diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdkfd')
25 files changed, 364 insertions, 560 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..f187596faf66 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_crat.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_crat.c @@ -1340,7 +1340,7 @@ 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 +1377,71 @@ 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 +1967,6 @@ 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 +1983,7 @@ 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 +1992,19 @@ 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 +2046,11 @@ 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 +2116,7 @@ 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 +2147,7 @@ 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..e1294fba0c26 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c @@ -55,7 +55,6 @@ 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, @@ -69,13 +68,10 @@ static const struct kfd_device_info kaveri_device_info = { .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, @@ -89,13 +85,10 @@ static const struct kfd_device_info carrizo_device_info = { .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, @@ -108,15 +101,12 @@ static const struct kfd_device_info raven_device_info = { .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, @@ -130,14 +120,11 @@ static const struct kfd_device_info hawaii_device_info = { .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, @@ -150,13 +137,10 @@ static const struct kfd_device_info tonga_device_info = { .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, @@ -169,13 +153,10 @@ static const struct kfd_device_info fiji_device_info = { .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, @@ -188,14 +169,11 @@ static const struct kfd_device_info fiji_vf_device_info = { .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, @@ -208,13 +186,10 @@ static const struct kfd_device_info polaris10_device_info = { .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, @@ -227,13 +202,10 @@ static const struct kfd_device_info polaris10_vf_device_info = { .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, @@ -246,13 +218,10 @@ static const struct kfd_device_info polaris11_device_info = { .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, @@ -265,13 +234,10 @@ static const struct kfd_device_info polaris12_device_info = { .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, @@ -284,13 +250,10 @@ static const struct kfd_device_info vegam_device_info = { .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, @@ -303,13 +266,10 @@ static const struct kfd_device_info vega10_device_info = { .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, @@ -322,13 +282,10 @@ static const struct kfd_device_info vega10_vf_device_info = { .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, @@ -341,13 +298,10 @@ static const struct kfd_device_info vega12_device_info = { .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, @@ -360,13 +314,10 @@ static const struct kfd_device_info vega20_device_info = { .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, @@ -379,13 +330,10 @@ static const struct kfd_device_info arcturus_device_info = { .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, @@ -398,13 +346,10 @@ static const struct kfd_device_info aldebaran_device_info = { .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, @@ -417,13 +362,10 @@ static const struct kfd_device_info renoir_device_info = { .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, @@ -437,13 +379,10 @@ static const struct kfd_device_info navi10_device_info = { .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, @@ -457,13 +396,10 @@ static const struct kfd_device_info navi12_device_info = { .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, @@ -477,13 +413,10 @@ static const struct kfd_device_info navi14_device_info = { .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, @@ -497,13 +430,10 @@ static const struct kfd_device_info sienna_cichlid_device_info = { .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, @@ -517,13 +447,10 @@ static const struct kfd_device_info navy_flounder_device_info = { .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, @@ -537,13 +464,10 @@ static const struct kfd_device_info vangogh_device_info = { .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, @@ -557,13 +481,10 @@ static const struct kfd_device_info dimgrey_cavefish_device_info = { .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, @@ -577,13 +498,10 @@ static const struct kfd_device_info beige_goby_device_info = { .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, @@ -597,13 +515,10 @@ static const struct kfd_device_info yellow_carp_device_info = { .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, @@ -616,8 +531,6 @@ static const struct kfd_device_info cyan_skillfish_device_info = { .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, }; @@ -627,12 +540,11 @@ 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) +struct kfd_dev *kgd2kfd_probe(struct amdgpu_device *adev, bool vf) { struct kfd_dev *kfd; const struct kfd_device_info *device_info; const struct kfd2kgd_calls *f2g; - struct amdgpu_device *adev = (struct amdgpu_device *)kgd; struct pci_dev *pdev = adev->pdev; switch (adev->asic_type) { @@ -815,8 +727,12 @@ struct kfd_dev *kgd2kfd_probe(struct kgd_dev *kgd, bool vf) } if (!device_info || !f2g) { - dev_err(kfd_device, "%s %s not supported in kfd\n", - amdgpu_asic_name[adev->asic_type], vf ? "VF" : ""); + 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; } @@ -824,7 +740,7 @@ struct kfd_dev *kgd2kfd_probe(struct kgd_dev *kgd, bool vf) if (!kfd) return NULL; - kfd->kgd = kgd; + kfd->adev = adev; kfd->device_info = device_info; kfd->pdev = pdev; kfd->init_complete = false; @@ -845,23 +761,23 @@ 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 (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 +798,17 @@ 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 +825,11 @@ 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; @@ -927,7 +842,7 @@ 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 || @@ -959,10 +874,9 @@ bool kgd2kfd_device_init(struct kfd_dev *kfd, * 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 +888,7 @@ 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 +909,9 @@ 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 +929,7 @@ 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 +944,7 @@ 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 +982,10 @@ 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 +1002,9 @@ 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); @@ -1526,7 +1440,7 @@ 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 +1448,7 @@ 