diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdkfd')
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 31 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_device.c | 56 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c | 7 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_migrate.c | 93 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 11 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_process.c | 4 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 40 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_topology.c | 5 |
8 files changed, 170 insertions, 77 deletions
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c index 828a9ceef1e7..0f0719528bcc 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c @@ -521,15 +521,10 @@ static int kfd_ioctl_set_cu_mask(struct file *filp, struct kfd_process *p, cu_mask_size = sizeof(uint32_t) * (max_num_cus/32); } - minfo.cu_mask.ptr = kzalloc(cu_mask_size, GFP_KERNEL); - if (!minfo.cu_mask.ptr) - return -ENOMEM; - - retval = copy_from_user(minfo.cu_mask.ptr, cu_mask_ptr, cu_mask_size); - if (retval) { + minfo.cu_mask.ptr = memdup_user(cu_mask_ptr, cu_mask_size); + if (IS_ERR(minfo.cu_mask.ptr)) { pr_debug("Could not copy CU mask from userspace"); - retval = -EFAULT; - goto out; + return PTR_ERR(minfo.cu_mask.ptr); } mutex_lock(&p->mutex); @@ -538,7 +533,6 @@ static int kfd_ioctl_set_cu_mask(struct file *filp, struct kfd_process *p, mutex_unlock(&p->mutex); -out: kfree(minfo.cu_mask.ptr); return retval; } @@ -1070,7 +1064,12 @@ static int kfd_ioctl_alloc_memory_of_gpu(struct file *filep, svm_range_list_lock_and_flush_work(&p->svms, current->mm); mutex_lock(&p->svms.lock); mmap_write_unlock(current->mm); - if (interval_tree_iter_first(&p->svms.objects, + + /* Skip a special case that allocates VRAM without VA, + * VA will be invalid of 0. + */ + if (!(!args->va_addr && (flags & KFD_IOC_ALLOC_MEM_FLAGS_VRAM)) && + interval_tree_iter_first(&p->svms.objects, args->va_addr >> PAGE_SHIFT, (args->va_addr + args->size - 1) >> PAGE_SHIFT)) { pr_err("Address: 0x%llx already allocated by SVM\n", @@ -2566,8 +2565,8 @@ static int criu_restore(struct file *filep, pr_debug("CRIU restore (num_devices:%u num_bos:%u num_objects:%u priv_data_size:%llu)\n", args->num_devices, args->num_bos, args->num_objects, args->priv_data_size); - if (!args->bos || !args->devices || !args->priv_data || !args->priv_data_size || - !args->num_devices || !args->num_bos) + if ((args->num_bos > 0 && !args->bos) || !args->devices || !args->priv_data || + !args->priv_data_size || !args->num_devices) return -EINVAL; mutex_lock(&p->mutex); @@ -3252,8 +3251,10 @@ static long kfd_ioctl(struct file *filep, unsigned int cmd, unsigned long arg) int retcode = -EINVAL; bool ptrace_attached = false; - if (nr >= AMDKFD_CORE_IOCTL_COUNT) + if (nr >= AMDKFD_CORE_IOCTL_COUNT) { + retcode = -ENOTTY; goto err_i1; + } if ((nr >= AMDKFD_COMMAND_START) && (nr < AMDKFD_COMMAND_END)) { u32 amdkfd_size; @@ -3266,8 +3267,10 @@ static long kfd_ioctl(struct file *filep, unsigned int cmd, unsigned long arg) asize = amdkfd_size; cmd = ioctl->cmd; - } else + } else { + retcode = -ENOTTY; goto err_i1; + } dev_dbg(kfd_device, "ioctl cmd 0x%x (#0x%x), arg 0x%lx\n", cmd, nr, arg); diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c index 