diff options
Diffstat (limited to 'drivers/gpu/drm/amd/amdkfd')
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 19 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdkfd/kfd_device.c | 20 | ||||
| -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 | 25 |
5 files changed, 72 insertions, 7 deletions
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c index 828a9ceef1e7..8535a52a62ca 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c @@ -1070,7 +1070,12 @@ 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 +2571,8 @@ 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 +3257,10 @@ 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 +3273,10 @@ 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 349c351e242b..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; 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 3d8b20828c06..827507cfed7a 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c @@ -1714,6 +1714,29 @@ 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, @@ -3023,6 +3046,8 @@ retry_write_locked: if (svms->checkpoint_ts[gpuidx] != 0) { if (amdgpu_ih_ts_after_or_equal(ts, svms->checkpoint_ts[gpuidx])) { pr_debug("draining retry fault, drop fault 0x%llx\n", addr); + if (write_locked) + mmap_write_downgrade(mm); r = -EAGAIN; goto out_unlock_svms; } else { |