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 +1458,26 @@ 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 93e33dd84dd4..2af2b3268171 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c @@ -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"); @@ -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) @@ -1017,7 +1007,7 @@ static int start_nocpsch(struct device_queue_manager *dqm) pr_info("SW scheduler is used"); init_interrupts(dqm); - if (dqm->dev->device_info->asic_family == CHIP_HAWAII) + if (dqm->dev->adev->asic_type == CHIP_HAWAII) return pm_init(&dqm->packet_mgr, dqm); dqm->sched_running = true; @@ -1026,7 +1016,7 @@ static int start_nocpsch(struct device_queue_manager *dqm) 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 +1045,9 @@ 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 +1062,11 @@ 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 +1123,7 @@ 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; @@ -1850,7 +1841,7 @@ static int allocate_hiq_sdma_mqd(struct device_queue_manager *dqm) 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); @@ -1867,7 +1858,7 @@ 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 @@ -1930,7 +1921,7 @@ 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; @@ -1952,31 +1943,16 @@ 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)) @@ -2000,7 +1976,7 @@ 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) @@ -2031,7 +2007,7 @@ 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) @@ -2070,7 +2046,7 @@ 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) { @@ -2092,7 +2068,7 @@ 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; @@ -2109,7 +2085,7 @@ int dqm_debugfs_hqds(struct seq_file *m, void *data) 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_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_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..20512a4e9a91 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_int_process_v9.c @@ -231,7 +231,7 @@ static void event_interrupt_wq_v9(struct kfd_dev *dev, 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); + amdgpu_amdkfd_ras_poison_consumption_handler(dev->adev); return; } break; @@ -253,7 +253,7 @@ static void event_interrupt_wq_v9(struct kfd_dev *dev, 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); + amdgpu_amdkfd_ras_poison_consumption_handler(dev->adev); return; } } else if (client_id == SOC15_IH_CLIENTID_VMC || diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c b/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c index 64b4ac339904..406479a369a9 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; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c index 9b9c2b9bf2ef..d84cec0022b1 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c @@ -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..7b4118915bf6 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c @@ -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_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h index 8fd48d0ed240..7ea528941951 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, @@ -194,7 +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; @@ -208,11 +208,12 @@ 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; @@ -228,7 +229,7 @@ struct kfd_vmid_info { }; struct kfd_dev { - struct kgd_dev *kgd; + struct amdgpu_device *adev; const struct kfd_device_info *device_info; struct pci_dev *pdev; @@ -891,7 +892,7 @@ 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 +985,7 @@ 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..d4c8a6948a9f 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c @@ -288,7 +288,7 @@ 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 */ @@ -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); } @@ -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) @@ -1431,7 +1430,7 @@ 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 = @@ -1547,7 +1546,7 @@ 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 +1778,13 @@ 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 +1949,10 @@ 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..4f8464658daf 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); 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 3cb4681c5f53..755265f6c53b 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c @@ -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; @@ -581,7 +579,7 @@ 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 +591,7 @@ 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; @@ -1053,8 +1051,8 @@ 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 +1068,7 @@ 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 +1127,6 @@ 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 +1142,9 @@ 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 +1156,7 @@ 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); } @@ -1243,8 +1240,7 @@ 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 +1253,6 @@ 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 +1271,18 @@ 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 +1316,6 @@ 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 +1327,6 @@ 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 +1348,9 @@ 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 +1373,10 @@ 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); } /* @@ -1962,7 +1952,6 @@ 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; @@ -1980,9 +1969,9 @@ restart: 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(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) @@ -2304,7 +2293,7 @@ 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; @@ -2481,7 +2470,7 @@ 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; @@ -2548,7 +2537,7 @@ 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; } @@ -2980,7 +2969,6 @@ 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; @@ -3008,12 +2996,11 @@ 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; } diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c index dd593ad0614a..2d44b26b6657 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; } @@ -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 +1285,6 @@ 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; INIT_LIST_HEAD(&temp_topology_device_list); @@ -1296,10 +1294,8 @@ 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); @@ -1372,7 +1368,7 @@ 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); @@ -1384,33 +1380,32 @@ int kfd_topology_add_device(struct kfd_dev *gpu) 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; 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 +1424,14 @@ 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 +1448,7 @@ 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 +1456,17 @@ 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(); 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; |