diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdkfd')
31 files changed, 951 insertions, 1399 deletions
diff --git a/drivers/gpu/drm/amd/amdkfd/cik_event_interrupt.c b/drivers/gpu/drm/amd/amdkfd/cik_event_interrupt.c index f6233019f042..d60576ce10cd 100644 --- a/drivers/gpu/drm/amd/amdkfd/cik_event_interrupt.c +++ b/drivers/gpu/drm/amd/amdkfd/cik_event_interrupt.c @@ -43,15 +43,15 @@ static bool cik_event_interrupt_isr(struct kfd_dev *dev, */ if ((ihre->source_id == CIK_INTSRC_GFX_PAGE_INV_FAULT || ihre->source_id == CIK_INTSRC_GFX_MEM_PROT_FAULT) && - dev->device_info->asic_family == CHIP_HAWAII) { + dev->adev->asic_type == CHIP_HAWAII) { struct cik_ih_ring_entry *tmp_ihre = (struct cik_ih_ring_entry *)patched_ihre; *patched_flag = true; *tmp_ihre = *ihre; - vmid = f2g->read_vmid_from_vmfault_reg(dev->kgd); - ret = f2g->get_atc_vmid_pasid_mapping_info(dev->kgd, vmid, &pasid); + vmid = f2g->read_vmid_from_vmfault_reg(dev->adev); + ret = f2g->get_atc_vmid_pasid_mapping_info(dev->adev, vmid, &pasid); tmp_ihre->ring_id &= 0x000000ff; tmp_ihre->ring_id |= vmid << 8; @@ -113,7 +113,7 @@ static void cik_event_interrupt_wq(struct kfd_dev *dev, kfd_process_vm_fault(dev->dqm, pasid); memset(&info, 0, sizeof(info)); - amdgpu_amdkfd_gpuvm_get_vm_fault_info(dev->kgd, &info); + amdgpu_amdkfd_gpuvm_get_vm_fault_info(dev->adev, &info); if (!info.page_addr && !info.status) return; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c index 24ebd61395d8..4bfc0c8ab764 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c @@ -321,7 +321,7 @@ static int kfd_ioctl_create_queue(struct file *filep, struct kfd_process *p, /* Return gpu_id as doorbell offset for mmap usage */ args->doorbell_offset = KFD_MMAP_TYPE_DOORBELL; args->doorbell_offset |= KFD_MMAP_GPU_ID(args->gpu_id); - if (KFD_IS_SOC15(dev->device_info->asic_family)) + if (KFD_IS_SOC15(dev)) /* On SOC15 ASICs, include the doorbell offset within the * process doorbell frame, which is 2 pages. */ @@ -580,7 +580,7 @@ static int kfd_ioctl_dbg_register(struct file *filep, if (!dev) return -EINVAL; - if (dev->device_info->asic_family == CHIP_CARRIZO) { + if (dev->adev->asic_type == CHIP_CARRIZO) { pr_debug("kfd_ioctl_dbg_register not supported on CZ\n"); return -EINVAL; } @@ -631,7 +631,7 @@ static int kfd_ioctl_dbg_unregister(struct file *filep, if (!dev || !dev->dbgmgr) return -EINVAL; - if (dev->device_info->asic_family == CHIP_CARRIZO) { + if (dev->adev->asic_type == CHIP_CARRIZO) { pr_debug("kfd_ioctl_dbg_unregister not supported on CZ\n"); return -EINVAL; } @@ -676,7 +676,7 @@ static int kfd_ioctl_dbg_address_watch(struct file *filep, if (!dev) return -EINVAL; - if (dev->device_info->asic_family == CHIP_CARRIZO) { + if (dev->adev->asic_type == CHIP_CARRIZO) { pr_debug("kfd_ioctl_dbg_wave_control not supported on CZ\n"); return -EINVAL; } @@ -784,7 +784,7 @@ static int kfd_ioctl_dbg_wave_control(struct file *filep, if (!dev) return -EINVAL; - if (dev->device_info->asic_family == CHIP_CARRIZO) { + if (dev->adev->asic_type == CHIP_CARRIZO) { pr_debug("kfd_ioctl_dbg_wave_control not supported on CZ\n"); return -EINVAL; } @@ -851,7 +851,7 @@ static int kfd_ioctl_get_clock_counters(struct file *filep, dev = kfd_device_by_id(args->gpu_id); if (dev) /* Reading GPU clock counter from KGD */ - args->gpu_clock_counter = amdgpu_amdkfd_get_gpu_clock_counter(dev->kgd); + args->gpu_clock_counter = amdgpu_amdkfd_get_gpu_clock_counter(dev->adev); else /* Node without GPU resource */ args->gpu_clock_counter = 0; @@ -1041,7 +1041,7 @@ static int kfd_ioctl_create_event(struct file *filp, struct kfd_process *p, goto out_unlock; } - err = amdgpu_amdkfd_gpuvm_map_gtt_bo_to_kernel(kfd->kgd, + err = amdgpu_amdkfd_gpuvm_map_gtt_bo_to_kernel(kfd->adev, mem, &kern_addr, &size); if (err) { pr_err("Failed to map event page to kernel\n"); @@ -1051,7 +1051,7 @@ static int kfd_ioctl_create_event(struct file *filp, struct kfd_process *p, err = kfd_event_page_set(p, kern_addr, size); if (err) { pr_err("Failed to set event page\n"); - amdgpu_amdkfd_gpuvm_unmap_gtt_bo_from_kernel(kfd->kgd, mem); + amdgpu_amdkfd_gpuvm_unmap_gtt_bo_from_kernel(kfd->adev, mem); goto out_unlock; } @@ -1137,7 +1137,7 @@ static int kfd_ioctl_set_scratch_backing_va(struct file *filep, if (dev->dqm->sched_policy == KFD_SCHED_POLICY_NO_HWS && pdd->qpd.vmid != 0 && dev->kfd2kgd->set_scratch_backing_va) dev->kfd2kgd->set_scratch_backing_va( - dev->kgd, args->va_addr, pdd->qpd.vmid); + dev->adev, args->va_addr, pdd->qpd.vmid); return 0; @@ -1158,7 +1158,7 @@ static int kfd_ioctl_get_tile_config(struct file *filep, if (!dev) return -EINVAL; - amdgpu_amdkfd_get_tile_config(dev->kgd, &config); + amdgpu_amdkfd_get_tile_config(dev->adev, &config); args->gb_addr_config = config.gb_addr_config; args->num_banks = config.num_banks; @@ -1244,7 +1244,7 @@ bool kfd_dev_is_large_bar(struct kfd_dev *dev) if (dev->use_iommu_v2) return false; - amdgpu_amdkfd_get_local_mem_info(dev->kgd, &mem_info); + amdgpu_amdkfd_get_local_mem_info(dev->adev, &mem_info); if (mem_info.local_mem_size_private == 0 && mem_info.local_mem_size_public > 0) return true; @@ -1313,7 +1313,7 @@ static int kfd_ioctl_alloc_memory_of_gpu(struct file *filep, err = -EINVAL; goto err_unlock; } - offset = amdgpu_amdkfd_get_mmio_remap_phys_addr(dev->kgd); + offset = dev->adev->rmmio_remap.bus_addr; if (!offset) { err = -ENOMEM; goto err_unlock; @@ -1321,7 +1321,7 @@ static int kfd_ioctl_alloc_memory_of_gpu(struct file *filep, } err = amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu( - dev->kgd, args->va_addr, args->size, + dev->adev, args->va_addr, args->size, pdd->drm_priv, (struct kgd_mem **) &mem, &offset, flags); @@ -1353,7 +1353,7 @@ static int kfd_ioctl_alloc_memory_of_gpu(struct file *filep, return 0; err_free: - amdgpu_amdkfd_gpuvm_free_memory_of_gpu(dev->kgd, (struct kgd_mem *)mem, + amdgpu_amdkfd_gpuvm_free_memory_of_gpu(dev->adev, (struct kgd_mem *)mem, pdd->drm_priv, NULL); err_unlock: mutex_unlock(&p->mutex); @@ -1399,7 +1399,7 @@ static int kfd_ioctl_free_memory_of_gpu(struct file *filep, goto err_unlock; } - ret = amdgpu_amdkfd_gpuvm_free_memory_of_gpu(dev->kgd, + ret = amdgpu_amdkfd_gpuvm_free_memory_of_gpu(dev->adev, (struct kgd_mem *)mem, pdd->drm_priv, &size); /* If freeing the buffer failed, leave the handle in place for @@ -1484,7 +1484,7 @@ static int kfd_ioctl_map_memory_to_gpu(struct file *filep, goto get_mem_obj_from_handle_failed; } err = amdgpu_amdkfd_gpuvm_map_memory_to_gpu( - peer->kgd, (struct kgd_mem *)mem, + peer->adev, (struct kgd_mem *)mem, peer_pdd->drm_priv, &table_freed); if (err) { pr_err("Failed to map to gpu %d/%d\n", @@ -1496,7 +1496,7 @@ static int kfd_ioctl_map_memory_to_gpu(struct file *filep, mutex_unlock(&p->mutex); - err = amdgpu_amdkfd_gpuvm_sync_memory(dev->kgd, (struct kgd_mem *) mem, true); + err = amdgpu_amdkfd_gpuvm_sync_memory(dev->adev, (struct kgd_mem *) mem, true); if (err) { pr_debug("Sync memory failed, wait interrupted by user signal\n"); goto sync_memory_failed; @@ -1593,7 +1593,7 @@ static int kfd_ioctl_unmap_memory_from_gpu(struct file *filep, goto get_mem_obj_from_handle_failed; } err = amdgpu_amdkfd_gpuvm_unmap_memory_from_gpu( - peer->kgd, (struct kgd_mem *)mem, peer_pdd->drm_priv); + peer->adev, (struct kgd_mem *)mem, peer_pdd->drm_priv); if (err) { pr_err("Failed to unmap from gpu %d/%d\n", i, args->n_devices); @@ -1603,8 +1603,8 @@ static int kfd_ioctl_unmap_memory_from_gpu(struct file *filep, } mutex_unlock(&p->mutex); - if (dev->device_info->asic_family == CHIP_ALDEBARAN) { - err = amdgpu_amdkfd_gpuvm_sync_memory(dev->kgd, + if (KFD_GC_VERSION(dev) == IP_VERSION(9, 4, 2)) { + err = amdgpu_amdkfd_gpuvm_sync_memory(dev->adev, (struct kgd_mem *) mem, true); if (err) { pr_debug("Sync memory failed, wait interrupted by user signal\n"); @@ -1680,7 +1680,7 @@ static int kfd_ioctl_get_dmabuf_info(struct file *filep, { struct kfd_ioctl_get_dmabuf_info_args *args = data; struct kfd_dev *dev = NULL; - struct kgd_dev *dma_buf_kgd; + struct amdgpu_device *dmabuf_adev; void *metadata_buffer = NULL; uint32_t flags; unsigned int i; @@ -1700,15 +1700,15 @@ static int kfd_ioctl_get_dmabuf_info(struct file *filep, } /* Get dmabuf info from KGD */ - r = amdgpu_amdkfd_get_dmabuf_info(dev->kgd, args->dmabuf_fd, - &dma_buf_kgd, &args->size, + r = amdgpu_amdkfd_get_dmabuf_info(dev->adev, args->dmabuf_fd, + &dmabuf_adev, &args->size, metadata_buffer, args->metadata_size, &args->metadata_size, &flags); if (r) goto exit; /* Reverse-lookup gpu_id from kgd pointer */ - dev = kfd_device_by_kgd(dma_buf_kgd); + dev = kfd_device_by_adev(dmabuf_adev); if (!dev) { r = -EINVAL; goto exit; @@ -1758,7 +1758,7 @@ static int kfd_ioctl_import_dmabuf(struct file *filep, goto err_unlock; } - r = amdgpu_amdkfd_gpuvm_import_dmabuf(dev->kgd, dmabuf, + r = amdgpu_amdkfd_gpuvm_import_dmabuf(dev->adev, dmabuf, args->va_addr, pdd->drm_priv, (struct kgd_mem **)&mem, &size, NULL); @@ -1779,7 +1779,7 @@ static int kfd_ioctl_import_dmabuf(struct file *filep, return 0; err_free: - amdgpu_amdkfd_gpuvm_free_memory_of_gpu(dev->kgd, (struct kgd_mem *)mem, + amdgpu_amdkfd_gpuvm_free_memory_of_gpu(dev->adev, (struct kgd_mem *)mem, pdd->drm_priv, NULL); err_unlock: mutex_unlock(&p->mutex); @@ -2066,7 +2066,7 @@ static int kfd_mmio_mmap(struct kfd_dev *dev, struct kfd_process *process, if (vma->vm_end - vma->vm_start != PAGE_SIZE) return -EINVAL; - address = amdgpu_amdkfd_get_mmio_remap_phys_addr(dev->kgd); + address = dev->adev->rmmio_remap.bus_addr; vma->vm_flags |= VM_IO | VM_DONTCOPY | VM_DONTEXPAND | VM_NORESERVE | VM_DONTDUMP | VM_PFNMAP; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_crat.c b/drivers/gpu/drm/amd/amdkfd/kfd_crat.c index cfedfb1e8596..9624bbe8b501 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_crat.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_crat.c @@ -1060,6 +1060,9 @@ static int kfd_parse_subtype_iolink(struct crat_subtype_iolink *iolink, return -ENODEV; /* same everything but the other direction */ props2 = kmemdup(props, sizeof(*props2), GFP_KERNEL); + if (!props2) + return -ENOMEM; + props2->node_from = id_to; props2->node_to = id_from; props2->kobj = NULL; @@ -1340,7 +1343,7 @@ static int kfd_fill_gpu_cache_info(struct kfd_dev *kdev, int ret; unsigned int num_cu_shared; - switch (kdev->device_info->asic_family) { + switch (kdev->adev->asic_type) { case CHIP_KAVERI: pcache_info = kaveri_cache_info; num_of_cache_types = ARRAY_SIZE(kaveri_cache_info); @@ -1377,67 +1380,71 @@ static int kfd_fill_gpu_cache_info(struct kfd_dev *kdev, pcache_info = vegam_cache_info; num_of_cache_types = ARRAY_SIZE(vegam_cache_info); break; - case CHIP_VEGA10: - pcache_info = vega10_cache_info; - num_of_cache_types = ARRAY_SIZE(vega10_cache_info); - break; - case CHIP_VEGA12: - pcache_info = vega12_cache_info; - num_of_cache_types = ARRAY_SIZE(vega12_cache_info); - break; - case CHIP_VEGA20: - case CHIP_ARCTURUS: - pcache_info = vega20_cache_info; - num_of_cache_types = ARRAY_SIZE(vega20_cache_info); - break; - case CHIP_ALDEBARAN: - pcache_info = aldebaran_cache_info; - num_of_cache_types = ARRAY_SIZE(aldebaran_cache_info); - break; - case CHIP_RAVEN: - pcache_info = raven_cache_info; - num_of_cache_types = ARRAY_SIZE(raven_cache_info); - break; - case CHIP_RENOIR: - pcache_info = renoir_cache_info; - num_of_cache_types = ARRAY_SIZE(renoir_cache_info); - break; - case CHIP_NAVI10: - case CHIP_NAVI12: - case CHIP_CYAN_SKILLFISH: - pcache_info = navi10_cache_info; - num_of_cache_types = ARRAY_SIZE(navi10_cache_info); - break; - case CHIP_NAVI14: - pcache_info = navi14_cache_info; - num_of_cache_types = ARRAY_SIZE(navi14_cache_info); - break; - case CHIP_SIENNA_CICHLID: - pcache_info = sienna_cichlid_cache_info; - num_of_cache_types = ARRAY_SIZE(sienna_cichlid_cache_info); - break; - case CHIP_NAVY_FLOUNDER: - pcache_info = navy_flounder_cache_info; - num_of_cache_types = ARRAY_SIZE(navy_flounder_cache_info); - break; - case CHIP_DIMGREY_CAVEFISH: - pcache_info = dimgrey_cavefish_cache_info; - num_of_cache_types = ARRAY_SIZE(dimgrey_cavefish_cache_info); - break; - case CHIP_VANGOGH: - pcache_info = vangogh_cache_info; - num_of_cache_types = ARRAY_SIZE(vangogh_cache_info); - break; - case CHIP_BEIGE_GOBY: - pcache_info = beige_goby_cache_info; - num_of_cache_types = ARRAY_SIZE(beige_goby_cache_info); - break; - case CHIP_YELLOW_CARP: - pcache_info = yellow_carp_cache_info; - num_of_cache_types = ARRAY_SIZE(yellow_carp_cache_info); - break; default: - return -EINVAL; + switch(KFD_GC_VERSION(kdev)) { + case IP_VERSION(9, 0, 1): + pcache_info = vega10_cache_info; + num_of_cache_types = ARRAY_SIZE(vega10_cache_info); + break; + case IP_VERSION(9, 2, 1): + pcache_info = vega12_cache_info; + num_of_cache_types = ARRAY_SIZE(vega12_cache_info); + break; + case IP_VERSION(9, 4, 0): + case IP_VERSION(9, 4, 1): + pcache_info = vega20_cache_info; + num_of_cache_types = ARRAY_SIZE(vega20_cache_info); + break; + case IP_VERSION(9, 4, 2): + pcache_info = aldebaran_cache_info; + num_of_cache_types = ARRAY_SIZE(aldebaran_cache_info); + break; + case IP_VERSION(9, 1, 0): + case IP_VERSION(9, 2, 2): + pcache_info = raven_cache_info; + num_of_cache_types = ARRAY_SIZE(raven_cache_info); + break; + case IP_VERSION(9, 3, 0): + pcache_info = renoir_cache_info; + num_of_cache_types = ARRAY_SIZE(renoir_cache_info); + break; + case IP_VERSION(10, 1, 10): + case IP_VERSION(10, 1, 2): + case IP_VERSION(10, 1, 3): + pcache_info = navi10_cache_info; + num_of_cache_types = ARRAY_SIZE(navi10_cache_info); + break; + case IP_VERSION(10, 1, 1): + pcache_info = navi14_cache_info; + num_of_cache_types = ARRAY_SIZE(navi14_cache_info); + break; + case IP_VERSION(10, 3, 0): + pcache_info = sienna_cichlid_cache_info; + num_of_cache_types = ARRAY_SIZE(sienna_cichlid_cache_info); + break; + case IP_VERSION(10, 3, 2): + pcache_info = navy_flounder_cache_info; + num_of_cache_types = ARRAY_SIZE(navy_flounder_cache_info); + break; + case IP_VERSION(10, 3, 4): + pcache_info = dimgrey_cavefish_cache_info; + num_of_cache_types = ARRAY_SIZE(dimgrey_cavefish_cache_info); + break; + case IP_VERSION(10, 3, 1): + pcache_info = vangogh_cache_info; + num_of_cache_types = ARRAY_SIZE(vangogh_cache_info); + break; + case IP_VERSION(10, 3, 5): + pcache_info = beige_goby_cache_info; + num_of_cache_types = ARRAY_SIZE(beige_goby_cache_info); + break; + case IP_VERSION(10, 3, 3): + pcache_info = yellow_carp_cache_info; + num_of_cache_types = ARRAY_SIZE(yellow_carp_cache_info); + break; + default: + return -EINVAL; + } } *size_filled = 0; @@ -1963,8 +1970,6 @@ static int kfd_fill_gpu_direct_io_link_to_cpu(int *avail_size, struct crat_subtype_iolink *sub_type_hdr, uint32_t proximity_domain) { - struct amdgpu_device *adev = (struct amdgpu_device *)kdev->kgd; - *avail_size -= sizeof(struct crat_subtype_iolink); if (*avail_size < 0) return -ENOMEM; @@ -1981,7 +1986,7 @@ static int kfd_fill_gpu_direct_io_link_to_cpu(int *avail_size, /* Fill in IOLINK subtype. * TODO: Fill-in other fields of iolink subtype */ - if (adev->gmc.xgmi.connected_to_cpu) { + if (kdev->adev->gmc.xgmi.connected_to_cpu) { /* * with host gpu xgmi link, host can access gpu memory whether * or not pcie bar type is large, so always create bidirectional @@ -1990,19 +1995,19 @@ static int kfd_fill_gpu_direct_io_link_to_cpu(int *avail_size, sub_type_hdr->flags |= CRAT_IOLINK_FLAGS_BI_DIRECTIONAL; sub_type_hdr->io_interface_type = CRAT_IOLINK_TYPE_XGMI; sub_type_hdr->num_hops_xgmi = 1; - if (adev->asic_type == CHIP_ALDEBARAN) { + if (KFD_GC_VERSION(kdev) == IP_VERSION(9, 4, 2)) { sub_type_hdr->minimum_bandwidth_mbs = amdgpu_amdkfd_get_xgmi_bandwidth_mbytes( - kdev->kgd, NULL, true); + kdev->adev, NULL, true); sub_type_hdr->maximum_bandwidth_mbs = sub_type_hdr->minimum_bandwidth_mbs; } } else { sub_type_hdr->io_interface_type = CRAT_IOLINK_TYPE_PCIEXPRESS; sub_type_hdr->minimum_bandwidth_mbs = - amdgpu_amdkfd_get_pcie_bandwidth_mbytes(kdev->kgd, true); + amdgpu_amdkfd_get_pcie_bandwidth_mbytes(kdev->adev, true); sub_type_hdr->maximum_bandwidth_mbs = - amdgpu_amdkfd_get_pcie_bandwidth_mbytes(kdev->kgd, false); + amdgpu_amdkfd_get_pcie_bandwidth_mbytes(kdev->adev, false); } sub_type_hdr->proximity_domain_from = proximity_domain; @@ -2044,11 +2049,11 @@ static int kfd_fill_gpu_xgmi_link_to_gpu(int *avail_size, sub_type_hdr->proximity_domain_from = proximity_domain_from; sub_type_hdr->proximity_domain_to = proximity_domain_to; sub_type_hdr->num_hops_xgmi = - amdgpu_amdkfd_get_xgmi_hops_count(kdev->kgd, peer_kdev->kgd); + amdgpu_amdkfd_get_xgmi_hops_count(kdev->adev, peer_kdev->adev); sub_type_hdr->maximum_bandwidth_mbs = - amdgpu_amdkfd_get_xgmi_bandwidth_mbytes(kdev->kgd, peer_kdev->kgd, false); + amdgpu_amdkfd_get_xgmi_bandwidth_mbytes(kdev->adev, peer_kdev->adev, false); sub_type_hdr->minimum_bandwidth_mbs = sub_type_hdr->maximum_bandwidth_mbs ? - amdgpu_amdkfd_get_xgmi_bandwidth_mbytes(kdev->kgd, NULL, true) : 0; + amdgpu_amdkfd_get_xgmi_bandwidth_mbytes(kdev->adev, NULL, true) : 0; return 0; } @@ -2114,7 +2119,7 @@ static int kfd_create_vcrat_image_gpu(void *pcrat_image, cu->flags |= CRAT_CU_FLAGS_GPU_PRESENT; cu->proximity_domain = proximity_domain; - amdgpu_amdkfd_get_cu_info(kdev->kgd, &cu_info); + amdgpu_amdkfd_get_cu_info(kdev->adev, &cu_info); cu->num_simd_per_cu = cu_info.simd_per_cu; cu->num_simd_cores = cu_info.simd_per_cu * cu_info.cu_active_number; cu->max_waves_simd = cu_info.max_waves_per_simd; @@ -2145,7 +2150,7 @@ static int kfd_create_vcrat_image_gpu(void *pcrat_image, * report the total FB size (public+private) as a single * private heap. */ - amdgpu_amdkfd_get_local_mem_info(kdev->kgd, &local_mem_info); + amdgpu_amdkfd_get_local_mem_info(kdev->adev, &local_mem_info); sub_type_hdr = (typeof(sub_type_hdr))((char *)sub_type_hdr + sub_type_hdr->length); diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_dbgdev.c b/drivers/gpu/drm/amd/amdkfd/kfd_dbgdev.c index 159add0f5aaa..1e30717b5253 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_dbgdev.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_dbgdev.c @@ -41,7 +41,7 @@ static void dbgdev_address_watch_disable_nodiq(struct kfd_dev *dev) { - dev->kfd2kgd->address_watch_disable(dev->kgd); + dev->kfd2kgd->address_watch_disable(dev->adev); } static int dbgdev_diq_submit_ib(struct kfd_dbgdev *dbgdev, @@ -322,7 +322,7 @@ static int dbgdev_address_watch_nodiq(struct kfd_dbgdev *dbgdev, pr_debug("\t\t%30s\n", "* * * * * * * * * * * * * * * * * *"); pdd->dev->kfd2kgd->address_watch_execute( - dbgdev->dev->kgd, + dbgdev->dev->adev, i, cntl.u32All, addrHi.u32All, @@ -420,7 +420,7 @@ static int dbgdev_address_watch_diq(struct kfd_dbgdev *dbgdev, aw_reg_add_dword = dbgdev->dev->kfd2kgd->address_watch_get_offset( - dbgdev->dev->kgd, + dbgdev->dev->adev, i, ADDRESS_WATCH_REG_CNTL); @@ -431,7 +431,7 @@ static int dbgdev_address_watch_diq(struct kfd_dbgdev *dbgdev, aw_reg_add_dword = dbgdev->dev->kfd2kgd->address_watch_get_offset( - dbgdev->dev->kgd, + dbgdev->dev->adev, i, ADDRESS_WATCH_REG_ADDR_HI); @@ -441,7 +441,7 @@ static int dbgdev_address_watch_diq(struct kfd_dbgdev *dbgdev, aw_reg_add_dword = dbgdev->dev->kfd2kgd->address_watch_get_offset( - dbgdev->dev->kgd, + dbgdev->dev->adev, i, ADDRESS_WATCH_REG_ADDR_LO); @@ -457,7 +457,7 @@ static int dbgdev_address_watch_diq(struct kfd_dbgdev *dbgdev, aw_reg_add_dword = dbgdev->dev->kfd2kgd->address_watch_get_offset( - dbgdev->dev->kgd, + dbgdev->dev->adev, i, ADDRESS_WATCH_REG_CNTL); @@ -752,7 +752,7 @@ static int dbgdev_wave_control_nodiq(struct kfd_dbgdev *dbgdev, pr_debug("\t\t %30s\n", "* * * * * * * * * * * * * * * * * *"); - return dbgdev->dev->kfd2kgd->wave_control_execute(dbgdev->dev->kgd, + return dbgdev->dev->kfd2kgd->wave_control_execute(dbgdev->dev->adev, reg_gfx_index.u32All, reg_sq_cmd.u32All); } @@ -784,7 +784,7 @@ int dbgdev_wave_reset_wavefronts(struct kfd_dev *dev, struct kfd_process *p) for (vmid = first_vmid_to_scan; vmid <= last_vmid_to_scan; vmid++) { status = dev->kfd2kgd->get_atc_vmid_pasid_mapping_info - (dev->kgd, vmid, &queried_pasid); + (dev->adev, vmid, &queried_pasid); if (status && queried_pasid == p->pasid) { pr_debug("Killing wave fronts of vmid %d and pasid 0x%x\n", @@ -811,7 +811,7 @@ int dbgdev_wave_reset_wavefronts(struct kfd_dev *dev, struct kfd_process *p) /* for non DIQ we need to patch the VMID: */ reg_sq_cmd.bits.vm_id = vmid; - dev->kfd2kgd->wave_control_execute(dev->kgd, + dev->kfd2kgd->wave_control_execute(dev->adev, reg_gfx_index.u32All, reg_sq_cmd.u32All); diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c index 3b119db16003..2b65d0acae2c 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c @@ -53,770 +53,310 @@ 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_family = CHIP_KAVERI, - .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_engines = 2, - .num_xgmi_sdma_engines = 0, - .num_sdma_queues_per_engine = 2, -}; - -static const struct kfd_device_info carrizo_device_info = { - .asic_family = CHIP_CARRIZO, - .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_engines = 2, - .num_xgmi_sdma_engines = 0, - .num_sdma_queues_per_engine = 2, -}; - -static const struct kfd_device_info raven_device_info = { - .asic_family = CHIP_RAVEN, - .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_engines = 1, - .num_xgmi_sdma_engines = 0, - .num_sdma_queues_per_engine = 2, -}; -#endif - -#ifdef CONFIG_DRM_AMDGPU_CIK -static const struct kfd_device_info hawaii_device_info = { - .asic_family = CHIP_HAWAII, - .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_engines = 2, - .num_xgmi_sdma_engines = 0, - .num_sdma_queues_per_engine = 2, -}; -#endif - -static const struct kfd_device_info tonga_device_info = { - .asic_family = CHIP_TONGA, - .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_engines = 2, - .num_xgmi_sdma_engines = 0, - .num_sdma_queues_per_engine = 2, -}; - -static const struct kfd_device_info fiji_device_info = { - .asic_family = CHIP_FIJI, - .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_engines = 2, - .num_xgmi_sdma_engines = 0, - .num_sdma_queues_per_engine = 2, -}; - -static const struct kfd_device_info fiji_vf_device_info = { - .asic_family = CHIP_FIJI, - .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_engines = 2, - .num_xgmi_sdma_engines = 0, - .num_sdma_queues_per_engine = 2, -}; - - -static const struct kfd_device_info polaris10_device_info = { - .asic_family = CHIP_POLARIS10, - .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_engines = 2, - .num_xgmi_sdma_engines = 0, - .num_sdma_queues_per_engine = 2, -}; - -static const struct kfd_device_info polaris10_vf_device_info = { - .asic_family = CHIP_POLARIS10, - .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_engines = 2, - .num_xgmi_sdma_engines = 0, - .num_sdma_queues_per_engine = 2, -}; - -static const struct kfd_device_info polaris11_device_info = { - .asic_family = CHIP_POLARIS11, - .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_engines = 2, - .num_xgmi_sdma_engines = 0, - .num_sdma_queues_per_engine = 2, -}; - -static const struct kfd_device_info polaris12_device_info = { - .asic_family = CHIP_POLARIS12, - .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_engines = 2, - .num_xgmi_sdma_engines = 0, - .num_sdma_queues_per_engine = 2, -}; - -static const struct kfd_device_info vegam_device_info = { - .asic_family = CHIP_VEGAM, - .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_engines = 2, - .num_xgmi_sdma_engines = 0, - .num_sdma_queues_per_engine = 2, -}; - -static const struct kfd_device_info vega10_device_info = { - .asic_family = CHIP_VEGA10, - .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_engines = 2, - .num_xgmi_sdma_engines = 0, - .num_sdma_queues_per_engine = 2, -}; - -static const struct kfd_device_info vega10_vf_device_info = { - .asic_family = CHIP_VEGA10, - .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_engines = 2, - .num_xgmi_sdma_engines = 0, - .num_sdma_queues_per_engine = 2, -}; - -static const struct kfd_device_info vega12_device_info = { - .asic_family = CHIP_VEGA12, - .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_engines = 2, - .num_xgmi_sdma_engines = 0, - .num_sdma_queues_per_engine = 2, -}; - -static const struct kfd_device_info vega20_device_info = { - .asic_family = CHIP_VEGA20, - .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_engines = 2, - .num_xgmi_sdma_engines = 0, - .num_sdma_queues_per_engine = 8, -}; - -static const struct kfd_device_info arcturus_device_info = { - .asic_family = CHIP_ARCTURUS, - .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_engines = 2, - .num_xgmi_sdma_engines = 6, - .num_sdma_queues_per_engine = 8, -}; - -static const struct kfd_device_info aldebaran_device_info = { - .asic_family = CHIP_ALDEBARAN, - .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_engines = 2, - .num_xgmi_sdma_engines = 3, - .num_sdma_queues_per_engine = 8, -}; - -static const struct kfd_device_info renoir_device_info = { - .asic_family = CHIP_RENOIR, - .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_engines = 1, - .num_xgmi_sdma_engines = 0, - .num_sdma_queues_per_engine = 2, -}; - -static const struct kfd_device_info navi10_device_info = { - .asic_family = CHIP_NAVI10, - .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_engines = 2, - .num_xgmi_sdma_engines = 0, - .num_sdma_queues_per_engine = 8, -}; - -static const struct kfd_device_info navi12_device_info = { - .asic_family = CHIP_NAVI12, - .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_engines = 2, - .num_xgmi_sdma_engines = 0, - .num_sdma_queues_per_engine = 8, -}; - -static const struct kfd_device_info navi14_device_info = { - .asic_family = CHIP_NAVI14, - .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_engines = 2, - .num_xgmi_sdma_engines = 0, - .num_sdma_queues_per_engine = 8, -}; - -static const struct kfd_device_info sienna_cichlid_device_info = { - .asic_family = CHIP_SIENNA_CICHLID, - .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_engines = 4, - .num_xgmi_sdma_engines = 0, - .num_sdma_queues_per_engine = 8, -}; - -static const struct kfd_device_info navy_flounder_device_info = { - .asic_family = CHIP_NAVY_FLOUNDER, - .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_engines = 2, - .num_xgmi_sdma_engines = 0, - .num_sdma_queues_per_engine = 8, -}; - -static const struct kfd_device_info vangogh_device_info = { - .asic_family = CHIP_VANGOGH, - .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_engines = 1, - .num_xgmi_sdma_engines = 0, - .num_sdma_queues_per_engine = 2, -}; - -static const struct kfd_device_info dimgrey_cavefish_device_info = { - .asic_family = CHIP_DIMGREY_CAVEFISH, - .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_engines = 2, - .num_xgmi_sdma_engines = 0, - .num_sdma_queues_per_engine = 8, -}; - -static const struct kfd_device_info beige_goby_device_info = { - .asic_family = CHIP_BEIGE_GOBY, - .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_engines = 1, - .num_xgmi_sdma_engines = 0, - .num_sdma_queues_per_engine = 8, -}; - -static const struct kfd_device_info yellow_carp_device_info = { - .asic_family = CHIP_YELLOW_CARP, - .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_engines = 1, - .num_xgmi_sdma_engines = 0, - .num_sdma_queues_per_engine = 2, -}; - -static const struct kfd_device_info cyan_skillfish_device_info = { - .asic_family = CHIP_CYAN_SKILLFISH, - .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_engines = 2, - .num_xgmi_sdma_engines = 0, - .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); -struct kfd_dev *kgd2kfd_probe(struct kgd_dev *kgd, bool vf) +static void kfd_device_info_set_sdma_queue_num(struct kfd_dev *kfd) { - struct kfd_dev *kfd; - const struct kfd_device_info *device_info; - const struct kfd2kgd_calls *f2g; - struct amdgpu_device *adev = (struct amdgpu_device *)kgd; + uint32_t sdma_version = kfd->adev->ip_versions[SDMA0_HWIP][0]; + + switch (sdma_version) { + case IP_VERSION(4, 0, 0):/* VEGA10 */ + case IP_VERSION(4, 0, 1):/* VEGA12 */ + case IP_VERSION(4, 1, 0):/* RAVEN */ + case IP_VERSION(4, 1, 1):/* RAVEN */ + case IP_VERSION(4, 1, 2):/* RENOIR */ + case IP_VERSION(5, 2, 1):/* VANGOGH */ + case IP_VERSION(5, 2, 3):/* YELLOW_CARP */ + kfd->device_info.num_sdma_queues_per_engine = 2; + break; + case IP_VERSION(4, 2, 0):/* VEGA20 */ + case IP_VERSION(4, 2, 2):/* ARCTURUS */ + case IP_VERSION(4, 4, 0):/* ALDEBARAN */ + case IP_VERSION(5, 0, 0):/* NAVI10 */ + case IP_VERSION(5, 0, 1):/* CYAN_SKILLFISH */ + case IP_VERSION(5, 0, 2):/* NAVI14 */ + case IP_VERSION(5, 0, 5):/* NAVI12 */ + case IP_VERSION(5, 2, 0):/* SIENNA_CICHLID */ + case IP_VERSION(5, 2, 2):/* NAVY_FLOUNDER */ + case IP_VERSION(5, 2, 4):/* DIMGREY_CAVEFISH */ + case IP_VERSION(5, 2, 5):/* BEIGE_GOBY */ + kfd->device_info.num_sdma_queues_per_engine = 8; + break; + default: + dev_warn(kfd_device, + "Default sdma queue per engine(8) is set due to " + "mismatch of sdma ip block(SDMA_HWIP:0x%x).