chunk_size); static void kfd_gtt_sa_fini(struct kfd_dev *kfd); static int kfd_resume(struct kfd_dev *kfd); +static void kfd_device_info_init(struct kfd_dev *kfd, + bool vf, uint32_t gfx_target_version) +{ + uint32_t gc_version = KFD_GC_VERSION(kfd); + uint32_t sdma_version = kfd->adev->ip_versions[SDMA0_HWIP][0]; + uint32_t asic_type = kfd->adev->asic_type; + + kfd->device_info.max_pasid_bits = 16; + kfd->device_info.max_no_of_hqd = 24; + kfd->device_info.num_of_watch_points = 4; + kfd->device_info.mqd_size_aligned = MQD_SIZE_ALIGNED; + kfd->device_info.gfx_target_version = gfx_target_version; + + if (KFD_IS_SOC15(kfd)) { + kfd->device_info.doorbell_size = 8; + kfd->device_info.ih_ring_entry_size = 8 * sizeof(uint32_t); + kfd->device_info.event_interrupt_class = &event_interrupt_class_v9; + kfd->device_info.supports_cwsr = true; + + if ((sdma_version >= IP_VERSION(4, 0, 0) && + sdma_version <= IP_VERSION(4, 2, 0)) || + sdma_version == IP_VERSION(5, 2, 1) || + sdma_version == IP_VERSION(5, 2, 3)) + kfd->device_info.num_sdma_queues_per_engine = 2; + else + kfd->device_info.num_sdma_queues_per_engine = 8; + + /* Raven */ + if (gc_version == IP_VERSION(9, 1, 0) || + gc_version == IP_VERSION(9, 2, 2)) + kfd->device_info.needs_iommu_device = true; + + if (gc_version < IP_VERSION(11, 0, 0)) { + /* Navi2x+, Navi1x+ */ + if (gc_version >= IP_VERSION(10, 3, 0)) + kfd->device_info.no_atomic_fw_version = 92; + else if (gc_version >= IP_VERSION(10, 1, 1)) + kfd->device_info.no_atomic_fw_version = 145; + + /* Navi1x+ */ + if (gc_version >= IP_VERSION(10, 1, 1)) + kfd->device_info.needs_pci_atomics = true; + } + } else { + kfd->device_info.doorbell_size = 4; + kfd->device_info.ih_ring_entry_size = 4 * sizeof(uint32_t); + kfd->device_info.event_interrupt_class = &event_interrupt_class_cik; + kfd->device_info.num_sdma_queues_per_engine = 2; + + if (asic_type != CHIP_KAVERI && + asic_type != CHIP_HAWAII && + asic_type != CHIP_TONGA) + kfd->device_info.supports_cwsr = true; + + if (asic_type == CHIP_KAVERI || + asic_type == CHIP_CARRIZO) + kfd->device_info.needs_iommu_device = true; + + if (asic_type != CHIP_HAWAII && !vf) + kfd->device_info.needs_pci_atomics = true; + } +} + struct kfd_dev *kgd2kfd_probe(struct amdgpu_device *adev, bool vf) { - struct kfd_dev *kfd; - const struct kfd_device_info *device_info; - const struct kfd2kgd_calls *f2g; + struct kfd_dev *kfd = NULL; + const struct kfd2kgd_calls *f2g = NULL; struct pci_dev *pdev = adev->pdev; + uint32_t gfx_target_version = 0; switch (adev->asic_type) { #ifdef KFD_SUPPORT_IOMMU_V2 #ifdef CONFIG_DRM_AMDGPU_CIK case CHIP_KAVERI: - if (vf) - device_info = NULL; - else - device_info = &kaveri_device_info; - f2g = &gfx_v7_kfd2kgd; + gfx_target_version = 70000; + if (!vf) + f2g = &gfx_v7_kfd2kgd; break; #endif case CHIP_CARRIZO: - if (vf) - device_info = NULL; - else - device_info = &carrizo_device_info; - f2g = &gfx_v8_kfd2kgd; + gfx_target_version = 80001; + if (!vf) + f2g = &gfx_v8_kfd2kgd; break; #endif #ifdef CONFIG_DRM_AMDGPU_CIK case CHIP_HAWAII: - if (vf) - device_info = NULL; - else - device_info = &hawaii_device_info; - f2g = &gfx_v7_kfd2kgd; + gfx_target_version = 70001; + if (!amdgpu_exp_hw_support) + pr_info( + "KFD support on Hawaii is experimental. See modparam exp_hw_support\n" + ); + else if (!vf) + f2g = &gfx_v7_kfd2kgd; break; #endif case CHIP_TONGA: - if (vf) - device_info = NULL; - else - device_info = &tonga_device_info; - f2g = &gfx_v8_kfd2kgd; + gfx_target_version = 80002; + if (!vf) + f2g = &gfx_v8_kfd2kgd; break; case CHIP_FIJI: - if (vf) - device_info = &fiji_vf_device_info; - else - device_info = &fiji_device_info; + gfx_target_version = 80003; f2g = &gfx_v8_kfd2kgd; break; case CHIP_POLARIS10: - if (vf) - device_info = &polaris10_vf_device_info; - else - device_info = &polaris10_device_info; + gfx_target_version = 80003; f2g = &gfx_v8_kfd2kgd; break; case CHIP_POLARIS11: - if (vf) - device_info = NULL; - else - device_info = &polaris11_device_info; - f2g = &gfx_v8_kfd2kgd; + gfx_target_version = 80003; + if (!vf) + f2g = &gfx_v8_kfd2kgd; break; case CHIP_POLARIS12: - if (vf) - device_info = NULL; - else - device_info = &polaris12_device_info; - f2g = &gfx_v8_kfd2kgd; + gfx_target_version = 80003; + if (!vf) + f2g = &gfx_v8_kfd2kgd; break; case CHIP_VEGAM: - if (vf) - device_info = NULL; - else - device_info = &vegam_device_info; - f2g = &gfx_v8_kfd2kgd; + gfx_target_version = 80003; + if (!vf) + f2g = &gfx_v8_kfd2kgd; break; default: switch (adev->ip_versions[GC_HWIP][0]) { + /* Vega 10 */ case IP_VERSION(9, 0, 1): - if (vf) - device_info = &vega10_vf_device_info; - else - device_info = &vega10_device_info; + gfx_target_version = 90000; f2g = &gfx_v9_kfd2kgd; break; #ifdef