7e749f9b6d69..e9cfb80bd436 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c @@ -495,6 +495,7 @@ struct kfd_dev *kgd2kfd_probe(struct amdgpu_device *adev, bool vf) mutex_init(&kfd->doorbell_mutex); ida_init(&kfd->doorbell_ida); + atomic_set(&kfd->kfd_processes_count, 0); return kfd; } @@ -1133,7 +1134,15 @@ void kgd2kfd_interrupt(struct kfd_dev *kfd, const void *ih_ring_entry) } for (i = 0; i < kfd->num_nodes; i++) { - node = kfd->nodes[i]; + /* Race if another thread in b/w + * kfd_cleanup_nodes and kfree(kfd), + * when kfd->nodes[i] = NULL + */ + if (kfd->nodes[i]) + node = kfd->nodes[i]; + else + return; + spin_lock_irqsave(&node->interrupt_lock, flags); if (node->interrupts_active @@ -1485,6 +1494,15 @@ int kgd2kfd_check_and_lock_kfd(struct kfd_dev *kfd) mutex_lock(&kfd_processes_mutex); + /* kfd_processes_count is per kfd_dev, return -EBUSY without + * further check + */ + if (!!atomic_read(&kfd->kfd_processes_count)) { + pr_debug("process_wq_release not finished\n"); + r = -EBUSY; + goto out; + } + if (hash_empty(kfd_processes_table) && !kfd_is_locked(kfd)) goto out; @@ -1550,6 +1568,25 @@ int kgd2kfd_start_sched(struct kfd_dev *kfd, uint32_t node_id) return ret; } +int kgd2kfd_start_sched_all_nodes(struct kfd_dev *kfd) +{ + struct kfd_node *node; + int i, r; + + if (!kfd->init_complete) + return 0; + + for (i = 0; i < kfd->num_nodes; i++) { + node = kfd->nodes[i]; + r = node->dqm->ops.unhalt(node->dqm); + if (r) { + dev_err(kfd_device, "Error in starting scheduler\n"); + return r; + } + } + return 0; +} + int kgd2kfd_stop_sched(struct kfd_dev *kfd, uint32_t node_id) { struct kfd_node *node; @@ -1567,6 +1604,23 @@ int kgd2kfd_stop_sched(struct kfd_dev *kfd, uint32_t node_id) return node->dqm->ops.halt(node->dqm); } +int kgd2kfd_stop_sched_all_nodes(struct kfd_dev *kfd) +{ + struct kfd_node *node; + int i, r; + + if (!kfd->init_complete) + return 0; + + for (i = 0; i < kfd->num_nodes; i++) { + node = kfd->nodes[i]; + r = node->dqm->ops.halt(node->dqm); + if (r) + return r; + } + return 0; +} + bool kgd2kfd_compute_active(struct kfd_dev *kfd, uint32_t node_id) { struct kfd_node *node; diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c b/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c index 2b0a830f5b29..fb3129883a4c 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c @@ -46,11 +46,7 @@ static bool kq_initialize(struct kernel_queue *kq, struct kfd_node *dev, int retval; union PM4_MES_TYPE_3_HEADER nop; - if (WARN_ON(type != KFD_QUEUE_TYPE_DIQ && type != KFD_QUEUE_TYPE_HIQ)) - return false; - - pr_debug("Initializing queue type %d size %d\n", KFD_QUEUE_TYPE_HIQ, - queue_size); + pr_debug("Initializing queue type %d size %d\n", type, queue_size); memset(&prop, 0, sizeof(prop)); memset(&nop, 0, sizeof(nop)); @@ -69,6 +65,7 @@ static bool kq_initialize(struct kernel_queue *kq, struct kfd_node *dev, kq->mqd_mgr = dev->dqm->mqd_mgrs[KFD_MQD_TYPE_HIQ]; break; default: + WARN(1, "Invalid queue type %d\n", type); dev_err(dev->adev->dev, "Invalid queue type %d\n", type); return false; } diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c index 79251f22b702..59a5a3fea65d 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c @@ -39,22 +39,22 @@ #endif #define dev_fmt(fmt) "kfd_migrate: " fmt -static uint64_t -svm_migrate_direct_mapping_addr(struct amdgpu_device *adev, uint64_t addr) +static u64 +svm_migrate_direct_mapping_addr(struct amdgpu_device *adev, u64 addr) { return addr + amdgpu_ttm_domain_start(adev, TTM_PL_VRAM); } static int -svm_migrate_gart_map(struct amdgpu_ring *ring, uint64_t npages, - dma_addr_t *addr, uint64_t *gart_addr, uint64_t flags) +svm_migrate_gart_map(struct amdgpu_ring *ring, u64 npages, + dma_addr_t *addr, u64 *gart_addr, u64 flags) { struct amdgpu_device *adev = ring->adev; struct amdgpu_job *job; unsigned int num_dw, num_bytes; struct dma_fence *fence; - uint64_t src_addr, dst_addr; - uint64_t pte_flags; + u64 src_addr, dst_addr; + u64 pte_flags; void *cpu_addr; int r; @@ -68,7 +68,8 @@ svm_migrate_gart_map(struct amdgpu_ring *ring, uint64_t npages, AMDGPU_FENCE_OWNER_UNDEFINED, num_dw * 4 + num_bytes, AMDGPU_IB_POOL_DELAYED, - &job); + &job, + AMDGPU_KERNEL_JOB_ID_KFD_GART_MAP); if (r) return r; @@ -122,15 +123,15 @@ svm_migrate_gart_map(struct amdgpu_ring *ring, uint64_t npages, static int svm_migrate_copy_memory_gart(struct amdgpu_device *adev, dma_addr_t *sys, - uint64_t *vram, uint64_t npages, + u64 *vram, u64 npages, enum MIGRATION_COPY_DIR direction, struct dma_fence **mfence) { - const uint64_t GTT_MAX_PAGES = AMDGPU_GTT_MAX_TRANSFER_SIZE; + const u64 GTT_MAX_PAGES = AMDGPU_GTT_MAX_TRANSFER_SIZE; struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring; - uint64_t gart_s, gart_d; + u64 gart_s, gart_d; struct dma_fence *next; - uint64_t size; + u64 size; int r; mutex_lock(&adev->mman.gtt_window_lock); @@ -260,39 +261,39 @@ static void svm_migrate_put_sys_page(unsigned long addr) put_page(page); } -static unsigned long svm_migrate_unsuccessful_pages(struct migrate_vma *migrate) +static unsigned long svm_migrate_successful_pages(struct migrate_vma *migrate) { - unsigned long upages = 0; + unsigned long mpages = 0; unsigned long i; for (i = 0; i < migrate->npages; i++) { - if (migrate->src[i] & MIGRATE_PFN_VALID && - !(migrate->src[i] & MIGRATE_PFN_MIGRATE)) - upages++; + if (migrate->dst[i] & MIGRATE_PFN_VALID && + migrate->src[i] & MIGRATE_PFN_MIGRATE) + mpages++; } - return upages; + return mpages; } static int svm_migrate_copy_to_vram(struct kfd_node *node, struct svm_range *prange, struct migrate_vma *migrate, struct dma_fence **mfence, - dma_addr_t *scratch, uint64_t ttm_res_offset) + dma_addr_t *scratch, u64 ttm_res_offset) { - uint64_t npages = migrate->npages; + u64 npages = migrate->npages; struct amdgpu_device *adev = node->adev; struct device *dev = adev->dev; struct amdgpu_res_cursor cursor; - uint64_t mpages = 0; + u64 mpages = 0; dma_addr_t *src; - uint64_t *dst; - uint64_t i, j; + u64 *dst; + u64 i, j; int r; pr_debug("svms 0x%p [0x%lx 0x%lx 0x%llx]\n", prange->svms, prange->start, prange->last, ttm_res_offset); src = scratch; - dst = (uint64_t *)(scratch + npages); + dst = (u64 *)(scratch + npages); amdgpu_res_first(prange->ttm_res, ttm_res_offset, npages << PAGE_SHIFT, &cursor); @@ -385,11 +386,11 @@ out_free_vram_pages: static long svm_migrate_vma_to_vram(struct kfd_node *node, struct svm_range *prange, - struct vm_area_struct *vma, uint64_t start, - uint64_t end, uint32_t trigger, uint64_t ttm_res_offset) + struct vm_area_struct *vma, u64 start, + u64 end, uint32_t trigger, u64 ttm_res_offset) { struct kfd_process *p = container_of(prange->svms, struct kfd_process, svms); - uint64_t npages = (end - start) >> PAGE_SHIFT; + u64 npages = (end - start) >> PAGE_SHIFT; struct amdgpu_device *adev = node->adev; struct kfd_process_device *pdd; struct dma_fence *mfence = NULL; @@ -408,7 +409,7 @@ svm_migrate_vma_to_vram(struct kfd_node *node, struct svm_range *prange, migrate.pgmap_owner = SVM_ADEV_PGMAP_OWNER(adev); buf = kvcalloc(npages, - 2 * sizeof(*migrate.src) + sizeof(uint64_t) + sizeof(dma_addr_t), + 2 * sizeof(*migrate.src) + sizeof(u64) + sizeof(dma_addr_t), GFP_KERNEL); if (!buf) goto out; @@ -447,9 +448,9 @@ svm_migrate_vma_to_vram(struct kfd_node *node, struct svm_range *prange, svm_migrate_copy_done(adev, mfence); migrate_vma_finalize(&migrate); - mpages = cpages - svm_migrate_unsuccessful_pages(&migrate); - pr_debug("successful/cpages/npages 0x%lx/0x%lx/0x%lx\n", - mpages, cpages, migrate.npages); + mpages = svm_migrate_successful_pages(&migrate); + pr_debug("migrated/collected/requested 0x%lx/0x%lx/0x%lx\n", + mpages, cpages, migrate.npages); svm_range_dma_unmap_dev(adev->dev, scratch, 0, npages); @@ -490,7 +491,7 @@ svm_migrate_ram_to_vram(struct svm_range *prange, uint32_t best_loc, { unsigned long addr, start, end; struct vm_area_struct *vma; - uint64_t ttm_res_offset; + u64 ttm_res_offset; struct kfd_node *node; unsigned long mpages = 0; long r = 0; @@ -580,14 +581,14 @@ static void svm_migrate_page_free(struct page *page) static int svm_migrate_copy_to_ram(struct amdgpu_device *adev, struct svm_range *prange, struct migrate_vma *migrate, struct dma_fence **mfence, - dma_addr_t *scratch, uint64_t npages) + dma_addr_t *scratch, u64 npages) { struct device *dev = adev->dev; - uint64_t *src; + u64 *src; dma_addr_t *dst; struct page *dpage; - uint64_t i = 0, j; - uint64_t addr; + u64 i = 0, j; + u64 addr; int r = 0; pr_debug("svms 0x%p [0x%lx 0x%lx]\n", prange->svms, prange->start, @@ -595,7 +596,7 @@ svm_migrate_copy_to_ram(struct amdgpu_device *adev, struct svm_range *prange, addr = migrate->start; - src = (uint64_t *)(scratch + npages); + src = (u64 *)(scratch + npages); dst = scratch; for (i = 0, j = 0; i < npages; i++, addr += PAGE_SIZE) { @@ -683,12 +684,11 @@ out_oom: */ static long svm_migrate_vma_to_ram(struct kfd_node *node, struct svm_range *prange, - struct vm_area_struct *vma, uint64_t start, uint64_t end, + struct vm_area_struct *vma, u64 start, u64 end, uint32_t trigger, struct page *fault_page) { struct kfd_process *p = container_of(prange->svms, struct kfd_process, svms); - uint64_t npages = (end - start) >> PAGE_SHIFT; - unsigned long upages = npages; + u64 npages = (end - start) >> PAGE_SHIFT; unsigned long cpages = 0; unsigned long mpages = 0; struct amdgpu_device *adev = node->adev; @@ -710,7 +710,7 @@ svm_migrate_vma_to_ram(struct kfd_node *node, struct svm_range *prange, migrate.flags = MIGRATE_VMA_SELECT_DEVICE_PRIVATE; buf = kvcalloc(npages, - 2 * sizeof(*migrate.src) + sizeof(uint64_t) + sizeof(dma_addr_t), + 2 * sizeof(*migrate.src) + sizeof(u64) + sizeof(dma_addr_t), GFP_KERNEL); if (!buf) goto out; @@ -736,7 +736,6 @@ svm_migrate_vma_to_ram(struct kfd_node *node, struct svm_range *prange, if (!cpages) { pr_debug("failed collect migrate device pages [0x%lx 0x%lx]\n", prange->start, prange->last); - upages = svm_migrate_unsuccessful_pages(&migrate); goto out_free; } if (cpages != npages) @@ -749,9 +748,9 @@ svm_migrate_vma_to_ram(struct kfd_node *node, struct svm_range *prange, scratch, npages); migrate_vma_pages(&migrate); - upages = svm_migrate_unsuccessful_pages(&migrate); - pr_debug("unsuccessful/cpages/npages 0x%lx/0x%lx/0x%lx\n", - upages, cpages, migrate.npages); + mpages = svm_migrate_successful_pages(&migrate); + pr_debug("migrated/collected/requested 0x%lx/0x%lx/0x%lx\n", + mpages, cpages, migrate.npages); svm_migrate_copy_done(adev, mfence); migrate_vma_finalize(&migrate); @@ -764,8 +763,7 @@ out_free: start >> PAGE_SHIFT, end >> PAGE_SHIFT, node->id, 0, trigger, r); out: - if (!r && cpages) { - mpages = cpages - upages; + if (!r && mpages) { pdd = svm_range_get_pdd_by_node(prange, node); if (pdd) WRITE_ONCE(pdd->page_out, pdd->page_out + mpages); @@ -848,6 +846,9 @@ int svm_migrate_vram_to_ram(struct svm_range *prange, struct mm_struct *mm, } if (r >= 0) { + WARN_ONCE(prange->vram_pages < mpages, + "Recorded vram pages(0x%llx) should not be less than migration pages(0x%lx).", + prange->vram_pages, mpages); prange->vram_pages -= mpages; /* prange does not have vram page set its actual_loc to system diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h index 67694bcd9464..70ef051511bb 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h +++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h @@ -111,7 +111,14 @@ #define KFD_KERNEL_QUEUE_SIZE 2048 -#define KFD_UNMAP_LATENCY_MS (4000) +/* KFD_UNMAP_LATENCY_MS is the timeout CP waiting for SDMA preemption. One XCC + * can be associated to 2 SDMA engines. queue_preemption_timeout_ms is the time + * driver waiting for CP returning the UNMAP_QUEUE fence. Thus the math is + * queue_preemption_timeout_ms = sdma_preemption_time * 2 + cp workload + * The format here makes CP workload 10% of total timeout + */ +#define KFD_UNMAP_LATENCY_MS \ + ((queue_preemption_timeout_ms - queue_preemption_timeout_ms / 10) >> 1) #define KFD_MAX_SDMA_QUEUES 128 @@ -375,6 +382,8 @@ struct kfd_dev { /* for dynamic partitioning */ int kfd_dev_lock; + + atomic_t kfd_processes_count; }; enum kfd_mempool { diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c index 5be28c6c4f6a..ddfe30c13e9d 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c @@ -1088,6 +1088,8 @@ static void kfd_process_destroy_pdds(struct kfd_process *p) pdd->runtime_inuse = false; } + atomic_dec(&pdd->dev->kfd->kfd_processes_count); + kfree(pdd); p->pdds[i] = NULL; } @@ -1649,6 +1651,8 @@ struct kfd_process_device *kfd_create_process_device_data(struct kfd_node *dev, /* Init idr used for memory handle translation */ idr_init(&pdd->alloc_idr); + atomic_inc(&dev->kfd->kfd_processes_count); + return pdd; } diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c index a0f22ea6d15a..273f42e3afdd 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c @@ -1189,7 +1189,7 @@ svm_nodes_in_same_hive(struct kfd_node *node_a, struct kfd_node *node_b) } static uint64_t -svm_range_get_pte_flags(struct kfd_node *node, +svm_range_get_pte_flags(struct kfd_node *node, struct amdgpu_vm *vm, struct svm_range *prange, int domain) { struct kfd_node *bo_node; @@ -1292,10 +1292,6 @@ svm_range_get_pte_flags(struct kfd_node *node, AMDGPU_VM_MTYPE_UC : AMDGPU_VM_MTYPE_NC; } - mapping_flags |= AMDGPU_VM_PAGE_READABLE | AMDGPU_VM_PAGE_WRITEABLE; - - if (flags & KFD_IOCTL_SVM_FLAG_GPU_RO) - mapping_flags &= ~AMDGPU_VM_PAGE_WRITEABLE; if (flags & KFD_IOCTL_SVM_FLAG_GPU_EXEC) mapping_flags |= AMDGPU_VM_PAGE_EXECUTABLE; @@ -1305,7 +1301,10 @@ svm_range_get_pte_flags(struct kfd_node *node, if (gc_ip_version >= IP_VERSION(12, 0, 0)) pte_flags |= AMDGPU_PTE_IS_PTE; - pte_flags |= amdgpu_gem_va_map_flags(node->adev, mapping_flags); + amdgpu_gmc_get_vm_pte(node->adev, vm, NULL, mapping_flags, &pte_flags); + pte_flags |= AMDGPU_PTE_READABLE; + if (!(flags & KFD_IOCTL_SVM_FLAG_GPU_RO)) + pte_flags |= AMDGPU_PTE_WRITEABLE; return pte_flags; } @@ -1412,7 +1411,7 @@ svm_range_map_to_gpu(struct kfd_process_device *pdd, struct svm_range *prange, pr_debug("Mapping range [0x%lx 0x%llx] on domain: %s\n", last_start, prange->start + i, last_domain ? "GPU" : "CPU"); - pte_flags = svm_range_get_pte_flags(pdd->dev, prange, last_domain); + pte_flags = svm_range_get_pte_flags(pdd->dev, vm, prange, last_domain); if (readonly) pte_flags &= ~AMDGPU_PTE_WRITEABLE; @@ -1714,9 +1713,32 @@ static int svm_range_validate_and_map(struct mm_struct *mm, next = min(vma->vm_end, end); npages = (next - addr) >> PAGE_SHIFT; + /* HMM requires at least READ permissions. If provided with PROT_NONE, + * unmap the memory. If it's not already mapped, this is a no-op + * If PROT_WRITE is provided without READ, warn first then unmap + */ + if (!(vma->vm_flags & VM_READ)) { + unsigned long e, s; + + svm_range_lock(prange); + if (vma->vm_flags & VM_WRITE) + pr_debug("VM_WRITE without VM_READ is not supported"); + s = max(start, prange->start); + e = min(end, prange->last); + if (e >= s) + r = svm_range_unmap_from_gpus(prange, s, e, + KFD_SVM_UNMAP_TRIGGER_UNMAP_FROM_CPU); + svm_range_unlock(prange); + /* If unmap returns non-zero, we'll bail on the next for loop + * iteration, so just leave r and continue + */ + addr = next; + continue; + } + WRITE_ONCE(p->svms.faulting_task, current); r = amdgpu_hmm_range_get_pages(&prange->notifier, addr, npages, - readonly, owner, NULL, + readonly, owner, &hmm_range); WRITE_ONCE(p->svms.faulting_task, NULL); if (r) @@ -4239,7 +4261,7 @@ svm_ioctl(struct kfd_process *p, enum kfd_ioctl_svm_op op, uint64_t start, r = svm_range_get_attr(p, mm, start, size, nattrs, attrs); break; default: - r = EINVAL; + r = -EINVAL; break; } diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c index 4ec73f33535e..5c98746eb72d 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c @@ -530,6 +530,8 @@ static ssize_t node_show(struct kobject *kobj, struct attribute *attr, sysfs_show_32bit_prop(buffer, offs, "sdma_fw_version", dev->gpu->kfd->sdma_fw_version); sysfs_show_64bit_prop(buffer, offs, "unique_id", + dev->gpu->xcp ? + dev->gpu->xcp->unique_id : dev->gpu->adev->unique_id); sysfs_show_32bit_prop(buffer, offs, "num_xcc", NUM_XCC(dev->gpu->xcc_mask)); @@ -1587,7 +1589,8 @@ static int kfd_dev_create_p2p_links(void) break; if (!dev->gpu || !dev->gpu->adev || (dev->gpu->kfd->hive_id && - dev->gpu->kfd->hive_id == new_dev->gpu->kfd->hive_id)) + dev->gpu->kfd->hive_id == new_dev->gpu->kfd->hive_id && + amdgpu_xgmi_get_is_sharing_enabled(dev->gpu->adev, new_dev->gpu->adev))) goto next; /* check if node(s) is/are peer accessible in one direction or bi-direction */ |