\n", + sdma_version); + kfd->device_info.num_sdma_queues_per_engine = 8; + } +} + +static void kfd_device_info_set_event_interrupt_class(struct kfd_dev *kfd) +{ + uint32_t gc_version = KFD_GC_VERSION(kfd); + + switch (gc_version) { + case IP_VERSION(9, 0, 1): /* VEGA10 */ + case IP_VERSION(9, 1, 0): /* RAVEN */ + case IP_VERSION(9, 2, 1): /* VEGA12 */ + case IP_VERSION(9, 2, 2): /* RAVEN */ + case IP_VERSION(9, 3, 0): /* RENOIR */ + case IP_VERSION(9, 4, 0): /* VEGA20 */ + case IP_VERSION(9, 4, 1): /* ARCTURUS */ + case IP_VERSION(9, 4, 2): /* ALDEBARAN */ + case IP_VERSION(10, 3, 1): /* VANGOGH */ + case IP_VERSION(10, 3, 3): /* YELLOW_CARP */ + case IP_VERSION(10, 1, 3): /* CYAN_SKILLFISH */ + case IP_VERSION(10, 1, 10): /* NAVI10 */ + case IP_VERSION(10, 1, 2): /* NAVI12 */ + case IP_VERSION(10, 1, 1): /* NAVI14 */ + case IP_VERSION(10, 3, 0): /* SIENNA_CICHLID */ + case IP_VERSION(10, 3, 2): /* NAVY_FLOUNDER */ + case IP_VERSION(10, 3, 4): /* DIMGREY_CAVEFISH */ + case IP_VERSION(10, 3, 5): /* BEIGE_GOBY */ + kfd->device_info.event_interrupt_class = &event_interrupt_class_v9; + break; + default: + dev_warn(kfd_device, "v9 event interrupt handler is set due to " + "mismatch of gc ip block(GC_HWIP:0x%x).\n", gc_version); + kfd->device_info.event_interrupt_class = &event_interrupt_class_v9; + } +} + +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 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.supports_cwsr = true; + + kfd_device_info_set_sdma_queue_num(kfd); + + kfd_device_info_set_event_interrupt_class(kfd); + + /* 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 = 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) { - dev_err(kfd_device, "%s %s not supported in kfd\n", - amdgpu_asic_name[adev->asic_type], vf ? "VF" : ""); + 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" : ""); + else + dev_err(kfd_device, "%s %s not supported in kfd\n", + amdgpu_asic_name[adev->asic_type], vf ? "VF" : ""); return NULL; } @@ -824,8 +364,8 @@ struct kfd_dev *kgd2kfd_probe(struct kgd_dev *kgd, bool vf) if (!kfd) return NULL; - kfd->kgd = kgd; - kfd->device_info = device_info; + kfd->adev = adev; + kfd_device_info_init(kfd, vf, gfx_target_version); kfd->pdev = pdev; kfd->init_complete = false; kfd->kfd2kgd = f2g; @@ -844,24 +384,24 @@ struct kfd_dev *kgd2kfd_probe(struct kgd_dev *kgd, bool vf) static void kfd_cwsr_init(struct kfd_dev *kfd) { - if (cwsr_enable && kfd->device_info->supports_cwsr) { - if (kfd->device_info->asic_family < CHIP_VEGA10) { + 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; kfd->cwsr_isa_size = sizeof(cwsr_trap_gfx8_hex); - } else if (kfd->device_info->asic_family == CHIP_ARCTURUS) { + } else if (KFD_GC_VERSION(kfd) == IP_VERSION(9, 4, 1)) { BUILD_BUG_ON(sizeof(cwsr_trap_arcturus_hex) > PAGE_SIZE); kfd->cwsr_isa = cwsr_trap_arcturus_hex; kfd->cwsr_isa_size = sizeof(cwsr_trap_arcturus_hex); - } else if (kfd->device_info->asic_family == CHIP_ALDEBARAN) { + } else if (KFD_GC_VERSION(kfd) == IP_VERSION(9, 4, 2)) { BUILD_BUG_ON(sizeof(cwsr_trap_aldebaran_hex) > PAGE_SIZE); kfd->cwsr_isa = cwsr_trap_aldebaran_hex; kfd->cwsr_isa_size = sizeof(cwsr_trap_aldebaran_hex); - } else if (kfd->device_info->asic_family < CHIP_NAVI10) { + } else if (KFD_GC_VERSION(kfd) < IP_VERSION(10, 1, 1)) { BUILD_BUG_ON(sizeof(cwsr_trap_gfx9_hex) > PAGE_SIZE); kfd->cwsr_isa = cwsr_trap_gfx9_hex; kfd->cwsr_isa_size = sizeof(cwsr_trap_gfx9_hex); - } else if (kfd->device_info->asic_family < CHIP_SIENNA_CICHLID) { + } else if (KFD_GC_VERSION(kfd) < IP_VERSION(10, 3, 0)) { BUILD_BUG_ON(sizeof(cwsr_trap_nv1x_hex) > PAGE_SIZE); kfd->cwsr_isa = cwsr_trap_nv1x_hex; kfd->cwsr_isa_size = sizeof(cwsr_trap_nv1x_hex); @@ -882,18 +422,17 @@ static int kfd_gws_init(struct kfd_dev *kfd) if (kfd->dqm->sched_policy == KFD_SCHED_POLICY_NO_HWS) return 0; - if (hws_gws_support - || (kfd->device_info->asic_family == CHIP_VEGA10 - && kfd->mec2_fw_version >= 0x81b3) - || (kfd->device_info->asic_family >= CHIP_VEGA12 - && kfd->device_info->asic_family <= CHIP_RAVEN - && kfd->mec2_fw_version >= 0x1b3) - || (kfd->device_info->asic_family == CHIP_ARCTURUS - && kfd->mec2_fw_version >= 0x30) - || (kfd->device_info->asic_family == CHIP_ALDEBARAN - && kfd->mec2_fw_version >= 0x28)) - ret = amdgpu_amdkfd_alloc_gws(kfd->kgd, - amdgpu_amdkfd_get_num_gws(kfd->kgd), &kfd->gws); + if (hws_gws_support || (KFD_IS_SOC15(kfd) && + ((KFD_GC_VERSION(kfd) == IP_VERSION(9, 0, 1) + && kfd->mec2_fw_version >= 0x81b3) || + (KFD_GC_VERSION(kfd) <= IP_VERSION(9, 4, 0) + && kfd->mec2_fw_version >= 0x1b3) || + (KFD_GC_VERSION(kfd) == IP_VERSION(9, 4, 1) + && kfd->mec2_fw_version >= 0x30) || + (KFD_GC_VERSION(kfd) == IP_VERSION(9, 4, 2) + && kfd->mec2_fw_version >= 0x28)))) + ret = amdgpu_amdkfd_alloc_gws(kfd->adev, + kfd->adev->gds.gws_size, &kfd->gws); return ret; } @@ -910,11 +449,11 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd, unsigned int size, map_process_packet_size; kfd->ddev = ddev; - kfd->mec_fw_version = amdgpu_amdkfd_get_fw_version(kfd->kgd, + kfd->mec_fw_version = amdgpu_amdkfd_get_fw_version(kfd->adev, KGD_ENGINE_MEC1); - kfd->mec2_fw_version = amdgpu_amdkfd_get_fw_version(kfd->kgd, + kfd->mec2_fw_version = amdgpu_amdkfd_get_fw_version(kfd->adev, KGD_ENGINE_MEC2); - kfd->sdma_fw_version = amdgpu_amdkfd_get_fw_version(kfd->kgd, + kfd->sdma_fw_version = amdgpu_amdkfd_get_fw_version(kfd->adev, KGD_ENGINE_SDMA1); kfd->shared_resources = *gpu_resources; @@ -927,16 +466,16 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd, * 32 and 64-bit requests are possible and must be * supported. */ - kfd->pci_atomic_requested = amdgpu_amdkfd_have_atomics_support(kfd->kgd); + 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; } @@ -953,16 +492,15 @@ 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. * There can be only 2 packets at once */ - map_process_packet_size = - kfd->device_info->asic_family == CHIP_ALDEBARAN ? + map_process_packet_size = KFD_GC_VERSION(kfd) == IP_VERSION(9, 4, 2) ? sizeof(struct pm4_mes_map_process_aldebaran) : - sizeof(struct pm4_mes_map_process); + sizeof(struct pm4_mes_map_process); size += (KFD_MAX_NUM_OF_PROCESSES * map_process_packet_size + max_num_of_queues_per_device * sizeof(struct pm4_mes_map_queues) + sizeof(struct pm4_mes_runlist)) * 2; @@ -974,7 +512,7 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd, size += 512 * 1024; if (amdgpu_amdkfd_alloc_gtt_mem( - kfd->kgd, size, &kfd->gtt_mem, + kfd->adev, size, &kfd->gtt_mem, &kfd->gtt_start_gpu_addr, &kfd->gtt_start_cpu_ptr, false)) { dev_err(kfd_device, "Could not allocate %d bytes\n", size); @@ -995,9 +533,9 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd, goto kfd_doorbell_error; } - kfd->hive_id = amdgpu_amdkfd_get_hive_id(kfd->kgd); + kfd->hive_id = kfd->adev->gmc.xgmi.hive_id; - kfd->noretry = amdgpu_amdkfd_get_noretry(kfd->kgd); + kfd->noretry = kfd->adev->gmc.noretry; if (kfd_interrupt_init(kfd)) { dev_err(kfd_device, "Error initializing interrupts\n"); @@ -1015,7 +553,7 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd, */ if (kfd_gws_init(kfd)) { dev_err(kfd_device, "Could not allocate %d gws\n", - amdgpu_amdkfd_get_num_gws(kfd->kgd)); + kfd->adev->gds.gws_size); goto gws_error; } @@ -1030,7 +568,7 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd, kfd_cwsr_init(kfd); - svm_migrate_init((struct amdgpu_device *)kfd->kgd); + svm_migrate_init(kfd->adev); if(kgd2kfd_resume_iommu(kfd)) goto device_iommu_error; @@ -1068,10 +606,10 @@ kfd_interrupt_error: kfd_doorbell_error: kfd_gtt_sa_fini(kfd); kfd_gtt_sa_init_error: - amdgpu_amdkfd_free_gtt_mem(kfd->kgd, kfd->gtt_mem); + amdgpu_amdkfd_free_gtt_mem(kfd->adev, kfd->gtt_mem); alloc_gtt_mem_failure: if (kfd->gws) - amdgpu_amdkfd_free_gws(kfd->kgd, kfd->gws); + amdgpu_amdkfd_free_gws(kfd->adev, kfd->gws); dev_err(kfd_device, "device %x:%x NOT added due to errors\n", kfd->pdev->vendor, kfd->pdev->device); @@ -1088,9 +626,9 @@ void kgd2kfd_device_exit(struct kfd_dev *kfd) kfd_doorbell_fini(kfd); ida_destroy(&kfd->doorbell_ida); kfd_gtt_sa_fini(kfd); - amdgpu_amdkfd_free_gtt_mem(kfd->kgd, kfd->gtt_mem); + amdgpu_amdkfd_free_gtt_mem(kfd->adev, kfd->gtt_mem); if (kfd->gws) - amdgpu_amdkfd_free_gws(kfd->kgd, kfd->gws); + amdgpu_amdkfd_free_gws(kfd->adev, kfd->gws); } kfree(kfd); @@ -1229,7 +767,7 @@ 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; } @@ -1526,7 +1064,7 @@ void kgd2kfd_set_sram_ecc_flag(struct kfd_dev *kfd) void kfd_inc_compute_active(struct kfd_dev *kfd) { if (atomic_inc_return(&kfd->compute_profile) == 1) - amdgpu_amdkfd_set_compute_idle(kfd->kgd, false); + amdgpu_amdkfd_set_compute_idle(kfd->adev, false); } void kfd_dec_compute_active(struct kfd_dev *kfd) @@ -1534,7 +1072,7 @@ void kfd_dec_compute_active(struct kfd_dev *kfd) int count = atomic_dec_return(&kfd->compute_profile); if (count == 0) - amdgpu_amdkfd_set_compute_idle(kfd->kgd, true); + amdgpu_amdkfd_set_compute_idle(kfd->adev, true); WARN_ONCE(count < 0, "Compute profile ref. count error"); } @@ -1544,6 +1082,26 @@ void kgd2kfd_smi_event_throttle(struct kfd_dev *kfd, uint64_t throttle_bitmask) kfd_smi_event_update_thermal_throttling(kfd, throttle_bitmask); } +/* kfd_get_num_sdma_engines returns the number of PCIe optimized SDMA and + * kfd_get_num_xgmi_sdma_engines returns the number of XGMI SDMA. + * When the device has more than two engines, we reserve two for PCIe to enable + * full-duplex and the rest are used as XGMI. + */ +unsigned int kfd_get_num_sdma_engines(struct kfd_dev *kdev) +{ + /* If XGMI is not supported, all SDMA engines are PCIe */ + if (!kdev->adev->gmc.xgmi.supported) + return kdev->adev->sdma.num_instances; + + return min(kdev->adev->sdma.num_instances, 2); +} + +unsigned int kfd_get_num_xgmi_sdma_engines(struct kfd_dev *kdev) +{ + /* After reserved for PCIe, the rest of engines are XGMI */ + return kdev->adev->sdma.num_instances - kfd_get_num_sdma_engines(kdev); +} + #if defined(CONFIG_DEBUG_FS) /* This function will send a package to HIQ to hang the HWS 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 003ba6a373ff..4b6814949aad 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c @@ -47,7 +47,7 @@ static int execute_queues_cpsch(struct device_queue_manager *dqm, uint32_t filter_param); static int unmap_queues_cpsch(struct device_queue_manager *dqm, enum kfd_unmap_queues_filter filter, - uint32_t filter_param); + uint32_t filter_param, bool reset); static int map_queues_cpsch(struct device_queue_manager *dqm); @@ -99,38 +99,29 @@ unsigned int get_pipes_per_mec(struct device_queue_manager *dqm) return dqm->dev->shared_resources.num_pipe_per_mec; } -static unsigned int get_num_sdma_engines(struct device_queue_manager *dqm) -{ - return dqm->dev->device_info->num_sdma_engines; -} - -static unsigned int get_num_xgmi_sdma_engines(struct device_queue_manager *dqm) -{ - return dqm->dev->device_info->num_xgmi_sdma_engines; -} - static unsigned int get_num_all_sdma_engines(struct device_queue_manager *dqm) { - return get_num_sdma_engines(dqm) + get_num_xgmi_sdma_engines(dqm); + return kfd_get_num_sdma_engines(dqm->dev) + + kfd_get_num_xgmi_sdma_engines(dqm->dev); } unsigned int get_num_sdma_queues(struct device_queue_manager *dqm) { - return dqm->dev->device_info->num_sdma_engines - * dqm->dev->device_info->num_sdma_queues_per_engine; + return kfd_get_num_sdma_engines(dqm->dev) * + dqm->dev->device_info.num_sdma_queues_per_engine; } unsigned int get_num_xgmi_sdma_queues(struct device_queue_manager *dqm) { - return dqm->dev->device_info->num_xgmi_sdma_engines - * dqm->dev->device_info->num_sdma_queues_per_engine; + return kfd_get_num_xgmi_sdma_engines(dqm->dev) * + dqm->dev->device_info.num_sdma_queues_per_engine; } void program_sh_mem_settings(struct device_queue_manager *dqm, struct qcm_process_device *qpd) { return dqm->dev->kfd2kgd->program_sh_mem_settings( - dqm->dev->kgd, qpd->vmid, + dqm->dev->adev, qpd->vmid, qpd->sh_mem_config, qpd->sh_mem_ape1_base, qpd->sh_mem_ape1_limit, @@ -157,7 +148,7 @@ static int allocate_doorbell(struct qcm_process_device *qpd, struct queue *q) { struct kfd_dev *dev = qpd->dqm->dev; - if (!KFD_IS_SOC15(dev->device_info->asic_family)) { + if (!KFD_IS_SOC15(dev)) { /* On pre-SOC15 chips we need to use the queue ID to * preserve the user mode ABI. */ @@ -202,7 +193,7 @@ static void deallocate_doorbell(struct qcm_process_device *qpd, unsigned int old; struct kfd_dev *dev = qpd->dqm->dev; - if (!KFD_IS_SOC15(dev->device_info->asic_family) || + if (!KFD_IS_SOC15(dev) || q->properties.type == KFD_QUEUE_TYPE_SDMA || q->properties.type == KFD_QUEUE_TYPE_SDMA_XGMI) return; @@ -216,7 +207,7 @@ static void program_trap_handler_settings(struct device_queue_manager *dqm, { if (dqm->dev->kfd2kgd->program_trap_handler_settings) dqm->dev->kfd2kgd->program_trap_handler_settings( - dqm->dev->kgd, qpd->vmid, + dqm->dev->adev, qpd->vmid, qpd->tba_addr, qpd->tma_addr); } @@ -250,21 +241,20 @@ static int allocate_vmid(struct device_queue_manager *dqm, program_sh_mem_settings(dqm, qpd); - if (dqm->dev->device_info->asic_family >= CHIP_VEGA10 && - dqm->dev->cwsr_enabled) + if (KFD_IS_SOC15(dqm->dev) && dqm->dev->cwsr_enabled) program_trap_handler_settings(dqm, qpd); /* qpd->page_table_base is set earlier when register_process() * is called, i.e. when the first queue is created. */ - dqm->dev->kfd2kgd->set_vm_context_page_table_base(dqm->dev->kgd, + dqm->dev->kfd2kgd->set_vm_context_page_table_base(dqm->dev->adev, qpd->vmid, qpd->page_table_base); /* invalidate the VM context after pasid and vmid mapping is set up */ kfd_flush_tlb(qpd_to_pdd(qpd), TLB_FLUSH_LEGACY); if (dqm->dev->kfd2kgd->set_scratch_backing_va) - dqm->dev->kfd2kgd->set_scratch_backing_va(dqm->dev->kgd, + dqm->dev->kfd2kgd->set_scratch_backing_va(dqm->dev->adev, qpd->sh_hidden_private_base, qpd->vmid); return 0; @@ -283,7 +273,7 @@ static int flush_texture_cache_nocpsch(struct kfd_dev *kdev, if (ret) return ret; - return amdgpu_amdkfd_submit_ib(kdev->kgd, KGD_ENGINE_MEC1, qpd->vmid, + return amdgpu_amdkfd_submit_ib(kdev->adev, KGD_ENGINE_MEC1, qpd->vmid, qpd->ib_base, (uint32_t *)qpd->ib_kaddr, pmf->release_mem_size / sizeof(uint32_t)); } @@ -293,7 +283,7 @@ static void deallocate_vmid(struct device_queue_manager *dqm, struct queue *q) { /* On GFX v7, CP doesn't flush TC at dequeue */ - if (q->device->device_info->asic_family == CHIP_HAWAII) + if (q->device->adev->asic_type == CHIP_HAWAII) if (flush_texture_cache_nocpsch(q->device, qpd)) pr_err("Failed to flush TC\n"); @@ -580,7 +570,7 @@ static int update_queue(struct device_queue_manager *dqm, struct queue *q, /* Make sure the queue is unmapped before updating the MQD */ if (dqm->sched_policy != KFD_SCHED_POLICY_NO_HWS) { retval = unmap_queues_cpsch(dqm, - KFD_UNMAP_QUEUES_FILTER_DYNAMIC_QUEUES, 0); + KFD_UNMAP_QUEUES_FILTER_DYNAMIC_QUEUES, 0, false); if (retval) { pr_err("unmap queue failed\n"); goto out_unlock; @@ -776,7 +766,7 @@ static int