KFD_SUPPORT_IOMMU_V2 + /* Raven */ case IP_VERSION(9, 1, 0): case IP_VERSION(9, 2, 2): - if (vf) - device_info = NULL; - else - device_info = &raven_device_info; - f2g = &gfx_v9_kfd2kgd; + gfx_target_version = 90002; + if (!vf) + f2g = &gfx_v9_kfd2kgd; break; #endif + /* Vega12 */ case IP_VERSION(9, 2, 1): - if (vf) - device_info = NULL; - else - device_info = &vega12_device_info; - f2g = &gfx_v9_kfd2kgd; + gfx_target_version = 90004; + if (!vf) + f2g = &gfx_v9_kfd2kgd; break; + /* Renoir */ case IP_VERSION(9, 3, 0): - if (vf) - device_info = NULL; - else - device_info = &renoir_device_info; - f2g = &gfx_v9_kfd2kgd; + gfx_target_version = 90012; + if (!vf) + f2g = &gfx_v9_kfd2kgd; break; + /* Vega20 */ case IP_VERSION(9, 4, 0): - if (vf) - device_info = NULL; - else - device_info = &vega20_device_info; - f2g = &gfx_v9_kfd2kgd; + gfx_target_version = 90006; + if (!vf) + f2g = &gfx_v9_kfd2kgd; break; + /* Arcturus */ case IP_VERSION(9, 4, 1): - device_info = &arcturus_device_info; + gfx_target_version = 90008; f2g = &arcturus_kfd2kgd; break; + /* Aldebaran */ case IP_VERSION(9, 4, 2): - device_info = &aldebaran_device_info; + gfx_target_version = 90010; f2g = &aldebaran_kfd2kgd; break; + /* Navi10 */ case IP_VERSION(10, 1, 10): - if (vf) - device_info = NULL; - else - device_info = &navi10_device_info; - f2g = &gfx_v10_kfd2kgd; + gfx_target_version = 100100; + if (!vf) + f2g = &gfx_v10_kfd2kgd; break; + /* Navi12 */ case IP_VERSION(10, 1, 2): - device_info = &navi12_device_info; + gfx_target_version = 100101; f2g = &gfx_v10_kfd2kgd; break; + /* Navi14 */ case IP_VERSION(10, 1, 1): - if (vf) - device_info = NULL; - else - device_info = &navi14_device_info; - f2g = &gfx_v10_kfd2kgd; + gfx_target_version = 100102; + if (!vf) + f2g = &gfx_v10_kfd2kgd; break; + /* Cyan Skillfish */ case IP_VERSION(10, 1, 3): - if (vf) - device_info = NULL; - else - device_info = &cyan_skillfish_device_info; - f2g = &gfx_v10_kfd2kgd; + gfx_target_version = 100103; + if (!vf) + f2g = &gfx_v10_kfd2kgd; break; + /* Sienna Cichlid */ case IP_VERSION(10, 3, 0): - device_info = &sienna_cichlid_device_info; + gfx_target_version = 100300; f2g = &gfx_v10_3_kfd2kgd; break; + /* Navy Flounder */ case IP_VERSION(10, 3, 2): - device_info = &navy_flounder_device_info; + gfx_target_version = 100301; f2g = &gfx_v10_3_kfd2kgd; break; + /* Van Gogh */ case IP_VERSION(10, 3, 1): - if (vf) - device_info = NULL; - else - device_info = &vangogh_device_info; - f2g = &gfx_v10_3_kfd2kgd; + gfx_target_version = 100303; + if (!vf) + f2g = &gfx_v10_3_kfd2kgd; break; + /* Dimgrey Cavefish */ case IP_VERSION(10, 3, 4): - device_info = &dimgrey_cavefish_device_info; + gfx_target_version = 100302; f2g = &gfx_v10_3_kfd2kgd; break; + /* Beige Goby */ case IP_VERSION(10, 3, 5): - device_info = &beige_goby_device_info; + gfx_target_version = 100304; f2g = &gfx_v10_3_kfd2kgd; break; + /* Yellow Carp */ case IP_VERSION(10, 3, 3): - if (vf) - device_info = NULL; - else - device_info = &yellow_carp_device_info; - f2g = &gfx_v10_3_kfd2kgd; + gfx_target_version = 100305; + if (!vf) + f2g = &gfx_v10_3_kfd2kgd; break; default: - return NULL; + break; } break; } - if (!device_info || !f2g) { + if (!f2g) { if (adev->ip_versions[GC_HWIP][0]) dev_err(kfd_device, "GC IP %06x %s not supported in kfd\n", adev->ip_versions[GC_HWIP][0], vf ? "VF" : ""); @@ -741,7 +303,7 @@ struct kfd_dev *kgd2kfd_probe(struct amdgpu_device *adev, bool vf) return NULL; kfd->adev = adev; - kfd->device_info = device_info; + kfd_device_info_init(kfd, vf, gfx_target_version); kfd->pdev = pdev; kfd->init_complete = false; kfd->kfd2kgd = f2g; @@ -760,7 +322,7 @@ struct kfd_dev *kgd2kfd_probe(struct amdgpu_device *adev, bool vf) static void kfd_cwsr_init(struct kfd_dev *kfd) { - if (cwsr_enable && kfd->device_info->supports_cwsr) { + if (cwsr_enable && kfd->device_info.supports_cwsr) { if (KFD_GC_VERSION(kfd) < IP_VERSION(9, 0, 1)) { BUILD_BUG_ON(sizeof(cwsr_trap_gfx8_hex) > PAGE_SIZE); kfd->cwsr_isa = cwsr_trap_gfx8_hex; @@ -844,14 +406,14 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd, */ kfd->pci_atomic_requested = amdgpu_amdkfd_have_atomics_support(kfd->adev); if (!kfd->pci_atomic_requested && - kfd->device_info->needs_pci_atomics && - (!kfd->device_info->no_atomic_fw_version || - kfd->mec_fw_version < kfd->device_info->no_atomic_fw_version)) { + kfd->device_info.needs_pci_atomics && + (!kfd->device_info.no_atomic_fw_version || + kfd->mec_fw_version < kfd->device_info.no_atomic_fw_version)) { dev_info(kfd_device, "skipped device %x:%x, PCI rejects atomics %d<%d\n", kfd->pdev->vendor, kfd->pdev->device, kfd->mec_fw_version, - kfd->device_info->no_atomic_fw_version); + kfd->device_info.no_atomic_fw_version); return