restore_process_queues_nocpsch(struct device_queue_manager *dqm, if (!list_empty(&qpd->queues_list)) { dqm->dev->kfd2kgd->set_vm_context_page_table_base( - dqm->dev->kgd, + dqm->dev->adev, qpd->vmid, qpd->page_table_base); kfd_flush_tlb(pdd, TLB_FLUSH_LEGACY); @@ -954,7 +944,7 @@ set_pasid_vmid_mapping(struct device_queue_manager *dqm, u32 pasid, unsigned int vmid) { return dqm->dev->kfd2kgd->set_pasid_vmid_mapping( - dqm->dev->kgd, pasid, vmid); + dqm->dev->adev, pasid, vmid); } static void init_interrupts(struct device_queue_manager *dqm) @@ -963,7 +953,7 @@ static void init_interrupts(struct device_queue_manager *dqm) for (i = 0 ; i < get_pipes_per_mec(dqm) ; i++) if (is_pipe_enabled(dqm, 0, i)) - dqm->dev->kfd2kgd->init_interrupts(dqm->dev->kgd, i); + dqm->dev->kfd2kgd->init_interrupts(dqm->dev->adev, i); } static int initialize_nocpsch(struct device_queue_manager *dqm) @@ -1014,19 +1004,22 @@ static void uninitialize(struct device_queue_manager *dqm) static int start_nocpsch(struct device_queue_manager *dqm) { + int r = 0; + pr_info("SW scheduler is used"); init_interrupts(dqm); - if (dqm->dev->device_info->asic_family == CHIP_HAWAII) - return pm_init(&dqm->packet_mgr, dqm); - dqm->sched_running = true; + if (dqm->dev->adev->asic_type == CHIP_HAWAII) + r = pm_init(&dqm->packet_mgr, dqm); + if (!r) + dqm->sched_running = true; - return 0; + return r; } static int stop_nocpsch(struct device_queue_manager *dqm) { - if (dqm->dev->device_info->asic_family == CHIP_HAWAII) + if (dqm->dev->adev->asic_type == CHIP_HAWAII) pm_uninit(&dqm->packet_mgr, false); dqm->sched_running = false; @@ -1055,9 +1048,9 @@ static int allocate_sdma_queue(struct device_queue_manager *dqm, dqm->sdma_bitmap &= ~(1ULL << bit); q->sdma_id = bit; q->properties.sdma_engine_id = q->sdma_id % - get_num_sdma_engines(dqm); + kfd_get_num_sdma_engines(dqm->dev); q->properties.sdma_queue_id = q->sdma_id / - get_num_sdma_engines(dqm); + kfd_get_num_sdma_engines(dqm->dev); } else if (q->properties.type == KFD_QUEUE_TYPE_SDMA_XGMI) { if (dqm->xgmi_sdma_bitmap == 0) { pr_err("No more XGMI SDMA queue to allocate\n"); @@ -1072,10 +1065,11 @@ static int allocate_sdma_queue(struct device_queue_manager *dqm, * assumes the first N engines are always * PCIe-optimized ones */ - q->properties.sdma_engine_id = get_num_sdma_engines(dqm) + - q->sdma_id % get_num_xgmi_sdma_engines(dqm); + q->properties.sdma_engine_id = + kfd_get_num_sdma_engines(dqm->dev) + + q->sdma_id % kfd_get_num_xgmi_sdma_engines(dqm->dev); q->properties.sdma_queue_id = q->sdma_id / - get_num_xgmi_sdma_engines(dqm); + kfd_get_num_xgmi_sdma_engines(dqm->dev); } pr_debug("SDMA engine id: %d\n", q->properties.sdma_engine_id); @@ -1132,7 +1126,7 @@ static int set_sched_resources(struct device_queue_manager *dqm) res.queue_mask |= 1ull << amdgpu_queue_mask_bit_to_set_resource_bit( - (struct amdgpu_device *)dqm->dev->kgd, i); + dqm->dev->adev, i); } res.gws_mask = ~0ull; res.oac_mask = res.gds_heap_base = res.gds_heap_size = 0; @@ -1226,8 +1220,13 @@ static int stop_cpsch(struct device_queue_manager *dqm) bool hanging; dqm_lock(dqm); + if (!dqm->sched_running) { + dqm_unlock(dqm); + return 0; + } + if (!dqm->is_hws_hang) - unmap_queues_cpsch(dqm, KFD_UNMAP_QUEUES_FILTER_ALL_QUEUES, 0); + unmap_queues_cpsch(dqm, KFD_UNMAP_QUEUES_FILTER_ALL_QUEUES, 0, false); hanging = dqm->is_hws_hang || dqm->is_resetting; dqm->sched_running = false; @@ -1423,7 +1422,7 @@ static int map_queues_cpsch(struct device_queue_manager *dqm) /* dqm->lock mutex has to be locked before calling this function */ static int unmap_queues_cpsch(struct device_queue_manager *dqm, enum kfd_unmap_queues_filter filter, - uint32_t filter_param) + uint32_t filter_param, bool reset) { int retval = 0; struct mqd_manager *mqd_mgr; @@ -1436,7 +1435,7 @@ static int unmap_queues_cpsch(struct device_queue_manager *dqm, return retval; retval = pm_send_unmap_queue(&dqm->packet_mgr, KFD_QUEUE_TYPE_COMPUTE, - filter, filter_param, false, 0); + filter, filter_param, reset, 0); if (retval) return retval; @@ -1480,6 +1479,21 @@ static int unmap_queues_cpsch(struct device_queue_manager *dqm, return retval; } +/* only for compute queue */ +static int reset_queues_cpsch(struct device_queue_manager *dqm, + uint16_t pasid) +{ + int retval; + + dqm_lock(dqm); + + retval = unmap_queues_cpsch(dqm, KFD_UNMAP_QUEUES_FILTER_BY_PASID, + pasid, true); + + dqm_unlock(dqm); + return retval; +} + /* dqm->lock mutex has to be locked before calling this function */ static int execute_queues_cpsch(struct device_queue_manager *dqm, enum kfd_unmap_queues_filter filter, @@ -1489,7 +1503,7 @@ static int execute_queues_cpsch(struct device_queue_manager *dqm, if (dqm->is_hws_hang) return -EIO; - retval = unmap_queues_cpsch(dqm, filter, filter_param); + retval = unmap_queues_cpsch(dqm, filter, filter_param, false); if (retval) return retval; @@ -1842,10 +1856,10 @@ 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->kgd, size, + retval = amdgpu_amdkfd_alloc_gtt_mem(dev->adev, size, &(mem_obj->gtt_mem), &(mem_obj->gpu_addr), (void *)&(mem_obj->cpu_ptr), false); @@ -1862,7 +1876,7 @@ struct device_queue_manager *device_queue_manager_init(struct kfd_dev *dev) if (!dqm) return NULL; - switch (dev->device_info->asic_family) { + switch (dev->adev->asic_type) { /* HWS is not available on Hawaii. */ case CHIP_HAWAII: /* HWS depends on CWSR for timely dequeue. CWSR is not @@ -1900,6 +1914,7 @@ struct device_queue_manager *device_queue_manager_init(struct kfd_dev *dev) dqm->ops.evict_process_queues = evict_process_queues_cpsch; dqm->ops.restore_process_queues = restore_process_queues_cpsch; dqm->ops.get_wave_state = get_wave_state; + dqm->ops.reset_queues = reset_queues_cpsch; break; case KFD_SCHED_POLICY_NO_HWS: /* initialize dqm for no cp scheduling */ @@ -1925,7 +1940,7 @@ struct device_queue_manager *device_queue_manager_init(struct kfd_dev *dev) goto out_free; } - switch (dev->device_info->asic_family) { + switch (dev->adev->asic_type) { case CHIP_CARRIZO: device_queue_manager_init_vi(&dqm->asic_ops); break; @@ -1947,31 +1962,16 @@ struct device_queue_manager *device_queue_manager_init(struct kfd_dev *dev) device_queue_manager_init_vi_tonga(&dqm->asic_ops); break; - case CHIP_VEGA10: - case CHIP_VEGA12: - case CHIP_VEGA20: - case CHIP_RAVEN: - case CHIP_RENOIR: - case CHIP_ARCTURUS: - case CHIP_ALDEBARAN: - device_queue_manager_init_v9(&dqm->asic_ops); - break; - case CHIP_NAVI10: - case CHIP_NAVI12: - case CHIP_NAVI14: - case CHIP_SIENNA_CICHLID: - case CHIP_NAVY_FLOUNDER: - case CHIP_VANGOGH: - case CHIP_DIMGREY_CAVEFISH: - case CHIP_BEIGE_GOBY: - case CHIP_YELLOW_CARP: - case CHIP_CYAN_SKILLFISH: - device_queue_manager_init_v10_navi10(&dqm->asic_ops); - break; default: - WARN(1, "Unexpected ASIC family %u", - dev->device_info->asic_family); - goto out_free; + if (KFD_GC_VERSION(dev) >= IP_VERSION(10, 1, 1)) + device_queue_manager_init_v10_navi10(&dqm->asic_ops); + else if (KFD_GC_VERSION(dev) >= IP_VERSION(9, 0, 1)) + device_queue_manager_init_v9(&dqm->asic_ops); + else { + WARN(1, "Unexpected ASIC family %u", + dev->adev->asic_type); + goto out_free; + } } if (init_mqd_managers(dqm)) @@ -1995,7 +1995,7 @@ static void deallocate_hiq_sdma_mqd(struct kfd_dev *dev, { WARN(!mqd, "No hiq sdma mqd trunk to free"); - amdgpu_amdkfd_free_gtt_mem(dev->kgd, mqd->gtt_mem); + amdgpu_amdkfd_free_gtt_mem(dev->adev, mqd->gtt_mem); } void device_queue_manager_uninit(struct device_queue_manager *dqm) @@ -2026,7 +2026,7 @@ static void kfd_process_hw_exception(struct work_struct *work) { struct device_queue_manager *dqm = container_of(work, struct device_queue_manager, hw_exception_work); - amdgpu_amdkfd_gpu_reset(dqm->dev->kgd); + amdgpu_amdkfd_gpu_reset(dqm->dev->adev); } #if defined(CONFIG_DEBUG_FS) @@ -2065,7 +2065,7 @@ int dqm_debugfs_hqds(struct seq_file *m, void *data) return 0; } - r = dqm->dev->kfd2kgd->hqd_dump(dqm->dev->kgd, + r = dqm->dev->kfd2kgd->hqd_dump(dqm->dev->adev, KFD_CIK_HIQ_PIPE, KFD_CIK_HIQ_QUEUE, &dump, &n_regs); if (!r) { @@ -2087,7 +2087,7 @@ int dqm_debugfs_hqds(struct seq_file *m, void *data) continue; r = dqm->dev->kfd2kgd->hqd_dump( - dqm->dev->kgd, pipe, queue, &dump, &n_regs); + dqm->dev->adev, pipe, queue, &dump, &n_regs); if (r) break; @@ -2101,10 +2101,10 @@ 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->kgd, pipe, queue, &dump, &n_regs); + dqm->dev->adev, pipe, queue, &dump, &n_regs); if (r) break; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h index 499fc0ea387f..e145e4deb53a 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h @@ -81,6 +81,8 @@ struct device_process_node { * * @get_wave_state: Retrieves context save state and optionally copies the * control stack, if kept in the MQD, to the given userspace address. + * + * @reset_queues: reset queues which consume RAS poison */ struct device_queue_manager_ops { @@ -134,6 +136,9 @@ struct device_queue_manager_ops { void __user *ctl_stack, u32 *ctl_stack_used_size, u32 *save_area_used_size); + + int (*reset_queues)(struct device_queue_manager *dqm, + uint16_t pasid); }; struct device_queue_manager_asic_ops { diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v9.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v9.c index b5c3d13643f1..f20434d9980e 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v9.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v9.c @@ -62,7 +62,7 @@ static int update_qpd_v9(struct device_queue_manager *dqm, SH_MEM_ALIGNMENT_MODE_UNALIGNED << SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT; - if (dqm->dev->device_info->asic_family == CHIP_ALDEBARAN) { + if (KFD_GC_VERSION(dqm->dev) == IP_VERSION(9, 4, 2)) { /* Aldebaran can safely support different XNACK modes * per process */ 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_events.c b/drivers/gpu/drm/amd/amdkfd/kfd_events.c index 3eea4edee355..afe72dd11325 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_events.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_events.c @@ -935,8 +935,10 @@ void kfd_signal_iommu_event(struct kfd_dev *dev, u32 pasid, /* Workaround on Raven to not kill the process when memory is freed * before IOMMU is able to finish processing all the excessive PPRs */ - if (dev->device_info->asic_family != CHIP_RAVEN && - dev->device_info->asic_family != CHIP_RENOIR) { + + if (KFD_GC_VERSION(dev) != IP_VERSION(9, 1, 0) && + KFD_GC_VERSION(dev) != IP_VERSION(9, 2, 2) && + KFD_GC_VERSION(dev) != IP_VERSION(9, 3, 0)) { mutex_lock(&p->event_mutex); /* Lookup events by type and signal them */ diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c b/drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c index d1388896f9c1..2e2b7ceb71db 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c @@ -394,7 +394,7 @@ int kfd_init_apertures(struct kfd_process *process) pdd->gpuvm_base = pdd->gpuvm_limit = 0; pdd->scratch_base = pdd->scratch_limit = 0; } else { - switch (dev->device_info->asic_family) { + switch (dev->adev->asic_type) { case CHIP_KAVERI: case CHIP_HAWAII: case CHIP_CARRIZO: @@ -406,29 +406,14 @@ int kfd_init_apertures(struct kfd_process *process) case CHIP_VEGAM: kfd_init_apertures_vi(pdd, id); break; - case CHIP_VEGA10: - case CHIP_VEGA12: - case CHIP_VEGA20: - case CHIP_RAVEN: - case CHIP_RENOIR: - case CHIP_ARCTURUS: - case CHIP_ALDEBARAN: - case CHIP_NAVI10: - case CHIP_NAVI12: - case CHIP_NAVI14: - case CHIP_SIENNA_CICHLID: - case CHIP_NAVY_FLOUNDER: - case CHIP_VANGOGH: - case CHIP_DIMGREY_CAVEFISH: - case CHIP_BEIGE_GOBY: - case CHIP_YELLOW_CARP: - case CHIP_CYAN_SKILLFISH: - kfd_init_apertures_v9(pdd, id); - break; default: - WARN(1, "Unexpected ASIC family %u", - dev->device_info->asic_family); - return -EINVAL; + if (KFD_GC_VERSION(dev) >= IP_VERSION(9, 0, 1)) + kfd_init_apertures_v9(pdd, id); + else { + WARN(1, "Unexpected ASIC family %u", + dev->adev->asic_type); + return -EINVAL; + } } if (!dev->use_iommu_v2) { 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 543e7ea75593..e8bc28009c22 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c @@ -89,6 +89,44 @@ enum SQ_INTERRUPT_ERROR_TYPE { #define KFD_SQ_INT_DATA__ERR_TYPE_MASK 0xF00000 #define KFD_SQ_INT_DATA__ERR_TYPE__SHIFT 20 +static void event_interrupt_poison_consumption(struct kfd_dev *dev, + uint16_t pasid, uint16_t source_id) +{ + int ret = -EINVAL; + struct kfd_process *p = kfd_lookup_process_by_pasid(pasid); + + if (!p) + return; + + /* all queues of a process will be unmapped in one time */ + if (atomic_read(&p->poison)) { + kfd_unref_process(p); + return; + } + + atomic_set(&p->poison, 1); + kfd_unref_process(p); + + switch (source_id) { + case SOC15_INTSRC_SQ_INTERRUPT_MSG: + if (dev->dqm->ops.reset_queues) + ret = dev->dqm->ops.reset_queues(dev->dqm, pasid); + break; + case SOC15_INTSRC_SDMA_ECC: + default: + break; + } + + kfd_signal_poison_consumed_event(dev, pasid); + + /* resetting queue passes, do page retirement without gpu reset + resetting queue fails, fallback to gpu reset solution */ + if (!ret) + amdgpu_amdkfd_ras_poison_consumption_handler(dev->adev, false); + else + amdgpu_amdkfd_ras_poison_consumption_handler(dev->adev, true); +} + static bool event_interrupt_isr_v9(struct kfd_dev *dev, const uint32_t *ih_ring_entry, uint32_t *patched_ihre, @@ -135,7 +173,7 @@ 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]; @@ -159,6 +197,7 @@ static bool event_interrupt_isr_v9(struct kfd_dev *dev, */ return source_id == SOC15_INTSRC_CP_END_OF_PIPE || source_id == SOC15_INTSRC_SDMA_TRAP || + source_id == SOC15_INTSRC_SDMA_ECC || source_id == SOC15_INTSRC_SQ_INTERRUPT_MSG || source_id == SOC15_INTSRC_CP_BAD_OPCODE || ((client_id == SOC15_IH_CLIENTID_VMC || @@ -230,8 +269,7 @@ static void event_interrupt_wq_v9(struct kfd_dev *dev, sq_intr_err); if (sq_intr_err != SQ_INTERRUPT_ERROR_TYPE_ILLEGAL_INST && sq_intr_err != SQ_INTERRUPT_ERROR_TYPE_MEMVIOL) { - kfd_signal_poison_consumed_event(dev, pasid); - amdgpu_amdkfd_ras_poison_consumption_handler(dev->kgd); + event_interrupt_poison_consumption(dev, pasid, source_id); return; } break; @@ -252,8 +290,7 @@ static void event_interrupt_wq_v9(struct kfd_dev *dev, if (source_id == SOC15_INTSRC_SDMA_TRAP) { kfd_signal_event_interrupt(pasid, context_id0 & 0xfffffff, 28); } else if (source_id == SOC15_INTSRC_SDMA_ECC) { - kfd_signal_poison_consumed_event(dev, pasid); - amdgpu_amdkfd_ras_poison_consumption_handler(dev->kgd); + event_interrupt_poison_consumption(dev, pasid, source_id); return; } } else if (client_id == SOC15_IH_CLIENTID_VMC || 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 64b4ac339904..16f8bc4ca7f6 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c @@ -91,7 +91,7 @@ static bool kq_initialize(struct kernel_queue *kq, struct kfd_dev *dev, kq->pq_gpu_addr = kq->pq->gpu_addr; /* For CIK family asics, kq->eop_mem is not needed */ - if (dev->device_info->asic_family > CHIP_MULLINS) { + if (dev->adev->asic_type > CHIP_MULLINS) { retval = kfd_gtt_sa_allocate(dev, PAGE_SIZE, &kq->eop_mem); if (retval != 0) goto err_eop_allocate_vidmem; @@ -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 9b9c2b9bf2ef..ed5385137f48 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 @@ -549,7 +549,7 @@ static void svm_migrate_page_free(struct page *page) if (svm_bo) { pr_debug_ratelimited("ref: %d\n", kref_read(&svm_bo->kref)); - svm_range_bo_unref(svm_bo); + svm_range_bo_unref_async(svm_bo); } } @@ -938,7 +938,7 @@ int svm_migrate_init(struct amdgpu_device *adev) void *r; /* Page migration works on Vega10 or newer */ - if (kfddev->device_info->asic_family < CHIP_VEGA10) + if (!KFD_IS_SOC15(kfddev)) return -EINVAL; pgmap = &kfddev->pgmap; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c index c021519af810..