false; } @@ -868,7 +430,7 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd, /* calculate max size of mqds needed for queues */ size = max_num_of_queues_per_device * - kfd->device_info->mqd_size_aligned; + kfd->device_info.mqd_size_aligned; /* * calculate max size of runlist packet. @@ -1143,7 +705,7 @@ void kgd2kfd_interrupt(struct kfd_dev *kfd, const void *ih_ring_entry) if (!kfd->init_complete) return; - if (kfd->device_info->ih_ring_entry_size > sizeof(patched_ihre)) { + if (kfd->device_info.ih_ring_entry_size > sizeof(patched_ihre)) { dev_err_once(kfd_device, "Ring entry too small\n"); return; } diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c index 2af2b3268171..dd0b952f0173 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c @@ -108,13 +108,13 @@ static unsigned int get_num_all_sdma_engines(struct device_queue_manager *dqm) unsigned int get_num_sdma_queues(struct device_queue_manager *dqm) { return kfd_get_num_sdma_engines(dqm->dev) * - dqm->dev->device_info->num_sdma_queues_per_engine; + dqm->dev->device_info.num_sdma_queues_per_engine; } unsigned int get_num_xgmi_sdma_queues(struct device_queue_manager *dqm) { return kfd_get_num_xgmi_sdma_engines(dqm->dev) * - dqm->dev->device_info->num_sdma_queues_per_engine; + dqm->dev->device_info.num_sdma_queues_per_engine; } void program_sh_mem_settings(struct device_queue_manager *dqm, @@ -1838,7 +1838,7 @@ static int allocate_hiq_sdma_mqd(struct device_queue_manager *dqm) struct kfd_mem_obj *mem_obj = &dqm->hiq_sdma_mqd; uint32_t size = dqm->mqd_mgrs[KFD_MQD_TYPE_SDMA]->mqd_size * get_num_all_sdma_engines(dqm) * - dev->device_info->num_sdma_queues_per_engine + + dev->device_info.num_sdma_queues_per_engine + dqm->mqd_mgrs[KFD_MQD_TYPE_HIQ]->mqd_size; retval = amdgpu_amdkfd_alloc_gtt_mem(dev->adev, size, @@ -2082,7 +2082,7 @@ int dqm_debugfs_hqds(struct seq_file *m, void *data) for (pipe = 0; pipe < get_num_all_sdma_engines(dqm); pipe++) { for (queue = 0; - queue < dqm->dev->device_info->num_sdma_queues_per_engine; + queue < dqm->dev->device_info.num_sdma_queues_per_engine; queue++) { r = dqm->dev->kfd2kgd->hqd_sdma_dump( dqm->dev->adev, pipe, queue, &dump, &n_regs); diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_doorbell.c b/drivers/gpu/drm/amd/amdkfd/kfd_doorbell.c index 768d153acff4..0dbcf54657ed 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_doorbell.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_doorbell.c @@ -48,7 +48,7 @@ /* # of doorbell bytes allocated for each process. */ size_t kfd_doorbell_process_slice(struct kfd_dev *kfd) { - return roundup(kfd->device_info->doorbell_size * + return roundup(kfd->device_info.doorbell_size * KFD_MAX_NUM_OF_QUEUES_PER_PROCESS, PAGE_SIZE); } @@ -180,7 +180,7 @@ void __iomem *kfd_get_kernel_doorbell(struct kfd_dev *kfd, if (inx >= KFD_MAX_NUM_OF_QUEUES_PER_PROCESS) return NULL; - inx *= kfd->device_info->doorbell_size / sizeof(u32); + inx *= kfd->device_info.doorbell_size / sizeof(u32); /* * Calculating the kernel doorbell offset using the first @@ -201,7 +201,7 @@ void kfd_release_kernel_doorbell(struct kfd_dev *kfd, u32 __iomem *db_addr) unsigned int inx; inx = (unsigned int)(db_addr - kfd->doorbell_kernel_ptr) - * sizeof(u32) / kfd->device_info->doorbell_size; + * sizeof(u32) / kfd->device_info.doorbell_size; mutex_lock(&kfd->doorbell_mutex); __clear_bit(inx, kfd->doorbell_available_index); @@ -239,7 +239,7 @@ unsigned int kfd_get_doorbell_dw_offset_in_bar(struct kfd_dev *kfd, return kfd->doorbell_base_dw_offset + pdd->doorbell_index * kfd_doorbell_process_slice(kfd) / sizeof(u32) + - doorbell_id * kfd->device_info->doorbell_size / sizeof(u32); + doorbell_id * kfd->device_info.doorbell_size / sizeof(u32); } uint64_t kfd_get_number_elems(struct kfd_dev *kfd) diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c index 20512a4e9a91..deb64168c9e8 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c @@ -135,7 +135,7 @@ static bool event_interrupt_isr_v9(struct kfd_dev *dev, *patched_flag = true; memcpy(patched_ihre, ih_ring_entry, - dev->device_info->ih_ring_entry_size); + dev->device_info.ih_ring_entry_size); pasid = dev->dqm->vmid_pasid[vmid]; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c b/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c index bc47f6a44456..81887c2013c9 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_interrupt.c @@ -54,7 +54,7 @@ int kfd_interrupt_init(struct kfd_dev *kfd) int r; r = kfifo_alloc(&kfd->ih_fifo, - KFD_IH_NUM_ENTRIES * kfd->device_info->ih_ring_entry_size, + KFD_IH_NUM_ENTRIES * kfd->device_info.ih_ring_entry_size, GFP_KERNEL); if (r) { dev_err(kfd_chardev(), "Failed to allocate IH fifo\n"); @@ -114,8 +114,8 @@ bool enqueue_ih_ring_entry(struct kfd_dev *kfd, const void *ih_ring_entry) int count; count = kfifo_in(&kfd->ih_fifo, ih_ring_entry, - kfd->device_info->ih_ring_entry_size); - if (count != kfd->device_info->ih_ring_entry_size) { + kfd->device_info.ih_ring_entry_size); + if (count != kfd->device_info.ih_ring_entry_size) { dev_err_ratelimited(kfd_chardev(), "Interrupt ring overflow, dropping interrupt %d\n", count); @@ -133,11 +133,11 @@ static bool dequeue_ih_ring_entry(struct kfd_dev *kfd, void *ih_ring_entry) int count; count = kfifo_out(&kfd->ih_fifo, ih_ring_entry, - kfd->device_info->ih_ring_entry_size); + kfd->device_info.ih_ring_entry_size); - WARN_ON(count && count != kfd->device_info->ih_ring_entry_size); + WARN_ON(count && count != kfd->device_info.ih_ring_entry_size); - return count == kfd->device_info->ih_ring_entry_size; + return count == kfd->device_info.ih_ring_entry_size; } static void interrupt_wq(struct work_struct *work) @@ -146,13 +146,13 @@ static void interrupt_wq(struct work_struct *work) interrupt_work); uint32_t ih_ring_entry[KFD_MAX_RING_ENTRY_SIZE]; - if (dev->device_info->ih_ring_entry_size > sizeof(ih_ring_entry)) { + if (dev->device_info.ih_ring_entry_size > sizeof(ih_ring_entry)) { dev_err_once(kfd_chardev(), "Ring entry too small\n"); return; } while (dequeue_ih_ring_entry(dev, ih_ring_entry)) - dev->device_info->event_interrupt_class->interrupt_wq(dev, + dev->device_info.event_interrupt_class->interrupt_wq(dev, ih_ring_entry); } @@ -163,7 +163,7 @@ bool interrupt_is_wanted(struct kfd_dev *dev, /* integer and bitwise OR so there is no boolean short-circuiting */ unsigned int wanted = 0; - wanted |= dev->device_info->event_interrupt_class->interrupt_isr(dev, + wanted |= dev->device_info.event_interrupt_class->interrupt_isr(dev, ih_ring_entry, patched_ihre, flag); return wanted != 0; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_iommu.c b/drivers/gpu/drm/amd/amdkfd/kfd_iommu.c index 73f2257acc23..66ad8d0b8f7f 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_iommu.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_iommu.c @@ -89,7 +89,7 @@ int kfd_iommu_device_init(struct kfd_dev *kfd) } pasid_limit = min_t(unsigned int, - (unsigned int)(1 << kfd->device_info->max_pasid_bits), + (unsigned int)(1 << kfd->device_info.max_pasid_bits), iommu_info.max_pasids); if (!kfd_set_pasid_limit(pasid_limit)) { diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c b/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c index 406479a369a9..16f8bc4ca7f6 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c @@ -111,7 +111,7 @@ static bool kq_initialize(struct kernel_queue *kq, struct kfd_dev *dev, kq->rptr_kernel = kq->rptr_mem->cpu_ptr; kq->rptr_gpu_addr = kq->rptr_mem->gpu_addr; - retval = kfd_gtt_sa_allocate(dev, dev->device_info->doorbell_size, + retval = kfd_gtt_sa_allocate(dev, dev->device_info.doorbell_size, &kq->wptr_mem); if (retval != 0) @@ -297,7 +297,7 @@ void kq_submit_packet(struct kernel_queue *kq) } pr_debug("\n"); #endif - if (kq->dev->device_info->doorbell_size == 8) { + if (kq->dev->device_info.doorbell_size == 8) { *kq->wptr64_kernel = kq->pending_wptr64; write_kernel_doorbell64(kq->queue->properties.doorbell_ptr, kq->pending_wptr64); @@ -310,7 +310,7 @@ void kq_submit_packet(struct kernel_queue *kq) void kq_rollback_packet(struct kernel_queue *kq) { - if (kq->dev->device_info->doorbell_size == 8) { + if (kq->dev->device_info.doorbell_size == 8) { kq->pending_wptr64 = *kq->wptr64_kernel; kq->pending_wptr = *kq->wptr_kernel % (kq->queue->properties.queue_size / 4); diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c index d84cec0022b1..48c2f2b6e217 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c @@ -108,8 +108,8 @@ error_free: * svm_migrate_copy_memory_gart - sdma copy data between ram and vram * * @adev: amdgpu device the sdma ring running - * @src: source page address array - * @dst: destination page address array + * @sys: system DMA pointer to be copied + * @vram: vram destination DMA pointer * @npages: number of pages to copy * @direction: enum MIGRATION_COPY_DIR * @mfence: output, sdma fence to signal after sdma is done diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c index 7b4118915bf6..e2825ad4d699 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c @@ -71,7 +71,7 @@ struct kfd_mem_obj *allocate_sdma_mqd(struct kfd_dev *dev, return NULL; offset = (q->sdma_engine_id * - dev->device_info->num_sdma_queues_per_engine + + dev->device_info.num_sdma_queues_per_engine + q->sdma_queue_id) * dev->dqm->mqd_mgrs[KFD_MQD_TYPE_SDMA]->mqd_size; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_vi.c b/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_vi.c index 08442e7d9944..3c0658e32e93 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_vi.