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; @@ -100,7 +100,7 @@ void mqd_symmetrically_map_cu_mask(struct mqd_manager *mm, struct kfd_cu_info cu_info; uint32_t cu_per_sh[KFD_MAX_NUM_SE][KFD_MAX_NUM_SH_PER_SE] = {0}; int i, se, sh, cu; - amdgpu_amdkfd_get_cu_info(mm->dev->kgd, &cu_info); + amdgpu_amdkfd_get_cu_info(mm->dev->adev, &cu_info); if (cu_mask_count > cu_info.cu_active_number) cu_mask_count = cu_info.cu_active_number; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_cik.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_cik.c index 8128f4d312f1..e9a8e21e144e 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_cik.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_cik.c @@ -171,7 +171,7 @@ static int load_mqd(struct mqd_manager *mm, void *mqd, uint32_t pipe_id, uint32_t wptr_shift = (p->format == KFD_QUEUE_FORMAT_AQL ? 4 : 0); uint32_t wptr_mask = (uint32_t)((p->queue_size / 4) - 1); - return mm->dev->kfd2kgd->hqd_load(mm->dev->kgd, mqd, pipe_id, queue_id, + return mm->dev->kfd2kgd->hqd_load(mm->dev->adev, mqd, pipe_id, queue_id, (uint32_t __user *)p->write_ptr, wptr_shift, wptr_mask, mms); } @@ -180,7 +180,7 @@ static int load_mqd_sdma(struct mqd_manager *mm, void *mqd, uint32_t pipe_id, uint32_t queue_id, struct queue_properties *p, struct mm_struct *mms) { - return mm->dev->kfd2kgd->hqd_sdma_load(mm->dev->kgd, mqd, + return mm->dev->kfd2kgd->hqd_sdma_load(mm->dev->adev, mqd, (uint32_t __user *)p->write_ptr, mms); } @@ -276,7 +276,7 @@ static int destroy_mqd(struct mqd_manager *mm, void *mqd, unsigned int timeout, uint32_t pipe_id, uint32_t queue_id) { - return mm->dev->kfd2kgd->hqd_destroy(mm->dev->kgd, mqd, type, timeout, + return mm->dev->kfd2kgd->hqd_destroy(mm->dev->adev, mqd, type, timeout, pipe_id, queue_id); } @@ -289,7 +289,7 @@ static int destroy_mqd_sdma(struct mqd_manager *mm, void *mqd, unsigned int timeout, uint32_t pipe_id, uint32_t queue_id) { - return mm->dev->kfd2kgd->hqd_sdma_destroy(mm->dev->kgd, mqd, timeout); + return mm->dev->kfd2kgd->hqd_sdma_destroy(mm->dev->adev, mqd, timeout); } static bool is_occupied(struct mqd_manager *mm, void *mqd, @@ -297,7 +297,7 @@ static bool is_occupied(struct mqd_manager *mm, void *mqd, uint32_t queue_id) { - return mm->dev->kfd2kgd->hqd_is_occupied(mm->dev->kgd, queue_address, + return mm->dev->kfd2kgd->hqd_is_occupied(mm->dev->adev, queue_address, pipe_id, queue_id); } @@ -306,7 +306,7 @@ static bool is_occupied_sdma(struct mqd_manager *mm, void *mqd, uint64_t queue_address, uint32_t pipe_id, uint32_t queue_id) { - return mm->dev->kfd2kgd->hqd_sdma_is_occupied(mm->dev->kgd, mqd); + return mm->dev->kfd2kgd->hqd_sdma_is_occupied(mm->dev->adev, mqd); } /* diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c index 270160fc401b..d74d8a6ac27a 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c @@ -148,7 +148,7 @@ static int load_mqd(struct mqd_manager *mm, void *mqd, /* AQL write pointer counts in 64B packets, PM4/CP counts in dwords. */ uint32_t wptr_shift = (p->format == KFD_QUEUE_FORMAT_AQL ? 4 : 0); - r = mm->dev->kfd2kgd->hqd_load(mm->dev->kgd, mqd, pipe_id, queue_id, + r = mm->dev->kfd2kgd->hqd_load(mm->dev->adev, mqd, pipe_id, queue_id, (uint32_t __user *)p->write_ptr, wptr_shift, 0, mms); return r; @@ -158,7 +158,7 @@ static int hiq_load_mqd_kiq(struct mqd_manager *mm, void *mqd, uint32_t pipe_id, uint32_t queue_id, struct queue_properties *p, struct mm_struct *mms) { - return mm->dev->kfd2kgd->hiq_mqd_load(mm->dev->kgd, mqd, pipe_id, + return mm->dev->kfd2kgd->hiq_mqd_load(mm->dev->adev, mqd, pipe_id, queue_id, p->doorbell_off); } @@ -239,7 +239,7 @@ static int destroy_mqd(struct mqd_manager *mm, void *mqd, uint32_t queue_id) { return mm->dev->kfd2kgd->hqd_destroy - (mm->dev->kgd, mqd, type, timeout, + (mm->dev->adev, mqd, type, timeout, pipe_id, queue_id); } @@ -254,7 +254,7 @@ static bool is_occupied(struct mqd_manager *mm, void *mqd, uint32_t queue_id) { return mm->dev->kfd2kgd->hqd_is_occupied( - mm->dev->kgd, queue_address, + mm->dev->adev, queue_address, pipe_id, queue_id); } @@ -320,7 +320,7 @@ static int load_mqd_sdma(struct mqd_manager *mm, void *mqd, uint32_t pipe_id, uint32_t queue_id, struct queue_properties *p, struct mm_struct *mms) { - return mm->dev->kfd2kgd->hqd_sdma_load(mm->dev->kgd, mqd, + return mm->dev->kfd2kgd->hqd_sdma_load(mm->dev->adev, mqd, (uint32_t __user *)p->write_ptr, mms); } @@ -363,14 +363,14 @@ static int destroy_mqd_sdma(struct mqd_manager *mm, void *mqd, unsigned int timeout, uint32_t pipe_id, uint32_t queue_id) { - return mm->dev->kfd2kgd->hqd_sdma_destroy(mm->dev->kgd, mqd, timeout); + return mm->dev->kfd2kgd->hqd_sdma_destroy(mm->dev->adev, mqd, timeout); } static bool is_occupied_sdma(struct mqd_manager *mm, void *mqd, uint64_t queue_address, uint32_t pipe_id, uint32_t queue_id) { - return mm->dev->kfd2kgd->hqd_sdma_is_occupied(mm->dev->kgd, mqd); + return mm->dev->kfd2kgd->hqd_sdma_is_occupied(mm->dev->adev, mqd); } #if defined(CONFIG_DEBUG_FS) diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c index 4e5932f54b5a..326eb2285029 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c @@ -108,7 +108,7 @@ static struct kfd_mem_obj *allocate_mqd(struct kfd_dev *kfd, mqd_mem_obj = kzalloc(sizeof(struct kfd_mem_obj), GFP_KERNEL); if (!mqd_mem_obj) return NULL; - retval = amdgpu_amdkfd_alloc_gtt_mem(kfd->kgd, + retval = amdgpu_amdkfd_alloc_gtt_mem(kfd->adev, ALIGN(q->ctl_stack_size, PAGE_SIZE) + ALIGN(sizeof(struct v9_mqd), PAGE_SIZE), &(mqd_mem_obj->gtt_mem), @@ -199,7 +199,7 @@ static int load_mqd(struct mqd_manager *mm, void *mqd, /* AQL write pointer counts in 64B packets, PM4/CP counts in dwords. */ uint32_t wptr_shift = (p->format == KFD_QUEUE_FORMAT_AQL ? 4 : 0); - return mm->dev->kfd2kgd->hqd_load(mm->dev->kgd, mqd, pipe_id, queue_id, + return mm->dev->kfd2kgd->hqd_load(mm->dev->adev, mqd, pipe_id, queue_id, (uint32_t __user *)p->write_ptr, wptr_shift, 0, mms); } @@ -208,7 +208,7 @@ static int hiq_load_mqd_kiq(struct mqd_manager *mm, void *mqd, uint32_t pipe_id, uint32_t queue_id, struct queue_properties *p, struct mm_struct *mms) { - return mm->dev->kfd2kgd->hiq_mqd_load(mm->dev->kgd, mqd, pipe_id, + return mm->dev->kfd2kgd->hiq_mqd_load(mm->dev->adev, mqd, pipe_id, queue_id, p->doorbell_off); } @@ -291,7 +291,7 @@ static int destroy_mqd(struct mqd_manager *mm, void *mqd, uint32_t queue_id) { return mm->dev->kfd2kgd->hqd_destroy - (mm->dev->kgd, mqd, type, timeout, + (mm->dev->adev, mqd, type, timeout, pipe_id, queue_id); } @@ -301,7 +301,7 @@ static void free_mqd(struct mqd_manager *mm, void *mqd, struct kfd_dev *kfd = mm->dev; if (mqd_mem_obj->gtt_mem) { - amdgpu_amdkfd_free_gtt_mem(kfd->kgd, mqd_mem_obj->gtt_mem); + amdgpu_amdkfd_free_gtt_mem(kfd->adev, mqd_mem_obj->gtt_mem); kfree(mqd_mem_obj); } else { kfd_gtt_sa_free(mm->dev, mqd_mem_obj); @@ -313,7 +313,7 @@ static bool is_occupied(struct mqd_manager *mm, void *mqd, uint32_t queue_id) { return mm->dev->kfd2kgd->hqd_is_occupied( - mm->dev->kgd, queue_address, + mm->dev->adev, queue_address, pipe_id, queue_id); } @@ -375,7 +375,7 @@ static int load_mqd_sdma(struct mqd_manager *mm, void *mqd, uint32_t pipe_id, uint32_t queue_id, struct queue_properties *p, struct mm_struct *mms) { - return mm->dev->kfd2kgd->hqd_sdma_load(mm->dev->kgd, mqd, + return mm->dev->kfd2kgd->hqd_sdma_load(mm->dev->adev, mqd, (uint32_t __user *)p->write_ptr, mms); } @@ -418,14 +418,14 @@ static int destroy_mqd_sdma(struct mqd_manager *mm, void *mqd, unsigned int timeout, uint32_t pipe_id, uint32_t queue_id) { - return mm->dev->kfd2kgd->hqd_sdma_destroy(mm->dev->kgd, mqd, timeout); + return mm->dev->kfd2kgd->hqd_sdma_destroy(mm->dev->adev, mqd, timeout); } static bool is_occupied_sdma(struct mqd_manager *mm, void *mqd, uint64_t queue_address, uint32_t pipe_id, uint32_t queue_id) { - return mm->dev->kfd2kgd->hqd_sdma_is_occupied(mm->dev->kgd, mqd); + return mm->dev->kfd2kgd->hqd_sdma_is_occupied(mm->dev->adev, mqd); } #if defined(CONFIG_DEBUG_FS) diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c index cd9220eb8a7a..d456e950ce1d 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c @@ -162,7 +162,7 @@ static int load_mqd(struct mqd_manager *mm, void *mqd, uint32_t wptr_shift = (p->format == KFD_QUEUE_FORMAT_AQL ? 4 : 0); uint32_t wptr_mask = (uint32_t)((p->queue_size / 4) - 1); - return mm->dev->kfd2kgd->hqd_load(mm->dev->kgd, mqd, pipe_id, queue_id, + return mm->dev->kfd2kgd->hqd_load(mm->dev->adev, mqd, pipe_id, queue_id, (uint32_t __user *)p->write_ptr, wptr_shift, wptr_mask, mms); } @@ -265,7 +265,7 @@ static int destroy_mqd(struct mqd_manager *mm, void *mqd, uint32_t queue_id) { return mm->dev->kfd2kgd->hqd_destroy - (mm->dev->kgd, mqd, type, timeout, + (mm->dev->adev, mqd, type, timeout, pipe_id, queue_id); } @@ -280,7 +280,7 @@ static bool is_occupied(struct mqd_manager *mm, void *mqd, uint32_t queue_id) { return mm->dev->kfd2kgd->hqd_is_occupied( - mm->dev->kgd, queue_address, + mm->dev->adev, queue_address, pipe_id, queue_id); } @@ -347,7 +347,7 @@ static int load_mqd_sdma(struct mqd_manager *mm, void *mqd, uint32_t pipe_id, uint32_t queue_id, struct queue_properties *p, struct mm_struct *mms) { - return mm->dev->kfd2kgd->hqd_sdma_load(mm->dev->kgd, mqd, + return mm->dev->kfd2kgd->hqd_sdma_load(mm->dev->adev, mqd, (uint32_t __user *)p->write_ptr, mms); } @@ -389,14 +389,14 @@ static int destroy_mqd_sdma(struct mqd_manager *mm, void *mqd, unsigned int timeout, uint32_t pipe_id, uint32_t queue_id) { - return mm->dev->kfd2kgd->hqd_sdma_destroy(mm->dev->kgd, mqd, timeout); + return mm->dev->kfd2kgd->hqd_sdma_destroy(mm->dev->adev, mqd, timeout); } static bool is_occupied_sdma(struct mqd_manager *mm, void *mqd, uint64_t queue_address, uint32_t pipe_id, uint32_t queue_id) { - return mm->dev->kfd2kgd->hqd_sdma_is_occupied(mm->dev->kgd, mqd); + return mm->dev->kfd2kgd->hqd_sdma_is_occupied(mm->dev->adev, mqd); } #if defined(CONFIG_DEBUG_FS) diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c index e547f1f8c49f..1439420925a0 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c @@ -223,7 +223,7 @@ static int pm_create_runlist_ib(struct packet_manager *pm, int pm_init(struct packet_manager *pm, struct device_queue_manager *dqm) { - switch (dqm->dev->device_info->asic_family) { + switch (dqm->dev->adev->asic_type) { case CHIP_KAVERI: case CHIP_HAWAII: /* PM4 packet structures on CIK are the same as on VI */ @@ -236,31 +236,16 @@ int pm_init(struct packet_manager *pm, struct device_queue_manager *dqm) case CHIP_VEGAM: pm->pmf = &kfd_vi_pm_funcs; break; - case CHIP_VEGA10: - case CHIP_VEGA12: - case CHIP_VEGA20: - case CHIP_RAVEN: - case CHIP_RENOIR: - case CHIP_ARCTURUS: - case CHIP_NAVI10: - case CHIP_NAVI12: - case CHIP_NAVI14: - case CHIP_SIENNA_CICHLID: - case CHIP_NAVY_FLOUNDER: - case CHIP_VANGOGH: - case CHIP_DIMGREY_CAVEFISH: - case CHIP_BEIGE_GOBY: - case CHIP_YELLOW_CARP: - case CHIP_CYAN_SKILLFISH: - pm->pmf = &kfd_v9_pm_funcs; - break; - case CHIP_ALDEBARAN: - pm->pmf = &kfd_aldebaran_pm_funcs; - break; default: - WARN(1, "Unexpected ASIC family %u", - dqm->dev->device_info->asic_family); - return -EINVAL; + if (KFD_GC_VERSION(dqm->dev) == IP_VERSION(9, 4, 2)) + pm->pmf = &kfd_aldebaran_pm_funcs; + else if (KFD_GC_VERSION(dqm->dev) >= IP_VERSION(9, 0, 1)) + pm->pmf = &kfd_v9_pm_funcs; + else { + WARN(1, "Unexpected ASIC family %u", + dqm->dev->adev->asic_type); + return -EINVAL; + } } pm->dqm = dqm; 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 94e92c0812db..ea68f3b3a4e9 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h +++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h @@ -183,7 +183,8 @@ enum cache_policy { cache_policy_noncoherent }; -#define KFD_IS_SOC15(chip) ((chip) >= CHIP_VEGA10) +#define KFD_GC_VERSION(dev) ((dev)->adev->ip_versions[GC_HWIP][0]) +#define KFD_IS_SOC15(dev) ((KFD_GC_VERSION(dev)) >= (IP_VERSION(9, 0, 1))) struct kfd_event_interrupt_class { bool (*interrupt_isr)(struct kfd_dev *dev, @@ -194,8 +195,6 @@ struct kfd_event_interrupt_class { }; struct kfd_device_info { - enum amd_asic_type asic_family; - const char *asic_name; uint32_t gfx_target_version; const struct kfd_event_interrupt_class *event_interrupt_class; unsigned int max_pasid_bits; @@ -208,11 +207,12 @@ struct kfd_device_info { bool needs_iommu_device; bool needs_pci_atomics; uint32_t no_atomic_fw_version; - unsigned int num_sdma_engines; - unsigned int num_xgmi_sdma_engines; unsigned int num_sdma_queues_per_engine; }; +unsigned int kfd_get_num_sdma_engines(struct kfd_dev *kdev); +unsigned int kfd_get_num_xgmi_sdma_engines(struct kfd_dev *kdev); + struct kfd_mem_obj { uint32_t range_start; uint32_t range_end; @@ -228,9 +228,9 @@ struct kfd_vmid_info { }; struct kfd_dev { - struct kgd_dev *kgd; + 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; @@ -766,7 +766,7 @@ struct svm_range_list { struct list_head deferred_range_list; spinlock_t deferred_list_lock; atomic_t evicted_ranges; - bool drain_pagefaults; + atomic_t drain_pagefaults; struct delayed_work restore_work; DECLARE_BITMAP(bitmap_supported, MAX_GPU_INSTANCE); struct task_struct *faulting_task; @@ -856,6 +856,8 @@ struct kfd_process { struct svm_range_list svms; bool xnack_enabled; + + atomic_t poison; }; #define KFD_PROCESS_TABLE_SIZE 5 /* bits: 32 entries */ @@ -891,7 +893,7 @@ struct kfd_process *kfd_lookup_process_by_pasid(u32 pasid); struct kfd_process *kfd_lookup_process_by_mm(const struct mm_struct *mm); int kfd_process_gpuidx_from_gpuid(struct kfd_process *p, uint32_t gpu_id); -int kfd_process_gpuid_from_kgd(struct