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_vi.c @@ -110,8 +110,8 @@ static int pm_runlist_vi(struct packet_manager *pm, uint32_t *buffer, return 0; } -int pm_set_resources_vi(struct packet_manager *pm, uint32_t *buffer, - struct scheduling_resources *res) +static int pm_set_resources_vi(struct packet_manager *pm, uint32_t *buffer, + struct scheduling_resources *res) { struct pm4_mes_set_resources *packet; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h index 7ea528941951..0c3f911e3bf4 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h +++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h @@ -195,7 +195,6 @@ struct kfd_event_interrupt_class { }; struct kfd_device_info { - const char *asic_name; uint32_t gfx_target_version; const struct kfd_event_interrupt_class *event_interrupt_class; unsigned int max_pasid_bits; @@ -231,7 +230,7 @@ struct kfd_vmid_info { struct kfd_dev { struct amdgpu_device *adev; - const struct kfd_device_info *device_info; + struct kfd_device_info device_info; struct pci_dev *pdev; struct drm_device *ddev; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c index d4c8a6948a9f..f1930ff2c74a 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c @@ -251,14 +251,13 @@ cleanup: } /** - * @kfd_get_cu_occupancy - Collect number of waves in-flight on this device + * kfd_get_cu_occupancy - Collect number of waves in-flight on this device * by current process. Translates acquired wave count into number of compute units * that are occupied. * - * @atr: Handle of attribute that allows reporting of wave count. The attribute + * @attr: Handle of attribute that allows reporting of wave count. The attribute * handle encapsulates GPU device it is associated with, thereby allowing collection * of waves in flight, etc - * * @buffer: Handle of user provided buffer updated with wave count * * Return: Number of bytes written to user buffer or an error value @@ -1011,7 +1010,7 @@ static void kfd_process_destroy_pdds(struct kfd_process *p) free_pages((unsigned long)pdd->qpd.cwsr_kaddr, get_order(KFD_CWSR_TBA_TMA_SIZE)); - kfree(pdd->qpd.doorbell_bitmap); + bitmap_free(pdd->qpd.doorbell_bitmap); idr_destroy(&pdd->alloc_idr); kfd_free_process_doorbells(pdd->dev, pdd->doorbell_index); @@ -1433,9 +1432,8 @@ static int init_doorbell_bitmap(struct qcm_process_device *qpd, if (!KFD_IS_SOC15(dev)) return 0; - qpd->doorbell_bitmap = - kzalloc(DIV_ROUND_UP(KFD_MAX_NUM_OF_QUEUES_PER_PROCESS, - BITS_PER_BYTE), GFP_KERNEL); + qpd->doorbell_bitmap = bitmap_zalloc(KFD_MAX_NUM_OF_QUEUES_PER_PROCESS, + GFP_KERNEL); if (!qpd->doorbell_bitmap) return -ENOMEM; @@ -1447,9 +1445,9 @@ static int init_doorbell_bitmap(struct qcm_process_device *qpd, for (i = 0; i < KFD_MAX_NUM_OF_QUEUES_PER_PROCESS / 2; i++) { if (i >= range_start && i <= range_end) { - set_bit(i, qpd->doorbell_bitmap); - set_bit(i + KFD_QUEUE_DOORBELL_MIRROR_OFFSET, - qpd->doorbell_bitmap); + __set_bit(i, qpd->doorbell_bitmap); + __set_bit(i + KFD_QUEUE_DOORBELL_MIRROR_OFFSET, + qpd->doorbell_bitmap); } } diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c index 4f8464658daf..5e5c84a8e1ef 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c @@ -135,9 +135,8 @@ void kfd_process_dequeue_from_all_devices(struct kfd_process *p) int pqm_init(struct process_queue_manager *pqm, struct kfd_process *p) { INIT_LIST_HEAD(&pqm->queues); - pqm->queue_slot_bitmap = - kzalloc(DIV_ROUND_UP(KFD_MAX_NUM_OF_QUEUES_PER_PROCESS, - BITS_PER_BYTE), GFP_KERNEL); + pqm->queue_slot_bitmap = bitmap_zalloc(KFD_MAX_NUM_OF_QUEUES_PER_PROCESS, + GFP_KERNEL); if (!pqm->queue_slot_bitmap) return -ENOMEM; pqm->process = p; @@ -159,7 +158,7 @@ void pqm_uninit(struct process_queue_manager *pqm) kfree(pqn); } - kfree(pqm->queue_slot_bitmap); + bitmap_free(pqm->queue_slot_bitmap); pqm->queue_slot_bitmap = NULL; } @@ -220,7 +219,7 @@ int pqm_create_queue(struct process_queue_manager *pqm, * Hence we also check the type as well */ if ((pdd->qpd.is_debug) || (type == KFD_QUEUE_TYPE_DIQ)) - max_queues = dev->device_info->max_no_of_hqd/2; + max_queues = dev->device_info.max_no_of_hqd/2; if (pdd->qpd.queue_count >= max_queues) return -ENOSPC; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c index 755265f6c53b..7e92dcea4ce8 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c @@ -704,6 +704,61 @@ svm_range_apply_attrs(struct kfd_process *p, struct svm_range *prange, } } +static bool +svm_range_is_same_attrs(struct kfd_process *p, struct svm_range *prange, + uint32_t nattr, struct kfd_ioctl_svm_attribute *attrs) +{ + uint32_t i; + int gpuidx; + + for (i = 0; i < nattr; i++) { + switch (attrs[i].type) { + case KFD_IOCTL_SVM_ATTR_PREFERRED_LOC: + if (prange->preferred_loc != attrs[i].value) + return false; + break; + case KFD_IOCTL_SVM_ATTR_PREFETCH_LOC: + /* Prefetch should always trigger a migration even + * if the value of the attribute didn't change. + */ + return false; + case KFD_IOCTL_SVM_ATTR_ACCESS: + case KFD_IOCTL_SVM_ATTR_ACCESS_IN_PLACE: + case KFD_IOCTL_SVM_ATTR_NO_ACCESS: + gpuidx = kfd_process_gpuidx_from_gpuid(p, + attrs[i].value); + if (attrs[i].type == KFD_IOCTL_SVM_ATTR_NO_ACCESS) { + if (test_bit(gpuidx, prange->bitmap_access) || + test_bit(gpuidx, prange->bitmap_aip)) + return false; + } else if (attrs[i].type == KFD_IOCTL_SVM_ATTR_ACCESS) { + if (!test_bit(gpuidx, prange->bitmap_access)) + return false; + } else { + if (!test_bit(gpuidx, prange->bitmap_aip)) + return false; + } + break; + case KFD_IOCTL_SVM_ATTR_SET_FLAGS: + if ((prange->flags & attrs[i].value) != attrs[i].value) + return false; + break; + case KFD_IOCTL_SVM_ATTR_CLR_FLAGS: + if ((prange->flags & attrs[i].value) != 0) + return false; + break; + case KFD_IOCTL_SVM_ATTR_GRANULARITY: + if (prange->granularity != attrs[i].value) + return false; + break; + default: + WARN_ONCE(1, "svm_range_check_attrs wasn't called?"); + } + } + + return true; +} + /** * svm_range_debug_dump - print all range information from svms * @svms: svm range list header @@ -741,14 +796,6 @@ static void svm_range_debug_dump(struct svm_range_list *svms) } } -static bool -svm_range_is_same_attrs(struct svm_range *old, struct svm_range *new) -{ - return (old->prefetch_loc == new->prefetch_loc && - old->flags == new->flags && - old->granularity == new->granularity); -} - static int svm_range_split_array(void *ppnew, void *ppold, size_t size, uint64_t old_start, uint64_t old_n, @@ -941,7 +988,7 @@ svm_range_split(struct svm_range *prange, uint64_t start, uint64_t last, } static int -svm_range_split_tail(struct svm_range *prange, struct svm_range *new, +svm_range_split_tail(struct svm_range *prange, uint64_t new_last, struct list_head *insert_list) { struct svm_range *tail; @@ -953,7 +1000,7 @@ svm_range_split_tail(struct svm_range *prange, struct svm_range *new, } static int -svm_range_split_head(struct svm_range *prange, struct svm_range *new, +svm_range_split_head(struct svm_range *prange, uint64_t new_start, struct list_head *insert_list) { struct svm_range *head; @@ -1169,7 +1216,6 @@ svm_range_map_to_gpu(struct amdgpu_device *adev, struct amdgpu_vm *vm, unsigned long npages, bool readonly, dma_addr_t *dma_addr, struct amdgpu_device *bo_adev, struct dma_fence **fence) { - struct amdgpu_bo_va bo_va; bool table_freed = false; uint64_t pte_flags; unsigned long last_start; @@ -1182,9 +1228,6 @@ svm_range_map_to_gpu(struct amdgpu_device *adev, struct amdgpu_vm *vm, pr_debug("svms 0x%p [0x%lx 0x%lx] readonly %d\n", prange->svms, last_start, last_start + npages - 1, readonly); - if (prange->svm_bo && prange->ttm_res) - bo_va.is_xgmi = amdgpu_xgmi_same_hive(adev, bo_adev); - for (i = offset; i < offset + npages; i++) { last_domain = dma_addr[i] & SVM_RANGE_VRAM_DOMAIN; dma_addr[i] &= ~SVM_RANGE_VRAM_DOMAIN; @@ -1650,6 +1693,10 @@ out_reschedule: /** * svm_range_evict - evict svm range + * @prange: svm range structure + * @mm: current process mm_struct + * @start: starting process queue number + * @last: last process queue number * * Stop all queues of the process to ensure GPU doesn't access the memory, then * return to let CPU evict the buffer and proceed CPU pagetable update. @@ -1754,46 +1801,49 @@ static struct svm_range *svm_range_clone(struct svm_range *old) } /** - * svm_range_handle_overlap - split overlap ranges - * @svms: svm range list header - * @new: range added with this attributes - * @start: range added start address, in pages - * @last: range last address, in pages - * @update_list: output, the ranges attributes are updated. For set_attr, this - * will do validation and map to GPUs. For unmap, this will be - * removed and unmap from GPUs - * @insert_list: output, the ranges will be inserted into svms, attributes are - * not changes. For set_attr, this will add into svms. - * @remove_list:output, the ranges will be removed from svms - * @left: the remaining range after overlap, For set_attr, this will be added - * as new range. + * svm_range_add - add svm range and handle overlap + * @p: the range add to this process svms + * @start: page size aligned + * @size: page size aligned + * @nattr: number of attributes + * @attrs: array of attributes + * @update_list: output, the ranges need validate and update GPU mapping + * @insert_list: output, the ranges need insert to svms + * @remove_list: output, the ranges are replaced and need remove from svms * - * Total have 5 overlap cases. + * Check