kfd_process *p, +int kfd_process_gpuid_from_adev(struct kfd_process *p, struct amdgpu_device *adev, uint32_t *gpuid, uint32_t *gpuidx); static inline int kfd_process_gpuid_from_gpuidx(struct kfd_process *p, @@ -984,7 +986,7 @@ struct kfd_topology_device *kfd_topology_device_by_proximity_domain( struct kfd_topology_device *kfd_topology_device_by_id(uint32_t gpu_id); struct kfd_dev *kfd_device_by_id(uint32_t gpu_id); struct kfd_dev *kfd_device_by_pci_dev(const struct pci_dev *pdev); -struct kfd_dev *kfd_device_by_kgd(const struct kgd_dev *kgd); +struct kfd_dev *kfd_device_by_adev(const struct amdgpu_device *adev); int kfd_topology_enum_kfd_devices(uint8_t idx, struct kfd_dev **kdev); int kfd_numa_node_to_apic_id(int numa_node_id); void kfd_double_confirm_iommu_support(struct kfd_dev *gpu); diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c index b993011cfa64..d1145da5348f 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 @@ -288,7 +287,7 @@ static int kfd_get_cu_occupancy(struct attribute *attr, char *buffer) /* Collect wave count from device if it supports */ wave_cnt = 0; max_waves_per_cu = 0; - dev->kfd2kgd->get_cu_occupancy(dev->kgd, proc->pasid, &wave_cnt, + dev->kfd2kgd->get_cu_occupancy(dev->adev, proc->pasid, &wave_cnt, &max_waves_per_cu); /* Translate wave count to number of compute units */ @@ -462,6 +461,7 @@ static struct attribute *procfs_queue_attrs[] = { &attr_queue_gpuid, NULL }; +ATTRIBUTE_GROUPS(procfs_queue); static const struct sysfs_ops procfs_queue_ops = { .show = kfd_procfs_queue_show, @@ -469,7 +469,7 @@ static const struct sysfs_ops procfs_queue_ops = { static struct kobj_type procfs_queue_type = { .sysfs_ops = &procfs_queue_ops, - .default_attrs = procfs_queue_attrs, + .default_groups = procfs_queue_groups, }; static const struct sysfs_ops procfs_stats_ops = { @@ -692,12 +692,12 @@ static void kfd_process_free_gpuvm(struct kgd_mem *mem, struct kfd_dev *dev = pdd->dev; if (kptr) { - amdgpu_amdkfd_gpuvm_unmap_gtt_bo_from_kernel(dev->kgd, mem); + amdgpu_amdkfd_gpuvm_unmap_gtt_bo_from_kernel(dev->adev, mem); kptr = NULL; } - amdgpu_amdkfd_gpuvm_unmap_memory_from_gpu(dev->kgd, mem, pdd->drm_priv); - amdgpu_amdkfd_gpuvm_free_memory_of_gpu(dev->kgd, mem, pdd->drm_priv, + amdgpu_amdkfd_gpuvm_unmap_memory_from_gpu(dev->adev, mem, pdd->drm_priv); + amdgpu_amdkfd_gpuvm_free_memory_of_gpu(dev->adev, mem, pdd->drm_priv, NULL); } @@ -714,24 +714,24 @@ static int kfd_process_alloc_gpuvm(struct kfd_process_device *pdd, struct kfd_dev *kdev = pdd->dev; int err; - err = amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(kdev->kgd, gpu_va, size, + err = amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(kdev->adev, gpu_va, size, pdd->drm_priv, mem, NULL, flags); if (err) goto err_alloc_mem; - err = amdgpu_amdkfd_gpuvm_map_memory_to_gpu(kdev->kgd, *mem, + err = amdgpu_amdkfd_gpuvm_map_memory_to_gpu(kdev->adev, *mem, pdd->drm_priv, NULL); if (err) goto err_map_mem; - err = amdgpu_amdkfd_gpuvm_sync_memory(kdev->kgd, *mem, true); + err = amdgpu_amdkfd_gpuvm_sync_memory(kdev->adev, *mem, true); if (err) { pr_debug("Sync memory failed, wait interrupted by user signal\n"); goto sync_memory_failed; } if (kptr) { - err = amdgpu_amdkfd_gpuvm_map_gtt_bo_to_kernel(kdev->kgd, + err = amdgpu_amdkfd_gpuvm_map_gtt_bo_to_kernel(kdev->adev, (struct kgd_mem *)*mem, kptr, NULL); if (err) { pr_debug("Map GTT BO to kernel failed\n"); @@ -742,10 +742,10 @@ static int kfd_process_alloc_gpuvm(struct kfd_process_device *pdd, return err; sync_memory_failed: - amdgpu_amdkfd_gpuvm_unmap_memory_from_gpu(kdev->kgd, *mem, pdd->drm_priv); + amdgpu_amdkfd_gpuvm_unmap_memory_from_gpu(kdev->adev, *mem, pdd->drm_priv); err_map_mem: - amdgpu_amdkfd_gpuvm_free_memory_of_gpu(kdev->kgd, *mem, pdd->drm_priv, + amdgpu_amdkfd_gpuvm_free_memory_of_gpu(kdev->adev, *mem, pdd->drm_priv, NULL); err_alloc_mem: *mem = NULL; @@ -940,10 +940,10 @@ static void kfd_process_device_free_bos(struct kfd_process_device *pdd) if (!peer_pdd->drm_priv) continue; amdgpu_amdkfd_gpuvm_unmap_memory_from_gpu( - peer_pdd->dev->kgd, mem, peer_pdd->drm_priv); + peer_pdd->dev->adev, mem, peer_pdd->drm_priv); } - amdgpu_amdkfd_gpuvm_free_memory_of_gpu(pdd->dev->kgd, mem, + amdgpu_amdkfd_gpuvm_free_memory_of_gpu(pdd->dev->adev, mem, pdd->drm_priv, NULL); kfd_process_device_remove_obj_handle(pdd, id); } @@ -974,7 +974,7 @@ static void kfd_process_kunmap_signal_bo(struct kfd_process *p) if (!mem) goto out; - amdgpu_amdkfd_gpuvm_unmap_gtt_bo_from_kernel(kdev->kgd, mem); + amdgpu_amdkfd_gpuvm_unmap_gtt_bo_from_kernel(kdev->adev, mem); out: mutex_unlock(&p->mutex); @@ -1003,7 +1003,7 @@ static void kfd_process_destroy_pdds(struct kfd_process *p) if (pdd->drm_file) { amdgpu_amdkfd_gpuvm_release_process_vm( - pdd->dev->kgd, pdd->drm_priv); + pdd->dev->adev, pdd->drm_priv); fput(pdd->drm_file); } @@ -1011,7 +1011,7 @@ 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); @@ -1317,14 +1317,13 @@ bool kfd_process_xnack_mode(struct kfd_process *p, bool supported) * support the SVM APIs and don't need to be considered * for the XNACK mode selection. */ - if (dev->device_info->asic_family < CHIP_VEGA10) + if (!KFD_IS_SOC15(dev)) continue; /* Aldebaran can always support XNACK because it can support * per-process XNACK mode selection. But let the dev->noretry * setting still influence the default XNACK mode. */ - if (supported && - dev->device_info->asic_family == CHIP_ALDEBARAN) + if (supported && KFD_GC_VERSION(dev) == IP_VERSION(9, 4, 2)) continue; /* GFXv10 and later GPUs do not support shader preemption @@ -1332,7 +1331,7 @@ bool kfd_process_xnack_mode(struct kfd_process *p, bool supported) * management and memory-manager-related preemptions or * even deadlocks. */ - if (dev->device_info->asic_family >= CHIP_NAVI10) + if (KFD_GC_VERSION(dev) >= IP_VERSION(10, 1, 1)) return false; if (dev->noretry) @@ -1431,12 +1430,11 @@ static int init_doorbell_bitmap(struct qcm_process_device *qpd, int range_start = dev->shared_resources.non_cp_doorbells_start; int range_end = dev->shared_resources.non_cp_doorbells_end; - if (!KFD_IS_SOC15(dev->device_info->asic_family)) + 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; @@ -1448,9 +1446,9 @@ 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); } } @@ -1547,7 +1545,7 @@ int kfd_process_device_init_vm(struct kfd_process_device *pdd, dev = pdd->dev; ret = amdgpu_amdkfd_gpuvm_acquire_process_vm( - dev->kgd, drm_file, p->pasid, + dev->adev, drm_file, p->pasid, &p->kgd_process_info, &p->ef); if (ret) { pr_err("Failed to create process VM object\n"); @@ -1779,14 +1777,13 @@ int kfd_process_gpuidx_from_gpuid(struct kfd_process *p, uint32_t gpu_id) } int -kfd_process_gpuid_from_kgd(struct kfd_process *p, struct amdgpu_device *adev, +kfd_process_gpuid_from_adev(struct kfd_process *p, struct amdgpu_device *adev, uint32_t *gpuid, uint32_t *gpuidx) { - struct kgd_dev *kgd = (struct kgd_dev *)adev; int i; for (i = 0; i < p->n_pdds; i++) - if (p->pdds[i] && p->pdds[i]->dev->kgd == kgd) { + if (p->pdds[i] && p->pdds[i]->dev->adev == adev) { *gpuid = p->pdds[i]->dev->id; *gpuidx = i; return 0; @@ -1951,10 +1948,10 @@ void kfd_flush_tlb(struct kfd_process_device *pdd, enum TLB_FLUSH_TYPE type) * only happens when the first queue is created. */ if (pdd->qpd.vmid) - amdgpu_amdkfd_flush_gpu_tlb_vmid(dev->kgd, + amdgpu_amdkfd_flush_gpu_tlb_vmid(dev->adev, pdd->qpd.vmid); } else { - amdgpu_amdkfd_flush_gpu_tlb_pasid(dev->kgd, + amdgpu_amdkfd_flush_gpu_tlb_pasid(dev->adev, pdd->process->pasid, type); } } 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 3627e7ac161b..5e5c84a8e1ef 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c @@ -118,7 +118,7 @@ int pqm_set_gws(struct process_queue_manager *pqm, unsigned int qid, return ret; pqn->q->gws = mem; - pdd->qpd.num_gws = gws ? amdgpu_amdkfd_get_num_gws(dev->kgd) : 0; + pdd->qpd.num_gws = gws ? dev->adev->gds.gws_size : 0; return pqn->q->device->dqm->ops.update_queue(pqn->q->device->dqm, pqn->q, NULL); @@ -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_smi_events.c b/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c index ed4bc5f844ce..deae12dc777d 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c @@ -207,7 +207,6 @@ void kfd_smi_event_update_gpu_reset(struct kfd_dev *dev, bool post_reset) void kfd_smi_event_update_thermal_throttling(struct kfd_dev *dev, uint64_t throttle_bitmask) { - struct amdgpu_device *adev = (struct amdgpu_device *)dev->kgd; /* * ThermalThrottle msg = throttle_bitmask(8): * thermal_interrupt_count(16): @@ -223,14 +222,13 @@ void kfd_smi_event_update_thermal_throttling(struct kfd_dev *dev, len = snprintf(fifo_in, sizeof(fifo_in), "%x %llx:%llx\n", KFD_SMI_EVENT_THERMAL_THROTTLE, throttle_bitmask, - atomic64_read(&adev->smu.throttle_int_counter)); + atomic64_read(&dev->adev->smu.throttle_int_counter)); add_event_to_kfifo(dev, KFD_SMI_EVENT_THERMAL_THROTTLE, fifo_in, len); } void kfd_smi_event_update_vmfault(struct kfd_dev *dev, uint16_t pasid) { - struct amdgpu_device *adev = (struct amdgpu_device *)dev->kgd; struct amdgpu_task_info task_info; /* VmFault msg = (hex)uint32_pid(8) + :(1) + task name(16) = 25 */ /* 1 byte event + 1 byte space + 25 bytes msg + 1 byte \n + @@ -243,7 +241,7 @@ void kfd_smi_event_update_vmfault(struct kfd_dev *dev, uint16_t pasid) return; memset(&task_info, 0, sizeof(struct amdgpu_task_info)); - amdgpu_vm_get_task_info(adev, pasid, &task_info); + amdgpu_vm_get_task_info(dev->adev, pasid, &task_info); /* Report VM faults from user applications, not retry from kernel */ if (!task_info.pid) return; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c index 16137c4247bb..f2805ba74c80 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c @@ -107,7 +107,7 @@ static void svm_range_add_to_svms(struct svm_range *prange) pr_debug("svms 0x%p prange 0x%p [0x%lx 0x%lx]\n", prange->svms, prange, prange->start, prange->last); - list_add_tail(&prange->list, &prange->svms->list); + list_move_tail(&prange->list, &prange->svms->list); prange->it_node.start = prange->start; prange->it_node.last = prange->last; interval_tree_insert(&prange->it_node, &prange->svms->objects); @@ -193,7 +193,6 @@ svm_range_dma_map(struct svm_range *prange, unsigned long *bitmap, for_each_set_bit(gpuidx, bitmap, MAX_GPU_INSTANCE) { struct kfd_process_device *pdd; - struct amdgpu_device *adev; pr_debug("mapping to gpu idx 0x%x\n", gpuidx); pdd = kfd_process_device_from_gpuidx(p, gpuidx); @@ -201,9 +200,8 @@ svm_range_dma_map(struct svm_range *prange, unsigned long *bitmap, pr_debug("failed to find device idx %d\n", gpuidx); return -EINVAL; } - adev = (struct amdgpu_device *)pdd->dev->kgd; - r = svm_range_dma_map_dev(adev, prange, offset, npages, + r = svm_range_dma_map_dev(pdd->dev->adev, prange, offset, npages, hmm_pfns, gpuidx); if (r) break; @@ -297,8 +295,6 @@ svm_range *svm_range_new(struct svm_range_list *svms, uint64_t start, prange->last = last; INIT_LIST_HEAD(&prange->list); INIT_LIST_HEAD(&prange->update_list); - INIT_LIST_HEAD(&prange->remove_list); - INIT_LIST_HEAD(&prange->insert_list); INIT_LIST_HEAD(&prange->svm_bo_list); INIT_LIST_HEAD(&prange->deferred_list); INIT_LIST_HEAD(&prange->child_list); @@ -334,6 +330,8 @@ static void svm_range_bo_release(struct kref *kref) struct svm_range_bo *svm_bo; svm_bo = container_of(kref, struct svm_range_bo, kref); + pr_debug("svm_bo 0x%p\n", svm_bo); + spin_lock(&svm_bo->list_lock); while (!list_empty(&svm_bo->range_list)) { struct svm_range *prange = @@ -367,12 +365,33 @@ static void svm_range_bo_release(struct kref *kref) kfree(svm_bo); } -void svm_range_bo_unref(struct svm_range_bo *svm_bo) +static void svm_range_bo_wq_release(struct work_struct *work) { - if (!svm_bo) - return; + struct svm_range_bo *svm_bo; + + svm_bo = container_of(work, struct svm_range_bo, release_work); + svm_range_bo_release(&svm_bo->kref); +} + +static void svm_range_bo_release_async(struct kref *kref) +{ + struct svm_range_bo *svm_bo; + + svm_bo = container_of(kref, struct svm_range_bo, kref); + pr_debug("svm_bo 0x%p\n", svm_bo); + INIT_WORK(&svm_bo->release_work, svm_range_bo_wq_release); + schedule_work(&svm_bo->release_work); +} + +void svm_range_bo_unref_async(struct svm_range_bo *svm_bo) +{ + kref_put(&svm_bo->kref, svm_range_bo_release_async); +} - kref_put(&svm_bo->kref, svm_range_bo_release); +static void svm_range_bo_unref(struct svm_range_bo *svm_bo) +{ + if (svm_bo) + kref_put(&svm_bo->kref, svm_range_bo_release); } static bool @@ -581,7 +600,7 @@ svm_range_get_adev_by_id(struct svm_range *prange, uint32_t gpu_id) return NULL; } - return (struct amdgpu_device *)pdd->dev->kgd; + return pdd->dev->adev; } struct kfd_process_device * @@ -593,7 +612,7 @@ svm_range_get_pdd_by_adev(struct svm_range *prange, struct amdgpu_device *adev) p = container_of(prange->svms, struct kfd_process, svms); - r = kfd_process_gpuid_from_kgd(p, adev, &gpuid, &gpu_idx); + r = kfd_process_gpuid_from_adev(p, adev, &gpuid, &gpu_idx); if (r) { pr_debug("failed to get device id by adev %p\n", adev); return NULL; @@ -706,6 +725,61 @@ 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 @@ -743,14 +817,6 @@ 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, @@ -943,26 +1009,26 @@ 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; int r = svm_range_split(prange, prange->start, new_last, &tail); if (!r) - list_add(&tail->insert_list, insert_list); + list_add(&tail->list, insert_list); return r; } 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; int r = svm_range_split(prange, new_start, prange->last, &head); if (!r) - list_add(&head->insert_list, insert_list); + list_add(&head->list, insert_list); return r; } @@ -1053,8 +1119,8 @@ svm_range_get_pte_flags(struct amdgpu_device *adev, struct svm_range *prange, if (domain == SVM_RANGE_VRAM_DOMAIN) bo_adev = amdgpu_ttm_adev(prange->svm_bo->bo->tbo.bdev); - switch (adev->asic_type) { - case CHIP_ARCTURUS: + switch (KFD_GC_VERSION(adev->kfd.dev)) { + case IP_VERSION(9, 4, 1): if (domain == SVM_RANGE_VRAM_DOMAIN) { if (bo_adev == adev) { mapping_flags |= coherent ? @@ -1070,7 +1136,7 @@ svm_range_get_pte_flags(struct amdgpu_device *adev, struct svm_range *prange, AMDGPU_VM_MTYPE_UC : AMDGPU_VM_MTYPE_NC; } break; - case CHIP_ALDEBARAN: + case IP_VERSION(9, 4, 2): if (domain == SVM_RANGE_VRAM_DOMAIN) { if (bo_adev == adev) { mapping_flags |= coherent ? @@ -1129,7 +1195,6 @@ svm_range_unmap_from_gpus(struct svm_range *prange, unsigned long start, DECLARE_BITMAP(bitmap, MAX_GPU_INSTANCE); struct kfd_process_device *pdd; struct dma_fence *fence = NULL; - struct amdgpu_device *adev; struct kfd_process *p; uint32_t gpuidx; int r = 0; @@ -1145,9 +1210,9 @@ svm_range_unmap_from_gpus(struct svm_range *prange, unsigned long start, pr_debug("failed to find device idx %d\n", gpuidx); return -EINVAL; } - adev = (struct amdgpu_device *)pdd->dev->kgd; - r = svm_range_unmap_from_gpu(adev, drm_priv_to_vm(pdd->drm_priv), + r = svm_range_unmap_from_gpu(pdd->dev->adev, + drm_priv_to_vm(pdd->drm_priv), start, last, &fence); if (r) break; @@ -1159,7 +1224,7 @@ svm_range_unmap_from_gpus(struct svm_range *prange, unsigned long start, if (r) break; } - amdgpu_amdkfd_flush_gpu_tlb_pasid((struct kgd_dev *)adev, + amdgpu_amdkfd_flush_gpu_tlb_pasid(pdd->dev->adev, p->pasid, TLB_FLUSH_HEAVYWEIGHT); } @@ -1172,7 +1237,6 @@ 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; @@ -1185,9 +1249,6 @@ 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; @@ -1243,8 +1304,7 @@ svm_range_map_to_gpu(struct amdgpu_device *adev, struct amdgpu_vm *vm, struct kfd_process *p; p = container_of(prange->svms, struct kfd_process, svms); - amdgpu_amdkfd_flush_gpu_tlb_pasid((struct kgd_dev *)adev, - p->pasid, TLB_FLUSH_LEGACY); + amdgpu_amdkfd_flush_gpu_tlb_pasid(adev, p->pasid, TLB_FLUSH_LEGACY); } out: return r; @@ -1257,7 +1317,6 @@ svm_range_map_to_gpus(struct svm_range *prange, unsigned long offset, { struct kfd_process_device *pdd; struct amdgpu_device *bo_adev; - struct amdgpu_device *adev; struct kfd_process *p; struct dma_fence *fence = NULL; uint32_t gpuidx; @@ -1276,19 +1335,18 @@ svm_range_map_to_gpus(struct svm_range *prange, unsigned long offset, pr_debug("failed to find device idx %d\n", gpuidx); return -EINVAL; } - adev = (struct amdgpu_device *)pdd->dev->kgd; pdd = kfd_bind_process_to_device(pdd->dev, p); if (IS_ERR(pdd)) return -EINVAL; - if (bo_adev && adev != bo_adev && - !amdgpu_xgmi_same_hive(adev, bo_adev)) { + if (bo_adev && pdd->dev->adev != bo_adev && + !amdgpu_xgmi_same_hive(pdd->dev->adev, bo_adev)) { pr_debug("cannot map to device idx %d\n", gpuidx); continue; } - r = svm_range_map_to_gpu(adev, drm_priv_to_vm(pdd->drm_priv), + r = svm_range_map_to_gpu(pdd->dev->adev, drm_priv_to_vm(pdd->drm_priv), prange, offset, npages, readonly, prange->dma_addr[gpuidx], bo_adev, wait ? &fence : NULL); @@ -1322,7 +1380,6 @@ struct svm_validate_context { static int svm_range_reserve_bos(struct svm_validate_context *ctx) { struct kfd_process_device *pdd; - struct amdgpu_device *adev; struct amdgpu_vm *vm; uint32_t gpuidx; int r; @@ -1334,7 +1391,6 @@ static int svm_range_reserve_bos(struct svm_validate_context *ctx) pr_debug("failed to find device idx %d\n", gpuidx); return -EINVAL; } - adev = (struct amdgpu_device *)pdd->dev->kgd; vm = drm_priv_to_vm(pdd->drm_priv); ctx->tv[gpuidx].bo = &vm->root.bo->tbo; @@ -1356,9 +1412,9 @@ static int svm_range_reserve_bos(struct svm_validate_context *ctx) r = -EINVAL; goto unreserve_out; } - adev = (struct amdgpu_device *)pdd->dev->kgd; - r = amdgpu_vm_validate_pt_bos(adev, drm_priv_to_vm(pdd->drm_priv), + r = amdgpu_vm_validate_pt_bos(pdd->dev->adev, + drm_priv_to_vm(pdd->drm_priv), svm_range_bo_validate, NULL); if (r) { pr_debug("failed %d validate pt bos\n", r); @@ -1381,12 +1437,10 @@ static void svm_range_unreserve_bos(struct svm_validate_context *ctx) static void *kfd_svm_page_owner(struct kfd_process *p, int32_t gpuidx) { struct kfd_process_device *pdd; - struct amdgpu_device *adev; pdd = kfd_process_device_from_gpuidx(p, gpuidx); - adev = (struct amdgpu_device *)pdd->dev->kgd; - return SVM_ADEV_PGMAP_OWNER(adev); + return SVM_ADEV_PGMAP_OWNER(pdd->dev->adev); } /* @@ -1574,7 +1628,6 @@ retry_flush_work: static void svm_range_restore_work(struct work_struct *work) { struct delayed_work *dwork = to_delayed_work(work); - struct amdkfd_process_info *process_info; struct svm_range_list *svms; struct svm_range *prange; struct kfd_process *p; @@ -1594,12 +1647,10 @@ static void svm_range_restore_work(struct work_struct *work) * the lifetime of this thread, kfd_process and mm will be valid. */ p = container_of(svms, struct kfd_process, svms); - process_info = p->kgd_process_info; mm = p->mm; if (!mm) return; - mutex_lock(&process_info->lock); svm_range_list_lock_and_flush_work(svms, mm); mutex_lock(&svms->lock); @@ -1652,7 +1703,6 @@ static void svm_range_restore_work(struct work_struct *work) out_reschedule: mutex_unlock(&svms->lock); mmap_write_unlock(mm); - mutex_unlock(&process_info->lock); /* If validation failed, reschedule another attempt */ if (evicted_ranges) { @@ -1664,6 +1714,10 @@ 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. @@ -1768,46 +1822,49 @@ 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); @@ -1815,37 +1872,44 @@ 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; goto out; } - list_add(&old->remove_list, remove_list); - list_add(&prange->insert_list, insert_list); + list_add(&old->update_list, remove_list); + list_add(&prange->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; @@ -1854,22 +1918,18 @@ 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; } - list_add(&prange->insert_list, insert_list); + list_add(&prange->list, insert_list); list_add(&prange->update_list, update_list); } @@ -1877,12 +1937,20 @@ 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->list, insert_list); + list_add(&prange->update_list, update_list); + } out: if (r) - list_for_each_entry_safe(prange, tmp, insert_list, insert_list) + list_for_each_entry_safe(prange, tmp, insert_list, list) svm_range_free(prange); return r; @@ -1966,23 +2034,30 @@ svm_range_handle_list_op(struct svm_range_list *svms, struct svm_range *prange) static void svm_range_drain_retry_fault(struct svm_range_list *svms) { struct kfd_process_device *pdd; - struct amdgpu_device *adev; struct kfd_process *p; + int drain; uint32_t i; p = container_of(svms, struct kfd_process, svms); +restart: + drain = atomic_read(&svms->drain_pagefaults); + if (!drain) + return; + for_each_set_bit(i, svms->bitmap_supported, p->n_pdds) { pdd = p->pdds[i]; if (!pdd) continue; pr_debug("drain retry fault gpu %d svms %p\n", i, svms); - adev = (struct amdgpu_device *)pdd->dev->kgd; - amdgpu_ih_wait_on_checkpoint_process(adev, &adev->irq.ih1); + 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); } + if (atomic_cmpxchg(&svms->drain_pagefaults, drain, 0) != drain) + goto restart; } static void svm_range_deferred_list_work(struct work_struct *work) @@ -1990,43 +2065,41 @@ static void svm_range_deferred_list_work(struct work_struct *work) struct svm_range_list *svms; struct svm_range *prange; struct mm_struct *mm; + struct kfd_process *p; svms = container_of(work, struct svm_range_list, deferred_list_work); pr_debug("enter svms 0x%p\n", svms); + p = container_of(svms, struct kfd_process, svms); + /* Avoid mm is gone when inserting mmu notifier */ + mm = get_task_mm(p->lead_thread); + if (!mm) { + pr_debug("svms 0x%p process mm gone\n", svms); + return; + } +retry: + mmap_write_lock(mm); + + /* Checking for the need to drain retry faults must be inside + * mmap write lock to serialize with munmap notifiers. + */ + if (unlikely(atomic_read(&svms->drain_pagefaults))) { + mmap_write_unlock(mm); + svm_range_drain_retry_fault(svms); + goto retry; + } + spin_lock(&svms->deferred_list_lock); while (!list_empty(&svms->deferred_range_list)) { prange = list_first_entry(&svms->deferred_range_list, struct svm_range, deferred_list); + list_del_init(&prange->deferred_list); spin_unlock(&svms->deferred_list_lock); + pr_debug("prange 0x%p [0x%lx 0x%lx] op %d\n", prange, prange->start, prange->last, prange->work_item.op); - mm = prange->work_item.mm; -retry: - mmap_write_lock(mm); mutex_lock(&svms->lock); - - /* Checking for the need to drain retry faults must be in - * mmap write lock to serialize with munmap notifiers. - * - * Remove from deferred_list must be inside mmap write lock, - * otherwise, svm_range_list_lock_and_flush_work may hold mmap - * write lock, and continue because deferred_list is empty, then - * deferred_list handle is blocked by mmap write lock. - */ - spin_lock(&svms->deferred_list_lock); - if (unlikely(svms->drain_pagefaults)) { - svms->drain_pagefaults = false; - spin_unlock(&svms->deferred_list_lock); - mutex_unlock(&svms->lock); - mmap_write_unlock(mm); - svm_range_drain_retry_fault(svms); - goto retry; - } - list_del_init(&prange->deferred_list); - spin_unlock(&svms->deferred_list_lock); - mutex_lock(&prange->migrate_mutex); while (!list_empty(&prange->child_list)) { struct svm_range *pchild; @@ -2042,12 +2115,13 @@ retry: svm_range_handle_list_op(svms, prange); mutex_unlock(&svms->lock); - mmap_write_unlock(mm); spin_lock(&svms->deferred_list_lock); } spin_unlock(&svms->deferred_list_lock); + mmap_write_unlock(mm); + mmput(mm); pr_debug("exit svms 0x%p\n", svms); } @@ -2056,12 +2130,6 @@ svm_range_add_list_work(struct svm_range_list *svms, struct svm_range *prange, struct mm_struct *mm, enum svm_work_list_ops op) { spin_lock(&svms->deferred_list_lock); - /* Make sure pending page faults are drained in the deferred worker - * before the range is freed to avoid straggler interrupts on - * unmapped memory causing "phantom faults". - */ - if (op == SVM_OP_UNMAP_RANGE) - svms->drain_pagefaults = true; /* if prange is on the deferred list */ if (!list_empty(&prange->deferred_list)) { pr_debug("update exist prange 0x%p work op %d\n", prange, op); @@ -2140,6 +2208,12 @@ svm_range_unmap_from_cpu(struct mm_struct *mm, struct svm_range *prange, pr_debug("svms 0x%p prange 0x%p [0x%lx 0x%lx] [0x%lx 0x%lx]\n", svms, prange, prange->start, prange->last, start, last); + /* Make sure pending page faults are drained in the deferred worker + * before the range is freed to avoid straggler interrupts on + * unmapped memory causing "phantom faults". + */ + atomic_inc(&svms->drain_pagefaults); + unmap_parent = start <= prange->start && last >= prange->last; list_for_each_entry(pchild, &prange->child_list, child_list) { @@ -2169,6 +2243,9 @@ 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. @@ -2198,8 +2275,8 @@ svm_range_cpu_invalidate_pagetables(struct mmu_interval_notifier *mni, start = mni->interval_tree.start; last = mni->interval_tree.last; - start = (start > range->start ? start : range->start) >> PAGE_SHIFT; - last = (last < (range->end - 1) ? last : range->end - 1) >> PAGE_SHIFT; + start = max(start, range->start) >> PAGE_SHIFT; + last = min(last, range->end - 1) >> PAGE_SHIFT; pr_debug("[0x%lx 0x%lx] range[0x%lx 0x%lx] notifier[0x%lx 0x%lx] %d\n", start, last, range->start >> PAGE_SHIFT, (range->end - 1) >> PAGE_SHIFT, @@ -2301,7 +2378,7 @@ svm_range_best_restore_location(struct svm_range *prange, p = container_of(prange->svms, struct kfd_process, svms); - r = kfd_process_gpuid_from_kgd(p, adev, &gpuid, gpuidx); + r = kfd_process_gpuid_from_adev(p, adev, &gpuid, gpuidx); if (r < 0) { pr_debug("failed to get gpuid from kgd\n"); return -1; @@ -2478,7 +2555,7 @@ svm_range *svm_range_create_unregistered_range(struct amdgpu_device *adev, pr_debug("Failed to create prange in address [0x%llx]\n", addr); return NULL; } - if (kfd_process_gpuid_from_kgd(p, adev, &gpuid, &gpuidx)) { + if (kfd_process_gpuid_from_adev(p, adev, &gpuid, &gpuidx)) { pr_debug("failed to get gpuid from kgd\n"); svm_range_free(prange); return NULL; @@ -2545,7 +2622,7 @@ svm_range_count_fault(struct amdgpu_device *adev, struct kfd_process *p, uint32_t gpuid; int r; - r = kfd_process_gpuid_from_kgd(p, adev, &gpuid, &gpuidx); + r = kfd_process_gpuid_from_adev(p, adev, &gpuid, &gpuidx); if (r < 0) return; } @@ -2559,20 +2636,13 @@ svm_range_count_fault(struct amdgpu_device *adev, struct kfd_process *p, } static bool -svm_fault_allowed(struct mm_struct *mm, uint64_t addr, bool write_fault) +svm_fault_allowed(struct vm_area_struct *vma, bool write_fault) { unsigned long requested = VM_READ; - struct vm_area_struct *vma; if (write_fault) requested |= VM_WRITE; - vma = find_vma(mm, addr << PAGE_SHIFT); - if (!vma || (addr << PAGE_SHIFT) < vma->vm_start) { - pr_debug("address 0x%llx VMA is removed\n", addr); - return true; - } - pr_debug("requested 0x%lx, vma permission flags 0x%lx\n", requested, vma->vm_flags); return (vma->vm_flags & requested) == requested; @@ -2590,6 +2660,7 @@ svm_range_restore_pages(struct amdgpu_device *adev, unsigned int pasid, int32_t best_loc; int32_t gpuidx = MAX_GPU_INSTANCE; bool write_locked = false; + struct vm_area_struct *vma; int r = 0; if (!KFD_IS_SVM_API_SUPPORTED(adev->kfd.dev)) { @@ -2600,7 +2671,7 @@ svm_range_restore_pages(struct amdgpu_device *adev, unsigned int pasid, p = kfd_lookup_process_by_pasid(pasid); if (!p) { pr_debug("kfd process not founded pasid 0x%x\n", pasid); - return -ESRCH; + return 0; } if (!p->xnack_enabled) { pr_debug("XNACK not enabled for pasid 0x%x\n", pasid); @@ -2611,10 +2682,19 @@ svm_range_restore_pages(struct amdgpu_device *adev, unsigned int pasid, pr_debug("restoring svms 0x%p fault address 0x%llx\n", svms, addr); + if (atomic_read(&svms->drain_pagefaults)) { + pr_debug("draining retry fault, drop fault 0x%llx\n", addr); + r = 0; + goto out; + } + + /* p->lead_thread is available as kfd_process_wq_release flush the work + * before releasing task ref. + */ mm = get_task_mm(p->lead_thread); if (!mm) { pr_debug("svms 0x%p failed to get mm\n", svms); - r = -ESRCH; + r = 0; goto out; } @@ -2652,6 +2732,7 @@ retry_write_locked: if (svm_range_skip_recover(prange)) { amdgpu_gmc_filter_faults_remove(adev, addr, pasid); + r = 0; goto out_unlock_range; } @@ -2660,10 +2741,21 @@ retry_write_locked: if (timestamp < AMDGPU_SVM_RANGE_RETRY_FAULT_PENDING) { pr_debug("svms 0x%p [0x%lx %lx] already restored\n", svms, prange->start, prange->last); + r = 0; + goto out_unlock_range; + } + + /* __do_munmap removed VMA, return success as we are handling stale + * retry fault. + */ + vma = find_vma(mm, addr << PAGE_SHIFT); + if (!vma || (addr << PAGE_SHIFT) < vma->vm_start) { + pr_debug("address 0x%llx VMA is removed\n", addr); + r = 0; goto out_unlock_range; } - if (!svm_fault_allowed(mm, addr, write_fault)) { + if (!svm_fault_allowed(vma, write_fault)) { pr_debug("fault addr 0x%llx no %s permission\n", addr, write_fault ? "write" : "read"); r = -EPERM; @@ -2741,6 +2833,14 @@ void svm_range_list_fini(struct kfd_process *p) /* Ensure list work is finished before process is destroyed */ flush_work(&p->svms.deferred_list_work); + /* + * Ensure no retry fault comes in afterwards, as page fault handler will + * not find kfd process and take mm lock to recover fault. + */ + atomic_inc(&p->svms.drain_pagefaults); + svm_range_drain_retry_fault(&p->svms); + + list_for_each_entry_safe(prange, next, &p->svms.list, list) { svm_range_unlink(prange); svm_range_remove_notifier(prange); @@ -2761,6 +2861,7 @@ int svm_range_list_init(struct kfd_process *p) mutex_init(&svms->lock); INIT_LIST_HEAD(&svms->list); atomic_set(&svms->evicted_ranges, 0); + atomic_set(&svms->drain_pagefaults, 0); INIT_DELAYED_WORK(&svms->restore_work, svm_range_restore_work); INIT_WORK(&svms->deferred_list_work, svm_range_deferred_list_work); INIT_LIST_HEAD(&svms->deferred_range_list); @@ -2868,59 +2969,6 @@ 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 * @@ -2953,7 +3001,6 @@ svm_range_best_prefetch_location(struct svm_range *prange) uint32_t best_loc = prange->prefetch_loc; struct kfd_process_device *pdd; struct amdgpu_device *bo_adev; - struct amdgpu_device *adev; struct kfd_process *p; uint32_t gpuidx; @@ -2981,12 +3028,11 @@ svm_range_best_prefetch_location(struct svm_range *prange) pr_debug("failed to get device by idx 0x%x\n", gpuidx); continue; } - adev = (struct amdgpu_device *)pdd->dev->kgd; - if (adev == bo_adev) + if (pdd->dev->adev == bo_adev) continue; - if (!amdgpu_xgmi_same_hive(adev, bo_adev)) { + if (!amdgpu_xgmi_same_hive(pdd->dev->adev, bo_adev)) { best_loc = 0; break; } @@ -3150,7 +3196,6 @@ static int svm_range_set_attr(struct kfd_process *p, uint64_t start, uint64_t size, uint32_t nattr, struct kfd_ioctl_svm_attribute *attrs) { - struct amdkfd_process_info *process_info = p->kgd_process_info; struct mm_struct *mm = current->mm; struct list_head update_list; struct list_head insert_list; @@ -3169,8 +3214,6 @@ svm_range_set_attr(struct kfd_process *p, uint64_t start, uint64_t size, svms = &p->svms; - mutex_lock(&process_info->lock); - svm_range_list_lock_and_flush_work(svms, mm); r = svm_range_is_valid(p, start, size); @@ -3191,7 +3234,7 @@ svm_range_set_attr(struct kfd_process *p, uint64_t start, uint64_t size, goto out; } /* Apply changes as a transaction */ - list_for_each_entry_safe(prange, next, &insert_list, insert_list) { + list_for_each_entry_safe(prange, next, &insert_list, list) { svm_range_add_to_svms(prange); svm_range_add_notifier_locked(mm, prange); } @@ -3199,8 +3242,7 @@ svm_range_set_attr(struct kfd_process *p, uint64_t start, uint64_t size, svm_range_apply_attrs(p, prange, nattr, attrs); /* TODO: unmap ranges from GPU that lost access */ } - list_for_each_entry_safe(prange, next, &remove_list, - remove_list) { + list_for_each_entry_safe(prange, next, &remove_list, update_list) { pr_debug("unlink old 0x%p prange 0x%p [0x%lx 0x%lx]\n", prange->svms, prange, prange->start, prange->last); @@ -3246,8 +3288,6 @@ out_unlock_range: mutex_unlock(&svms->lock); mmap_read_unlock(mm); out: - mutex_unlock(&process_info->lock); - pr_debug("pasid 0x%x svms 0x%p [0x%llx 0x%llx] done, r=%d\n", p->pasid, &p->svms, start, start + size - 1, r); diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.h b/drivers/gpu/drm/amd/amdkfd/kfd_svm.h index 6dc91c33e80f..949b477e2f4c 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.h +++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.h @@ -48,6 +48,7 @@ struct svm_range_bo { struct work_struct eviction_work; struct svm_range_list *svms; uint32_t evicting; + struct work_struct release_work; }; enum svm_work_list_ops { @@ -75,8 +76,6 @@ struct svm_work_list_item { * aligned, page size is (last - start + 1) * @list: link list node, used to scan all ranges of svms * @update_list:link list node used to add to update_list - * @remove_list:link list node used to add to remove list - * @insert_list:link list node used to add to insert list * @mapping: bo_va mapping structure to create and update GPU page table * @npages: number of pages * @dma_addr: dma mapping address on each GPU for system memory physical page @@ -112,8 +111,6 @@ struct svm_range { struct interval_tree_node it_node; struct list_head list; struct list_head update_list; - struct list_head remove_list; - struct list_head insert_list; uint64_t npages; dma_addr_t *dma_addr[MAX_GPU_INSTANCE]; struct ttm_resource *ttm_res; @@ -195,7 +192,7 @@ void svm_range_list_lock_and_flush_work(struct svm_range_list *svms, struct mm_s */ #define KFD_IS_SVM_API_SUPPORTED(dev) ((dev)->pgmap.type != 0) -void svm_range_bo_unref(struct svm_range_bo *svm_bo); +void svm_range_bo_unref_async(struct svm_range_bo *svm_bo); #else struct kfd_process; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c index dd593ad0614a..948fbb39336e 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c @@ -113,7 +113,7 @@ struct kfd_dev *kfd_device_by_pci_dev(const struct pci_dev *pdev) return device; } -struct kfd_dev *kfd_device_by_kgd(const struct kgd_dev *kgd) +struct kfd_dev *kfd_device_by_adev(const struct amdgpu_device *adev) { struct kfd_topology_device *top_dev; struct kfd_dev *device = NULL; @@ -121,7 +121,7 @@ struct kfd_dev *kfd_device_by_kgd(const struct kgd_dev *kgd) down_read(&topology_lock); list_for_each_entry(top_dev, &topology_device_list, list) - if (top_dev->gpu && top_dev->gpu->kgd == kgd) { + if (top_dev->gpu && top_dev->gpu->adev == adev) { device = top_dev->gpu; break; } @@ -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 |= @@ -515,7 +515,7 @@ static ssize_t node_show(struct kobject *kobj, struct attribute *attr, HSA_CAP_WATCH_POINTS_TOTALBITS_MASK); } - if (dev->gpu->device_info->asic_family == CHIP_TONGA) + if (dev->gpu->adev->asic_type == CHIP_TONGA) dev->node_props.capability |= HSA_CAP_AQL_QUEUE_DOUBLE_MAP; @@ -531,7 +531,7 @@ static ssize_t node_show(struct kobject *kobj, struct attribute *attr, sysfs_show_32bit_prop(buffer, offs, "sdma_fw_version", dev->gpu->sdma_fw_version); sysfs_show_64bit_prop(buffer, offs, "unique_id", - amdgpu_amdkfd_get_unique_id(dev->gpu->kgd)); + dev->gpu->adev->unique_id); } @@ -1106,7 +1106,7 @@ static uint32_t kfd_generate_gpu_id(struct kfd_dev *gpu) if (!gpu) return 0; - amdgpu_amdkfd_get_local_mem_info(gpu->kgd, &local_mem_info); + amdgpu_amdkfd_get_local_mem_info(gpu->adev, &local_mem_info); local_mem_size = local_mem_info.local_mem_size_private + local_mem_info.local_mem_size_public; @@ -1189,7 +1189,7 @@ static void kfd_fill_mem_clk_max_info(struct kfd_topology_device *dev) * for APUs - If CRAT from ACPI reports more than one bank, then * all the banks will report the same mem_clk_max information */ - amdgpu_amdkfd_get_local_mem_info(dev->gpu->kgd, &local_mem_info); + amdgpu_amdkfd_get_local_mem_info(dev->gpu->adev, &local_mem_info); list_for_each_entry(mem, &dev->mem_props, list) mem->mem_clk_max = local_mem_info.mem_clk_max; @@ -1217,8 +1217,7 @@ static void kfd_set_iolink_no_atomics(struct kfd_topology_device *dev, /* set gpu (dev) flags. */ } else { if (!dev->gpu->pci_atomic_requested || - dev->gpu->device_info->asic_family == - CHIP_HAWAII) + dev->gpu->adev->asic_type == CHIP_HAWAII) link->flags |= CRAT_IOLINK_FLAGS_NO_ATOMICS_32_BIT | CRAT_IOLINK_FLAGS_NO_ATOMICS_64_BIT; } @@ -1239,7 +1238,7 @@ static void kfd_set_iolink_non_coherent(struct kfd_topology_device *to_dev, */ if (inbound_link->iolink_type == CRAT_IOLINK_TYPE_PCIEXPRESS || (inbound_link->iolink_type == CRAT_IOLINK_TYPE_XGMI && - to_dev->gpu->device_info->asic_family == CHIP_VEGA20)) { + KFD_GC_VERSION(to_dev->gpu) == IP_VERSION(9, 4, 0))) { outbound_link->flags |= CRAT_IOLINK_FLAGS_NON_COHERENT; inbound_link->flags |= CRAT_IOLINK_FLAGS_NON_COHERENT; } @@ -1286,7 +1285,8 @@ int kfd_topology_add_device(struct kfd_dev *gpu) void *crat_image = NULL; size_t image_size = 0; int proximity_domain; - struct amdgpu_device *adev; + int i; + const char *asic_name = amdgpu_asic_name[gpu->adev->asic_type]; INIT_LIST_HEAD(&temp_topology_device_list); @@ -1296,10 +1296,8 @@ int kfd_topology_add_device(struct kfd_dev *gpu) proximity_domain = atomic_inc_return(&topology_crat_proximity_domain); - adev = (struct amdgpu_device *)(gpu->kgd); - /* Include the CPU in xGMI hive if xGMI connected by assigning it the hive ID. */ - if (gpu->hive_id && adev->gmc.xgmi.connected_to_cpu) { + if (gpu->hive_id && gpu->adev->gmc.xgmi.connected_to_cpu) { struct kfd_topology_device *top_dev; down_read(&topology_lock); @@ -1372,45 +1370,48 @@ int kfd_topology_add_device(struct kfd_dev *gpu) * needed for the topology */ - amdgpu_amdkfd_get_cu_info(dev->gpu->kgd, &cu_info); + 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 |= - ((amdgpu_amdkfd_get_asic_rev_id(dev->gpu->kgd) << - HSA_CAP_ASIC_REVISION_SHIFT) & + ((dev->gpu->adev->rev_id << HSA_CAP_ASIC_REVISION_SHIFT) & HSA_CAP_ASIC_REVISION_MASK); dev->node_props.location_id = pci_dev_id(gpu->pdev); dev->node_props.domain = pci_domain_nr(gpu->pdev->bus); dev->node_props.max_engine_clk_fcompute = - amdgpu_amdkfd_get_max_engine_clock_in_mhz(dev->gpu->kgd); + amdgpu_amdkfd_get_max_engine_clock_in_mhz(dev->gpu->adev); dev->node_props.max_engine_clk_ccompute = cpufreq_quick_get_max(0) / 1000; dev->node_props.drm_render_minor = gpu->shared_resources.drm_render_minor; dev->node_props.hive_id = gpu->hive_id; - dev->node_props.num_sdma_engines = gpu->device_info->num_sdma_engines; + dev->node_props.num_sdma_engines = kfd_get_num_sdma_engines(gpu); dev->node_props.num_sdma_xgmi_engines = - gpu->device_info->num_xgmi_sdma_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) ? - amdgpu_amdkfd_get_num_gws(dev->gpu->kgd) : 0; + dev->gpu->adev->gds.gws_size : 0; dev->node_props.num_cp_queues = get_cp_queues_num(dev->gpu->dqm); kfd_fill_mem_clk_max_info(dev); kfd_fill_iolink_non_crat_info(dev); - switch (dev->gpu->device_info->asic_family) { + switch (dev->gpu->adev->asic_type) { case CHIP_KAVERI: case CHIP_HAWAII: case CHIP_TONGA: @@ -1429,30 +1430,14 @@ int kfd_topology_add_device(struct kfd_dev *gpu) HSA_CAP_DOORBELL_TYPE_TOTALBITS_SHIFT) & HSA_CAP_DOORBELL_TYPE_TOTALBITS_MASK); break; - case CHIP_VEGA10: - case CHIP_VEGA12: - case CHIP_VEGA20: - case CHIP_RAVEN: - case CHIP_RENOIR: - case CHIP_ARCTURUS: - case CHIP_ALDEBARAN: - case CHIP_NAVI10: - case CHIP_NAVI12: - case CHIP_NAVI14: - case CHIP_SIENNA_CICHLID: - case CHIP_NAVY_FLOUNDER: - case CHIP_VANGOGH: - case CHIP_DIMGREY_CAVEFISH: - case CHIP_BEIGE_GOBY: - case CHIP_YELLOW_CARP: - case CHIP_CYAN_SKILLFISH: - dev->node_props.capability |= ((HSA_CAP_DOORBELL_TYPE_2_0 << - HSA_CAP_DOORBELL_TYPE_TOTALBITS_SHIFT) & - HSA_CAP_DOORBELL_TYPE_TOTALBITS_MASK); - break; default: - WARN(1, "Unexpected ASIC family %u", - dev->gpu->device_info->asic_family); + if (KFD_GC_VERSION(dev->gpu) >= IP_VERSION(9, 0, 1)) + dev->node_props.capability |= ((HSA_CAP_DOORBELL_TYPE_2_0 << + HSA_CAP_DOORBELL_TYPE_TOTALBITS_SHIFT) & + HSA_CAP_DOORBELL_TYPE_TOTALBITS_MASK); + else + WARN(1, "Unexpected ASIC family %u", + dev->gpu->adev->asic_type); } /* @@ -1469,7 +1454,7 @@ int kfd_topology_add_device(struct kfd_dev *gpu) * because it doesn't consider masked out CUs * max_waves_per_simd: Carrizo reports wrong max_waves_per_simd */ - if (dev->gpu->device_info->asic_family == CHIP_CARRIZO) { + if (dev->gpu->adev->asic_type == CHIP_CARRIZO) { dev->node_props.simd_count = cu_info.simd_per_cu * cu_info.cu_active_number; dev->node_props.max_waves_per_simd = 10; @@ -1477,16 +1462,17 @@ int kfd_topology_add_device(struct kfd_dev *gpu) /* kfd only concerns sram ecc on GFX and HBM ecc on UMC */ dev->node_props.capability |= - ((adev->ras_enabled & BIT(AMDGPU_RAS_BLOCK__GFX)) != 0) ? + ((dev->gpu->adev->ras_enabled & BIT(AMDGPU_RAS_BLOCK__GFX)) != 0) ? HSA_CAP_SRAM_EDCSUPPORTED : 0; - dev->node_props.capability |= ((adev->ras_enabled & BIT(AMDGPU_RAS_BLOCK__UMC)) != 0) ? + dev->node_props.capability |= + ((dev->gpu->adev->ras_enabled & BIT(AMDGPU_RAS_BLOCK__UMC)) != 0) ? HSA_CAP_MEM_EDCSUPPORTED : 0; - if (adev->asic_type != CHIP_VEGA10) - dev->node_props.capability |= (adev->ras_enabled != 0) ? + if (KFD_GC_VERSION(dev->gpu) != IP_VERSION(9, 0, 1)) + dev->node_props.capability |= (dev->gpu->adev->ras_enabled != 0) ? HSA_CAP_RASEVENTNOTIFY : 0; - if (KFD_IS_SVM_API_SUPPORTED(adev->kfd.dev)) + if (KFD_IS_SVM_API_SUPPORTED(dev->gpu->adev->kfd.dev)) dev->node_props.capability |= HSA_CAP_SVMAPI_SUPPORTED; kfd_debug_print_topology(); @@ -1592,7 +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); diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.h b/drivers/gpu/drm/amd/amdkfd/kfd_topology.h index a8db017c9b8e..f0cc59d2fd5d 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_topology.h +++ b/drivers/gpu/drm/amd/amdkfd/kfd_topology.h @@ -25,38 +25,11 @@ #include <linux/types.h> #include <linux/list.h> +#include <linux/kfd_sysfs.h> #include "kfd_crat.h" #define KFD_TOPOLOGY_PUBLIC_NAME_SIZE 32 -#define HSA_CAP_HOT_PLUGGABLE 0x00000001 -#define HSA_CAP_ATS_PRESENT 0x00000002 -#define HSA_CAP_SHARED_WITH_GRAPHICS 0x00000004 -#define HSA_CAP_QUEUE_SIZE_POW2 0x00000008 -#define HSA_CAP_QUEUE_SIZE_32BIT 0x00000010 -#define HSA_CAP_QUEUE_IDLE_EVENT 0x00000020 -#define HSA_CAP_VA_LIMIT 0x00000040 -#define HSA_CAP_WATCH_POINTS_SUPPORTED 0x00000080 -#define HSA_CAP_WATCH_POINTS_TOTALBITS_MASK 0x00000f00 -#define HSA_CAP_WATCH_POINTS_TOTALBITS_SHIFT 8 -#define HSA_CAP_DOORBELL_TYPE_TOTALBITS_MASK 0x00003000 -#define HSA_CAP_DOORBELL_TYPE_TOTALBITS_SHIFT 12 - -#define HSA_CAP_DOORBELL_TYPE_PRE_1_0 0x0 -#define HSA_CAP_DOORBELL_TYPE_1_0 0x1 -#define HSA_CAP_DOORBELL_TYPE_2_0 0x2 -#define HSA_CAP_AQL_QUEUE_DOUBLE_MAP 0x00004000 - -#define HSA_CAP_RESERVED_WAS_SRAM_EDCSUPPORTED 0x00080000 /* Old buggy user mode depends on this being 0 */ -#define HSA_CAP_MEM_EDCSUPPORTED 0x00100000 -#define HSA_CAP_RASEVENTNOTIFY 0x00200000 -#define HSA_CAP_ASIC_REVISION_MASK 0x03c00000 -#define HSA_CAP_ASIC_REVISION_SHIFT 22 -#define HSA_CAP_SRAM_EDCSUPPORTED 0x04000000 -#define HSA_CAP_SVMAPI_SUPPORTED 0x08000000 -#define HSA_CAP_FLAGS_COHERENTHOSTACCESS 0x10000000 -#define HSA_CAP_RESERVED 0xe00f8000 - struct kfd_node_properties { uint64_t hive_id; uint32_t cpu_cores_count; @@ -93,17 +66,6 @@ struct kfd_node_properties { char name[KFD_TOPOLOGY_PUBLIC_NAME_SIZE]; }; -#define HSA_MEM_HEAP_TYPE_SYSTEM 0 -#define HSA_MEM_HEAP_TYPE_FB_PUBLIC 1 -#define HSA_MEM_HEAP_TYPE_FB_PRIVATE 2 -#define HSA_MEM_HEAP_TYPE_GPU_GDS 3 -#define HSA_MEM_HEAP_TYPE_GPU_LDS 4 -#define HSA_MEM_HEAP_TYPE_GPU_SCRATCH 5 - -#define HSA_MEM_FLAGS_HOT_PLUGGABLE 0x00000001 -#define HSA_MEM_FLAGS_NON_VOLATILE 0x00000002 -#define HSA_MEM_FLAGS_RESERVED 0xfffffffc - struct kfd_mem_properties { struct list_head list; uint32_t heap_type; @@ -116,12 +78,6 @@ struct kfd_mem_properties { struct attribute attr; }; -#define HSA_CACHE_TYPE_DATA 0x00000001 -#define HSA_CACHE_TYPE_INSTRUCTION 0x00000002 -#define HSA_CACHE_TYPE_CPU 0x00000004 -#define HSA_CACHE_TYPE_HSACU 0x00000008 -#define HSA_CACHE_TYPE_RESERVED 0xfffffff0 - struct kfd_cache_properties { struct list_head list; uint32_t processor_id_low; |