if the virtual address range has overlap with any existing ranges, + * split partly overlapping ranges and add new ranges in the gaps. All changes + * should be applied to the range_list and interval tree transactionally. If + * any range split or allocation fails, the entire update fails. Therefore any + * existing overlapping svm_ranges are cloned and the original svm_ranges left + * unchanged. * - * This function handles overlap of an address interval with existing - * struct svm_ranges for applying new attributes. This may require - * splitting existing struct svm_ranges. All changes should be applied to - * the range_list and interval tree transactionally. If any split operation - * fails, the entire update fails. Therefore the existing overlapping - * svm_ranges are cloned and the original svm_ranges left unchanged. If the - * transaction succeeds, the modified clones are added and the originals - * freed. Otherwise the clones are removed and the old svm_ranges remain. + * If the transaction succeeds, the caller can update and insert clones and + * new ranges, then free the originals. * - * Context: The caller must hold svms->lock + * Otherwise the caller can free the clones and new ranges, while the old + * svm_ranges remain unchanged. + * + * Context: Process context, caller must hold svms->lock + * + * Return: + * 0 - OK, otherwise error code */ static int -svm_range_handle_overlap(struct svm_range_list *svms, struct svm_range *new, - unsigned long start, unsigned long last, - struct list_head *update_list, - struct list_head *insert_list, - struct list_head *remove_list, - unsigned long *left) +svm_range_add(struct kfd_process *p, uint64_t start, uint64_t size, + uint32_t nattr, struct kfd_ioctl_svm_attribute *attrs, + struct list_head *update_list, struct list_head *insert_list, + struct list_head *remove_list) { + unsigned long last = start + size - 1UL; + struct svm_range_list *svms = &p->svms; struct interval_tree_node *node; struct svm_range *prange; struct svm_range *tmp; int r = 0; + pr_debug("svms 0x%p [0x%llx 0x%lx]\n", &p->svms, start, last); + INIT_LIST_HEAD(update_list); INIT_LIST_HEAD(insert_list); INIT_LIST_HEAD(remove_list); @@ -1801,18 +1851,24 @@ svm_range_handle_overlap(struct svm_range_list *svms, struct svm_range *new, node = interval_tree_iter_first(&svms->objects, start, last); while (node) { struct interval_tree_node *next; - struct svm_range *old; unsigned long next_start; pr_debug("found overlap node [0x%lx 0x%lx]\n", node->start, node->last); - old = container_of(node, struct svm_range, it_node); + prange = container_of(node, struct svm_range, it_node); next = interval_tree_iter_next(node, start, last); next_start = min(node->last, last) + 1; - if (node->start < start || node->last > last) { - /* node intersects the updated range, clone+split it */ + if (svm_range_is_same_attrs(p, prange, nattr, attrs)) { + /* nothing to do */ + } else if (node->start < start || node->last > last) { + /* node intersects the update range and its attributes + * will change. Clone and split it, apply updates only + * to the overlapping part + */ + struct svm_range *old = prange; + prange = svm_range_clone(old); if (!prange) { r = -ENOMEM; @@ -1821,17 +1877,18 @@ svm_range_handle_overlap(struct svm_range_list *svms, struct svm_range *new, list_add(&old->remove_list, remove_list); list_add(&prange->insert_list, insert_list); + list_add(&prange->update_list, update_list); if (node->start < start) { pr_debug("change old range start\n"); - r = svm_range_split_head(prange, new, start, + r = svm_range_split_head(prange, start, insert_list); if (r) goto out; } if (node->last > last) { pr_debug("change old range last\n"); - r = svm_range_split_tail(prange, new, last, + r = svm_range_split_tail(prange, last, insert_list); if (r) goto out; @@ -1840,16 +1897,12 @@ svm_range_handle_overlap(struct svm_range_list *svms, struct svm_range *new, /* The node is contained within start..last, * just update it */ - prange = old; - } - - if (!svm_range_is_same_attrs(prange, new)) list_add(&prange->update_list, update_list); + } /* insert a new node if needed */ if (node->start > start) { - prange = svm_range_new(prange->svms, start, - node->start - 1); + prange = svm_range_new(svms, start, node->start - 1); if (!prange) { r = -ENOMEM; goto out; @@ -1863,8 +1916,16 @@ svm_range_handle_overlap(struct svm_range_list *svms, struct svm_range *new, start = next_start; } - if (left && start <= last) - *left = last - start + 1; + /* add a final range at the end if needed */ + if (start <= last) { + prange = svm_range_new(svms, start, last); + if (!prange) { + r = -ENOMEM; + goto out; + } + list_add(&prange->insert_list, insert_list); + list_add(&prange->update_list, update_list); + } out: if (r) @@ -1970,7 +2031,7 @@ restart: pr_debug("drain retry fault gpu %d svms %p\n", i, svms); - amdgpu_ih_wait_on_checkpoint_process(pdd->dev->adev, + amdgpu_ih_wait_on_checkpoint_process_ts(pdd->dev->adev, &pdd->dev->adev->irq.ih1); pr_debug("drain retry fault gpu %d svms 0x%p done\n", i, svms); } @@ -2161,6 +2222,9 @@ svm_range_unmap_from_cpu(struct mm_struct *mm, struct svm_range *prange, /** * svm_range_cpu_invalidate_pagetables - interval notifier callback + * @mni: mmu_interval_notifier struct + * @range: mmu_notifier_range struct + * @cur_seq: value to pass to mmu_interval_set_seq() * * If event is MMU_NOTIFY_UNMAP, this is from CPU unmap range, otherwise, it * is from migration, or CPU page invalidation callback. @@ -2884,59 +2948,6 @@ svm_range_is_valid(struct kfd_process *p, uint64_t start, uint64_t size) } /** - * svm_range_add - add svm range and handle overlap - * @p: the range add to this process svms - * @start: page size aligned - * @size: page size aligned - * @nattr: number of attributes - * @attrs: array of attributes - * @update_list: output, the ranges need validate and update GPU mapping - * @insert_list: output, the ranges need insert to svms - * @remove_list: output, the ranges are replaced and need remove from svms - * - * Check if the virtual address range has overlap with the registered ranges, - * split the overlapped range, copy and adjust pages address and vram nodes in - * old and new ranges. - * - * Context: Process context, caller must hold svms->lock - * - * Return: - * 0 - OK, otherwise error code - */ -static int -svm_range_add(struct kfd_process *p, uint64_t start, uint64_t size, - uint32_t nattr, struct kfd_ioctl_svm_attribute *attrs, - struct list_head *update_list, struct list_head *insert_list, - struct list_head *remove_list) -{ - uint64_t last = start + size - 1UL; - struct svm_range_list *svms; - struct svm_range new = {0}; - struct svm_range *prange; - unsigned long left = 0; - int r = 0; - - pr_debug("svms 0x%p [0x%llx 0x%llx]\n", &p->svms, start, last); - - svm_range_apply_attrs(p, &new, nattr, attrs); - - svms = &p->svms; - - r = svm_range_handle_overlap(svms, &new, start, last, update_list, - insert_list, remove_list, &left); - if (r) - return r; - - if (left) { - prange = svm_range_new(svms, last - left + 1, last); - list_add(&prange->insert_list, insert_list); - list_add(&prange->update_list, update_list); - } - - return 0; -} - -/** * svm_range_best_prefetch_location - decide the best prefetch location * @prange: svm range structure * diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c index 2d44b26b6657..948fbb39336e 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c @@ -503,7 +503,7 @@ static ssize_t node_show(struct kobject *kobj, struct attribute *attr, if (dev->gpu) { log_max_watch_addr = - __ilog2_u32(dev->gpu->device_info->num_of_watch_points); + __ilog2_u32(dev->gpu->device_info.num_of_watch_points); if (log_max_watch_addr) { dev->node_props.capability |= @@ -1285,6 +1285,8 @@ int kfd_topology_add_device(struct kfd_dev *gpu) void *crat_image = NULL; size_t image_size = 0; int proximity_domain; + int i; + const char *asic_name = amdgpu_asic_name[gpu->adev->asic_type]; INIT_LIST_HEAD(&temp_topology_device_list); @@ -1370,13 +1372,17 @@ int kfd_topology_add_device(struct kfd_dev *gpu) amdgpu_amdkfd_get_cu_info(dev->gpu->adev, &cu_info); - strncpy(dev->node_props.name, gpu->device_info->asic_name, - KFD_TOPOLOGY_PUBLIC_NAME_SIZE); + for (i = 0; i < KFD_TOPOLOGY_PUBLIC_NAME_SIZE-1; i++) { + dev->node_props.name[i] = __tolower(asic_name[i]); + if (asic_name[i] == '\0') + break; + } + dev->node_props.name[i] = '\0'; dev->node_props.simd_arrays_per_engine = cu_info.num_shader_arrays_per_engine; - dev->node_props.gfx_target_version = gpu->device_info->gfx_target_version; + dev->node_props.gfx_target_version = gpu->device_info.gfx_target_version; dev->node_props.vendor_id = gpu->pdev->vendor; dev->node_props.device_id = gpu->pdev->device; dev->node_props.capability |= @@ -1396,7 +1402,7 @@ int kfd_topology_add_device(struct kfd_dev *gpu) dev->node_props.num_sdma_xgmi_engines = kfd_get_num_xgmi_sdma_engines(gpu); dev->node_props.num_sdma_queues_per_engine = - gpu->device_info->num_sdma_queues_per_engine; + gpu->device_info.num_sdma_queues_per_engine; dev->node_props.num_gws = (dev->gpu->gws && dev->gpu->dqm->sched_policy != KFD_SCHED_POLICY_NO_HWS) ? dev->gpu->adev->gds.gws_size : 0; @@ -1572,7 +1578,7 @@ void kfd_double_confirm_iommu_support(struct kfd_dev *gpu) gpu->use_iommu_v2 = false; - if (!gpu->device_info->needs_iommu_device) + if (!gpu->device_info.needs_iommu_device) return; down_read(&topology_lock); |